diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c index fd86a75416e6..a6930cec517b 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c @@ -1,303 +1,303 @@ /* * Copyright (c) 2020 iXsystems, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include int zfs_file_open(const char *path, int flags, int mode, zfs_file_t **fpp) { struct thread *td; int rc, fd; td = curthread; pwd_ensure_dirs(); /* 12.x doesn't take a const char * */ rc = kern_openat(td, AT_FDCWD, __DECONST(char *, path), UIO_SYSSPACE, flags, mode); if (rc) return (SET_ERROR(rc)); fd = td->td_retval[0]; td->td_retval[0] = 0; if (fget(curthread, fd, &cap_no_rights, fpp)) kern_close(td, fd); return (0); } void zfs_file_close(zfs_file_t *fp) { fo_close(fp, curthread); } static int zfs_file_write_impl(zfs_file_t *fp, const void *buf, size_t count, loff_t *offp, ssize_t *resid) { ssize_t rc; struct uio auio; struct thread *td; struct iovec aiov; td = curthread; aiov.iov_base = (void *)(uintptr_t)buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_segflg = UIO_SYSSPACE; auio.uio_resid = count; auio.uio_rw = UIO_WRITE; auio.uio_td = td; auio.uio_offset = *offp; if ((fp->f_flag & FWRITE) == 0) return (SET_ERROR(EBADF)); if (fp->f_type == DTYPE_VNODE) bwillwrite(); rc = fo_write(fp, &auio, td->td_ucred, FOF_OFFSET, td); if (rc) return (SET_ERROR(rc)); if (resid) *resid = auio.uio_resid; else if (auio.uio_resid) return (SET_ERROR(EIO)); *offp += count - auio.uio_resid; return (rc); } int zfs_file_write(zfs_file_t *fp, const void *buf, size_t count, ssize_t *resid) { loff_t off = fp->f_offset; ssize_t rc; rc = zfs_file_write_impl(fp, buf, count, &off, resid); if (rc == 0) fp->f_offset = off; return (SET_ERROR(rc)); } int zfs_file_pwrite(zfs_file_t *fp, const void *buf, size_t count, loff_t off, ssize_t *resid) { return (zfs_file_write_impl(fp, buf, count, &off, resid)); } static int zfs_file_read_impl(zfs_file_t *fp, void *buf, size_t count, loff_t *offp, ssize_t *resid) { ssize_t rc; struct uio auio; struct thread *td; struct iovec aiov; td = curthread; aiov.iov_base = (void *)(uintptr_t)buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_segflg = UIO_SYSSPACE; auio.uio_resid = count; auio.uio_rw = UIO_READ; auio.uio_td = td; auio.uio_offset = *offp; if ((fp->f_flag & FREAD) == 0) return (SET_ERROR(EBADF)); rc = fo_read(fp, &auio, td->td_ucred, FOF_OFFSET, td); if (rc) return (SET_ERROR(rc)); if (resid) *resid = auio.uio_resid; *offp += count - auio.uio_resid; return (SET_ERROR(0)); } int zfs_file_read(zfs_file_t *fp, void *buf, size_t count, ssize_t *resid) { loff_t off = fp->f_offset; ssize_t rc; rc = zfs_file_read_impl(fp, buf, count, &off, resid); if (rc == 0) fp->f_offset = off; return (rc); } int zfs_file_pread(zfs_file_t *fp, void *buf, size_t count, loff_t off, ssize_t *resid) { return (zfs_file_read_impl(fp, buf, count, &off, resid)); } int zfs_file_seek(zfs_file_t *fp, loff_t *offp, int whence) { int rc; struct thread *td; td = curthread; if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) return (SET_ERROR(ESPIPE)); rc = fo_seek(fp, *offp, whence, td); if (rc == 0) *offp = td->td_uretoff.tdu_off; return (SET_ERROR(rc)); } int zfs_file_getattr(zfs_file_t *fp, zfs_file_attr_t *zfattr) { struct thread *td; struct stat sb; int rc; td = curthread; #if __FreeBSD_version < 1400037 rc = fo_stat(fp, &sb, td->td_ucred, td); #else rc = fo_stat(fp, &sb, td->td_ucred); #endif if (rc) return (SET_ERROR(rc)); zfattr->zfa_size = sb.st_size; zfattr->zfa_mode = sb.st_mode; return (0); } static __inline int zfs_vop_fsync(vnode_t *vp) { struct mount *mp; int error; - if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) + if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0) goto drop; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = VOP_FSYNC(vp, MNT_WAIT, curthread); VOP_UNLOCK1(vp); vn_finished_write(mp); drop: return (SET_ERROR(error)); } int zfs_file_fsync(zfs_file_t *fp, int flags) { if (fp->f_type != DTYPE_VNODE) return (EINVAL); return (zfs_vop_fsync(fp->f_vnode)); } zfs_file_t * zfs_file_get(int fd) { struct file *fp; if (fget(curthread, fd, &cap_no_rights, &fp)) return (NULL); return (fp); } void zfs_file_put(zfs_file_t *fp) { fdrop(fp, curthread); } loff_t zfs_file_off(zfs_file_t *fp) { return (fp->f_offset); } void * zfs_file_private(zfs_file_t *fp) { file_t *tmpfp; void *data; int error; tmpfp = curthread->td_fpop; curthread->td_fpop = fp; error = devfs_get_cdevpriv(&data); curthread->td_fpop = tmpfp; if (error != 0) return (NULL); return (data); } int zfs_file_unlink(const char *fnamep) { zfs_uio_seg_t seg = UIO_SYSSPACE; int rc; #if __FreeBSD_version >= 1300018 rc = kern_funlinkat(curthread, AT_FDCWD, fnamep, FD_NONE, seg, 0, 0); #elif __FreeBSD_version >= 1202504 || defined(AT_BENEATH) rc = kern_unlinkat(curthread, AT_FDCWD, __DECONST(char *, fnamep), seg, 0, 0); #else rc = kern_unlinkat(curthread, AT_FDCWD, __DECONST(char *, fnamep), seg, 0); #endif return (SET_ERROR(rc)); } diff --git a/sys/fs/fdescfs/fdesc_vnops.c b/sys/fs/fdescfs/fdesc_vnops.c index ae6d60d0fc55..7046bb6bf244 100644 --- a/sys/fs/fdescfs/fdesc_vnops.c +++ b/sys/fs/fdescfs/fdesc_vnops.c @@ -1,664 +1,664 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software donated to Berkeley by * Jan-Simon Pendry. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)fdesc_vnops.c 8.9 (Berkeley) 1/21/94 * * $FreeBSD$ */ /* * /dev/fd Filesystem */ #include #include #include #include #include #include #include /* boottime */ #include #include #include #include /* Must come after sys/malloc.h */ #include #include #include #include #include #include #include #include #define NFDCACHE 4 #define FD_NHASH(ix) \ (&fdhashtbl[(ix) & fdhash]) static LIST_HEAD(fdhashhead, fdescnode) *fdhashtbl; static u_long fdhash; struct mtx fdesc_hashmtx; static vop_getattr_t fdesc_getattr; static vop_lookup_t fdesc_lookup; static vop_open_t fdesc_open; static vop_pathconf_t fdesc_pathconf; static vop_readdir_t fdesc_readdir; static vop_readlink_t fdesc_readlink; static vop_reclaim_t fdesc_reclaim; static vop_setattr_t fdesc_setattr; static struct vop_vector fdesc_vnodeops = { .vop_default = &default_vnodeops, .vop_access = VOP_NULL, .vop_getattr = fdesc_getattr, .vop_lookup = fdesc_lookup, .vop_open = fdesc_open, .vop_pathconf = fdesc_pathconf, .vop_readdir = fdesc_readdir, .vop_readlink = fdesc_readlink, .vop_reclaim = fdesc_reclaim, .vop_setattr = fdesc_setattr, }; VFS_VOP_VECTOR_REGISTER(fdesc_vnodeops); static void fdesc_remove_entry(struct fdescnode *); /* * Initialise cache headers */ int fdesc_init(struct vfsconf *vfsp) { mtx_init(&fdesc_hashmtx, "fdescfs_hash", NULL, MTX_DEF); fdhashtbl = hashinit(NFDCACHE, M_CACHE, &fdhash); return (0); } /* * Uninit ready for unload. */ int fdesc_uninit(struct vfsconf *vfsp) { hashdestroy(fdhashtbl, M_CACHE, fdhash); mtx_destroy(&fdesc_hashmtx); return (0); } /* * Remove an entry from the hash if it exists. */ static void fdesc_remove_entry(struct fdescnode *fd) { struct fdhashhead *fc; struct fdescnode *fd2; fc = FD_NHASH(fd->fd_ix); mtx_lock(&fdesc_hashmtx); LIST_FOREACH(fd2, fc, fd_hash) { if (fd == fd2) { LIST_REMOVE(fd, fd_hash); break; } } mtx_unlock(&fdesc_hashmtx); } int fdesc_allocvp(fdntype ftype, unsigned fd_fd, int ix, struct mount *mp, struct vnode **vpp) { struct fdescmount *fmp; struct fdhashhead *fc; struct fdescnode *fd, *fd2; struct vnode *vp, *vp2; int error; fc = FD_NHASH(ix); loop: mtx_lock(&fdesc_hashmtx); /* * If a forced unmount is progressing, we need to drop it. The flags are * protected by the hashmtx. */ fmp = mp->mnt_data; if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) { mtx_unlock(&fdesc_hashmtx); return (-1); } LIST_FOREACH(fd, fc, fd_hash) { if (fd->fd_ix == ix && fd->fd_vnode->v_mount == mp) { /* Get reference to vnode in case it's being free'd */ vp = fd->fd_vnode; VI_LOCK(vp); mtx_unlock(&fdesc_hashmtx); if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) goto loop; *vpp = vp; return (0); } } mtx_unlock(&fdesc_hashmtx); fd = malloc(sizeof(struct fdescnode), M_TEMP, M_WAITOK); error = getnewvnode("fdescfs", mp, &fdesc_vnodeops, &vp); if (error) { free(fd, M_TEMP); return (error); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_data = fd; fd->fd_vnode = vp; fd->fd_type = ftype; fd->fd_fd = fd_fd; fd->fd_ix = ix; if (ftype == Fdesc && fmp->flags & FMNT_LINRDLNKF) vp->v_vflag |= VV_READLINK; error = insmntque1(vp, mp); if (error != 0) { vgone(vp); vput(vp); *vpp = NULLVP; return (error); } /* Make sure that someone didn't beat us when inserting the vnode. */ mtx_lock(&fdesc_hashmtx); /* * If a forced unmount is progressing, we need to drop it. The flags are * protected by the hashmtx. */ fmp = mp->mnt_data; if (fmp == NULL || fmp->flags & FMNT_UNMOUNTF) { mtx_unlock(&fdesc_hashmtx); vgone(vp); vput(vp); *vpp = NULLVP; return (-1); } LIST_FOREACH(fd2, fc, fd_hash) { if (fd2->fd_ix == ix && fd2->fd_vnode->v_mount == mp) { /* Get reference to vnode in case it's being free'd */ vp2 = fd2->fd_vnode; VI_LOCK(vp2); mtx_unlock(&fdesc_hashmtx); error = vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK); /* Someone beat us, dec use count and wait for reclaim */ vgone(vp); vput(vp); /* If we didn't get it, return no vnode. */ if (error) vp2 = NULLVP; *vpp = vp2; return (error); } } /* If we came here, we can insert it safely. */ LIST_INSERT_HEAD(fc, fd, fd_hash); mtx_unlock(&fdesc_hashmtx); *vpp = vp; return (0); } struct fdesc_get_ino_args { fdntype ftype; unsigned fd_fd; int ix; struct file *fp; struct thread *td; }; static int fdesc_get_ino_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { struct fdesc_get_ino_args *a; struct fdescmount *fdm; struct vnode *vp; int error; a = arg; fdm = VFSTOFDESC(mp); if ((fdm->flags & FMNT_NODUP) != 0 && a->fp->f_type == DTYPE_VNODE) { vp = a->fp->f_vnode; vget(vp, lkflags | LK_RETRY); *rvp = vp; error = 0; } else { error = fdesc_allocvp(a->ftype, a->fd_fd, a->ix, mp, rvp); } fdrop(a->fp, a->td); return (error); } /* * vp is the current namei directory * ndp is the name to locate in that directory... */ static int fdesc_lookup(struct vop_lookup_args *ap) { struct vnode **vpp = ap->a_vpp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; char *pname = cnp->cn_nameptr; struct thread *td = curthread; struct file *fp; struct fdesc_get_ino_args arg; int nlen = cnp->cn_namelen; u_int fd, fd1; int error; struct vnode *fvp; if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = EROFS; goto bad; } if (cnp->cn_namelen == 1 && *pname == '.') { *vpp = dvp; VREF(dvp); return (0); } if (VTOFDESC(dvp)->fd_type != Froot) { error = ENOTDIR; goto bad; } fd = 0; /* the only time a leading 0 is acceptable is if it's "0" */ if (*pname == '0' && nlen != 1) { error = ENOENT; goto bad; } while (nlen--) { if (*pname < '0' || *pname > '9') { error = ENOENT; goto bad; } fd1 = 10 * fd + *pname++ - '0'; if (fd1 < fd) { error = ENOENT; goto bad; } fd = fd1; } /* * No rights to check since 'fp' isn't actually used. */ if ((error = fget(td, fd, &cap_no_rights, &fp)) != 0) goto bad; /* Check if we're looking up ourselves. */ if (VTOFDESC(dvp)->fd_ix == FD_DESC + fd) { /* * In case we're holding the last reference to the file, the dvp * will be re-acquired. */ vhold(dvp); VOP_UNLOCK(dvp); fdrop(fp, td); /* Re-aquire the lock afterwards. */ vn_lock(dvp, LK_RETRY | LK_EXCLUSIVE); vdrop(dvp); fvp = dvp; if (VN_IS_DOOMED(dvp)) error = ENOENT; } else { /* * Unlock our root node (dvp) when doing this, since we might * deadlock since the vnode might be locked by another thread * and the root vnode lock will be obtained afterwards (in case * we're looking up the fd of the root vnode), which will be the * opposite lock order. Vhold the root vnode first so we don't * lose it. */ arg.ftype = Fdesc; arg.fd_fd = fd; arg.ix = FD_DESC + fd; arg.fp = fp; arg.td = td; error = vn_vget_ino_gen(dvp, fdesc_get_ino_alloc, &arg, LK_EXCLUSIVE, &fvp); } if (error) goto bad; *vpp = fvp; return (0); bad: *vpp = NULL; return (error); } static int fdesc_open(struct vop_open_args *ap) { struct vnode *vp = ap->a_vp; if (VTOFDESC(vp)->fd_type == Froot) return (0); /* * XXX Kludge: set td->td_proc->p_dupfd to contain the value of the file * descriptor being sought for duplication. The error return ensures * that the vnode for this device will be released by vn_open. Open * will detect this special error and take the actions in dupfdopen. * Other callers of vn_open or VOP_OPEN will simply report the * error. */ ap->a_td->td_dupfd = VTOFDESC(vp)->fd_fd; /* XXX */ return (ENODEV); } static int fdesc_pathconf(struct vop_pathconf_args *ap) { struct vnode *vp = ap->a_vp; int error; switch (ap->a_name) { case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; return (0); case _PC_LINK_MAX: if (VTOFDESC(vp)->fd_type == Froot) *ap->a_retval = 2; else *ap->a_retval = 1; return (0); default: if (VTOFDESC(vp)->fd_type == Froot) return (vop_stdpathconf(ap)); vref(vp); VOP_UNLOCK(vp); error = kern_fpathconf(curthread, VTOFDESC(vp)->fd_fd, ap->a_name, ap->a_retval); vn_lock(vp, LK_SHARED | LK_RETRY); vunref(vp); return (error); } } static int fdesc_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct vattr *vap = ap->a_vap; struct timeval boottime; getboottime(&boottime); vap->va_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH; vap->va_fileid = VTOFDESC(vp)->fd_ix; vap->va_uid = 0; vap->va_gid = 0; vap->va_blocksize = DEV_BSIZE; vap->va_atime.tv_sec = boottime.tv_sec; vap->va_atime.tv_nsec = 0; vap->va_mtime = vap->va_atime; vap->va_ctime = vap->va_mtime; vap->va_gen = 0; vap->va_flags = 0; vap->va_bytes = 0; vap->va_filerev = 0; switch (VTOFDESC(vp)->fd_type) { case Froot: vap->va_type = VDIR; vap->va_nlink = 2; vap->va_size = DEV_BSIZE; vap->va_rdev = NODEV; break; case Fdesc: vap->va_type = (vp->v_vflag & VV_READLINK) == 0 ? VCHR : VLNK; vap->va_nlink = 1; vap->va_size = 0; vap->va_rdev = makedev(0, vap->va_fileid); break; default: panic("fdesc_getattr"); break; } vp->v_type = vap->va_type; return (0); } static int fdesc_setattr(struct vop_setattr_args *ap) { struct vattr *vap = ap->a_vap; struct vnode *vp; struct mount *mp; struct file *fp; struct thread *td = curthread; cap_rights_t rights; unsigned fd; int error; /* * Can't mess with the root vnode */ if (VTOFDESC(ap->a_vp)->fd_type == Froot) return (EACCES); fd = VTOFDESC(ap->a_vp)->fd_fd; /* * Allow setattr where there is an underlying vnode. * For O_PATH descriptors, disallow truncate. */ if (vap->va_size != VNOVAL) { error = getvnode(td, fd, cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); } else { error = getvnode_path(td, fd, cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); } if (error) { /* * getvnode() returns EINVAL if the file descriptor is not * backed by a vnode. Silently drop all changes except * chflags(2) in this case. */ if (error == EINVAL) { if (vap->va_flags != VNOVAL) error = EOPNOTSUPP; else error = 0; } return (error); } vp = fp->f_vnode; - if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) == 0) { + if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) == 0) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = VOP_SETATTR(vp, ap->a_vap, ap->a_cred); VOP_UNLOCK(vp); vn_finished_write(mp); } fdrop(fp, td); return (error); } #define UIO_MX _GENERIC_DIRLEN(10) /* number of symbols in INT_MAX printout */ static int fdesc_readdir(struct vop_readdir_args *ap) { struct fdescmount *fmp; struct uio *uio = ap->a_uio; struct filedesc *fdp; struct dirent d; struct dirent *dp = &d; int error, i, off, fcnt; if (VTOFDESC(ap->a_vp)->fd_type != Froot) panic("fdesc_readdir: not dir"); fmp = VFSTOFDESC(ap->a_vp->v_mount); if (ap->a_ncookies != NULL) *ap->a_ncookies = 0; off = (int)uio->uio_offset; if (off != uio->uio_offset || off < 0 || (u_int)off % UIO_MX != 0 || uio->uio_resid < UIO_MX) return (EINVAL); i = (u_int)off / UIO_MX; fdp = uio->uio_td->td_proc->p_fd; error = 0; fcnt = i - 2; /* The first two nodes are `.' and `..' */ FILEDESC_SLOCK(fdp); while (i < fdp->fd_nfiles + 2 && uio->uio_resid >= UIO_MX) { bzero((caddr_t)dp, UIO_MX); switch (i) { case 0: /* `.' */ case 1: /* `..' */ dp->d_fileno = i + FD_ROOT; dp->d_namlen = i + 1; dp->d_reclen = UIO_MX; bcopy("..", dp->d_name, dp->d_namlen); dp->d_type = DT_DIR; dirent_terminate(dp); break; default: if (fdp->fd_ofiles[fcnt].fde_file == NULL) break; dp->d_namlen = sprintf(dp->d_name, "%d", fcnt); dp->d_reclen = UIO_MX; dp->d_type = (fmp->flags & FMNT_LINRDLNKF) == 0 ? DT_CHR : DT_LNK; dp->d_fileno = i + FD_DESC; dirent_terminate(dp); break; } /* NOTE: d_off is the offset of the *next* entry. */ dp->d_off = UIO_MX * (i + 1); if (dp->d_namlen != 0) { /* * And ship to userland */ FILEDESC_SUNLOCK(fdp); error = uiomove(dp, UIO_MX, uio); if (error) goto done; FILEDESC_SLOCK(fdp); } i++; fcnt++; } FILEDESC_SUNLOCK(fdp); done: uio->uio_offset = i * UIO_MX; return (error); } static int fdesc_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp; struct fdescnode *fd; vp = ap->a_vp; fd = VTOFDESC(vp); fdesc_remove_entry(fd); free(vp->v_data, M_TEMP); vp->v_data = NULL; return (0); } static int fdesc_readlink(struct vop_readlink_args *va) { struct vnode *vp, *vn; struct thread *td; struct uio *uio; struct file *fp; char *freepath, *fullpath; size_t pathlen; int lockflags, fd_fd; int error; freepath = NULL; vn = va->a_vp; if (VTOFDESC(vn)->fd_type != Fdesc) panic("fdesc_readlink: not fdescfs link"); fd_fd = ((struct fdescnode *)vn->v_data)->fd_fd; lockflags = VOP_ISLOCKED(vn); VOP_UNLOCK(vn); td = curthread; error = fget_cap(td, fd_fd, &cap_no_rights, &fp, NULL); if (error != 0) goto out; switch (fp->f_type) { case DTYPE_VNODE: vp = fp->f_vnode; error = vn_fullpath(vp, &fullpath, &freepath); break; default: fullpath = "anon_inode:[unknown]"; break; } if (error == 0) { uio = va->a_uio; pathlen = strlen(fullpath); error = uiomove(fullpath, pathlen, uio); } if (freepath != NULL) free(freepath, M_TEMP); fdrop(fp, td); out: vn_lock(vn, lockflags | LK_RETRY); return (error); } diff --git a/sys/fs/nfsserver/nfs_nfsdstate.c b/sys/fs/nfsserver/nfs_nfsdstate.c index d5d7f9fbd363..84da1d9561cd 100644 --- a/sys/fs/nfsserver/nfs_nfsdstate.c +++ b/sys/fs/nfsserver/nfs_nfsdstate.c @@ -1,8864 +1,8864 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009 Rick Macklem, University of Guelph * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include struct nfsrv_stablefirst nfsrv_stablefirst; int nfsrv_issuedelegs = 0; int nfsrv_dolocallocks = 0; struct nfsv4lock nfsv4rootfs_lock; time_t nfsdev_time = 0; int nfsrv_layouthashsize; volatile int nfsrv_layoutcnt = 0; extern uint32_t nfs_srvmaxio; extern int newnfs_numnfsd; extern struct nfsstatsv1 nfsstatsv1; extern int nfsrv_lease; extern struct timeval nfsboottime; extern u_int32_t newnfs_true, newnfs_false; extern struct mtx nfsrv_dslock_mtx; extern struct mtx nfsrv_recalllock_mtx; extern struct mtx nfsrv_dontlistlock_mtx; extern int nfsd_debuglevel; extern u_int nfsrv_dsdirsize; extern struct nfsdevicehead nfsrv_devidhead; extern int nfsrv_doflexfile; extern int nfsrv_maxpnfsmirror; NFSV4ROOTLOCKMUTEX; NFSSTATESPINLOCK; extern struct nfsdontlisthead nfsrv_dontlisthead; extern volatile int nfsrv_devidcnt; extern struct nfslayouthead nfsrv_recalllisthead; extern char *nfsrv_zeropnfsdat; SYSCTL_DECL(_vfs_nfsd); int nfsrv_statehashsize = NFSSTATEHASHSIZE; SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN, &nfsrv_statehashsize, 0, "Size of state hash table set via loader.conf"); int nfsrv_clienthashsize = NFSCLIENTHASHSIZE; SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN, &nfsrv_clienthashsize, 0, "Size of client hash table set via loader.conf"); int nfsrv_lockhashsize = NFSLOCKHASHSIZE; SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN, &nfsrv_lockhashsize, 0, "Size of file handle hash table set via loader.conf"); int nfsrv_sessionhashsize = NFSSESSIONHASHSIZE; SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN, &nfsrv_sessionhashsize, 0, "Size of session hash table set via loader.conf"); int nfsrv_layouthighwater = NFSLAYOUTHIGHWATER; SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN, &nfsrv_layouthighwater, 0, "High water mark for number of layouts set via loader.conf"); static int nfsrv_v4statelimit = NFSRV_V4STATELIMIT; SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN, &nfsrv_v4statelimit, 0, "High water limit for NFSv4 opens+locks+delegations"); static int nfsrv_writedelegifpos = 0; SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW, &nfsrv_writedelegifpos, 0, "Issue a write delegation for read opens if possible"); static int nfsrv_allowreadforwriteopen = 1; SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW, &nfsrv_allowreadforwriteopen, 0, "Allow Reads to be done with Write Access StateIDs"); int nfsrv_pnfsatime = 0; SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW, &nfsrv_pnfsatime, 0, "For pNFS service, do Getattr ops to keep atime up-to-date"); int nfsrv_flexlinuxhack = 0; SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW, &nfsrv_flexlinuxhack, 0, "For Linux clients, hack around Flex File Layout bug"); /* * Hash lists for nfs V4. */ struct nfsclienthashhead *nfsclienthash; struct nfslockhashhead *nfslockhash; struct nfssessionhash *nfssessionhash; struct nfslayouthash *nfslayouthash; volatile int nfsrv_dontlistlen = 0; static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0; static time_t nfsrvboottime; static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0; static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER; static int nfsrv_nogsscallback = 0; static volatile int nfsrv_writedelegcnt = 0; static int nfsrv_faildscnt; /* local functions */ static void nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp); static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p); static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p); static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p); static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p); static void nfsrv_freenfslock(struct nfslock *lop); static void nfsrv_freenfslockfile(struct nfslockfile *lfp); static void nfsrv_freedeleg(struct nfsstate *); static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, u_int32_t flags, struct nfsstate **stpp); static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp, struct nfsstate **stpp); static int nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p); static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp, struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit); static void nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp); static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp, struct nfslock **other_lopp, struct nfslockfile *lfp); static int nfsrv_getipnumber(u_char *cp); static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags, nfsv4stateid_t *stateidp, int specialid); static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp, u_int32_t flags); static int nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p); static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp, uint32_t callback, int op, const char *optag, struct nfsdsession **sepp, int *slotposp); static u_int32_t nfsrv_nextclientindex(void); static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp); static void nfsrv_markstable(struct nfsclient *clp); static void nfsrv_markreclaim(struct nfsclient *clp); static int nfsrv_checkstable(struct nfsclient *clp); static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct vnode *vp, NFSPROC_T *p); static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p, vnode_t vp); static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp, struct nfsclient *clp, int *haslockp, NFSPROC_T *p); static int nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp); static time_t nfsrv_leaseexpiry(void); static void nfsrv_delaydelegtimeout(struct nfsstate *stp); static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid, struct nfsstate *stp, struct nfsrvcache *op); static int nfsrv_nootherstate(struct nfsstate *stp); static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags, uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p); static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first, uint64_t init_end, NFSPROC_T *p); static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p); static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p); static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first, uint64_t end); static void nfsrv_locklf(struct nfslockfile *lfp); static void nfsrv_unlocklf(struct nfslockfile *lfp); static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid); static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid); static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp, int dont_replycache, struct nfsdsession **sepp, int *slotposp); static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp); static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp, nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p); static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp); static void nfsrv_freelayoutlist(nfsquad_t clientid); static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode); static void nfsrv_freealllayouts(void); static void nfsrv_freedevid(struct nfsdevice *ds); static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p, struct nfsdevice **dsp); static void nfsrv_deleteds(struct nfsdevice *fndds); static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost); static void nfsrv_freealldevids(void); static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt, NFSPROC_T *p); static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p); static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype, NFSPROC_T *, struct nfslayout **lypp); static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt); static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs); static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs); static int nfsrv_dontlayout(fhandle_t *fhp); static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf, vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p, vnode_t *tvpp); static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp); /* * Scan the client list for a match and either return the current one, * create a new entry or return an error. * If returning a non-error, the clp structure must either be linked into * the client list or free'd. */ int nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp, nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p) { struct nfsclient *clp = NULL, *new_clp = *new_clpp; int i, error = 0, ret; struct nfsstate *stp, *tstp; #ifdef INET struct sockaddr_in *sin, *rin; #endif #ifdef INET6 struct sockaddr_in6 *sin6, *rin6; #endif struct nfsdsession *sep, *nsep; int zapit = 0, gotit, hasstate = 0, igotlock; static u_int64_t confirm_index = 0; /* * Check for state resource limit exceeded. */ if (nfsrv_openpluslock > nfsrv_v4statelimit) { error = NFSERR_RESOURCE; goto out; } if (nfsrv_issuedelegs == 0 || ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0)) /* * Don't do callbacks when delegations are disabled or * for AUTH_GSS unless enabled via nfsrv_nogsscallback. * If establishing a callback connection is attempted * when a firewall is blocking the callback path, the * server may wait too long for the connect attempt to * succeed during the Open. Some clients, such as Linux, * may timeout and give up on the Open before the server * replies. Also, since AUTH_GSS callbacks are not * yet interoperability tested, they might cause the * server to crap out, if they get past the Init call to * the client. */ new_clp->lc_program = 0; /* Lock out other nfsd threads */ NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!igotlock); NFSUNLOCKV4ROOTMUTEX(); /* * Search for a match in the client list. */ gotit = i = 0; while (i < nfsrv_clienthashsize && !gotit) { LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { if (new_clp->lc_idlen == clp->lc_idlen && !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) { gotit = 1; break; } } if (gotit == 0) i++; } if (!gotit || (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) { if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) { /* * For NFSv4.1, if confirmp->lval[1] is non-zero, the * client is trying to update a confirmed clientid. */ NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); confirmp->lval[1] = 0; error = NFSERR_NOENT; goto out; } /* * Get rid of the old one. */ if (i != nfsrv_clienthashsize) { LIST_REMOVE(clp, lc_hash); nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); zapit = 1; } /* * Add it after assigning a client id to it. */ new_clp->lc_flags |= LCL_NEEDSCONFIRM; if ((nd->nd_flag & ND_NFSV41) != 0) { confirmp->lval[0] = ++confirm_index; new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1; } else confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index; clientidp->lval[0] = new_clp->lc_clientid.lval[0] = (u_int32_t)nfsrvboottime; clientidp->lval[1] = new_clp->lc_clientid.lval[1] = nfsrv_nextclientindex(); new_clp->lc_stateindex = 0; new_clp->lc_statemaxindex = 0; new_clp->lc_prevsess = 0; new_clp->lc_cbref = 0; new_clp->lc_expiry = nfsrv_leaseexpiry(); LIST_INIT(&new_clp->lc_open); LIST_INIT(&new_clp->lc_deleg); LIST_INIT(&new_clp->lc_olddeleg); LIST_INIT(&new_clp->lc_session); for (i = 0; i < nfsrv_statehashsize; i++) LIST_INIT(&new_clp->lc_stateid[i]); LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, lc_hash); nfsstatsv1.srvclients++; nfsrv_openpluslock++; nfsrv_clients++; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); if (zapit) nfsrv_zapclient(clp, p); *new_clpp = NULL; goto out; } /* * Now, handle the cases where the id is already issued. */ if (nfsrv_notsamecredname(nd, clp)) { /* * Check to see if there is expired state that should go away. */ if (clp->lc_expiry < NFSD_MONOSEC && (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) { nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); } /* * If there is outstanding state, then reply NFSERR_CLIDINUSE per * RFC3530 Sec. 8.1.2 last para. */ if (!LIST_EMPTY(&clp->lc_deleg)) { hasstate = 1; } else if (LIST_EMPTY(&clp->lc_open)) { hasstate = 0; } else { hasstate = 0; /* Look for an Open on the OpenOwner */ LIST_FOREACH(stp, &clp->lc_open, ls_list) { if (!LIST_EMPTY(&stp->ls_open)) { hasstate = 1; break; } } } if (hasstate) { /* * If the uid doesn't match, return NFSERR_CLIDINUSE after * filling out the correct ipaddr and portnum. */ switch (clp->lc_req.nr_nam->sa_family) { #ifdef INET case AF_INET: sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam; rin = (struct sockaddr_in *)clp->lc_req.nr_nam; sin->sin_addr.s_addr = rin->sin_addr.s_addr; sin->sin_port = rin->sin_port; break; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam; rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; sin6->sin6_addr = rin6->sin6_addr; sin6->sin6_port = rin6->sin6_port; break; #endif } NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); error = NFSERR_CLIDINUSE; goto out; } } if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) { /* * If the verifier has changed, the client has rebooted * and a new client id is issued. The old state info * can be thrown away once the SETCLIENTID_CONFIRM occurs. */ LIST_REMOVE(clp, lc_hash); /* Get rid of all sessions on this clientid. */ LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) { ret = nfsrv_freesession(sep, NULL); if (ret != 0) printf("nfsrv_setclient: verifier changed free" " session failed=%d\n", ret); } new_clp->lc_flags |= LCL_NEEDSCONFIRM; if ((nd->nd_flag & ND_NFSV41) != 0) { confirmp->lval[0] = ++confirm_index; new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1; } else confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index; clientidp->lval[0] = new_clp->lc_clientid.lval[0] = nfsrvboottime; clientidp->lval[1] = new_clp->lc_clientid.lval[1] = nfsrv_nextclientindex(); new_clp->lc_stateindex = 0; new_clp->lc_statemaxindex = 0; new_clp->lc_prevsess = 0; new_clp->lc_cbref = 0; new_clp->lc_expiry = nfsrv_leaseexpiry(); /* * Save the state until confirmed. */ LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list); LIST_FOREACH(tstp, &new_clp->lc_open, ls_list) tstp->ls_clp = new_clp; LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list); LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list) tstp->ls_clp = new_clp; LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list); LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list) tstp->ls_clp = new_clp; for (i = 0; i < nfsrv_statehashsize; i++) { LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i], ls_hash); LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash) tstp->ls_clp = new_clp; } LIST_INIT(&new_clp->lc_session); LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, lc_hash); nfsstatsv1.srvclients++; nfsrv_openpluslock++; nfsrv_clients++; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); /* * Must wait until any outstanding callback on the old clp * completes. */ NFSLOCKSTATE(); while (clp->lc_cbref) { clp->lc_flags |= LCL_WAKEUPWANTED; (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, "nfsd clp", 10 * hz); } NFSUNLOCKSTATE(); nfsrv_zapclient(clp, p); *new_clpp = NULL; goto out; } /* For NFSv4.1, mark that we found a confirmed clientid. */ if ((nd->nd_flag & ND_NFSV41) != 0) { clientidp->lval[0] = clp->lc_clientid.lval[0]; clientidp->lval[1] = clp->lc_clientid.lval[1]; confirmp->lval[0] = 0; /* Ignored by client */ confirmp->lval[1] = 1; } else { /* * id and verifier match, so update the net address info * and get rid of any existing callback authentication * handle, so a new one will be acquired. */ LIST_REMOVE(clp, lc_hash); new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN); new_clp->lc_expiry = nfsrv_leaseexpiry(); confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index; clientidp->lval[0] = new_clp->lc_clientid.lval[0] = clp->lc_clientid.lval[0]; clientidp->lval[1] = new_clp->lc_clientid.lval[1] = clp->lc_clientid.lval[1]; new_clp->lc_delegtime = clp->lc_delegtime; new_clp->lc_stateindex = clp->lc_stateindex; new_clp->lc_statemaxindex = clp->lc_statemaxindex; new_clp->lc_cbref = 0; LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list); LIST_FOREACH(tstp, &new_clp->lc_open, ls_list) tstp->ls_clp = new_clp; LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list); LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list) tstp->ls_clp = new_clp; LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list); LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list) tstp->ls_clp = new_clp; for (i = 0; i < nfsrv_statehashsize; i++) { LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i], ls_hash); LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash) tstp->ls_clp = new_clp; } LIST_INIT(&new_clp->lc_session); LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, lc_hash); nfsstatsv1.srvclients++; nfsrv_openpluslock++; nfsrv_clients++; } NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); if ((nd->nd_flag & ND_NFSV41) == 0) { /* * Must wait until any outstanding callback on the old clp * completes. */ NFSLOCKSTATE(); while (clp->lc_cbref) { clp->lc_flags |= LCL_WAKEUPWANTED; (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, "nfsdclp", 10 * hz); } NFSUNLOCKSTATE(); nfsrv_zapclient(clp, p); *new_clpp = NULL; } out: NFSEXITCODE2(error, nd); return (error); } /* * Check to see if the client id exists and optionally confirm it. */ int nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp, struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram, struct nfsrv_descript *nd, NFSPROC_T *p) { struct nfsclient *clp; struct nfsstate *stp; int i; struct nfsclienthashhead *hp; int error = 0, igotlock, doneok; struct nfssessionhash *shp; struct nfsdsession *sep; uint64_t sessid[2]; bool sess_replay; static uint64_t next_sess = 0; if (clpp) *clpp = NULL; if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 || opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) { error = NFSERR_STALECLIENTID; goto out; } /* * If called with opflags == CLOPS_RENEW, the State Lock is * already held. Otherwise, we need to get either that or, * for the case of Confirm, lock out the nfsd threads. */ if (opflags & CLOPS_CONFIRM) { NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!igotlock); /* * Create a new sessionid here, since we need to do it where * there is a mutex held to serialize update of next_sess. */ if ((nd->nd_flag & ND_NFSV41) != 0) { sessid[0] = ++next_sess; sessid[1] = clientid.qval; } NFSUNLOCKV4ROOTMUTEX(); } else if (opflags != CLOPS_RENEW) { NFSLOCKSTATE(); } /* For NFSv4.1, the clp is acquired from the associated session. */ if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 && opflags == CLOPS_RENEW) { clp = NULL; if ((nd->nd_flag & ND_HASSEQUENCE) != 0) { shp = NFSSESSIONHASH(nd->nd_sessionid); NFSLOCKSESSION(shp); sep = nfsrv_findsession(nd->nd_sessionid); if (sep != NULL) clp = sep->sess_clp; NFSUNLOCKSESSION(shp); } } else { hp = NFSCLIENTHASH(clientid); LIST_FOREACH(clp, hp, lc_hash) { if (clp->lc_clientid.lval[1] == clientid.lval[1]) break; } } if (clp == NULL) { if (opflags & CLOPS_CONFIRM) error = NFSERR_STALECLIENTID; else error = NFSERR_EXPIRED; } else if (clp->lc_flags & LCL_ADMINREVOKED) { /* * If marked admin revoked, just return the error. */ error = NFSERR_ADMINREVOKED; } if (error) { if (opflags & CLOPS_CONFIRM) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } else if (opflags != CLOPS_RENEW) { NFSUNLOCKSTATE(); } goto out; } /* * Perform any operations specified by the opflags. */ if (opflags & CLOPS_CONFIRM) { sess_replay = false; if ((nd->nd_flag & ND_NFSV41) != 0) { /* * For the case where lc_confirm.lval[0] == confirm.lval[0], * use the new session, but with the previous sessionid. * This is not exactly what the RFC describes, but should * result in the same reply as the previous CreateSession. */ if (clp->lc_confirm.lval[0] + 1 == confirm.lval[0]) { clp->lc_confirm.lval[0] = confirm.lval[0]; clp->lc_prevsess = sessid[0]; } else if (clp->lc_confirm.lval[0] == confirm.lval[0]) { if (clp->lc_prevsess == 0) error = NFSERR_SEQMISORDERED; else sessid[0] = clp->lc_prevsess; sess_replay = true; } else error = NFSERR_SEQMISORDERED; } else if ((nd->nd_flag & ND_NFSV41) == 0 && clp->lc_confirm.qval != confirm.qval) error = NFSERR_STALECLIENTID; if (error == 0 && nfsrv_notsamecredname(nd, clp)) error = NFSERR_CLIDINUSE; if (!error) { if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) == LCL_NEEDSCONFIRM) { /* * Hang onto the delegations (as old delegations) * for an Open with CLAIM_DELEGATE_PREV unless in * grace, but get rid of the rest of the state. */ nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_olddeleg); if (nfsrv_checkgrace(nd, clp, 0)) { /* In grace, so just delete delegations */ nfsrv_freedeleglist(&clp->lc_deleg); } else { LIST_FOREACH(stp, &clp->lc_deleg, ls_list) stp->ls_flags |= NFSLCK_OLDDELEG; clp->lc_delegtime = NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA; LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg, ls_list); } if ((nd->nd_flag & ND_NFSV41) != 0) clp->lc_program = cbprogram; } clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN); if (clp->lc_program) clp->lc_flags |= LCL_NEEDSCBNULL; /* For NFSv4.1, link the session onto the client. */ if (nsep != NULL) { /* Hold a reference on the xprt for a backchannel. */ if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 && !sess_replay) { if (clp->lc_req.nr_client == NULL) clp->lc_req.nr_client = (struct __rpc_client *) clnt_bck_create(nd->nd_xprt->xp_socket, cbprogram, NFSV4_CBVERS); if (clp->lc_req.nr_client != NULL) { SVC_ACQUIRE(nd->nd_xprt); CLNT_ACQUIRE(clp->lc_req.nr_client); nd->nd_xprt->xp_p2 = clp->lc_req.nr_client; /* Disable idle timeout. */ nd->nd_xprt->xp_idletimeout = 0; nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt; } else nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN; } NFSBCOPY(sessid, nsep->sess_sessionid, NFSX_V4SESSIONID); NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid, NFSX_V4SESSIONID); if (!sess_replay) { shp = NFSSESSIONHASH(nsep->sess_sessionid); NFSLOCKSTATE(); NFSLOCKSESSION(shp); LIST_INSERT_HEAD(&shp->list, nsep, sess_hash); LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list); nsep->sess_clp = clp; NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); } } } } else if (clp->lc_flags & LCL_NEEDSCONFIRM) { error = NFSERR_EXPIRED; } /* * If called by the Renew Op, we must check the principal. */ if (!error && (opflags & CLOPS_RENEWOP)) { if (nfsrv_notsamecredname(nd, clp)) { doneok = 0; for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) { LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) { if ((stp->ls_flags & NFSLCK_OPEN) && stp->ls_uid == nd->nd_cred->cr_uid) { doneok = 1; break; } } } if (!doneok) error = NFSERR_ACCES; } if (!error && (clp->lc_flags & LCL_CBDOWN)) error = NFSERR_CBPATHDOWN; } if ((!error || error == NFSERR_CBPATHDOWN) && (opflags & CLOPS_RENEW)) { clp->lc_expiry = nfsrv_leaseexpiry(); } if (opflags & CLOPS_CONFIRM) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } else if (opflags != CLOPS_RENEW) { NFSUNLOCKSTATE(); } if (clpp) *clpp = clp; out: NFSEXITCODE2(error, nd); return (error); } /* * Perform the NFSv4.1 destroy clientid. */ int nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p) { struct nfsclient *clp; struct nfsclienthashhead *hp; int error = 0, i, igotlock; if (nfsrvboottime != clientid.lval[0]) { error = NFSERR_STALECLIENTID; goto out; } /* Lock out other nfsd threads */ NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (igotlock == 0); NFSUNLOCKV4ROOTMUTEX(); hp = NFSCLIENTHASH(clientid); LIST_FOREACH(clp, hp, lc_hash) { if (clp->lc_clientid.lval[1] == clientid.lval[1]) break; } if (clp == NULL) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); /* Just return ok, since it is gone. */ goto out; } /* * Free up all layouts on the clientid. Should the client return the * layouts? */ nfsrv_freelayoutlist(clientid); /* Scan for state on the clientid. */ for (i = 0; i < nfsrv_statehashsize; i++) if (!LIST_EMPTY(&clp->lc_stateid[i])) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); error = NFSERR_CLIENTIDBUSY; goto out; } if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); error = NFSERR_CLIENTIDBUSY; goto out; } /* Destroy the clientid and return ok. */ nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); LIST_REMOVE(clp, lc_hash); NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); nfsrv_zapclient(clp, p); out: NFSEXITCODE2(error, nd); return (error); } /* * Called from the new nfssvc syscall to admin revoke a clientid. * Returns 0 for success, error otherwise. */ int nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p) { struct nfsclient *clp = NULL; int i, error = 0; int gotit, igotlock; /* * First, lock out the nfsd so that state won't change while the * revocation record is being written to the stable storage restart * file. */ NFSLOCKV4ROOTMUTEX(); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!igotlock); NFSUNLOCKV4ROOTMUTEX(); /* * Search for a match in the client list. */ gotit = i = 0; while (i < nfsrv_clienthashsize && !gotit) { LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { if (revokep->nclid_idlen == clp->lc_idlen && !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) { gotit = 1; break; } } i++; } if (!gotit) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 0); NFSUNLOCKV4ROOTMUTEX(); error = EPERM; goto out; } /* * Now, write out the revocation record */ nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); nfsrv_backupstable(); /* * and clear out the state, marking the clientid revoked. */ clp->lc_flags &= ~LCL_CALLBACKSON; clp->lc_flags |= LCL_ADMINREVOKED; nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 0); NFSUNLOCKV4ROOTMUTEX(); out: NFSEXITCODE(error); return (error); } /* * Dump out stats for all clients. Called from nfssvc(2), that is used * nfsstatsv1. */ void nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt) { struct nfsclient *clp; int i = 0, cnt = 0; /* * First, get a reference on the nfsv4rootfs_lock so that an * exclusive lock cannot be acquired while dumping the clients. */ NFSLOCKV4ROOTMUTEX(); nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); NFSUNLOCKV4ROOTMUTEX(); NFSLOCKSTATE(); /* * Rattle through the client lists until done. */ while (i < nfsrv_clienthashsize && cnt < maxcnt) { clp = LIST_FIRST(&nfsclienthash[i]); while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) { nfsrv_dumpaclient(clp, &dumpp[cnt]); cnt++; clp = LIST_NEXT(clp, lc_hash); } i++; } if (cnt < maxcnt) dumpp[cnt].ndcl_clid.nclid_idlen = 0; NFSUNLOCKSTATE(); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); } /* * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd. */ static void nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp) { struct nfsstate *stp, *openstp, *lckownstp; struct nfslock *lop; sa_family_t af; #ifdef INET struct sockaddr_in *rin; #endif #ifdef INET6 struct sockaddr_in6 *rin6; #endif dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0; dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0; dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0; dumpp->ndcl_flags = clp->lc_flags; dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen; NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen); af = clp->lc_req.nr_nam->sa_family; dumpp->ndcl_addrfam = af; switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)clp->lc_req.nr_nam; dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr; break; #endif } /* * Now, scan the state lists and total up the opens and locks. */ LIST_FOREACH(stp, &clp->lc_open, ls_list) { dumpp->ndcl_nopenowners++; LIST_FOREACH(openstp, &stp->ls_open, ls_list) { dumpp->ndcl_nopens++; LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) { dumpp->ndcl_nlockowners++; LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) { dumpp->ndcl_nlocks++; } } } } /* * and the delegation lists. */ LIST_FOREACH(stp, &clp->lc_deleg, ls_list) { dumpp->ndcl_ndelegs++; } LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) { dumpp->ndcl_nolddelegs++; } } /* * Dump out lock stats for a file. */ void nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt, NFSPROC_T *p) { struct nfsstate *stp; struct nfslock *lop; int cnt = 0; struct nfslockfile *lfp; sa_family_t af; #ifdef INET struct sockaddr_in *rin; #endif #ifdef INET6 struct sockaddr_in6 *rin6; #endif int ret; fhandle_t nfh; ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p); /* * First, get a reference on the nfsv4rootfs_lock so that an * exclusive lock on it cannot be acquired while dumping the locks. */ NFSLOCKV4ROOTMUTEX(); nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); NFSUNLOCKV4ROOTMUTEX(); NFSLOCKSTATE(); if (!ret) ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0); if (ret) { ldumpp[0].ndlck_clid.nclid_idlen = 0; NFSUNLOCKSTATE(); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); return; } /* * For each open share on file, dump it out. */ stp = LIST_FIRST(&lfp->lf_open); while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) { ldumpp[cnt].ndlck_flags = stp->ls_flags; ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid; ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0]; ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1]; ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2]; ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_openowner->ls_ownerlen; NFSBCOPY(stp->ls_openowner->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id, stp->ls_openowner->ls_ownerlen); ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen; NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id, stp->ls_clp->lc_idlen); af = stp->ls_clp->lc_req.nr_nam->sa_family; ldumpp[cnt].ndlck_addrfam = af; switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *) stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; break; #endif } stp = LIST_NEXT(stp, ls_file); cnt++; } /* * and all locks. */ lop = LIST_FIRST(&lfp->lf_lock); while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) { stp = lop->lo_stp; ldumpp[cnt].ndlck_flags = lop->lo_flags; ldumpp[cnt].ndlck_first = lop->lo_first; ldumpp[cnt].ndlck_end = lop->lo_end; ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid; ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0]; ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1]; ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2]; ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen; NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id, stp->ls_ownerlen); ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen; NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id, stp->ls_clp->lc_idlen); af = stp->ls_clp->lc_req.nr_nam->sa_family; ldumpp[cnt].ndlck_addrfam = af; switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *) stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; break; #endif } lop = LIST_NEXT(lop, lo_lckfile); cnt++; } /* * and the delegations. */ stp = LIST_FIRST(&lfp->lf_deleg); while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) { ldumpp[cnt].ndlck_flags = stp->ls_flags; ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid; ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0]; ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1]; ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2]; ldumpp[cnt].ndlck_owner.nclid_idlen = 0; ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen; NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id, stp->ls_clp->lc_idlen); af = stp->ls_clp->lc_req.nr_nam->sa_family; ldumpp[cnt].ndlck_addrfam = af; switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *) stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; break; #endif } stp = LIST_NEXT(stp, ls_file); cnt++; } /* * If list isn't full, mark end of list by setting the client name * to zero length. */ if (cnt < maxcnt) ldumpp[cnt].ndlck_clid.nclid_idlen = 0; NFSUNLOCKSTATE(); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); } /* * Server timer routine. It can scan any linked list, so long * as it holds the spin/mutex lock and there is no exclusive lock on * nfsv4rootfs_lock. * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok * to do this from a callout, since the spin locks work. For * Darwin, I'm not sure what will work correctly yet.) * Should be called once per second. */ void nfsrv_servertimer(void) { struct nfsclient *clp, *nclp; struct nfsstate *stp, *nstp; int got_ref, i; /* * Make sure nfsboottime is set. This is used by V3 as well * as V4. Note that nfsboottime is not nfsrvboottime, which is * only used by the V4 server for leases. */ if (nfsboottime.tv_sec == 0) NFSSETBOOTTIME(nfsboottime); /* * If server hasn't started yet, just return. */ NFSLOCKSTATE(); if (nfsrv_stablefirst.nsf_eograce == 0) { NFSUNLOCKSTATE(); return; } if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) { if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) && NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce) nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK); NFSUNLOCKSTATE(); return; } /* * Try and get a reference count on the nfsv4rootfs_lock so that * no nfsd thread can acquire an exclusive lock on it before this * call is done. If it is already exclusively locked, just return. */ NFSLOCKV4ROOTMUTEX(); got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); if (got_ref == 0) { NFSUNLOCKSTATE(); return; } /* * For each client... */ for (i = 0; i < nfsrv_clienthashsize; i++) { clp = LIST_FIRST(&nfsclienthash[i]); while (clp != LIST_END(&nfsclienthash[i])) { nclp = LIST_NEXT(clp, lc_hash); if (!(clp->lc_flags & LCL_EXPIREIT)) { if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC && ((LIST_EMPTY(&clp->lc_deleg) && LIST_EMPTY(&clp->lc_open)) || nfsrv_clients > nfsrv_clienthighwater)) || (clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC || (clp->lc_expiry < NFSD_MONOSEC && (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) { /* * Lease has expired several nfsrv_lease times ago: * PLUS * - no state is associated with it * OR * - above high water mark for number of clients * (nfsrv_clienthighwater should be large enough * that this only occurs when clients fail to * use the same nfs_client_id4.id. Maybe somewhat * higher that the maximum number of clients that * will mount this server?) * OR * Lease has expired a very long time ago * OR * Lease has expired PLUS the number of opens + locks * has exceeded 90% of capacity * * --> Mark for expiry. The actual expiry will be done * by an nfsd sometime soon. */ clp->lc_flags |= LCL_EXPIREIT; nfsrv_stablefirst.nsf_flags |= (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT); } else { /* * If there are no opens, increment no open tick cnt * If time exceeds NFSNOOPEN, mark it to be thrown away * otherwise, if there is an open, reset no open time * Hopefully, this will avoid excessive re-creation * of open owners and subsequent open confirms. */ stp = LIST_FIRST(&clp->lc_open); while (stp != LIST_END(&clp->lc_open)) { nstp = LIST_NEXT(stp, ls_list); if (LIST_EMPTY(&stp->ls_open)) { stp->ls_noopens++; if (stp->ls_noopens > NFSNOOPEN || (nfsrv_openpluslock * 2) > nfsrv_v4statelimit) nfsrv_stablefirst.nsf_flags |= NFSNSF_NOOPENS; } else { stp->ls_noopens = 0; } stp = nstp; } } } clp = nclp; } } NFSUNLOCKSTATE(); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); } /* * The following set of functions free up the various data structures. */ /* * Clear out all open/lock state related to this nfsclient. * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that * there are no other active nfsd threads. */ void nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p) { struct nfsstate *stp, *nstp; struct nfsdsession *sep, *nsep; LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) nfsrv_freeopenowner(stp, 1, p); if ((clp->lc_flags & LCL_ADMINREVOKED) == 0) LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) (void)nfsrv_freesession(sep, NULL); } /* * Free a client that has been cleaned. It should also already have been * removed from the lists. * (Just to be safe w.r.t. newnfs_disconnect(), call this function when * softclock interrupts are enabled.) */ void nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p) { #ifdef notyet if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) == (LCL_GSS | LCL_CALLBACKSON) && (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) && clp->lc_handlelen > 0) { clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE; clp->lc_hand.nfsh_flag |= NFSG_DESTROYED; (void) nfsrv_docallback(clp, NFSV4PROC_CBNULL, NULL, 0, NULL, NULL, NULL, 0, p); } #endif newnfs_disconnect(NULL, &clp->lc_req); free(clp->lc_req.nr_nam, M_SONAME); NFSFREEMUTEX(&clp->lc_req.nr_mtx); free(clp->lc_stateid, M_NFSDCLIENT); free(clp, M_NFSDCLIENT); NFSLOCKSTATE(); nfsstatsv1.srvclients--; nfsrv_openpluslock--; nfsrv_clients--; NFSUNLOCKSTATE(); } /* * Free a list of delegation state structures. * (This function will also free all nfslockfile structures that no * longer have associated state.) */ void nfsrv_freedeleglist(struct nfsstatehead *sthp) { struct nfsstate *stp, *nstp; LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) { nfsrv_freedeleg(stp); } LIST_INIT(sthp); } /* * Free up a delegation. */ static void nfsrv_freedeleg(struct nfsstate *stp) { struct nfslockfile *lfp; LIST_REMOVE(stp, ls_hash); LIST_REMOVE(stp, ls_list); LIST_REMOVE(stp, ls_file); if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0) nfsrv_writedelegcnt--; lfp = stp->ls_lfp; if (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) && lfp->lf_usecount == 0 && nfsv4_testlock(&lfp->lf_locallock_lck) == 0) nfsrv_freenfslockfile(lfp); free(stp, M_NFSDSTATE); nfsstatsv1.srvdelegates--; nfsrv_openpluslock--; nfsrv_delegatecnt--; } /* * This function frees an open owner and all associated opens. */ static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p) { struct nfsstate *nstp, *tstp; LIST_REMOVE(stp, ls_list); /* * Now, free all associated opens. */ nstp = LIST_FIRST(&stp->ls_open); while (nstp != LIST_END(&stp->ls_open)) { tstp = nstp; nstp = LIST_NEXT(nstp, ls_list); (void) nfsrv_freeopen(tstp, NULL, cansleep, p); } if (stp->ls_op) nfsrvd_derefcache(stp->ls_op); free(stp, M_NFSDSTATE); nfsstatsv1.srvopenowners--; nfsrv_openpluslock--; } /* * This function frees an open (nfsstate open structure) with all associated * lock_owners and locks. It also frees the nfslockfile structure iff there * are no other opens on the file. * Returns 1 if it free'd the nfslockfile, 0 otherwise. */ static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p) { struct nfsstate *nstp, *tstp; struct nfslockfile *lfp; int ret; LIST_REMOVE(stp, ls_hash); LIST_REMOVE(stp, ls_list); LIST_REMOVE(stp, ls_file); lfp = stp->ls_lfp; /* * Now, free all lockowners associated with this open. */ LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp) nfsrv_freelockowner(tstp, vp, cansleep, p); /* * The nfslockfile is freed here if there are no locks * associated with the open. * If there are locks associated with the open, the * nfslockfile structure can be freed via nfsrv_freelockowner(). * Acquire the state mutex to avoid races with calls to * nfsrv_getlockfile(). */ if (cansleep != 0) NFSLOCKSTATE(); if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) && lfp->lf_usecount == 0 && (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) { nfsrv_freenfslockfile(lfp); ret = 1; } else ret = 0; if (cansleep != 0) NFSUNLOCKSTATE(); free(stp, M_NFSDSTATE); nfsstatsv1.srvopens--; nfsrv_openpluslock--; return (ret); } /* * Frees a lockowner and all associated locks. */ static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p) { LIST_REMOVE(stp, ls_hash); LIST_REMOVE(stp, ls_list); nfsrv_freeallnfslocks(stp, vp, cansleep, p); if (stp->ls_op) nfsrvd_derefcache(stp->ls_op); free(stp, M_NFSDSTATE); nfsstatsv1.srvlockowners--; nfsrv_openpluslock--; } /* * Free all the nfs locks on a lockowner. */ static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p) { struct nfslock *lop, *nlop; struct nfsrollback *rlp, *nrlp; struct nfslockfile *lfp = NULL; int gottvp = 0; vnode_t tvp = NULL; uint64_t first, end; if (vp != NULL) ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked"); lop = LIST_FIRST(&stp->ls_lock); while (lop != LIST_END(&stp->ls_lock)) { nlop = LIST_NEXT(lop, lo_lckowner); /* * Since all locks should be for the same file, lfp should * not change. */ if (lfp == NULL) lfp = lop->lo_lfp; else if (lfp != lop->lo_lfp) panic("allnfslocks"); /* * If vp is NULL and cansleep != 0, a vnode must be acquired * from the file handle. This only occurs when called from * nfsrv_cleanclient(). */ if (gottvp == 0) { if (nfsrv_dolocallocks == 0) tvp = NULL; else if (vp == NULL && cansleep != 0) { tvp = nfsvno_getvp(&lfp->lf_fh); if (tvp != NULL) NFSVOPUNLOCK(tvp); } else tvp = vp; gottvp = 1; } if (tvp != NULL) { if (cansleep == 0) panic("allnfs2"); first = lop->lo_first; end = lop->lo_end; nfsrv_freenfslock(lop); nfsrv_localunlock(tvp, lfp, first, end, p); LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) free(rlp, M_NFSDROLLBACK); LIST_INIT(&lfp->lf_rollback); } else nfsrv_freenfslock(lop); lop = nlop; } if (vp == NULL && tvp != NULL) vrele(tvp); } /* * Free an nfslock structure. */ static void nfsrv_freenfslock(struct nfslock *lop) { if (lop->lo_lckfile.le_prev != NULL) { LIST_REMOVE(lop, lo_lckfile); nfsstatsv1.srvlocks--; nfsrv_openpluslock--; } LIST_REMOVE(lop, lo_lckowner); free(lop, M_NFSDLOCK); } /* * This function frees an nfslockfile structure. */ static void nfsrv_freenfslockfile(struct nfslockfile *lfp) { LIST_REMOVE(lfp, lf_hash); free(lfp, M_NFSDLOCKFILE); } /* * This function looks up an nfsstate structure via stateid. */ static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags, struct nfsstate **stpp) { struct nfsstate *stp; struct nfsstatehead *hp; int error = 0; *stpp = NULL; hp = NFSSTATEHASH(clp, *stateidp); LIST_FOREACH(stp, hp, ls_hash) { if (!NFSBCMP(stp->ls_stateid.other, stateidp->other, NFSX_STATEIDOTHER)) break; } /* * If no state id in list, return NFSERR_BADSTATEID. */ if (stp == LIST_END(hp)) { error = NFSERR_BADSTATEID; goto out; } *stpp = stp; out: NFSEXITCODE(error); return (error); } /* * This function gets an nfsstate structure via owner string. */ static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp, struct nfsstate **stpp) { struct nfsstate *stp; *stpp = NULL; LIST_FOREACH(stp, hp, ls_list) { if (new_stp->ls_ownerlen == stp->ls_ownerlen && !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) { *stpp = stp; return; } } } /* * Lock control function called to update lock status. * Returns 0 upon success, -1 if there is no lock and the flags indicate * that one isn't to be created and an NFSERR_xxx for other errors. * The structures new_stp and new_lop are passed in as pointers that should * be set to NULL if the structure is used and shouldn't be free'd. * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are * never used and can safely be allocated on the stack. For all other * cases, *new_stpp and *new_lopp should be malloc'd before the call, * in case they are used. */ int nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp, struct nfslock **new_lopp, struct nfslockconflict *cfp, nfsquad_t clientid, nfsv4stateid_t *stateidp, __unused struct nfsexstuff *exp, struct nfsrv_descript *nd, NFSPROC_T *p) { struct nfslock *lop; struct nfsstate *new_stp = *new_stpp; struct nfslock *new_lop = *new_lopp; struct nfsstate *tstp, *mystp, *nstp; int specialid = 0; struct nfslockfile *lfp; struct nfslock *other_lop = NULL; struct nfsstate *stp, *lckstp = NULL; struct nfsclient *clp = NULL; u_int32_t bits; int error = 0, haslock = 0, ret, reterr; int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0; fhandle_t nfh; uint64_t first, end; uint32_t lock_flags; if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) { /* * Note the special cases of "all 1s" or "all 0s" stateids and * let reads with all 1s go ahead. */ if (new_stp->ls_stateid.seqid == 0x0 && new_stp->ls_stateid.other[0] == 0x0 && new_stp->ls_stateid.other[1] == 0x0 && new_stp->ls_stateid.other[2] == 0x0) specialid = 1; else if (new_stp->ls_stateid.seqid == 0xffffffff && new_stp->ls_stateid.other[0] == 0xffffffff && new_stp->ls_stateid.other[1] == 0xffffffff && new_stp->ls_stateid.other[2] == 0xffffffff) specialid = 2; } /* * Check for restart conditions (client and server). */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, specialid); if (error) goto out; /* * Check for state resource limit exceeded. */ if ((new_stp->ls_flags & NFSLCK_LOCK) && nfsrv_openpluslock > nfsrv_v4statelimit) { error = NFSERR_RESOURCE; goto out; } /* * For the lock case, get another nfslock structure, * just in case we need it. * Malloc now, before we start sifting through the linked lists, * in case we have to wait for memory. */ tryagain: if (new_stp->ls_flags & NFSLCK_LOCK) other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK); filestruct_locked = 0; reterr = 0; lfp = NULL; /* * Get the lockfile structure for CFH now, so we can do a sanity * check against the stateid, before incrementing the seqid#, since * we want to return NFSERR_BADSTATEID on failure and the seqid# * shouldn't be incremented for this case. * If nfsrv_getlockfile() returns -1, it means "not found", which * will be handled later. * If we are doing Lock/LockU and local locking is enabled, sleep * lock the nfslockfile structure. */ getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p); NFSLOCKSTATE(); if (getlckret == 0) { if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 && nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) { getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL, &lfp, &nfh, 1); if (getlckret == 0) filestruct_locked = 1; } else getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL, &lfp, &nfh, 0); } if (getlckret != 0 && getlckret != -1) reterr = getlckret; if (filestruct_locked != 0) { LIST_INIT(&lfp->lf_rollback); if ((new_stp->ls_flags & NFSLCK_LOCK)) { /* * For local locking, do the advisory locking now, so * that any conflict can be detected. A failure later * can be rolled back locally. If an error is returned, * struct nfslockfile has been unlocked and any local * locking rolled back. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } reterr = nfsrv_locallock(vp, lfp, (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)), new_lop->lo_first, new_lop->lo_end, cfp, p); NFSLOCKSTATE(); } } if (specialid == 0) { if (new_stp->ls_flags & NFSLCK_TEST) { /* * RFC 3530 does not list LockT as an op that renews a * lease, but the consensus seems to be that it is ok * for a server to do so. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); /* * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid * error returns for LockT, just go ahead and test for a lock, * since there are no locks for this client, but other locks * can conflict. (ie. same client will always be false) */ if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED) error = 0; lckstp = new_stp; } else { error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error == 0) /* * Look up the stateid */ error = nfsrv_getstate(clp, &new_stp->ls_stateid, new_stp->ls_flags, &stp); /* * do some sanity checks for an unconfirmed open or a * stateid that refers to the wrong file, for an open stateid */ if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) && ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) || (getlckret == 0 && stp->ls_lfp != lfp))){ /* * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID * The only exception is using SETATTR with SIZE. * */ if ((new_stp->ls_flags & (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR) error = NFSERR_BADSTATEID; } if (error == 0 && (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) && getlckret == 0 && stp->ls_lfp != lfp) error = NFSERR_BADSTATEID; /* * If the lockowner stateid doesn't refer to the same file, * I believe that is considered ok, since some clients will * only create a single lockowner and use that for all locks * on all files. * For now, log it as a diagnostic, instead of considering it * a BadStateid. */ if (error == 0 && (stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 && getlckret == 0 && stp->ls_lfp != lfp) { #ifdef DIAGNOSTIC printf("Got a lock statid for different file open\n"); #endif /* error = NFSERR_BADSTATEID; */ } if (error == 0) { if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) { /* * If haslock set, we've already checked the seqid. */ if (!haslock) { if (stp->ls_flags & NFSLCK_OPEN) error = nfsrv_checkseqid(nd, new_stp->ls_seq, stp->ls_openowner, new_stp->ls_op); else error = NFSERR_BADSTATEID; } if (!error) nfsrv_getowner(&stp->ls_open, new_stp, &lckstp); if (lckstp) { /* * For NFSv4.1 and NFSv4.2 allow an * open_to_lock_owner when the lock_owner already * exists. Just clear NFSLCK_OPENTOLOCK so that * a new lock_owner will not be created. * RFC7530 states that the error for NFSv4.0 * is NFS4ERR_BAD_SEQID. */ if ((nd->nd_flag & ND_NFSV41) != 0) new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK; else error = NFSERR_BADSEQID; } else lckstp = new_stp; } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) { /* * If haslock set, ditto above. */ if (!haslock) { if (stp->ls_flags & NFSLCK_OPEN) error = NFSERR_BADSTATEID; else error = nfsrv_checkseqid(nd, new_stp->ls_seq, stp, new_stp->ls_op); } lckstp = stp; } else { lckstp = stp; } } /* * If the seqid part of the stateid isn't the same, return * NFSERR_OLDSTATEID for cases other than I/O Ops. * For I/O Ops, only return NFSERR_OLDSTATEID if * nfsrv_returnoldstateid is set. (The consensus on the email * list was that most clients would prefer to not receive * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that * is what will happen, so I use the nfsrv_returnoldstateid to * allow for either server configuration.) */ if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid && (((nd->nd_flag & ND_NFSV41) == 0 && (!(new_stp->ls_flags & NFSLCK_CHECK) || nfsrv_returnoldstateid)) || ((nd->nd_flag & ND_NFSV41) != 0 && new_stp->ls_stateid.seqid != 0))) error = NFSERR_OLDSTATEID; } } /* * Now we can check for grace. */ if (!error) error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags); if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error && nfsrv_checkstable(clp)) error = NFSERR_NOGRACE; /* * If we successfully Reclaimed state, note that. */ if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error) nfsrv_markstable(clp); /* * At this point, either error == NFSERR_BADSTATEID or the * seqid# has been updated, so we can return any error. * If error == 0, there may be an error in: * nd_repstat - Set by the calling function. * reterr - Set above, if getting the nfslockfile structure * or acquiring the local lock failed. * (If both of these are set, nd_repstat should probably be * returned, since that error was detected before this * function call.) */ if (error != 0 || nd->nd_repstat != 0 || reterr != 0) { if (error == 0) { if (nd->nd_repstat != 0) error = nd->nd_repstat; else error = reterr; } if (filestruct_locked != 0) { /* Roll back local locks. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } NFSUNLOCKSTATE(); goto out; } /* * Check the nfsrv_getlockfile return. * Returned -1 if no structure found. */ if (getlckret == -1) { error = NFSERR_EXPIRED; /* * Called from lockt, so no lock is OK. */ if (new_stp->ls_flags & NFSLCK_TEST) { error = 0; } else if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) { /* * Called to check for a lock, OK if the stateid is all * 1s or all 0s, but there should be an nfsstate * otherwise. * (ie. If there is no open, I'll assume no share * deny bits.) */ if (specialid) error = 0; else error = NFSERR_BADSTATEID; } NFSUNLOCKSTATE(); goto out; } /* * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict. * For NFSLCK_CHECK, allow a read if write access is granted, * but check for a deny. For NFSLCK_LOCK, require correct access, * which implies a conflicting deny can't exist. */ if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) { /* * Four kinds of state id: * - specialid (all 0s or all 1s), only for NFSLCK_CHECK * - stateid for an open * - stateid for a delegation * - stateid for a lock owner */ if (!specialid) { if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) { delegation = 1; mystp = stp; nfsrv_delaydelegtimeout(stp); } else if (stp->ls_flags & NFSLCK_OPEN) { mystp = stp; } else { mystp = stp->ls_openstp; } /* * If locking or checking, require correct access * bit set. */ if (((new_stp->ls_flags & NFSLCK_LOCK) && !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) & mystp->ls_flags & NFSLCK_ACCESSBITS)) || ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) == (NFSLCK_CHECK | NFSLCK_READACCESS) && !(mystp->ls_flags & NFSLCK_READACCESS) && nfsrv_allowreadforwriteopen == 0) || ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) == (NFSLCK_CHECK | NFSLCK_WRITEACCESS) && !(mystp->ls_flags & NFSLCK_WRITEACCESS))) { if (filestruct_locked != 0) { /* Roll back local locks. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl3"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } NFSUNLOCKSTATE(); error = NFSERR_OPENMODE; goto out; } } else mystp = NULL; if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) { /* * Check for a conflicting deny bit. */ LIST_FOREACH(tstp, &lfp->lf_open, ls_file) { if (tstp != mystp) { bits = tstp->ls_flags; bits >>= NFSLCK_SHIFT; if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) { KASSERT(vnode_unlocked == 0, ("nfsrv_lockctrl: vnode unlocked1")); ret = nfsrv_clientconflict(tstp->ls_clp, &haslock, vp, p); if (ret == 1) { /* * nfsrv_clientconflict unlocks state * when it returns non-zero. */ lckstp = NULL; goto tryagain; } if (ret == 0) NFSUNLOCKSTATE(); if (ret == 2) error = NFSERR_PERM; else error = NFSERR_OPENMODE; goto out; } } } /* We're outta here */ NFSUNLOCKSTATE(); goto out; } } /* * For setattr, just get rid of all the Delegations for other clients. */ if (new_stp->ls_flags & NFSLCK_SETATTR) { KASSERT(vnode_unlocked == 0, ("nfsrv_lockctrl: vnode unlocked2")); ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p); if (ret) { /* * nfsrv_cleandeleg() unlocks state when it * returns non-zero. */ if (ret == -1) { lckstp = NULL; goto tryagain; } error = ret; goto out; } if (!(new_stp->ls_flags & NFSLCK_CHECK) || (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg))) { NFSUNLOCKSTATE(); goto out; } } /* * Check for a conflicting delegation. If one is found, call * nfsrv_delegconflict() to handle it. If the v4root lock hasn't * been set yet, it will get the lock. Otherwise, it will recall * the delegation. Then, we try try again... * I currently believe the conflict algorithm to be: * For Lock Ops (Lock/LockT/LockU) * - there is a conflict iff a different client has a write delegation * For Reading (Read Op) * - there is a conflict iff a different client has a write delegation * (the specialids are always a different client) * For Writing (Write/Setattr of size) * - there is a conflict if a different client has any delegation * - there is a conflict if the same client has a read delegation * (I don't understand why this isn't allowed, but that seems to be * the current consensus?) */ tstp = LIST_FIRST(&lfp->lf_deleg); while (tstp != LIST_END(&lfp->lf_deleg)) { nstp = LIST_NEXT(tstp, ls_file); if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))|| ((new_stp->ls_flags & NFSLCK_CHECK) && (new_lop->lo_flags & NFSLCK_READ))) && clp != tstp->ls_clp && (tstp->ls_flags & NFSLCK_DELEGWRITE)) || ((new_stp->ls_flags & NFSLCK_CHECK) && (new_lop->lo_flags & NFSLCK_WRITE) && (clp != tstp->ls_clp || (tstp->ls_flags & NFSLCK_DELEGREAD)))) { ret = 0; if (filestruct_locked != 0) { /* Roll back local locks. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4"); NFSVOPUNLOCK(vp); } nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); NFSUNLOCKSTATE(); NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); vnode_unlocked = 0; if (VN_IS_DOOMED(vp)) ret = NFSERR_SERVERFAULT; NFSLOCKSTATE(); } if (ret == 0) ret = nfsrv_delegconflict(tstp, &haslock, p, vp); if (ret) { /* * nfsrv_delegconflict unlocks state when it * returns non-zero, which it always does. */ if (other_lop) { free(other_lop, M_NFSDLOCK); other_lop = NULL; } if (ret == -1) { lckstp = NULL; goto tryagain; } error = ret; goto out; } /* Never gets here. */ } tstp = nstp; } /* * Handle the unlock case by calling nfsrv_updatelock(). * (Should I have done some access checking above for unlock? For now, * just let it happen.) */ if (new_stp->ls_flags & NFSLCK_UNLOCK) { first = new_lop->lo_first; end = new_lop->lo_end; nfsrv_updatelock(stp, new_lopp, &other_lop, lfp); stateidp->seqid = ++(stp->ls_stateid.seqid); if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) stateidp->seqid = stp->ls_stateid.seqid = 1; stateidp->other[0] = stp->ls_stateid.other[0]; stateidp->other[1] = stp->ls_stateid.other[1]; stateidp->other[2] = stp->ls_stateid.other[2]; if (filestruct_locked != 0) { NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } /* Update the local locks. */ nfsrv_localunlock(vp, lfp, first, end, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } NFSUNLOCKSTATE(); goto out; } /* * Search for a conflicting lock. A lock conflicts if: * - the lock range overlaps and * - at least one lock is a write lock and * - it is not owned by the same lock owner */ if (!delegation) { LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) { if (new_lop->lo_end > lop->lo_first && new_lop->lo_first < lop->lo_end && (new_lop->lo_flags == NFSLCK_WRITE || lop->lo_flags == NFSLCK_WRITE) && lckstp != lop->lo_stp && (clp != lop->lo_stp->ls_clp || lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen || NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner, lckstp->ls_ownerlen))) { if (other_lop) { free(other_lop, M_NFSDLOCK); other_lop = NULL; } if (vnode_unlocked != 0) ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock, NULL, p); else ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock, vp, p); if (ret == 1) { if (filestruct_locked != 0) { if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6"); NFSVOPUNLOCK(vp); } /* Roll back local locks. */ nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); NFSUNLOCKSTATE(); NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); vnode_unlocked = 0; if (VN_IS_DOOMED(vp)) { error = NFSERR_SERVERFAULT; goto out; } } /* * nfsrv_clientconflict() unlocks state when it * returns non-zero. */ lckstp = NULL; goto tryagain; } /* * Found a conflicting lock, so record the conflict and * return the error. */ if (cfp != NULL && ret == 0) { cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0]; cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1]; cfp->cl_first = lop->lo_first; cfp->cl_end = lop->lo_end; cfp->cl_flags = lop->lo_flags; cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen; NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner, cfp->cl_ownerlen); } if (ret == 2) error = NFSERR_PERM; else if (new_stp->ls_flags & NFSLCK_RECLAIM) error = NFSERR_RECLAIMCONFLICT; else if (new_stp->ls_flags & NFSLCK_CHECK) error = NFSERR_LOCKED; else error = NFSERR_DENIED; if (filestruct_locked != 0 && ret == 0) { /* Roll back local locks. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } if (ret == 0) NFSUNLOCKSTATE(); goto out; } } } /* * We only get here if there was no lock that conflicted. */ if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) { NFSUNLOCKSTATE(); goto out; } /* * We only get here when we are creating or modifying a lock. * There are two variants: * - exist_lock_owner where lock_owner exists * - open_to_lock_owner with new lock_owner */ first = new_lop->lo_first; end = new_lop->lo_end; lock_flags = new_lop->lo_flags; if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) { nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp); stateidp->seqid = ++(lckstp->ls_stateid.seqid); if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) stateidp->seqid = lckstp->ls_stateid.seqid = 1; stateidp->other[0] = lckstp->ls_stateid.other[0]; stateidp->other[1] = lckstp->ls_stateid.other[1]; stateidp->other[2] = lckstp->ls_stateid.other[2]; } else { /* * The new open_to_lock_owner case. * Link the new nfsstate into the lists. */ new_stp->ls_seq = new_stp->ls_opentolockseq; nfsrvd_refcache(new_stp->ls_op); stateidp->seqid = new_stp->ls_stateid.seqid = 1; stateidp->other[0] = new_stp->ls_stateid.other[0] = clp->lc_clientid.lval[0]; stateidp->other[1] = new_stp->ls_stateid.other[1] = clp->lc_clientid.lval[1]; stateidp->other[2] = new_stp->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_stp->ls_clp = clp; LIST_INIT(&new_stp->ls_lock); new_stp->ls_openstp = stp; new_stp->ls_lfp = lfp; nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp, lfp); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid), new_stp, ls_hash); LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list); *new_lopp = NULL; *new_stpp = NULL; nfsstatsv1.srvlockowners++; nfsrv_openpluslock++; } if (filestruct_locked != 0) { NFSUNLOCKSTATE(); nfsrv_locallock_commit(lfp, lock_flags, first, end); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } NFSUNLOCKSTATE(); out: if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } if (vnode_unlocked != 0) { NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (error == 0 && VN_IS_DOOMED(vp)) error = NFSERR_SERVERFAULT; } if (other_lop) free(other_lop, M_NFSDLOCK); NFSEXITCODE2(error, nd); return (error); } /* * Check for state errors for Open. * repstat is passed back out as an error if more critical errors * are not detected. */ int nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd, NFSPROC_T *p, int repstat) { struct nfsstate *stp, *nstp; struct nfsclient *clp; struct nfsstate *ownerstp; struct nfslockfile *lfp, *new_lfp; int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0; if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) readonly = 1; /* * Check for restart conditions (client and server). */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, 0); if (error) goto out; /* * Check for state resource limit exceeded. * Technically this should be SMP protected, but the worst * case error is "out by one or two" on the count when it * returns NFSERR_RESOURCE and the limit is just a rather * arbitrary high water mark, so no harm is done. */ if (nfsrv_openpluslock > nfsrv_v4statelimit) { error = NFSERR_RESOURCE; goto out; } tryagain: new_lfp = malloc(sizeof (struct nfslockfile), M_NFSDLOCKFILE, M_WAITOK); if (vp) getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp, NULL, p); NFSLOCKSTATE(); /* * Get the nfsclient structure. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); /* * Look up the open owner. See if it needs confirmation and * check the seq#, as required. */ if (!error) nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp); if (!error && ownerstp) { error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp, new_stp->ls_op); /* * If the OpenOwner hasn't been confirmed, assume the * old one was a replay and this one is ok. * See: RFC3530 Sec. 14.2.18. */ if (error == NFSERR_BADSEQID && (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM)) error = 0; } /* * Check for grace. */ if (!error) error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags); if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error && nfsrv_checkstable(clp)) error = NFSERR_NOGRACE; /* * If none of the above errors occurred, let repstat be * returned. */ if (repstat && !error) error = repstat; if (error) { NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } free(new_lfp, M_NFSDLOCKFILE); goto out; } /* * If vp == NULL, the file doesn't exist yet, so return ok. * (This always happens on the first pass, so haslock must be 0.) */ if (vp == NULL) { NFSUNLOCKSTATE(); free(new_lfp, M_NFSDLOCKFILE); goto out; } /* * Get the structure for the underlying file. */ if (getfhret) error = getfhret; else error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp, NULL, 0); if (new_lfp) free(new_lfp, M_NFSDLOCKFILE); if (error) { NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } goto out; } /* * Search for a conflicting open/share. */ if (new_stp->ls_flags & NFSLCK_DELEGCUR) { /* * For Delegate_Cur, search for the matching Delegation, * which indicates no conflict. * An old delegation should have been recovered by the * client doing a Claim_DELEGATE_Prev, so I won't let * it match and return NFSERR_EXPIRED. Should I let it * match? */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if (!(stp->ls_flags & NFSLCK_OLDDELEG) && (((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) || stateidp->seqid == stp->ls_stateid.seqid) && !NFSBCMP(stateidp->other, stp->ls_stateid.other, NFSX_STATEIDOTHER)) break; } if (stp == LIST_END(&lfp->lf_deleg) || ((new_stp->ls_flags & NFSLCK_WRITEACCESS) && (stp->ls_flags & NFSLCK_DELEGREAD))) { NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_EXPIRED; goto out; } } /* * Check for access/deny bit conflicts. I check for the same * owner as well, in case the client didn't bother. */ LIST_FOREACH(stp, &lfp->lf_open, ls_file) { if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) && (((new_stp->ls_flags & NFSLCK_ACCESSBITS) & ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))|| ((stp->ls_flags & NFSLCK_ACCESSBITS) & ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){ ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p); if (ret == 1) { /* * nfsrv_clientconflict() unlocks * state when it returns non-zero. */ goto tryagain; } if (ret == 2) error = NFSERR_PERM; else if (new_stp->ls_flags & NFSLCK_RECLAIM) error = NFSERR_RECLAIMCONFLICT; else error = NFSERR_SHAREDENIED; if (ret == 0) NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } goto out; } } /* * Check for a conflicting delegation. If one is found, call * nfsrv_delegconflict() to handle it. If the v4root lock hasn't * been set yet, it will get the lock. Otherwise, it will recall * the delegation. Then, we try try again... * (If NFSLCK_DELEGCUR is set, it has a delegation, so there * isn't a conflict.) * I currently believe the conflict algorithm to be: * For Open with Read Access and Deny None * - there is a conflict iff a different client has a write delegation * For Open with other Write Access or any Deny except None * - there is a conflict if a different client has any delegation * - there is a conflict if the same client has a read delegation * (The current consensus is that this last case should be * considered a conflict since the client with a read delegation * could have done an Open with ReadAccess and WriteDeny * locally and then not have checked for the WriteDeny.) * Don't check for a Reclaim, since that will be dealt with * by nfsrv_openctrl(). */ if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) { stp = LIST_FIRST(&lfp->lf_deleg); while (stp != LIST_END(&lfp->lf_deleg)) { nstp = LIST_NEXT(stp, ls_file); if ((readonly && stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGWRITE)) || (!readonly && (stp->ls_clp != clp || (stp->ls_flags & NFSLCK_DELEGREAD)))) { ret = nfsrv_delegconflict(stp, &haslock, p, vp); if (ret) { /* * nfsrv_delegconflict() unlocks state * when it returns non-zero. */ if (ret == -1) goto tryagain; error = ret; goto out; } } stp = nstp; } } NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } out: NFSEXITCODE2(error, nd); return (error); } /* * Open control function to create/update open state for an open. */ int nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp, struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp, nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp, NFSPROC_T *p, u_quad_t filerev) { struct nfsstate *new_stp = *new_stpp; struct nfsstate *stp, *nstp; struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg; struct nfslockfile *lfp, *new_lfp; struct nfsclient *clp; int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1; int readonly = 0, cbret = 1, getfhret = 0; int gotstate = 0, len = 0; u_char *clidp = NULL; if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) readonly = 1; /* * Check for restart conditions (client and server). * (Paranoia, should have been detected by nfsrv_opencheck().) * If an error does show up, return NFSERR_EXPIRED, since the * the seqid# has already been incremented. */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, 0); if (error) { printf("Nfsd: openctrl unexpected restart err=%d\n", error); error = NFSERR_EXPIRED; goto out; } clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK); tryagain: new_lfp = malloc(sizeof (struct nfslockfile), M_NFSDLOCKFILE, M_WAITOK); new_open = malloc(sizeof (struct nfsstate), M_NFSDSTATE, M_WAITOK); new_deleg = malloc(sizeof (struct nfsstate), M_NFSDSTATE, M_WAITOK); getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp, NULL, p); NFSLOCKSTATE(); /* * Get the client structure. Since the linked lists could be changed * by other nfsd processes if this process does a tsleep(), one of * two things must be done. * 1 - don't tsleep() * or * 2 - get the nfsv4_lock() { indicated by haslock == 1 } * before using the lists, since this lock stops the other * nfsd. This should only be used for rare cases, since it * essentially single threads the nfsd. * At this time, it is only done for cases where the stable * storage file must be written prior to completion of state * expiration. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) && clp->lc_program) { /* * This happens on the first open for a client * that supports callbacks. */ NFSUNLOCKSTATE(); /* * Although nfsrv_docallback() will sleep, clp won't * go away, since they are only removed when the * nfsv4_lock() has blocked the nfsd threads. The * fields in clp can change, but having multiple * threads do this Null callback RPC should be * harmless. */ cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL, NULL, 0, NULL, NULL, NULL, 0, p); NFSLOCKSTATE(); clp->lc_flags &= ~LCL_NEEDSCBNULL; if (!cbret) clp->lc_flags |= LCL_CALLBACKSON; } /* * Look up the open owner. See if it needs confirmation and * check the seq#, as required. */ if (!error) nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp); if (error) { NFSUNLOCKSTATE(); printf("Nfsd: openctrl unexpected state err=%d\n", error); free(new_lfp, M_NFSDLOCKFILE); free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_EXPIRED; goto out; } if (new_stp->ls_flags & NFSLCK_RECLAIM) nfsrv_markstable(clp); /* * Get the structure for the underlying file. */ if (getfhret) error = getfhret; else error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp, NULL, 0); if (new_lfp) free(new_lfp, M_NFSDLOCKFILE); if (error) { NFSUNLOCKSTATE(); printf("Nfsd openctrl unexpected getlockfile err=%d\n", error); free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } goto out; } /* * Search for a conflicting open/share. */ if (new_stp->ls_flags & NFSLCK_DELEGCUR) { /* * For Delegate_Cur, search for the matching Delegation, * which indicates no conflict. * An old delegation should have been recovered by the * client doing a Claim_DELEGATE_Prev, so I won't let * it match and return NFSERR_EXPIRED. Should I let it * match? */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if (!(stp->ls_flags & NFSLCK_OLDDELEG) && (((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) || stateidp->seqid == stp->ls_stateid.seqid) && !NFSBCMP(stateidp->other, stp->ls_stateid.other, NFSX_STATEIDOTHER)) break; } if (stp == LIST_END(&lfp->lf_deleg) || ((new_stp->ls_flags & NFSLCK_WRITEACCESS) && (stp->ls_flags & NFSLCK_DELEGREAD))) { NFSUNLOCKSTATE(); printf("Nfsd openctrl unexpected expiry\n"); free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_EXPIRED; goto out; } /* * Don't issue a Delegation, since one already exists and * delay delegation timeout, as required. */ delegate = 0; nfsrv_delaydelegtimeout(stp); } /* * Check for access/deny bit conflicts. I also check for the * same owner, since the client might not have bothered to check. * Also, note an open for the same file and owner, if found, * which is all we do here for Delegate_Cur, since conflict * checking is already done. */ LIST_FOREACH(stp, &lfp->lf_open, ls_file) { if (ownerstp && stp->ls_openowner == ownerstp) openstp = stp; if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) { /* * If another client has the file open, the only * delegation that can be issued is a Read delegation * and only if it is a Read open with Deny none. */ if (clp != stp->ls_clp) { if ((stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) writedeleg = 0; else delegate = 0; } if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) & ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))|| ((stp->ls_flags & NFSLCK_ACCESSBITS) & ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){ ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p); if (ret == 1) { /* * nfsrv_clientconflict() unlocks state * when it returns non-zero. */ free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); openstp = NULL; goto tryagain; } if (ret == 2) error = NFSERR_PERM; else if (new_stp->ls_flags & NFSLCK_RECLAIM) error = NFSERR_RECLAIMCONFLICT; else error = NFSERR_SHAREDENIED; if (ret == 0) NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); printf("nfsd openctrl unexpected client cnfl\n"); goto out; } } } /* * Check for a conflicting delegation. If one is found, call * nfsrv_delegconflict() to handle it. If the v4root lock hasn't * been set yet, it will get the lock. Otherwise, it will recall * the delegation. Then, we try try again... * (If NFSLCK_DELEGCUR is set, it has a delegation, so there * isn't a conflict.) * I currently believe the conflict algorithm to be: * For Open with Read Access and Deny None * - there is a conflict iff a different client has a write delegation * For Open with other Write Access or any Deny except None * - there is a conflict if a different client has any delegation * - there is a conflict if the same client has a read delegation * (The current consensus is that this last case should be * considered a conflict since the client with a read delegation * could have done an Open with ReadAccess and WriteDeny * locally and then not have checked for the WriteDeny.) */ if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) { stp = LIST_FIRST(&lfp->lf_deleg); while (stp != LIST_END(&lfp->lf_deleg)) { nstp = LIST_NEXT(stp, ls_file); if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD)) writedeleg = 0; else delegate = 0; if ((readonly && stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGWRITE)) || (!readonly && (stp->ls_clp != clp || (stp->ls_flags & NFSLCK_DELEGREAD)))) { if (new_stp->ls_flags & NFSLCK_RECLAIM) { delegate = 2; } else { ret = nfsrv_delegconflict(stp, &haslock, p, vp); if (ret) { /* * nfsrv_delegconflict() unlocks state * when it returns non-zero. */ printf("Nfsd openctrl unexpected deleg cnfl\n"); free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); if (ret == -1) { openstp = NULL; goto tryagain; } error = ret; goto out; } } } stp = nstp; } } /* * We only get here if there was no open that conflicted. * If an open for the owner exists, or in the access/deny bits. * Otherwise it is a new open. If the open_owner hasn't been * confirmed, replace the open with the new one needing confirmation, * otherwise add the open. */ if (new_stp->ls_flags & NFSLCK_DELEGPREV) { /* * Handle NFSLCK_DELEGPREV by searching the old delegations for * a match. If found, just move the old delegation to the current * delegation list and issue open. If not found, return * NFSERR_EXPIRED. */ LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) { if (stp->ls_lfp == lfp) { /* Found it */ if (stp->ls_clp != clp) panic("olddeleg clp"); LIST_REMOVE(stp, ls_list); LIST_REMOVE(stp, ls_hash); stp->ls_flags &= ~NFSLCK_OLDDELEG; stp->ls_stateid.seqid = delegstateidp->seqid = 1; stp->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; stp->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; stp->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); stp->ls_compref = nd->nd_compref; LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list); LIST_INSERT_HEAD(NFSSTATEHASH(clp, stp->ls_stateid), stp, ls_hash); if (stp->ls_flags & NFSLCK_DELEGWRITE) *rflagsp |= NFSV4OPEN_WRITEDELEGATE; else *rflagsp |= NFSV4OPEN_READDELEGATE; clp->lc_delegtime = NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA; /* * Now, do the associated open. */ new_open->ls_stateid.seqid = 1; new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)| NFSLCK_OPEN; if (stp->ls_flags & NFSLCK_DELEGWRITE) new_open->ls_flags |= (NFSLCK_READACCESS | NFSLCK_WRITEACCESS); else new_open->ls_flags |= NFSLCK_READACCESS; new_open->ls_uid = new_stp->ls_uid; new_open->ls_lfp = lfp; new_open->ls_clp = clp; LIST_INIT(&new_open->ls_open); LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), new_open, ls_hash); /* * and handle the open owner */ if (ownerstp) { new_open->ls_openowner = ownerstp; LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list); } else { new_open->ls_openowner = new_stp; new_stp->ls_flags = 0; nfsrvd_refcache(new_stp->ls_op); new_stp->ls_noopens = 0; LIST_INIT(&new_stp->ls_open); LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list); LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list); *new_stpp = NULL; nfsstatsv1.srvopenowners++; nfsrv_openpluslock++; } openstp = new_open; new_open = NULL; nfsstatsv1.srvopens++; nfsrv_openpluslock++; break; } } if (stp == LIST_END(&clp->lc_olddeleg)) error = NFSERR_EXPIRED; } else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) { /* * Scan to see that no delegation for this client and file * doesn't already exist. * There also shouldn't yet be an Open for this file and * openowner. */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if (stp->ls_clp == clp) break; } if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) { /* * This is the Claim_Previous case with a delegation * type != Delegate_None. */ /* * First, add the delegation. (Although we must issue the * delegation, we can also ask for an immediate return.) */ new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; new_deleg->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; new_deleg->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; new_deleg->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); if (new_stp->ls_flags & NFSLCK_DELEGWRITE) { new_deleg->ls_flags = (NFSLCK_DELEGWRITE | NFSLCK_READACCESS | NFSLCK_WRITEACCESS); *rflagsp |= NFSV4OPEN_WRITEDELEGATE; nfsrv_writedelegcnt++; } else { new_deleg->ls_flags = (NFSLCK_DELEGREAD | NFSLCK_READACCESS); *rflagsp |= NFSV4OPEN_READDELEGATE; } new_deleg->ls_uid = new_stp->ls_uid; new_deleg->ls_lfp = lfp; new_deleg->ls_clp = clp; new_deleg->ls_filerev = filerev; new_deleg->ls_compref = nd->nd_compref; new_deleg->ls_lastrecall = 0; LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid), new_deleg, ls_hash); LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); new_deleg = NULL; if (delegate == 2 || nfsrv_issuedelegs == 0 || (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != LCL_CALLBACKSON || NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) || !NFSVNO_DELEGOK(vp)) *rflagsp |= NFSV4OPEN_RECALL; nfsstatsv1.srvdelegates++; nfsrv_openpluslock++; nfsrv_delegatecnt++; /* * Now, do the associated open. */ new_open->ls_stateid.seqid = 1; new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) | NFSLCK_OPEN; if (new_stp->ls_flags & NFSLCK_DELEGWRITE) new_open->ls_flags |= (NFSLCK_READACCESS | NFSLCK_WRITEACCESS); else new_open->ls_flags |= NFSLCK_READACCESS; new_open->ls_uid = new_stp->ls_uid; new_open->ls_lfp = lfp; new_open->ls_clp = clp; LIST_INIT(&new_open->ls_open); LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), new_open, ls_hash); /* * and handle the open owner */ if (ownerstp) { new_open->ls_openowner = ownerstp; LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list); } else { new_open->ls_openowner = new_stp; new_stp->ls_flags = 0; nfsrvd_refcache(new_stp->ls_op); new_stp->ls_noopens = 0; LIST_INIT(&new_stp->ls_open); LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list); LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list); *new_stpp = NULL; nfsstatsv1.srvopenowners++; nfsrv_openpluslock++; } openstp = new_open; new_open = NULL; nfsstatsv1.srvopens++; nfsrv_openpluslock++; } else { error = NFSERR_RECLAIMCONFLICT; } } else if (ownerstp) { if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) { /* Replace the open */ if (ownerstp->ls_op) nfsrvd_derefcache(ownerstp->ls_op); ownerstp->ls_op = new_stp->ls_op; nfsrvd_refcache(ownerstp->ls_op); ownerstp->ls_seq = new_stp->ls_seq; *rflagsp |= NFSV4OPEN_RESULTCONFIRM; stp = LIST_FIRST(&ownerstp->ls_open); stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) | NFSLCK_OPEN; stp->ls_stateid.seqid = 1; stp->ls_uid = new_stp->ls_uid; if (lfp != stp->ls_lfp) { LIST_REMOVE(stp, ls_file); LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file); stp->ls_lfp = lfp; } openstp = stp; } else if (openstp) { openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS); openstp->ls_stateid.seqid++; if ((nd->nd_flag & ND_NFSV41) != 0 && openstp->ls_stateid.seqid == 0) openstp->ls_stateid.seqid = 1; /* * This is where we can choose to issue a delegation. */ if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0) *rflagsp |= NFSV4OPEN_WDNOTWANTED; else if (nfsrv_issuedelegs == 0) *rflagsp |= NFSV4OPEN_WDSUPPFTYPE; else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt)) *rflagsp |= NFSV4OPEN_WDRESOURCE; else if (delegate == 0 || writedeleg == 0 || NFSVNO_EXRDONLY(exp) || (readonly != 0 && nfsrv_writedelegifpos == 0) || !NFSVNO_DELEGOK(vp) || (new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 || (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != LCL_CALLBACKSON) *rflagsp |= NFSV4OPEN_WDCONTENTION; else { new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; new_deleg->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; new_deleg->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; new_deleg->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); new_deleg->ls_flags = (NFSLCK_DELEGWRITE | NFSLCK_READACCESS | NFSLCK_WRITEACCESS); *rflagsp |= NFSV4OPEN_WRITEDELEGATE; new_deleg->ls_uid = new_stp->ls_uid; new_deleg->ls_lfp = lfp; new_deleg->ls_clp = clp; new_deleg->ls_filerev = filerev; new_deleg->ls_compref = nd->nd_compref; new_deleg->ls_lastrecall = 0; nfsrv_writedelegcnt++; LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid), new_deleg, ls_hash); LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); new_deleg = NULL; nfsstatsv1.srvdelegates++; nfsrv_openpluslock++; nfsrv_delegatecnt++; } } else { new_open->ls_stateid.seqid = 1; new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)| NFSLCK_OPEN; new_open->ls_uid = new_stp->ls_uid; new_open->ls_openowner = ownerstp; new_open->ls_lfp = lfp; new_open->ls_clp = clp; LIST_INIT(&new_open->ls_open); LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), new_open, ls_hash); openstp = new_open; new_open = NULL; nfsstatsv1.srvopens++; nfsrv_openpluslock++; /* * This is where we can choose to issue a delegation. */ if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0) *rflagsp |= NFSV4OPEN_WDNOTWANTED; else if (nfsrv_issuedelegs == 0) *rflagsp |= NFSV4OPEN_WDSUPPFTYPE; else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt)) *rflagsp |= NFSV4OPEN_WDRESOURCE; else if (delegate == 0 || (writedeleg == 0 && readonly == 0) || !NFSVNO_DELEGOK(vp) || (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != LCL_CALLBACKSON) *rflagsp |= NFSV4OPEN_WDCONTENTION; else { new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; new_deleg->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; new_deleg->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; new_deleg->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); if (writedeleg && !NFSVNO_EXRDONLY(exp) && (nfsrv_writedelegifpos || !readonly) && (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) { new_deleg->ls_flags = (NFSLCK_DELEGWRITE | NFSLCK_READACCESS | NFSLCK_WRITEACCESS); *rflagsp |= NFSV4OPEN_WRITEDELEGATE; nfsrv_writedelegcnt++; } else { new_deleg->ls_flags = (NFSLCK_DELEGREAD | NFSLCK_READACCESS); *rflagsp |= NFSV4OPEN_READDELEGATE; } new_deleg->ls_uid = new_stp->ls_uid; new_deleg->ls_lfp = lfp; new_deleg->ls_clp = clp; new_deleg->ls_filerev = filerev; new_deleg->ls_compref = nd->nd_compref; new_deleg->ls_lastrecall = 0; LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid), new_deleg, ls_hash); LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); new_deleg = NULL; nfsstatsv1.srvdelegates++; nfsrv_openpluslock++; nfsrv_delegatecnt++; } } } else { /* * New owner case. Start the open_owner sequence with a * Needs confirmation (unless a reclaim) and hang the * new open off it. */ new_open->ls_stateid.seqid = 1; new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) | NFSLCK_OPEN; new_open->ls_uid = new_stp->ls_uid; LIST_INIT(&new_open->ls_open); new_open->ls_openowner = new_stp; new_open->ls_lfp = lfp; new_open->ls_clp = clp; LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); if (new_stp->ls_flags & NFSLCK_RECLAIM) { new_stp->ls_flags = 0; } else if ((nd->nd_flag & ND_NFSV41) != 0) { /* NFSv4.1 never needs confirmation. */ new_stp->ls_flags = 0; /* * This is where we can choose to issue a delegation. */ if (delegate && nfsrv_issuedelegs && (writedeleg || readonly) && (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) == LCL_CALLBACKSON && !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) && NFSVNO_DELEGOK(vp) && ((nd->nd_flag & ND_NFSV41) == 0 || (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) { new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; new_deleg->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; new_deleg->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; new_deleg->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); if (writedeleg && !NFSVNO_EXRDONLY(exp) && (nfsrv_writedelegifpos || !readonly) && ((nd->nd_flag & ND_NFSV41) == 0 || (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0)) { new_deleg->ls_flags = (NFSLCK_DELEGWRITE | NFSLCK_READACCESS | NFSLCK_WRITEACCESS); *rflagsp |= NFSV4OPEN_WRITEDELEGATE; nfsrv_writedelegcnt++; } else { new_deleg->ls_flags = (NFSLCK_DELEGREAD | NFSLCK_READACCESS); *rflagsp |= NFSV4OPEN_READDELEGATE; } new_deleg->ls_uid = new_stp->ls_uid; new_deleg->ls_lfp = lfp; new_deleg->ls_clp = clp; new_deleg->ls_filerev = filerev; new_deleg->ls_compref = nd->nd_compref; new_deleg->ls_lastrecall = 0; LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid), new_deleg, ls_hash); LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); new_deleg = NULL; nfsstatsv1.srvdelegates++; nfsrv_openpluslock++; nfsrv_delegatecnt++; } /* * Since NFSv4.1 never does an OpenConfirm, the first * open state will be acquired here. */ if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) { clp->lc_flags |= LCL_STAMPEDSTABLE; len = clp->lc_idlen; NFSBCOPY(clp->lc_id, clidp, len); gotstate = 1; } } else { *rflagsp |= NFSV4OPEN_RESULTCONFIRM; new_stp->ls_flags = NFSLCK_NEEDSCONFIRM; } nfsrvd_refcache(new_stp->ls_op); new_stp->ls_noopens = 0; LIST_INIT(&new_stp->ls_open); LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list); LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), new_open, ls_hash); openstp = new_open; new_open = NULL; *new_stpp = NULL; nfsstatsv1.srvopens++; nfsrv_openpluslock++; nfsstatsv1.srvopenowners++; nfsrv_openpluslock++; } if (!error) { stateidp->seqid = openstp->ls_stateid.seqid; stateidp->other[0] = openstp->ls_stateid.other[0]; stateidp->other[1] = openstp->ls_stateid.other[1]; stateidp->other[2] = openstp->ls_stateid.other[2]; } NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } if (new_open) free(new_open, M_NFSDSTATE); if (new_deleg) free(new_deleg, M_NFSDSTATE); /* * If the NFSv4.1 client just acquired its first open, write a timestamp * to the stable storage file. */ if (gotstate != 0) { nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p); nfsrv_backupstable(); } out: free(clidp, M_TEMP); NFSEXITCODE2(error, nd); return (error); } /* * Open update. Does the confirm, downgrade and close. */ int nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p, int *retwriteaccessp) { struct nfsstate *stp; struct nfsclient *clp; struct nfslockfile *lfp; u_int32_t bits; int error = 0, gotstate = 0, len = 0; u_char *clidp = NULL; /* * Check for restart conditions (client and server). */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, 0); if (error) goto out; clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK); NFSLOCKSTATE(); /* * Get the open structure via clientid and stateid. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (!error) error = nfsrv_getstate(clp, &new_stp->ls_stateid, new_stp->ls_flags, &stp); /* * Sanity check the open. */ if (!error && (!(stp->ls_flags & NFSLCK_OPEN) || (!(new_stp->ls_flags & NFSLCK_CONFIRM) && (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) || ((new_stp->ls_flags & NFSLCK_CONFIRM) && (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))))) error = NFSERR_BADSTATEID; if (!error) error = nfsrv_checkseqid(nd, new_stp->ls_seq, stp->ls_openowner, new_stp->ls_op); if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid && (((nd->nd_flag & ND_NFSV41) == 0 && !(new_stp->ls_flags & NFSLCK_CONFIRM)) || ((nd->nd_flag & ND_NFSV41) != 0 && new_stp->ls_stateid.seqid != 0))) error = NFSERR_OLDSTATEID; if (!error && vp->v_type != VREG) { if (vp->v_type == VDIR) error = NFSERR_ISDIR; else error = NFSERR_INVAL; } if (error) { /* * If a client tries to confirm an Open with a bad * seqid# and there are no byte range locks or other Opens * on the openowner, just throw it away, so the next use of the * openowner will start a fresh seq#. */ if (error == NFSERR_BADSEQID && (new_stp->ls_flags & NFSLCK_CONFIRM) && nfsrv_nootherstate(stp)) nfsrv_freeopenowner(stp->ls_openowner, 0, p); NFSUNLOCKSTATE(); goto out; } /* * Set the return stateid. */ stateidp->seqid = stp->ls_stateid.seqid + 1; if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) stateidp->seqid = 1; stateidp->other[0] = stp->ls_stateid.other[0]; stateidp->other[1] = stp->ls_stateid.other[1]; stateidp->other[2] = stp->ls_stateid.other[2]; /* * Now, handle the three cases. */ if (new_stp->ls_flags & NFSLCK_CONFIRM) { /* * If the open doesn't need confirmation, it seems to me that * there is a client error, but I'll just log it and keep going? */ if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) printf("Nfsv4d: stray open confirm\n"); stp->ls_openowner->ls_flags = 0; stp->ls_stateid.seqid++; if ((nd->nd_flag & ND_NFSV41) != 0 && stp->ls_stateid.seqid == 0) stp->ls_stateid.seqid = 1; if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) { clp->lc_flags |= LCL_STAMPEDSTABLE; len = clp->lc_idlen; NFSBCOPY(clp->lc_id, clidp, len); gotstate = 1; } NFSUNLOCKSTATE(); } else if (new_stp->ls_flags & NFSLCK_CLOSE) { lfp = stp->ls_lfp; if (retwriteaccessp != NULL) { if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0) *retwriteaccessp = 1; else *retwriteaccessp = 0; } if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) { /* Get the lf lock */ nfsrv_locklf(lfp); NFSUNLOCKSTATE(); ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate"); NFSVOPUNLOCK(vp); if (nfsrv_freeopen(stp, vp, 1, p) == 0) { NFSLOCKSTATE(); nfsrv_unlocklf(lfp); NFSUNLOCKSTATE(); } NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); } else { (void) nfsrv_freeopen(stp, NULL, 0, p); NFSUNLOCKSTATE(); } } else { /* * Update the share bits, making sure that the new set are a * subset of the old ones. */ bits = (new_stp->ls_flags & NFSLCK_SHAREBITS); if (~(stp->ls_flags) & bits) { NFSUNLOCKSTATE(); error = NFSERR_INVAL; goto out; } stp->ls_flags = (bits | NFSLCK_OPEN); stp->ls_stateid.seqid++; if ((nd->nd_flag & ND_NFSV41) != 0 && stp->ls_stateid.seqid == 0) stp->ls_stateid.seqid = 1; NFSUNLOCKSTATE(); } /* * If the client just confirmed its first open, write a timestamp * to the stable storage file. */ if (gotstate != 0) { nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p); nfsrv_backupstable(); } out: free(clidp, M_TEMP); NFSEXITCODE2(error, nd); return (error); } /* * Delegation update. Does the purge and return. */ int nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid, nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred, NFSPROC_T *p, int *retwriteaccessp) { struct nfsstate *stp; struct nfsclient *clp; int error = 0; fhandle_t fh; /* * Do a sanity check against the file handle for DelegReturn. */ if (vp) { error = nfsvno_getfh(vp, &fh, p); if (error) goto out; } /* * Check for restart conditions (client and server). */ if (op == NFSV4OP_DELEGRETURN) error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN, stateidp, 0); else error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE, stateidp, 0); NFSLOCKSTATE(); /* * Get the open structure via clientid and stateid. */ if (!error) error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error) { if (error == NFSERR_CBPATHDOWN) error = 0; if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN) error = NFSERR_STALESTATEID; } if (!error && op == NFSV4OP_DELEGRETURN) { error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp); if (!error && stp->ls_stateid.seqid != stateidp->seqid && ((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0)) error = NFSERR_OLDSTATEID; } /* * NFSERR_EXPIRED means that the state has gone away, * so Delegations have been purged. Just return ok. */ if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) { NFSUNLOCKSTATE(); error = 0; goto out; } if (error) { NFSUNLOCKSTATE(); goto out; } if (op == NFSV4OP_DELEGRETURN) { if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh, sizeof (fhandle_t))) { NFSUNLOCKSTATE(); error = NFSERR_BADSTATEID; goto out; } if (retwriteaccessp != NULL) { if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0) *retwriteaccessp = 1; else *retwriteaccessp = 0; } nfsrv_freedeleg(stp); } else { nfsrv_freedeleglist(&clp->lc_olddeleg); } NFSUNLOCKSTATE(); error = 0; out: NFSEXITCODE(error); return (error); } /* * Release lock owner. */ int nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid, NFSPROC_T *p) { struct nfsstate *stp, *nstp, *openstp, *ownstp; struct nfsclient *clp; int error = 0; /* * Check for restart conditions (client and server). */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, 0); if (error) goto out; NFSLOCKSTATE(); /* * Get the lock owner by name. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, NULL, p); if (error) { NFSUNLOCKSTATE(); goto out; } LIST_FOREACH(ownstp, &clp->lc_open, ls_list) { LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) { stp = LIST_FIRST(&openstp->ls_open); while (stp != LIST_END(&openstp->ls_open)) { nstp = LIST_NEXT(stp, ls_list); /* * If the owner matches, check for locks and * then free or return an error. */ if (stp->ls_ownerlen == new_stp->ls_ownerlen && !NFSBCMP(stp->ls_owner, new_stp->ls_owner, stp->ls_ownerlen)){ if (LIST_EMPTY(&stp->ls_lock)) { nfsrv_freelockowner(stp, NULL, 0, p); } else { NFSUNLOCKSTATE(); error = NFSERR_LOCKSHELD; goto out; } } stp = nstp; } } } NFSUNLOCKSTATE(); out: NFSEXITCODE(error); return (error); } /* * Get the file handle for a lock structure. */ static int nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p) { fhandle_t *fhp = NULL; int error; /* * For lock, use the new nfslock structure, otherwise just * a fhandle_t on the stack. */ if (flags & NFSLCK_OPEN) { KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL")); fhp = &new_lfp->lf_fh; } else if (nfhp) { fhp = nfhp; } else { panic("nfsrv_getlockfh"); } error = nfsvno_getfh(vp, fhp, p); NFSEXITCODE(error); return (error); } /* * Get an nfs lock structure. Allocate one, as required, and return a * pointer to it. * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock. */ static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp, struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit) { struct nfslockfile *lfp; fhandle_t *fhp = NULL, *tfhp; struct nfslockhashhead *hp; struct nfslockfile *new_lfp = NULL; /* * For lock, use the new nfslock structure, otherwise just * a fhandle_t on the stack. */ if (flags & NFSLCK_OPEN) { new_lfp = *new_lfpp; fhp = &new_lfp->lf_fh; } else if (nfhp) { fhp = nfhp; } else { panic("nfsrv_getlockfile"); } hp = NFSLOCKHASH(fhp); LIST_FOREACH(lfp, hp, lf_hash) { tfhp = &lfp->lf_fh; if (NFSVNO_CMPFH(fhp, tfhp)) { if (lockit) nfsrv_locklf(lfp); *lfpp = lfp; return (0); } } if (!(flags & NFSLCK_OPEN)) return (-1); /* * No match, so chain the new one into the list. */ LIST_INIT(&new_lfp->lf_open); LIST_INIT(&new_lfp->lf_lock); LIST_INIT(&new_lfp->lf_deleg); LIST_INIT(&new_lfp->lf_locallock); LIST_INIT(&new_lfp->lf_rollback); new_lfp->lf_locallock_lck.nfslock_usecnt = 0; new_lfp->lf_locallock_lck.nfslock_lock = 0; new_lfp->lf_usecount = 0; LIST_INSERT_HEAD(hp, new_lfp, lf_hash); *lfpp = new_lfp; *new_lfpp = NULL; return (0); } /* * This function adds a nfslock lock structure to the list for the associated * nfsstate and nfslockfile structures. It will be inserted after the * entry pointed at by insert_lop. */ static void nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp) { struct nfslock *lop, *nlop; new_lop->lo_stp = stp; new_lop->lo_lfp = lfp; if (stp != NULL) { /* Insert in increasing lo_first order */ lop = LIST_FIRST(&lfp->lf_lock); if (lop == LIST_END(&lfp->lf_lock) || new_lop->lo_first <= lop->lo_first) { LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile); } else { nlop = LIST_NEXT(lop, lo_lckfile); while (nlop != LIST_END(&lfp->lf_lock) && nlop->lo_first < new_lop->lo_first) { lop = nlop; nlop = LIST_NEXT(lop, lo_lckfile); } LIST_INSERT_AFTER(lop, new_lop, lo_lckfile); } } else { new_lop->lo_lckfile.le_prev = NULL; /* list not used */ } /* * Insert after insert_lop, which is overloaded as stp or lfp for * an empty list. */ if (stp == NULL && (struct nfslockfile *)insert_lop == lfp) LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner); else if ((struct nfsstate *)insert_lop == stp) LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner); else LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner); if (stp != NULL) { nfsstatsv1.srvlocks++; nfsrv_openpluslock++; } } /* * This function updates the locking for a lock owner and given file. It * maintains a list of lock ranges ordered on increasing file offset that * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style). * It always adds new_lop to the list and sometimes uses the one pointed * at by other_lopp. */ static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp, struct nfslock **other_lopp, struct nfslockfile *lfp) { struct nfslock *new_lop = *new_lopp; struct nfslock *lop, *tlop, *ilop; struct nfslock *other_lop = *other_lopp; int unlock = 0, myfile = 0; u_int64_t tmp; /* * Work down the list until the lock is merged. */ if (new_lop->lo_flags & NFSLCK_UNLOCK) unlock = 1; if (stp != NULL) { ilop = (struct nfslock *)stp; lop = LIST_FIRST(&stp->ls_lock); } else { ilop = (struct nfslock *)lfp; lop = LIST_FIRST(&lfp->lf_locallock); } while (lop != NULL) { /* * Only check locks for this file that aren't before the start of * new lock's range. */ if (lop->lo_lfp == lfp) { myfile = 1; if (lop->lo_end >= new_lop->lo_first) { if (new_lop->lo_end < lop->lo_first) { /* * If the new lock ends before the start of the * current lock's range, no merge, just insert * the new lock. */ break; } if (new_lop->lo_flags == lop->lo_flags || (new_lop->lo_first <= lop->lo_first && new_lop->lo_end >= lop->lo_end)) { /* * This lock can be absorbed by the new lock/unlock. * This happens when it covers the entire range * of the old lock or is contiguous * with the old lock and is of the same type or an * unlock. */ if (lop->lo_first < new_lop->lo_first) new_lop->lo_first = lop->lo_first; if (lop->lo_end > new_lop->lo_end) new_lop->lo_end = lop->lo_end; tlop = lop; lop = LIST_NEXT(lop, lo_lckowner); nfsrv_freenfslock(tlop); continue; } /* * All these cases are for contiguous locks that are not the * same type, so they can't be merged. */ if (new_lop->lo_first <= lop->lo_first) { /* * This case is where the new lock overlaps with the * first part of the old lock. Move the start of the * old lock to just past the end of the new lock. The * new lock will be inserted in front of the old, since * ilop hasn't been updated. (We are done now.) */ lop->lo_first = new_lop->lo_end; break; } if (new_lop->lo_end >= lop->lo_end) { /* * This case is where the new lock overlaps with the * end of the old lock's range. Move the old lock's * end to just before the new lock's first and insert * the new lock after the old lock. * Might not be done yet, since the new lock could * overlap further locks with higher ranges. */ lop->lo_end = new_lop->lo_first; ilop = lop; lop = LIST_NEXT(lop, lo_lckowner); continue; } /* * The final case is where the new lock's range is in the * middle of the current lock's and splits the current lock * up. Use *other_lopp to handle the second part of the * split old lock range. (We are done now.) * For unlock, we use new_lop as other_lop and tmp, since * other_lop and new_lop are the same for this case. * We noted the unlock case above, so we don't need * new_lop->lo_flags any longer. */ tmp = new_lop->lo_first; if (other_lop == NULL) { if (!unlock) panic("nfsd srv update unlock"); other_lop = new_lop; *new_lopp = NULL; } other_lop->lo_first = new_lop->lo_end; other_lop->lo_end = lop->lo_end; other_lop->lo_flags = lop->lo_flags; other_lop->lo_stp = stp; other_lop->lo_lfp = lfp; lop->lo_end = tmp; nfsrv_insertlock(other_lop, lop, stp, lfp); *other_lopp = NULL; ilop = lop; break; } } ilop = lop; lop = LIST_NEXT(lop, lo_lckowner); if (myfile && (lop == NULL || lop->lo_lfp != lfp)) break; } /* * Insert the new lock in the list at the appropriate place. */ if (!unlock) { nfsrv_insertlock(new_lop, ilop, stp, lfp); *new_lopp = NULL; } } /* * This function handles sequencing of locks, etc. * It returns an error that indicates what the caller should do. */ static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid, struct nfsstate *stp, struct nfsrvcache *op) { int error = 0; if ((nd->nd_flag & ND_NFSV41) != 0) /* NFSv4.1 ignores the open_seqid and lock_seqid. */ goto out; if (op != nd->nd_rp) panic("nfsrvstate checkseqid"); if (!(op->rc_flag & RC_INPROG)) panic("nfsrvstate not inprog"); if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) { printf("refcnt=%d\n", stp->ls_op->rc_refcnt); panic("nfsrvstate op refcnt"); } /* If ND_ERELOOKUP is set, the seqid has already been handled. */ if ((nd->nd_flag & ND_ERELOOKUP) != 0) goto out; if ((stp->ls_seq + 1) == seqid) { if (stp->ls_op) nfsrvd_derefcache(stp->ls_op); stp->ls_op = op; nfsrvd_refcache(op); stp->ls_seq = seqid; goto out; } else if (stp->ls_seq == seqid && stp->ls_op && op->rc_xid == stp->ls_op->rc_xid && op->rc_refcnt == 0 && op->rc_reqlen == stp->ls_op->rc_reqlen && op->rc_cksum == stp->ls_op->rc_cksum) { if (stp->ls_op->rc_flag & RC_INPROG) { error = NFSERR_DONTREPLY; goto out; } nd->nd_rp = stp->ls_op; nd->nd_rp->rc_flag |= RC_INPROG; nfsrvd_delcache(op); error = NFSERR_REPLYFROMCACHE; goto out; } error = NFSERR_BADSEQID; out: NFSEXITCODE2(error, nd); return (error); } /* * Get the client ip address for callbacks. If the strings can't be parsed, * just set lc_program to 0 to indicate no callbacks are possible. * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set * the address to the client's transport address. This won't be used * for callbacks, but can be printed out by nfsstats for info.) * Return error if the xdr can't be parsed, 0 otherwise. */ int nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp) { u_int32_t *tl; u_char *cp, *cp2; int i, j, maxalen = 0, minalen = 0; sa_family_t af; #ifdef INET struct sockaddr_in *rin = NULL, *sin; #endif #ifdef INET6 struct sockaddr_in6 *rin6 = NULL, *sin6; #endif u_char *addr; int error = 0, cantparse = 0; union { in_addr_t ival; u_char cval[4]; } ip; union { in_port_t sval; u_char cval[2]; } port; /* 8 is the maximum length of the port# string. */ addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK); clp->lc_req.nr_client = NULL; clp->lc_req.nr_lock = 0; af = AF_UNSPEC; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); if (i >= 3 && i <= 4) { error = nfsrv_mtostr(nd, addr, i); if (error) goto nfsmout; #ifdef INET if (!strcmp(addr, "tcp")) { clp->lc_flags |= LCL_TCPCALLBACK; clp->lc_req.nr_sotype = SOCK_STREAM; clp->lc_req.nr_soproto = IPPROTO_TCP; af = AF_INET; } else if (!strcmp(addr, "udp")) { clp->lc_req.nr_sotype = SOCK_DGRAM; clp->lc_req.nr_soproto = IPPROTO_UDP; af = AF_INET; } #endif #ifdef INET6 if (af == AF_UNSPEC) { if (!strcmp(addr, "tcp6")) { clp->lc_flags |= LCL_TCPCALLBACK; clp->lc_req.nr_sotype = SOCK_STREAM; clp->lc_req.nr_soproto = IPPROTO_TCP; af = AF_INET6; } else if (!strcmp(addr, "udp6")) { clp->lc_req.nr_sotype = SOCK_DGRAM; clp->lc_req.nr_soproto = IPPROTO_UDP; af = AF_INET6; } } #endif if (af == AF_UNSPEC) { cantparse = 1; } } else { cantparse = 1; if (i > 0) { error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; } } /* * The caller has allocated clp->lc_req.nr_nam to be large enough * for either AF_INET or AF_INET6 and zeroed out the contents. * maxalen is set to the maximum length of the host IP address string * plus 8 for the maximum length of the port#. * minalen is set to the minimum length of the host IP address string * plus 4 for the minimum length of the port#. * These lengths do not include NULL termination, * so INET[6]_ADDRSTRLEN - 1 is used in the calculations. */ switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)clp->lc_req.nr_nam; rin->sin_family = AF_INET; rin->sin_len = sizeof(struct sockaddr_in); maxalen = INET_ADDRSTRLEN - 1 + 8; minalen = 7 + 4; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; rin6->sin6_family = AF_INET6; rin6->sin6_len = sizeof(struct sockaddr_in6); maxalen = INET6_ADDRSTRLEN - 1 + 8; minalen = 3 + 4; break; #endif } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); if (i < 0) { error = NFSERR_BADXDR; goto nfsmout; } else if (i == 0) { cantparse = 1; } else if (!cantparse && i <= maxalen && i >= minalen) { error = nfsrv_mtostr(nd, addr, i); if (error) goto nfsmout; /* * Parse out the address fields. We expect 6 decimal numbers * separated by '.'s for AF_INET and two decimal numbers * preceeded by '.'s for AF_INET6. */ cp = NULL; switch (af) { #ifdef INET6 /* * For AF_INET6, first parse the host address. */ case AF_INET6: cp = strchr(addr, '.'); if (cp != NULL) { *cp++ = '\0'; if (inet_pton(af, addr, &rin6->sin6_addr) == 1) i = 4; else { cp = NULL; cantparse = 1; } } break; #endif #ifdef INET case AF_INET: cp = addr; i = 0; break; #endif } while (cp != NULL && *cp && i < 6) { cp2 = cp; while (*cp2 && *cp2 != '.') cp2++; if (*cp2) *cp2++ = '\0'; else if (i != 5) { cantparse = 1; break; } j = nfsrv_getipnumber(cp); if (j >= 0) { if (i < 4) ip.cval[3 - i] = j; else port.cval[5 - i] = j; } else { cantparse = 1; break; } cp = cp2; i++; } if (!cantparse) { /* * The host address INADDR_ANY is (mis)used to indicate * "there is no valid callback address". */ switch (af) { #ifdef INET6 case AF_INET6: if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr, &in6addr_any)) rin6->sin6_port = htons(port.sval); else cantparse = 1; break; #endif #ifdef INET case AF_INET: if (ip.ival != INADDR_ANY) { rin->sin_addr.s_addr = htonl(ip.ival); rin->sin_port = htons(port.sval); } else { cantparse = 1; } break; #endif } } } else { cantparse = 1; if (i > 0) { error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; } } if (cantparse) { switch (nd->nd_nam->sa_family) { #ifdef INET case AF_INET: sin = (struct sockaddr_in *)nd->nd_nam; rin = (struct sockaddr_in *)clp->lc_req.nr_nam; rin->sin_family = AF_INET; rin->sin_len = sizeof(struct sockaddr_in); rin->sin_addr.s_addr = sin->sin_addr.s_addr; rin->sin_port = 0x0; break; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)nd->nd_nam; rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; rin6->sin6_family = AF_INET6; rin6->sin6_len = sizeof(struct sockaddr_in6); rin6->sin6_addr = sin6->sin6_addr; rin6->sin6_port = 0x0; break; #endif } clp->lc_program = 0; } nfsmout: free(addr, M_TEMP); NFSEXITCODE2(error, nd); return (error); } /* * Turn a string of up to three decimal digits into a number. Return -1 upon * error. */ static int nfsrv_getipnumber(u_char *cp) { int i = 0, j = 0; while (*cp) { if (j > 2 || *cp < '0' || *cp > '9') return (-1); i *= 10; i += (*cp - '0'); cp++; j++; } if (i < 256) return (i); return (-1); } /* * This function checks for restart conditions. */ static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags, nfsv4stateid_t *stateidp, int specialid) { int ret = 0; /* * First check for a server restart. Open, LockT, ReleaseLockOwner * and DelegPurge have a clientid, the rest a stateid. */ if (flags & (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) { if (clientid.lval[0] != nfsrvboottime) { ret = NFSERR_STALECLIENTID; goto out; } } else if (stateidp->other[0] != nfsrvboottime && specialid == 0) { ret = NFSERR_STALESTATEID; goto out; } /* * Read, Write, Setattr and LockT can return NFSERR_GRACE and do * not use a lock/open owner seqid#, so the check can be done now. * (The others will be checked, as required, later.) */ if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST))) goto out; NFSLOCKSTATE(); ret = nfsrv_checkgrace(NULL, NULL, flags); NFSUNLOCKSTATE(); out: NFSEXITCODE(ret); return (ret); } /* * Check for grace. */ static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp, u_int32_t flags) { int error = 0, notreclaimed; struct nfsrv_stable *sp; if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE | NFSNSF_GRACEOVER)) == 0) { /* * First, check to see if all of the clients have done a * ReclaimComplete. If so, grace can end now. */ notreclaimed = 0; LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) { notreclaimed = 1; break; } } if (notreclaimed == 0) nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK); } if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) { if (flags & NFSLCK_RECLAIM) { error = NFSERR_NOGRACE; goto out; } } else { if (!(flags & NFSLCK_RECLAIM)) { error = NFSERR_GRACE; goto out; } if (nd != NULL && clp != NULL && (nd->nd_flag & ND_NFSV41) != 0 && (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) { error = NFSERR_NOGRACE; goto out; } /* * If grace is almost over and we are still getting Reclaims, * extend grace a bit. */ if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) > nfsrv_stablefirst.nsf_eograce) nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA; } out: NFSEXITCODE(error); return (error); } /* * Do a server callback. * The "trunc" argument is slightly overloaded and refers to different * boolean arguments for CBRECALL and CBLAYOUTRECALL. */ static int nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p) { struct mbuf *m; u_int32_t *tl; struct nfsrv_descript *nd; struct ucred *cred; int error = 0, slotpos; u_int32_t callback; struct nfsdsession *sep = NULL; uint64_t tval; bool dotls; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); cred = newnfs_getcred(); NFSLOCKSTATE(); /* mostly for lc_cbref++ */ if (clp->lc_flags & LCL_NEEDSCONFIRM) { NFSUNLOCKSTATE(); panic("docallb"); } clp->lc_cbref++; /* * Fill the callback program# and version into the request * structure for newnfs_connect() to use. */ clp->lc_req.nr_prog = clp->lc_program; #ifdef notnow if ((clp->lc_flags & LCL_NFSV41) != 0) clp->lc_req.nr_vers = NFSV41_CBVERS; else #endif clp->lc_req.nr_vers = NFSV4_CBVERS; /* * First, fill in some of the fields of nd and cr. */ nd->nd_flag = ND_NFSV4; if (clp->lc_flags & LCL_GSS) nd->nd_flag |= ND_KERBV; if ((clp->lc_flags & LCL_NFSV41) != 0) nd->nd_flag |= ND_NFSV41; if ((clp->lc_flags & LCL_NFSV42) != 0) nd->nd_flag |= ND_NFSV42; nd->nd_repstat = 0; cred->cr_uid = clp->lc_uid; cred->cr_gid = clp->lc_gid; callback = clp->lc_callback; NFSUNLOCKSTATE(); cred->cr_ngroups = 1; /* * Get the first mbuf for the request. */ MGET(m, M_WAITOK, MT_DATA); m->m_len = 0; nd->nd_mreq = nd->nd_mb = m; nd->nd_bpos = mtod(m, caddr_t); /* * and build the callback request. */ if (procnum == NFSV4OP_CBGETATTR) { nd->nd_procnum = NFSV4PROC_CBCOMPOUND; error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR, "CB Getattr", &sep, &slotpos); if (error != 0) { m_freem(nd->nd_mreq); goto errout; } (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0); (void)nfsrv_putattrbit(nd, attrbitp); } else if (procnum == NFSV4OP_CBRECALL) { nd->nd_procnum = NFSV4PROC_CBCOMPOUND; error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL, "CB Recall", &sep, &slotpos); if (error != 0) { m_freem(nd->nd_mreq); goto errout; } NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID); *tl++ = txdr_unsigned(stateidp->seqid); NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl, NFSX_STATEIDOTHER); tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED); if (trunc) *tl = newnfs_true; else *tl = newnfs_false; (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0); } else if (procnum == NFSV4OP_CBLAYOUTRECALL) { NFSD_DEBUG(4, "docallback layout recall\n"); nd->nd_procnum = NFSV4PROC_CBCOMPOUND; error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos); NFSD_DEBUG(4, "aft cbcallargs=%d\n", error); if (error != 0) { m_freem(nd->nd_mreq); goto errout; } NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(laytype); *tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY); if (trunc) *tl++ = newnfs_true; else *tl++ = newnfs_false; *tl = txdr_unsigned(NFSV4LAYOUTRET_FILE); nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID); tval = 0; txdr_hyper(tval, tl); tl += 2; tval = UINT64_MAX; txdr_hyper(tval, tl); tl += 2; *tl++ = txdr_unsigned(stateidp->seqid); NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER); tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED); NFSD_DEBUG(4, "aft args\n"); } else if (procnum == NFSV4PROC_CBNULL) { nd->nd_procnum = NFSV4PROC_CBNULL; if ((clp->lc_flags & LCL_NFSV41) != 0) { error = nfsv4_getcbsession(clp, &sep); if (error != 0) { m_freem(nd->nd_mreq); goto errout; } } } else { error = NFSERR_SERVERFAULT; m_freem(nd->nd_mreq); goto errout; } /* * Call newnfs_connect(), as required, and then newnfs_request(). */ dotls = false; if ((clp->lc_flags & LCL_TLSCB) != 0) dotls = true; (void) newnfs_sndlock(&clp->lc_req.nr_lock); if (clp->lc_req.nr_client == NULL) { if ((clp->lc_flags & LCL_NFSV41) != 0) { error = ECONNREFUSED; if (procnum != NFSV4PROC_CBNULL) nfsv4_freeslot(&sep->sess_cbsess, slotpos, true); nfsrv_freesession(sep, NULL); } else if (nd->nd_procnum == NFSV4PROC_CBNULL) error = newnfs_connect(NULL, &clp->lc_req, cred, NULL, 1, dotls, &clp->lc_req.nr_client); else error = newnfs_connect(NULL, &clp->lc_req, cred, NULL, 3, dotls, &clp->lc_req.nr_client); } newnfs_sndunlock(&clp->lc_req.nr_lock); NFSD_DEBUG(4, "aft sndunlock=%d\n", error); if (!error) { if ((nd->nd_flag & ND_NFSV41) != 0) { KASSERT(sep != NULL, ("sep NULL")); if (sep->sess_cbsess.nfsess_xprt != NULL) error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL, NULL, cred, clp->lc_program, clp->lc_req.nr_vers, NULL, 1, NULL, &sep->sess_cbsess); else { /* * This should probably never occur, but if a * client somehow does an RPC without a * SequenceID Op that causes a callback just * after the nfsd threads have been terminated * and restared we could conceivably get here * without a backchannel xprt. */ printf("nfsrv_docallback: no xprt\n"); error = ECONNREFUSED; } NFSD_DEBUG(4, "aft newnfs_request=%d\n", error); if (error != 0 && procnum != NFSV4PROC_CBNULL) { /* * It is likely that the callback was never * processed by the client and, as such, * the sequence# for the session slot needs * to be backed up by one to avoid a * NFSERR_SEQMISORDERED error reply. * For the unlikely case where the callback * was processed by the client, this will * make the next callback on the slot * appear to be a retry. * Since callbacks never specify that the * reply be cached, this "apparent retry" * should not be a problem. */ nfsv4_freeslot(&sep->sess_cbsess, slotpos, true); } nfsrv_freesession(sep, NULL); } else error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL, NULL, cred, clp->lc_program, clp->lc_req.nr_vers, NULL, 1, NULL, NULL); } errout: NFSFREECRED(cred); /* * If error is set here, the Callback path isn't working * properly, so twiddle the appropriate LCL_ flags. * (nd_repstat != 0 indicates the Callback path is working, * but the callback failed on the client.) */ if (error) { /* * Mark the callback pathway down, which disabled issuing * of delegations and gets Renew to return NFSERR_CBPATHDOWN. */ NFSLOCKSTATE(); clp->lc_flags |= LCL_CBDOWN; NFSUNLOCKSTATE(); } else { /* * Callback worked. If the callback path was down, disable * callbacks, so no more delegations will be issued. (This * is done on the assumption that the callback pathway is * flakey.) */ NFSLOCKSTATE(); if (clp->lc_flags & LCL_CBDOWN) clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON); NFSUNLOCKSTATE(); if (nd->nd_repstat) { error = nd->nd_repstat; NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n", procnum, error); } else if (error == 0 && procnum == NFSV4OP_CBGETATTR) error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, p, NULL); m_freem(nd->nd_mrep); } NFSLOCKSTATE(); clp->lc_cbref--; if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) { clp->lc_flags &= ~LCL_WAKEUPWANTED; wakeup(clp); } NFSUNLOCKSTATE(); free(nd, M_TEMP); NFSEXITCODE(error); return (error); } /* * Set up the compound RPC for the callback. */ static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp, uint32_t callback, int op, const char *optag, struct nfsdsession **sepp, int *slotposp) { uint32_t *tl; int error, len; len = strlen(optag); (void)nfsm_strtom(nd, optag, len); NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED); if ((nd->nd_flag & ND_NFSV41) != 0) { if ((nd->nd_flag & ND_NFSV42) != 0) *tl++ = txdr_unsigned(NFSV42_MINORVERSION); else *tl++ = txdr_unsigned(NFSV41_MINORVERSION); *tl++ = txdr_unsigned(callback); *tl++ = txdr_unsigned(2); *tl = txdr_unsigned(NFSV4OP_CBSEQUENCE); error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp); if (error != 0) return (error); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(op); } else { *tl++ = txdr_unsigned(NFSV4_MINORVERSION); *tl++ = txdr_unsigned(callback); *tl++ = txdr_unsigned(1); *tl = txdr_unsigned(op); } return (0); } /* * Return the next index# for a clientid. Mostly just increment and return * the next one, but... if the 32bit unsigned does actually wrap around, * it should be rebooted. * At an average rate of one new client per second, it will wrap around in * approximately 136 years. (I think the server will have been shut * down or rebooted before then.) */ static u_int32_t nfsrv_nextclientindex(void) { static u_int32_t client_index = 0; client_index++; if (client_index != 0) return (client_index); printf("%s: out of clientids\n", __func__); return (client_index); } /* * Return the next index# for a stateid. Mostly just increment and return * the next one, but... if the 32bit unsigned does actually wrap around * (will a BSD server stay up that long?), find * new start and end values. */ static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp) { struct nfsstate *stp; int i; u_int32_t canuse, min_index, max_index; if (!(clp->lc_flags & LCL_INDEXNOTOK)) { clp->lc_stateindex++; if (clp->lc_stateindex != clp->lc_statemaxindex) return (clp->lc_stateindex); } /* * Yuck, we've hit the end. * Look for a new min and max. */ min_index = 0; max_index = 0xffffffff; for (i = 0; i < nfsrv_statehashsize; i++) { LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) { if (stp->ls_stateid.other[2] > 0x80000000) { if (stp->ls_stateid.other[2] < max_index) max_index = stp->ls_stateid.other[2]; } else { if (stp->ls_stateid.other[2] > min_index) min_index = stp->ls_stateid.other[2]; } } } /* * Yikes, highly unlikely, but I'll handle it anyhow. */ if (min_index == 0x80000000 && max_index == 0x80000001) { canuse = 0; /* * Loop around until we find an unused entry. Return that * and set LCL_INDEXNOTOK, so the search will continue next time. * (This is one of those rare cases where a goto is the * cleanest way to code the loop.) */ tryagain: for (i = 0; i < nfsrv_statehashsize; i++) { LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) { if (stp->ls_stateid.other[2] == canuse) { canuse++; goto tryagain; } } } clp->lc_flags |= LCL_INDEXNOTOK; return (canuse); } /* * Ok to start again from min + 1. */ clp->lc_stateindex = min_index + 1; clp->lc_statemaxindex = max_index; clp->lc_flags &= ~LCL_INDEXNOTOK; return (clp->lc_stateindex); } /* * The following functions handle the stable storage file that deals with * the edge conditions described in RFC3530 Sec. 8.6.3. * The file is as follows: * - a single record at the beginning that has the lease time of the * previous server instance (before the last reboot) and the nfsrvboottime * values for the previous server boots. * These previous boot times are used to ensure that the current * nfsrvboottime does not, somehow, get set to a previous one. * (This is important so that Stale ClientIDs and StateIDs can * be recognized.) * The number of previous nfsvrboottime values precedes the list. * - followed by some number of appended records with: * - client id string * - flag that indicates it is a record revoking state via lease * expiration or similar * OR has successfully acquired state. * These structures vary in length, with the client string at the end, up * to NFSV4_OPAQUELIMIT in size. * * At the end of the grace period, the file is truncated, the first * record is rewritten with updated information and any acquired state * records for successful reclaims of state are written. * * Subsequent records are appended when the first state is issued to * a client and when state is revoked for a client. * * When reading the file in, state issued records that come later in * the file override older ones, since the append log is in cronological order. * If, for some reason, the file can't be read, the grace period is * immediately terminated and all reclaims get NFSERR_NOGRACE. */ /* * Read in the stable storage file. Called by nfssvc() before the nfsd * processes start servicing requests. */ void nfsrv_setupstable(NFSPROC_T *p) { struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; struct nfsrv_stable *sp, *nsp; struct nfst_rec *tsp; int error, i, tryagain; off_t off = 0; ssize_t aresid, len; /* * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without * a reboot, so state has not been lost. */ if (sf->nsf_flags & NFSNSF_UPDATEDONE) return; /* * Set Grace over just until the file reads successfully. */ nfsrvboottime = time_second; LIST_INIT(&sf->nsf_head); sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK); sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA; if (sf->nsf_fp == NULL) return; error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp), (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p); if (error || aresid || sf->nsf_numboots == 0 || sf->nsf_numboots > NFSNSF_MAXNUMBOOTS) return; /* * Now, read in the boottimes. */ sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) * sizeof (time_t), M_TEMP, M_WAITOK); off = sizeof (struct nfsf_rec); error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp), (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p); if (error || aresid) { free(sf->nsf_bootvals, M_TEMP); sf->nsf_bootvals = NULL; return; } /* * Make sure this nfsrvboottime is different from all recorded * previous ones. */ do { tryagain = 0; for (i = 0; i < sf->nsf_numboots; i++) { if (nfsrvboottime == sf->nsf_bootvals[i]) { nfsrvboottime++; tryagain = 1; break; } } } while (tryagain); sf->nsf_flags |= NFSNSF_OK; off += (sf->nsf_numboots * sizeof (time_t)); /* * Read through the file, building a list of records for grace * checking. * Each record is between sizeof (struct nfst_rec) and * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1 * and is actually sizeof (struct nfst_rec) + nst_len - 1. */ tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK); do { error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp), (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1, off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p); len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid; if (error || (len > 0 && (len < sizeof (struct nfst_rec) || len < (sizeof (struct nfst_rec) + tsp->len - 1)))) { /* * Yuck, the file has been corrupted, so just return * after clearing out any restart state, so the grace period * is over. */ LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) { LIST_REMOVE(sp, nst_list); free(sp, M_TEMP); } free(tsp, M_TEMP); sf->nsf_flags &= ~NFSNSF_OK; free(sf->nsf_bootvals, M_TEMP); sf->nsf_bootvals = NULL; return; } if (len > 0) { off += sizeof (struct nfst_rec) + tsp->len - 1; /* * Search the list for a matching client. */ LIST_FOREACH(sp, &sf->nsf_head, nst_list) { if (tsp->len == sp->nst_len && !NFSBCMP(tsp->client, sp->nst_client, tsp->len)) break; } if (sp == LIST_END(&sf->nsf_head)) { sp = (struct nfsrv_stable *)malloc(tsp->len + sizeof (struct nfsrv_stable) - 1, M_TEMP, M_WAITOK); NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec, sizeof (struct nfst_rec) + tsp->len - 1); LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list); } else { if (tsp->flag == NFSNST_REVOKE) sp->nst_flag |= NFSNST_REVOKE; else /* * A subsequent timestamp indicates the client * did a setclientid/confirm and any previous * revoke is no longer relevant. */ sp->nst_flag &= ~NFSNST_REVOKE; } } } while (len > 0); free(tsp, M_TEMP); sf->nsf_flags = NFSNSF_OK; sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease + NFSRV_LEASEDELTA; } /* * Update the stable storage file, now that the grace period is over. */ void nfsrv_updatestable(NFSPROC_T *p) { struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; struct nfsrv_stable *sp, *nsp; int i; struct nfsvattr nva; vnode_t vp; #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000) mount_t mp = NULL; #endif int error; if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE)) return; sf->nsf_flags |= NFSNSF_UPDATEDONE; /* * Ok, we need to rewrite the stable storage file. * - truncate to 0 length * - write the new first structure * - loop through the data structures, writing out any that * have timestamps older than the old boot */ if (sf->nsf_bootvals) { sf->nsf_numboots++; for (i = sf->nsf_numboots - 2; i >= 0; i--) sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i]; } else { sf->nsf_numboots = 1; sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t), M_TEMP, M_WAITOK); } sf->nsf_bootvals[0] = nfsrvboottime; sf->nsf_lease = nfsrv_lease; NFSVNO_ATTRINIT(&nva); NFSVNO_SETATTRVAL(&nva, size, 0); vp = NFSFPVNODE(sf->nsf_fp); vn_start_write(vp, &mp, V_WAIT); if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) { error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p, NULL); NFSVOPUNLOCK(vp); } else error = EPERM; vn_finished_write(mp); if (!error) error = NFSD_RDWR(UIO_WRITE, vp, (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0, UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p); if (!error) error = NFSD_RDWR(UIO_WRITE, vp, (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), (off_t)(sizeof (struct nfsf_rec)), UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p); free(sf->nsf_bootvals, M_TEMP); sf->nsf_bootvals = NULL; if (error) { sf->nsf_flags &= ~NFSNSF_OK; printf("EEK! Can't write NfsV4 stable storage file\n"); return; } sf->nsf_flags |= NFSNSF_OK; /* * Loop through the list and write out timestamp records for * any clients that successfully reclaimed state. */ LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) { if (sp->nst_flag & NFSNST_GOTSTATE) { nfsrv_writestable(sp->nst_client, sp->nst_len, NFSNST_NEWSTATE, p); sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE; } LIST_REMOVE(sp, nst_list); free(sp, M_TEMP); } nfsrv_backupstable(); } /* * Append a record to the stable storage file. */ void nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p) { struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; struct nfst_rec *sp; int error; if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL) return; sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) + len - 1, M_TEMP, M_WAITOK); sp->len = len; NFSBCOPY(client, sp->client, len); sp->flag = flag; error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp), (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0, UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p); free(sp, M_TEMP); if (error) { sf->nsf_flags &= ~NFSNSF_OK; printf("EEK! Can't write NfsV4 stable storage file\n"); } } /* * This function is called during the grace period to mark a client * that successfully reclaimed state. */ static void nfsrv_markstable(struct nfsclient *clp) { struct nfsrv_stable *sp; /* * First find the client structure. */ LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { if (sp->nst_len == clp->lc_idlen && !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len)) break; } if (sp == LIST_END(&nfsrv_stablefirst.nsf_head)) return; /* * Now, just mark it and set the nfsclient back pointer. */ sp->nst_flag |= NFSNST_GOTSTATE; sp->nst_clp = clp; } /* * This function is called when a NFSv4.1 client does a ReclaimComplete. * Very similar to nfsrv_markstable(), except for the flag being set. */ static void nfsrv_markreclaim(struct nfsclient *clp) { struct nfsrv_stable *sp; /* * First find the client structure. */ LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { if (sp->nst_len == clp->lc_idlen && !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len)) break; } if (sp == LIST_END(&nfsrv_stablefirst.nsf_head)) return; /* * Now, just set the flag. */ sp->nst_flag |= NFSNST_RECLAIMED; } /* * This function is called for a reclaim, to see if it gets grace. * It returns 0 if a reclaim is allowed, 1 otherwise. */ static int nfsrv_checkstable(struct nfsclient *clp) { struct nfsrv_stable *sp; /* * First, find the entry for the client. */ LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { if (sp->nst_len == clp->lc_idlen && !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len)) break; } /* * If not in the list, state was revoked or no state was issued * since the previous reboot, a reclaim is denied. */ if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) || (sp->nst_flag & NFSNST_REVOKE) || !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK)) return (1); return (0); } /* * Test for and try to clear out a conflicting client. This is called by * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients * a found. * The trick here is that it can't revoke a conflicting client with an * expired lease unless it holds the v4root lock, so... * If no v4root lock, get the lock and return 1 to indicate "try again". * Return 0 to indicate the conflict can't be revoked and 1 to indicate * the revocation worked and the conflicting client is "bye, bye", so it * can be tried again. * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK(). * Unlocks State before a non-zero value is returned. */ static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp, NFSPROC_T *p) { int gotlock, lktype = 0; /* * If lease hasn't expired, we can't fix it. */ if (clp->lc_expiry >= NFSD_MONOSEC || !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) return (0); if (*haslockp == 0) { NFSUNLOCKSTATE(); if (vp != NULL) { lktype = NFSVOPISLOCKED(vp); NFSVOPUNLOCK(vp); } NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!gotlock); NFSUNLOCKV4ROOTMUTEX(); *haslockp = 1; if (vp != NULL) { NFSVOPLOCK(vp, lktype | LK_RETRY); if (VN_IS_DOOMED(vp)) return (2); } return (1); } NFSUNLOCKSTATE(); /* * Ok, we can expire the conflicting client. */ nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); nfsrv_backupstable(); nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); LIST_REMOVE(clp, lc_hash); nfsrv_zapclient(clp, p); return (1); } /* * Resolve a delegation conflict. * Returns 0 to indicate the conflict was resolved without sleeping. * Return -1 to indicate that the caller should check for conflicts again. * Return > 0 for an error that should be returned, normally NFSERR_DELAY. * * Also, manipulate the nfsv4root_lock, as required. It isn't changed * for a return of 0, since there was no sleep and it could be required * later. It is released for a return of NFSERR_DELAY, since the caller * will return that error. It is released when a sleep was done waiting * for the delegation to be returned or expire (so that other nfsds can * handle ops). Then, it must be acquired for the write to stable storage. * (This function is somewhat similar to nfsrv_clientconflict(), but * the semantics differ in a couple of subtle ways. The return of 0 * indicates the conflict was resolved without sleeping here, not * that the conflict can't be resolved and the handling of nfsv4root_lock * differs, as noted above.) * Unlocks State before returning a non-zero value. */ static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p, vnode_t vp) { struct nfsclient *clp = stp->ls_clp; int gotlock, error, lktype = 0, retrycnt, zapped_clp; nfsv4stateid_t tstateid; fhandle_t tfh; /* * If the conflict is with an old delegation... */ if (stp->ls_flags & NFSLCK_OLDDELEG) { /* * You can delete it, if it has expired. */ if (clp->lc_delegtime < NFSD_MONOSEC) { nfsrv_freedeleg(stp); NFSUNLOCKSTATE(); error = -1; goto out; } NFSUNLOCKSTATE(); /* * During this delay, the old delegation could expire or it * could be recovered by the client via an Open with * CLAIM_DELEGATE_PREV. * Release the nfsv4root_lock, if held. */ if (*haslockp) { *haslockp = 0; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_DELAY; goto out; } /* * It's a current delegation, so: * - check to see if the delegation has expired * - if so, get the v4root lock and then expire it */ if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall < NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC && stp->ls_delegtime >= NFSD_MONOSEC)) { /* * - do a recall callback, since not yet done * For now, never allow truncate to be set. To use * truncate safely, it must be guaranteed that the * Remove, Rename or Setattr with size of 0 will * succeed and that would require major changes to * the VFS/Vnode OPs. * Set the expiry time large enough so that it won't expire * until after the callback, then set it correctly, once * the callback is done. (The delegation will now time * out whether or not the Recall worked ok. The timeout * will be extended when ops are done on the delegation * stateid, up to the timelimit.) */ if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) { stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) + NFSRV_LEASEDELTA; stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) + NFSRV_LEASEDELTA; stp->ls_flags |= NFSLCK_DELEGRECALL; } stp->ls_lastrecall = time_uptime + 1; /* * Loop NFSRV_CBRETRYCNT times while the CBRecall replies * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done * in order to try and avoid a race that could happen * when a CBRecall request passed the Open reply with * the delegation in it when transitting the network. * Since nfsrv_docallback will sleep, don't use stp after * the call. */ NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid, sizeof (tstateid)); NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh, sizeof (tfh)); NFSUNLOCKSTATE(); if (*haslockp) { *haslockp = 0; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } retrycnt = 0; do { error = nfsrv_docallback(clp, NFSV4OP_CBRECALL, &tstateid, 0, &tfh, NULL, NULL, 0, p); retrycnt++; } while ((error == NFSERR_BADSTATEID || error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT); error = NFSERR_DELAY; goto out; } if (clp->lc_expiry >= NFSD_MONOSEC && stp->ls_delegtime >= NFSD_MONOSEC) { NFSUNLOCKSTATE(); /* * A recall has been done, but it has not yet expired. * So, RETURN_DELAY. */ if (*haslockp) { *haslockp = 0; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_DELAY; goto out; } /* * If we don't yet have the lock, just get it and then return, * since we need that before deleting expired state, such as * this delegation. * When getting the lock, unlock the vnode, so other nfsds that * are in progress, won't get stuck waiting for the vnode lock. */ if (*haslockp == 0) { NFSUNLOCKSTATE(); if (vp != NULL) { lktype = NFSVOPISLOCKED(vp); NFSVOPUNLOCK(vp); } NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!gotlock); NFSUNLOCKV4ROOTMUTEX(); *haslockp = 1; if (vp != NULL) { NFSVOPLOCK(vp, lktype | LK_RETRY); if (VN_IS_DOOMED(vp)) { *haslockp = 0; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); error = NFSERR_PERM; goto out; } } error = -1; goto out; } NFSUNLOCKSTATE(); /* * Ok, we can delete the expired delegation. * First, write the Revoke record to stable storage and then * clear out the conflict. * Since all other nfsd threads are now blocked, we can safely * sleep without the state changing. */ nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); nfsrv_backupstable(); if (clp->lc_expiry < NFSD_MONOSEC) { nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); LIST_REMOVE(clp, lc_hash); zapped_clp = 1; } else { nfsrv_freedeleg(stp); zapped_clp = 0; } if (zapped_clp) nfsrv_zapclient(clp, p); error = -1; out: NFSEXITCODE(error); return (error); } /* * Check for a remove allowed, if remove is set to 1 and get rid of * delegations. */ int nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd, nfsquad_t clientid, NFSPROC_T *p) { struct nfsclient *clp; struct nfsstate *stp; struct nfslockfile *lfp; int error, haslock = 0; fhandle_t nfh; clp = NULL; /* * First, get the lock file structure. * (A return of -1 means no associated state, so remove ok.) */ error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p); tryagain: NFSLOCKSTATE(); if (error == 0 && clientid.qval != 0) error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (!error) error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0); if (error) { NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } if (error == -1) error = 0; goto out; } /* * Now, we must Recall any delegations. */ error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p); if (error) { /* * nfsrv_cleandeleg() unlocks state for non-zero * return. */ if (error == -1) goto tryagain; if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } goto out; } /* * Now, look for a conflicting open share. */ if (remove) { /* * If the entry in the directory was the last reference to the * corresponding filesystem object, the object can be destroyed * */ if(lfp->lf_usecount>1) LIST_FOREACH(stp, &lfp->lf_open, ls_file) { if (stp->ls_flags & NFSLCK_WRITEDENY) { error = NFSERR_FILEOPEN; break; } } } NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } out: NFSEXITCODE(error); return (error); } /* * Clear out all delegations for the file referred to by lfp. * May return NFSERR_DELAY, if there will be a delay waiting for * delegations to expire. * Returns -1 to indicate it slept while recalling a delegation. * This function has the side effect of deleting the nfslockfile structure, * if it no longer has associated state and didn't have to sleep. * Unlocks State before a non-zero value is returned. */ static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp, struct nfsclient *clp, int *haslockp, NFSPROC_T *p) { struct nfsstate *stp, *nstp; int ret = 0; stp = LIST_FIRST(&lfp->lf_deleg); while (stp != LIST_END(&lfp->lf_deleg)) { nstp = LIST_NEXT(stp, ls_file); if (stp->ls_clp != clp) { ret = nfsrv_delegconflict(stp, haslockp, p, vp); if (ret) { /* * nfsrv_delegconflict() unlocks state * when it returns non-zero. */ goto out; } } stp = nstp; } out: NFSEXITCODE(ret); return (ret); } /* * There are certain operations that, when being done outside of NFSv4, * require that any NFSv4 delegation for the file be recalled. * This function is to be called for those cases: * VOP_RENAME() - When a delegation is being recalled for any reason, * the client may have to do Opens against the server, using the file's * final component name. If the file has been renamed on the server, * that component name will be incorrect and the Open will fail. * VOP_REMOVE() - Theoretically, a client could Open a file after it has * been removed on the server, if there is a delegation issued to * that client for the file. I say "theoretically" since clients * normally do an Access Op before the Open and that Access Op will * fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so * they will detect the file's removal in the same manner. (There is * one case where RFC3530 allows a client to do an Open without first * doing an Access Op, which is passage of a check against the ACE * returned with a Write delegation, but current practice is to ignore * the ACE and always do an Access Op.) * Since the functions can only be called with an unlocked vnode, this * can't be done at this time. * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range * locks locally in the client, which are not visible to the server. To * deal with this, issuing of delegations for a vnode must be disabled * and all delegations for the vnode recalled. This is done via the * second function, using the VV_DISABLEDELEG vflag on the vnode. */ void nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p) { time_t starttime; int error; /* * First, check to see if the server is currently running and it has * been called for a regular file when issuing delegations. */ if (newnfs_numnfsd == 0 || vp->v_type != VREG || nfsrv_issuedelegs == 0) return; KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp)); /* * First, get a reference on the nfsv4rootfs_lock so that an * exclusive lock cannot be acquired by another thread. */ NFSLOCKV4ROOTMUTEX(); nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); NFSUNLOCKV4ROOTMUTEX(); /* * Now, call nfsrv_checkremove() in a loop while it returns * NFSERR_DELAY. Return upon any other error or when timed out. */ starttime = NFSD_MONOSEC; do { if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) { error = nfsrv_checkremove(vp, 0, NULL, (nfsquad_t)((u_quad_t)0), p); NFSVOPUNLOCK(vp); } else error = EPERM; if (error == NFSERR_DELAY) { if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO) break; /* Sleep for a short period of time */ (void) nfs_catnap(PZERO, 0, "nfsremove"); } } while (error == NFSERR_DELAY); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); } void nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p) { #ifdef VV_DISABLEDELEG /* * First, flag issuance of delegations disabled. */ atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG); #endif /* * Then call nfsd_recalldelegation() to get rid of all extant * delegations. */ nfsd_recalldelegation(vp, p); } /* * Check for conflicting locks, etc. and then get rid of delegations. * (At one point I thought that I should get rid of delegations for any * Setattr, since it could potentially disallow the I/O op (read or write) * allowed by the delegation. However, Setattr Ops that aren't changing * the size get a stateid of all 0s, so you can't tell if it is a delegation * for the same client or a different one, so I decided to only get rid * of delegations for other clients when the size is being changed.) * In general, a Setattr can disable NFS I/O Ops that are outstanding, such * as Write backs, even if there is no delegation, so it really isn't any * different?) */ int nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct nfsexstuff *exp, NFSPROC_T *p) { struct nfsstate st, *stp = &st; struct nfslock lo, *lop = &lo; int error = 0; nfsquad_t clientid; if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) { stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS); lop->lo_first = nvap->na_size; } else { stp->ls_flags = 0; lop->lo_first = 0; } if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL)) stp->ls_flags |= NFSLCK_SETATTR; if (stp->ls_flags == 0) goto out; lop->lo_end = NFS64BITSSET; lop->lo_flags = NFSLCK_WRITE; stp->ls_ownerlen = 0; stp->ls_op = NULL; stp->ls_uid = nd->nd_cred->cr_uid; stp->ls_stateid.seqid = stateidp->seqid; clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0]; clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1]; stp->ls_stateid.other[2] = stateidp->other[2]; error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid, stateidp, exp, nd, p); out: NFSEXITCODE2(error, nd); return (error); } /* * Check for a write delegation and do a CBGETATTR if there is one, updating * the attributes, as required. * Should I return an error if I can't get the attributes? (For now, I'll * just return ok. */ int nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p) { struct nfsstate *stp; struct nfslockfile *lfp; struct nfsclient *clp; struct nfsvattr nva; fhandle_t nfh; int error = 0; nfsattrbit_t cbbits; u_quad_t delegfilerev; NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits); if (!NFSNONZERO_ATTRBIT(&cbbits)) goto out; if (nfsrv_writedelegcnt == 0) goto out; /* * Get the lock file structure. * (A return of -1 means no associated state, so return ok.) */ error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p); NFSLOCKSTATE(); if (!error) error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0); if (error) { NFSUNLOCKSTATE(); if (error == -1) error = 0; goto out; } /* * Now, look for a write delegation. */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if (stp->ls_flags & NFSLCK_DELEGWRITE) break; } if (stp == LIST_END(&lfp->lf_deleg)) { NFSUNLOCKSTATE(); goto out; } clp = stp->ls_clp; /* If the clientid is not confirmed, ignore the delegation. */ if (clp->lc_flags & LCL_NEEDSCONFIRM) { NFSUNLOCKSTATE(); goto out; } delegfilerev = stp->ls_filerev; /* * If the Write delegation was issued as a part of this Compound RPC * or if we have an Implied Clientid (used in a previous Op in this * compound) and it is the client the delegation was issued to, * just return ok. * I also assume that it is from the same client iff the network * host IP address is the same as the callback address. (Not * exactly correct by the RFC, but avoids a lot of Getattr * callbacks.) */ if (nd->nd_compref == stp->ls_compref || ((nd->nd_flag & ND_IMPLIEDCLID) && clp->lc_clientid.qval == nd->nd_clientid.qval) || nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) { NFSUNLOCKSTATE(); goto out; } /* * We are now done with the delegation state structure, * so the statelock can be released and we can now tsleep(). */ /* * Now, we must do the CB Getattr callback, to see if Change or Size * has changed. */ if (clp->lc_expiry >= NFSD_MONOSEC) { NFSUNLOCKSTATE(); NFSVNO_ATTRINIT(&nva); nva.na_filerev = NFS64BITSSET; error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL, 0, &nfh, &nva, &cbbits, 0, p); if (!error) { if ((nva.na_filerev != NFS64BITSSET && nva.na_filerev > delegfilerev) || (NFSVNO_ISSETSIZE(&nva) && nva.na_size != nvap->na_size)) { error = nfsvno_updfilerev(vp, nvap, nd, p); if (NFSVNO_ISSETSIZE(&nva)) nvap->na_size = nva.na_size; } } else error = 0; /* Ignore callback errors for now. */ } else { NFSUNLOCKSTATE(); } out: NFSEXITCODE2(error, nd); return (error); } /* * This function looks for openowners that haven't had any opens for * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS * is set. */ void nfsrv_throwawayopens(NFSPROC_T *p) { struct nfsclient *clp, *nclp; struct nfsstate *stp, *nstp; int i; NFSLOCKSTATE(); nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS; /* * For each client... */ for (i = 0; i < nfsrv_clienthashsize; i++) { LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) { LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) { if (LIST_EMPTY(&stp->ls_open) && (stp->ls_noopens > NFSNOOPEN || (nfsrv_openpluslock * 2) > nfsrv_v4statelimit)) nfsrv_freeopenowner(stp, 0, p); } } } NFSUNLOCKSTATE(); } /* * This function checks to see if the credentials are the same. * Returns 1 for not same, 0 otherwise. */ static int nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp) { if (nd->nd_flag & ND_GSS) { if (!(clp->lc_flags & LCL_GSS)) return (1); if (clp->lc_flags & LCL_NAME) { if (nd->nd_princlen != clp->lc_namelen || NFSBCMP(nd->nd_principal, clp->lc_name, clp->lc_namelen)) return (1); else return (0); } if (nd->nd_cred->cr_uid == clp->lc_uid) return (0); else return (1); } else if (clp->lc_flags & LCL_GSS) return (1); /* * For AUTH_SYS, allow the same uid or root. (This is underspecified * in RFC3530, which talks about principals, but doesn't say anything * about uids for AUTH_SYS.) */ if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0) return (0); else return (1); } /* * Calculate the lease expiry time. */ static time_t nfsrv_leaseexpiry(void) { if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC) return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA)); return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA); } /* * Delay the delegation timeout as far as ls_delegtimelimit, as required. */ static void nfsrv_delaydelegtimeout(struct nfsstate *stp) { if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) return; if ((stp->ls_delegtime + 15) > NFSD_MONOSEC && stp->ls_delegtime < stp->ls_delegtimelimit) { stp->ls_delegtime += nfsrv_lease; if (stp->ls_delegtime > stp->ls_delegtimelimit) stp->ls_delegtime = stp->ls_delegtimelimit; } } /* * This function checks to see if there is any other state associated * with the openowner for this Open. * It returns 1 if there is no other state, 0 otherwise. */ static int nfsrv_nootherstate(struct nfsstate *stp) { struct nfsstate *tstp; LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) { if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock)) return (0); } return (1); } /* * Create a list of lock deltas (changes to local byte range locking * that can be rolled back using the list) and apply the changes via * nfsvno_advlock(). Optionally, lock the list. It is expected that either * the rollback or update function will be called after this. * It returns an error (and rolls back, as required), if any nfsvno_advlock() * call fails. If it returns an error, it will unlock the list. */ static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags, uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p) { struct nfslock *lop, *nlop; int error = 0; /* Loop through the list of locks. */ lop = LIST_FIRST(&lfp->lf_locallock); while (first < end && lop != NULL) { nlop = LIST_NEXT(lop, lo_lckowner); if (first >= lop->lo_end) { /* not there yet */ lop = nlop; } else if (first < lop->lo_first) { /* new one starts before entry in list */ if (end <= lop->lo_first) { /* no overlap between old and new */ error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first, end, cfp, p); if (error != 0) break; first = end; } else { /* handle fragment overlapped with new one */ error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first, lop->lo_first, cfp, p); if (error != 0) break; first = lop->lo_first; } } else { /* new one overlaps this entry in list */ if (end <= lop->lo_end) { /* overlaps all of new one */ error = nfsrv_dolocal(vp, lfp, flags, lop->lo_flags, first, end, cfp, p); if (error != 0) break; first = end; } else { /* handle fragment overlapped with new one */ error = nfsrv_dolocal(vp, lfp, flags, lop->lo_flags, first, lop->lo_end, cfp, p); if (error != 0) break; first = lop->lo_end; lop = nlop; } } } if (first < end && error == 0) /* handle fragment past end of list */ error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first, end, cfp, p); NFSEXITCODE(error); return (error); } /* * Local lock unlock. Unlock all byte ranges that are no longer locked * by NFSv4. To do this, unlock any subranges of first-->end that * do not overlap with the byte ranges of any lock in the lfp->lf_lock * list. This list has all locks for the file held by other * tuples. The list is ordered by increasing * lo_first value, but may have entries that overlap each other, for * the case of read locks. */ static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first, uint64_t init_end, NFSPROC_T *p) { struct nfslock *lop; uint64_t first, end, prevfirst __unused; first = init_first; end = init_end; while (first < init_end) { /* Loop through all nfs locks, adjusting first and end */ prevfirst = 0; LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) { KASSERT(prevfirst <= lop->lo_first, ("nfsv4 locks out of order")); KASSERT(lop->lo_first < lop->lo_end, ("nfsv4 bogus lock")); prevfirst = lop->lo_first; if (first >= lop->lo_first && first < lop->lo_end) /* * Overlaps with initial part, so trim * off that initial part by moving first past * it. */ first = lop->lo_end; else if (end > lop->lo_first && lop->lo_first > first) { /* * This lock defines the end of the * segment to unlock, so set end to the * start of it and break out of the loop. */ end = lop->lo_first; break; } if (first >= end) /* * There is no segment left to do, so * break out of this loop and then exit * the outer while() since first will be set * to end, which must equal init_end here. */ break; } if (first < end) { /* Unlock this segment */ (void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK, NFSLCK_READ, first, end, NULL, p); nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK, first, end); } /* * Now move past this segment and look for any further * segment in the range, if there is one. */ first = end; end = init_end; } } /* * Do the local lock operation and update the rollback list, as required. * Perform the rollback and return the error if nfsvno_advlock() fails. */ static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p) { struct nfsrollback *rlp; int error = 0, ltype, oldltype; if (flags & NFSLCK_WRITE) ltype = F_WRLCK; else if (flags & NFSLCK_READ) ltype = F_RDLCK; else ltype = F_UNLCK; if (oldflags & NFSLCK_WRITE) oldltype = F_WRLCK; else if (oldflags & NFSLCK_READ) oldltype = F_RDLCK; else oldltype = F_UNLCK; if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK)) /* nothing to do */ goto out; error = nfsvno_advlock(vp, ltype, first, end, p); if (error != 0) { if (cfp != NULL) { cfp->cl_clientid.lval[0] = 0; cfp->cl_clientid.lval[1] = 0; cfp->cl_first = 0; cfp->cl_end = NFS64BITSSET; cfp->cl_flags = NFSLCK_WRITE; cfp->cl_ownerlen = 5; NFSBCOPY("LOCAL", cfp->cl_owner, 5); } nfsrv_locallock_rollback(vp, lfp, p); } else if (ltype != F_UNLCK) { rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK, M_WAITOK); rlp->rlck_first = first; rlp->rlck_end = end; rlp->rlck_type = oldltype; LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list); } out: NFSEXITCODE(error); return (error); } /* * Roll back local lock changes and free up the rollback list. */ static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p) { struct nfsrollback *rlp, *nrlp; LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) { (void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first, rlp->rlck_end, p); free(rlp, M_NFSDROLLBACK); } LIST_INIT(&lfp->lf_rollback); } /* * Update local lock list and delete rollback list (ie now committed to the * local locks). Most of the work is done by the internal function. */ static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first, uint64_t end) { struct nfsrollback *rlp, *nrlp; struct nfslock *new_lop, *other_lop; new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK); if (flags & (NFSLCK_READ | NFSLCK_WRITE)) other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK); else other_lop = NULL; new_lop->lo_flags = flags; new_lop->lo_first = first; new_lop->lo_end = end; nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp); if (new_lop != NULL) free(new_lop, M_NFSDLOCK); if (other_lop != NULL) free(other_lop, M_NFSDLOCK); /* and get rid of the rollback list */ LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) free(rlp, M_NFSDROLLBACK); LIST_INIT(&lfp->lf_rollback); } /* * Lock the struct nfslockfile for local lock updating. */ static void nfsrv_locklf(struct nfslockfile *lfp) { int gotlock; /* lf_usecount ensures *lfp won't be free'd */ lfp->lf_usecount++; do { gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL, NFSSTATEMUTEXPTR, NULL); } while (gotlock == 0); lfp->lf_usecount--; } /* * Unlock the struct nfslockfile after local lock updating. */ static void nfsrv_unlocklf(struct nfslockfile *lfp) { nfsv4_unlock(&lfp->lf_locallock_lck, 0); } /* * Clear out all state for the NFSv4 server. * Must be called by a thread that can sleep when no nfsds are running. */ void nfsrv_throwawayallstate(NFSPROC_T *p) { struct nfsclient *clp, *nclp; struct nfslockfile *lfp, *nlfp; int i; /* * For each client, clean out the state and then free the structure. */ for (i = 0; i < nfsrv_clienthashsize; i++) { LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) { nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); free(clp->lc_stateid, M_NFSDCLIENT); free(clp, M_NFSDCLIENT); } } /* * Also, free up any remaining lock file structures. */ for (i = 0; i < nfsrv_lockhashsize; i++) { LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) { printf("nfsd unload: fnd a lock file struct\n"); nfsrv_freenfslockfile(lfp); } } /* And get rid of the deviceid structures and layouts. */ nfsrv_freealllayoutsanddevids(); } /* * Check the sequence# for the session and slot provided as an argument. * Also, renew the lease if the session will return NFS_OK. */ int nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid, uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this, uint32_t *sflagsp, NFSPROC_T *p) { struct nfsdsession *sep; struct nfssessionhash *shp; int error; shp = NFSSESSIONHASH(nd->nd_sessionid); NFSLOCKSESSION(shp); sep = nfsrv_findsession(nd->nd_sessionid); if (sep == NULL) { NFSUNLOCKSESSION(shp); return (NFSERR_BADSESSION); } error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp, sep->sess_slots, NULL, NFSV4_SLOTS - 1); if (error != 0) { NFSUNLOCKSESSION(shp); return (error); } if (cache_this != 0) nd->nd_flag |= ND_SAVEREPLY; /* Renew the lease. */ sep->sess_clp->lc_expiry = nfsrv_leaseexpiry(); nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval; nd->nd_flag |= ND_IMPLIEDCLID; /* Save maximum request and reply sizes. */ nd->nd_maxreq = sep->sess_maxreq; nd->nd_maxresp = sep->sess_maxresp; *sflagsp = 0; if (sep->sess_clp->lc_req.nr_client == NULL || (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0) *sflagsp |= NFSV4SEQ_CBPATHDOWN; NFSUNLOCKSESSION(shp); if (error == NFSERR_EXPIRED) { *sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED; error = 0; } else if (error == NFSERR_ADMINREVOKED) { *sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED; error = 0; } *highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1; return (0); } /* * Check/set reclaim complete for this session/clientid. */ int nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs) { struct nfsdsession *sep; struct nfssessionhash *shp; int error = 0; shp = NFSSESSIONHASH(nd->nd_sessionid); NFSLOCKSTATE(); NFSLOCKSESSION(shp); sep = nfsrv_findsession(nd->nd_sessionid); if (sep == NULL) { NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); return (NFSERR_BADSESSION); } if (onefs != 0) sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS; /* Check to see if reclaim complete has already happened. */ else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) error = NFSERR_COMPLETEALREADY; else { sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE; nfsrv_markreclaim(sep->sess_clp); } NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); return (error); } /* * Cache the reply in a session slot. */ void nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m) { struct nfsdsession *sep; struct nfssessionhash *shp; char *buf, *cp; #ifdef INET struct sockaddr_in *sin; #endif #ifdef INET6 struct sockaddr_in6 *sin6; #endif shp = NFSSESSIONHASH(nd->nd_sessionid); NFSLOCKSESSION(shp); sep = nfsrv_findsession(nd->nd_sessionid); if (sep == NULL) { NFSUNLOCKSESSION(shp); if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) { buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK); switch (nd->nd_nam->sa_family) { #ifdef INET case AF_INET: sin = (struct sockaddr_in *)nd->nd_nam; cp = inet_ntop(sin->sin_family, &sin->sin_addr.s_addr, buf, INET6_ADDRSTRLEN); break; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)nd->nd_nam; cp = inet_ntop(sin6->sin6_family, &sin6->sin6_addr, buf, INET6_ADDRSTRLEN); break; #endif default: cp = NULL; } if (cp != NULL) printf("nfsrv_cache_session: no session " "IPaddr=%s, check NFS clients for unique " "/etc/hostid's\n", cp); else printf("nfsrv_cache_session: no session, " "check NFS clients for unique " "/etc/hostid's\n"); free(buf, M_TEMP); } m_freem(*m); return; } nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat, m); NFSUNLOCKSESSION(shp); } /* * Search for a session that matches the sessionid. */ static struct nfsdsession * nfsrv_findsession(uint8_t *sessionid) { struct nfsdsession *sep; struct nfssessionhash *shp; shp = NFSSESSIONHASH(sessionid); LIST_FOREACH(sep, &shp->list, sess_hash) { if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID)) break; } return (sep); } /* * Destroy a session. */ int nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid) { int error, igotlock, samesess; samesess = 0; if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) && (nd->nd_flag & ND_HASSEQUENCE) != 0) { samesess = 1; if ((nd->nd_flag & ND_LASTOP) == 0) return (NFSERR_BADSESSION); } /* Lock out other nfsd threads */ NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (igotlock == 0); NFSUNLOCKV4ROOTMUTEX(); error = nfsrv_freesession(NULL, sessionid); if (error == 0 && samesess != 0) nd->nd_flag &= ~ND_HASSEQUENCE; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); return (error); } /* * Bind a connection to a session. * For now, only certain variants are supported, since the current session * structure can only handle a single backchannel entry, which will be * applied to all connections if it is set. */ int nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp) { struct nfssessionhash *shp; struct nfsdsession *sep; struct nfsclient *clp; SVCXPRT *savxprt; int error; error = 0; savxprt = NULL; shp = NFSSESSIONHASH(sessionid); NFSLOCKSTATE(); NFSLOCKSESSION(shp); sep = nfsrv_findsession(sessionid); if (sep != NULL) { clp = sep->sess_clp; if (*foreaftp == NFSCDFC4_BACK || *foreaftp == NFSCDFC4_BACK_OR_BOTH || *foreaftp == NFSCDFC4_FORE_OR_BOTH) { /* Try to set up a backchannel. */ if (clp->lc_req.nr_client == NULL) { NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire " "backchannel\n"); clp->lc_req.nr_client = (struct __rpc_client *) clnt_bck_create(nd->nd_xprt->xp_socket, sep->sess_cbprogram, NFSV4_CBVERS); } if (clp->lc_req.nr_client != NULL) { NFSD_DEBUG(2, "nfsrv_bindconnsess: set up " "backchannel\n"); savxprt = sep->sess_cbsess.nfsess_xprt; SVC_ACQUIRE(nd->nd_xprt); CLNT_ACQUIRE(clp->lc_req.nr_client); nd->nd_xprt->xp_p2 = clp->lc_req.nr_client; /* Disable idle timeout. */ nd->nd_xprt->xp_idletimeout = 0; sep->sess_cbsess.nfsess_xprt = nd->nd_xprt; sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN; clp->lc_flags |= LCL_DONEBINDCONN | LCL_NEEDSCBNULL; clp->lc_flags &= ~LCL_CBDOWN; if (*foreaftp == NFSCDFS4_BACK) *foreaftp = NFSCDFS4_BACK; else *foreaftp = NFSCDFS4_BOTH; } else if (*foreaftp != NFSCDFC4_BACK) { NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set " "up backchannel\n"); sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN; clp->lc_flags |= LCL_DONEBINDCONN; *foreaftp = NFSCDFS4_FORE; } else { error = NFSERR_NOTSUPP; printf("nfsrv_bindconnsess: Can't add " "backchannel\n"); } } else { NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n"); clp->lc_flags |= LCL_DONEBINDCONN; *foreaftp = NFSCDFS4_FORE; } } else error = NFSERR_BADSESSION; NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); if (savxprt != NULL) SVC_RELEASE(savxprt); return (error); } /* * Free up a session structure. */ static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid) { struct nfssessionhash *shp; int i; NFSLOCKSTATE(); if (sep == NULL) { shp = NFSSESSIONHASH(sessionid); NFSLOCKSESSION(shp); sep = nfsrv_findsession(sessionid); } else { shp = NFSSESSIONHASH(sep->sess_sessionid); NFSLOCKSESSION(shp); } if (sep != NULL) { sep->sess_refcnt--; if (sep->sess_refcnt > 0) { NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); return (NFSERR_BACKCHANBUSY); } LIST_REMOVE(sep, sess_hash); LIST_REMOVE(sep, sess_list); } NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); if (sep == NULL) return (NFSERR_BADSESSION); for (i = 0; i < NFSV4_SLOTS; i++) if (sep->sess_slots[i].nfssl_reply != NULL) m_freem(sep->sess_slots[i].nfssl_reply); if (sep->sess_cbsess.nfsess_xprt != NULL) SVC_RELEASE(sep->sess_cbsess.nfsess_xprt); free(sep, M_NFSDSESSION); return (0); } /* * Free a stateid. * RFC5661 says that it should fail when there are associated opens, locks * or delegations. Since stateids represent opens, I don't see how you can * free an open stateid (it will be free'd when closed), so this function * only works for lock stateids (freeing the lock_owner) or delegations. */ int nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, NFSPROC_T *p) { struct nfsclient *clp; struct nfsstate *stp; int error; NFSLOCKSTATE(); /* * Look up the stateid */ error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error == 0) { /* First, check for a delegation. */ LIST_FOREACH(stp, &clp->lc_deleg, ls_list) { if (!NFSBCMP(stp->ls_stateid.other, stateidp->other, NFSX_STATEIDOTHER)) break; } if (stp != NULL) { nfsrv_freedeleg(stp); NFSUNLOCKSTATE(); return (error); } } /* Not a delegation, try for a lock_owner. */ if (error == 0) error = nfsrv_getstate(clp, stateidp, 0, &stp); if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0)) /* Not a lock_owner stateid. */ error = NFSERR_LOCKSHELD; if (error == 0 && !LIST_EMPTY(&stp->ls_lock)) error = NFSERR_LOCKSHELD; if (error == 0) nfsrv_freelockowner(stp, NULL, 0, p); NFSUNLOCKSTATE(); return (error); } /* * Test a stateid. */ int nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, NFSPROC_T *p) { struct nfsclient *clp; struct nfsstate *stp; int error; NFSLOCKSTATE(); /* * Look up the stateid */ error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error == 0) error = nfsrv_getstate(clp, stateidp, 0, &stp); if (error == 0 && stateidp->seqid != 0 && SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid)) error = NFSERR_OLDSTATEID; NFSUNLOCKSTATE(); return (error); } /* * Generate the xdr for an NFSv4.1 CBSequence Operation. */ static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp, int dont_replycache, struct nfsdsession **sepp, int *slotposp) { struct nfsdsession *sep; uint32_t *tl, slotseq = 0; int maxslot; uint8_t sessionid[NFSX_V4SESSIONID]; int error; error = nfsv4_getcbsession(clp, sepp); if (error != 0) return (error); sep = *sepp; nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot, &slotseq, sessionid, true); KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot")); /* Build the Sequence arguments. */ NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED); bcopy(sessionid, tl, NFSX_V4SESSIONID); tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; nd->nd_slotseq = tl; nd->nd_slotid = *slotposp; nd->nd_flag |= ND_HASSLOTID; *tl++ = txdr_unsigned(slotseq); *tl++ = txdr_unsigned(*slotposp); *tl++ = txdr_unsigned(maxslot); if (dont_replycache == 0) *tl++ = newnfs_true; else *tl++ = newnfs_false; *tl = 0; /* No referring call list, for now. */ nd->nd_flag |= ND_HASSEQUENCE; return (0); } /* * Get a session for the callback. */ static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp) { struct nfsdsession *sep; NFSLOCKSTATE(); LIST_FOREACH(sep, &clp->lc_session, sess_list) { if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) break; } if (sep == NULL) { NFSUNLOCKSTATE(); return (NFSERR_BADSESSION); } sep->sess_refcnt++; *sepp = sep; NFSUNLOCKSTATE(); return (0); } /* * Free up all backchannel xprts. This needs to be done when the nfsd threads * exit, since those transports will all be going away. * This is only called after all the nfsd threads are done performing RPCs, * so locking shouldn't be an issue. */ void nfsrv_freeallbackchannel_xprts(void) { struct nfsdsession *sep; struct nfsclient *clp; SVCXPRT *xprt; int i; for (i = 0; i < nfsrv_clienthashsize; i++) { LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { LIST_FOREACH(sep, &clp->lc_session, sess_list) { xprt = sep->sess_cbsess.nfsess_xprt; sep->sess_cbsess.nfsess_xprt = NULL; if (xprt != NULL) SVC_RELEASE(xprt); } } } } /* * Do a layout commit. Actually just call nfsrv_updatemdsattr(). * I have no idea if the rest of these arguments will ever be useful? */ int nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype, int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len, int hasnewmtime, struct timespec *newmtimep, int reclaim, nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep, uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p) { struct nfsvattr na; int error; error = nfsrv_updatemdsattr(vp, &na, p); if (error == 0) { *hasnewsizep = 1; *newsizep = na.na_size; } return (error); } /* * Try and get a layout. */ int nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp, int layouttype, int *iomode, uint64_t *offset, uint64_t *len, uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose, int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p) { struct nfslayouthash *lhyp; struct nfslayout *lyp; char *devid; fhandle_t fh, *dsfhp; int error, mirrorcnt; if (nfsrv_devidcnt == 0) return (NFSERR_UNKNLAYOUTTYPE); if (*offset != 0) printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset, (uintmax_t)*len); error = nfsvno_getfh(vp, &fh, p); NFSD_DEBUG(4, "layoutget getfh=%d\n", error); if (error != 0) return (error); /* * For now, all layouts are for entire files. * Only issue Read/Write layouts if requested for a non-readonly fs. */ if (NFSVNO_EXRDONLY(exp)) { if (*iomode == NFSLAYOUTIOMODE_RW) return (NFSERR_LAYOUTTRYLATER); *iomode = NFSLAYOUTIOMODE_READ; } if (*iomode != NFSLAYOUTIOMODE_RW) *iomode = NFSLAYOUTIOMODE_READ; /* * Check to see if a write layout can be issued for this file. * This is used during mirror recovery to avoid RW layouts being * issued for a file while it is being copied to the recovered * mirror. */ if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0) return (NFSERR_LAYOUTTRYLATER); *retonclose = 0; *offset = 0; *len = UINT64_MAX; /* First, see if a layout already exists and return if found. */ lhyp = NFSLAYOUTHASH(&fh); NFSLOCKLAYOUT(lhyp); error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp); NFSD_DEBUG(4, "layoutget findlay=%d\n", error); /* * Not sure if the seqid must be the same, so I won't check it. */ if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] || stateidp->other[1] != lyp->lay_stateid.other[1] || stateidp->other[2] != lyp->lay_stateid.other[2])) { if ((lyp->lay_flags & NFSLAY_CALLB) == 0) { NFSUNLOCKLAYOUT(lhyp); NFSD_DEBUG(1, "ret bad stateid\n"); return (NFSERR_BADSTATEID); } /* * I believe we get here because there is a race between * the client processing the CBLAYOUTRECALL and the layout * being deleted here on the server. * The client has now done a LayoutGet with a non-layout * stateid, as it would when there is no layout. * As such, free this layout and set error == NFSERR_BADSTATEID * so the code below will create a new layout structure as * would happen if no layout was found. * "lyp" will be set before being used below, but set it NULL * as a safety belt. */ nfsrv_freelayout(&lhyp->list, lyp); lyp = NULL; error = NFSERR_BADSTATEID; } if (error == 0) { if (lyp->lay_layoutlen > maxcnt) { NFSUNLOCKLAYOUT(lhyp); NFSD_DEBUG(1, "ret layout too small\n"); return (NFSERR_TOOSMALL); } if (*iomode == NFSLAYOUTIOMODE_RW) { if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) { NFSUNLOCKLAYOUT(lhyp); NFSD_DEBUG(1, "ret layout nospace\n"); return (NFSERR_NOSPC); } lyp->lay_flags |= NFSLAY_RW; } else lyp->lay_flags |= NFSLAY_READ; NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen); *layoutlenp = lyp->lay_layoutlen; if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; stateidp->seqid = lyp->lay_stateid.seqid; NFSUNLOCKLAYOUT(lhyp); NFSD_DEBUG(4, "ret fnd layout\n"); return (0); } NFSUNLOCKLAYOUT(lhyp); /* Find the device id and file handle. */ dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK); devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK); error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid); NFSD_DEBUG(4, "layoutget devandfh=%d\n", error); if (error == 0) { if (layouttype == NFSLAYOUT_NFSV4_1_FILES) { if (NFSX_V4FILELAYOUT > maxcnt) error = NFSERR_TOOSMALL; else lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp, devid, vp->v_mount->mnt_stat.f_fsid); } else { if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt) error = NFSERR_TOOSMALL; else lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt, &fh, dsfhp, devid, vp->v_mount->mnt_stat.f_fsid); } } free(dsfhp, M_TEMP); free(devid, M_TEMP); if (error != 0) return (error); /* * Now, add this layout to the list. */ error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p); NFSD_DEBUG(4, "layoutget addl=%d\n", error); /* * The lyp will be set to NULL by nfsrv_addlayout() if it * linked the new structure into the lists. */ free(lyp, M_NFSDSTATE); return (error); } /* * Generate a File Layout. */ static struct nfslayout * nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs) { uint32_t *tl; struct nfslayout *lyp; uint64_t pattern_offset; lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE, M_WAITOK | M_ZERO); lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES; if (iomode == NFSLAYOUTIOMODE_RW) lyp->lay_flags = NFSLAY_RW; else lyp->lay_flags = NFSLAY_READ; NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp)); lyp->lay_clientid.qval = nd->nd_clientid.qval; lyp->lay_fsid = fs; NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID); /* Fill in the xdr for the files layout. */ tl = (uint32_t *)lyp->lay_xdr; NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */ tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); /* Set the stripe size to the maximum I/O size. */ *tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK); *tl++ = 0; /* 1st stripe index. */ pattern_offset = 0; txdr_hyper(pattern_offset, tl); tl += 2; /* Pattern offset. */ *tl++ = txdr_unsigned(1); /* 1 file handle. */ *tl++ = txdr_unsigned(NFSX_V4PNFSFH); NFSBCOPY(dsfhp, tl, sizeof(*dsfhp)); lyp->lay_layoutlen = NFSX_V4FILELAYOUT; return (lyp); } #define FLEX_OWNERID "999" #define FLEX_UID0 "0" /* * Generate a Flex File Layout. * The FLEX_OWNERID can be any string of 3 decimal digits. Although this * string goes on the wire, it isn't supposed to be used by the client, * since this server uses tight coupling. * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use * a string of "0". This works around the Linux Flex File Layout driver bug * which uses the synthetic uid/gid strings for the "tightly coupled" case. */ static struct nfslayout * nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs) { uint32_t *tl; struct nfslayout *lyp; uint64_t lenval; int i; lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt), M_NFSDSTATE, M_WAITOK | M_ZERO); lyp->lay_type = NFSLAYOUT_FLEXFILE; if (iomode == NFSLAYOUTIOMODE_RW) lyp->lay_flags = NFSLAY_RW; else lyp->lay_flags = NFSLAY_READ; NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp)); lyp->lay_clientid.qval = nd->nd_clientid.qval; lyp->lay_fsid = fs; lyp->lay_mirrorcnt = mirrorcnt; NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID); /* Fill in the xdr for the files layout. */ tl = (uint32_t *)lyp->lay_xdr; lenval = 0; txdr_hyper(lenval, tl); tl += 2; /* Stripe unit. */ *tl++ = txdr_unsigned(mirrorcnt); /* # of mirrors. */ for (i = 0; i < mirrorcnt; i++) { *tl++ = txdr_unsigned(1); /* One stripe. */ NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */ tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); devid += NFSX_V4DEVICEID; *tl++ = txdr_unsigned(1); /* Efficiency. */ *tl++ = 0; /* Proxy Stateid. */ *tl++ = 0x55555555; *tl++ = 0x55555555; *tl++ = 0x55555555; *tl++ = txdr_unsigned(1); /* 1 file handle. */ *tl++ = txdr_unsigned(NFSX_V4PNFSFH); NFSBCOPY(dsfhp, tl, sizeof(*dsfhp)); tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED); dsfhp++; if (nfsrv_flexlinuxhack != 0) { *tl++ = txdr_unsigned(strlen(FLEX_UID0)); *tl = 0; /* 0 pad string. */ NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0)); *tl++ = txdr_unsigned(strlen(FLEX_UID0)); *tl = 0; /* 0 pad string. */ NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0)); } else { *tl++ = txdr_unsigned(strlen(FLEX_OWNERID)); NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED); *tl++ = txdr_unsigned(strlen(FLEX_OWNERID)); NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED); } } *tl++ = txdr_unsigned(0); /* ff_flags. */ *tl = txdr_unsigned(60); /* Status interval hint. */ lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt); return (lyp); } /* * Parse and process Flex File errors returned via LayoutReturn. */ static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt, NFSPROC_T *p) { uint32_t *tl; int cnt, errcnt, i, j, opnum, stat; char devid[NFSX_V4DEVICEID]; tl = layp; maxcnt -= NFSX_UNSIGNED; if (maxcnt > 0) cnt = fxdr_unsigned(int, *tl++); else cnt = 0; NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt); for (i = 0; i < cnt; i++) { maxcnt -= NFSX_STATEID + 2 * NFSX_HYPER + NFSX_UNSIGNED; if (maxcnt <= 0) break; /* Skip offset, length and stateid for now. */ tl += (4 + NFSX_STATEID / NFSX_UNSIGNED); errcnt = fxdr_unsigned(int, *tl++); NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt); for (j = 0; j < errcnt; j++) { maxcnt -= NFSX_V4DEVICEID + 2 * NFSX_UNSIGNED; if (maxcnt < 0) break; NFSBCOPY(tl, devid, NFSX_V4DEVICEID); tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); stat = fxdr_unsigned(int, *tl++); opnum = fxdr_unsigned(int, *tl++); NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum, stat); /* * Except for NFSERR_ACCES, NFSERR_STALE and * NFSERR_NOSPC errors, disable the mirror. */ if (stat != NFSERR_ACCES && stat != NFSERR_STALE && stat != NFSERR_NOSPC) nfsrv_delds(devid, p); /* For NFSERR_NOSPC, mark all devids and layouts. */ if (stat == NFSERR_NOSPC) nfsrv_marknospc(devid, true); } } } /* * This function removes all flex file layouts which has a mirror with * a device id that matches the argument. * Called when the DS represented by the device id has failed. */ void nfsrv_flexmirrordel(char *devid, NFSPROC_T *p) { uint32_t *tl; struct nfslayout *lyp, *nlyp; struct nfslayouthash *lhyp; struct nfslayouthead loclyp; int i, j; NFSD_DEBUG(4, "flexmirrordel\n"); /* Move all layouts found onto a local list. */ TAILQ_INIT(&loclyp); for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (lyp->lay_type == NFSLAYOUT_FLEXFILE && lyp->lay_mirrorcnt > 1) { NFSD_DEBUG(4, "possible match\n"); tl = lyp->lay_xdr; tl += 3; for (j = 0; j < lyp->lay_mirrorcnt; j++) { tl++; if (NFSBCMP(devid, tl, NFSX_V4DEVICEID) == 0) { /* Found one. */ NFSD_DEBUG(4, "fnd one\n"); TAILQ_REMOVE(&lhyp->list, lyp, lay_list); TAILQ_INSERT_HEAD(&loclyp, lyp, lay_list); break; } tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED + NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED + 11 * NFSX_UNSIGNED); } } } NFSUNLOCKLAYOUT(lhyp); } /* Now, try to do a Layout recall for each one found. */ TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) { NFSD_DEBUG(4, "do layout recall\n"); /* * The layout stateid.seqid needs to be incremented * before doing a LAYOUT_RECALL callback. */ if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid, &lyp->lay_fh, lyp, 1, lyp->lay_type, p); nfsrv_freelayout(&loclyp, lyp); } } /* * Do a recall callback to the client for this layout. */ static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p) { struct nfsclient *clp; int error; NFSD_DEBUG(4, "nfsrv_recalllayout\n"); error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, NULL, p); NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error); if (error != 0) { printf("nfsrv_recalllayout: getclient err=%d\n", error); return (error); } if ((clp->lc_flags & LCL_NFSV41) != 0) { error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL, stateidp, changed, fhp, NULL, NULL, laytype, p); /* If lyp != NULL, handle an error return here. */ if (error != 0 && lyp != NULL) { NFSDRECALLLOCK(); /* * Mark it returned, since no layout recall * has been done. * All errors seem to be non-recoverable, although * NFSERR_NOMATCHLAYOUT is a normal event. */ if ((lyp->lay_flags & NFSLAY_RECALL) != 0) { lyp->lay_flags |= NFSLAY_RETURNED; wakeup(lyp); } NFSDRECALLUNLOCK(); if (error != NFSERR_NOMATCHLAYOUT) printf("nfsrv_recalllayout: err=%d\n", error); } } else printf("nfsrv_recalllayout: clp not NFSv4.1\n"); return (error); } /* * Find a layout to recall when we exceed our high water mark. */ void nfsrv_recalloldlayout(NFSPROC_T *p) { struct nfslayouthash *lhyp; struct nfslayout *lyp; nfsquad_t clientid; nfsv4stateid_t stateid; fhandle_t fh; int error, laytype = 0, ret; lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) { if ((lyp->lay_flags & NFSLAY_CALLB) == 0) { lyp->lay_flags |= NFSLAY_CALLB; /* * The layout stateid.seqid needs to be incremented * before doing a LAYOUT_RECALL callback. */ if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; clientid = lyp->lay_clientid; stateid = lyp->lay_stateid; NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh)); laytype = lyp->lay_type; break; } } NFSUNLOCKLAYOUT(lhyp); if (lyp != NULL) { error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0, laytype, p); if (error != 0 && error != NFSERR_NOMATCHLAYOUT) NFSD_DEBUG(4, "recallold=%d\n", error); if (error != 0) { NFSLOCKLAYOUT(lhyp); /* * Since the hash list was unlocked, we need to * find it again. */ ret = nfsrv_findlayout(&clientid, &fh, laytype, p, &lyp); if (ret == 0 && (lyp->lay_flags & NFSLAY_CALLB) != 0 && lyp->lay_stateid.other[0] == stateid.other[0] && lyp->lay_stateid.other[1] == stateid.other[1] && lyp->lay_stateid.other[2] == stateid.other[2]) { /* * The client no longer knows this layout, so * it can be free'd now. */ if (error == NFSERR_NOMATCHLAYOUT) nfsrv_freelayout(&lhyp->list, lyp); else { /* * Leave it to be tried later by * clearing NFSLAY_CALLB and moving * it to the head of the list, so it * won't be tried again for a while. */ lyp->lay_flags &= ~NFSLAY_CALLB; TAILQ_REMOVE(&lhyp->list, lyp, lay_list); TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list); } } NFSUNLOCKLAYOUT(lhyp); } } } /* * Try and return layout(s). */ int nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp, int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim, int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp, struct ucred *cred, NFSPROC_T *p) { struct nfsvattr na; struct nfslayouthash *lhyp; struct nfslayout *lyp; fhandle_t fh; int error = 0; *fndp = 0; if (kind == NFSV4LAYOUTRET_FILE) { error = nfsvno_getfh(vp, &fh, p); if (error == 0) { error = nfsrv_updatemdsattr(vp, &na, p); if (error != 0) printf("nfsrv_layoutreturn: updatemdsattr" " failed=%d\n", error); } if (error == 0) { if (reclaim == newnfs_true) { error = nfsrv_checkgrace(NULL, NULL, NFSLCK_RECLAIM); if (error != NFSERR_NOGRACE) error = 0; return (error); } lhyp = NFSLAYOUTHASH(&fh); NFSDRECALLLOCK(); NFSLOCKLAYOUT(lhyp); error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp); NFSD_DEBUG(4, "layoutret findlay=%d\n", error); if (error == 0 && stateidp->other[0] == lyp->lay_stateid.other[0] && stateidp->other[1] == lyp->lay_stateid.other[1] && stateidp->other[2] == lyp->lay_stateid.other[2]) { NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d" " %x %x %x laystateid %d %x %x %x" " off=%ju len=%ju flgs=0x%x\n", stateidp->seqid, stateidp->other[0], stateidp->other[1], stateidp->other[2], lyp->lay_stateid.seqid, lyp->lay_stateid.other[0], lyp->lay_stateid.other[1], lyp->lay_stateid.other[2], (uintmax_t)offset, (uintmax_t)len, lyp->lay_flags); if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; stateidp->seqid = lyp->lay_stateid.seqid; if (offset == 0 && len == UINT64_MAX) { if ((iomode & NFSLAYOUTIOMODE_READ) != 0) lyp->lay_flags &= ~NFSLAY_READ; if ((iomode & NFSLAYOUTIOMODE_RW) != 0) lyp->lay_flags &= ~NFSLAY_RW; if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0) nfsrv_freelayout(&lhyp->list, lyp); else *fndp = 1; } else *fndp = 1; } NFSUNLOCKLAYOUT(lhyp); /* Search the nfsrv_recalllist for a match. */ TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && lyp->lay_clientid.qval == nd->nd_clientid.qval && stateidp->other[0] == lyp->lay_stateid.other[0] && stateidp->other[1] == lyp->lay_stateid.other[1] && stateidp->other[2] == lyp->lay_stateid.other[2]) { lyp->lay_flags |= NFSLAY_RETURNED; wakeup(lyp); error = 0; } } NFSDRECALLUNLOCK(); } if (layouttype == NFSLAYOUT_FLEXFILE && layp != NULL) nfsrv_flexlayouterr(nd, layp, maxcnt, p); } else if (kind == NFSV4LAYOUTRET_FSID) nfsrv_freelayouts(&nd->nd_clientid, &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode); else if (kind == NFSV4LAYOUTRET_ALL) nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode); else error = NFSERR_INVAL; if (error == -1) error = 0; return (error); } /* * Look for an existing layout. */ static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype, NFSPROC_T *p, struct nfslayout **lypp) { struct nfslayouthash *lhyp; struct nfslayout *lyp; int ret; *lypp = NULL; ret = 0; lhyp = NFSLAYOUTHASH(fhp); TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 && lyp->lay_clientid.qval == clientidp->qval && lyp->lay_type == laytype) break; } if (lyp != NULL) *lypp = lyp; else ret = -1; return (ret); } /* * Add the new layout, as required. */ static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp, nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p) { struct nfsclient *clp; struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; fhandle_t *fhp; int error; KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0, ("nfsrv_layoutget: no nd_clientid\n")); lyp = *lypp; fhp = &lyp->lay_fh; NFSLOCKSTATE(); error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error != 0) { NFSUNLOCKSTATE(); return (error); } lyp->lay_stateid.seqid = stateidp->seqid = 1; lyp->lay_stateid.other[0] = stateidp->other[0] = clp->lc_clientid.lval[0]; lyp->lay_stateid.other[1] = stateidp->other[1] = clp->lc_clientid.lval[1]; lyp->lay_stateid.other[2] = stateidp->other[2] = nfsrv_nextstateindex(clp); NFSUNLOCKSTATE(); lhyp = NFSLAYOUTHASH(fhp); NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) { if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 && nlyp->lay_clientid.qval == nd->nd_clientid.qval) break; } if (nlyp != NULL) { /* A layout already exists, so use it. */ nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)); NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen); *layoutlenp = nlyp->lay_layoutlen; if (++nlyp->lay_stateid.seqid == 0) nlyp->lay_stateid.seqid = 1; stateidp->seqid = nlyp->lay_stateid.seqid; stateidp->other[0] = nlyp->lay_stateid.other[0]; stateidp->other[1] = nlyp->lay_stateid.other[1]; stateidp->other[2] = nlyp->lay_stateid.other[2]; NFSUNLOCKLAYOUT(lhyp); return (0); } /* Insert the new layout in the lists. */ *lypp = NULL; atomic_add_int(&nfsrv_layoutcnt, 1); nfsstatsv1.srvlayouts++; NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen); *layoutlenp = lyp->lay_layoutlen; TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list); NFSUNLOCKLAYOUT(lhyp); return (0); } /* * Get the devinfo for a deviceid. */ int nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt, uint32_t *notify, int *devaddrlen, char **devaddr) { struct nfsdevice *ds; if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype != NFSLAYOUT_FLEXFILE) || (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES)) return (NFSERR_UNKNLAYOUTTYPE); /* * Now, search for the device id. Note that the structures won't go * away, but the order changes in the list. As such, the lock only * needs to be held during the search through the list. */ NFSDDSLOCK(); TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 && ds->nfsdev_nmp != NULL) break; } NFSDDSUNLOCK(); if (ds == NULL) return (NFSERR_NOENT); /* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */ *devaddrlen = 0; if (layouttype == NFSLAYOUT_NFSV4_1_FILES) { *devaddrlen = ds->nfsdev_fileaddrlen; *devaddr = ds->nfsdev_fileaddr; } else if (layouttype == NFSLAYOUT_FLEXFILE) { *devaddrlen = ds->nfsdev_flexaddrlen; *devaddr = ds->nfsdev_flexaddr; } if (*devaddrlen == 0) return (NFSERR_UNKNLAYOUTTYPE); /* * The XDR overhead is 3 unsigned values: layout_type, * length_of_address and notify bitmap. * If the notify array is changed to not all zeros, the * count of unsigned values must be increased. */ if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED) { *maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED; return (NFSERR_TOOSMALL); } return (0); } /* * Free a list of layout state structures. */ static void nfsrv_freelayoutlist(nfsquad_t clientid) { struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; int i; for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (lyp->lay_clientid.qval == clientid.qval) nfsrv_freelayout(&lhyp->list, lyp); } NFSUNLOCKLAYOUT(lhyp); } } /* * Free up a layout. */ static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp) { NFSD_DEBUG(4, "Freelayout=%p\n", lyp); atomic_add_int(&nfsrv_layoutcnt, -1); nfsstatsv1.srvlayouts--; TAILQ_REMOVE(lhp, lyp, lay_list); free(lyp, M_NFSDSTATE); } /* * Free up a device id. */ void nfsrv_freeonedevid(struct nfsdevice *ds) { int i; atomic_add_int(&nfsrv_devidcnt, -1); vrele(ds->nfsdev_dvp); for (i = 0; i < nfsrv_dsdirsize; i++) if (ds->nfsdev_dsdir[i] != NULL) vrele(ds->nfsdev_dsdir[i]); free(ds->nfsdev_fileaddr, M_NFSDSTATE); free(ds->nfsdev_flexaddr, M_NFSDSTATE); free(ds->nfsdev_host, M_NFSDSTATE); free(ds, M_NFSDSTATE); } /* * Free up a device id and its mirrors. */ static void nfsrv_freedevid(struct nfsdevice *ds) { TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); nfsrv_freeonedevid(ds); } /* * Free all layouts and device ids. * Done when the nfsd threads are shut down since there may be a new * modified device id list created when the nfsd is restarted. */ void nfsrv_freealllayoutsanddevids(void) { struct nfsdontlist *mrp, *nmrp; struct nfslayout *lyp, *nlyp; /* Get rid of the deviceid structures. */ nfsrv_freealldevids(); TAILQ_INIT(&nfsrv_devidhead); nfsrv_devidcnt = 0; /* Get rid of all layouts. */ nfsrv_freealllayouts(); /* Get rid of any nfsdontlist entries. */ LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp) free(mrp, M_NFSDSTATE); LIST_INIT(&nfsrv_dontlisthead); nfsrv_dontlistlen = 0; /* Free layouts in the recall list. */ TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp) nfsrv_freelayout(&nfsrv_recalllisthead, lyp); TAILQ_INIT(&nfsrv_recalllisthead); } /* * Free layouts that match the arguments. */ static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode) { struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; int i; for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (clid->qval != lyp->lay_clientid.qval) continue; if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0) continue; if (laytype != lyp->lay_type) continue; if ((iomode & NFSLAYOUTIOMODE_READ) != 0) lyp->lay_flags &= ~NFSLAY_READ; if ((iomode & NFSLAYOUTIOMODE_RW) != 0) lyp->lay_flags &= ~NFSLAY_RW; if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0) nfsrv_freelayout(&lhyp->list, lyp); } NFSUNLOCKLAYOUT(lhyp); } } /* * Free all layouts for the argument file. */ void nfsrv_freefilelayouts(fhandle_t *fhp) { struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; lhyp = NFSLAYOUTHASH(fhp); NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0) nfsrv_freelayout(&lhyp->list, lyp); } NFSUNLOCKLAYOUT(lhyp); } /* * Free all layouts. */ static void nfsrv_freealllayouts(void) { struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; int i; for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) nfsrv_freelayout(&lhyp->list, lyp); NFSUNLOCKLAYOUT(lhyp); } } /* * Look up the mount path for the DS server. */ static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p, struct nfsdevice **dsp) { struct nameidata nd; struct nfsdevice *ds; struct mount *mp; int error, i; char *dsdirpath; size_t dsdirsize; NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp); *dsp = NULL; NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, dspathp); error = namei(&nd); NFSD_DEBUG(4, "lookup=%d\n", error); if (error != 0) return (error); if (nd.ni_vp->v_type != VDIR) { vput(nd.ni_vp); NFSD_DEBUG(4, "dspath not dir\n"); return (ENOTDIR); } if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); NFSD_DEBUG(4, "dspath not an NFS mount\n"); return (ENXIO); } /* * Allocate a DS server structure with the NFS mounted directory * vnode reference counted, so that a non-forced dismount will * fail with EBUSY. * This structure is always linked into the list, even if an error * is being returned. The caller will free the entire list upon * an error return. */ *dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t), M_NFSDSTATE, M_WAITOK | M_ZERO); ds->nfsdev_dvp = nd.ni_vp; ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount); NFSVOPUNLOCK(nd.ni_vp); dsdirsize = strlen(dspathp) + 16; dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK); /* Now, create the DS directory structures. */ for (i = 0; i < nfsrv_dsdirsize; i++) { snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, dsdirpath); error = namei(&nd); NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error); if (error != 0) break; if (nd.ni_vp->v_type != VDIR) { vput(nd.ni_vp); error = ENOTDIR; NFSD_DEBUG(4, "dsdirpath not a VDIR\n"); break; } if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); error = ENXIO; NFSD_DEBUG(4, "dsdirpath not an NFS mount\n"); break; } ds->nfsdev_dsdir[i] = nd.ni_vp; NFSVOPUNLOCK(nd.ni_vp); } free(dsdirpath, M_TEMP); if (strlen(mdspathp) > 0) { /* * This DS stores file for a specific MDS exported file * system. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, mdspathp); error = namei(&nd); NFSD_DEBUG(4, "mds lookup=%d\n", error); if (error != 0) goto out; if (nd.ni_vp->v_type != VDIR) { vput(nd.ni_vp); error = ENOTDIR; NFSD_DEBUG(4, "mdspath not dir\n"); goto out; } mp = nd.ni_vp->v_mount; if ((mp->mnt_flag & MNT_EXPORTED) == 0) { vput(nd.ni_vp); error = ENXIO; NFSD_DEBUG(4, "mdspath not an exported fs\n"); goto out; } ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid; ds->nfsdev_mdsisset = 1; vput(nd.ni_vp); } out: TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); atomic_add_int(&nfsrv_devidcnt, 1); return (error); } /* * Look up the mount path for the DS server and delete it. */ int nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p) { struct mount *mp; struct nfsmount *nmp; struct nfsdevice *ds; int error; NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp); /* * Search for the path in the mount list. Avoid looking the path * up, since this mount point may be hung, with associated locked * vnodes, etc. * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked * until this completes. * As noted in the man page, this should be done before any forced * dismount on the mount point, but at least the handshake on * NFSMNTP_CANCELRPCS should make it safe. */ error = 0; ds = NULL; nmp = NULL; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 && strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 && mp->mnt_data != NULL) { nmp = VFSTONFS(mp); NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); } else { NFSUNLOCKMNT(nmp); nmp = NULL; } break; } } mtx_unlock(&mountlist_mtx); if (nmp != NULL) { ds = nfsrv_deldsnmp(op, nmp, p); NFSD_DEBUG(4, "deldsnmp=%p\n", ds); if (ds != NULL) { nfsrv_killrpcs(nmp); NFSD_DEBUG(4, "aft killrpcs\n"); } else error = ENXIO; NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); NFSUNLOCKMNT(nmp); } else error = EINVAL; return (error); } /* * Search for and remove a DS entry which matches the "nmp" argument. * The nfsdevice structure pointer is returned so that the caller can * free it via nfsrv_freeonedevid(). * For the forced case, do not try to do LayoutRecalls, since the server * must be shut down now anyhow. */ struct nfsdevice * nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p) { struct nfsdevice *fndds; NFSD_DEBUG(4, "deldsdvp\n"); NFSDDSLOCK(); if (op == PNFSDOP_FORCEDELDS) fndds = nfsv4_findmirror(nmp); else fndds = nfsrv_findmirroredds(nmp); if (fndds != NULL) nfsrv_deleteds(fndds); NFSDDSUNLOCK(); if (fndds != NULL) { if (op != PNFSDOP_FORCEDELDS) nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p); printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host); } return (fndds); } /* * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid. * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount * point. * Also, returns an error instead of the nfsdevice found. */ int nfsrv_delds(char *devid, NFSPROC_T *p) { struct nfsdevice *ds, *fndds; struct nfsmount *nmp; int fndmirror; NFSD_DEBUG(4, "delds\n"); /* * Search the DS server list for a match with devid. * Remove the DS entry if found and there is a mirror. */ fndds = NULL; nmp = NULL; fndmirror = 0; NFSDDSLOCK(); TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 && ds->nfsdev_nmp != NULL) { NFSD_DEBUG(4, "fnd main ds\n"); fndds = ds; break; } } if (fndds == NULL) { NFSDDSUNLOCK(); return (ENXIO); } if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0) fndmirror = 1; else if (fndds->nfsdev_mdsisset != 0) { /* For the fsid is set case, search for a mirror. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds != fndds && ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && fsidcmp(&ds->nfsdev_mdsfsid, &fndds->nfsdev_mdsfsid) == 0) { fndmirror = 1; break; } } } if (fndmirror != 0) { nmp = fndds->nfsdev_nmp; NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); nfsrv_deleteds(fndds); } else { NFSUNLOCKMNT(nmp); nmp = NULL; } } NFSDDSUNLOCK(); if (nmp != NULL) { nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p); printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host); nfsrv_killrpcs(nmp); NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); NFSUNLOCKMNT(nmp); return (0); } return (ENXIO); } /* * Mark a DS as disabled by setting nfsdev_nmp = NULL. */ static void nfsrv_deleteds(struct nfsdevice *fndds) { NFSD_DEBUG(4, "deleteds: deleting a mirror\n"); fndds->nfsdev_nmp = NULL; if (fndds->nfsdev_mdsisset == 0) nfsrv_faildscnt--; } /* * Fill in the addr structures for the File and Flex File layouts. */ static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost) { uint32_t *tl; char *netprot; int addrlen; static uint64_t new_devid = 0; if (strchr(addr, ':') != NULL) netprot = "tcp6"; else netprot = "tcp"; /* Fill in the device id. */ NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time)); new_devid++; NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)], sizeof(new_devid)); /* * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4 * as defined in RFC5661) in XDR. */ addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) + 6 * NFSX_UNSIGNED; NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot); ds->nfsdev_fileaddrlen = addrlen; tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO); ds->nfsdev_fileaddr = (char *)tl; *tl++ = txdr_unsigned(1); /* One stripe with index 0. */ *tl++ = 0; *tl++ = txdr_unsigned(1); /* One multipath list */ *tl++ = txdr_unsigned(1); /* with one entry in it. */ /* The netaddr for this one entry. */ *tl++ = txdr_unsigned(strlen(netprot)); NFSBCOPY(netprot, tl, strlen(netprot)); tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED); *tl++ = txdr_unsigned(strlen(addr)); NFSBCOPY(addr, tl, strlen(addr)); /* * Fill in the flex file addr (actually the ff_device_addr4 * as defined for Flexible File Layout) in XDR. */ addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) + 14 * NFSX_UNSIGNED; ds->nfsdev_flexaddrlen = addrlen; tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO); ds->nfsdev_flexaddr = (char *)tl; *tl++ = txdr_unsigned(1); /* One multipath entry. */ /* The netaddr for this one entry. */ *tl++ = txdr_unsigned(strlen(netprot)); NFSBCOPY(netprot, tl, strlen(netprot)); tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED); *tl++ = txdr_unsigned(strlen(addr)); NFSBCOPY(addr, tl, strlen(addr)); tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED); *tl++ = txdr_unsigned(2); /* Two NFS Versions. */ *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */ *tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */ *tl++ = txdr_unsigned(nfs_srvmaxio); /* DS max rsize. */ *tl++ = txdr_unsigned(nfs_srvmaxio); /* DS max wsize. */ *tl++ = newnfs_true; /* Tightly coupled. */ *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */ *tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */ *tl++ = txdr_unsigned(nfs_srvmaxio); /* DS max rsize. */ *tl++ = txdr_unsigned(nfs_srvmaxio); /* DS max wsize. */ *tl = newnfs_true; /* Tightly coupled. */ ds->nfsdev_hostnamelen = strlen(dnshost); ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE, M_WAITOK); NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1); } /* * Create the device id list. * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument * is misconfigured. */ int nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p) { struct nfsdevice *ds; char *addrp, *dnshostp, *dspathp, *mdspathp; int error, i; addrp = args->addr; dnshostp = args->dnshost; dspathp = args->dspath; mdspathp = args->mdspath; nfsrv_maxpnfsmirror = args->mirrorcnt; if (addrp == NULL || dnshostp == NULL || dspathp == NULL || mdspathp == NULL) return (0); /* * Loop around for each nul-terminated string in args->addr, * args->dnshost, args->dnspath and args->mdspath. */ while (addrp < (args->addr + args->addrlen) && dnshostp < (args->dnshost + args->dnshostlen) && dspathp < (args->dspath + args->dspathlen) && mdspathp < (args->mdspath + args->mdspathlen)) { error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds); if (error != 0) { /* Free all DS servers. */ nfsrv_freealldevids(); nfsrv_devidcnt = 0; return (ENXIO); } nfsrv_allocdevid(ds, addrp, dnshostp); addrp += (strlen(addrp) + 1); dnshostp += (strlen(dnshostp) + 1); dspathp += (strlen(dspathp) + 1); mdspathp += (strlen(mdspathp) + 1); } if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) { /* Free all DS servers. */ nfsrv_freealldevids(); nfsrv_devidcnt = 0; nfsrv_maxpnfsmirror = 1; return (ENXIO); } /* We can fail at most one less DS than the mirror level. */ nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1; /* * Allocate the nfslayout hash table now, since this is a pNFS server. * Make it 1% of the high water mark and at least 100. */ if (nfslayouthash == NULL) { nfsrv_layouthashsize = nfsrv_layouthighwater / 100; if (nfsrv_layouthashsize < 100) nfsrv_layouthashsize = 100; nfslayouthash = mallocarray(nfsrv_layouthashsize, sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK | M_ZERO); for (i = 0; i < nfsrv_layouthashsize; i++) { mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF); TAILQ_INIT(&nfslayouthash[i].list); } } return (0); } /* * Free all device ids. */ static void nfsrv_freealldevids(void) { struct nfsdevice *ds, *nds; TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds) nfsrv_freedevid(ds); } /* * Check to see if there is a Read/Write Layout plus either: * - A Write Delegation * or * - An Open with Write_access. * Return 1 if this is the case and 0 otherwise. * This function is used by nfsrv_proxyds() to decide if doing a Proxy * Getattr RPC to the Data Server (DS) is necessary. */ #define NFSCLIDVECSIZE 6 int nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p) { fhandle_t fh, *tfhp; struct nfsstate *stp; struct nfslayout *lyp; struct nfslayouthash *lhyp; struct nfslockhashhead *hp; struct nfslockfile *lfp; nfsquad_t clid[NFSCLIDVECSIZE]; int clidcnt, ret; ret = nfsvno_getfh(vp, &fh, p); if (ret != 0) return (0); /* First check for a Read/Write Layout. */ clidcnt = 0; lhyp = NFSLAYOUTHASH(&fh); NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && ((lyp->lay_flags & NFSLAY_RW) != 0 || ((lyp->lay_flags & NFSLAY_READ) != 0 && nfsrv_pnfsatime != 0))) { if (clidcnt < NFSCLIDVECSIZE) clid[clidcnt].qval = lyp->lay_clientid.qval; clidcnt++; } } NFSUNLOCKLAYOUT(lhyp); if (clidcnt == 0) { /* None found, so return 0. */ return (0); } /* Get the nfslockfile for this fh. */ NFSLOCKSTATE(); hp = NFSLOCKHASH(&fh); LIST_FOREACH(lfp, hp, lf_hash) { tfhp = &lfp->lf_fh; if (NFSVNO_CMPFH(&fh, tfhp)) break; } if (lfp == NULL) { /* None found, so return 0. */ NFSUNLOCKSTATE(); return (0); } /* Now, look for a Write delegation for this clientid. */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 && nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0) break; } if (stp != NULL) { /* Found one, so return 1. */ NFSUNLOCKSTATE(); return (1); } /* No Write delegation, so look for an Open with Write_access. */ LIST_FOREACH(stp, &lfp->lf_open, ls_file) { KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0, ("nfsrv_checkdsattr: Non-open in Open list\n")); if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 && nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0) break; } NFSUNLOCKSTATE(); if (stp != NULL) return (1); return (0); } /* * Look for a matching clientid in the vector. Return 1 if one might match. */ static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt) { int i; /* If too many for the vector, return 1 since there might be a match. */ if (clidcnt > NFSCLIDVECSIZE) return (1); for (i = 0; i < clidcnt; i++) if (clidvec[i].qval == clid.qval) return (1); return (0); } /* * Check the don't list for "vp" and see if issuing an rw layout is allowed. * Return 1 if issuing an rw layout isn't allowed, 0 otherwise. */ static int nfsrv_dontlayout(fhandle_t *fhp) { struct nfsdontlist *mrp; int ret; if (nfsrv_dontlistlen == 0) return (0); ret = 0; NFSDDONTLISTLOCK(); LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) { if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 && (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) { ret = 1; break; } } NFSDDONTLISTUNLOCK(); return (ret); } #define PNFSDS_COPYSIZ 65536 /* * Create a new file on a DS and copy the contents of an extant DS file to it. * This can be used for recovery of a DS file onto a recovered DS. * The steps are: * - When called, the MDS file's vnode is locked, blocking LayoutGet operations. * - Disable issuing of read/write layouts for the file via the nfsdontlist, * so that they will be disabled after the MDS file's vnode is unlocked. * - Set up the nfsrv_recalllist so that recall of read/write layouts can * be done. * - Unlock the MDS file's vnode, so that the client(s) can perform proxied * writes, LayoutCommits and LayoutReturns for the file when completing the * LayoutReturn requested by the LayoutRecall callback. * - Issue a LayoutRecall callback for all read/write layouts and wait for * them to be returned. (If the LayoutRecall callback replies * NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.) * - Exclusively lock the MDS file's vnode. This ensures that no proxied * writes are in progress or can occur during the DS file copy. * It also blocks Setattr operations. * - Create the file on the recovered mirror. * - Copy the file from the operational DS. * - Copy any ACL from the MDS file to the new DS file. * - Set the modify time of the new DS file to that of the MDS file. * - Update the extended attribute for the MDS file. * - Enable issuing of rw layouts by deleting the nfsdontlist entry. * - The caller will unlock the MDS file's vnode allowing operations * to continue normally, since it is now on the mirror again. */ int nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds, struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt, struct ucred *cred, NFSPROC_T *p) { struct nfsdontlist *mrp, *nmrp; struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; struct nfslayouthead thl; struct mount *mp, *tvmp; struct acl *aclp; struct vattr va; struct timespec mtime; fhandle_t fh; vnode_t tvp; off_t rdpos, wrpos; ssize_t aresid; char *dat; int didprintf, ret, retacl, xfer; ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp"); ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp"); /* * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag * so that no more RW layouts will get issued. */ ret = nfsvno_getfh(vp, &fh, p); if (ret != 0) { NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret); return (ret); } nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK); nmrp->nfsmr_flags = NFSMR_DONTLAYOUT; NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh)); NFSDDONTLISTLOCK(); LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) { if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0) break; } if (mrp == NULL) { LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list); mrp = nmrp; nmrp = NULL; nfsrv_dontlistlen++; NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n"); } else { NFSDDONTLISTUNLOCK(); free(nmrp, M_NFSDSTATE); NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n"); return (ENXIO); } NFSDDONTLISTUNLOCK(); /* * Search for all RW layouts for this file. Move them to the * recall list, so they can be recalled and their return noted. */ lhyp = NFSLAYOUTHASH(&fh); NFSDRECALLLOCK(); NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && (lyp->lay_flags & NFSLAY_RW) != 0) { TAILQ_REMOVE(&lhyp->list, lyp, lay_list); TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list); lyp->lay_trycnt = 0; } } NFSUNLOCKLAYOUT(lhyp); NFSDRECALLUNLOCK(); ret = 0; mp = tvmp = NULL; didprintf = 0; TAILQ_INIT(&thl); /* Unlock the MDS vp, so that a LayoutReturn can be done on it. */ NFSVOPUNLOCK(vp); /* Now, do a recall for all layouts not yet recalled. */ tryagain: NFSDRECALLLOCK(); TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && (lyp->lay_flags & NFSLAY_RECALL) == 0) { lyp->lay_flags |= NFSLAY_RECALL; /* * The layout stateid.seqid needs to be incremented * before doing a LAYOUT_RECALL callback. */ if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; NFSDRECALLUNLOCK(); nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid, &lyp->lay_fh, lyp, 0, lyp->lay_type, p); NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n"); goto tryagain; } } /* Now wait for them to be returned. */ tryagain2: TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) { if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) { TAILQ_REMOVE(&nfsrv_recalllisthead, lyp, lay_list); TAILQ_INSERT_HEAD(&thl, lyp, lay_list); NFSD_DEBUG(4, "nfsrv_copymr: layout returned\n"); } else { lyp->lay_trycnt++; ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR, PVFS | PCATCH, "nfsmrl", hz); NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n", ret); if (ret == EINTR || ret == ERESTART) break; if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) { /* * Give up after 60sec and return * ENXIO, failing the copymr. * This layout will remain on the * recalllist. It can only be cleared * by restarting the nfsd. * This seems the safe way to handle * it, since it cannot be safely copied * with an outstanding RW layout. */ if (lyp->lay_trycnt >= 60) { ret = ENXIO; break; } if (didprintf == 0) { printf("nfsrv_copymr: layout " "not returned\n"); didprintf = 1; } } } goto tryagain2; } } NFSDRECALLUNLOCK(); /* We can now get rid of the layouts that have been returned. */ TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp) nfsrv_freelayout(&thl, lyp); /* * Do the vn_start_write() calls here, before the MDS vnode is * locked and the tvp is created (locked) in the NFS file system * that dvp is in. * For tvmp, this probably isn't necessary, since it will be an * NFS mount and they are not suspendable at this time. */ if (ret == 0) - ret = vn_start_write(vp, &mp, V_WAIT | PCATCH); + ret = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (ret == 0) { tvmp = dvp->v_mount; - ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH); + ret = vn_start_write(NULL, &tvmp, V_WAIT | V_PCATCH); } /* * LK_EXCLUSIVE lock the MDS vnode, so that any * proxied writes through the MDS will be blocked until we have * completed the copy and update of the extended attributes. * This will also ensure that any attributes and ACL will not be * changed until the copy is complete. */ NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (ret == 0 && VN_IS_DOOMED(vp)) { NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n"); ret = ESTALE; } /* Create the data file on the recovered DS. */ if (ret == 0) ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp); /* Copy the DS file, if created successfully. */ if (ret == 0) { /* * Get any NFSv4 ACL on the MDS file, so that it can be set * on the new DS file. */ aclp = acl_alloc(M_WAITOK | M_ZERO); retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); if (retacl != 0 && retacl != ENOATTR) NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl); dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK); /* Malloc a block of 0s used to check for holes. */ if (nfsrv_zeropnfsdat == NULL) nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK | M_ZERO); rdpos = wrpos = 0; ret = VOP_GETATTR(fvp, &va, cred); aresid = 0; while (ret == 0 && aresid == 0) { ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ, rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL, &aresid, p); xfer = PNFSDS_COPYSIZ - aresid; if (ret == 0 && xfer > 0) { rdpos += xfer; /* * Skip the write for holes, except for the * last block. */ if (xfer < PNFSDS_COPYSIZ || rdpos == va.va_size || NFSBCMP(dat, nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0) ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer, wrpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL, NULL, p); if (ret == 0) wrpos += xfer; } } /* If there is an ACL and the copy succeeded, set the ACL. */ if (ret == 0 && retacl == 0) { ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p); /* * Don't consider these as errors, since VOP_GETACL() * can return an ACL when they are not actually * supported. For example, for UFS, VOP_GETACL() * will return a trivial ACL based on the uid/gid/mode * when there is no ACL on the file. * This case should be recognized as a trivial ACL * by UFS's VOP_SETACL() and succeed, but... */ if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM) ret = 0; } if (ret == 0) ret = VOP_FSYNC(tvp, MNT_WAIT, p); /* Set the DS data file's modify time that of the MDS file. */ if (ret == 0) ret = VOP_GETATTR(vp, &va, cred); if (ret == 0) { mtime = va.va_mtime; VATTR_NULL(&va); va.va_mtime = mtime; ret = VOP_SETATTR(tvp, &va, cred); } vput(tvp); acl_free(aclp); free(dat, M_TEMP); } if (tvmp != NULL) vn_finished_write(tvmp); /* Update the extended attributes for the newly created DS file. */ if (ret == 0) ret = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*wpf) * mirrorcnt, (char *)wpf, p); if (mp != NULL) vn_finished_write(mp); /* Get rid of the dontlist entry, so that Layouts can be issued. */ NFSDDONTLISTLOCK(); LIST_REMOVE(mrp, nfsmr_list); NFSDDONTLISTUNLOCK(); free(mrp, M_NFSDSTATE); return (ret); } /* * Create a data storage file on the recovered DS. */ static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf, vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p, vnode_t *tvpp) { struct vattr va, nva; int error; /* Make data file name based on FH. */ error = VOP_GETATTR(vp, &va, cred); if (error == 0) { /* Set the attributes for "vp" to Setattr the DS vp. */ VATTR_NULL(&nva); nva.va_uid = va.va_uid; nva.va_gid = va.va_gid; nva.va_mode = va.va_mode; nva.va_size = 0; VATTR_NULL(&va); va.va_type = VREG; va.va_mode = nva.va_mode; NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf); error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL, pf->dsf_filename, cred, p, tvpp); } return (error); } /* * Look up the MDS file shared locked, and then get the extended attribute * to find the extant DS file to be copied to the new mirror. * If successful, *vpp is set to the MDS file's vp and *nvpp is * set to a DS data file for the MDS file, both exclusively locked. * The "buf" argument has the pnfsdsfile structure from the MDS file * in it and buflen is set to its length. */ int nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf, int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp, struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp, struct nfsdevice **fdsp) { struct nameidata nd; struct vnode *vp, *curvp; struct pnfsdsfile *pf; struct nfsmount *nmp, *curnmp; int dsdir, error, mirrorcnt, ippos; vp = NULL; curvp = NULL; curnmp = NULL; *dsp = NULL; *fdsp = NULL; if (dspathp == NULL && curdspathp != NULL) return (EPERM); /* * Look up the MDS file shared locked. The lock will be upgraded * to an exclusive lock after any rw layouts have been returned. */ NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, mdspathp); error = namei(&nd); NFSD_DEBUG(4, "lookup=%d\n", error); if (error != 0) return (error); if (nd.ni_vp->v_type != VREG) { vput(nd.ni_vp); NFSD_DEBUG(4, "mdspath not reg\n"); return (EISDIR); } vp = nd.ni_vp; if (curdspathp != NULL) { /* * Look up the current DS path and find the nfsdev structure for * it. */ NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, curdspathp); error = namei(&nd); NFSD_DEBUG(4, "ds lookup=%d\n", error); if (error != 0) { vput(vp); return (error); } if (nd.ni_vp->v_type != VDIR) { vput(nd.ni_vp); vput(vp); NFSD_DEBUG(4, "curdspath not dir\n"); return (ENOTDIR); } if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); vput(vp); NFSD_DEBUG(4, "curdspath not an NFS mount\n"); return (ENXIO); } curnmp = VFSTONFS(nd.ni_vp->v_mount); /* Search the nfsdev list for a match. */ NFSDDSLOCK(); *fdsp = nfsv4_findmirror(curnmp); NFSDDSUNLOCK(); if (*fdsp == NULL) curnmp = NULL; if (curnmp == NULL) { vput(nd.ni_vp); vput(vp); NFSD_DEBUG(4, "mdscopymr: no current ds\n"); return (ENXIO); } curvp = nd.ni_vp; } if (dspathp != NULL) { /* Look up the nfsdev path and find the nfsdev structure. */ NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, dspathp); error = namei(&nd); NFSD_DEBUG(4, "ds lookup=%d\n", error); if (error != 0) { vput(vp); if (curvp != NULL) vput(curvp); return (error); } if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) { vput(nd.ni_vp); vput(vp); if (curvp != NULL) vput(curvp); NFSD_DEBUG(4, "dspath not dir\n"); if (nd.ni_vp == curvp) return (EPERM); return (ENOTDIR); } if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); vput(vp); if (curvp != NULL) vput(curvp); NFSD_DEBUG(4, "dspath not an NFS mount\n"); return (ENXIO); } nmp = VFSTONFS(nd.ni_vp->v_mount); /* * Search the nfsdevice list for a match. If curnmp == NULL, * this is a recovery and there must be a mirror. */ NFSDDSLOCK(); if (curnmp == NULL) *dsp = nfsrv_findmirroredds(nmp); else *dsp = nfsv4_findmirror(nmp); NFSDDSUNLOCK(); if (*dsp == NULL) { vput(nd.ni_vp); vput(vp); if (curvp != NULL) vput(curvp); NFSD_DEBUG(4, "mdscopymr: no ds\n"); return (ENXIO); } } else { nd.ni_vp = NULL; nmp = NULL; } /* * Get a vp for an available DS data file using the extended * attribute on the MDS file. * If there is a valid entry for the new DS in the extended attribute * on the MDS file (as checked via the nmp argument), * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur. */ error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p, NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir); if (curvp != NULL) vput(curvp); if (nd.ni_vp == NULL) { if (error == 0 && nmp != NULL) { /* Search the nfsdev list for a match. */ NFSDDSLOCK(); *dsp = nfsrv_findmirroredds(nmp); NFSDDSUNLOCK(); } if (error == 0 && (nmp == NULL || *dsp == NULL)) { if (nvpp != NULL && *nvpp != NULL) { vput(*nvpp); *nvpp = NULL; } error = ENXIO; } } else vput(nd.ni_vp); /* * When dspathp != NULL and curdspathp == NULL, this is a recovery * and is only allowed if there is a 0.0.0.0 IP address entry. * When curdspathp != NULL, the ippos will be set to that entry. */ if (error == 0 && dspathp != NULL && ippos == -1) { if (nvpp != NULL && *nvpp != NULL) { vput(*nvpp); *nvpp = NULL; } error = ENXIO; } if (error == 0) { *vpp = vp; pf = (struct pnfsdsfile *)buf; if (ippos == -1) { /* If no zeroip pnfsdsfile, add one. */ ippos = *buflenp / sizeof(*pf); *buflenp += sizeof(*pf); pf += ippos; pf->dsf_dir = dsdir; strlcpy(pf->dsf_filename, fname, sizeof(pf->dsf_filename)); } else pf += ippos; *pfp = pf; } else vput(vp); return (error); } /* * Search for a matching pnfsd mirror device structure, base on the nmp arg. * Return one if found, NULL otherwise. */ static struct nfsdevice * nfsrv_findmirroredds(struct nfsmount *nmp) { struct nfsdevice *ds, *fndds; int fndmirror; mtx_assert(NFSDDSMUTEXPTR, MA_OWNED); /* * Search the DS server list for a match with nmp. * Remove the DS entry if found and there is a mirror. */ fndds = NULL; fndmirror = 0; if (nfsrv_devidcnt == 0) return (fndds); TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp == nmp) { NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n"); fndds = ds; break; } } if (fndds == NULL) return (fndds); if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0) fndmirror = 1; else if (fndds->nfsdev_mdsisset != 0) { /* For the fsid is set case, search for a mirror. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds != fndds && ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && fsidcmp(&ds->nfsdev_mdsfsid, &fndds->nfsdev_mdsfsid) == 0) { fndmirror = 1; break; } } } if (fndmirror == 0) { NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n"); return (NULL); } return (fndds); } /* * Mark the appropriate devid and all associated layout as "out of space". */ void nfsrv_marknospc(char *devid, bool setit) { struct nfsdevice *ds; struct nfslayout *lyp; struct nfslayouthash *lhyp; int i; NFSDDSLOCK(); TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) { NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit); ds->nfsdev_nospc = setit; } } NFSDDSUNLOCK(); for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { if (NFSBCMP(lyp->lay_deviceid, devid, NFSX_V4DEVICEID) == 0) { NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n", setit); if (setit) lyp->lay_flags |= NFSLAY_NOSPC; else lyp->lay_flags &= ~NFSLAY_NOSPC; } } NFSUNLOCKLAYOUT(lhyp); } } diff --git a/sys/fs/unionfs/union_subr.c b/sys/fs/unionfs/union_subr.c index ac00cc1bc093..0b9e8fde4b6a 100644 --- a/sys/fs/unionfs/union_subr.c +++ b/sys/fs/unionfs/union_subr.c @@ -1,1319 +1,1319 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1994 Jan-Simon Pendry * Copyright (c) 1994 * The Regents of the University of California. All rights reserved. * Copyright (c) 2005, 2006, 2012 Masanori Ozawa , ONGS Inc. * Copyright (c) 2006, 2012 Daichi Goto * * This code is derived from software contributed to Berkeley by * Jan-Simon Pendry. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)union_subr.c 8.20 (Berkeley) 5/20/95 * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define NUNIONFSNODECACHE 16 #define UNIONFSHASHMASK (NUNIONFSNODECACHE - 1) static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table"); MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part"); MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part"); static struct task unionfs_deferred_rele_task; static struct mtx unionfs_deferred_rele_lock; static STAILQ_HEAD(, unionfs_node) unionfs_deferred_rele_list = STAILQ_HEAD_INITIALIZER(unionfs_deferred_rele_list); static TASKQUEUE_DEFINE_THREAD(unionfs_rele); unsigned int unionfs_ndeferred = 0; SYSCTL_UINT(_vfs, OID_AUTO, unionfs_ndeferred, CTLFLAG_RD, &unionfs_ndeferred, 0, "unionfs deferred vnode release"); static void unionfs_deferred_rele(void *, int); /* * Initialize */ int unionfs_init(struct vfsconf *vfsp) { UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */ TASK_INIT(&unionfs_deferred_rele_task, 0, unionfs_deferred_rele, NULL); mtx_init(&unionfs_deferred_rele_lock, "uniondefr", NULL, MTX_DEF); return (0); } /* * Uninitialize */ int unionfs_uninit(struct vfsconf *vfsp) { taskqueue_quiesce(taskqueue_unionfs_rele); taskqueue_free(taskqueue_unionfs_rele); mtx_destroy(&unionfs_deferred_rele_lock); return (0); } static void unionfs_deferred_rele(void *arg __unused, int pending __unused) { STAILQ_HEAD(, unionfs_node) local_rele_list; struct unionfs_node *unp, *tunp; unsigned int ndeferred; ndeferred = 0; STAILQ_INIT(&local_rele_list); mtx_lock(&unionfs_deferred_rele_lock); STAILQ_CONCAT(&local_rele_list, &unionfs_deferred_rele_list); mtx_unlock(&unionfs_deferred_rele_lock); STAILQ_FOREACH_SAFE(unp, &local_rele_list, un_rele, tunp) { ++ndeferred; MPASS(unp->un_dvp != NULL); vrele(unp->un_dvp); free(unp, M_UNIONFSNODE); } /* We expect this function to be single-threaded, thus no atomic */ unionfs_ndeferred += ndeferred; } static struct unionfs_node_hashhead * unionfs_get_hashhead(struct vnode *dvp, struct vnode *lookup) { struct unionfs_node *unp; unp = VTOUNIONFS(dvp); return (&(unp->un_hashtbl[vfs_hash_index(lookup) & UNIONFSHASHMASK])); } /* * Attempt to lookup a cached unionfs vnode by upper/lower vp * from dvp, with dvp's interlock held. */ static struct vnode * unionfs_get_cached_vnode_locked(struct vnode *lookup, struct vnode *dvp) { struct unionfs_node *unp; struct unionfs_node_hashhead *hd; struct vnode *vp; hd = unionfs_get_hashhead(dvp, lookup); LIST_FOREACH(unp, hd, un_hash) { if (unp->un_uppervp == lookup || unp->un_lowervp == lookup) { vp = UNIONFSTOV(unp); VI_LOCK_FLAGS(vp, MTX_DUPOK); vp->v_iflag &= ~VI_OWEINACT; if (VN_IS_DOOMED(vp) || ((vp->v_iflag & VI_DOINGINACT) != 0)) { VI_UNLOCK(vp); vp = NULLVP; } else { vrefl(vp); VI_UNLOCK(vp); } return (vp); } } return (NULLVP); } /* * Get the cached vnode. */ static struct vnode * unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp, struct vnode *dvp) { struct vnode *vp; vp = NULLVP; VI_LOCK(dvp); if (uvp != NULLVP) vp = unionfs_get_cached_vnode_locked(uvp, dvp); else if (lvp != NULLVP) vp = unionfs_get_cached_vnode_locked(lvp, dvp); VI_UNLOCK(dvp); return (vp); } /* * Add the new vnode into cache. */ static struct vnode * unionfs_ins_cached_vnode(struct unionfs_node *uncp, struct vnode *dvp) { struct unionfs_node_hashhead *hd; struct vnode *vp; ASSERT_VOP_ELOCKED(uncp->un_uppervp, __func__); ASSERT_VOP_ELOCKED(uncp->un_lowervp, __func__); KASSERT(uncp->un_uppervp == NULLVP || uncp->un_uppervp->v_type == VDIR, ("%s: v_type != VDIR", __func__)); KASSERT(uncp->un_lowervp == NULLVP || uncp->un_lowervp->v_type == VDIR, ("%s: v_type != VDIR", __func__)); vp = NULLVP; VI_LOCK(dvp); if (uncp->un_uppervp != NULL) vp = unionfs_get_cached_vnode_locked(uncp->un_uppervp, dvp); else if (uncp->un_lowervp != NULL) vp = unionfs_get_cached_vnode_locked(uncp->un_lowervp, dvp); if (vp == NULLVP) { hd = unionfs_get_hashhead(dvp, (uncp->un_uppervp != NULLVP ? uncp->un_uppervp : uncp->un_lowervp)); LIST_INSERT_HEAD(hd, uncp, un_hash); } VI_UNLOCK(dvp); return (vp); } /* * Remove the vnode. */ static void unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp) { KASSERT(unp != NULL, ("%s: null node", __func__)); KASSERT(dvp != NULLVP, ("%s: null parent vnode", __func__)); VI_LOCK(dvp); if (unp->un_hash.le_prev != NULL) { LIST_REMOVE(unp, un_hash); unp->un_hash.le_next = NULL; unp->un_hash.le_prev = NULL; } VI_UNLOCK(dvp); } /* * Common cleanup handling for unionfs_nodeget * Upper, lower, and parent directory vnodes are expected to be referenced by * the caller. Upper and lower vnodes, if non-NULL, are also expected to be * exclusively locked by the caller. * This function will return with the caller's locks and references undone. */ static void unionfs_nodeget_cleanup(struct vnode *vp, struct unionfs_node *unp) { /* * Lock and reset the default vnode lock; vgone() expects a locked * vnode, and we're going to reset the vnode ops. */ lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); /* * Clear out private data and reset the vnode ops to avoid use of * unionfs vnode ops on a partially constructed vnode. */ VI_LOCK(vp); vp->v_data = NULL; vp->v_vnlock = &vp->v_lock; vp->v_op = &dead_vnodeops; VI_UNLOCK(vp); vgone(vp); vput(vp); if (unp->un_dvp != NULLVP) vrele(unp->un_dvp); if (unp->un_uppervp != NULLVP) vput(unp->un_uppervp); if (unp->un_lowervp != NULLVP) vput(unp->un_lowervp); if (unp->un_hashtbl != NULL) hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK); free(unp->un_path, M_UNIONFSPATH); free(unp, M_UNIONFSNODE); } /* * Make a new or get existing unionfs node. * * uppervp and lowervp should be unlocked. Because if new unionfs vnode is * locked, uppervp or lowervp is locked too. In order to prevent dead lock, * you should not lock plurality simultaneously. */ int unionfs_nodeget(struct mount *mp, struct vnode *uppervp, struct vnode *lowervp, struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) { char *path; struct unionfs_mount *ump; struct unionfs_node *unp; struct vnode *vp; u_long hashmask; int error; int lkflags; enum vtype vt; error = 0; ump = MOUNTTOUNIONFSMOUNT(mp); lkflags = (cnp ? cnp->cn_lkflags : 0); path = (cnp ? cnp->cn_nameptr : NULL); *vpp = NULLVP; if (uppervp == NULLVP && lowervp == NULLVP) panic("%s: upper and lower is null", __func__); vt = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type); /* If it has no ISLASTCN flag, path check is skipped. */ if (cnp && !(cnp->cn_flags & ISLASTCN)) path = NULL; /* check the cache */ if (dvp != NULLVP && vt == VDIR) { vp = unionfs_get_cached_vnode(uppervp, lowervp, dvp); if (vp != NULLVP) { *vpp = vp; goto unionfs_nodeget_out; } } unp = malloc(sizeof(struct unionfs_node), M_UNIONFSNODE, M_WAITOK | M_ZERO); error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp); if (error != 0) { free(unp, M_UNIONFSNODE); return (error); } if (dvp != NULLVP) vref(dvp); if (uppervp != NULLVP) vref(uppervp); if (lowervp != NULLVP) vref(lowervp); if (vt == VDIR) { unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH, &hashmask); KASSERT(hashmask == UNIONFSHASHMASK, ("unexpected unionfs hash mask 0x%lx", hashmask)); } unp->un_vnode = vp; unp->un_uppervp = uppervp; unp->un_lowervp = lowervp; unp->un_dvp = dvp; if (uppervp != NULLVP) vp->v_vnlock = uppervp->v_vnlock; else vp->v_vnlock = lowervp->v_vnlock; if (path != NULL) { unp->un_path = malloc(cnp->cn_namelen + 1, M_UNIONFSPATH, M_WAITOK | M_ZERO); bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen); unp->un_path[cnp->cn_namelen] = '\0'; unp->un_pathlen = cnp->cn_namelen; } vp->v_type = vt; vp->v_data = unp; /* * TODO: This is an imperfect check, as there's no guarantee that * the underlying filesystems will always return vnode pointers * for the root inodes that match our cached values. To reduce * the likelihood of failure, for example in the case where either * vnode has been forcibly doomed, we check both pointers and set * VV_ROOT if either matches. */ if (ump->um_uppervp == uppervp || ump->um_lowervp == lowervp) vp->v_vflag |= VV_ROOT; KASSERT(dvp != NULL || (vp->v_vflag & VV_ROOT) != 0, ("%s: NULL dvp for non-root vp %p", __func__, vp)); vn_lock_pair(lowervp, false, uppervp, false); error = insmntque1(vp, mp); if (error != 0) { unionfs_nodeget_cleanup(vp, unp); return (error); } if (lowervp != NULL && VN_IS_DOOMED(lowervp)) { vput(lowervp); unp->un_lowervp = NULL; } if (uppervp != NULL && VN_IS_DOOMED(uppervp)) { vput(uppervp); unp->un_uppervp = NULL; } if (unp->un_lowervp == NULL && unp->un_uppervp == NULL) { unionfs_nodeget_cleanup(vp, unp); return (ENOENT); } if (dvp != NULLVP && vt == VDIR) *vpp = unionfs_ins_cached_vnode(unp, dvp); if (*vpp != NULLVP) { unionfs_nodeget_cleanup(vp, unp); vp = *vpp; } else { if (uppervp != NULL) VOP_UNLOCK(uppervp); if (lowervp != NULL) VOP_UNLOCK(lowervp); *vpp = vp; } unionfs_nodeget_out: if (lkflags & LK_TYPE_MASK) vn_lock(vp, lkflags | LK_RETRY); return (0); } /* * Clean up the unionfs node. */ void unionfs_noderem(struct vnode *vp) { struct unionfs_node *unp, *unp_t1, *unp_t2; struct unionfs_node_hashhead *hd; struct unionfs_node_status *unsp, *unsp_tmp; struct vnode *lvp; struct vnode *uvp; struct vnode *dvp; int count; int writerefs; /* * The root vnode lock may be recursed during unmount, because * it may share the same lock as the unionfs mount's covered vnode, * which is locked across VFS_UNMOUNT(). This lock will then be * recursively taken during the vflush() issued by unionfs_unmount(). * But we still only need to lock the unionfs lock once, because only * one of those lock operations was taken against a unionfs vnode and * will be undone against a unionfs vnode. */ KASSERT(vp->v_vnlock->lk_recurse == 0 || (vp->v_vflag & VV_ROOT) != 0, ("%s: vnode %p locked recursively", __func__, vp)); if (lockmgr(&vp->v_lock, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) panic("%s: failed to acquire lock for vnode lock", __func__); /* * Use the interlock to protect the clearing of v_data to * prevent faults in unionfs_lock(). */ VI_LOCK(vp); unp = VTOUNIONFS(vp); lvp = unp->un_lowervp; uvp = unp->un_uppervp; dvp = unp->un_dvp; unp->un_lowervp = unp->un_uppervp = NULLVP; vp->v_vnlock = &(vp->v_lock); vp->v_data = NULL; vp->v_object = NULL; if (unp->un_hashtbl != NULL) { /* * Clear out any cached child vnodes. This should only * be necessary during forced unmount, when the vnode may * be reclaimed with a non-zero use count. Otherwise the * reference held by each child should prevent reclamation. */ for (count = 0; count <= UNIONFSHASHMASK; count++) { hd = unp->un_hashtbl + count; LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) { LIST_REMOVE(unp_t1, un_hash); unp_t1->un_hash.le_next = NULL; unp_t1->un_hash.le_prev = NULL; } } } VI_UNLOCK(vp); writerefs = atomic_load_int(&vp->v_writecount); VNASSERT(writerefs >= 0, vp, ("%s: write count %d, unexpected text ref", __func__, writerefs)); /* * If we were opened for write, we leased the write reference * to the lower vnode. If this is a reclamation due to the * forced unmount, undo the reference now. */ if (writerefs > 0) { VNASSERT(uvp != NULL, vp, ("%s: write reference without upper vnode", __func__)); VOP_ADD_WRITECOUNT(uvp, -writerefs); } if (lvp != NULLVP) VOP_UNLOCK(lvp); if (uvp != NULLVP) VOP_UNLOCK(uvp); if (dvp != NULLVP) unionfs_rem_cached_vnode(unp, dvp); if (lvp != NULLVP) vrele(lvp); if (uvp != NULLVP) vrele(uvp); if (unp->un_path != NULL) { free(unp->un_path, M_UNIONFSPATH); unp->un_path = NULL; unp->un_pathlen = 0; } if (unp->un_hashtbl != NULL) { hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK); } LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) { LIST_REMOVE(unsp, uns_list); free(unsp, M_TEMP); } if (dvp != NULLVP) { mtx_lock(&unionfs_deferred_rele_lock); STAILQ_INSERT_TAIL(&unionfs_deferred_rele_list, unp, un_rele); mtx_unlock(&unionfs_deferred_rele_lock); taskqueue_enqueue(taskqueue_unionfs_rele, &unionfs_deferred_rele_task); } else free(unp, M_UNIONFSNODE); } /* * Get the unionfs node status object for the vnode corresponding to unp, * for the process that owns td. Allocate a new status object if one * does not already exist. */ void unionfs_get_node_status(struct unionfs_node *unp, struct thread *td, struct unionfs_node_status **unspp) { struct unionfs_node_status *unsp; pid_t pid; pid = td->td_proc->p_pid; KASSERT(NULL != unspp, ("%s: NULL status", __func__)); ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__); LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) { if (unsp->uns_pid == pid) { *unspp = unsp; return; } } /* create a new unionfs node status */ unsp = malloc(sizeof(struct unionfs_node_status), M_TEMP, M_WAITOK | M_ZERO); unsp->uns_pid = pid; LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list); *unspp = unsp; } /* * Remove the unionfs node status, if you can. * You need exclusive lock this vnode. */ void unionfs_tryrem_node_status(struct unionfs_node *unp, struct unionfs_node_status *unsp) { KASSERT(NULL != unsp, ("%s: NULL status", __func__)); ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__); if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt) return; LIST_REMOVE(unsp, uns_list); free(unsp, M_TEMP); } /* * Create upper node attr. */ void unionfs_create_uppervattr_core(struct unionfs_mount *ump, struct vattr *lva, struct vattr *uva, struct thread *td) { VATTR_NULL(uva); uva->va_type = lva->va_type; uva->va_atime = lva->va_atime; uva->va_mtime = lva->va_mtime; uva->va_ctime = lva->va_ctime; switch (ump->um_copymode) { case UNIONFS_TRANSPARENT: uva->va_mode = lva->va_mode; uva->va_uid = lva->va_uid; uva->va_gid = lva->va_gid; break; case UNIONFS_MASQUERADE: if (ump->um_uid == lva->va_uid) { uva->va_mode = lva->va_mode & 077077; uva->va_mode |= (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile) & 0700; uva->va_uid = lva->va_uid; uva->va_gid = lva->va_gid; } else { uva->va_mode = (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile); uva->va_uid = ump->um_uid; uva->va_gid = ump->um_gid; } break; default: /* UNIONFS_TRADITIONAL */ uva->va_mode = 0777 & ~td->td_proc->p_pd->pd_cmask; uva->va_uid = ump->um_uid; uva->va_gid = ump->um_gid; break; } } /* * Create upper node attr. */ int unionfs_create_uppervattr(struct unionfs_mount *ump, struct vnode *lvp, struct vattr *uva, struct ucred *cred, struct thread *td) { struct vattr lva; int error; if ((error = VOP_GETATTR(lvp, &lva, cred))) return (error); unionfs_create_uppervattr_core(ump, &lva, uva, td); return (error); } /* * relookup * * dvp should be locked on entry and will be locked on return. * * If an error is returned, *vpp will be invalid, otherwise it will hold a * locked, referenced vnode. If *vpp == dvp then remember that only one * LK_EXCLUSIVE lock is held. */ int unionfs_relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct componentname *cn, struct thread *td, char *path, int pathlen, u_long nameiop) { int error; cn->cn_namelen = pathlen; cn->cn_pnbuf = path; cn->cn_nameiop = nameiop; cn->cn_flags = (LOCKPARENT | LOCKLEAF | ISLASTCN); cn->cn_lkflags = LK_EXCLUSIVE; cn->cn_cred = cnp->cn_cred; cn->cn_nameptr = cn->cn_pnbuf; if (nameiop == DELETE) cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART)); else if (RENAME == nameiop) cn->cn_flags |= (cnp->cn_flags & SAVESTART); else if (nameiop == CREATE) cn->cn_flags |= NOCACHE; vref(dvp); VOP_UNLOCK(dvp); if ((error = vfs_relookup(dvp, vpp, cn))) { vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); } else vrele(dvp); KASSERT(cn->cn_pnbuf == path, ("%s: cn_pnbuf changed", __func__)); return (error); } /* * relookup for CREATE namei operation. * * dvp is unionfs vnode. dvp should be locked. * * If it called 'unionfs_copyfile' function by unionfs_link etc, * VOP_LOOKUP information is broken. * So it need relookup in order to create link etc. */ int unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp, struct thread *td) { struct vnode *udvp; struct vnode *vp; struct componentname cn; int error; udvp = UNIONFSVPTOUPPERVP(dvp); vp = NULLVP; error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE); if (error) return (error); if (vp != NULLVP) { if (udvp == vp) vrele(vp); else vput(vp); error = EEXIST; } return (error); } /* * relookup for DELETE namei operation. * * dvp is unionfs vnode. dvp should be locked. */ int unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp, struct thread *td) { struct vnode *udvp; struct vnode *vp; struct componentname cn; int error; udvp = UNIONFSVPTOUPPERVP(dvp); vp = NULLVP; error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, DELETE); if (error) return (error); if (vp == NULLVP) error = ENOENT; else { if (udvp == vp) vrele(vp); else vput(vp); } return (error); } /* * relookup for RENAME namei operation. * * dvp is unionfs vnode. dvp should be locked. */ int unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp, struct thread *td) { struct vnode *udvp; struct vnode *vp; struct componentname cn; int error; udvp = UNIONFSVPTOUPPERVP(dvp); vp = NULLVP; error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, RENAME); if (error) return (error); if (vp != NULLVP) { if (udvp == vp) vrele(vp); else vput(vp); } return (error); } /* * Update the unionfs_node. * * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the * uvp's lock and lower's lock will be unlocked. */ static void unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp, struct thread *td) { struct unionfs_node_hashhead *hd; struct vnode *vp; struct vnode *lvp; struct vnode *dvp; unsigned count, lockrec; vp = UNIONFSTOV(unp); lvp = unp->un_lowervp; ASSERT_VOP_ELOCKED(lvp, __func__); ASSERT_VOP_ELOCKED(uvp, __func__); dvp = unp->un_dvp; VNASSERT(vp->v_writecount == 0, vp, ("%s: non-zero writecount", __func__)); /* * Update the upper vnode's lock state to match the lower vnode, * and then switch the unionfs vnode's lock to the upper vnode. */ lockrec = lvp->v_vnlock->lk_recurse; for (count = 0; count < lockrec; count++) vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY); VI_LOCK(vp); unp->un_uppervp = uvp; vp->v_vnlock = uvp->v_vnlock; VI_UNLOCK(vp); /* * Re-cache the unionfs vnode against the upper vnode */ if (dvp != NULLVP && vp->v_type == VDIR) { VI_LOCK(dvp); if (unp->un_hash.le_prev != NULL) { LIST_REMOVE(unp, un_hash); hd = unionfs_get_hashhead(dvp, uvp); LIST_INSERT_HEAD(hd, unp, un_hash); } VI_UNLOCK(unp->un_dvp); } } /* * Create a new shadow dir. * * udvp should be locked on entry and will be locked on return. * * If no error returned, unp will be updated. */ int unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp, struct unionfs_node *unp, struct componentname *cnp, struct thread *td) { struct vnode *lvp; struct vnode *uvp; struct vattr va; struct vattr lva; struct nameidata nd; struct mount *mp; struct ucred *cred; struct ucred *credbk; struct uidinfo *rootinfo; int error; if (unp->un_uppervp != NULLVP) return (EEXIST); lvp = unp->un_lowervp; uvp = NULLVP; credbk = cnp->cn_cred; /* Authority change to root */ rootinfo = uifind((uid_t)0); cred = crdup(cnp->cn_cred); /* * The calls to chgproccnt() are needed to compensate for change_ruid() * calling chgproccnt(). */ chgproccnt(cred->cr_ruidinfo, 1, 0); change_euid(cred, rootinfo); change_ruid(cred, rootinfo); change_svuid(cred, (uid_t)0); uifree(rootinfo); cnp->cn_cred = cred; memset(&nd.ni_cnd, 0, sizeof(struct componentname)); NDPREINIT(&nd); if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred))) goto unionfs_mkshadowdir_abort; if ((error = unionfs_relookup(udvp, &uvp, cnp, &nd.ni_cnd, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE))) goto unionfs_mkshadowdir_abort; if (uvp != NULLVP) { if (udvp == uvp) vrele(uvp); else vput(uvp); error = EEXIST; goto unionfs_mkshadowdir_abort; } - if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH))) + if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH))) goto unionfs_mkshadowdir_abort; unionfs_create_uppervattr_core(ump, &lva, &va, td); error = VOP_MKDIR(udvp, &uvp, &nd.ni_cnd, &va); if (!error) { unionfs_node_update(unp, uvp, td); /* * XXX The bug which cannot set uid/gid was corrected. * Ignore errors. */ va.va_type = VNON; VOP_SETATTR(uvp, &va, nd.ni_cnd.cn_cred); } vn_finished_write(mp); unionfs_mkshadowdir_abort: cnp->cn_cred = credbk; chgproccnt(cred->cr_ruidinfo, -1, 0); crfree(cred); return (error); } /* * Create a new whiteout. * * dvp should be locked on entry and will be locked on return. */ int unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp, struct thread *td, char *path, int pathlen) { struct vnode *wvp; struct nameidata nd; struct mount *mp; int error; wvp = NULLVP; NDPREINIT(&nd); if ((error = unionfs_relookup(dvp, &wvp, cnp, &nd.ni_cnd, td, path, pathlen, CREATE))) { return (error); } if (wvp != NULLVP) { if (dvp == wvp) vrele(wvp); else vput(wvp); return (EEXIST); } - if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH))) + if ((error = vn_start_write(dvp, &mp, V_WAIT | V_PCATCH))) goto unionfs_mkwhiteout_free_out; error = VOP_WHITEOUT(dvp, &nd.ni_cnd, CREATE); vn_finished_write(mp); unionfs_mkwhiteout_free_out: return (error); } /* * Create a new vnode for create a new shadow file. * * If an error is returned, *vpp will be invalid, otherwise it will hold a * locked, referenced and opened vnode. * * unp is never updated. */ static int unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp, struct unionfs_node *unp, struct vattr *uvap, struct thread *td) { struct unionfs_mount *ump; struct vnode *vp; struct vnode *lvp; struct ucred *cred; struct vattr lva; struct nameidata nd; int fmode; int error; ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount); vp = NULLVP; lvp = unp->un_lowervp; cred = td->td_ucred; fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL); error = 0; if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0) return (error); unionfs_create_uppervattr_core(ump, &lva, uvap, td); if (unp->un_path == NULL) panic("%s: NULL un_path", __func__); nd.ni_cnd.cn_namelen = unp->un_pathlen; nd.ni_cnd.cn_pnbuf = unp->un_path; nd.ni_cnd.cn_nameiop = CREATE; nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | ISLASTCN; nd.ni_cnd.cn_lkflags = LK_EXCLUSIVE; nd.ni_cnd.cn_cred = cred; nd.ni_cnd.cn_nameptr = nd.ni_cnd.cn_pnbuf; NDPREINIT(&nd); vref(udvp); if ((error = vfs_relookup(udvp, &vp, &nd.ni_cnd)) != 0) goto unionfs_vn_create_on_upper_free_out2; vrele(udvp); if (vp != NULLVP) { if (vp == udvp) vrele(vp); else vput(vp); error = EEXIST; goto unionfs_vn_create_on_upper_free_out1; } if ((error = VOP_CREATE(udvp, &vp, &nd.ni_cnd, uvap)) != 0) goto unionfs_vn_create_on_upper_free_out1; if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) { vput(vp); goto unionfs_vn_create_on_upper_free_out1; } error = VOP_ADD_WRITECOUNT(vp, 1); CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); if (error == 0) { *vpp = vp; } else { VOP_CLOSE(vp, fmode, cred, td); } unionfs_vn_create_on_upper_free_out1: VOP_UNLOCK(udvp); unionfs_vn_create_on_upper_free_out2: KASSERT(nd.ni_cnd.cn_pnbuf == unp->un_path, ("%s: cn_pnbuf changed", __func__)); return (error); } /* * Copy from lvp to uvp. * * lvp and uvp should be locked and opened on entry and will be locked and * opened on return. */ static int unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp, struct ucred *cred, struct thread *td) { char *buf; struct uio uio; struct iovec iov; off_t offset; int count; int error; int bufoffset; error = 0; memset(&uio, 0, sizeof(uio)); uio.uio_td = td; uio.uio_segflg = UIO_SYSSPACE; uio.uio_offset = 0; buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK); while (error == 0) { offset = uio.uio_offset; uio.uio_iov = &iov; uio.uio_iovcnt = 1; iov.iov_base = buf; iov.iov_len = MAXBSIZE; uio.uio_resid = iov.iov_len; uio.uio_rw = UIO_READ; if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0) break; if ((count = MAXBSIZE - uio.uio_resid) == 0) break; bufoffset = 0; while (bufoffset < count) { uio.uio_iov = &iov; uio.uio_iovcnt = 1; iov.iov_base = buf + bufoffset; iov.iov_len = count - bufoffset; uio.uio_offset = offset + bufoffset; uio.uio_resid = iov.iov_len; uio.uio_rw = UIO_WRITE; if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0) break; bufoffset += (count - bufoffset) - uio.uio_resid; } uio.uio_offset = offset + bufoffset; } free(buf, M_TEMP); return (error); } /* * Copy file from lower to upper. * * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to * docopy. * * If no error returned, unp will be updated. */ int unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred, struct thread *td) { struct mount *mp; struct vnode *udvp; struct vnode *lvp; struct vnode *uvp; struct vattr uva; int error; lvp = unp->un_lowervp; uvp = NULLVP; if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY)) return (EROFS); if (unp->un_dvp == NULLVP) return (EINVAL); if (unp->un_uppervp != NULLVP) return (EEXIST); udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp; if (udvp == NULLVP) return (EROFS); if ((udvp->v_mount->mnt_flag & MNT_RDONLY)) return (EROFS); error = VOP_ACCESS(lvp, VREAD, cred, td); if (error != 0) return (error); - if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)) != 0) + if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)) != 0) return (error); error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td); if (error != 0) { vn_finished_write(mp); return (error); } if (docopy != 0) { error = VOP_OPEN(lvp, FREAD, cred, td, NULL); if (error == 0) { error = unionfs_copyfile_core(lvp, uvp, cred, td); VOP_CLOSE(lvp, FREAD, cred, td); } } VOP_CLOSE(uvp, FWRITE, cred, td); VOP_ADD_WRITECOUNT_CHECKED(uvp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, uvp, uvp->v_writecount); vn_finished_write(mp); if (error == 0) { /* Reset the attributes. Ignore errors. */ uva.va_type = VNON; VOP_SETATTR(uvp, &uva, cred); } unionfs_node_update(unp, uvp, td); return (error); } /* * It checks whether vp can rmdir. (check empty) * * vp is unionfs vnode. * vp should be locked. */ int unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td) { struct vnode *uvp; struct vnode *lvp; struct vnode *tvp; struct dirent *dp; struct dirent *edp; struct componentname cn; struct iovec iov; struct uio uio; struct vattr va; int error; int eofflag; int lookuperr; /* * The size of buf needs to be larger than DIRBLKSIZ. */ char buf[256 * 6]; ASSERT_VOP_ELOCKED(vp, __func__); eofflag = 0; uvp = UNIONFSVPTOUPPERVP(vp); lvp = UNIONFSVPTOLOWERVP(vp); /* check opaque */ if ((error = VOP_GETATTR(uvp, &va, cred)) != 0) return (error); if (va.va_flags & OPAQUE) return (0); /* open vnode */ #ifdef MAC if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0) return (error); #endif if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0) return (error); if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0) return (error); uio.uio_rw = UIO_READ; uio.uio_segflg = UIO_SYSSPACE; uio.uio_td = td; uio.uio_offset = 0; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, lvp); #endif while (!error && !eofflag) { iov.iov_base = buf; iov.iov_len = sizeof(buf); uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_resid = iov.iov_len; error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL); if (error != 0) break; KASSERT(eofflag != 0 || uio.uio_resid < sizeof(buf), ("%s: empty read from lower FS", __func__)); edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid]; for (dp = (struct dirent*)buf; !error && dp < edp; dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) { if (dp->d_type == DT_WHT || dp->d_fileno == 0 || (dp->d_namlen == 1 && dp->d_name[0] == '.') || (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2))) continue; cn.cn_namelen = dp->d_namlen; cn.cn_pnbuf = NULL; cn.cn_nameptr = dp->d_name; cn.cn_nameiop = LOOKUP; cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN; cn.cn_lkflags = LK_EXCLUSIVE; cn.cn_cred = cred; /* * check entry in lower. * Sometimes, readdir function returns * wrong entry. */ lookuperr = VOP_LOOKUP(lvp, &tvp, &cn); if (!lookuperr) vput(tvp); else continue; /* skip entry */ /* * check entry * If it has no exist/whiteout entry in upper, * directory is not empty. */ cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN; lookuperr = VOP_LOOKUP(uvp, &tvp, &cn); if (!lookuperr) vput(tvp); /* ignore exist or whiteout entry */ if (!lookuperr || (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT))) continue; error = ENOTEMPTY; } } /* close vnode */ VOP_CLOSE(vp, FREAD, cred, td); return (error); } diff --git a/sys/kern/uipc_usrreq.c b/sys/kern/uipc_usrreq.c index 1a4d2d5adc0a..00d7aa1fc968 100644 --- a/sys/kern/uipc_usrreq.c +++ b/sys/kern/uipc_usrreq.c @@ -1,3565 +1,3565 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All Rights Reserved. * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved. * Copyright (c) 2018 Matthew Macy * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 */ /* * UNIX Domain (Local) Sockets * * This is an implementation of UNIX (local) domain sockets. Each socket has * an associated struct unpcb (UNIX protocol control block). Stream sockets * may be connected to 0 or 1 other socket. Datagram sockets may be * connected to 0, 1, or many other sockets. Sockets may be created and * connected in pairs (socketpair(2)), or bound/connected to using the file * system name space. For most purposes, only the receive socket buffer is * used, as sending on one socket delivers directly to the receive socket * buffer of a second socket. * * The implementation is substantially complicated by the fact that * "ancillary data", such as file descriptors or credentials, may be passed * across UNIX domain sockets. The potential for passing UNIX domain sockets * over other UNIX domain sockets requires the implementation of a simple * garbage collector to find and tear down cycles of disconnected sockets. * * TODO: * RDM * rethink name space problems * need a proper out-of-band */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include MALLOC_DECLARE(M_FILECAPS); static struct domain localdomain; static uma_zone_t unp_zone; static unp_gen_t unp_gencnt; /* (l) */ static u_int unp_count; /* (l) Count of local sockets. */ static ino_t unp_ino; /* Prototype for fake inode numbers. */ static int unp_rights; /* (g) File descriptors in flight. */ static struct unp_head unp_shead; /* (l) List of stream sockets. */ static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ struct unp_defer { SLIST_ENTRY(unp_defer) ud_link; struct file *ud_fp; }; static SLIST_HEAD(, unp_defer) unp_defers; static int unp_defers_count; static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; /* * Garbage collection of cyclic file descriptor/socket references occurs * asynchronously in a taskqueue context in order to avoid recursion and * reentrance in the UNIX domain socket, file descriptor, and socket layer * code. See unp_gc() for a full description. */ static struct timeout_task unp_gc_task; /* * The close of unix domain sockets attached as SCM_RIGHTS is * postponed to the taskqueue, to avoid arbitrary recursion depth. * The attached sockets might have another sockets attached. */ static struct task unp_defer_task; /* * Both send and receive buffers are allocated PIPSIZ bytes of buffering for * stream sockets, although the total for sender and receiver is actually * only PIPSIZ. * * Datagram sockets really use the sendspace as the maximum datagram size, * and don't really want to reserve the sendspace. Their recvspace should be * large enough for at least one max-size datagram plus address. */ #ifndef PIPSIZ #define PIPSIZ 8192 #endif static u_long unpst_sendspace = PIPSIZ; static u_long unpst_recvspace = PIPSIZ; static u_long unpdg_maxdgram = 2*1024; static u_long unpdg_recvspace = 16*1024; /* support 8KB syslog msgs */ static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ static u_long unpsp_recvspace = PIPSIZ; static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Local domain"); static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "SOCK_STREAM"); static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "SOCK_DGRAM"); static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "SOCK_SEQPACKET"); SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, &unpst_sendspace, 0, "Default stream send space."); SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, &unpst_recvspace, 0, "Default stream receive space."); SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, &unpdg_maxdgram, 0, "Maximum datagram size."); SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, &unpdg_recvspace, 0, "Default datagram receive space."); SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, &unpsp_sendspace, 0, "Default seqpacket send space."); SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, &unpsp_recvspace, 0, "Default seqpacket receive space."); SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "File descriptors in flight."); SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, &unp_defers_count, 0, "File descriptors deferred to taskqueue for close."); /* * Locking and synchronization: * * Several types of locks exist in the local domain socket implementation: * - a global linkage lock * - a global connection list lock * - the mtxpool lock * - per-unpcb mutexes * * The linkage lock protects the global socket lists, the generation number * counter and garbage collector state. * * The connection list lock protects the list of referring sockets in a datagram * socket PCB. This lock is also overloaded to protect a global list of * sockets whose buffers contain socket references in the form of SCM_RIGHTS * messages. To avoid recursion, such references are released by a dedicated * thread. * * The mtxpool lock protects the vnode from being modified while referenced. * Lock ordering rules require that it be acquired before any PCB locks. * * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the * unpcb. This includes the unp_conn field, which either links two connected * PCBs together (for connected socket types) or points at the destination * socket (for connectionless socket types). The operations of creating or * destroying a connection therefore involve locking multiple PCBs. To avoid * lock order reversals, in some cases this involves dropping a PCB lock and * using a reference counter to maintain liveness. * * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, * allocated in pr_attach() and freed in pr_detach(). The validity of that * pointer is an invariant, so no lock is required to dereference the so_pcb * pointer if a valid socket reference is held by the caller. In practice, * this is always true during operations performed on a socket. Each unpcb * has a back-pointer to its socket, unp_socket, which will be stable under * the same circumstances. * * This pointer may only be safely dereferenced as long as a valid reference * to the unpcb is held. Typically, this reference will be from the socket, * or from another unpcb when the referring unpcb's lock is held (in order * that the reference not be invalidated during use). For example, to follow * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee * that detach is not run clearing unp_socket. * * Blocking with UNIX domain sockets is a tricky issue: unlike most network * protocols, bind() is a non-atomic operation, and connect() requires * potential sleeping in the protocol, due to potentially waiting on local or * distributed file systems. We try to separate "lookup" operations, which * may sleep, and the IPC operations themselves, which typically can occur * with relative atomicity as locks can be held over the entire operation. * * Another tricky issue is simultaneous multi-threaded or multi-process * access to a single UNIX domain socket. These are handled by the flags * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or * binding, both of which involve dropping UNIX domain socket locks in order * to perform namei() and other file system operations. */ static struct rwlock unp_link_rwlock; static struct mtx unp_defers_lock; #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ "unp_link_rwlock") #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_LOCKED) #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_UNLOCKED) #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_WLOCKED) #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ "unp_defer", NULL, MTX_DEF) #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ "unp", "unp", \ MTX_DUPOK|MTX_DEF) #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx) #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) static int uipc_connect2(struct socket *, struct socket *); static int uipc_ctloutput(struct socket *, struct sockopt *); static int unp_connect(struct socket *, struct sockaddr *, struct thread *); static int unp_connectat(int, struct socket *, struct sockaddr *, struct thread *, bool); typedef enum { PRU_CONNECT, PRU_CONNECT2 } conn2_how; static void unp_connect2(struct socket *so, struct socket *so2, conn2_how); static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); static void unp_dispose(struct socket *so); static void unp_shutdown(struct unpcb *); static void unp_drop(struct unpcb *); static void unp_gc(__unused void *, int); static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); static void unp_discard(struct file *); static void unp_freerights(struct filedescent **, int); static int unp_internalize(struct mbuf **, struct thread *, struct mbuf **, u_int *, u_int *); static void unp_internalize_fp(struct file *); static int unp_externalize(struct mbuf *, struct mbuf **, int); static int unp_externalize_fp(struct file *); static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, int, struct mbuf **, u_int *, u_int *); static void unp_process_defers(void * __unused, int); static void unp_pcb_hold(struct unpcb *unp) { u_int old __unused; old = refcount_acquire(&unp->unp_refcount); KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); } static __result_use_check bool unp_pcb_rele(struct unpcb *unp) { bool ret; UNP_PCB_LOCK_ASSERT(unp); if ((ret = refcount_release(&unp->unp_refcount))) { UNP_PCB_UNLOCK(unp); UNP_PCB_LOCK_DESTROY(unp); uma_zfree(unp_zone, unp); } return (ret); } static void unp_pcb_rele_notlast(struct unpcb *unp) { bool ret __unused; ret = refcount_release(&unp->unp_refcount); KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); } static void unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) { UNP_PCB_UNLOCK_ASSERT(unp); UNP_PCB_UNLOCK_ASSERT(unp2); if (unp == unp2) { UNP_PCB_LOCK(unp); } else if ((uintptr_t)unp2 > (uintptr_t)unp) { UNP_PCB_LOCK(unp); UNP_PCB_LOCK(unp2); } else { UNP_PCB_LOCK(unp2); UNP_PCB_LOCK(unp); } } static void unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) { UNP_PCB_UNLOCK(unp); if (unp != unp2) UNP_PCB_UNLOCK(unp2); } /* * Try to lock the connected peer of an already locked socket. In some cases * this requires that we unlock the current socket. The pairbusy counter is * used to block concurrent connection attempts while the lock is dropped. The * caller must be careful to revalidate PCB state. */ static struct unpcb * unp_pcb_lock_peer(struct unpcb *unp) { struct unpcb *unp2; UNP_PCB_LOCK_ASSERT(unp); unp2 = unp->unp_conn; if (unp2 == NULL) return (NULL); if (__predict_false(unp == unp2)) return (unp); UNP_PCB_UNLOCK_ASSERT(unp2); if (__predict_true(UNP_PCB_TRYLOCK(unp2))) return (unp2); if ((uintptr_t)unp2 > (uintptr_t)unp) { UNP_PCB_LOCK(unp2); return (unp2); } unp->unp_pairbusy++; unp_pcb_hold(unp2); UNP_PCB_UNLOCK(unp); UNP_PCB_LOCK(unp2); UNP_PCB_LOCK(unp); KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, ("%s: socket %p was reconnected", __func__, unp)); if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { unp->unp_flags &= ~UNP_WAITING; wakeup(unp); } if (unp_pcb_rele(unp2)) { /* unp2 is unlocked. */ return (NULL); } if (unp->unp_conn == NULL) { UNP_PCB_UNLOCK(unp2); return (NULL); } return (unp2); } static void uipc_abort(struct socket *so) { struct unpcb *unp, *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); UNP_PCB_UNLOCK_ASSERT(unp); UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 != NULL) { unp_pcb_hold(unp2); UNP_PCB_UNLOCK(unp); unp_drop(unp2); } else UNP_PCB_UNLOCK(unp); } static int uipc_accept(struct socket *so, struct sockaddr **nam) { struct unpcb *unp, *unp2; const struct sockaddr *sa; /* * Pass back name of connected socket, if it was bound and we are * still connected (our peer may have closed already!). */ unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_PCB_LOCK(unp); unp2 = unp_pcb_lock_peer(unp); if (unp2 != NULL && unp2->unp_addr != NULL) sa = (struct sockaddr *)unp2->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); if (unp2 != NULL) unp_pcb_unlock_pair(unp, unp2); else UNP_PCB_UNLOCK(unp); return (0); } static int uipc_attach(struct socket *so, int proto, struct thread *td) { u_long sendspace, recvspace; struct unpcb *unp; int error; bool locked; KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { switch (so->so_type) { case SOCK_STREAM: sendspace = unpst_sendspace; recvspace = unpst_recvspace; break; case SOCK_DGRAM: STAILQ_INIT(&so->so_rcv.uxdg_mb); STAILQ_INIT(&so->so_snd.uxdg_mb); TAILQ_INIT(&so->so_rcv.uxdg_conns); /* * Since send buffer is either bypassed or is a part * of one-to-many receive buffer, we assign both space * limits to unpdg_recvspace. */ sendspace = recvspace = unpdg_recvspace; break; case SOCK_SEQPACKET: sendspace = unpsp_sendspace; recvspace = unpsp_recvspace; break; default: panic("uipc_attach"); } error = soreserve(so, sendspace, recvspace); if (error) return (error); } unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); if (unp == NULL) return (ENOBUFS); LIST_INIT(&unp->unp_refs); UNP_PCB_LOCK_INIT(unp); unp->unp_socket = so; so->so_pcb = unp; refcount_init(&unp->unp_refcount, 1); if ((locked = UNP_LINK_WOWNED()) == false) UNP_LINK_WLOCK(); unp->unp_gencnt = ++unp_gencnt; unp->unp_ino = ++unp_ino; unp_count++; switch (so->so_type) { case SOCK_STREAM: LIST_INSERT_HEAD(&unp_shead, unp, unp_link); break; case SOCK_DGRAM: LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); break; case SOCK_SEQPACKET: LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); break; default: panic("uipc_attach"); } if (locked == false) UNP_LINK_WUNLOCK(); return (0); } static int uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { struct sockaddr_un *soun = (struct sockaddr_un *)nam; struct vattr vattr; int error, namelen; struct nameidata nd; struct unpcb *unp; struct vnode *vp; struct mount *mp; cap_rights_t rights; char *buf; if (nam->sa_family != AF_UNIX) return (EAFNOSUPPORT); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); if (soun->sun_len > sizeof(struct sockaddr_un)) return (EINVAL); namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); if (namelen <= 0) return (EINVAL); /* * We don't allow simultaneous bind() calls on a single UNIX domain * socket, so flag in-progress operations, and return an error if an * operation is already in progress. * * Historically, we have not allowed a socket to be rebound, so this * also returns an error. Not allowing re-binding simplifies the * implementation and avoids a great many possible failure modes. */ UNP_PCB_LOCK(unp); if (unp->unp_vnode != NULL) { UNP_PCB_UNLOCK(unp); return (EINVAL); } if (unp->unp_flags & UNP_BINDING) { UNP_PCB_UNLOCK(unp); return (EALREADY); } unp->unp_flags |= UNP_BINDING; UNP_PCB_UNLOCK(unp); buf = malloc(namelen + 1, M_TEMP, M_WAITOK); bcopy(soun->sun_path, buf, namelen); buf[namelen] = 0; restart: NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | NOCACHE, UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT)); /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ error = namei(&nd); if (error) goto error; vp = nd.ni_vp; if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE_PNBUF(&nd); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if (vp != NULL) { vrele(vp); error = EADDRINUSE; goto error; } - error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); + error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH); if (error) goto error; goto restart; } VATTR_NULL(&vattr); vattr.va_type = VSOCK; vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask); #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); #endif if (error == 0) error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); NDFREE_PNBUF(&nd); if (error) { VOP_VPUT_PAIR(nd.ni_dvp, NULL, true); vn_finished_write(mp); if (error == ERELOOKUP) goto restart; goto error; } vp = nd.ni_vp; ASSERT_VOP_ELOCKED(vp, "uipc_bind"); soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); UNP_PCB_LOCK(unp); VOP_UNP_BIND(vp, unp); unp->unp_vnode = vp; unp->unp_addr = soun; unp->unp_flags &= ~UNP_BINDING; UNP_PCB_UNLOCK(unp); vref(vp); VOP_VPUT_PAIR(nd.ni_dvp, &vp, true); vn_finished_write(mp); free(buf, M_TEMP); return (0); error: UNP_PCB_LOCK(unp); unp->unp_flags &= ~UNP_BINDING; UNP_PCB_UNLOCK(unp); free(buf, M_TEMP); return (error); } static int uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) { return (uipc_bindat(AT_FDCWD, so, nam, td)); } static int uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { int error; KASSERT(td == curthread, ("uipc_connect: td != curthread")); error = unp_connect(so, nam, td); return (error); } static int uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { int error; KASSERT(td == curthread, ("uipc_connectat: td != curthread")); error = unp_connectat(fd, so, nam, td, false); return (error); } static void uipc_close(struct socket *so) { struct unpcb *unp, *unp2; struct vnode *vp = NULL; struct mtx *vplock; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_close: unp == NULL")); vplock = NULL; if ((vp = unp->unp_vnode) != NULL) { vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); } UNP_PCB_LOCK(unp); if (vp && unp->unp_vnode == NULL) { mtx_unlock(vplock); vp = NULL; } if (vp != NULL) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; } if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) unp_disconnect(unp, unp2); else UNP_PCB_UNLOCK(unp); if (vp) { mtx_unlock(vplock); vrele(vp); } } static int uipc_connect2(struct socket *so1, struct socket *so2) { struct unpcb *unp, *unp2; if (so1->so_type != so2->so_type) return (EPROTOTYPE); unp = so1->so_pcb; KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); unp2 = so2->so_pcb; KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); unp_pcb_lock_pair(unp, unp2); unp_connect2(so1, so2, PRU_CONNECT2); unp_pcb_unlock_pair(unp, unp2); return (0); } static void uipc_detach(struct socket *so) { struct unpcb *unp, *unp2; struct mtx *vplock; struct vnode *vp; int local_unp_rights; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); vp = NULL; vplock = NULL; UNP_LINK_WLOCK(); LIST_REMOVE(unp, unp_link); if (unp->unp_gcflag & UNPGC_DEAD) LIST_REMOVE(unp, unp_dead); unp->unp_gencnt = ++unp_gencnt; --unp_count; UNP_LINK_WUNLOCK(); UNP_PCB_UNLOCK_ASSERT(unp); restart: if ((vp = unp->unp_vnode) != NULL) { vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); } UNP_PCB_LOCK(unp); if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { if (vplock) mtx_unlock(vplock); UNP_PCB_UNLOCK(unp); goto restart; } if ((vp = unp->unp_vnode) != NULL) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; } if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) unp_disconnect(unp, unp2); else UNP_PCB_UNLOCK(unp); UNP_REF_LIST_LOCK(); while (!LIST_EMPTY(&unp->unp_refs)) { struct unpcb *ref = LIST_FIRST(&unp->unp_refs); unp_pcb_hold(ref); UNP_REF_LIST_UNLOCK(); MPASS(ref != unp); UNP_PCB_UNLOCK_ASSERT(ref); unp_drop(ref); UNP_REF_LIST_LOCK(); } UNP_REF_LIST_UNLOCK(); UNP_PCB_LOCK(unp); local_unp_rights = unp_rights; unp->unp_socket->so_pcb = NULL; unp->unp_socket = NULL; free(unp->unp_addr, M_SONAME); unp->unp_addr = NULL; if (!unp_pcb_rele(unp)) UNP_PCB_UNLOCK(unp); if (vp) { mtx_unlock(vplock); vrele(vp); } if (local_unp_rights) taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); switch (so->so_type) { case SOCK_DGRAM: /* * Everything should have been unlinked/freed by unp_dispose() * and/or unp_disconnect(). */ MPASS(so->so_rcv.uxdg_peeked == NULL); MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb)); MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns)); MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb)); } } static int uipc_disconnect(struct socket *so) { struct unpcb *unp, *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); UNP_PCB_LOCK(unp); if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) unp_disconnect(unp, unp2); else UNP_PCB_UNLOCK(unp); return (0); } static int uipc_listen(struct socket *so, int backlog, struct thread *td) { struct unpcb *unp; int error; MPASS(so->so_type != SOCK_DGRAM); /* * Synchronize with concurrent connection attempts. */ error = 0; unp = sotounpcb(so); UNP_PCB_LOCK(unp); if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0) error = EINVAL; else if (unp->unp_vnode == NULL) error = EDESTADDRREQ; if (error != 0) { UNP_PCB_UNLOCK(unp); return (error); } SOCK_LOCK(so); error = solisten_proto_check(so); if (error == 0) { cru2xt(td, &unp->unp_peercred); solisten_proto(so, backlog); } SOCK_UNLOCK(so); UNP_PCB_UNLOCK(unp); return (error); } static int uipc_peeraddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp, *unp2; const struct sockaddr *sa; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_LINK_RLOCK(); /* * XXX: It seems that this test always fails even when connection is * established. So, this else clause is added as workaround to * return PF_LOCAL sockaddr. */ unp2 = unp->unp_conn; if (unp2 != NULL) { UNP_PCB_LOCK(unp2); if (unp2->unp_addr != NULL) sa = (struct sockaddr *) unp2->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); UNP_PCB_UNLOCK(unp2); } else { sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); } UNP_LINK_RUNLOCK(); return (0); } static int uipc_rcvd(struct socket *so, int flags) { struct unpcb *unp, *unp2; struct socket *so2; u_int mbcnt, sbcc; unp = sotounpcb(so); KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, ("%s: socktype %d", __func__, so->so_type)); /* * Adjust backpressure on sender and wakeup any waiting to write. * * The unp lock is acquired to maintain the validity of the unp_conn * pointer; no lock on unp2 is required as unp2->unp_socket will be * static as long as we don't permit unp2 to disconnect from unp, * which is prevented by the lock on unp. We cache values from * so_rcv to avoid holding the so_rcv lock over the entire * transaction on the remote so_snd. */ SOCKBUF_LOCK(&so->so_rcv); mbcnt = so->so_rcv.sb_mbcnt; sbcc = sbavail(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); /* * There is a benign race condition at this point. If we're planning to * clear SB_STOP, but uipc_send is called on the connected socket at * this instant, it might add data to the sockbuf and set SB_STOP. Then * we would erroneously clear SB_STOP below, even though the sockbuf is * full. The race is benign because the only ill effect is to allow the * sockbuf to exceed its size limit, and the size limits are not * strictly guaranteed anyway. */ UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 == NULL) { UNP_PCB_UNLOCK(unp); return (0); } so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_snd); if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) so2->so_snd.sb_flags &= ~SB_STOP; sowwakeup_locked(so2); UNP_PCB_UNLOCK(unp); return (0); } static int uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, struct mbuf *control, struct thread *td) { struct unpcb *unp, *unp2; struct socket *so2; u_int mbcnt, sbcc; int error; unp = sotounpcb(so); KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, ("%s: socktype %d", __func__, so->so_type)); error = 0; if (flags & PRUS_OOB) { error = EOPNOTSUPP; goto release; } if (control != NULL && (error = unp_internalize(&control, td, NULL, NULL, NULL))) goto release; unp2 = NULL; if ((so->so_state & SS_ISCONNECTED) == 0) { if (nam != NULL) { if ((error = unp_connect(so, nam, td)) != 0) goto out; } else { error = ENOTCONN; goto out; } } UNP_PCB_LOCK(unp); if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { UNP_PCB_UNLOCK(unp); error = ENOTCONN; goto out; } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { unp_pcb_unlock_pair(unp, unp2); error = EPIPE; goto out; } UNP_PCB_UNLOCK(unp); if ((so2 = unp2->unp_socket) == NULL) { UNP_PCB_UNLOCK(unp2); error = ENOTCONN; goto out; } SOCKBUF_LOCK(&so2->so_rcv); if (unp2->unp_flags & UNP_WANTCRED_MASK) { /* * Credentials are passed only once on SOCK_STREAM and * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). */ control = unp_addsockcred(td, control, unp2->unp_flags, NULL, NULL, NULL); unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; } /* * Send to paired receive port and wake up readers. Don't * check for space available in the receive buffer if we're * attaching ancillary data; Unix domain sockets only check * for space in the sending sockbuf, and that check is * performed one level up the stack. At that level we cannot * precisely account for the amount of buffer space used * (e.g., because control messages are not yet internalized). */ switch (so->so_type) { case SOCK_STREAM: if (control != NULL) { sbappendcontrol_locked(&so2->so_rcv, m, control, flags); control = NULL; } else sbappend_locked(&so2->so_rcv, m, flags); break; case SOCK_SEQPACKET: if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, &sun_noname, m, control)) control = NULL; break; } mbcnt = so2->so_rcv.sb_mbcnt; sbcc = sbavail(&so2->so_rcv); if (sbcc) sorwakeup_locked(so2); else SOCKBUF_UNLOCK(&so2->so_rcv); /* * The PCB lock on unp2 protects the SB_STOP flag. Without it, * it would be possible for uipc_rcvd to be called at this * point, drain the receiving sockbuf, clear SB_STOP, and then * we would set SB_STOP below. That could lead to an empty * sockbuf having SB_STOP set */ SOCKBUF_LOCK(&so->so_snd); if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) so->so_snd.sb_flags |= SB_STOP; SOCKBUF_UNLOCK(&so->so_snd); UNP_PCB_UNLOCK(unp2); m = NULL; out: /* * PRUS_EOF is equivalent to pr_send followed by pr_shutdown. */ if (flags & PRUS_EOF) { UNP_PCB_LOCK(unp); socantsendmore(so); unp_shutdown(unp); UNP_PCB_UNLOCK(unp); } if (control != NULL && error != 0) unp_scan(control, unp_freerights); release: if (control != NULL) m_freem(control); /* * In case of PRUS_NOTREADY, uipc_ready() is responsible * for freeing memory. */ if (m != NULL && (flags & PRUS_NOTREADY) == 0) m_freem(m); return (error); } /* PF_UNIX/SOCK_DGRAM version of sbspace() */ static inline bool uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt) { u_int bleft, mleft; /* * Negative space may happen if send(2) is followed by * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum. */ if (__predict_false(sb->sb_hiwat < sb->uxdg_cc || sb->sb_mbmax < sb->uxdg_mbcnt)) return (false); if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) return (false); bleft = sb->sb_hiwat - sb->uxdg_cc; mleft = sb->sb_mbmax - sb->uxdg_mbcnt; return (bleft >= cc && mleft >= mbcnt); } /* * PF_UNIX/SOCK_DGRAM send * * Allocate a record consisting of 3 mbufs in the sequence of * from -> control -> data and append it to the socket buffer. * * The first mbuf carries sender's name and is a pkthdr that stores * overall length of datagram, its memory consumption and control length. */ #define ctllen PH_loc.thirtytwo[1] _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <= offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen"); static int uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *m, struct mbuf *c, int flags, struct thread *td) { struct unpcb *unp, *unp2; const struct sockaddr *from; struct socket *so2; struct sockbuf *sb; struct mbuf *f, *clast; u_int cc, ctl, mbcnt; u_int dcc __diagused, dctl __diagused, dmbcnt __diagused; int error; MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL)); error = 0; f = NULL; ctl = 0; if (__predict_false(flags & MSG_OOB)) { error = EOPNOTSUPP; goto out; } if (m == NULL) { if (__predict_false(uio->uio_resid > unpdg_maxdgram)) { error = EMSGSIZE; goto out; } m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR); if (__predict_false(m == NULL)) { error = EFAULT; goto out; } f = m_gethdr(M_WAITOK, MT_SONAME); cc = m->m_pkthdr.len; mbcnt = MSIZE + m->m_pkthdr.memlen; if (c != NULL && (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt))) goto out; } else { /* pr_sosend() with mbuf usually is a kernel thread. */ M_ASSERTPKTHDR(m); if (__predict_false(c != NULL)) panic("%s: control from a kernel thread", __func__); if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) { error = EMSGSIZE; goto out; } if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) { error = ENOBUFS; goto out; } /* Condition the foreign mbuf to our standards. */ m_clrprotoflags(m); m_tag_delete_chain(m, NULL); m->m_pkthdr.rcvif = NULL; m->m_pkthdr.flowid = 0; m->m_pkthdr.csum_flags = 0; m->m_pkthdr.fibnum = 0; m->m_pkthdr.rsstype = 0; cc = m->m_pkthdr.len; mbcnt = MSIZE; for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) { mbcnt += MSIZE; if (mb->m_flags & M_EXT) mbcnt += mb->m_ext.ext_size; } } unp = sotounpcb(so); MPASS(unp); /* * XXXGL: would be cool to fully remove so_snd out of the equation * and avoid this lock, which is not only extraneous, but also being * released, thus still leaving possibility for a race. We can easily * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it * is more difficult to invent something to handle so_error. */ error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); if (error) goto out2; SOCK_SENDBUF_LOCK(so); if (so->so_snd.sb_state & SBS_CANTSENDMORE) { SOCK_SENDBUF_UNLOCK(so); error = EPIPE; goto out3; } if (so->so_error != 0) { error = so->so_error; so->so_error = 0; SOCK_SENDBUF_UNLOCK(so); goto out3; } if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) { SOCK_SENDBUF_UNLOCK(so); error = EDESTADDRREQ; goto out3; } SOCK_SENDBUF_UNLOCK(so); if (addr != NULL) { if ((error = unp_connectat(AT_FDCWD, so, addr, td, true))) goto out3; UNP_PCB_LOCK_ASSERT(unp); unp2 = unp->unp_conn; UNP_PCB_LOCK_ASSERT(unp2); } else { UNP_PCB_LOCK(unp); unp2 = unp_pcb_lock_peer(unp); if (unp2 == NULL) { UNP_PCB_UNLOCK(unp); error = ENOTCONN; goto out3; } } if (unp2->unp_flags & UNP_WANTCRED_MASK) c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl, &mbcnt); if (unp->unp_addr != NULL) from = (struct sockaddr *)unp->unp_addr; else from = &sun_noname; f->m_len = from->sa_len; MPASS(from->sa_len <= MLEN); bcopy(from, mtod(f, void *), from->sa_len); ctl += f->m_len; /* * Concatenate mbufs: from -> control -> data. * Save overall cc and mbcnt in "from" mbuf. */ if (c != NULL) { #ifdef INVARIANTS struct mbuf *mc; for (mc = c; mc->m_next != NULL; mc = mc->m_next); MPASS(mc == clast); #endif f->m_next = c; clast->m_next = m; c = NULL; } else f->m_next = m; m = NULL; #ifdef INVARIANTS dcc = dctl = dmbcnt = 0; for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) { if (mb->m_type == MT_DATA) dcc += mb->m_len; else dctl += mb->m_len; dmbcnt += MSIZE; if (mb->m_flags & M_EXT) dmbcnt += mb->m_ext.ext_size; } MPASS(dcc == cc); MPASS(dctl == ctl); MPASS(dmbcnt == mbcnt); #endif f->m_pkthdr.len = cc + ctl; f->m_pkthdr.memlen = mbcnt; f->m_pkthdr.ctllen = ctl; /* * Destination socket buffer selection. * * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the * destination address is supplied, create a temporary connection for * the run time of the function (see call to unp_connectat() above and * to unp_disconnect() below). We distinguish them by condition of * (addr != NULL). We intentionally avoid adding 'bool connected' for * that condition, since, again, through the run time of this code we * are always connected. For such "unconnected" sends, the destination * buffer would be the receive buffer of destination socket so2. * * For connected sends, data lands on the send buffer of the sender's * socket "so". Then, if we just added the very first datagram * on this send buffer, we need to add the send buffer on to the * receiving socket's buffer list. We put ourselves on top of the * list. Such logic gives infrequent senders priority over frequent * senders. * * Note on byte count management. As long as event methods kevent(2), * select(2) are not protocol specific (yet), we need to maintain * meaningful values on the receive buffer. So, the receive buffer * would accumulate counters from all connected buffers potentially * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax. */ so2 = unp2->unp_socket; sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv; SOCK_RECVBUF_LOCK(so2); if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) { if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb)) TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd, uxdg_clist); STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt); sb->uxdg_cc += cc + ctl; sb->uxdg_ctl += ctl; sb->uxdg_mbcnt += mbcnt; so2->so_rcv.sb_acc += cc + ctl; so2->so_rcv.sb_ccc += cc + ctl; so2->so_rcv.sb_ctl += ctl; so2->so_rcv.sb_mbcnt += mbcnt; sorwakeup_locked(so2); f = NULL; } else { soroverflow_locked(so2); error = (so->so_state & SS_NBIO) ? EAGAIN : ENOBUFS; } if (addr != NULL) unp_disconnect(unp, unp2); else unp_pcb_unlock_pair(unp, unp2); td->td_ru.ru_msgsnd++; out3: SOCK_IO_SEND_UNLOCK(so); out2: if (c) unp_scan(c, unp_freerights); out: if (f) m_freem(f); if (c) m_freem(c); if (m) m_freem(m); return (error); } /* * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. * The mbuf has already been unlinked from the uxdg_mb of socket buffer * and needs to be linked onto uxdg_peeked of receive socket buffer. */ static int uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, struct uio *uio, struct mbuf **controlp, int *flagsp) { ssize_t len = 0; int error; so->so_rcv.uxdg_peeked = m; so->so_rcv.uxdg_cc += m->m_pkthdr.len; so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; SOCK_RECVBUF_UNLOCK(so); KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); if (psa != NULL) *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); m = m->m_next; KASSERT(m, ("%s: no data or control after soname", __func__)); /* * With MSG_PEEK the control isn't executed, just copied. */ while (m != NULL && m->m_type == MT_CONTROL) { if (controlp != NULL) { *controlp = m_copym(m, 0, m->m_len, M_WAITOK); controlp = &(*controlp)->m_next; } m = m->m_next; } KASSERT(m == NULL || m->m_type == MT_DATA, ("%s: not MT_DATA mbuf %p", __func__, m)); while (m != NULL && uio->uio_resid > 0) { len = uio->uio_resid; if (len > m->m_len) len = m->m_len; error = uiomove(mtod(m, char *), (int)len, uio); if (error) { SOCK_IO_RECV_UNLOCK(so); return (error); } if (len == m->m_len) m = m->m_next; } SOCK_IO_RECV_UNLOCK(so); if (flagsp != NULL) { if (m != NULL) { if (*flagsp & MSG_TRUNC) { /* Report real length of the packet */ uio->uio_resid -= m_length(m, NULL) - len; } *flagsp |= MSG_TRUNC; } else *flagsp &= ~MSG_TRUNC; } return (0); } /* * PF_UNIX/SOCK_DGRAM receive */ static int uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { struct sockbuf *sb = NULL; struct mbuf *m; int flags, error; ssize_t len = 0; bool nonblock; MPASS(mp0 == NULL); if (psa != NULL) *psa = NULL; if (controlp != NULL) *controlp = NULL; flags = flagsp != NULL ? *flagsp : 0; nonblock = (so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT | MSG_NBIO)); error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); if (__predict_false(error)) return (error); /* * Loop blocking while waiting for a datagram. Prioritize connected * peers over unconnected sends. Set sb to selected socket buffer * containing an mbuf on exit from the wait loop. A datagram that * had already been peeked at has top priority. */ SOCK_RECVBUF_LOCK(so); while ((m = so->so_rcv.uxdg_peeked) == NULL && (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { if (so->so_error) { error = so->so_error; so->so_error = 0; SOCK_RECVBUF_UNLOCK(so); SOCK_IO_RECV_UNLOCK(so); return (error); } if (so->so_rcv.sb_state & SBS_CANTRCVMORE || uio->uio_resid == 0) { SOCK_RECVBUF_UNLOCK(so); SOCK_IO_RECV_UNLOCK(so); return (0); } if (nonblock) { SOCK_RECVBUF_UNLOCK(so); SOCK_IO_RECV_UNLOCK(so); return (EWOULDBLOCK); } error = sbwait(so, SO_RCV); if (error) { SOCK_RECVBUF_UNLOCK(so); SOCK_IO_RECV_UNLOCK(so); return (error); } } if (sb == NULL) sb = &so->so_rcv; else if (m == NULL) m = STAILQ_FIRST(&sb->uxdg_mb); else MPASS(m == so->so_rcv.uxdg_peeked); MPASS(sb->uxdg_cc > 0); M_ASSERTPKTHDR(m); KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); if (uio->uio_td) uio->uio_td->td_ru.ru_msgrcv++; if (__predict_true(m != so->so_rcv.uxdg_peeked)) { STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); } else so->so_rcv.uxdg_peeked = NULL; sb->uxdg_cc -= m->m_pkthdr.len; sb->uxdg_ctl -= m->m_pkthdr.ctllen; sb->uxdg_mbcnt -= m->m_pkthdr.memlen; if (__predict_false(flags & MSG_PEEK)) return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); so->so_rcv.sb_acc -= m->m_pkthdr.len; so->so_rcv.sb_ccc -= m->m_pkthdr.len; so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; SOCK_RECVBUF_UNLOCK(so); if (psa != NULL) *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); m = m_free(m); KASSERT(m, ("%s: no data or control after soname", __func__)); /* * Packet to copyout() is now in 'm' and it is disconnected from the * queue. * * Process one or more MT_CONTROL mbufs present before any data mbufs * in the first mbuf chain on the socket buffer. We call into the * unp_externalize() to perform externalization (or freeing if * controlp == NULL). In some cases there can be only MT_CONTROL mbufs * without MT_DATA mbufs. */ while (m != NULL && m->m_type == MT_CONTROL) { struct mbuf *cm; /* XXXGL: unp_externalize() is also dom_externalize() KBI and * it frees whole chain, so we must disconnect the mbuf. */ cm = m; m = m->m_next; cm->m_next = NULL; error = unp_externalize(cm, controlp, flags); if (error != 0) { SOCK_IO_RECV_UNLOCK(so); unp_scan(m, unp_freerights); m_freem(m); return (error); } if (controlp != NULL) { while (*controlp != NULL) controlp = &(*controlp)->m_next; } } KASSERT(m == NULL || m->m_type == MT_DATA, ("%s: not MT_DATA mbuf %p", __func__, m)); while (m != NULL && uio->uio_resid > 0) { len = uio->uio_resid; if (len > m->m_len) len = m->m_len; error = uiomove(mtod(m, char *), (int)len, uio); if (error) { SOCK_IO_RECV_UNLOCK(so); m_freem(m); return (error); } if (len == m->m_len) m = m_free(m); else { m->m_data += len; m->m_len -= len; } } SOCK_IO_RECV_UNLOCK(so); if (m != NULL) { if (flagsp != NULL) { if (flags & MSG_TRUNC) { /* Report real length of the packet */ uio->uio_resid -= m_length(m, NULL); } *flagsp |= MSG_TRUNC; } m_freem(m); } else if (flagsp != NULL) *flagsp &= ~MSG_TRUNC; return (0); } static bool uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) { struct mbuf *mb, *n; struct sockbuf *sb; SOCK_LOCK(so); if (SOLISTENING(so)) { SOCK_UNLOCK(so); return (false); } mb = NULL; sb = &so->so_rcv; SOCKBUF_LOCK(sb); if (sb->sb_fnrdy != NULL) { for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { if (mb == m) { *errorp = sbready(sb, m, count); break; } mb = mb->m_next; if (mb == NULL) { mb = n; if (mb != NULL) n = mb->m_nextpkt; } } } SOCKBUF_UNLOCK(sb); SOCK_UNLOCK(so); return (mb != NULL); } static int uipc_ready(struct socket *so, struct mbuf *m, int count) { struct unpcb *unp, *unp2; struct socket *so2; int error, i; unp = sotounpcb(so); KASSERT(so->so_type == SOCK_STREAM, ("%s: unexpected socket type for %p", __func__, so)); UNP_PCB_LOCK(unp); if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { UNP_PCB_UNLOCK(unp); so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_rcv); if ((error = sbready(&so2->so_rcv, m, count)) == 0) sorwakeup_locked(so2); else SOCKBUF_UNLOCK(&so2->so_rcv); UNP_PCB_UNLOCK(unp2); return (error); } UNP_PCB_UNLOCK(unp); /* * The receiving socket has been disconnected, but may still be valid. * In this case, the now-ready mbufs are still present in its socket * buffer, so perform an exhaustive search before giving up and freeing * the mbufs. */ UNP_LINK_RLOCK(); LIST_FOREACH(unp, &unp_shead, unp_link) { if (uipc_ready_scan(unp->unp_socket, m, count, &error)) break; } UNP_LINK_RUNLOCK(); if (unp == NULL) { for (i = 0; i < count; i++) m = m_free(m); error = ECONNRESET; } return (error); } static int uipc_sense(struct socket *so, struct stat *sb) { struct unpcb *unp; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); sb->st_blksize = so->so_snd.sb_hiwat; sb->st_dev = NODEV; sb->st_ino = unp->unp_ino; return (0); } static int uipc_shutdown(struct socket *so) { struct unpcb *unp; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); UNP_PCB_LOCK(unp); socantsendmore(so); unp_shutdown(unp); UNP_PCB_UNLOCK(unp); return (0); } static int uipc_sockaddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp; const struct sockaddr *sa; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_PCB_LOCK(unp); if (unp->unp_addr != NULL) sa = (struct sockaddr *) unp->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); UNP_PCB_UNLOCK(unp); return (0); } static int uipc_ctloutput(struct socket *so, struct sockopt *sopt) { struct unpcb *unp; struct xucred xu; int error, optval; if (sopt->sopt_level != SOL_LOCAL) return (EINVAL); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); error = 0; switch (sopt->sopt_dir) { case SOPT_GET: switch (sopt->sopt_name) { case LOCAL_PEERCRED: UNP_PCB_LOCK(unp); if (unp->unp_flags & UNP_HAVEPC) xu = unp->unp_peercred; else { if (so->so_type == SOCK_STREAM) error = ENOTCONN; else error = EINVAL; } UNP_PCB_UNLOCK(unp); if (error == 0) error = sooptcopyout(sopt, &xu, sizeof(xu)); break; case LOCAL_CREDS: /* Unlocked read. */ optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; case LOCAL_CREDS_PERSISTENT: /* Unlocked read. */ optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; case LOCAL_CONNWAIT: /* Unlocked read. */ optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; default: error = EOPNOTSUPP; break; } break; case SOPT_SET: switch (sopt->sopt_name) { case LOCAL_CREDS: case LOCAL_CREDS_PERSISTENT: case LOCAL_CONNWAIT: error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error) break; #define OPTSET(bit, exclusive) do { \ UNP_PCB_LOCK(unp); \ if (optval) { \ if ((unp->unp_flags & (exclusive)) != 0) { \ UNP_PCB_UNLOCK(unp); \ error = EINVAL; \ break; \ } \ unp->unp_flags |= (bit); \ } else \ unp->unp_flags &= ~(bit); \ UNP_PCB_UNLOCK(unp); \ } while (0) switch (sopt->sopt_name) { case LOCAL_CREDS: OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); break; case LOCAL_CREDS_PERSISTENT: OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); break; case LOCAL_CONNWAIT: OPTSET(UNP_CONNWAIT, 0); break; default: break; } break; #undef OPTSET default: error = ENOPROTOOPT; break; } break; default: error = EOPNOTSUPP; break; } return (error); } static int unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { return (unp_connectat(AT_FDCWD, so, nam, td, false)); } static int unp_connectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td, bool return_locked) { struct mtx *vplock; struct sockaddr_un *soun; struct vnode *vp; struct socket *so2; struct unpcb *unp, *unp2, *unp3; struct nameidata nd; char buf[SOCK_MAXADDRLEN]; struct sockaddr *sa; cap_rights_t rights; int error, len; bool connreq; if (nam->sa_family != AF_UNIX) return (EAFNOSUPPORT); if (nam->sa_len > sizeof(struct sockaddr_un)) return (EINVAL); len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); if (len <= 0) return (EINVAL); soun = (struct sockaddr_un *)nam; bcopy(soun->sun_path, buf, len); buf[len] = 0; error = 0; unp = sotounpcb(so); UNP_PCB_LOCK(unp); for (;;) { /* * Wait for connection state to stabilize. If a connection * already exists, give up. For datagram sockets, which permit * multiple consecutive connect(2) calls, upper layers are * responsible for disconnecting in advance of a subsequent * connect(2), but this is not synchronized with PCB connection * state. * * Also make sure that no threads are currently attempting to * lock the peer socket, to ensure that unp_conn cannot * transition between two valid sockets while locks are dropped. */ if (SOLISTENING(so)) error = EOPNOTSUPP; else if (unp->unp_conn != NULL) error = EISCONN; else if ((unp->unp_flags & UNP_CONNECTING) != 0) { error = EALREADY; } if (error != 0) { UNP_PCB_UNLOCK(unp); return (error); } if (unp->unp_pairbusy > 0) { unp->unp_flags |= UNP_WAITING; mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); continue; } break; } unp->unp_flags |= UNP_CONNECTING; UNP_PCB_UNLOCK(unp); connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; if (connreq) sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); else sa = NULL; NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); error = namei(&nd); if (error) vp = NULL; else vp = nd.ni_vp; ASSERT_VOP_LOCKED(vp, "unp_connect"); if (error) goto bad; NDFREE_PNBUF(&nd); if (vp->v_type != VSOCK) { error = ENOTSOCK; goto bad; } #ifdef MAC error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); if (error) goto bad; #endif error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); if (error) goto bad; unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect: unp == NULL")); vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); VOP_UNP_CONNECT(vp, &unp2); if (unp2 == NULL) { error = ECONNREFUSED; goto bad2; } so2 = unp2->unp_socket; if (so->so_type != so2->so_type) { error = EPROTOTYPE; goto bad2; } if (connreq) { if (SOLISTENING(so2)) { CURVNET_SET(so2->so_vnet); so2 = sonewconn(so2, 0); CURVNET_RESTORE(); } else so2 = NULL; if (so2 == NULL) { error = ECONNREFUSED; goto bad2; } unp3 = sotounpcb(so2); unp_pcb_lock_pair(unp2, unp3); if (unp2->unp_addr != NULL) { bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); unp3->unp_addr = (struct sockaddr_un *) sa; sa = NULL; } unp_copy_peercred(td, unp3, unp, unp2); UNP_PCB_UNLOCK(unp2); unp2 = unp3; /* * It is safe to block on the PCB lock here since unp2 is * nascent and cannot be connected to any other sockets. */ UNP_PCB_LOCK(unp); #ifdef MAC mac_socketpeer_set_from_socket(so, so2); mac_socketpeer_set_from_socket(so2, so); #endif } else { unp_pcb_lock_pair(unp, unp2); } KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && sotounpcb(so2) == unp2, ("%s: unp2 %p so2 %p", __func__, unp2, so2)); unp_connect2(so, so2, PRU_CONNECT); KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); unp->unp_flags &= ~UNP_CONNECTING; if (!return_locked) unp_pcb_unlock_pair(unp, unp2); bad2: mtx_unlock(vplock); bad: if (vp != NULL) { /* * If we are returning locked (called via uipc_sosend_dgram()), * we need to be sure that vput() won't sleep. This is * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. * SOCK_STREAM/SEQPACKET can't request return_locked (yet). */ MPASS(!(return_locked && connreq)); vput(vp); } free(sa, M_SONAME); if (__predict_false(error)) { UNP_PCB_LOCK(unp); KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); unp->unp_flags &= ~UNP_CONNECTING; UNP_PCB_UNLOCK(unp); } return (error); } /* * Set socket peer credentials at connection time. * * The client's PCB credentials are copied from its process structure. The * server's PCB credentials are copied from the socket on which it called * listen(2). uipc_listen cached that process's credentials at the time. */ void unp_copy_peercred(struct thread *td, struct unpcb *client_unp, struct unpcb *server_unp, struct unpcb *listen_unp) { cru2xt(td, &client_unp->unp_peercred); client_unp->unp_flags |= UNP_HAVEPC; memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, sizeof(server_unp->unp_peercred)); server_unp->unp_flags |= UNP_HAVEPC; client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); } static void unp_connect2(struct socket *so, struct socket *so2, conn2_how req) { struct unpcb *unp; struct unpcb *unp2; MPASS(so2->so_type == so->so_type); unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); unp2 = sotounpcb(so2); KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); UNP_PCB_LOCK_ASSERT(unp); UNP_PCB_LOCK_ASSERT(unp2); KASSERT(unp->unp_conn == NULL, ("%s: socket %p is already connected", __func__, unp)); unp->unp_conn = unp2; unp_pcb_hold(unp2); unp_pcb_hold(unp); switch (so->so_type) { case SOCK_DGRAM: UNP_REF_LIST_LOCK(); LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); UNP_REF_LIST_UNLOCK(); soisconnected(so); break; case SOCK_STREAM: case SOCK_SEQPACKET: KASSERT(unp2->unp_conn == NULL, ("%s: socket %p is already connected", __func__, unp2)); unp2->unp_conn = unp; if (req == PRU_CONNECT && ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) soisconnecting(so); else soisconnected(so); soisconnected(so2); break; default: panic("unp_connect2"); } } static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2) { struct socket *so, *so2; struct mbuf *m = NULL; #ifdef INVARIANTS struct unpcb *unptmp; #endif UNP_PCB_LOCK_ASSERT(unp); UNP_PCB_LOCK_ASSERT(unp2); KASSERT(unp->unp_conn == unp2, ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); unp->unp_conn = NULL; so = unp->unp_socket; so2 = unp2->unp_socket; switch (unp->unp_socket->so_type) { case SOCK_DGRAM: /* * Remove our send socket buffer from the peer's receive buffer. * Move the data to the receive buffer only if it is empty. * This is a protection against a scenario where a peer * connects, floods and disconnects, effectively blocking * sendto() from unconnected sockets. */ SOCK_RECVBUF_LOCK(so2); if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, uxdg_clist); if (__predict_true((so2->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) && STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, &so->so_snd.uxdg_mb); so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; } else { m = STAILQ_FIRST(&so->so_snd.uxdg_mb); STAILQ_INIT(&so->so_snd.uxdg_mb); so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; } /* Note: so may reconnect. */ so->so_snd.uxdg_cc = 0; so->so_snd.uxdg_ctl = 0; so->so_snd.uxdg_mbcnt = 0; } SOCK_RECVBUF_UNLOCK(so2); UNP_REF_LIST_LOCK(); #ifdef INVARIANTS LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { if (unptmp == unp) break; } KASSERT(unptmp != NULL, ("%s: %p not found in reflist of %p", __func__, unp, unp2)); #endif LIST_REMOVE(unp, unp_reflink); UNP_REF_LIST_UNLOCK(); if (so) { SOCK_LOCK(so); so->so_state &= ~SS_ISCONNECTED; SOCK_UNLOCK(so); } break; case SOCK_STREAM: case SOCK_SEQPACKET: if (so) soisdisconnected(so); MPASS(unp2->unp_conn == unp); unp2->unp_conn = NULL; if (so2) soisdisconnected(so2); break; } if (unp == unp2) { unp_pcb_rele_notlast(unp); if (!unp_pcb_rele(unp)) UNP_PCB_UNLOCK(unp); } else { if (!unp_pcb_rele(unp)) UNP_PCB_UNLOCK(unp); if (!unp_pcb_rele(unp2)) UNP_PCB_UNLOCK(unp2); } if (m != NULL) { unp_scan(m, unp_freerights); m_freem(m); } } /* * unp_pcblist() walks the global list of struct unpcb's to generate a * pointer list, bumping the refcount on each unpcb. It then copies them out * sequentially, validating the generation number on each to see if it has * been detached. All of this is necessary because copyout() may sleep on * disk I/O. */ static int unp_pcblist(SYSCTL_HANDLER_ARGS) { struct unpcb *unp, **unp_list; unp_gen_t gencnt; struct xunpgen *xug; struct unp_head *head; struct xunpcb *xu; u_int i; int error, n; switch ((intptr_t)arg1) { case SOCK_STREAM: head = &unp_shead; break; case SOCK_DGRAM: head = &unp_dhead; break; case SOCK_SEQPACKET: head = &unp_sphead; break; default: panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); } /* * The process of preparing the PCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ if (req->oldptr == NULL) { n = unp_count; req->oldidx = 2 * (sizeof *xug) + (n + n/8) * sizeof(struct xunpcb); return (0); } if (req->newptr != NULL) return (EPERM); /* * OK, now we're committed to doing something. */ xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); UNP_LINK_RLOCK(); gencnt = unp_gencnt; n = unp_count; UNP_LINK_RUNLOCK(); xug->xug_len = sizeof *xug; xug->xug_count = n; xug->xug_gen = gencnt; xug->xug_sogen = so_gencnt; error = SYSCTL_OUT(req, xug, sizeof *xug); if (error) { free(xug, M_TEMP); return (error); } unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); UNP_LINK_RLOCK(); for (unp = LIST_FIRST(head), i = 0; unp && i < n; unp = LIST_NEXT(unp, unp_link)) { UNP_PCB_LOCK(unp); if (unp->unp_gencnt <= gencnt) { if (cr_cansee(req->td->td_ucred, unp->unp_socket->so_cred)) { UNP_PCB_UNLOCK(unp); continue; } unp_list[i++] = unp; unp_pcb_hold(unp); } UNP_PCB_UNLOCK(unp); } UNP_LINK_RUNLOCK(); n = i; /* In case we lost some during malloc. */ error = 0; xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); for (i = 0; i < n; i++) { unp = unp_list[i]; UNP_PCB_LOCK(unp); if (unp_pcb_rele(unp)) continue; if (unp->unp_gencnt <= gencnt) { xu->xu_len = sizeof *xu; xu->xu_unpp = (uintptr_t)unp; /* * XXX - need more locking here to protect against * connect/disconnect races for SMP. */ if (unp->unp_addr != NULL) bcopy(unp->unp_addr, &xu->xu_addr, unp->unp_addr->sun_len); else bzero(&xu->xu_addr, sizeof(xu->xu_addr)); if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) bcopy(unp->unp_conn->unp_addr, &xu->xu_caddr, unp->unp_conn->unp_addr->sun_len); else bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); xu->unp_vnode = (uintptr_t)unp->unp_vnode; xu->unp_conn = (uintptr_t)unp->unp_conn; xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); xu->unp_gencnt = unp->unp_gencnt; sotoxsocket(unp->unp_socket, &xu->xu_socket); UNP_PCB_UNLOCK(unp); error = SYSCTL_OUT(req, xu, sizeof *xu); } else { UNP_PCB_UNLOCK(unp); } } free(xu, M_TEMP); if (!error) { /* * Give the user an updated idea of our state. If the * generation differs from what we told her before, she knows * that something happened while we were processing this * request, and it might be necessary to retry. */ xug->xug_gen = unp_gencnt; xug->xug_sogen = so_gencnt; xug->xug_count = unp_count; error = SYSCTL_OUT(req, xug, sizeof *xug); } free(unp_list, M_TEMP); free(xug, M_TEMP); return (error); } SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", "List of active local datagram sockets"); SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", "List of active local stream sockets"); SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", "List of active local seqpacket sockets"); static void unp_shutdown(struct unpcb *unp) { struct unpcb *unp2; struct socket *so; UNP_PCB_LOCK_ASSERT(unp); unp2 = unp->unp_conn; if ((unp->unp_socket->so_type == SOCK_STREAM || (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { so = unp2->unp_socket; if (so != NULL) socantrcvmore(so); } } static void unp_drop(struct unpcb *unp) { struct socket *so; struct unpcb *unp2; /* * Regardless of whether the socket's peer dropped the connection * with this socket by aborting or disconnecting, POSIX requires * that ECONNRESET is returned. */ UNP_PCB_LOCK(unp); so = unp->unp_socket; if (so) so->so_error = ECONNRESET; if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { /* Last reference dropped in unp_disconnect(). */ unp_pcb_rele_notlast(unp); unp_disconnect(unp, unp2); } else if (!unp_pcb_rele(unp)) { UNP_PCB_UNLOCK(unp); } } static void unp_freerights(struct filedescent **fdep, int fdcount) { struct file *fp; int i; KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; filecaps_free(&fdep[i]->fde_caps); unp_discard(fp); } free(fdep[0], M_FILECAPS); } static int unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) { struct thread *td = curthread; /* XXX */ struct cmsghdr *cm = mtod(control, struct cmsghdr *); int i; int *fdp; struct filedesc *fdesc = td->td_proc->p_fd; struct filedescent **fdep; void *data; socklen_t clen = control->m_len, datalen; int error, newfds; u_int newlen; UNP_LINK_UNLOCK_ASSERT(); error = 0; if (controlp != NULL) /* controlp == NULL => free control messages */ *controlp = NULL; while (cm != NULL) { MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_RIGHTS) { newfds = datalen / sizeof(*fdep); if (newfds == 0) goto next; fdep = data; /* If we're not outputting the descriptors free them. */ if (error || controlp == NULL) { unp_freerights(fdep, newfds); goto next; } FILEDESC_XLOCK(fdesc); /* * Now change each pointer to an fd in the global * table to an integer that is the index to the local * fd table entry that we set up to point to the * global one we are transferring. */ newlen = newfds * sizeof(int); *controlp = sbcreatecontrol(NULL, newlen, SCM_RIGHTS, SOL_SOCKET, M_WAITOK); fdp = (int *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); if ((error = fdallocn(td, 0, fdp, newfds))) { FILEDESC_XUNLOCK(fdesc); unp_freerights(fdep, newfds); m_freem(*controlp); *controlp = NULL; goto next; } for (i = 0; i < newfds; i++, fdp++) { _finstall(fdesc, fdep[i]->fde_file, *fdp, (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, &fdep[i]->fde_caps); unp_externalize_fp(fdep[i]->fde_file); } /* * The new type indicates that the mbuf data refers to * kernel resources that may need to be released before * the mbuf is freed. */ m_chtype(*controlp, MT_EXTCONTROL); FILEDESC_XUNLOCK(fdesc); free(fdep[0], M_FILECAPS); } else { /* We can just copy anything else across. */ if (error || controlp == NULL) goto next; *controlp = sbcreatecontrol(NULL, datalen, cm->cmsg_type, cm->cmsg_level, M_WAITOK); bcopy(data, CMSG_DATA(mtod(*controlp, struct cmsghdr *)), datalen); } controlp = &(*controlp)->m_next; next: if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } m_freem(control); return (error); } static void unp_zone_change(void *tag) { uma_zone_set_max(unp_zone, maxsockets); } #ifdef INVARIANTS static void unp_zdtor(void *mem, int size __unused, void *arg __unused) { struct unpcb *unp; unp = mem; KASSERT(LIST_EMPTY(&unp->unp_refs), ("%s: unpcb %p has lingering refs", __func__, unp)); KASSERT(unp->unp_socket == NULL, ("%s: unpcb %p has socket backpointer", __func__, unp)); KASSERT(unp->unp_vnode == NULL, ("%s: unpcb %p has vnode references", __func__, unp)); KASSERT(unp->unp_conn == NULL, ("%s: unpcb %p is still connected", __func__, unp)); KASSERT(unp->unp_addr == NULL, ("%s: unpcb %p has leaked addr", __func__, unp)); } #endif static void unp_init(void *arg __unused) { uma_dtor dtor; #ifdef INVARIANTS dtor = unp_zdtor; #else dtor = NULL; #endif unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, NULL, NULL, UMA_ALIGN_CACHE, 0); uma_zone_set_max(unp_zone, maxsockets); uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, NULL, EVENTHANDLER_PRI_ANY); LIST_INIT(&unp_dhead); LIST_INIT(&unp_shead); LIST_INIT(&unp_sphead); SLIST_INIT(&unp_defers); TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); UNP_LINK_LOCK_INIT(); UNP_DEFERRED_LOCK_INIT(); } SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); static void unp_internalize_cleanup_rights(struct mbuf *control) { struct cmsghdr *cp; struct mbuf *m; void *data; socklen_t datalen; for (m = control; m != NULL; m = m->m_next) { cp = mtod(m, struct cmsghdr *); if (cp->cmsg_level != SOL_SOCKET || cp->cmsg_type != SCM_RIGHTS) continue; data = CMSG_DATA(cp); datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; unp_freerights(data, datalen / sizeof(struct filedesc *)); } } static int unp_internalize(struct mbuf **controlp, struct thread *td, struct mbuf **clast, u_int *space, u_int *mbcnt) { struct mbuf *control, **initial_controlp; struct proc *p; struct filedesc *fdesc; struct bintime *bt; struct cmsghdr *cm; struct cmsgcred *cmcred; struct filedescent *fde, **fdep, *fdev; struct file *fp; struct timeval *tv; struct timespec *ts; void *data; socklen_t clen, datalen; int i, j, error, *fdp, oldfds; u_int newlen; MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ UNP_LINK_UNLOCK_ASSERT(); p = td->td_proc; fdesc = p->p_fd; error = 0; control = *controlp; *controlp = NULL; initial_controlp = controlp; for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), data = CMSG_DATA(cm); clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && (char *)cm + cm->cmsg_len >= (char *)data; clen -= min(CMSG_SPACE(datalen), clen), cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), data = CMSG_DATA(cm)) { datalen = (char *)cm + cm->cmsg_len - (char *)data; switch (cm->cmsg_type) { case SCM_CREDS: *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), SCM_CREDS, SOL_SOCKET, M_WAITOK); cmcred = (struct cmsgcred *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); cmcred->cmcred_pid = p->p_pid; cmcred->cmcred_uid = td->td_ucred->cr_ruid; cmcred->cmcred_gid = td->td_ucred->cr_rgid; cmcred->cmcred_euid = td->td_ucred->cr_uid; cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); for (i = 0; i < cmcred->cmcred_ngroups; i++) cmcred->cmcred_groups[i] = td->td_ucred->cr_groups[i]; break; case SCM_RIGHTS: oldfds = datalen / sizeof (int); if (oldfds == 0) continue; /* On some machines sizeof pointer is bigger than * sizeof int, so we need to check if data fits into * single mbuf. We could allocate several mbufs, and * unp_externalize() should even properly handle that. * But it is not worth to complicate the code for an * insane scenario of passing over 200 file descriptors * at once. */ newlen = oldfds * sizeof(fdep[0]); if (CMSG_SPACE(newlen) > MCLBYTES) { error = EMSGSIZE; goto out; } /* * Check that all the FDs passed in refer to legal * files. If not, reject the entire operation. */ fdp = data; FILEDESC_SLOCK(fdesc); for (i = 0; i < oldfds; i++, fdp++) { fp = fget_noref(fdesc, *fdp); if (fp == NULL) { FILEDESC_SUNLOCK(fdesc); error = EBADF; goto out; } if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { FILEDESC_SUNLOCK(fdesc); error = EOPNOTSUPP; goto out; } } /* * Now replace the integer FDs with pointers to the * file structure and capability rights. */ *controlp = sbcreatecontrol(NULL, newlen, SCM_RIGHTS, SOL_SOCKET, M_WAITOK); fdp = data; for (i = 0; i < oldfds; i++, fdp++) { if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { fdp = data; for (j = 0; j < i; j++, fdp++) { fdrop(fdesc->fd_ofiles[*fdp]. fde_file, td); } FILEDESC_SUNLOCK(fdesc); error = EBADF; goto out; } } fdp = data; fdep = (struct filedescent **) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, M_WAITOK); for (i = 0; i < oldfds; i++, fdev++, fdp++) { fde = &fdesc->fd_ofiles[*fdp]; fdep[i] = fdev; fdep[i]->fde_file = fde->fde_file; filecaps_copy(&fde->fde_caps, &fdep[i]->fde_caps, true); unp_internalize_fp(fdep[i]->fde_file); } FILEDESC_SUNLOCK(fdesc); break; case SCM_TIMESTAMP: *controlp = sbcreatecontrol(NULL, sizeof(*tv), SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); tv = (struct timeval *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); microtime(tv); break; case SCM_BINTIME: *controlp = sbcreatecontrol(NULL, sizeof(*bt), SCM_BINTIME, SOL_SOCKET, M_WAITOK); bt = (struct bintime *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); bintime(bt); break; case SCM_REALTIME: *controlp = sbcreatecontrol(NULL, sizeof(*ts), SCM_REALTIME, SOL_SOCKET, M_WAITOK); ts = (struct timespec *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); nanotime(ts); break; case SCM_MONOTONIC: *controlp = sbcreatecontrol(NULL, sizeof(*ts), SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); ts = (struct timespec *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); nanouptime(ts); break; default: error = EINVAL; goto out; } if (space != NULL) { *space += (*controlp)->m_len; *mbcnt += MSIZE; if ((*controlp)->m_flags & M_EXT) *mbcnt += (*controlp)->m_ext.ext_size; *clast = *controlp; } controlp = &(*controlp)->m_next; } if (clen > 0) error = EINVAL; out: if (error != 0 && initial_controlp != NULL) unp_internalize_cleanup_rights(*initial_controlp); m_freem(control); return (error); } static struct mbuf * unp_addsockcred(struct thread *td, struct mbuf *control, int mode, struct mbuf **clast, u_int *space, u_int *mbcnt) { struct mbuf *m, *n, *n_prev; const struct cmsghdr *cm; int ngroups, i, cmsgtype; size_t ctrlsz; ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); if (mode & UNP_WANTCRED_ALWAYS) { ctrlsz = SOCKCRED2SIZE(ngroups); cmsgtype = SCM_CREDS2; } else { ctrlsz = SOCKCREDSIZE(ngroups); cmsgtype = SCM_CREDS; } m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); if (m == NULL) return (control); MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); if (mode & UNP_WANTCRED_ALWAYS) { struct sockcred2 *sc; sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); sc->sc_version = 0; sc->sc_pid = td->td_proc->p_pid; sc->sc_uid = td->td_ucred->cr_ruid; sc->sc_euid = td->td_ucred->cr_uid; sc->sc_gid = td->td_ucred->cr_rgid; sc->sc_egid = td->td_ucred->cr_gid; sc->sc_ngroups = ngroups; for (i = 0; i < sc->sc_ngroups; i++) sc->sc_groups[i] = td->td_ucred->cr_groups[i]; } else { struct sockcred *sc; sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); sc->sc_uid = td->td_ucred->cr_ruid; sc->sc_euid = td->td_ucred->cr_uid; sc->sc_gid = td->td_ucred->cr_rgid; sc->sc_egid = td->td_ucred->cr_gid; sc->sc_ngroups = ngroups; for (i = 0; i < sc->sc_ngroups; i++) sc->sc_groups[i] = td->td_ucred->cr_groups[i]; } /* * Unlink SCM_CREDS control messages (struct cmsgcred), since just * created SCM_CREDS control message (struct sockcred) has another * format. */ if (control != NULL && cmsgtype == SCM_CREDS) for (n = control, n_prev = NULL; n != NULL;) { cm = mtod(n, struct cmsghdr *); if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_CREDS) { if (n_prev == NULL) control = n->m_next; else n_prev->m_next = n->m_next; if (space != NULL) { MPASS(*space >= n->m_len); *space -= n->m_len; MPASS(*mbcnt >= MSIZE); *mbcnt -= MSIZE; if (n->m_flags & M_EXT) { MPASS(*mbcnt >= n->m_ext.ext_size); *mbcnt -= n->m_ext.ext_size; } MPASS(clast); if (*clast == n) { MPASS(n->m_next == NULL); if (n_prev == NULL) *clast = m; else *clast = n_prev; } } n = m_free(n); } else { n_prev = n; n = n->m_next; } } /* Prepend it to the head. */ m->m_next = control; if (space != NULL) { *space += m->m_len; *mbcnt += MSIZE; if (control == NULL) *clast = m; } return (m); } static struct unpcb * fptounp(struct file *fp) { struct socket *so; if (fp->f_type != DTYPE_SOCKET) return (NULL); if ((so = fp->f_data) == NULL) return (NULL); if (so->so_proto->pr_domain != &localdomain) return (NULL); return sotounpcb(so); } static void unp_discard(struct file *fp) { struct unp_defer *dr; if (unp_externalize_fp(fp)) { dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); dr->ud_fp = fp; UNP_DEFERRED_LOCK(); SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); UNP_DEFERRED_UNLOCK(); atomic_add_int(&unp_defers_count, 1); taskqueue_enqueue(taskqueue_thread, &unp_defer_task); } else closef_nothread(fp); } static void unp_process_defers(void *arg __unused, int pending) { struct unp_defer *dr; SLIST_HEAD(, unp_defer) drl; int count; SLIST_INIT(&drl); for (;;) { UNP_DEFERRED_LOCK(); if (SLIST_FIRST(&unp_defers) == NULL) { UNP_DEFERRED_UNLOCK(); break; } SLIST_SWAP(&unp_defers, &drl, unp_defer); UNP_DEFERRED_UNLOCK(); count = 0; while ((dr = SLIST_FIRST(&drl)) != NULL) { SLIST_REMOVE_HEAD(&drl, ud_link); closef_nothread(dr->ud_fp); free(dr, M_TEMP); count++; } atomic_add_int(&unp_defers_count, -count); } } static void unp_internalize_fp(struct file *fp) { struct unpcb *unp; UNP_LINK_WLOCK(); if ((unp = fptounp(fp)) != NULL) { unp->unp_file = fp; unp->unp_msgcount++; } unp_rights++; UNP_LINK_WUNLOCK(); } static int unp_externalize_fp(struct file *fp) { struct unpcb *unp; int ret; UNP_LINK_WLOCK(); if ((unp = fptounp(fp)) != NULL) { unp->unp_msgcount--; ret = 1; } else ret = 0; unp_rights--; UNP_LINK_WUNLOCK(); return (ret); } /* * unp_defer indicates whether additional work has been defered for a future * pass through unp_gc(). It is thread local and does not require explicit * synchronization. */ static int unp_marked; static void unp_remove_dead_ref(struct filedescent **fdep, int fdcount) { struct unpcb *unp; struct file *fp; int i; /* * This function can only be called from the gc task. */ KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, ("%s: not on gc callout", __func__)); UNP_LINK_LOCK_ASSERT(); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; if ((unp = fptounp(fp)) == NULL) continue; if ((unp->unp_gcflag & UNPGC_DEAD) == 0) continue; unp->unp_gcrefs--; } } static void unp_restore_undead_ref(struct filedescent **fdep, int fdcount) { struct unpcb *unp; struct file *fp; int i; /* * This function can only be called from the gc task. */ KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, ("%s: not on gc callout", __func__)); UNP_LINK_LOCK_ASSERT(); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; if ((unp = fptounp(fp)) == NULL) continue; if ((unp->unp_gcflag & UNPGC_DEAD) == 0) continue; unp->unp_gcrefs++; unp_marked++; } } static void unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) { struct sockbuf *sb; SOCK_LOCK_ASSERT(so); if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) return; SOCK_RECVBUF_LOCK(so); switch (so->so_type) { case SOCK_DGRAM: unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); unp_scan(so->so_rcv.uxdg_peeked, op); TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); break; case SOCK_STREAM: case SOCK_SEQPACKET: unp_scan(so->so_rcv.sb_mb, op); break; } SOCK_RECVBUF_UNLOCK(so); } static void unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) { struct socket *so, *soa; so = unp->unp_socket; SOCK_LOCK(so); if (SOLISTENING(so)) { /* * Mark all sockets in our accept queue. */ TAILQ_FOREACH(soa, &so->sol_comp, so_list) unp_scan_socket(soa, op); } else { /* * Mark all sockets we reference with RIGHTS. */ unp_scan_socket(so, op); } SOCK_UNLOCK(so); } static int unp_recycled; SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "Number of unreachable sockets claimed by the garbage collector."); static int unp_taskcount; SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "Number of times the garbage collector has run."); SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, "Number of active local sockets."); static void unp_gc(__unused void *arg, int pending) { struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, NULL }; struct unp_head **head; struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ struct file *f, **unref; struct unpcb *unp, *unptmp; int i, total, unp_unreachable; LIST_INIT(&unp_deadhead); unp_taskcount++; UNP_LINK_RLOCK(); /* * First determine which sockets may be in cycles. */ unp_unreachable = 0; for (head = heads; *head != NULL; head++) LIST_FOREACH(unp, *head, unp_link) { KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, ("%s: unp %p has unexpected gc flags 0x%x", __func__, unp, (unsigned int)unp->unp_gcflag)); f = unp->unp_file; /* * Check for an unreachable socket potentially in a * cycle. It must be in a queue as indicated by * msgcount, and this must equal the file reference * count. Note that when msgcount is 0 the file is * NULL. */ if (f != NULL && unp->unp_msgcount != 0 && refcount_load(&f->f_count) == unp->unp_msgcount) { LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); unp->unp_gcflag |= UNPGC_DEAD; unp->unp_gcrefs = unp->unp_msgcount; unp_unreachable++; } } /* * Scan all sockets previously marked as potentially being in a cycle * and remove the references each socket holds on any UNPGC_DEAD * sockets in its queue. After this step, all remaining references on * sockets marked UNPGC_DEAD should not be part of any cycle. */ LIST_FOREACH(unp, &unp_deadhead, unp_dead) unp_gc_scan(unp, unp_remove_dead_ref); /* * If a socket still has a non-negative refcount, it cannot be in a * cycle. In this case increment refcount of all children iteratively. * Stop the scan once we do a complete loop without discovering * a new reachable socket. */ do { unp_marked = 0; LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) if (unp->unp_gcrefs > 0) { unp->unp_gcflag &= ~UNPGC_DEAD; LIST_REMOVE(unp, unp_dead); KASSERT(unp_unreachable > 0, ("%s: unp_unreachable underflow.", __func__)); unp_unreachable--; unp_gc_scan(unp, unp_restore_undead_ref); } } while (unp_marked); UNP_LINK_RUNLOCK(); if (unp_unreachable == 0) return; /* * Allocate space for a local array of dead unpcbs. * TODO: can this path be simplified by instead using the local * dead list at unp_deadhead, after taking out references * on the file object and/or unpcb and dropping the link lock? */ unref = malloc(unp_unreachable * sizeof(struct file *), M_TEMP, M_WAITOK); /* * Iterate looking for sockets which have been specifically marked * as unreachable and store them locally. */ UNP_LINK_RLOCK(); total = 0; LIST_FOREACH(unp, &unp_deadhead, unp_dead) { KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); unp->unp_gcflag &= ~UNPGC_DEAD; f = unp->unp_file; if (unp->unp_msgcount == 0 || f == NULL || refcount_load(&f->f_count) != unp->unp_msgcount || !fhold(f)) continue; unref[total++] = f; KASSERT(total <= unp_unreachable, ("%s: incorrect unreachable count.", __func__)); } UNP_LINK_RUNLOCK(); /* * Now flush all sockets, free'ing rights. This will free the * struct files associated with these sockets but leave each socket * with one remaining ref. */ for (i = 0; i < total; i++) { struct socket *so; so = unref[i]->f_data; CURVNET_SET(so->so_vnet); sorflush(so); CURVNET_RESTORE(); } /* * And finally release the sockets so they can be reclaimed. */ for (i = 0; i < total; i++) fdrop(unref[i], NULL); unp_recycled += total; free(unref, M_TEMP); } /* * Synchronize against unp_gc, which can trip over data as we are freeing it. */ static void unp_dispose(struct socket *so) { struct sockbuf *sb; struct unpcb *unp; struct mbuf *m; MPASS(!SOLISTENING(so)); unp = sotounpcb(so); UNP_LINK_WLOCK(); unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; UNP_LINK_WUNLOCK(); /* * Grab our special mbufs before calling sbrelease(). */ SOCK_RECVBUF_LOCK(so); switch (so->so_type) { case SOCK_DGRAM: while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); /* Note: socket of sb may reconnect. */ sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; } sb = &so->so_rcv; if (sb->uxdg_peeked != NULL) { STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, m_stailqpkt); sb->uxdg_peeked = NULL; } m = STAILQ_FIRST(&sb->uxdg_mb); STAILQ_INIT(&sb->uxdg_mb); /* XXX: our shortened sbrelease() */ (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY); /* * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to * prevent uipc_sosend_dgram() or unp_disconnect() adding more * data to the socket. * We are now in dom_dispose and it could be a call from * soshutdown() or from the final sofree(). The sofree() case * is simple as it guarantees that no more sends will happen, * however we can race with unp_disconnect() from our peer. * The shutdown(2) case is more exotic. It would call into * dom_dispose() only if socket is SS_ISCONNECTED. This is * possible if we did connect(2) on this socket and we also * had it bound with bind(2) and receive connections from other * sockets. Because soshutdown() violates POSIX (see comment * there) we will end up here shutting down our receive side. * Of course this will have affect not only on the peer we * connect(2)ed to, but also on all of the peers who had * connect(2)ed to us. Their sends would end up with ENOBUFS. */ sb->sb_state |= SBS_CANTRCVMORE; break; case SOCK_STREAM: case SOCK_SEQPACKET: sb = &so->so_rcv; m = sbcut_locked(sb, sb->sb_ccc); KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, ("%s: ccc %u mb %p mbcnt %u", __func__, sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); sbrelease_locked(so, SO_RCV); break; } SOCK_RECVBUF_UNLOCK(so); if (SOCK_IO_RECV_OWNED(so)) SOCK_IO_RECV_UNLOCK(so); if (m != NULL) { unp_scan(m, unp_freerights); m_freem(m); } } static void unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) { struct mbuf *m; struct cmsghdr *cm; void *data; socklen_t clen, datalen; while (m0 != NULL) { for (m = m0; m; m = m->m_next) { if (m->m_type != MT_CONTROL) continue; cm = mtod(m, struct cmsghdr *); clen = m->m_len; while (cm != NULL) { if (sizeof(*cm) > clen || cm->cmsg_len > clen) break; data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_RIGHTS) { (*op)(data, datalen / sizeof(struct filedescent *)); } if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } } m0 = m0->m_nextpkt; } } /* * Definitions of protocols supported in the LOCAL domain. */ static struct protosw streamproto = { .pr_type = SOCK_STREAM, .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS| PR_CAPATTACH, .pr_ctloutput = &uipc_ctloutput, .pr_abort = uipc_abort, .pr_accept = uipc_accept, .pr_attach = uipc_attach, .pr_bind = uipc_bind, .pr_bindat = uipc_bindat, .pr_connect = uipc_connect, .pr_connectat = uipc_connectat, .pr_connect2 = uipc_connect2, .pr_detach = uipc_detach, .pr_disconnect = uipc_disconnect, .pr_listen = uipc_listen, .pr_peeraddr = uipc_peeraddr, .pr_rcvd = uipc_rcvd, .pr_send = uipc_send, .pr_ready = uipc_ready, .pr_sense = uipc_sense, .pr_shutdown = uipc_shutdown, .pr_sockaddr = uipc_sockaddr, .pr_soreceive = soreceive_generic, .pr_close = uipc_close, }; static struct protosw dgramproto = { .pr_type = SOCK_DGRAM, .pr_flags = PR_ATOMIC | PR_ADDR |PR_RIGHTS | PR_CAPATTACH | PR_SOCKBUF, .pr_ctloutput = &uipc_ctloutput, .pr_abort = uipc_abort, .pr_accept = uipc_accept, .pr_attach = uipc_attach, .pr_bind = uipc_bind, .pr_bindat = uipc_bindat, .pr_connect = uipc_connect, .pr_connectat = uipc_connectat, .pr_connect2 = uipc_connect2, .pr_detach = uipc_detach, .pr_disconnect = uipc_disconnect, .pr_peeraddr = uipc_peeraddr, .pr_sosend = uipc_sosend_dgram, .pr_sense = uipc_sense, .pr_shutdown = uipc_shutdown, .pr_sockaddr = uipc_sockaddr, .pr_soreceive = uipc_soreceive_dgram, .pr_close = uipc_close, }; static struct protosw seqpacketproto = { .pr_type = SOCK_SEQPACKET, /* * XXXRW: For now, PR_ADDR because soreceive will bump into them * due to our use of sbappendaddr. A new sbappend variants is needed * that supports both atomic record writes and control data. */ .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED| PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH, .pr_ctloutput = &uipc_ctloutput, .pr_abort = uipc_abort, .pr_accept = uipc_accept, .pr_attach = uipc_attach, .pr_bind = uipc_bind, .pr_bindat = uipc_bindat, .pr_connect = uipc_connect, .pr_connectat = uipc_connectat, .pr_connect2 = uipc_connect2, .pr_detach = uipc_detach, .pr_disconnect = uipc_disconnect, .pr_listen = uipc_listen, .pr_peeraddr = uipc_peeraddr, .pr_rcvd = uipc_rcvd, .pr_send = uipc_send, .pr_sense = uipc_sense, .pr_shutdown = uipc_shutdown, .pr_sockaddr = uipc_sockaddr, .pr_soreceive = soreceive_generic, /* XXX: or...? */ .pr_close = uipc_close, }; static struct domain localdomain = { .dom_family = AF_LOCAL, .dom_name = "local", .dom_externalize = unp_externalize, .dom_dispose = unp_dispose, .dom_nprotosw = 3, .dom_protosw = { &streamproto, &dgramproto, &seqpacketproto, } }; DOMAIN_SET(local); /* * A helper function called by VFS before socket-type vnode reclamation. * For an active vnode it clears unp_vnode pointer and decrements unp_vnode * use count. */ void vfs_unp_reclaim(struct vnode *vp) { struct unpcb *unp; int active; struct mtx *vplock; ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); KASSERT(vp->v_type == VSOCK, ("vfs_unp_reclaim: vp->v_type != VSOCK")); active = 0; vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); VOP_UNP_CONNECT(vp, &unp); if (unp == NULL) goto done; UNP_PCB_LOCK(unp); if (unp->unp_vnode == vp) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; active = 1; } UNP_PCB_UNLOCK(unp); done: mtx_unlock(vplock); if (active) vunref(vp); } #ifdef DDB static void db_print_indent(int indent) { int i; for (i = 0; i < indent; i++) db_printf(" "); } static void db_print_unpflags(int unp_flags) { int comma; comma = 0; if (unp_flags & UNP_HAVEPC) { db_printf("%sUNP_HAVEPC", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_WANTCRED_ALWAYS) { db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_WANTCRED_ONESHOT) { db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_CONNWAIT) { db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_CONNECTING) { db_printf("%sUNP_CONNECTING", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_BINDING) { db_printf("%sUNP_BINDING", comma ? ", " : ""); comma = 1; } } static void db_print_xucred(int indent, struct xucred *xu) { int comma, i; db_print_indent(indent); db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); db_print_indent(indent); db_printf("cr_groups: "); comma = 0; for (i = 0; i < xu->cr_ngroups; i++) { db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); comma = 1; } db_printf("\n"); } static void db_print_unprefs(int indent, struct unp_head *uh) { struct unpcb *unp; int counter; counter = 0; LIST_FOREACH(unp, uh, unp_reflink) { if (counter % 4 == 0) db_print_indent(indent); db_printf("%p ", unp); if (counter % 4 == 3) db_printf("\n"); counter++; } if (counter != 0 && counter % 4 != 0) db_printf("\n"); } DB_SHOW_COMMAND(unpcb, db_show_unpcb) { struct unpcb *unp; if (!have_addr) { db_printf("usage: show unpcb \n"); return; } unp = (struct unpcb *)addr; db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, unp->unp_vnode); db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, unp->unp_conn); db_printf("unp_refs:\n"); db_print_unprefs(2, &unp->unp_refs); /* XXXRW: Would be nice to print the full address, if any. */ db_printf("unp_addr: %p\n", unp->unp_addr); db_printf("unp_gencnt: %llu\n", (unsigned long long)unp->unp_gencnt); db_printf("unp_flags: %x (", unp->unp_flags); db_print_unpflags(unp->unp_flags); db_printf(")\n"); db_printf("unp_peercred:\n"); db_print_xucred(2, &unp->unp_peercred); db_printf("unp_refcount: %u\n", unp->unp_refcount); } #endif diff --git a/sys/kern/vfs_acl.c b/sys/kern/vfs_acl.c index 2e3abe007c2a..dffce9d29291 100644 --- a/sys/kern/vfs_acl.c +++ b/sys/kern/vfs_acl.c @@ -1,598 +1,598 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1999-2006, 2016-2017 Robert N. M. Watson * All rights reserved. * * This software was developed by Robert Watson for the TrustedBSD Project. * * Portions of this software were developed by BAE Systems, the University of * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent * Computing (TC) research program. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * Developed by the TrustedBSD Project. * * ACL system calls and other functions common across different ACL types. * Type-specific routines go into subr_acl_.c. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include CTASSERT(ACL_MAX_ENTRIES >= OLDACL_MAX_ENTRIES); MALLOC_DEFINE(M_ACL, "acl", "Access Control Lists"); static int kern___acl_aclcheck_path(struct thread *td, const char *path, acl_type_t type, struct acl *aclp, int follow); static int kern___acl_delete_path(struct thread *td, const char *path, acl_type_t type, int follow); static int kern___acl_get_path(struct thread *td, const char *path, acl_type_t type, struct acl *aclp, int follow); static int kern___acl_set_path(struct thread *td, const char *path, acl_type_t type, const struct acl *aclp, int follow); static int vacl_set_acl(struct thread *td, struct vnode *vp, acl_type_t type, const struct acl *aclp); static int vacl_get_acl(struct thread *td, struct vnode *vp, acl_type_t type, struct acl *aclp); static int vacl_aclcheck(struct thread *td, struct vnode *vp, acl_type_t type, const struct acl *aclp); int acl_copy_oldacl_into_acl(const struct oldacl *source, struct acl *dest) { int i; if (source->acl_cnt < 0 || source->acl_cnt > OLDACL_MAX_ENTRIES) return (EINVAL); bzero(dest, sizeof(*dest)); dest->acl_cnt = source->acl_cnt; dest->acl_maxcnt = ACL_MAX_ENTRIES; for (i = 0; i < dest->acl_cnt; i++) { dest->acl_entry[i].ae_tag = source->acl_entry[i].ae_tag; dest->acl_entry[i].ae_id = source->acl_entry[i].ae_id; dest->acl_entry[i].ae_perm = source->acl_entry[i].ae_perm; } return (0); } int acl_copy_acl_into_oldacl(const struct acl *source, struct oldacl *dest) { int i; if (source->acl_cnt > OLDACL_MAX_ENTRIES) return (EINVAL); bzero(dest, sizeof(*dest)); dest->acl_cnt = source->acl_cnt; for (i = 0; i < dest->acl_cnt; i++) { dest->acl_entry[i].ae_tag = source->acl_entry[i].ae_tag; dest->acl_entry[i].ae_id = source->acl_entry[i].ae_id; dest->acl_entry[i].ae_perm = source->acl_entry[i].ae_perm; } return (0); } /* * At one time, "struct ACL" was extended in order to add support for NFSv4 * ACLs. Instead of creating compatibility versions of all the ACL-related * syscalls, they were left intact. It's possible to find out what the code * calling these syscalls (libc) expects basing on "type" argument - if it's * either ACL_TYPE_ACCESS_OLD or ACL_TYPE_DEFAULT_OLD (which previously were * known as ACL_TYPE_ACCESS and ACL_TYPE_DEFAULT), then it's the "struct * oldacl". If it's something else, then it's the new "struct acl". In the * latter case, the routines below just copyin/copyout the contents. In the * former case, they copyin the "struct oldacl" and convert it to the new * format. */ static int acl_copyin(const void *user_acl, struct acl *kernel_acl, acl_type_t type) { int error; struct oldacl old; switch (type) { case ACL_TYPE_ACCESS_OLD: case ACL_TYPE_DEFAULT_OLD: error = copyin(user_acl, &old, sizeof(old)); if (error != 0) break; acl_copy_oldacl_into_acl(&old, kernel_acl); break; default: error = copyin(user_acl, kernel_acl, sizeof(*kernel_acl)); if (kernel_acl->acl_maxcnt != ACL_MAX_ENTRIES) return (EINVAL); } return (error); } static int acl_copyout(const struct acl *kernel_acl, void *user_acl, acl_type_t type) { uint32_t am; int error; struct oldacl old; switch (type) { case ACL_TYPE_ACCESS_OLD: case ACL_TYPE_DEFAULT_OLD: error = acl_copy_acl_into_oldacl(kernel_acl, &old); if (error != 0) break; error = copyout(&old, user_acl, sizeof(old)); break; default: error = fueword32((char *)user_acl + offsetof(struct acl, acl_maxcnt), &am); if (error == -1) return (EFAULT); if (am != ACL_MAX_ENTRIES) return (EINVAL); error = copyout(kernel_acl, user_acl, sizeof(*kernel_acl)); } return (error); } /* * Convert "old" type - ACL_TYPE_{ACCESS,DEFAULT}_OLD - into its "new" * counterpart. It's required for old (pre-NFSv4 ACLs) libc to work * with new kernel. Fixing 'type' for old binaries with new libc * is being done in lib/libc/posix1e/acl_support.c:_acl_type_unold(). */ static int acl_type_unold(int type) { switch (type) { case ACL_TYPE_ACCESS_OLD: return (ACL_TYPE_ACCESS); case ACL_TYPE_DEFAULT_OLD: return (ACL_TYPE_DEFAULT); default: return (type); } } /* * These calls wrap the real vnode operations, and are called by the syscall * code once the syscall has converted the path or file descriptor to a vnode * (unlocked). The aclp pointer is assumed still to point to userland, so * this should not be consumed within the kernel except by syscall code. * Other code should directly invoke VOP_{SET,GET}ACL. */ /* * Given a vnode, set its ACL. */ static int vacl_set_acl(struct thread *td, struct vnode *vp, acl_type_t type, const struct acl *aclp) { struct acl *inkernelacl; struct mount *mp; int error; AUDIT_ARG_VALUE(type); inkernelacl = acl_alloc(M_WAITOK); error = acl_copyin(aclp, inkernelacl, type); if (error != 0) goto out; - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error != 0) goto out; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(vp); #ifdef MAC error = mac_vnode_check_setacl(td->td_ucred, vp, type, inkernelacl); if (error != 0) goto out_unlock; #endif error = VOP_SETACL(vp, acl_type_unold(type), inkernelacl, td->td_ucred, td); #ifdef MAC out_unlock: #endif VOP_UNLOCK(vp); vn_finished_write(mp); out: acl_free(inkernelacl); return (error); } /* * Given a vnode, get its ACL. */ static int vacl_get_acl(struct thread *td, struct vnode *vp, acl_type_t type, struct acl *aclp) { struct acl *inkernelacl; int error; AUDIT_ARG_VALUE(type); inkernelacl = acl_alloc(M_WAITOK | M_ZERO); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(vp); #ifdef MAC error = mac_vnode_check_getacl(td->td_ucred, vp, type); if (error != 0) goto out; #endif error = VOP_GETACL(vp, acl_type_unold(type), inkernelacl, td->td_ucred, td); #ifdef MAC out: #endif VOP_UNLOCK(vp); if (error == 0) error = acl_copyout(inkernelacl, aclp, type); acl_free(inkernelacl); return (error); } /* * Given a vnode, delete its ACL. */ static int vacl_delete(struct thread *td, struct vnode *vp, acl_type_t type) { struct mount *mp; int error; AUDIT_ARG_VALUE(type); - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(vp); #ifdef MAC error = mac_vnode_check_deleteacl(td->td_ucred, vp, type); if (error != 0) goto out; #endif error = VOP_SETACL(vp, acl_type_unold(type), 0, td->td_ucred, td); #ifdef MAC out: #endif VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } /* * Given a vnode, check whether an ACL is appropriate for it * * XXXRW: No vnode lock held so can't audit vnode state...? */ static int vacl_aclcheck(struct thread *td, struct vnode *vp, acl_type_t type, const struct acl *aclp) { struct acl *inkernelacl; int error; inkernelacl = acl_alloc(M_WAITOK); error = acl_copyin(aclp, inkernelacl, type); if (error != 0) goto out; error = VOP_ACLCHECK(vp, acl_type_unold(type), inkernelacl, td->td_ucred, td); out: acl_free(inkernelacl); return (error); } /* * syscalls -- convert the path/fd to a vnode, and call vacl_whatever. Don't * need to lock, as the vacl_ code will get/release any locks required. */ /* * Given a file path, get an ACL for it */ int sys___acl_get_file(struct thread *td, struct __acl_get_file_args *uap) { return (kern___acl_get_path(td, uap->path, uap->type, uap->aclp, FOLLOW)); } /* * Given a file path, get an ACL for it; don't follow links. */ int sys___acl_get_link(struct thread *td, struct __acl_get_link_args *uap) { return(kern___acl_get_path(td, uap->path, uap->type, uap->aclp, NOFOLLOW)); } static int kern___acl_get_path(struct thread *td, const char *path, acl_type_t type, struct acl *aclp, int follow) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, follow | AUDITVNODE1, UIO_USERSPACE, path); error = namei(&nd); if (error == 0) { error = vacl_get_acl(td, nd.ni_vp, type, aclp); NDFREE(&nd, 0); } return (error); } /* * Given a file path, set an ACL for it. */ int sys___acl_set_file(struct thread *td, struct __acl_set_file_args *uap) { return(kern___acl_set_path(td, uap->path, uap->type, uap->aclp, FOLLOW)); } /* * Given a file path, set an ACL for it; don't follow links. */ int sys___acl_set_link(struct thread *td, struct __acl_set_link_args *uap) { return(kern___acl_set_path(td, uap->path, uap->type, uap->aclp, NOFOLLOW)); } static int kern___acl_set_path(struct thread *td, const char *path, acl_type_t type, const struct acl *aclp, int follow) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, follow | AUDITVNODE1, UIO_USERSPACE, path); error = namei(&nd); if (error == 0) { error = vacl_set_acl(td, nd.ni_vp, type, aclp); NDFREE(&nd, 0); } return (error); } /* * Given a file descriptor, get an ACL for it. */ int sys___acl_get_fd(struct thread *td, struct __acl_get_fd_args *uap) { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->filedes); error = getvnode(td, uap->filedes, cap_rights_init_one(&rights, CAP_ACL_GET), &fp); if (error == 0) { error = vacl_get_acl(td, fp->f_vnode, uap->type, uap->aclp); fdrop(fp, td); } return (error); } /* * Given a file descriptor, set an ACL for it. */ int sys___acl_set_fd(struct thread *td, struct __acl_set_fd_args *uap) { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->filedes); error = getvnode(td, uap->filedes, cap_rights_init_one(&rights, CAP_ACL_SET), &fp); if (error == 0) { error = vacl_set_acl(td, fp->f_vnode, uap->type, uap->aclp); fdrop(fp, td); } return (error); } /* * Given a file path, delete an ACL from it. */ int sys___acl_delete_file(struct thread *td, struct __acl_delete_file_args *uap) { return (kern___acl_delete_path(td, uap->path, uap->type, FOLLOW)); } /* * Given a file path, delete an ACL from it; don't follow links. */ int sys___acl_delete_link(struct thread *td, struct __acl_delete_link_args *uap) { return (kern___acl_delete_path(td, uap->path, uap->type, NOFOLLOW)); } static int kern___acl_delete_path(struct thread *td, const char *path, acl_type_t type, int follow) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, follow, UIO_USERSPACE, path); error = namei(&nd); if (error == 0) { error = vacl_delete(td, nd.ni_vp, type); NDFREE(&nd, 0); } return (error); } /* * Given a file path, delete an ACL from it. */ int sys___acl_delete_fd(struct thread *td, struct __acl_delete_fd_args *uap) { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->filedes); error = getvnode(td, uap->filedes, cap_rights_init_one(&rights, CAP_ACL_DELETE), &fp); if (error == 0) { error = vacl_delete(td, fp->f_vnode, uap->type); fdrop(fp, td); } return (error); } /* * Given a file path, check an ACL for it. */ int sys___acl_aclcheck_file(struct thread *td, struct __acl_aclcheck_file_args *uap) { return (kern___acl_aclcheck_path(td, uap->path, uap->type, uap->aclp, FOLLOW)); } /* * Given a file path, check an ACL for it; don't follow links. */ int sys___acl_aclcheck_link(struct thread *td, struct __acl_aclcheck_link_args *uap) { return (kern___acl_aclcheck_path(td, uap->path, uap->type, uap->aclp, NOFOLLOW)); } static int kern___acl_aclcheck_path(struct thread *td, const char *path, acl_type_t type, struct acl *aclp, int follow) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, follow, UIO_USERSPACE, path); error = namei(&nd); if (error == 0) { error = vacl_aclcheck(td, nd.ni_vp, type, aclp); NDFREE(&nd, 0); } return (error); } /* * Given a file descriptor, check an ACL for it. */ int sys___acl_aclcheck_fd(struct thread *td, struct __acl_aclcheck_fd_args *uap) { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->filedes); error = getvnode(td, uap->filedes, cap_rights_init_one(&rights, CAP_ACL_CHECK), &fp); if (error == 0) { error = vacl_aclcheck(td, fp->f_vnode, uap->type, uap->aclp); fdrop(fp, td); } return (error); } struct acl * acl_alloc(int flags) { struct acl *aclp; aclp = malloc(sizeof(*aclp), M_ACL, flags); if (aclp == NULL) return (NULL); aclp->acl_maxcnt = ACL_MAX_ENTRIES; return (aclp); } void acl_free(struct acl *aclp) { free(aclp, M_ACL); } diff --git a/sys/kern/vfs_aio.c b/sys/kern/vfs_aio.c index c9489ee5b4ad..43e54e1ad9bf 100644 --- a/sys/kern/vfs_aio.c +++ b/sys/kern/vfs_aio.c @@ -1,3164 +1,3164 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1997 John S. Dyson. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. John S. Dyson's name may not be used to endorse or promote products * derived from this software without specific prior written permission. * * DISCLAIMER: This code isn't warranted to do anything useful. Anything * bad that happens because of using this software isn't the responsibility * of the author. This software is distributed AS-IS. */ /* * This file contains support for the POSIX 1003.1B AIO/LIO facility. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Counter for allocating reference ids to new jobs. Wrapped to 1 on * overflow. (XXX will be removed soon.) */ static u_long jobrefid; /* * Counter for aio_fsync. */ static uint64_t jobseqno; #ifndef MAX_AIO_PER_PROC #define MAX_AIO_PER_PROC 32 #endif #ifndef MAX_AIO_QUEUE_PER_PROC #define MAX_AIO_QUEUE_PER_PROC 256 #endif #ifndef MAX_AIO_QUEUE #define MAX_AIO_QUEUE 1024 /* Bigger than MAX_AIO_QUEUE_PER_PROC */ #endif #ifndef MAX_BUF_AIO #define MAX_BUF_AIO 16 #endif FEATURE(aio, "Asynchronous I/O"); SYSCTL_DECL(_p1003_1b); static MALLOC_DEFINE(M_LIO, "lio", "listio aio control block list"); static MALLOC_DEFINE(M_AIO, "aio", "structures for asynchronous I/O"); static SYSCTL_NODE(_vfs, OID_AUTO, aio, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Async IO management"); static int enable_aio_unsafe = 0; SYSCTL_INT(_vfs_aio, OID_AUTO, enable_unsafe, CTLFLAG_RW, &enable_aio_unsafe, 0, "Permit asynchronous IO on all file types, not just known-safe types"); static unsigned int unsafe_warningcnt = 1; SYSCTL_UINT(_vfs_aio, OID_AUTO, unsafe_warningcnt, CTLFLAG_RW, &unsafe_warningcnt, 0, "Warnings that will be triggered upon failed IO requests on unsafe files"); static int max_aio_procs = MAX_AIO_PROCS; SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_procs, CTLFLAG_RW, &max_aio_procs, 0, "Maximum number of kernel processes to use for handling async IO "); static int num_aio_procs = 0; SYSCTL_INT(_vfs_aio, OID_AUTO, num_aio_procs, CTLFLAG_RD, &num_aio_procs, 0, "Number of presently active kernel processes for async IO"); /* * The code will adjust the actual number of AIO processes towards this * number when it gets a chance. */ static int target_aio_procs = TARGET_AIO_PROCS; SYSCTL_INT(_vfs_aio, OID_AUTO, target_aio_procs, CTLFLAG_RW, &target_aio_procs, 0, "Preferred number of ready kernel processes for async IO"); static int max_queue_count = MAX_AIO_QUEUE; SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_queue, CTLFLAG_RW, &max_queue_count, 0, "Maximum number of aio requests to queue, globally"); static int num_queue_count = 0; SYSCTL_INT(_vfs_aio, OID_AUTO, num_queue_count, CTLFLAG_RD, &num_queue_count, 0, "Number of queued aio requests"); static int num_buf_aio = 0; SYSCTL_INT(_vfs_aio, OID_AUTO, num_buf_aio, CTLFLAG_RD, &num_buf_aio, 0, "Number of aio requests presently handled by the buf subsystem"); static int num_unmapped_aio = 0; SYSCTL_INT(_vfs_aio, OID_AUTO, num_unmapped_aio, CTLFLAG_RD, &num_unmapped_aio, 0, "Number of aio requests presently handled by unmapped I/O buffers"); /* Number of async I/O processes in the process of being started */ /* XXX This should be local to aio_aqueue() */ static int num_aio_resv_start = 0; static int aiod_lifetime; SYSCTL_INT(_vfs_aio, OID_AUTO, aiod_lifetime, CTLFLAG_RW, &aiod_lifetime, 0, "Maximum lifetime for idle aiod"); static int max_aio_per_proc = MAX_AIO_PER_PROC; SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_per_proc, CTLFLAG_RW, &max_aio_per_proc, 0, "Maximum active aio requests per process"); static int max_aio_queue_per_proc = MAX_AIO_QUEUE_PER_PROC; SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_queue_per_proc, CTLFLAG_RW, &max_aio_queue_per_proc, 0, "Maximum queued aio requests per process"); static int max_buf_aio = MAX_BUF_AIO; SYSCTL_INT(_vfs_aio, OID_AUTO, max_buf_aio, CTLFLAG_RW, &max_buf_aio, 0, "Maximum buf aio requests per process"); /* * Though redundant with vfs.aio.max_aio_queue_per_proc, POSIX requires * sysconf(3) to support AIO_LISTIO_MAX, and we implement that with * vfs.aio.aio_listio_max. */ SYSCTL_INT(_p1003_1b, CTL_P1003_1B_AIO_LISTIO_MAX, aio_listio_max, CTLFLAG_RD | CTLFLAG_CAPRD, &max_aio_queue_per_proc, 0, "Maximum aio requests for a single lio_listio call"); #ifdef COMPAT_FREEBSD6 typedef struct oaiocb { int aio_fildes; /* File descriptor */ off_t aio_offset; /* File offset for I/O */ volatile void *aio_buf; /* I/O buffer in process space */ size_t aio_nbytes; /* Number of bytes for I/O */ struct osigevent aio_sigevent; /* Signal to deliver */ int aio_lio_opcode; /* LIO opcode */ int aio_reqprio; /* Request priority -- ignored */ struct __aiocb_private _aiocb_private; } oaiocb_t; #endif /* * Below is a key of locks used to protect each member of struct kaiocb * aioliojob and kaioinfo and any backends. * * * - need not protected * a - locked by kaioinfo lock * b - locked by backend lock, the backend lock can be null in some cases, * for example, BIO belongs to this type, in this case, proc lock is * reused. * c - locked by aio_job_mtx, the lock for the generic file I/O backend. */ /* * If the routine that services an AIO request blocks while running in an * AIO kernel process it can starve other I/O requests. BIO requests * queued via aio_qbio() complete asynchronously and do not use AIO kernel * processes at all. Socket I/O requests use a separate pool of * kprocs and also force non-blocking I/O. Other file I/O requests * use the generic fo_read/fo_write operations which can block. The * fsync and mlock operations can also block while executing. Ideally * none of these requests would block while executing. * * Note that the service routines cannot toggle O_NONBLOCK in the file * structure directly while handling a request due to races with * userland threads. */ /* jobflags */ #define KAIOCB_QUEUEING 0x01 #define KAIOCB_CANCELLED 0x02 #define KAIOCB_CANCELLING 0x04 #define KAIOCB_CHECKSYNC 0x08 #define KAIOCB_CLEARED 0x10 #define KAIOCB_FINISHED 0x20 /* * AIO process info */ #define AIOP_FREE 0x1 /* proc on free queue */ struct aioproc { int aioprocflags; /* (c) AIO proc flags */ TAILQ_ENTRY(aioproc) list; /* (c) list of processes */ struct proc *aioproc; /* (*) the AIO proc */ }; /* * data-structure for lio signal management */ struct aioliojob { int lioj_flags; /* (a) listio flags */ int lioj_count; /* (a) count of jobs */ int lioj_finished_count; /* (a) count of finished jobs */ struct sigevent lioj_signal; /* (a) signal on all I/O done */ TAILQ_ENTRY(aioliojob) lioj_list; /* (a) lio list */ struct knlist klist; /* (a) list of knotes */ ksiginfo_t lioj_ksi; /* (a) Realtime signal info */ }; #define LIOJ_SIGNAL 0x1 /* signal on all done (lio) */ #define LIOJ_SIGNAL_POSTED 0x2 /* signal has been posted */ #define LIOJ_KEVENT_POSTED 0x4 /* kevent triggered */ /* * per process aio data structure */ struct kaioinfo { struct mtx kaio_mtx; /* the lock to protect this struct */ int kaio_flags; /* (a) per process kaio flags */ int kaio_active_count; /* (c) number of currently used AIOs */ int kaio_count; /* (a) size of AIO queue */ int kaio_buffer_count; /* (a) number of bio buffers */ TAILQ_HEAD(,kaiocb) kaio_all; /* (a) all AIOs in a process */ TAILQ_HEAD(,kaiocb) kaio_done; /* (a) done queue for process */ TAILQ_HEAD(,aioliojob) kaio_liojoblist; /* (a) list of lio jobs */ TAILQ_HEAD(,kaiocb) kaio_jobqueue; /* (a) job queue for process */ TAILQ_HEAD(,kaiocb) kaio_syncqueue; /* (a) queue for aio_fsync */ TAILQ_HEAD(,kaiocb) kaio_syncready; /* (a) second q for aio_fsync */ struct task kaio_task; /* (*) task to kick aio processes */ struct task kaio_sync_task; /* (*) task to schedule fsync jobs */ }; #define AIO_LOCK(ki) mtx_lock(&(ki)->kaio_mtx) #define AIO_UNLOCK(ki) mtx_unlock(&(ki)->kaio_mtx) #define AIO_LOCK_ASSERT(ki, f) mtx_assert(&(ki)->kaio_mtx, (f)) #define AIO_MTX(ki) (&(ki)->kaio_mtx) #define KAIO_RUNDOWN 0x1 /* process is being run down */ #define KAIO_WAKEUP 0x2 /* wakeup process when AIO completes */ /* * Operations used to interact with userland aio control blocks. * Different ABIs provide their own operations. */ struct aiocb_ops { int (*aio_copyin)(struct aiocb *ujob, struct kaiocb *kjob, int ty); long (*fetch_status)(struct aiocb *ujob); long (*fetch_error)(struct aiocb *ujob); int (*store_status)(struct aiocb *ujob, long status); int (*store_error)(struct aiocb *ujob, long error); int (*store_kernelinfo)(struct aiocb *ujob, long jobref); int (*store_aiocb)(struct aiocb **ujobp, struct aiocb *ujob); }; static TAILQ_HEAD(,aioproc) aio_freeproc; /* (c) Idle daemons */ static struct sema aio_newproc_sem; static struct mtx aio_job_mtx; static TAILQ_HEAD(,kaiocb) aio_jobs; /* (c) Async job list */ static struct unrhdr *aiod_unr; static void aio_biocleanup(struct bio *bp); void aio_init_aioinfo(struct proc *p); static int aio_onceonly(void); static int aio_free_entry(struct kaiocb *job); static void aio_process_rw(struct kaiocb *job); static void aio_process_sync(struct kaiocb *job); static void aio_process_mlock(struct kaiocb *job); static void aio_schedule_fsync(void *context, int pending); static int aio_newproc(int *); int aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lio, int type, struct aiocb_ops *ops); static int aio_queue_file(struct file *fp, struct kaiocb *job); static void aio_biowakeup(struct bio *bp); static void aio_proc_rundown(void *arg, struct proc *p); static void aio_proc_rundown_exec(void *arg, struct proc *p, struct image_params *imgp); static int aio_qbio(struct proc *p, struct kaiocb *job); static void aio_daemon(void *param); static void aio_bio_done_notify(struct proc *userp, struct kaiocb *job); static bool aio_clear_cancel_function_locked(struct kaiocb *job); static int aio_kick(struct proc *userp); static void aio_kick_nowait(struct proc *userp); static void aio_kick_helper(void *context, int pending); static int filt_aioattach(struct knote *kn); static void filt_aiodetach(struct knote *kn); static int filt_aio(struct knote *kn, long hint); static int filt_lioattach(struct knote *kn); static void filt_liodetach(struct knote *kn); static int filt_lio(struct knote *kn, long hint); /* * Zones for: * kaio Per process async io info * aiocb async io jobs * aiolio list io jobs */ static uma_zone_t kaio_zone, aiocb_zone, aiolio_zone; /* kqueue filters for aio */ static struct filterops aio_filtops = { .f_isfd = 0, .f_attach = filt_aioattach, .f_detach = filt_aiodetach, .f_event = filt_aio, }; static struct filterops lio_filtops = { .f_isfd = 0, .f_attach = filt_lioattach, .f_detach = filt_liodetach, .f_event = filt_lio }; static eventhandler_tag exit_tag, exec_tag; TASKQUEUE_DEFINE_THREAD(aiod_kick); /* * Main operations function for use as a kernel module. */ static int aio_modload(struct module *module, int cmd, void *arg) { int error = 0; switch (cmd) { case MOD_LOAD: aio_onceonly(); break; case MOD_SHUTDOWN: break; default: error = EOPNOTSUPP; break; } return (error); } static moduledata_t aio_mod = { "aio", &aio_modload, NULL }; DECLARE_MODULE(aio, aio_mod, SI_SUB_VFS, SI_ORDER_ANY); MODULE_VERSION(aio, 1); /* * Startup initialization */ static int aio_onceonly(void) { exit_tag = EVENTHANDLER_REGISTER(process_exit, aio_proc_rundown, NULL, EVENTHANDLER_PRI_ANY); exec_tag = EVENTHANDLER_REGISTER(process_exec, aio_proc_rundown_exec, NULL, EVENTHANDLER_PRI_ANY); kqueue_add_filteropts(EVFILT_AIO, &aio_filtops); kqueue_add_filteropts(EVFILT_LIO, &lio_filtops); TAILQ_INIT(&aio_freeproc); sema_init(&aio_newproc_sem, 0, "aio_new_proc"); mtx_init(&aio_job_mtx, "aio_job", NULL, MTX_DEF); TAILQ_INIT(&aio_jobs); aiod_unr = new_unrhdr(1, INT_MAX, NULL); kaio_zone = uma_zcreate("AIO", sizeof(struct kaioinfo), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); aiocb_zone = uma_zcreate("AIOCB", sizeof(struct kaiocb), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); aiolio_zone = uma_zcreate("AIOLIO", sizeof(struct aioliojob), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); aiod_lifetime = AIOD_LIFETIME_DEFAULT; jobrefid = 1; p31b_setcfg(CTL_P1003_1B_ASYNCHRONOUS_IO, _POSIX_ASYNCHRONOUS_IO); p31b_setcfg(CTL_P1003_1B_AIO_MAX, MAX_AIO_QUEUE); p31b_setcfg(CTL_P1003_1B_AIO_PRIO_DELTA_MAX, 0); return (0); } /* * Init the per-process aioinfo structure. The aioinfo limits are set * per-process for user limit (resource) management. */ void aio_init_aioinfo(struct proc *p) { struct kaioinfo *ki; ki = uma_zalloc(kaio_zone, M_WAITOK); mtx_init(&ki->kaio_mtx, "aiomtx", NULL, MTX_DEF | MTX_NEW); ki->kaio_flags = 0; ki->kaio_active_count = 0; ki->kaio_count = 0; ki->kaio_buffer_count = 0; TAILQ_INIT(&ki->kaio_all); TAILQ_INIT(&ki->kaio_done); TAILQ_INIT(&ki->kaio_jobqueue); TAILQ_INIT(&ki->kaio_liojoblist); TAILQ_INIT(&ki->kaio_syncqueue); TAILQ_INIT(&ki->kaio_syncready); TASK_INIT(&ki->kaio_task, 0, aio_kick_helper, p); TASK_INIT(&ki->kaio_sync_task, 0, aio_schedule_fsync, ki); PROC_LOCK(p); if (p->p_aioinfo == NULL) { p->p_aioinfo = ki; PROC_UNLOCK(p); } else { PROC_UNLOCK(p); mtx_destroy(&ki->kaio_mtx); uma_zfree(kaio_zone, ki); } while (num_aio_procs < MIN(target_aio_procs, max_aio_procs)) aio_newproc(NULL); } static int aio_sendsig(struct proc *p, struct sigevent *sigev, ksiginfo_t *ksi, bool ext) { struct thread *td; int error; error = sigev_findtd(p, sigev, &td); if (error) return (error); if (!KSI_ONQ(ksi)) { ksiginfo_set_sigev(ksi, sigev); ksi->ksi_code = SI_ASYNCIO; ksi->ksi_flags |= ext ? (KSI_EXT | KSI_INS) : 0; tdsendsignal(p, td, ksi->ksi_signo, ksi); } PROC_UNLOCK(p); return (error); } /* * Free a job entry. Wait for completion if it is currently active, but don't * delay forever. If we delay, we return a flag that says that we have to * restart the queue scan. */ static int aio_free_entry(struct kaiocb *job) { struct kaioinfo *ki; struct aioliojob *lj; struct proc *p; p = job->userproc; MPASS(curproc == p); ki = p->p_aioinfo; MPASS(ki != NULL); AIO_LOCK_ASSERT(ki, MA_OWNED); MPASS(job->jobflags & KAIOCB_FINISHED); atomic_subtract_int(&num_queue_count, 1); ki->kaio_count--; MPASS(ki->kaio_count >= 0); TAILQ_REMOVE(&ki->kaio_done, job, plist); TAILQ_REMOVE(&ki->kaio_all, job, allist); lj = job->lio; if (lj) { lj->lioj_count--; lj->lioj_finished_count--; if (lj->lioj_count == 0) { TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list); /* lio is going away, we need to destroy any knotes */ knlist_delete(&lj->klist, curthread, 1); PROC_LOCK(p); sigqueue_take(&lj->lioj_ksi); PROC_UNLOCK(p); uma_zfree(aiolio_zone, lj); } } /* job is going away, we need to destroy any knotes */ knlist_delete(&job->klist, curthread, 1); PROC_LOCK(p); sigqueue_take(&job->ksi); PROC_UNLOCK(p); AIO_UNLOCK(ki); /* * The thread argument here is used to find the owning process * and is also passed to fo_close() which may pass it to various * places such as devsw close() routines. Because of that, we * need a thread pointer from the process owning the job that is * persistent and won't disappear out from under us or move to * another process. * * Currently, all the callers of this function call it to remove * a kaiocb from the current process' job list either via a * syscall or due to the current process calling exit() or * execve(). Thus, we know that p == curproc. We also know that * curthread can't exit since we are curthread. * * Therefore, we use curthread as the thread to pass to * knlist_delete(). This does mean that it is possible for the * thread pointer at close time to differ from the thread pointer * at open time, but this is already true of file descriptors in * a multithreaded process. */ if (job->fd_file) fdrop(job->fd_file, curthread); crfree(job->cred); if (job->uiop != &job->uio) free(job->uiop, M_IOV); uma_zfree(aiocb_zone, job); AIO_LOCK(ki); return (0); } static void aio_proc_rundown_exec(void *arg, struct proc *p, struct image_params *imgp __unused) { aio_proc_rundown(arg, p); } static int aio_cancel_job(struct proc *p, struct kaioinfo *ki, struct kaiocb *job) { aio_cancel_fn_t *func; int cancelled; AIO_LOCK_ASSERT(ki, MA_OWNED); if (job->jobflags & (KAIOCB_CANCELLED | KAIOCB_FINISHED)) return (0); MPASS((job->jobflags & KAIOCB_CANCELLING) == 0); job->jobflags |= KAIOCB_CANCELLED; func = job->cancel_fn; /* * If there is no cancel routine, just leave the job marked as * cancelled. The job should be in active use by a caller who * should complete it normally or when it fails to install a * cancel routine. */ if (func == NULL) return (0); /* * Set the CANCELLING flag so that aio_complete() will defer * completions of this job. This prevents the job from being * freed out from under the cancel callback. After the * callback any deferred completion (whether from the callback * or any other source) will be completed. */ job->jobflags |= KAIOCB_CANCELLING; AIO_UNLOCK(ki); func(job); AIO_LOCK(ki); job->jobflags &= ~KAIOCB_CANCELLING; if (job->jobflags & KAIOCB_FINISHED) { cancelled = job->uaiocb._aiocb_private.error == ECANCELED; TAILQ_REMOVE(&ki->kaio_jobqueue, job, plist); aio_bio_done_notify(p, job); } else { /* * The cancel callback might have scheduled an * operation to cancel this request, but it is * only counted as cancelled if the request is * cancelled when the callback returns. */ cancelled = 0; } return (cancelled); } /* * Rundown the jobs for a given process. */ static void aio_proc_rundown(void *arg, struct proc *p) { struct kaioinfo *ki; struct aioliojob *lj; struct kaiocb *job, *jobn; KASSERT(curthread->td_proc == p, ("%s: called on non-curproc", __func__)); ki = p->p_aioinfo; if (ki == NULL) return; AIO_LOCK(ki); ki->kaio_flags |= KAIO_RUNDOWN; restart: /* * Try to cancel all pending requests. This code simulates * aio_cancel on all pending I/O requests. */ TAILQ_FOREACH_SAFE(job, &ki->kaio_jobqueue, plist, jobn) { aio_cancel_job(p, ki, job); } /* Wait for all running I/O to be finished */ if (TAILQ_FIRST(&ki->kaio_jobqueue) || ki->kaio_active_count != 0) { ki->kaio_flags |= KAIO_WAKEUP; msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO, "aioprn", hz); goto restart; } /* Free all completed I/O requests. */ while ((job = TAILQ_FIRST(&ki->kaio_done)) != NULL) aio_free_entry(job); while ((lj = TAILQ_FIRST(&ki->kaio_liojoblist)) != NULL) { if (lj->lioj_count == 0) { TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list); knlist_delete(&lj->klist, curthread, 1); PROC_LOCK(p); sigqueue_take(&lj->lioj_ksi); PROC_UNLOCK(p); uma_zfree(aiolio_zone, lj); } else { panic("LIO job not cleaned up: C:%d, FC:%d\n", lj->lioj_count, lj->lioj_finished_count); } } AIO_UNLOCK(ki); taskqueue_drain(taskqueue_aiod_kick, &ki->kaio_task); taskqueue_drain(taskqueue_aiod_kick, &ki->kaio_sync_task); mtx_destroy(&ki->kaio_mtx); uma_zfree(kaio_zone, ki); p->p_aioinfo = NULL; } /* * Select a job to run (called by an AIO daemon). */ static struct kaiocb * aio_selectjob(struct aioproc *aiop) { struct kaiocb *job; struct kaioinfo *ki; struct proc *userp; mtx_assert(&aio_job_mtx, MA_OWNED); restart: TAILQ_FOREACH(job, &aio_jobs, list) { userp = job->userproc; ki = userp->p_aioinfo; if (ki->kaio_active_count < max_aio_per_proc) { TAILQ_REMOVE(&aio_jobs, job, list); if (!aio_clear_cancel_function(job)) goto restart; /* Account for currently active jobs. */ ki->kaio_active_count++; break; } } return (job); } /* * Move all data to a permanent storage device. This code * simulates the fsync and fdatasync syscalls. */ static int aio_fsync_vnode(struct thread *td, struct vnode *vp, int op) { struct mount *mp; vm_object_t obj; int error; for (;;) { - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error != 0) break; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); obj = vp->v_object; if (obj != NULL) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, 0); VM_OBJECT_WUNLOCK(obj); } if (op == LIO_DSYNC) error = VOP_FDATASYNC(vp, td); else error = VOP_FSYNC(vp, MNT_WAIT, td); VOP_UNLOCK(vp); vn_finished_write(mp); if (error != ERELOOKUP) break; } return (error); } /* * The AIO processing activity for LIO_READ/LIO_WRITE. This is the code that * does the I/O request for the non-bio version of the operations. The normal * vn operations are used, and this code should work in all instances for every * type of file, including pipes, sockets, fifos, and regular files. * * XXX I don't think it works well for socket, pipe, and fifo. */ static void aio_process_rw(struct kaiocb *job) { struct ucred *td_savedcred; struct thread *td; struct file *fp; ssize_t cnt; long msgsnd_st, msgsnd_end; long msgrcv_st, msgrcv_end; long oublock_st, oublock_end; long inblock_st, inblock_end; int error, opcode; KASSERT(job->uaiocb.aio_lio_opcode == LIO_READ || job->uaiocb.aio_lio_opcode == LIO_READV || job->uaiocb.aio_lio_opcode == LIO_WRITE || job->uaiocb.aio_lio_opcode == LIO_WRITEV, ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode)); aio_switch_vmspace(job); td = curthread; td_savedcred = td->td_ucred; td->td_ucred = job->cred; job->uiop->uio_td = td; fp = job->fd_file; opcode = job->uaiocb.aio_lio_opcode; cnt = job->uiop->uio_resid; msgrcv_st = td->td_ru.ru_msgrcv; msgsnd_st = td->td_ru.ru_msgsnd; inblock_st = td->td_ru.ru_inblock; oublock_st = td->td_ru.ru_oublock; /* * aio_aqueue() acquires a reference to the file that is * released in aio_free_entry(). */ if (opcode == LIO_READ || opcode == LIO_READV) { if (job->uiop->uio_resid == 0) error = 0; else error = fo_read(fp, job->uiop, fp->f_cred, FOF_OFFSET, td); } else { if (fp->f_type == DTYPE_VNODE) bwillwrite(); error = fo_write(fp, job->uiop, fp->f_cred, FOF_OFFSET, td); } msgrcv_end = td->td_ru.ru_msgrcv; msgsnd_end = td->td_ru.ru_msgsnd; inblock_end = td->td_ru.ru_inblock; oublock_end = td->td_ru.ru_oublock; job->msgrcv = msgrcv_end - msgrcv_st; job->msgsnd = msgsnd_end - msgsnd_st; job->inblock = inblock_end - inblock_st; job->outblock = oublock_end - oublock_st; if (error != 0 && job->uiop->uio_resid != cnt) { if (error == ERESTART || error == EINTR || error == EWOULDBLOCK) error = 0; if (error == EPIPE && (opcode & LIO_WRITE)) { PROC_LOCK(job->userproc); kern_psignal(job->userproc, SIGPIPE); PROC_UNLOCK(job->userproc); } } cnt -= job->uiop->uio_resid; td->td_ucred = td_savedcred; if (error) aio_complete(job, -1, error); else aio_complete(job, cnt, 0); } static void aio_process_sync(struct kaiocb *job) { struct thread *td = curthread; struct ucred *td_savedcred = td->td_ucred; struct file *fp = job->fd_file; int error = 0; KASSERT(job->uaiocb.aio_lio_opcode & LIO_SYNC, ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode)); td->td_ucred = job->cred; if (fp->f_vnode != NULL) { error = aio_fsync_vnode(td, fp->f_vnode, job->uaiocb.aio_lio_opcode); } td->td_ucred = td_savedcred; if (error) aio_complete(job, -1, error); else aio_complete(job, 0, 0); } static void aio_process_mlock(struct kaiocb *job) { struct aiocb *cb = &job->uaiocb; int error; KASSERT(job->uaiocb.aio_lio_opcode == LIO_MLOCK, ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode)); aio_switch_vmspace(job); error = kern_mlock(job->userproc, job->cred, __DEVOLATILE(uintptr_t, cb->aio_buf), cb->aio_nbytes); aio_complete(job, error != 0 ? -1 : 0, error); } static void aio_bio_done_notify(struct proc *userp, struct kaiocb *job) { struct aioliojob *lj; struct kaioinfo *ki; struct kaiocb *sjob, *sjobn; int lj_done; bool schedule_fsync; ki = userp->p_aioinfo; AIO_LOCK_ASSERT(ki, MA_OWNED); lj = job->lio; lj_done = 0; if (lj) { lj->lioj_finished_count++; if (lj->lioj_count == lj->lioj_finished_count) lj_done = 1; } TAILQ_INSERT_TAIL(&ki->kaio_done, job, plist); MPASS(job->jobflags & KAIOCB_FINISHED); if (ki->kaio_flags & KAIO_RUNDOWN) goto notification_done; if (job->uaiocb.aio_sigevent.sigev_notify == SIGEV_SIGNAL || job->uaiocb.aio_sigevent.sigev_notify == SIGEV_THREAD_ID) aio_sendsig(userp, &job->uaiocb.aio_sigevent, &job->ksi, true); KNOTE_LOCKED(&job->klist, 1); if (lj_done) { if (lj->lioj_signal.sigev_notify == SIGEV_KEVENT) { lj->lioj_flags |= LIOJ_KEVENT_POSTED; KNOTE_LOCKED(&lj->klist, 1); } if ((lj->lioj_flags & (LIOJ_SIGNAL | LIOJ_SIGNAL_POSTED)) == LIOJ_SIGNAL && (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL || lj->lioj_signal.sigev_notify == SIGEV_THREAD_ID)) { aio_sendsig(userp, &lj->lioj_signal, &lj->lioj_ksi, true); lj->lioj_flags |= LIOJ_SIGNAL_POSTED; } } notification_done: if (job->jobflags & KAIOCB_CHECKSYNC) { schedule_fsync = false; TAILQ_FOREACH_SAFE(sjob, &ki->kaio_syncqueue, list, sjobn) { if (job->fd_file != sjob->fd_file || job->seqno >= sjob->seqno) continue; if (--sjob->pending > 0) continue; TAILQ_REMOVE(&ki->kaio_syncqueue, sjob, list); if (!aio_clear_cancel_function_locked(sjob)) continue; TAILQ_INSERT_TAIL(&ki->kaio_syncready, sjob, list); schedule_fsync = true; } if (schedule_fsync) taskqueue_enqueue(taskqueue_aiod_kick, &ki->kaio_sync_task); } if (ki->kaio_flags & KAIO_WAKEUP) { ki->kaio_flags &= ~KAIO_WAKEUP; wakeup(&userp->p_aioinfo); } } static void aio_schedule_fsync(void *context, int pending) { struct kaioinfo *ki; struct kaiocb *job; ki = context; AIO_LOCK(ki); while (!TAILQ_EMPTY(&ki->kaio_syncready)) { job = TAILQ_FIRST(&ki->kaio_syncready); TAILQ_REMOVE(&ki->kaio_syncready, job, list); AIO_UNLOCK(ki); aio_schedule(job, aio_process_sync); AIO_LOCK(ki); } AIO_UNLOCK(ki); } bool aio_cancel_cleared(struct kaiocb *job) { /* * The caller should hold the same queue lock held when * aio_clear_cancel_function() was called and set this flag * ensuring this check sees an up-to-date value. However, * there is no way to assert that. */ return ((job->jobflags & KAIOCB_CLEARED) != 0); } static bool aio_clear_cancel_function_locked(struct kaiocb *job) { AIO_LOCK_ASSERT(job->userproc->p_aioinfo, MA_OWNED); MPASS(job->cancel_fn != NULL); if (job->jobflags & KAIOCB_CANCELLING) { job->jobflags |= KAIOCB_CLEARED; return (false); } job->cancel_fn = NULL; return (true); } bool aio_clear_cancel_function(struct kaiocb *job) { struct kaioinfo *ki; bool ret; ki = job->userproc->p_aioinfo; AIO_LOCK(ki); ret = aio_clear_cancel_function_locked(job); AIO_UNLOCK(ki); return (ret); } static bool aio_set_cancel_function_locked(struct kaiocb *job, aio_cancel_fn_t *func) { AIO_LOCK_ASSERT(job->userproc->p_aioinfo, MA_OWNED); if (job->jobflags & KAIOCB_CANCELLED) return (false); job->cancel_fn = func; return (true); } bool aio_set_cancel_function(struct kaiocb *job, aio_cancel_fn_t *func) { struct kaioinfo *ki; bool ret; ki = job->userproc->p_aioinfo; AIO_LOCK(ki); ret = aio_set_cancel_function_locked(job, func); AIO_UNLOCK(ki); return (ret); } void aio_complete(struct kaiocb *job, long status, int error) { struct kaioinfo *ki; struct proc *userp; job->uaiocb._aiocb_private.error = error; job->uaiocb._aiocb_private.status = status; userp = job->userproc; ki = userp->p_aioinfo; AIO_LOCK(ki); KASSERT(!(job->jobflags & KAIOCB_FINISHED), ("duplicate aio_complete")); job->jobflags |= KAIOCB_FINISHED; if ((job->jobflags & (KAIOCB_QUEUEING | KAIOCB_CANCELLING)) == 0) { TAILQ_REMOVE(&ki->kaio_jobqueue, job, plist); aio_bio_done_notify(userp, job); } AIO_UNLOCK(ki); } void aio_cancel(struct kaiocb *job) { aio_complete(job, -1, ECANCELED); } void aio_switch_vmspace(struct kaiocb *job) { vmspace_switch_aio(job->userproc->p_vmspace); } /* * The AIO daemon, most of the actual work is done in aio_process_*, * but the setup (and address space mgmt) is done in this routine. */ static void aio_daemon(void *_id) { struct kaiocb *job; struct aioproc *aiop; struct kaioinfo *ki; struct proc *p; struct vmspace *myvm; struct thread *td = curthread; int id = (intptr_t)_id; /* * Grab an extra reference on the daemon's vmspace so that it * doesn't get freed by jobs that switch to a different * vmspace. */ p = td->td_proc; myvm = vmspace_acquire_ref(p); KASSERT(p->p_textvp == NULL, ("kthread has a textvp")); /* * Allocate and ready the aio control info. There is one aiop structure * per daemon. */ aiop = malloc(sizeof(*aiop), M_AIO, M_WAITOK); aiop->aioproc = p; aiop->aioprocflags = 0; /* * Wakeup parent process. (Parent sleeps to keep from blasting away * and creating too many daemons.) */ sema_post(&aio_newproc_sem); mtx_lock(&aio_job_mtx); for (;;) { /* * Take daemon off of free queue */ if (aiop->aioprocflags & AIOP_FREE) { TAILQ_REMOVE(&aio_freeproc, aiop, list); aiop->aioprocflags &= ~AIOP_FREE; } /* * Check for jobs. */ while ((job = aio_selectjob(aiop)) != NULL) { mtx_unlock(&aio_job_mtx); ki = job->userproc->p_aioinfo; job->handle_fn(job); mtx_lock(&aio_job_mtx); /* Decrement the active job count. */ ki->kaio_active_count--; } /* * Disconnect from user address space. */ if (p->p_vmspace != myvm) { mtx_unlock(&aio_job_mtx); vmspace_switch_aio(myvm); mtx_lock(&aio_job_mtx); /* * We have to restart to avoid race, we only sleep if * no job can be selected. */ continue; } mtx_assert(&aio_job_mtx, MA_OWNED); TAILQ_INSERT_HEAD(&aio_freeproc, aiop, list); aiop->aioprocflags |= AIOP_FREE; /* * If daemon is inactive for a long time, allow it to exit, * thereby freeing resources. */ if (msleep(p, &aio_job_mtx, PRIBIO, "aiordy", aiod_lifetime) == EWOULDBLOCK && TAILQ_EMPTY(&aio_jobs) && (aiop->aioprocflags & AIOP_FREE) && num_aio_procs > target_aio_procs) break; } TAILQ_REMOVE(&aio_freeproc, aiop, list); num_aio_procs--; mtx_unlock(&aio_job_mtx); free(aiop, M_AIO); free_unr(aiod_unr, id); vmspace_free(myvm); KASSERT(p->p_vmspace == myvm, ("AIOD: bad vmspace for exiting daemon")); KASSERT(refcount_load(&myvm->vm_refcnt) > 1, ("AIOD: bad vm refcnt for exiting daemon: %d", refcount_load(&myvm->vm_refcnt))); kproc_exit(0); } /* * Create a new AIO daemon. This is mostly a kernel-thread fork routine. The * AIO daemon modifies its environment itself. */ static int aio_newproc(int *start) { int error; struct proc *p; int id; id = alloc_unr(aiod_unr); error = kproc_create(aio_daemon, (void *)(intptr_t)id, &p, RFNOWAIT, 0, "aiod%d", id); if (error == 0) { /* * Wait until daemon is started. */ sema_wait(&aio_newproc_sem); mtx_lock(&aio_job_mtx); num_aio_procs++; if (start != NULL) (*start)--; mtx_unlock(&aio_job_mtx); } else { free_unr(aiod_unr, id); } return (error); } /* * Try the high-performance, low-overhead bio method for eligible * VCHR devices. This method doesn't use an aio helper thread, and * thus has very low overhead. * * Assumes that the caller, aio_aqueue(), has incremented the file * structure's reference count, preventing its deallocation for the * duration of this call. */ static int aio_qbio(struct proc *p, struct kaiocb *job) { struct aiocb *cb; struct file *fp; struct buf *pbuf; struct vnode *vp; struct cdevsw *csw; struct cdev *dev; struct kaioinfo *ki; struct bio **bios = NULL; off_t offset; int bio_cmd, error, i, iovcnt, opcode, poff, ref; vm_prot_t prot; bool use_unmapped; cb = &job->uaiocb; fp = job->fd_file; opcode = cb->aio_lio_opcode; if (!(opcode == LIO_WRITE || opcode == LIO_WRITEV || opcode == LIO_READ || opcode == LIO_READV)) return (-1); if (fp == NULL || fp->f_type != DTYPE_VNODE) return (-1); vp = fp->f_vnode; if (vp->v_type != VCHR) return (-1); if (vp->v_bufobj.bo_bsize == 0) return (-1); bio_cmd = (opcode & LIO_WRITE) ? BIO_WRITE : BIO_READ; iovcnt = job->uiop->uio_iovcnt; if (iovcnt > max_buf_aio) return (-1); for (i = 0; i < iovcnt; i++) { if (job->uiop->uio_iov[i].iov_len % vp->v_bufobj.bo_bsize != 0) return (-1); if (job->uiop->uio_iov[i].iov_len > maxphys) { error = -1; return (-1); } } offset = cb->aio_offset; ref = 0; csw = devvn_refthread(vp, &dev, &ref); if (csw == NULL) return (ENXIO); if ((csw->d_flags & D_DISK) == 0) { error = -1; goto unref; } if (job->uiop->uio_resid > dev->si_iosize_max) { error = -1; goto unref; } ki = p->p_aioinfo; job->error = 0; use_unmapped = (dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed; if (!use_unmapped) { AIO_LOCK(ki); if (ki->kaio_buffer_count + iovcnt > max_buf_aio) { AIO_UNLOCK(ki); error = EAGAIN; goto unref; } ki->kaio_buffer_count += iovcnt; AIO_UNLOCK(ki); } bios = malloc(sizeof(struct bio *) * iovcnt, M_TEMP, M_WAITOK); atomic_store_int(&job->nbio, iovcnt); for (i = 0; i < iovcnt; i++) { struct vm_page** pages; struct bio *bp; void *buf; size_t nbytes; int npages; buf = job->uiop->uio_iov[i].iov_base; nbytes = job->uiop->uio_iov[i].iov_len; bios[i] = g_alloc_bio(); bp = bios[i]; poff = (vm_offset_t)buf & PAGE_MASK; if (use_unmapped) { pbuf = NULL; pages = malloc(sizeof(vm_page_t) * (atop(round_page( nbytes)) + 1), M_TEMP, M_WAITOK | M_ZERO); } else { pbuf = uma_zalloc(pbuf_zone, M_WAITOK); BUF_KERNPROC(pbuf); pages = pbuf->b_pages; } bp->bio_length = nbytes; bp->bio_bcount = nbytes; bp->bio_done = aio_biowakeup; bp->bio_offset = offset; bp->bio_cmd = bio_cmd; bp->bio_dev = dev; bp->bio_caller1 = job; bp->bio_caller2 = pbuf; prot = VM_PROT_READ; if (opcode == LIO_READ || opcode == LIO_READV) prot |= VM_PROT_WRITE; /* Less backwards than it looks */ npages = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map, (vm_offset_t)buf, bp->bio_length, prot, pages, atop(maxphys) + 1); if (npages < 0) { if (pbuf != NULL) uma_zfree(pbuf_zone, pbuf); else free(pages, M_TEMP); error = EFAULT; g_destroy_bio(bp); i--; goto destroy_bios; } if (pbuf != NULL) { pmap_qenter((vm_offset_t)pbuf->b_data, pages, npages); bp->bio_data = pbuf->b_data + poff; pbuf->b_npages = npages; atomic_add_int(&num_buf_aio, 1); } else { bp->bio_ma = pages; bp->bio_ma_n = npages; bp->bio_ma_offset = poff; bp->bio_data = unmapped_buf; bp->bio_flags |= BIO_UNMAPPED; atomic_add_int(&num_unmapped_aio, 1); } offset += nbytes; } /* Perform transfer. */ for (i = 0; i < iovcnt; i++) csw->d_strategy(bios[i]); free(bios, M_TEMP); dev_relthread(dev, ref); return (0); destroy_bios: for (; i >= 0; i--) aio_biocleanup(bios[i]); free(bios, M_TEMP); unref: dev_relthread(dev, ref); return (error); } #ifdef COMPAT_FREEBSD6 static int convert_old_sigevent(struct osigevent *osig, struct sigevent *nsig) { /* * Only SIGEV_NONE, SIGEV_SIGNAL, and SIGEV_KEVENT are * supported by AIO with the old sigevent structure. */ nsig->sigev_notify = osig->sigev_notify; switch (nsig->sigev_notify) { case SIGEV_NONE: break; case SIGEV_SIGNAL: nsig->sigev_signo = osig->__sigev_u.__sigev_signo; break; case SIGEV_KEVENT: nsig->sigev_notify_kqueue = osig->__sigev_u.__sigev_notify_kqueue; nsig->sigev_value.sival_ptr = osig->sigev_value.sival_ptr; break; default: return (EINVAL); } return (0); } static int aiocb_copyin_old_sigevent(struct aiocb *ujob, struct kaiocb *kjob, int type __unused) { struct oaiocb *ojob; struct aiocb *kcb = &kjob->uaiocb; int error; bzero(kcb, sizeof(struct aiocb)); error = copyin(ujob, kcb, sizeof(struct oaiocb)); if (error) return (error); /* No need to copyin aio_iov, because it did not exist in FreeBSD 6 */ ojob = (struct oaiocb *)kcb; return (convert_old_sigevent(&ojob->aio_sigevent, &kcb->aio_sigevent)); } #endif static int aiocb_copyin(struct aiocb *ujob, struct kaiocb *kjob, int type) { struct aiocb *kcb = &kjob->uaiocb; int error; error = copyin(ujob, kcb, sizeof(struct aiocb)); if (error) return (error); if (type == LIO_NOP) type = kcb->aio_lio_opcode; if (type & LIO_VECTORED) { /* malloc a uio and copy in the iovec */ error = copyinuio(__DEVOLATILE(struct iovec*, kcb->aio_iov), kcb->aio_iovcnt, &kjob->uiop); } return (error); } static long aiocb_fetch_status(struct aiocb *ujob) { return (fuword(&ujob->_aiocb_private.status)); } static long aiocb_fetch_error(struct aiocb *ujob) { return (fuword(&ujob->_aiocb_private.error)); } static int aiocb_store_status(struct aiocb *ujob, long status) { return (suword(&ujob->_aiocb_private.status, status)); } static int aiocb_store_error(struct aiocb *ujob, long error) { return (suword(&ujob->_aiocb_private.error, error)); } static int aiocb_store_kernelinfo(struct aiocb *ujob, long jobref) { return (suword(&ujob->_aiocb_private.kernelinfo, jobref)); } static int aiocb_store_aiocb(struct aiocb **ujobp, struct aiocb *ujob) { return (suword(ujobp, (long)ujob)); } static struct aiocb_ops aiocb_ops = { .aio_copyin = aiocb_copyin, .fetch_status = aiocb_fetch_status, .fetch_error = aiocb_fetch_error, .store_status = aiocb_store_status, .store_error = aiocb_store_error, .store_kernelinfo = aiocb_store_kernelinfo, .store_aiocb = aiocb_store_aiocb, }; #ifdef COMPAT_FREEBSD6 static struct aiocb_ops aiocb_ops_osigevent = { .aio_copyin = aiocb_copyin_old_sigevent, .fetch_status = aiocb_fetch_status, .fetch_error = aiocb_fetch_error, .store_status = aiocb_store_status, .store_error = aiocb_store_error, .store_kernelinfo = aiocb_store_kernelinfo, .store_aiocb = aiocb_store_aiocb, }; #endif /* * Queue a new AIO request. Choosing either the threaded or direct bio VCHR * technique is done in this code. */ int aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, int type, struct aiocb_ops *ops) { struct proc *p = td->td_proc; struct file *fp = NULL; struct kaiocb *job; struct kaioinfo *ki; struct kevent kev; int opcode; int error; int fd, kqfd; int jid; u_short evflags; if (p->p_aioinfo == NULL) aio_init_aioinfo(p); ki = p->p_aioinfo; ops->store_status(ujob, -1); ops->store_error(ujob, 0); ops->store_kernelinfo(ujob, -1); if (num_queue_count >= max_queue_count || ki->kaio_count >= max_aio_queue_per_proc) { error = EAGAIN; goto err1; } job = uma_zalloc(aiocb_zone, M_WAITOK | M_ZERO); knlist_init_mtx(&job->klist, AIO_MTX(ki)); error = ops->aio_copyin(ujob, job, type); if (error) goto err2; if (job->uaiocb.aio_nbytes > IOSIZE_MAX) { error = EINVAL; goto err2; } if (job->uaiocb.aio_sigevent.sigev_notify != SIGEV_KEVENT && job->uaiocb.aio_sigevent.sigev_notify != SIGEV_SIGNAL && job->uaiocb.aio_sigevent.sigev_notify != SIGEV_THREAD_ID && job->uaiocb.aio_sigevent.sigev_notify != SIGEV_NONE) { error = EINVAL; goto err2; } if ((job->uaiocb.aio_sigevent.sigev_notify == SIGEV_SIGNAL || job->uaiocb.aio_sigevent.sigev_notify == SIGEV_THREAD_ID) && !_SIG_VALID(job->uaiocb.aio_sigevent.sigev_signo)) { error = EINVAL; goto err2; } /* Get the opcode. */ if (type == LIO_NOP) { switch (job->uaiocb.aio_lio_opcode) { case LIO_WRITE: case LIO_WRITEV: case LIO_NOP: case LIO_READ: case LIO_READV: opcode = job->uaiocb.aio_lio_opcode; break; default: error = EINVAL; goto err2; } } else opcode = job->uaiocb.aio_lio_opcode = type; ksiginfo_init(&job->ksi); /* Save userspace address of the job info. */ job->ujob = ujob; /* * Validate the opcode and fetch the file object for the specified * file descriptor. * * XXXRW: Moved the opcode validation up here so that we don't * retrieve a file descriptor without knowing what the capabiltity * should be. */ fd = job->uaiocb.aio_fildes; switch (opcode) { case LIO_WRITE: case LIO_WRITEV: error = fget_write(td, fd, &cap_pwrite_rights, &fp); break; case LIO_READ: case LIO_READV: error = fget_read(td, fd, &cap_pread_rights, &fp); break; case LIO_SYNC: case LIO_DSYNC: error = fget(td, fd, &cap_fsync_rights, &fp); break; case LIO_MLOCK: break; case LIO_NOP: error = fget(td, fd, &cap_no_rights, &fp); break; default: error = EINVAL; } if (error) goto err3; if ((opcode & LIO_SYNC) && fp->f_vnode == NULL) { error = EINVAL; goto err3; } if ((opcode == LIO_READ || opcode == LIO_READV || opcode == LIO_WRITE || opcode == LIO_WRITEV) && job->uaiocb.aio_offset < 0 && (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) { error = EINVAL; goto err3; } if (fp != NULL && fp->f_ops == &path_fileops) { error = EBADF; goto err3; } job->fd_file = fp; mtx_lock(&aio_job_mtx); jid = jobrefid++; job->seqno = jobseqno++; mtx_unlock(&aio_job_mtx); error = ops->store_kernelinfo(ujob, jid); if (error) { error = EINVAL; goto err3; } job->uaiocb._aiocb_private.kernelinfo = (void *)(intptr_t)jid; if (opcode == LIO_NOP) { fdrop(fp, td); MPASS(job->uiop == &job->uio || job->uiop == NULL); uma_zfree(aiocb_zone, job); return (0); } if (job->uaiocb.aio_sigevent.sigev_notify != SIGEV_KEVENT) goto no_kqueue; evflags = job->uaiocb.aio_sigevent.sigev_notify_kevent_flags; if ((evflags & ~(EV_CLEAR | EV_DISPATCH | EV_ONESHOT)) != 0) { error = EINVAL; goto err3; } kqfd = job->uaiocb.aio_sigevent.sigev_notify_kqueue; memset(&kev, 0, sizeof(kev)); kev.ident = (uintptr_t)job->ujob; kev.filter = EVFILT_AIO; kev.flags = EV_ADD | EV_ENABLE | EV_FLAG1 | evflags; kev.data = (intptr_t)job; kev.udata = job->uaiocb.aio_sigevent.sigev_value.sival_ptr; error = kqfd_register(kqfd, &kev, td, M_WAITOK); if (error) goto err3; no_kqueue: ops->store_error(ujob, EINPROGRESS); job->uaiocb._aiocb_private.error = EINPROGRESS; job->userproc = p; job->cred = crhold(td->td_ucred); job->jobflags = KAIOCB_QUEUEING; job->lio = lj; if (opcode & LIO_VECTORED) { /* Use the uio copied in by aio_copyin */ MPASS(job->uiop != &job->uio && job->uiop != NULL); } else { /* Setup the inline uio */ job->iov[0].iov_base = (void *)(uintptr_t)job->uaiocb.aio_buf; job->iov[0].iov_len = job->uaiocb.aio_nbytes; job->uio.uio_iov = job->iov; job->uio.uio_iovcnt = 1; job->uio.uio_resid = job->uaiocb.aio_nbytes; job->uio.uio_segflg = UIO_USERSPACE; job->uiop = &job->uio; } switch (opcode & (LIO_READ | LIO_WRITE)) { case LIO_READ: job->uiop->uio_rw = UIO_READ; break; case LIO_WRITE: job->uiop->uio_rw = UIO_WRITE; break; } job->uiop->uio_offset = job->uaiocb.aio_offset; job->uiop->uio_td = td; if (opcode == LIO_MLOCK) { aio_schedule(job, aio_process_mlock); error = 0; } else if (fp->f_ops->fo_aio_queue == NULL) error = aio_queue_file(fp, job); else error = fo_aio_queue(fp, job); if (error) goto err4; AIO_LOCK(ki); job->jobflags &= ~KAIOCB_QUEUEING; TAILQ_INSERT_TAIL(&ki->kaio_all, job, allist); ki->kaio_count++; if (lj) lj->lioj_count++; atomic_add_int(&num_queue_count, 1); if (job->jobflags & KAIOCB_FINISHED) { /* * The queue callback completed the request synchronously. * The bulk of the completion is deferred in that case * until this point. */ aio_bio_done_notify(p, job); } else TAILQ_INSERT_TAIL(&ki->kaio_jobqueue, job, plist); AIO_UNLOCK(ki); return (0); err4: crfree(job->cred); err3: if (fp) fdrop(fp, td); knlist_delete(&job->klist, curthread, 0); err2: if (job->uiop != &job->uio) free(job->uiop, M_IOV); uma_zfree(aiocb_zone, job); err1: ops->store_error(ujob, error); return (error); } static void aio_cancel_daemon_job(struct kaiocb *job) { mtx_lock(&aio_job_mtx); if (!aio_cancel_cleared(job)) TAILQ_REMOVE(&aio_jobs, job, list); mtx_unlock(&aio_job_mtx); aio_cancel(job); } void aio_schedule(struct kaiocb *job, aio_handle_fn_t *func) { mtx_lock(&aio_job_mtx); if (!aio_set_cancel_function(job, aio_cancel_daemon_job)) { mtx_unlock(&aio_job_mtx); aio_cancel(job); return; } job->handle_fn = func; TAILQ_INSERT_TAIL(&aio_jobs, job, list); aio_kick_nowait(job->userproc); mtx_unlock(&aio_job_mtx); } static void aio_cancel_sync(struct kaiocb *job) { struct kaioinfo *ki; ki = job->userproc->p_aioinfo; AIO_LOCK(ki); if (!aio_cancel_cleared(job)) TAILQ_REMOVE(&ki->kaio_syncqueue, job, list); AIO_UNLOCK(ki); aio_cancel(job); } int aio_queue_file(struct file *fp, struct kaiocb *job) { struct kaioinfo *ki; struct kaiocb *job2; struct vnode *vp; struct mount *mp; int error; bool safe; ki = job->userproc->p_aioinfo; error = aio_qbio(job->userproc, job); if (error >= 0) return (error); safe = false; if (fp->f_type == DTYPE_VNODE) { vp = fp->f_vnode; if (vp->v_type == VREG || vp->v_type == VDIR) { mp = fp->f_vnode->v_mount; if (mp == NULL || (mp->mnt_flag & MNT_LOCAL) != 0) safe = true; } } if (!(safe || enable_aio_unsafe)) { counted_warning(&unsafe_warningcnt, "is attempting to use unsafe AIO requests"); return (EOPNOTSUPP); } if (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) { aio_schedule(job, aio_process_rw); error = 0; } else if (job->uaiocb.aio_lio_opcode & LIO_SYNC) { AIO_LOCK(ki); TAILQ_FOREACH(job2, &ki->kaio_jobqueue, plist) { if (job2->fd_file == job->fd_file && ((job2->uaiocb.aio_lio_opcode & LIO_SYNC) == 0) && job2->seqno < job->seqno) { job2->jobflags |= KAIOCB_CHECKSYNC; job->pending++; } } if (job->pending != 0) { if (!aio_set_cancel_function_locked(job, aio_cancel_sync)) { AIO_UNLOCK(ki); aio_cancel(job); return (0); } TAILQ_INSERT_TAIL(&ki->kaio_syncqueue, job, list); AIO_UNLOCK(ki); return (0); } AIO_UNLOCK(ki); aio_schedule(job, aio_process_sync); error = 0; } else { error = EINVAL; } return (error); } static void aio_kick_nowait(struct proc *userp) { struct kaioinfo *ki = userp->p_aioinfo; struct aioproc *aiop; mtx_assert(&aio_job_mtx, MA_OWNED); if ((aiop = TAILQ_FIRST(&aio_freeproc)) != NULL) { TAILQ_REMOVE(&aio_freeproc, aiop, list); aiop->aioprocflags &= ~AIOP_FREE; wakeup(aiop->aioproc); } else if (num_aio_resv_start + num_aio_procs < max_aio_procs && ki->kaio_active_count + num_aio_resv_start < max_aio_per_proc) { taskqueue_enqueue(taskqueue_aiod_kick, &ki->kaio_task); } } static int aio_kick(struct proc *userp) { struct kaioinfo *ki = userp->p_aioinfo; struct aioproc *aiop; int error, ret = 0; mtx_assert(&aio_job_mtx, MA_OWNED); retryproc: if ((aiop = TAILQ_FIRST(&aio_freeproc)) != NULL) { TAILQ_REMOVE(&aio_freeproc, aiop, list); aiop->aioprocflags &= ~AIOP_FREE; wakeup(aiop->aioproc); } else if (num_aio_resv_start + num_aio_procs < max_aio_procs && ki->kaio_active_count + num_aio_resv_start < max_aio_per_proc) { num_aio_resv_start++; mtx_unlock(&aio_job_mtx); error = aio_newproc(&num_aio_resv_start); mtx_lock(&aio_job_mtx); if (error) { num_aio_resv_start--; goto retryproc; } } else { ret = -1; } return (ret); } static void aio_kick_helper(void *context, int pending) { struct proc *userp = context; mtx_lock(&aio_job_mtx); while (--pending >= 0) { if (aio_kick(userp)) break; } mtx_unlock(&aio_job_mtx); } /* * Support the aio_return system call, as a side-effect, kernel resources are * released. */ static int kern_aio_return(struct thread *td, struct aiocb *ujob, struct aiocb_ops *ops) { struct proc *p = td->td_proc; struct kaiocb *job; struct kaioinfo *ki; long status, error; ki = p->p_aioinfo; if (ki == NULL) return (EINVAL); AIO_LOCK(ki); TAILQ_FOREACH(job, &ki->kaio_done, plist) { if (job->ujob == ujob) break; } if (job != NULL) { MPASS(job->jobflags & KAIOCB_FINISHED); status = job->uaiocb._aiocb_private.status; error = job->uaiocb._aiocb_private.error; td->td_retval[0] = status; td->td_ru.ru_oublock += job->outblock; td->td_ru.ru_inblock += job->inblock; td->td_ru.ru_msgsnd += job->msgsnd; td->td_ru.ru_msgrcv += job->msgrcv; aio_free_entry(job); AIO_UNLOCK(ki); ops->store_error(ujob, error); ops->store_status(ujob, status); } else { error = EINVAL; AIO_UNLOCK(ki); } return (error); } int sys_aio_return(struct thread *td, struct aio_return_args *uap) { return (kern_aio_return(td, uap->aiocbp, &aiocb_ops)); } /* * Allow a process to wakeup when any of the I/O requests are completed. */ static int kern_aio_suspend(struct thread *td, int njoblist, struct aiocb **ujoblist, struct timespec *ts) { struct proc *p = td->td_proc; struct timeval atv; struct kaioinfo *ki; struct kaiocb *firstjob, *job; int error, i, timo; timo = 0; if (ts) { if (ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000) return (EINVAL); TIMESPEC_TO_TIMEVAL(&atv, ts); if (itimerfix(&atv)) return (EINVAL); timo = tvtohz(&atv); } ki = p->p_aioinfo; if (ki == NULL) return (EAGAIN); if (njoblist == 0) return (0); AIO_LOCK(ki); for (;;) { firstjob = NULL; error = 0; TAILQ_FOREACH(job, &ki->kaio_all, allist) { for (i = 0; i < njoblist; i++) { if (job->ujob == ujoblist[i]) { if (firstjob == NULL) firstjob = job; if (job->jobflags & KAIOCB_FINISHED) goto RETURN; } } } /* All tasks were finished. */ if (firstjob == NULL) break; ki->kaio_flags |= KAIO_WAKEUP; error = msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO | PCATCH, "aiospn", timo); if (error == ERESTART) error = EINTR; if (error) break; } RETURN: AIO_UNLOCK(ki); return (error); } int sys_aio_suspend(struct thread *td, struct aio_suspend_args *uap) { struct timespec ts, *tsp; struct aiocb **ujoblist; int error; if (uap->nent < 0 || uap->nent > max_aio_queue_per_proc) return (EINVAL); if (uap->timeout) { /* Get timespec struct. */ if ((error = copyin(uap->timeout, &ts, sizeof(ts))) != 0) return (error); tsp = &ts; } else tsp = NULL; ujoblist = malloc(uap->nent * sizeof(ujoblist[0]), M_AIO, M_WAITOK); error = copyin(uap->aiocbp, ujoblist, uap->nent * sizeof(ujoblist[0])); if (error == 0) error = kern_aio_suspend(td, uap->nent, ujoblist, tsp); free(ujoblist, M_AIO); return (error); } /* * aio_cancel cancels any non-bio aio operations not currently in progress. */ int sys_aio_cancel(struct thread *td, struct aio_cancel_args *uap) { struct proc *p = td->td_proc; struct kaioinfo *ki; struct kaiocb *job, *jobn; struct file *fp; int error; int cancelled = 0; int notcancelled = 0; struct vnode *vp; /* Lookup file object. */ error = fget(td, uap->fd, &cap_no_rights, &fp); if (error) return (error); ki = p->p_aioinfo; if (ki == NULL) goto done; if (fp->f_type == DTYPE_VNODE) { vp = fp->f_vnode; if (vn_isdisk(vp)) { fdrop(fp, td); td->td_retval[0] = AIO_NOTCANCELED; return (0); } } AIO_LOCK(ki); TAILQ_FOREACH_SAFE(job, &ki->kaio_jobqueue, plist, jobn) { if ((uap->fd == job->uaiocb.aio_fildes) && ((uap->aiocbp == NULL) || (uap->aiocbp == job->ujob))) { if (aio_cancel_job(p, ki, job)) { cancelled++; } else { notcancelled++; } if (uap->aiocbp != NULL) break; } } AIO_UNLOCK(ki); done: fdrop(fp, td); if (uap->aiocbp != NULL) { if (cancelled) { td->td_retval[0] = AIO_CANCELED; return (0); } } if (notcancelled) { td->td_retval[0] = AIO_NOTCANCELED; return (0); } if (cancelled) { td->td_retval[0] = AIO_CANCELED; return (0); } td->td_retval[0] = AIO_ALLDONE; return (0); } /* * aio_error is implemented in the kernel level for compatibility purposes * only. For a user mode async implementation, it would be best to do it in * a userland subroutine. */ static int kern_aio_error(struct thread *td, struct aiocb *ujob, struct aiocb_ops *ops) { struct proc *p = td->td_proc; struct kaiocb *job; struct kaioinfo *ki; int status; ki = p->p_aioinfo; if (ki == NULL) { td->td_retval[0] = EINVAL; return (0); } AIO_LOCK(ki); TAILQ_FOREACH(job, &ki->kaio_all, allist) { if (job->ujob == ujob) { if (job->jobflags & KAIOCB_FINISHED) td->td_retval[0] = job->uaiocb._aiocb_private.error; else td->td_retval[0] = EINPROGRESS; AIO_UNLOCK(ki); return (0); } } AIO_UNLOCK(ki); /* * Hack for failure of aio_aqueue. */ status = ops->fetch_status(ujob); if (status == -1) { td->td_retval[0] = ops->fetch_error(ujob); return (0); } td->td_retval[0] = EINVAL; return (0); } int sys_aio_error(struct thread *td, struct aio_error_args *uap) { return (kern_aio_error(td, uap->aiocbp, &aiocb_ops)); } /* syscall - asynchronous read from a file (REALTIME) */ #ifdef COMPAT_FREEBSD6 int freebsd6_aio_read(struct thread *td, struct freebsd6_aio_read_args *uap) { return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READ, &aiocb_ops_osigevent)); } #endif int sys_aio_read(struct thread *td, struct aio_read_args *uap) { return (aio_aqueue(td, uap->aiocbp, NULL, LIO_READ, &aiocb_ops)); } int sys_aio_readv(struct thread *td, struct aio_readv_args *uap) { return (aio_aqueue(td, uap->aiocbp, NULL, LIO_READV, &aiocb_ops)); } /* syscall - asynchronous write to a file (REALTIME) */ #ifdef COMPAT_FREEBSD6 int freebsd6_aio_write(struct thread *td, struct freebsd6_aio_write_args *uap) { return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITE, &aiocb_ops_osigevent)); } #endif int sys_aio_write(struct thread *td, struct aio_write_args *uap) { return (aio_aqueue(td, uap->aiocbp, NULL, LIO_WRITE, &aiocb_ops)); } int sys_aio_writev(struct thread *td, struct aio_writev_args *uap) { return (aio_aqueue(td, uap->aiocbp, NULL, LIO_WRITEV, &aiocb_ops)); } int sys_aio_mlock(struct thread *td, struct aio_mlock_args *uap) { return (aio_aqueue(td, uap->aiocbp, NULL, LIO_MLOCK, &aiocb_ops)); } static int kern_lio_listio(struct thread *td, int mode, struct aiocb * const *uacb_list, struct aiocb **acb_list, int nent, struct sigevent *sig, struct aiocb_ops *ops) { struct proc *p = td->td_proc; struct aiocb *job; struct kaioinfo *ki; struct aioliojob *lj; struct kevent kev; int error; int nagain, nerror; int i; if ((mode != LIO_NOWAIT) && (mode != LIO_WAIT)) return (EINVAL); if (nent < 0 || nent > max_aio_queue_per_proc) return (EINVAL); if (p->p_aioinfo == NULL) aio_init_aioinfo(p); ki = p->p_aioinfo; lj = uma_zalloc(aiolio_zone, M_WAITOK); lj->lioj_flags = 0; lj->lioj_count = 0; lj->lioj_finished_count = 0; lj->lioj_signal.sigev_notify = SIGEV_NONE; knlist_init_mtx(&lj->klist, AIO_MTX(ki)); ksiginfo_init(&lj->lioj_ksi); /* * Setup signal. */ if (sig && (mode == LIO_NOWAIT)) { bcopy(sig, &lj->lioj_signal, sizeof(lj->lioj_signal)); if (lj->lioj_signal.sigev_notify == SIGEV_KEVENT) { /* Assume only new style KEVENT */ memset(&kev, 0, sizeof(kev)); kev.filter = EVFILT_LIO; kev.flags = EV_ADD | EV_ENABLE | EV_FLAG1; kev.ident = (uintptr_t)uacb_list; /* something unique */ kev.data = (intptr_t)lj; /* pass user defined sigval data */ kev.udata = lj->lioj_signal.sigev_value.sival_ptr; error = kqfd_register( lj->lioj_signal.sigev_notify_kqueue, &kev, td, M_WAITOK); if (error) { uma_zfree(aiolio_zone, lj); return (error); } } else if (lj->lioj_signal.sigev_notify == SIGEV_NONE) { ; } else if (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL || lj->lioj_signal.sigev_notify == SIGEV_THREAD_ID) { if (!_SIG_VALID(lj->lioj_signal.sigev_signo)) { uma_zfree(aiolio_zone, lj); return EINVAL; } lj->lioj_flags |= LIOJ_SIGNAL; } else { uma_zfree(aiolio_zone, lj); return EINVAL; } } AIO_LOCK(ki); TAILQ_INSERT_TAIL(&ki->kaio_liojoblist, lj, lioj_list); /* * Add extra aiocb count to avoid the lio to be freed * by other threads doing aio_waitcomplete or aio_return, * and prevent event from being sent until we have queued * all tasks. */ lj->lioj_count = 1; AIO_UNLOCK(ki); /* * Get pointers to the list of I/O requests. */ nagain = 0; nerror = 0; for (i = 0; i < nent; i++) { job = acb_list[i]; if (job != NULL) { error = aio_aqueue(td, job, lj, LIO_NOP, ops); if (error == EAGAIN) nagain++; else if (error != 0) nerror++; } } error = 0; AIO_LOCK(ki); if (mode == LIO_WAIT) { while (lj->lioj_count - 1 != lj->lioj_finished_count) { ki->kaio_flags |= KAIO_WAKEUP; error = msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO | PCATCH, "aiospn", 0); if (error == ERESTART) error = EINTR; if (error) break; } } else { if (lj->lioj_count - 1 == lj->lioj_finished_count) { if (lj->lioj_signal.sigev_notify == SIGEV_KEVENT) { lj->lioj_flags |= LIOJ_KEVENT_POSTED; KNOTE_LOCKED(&lj->klist, 1); } if ((lj->lioj_flags & (LIOJ_SIGNAL | LIOJ_SIGNAL_POSTED)) == LIOJ_SIGNAL && (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL || lj->lioj_signal.sigev_notify == SIGEV_THREAD_ID)) { aio_sendsig(p, &lj->lioj_signal, &lj->lioj_ksi, lj->lioj_count != 1); lj->lioj_flags |= LIOJ_SIGNAL_POSTED; } } } lj->lioj_count--; if (lj->lioj_count == 0) { TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list); knlist_delete(&lj->klist, curthread, 1); PROC_LOCK(p); sigqueue_take(&lj->lioj_ksi); PROC_UNLOCK(p); AIO_UNLOCK(ki); uma_zfree(aiolio_zone, lj); } else AIO_UNLOCK(ki); if (nerror) return (EIO); else if (nagain) return (EAGAIN); else return (error); } /* syscall - list directed I/O (REALTIME) */ #ifdef COMPAT_FREEBSD6 int freebsd6_lio_listio(struct thread *td, struct freebsd6_lio_listio_args *uap) { struct aiocb **acb_list; struct sigevent *sigp, sig; struct osigevent osig; int error, nent; if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT)) return (EINVAL); nent = uap->nent; if (nent < 0 || nent > max_aio_queue_per_proc) return (EINVAL); if (uap->sig && (uap->mode == LIO_NOWAIT)) { error = copyin(uap->sig, &osig, sizeof(osig)); if (error) return (error); error = convert_old_sigevent(&osig, &sig); if (error) return (error); sigp = &sig; } else sigp = NULL; acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK); error = copyin(uap->acb_list, acb_list, nent * sizeof(acb_list[0])); if (error == 0) error = kern_lio_listio(td, uap->mode, (struct aiocb * const *)uap->acb_list, acb_list, nent, sigp, &aiocb_ops_osigevent); free(acb_list, M_LIO); return (error); } #endif /* syscall - list directed I/O (REALTIME) */ int sys_lio_listio(struct thread *td, struct lio_listio_args *uap) { struct aiocb **acb_list; struct sigevent *sigp, sig; int error, nent; if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT)) return (EINVAL); nent = uap->nent; if (nent < 0 || nent > max_aio_queue_per_proc) return (EINVAL); if (uap->sig && (uap->mode == LIO_NOWAIT)) { error = copyin(uap->sig, &sig, sizeof(sig)); if (error) return (error); sigp = &sig; } else sigp = NULL; acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK); error = copyin(uap->acb_list, acb_list, nent * sizeof(acb_list[0])); if (error == 0) error = kern_lio_listio(td, uap->mode, uap->acb_list, acb_list, nent, sigp, &aiocb_ops); free(acb_list, M_LIO); return (error); } static void aio_biocleanup(struct bio *bp) { struct kaiocb *job = (struct kaiocb *)bp->bio_caller1; struct kaioinfo *ki; struct buf *pbuf = (struct buf *)bp->bio_caller2; /* Release mapping into kernel space. */ if (pbuf != NULL) { MPASS(pbuf->b_npages <= atop(maxphys) + 1); pmap_qremove((vm_offset_t)pbuf->b_data, pbuf->b_npages); vm_page_unhold_pages(pbuf->b_pages, pbuf->b_npages); uma_zfree(pbuf_zone, pbuf); atomic_subtract_int(&num_buf_aio, 1); ki = job->userproc->p_aioinfo; AIO_LOCK(ki); ki->kaio_buffer_count--; AIO_UNLOCK(ki); } else { MPASS(bp->bio_ma_n <= atop(maxphys) + 1); vm_page_unhold_pages(bp->bio_ma, bp->bio_ma_n); free(bp->bio_ma, M_TEMP); atomic_subtract_int(&num_unmapped_aio, 1); } g_destroy_bio(bp); } static void aio_biowakeup(struct bio *bp) { struct kaiocb *job = (struct kaiocb *)bp->bio_caller1; size_t nbytes; long bcount = bp->bio_bcount; long resid = bp->bio_resid; int opcode, nblks; int bio_error = bp->bio_error; uint16_t flags = bp->bio_flags; opcode = job->uaiocb.aio_lio_opcode; aio_biocleanup(bp); nbytes =bcount - resid; atomic_add_acq_long(&job->nbytes, nbytes); nblks = btodb(nbytes); /* * If multiple bios experienced an error, the job will reflect the * error of whichever failed bio completed last. */ if (flags & BIO_ERROR) atomic_set_int(&job->error, bio_error); if (opcode & LIO_WRITE) atomic_add_int(&job->outblock, nblks); else atomic_add_int(&job->inblock, nblks); atomic_subtract_int(&job->nbio, 1); if (atomic_load_int(&job->nbio) == 0) { if (atomic_load_int(&job->error)) aio_complete(job, -1, job->error); else aio_complete(job, atomic_load_long(&job->nbytes), 0); } } /* syscall - wait for the next completion of an aio request */ static int kern_aio_waitcomplete(struct thread *td, struct aiocb **ujobp, struct timespec *ts, struct aiocb_ops *ops) { struct proc *p = td->td_proc; struct timeval atv; struct kaioinfo *ki; struct kaiocb *job; struct aiocb *ujob; long error, status; int timo; ops->store_aiocb(ujobp, NULL); if (ts == NULL) { timo = 0; } else if (ts->tv_sec == 0 && ts->tv_nsec == 0) { timo = -1; } else { if ((ts->tv_nsec < 0) || (ts->tv_nsec >= 1000000000)) return (EINVAL); TIMESPEC_TO_TIMEVAL(&atv, ts); if (itimerfix(&atv)) return (EINVAL); timo = tvtohz(&atv); } if (p->p_aioinfo == NULL) aio_init_aioinfo(p); ki = p->p_aioinfo; error = 0; job = NULL; AIO_LOCK(ki); while ((job = TAILQ_FIRST(&ki->kaio_done)) == NULL) { if (timo == -1) { error = EWOULDBLOCK; break; } ki->kaio_flags |= KAIO_WAKEUP; error = msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO | PCATCH, "aiowc", timo); if (timo && error == ERESTART) error = EINTR; if (error) break; } if (job != NULL) { MPASS(job->jobflags & KAIOCB_FINISHED); ujob = job->ujob; status = job->uaiocb._aiocb_private.status; error = job->uaiocb._aiocb_private.error; td->td_retval[0] = status; td->td_ru.ru_oublock += job->outblock; td->td_ru.ru_inblock += job->inblock; td->td_ru.ru_msgsnd += job->msgsnd; td->td_ru.ru_msgrcv += job->msgrcv; aio_free_entry(job); AIO_UNLOCK(ki); ops->store_aiocb(ujobp, ujob); ops->store_error(ujob, error); ops->store_status(ujob, status); } else AIO_UNLOCK(ki); return (error); } int sys_aio_waitcomplete(struct thread *td, struct aio_waitcomplete_args *uap) { struct timespec ts, *tsp; int error; if (uap->timeout) { /* Get timespec struct. */ error = copyin(uap->timeout, &ts, sizeof(ts)); if (error) return (error); tsp = &ts; } else tsp = NULL; return (kern_aio_waitcomplete(td, uap->aiocbp, tsp, &aiocb_ops)); } static int kern_aio_fsync(struct thread *td, int op, struct aiocb *ujob, struct aiocb_ops *ops) { int listop; switch (op) { case O_SYNC: listop = LIO_SYNC; break; case O_DSYNC: listop = LIO_DSYNC; break; default: return (EINVAL); } return (aio_aqueue(td, ujob, NULL, listop, ops)); } int sys_aio_fsync(struct thread *td, struct aio_fsync_args *uap) { return (kern_aio_fsync(td, uap->op, uap->aiocbp, &aiocb_ops)); } /* kqueue attach function */ static int filt_aioattach(struct knote *kn) { struct kaiocb *job; job = (struct kaiocb *)(uintptr_t)kn->kn_sdata; /* * The job pointer must be validated before using it, so * registration is restricted to the kernel; the user cannot * set EV_FLAG1. */ if ((kn->kn_flags & EV_FLAG1) == 0) return (EPERM); kn->kn_ptr.p_aio = job; kn->kn_flags &= ~EV_FLAG1; knlist_add(&job->klist, kn, 0); return (0); } /* kqueue detach function */ static void filt_aiodetach(struct knote *kn) { struct knlist *knl; knl = &kn->kn_ptr.p_aio->klist; knl->kl_lock(knl->kl_lockarg); if (!knlist_empty(knl)) knlist_remove(knl, kn, 1); knl->kl_unlock(knl->kl_lockarg); } /* kqueue filter function */ /*ARGSUSED*/ static int filt_aio(struct knote *kn, long hint) { struct kaiocb *job = kn->kn_ptr.p_aio; kn->kn_data = job->uaiocb._aiocb_private.error; if (!(job->jobflags & KAIOCB_FINISHED)) return (0); kn->kn_flags |= EV_EOF; return (1); } /* kqueue attach function */ static int filt_lioattach(struct knote *kn) { struct aioliojob *lj; lj = (struct aioliojob *)(uintptr_t)kn->kn_sdata; /* * The aioliojob pointer must be validated before using it, so * registration is restricted to the kernel; the user cannot * set EV_FLAG1. */ if ((kn->kn_flags & EV_FLAG1) == 0) return (EPERM); kn->kn_ptr.p_lio = lj; kn->kn_flags &= ~EV_FLAG1; knlist_add(&lj->klist, kn, 0); return (0); } /* kqueue detach function */ static void filt_liodetach(struct knote *kn) { struct knlist *knl; knl = &kn->kn_ptr.p_lio->klist; knl->kl_lock(knl->kl_lockarg); if (!knlist_empty(knl)) knlist_remove(knl, kn, 1); knl->kl_unlock(knl->kl_lockarg); } /* kqueue filter function */ /*ARGSUSED*/ static int filt_lio(struct knote *kn, long hint) { struct aioliojob * lj = kn->kn_ptr.p_lio; return (lj->lioj_flags & LIOJ_KEVENT_POSTED); } #ifdef COMPAT_FREEBSD32 #include #include #include #include #include #include #include #include struct __aiocb_private32 { int32_t status; int32_t error; uint32_t kernelinfo; }; #ifdef COMPAT_FREEBSD6 typedef struct oaiocb32 { int aio_fildes; /* File descriptor */ uint64_t aio_offset __packed; /* File offset for I/O */ uint32_t aio_buf; /* I/O buffer in process space */ uint32_t aio_nbytes; /* Number of bytes for I/O */ struct osigevent32 aio_sigevent; /* Signal to deliver */ int aio_lio_opcode; /* LIO opcode */ int aio_reqprio; /* Request priority -- ignored */ struct __aiocb_private32 _aiocb_private; } oaiocb32_t; #endif typedef struct aiocb32 { int32_t aio_fildes; /* File descriptor */ uint64_t aio_offset __packed; /* File offset for I/O */ uint32_t aio_buf; /* I/O buffer in process space */ uint32_t aio_nbytes; /* Number of bytes for I/O */ int __spare__[2]; uint32_t __spare2__; int aio_lio_opcode; /* LIO opcode */ int aio_reqprio; /* Request priority -- ignored */ struct __aiocb_private32 _aiocb_private; struct sigevent32 aio_sigevent; /* Signal to deliver */ } aiocb32_t; #ifdef COMPAT_FREEBSD6 static int convert_old_sigevent32(struct osigevent32 *osig, struct sigevent *nsig) { /* * Only SIGEV_NONE, SIGEV_SIGNAL, and SIGEV_KEVENT are * supported by AIO with the old sigevent structure. */ CP(*osig, *nsig, sigev_notify); switch (nsig->sigev_notify) { case SIGEV_NONE: break; case SIGEV_SIGNAL: nsig->sigev_signo = osig->__sigev_u.__sigev_signo; break; case SIGEV_KEVENT: nsig->sigev_notify_kqueue = osig->__sigev_u.__sigev_notify_kqueue; PTRIN_CP(*osig, *nsig, sigev_value.sival_ptr); break; default: return (EINVAL); } return (0); } static int aiocb32_copyin_old_sigevent(struct aiocb *ujob, struct kaiocb *kjob, int type __unused) { struct oaiocb32 job32; struct aiocb *kcb = &kjob->uaiocb; int error; bzero(kcb, sizeof(struct aiocb)); error = copyin(ujob, &job32, sizeof(job32)); if (error) return (error); /* No need to copyin aio_iov, because it did not exist in FreeBSD 6 */ CP(job32, *kcb, aio_fildes); CP(job32, *kcb, aio_offset); PTRIN_CP(job32, *kcb, aio_buf); CP(job32, *kcb, aio_nbytes); CP(job32, *kcb, aio_lio_opcode); CP(job32, *kcb, aio_reqprio); CP(job32, *kcb, _aiocb_private.status); CP(job32, *kcb, _aiocb_private.error); PTRIN_CP(job32, *kcb, _aiocb_private.kernelinfo); return (convert_old_sigevent32(&job32.aio_sigevent, &kcb->aio_sigevent)); } #endif static int aiocb32_copyin(struct aiocb *ujob, struct kaiocb *kjob, int type) { struct aiocb32 job32; struct aiocb *kcb = &kjob->uaiocb; struct iovec32 *iov32; int error; error = copyin(ujob, &job32, sizeof(job32)); if (error) return (error); CP(job32, *kcb, aio_fildes); CP(job32, *kcb, aio_offset); CP(job32, *kcb, aio_lio_opcode); if (type == LIO_NOP) type = kcb->aio_lio_opcode; if (type & LIO_VECTORED) { iov32 = PTRIN(job32.aio_iov); CP(job32, *kcb, aio_iovcnt); /* malloc a uio and copy in the iovec */ error = freebsd32_copyinuio(iov32, kcb->aio_iovcnt, &kjob->uiop); if (error) return (error); } else { PTRIN_CP(job32, *kcb, aio_buf); CP(job32, *kcb, aio_nbytes); } CP(job32, *kcb, aio_reqprio); CP(job32, *kcb, _aiocb_private.status); CP(job32, *kcb, _aiocb_private.error); PTRIN_CP(job32, *kcb, _aiocb_private.kernelinfo); error = convert_sigevent32(&job32.aio_sigevent, &kcb->aio_sigevent); return (error); } static long aiocb32_fetch_status(struct aiocb *ujob) { struct aiocb32 *ujob32; ujob32 = (struct aiocb32 *)ujob; return (fuword32(&ujob32->_aiocb_private.status)); } static long aiocb32_fetch_error(struct aiocb *ujob) { struct aiocb32 *ujob32; ujob32 = (struct aiocb32 *)ujob; return (fuword32(&ujob32->_aiocb_private.error)); } static int aiocb32_store_status(struct aiocb *ujob, long status) { struct aiocb32 *ujob32; ujob32 = (struct aiocb32 *)ujob; return (suword32(&ujob32->_aiocb_private.status, status)); } static int aiocb32_store_error(struct aiocb *ujob, long error) { struct aiocb32 *ujob32; ujob32 = (struct aiocb32 *)ujob; return (suword32(&ujob32->_aiocb_private.error, error)); } static int aiocb32_store_kernelinfo(struct aiocb *ujob, long jobref) { struct aiocb32 *ujob32; ujob32 = (struct aiocb32 *)ujob; return (suword32(&ujob32->_aiocb_private.kernelinfo, jobref)); } static int aiocb32_store_aiocb(struct aiocb **ujobp, struct aiocb *ujob) { return (suword32(ujobp, (long)ujob)); } static struct aiocb_ops aiocb32_ops = { .aio_copyin = aiocb32_copyin, .fetch_status = aiocb32_fetch_status, .fetch_error = aiocb32_fetch_error, .store_status = aiocb32_store_status, .store_error = aiocb32_store_error, .store_kernelinfo = aiocb32_store_kernelinfo, .store_aiocb = aiocb32_store_aiocb, }; #ifdef COMPAT_FREEBSD6 static struct aiocb_ops aiocb32_ops_osigevent = { .aio_copyin = aiocb32_copyin_old_sigevent, .fetch_status = aiocb32_fetch_status, .fetch_error = aiocb32_fetch_error, .store_status = aiocb32_store_status, .store_error = aiocb32_store_error, .store_kernelinfo = aiocb32_store_kernelinfo, .store_aiocb = aiocb32_store_aiocb, }; #endif int freebsd32_aio_return(struct thread *td, struct freebsd32_aio_return_args *uap) { return (kern_aio_return(td, (struct aiocb *)uap->aiocbp, &aiocb32_ops)); } int freebsd32_aio_suspend(struct thread *td, struct freebsd32_aio_suspend_args *uap) { struct timespec32 ts32; struct timespec ts, *tsp; struct aiocb **ujoblist; uint32_t *ujoblist32; int error, i; if (uap->nent < 0 || uap->nent > max_aio_queue_per_proc) return (EINVAL); if (uap->timeout) { /* Get timespec struct. */ if ((error = copyin(uap->timeout, &ts32, sizeof(ts32))) != 0) return (error); CP(ts32, ts, tv_sec); CP(ts32, ts, tv_nsec); tsp = &ts; } else tsp = NULL; ujoblist = malloc(uap->nent * sizeof(ujoblist[0]), M_AIO, M_WAITOK); ujoblist32 = (uint32_t *)ujoblist; error = copyin(uap->aiocbp, ujoblist32, uap->nent * sizeof(ujoblist32[0])); if (error == 0) { for (i = uap->nent - 1; i >= 0; i--) ujoblist[i] = PTRIN(ujoblist32[i]); error = kern_aio_suspend(td, uap->nent, ujoblist, tsp); } free(ujoblist, M_AIO); return (error); } int freebsd32_aio_error(struct thread *td, struct freebsd32_aio_error_args *uap) { return (kern_aio_error(td, (struct aiocb *)uap->aiocbp, &aiocb32_ops)); } #ifdef COMPAT_FREEBSD6 int freebsd6_freebsd32_aio_read(struct thread *td, struct freebsd6_freebsd32_aio_read_args *uap) { return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READ, &aiocb32_ops_osigevent)); } #endif int freebsd32_aio_read(struct thread *td, struct freebsd32_aio_read_args *uap) { return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READ, &aiocb32_ops)); } int freebsd32_aio_readv(struct thread *td, struct freebsd32_aio_readv_args *uap) { return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READV, &aiocb32_ops)); } #ifdef COMPAT_FREEBSD6 int freebsd6_freebsd32_aio_write(struct thread *td, struct freebsd6_freebsd32_aio_write_args *uap) { return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITE, &aiocb32_ops_osigevent)); } #endif int freebsd32_aio_write(struct thread *td, struct freebsd32_aio_write_args *uap) { return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITE, &aiocb32_ops)); } int freebsd32_aio_writev(struct thread *td, struct freebsd32_aio_writev_args *uap) { return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITEV, &aiocb32_ops)); } int freebsd32_aio_mlock(struct thread *td, struct freebsd32_aio_mlock_args *uap) { return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_MLOCK, &aiocb32_ops)); } int freebsd32_aio_waitcomplete(struct thread *td, struct freebsd32_aio_waitcomplete_args *uap) { struct timespec32 ts32; struct timespec ts, *tsp; int error; if (uap->timeout) { /* Get timespec struct. */ error = copyin(uap->timeout, &ts32, sizeof(ts32)); if (error) return (error); CP(ts32, ts, tv_sec); CP(ts32, ts, tv_nsec); tsp = &ts; } else tsp = NULL; return (kern_aio_waitcomplete(td, (struct aiocb **)uap->aiocbp, tsp, &aiocb32_ops)); } int freebsd32_aio_fsync(struct thread *td, struct freebsd32_aio_fsync_args *uap) { return (kern_aio_fsync(td, uap->op, (struct aiocb *)uap->aiocbp, &aiocb32_ops)); } #ifdef COMPAT_FREEBSD6 int freebsd6_freebsd32_lio_listio(struct thread *td, struct freebsd6_freebsd32_lio_listio_args *uap) { struct aiocb **acb_list; struct sigevent *sigp, sig; struct osigevent32 osig; uint32_t *acb_list32; int error, i, nent; if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT)) return (EINVAL); nent = uap->nent; if (nent < 0 || nent > max_aio_queue_per_proc) return (EINVAL); if (uap->sig && (uap->mode == LIO_NOWAIT)) { error = copyin(uap->sig, &osig, sizeof(osig)); if (error) return (error); error = convert_old_sigevent32(&osig, &sig); if (error) return (error); sigp = &sig; } else sigp = NULL; acb_list32 = malloc(sizeof(uint32_t) * nent, M_LIO, M_WAITOK); error = copyin(uap->acb_list, acb_list32, nent * sizeof(uint32_t)); if (error) { free(acb_list32, M_LIO); return (error); } acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK); for (i = 0; i < nent; i++) acb_list[i] = PTRIN(acb_list32[i]); free(acb_list32, M_LIO); error = kern_lio_listio(td, uap->mode, (struct aiocb * const *)uap->acb_list, acb_list, nent, sigp, &aiocb32_ops_osigevent); free(acb_list, M_LIO); return (error); } #endif int freebsd32_lio_listio(struct thread *td, struct freebsd32_lio_listio_args *uap) { struct aiocb **acb_list; struct sigevent *sigp, sig; struct sigevent32 sig32; uint32_t *acb_list32; int error, i, nent; if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT)) return (EINVAL); nent = uap->nent; if (nent < 0 || nent > max_aio_queue_per_proc) return (EINVAL); if (uap->sig && (uap->mode == LIO_NOWAIT)) { error = copyin(uap->sig, &sig32, sizeof(sig32)); if (error) return (error); error = convert_sigevent32(&sig32, &sig); if (error) return (error); sigp = &sig; } else sigp = NULL; acb_list32 = malloc(sizeof(uint32_t) * nent, M_LIO, M_WAITOK); error = copyin(uap->acb_list, acb_list32, nent * sizeof(uint32_t)); if (error) { free(acb_list32, M_LIO); return (error); } acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK); for (i = 0; i < nent; i++) acb_list[i] = PTRIN(acb_list32[i]); free(acb_list32, M_LIO); error = kern_lio_listio(td, uap->mode, (struct aiocb * const *)uap->acb_list, acb_list, nent, sigp, &aiocb32_ops); free(acb_list, M_LIO); return (error); } #endif diff --git a/sys/kern/vfs_extattr.c b/sys/kern/vfs_extattr.c index 4c1851a52e88..d389b33d0634 100644 --- a/sys/kern/vfs_extattr.c +++ b/sys/kern/vfs_extattr.c @@ -1,755 +1,755 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1999-2001 Robert N. M. Watson * All rights reserved. * * This software was developed by Robert Watson for the TrustedBSD Project. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int kern_extattr_set_path(struct thread *td, const char *path, int attrnamespace, const char *attrname, void *data, size_t nbytes, int follow); static int kern_extattr_get_path(struct thread *td, const char *path, int attrnamespace, const char *attrname, void *data, size_t nbytes, int follow); static int kern_extattr_delete_path(struct thread *td, const char *path, int attrnamespace, const char *attrname, int follow); static int kern_extattr_list_path(struct thread *td, const char *path, int attrnamespace, void *data, size_t nbytes, int follow); /* * Syscall to push extended attribute configuration information into the VFS. * Accepts a path, which it converts to a mountpoint, as well as a command * (int cmd), and attribute name and misc data. * * Currently this is used only by UFS1 extended attributes. */ #ifndef _SYS_SYSPROTO_H_ struct extattrctl_args { const char *path; int cmd; const char *filename; int attrnamespace; const char *attrname; }; #endif int sys_extattrctl(struct thread *td, struct extattrctl_args *uap) { struct vnode *filename_vp; struct nameidata nd; struct mount *mp, *mp_writable; char attrname[EXTATTR_MAXNAMELEN + 1]; int error; AUDIT_ARG_CMD(uap->cmd); AUDIT_ARG_VALUE(uap->attrnamespace); /* * uap->attrname is not always defined. We check again later when we * invoke the VFS call so as to pass in NULL there if needed. */ if (uap->attrname != NULL) { error = copyinstr(uap->attrname, attrname, sizeof(attrname), NULL); if (error) return (error); } AUDIT_ARG_TEXT(attrname); mp = NULL; filename_vp = NULL; if (uap->filename != NULL) { NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE2, UIO_USERSPACE, uap->filename); error = namei(&nd); if (error) return (error); filename_vp = nd.ni_vp; NDFREE(&nd, NDF_NO_VP_RELE); } /* uap->path is always defined. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path); error = namei(&nd); if (error) goto out; mp = nd.ni_vp->v_mount; error = vfs_busy(mp, 0); if (error) { NDFREE(&nd, 0); mp = NULL; goto out; } VOP_UNLOCK(nd.ni_vp); - error = vn_start_write(nd.ni_vp, &mp_writable, V_WAIT | PCATCH); + error = vn_start_write(nd.ni_vp, &mp_writable, V_WAIT | V_PCATCH); NDFREE(&nd, NDF_NO_VP_UNLOCK); if (error) goto out; if (filename_vp != NULL) { /* * uap->filename is not always defined. If it is, * grab a vnode lock, which VFS_EXTATTRCTL() will * later release. */ error = vn_lock(filename_vp, LK_EXCLUSIVE); if (error) { vn_finished_write(mp_writable); goto out; } } error = VFS_EXTATTRCTL(mp, uap->cmd, filename_vp, uap->attrnamespace, uap->attrname != NULL ? attrname : NULL); vn_finished_write(mp_writable); out: if (mp != NULL) vfs_unbusy(mp); /* * VFS_EXTATTRCTL will have unlocked, but not de-ref'd, filename_vp, * so vrele it if it is defined. */ if (filename_vp != NULL) vrele(filename_vp); return (error); } /*- * Set a named extended attribute on a file or directory * * Arguments: unlocked vnode "vp", attribute namespace "attrnamespace", * kernelspace string pointer "attrname", userspace buffer * pointer "data", buffer length "nbytes", thread "td". * Returns: 0 on success, an error number otherwise * Locks: none * References: vp must be a valid reference for the duration of the call */ static int extattr_set_vp(struct vnode *vp, int attrnamespace, const char *attrname, void *data, size_t nbytes, struct thread *td) { struct mount *mp; struct uio auio; struct iovec aiov; ssize_t cnt; int error; if (nbytes > IOSIZE_MAX) return (EINVAL); - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); aiov.iov_base = data; aiov.iov_len = nbytes; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_resid = nbytes; auio.uio_rw = UIO_WRITE; auio.uio_segflg = UIO_USERSPACE; auio.uio_td = td; cnt = nbytes; #ifdef MAC error = mac_vnode_check_setextattr(td->td_ucred, vp, attrnamespace, attrname); if (error) goto done; #endif error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, td->td_ucred, td); cnt -= auio.uio_resid; td->td_retval[0] = cnt; #ifdef MAC done: #endif VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } #ifndef _SYS_SYSPROTO_H_ struct extattr_set_fd_args { int fd; int attrnamespace; const char *attrname; void *data; size_t nbytes; }; #endif int sys_extattr_set_fd(struct thread *td, struct extattr_set_fd_args *uap) { struct file *fp; char attrname[EXTATTR_MAXNAMELEN + 1]; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_VALUE(uap->attrnamespace); error = copyinstr(uap->attrname, attrname, sizeof(attrname), NULL); if (error) return (error); AUDIT_ARG_TEXT(attrname); error = getvnode_path(td, uap->fd, cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); if (error) return (error); error = extattr_set_vp(fp->f_vnode, uap->attrnamespace, attrname, uap->data, uap->nbytes, td); fdrop(fp, td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct extattr_set_file_args { const char *path; int attrnamespace; const char *attrname; void *data; size_t nbytes; }; #endif int sys_extattr_set_file(struct thread *td, struct extattr_set_file_args *uap) { return (kern_extattr_set_path(td, uap->path, uap->attrnamespace, uap->attrname, uap->data, uap->nbytes, FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct extattr_set_link_args { const char *path; int attrnamespace; const char *attrname; void *data; size_t nbytes; }; #endif int sys_extattr_set_link(struct thread *td, struct extattr_set_link_args *uap) { return (kern_extattr_set_path(td, uap->path, uap->attrnamespace, uap->attrname, uap->data, uap->nbytes, NOFOLLOW)); } static int kern_extattr_set_path(struct thread *td, const char *path, int attrnamespace, const char *uattrname, void *data, size_t nbytes, int follow) { struct nameidata nd; char attrname[EXTATTR_MAXNAMELEN + 1]; int error; AUDIT_ARG_VALUE(attrnamespace); error = copyinstr(uattrname, attrname, sizeof(attrname), NULL); if (error) return (error); AUDIT_ARG_TEXT(attrname); NDINIT(&nd, LOOKUP, follow | AUDITVNODE1, UIO_USERSPACE, path); error = namei(&nd); if (error) return (error); NDFREE_PNBUF(&nd); error = extattr_set_vp(nd.ni_vp, attrnamespace, attrname, data, nbytes, td); vrele(nd.ni_vp); return (error); } /*- * Get a named extended attribute on a file or directory * * Arguments: unlocked vnode "vp", attribute namespace "attrnamespace", * kernelspace string pointer "attrname", userspace buffer * pointer "data", buffer length "nbytes", thread "td". * Returns: 0 on success, an error number otherwise * Locks: none * References: vp must be a valid reference for the duration of the call */ static int extattr_get_vp(struct vnode *vp, int attrnamespace, const char *attrname, void *data, size_t nbytes, struct thread *td) { struct uio auio, *auiop; struct iovec aiov; ssize_t cnt; size_t size, *sizep; int error; if (nbytes > IOSIZE_MAX) return (EINVAL); vn_lock(vp, LK_SHARED | LK_RETRY); /* * Slightly unusual semantics: if the user provides a NULL data * pointer, they don't want to receive the data, just the maximum * read length. */ auiop = NULL; sizep = NULL; cnt = 0; if (data != NULL) { aiov.iov_base = data; aiov.iov_len = nbytes; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_resid = nbytes; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_USERSPACE; auio.uio_td = td; auiop = &auio; cnt = nbytes; } else sizep = &size; #ifdef MAC error = mac_vnode_check_getextattr(td->td_ucred, vp, attrnamespace, attrname); if (error) goto done; #endif error = VOP_GETEXTATTR(vp, attrnamespace, attrname, auiop, sizep, td->td_ucred, td); if (auiop != NULL) { cnt -= auio.uio_resid; td->td_retval[0] = cnt; } else td->td_retval[0] = size; #ifdef MAC done: #endif VOP_UNLOCK(vp); return (error); } #ifndef _SYS_SYSPROTO_H_ struct extattr_get_fd_args { int fd; int attrnamespace; const char *attrname; void *data; size_t nbytes; }; #endif int sys_extattr_get_fd(struct thread *td, struct extattr_get_fd_args *uap) { struct file *fp; char attrname[EXTATTR_MAXNAMELEN + 1]; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_VALUE(uap->attrnamespace); error = copyinstr(uap->attrname, attrname, sizeof(attrname), NULL); if (error) return (error); AUDIT_ARG_TEXT(attrname); error = getvnode_path(td, uap->fd, cap_rights_init_one(&rights, CAP_EXTATTR_GET), &fp); if (error) return (error); error = extattr_get_vp(fp->f_vnode, uap->attrnamespace, attrname, uap->data, uap->nbytes, td); fdrop(fp, td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct extattr_get_file_args { const char *path; int attrnamespace; const char *attrname; void *data; size_t nbytes; }; #endif int sys_extattr_get_file(struct thread *td, struct extattr_get_file_args *uap) { return (kern_extattr_get_path(td, uap->path, uap->attrnamespace, uap->attrname, uap->data, uap->nbytes, FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct extattr_get_link_args { const char *path; int attrnamespace; const char *attrname; void *data; size_t nbytes; }; #endif int sys_extattr_get_link(struct thread *td, struct extattr_get_link_args *uap) { return (kern_extattr_get_path(td, uap->path, uap->attrnamespace, uap->attrname, uap->data, uap->nbytes, NOFOLLOW)); } static int kern_extattr_get_path(struct thread *td, const char *path, int attrnamespace, const char *uattrname, void *data, size_t nbytes, int follow) { struct nameidata nd; char attrname[EXTATTR_MAXNAMELEN + 1]; int error; AUDIT_ARG_VALUE(attrnamespace); error = copyinstr(uattrname, attrname, sizeof(attrname), NULL); if (error) return (error); AUDIT_ARG_TEXT(attrname); NDINIT(&nd, LOOKUP, follow | AUDITVNODE1, UIO_USERSPACE, path); error = namei(&nd); if (error) return (error); NDFREE_PNBUF(&nd); error = extattr_get_vp(nd.ni_vp, attrnamespace, attrname, data, nbytes, td); vrele(nd.ni_vp); return (error); } /* * extattr_delete_vp(): Delete a named extended attribute on a file or * directory * * Arguments: unlocked vnode "vp", attribute namespace "attrnamespace", * kernelspace string pointer "attrname", proc "p" * Returns: 0 on success, an error number otherwise * Locks: none * References: vp must be a valid reference for the duration of the call */ static int extattr_delete_vp(struct vnode *vp, int attrnamespace, const char *attrname, struct thread *td) { struct mount *mp; int error; - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); #ifdef MAC error = mac_vnode_check_deleteextattr(td->td_ucred, vp, attrnamespace, attrname); if (error) goto done; #endif error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, td->td_ucred, td); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, td->td_ucred, td); #ifdef MAC done: #endif VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } #ifndef _SYS_SYSPROTO_H_ struct extattr_delete_fd_args { int fd; int attrnamespace; const char *attrname; }; #endif int sys_extattr_delete_fd(struct thread *td, struct extattr_delete_fd_args *uap) { struct file *fp; char attrname[EXTATTR_MAXNAMELEN + 1]; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_VALUE(uap->attrnamespace); error = copyinstr(uap->attrname, attrname, sizeof(attrname), NULL); if (error) return (error); AUDIT_ARG_TEXT(attrname); error = getvnode_path(td, uap->fd, cap_rights_init_one(&rights, CAP_EXTATTR_DELETE), &fp); if (error) return (error); error = extattr_delete_vp(fp->f_vnode, uap->attrnamespace, attrname, td); fdrop(fp, td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct extattr_delete_file_args { const char *path; int attrnamespace; const char *attrname; }; #endif int sys_extattr_delete_file(struct thread *td, struct extattr_delete_file_args *uap) { return (kern_extattr_delete_path(td, uap->path, uap->attrnamespace, uap->attrname, FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct extattr_delete_link_args { const char *path; int attrnamespace; const char *attrname; }; #endif int sys_extattr_delete_link(struct thread *td, struct extattr_delete_link_args *uap) { return (kern_extattr_delete_path(td, uap->path, uap->attrnamespace, uap->attrname, NOFOLLOW)); } static int kern_extattr_delete_path(struct thread *td, const char *path, int attrnamespace, const char *uattrname, int follow) { struct nameidata nd; char attrname[EXTATTR_MAXNAMELEN + 1]; int error; AUDIT_ARG_VALUE(attrnamespace); error = copyinstr(uattrname, attrname, sizeof(attrname), NULL); if (error) return(error); AUDIT_ARG_TEXT(attrname); NDINIT(&nd, LOOKUP, follow | AUDITVNODE1, UIO_USERSPACE, path); error = namei(&nd); if (error) return(error); NDFREE_PNBUF(&nd); error = extattr_delete_vp(nd.ni_vp, attrnamespace, attrname, td); vrele(nd.ni_vp); return(error); } /*- * Retrieve a list of extended attributes on a file or directory. * * Arguments: unlocked vnode "vp", attribute namespace 'attrnamespace", * userspace buffer pointer "data", buffer length "nbytes", * thread "td". * Returns: 0 on success, an error number otherwise * Locks: none * References: vp must be a valid reference for the duration of the call */ static int extattr_list_vp(struct vnode *vp, int attrnamespace, void *data, size_t nbytes, struct thread *td) { struct uio auio, *auiop; size_t size, *sizep; struct iovec aiov; ssize_t cnt; int error; if (nbytes > IOSIZE_MAX) return (EINVAL); auiop = NULL; sizep = NULL; cnt = 0; if (data != NULL) { aiov.iov_base = data; aiov.iov_len = nbytes; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_resid = nbytes; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_USERSPACE; auio.uio_td = td; auiop = &auio; cnt = nbytes; } else sizep = &size; vn_lock(vp, LK_SHARED | LK_RETRY); #ifdef MAC error = mac_vnode_check_listextattr(td->td_ucred, vp, attrnamespace); if (error) { VOP_UNLOCK(vp); return (error); } #endif error = VOP_LISTEXTATTR(vp, attrnamespace, auiop, sizep, td->td_ucred, td); VOP_UNLOCK(vp); if (auiop != NULL) { cnt -= auio.uio_resid; td->td_retval[0] = cnt; } else td->td_retval[0] = size; return (error); } #ifndef _SYS_SYSPROTO_H_ struct extattr_list_fd_args { int fd; int attrnamespace; void *data; size_t nbytes; }; #endif int sys_extattr_list_fd(struct thread *td, struct extattr_list_fd_args *uap) { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_VALUE(uap->attrnamespace); error = getvnode_path(td, uap->fd, cap_rights_init_one(&rights, CAP_EXTATTR_LIST), &fp); if (error) return (error); error = extattr_list_vp(fp->f_vnode, uap->attrnamespace, uap->data, uap->nbytes, td); fdrop(fp, td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct extattr_list_file_args { const char *path; int attrnamespace; void *data; size_t nbytes; } #endif int sys_extattr_list_file(struct thread *td, struct extattr_list_file_args *uap) { return (kern_extattr_list_path(td, uap->path, uap->attrnamespace, uap->data, uap->nbytes, FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct extattr_list_link_args { const char *path; int attrnamespace; void *data; size_t nbytes; }; #endif int sys_extattr_list_link(struct thread *td, struct extattr_list_link_args *uap) { return (kern_extattr_list_path(td, uap->path, uap->attrnamespace, uap->data, uap->nbytes, NOFOLLOW)); } static int kern_extattr_list_path(struct thread *td, const char *path, int attrnamespace, void *data, size_t nbytes, int follow) { struct nameidata nd; int error; AUDIT_ARG_VALUE(attrnamespace); NDINIT(&nd, LOOKUP, follow | AUDITVNODE1, UIO_USERSPACE, path); error = namei(&nd); if (error) return (error); NDFREE_PNBUF(&nd); error = extattr_list_vp(nd.ni_vp, attrnamespace, data, nbytes, td); vrele(nd.ni_vp); return (error); } diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c index 103f4dab519f..b3223e620962 100644 --- a/sys/kern/vfs_syscalls.c +++ b/sys/kern/vfs_syscalls.c @@ -1,5035 +1,5035 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_ktrace.h" #include #include #ifdef COMPAT_FREEBSD11 #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #include #include #include #include #include #include MALLOC_DEFINE(M_FADVISE, "fadvise", "posix_fadvise(2) information"); static int kern_chflagsat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, u_long flags, int atflag); static int setfflags(struct thread *td, struct vnode *, u_long); static int getutimes(const struct timeval *, enum uio_seg, struct timespec *); static int getutimens(const struct timespec *, enum uio_seg, struct timespec *, int *); static int setutimes(struct thread *td, struct vnode *, const struct timespec *, int, int); static int vn_access(struct vnode *vp, int user_flags, struct ucred *cred, struct thread *td); static int kern_fhlinkat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, fhandle_t *fhp); static int kern_readlink_vp(struct vnode *vp, char *buf, enum uio_seg bufseg, size_t count, struct thread *td); static int kern_linkat_vp(struct thread *td, struct vnode *vp, int fd, const char *path, enum uio_seg segflag); static uint64_t at2cnpflags(u_int at_flags, u_int mask) { uint64_t res; MPASS((at_flags & (AT_SYMLINK_FOLLOW | AT_SYMLINK_NOFOLLOW)) != (AT_SYMLINK_FOLLOW | AT_SYMLINK_NOFOLLOW)); res = 0; at_flags &= mask; if ((at_flags & AT_RESOLVE_BENEATH) != 0) res |= RBENEATH; if ((at_flags & AT_SYMLINK_FOLLOW) != 0) res |= FOLLOW; /* NOFOLLOW is pseudo flag */ if ((mask & AT_SYMLINK_NOFOLLOW) != 0) { res |= (at_flags & AT_SYMLINK_NOFOLLOW) != 0 ? NOFOLLOW : FOLLOW; } if ((mask & AT_EMPTY_PATH) != 0 && (at_flags & AT_EMPTY_PATH) != 0) res |= EMPTYPATH; return (res); } int kern_sync(struct thread *td) { struct mount *mp, *nmp; int save; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } if ((mp->mnt_flag & MNT_RDONLY) == 0 && vn_start_write(NULL, &mp, V_NOWAIT) == 0) { save = curthread_pflags_set(TDP_SYNCIO); vfs_periodic(mp, MNT_NOWAIT); VFS_SYNC(mp, MNT_NOWAIT); curthread_pflags_restore(save); vn_finished_write(mp); } mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); return (0); } /* * Sync each mounted filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct sync_args { int dummy; }; #endif /* ARGSUSED */ int sys_sync(struct thread *td, struct sync_args *uap) { return (kern_sync(td)); } /* * Change filesystem quotas. */ #ifndef _SYS_SYSPROTO_H_ struct quotactl_args { char *path; int cmd; int uid; caddr_t arg; }; #endif int sys_quotactl(struct thread *td, struct quotactl_args *uap) { struct mount *mp; struct nameidata nd; int error; bool mp_busy; AUDIT_ARG_CMD(uap->cmd); AUDIT_ARG_UID(uap->uid); if (!prison_allow(td->td_ucred, PR_ALLOW_QUOTAS)) return (EPERM); NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path); if ((error = namei(&nd)) != 0) return (error); NDFREE_PNBUF(&nd); mp = nd.ni_vp->v_mount; vfs_ref(mp); vput(nd.ni_vp); error = vfs_busy(mp, 0); if (error != 0) { vfs_rel(mp); return (error); } mp_busy = true; error = VFS_QUOTACTL(mp, uap->cmd, uap->uid, uap->arg, &mp_busy); /* * Since quota on/off operations typically need to open quota * files, the implementation may need to unbusy the mount point * before calling into namei. Otherwise, unmount might be * started between two vfs_busy() invocations (first is ours, * second is from mount point cross-walk code in lookup()), * causing deadlock. * * Avoid unbusying mp if the implementation indicates it has * already done so. */ if (mp_busy) vfs_unbusy(mp); vfs_rel(mp); return (error); } /* * Used by statfs conversion routines to scale the block size up if * necessary so that all of the block counts are <= 'max_size'. Note * that 'max_size' should be a bitmask, i.e. 2^n - 1 for some non-zero * value of 'n'. */ void statfs_scale_blocks(struct statfs *sf, long max_size) { uint64_t count; int shift; KASSERT(powerof2(max_size + 1), ("%s: invalid max_size", __func__)); /* * Attempt to scale the block counts to give a more accurate * overview to userland of the ratio of free space to used * space. To do this, find the largest block count and compute * a divisor that lets it fit into a signed integer <= max_size. */ if (sf->f_bavail < 0) count = -sf->f_bavail; else count = sf->f_bavail; count = MAX(sf->f_blocks, MAX(sf->f_bfree, count)); if (count <= max_size) return; count >>= flsl(max_size); shift = 0; while (count > 0) { shift++; count >>=1; } sf->f_bsize <<= shift; sf->f_blocks >>= shift; sf->f_bfree >>= shift; sf->f_bavail >>= shift; } static int kern_do_statfs(struct thread *td, struct mount *mp, struct statfs *buf) { int error; if (mp == NULL) return (EBADF); error = vfs_busy(mp, 0); vfs_rel(mp); if (error != 0) return (error); #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif error = VFS_STATFS(mp, buf); if (error != 0) goto out; if (priv_check_cred_vfs_generation(td->td_ucred)) { buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, buf); } out: vfs_unbusy(mp); return (error); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct statfs_args { char *path; struct statfs *buf; }; #endif int sys_statfs(struct thread *td, struct statfs_args *uap) { struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp); if (error == 0) error = copyout(sfp, uap->buf, sizeof(struct statfs)); free(sfp, M_STATFS); return (error); } int kern_statfs(struct thread *td, const char *path, enum uio_seg pathseg, struct statfs *buf) { struct mount *mp; struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path); error = namei(&nd); if (error != 0) return (error); NDFREE_PNBUF(&nd); mp = vfs_ref_from_vp(nd.ni_vp); vrele(nd.ni_vp); return (kern_do_statfs(td, mp, buf)); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct fstatfs_args { int fd; struct statfs *buf; }; #endif int sys_fstatfs(struct thread *td, struct fstatfs_args *uap) { struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fstatfs(td, uap->fd, sfp); if (error == 0) error = copyout(sfp, uap->buf, sizeof(struct statfs)); free(sfp, M_STATFS); return (error); } int kern_fstatfs(struct thread *td, int fd, struct statfs *buf) { struct file *fp; struct mount *mp; struct vnode *vp; int error; AUDIT_ARG_FD(fd); error = getvnode_path(td, fd, &cap_fstatfs_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; #ifdef AUDIT if (AUDITING_TD(td)) { vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp); } #endif mp = vfs_ref_from_vp(vp); fdrop(fp, td); return (kern_do_statfs(td, mp, buf)); } /* * Get statistics on all filesystems. */ #ifndef _SYS_SYSPROTO_H_ struct getfsstat_args { struct statfs *buf; long bufsize; int mode; }; #endif int sys_getfsstat(struct thread *td, struct getfsstat_args *uap) { size_t count; int error; if (uap->bufsize < 0 || uap->bufsize > SIZE_MAX) return (EINVAL); error = kern_getfsstat(td, &uap->buf, uap->bufsize, &count, UIO_USERSPACE, uap->mode); if (error == 0) td->td_retval[0] = count; return (error); } /* * If (bufsize > 0 && bufseg == UIO_SYSSPACE) * The caller is responsible for freeing memory which will be allocated * in '*buf'. */ int kern_getfsstat(struct thread *td, struct statfs **buf, size_t bufsize, size_t *countp, enum uio_seg bufseg, int mode) { struct mount *mp, *nmp; struct statfs *sfsp, *sp, *sptmp, *tofree; size_t count, maxcount; int error; switch (mode) { case MNT_WAIT: case MNT_NOWAIT: break; default: if (bufseg == UIO_SYSSPACE) *buf = NULL; return (EINVAL); } restart: maxcount = bufsize / sizeof(struct statfs); if (bufsize == 0) { sfsp = NULL; tofree = NULL; } else if (bufseg == UIO_USERSPACE) { sfsp = *buf; tofree = NULL; } else /* if (bufseg == UIO_SYSSPACE) */ { count = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { count++; } mtx_unlock(&mountlist_mtx); if (maxcount > count) maxcount = count; tofree = sfsp = *buf = malloc(maxcount * sizeof(struct statfs), M_STATFS, M_WAITOK); } count = 0; /* * If there is no target buffer they only want the count. * * This could be TAILQ_FOREACH but it is open-coded to match the original * code below. */ if (sfsp == NULL) { mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (prison_canseemount(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #ifdef MAC if (mac_mount_check_stat(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #endif count++; nmp = TAILQ_NEXT(mp, mnt_list); } mtx_unlock(&mountlist_mtx); *countp = count; return (0); } /* * They want the entire thing. * * Short-circuit the corner case of no room for anything, avoids * relocking below. */ if (maxcount < 1) { goto out; } mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (prison_canseemount(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #ifdef MAC if (mac_mount_check_stat(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #endif if (mode == MNT_WAIT) { if (vfs_busy(mp, MBF_MNTLSTLOCK) != 0) { /* * If vfs_busy() failed, and MBF_NOWAIT * wasn't passed, then the mp is gone. * Furthermore, because of MBF_MNTLSTLOCK, * the mountlist_mtx was dropped. We have * no other choice than to start over. */ mtx_unlock(&mountlist_mtx); free(tofree, M_STATFS); goto restart; } } else { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } } sp = &mp->mnt_stat; /* * If MNT_NOWAIT is specified, do not refresh * the fsstat cache. */ if (mode != MNT_NOWAIT) { error = VFS_STATFS(mp, sp); if (error != 0) { mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); continue; } } if (priv_check_cred_vfs_generation(td->td_ucred)) { sptmp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); *sptmp = *sp; sptmp->f_fsid.val[0] = sptmp->f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, sptmp); sp = sptmp; } else sptmp = NULL; if (bufseg == UIO_SYSSPACE) { bcopy(sp, sfsp, sizeof(*sp)); free(sptmp, M_STATFS); } else /* if (bufseg == UIO_USERSPACE) */ { error = copyout(sp, sfsp, sizeof(*sp)); free(sptmp, M_STATFS); if (error != 0) { vfs_unbusy(mp); return (error); } } sfsp++; count++; if (count == maxcount) { vfs_unbusy(mp); goto out; } mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); out: *countp = count; return (0); } #ifdef COMPAT_FREEBSD4 /* * Get old format filesystem statistics. */ static void freebsd4_cvtstatfs(struct statfs *, struct ostatfs *); #ifndef _SYS_SYSPROTO_H_ struct freebsd4_statfs_args { char *path; struct ostatfs *buf; }; #endif int freebsd4_statfs(struct thread *td, struct freebsd4_statfs_args *uap) { struct ostatfs osb; struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp); if (error == 0) { freebsd4_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_fstatfs_args { int fd; struct ostatfs *buf; }; #endif int freebsd4_fstatfs(struct thread *td, struct freebsd4_fstatfs_args *uap) { struct ostatfs osb; struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fstatfs(td, uap->fd, sfp); if (error == 0) { freebsd4_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Get statistics on all filesystems. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_getfsstat_args { struct ostatfs *buf; long bufsize; int mode; }; #endif int freebsd4_getfsstat(struct thread *td, struct freebsd4_getfsstat_args *uap) { struct statfs *buf, *sp; struct ostatfs osb; size_t count, size; int error; if (uap->bufsize < 0) return (EINVAL); count = uap->bufsize / sizeof(struct ostatfs); if (count > SIZE_MAX / sizeof(struct statfs)) return (EINVAL); size = count * sizeof(struct statfs); error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE, uap->mode); if (error == 0) td->td_retval[0] = count; if (size != 0) { sp = buf; while (count != 0 && error == 0) { freebsd4_cvtstatfs(sp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); sp++; uap->buf++; count--; } free(buf, M_STATFS); } return (error); } /* * Implement fstatfs() for (NFS) file handles. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_fhstatfs_args { struct fhandle *u_fhp; struct ostatfs *buf; }; #endif int freebsd4_fhstatfs(struct thread *td, struct freebsd4_fhstatfs_args *uap) { struct ostatfs osb; struct statfs *sfp; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fhstatfs(td, fh, sfp); if (error == 0) { freebsd4_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Convert a new format statfs structure to an old format statfs structure. */ static void freebsd4_cvtstatfs(struct statfs *nsp, struct ostatfs *osp) { statfs_scale_blocks(nsp, LONG_MAX); bzero(osp, sizeof(*osp)); osp->f_bsize = nsp->f_bsize; osp->f_iosize = MIN(nsp->f_iosize, LONG_MAX); osp->f_blocks = nsp->f_blocks; osp->f_bfree = nsp->f_bfree; osp->f_bavail = nsp->f_bavail; osp->f_files = MIN(nsp->f_files, LONG_MAX); osp->f_ffree = MIN(nsp->f_ffree, LONG_MAX); osp->f_owner = nsp->f_owner; osp->f_type = nsp->f_type; osp->f_flags = nsp->f_flags; osp->f_syncwrites = MIN(nsp->f_syncwrites, LONG_MAX); osp->f_asyncwrites = MIN(nsp->f_asyncwrites, LONG_MAX); osp->f_syncreads = MIN(nsp->f_syncreads, LONG_MAX); osp->f_asyncreads = MIN(nsp->f_asyncreads, LONG_MAX); strlcpy(osp->f_fstypename, nsp->f_fstypename, MIN(MFSNAMELEN, OMFSNAMELEN)); strlcpy(osp->f_mntonname, nsp->f_mntonname, MIN(MNAMELEN, OMNAMELEN)); strlcpy(osp->f_mntfromname, nsp->f_mntfromname, MIN(MNAMELEN, OMNAMELEN)); osp->f_fsid = nsp->f_fsid; } #endif /* COMPAT_FREEBSD4 */ #if defined(COMPAT_FREEBSD11) /* * Get old format filesystem statistics. */ static void freebsd11_cvtstatfs(struct statfs *, struct freebsd11_statfs *); int freebsd11_statfs(struct thread *td, struct freebsd11_statfs_args *uap) { struct freebsd11_statfs osb; struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp); if (error == 0) { freebsd11_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Get filesystem statistics. */ int freebsd11_fstatfs(struct thread *td, struct freebsd11_fstatfs_args *uap) { struct freebsd11_statfs osb; struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fstatfs(td, uap->fd, sfp); if (error == 0) { freebsd11_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Get statistics on all filesystems. */ int freebsd11_getfsstat(struct thread *td, struct freebsd11_getfsstat_args *uap) { return (kern_freebsd11_getfsstat(td, uap->buf, uap->bufsize, uap->mode)); } int kern_freebsd11_getfsstat(struct thread *td, struct freebsd11_statfs * ubuf, long bufsize, int mode) { struct freebsd11_statfs osb; struct statfs *buf, *sp; size_t count, size; int error; if (bufsize < 0) return (EINVAL); count = bufsize / sizeof(struct ostatfs); size = count * sizeof(struct statfs); error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE, mode); if (error == 0) td->td_retval[0] = count; if (size > 0) { sp = buf; while (count > 0 && error == 0) { freebsd11_cvtstatfs(sp, &osb); error = copyout(&osb, ubuf, sizeof(osb)); sp++; ubuf++; count--; } free(buf, M_STATFS); } return (error); } /* * Implement fstatfs() for (NFS) file handles. */ int freebsd11_fhstatfs(struct thread *td, struct freebsd11_fhstatfs_args *uap) { struct freebsd11_statfs osb; struct statfs *sfp; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error) return (error); sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fhstatfs(td, fh, sfp); if (error == 0) { freebsd11_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Convert a new format statfs structure to an old format statfs structure. */ static void freebsd11_cvtstatfs(struct statfs *nsp, struct freebsd11_statfs *osp) { bzero(osp, sizeof(*osp)); osp->f_version = FREEBSD11_STATFS_VERSION; osp->f_type = nsp->f_type; osp->f_flags = nsp->f_flags; osp->f_bsize = nsp->f_bsize; osp->f_iosize = nsp->f_iosize; osp->f_blocks = nsp->f_blocks; osp->f_bfree = nsp->f_bfree; osp->f_bavail = nsp->f_bavail; osp->f_files = nsp->f_files; osp->f_ffree = nsp->f_ffree; osp->f_syncwrites = nsp->f_syncwrites; osp->f_asyncwrites = nsp->f_asyncwrites; osp->f_syncreads = nsp->f_syncreads; osp->f_asyncreads = nsp->f_asyncreads; osp->f_namemax = nsp->f_namemax; osp->f_owner = nsp->f_owner; osp->f_fsid = nsp->f_fsid; strlcpy(osp->f_fstypename, nsp->f_fstypename, MIN(MFSNAMELEN, sizeof(osp->f_fstypename))); strlcpy(osp->f_mntonname, nsp->f_mntonname, MIN(MNAMELEN, sizeof(osp->f_mntonname))); strlcpy(osp->f_mntfromname, nsp->f_mntfromname, MIN(MNAMELEN, sizeof(osp->f_mntfromname))); } #endif /* COMPAT_FREEBSD11 */ /* * Change current working directory to a given file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchdir_args { int fd; }; #endif int sys_fchdir(struct thread *td, struct fchdir_args *uap) { struct vnode *vp, *tdp; struct mount *mp; struct file *fp; int error; AUDIT_ARG_FD(uap->fd); error = getvnode_path(td, uap->fd, &cap_fchdir_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; vrefact(vp); fdrop(fp, td); vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); error = change_dir(vp, td); while (!error && (mp = vp->v_mountedhere) != NULL) { if (vfs_busy(mp, 0)) continue; error = VFS_ROOT(mp, LK_SHARED, &tdp); vfs_unbusy(mp); if (error != 0) break; vput(vp); vp = tdp; } if (error != 0) { vput(vp); return (error); } VOP_UNLOCK(vp); pwd_chdir(td, vp); return (0); } /* * Change current working directory (``.''). */ #ifndef _SYS_SYSPROTO_H_ struct chdir_args { char *path; }; #endif int sys_chdir(struct thread *td, struct chdir_args *uap) { return (kern_chdir(td, uap->path, UIO_USERSPACE)); } int kern_chdir(struct thread *td, const char *path, enum uio_seg pathseg) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path); if ((error = namei(&nd)) != 0) return (error); if ((error = change_dir(nd.ni_vp, td)) != 0) { vput(nd.ni_vp); NDFREE_PNBUF(&nd); return (error); } VOP_UNLOCK(nd.ni_vp); NDFREE_PNBUF(&nd); pwd_chdir(td, nd.ni_vp); return (0); } static int unprivileged_chroot = 0; SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_chroot, CTLFLAG_RW, &unprivileged_chroot, 0, "Unprivileged processes can use chroot(2)"); /* * Change notion of root (``/'') directory. */ #ifndef _SYS_SYSPROTO_H_ struct chroot_args { char *path; }; #endif int sys_chroot(struct thread *td, struct chroot_args *uap) { struct nameidata nd; struct proc *p; int error; error = priv_check(td, PRIV_VFS_CHROOT); if (error != 0) { p = td->td_proc; PROC_LOCK(p); if (unprivileged_chroot == 0 || (p->p_flag2 & P2_NO_NEW_PRIVS) == 0) { PROC_UNLOCK(p); return (error); } PROC_UNLOCK(p); } NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path); error = namei(&nd); if (error != 0) return (error); NDFREE_PNBUF(&nd); error = change_dir(nd.ni_vp, td); if (error != 0) goto e_vunlock; #ifdef MAC error = mac_vnode_check_chroot(td->td_ucred, nd.ni_vp); if (error != 0) goto e_vunlock; #endif VOP_UNLOCK(nd.ni_vp); error = pwd_chroot(td, nd.ni_vp); vrele(nd.ni_vp); return (error); e_vunlock: vput(nd.ni_vp); return (error); } /* * Common routine for chroot and chdir. Callers must provide a locked vnode * instance. */ int change_dir(struct vnode *vp, struct thread *td) { #ifdef MAC int error; #endif ASSERT_VOP_LOCKED(vp, "change_dir(): vp not locked"); if (vp->v_type != VDIR) return (ENOTDIR); #ifdef MAC error = mac_vnode_check_chdir(td->td_ucred, vp); if (error != 0) return (error); #endif return (VOP_ACCESS(vp, VEXEC, td->td_ucred, td)); } static __inline void flags_to_rights(int flags, cap_rights_t *rightsp) { if (flags & O_EXEC) { cap_rights_set_one(rightsp, CAP_FEXECVE); if (flags & O_PATH) return; } else { switch ((flags & O_ACCMODE)) { case O_RDONLY: cap_rights_set_one(rightsp, CAP_READ); break; case O_RDWR: cap_rights_set_one(rightsp, CAP_READ); /* FALLTHROUGH */ case O_WRONLY: cap_rights_set_one(rightsp, CAP_WRITE); if (!(flags & (O_APPEND | O_TRUNC))) cap_rights_set_one(rightsp, CAP_SEEK); break; } } if (flags & O_CREAT) cap_rights_set_one(rightsp, CAP_CREATE); if (flags & O_TRUNC) cap_rights_set_one(rightsp, CAP_FTRUNCATE); if (flags & (O_SYNC | O_FSYNC)) cap_rights_set_one(rightsp, CAP_FSYNC); if (flags & (O_EXLOCK | O_SHLOCK)) cap_rights_set_one(rightsp, CAP_FLOCK); } /* * Check permissions, allocate an open file structure, and call the device * open routine if any. */ #ifndef _SYS_SYSPROTO_H_ struct open_args { char *path; int flags; int mode; }; #endif int sys_open(struct thread *td, struct open_args *uap) { return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct openat_args { int fd; char *path; int flag; int mode; }; #endif int sys_openat(struct thread *td, struct openat_args *uap) { AUDIT_ARG_FD(uap->fd); return (kern_openat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag, uap->mode)); } int kern_openat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int flags, int mode) { struct proc *p = td->td_proc; struct filedesc *fdp; struct pwddesc *pdp; struct file *fp; struct vnode *vp; struct nameidata nd; cap_rights_t rights; int cmode, error, indx; indx = -1; fdp = p->p_fd; pdp = p->p_pd; AUDIT_ARG_FFLAGS(flags); AUDIT_ARG_MODE(mode); cap_rights_init_one(&rights, CAP_LOOKUP); flags_to_rights(flags, &rights); /* * Only one of the O_EXEC, O_RDONLY, O_WRONLY and O_RDWR flags * may be specified. On the other hand, for O_PATH any mode * except O_EXEC is ignored. */ if ((flags & O_PATH) != 0) { flags &= ~(O_CREAT | O_ACCMODE); } else if ((flags & O_EXEC) != 0) { if (flags & O_ACCMODE) return (EINVAL); } else if ((flags & O_ACCMODE) == O_ACCMODE) { return (EINVAL); } else { flags = FFLAGS(flags); } /* * Allocate a file structure. The descriptor to reference it * is allocated and used by finstall_refed() below. */ error = falloc_noinstall(td, &fp); if (error != 0) return (error); /* Set the flags early so the finit in devfs can pick them up. */ fp->f_flag = flags & FMASK; cmode = ((mode & ~pdp->pd_cmask) & ALLPERMS) & ~S_ISTXT; NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1 | WANTIOCTLCAPS, pathseg, path, fd, &rights); td->td_dupfd = -1; /* XXX check for fdopen */ error = vn_open_cred(&nd, &flags, cmode, VN_OPEN_WANTIOCTLCAPS, td->td_ucred, fp); if (error != 0) { /* * If the vn_open replaced the method vector, something * wonderous happened deep below and we just pass it up * pretending we know what we do. */ if (error == ENXIO && fp->f_ops != &badfileops) { MPASS((flags & O_PATH) == 0); goto success; } /* * Handle special fdopen() case. bleh. * * Don't do this for relative (capability) lookups; we don't * understand exactly what would happen, and we don't think * that it ever should. */ if ((nd.ni_resflags & NIRES_STRICTREL) == 0 && (error == ENODEV || error == ENXIO) && td->td_dupfd >= 0) { error = dupfdopen(td, fdp, td->td_dupfd, flags, error, &indx); if (error == 0) goto success; } goto bad; } td->td_dupfd = 0; NDFREE_PNBUF(&nd); vp = nd.ni_vp; /* * Store the vnode, for any f_type. Typically, the vnode use * count is decremented by direct call to vn_closefile() for * files that switched type in the cdevsw fdopen() method. */ fp->f_vnode = vp; /* * If the file wasn't claimed by devfs bind it to the normal * vnode operations here. */ if (fp->f_ops == &badfileops) { KASSERT(vp->v_type != VFIFO || (flags & O_PATH) != 0, ("Unexpected fifo fp %p vp %p", fp, vp)); if ((flags & O_PATH) != 0) { finit(fp, (flags & FMASK) | (fp->f_flag & FKQALLOWED), DTYPE_VNODE, NULL, &path_fileops); } else { finit_vnode(fp, flags, NULL, &vnops); } } VOP_UNLOCK(vp); if (flags & O_TRUNC) { error = fo_truncate(fp, 0, td->td_ucred, td); if (error != 0) goto bad; } success: /* * If we haven't already installed the FD (for dupfdopen), do so now. */ if (indx == -1) { struct filecaps *fcaps; #ifdef CAPABILITIES if ((nd.ni_resflags & NIRES_STRICTREL) != 0) fcaps = &nd.ni_filecaps; else #endif fcaps = NULL; error = finstall_refed(td, fp, &indx, flags, fcaps); /* On success finstall_refed() consumes fcaps. */ if (error != 0) { goto bad; } } else { NDFREE_IOCTLCAPS(&nd); falloc_abort(td, fp); } td->td_retval[0] = indx; return (0); bad: KASSERT(indx == -1, ("indx=%d, should be -1", indx)); NDFREE_IOCTLCAPS(&nd); falloc_abort(td, fp); return (error); } #ifdef COMPAT_43 /* * Create a file. */ #ifndef _SYS_SYSPROTO_H_ struct ocreat_args { char *path; int mode; }; #endif int ocreat(struct thread *td, struct ocreat_args *uap) { return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE, O_WRONLY | O_CREAT | O_TRUNC, uap->mode)); } #endif /* COMPAT_43 */ /* * Create a special file. */ #ifndef _SYS_SYSPROTO_H_ struct mknodat_args { int fd; char *path; mode_t mode; dev_t dev; }; #endif int sys_mknodat(struct thread *td, struct mknodat_args *uap) { return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } #if defined(COMPAT_FREEBSD11) int freebsd11_mknod(struct thread *td, struct freebsd11_mknod_args *uap) { return (kern_mknodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } int freebsd11_mknodat(struct thread *td, struct freebsd11_mknodat_args *uap) { return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } #endif /* COMPAT_FREEBSD11 */ int kern_mknodat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int mode, dev_t dev) { struct vnode *vp; struct mount *mp; struct vattr vattr; struct nameidata nd; int error, whiteout = 0; AUDIT_ARG_MODE(mode); AUDIT_ARG_DEV(dev); switch (mode & S_IFMT) { case S_IFCHR: case S_IFBLK: error = priv_check(td, PRIV_VFS_MKNOD_DEV); if (error == 0 && dev == VNOVAL) error = EINVAL; break; case S_IFWHT: error = priv_check(td, PRIV_VFS_MKNOD_WHT); break; case S_IFIFO: if (dev == 0) return (kern_mkfifoat(td, fd, path, pathseg, mode)); /* FALLTHROUGH */ default: error = EINVAL; break; } if (error != 0) return (error); NDPREINIT(&nd); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | AUDITVNODE1 | NOCACHE, pathseg, path, fd, &cap_mknodat_rights); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; if (vp != NULL) { NDFREE_PNBUF(&nd); if (vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(vp); return (EEXIST); } else { VATTR_NULL(&vattr); vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_pd->pd_cmask; vattr.va_rdev = dev; whiteout = 0; switch (mode & S_IFMT) { case S_IFCHR: vattr.va_type = VCHR; break; case S_IFBLK: vattr.va_type = VBLK; break; case S_IFWHT: whiteout = 1; break; default: panic("kern_mknod: invalid mode"); } } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE_PNBUF(&nd); vput(nd.ni_dvp); - if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) + if ((error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH)) != 0) return (error); goto restart; } #ifdef MAC if (error == 0 && !whiteout) error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); #endif if (error == 0) { if (whiteout) error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, CREATE); else { error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); } } VOP_VPUT_PAIR(nd.ni_dvp, error == 0 && !whiteout ? &nd.ni_vp : NULL, true); vn_finished_write(mp); NDFREE_PNBUF(&nd); if (error == ERELOOKUP) goto restart; return (error); } /* * Create a named pipe. */ #ifndef _SYS_SYSPROTO_H_ struct mkfifo_args { char *path; int mode; }; #endif int sys_mkfifo(struct thread *td, struct mkfifo_args *uap) { return (kern_mkfifoat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct mkfifoat_args { int fd; char *path; mode_t mode; }; #endif int sys_mkfifoat(struct thread *td, struct mkfifoat_args *uap) { return (kern_mkfifoat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode)); } int kern_mkfifoat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int mode) { struct mount *mp; struct vattr vattr; struct nameidata nd; int error; AUDIT_ARG_MODE(mode); NDPREINIT(&nd); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | AUDITVNODE1 | NOCACHE, pathseg, path, fd, &cap_mkfifoat_rights); if ((error = namei(&nd)) != 0) return (error); if (nd.ni_vp != NULL) { NDFREE_PNBUF(&nd); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE_PNBUF(&nd); vput(nd.ni_dvp); - if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) + if ((error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH)) != 0) return (error); goto restart; } VATTR_NULL(&vattr); vattr.va_type = VFIFO; vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_pd->pd_cmask; #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out; #endif error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); #ifdef MAC out: #endif VOP_VPUT_PAIR(nd.ni_dvp, error == 0 ? &nd.ni_vp : NULL, true); vn_finished_write(mp); NDFREE_PNBUF(&nd); if (error == ERELOOKUP) goto restart; return (error); } /* * Make a hard file link. */ #ifndef _SYS_SYSPROTO_H_ struct link_args { char *path; char *link; }; #endif int sys_link(struct thread *td, struct link_args *uap) { return (kern_linkat(td, AT_FDCWD, AT_FDCWD, uap->path, uap->link, UIO_USERSPACE, AT_SYMLINK_FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct linkat_args { int fd1; char *path1; int fd2; char *path2; int flag; }; #endif int sys_linkat(struct thread *td, struct linkat_args *uap) { return (kern_linkat(td, uap->fd1, uap->fd2, uap->path1, uap->path2, UIO_USERSPACE, uap->flag)); } int hardlink_check_uid = 0; SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_uid, CTLFLAG_RW, &hardlink_check_uid, 0, "Unprivileged processes cannot create hard links to files owned by other " "users"); static int hardlink_check_gid = 0; SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_gid, CTLFLAG_RW, &hardlink_check_gid, 0, "Unprivileged processes cannot create hard links to files owned by other " "groups"); static int can_hardlink(struct vnode *vp, struct ucred *cred) { struct vattr va; int error; if (!hardlink_check_uid && !hardlink_check_gid) return (0); error = VOP_GETATTR(vp, &va, cred); if (error != 0) return (error); if (hardlink_check_uid && cred->cr_uid != va.va_uid) { error = priv_check_cred(cred, PRIV_VFS_LINK); if (error != 0) return (error); } if (hardlink_check_gid && !groupmember(va.va_gid, cred)) { error = priv_check_cred(cred, PRIV_VFS_LINK); if (error != 0) return (error); } return (0); } int kern_linkat(struct thread *td, int fd1, int fd2, const char *path1, const char *path2, enum uio_seg segflag, int flag) { struct nameidata nd; int error; if ((flag & ~(AT_SYMLINK_FOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH)) != 0) return (EINVAL); NDPREINIT(&nd); do { bwillwrite(); NDINIT_ATRIGHTS(&nd, LOOKUP, AUDITVNODE1 | at2cnpflags(flag, AT_SYMLINK_FOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH), segflag, path1, fd1, &cap_linkat_source_rights); if ((error = namei(&nd)) != 0) return (error); NDFREE_PNBUF(&nd); if ((nd.ni_resflags & NIRES_EMPTYPATH) != 0) { error = priv_check(td, PRIV_VFS_FHOPEN); if (error != 0) { vrele(nd.ni_vp); return (error); } } error = kern_linkat_vp(td, nd.ni_vp, fd2, path2, segflag); } while (error == EAGAIN || error == ERELOOKUP); return (error); } static int kern_linkat_vp(struct thread *td, struct vnode *vp, int fd, const char *path, enum uio_seg segflag) { struct nameidata nd; struct mount *mp; int error; if (vp->v_type == VDIR) { vrele(vp); return (EPERM); /* POSIX */ } NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | AUDITVNODE2 | NOCACHE, segflag, path, fd, &cap_linkat_target_rights); if ((error = namei(&nd)) == 0) { if (nd.ni_vp != NULL) { NDFREE_PNBUF(&nd); if (nd.ni_dvp == nd.ni_vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); vrele(vp); return (EEXIST); } else if (nd.ni_dvp->v_mount != vp->v_mount) { /* * Cross-device link. No need to recheck * vp->v_type, since it cannot change, except * to VBAD. */ NDFREE_PNBUF(&nd); vput(nd.ni_dvp); vrele(vp); return (EXDEV); } else if (vn_lock(vp, LK_EXCLUSIVE) == 0) { error = can_hardlink(vp, td->td_ucred); #ifdef MAC if (error == 0) error = mac_vnode_check_link(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); #endif if (error != 0) { vput(vp); vput(nd.ni_dvp); NDFREE_PNBUF(&nd); return (error); } error = vn_start_write(vp, &mp, V_NOWAIT); if (error != 0) { vput(vp); vput(nd.ni_dvp); NDFREE_PNBUF(&nd); error = vn_start_write(NULL, &mp, - V_XSLEEP | PCATCH); + V_XSLEEP | V_PCATCH); if (error != 0) return (error); return (EAGAIN); } error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); VOP_VPUT_PAIR(nd.ni_dvp, &vp, true); vn_finished_write(mp); NDFREE_PNBUF(&nd); vp = NULL; } else { vput(nd.ni_dvp); NDFREE_PNBUF(&nd); vrele(vp); return (EAGAIN); } } if (vp != NULL) vrele(vp); return (error); } /* * Make a symbolic link. */ #ifndef _SYS_SYSPROTO_H_ struct symlink_args { char *path; char *link; }; #endif int sys_symlink(struct thread *td, struct symlink_args *uap) { return (kern_symlinkat(td, uap->path, AT_FDCWD, uap->link, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct symlinkat_args { char *path; int fd; char *path2; }; #endif int sys_symlinkat(struct thread *td, struct symlinkat_args *uap) { return (kern_symlinkat(td, uap->path1, uap->fd, uap->path2, UIO_USERSPACE)); } int kern_symlinkat(struct thread *td, const char *path1, int fd, const char *path2, enum uio_seg segflg) { struct mount *mp; struct vattr vattr; const char *syspath; char *tmppath; struct nameidata nd; int error; if (segflg == UIO_SYSSPACE) { syspath = path1; } else { tmppath = uma_zalloc(namei_zone, M_WAITOK); if ((error = copyinstr(path1, tmppath, MAXPATHLEN, NULL)) != 0) goto out; syspath = tmppath; } AUDIT_ARG_TEXT(syspath); NDPREINIT(&nd); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | AUDITVNODE1 | NOCACHE, segflg, path2, fd, &cap_symlinkat_rights); if ((error = namei(&nd)) != 0) goto out; if (nd.ni_vp) { NDFREE_PNBUF(&nd); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); nd.ni_vp = NULL; error = EEXIST; goto out; } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE_PNBUF(&nd); vput(nd.ni_dvp); - if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) + if ((error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH)) != 0) goto out; goto restart; } VATTR_NULL(&vattr); vattr.va_mode = ACCESSPERMS &~ td->td_proc->p_pd->pd_cmask; #ifdef MAC vattr.va_type = VLNK; error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out2; #endif error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr, syspath); #ifdef MAC out2: #endif VOP_VPUT_PAIR(nd.ni_dvp, error == 0 ? &nd.ni_vp : NULL, true); vn_finished_write(mp); NDFREE_PNBUF(&nd); if (error == ERELOOKUP) goto restart; out: if (segflg != UIO_SYSSPACE) uma_zfree(namei_zone, tmppath); return (error); } /* * Delete a whiteout from the filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct undelete_args { char *path; }; #endif int sys_undelete(struct thread *td, struct undelete_args *uap) { struct mount *mp; struct nameidata nd; int error; NDPREINIT(&nd); restart: bwillwrite(); NDINIT(&nd, DELETE, LOCKPARENT | DOWHITEOUT | AUDITVNODE1, UIO_USERSPACE, uap->path); error = namei(&nd); if (error != 0) return (error); if (nd.ni_vp != NULLVP || !(nd.ni_cnd.cn_flags & ISWHITEOUT)) { NDFREE_PNBUF(&nd); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if (nd.ni_vp) vrele(nd.ni_vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE_PNBUF(&nd); vput(nd.ni_dvp); - if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) + if ((error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH)) != 0) return (error); goto restart; } error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, DELETE); NDFREE_PNBUF(&nd); vput(nd.ni_dvp); vn_finished_write(mp); if (error == ERELOOKUP) goto restart; return (error); } /* * Delete a name from the filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct unlink_args { char *path; }; #endif int sys_unlink(struct thread *td, struct unlink_args *uap) { return (kern_funlinkat(td, AT_FDCWD, uap->path, FD_NONE, UIO_USERSPACE, 0, 0)); } static int kern_funlinkat_ex(struct thread *td, int dfd, const char *path, int fd, int flag, enum uio_seg pathseg, ino_t oldinum) { if ((flag & ~(AT_REMOVEDIR | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); if ((flag & AT_REMOVEDIR) != 0) return (kern_frmdirat(td, dfd, path, fd, UIO_USERSPACE, 0)); return (kern_funlinkat(td, dfd, path, fd, UIO_USERSPACE, 0, 0)); } #ifndef _SYS_SYSPROTO_H_ struct unlinkat_args { int fd; char *path; int flag; }; #endif int sys_unlinkat(struct thread *td, struct unlinkat_args *uap) { return (kern_funlinkat_ex(td, uap->fd, uap->path, FD_NONE, uap->flag, UIO_USERSPACE, 0)); } #ifndef _SYS_SYSPROTO_H_ struct funlinkat_args { int dfd; const char *path; int fd; int flag; }; #endif int sys_funlinkat(struct thread *td, struct funlinkat_args *uap) { return (kern_funlinkat_ex(td, uap->dfd, uap->path, uap->fd, uap->flag, UIO_USERSPACE, 0)); } int kern_funlinkat(struct thread *td, int dfd, const char *path, int fd, enum uio_seg pathseg, int flag, ino_t oldinum) { struct mount *mp; struct file *fp; struct vnode *vp; struct nameidata nd; struct stat sb; int error; fp = NULL; if (fd != FD_NONE) { error = getvnode_path(td, fd, &cap_no_rights, &fp); if (error != 0) return (error); } NDPREINIT(&nd); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1 | at2cnpflags(flag, AT_RESOLVE_BENEATH), pathseg, path, dfd, &cap_unlinkat_rights); if ((error = namei(&nd)) != 0) { if (error == EINVAL) error = EPERM; goto fdout; } vp = nd.ni_vp; if (vp->v_type == VDIR && oldinum == 0) { error = EPERM; /* POSIX */ } else if (oldinum != 0 && ((error = VOP_STAT(vp, &sb, td->td_ucred, NOCRED)) == 0) && sb.st_ino != oldinum) { error = EIDRM; /* Identifier removed */ } else if (fp != NULL && fp->f_vnode != vp) { if (VN_IS_DOOMED(fp->f_vnode)) error = EBADF; else error = EDEADLK; } else { /* * The root of a mounted filesystem cannot be deleted. * * XXX: can this only be a VDIR case? */ if (vp->v_vflag & VV_ROOT) error = EBUSY; } if (error == 0) { if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE_PNBUF(&nd); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); if ((error = vn_start_write(NULL, &mp, - V_XSLEEP | PCATCH)) != 0) { + V_XSLEEP | V_PCATCH)) != 0) { goto fdout; } goto restart; } #ifdef MAC error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); if (error != 0) goto out; #endif vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK); error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); #ifdef MAC out: #endif vn_finished_write(mp); } NDFREE_PNBUF(&nd); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); if (error == ERELOOKUP) goto restart; fdout: if (fp != NULL) fdrop(fp, td); return (error); } /* * Reposition read/write file offset. */ #ifndef _SYS_SYSPROTO_H_ struct lseek_args { int fd; int pad; off_t offset; int whence; }; #endif int sys_lseek(struct thread *td, struct lseek_args *uap) { return (kern_lseek(td, uap->fd, uap->offset, uap->whence)); } int kern_lseek(struct thread *td, int fd, off_t offset, int whence) { struct file *fp; int error; AUDIT_ARG_FD(fd); error = fget(td, fd, &cap_seek_rights, &fp); if (error != 0) return (error); error = (fp->f_ops->fo_flags & DFLAG_SEEKABLE) != 0 ? fo_seek(fp, offset, whence, td) : ESPIPE; fdrop(fp, td); return (error); } #if defined(COMPAT_43) /* * Reposition read/write file offset. */ #ifndef _SYS_SYSPROTO_H_ struct olseek_args { int fd; long offset; int whence; }; #endif int olseek(struct thread *td, struct olseek_args *uap) { return (kern_lseek(td, uap->fd, uap->offset, uap->whence)); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD6) /* Version with the 'pad' argument */ int freebsd6_lseek(struct thread *td, struct freebsd6_lseek_args *uap) { return (kern_lseek(td, uap->fd, uap->offset, uap->whence)); } #endif /* * Check access permissions using passed credentials. */ static int vn_access(struct vnode *vp, int user_flags, struct ucred *cred, struct thread *td) { accmode_t accmode; int error; /* Flags == 0 means only check for existence. */ if (user_flags == 0) return (0); accmode = 0; if (user_flags & R_OK) accmode |= VREAD; if (user_flags & W_OK) accmode |= VWRITE; if (user_flags & X_OK) accmode |= VEXEC; #ifdef MAC error = mac_vnode_check_access(cred, vp, accmode); if (error != 0) return (error); #endif if ((accmode & VWRITE) == 0 || (error = vn_writechk(vp)) == 0) error = VOP_ACCESS(vp, accmode, cred, td); return (error); } /* * Check access permissions using "real" credentials. */ #ifndef _SYS_SYSPROTO_H_ struct access_args { char *path; int amode; }; #endif int sys_access(struct thread *td, struct access_args *uap) { return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE, 0, uap->amode)); } #ifndef _SYS_SYSPROTO_H_ struct faccessat_args { int dirfd; char *path; int amode; int flag; } #endif int sys_faccessat(struct thread *td, struct faccessat_args *uap) { return (kern_accessat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag, uap->amode)); } int kern_accessat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int flag, int amode) { struct ucred *cred, *usecred; struct vnode *vp; struct nameidata nd; int error; if ((flag & ~(AT_EACCESS | AT_RESOLVE_BENEATH | AT_EMPTY_PATH)) != 0) return (EINVAL); if (amode != F_OK && (amode & ~(R_OK | W_OK | X_OK)) != 0) return (EINVAL); /* * Create and modify a temporary credential instead of one that * is potentially shared (if we need one). */ cred = td->td_ucred; if ((flag & AT_EACCESS) == 0 && ((cred->cr_uid != cred->cr_ruid || cred->cr_rgid != cred->cr_groups[0]))) { usecred = crdup(cred); usecred->cr_uid = cred->cr_ruid; usecred->cr_groups[0] = cred->cr_rgid; td->td_ucred = usecred; } else usecred = cred; AUDIT_ARG_VALUE(amode); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1 | at2cnpflags(flag, AT_RESOLVE_BENEATH | AT_EMPTY_PATH), pathseg, path, fd, &cap_fstat_rights); if ((error = namei(&nd)) != 0) goto out; vp = nd.ni_vp; error = vn_access(vp, amode, usecred, td); NDFREE_PNBUF(&nd); vput(vp); out: if (usecred != cred) { td->td_ucred = cred; crfree(usecred); } return (error); } /* * Check access permissions using "effective" credentials. */ #ifndef _SYS_SYSPROTO_H_ struct eaccess_args { char *path; int amode; }; #endif int sys_eaccess(struct thread *td, struct eaccess_args *uap) { return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE, AT_EACCESS, uap->amode)); } #if defined(COMPAT_43) /* * Get file status; this version follows links. */ #ifndef _SYS_SYSPROTO_H_ struct ostat_args { char *path; struct ostat *ub; }; #endif int ostat(struct thread *td, struct ostat_args *uap) { struct stat sb; struct ostat osb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtstat(&sb, &osb); return (copyout(&osb, uap->ub, sizeof (osb))); } /* * Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct olstat_args { char *path; struct ostat *ub; }; #endif int olstat(struct thread *td, struct olstat_args *uap) { struct stat sb; struct ostat osb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtstat(&sb, &osb); return (copyout(&osb, uap->ub, sizeof (osb))); } /* * Convert from an old to a new stat structure. * XXX: many values are blindly truncated. */ void cvtstat(struct stat *st, struct ostat *ost) { bzero(ost, sizeof(*ost)); ost->st_dev = st->st_dev; ost->st_ino = st->st_ino; ost->st_mode = st->st_mode; ost->st_nlink = st->st_nlink; ost->st_uid = st->st_uid; ost->st_gid = st->st_gid; ost->st_rdev = st->st_rdev; ost->st_size = MIN(st->st_size, INT32_MAX); ost->st_atim = st->st_atim; ost->st_mtim = st->st_mtim; ost->st_ctim = st->st_ctim; ost->st_blksize = st->st_blksize; ost->st_blocks = st->st_blocks; ost->st_flags = st->st_flags; ost->st_gen = st->st_gen; } #endif /* COMPAT_43 */ #if defined(COMPAT_43) || defined(COMPAT_FREEBSD11) int ino64_trunc_error; SYSCTL_INT(_vfs, OID_AUTO, ino64_trunc_error, CTLFLAG_RW, &ino64_trunc_error, 0, "Error on truncation of device, file or inode number, or link count"); int freebsd11_cvtstat(struct stat *st, struct freebsd11_stat *ost) { ost->st_dev = st->st_dev; if (ost->st_dev != st->st_dev) { switch (ino64_trunc_error) { default: /* * Since dev_t is almost raw, don't clamp to the * maximum for case 2, but ignore the error. */ break; case 1: return (EOVERFLOW); } } ost->st_ino = st->st_ino; if (ost->st_ino != st->st_ino) { switch (ino64_trunc_error) { default: case 0: break; case 1: return (EOVERFLOW); case 2: ost->st_ino = UINT32_MAX; break; } } ost->st_mode = st->st_mode; ost->st_nlink = st->st_nlink; if (ost->st_nlink != st->st_nlink) { switch (ino64_trunc_error) { default: case 0: break; case 1: return (EOVERFLOW); case 2: ost->st_nlink = UINT16_MAX; break; } } ost->st_uid = st->st_uid; ost->st_gid = st->st_gid; ost->st_rdev = st->st_rdev; if (ost->st_rdev != st->st_rdev) { switch (ino64_trunc_error) { default: break; case 1: return (EOVERFLOW); } } ost->st_atim = st->st_atim; ost->st_mtim = st->st_mtim; ost->st_ctim = st->st_ctim; ost->st_size = st->st_size; ost->st_blocks = st->st_blocks; ost->st_blksize = st->st_blksize; ost->st_flags = st->st_flags; ost->st_gen = st->st_gen; ost->st_lspare = 0; ost->st_birthtim = st->st_birthtim; bzero((char *)&ost->st_birthtim + sizeof(ost->st_birthtim), sizeof(*ost) - offsetof(struct freebsd11_stat, st_birthtim) - sizeof(ost->st_birthtim)); return (0); } int freebsd11_stat(struct thread *td, struct freebsd11_stat_args* uap) { struct stat sb; struct freebsd11_stat osb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtstat(&sb, &osb); if (error == 0) error = copyout(&osb, uap->ub, sizeof(osb)); return (error); } int freebsd11_lstat(struct thread *td, struct freebsd11_lstat_args* uap) { struct stat sb; struct freebsd11_stat osb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtstat(&sb, &osb); if (error == 0) error = copyout(&osb, uap->ub, sizeof(osb)); return (error); } int freebsd11_fhstat(struct thread *td, struct freebsd11_fhstat_args* uap) { struct fhandle fh; struct stat sb; struct freebsd11_stat osb; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); error = kern_fhstat(td, fh, &sb); if (error != 0) return (error); error = freebsd11_cvtstat(&sb, &osb); if (error == 0) error = copyout(&osb, uap->sb, sizeof(osb)); return (error); } int freebsd11_fstatat(struct thread *td, struct freebsd11_fstatat_args* uap) { struct stat sb; struct freebsd11_stat osb; int error; error = kern_statat(td, uap->flag, uap->fd, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtstat(&sb, &osb); if (error == 0) error = copyout(&osb, uap->buf, sizeof(osb)); return (error); } #endif /* COMPAT_FREEBSD11 */ /* * Get file status */ #ifndef _SYS_SYSPROTO_H_ struct fstatat_args { int fd; char *path; struct stat *buf; int flag; } #endif int sys_fstatat(struct thread *td, struct fstatat_args *uap) { struct stat sb; int error; error = kern_statat(td, uap->flag, uap->fd, uap->path, UIO_USERSPACE, &sb, NULL); if (error == 0) error = copyout(&sb, uap->buf, sizeof (sb)); return (error); } int kern_statat(struct thread *td, int flag, int fd, const char *path, enum uio_seg pathseg, struct stat *sbp, void (*hook)(struct vnode *vp, struct stat *sbp)) { struct nameidata nd; int error; if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH)) != 0) return (EINVAL); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_RESOLVE_BENEATH | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH) | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, &cap_fstat_rights); if ((error = namei(&nd)) != 0) { if (error == ENOTDIR && (nd.ni_resflags & NIRES_EMPTYPATH) != 0) error = kern_fstat(td, fd, sbp); return (error); } error = VOP_STAT(nd.ni_vp, sbp, td->td_ucred, NOCRED); if (__predict_false(hook != NULL)) { if (error == 0) { hook(nd.ni_vp, sbp); } } NDFREE_PNBUF(&nd); vput(nd.ni_vp); #ifdef __STAT_TIME_T_EXT sbp->st_atim_ext = 0; sbp->st_mtim_ext = 0; sbp->st_ctim_ext = 0; sbp->st_btim_ext = 0; #endif #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrstat_error(sbp, error); #endif return (error); } #if defined(COMPAT_FREEBSD11) /* * Implementation of the NetBSD [l]stat() functions. */ int freebsd11_cvtnstat(struct stat *sb, struct nstat *nsb) { struct freebsd11_stat sb11; int error; error = freebsd11_cvtstat(sb, &sb11); if (error != 0) return (error); bzero(nsb, sizeof(*nsb)); CP(sb11, *nsb, st_dev); CP(sb11, *nsb, st_ino); CP(sb11, *nsb, st_mode); CP(sb11, *nsb, st_nlink); CP(sb11, *nsb, st_uid); CP(sb11, *nsb, st_gid); CP(sb11, *nsb, st_rdev); CP(sb11, *nsb, st_atim); CP(sb11, *nsb, st_mtim); CP(sb11, *nsb, st_ctim); CP(sb11, *nsb, st_size); CP(sb11, *nsb, st_blocks); CP(sb11, *nsb, st_blksize); CP(sb11, *nsb, st_flags); CP(sb11, *nsb, st_gen); CP(sb11, *nsb, st_birthtim); return (0); } #ifndef _SYS_SYSPROTO_H_ struct freebsd11_nstat_args { char *path; struct nstat *ub; }; #endif int freebsd11_nstat(struct thread *td, struct freebsd11_nstat_args *uap) { struct stat sb; struct nstat nsb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtnstat(&sb, &nsb); if (error == 0) error = copyout(&nsb, uap->ub, sizeof (nsb)); return (error); } /* * NetBSD lstat. Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd11_nlstat_args { char *path; struct nstat *ub; }; #endif int freebsd11_nlstat(struct thread *td, struct freebsd11_nlstat_args *uap) { struct stat sb; struct nstat nsb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtnstat(&sb, &nsb); if (error == 0) error = copyout(&nsb, uap->ub, sizeof (nsb)); return (error); } #endif /* COMPAT_FREEBSD11 */ /* * Get configurable pathname variables. */ #ifndef _SYS_SYSPROTO_H_ struct pathconf_args { char *path; int name; }; #endif int sys_pathconf(struct thread *td, struct pathconf_args *uap) { long value; int error; error = kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, FOLLOW, &value); if (error == 0) td->td_retval[0] = value; return (error); } #ifndef _SYS_SYSPROTO_H_ struct lpathconf_args { char *path; int name; }; #endif int sys_lpathconf(struct thread *td, struct lpathconf_args *uap) { long value; int error; error = kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, NOFOLLOW, &value); if (error == 0) td->td_retval[0] = value; return (error); } int kern_pathconf(struct thread *td, const char *path, enum uio_seg pathseg, int name, u_long flags, long *valuep) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, LOCKSHARED | LOCKLEAF | AUDITVNODE1 | flags, pathseg, path); if ((error = namei(&nd)) != 0) return (error); NDFREE_PNBUF(&nd); error = VOP_PATHCONF(nd.ni_vp, name, valuep); vput(nd.ni_vp); return (error); } /* * Return target name of a symbolic link. */ #ifndef _SYS_SYSPROTO_H_ struct readlink_args { char *path; char *buf; size_t count; }; #endif int sys_readlink(struct thread *td, struct readlink_args *uap) { return (kern_readlinkat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->buf, UIO_USERSPACE, uap->count)); } #ifndef _SYS_SYSPROTO_H_ struct readlinkat_args { int fd; char *path; char *buf; size_t bufsize; }; #endif int sys_readlinkat(struct thread *td, struct readlinkat_args *uap) { return (kern_readlinkat(td, uap->fd, uap->path, UIO_USERSPACE, uap->buf, UIO_USERSPACE, uap->bufsize)); } int kern_readlinkat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, char *buf, enum uio_seg bufseg, size_t count) { struct vnode *vp; struct nameidata nd; int error; if (count > IOSIZE_MAX) return (EINVAL); NDINIT_AT(&nd, LOOKUP, NOFOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1 | EMPTYPATH, pathseg, path, fd); if ((error = namei(&nd)) != 0) return (error); NDFREE_PNBUF(&nd); vp = nd.ni_vp; error = kern_readlink_vp(vp, buf, bufseg, count, td); vput(vp); return (error); } /* * Helper function to readlink from a vnode */ static int kern_readlink_vp(struct vnode *vp, char *buf, enum uio_seg bufseg, size_t count, struct thread *td) { struct iovec aiov; struct uio auio; int error; ASSERT_VOP_LOCKED(vp, "kern_readlink_vp(): vp not locked"); #ifdef MAC error = mac_vnode_check_readlink(td->td_ucred, vp); if (error != 0) return (error); #endif if (vp->v_type != VLNK && (vp->v_vflag & VV_READLINK) == 0) return (EINVAL); aiov.iov_base = buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = bufseg; auio.uio_td = td; auio.uio_resid = count; error = VOP_READLINK(vp, &auio, td->td_ucred); td->td_retval[0] = count - auio.uio_resid; return (error); } /* * Common implementation code for chflags() and fchflags(). */ static int setfflags(struct thread *td, struct vnode *vp, u_long flags) { struct mount *mp; struct vattr vattr; int error; /* We can't support the value matching VNOVAL. */ if (flags == VNOVAL) return (EOPNOTSUPP); /* * Prevent non-root users from setting flags on devices. When * a device is reused, users can retain ownership of the device * if they are allowed to set flags and programs assume that * chown can't fail when done as root. */ if (vp->v_type == VCHR || vp->v_type == VBLK) { error = priv_check(td, PRIV_VFS_CHFLAGS_DEV); if (error != 0) return (error); } - if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) + if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0) return (error); VATTR_NULL(&vattr); vattr.va_flags = flags; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); #ifdef MAC error = mac_vnode_check_setflags(td->td_ucred, vp, vattr.va_flags); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, td->td_ucred); VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } /* * Change flags of a file given a path name. */ #ifndef _SYS_SYSPROTO_H_ struct chflags_args { const char *path; u_long flags; }; #endif int sys_chflags(struct thread *td, struct chflags_args *uap) { return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, 0)); } #ifndef _SYS_SYSPROTO_H_ struct chflagsat_args { int fd; const char *path; u_long flags; int atflag; } #endif int sys_chflagsat(struct thread *td, struct chflagsat_args *uap) { return (kern_chflagsat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flags, uap->atflag)); } /* * Same as chflags() but doesn't follow symlinks. */ #ifndef _SYS_SYSPROTO_H_ struct lchflags_args { const char *path; u_long flags; }; #endif int sys_lchflags(struct thread *td, struct lchflags_args *uap) { return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, AT_SYMLINK_NOFOLLOW)); } static int kern_chflagsat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, u_long flags, int atflag) { struct nameidata nd; int error; if ((atflag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH)) != 0) return (EINVAL); AUDIT_ARG_FFLAGS(flags); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(atflag, AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path, fd, &cap_fchflags_rights); if ((error = namei(&nd)) != 0) return (error); NDFREE_PNBUF(&nd); error = setfflags(td, nd.ni_vp, flags); vrele(nd.ni_vp); return (error); } /* * Change flags of a file given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchflags_args { int fd; u_long flags; }; #endif int sys_fchflags(struct thread *td, struct fchflags_args *uap) { struct file *fp; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_FFLAGS(uap->flags); error = getvnode(td, uap->fd, &cap_fchflags_rights, &fp); if (error != 0) return (error); #ifdef AUDIT if (AUDITING_TD(td)) { vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode); } #endif error = setfflags(td, fp->f_vnode, uap->flags); fdrop(fp, td); return (error); } /* * Common implementation code for chmod(), lchmod() and fchmod(). */ int setfmode(struct thread *td, struct ucred *cred, struct vnode *vp, int mode) { struct mount *mp; struct vattr vattr; int error; - if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) + if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VATTR_NULL(&vattr); vattr.va_mode = mode & ALLPERMS; #ifdef MAC error = mac_vnode_check_setmode(cred, vp, vattr.va_mode); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } /* * Change mode of a file given path name. */ #ifndef _SYS_SYSPROTO_H_ struct chmod_args { char *path; int mode; }; #endif int sys_chmod(struct thread *td, struct chmod_args *uap) { return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, 0)); } #ifndef _SYS_SYSPROTO_H_ struct fchmodat_args { int dirfd; char *path; mode_t mode; int flag; } #endif int sys_fchmodat(struct thread *td, struct fchmodat_args *uap) { return (kern_fchmodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode, uap->flag)); } /* * Change mode of a file given path name (don't follow links.) */ #ifndef _SYS_SYSPROTO_H_ struct lchmod_args { char *path; int mode; }; #endif int sys_lchmod(struct thread *td, struct lchmod_args *uap) { return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, AT_SYMLINK_NOFOLLOW)); } int kern_fchmodat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, mode_t mode, int flag) { struct nameidata nd; int error; if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH)) != 0) return (EINVAL); AUDIT_ARG_MODE(mode); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path, fd, &cap_fchmod_rights); if ((error = namei(&nd)) != 0) return (error); NDFREE_PNBUF(&nd); error = setfmode(td, td->td_ucred, nd.ni_vp, mode); vrele(nd.ni_vp); return (error); } /* * Change mode of a file given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchmod_args { int fd; int mode; }; #endif int sys_fchmod(struct thread *td, struct fchmod_args *uap) { struct file *fp; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_MODE(uap->mode); error = fget(td, uap->fd, &cap_fchmod_rights, &fp); if (error != 0) return (error); error = fo_chmod(fp, uap->mode, td->td_ucred, td); fdrop(fp, td); return (error); } /* * Common implementation for chown(), lchown(), and fchown() */ int setfown(struct thread *td, struct ucred *cred, struct vnode *vp, uid_t uid, gid_t gid) { struct mount *mp; struct vattr vattr; int error; - if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) + if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VATTR_NULL(&vattr); vattr.va_uid = uid; vattr.va_gid = gid; #ifdef MAC error = mac_vnode_check_setowner(cred, vp, vattr.va_uid, vattr.va_gid); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } /* * Set ownership given a path name. */ #ifndef _SYS_SYSPROTO_H_ struct chown_args { char *path; int uid; int gid; }; #endif int sys_chown(struct thread *td, struct chown_args *uap) { return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid, uap->gid, 0)); } #ifndef _SYS_SYSPROTO_H_ struct fchownat_args { int fd; const char * path; uid_t uid; gid_t gid; int flag; }; #endif int sys_fchownat(struct thread *td, struct fchownat_args *uap) { return (kern_fchownat(td, uap->fd, uap->path, UIO_USERSPACE, uap->uid, uap->gid, uap->flag)); } int kern_fchownat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int uid, int gid, int flag) { struct nameidata nd; int error; if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH)) != 0) return (EINVAL); AUDIT_ARG_OWNER(uid, gid); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path, fd, &cap_fchown_rights); if ((error = namei(&nd)) != 0) return (error); NDFREE_PNBUF(&nd); error = setfown(td, td->td_ucred, nd.ni_vp, uid, gid); vrele(nd.ni_vp); return (error); } /* * Set ownership given a path name, do not cross symlinks. */ #ifndef _SYS_SYSPROTO_H_ struct lchown_args { char *path; int uid; int gid; }; #endif int sys_lchown(struct thread *td, struct lchown_args *uap) { return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid, uap->gid, AT_SYMLINK_NOFOLLOW)); } /* * Set ownership given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchown_args { int fd; int uid; int gid; }; #endif int sys_fchown(struct thread *td, struct fchown_args *uap) { struct file *fp; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_OWNER(uap->uid, uap->gid); error = fget(td, uap->fd, &cap_fchown_rights, &fp); if (error != 0) return (error); error = fo_chown(fp, uap->uid, uap->gid, td->td_ucred, td); fdrop(fp, td); return (error); } /* * Common implementation code for utimes(), lutimes(), and futimes(). */ static int getutimes(const struct timeval *usrtvp, enum uio_seg tvpseg, struct timespec *tsp) { struct timeval tv[2]; const struct timeval *tvp; int error; if (usrtvp == NULL) { vfs_timestamp(&tsp[0]); tsp[1] = tsp[0]; } else { if (tvpseg == UIO_SYSSPACE) { tvp = usrtvp; } else { if ((error = copyin(usrtvp, tv, sizeof(tv))) != 0) return (error); tvp = tv; } if (tvp[0].tv_usec < 0 || tvp[0].tv_usec >= 1000000 || tvp[1].tv_usec < 0 || tvp[1].tv_usec >= 1000000) return (EINVAL); TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]); TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]); } return (0); } /* * Common implementation code for futimens(), utimensat(). */ #define UTIMENS_NULL 0x1 #define UTIMENS_EXIT 0x2 static int getutimens(const struct timespec *usrtsp, enum uio_seg tspseg, struct timespec *tsp, int *retflags) { struct timespec tsnow; int error; vfs_timestamp(&tsnow); *retflags = 0; if (usrtsp == NULL) { tsp[0] = tsnow; tsp[1] = tsnow; *retflags |= UTIMENS_NULL; return (0); } if (tspseg == UIO_SYSSPACE) { tsp[0] = usrtsp[0]; tsp[1] = usrtsp[1]; } else if ((error = copyin(usrtsp, tsp, sizeof(*tsp) * 2)) != 0) return (error); if (tsp[0].tv_nsec == UTIME_OMIT && tsp[1].tv_nsec == UTIME_OMIT) *retflags |= UTIMENS_EXIT; if (tsp[0].tv_nsec == UTIME_NOW && tsp[1].tv_nsec == UTIME_NOW) *retflags |= UTIMENS_NULL; if (tsp[0].tv_nsec == UTIME_OMIT) tsp[0].tv_sec = VNOVAL; else if (tsp[0].tv_nsec == UTIME_NOW) tsp[0] = tsnow; else if (tsp[0].tv_nsec < 0 || tsp[0].tv_nsec >= 1000000000L) return (EINVAL); if (tsp[1].tv_nsec == UTIME_OMIT) tsp[1].tv_sec = VNOVAL; else if (tsp[1].tv_nsec == UTIME_NOW) tsp[1] = tsnow; else if (tsp[1].tv_nsec < 0 || tsp[1].tv_nsec >= 1000000000L) return (EINVAL); return (0); } /* * Common implementation code for utimes(), lutimes(), futimes(), futimens(), * and utimensat(). */ static int setutimes(struct thread *td, struct vnode *vp, const struct timespec *ts, int numtimes, int nullflag) { struct mount *mp; struct vattr vattr; int error; bool setbirthtime; setbirthtime = false; vattr.va_birthtime.tv_sec = VNOVAL; vattr.va_birthtime.tv_nsec = 0; - if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) + if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (numtimes < 3 && VOP_GETATTR(vp, &vattr, td->td_ucred) == 0 && timespeccmp(&ts[1], &vattr.va_birthtime, < )) setbirthtime = true; VATTR_NULL(&vattr); vattr.va_atime = ts[0]; vattr.va_mtime = ts[1]; if (setbirthtime) vattr.va_birthtime = ts[1]; if (numtimes > 2) vattr.va_birthtime = ts[2]; if (nullflag) vattr.va_vaflags |= VA_UTIMES_NULL; #ifdef MAC error = mac_vnode_check_setutimes(td->td_ucred, vp, vattr.va_atime, vattr.va_mtime); #endif if (error == 0) error = VOP_SETATTR(vp, &vattr, td->td_ucred); VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct utimes_args { char *path; struct timeval *tptr; }; #endif int sys_utimes(struct thread *td, struct utimes_args *uap) { return (kern_utimesat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->tptr, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct futimesat_args { int fd; const char * path; const struct timeval * times; }; #endif int sys_futimesat(struct thread *td, struct futimesat_args *uap) { return (kern_utimesat(td, uap->fd, uap->path, UIO_USERSPACE, uap->times, UIO_USERSPACE)); } int kern_utimesat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, const struct timeval *tptr, enum uio_seg tptrseg) { struct nameidata nd; struct timespec ts[2]; int error; if ((error = getutimes(tptr, tptrseg, ts)) != 0) return (error); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd, &cap_futimes_rights); if ((error = namei(&nd)) != 0) return (error); NDFREE_PNBUF(&nd); error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL); vrele(nd.ni_vp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct lutimes_args { char *path; struct timeval *tptr; }; #endif int sys_lutimes(struct thread *td, struct lutimes_args *uap) { return (kern_lutimes(td, uap->path, UIO_USERSPACE, uap->tptr, UIO_USERSPACE)); } int kern_lutimes(struct thread *td, const char *path, enum uio_seg pathseg, const struct timeval *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct nameidata nd; int error; if ((error = getutimes(tptr, tptrseg, ts)) != 0) return (error); NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1, pathseg, path); if ((error = namei(&nd)) != 0) return (error); NDFREE_PNBUF(&nd); error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL); vrele(nd.ni_vp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct futimes_args { int fd; struct timeval *tptr; }; #endif int sys_futimes(struct thread *td, struct futimes_args *uap) { return (kern_futimes(td, uap->fd, uap->tptr, UIO_USERSPACE)); } int kern_futimes(struct thread *td, int fd, const struct timeval *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct file *fp; int error; AUDIT_ARG_FD(fd); error = getutimes(tptr, tptrseg, ts); if (error != 0) return (error); error = getvnode(td, fd, &cap_futimes_rights, &fp); if (error != 0) return (error); #ifdef AUDIT if (AUDITING_TD(td)) { vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode); } #endif error = setutimes(td, fp->f_vnode, ts, 2, tptr == NULL); fdrop(fp, td); return (error); } int sys_futimens(struct thread *td, struct futimens_args *uap) { return (kern_futimens(td, uap->fd, uap->times, UIO_USERSPACE)); } int kern_futimens(struct thread *td, int fd, const struct timespec *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct file *fp; int error, flags; AUDIT_ARG_FD(fd); error = getutimens(tptr, tptrseg, ts, &flags); if (error != 0) return (error); if (flags & UTIMENS_EXIT) return (0); error = getvnode(td, fd, &cap_futimes_rights, &fp); if (error != 0) return (error); #ifdef AUDIT if (AUDITING_TD(td)) { vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode); } #endif error = setutimes(td, fp->f_vnode, ts, 2, flags & UTIMENS_NULL); fdrop(fp, td); return (error); } int sys_utimensat(struct thread *td, struct utimensat_args *uap) { return (kern_utimensat(td, uap->fd, uap->path, UIO_USERSPACE, uap->times, UIO_USERSPACE, uap->flag)); } int kern_utimensat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, const struct timespec *tptr, enum uio_seg tptrseg, int flag) { struct nameidata nd; struct timespec ts[2]; int error, flags; if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH)) != 0) return (EINVAL); if ((error = getutimens(tptr, tptrseg, ts, &flags)) != 0) return (error); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path, fd, &cap_futimes_rights); if ((error = namei(&nd)) != 0) return (error); /* * We are allowed to call namei() regardless of 2xUTIME_OMIT. * POSIX states: * "If both tv_nsec fields are UTIME_OMIT... EACCESS may be detected." * "Search permission is denied by a component of the path prefix." */ NDFREE_PNBUF(&nd); if ((flags & UTIMENS_EXIT) == 0) error = setutimes(td, nd.ni_vp, ts, 2, flags & UTIMENS_NULL); vrele(nd.ni_vp); return (error); } /* * Truncate a file given its path name. */ #ifndef _SYS_SYSPROTO_H_ struct truncate_args { char *path; int pad; off_t length; }; #endif int sys_truncate(struct thread *td, struct truncate_args *uap) { return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length)); } int kern_truncate(struct thread *td, const char *path, enum uio_seg pathseg, off_t length) { struct mount *mp; struct vnode *vp; void *rl_cookie; struct nameidata nd; int error; if (length < 0) return (EINVAL); NDPREINIT(&nd); retry: NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; NDFREE_PNBUF(&nd); rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); - if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) { + if ((error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0) { vn_rangelock_unlock(vp, rl_cookie); vrele(vp); return (error); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef MAC error = mac_vnode_check_write(td->td_ucred, NOCRED, vp); if (error != 0) goto out; #endif error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); if (error != 0) goto out; error = vn_truncate_locked(vp, length, false, td->td_ucred); out: VOP_UNLOCK(vp); vn_finished_write(mp); vn_rangelock_unlock(vp, rl_cookie); vrele(vp); if (error == ERELOOKUP) goto retry; return (error); } #if defined(COMPAT_43) /* * Truncate a file given its path name. */ #ifndef _SYS_SYSPROTO_H_ struct otruncate_args { char *path; long length; }; #endif int otruncate(struct thread *td, struct otruncate_args *uap) { return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length)); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD6) /* Versions with the pad argument */ int freebsd6_truncate(struct thread *td, struct freebsd6_truncate_args *uap) { return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length)); } int freebsd6_ftruncate(struct thread *td, struct freebsd6_ftruncate_args *uap) { return (kern_ftruncate(td, uap->fd, uap->length)); } #endif int kern_fsync(struct thread *td, int fd, bool fullsync) { struct vnode *vp; struct mount *mp; struct file *fp; int error; AUDIT_ARG_FD(fd); error = getvnode(td, fd, &cap_fsync_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; #if 0 if (!fullsync) /* XXXKIB: compete outstanding aio writes */; #endif retry: - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error != 0) goto drop; vn_lock(vp, vn_lktype_write(mp, vp) | LK_RETRY); AUDIT_ARG_VNODE1(vp); if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } error = fullsync ? VOP_FSYNC(vp, MNT_WAIT, td) : VOP_FDATASYNC(vp, td); VOP_UNLOCK(vp); vn_finished_write(mp); if (error == ERELOOKUP) goto retry; drop: fdrop(fp, td); return (error); } /* * Sync an open file. */ #ifndef _SYS_SYSPROTO_H_ struct fsync_args { int fd; }; #endif int sys_fsync(struct thread *td, struct fsync_args *uap) { return (kern_fsync(td, uap->fd, true)); } int sys_fdatasync(struct thread *td, struct fdatasync_args *uap) { return (kern_fsync(td, uap->fd, false)); } /* * Rename files. Source and destination must either both be directories, or * both not be directories. If target is a directory, it must be empty. */ #ifndef _SYS_SYSPROTO_H_ struct rename_args { char *from; char *to; }; #endif int sys_rename(struct thread *td, struct rename_args *uap) { return (kern_renameat(td, AT_FDCWD, uap->from, AT_FDCWD, uap->to, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct renameat_args { int oldfd; char *old; int newfd; char *new; }; #endif int sys_renameat(struct thread *td, struct renameat_args *uap) { return (kern_renameat(td, uap->oldfd, uap->old, uap->newfd, uap->new, UIO_USERSPACE)); } #ifdef MAC static int kern_renameat_mac(struct thread *td, int oldfd, const char *old, int newfd, const char *new, enum uio_seg pathseg, struct nameidata *fromnd) { int error; NDINIT_ATRIGHTS(fromnd, DELETE, LOCKPARENT | LOCKLEAF | SAVESTART | AUDITVNODE1, pathseg, old, oldfd, &cap_renameat_source_rights); if ((error = namei(fromnd)) != 0) return (error); error = mac_vnode_check_rename_from(td->td_ucred, fromnd->ni_dvp, fromnd->ni_vp, &fromnd->ni_cnd); VOP_UNLOCK(fromnd->ni_dvp); if (fromnd->ni_dvp != fromnd->ni_vp) VOP_UNLOCK(fromnd->ni_vp); if (error != 0) { NDFREE_PNBUF(fromnd); vrele(fromnd->ni_dvp); vrele(fromnd->ni_vp); if (fromnd->ni_startdir) vrele(fromnd->ni_startdir); } return (error); } #endif int kern_renameat(struct thread *td, int oldfd, const char *old, int newfd, const char *new, enum uio_seg pathseg) { struct mount *mp = NULL; struct vnode *tvp, *fvp, *tdvp; struct nameidata fromnd, tond; uint64_t tondflags; int error; again: bwillwrite(); #ifdef MAC if (mac_vnode_check_rename_from_enabled()) { error = kern_renameat_mac(td, oldfd, old, newfd, new, pathseg, &fromnd); if (error != 0) return (error); } else { #endif NDINIT_ATRIGHTS(&fromnd, DELETE, WANTPARENT | SAVESTART | AUDITVNODE1, pathseg, old, oldfd, &cap_renameat_source_rights); if ((error = namei(&fromnd)) != 0) return (error); #ifdef MAC } #endif fvp = fromnd.ni_vp; tondflags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | AUDITVNODE2; if (fromnd.ni_vp->v_type == VDIR) tondflags |= WILLBEDIR; NDINIT_ATRIGHTS(&tond, RENAME, tondflags, pathseg, new, newfd, &cap_renameat_target_rights); if ((error = namei(&tond)) != 0) { /* Translate error code for rename("dir1", "dir2/."). */ if (error == EISDIR && fvp->v_type == VDIR) error = EINVAL; NDFREE_PNBUF(&fromnd); vrele(fromnd.ni_dvp); vrele(fvp); goto out1; } tdvp = tond.ni_dvp; tvp = tond.ni_vp; error = vn_start_write(fvp, &mp, V_NOWAIT); if (error != 0) { NDFREE_PNBUF(&fromnd); NDFREE_PNBUF(&tond); if (tvp != NULL) vput(tvp); if (tdvp == tvp) vrele(tdvp); else vput(tdvp); vrele(fromnd.ni_dvp); vrele(fvp); vrele(tond.ni_startdir); if (fromnd.ni_startdir != NULL) vrele(fromnd.ni_startdir); - error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); + error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH); if (error != 0) return (error); goto again; } if (tvp != NULL) { if (fvp->v_type == VDIR && tvp->v_type != VDIR) { error = ENOTDIR; goto out; } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef CAPABILITIES if (newfd != AT_FDCWD && (tond.ni_resflags & NIRES_ABS) == 0) { /* * If the target already exists we require CAP_UNLINKAT * from 'newfd', when newfd was used for the lookup. */ error = cap_check(&tond.ni_filecaps.fc_rights, &cap_unlinkat_rights); if (error != 0) goto out; } #endif } if (fvp == tdvp) { error = EINVAL; goto out; } /* * If the source is the same as the destination (that is, if they * are links to the same vnode), then there is nothing to do. */ if (fvp == tvp) error = ERESTART; #ifdef MAC else error = mac_vnode_check_rename_to(td->td_ucred, tdvp, tond.ni_vp, fromnd.ni_dvp == tdvp, &tond.ni_cnd); #endif out: if (error == 0) { error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); NDFREE_PNBUF(&fromnd); NDFREE_PNBUF(&tond); } else { NDFREE_PNBUF(&fromnd); NDFREE_PNBUF(&tond); if (tvp != NULL) vput(tvp); if (tdvp == tvp) vrele(tdvp); else vput(tdvp); vrele(fromnd.ni_dvp); vrele(fvp); } vrele(tond.ni_startdir); vn_finished_write(mp); out1: if (fromnd.ni_startdir) vrele(fromnd.ni_startdir); if (error == ERESTART) return (0); if (error == ERELOOKUP) goto again; return (error); } /* * Make a directory file. */ #ifndef _SYS_SYSPROTO_H_ struct mkdir_args { char *path; int mode; }; #endif int sys_mkdir(struct thread *td, struct mkdir_args *uap) { return (kern_mkdirat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct mkdirat_args { int fd; char *path; mode_t mode; }; #endif int sys_mkdirat(struct thread *td, struct mkdirat_args *uap) { return (kern_mkdirat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode)); } int kern_mkdirat(struct thread *td, int fd, const char *path, enum uio_seg segflg, int mode) { struct mount *mp; struct vattr vattr; struct nameidata nd; int error; AUDIT_ARG_MODE(mode); NDPREINIT(&nd); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | AUDITVNODE1 | NC_NOMAKEENTRY | NC_KEEPPOSENTRY | FAILIFEXISTS | WILLBEDIR, segflg, path, fd, &cap_mkdirat_rights); if ((error = namei(&nd)) != 0) return (error); if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE_PNBUF(&nd); vput(nd.ni_dvp); - if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) + if ((error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH)) != 0) return (error); goto restart; } VATTR_NULL(&vattr); vattr.va_type = VDIR; vattr.va_mode = (mode & ACCESSPERMS) &~ td->td_proc->p_pd->pd_cmask; #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out; #endif error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); #ifdef MAC out: #endif NDFREE_PNBUF(&nd); VOP_VPUT_PAIR(nd.ni_dvp, error == 0 ? &nd.ni_vp : NULL, true); vn_finished_write(mp); if (error == ERELOOKUP) goto restart; return (error); } /* * Remove a directory file. */ #ifndef _SYS_SYSPROTO_H_ struct rmdir_args { char *path; }; #endif int sys_rmdir(struct thread *td, struct rmdir_args *uap) { return (kern_frmdirat(td, AT_FDCWD, uap->path, FD_NONE, UIO_USERSPACE, 0)); } int kern_frmdirat(struct thread *td, int dfd, const char *path, int fd, enum uio_seg pathseg, int flag) { struct mount *mp; struct vnode *vp; struct file *fp; struct nameidata nd; cap_rights_t rights; int error; fp = NULL; if (fd != FD_NONE) { error = getvnode(td, fd, cap_rights_init_one(&rights, CAP_LOOKUP), &fp); if (error != 0) return (error); } NDPREINIT(&nd); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1 | at2cnpflags(flag, AT_RESOLVE_BENEATH), pathseg, path, dfd, &cap_unlinkat_rights); if ((error = namei(&nd)) != 0) goto fdout; vp = nd.ni_vp; if (vp->v_type != VDIR) { error = ENOTDIR; goto out; } /* * No rmdir "." please. */ if (nd.ni_dvp == vp) { error = EINVAL; goto out; } /* * The root of a mounted filesystem cannot be deleted. */ if (vp->v_vflag & VV_ROOT) { error = EBUSY; goto out; } if (fp != NULL && fp->f_vnode != vp) { if (VN_IS_DOOMED(fp->f_vnode)) error = EBADF; else error = EDEADLK; goto out; } #ifdef MAC error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); if (error != 0) goto out; #endif if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE_PNBUF(&nd); vput(vp); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); - if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) + if ((error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH)) != 0) goto fdout; goto restart; } vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK); error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); vn_finished_write(mp); out: NDFREE_PNBUF(&nd); vput(vp); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if (error == ERELOOKUP) goto restart; fdout: if (fp != NULL) fdrop(fp, td); return (error); } #if defined(COMPAT_43) || defined(COMPAT_FREEBSD11) int freebsd11_kern_getdirentries(struct thread *td, int fd, char *ubuf, u_int count, long *basep, void (*func)(struct freebsd11_dirent *)) { struct freebsd11_dirent dstdp; struct dirent *dp, *edp; char *dirbuf; off_t base; ssize_t resid, ucount; int error; /* XXX arbitrary sanity limit on `count'. */ count = min(count, 64 * 1024); dirbuf = malloc(count, M_TEMP, M_WAITOK); error = kern_getdirentries(td, fd, dirbuf, count, &base, &resid, UIO_SYSSPACE); if (error != 0) goto done; if (basep != NULL) *basep = base; ucount = 0; for (dp = (struct dirent *)dirbuf, edp = (struct dirent *)&dirbuf[count - resid]; ucount < count && dp < edp; ) { if (dp->d_reclen == 0) break; MPASS(dp->d_reclen >= _GENERIC_DIRLEN(0)); if (dp->d_namlen >= sizeof(dstdp.d_name)) continue; dstdp.d_type = dp->d_type; dstdp.d_namlen = dp->d_namlen; dstdp.d_fileno = dp->d_fileno; /* truncate */ if (dstdp.d_fileno != dp->d_fileno) { switch (ino64_trunc_error) { default: case 0: break; case 1: error = EOVERFLOW; goto done; case 2: dstdp.d_fileno = UINT32_MAX; break; } } dstdp.d_reclen = sizeof(dstdp) - sizeof(dstdp.d_name) + ((dp->d_namlen + 1 + 3) &~ 3); bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen); bzero(dstdp.d_name + dstdp.d_namlen, dstdp.d_reclen - offsetof(struct freebsd11_dirent, d_name) - dstdp.d_namlen); MPASS(dstdp.d_reclen <= dp->d_reclen); MPASS(ucount + dstdp.d_reclen <= count); if (func != NULL) func(&dstdp); error = copyout(&dstdp, ubuf + ucount, dstdp.d_reclen); if (error != 0) break; dp = (struct dirent *)((char *)dp + dp->d_reclen); ucount += dstdp.d_reclen; } done: free(dirbuf, M_TEMP); if (error == 0) td->td_retval[0] = ucount; return (error); } #endif /* COMPAT */ #ifdef COMPAT_43 static void ogetdirentries_cvt(struct freebsd11_dirent *dp) { #if (BYTE_ORDER == LITTLE_ENDIAN) /* * The expected low byte of dp->d_namlen is our dp->d_type. * The high MBZ byte of dp->d_namlen is our dp->d_namlen. */ dp->d_type = dp->d_namlen; dp->d_namlen = 0; #else /* * The dp->d_type is the high byte of the expected dp->d_namlen, * so must be zero'ed. */ dp->d_type = 0; #endif } /* * Read a block of directory entries in a filesystem independent format. */ #ifndef _SYS_SYSPROTO_H_ struct ogetdirentries_args { int fd; char *buf; u_int count; long *basep; }; #endif int ogetdirentries(struct thread *td, struct ogetdirentries_args *uap) { long loff; int error; error = kern_ogetdirentries(td, uap, &loff); if (error == 0) error = copyout(&loff, uap->basep, sizeof(long)); return (error); } int kern_ogetdirentries(struct thread *td, struct ogetdirentries_args *uap, long *ploff) { long base; int error; /* XXX arbitrary sanity limit on `count'. */ if (uap->count > 64 * 1024) return (EINVAL); error = freebsd11_kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base, ogetdirentries_cvt); if (error == 0 && uap->basep != NULL) error = copyout(&base, uap->basep, sizeof(long)); return (error); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD11) #ifndef _SYS_SYSPROTO_H_ struct freebsd11_getdirentries_args { int fd; char *buf; u_int count; long *basep; }; #endif int freebsd11_getdirentries(struct thread *td, struct freebsd11_getdirentries_args *uap) { long base; int error; error = freebsd11_kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base, NULL); if (error == 0 && uap->basep != NULL) error = copyout(&base, uap->basep, sizeof(long)); return (error); } int freebsd11_getdents(struct thread *td, struct freebsd11_getdents_args *uap) { struct freebsd11_getdirentries_args ap; ap.fd = uap->fd; ap.buf = uap->buf; ap.count = uap->count; ap.basep = NULL; return (freebsd11_getdirentries(td, &ap)); } #endif /* COMPAT_FREEBSD11 */ /* * Read a block of directory entries in a filesystem independent format. */ int sys_getdirentries(struct thread *td, struct getdirentries_args *uap) { off_t base; int error; error = kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base, NULL, UIO_USERSPACE); if (error != 0) return (error); if (uap->basep != NULL) error = copyout(&base, uap->basep, sizeof(off_t)); return (error); } int kern_getdirentries(struct thread *td, int fd, char *buf, size_t count, off_t *basep, ssize_t *residp, enum uio_seg bufseg) { struct vnode *vp; struct file *fp; struct uio auio; struct iovec aiov; off_t loff; int error, eofflag; off_t foffset; AUDIT_ARG_FD(fd); if (count > IOSIZE_MAX) return (EINVAL); auio.uio_resid = count; error = getvnode(td, fd, &cap_read_rights, &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; foffset = foffset_lock(fp, 0); unionread: if (vp->v_type != VDIR) { error = EINVAL; goto fail; } if (__predict_false((vp->v_vflag & VV_UNLINKED) != 0)) { error = ENOENT; goto fail; } aiov.iov_base = buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = bufseg; auio.uio_td = td; vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); loff = auio.uio_offset = foffset; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error == 0) #endif error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL); foffset = auio.uio_offset; if (error != 0) { VOP_UNLOCK(vp); goto fail; } if (count == auio.uio_resid && (vp->v_vflag & VV_ROOT) && (vp->v_mount->mnt_flag & MNT_UNION)) { struct vnode *tvp = vp; vp = vp->v_mount->mnt_vnodecovered; VREF(vp); fp->f_vnode = vp; foffset = 0; vput(tvp); goto unionread; } VOP_UNLOCK(vp); *basep = loff; if (residp != NULL) *residp = auio.uio_resid; td->td_retval[0] = count - auio.uio_resid; fail: foffset_unlock(fp, foffset, 0); fdrop(fp, td); return (error); } /* * Set the mode mask for creation of filesystem nodes. */ #ifndef _SYS_SYSPROTO_H_ struct umask_args { int newmask; }; #endif int sys_umask(struct thread *td, struct umask_args *uap) { struct pwddesc *pdp; pdp = td->td_proc->p_pd; PWDDESC_XLOCK(pdp); td->td_retval[0] = pdp->pd_cmask; pdp->pd_cmask = uap->newmask & ALLPERMS; PWDDESC_XUNLOCK(pdp); return (0); } /* * Void all references to file by ripping underlying filesystem away from * vnode. */ #ifndef _SYS_SYSPROTO_H_ struct revoke_args { char *path; }; #endif int sys_revoke(struct thread *td, struct revoke_args *uap) { struct vnode *vp; struct vattr vattr; struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; NDFREE_PNBUF(&nd); if (vp->v_type != VCHR || vp->v_rdev == NULL) { error = EINVAL; goto out; } #ifdef MAC error = mac_vnode_check_revoke(td->td_ucred, vp); if (error != 0) goto out; #endif error = VOP_GETATTR(vp, &vattr, td->td_ucred); if (error != 0) goto out; if (td->td_ucred->cr_uid != vattr.va_uid) { error = priv_check(td, PRIV_VFS_ADMIN); if (error != 0) goto out; } if (devfs_usecount(vp) > 0) VOP_REVOKE(vp, REVOKEALL); out: vput(vp); return (error); } /* * This variant of getvnode() allows O_PATH files. Caller should * ensure that returned file and vnode are only used for compatible * semantics. */ int getvnode_path(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { struct file *fp; int error; error = fget_unlocked(td, fd, rightsp, &fp); if (error != 0) return (error); /* * The file could be not of the vnode type, or it may be not * yet fully initialized, in which case the f_vnode pointer * may be set, but f_ops is still badfileops. E.g., * devfs_open() transiently create such situation to * facilitate csw d_fdopen(). * * Dupfdopen() handling in kern_openat() installs the * half-baked file into the process descriptor table, allowing * other thread to dereference it. Guard against the race by * checking f_ops. */ if (__predict_false(fp->f_vnode == NULL || fp->f_ops == &badfileops)) { fdrop(fp, td); *fpp = NULL; return (EINVAL); } *fpp = fp; return (0); } /* * Convert a user file descriptor to a kernel file entry and check * that, if it is a capability, the correct rights are present. * A reference on the file entry is held upon returning. */ int getvnode(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { int error; error = getvnode_path(td, fd, rightsp, fpp); if (__predict_false(error != 0)) return (error); /* * Filter out O_PATH file descriptors, most getvnode() callers * do not call fo_ methods. */ if (__predict_false((*fpp)->f_ops == &path_fileops)) { fdrop(*fpp, td); *fpp = NULL; error = EBADF; } return (error); } /* * Get an (NFS) file handle. */ #ifndef _SYS_SYSPROTO_H_ struct lgetfh_args { char *fname; fhandle_t *fhp; }; #endif int sys_lgetfh(struct thread *td, struct lgetfh_args *uap) { return (kern_getfhat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->fname, UIO_USERSPACE, uap->fhp, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct getfh_args { char *fname; fhandle_t *fhp; }; #endif int sys_getfh(struct thread *td, struct getfh_args *uap) { return (kern_getfhat(td, 0, AT_FDCWD, uap->fname, UIO_USERSPACE, uap->fhp, UIO_USERSPACE)); } /* * syscall for the rpc.lockd to use to translate an open descriptor into * a NFS file handle. * * warning: do not remove the priv_check() call or this becomes one giant * security hole. */ #ifndef _SYS_SYSPROTO_H_ struct getfhat_args { int fd; char *path; fhandle_t *fhp; int flags; }; #endif int sys_getfhat(struct thread *td, struct getfhat_args *uap) { return (kern_getfhat(td, uap->flags, uap->fd, uap->path, UIO_USERSPACE, uap->fhp, UIO_USERSPACE)); } int kern_getfhat(struct thread *td, int flags, int fd, const char *path, enum uio_seg pathseg, fhandle_t *fhp, enum uio_seg fhseg) { struct nameidata nd; fhandle_t fh; struct vnode *vp; int error; if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); NDINIT_AT(&nd, LOOKUP, at2cnpflags(flags, AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH) | LOCKLEAF | AUDITVNODE1, pathseg, path, fd); error = namei(&nd); if (error != 0) return (error); NDFREE_PNBUF(&nd); vp = nd.ni_vp; bzero(&fh, sizeof(fh)); fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; error = VOP_VPTOFH(vp, &fh.fh_fid); vput(vp); if (error == 0) { if (fhseg == UIO_USERSPACE) error = copyout(&fh, fhp, sizeof (fh)); else memcpy(fhp, &fh, sizeof(fh)); } return (error); } #ifndef _SYS_SYSPROTO_H_ struct fhlink_args { fhandle_t *fhp; const char *to; }; #endif int sys_fhlink(struct thread *td, struct fhlink_args *uap) { return (kern_fhlinkat(td, AT_FDCWD, uap->to, UIO_USERSPACE, uap->fhp)); } #ifndef _SYS_SYSPROTO_H_ struct fhlinkat_args { fhandle_t *fhp; int tofd; const char *to; }; #endif int sys_fhlinkat(struct thread *td, struct fhlinkat_args *uap) { return (kern_fhlinkat(td, uap->tofd, uap->to, UIO_USERSPACE, uap->fhp)); } static int kern_fhlinkat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, fhandle_t *fhp) { fhandle_t fh; struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); error = copyin(fhp, &fh, sizeof(fh)); if (error != 0) return (error); do { bwillwrite(); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &vp); vfs_unbusy(mp); if (error != 0) return (error); VOP_UNLOCK(vp); error = kern_linkat_vp(td, vp, fd, path, pathseg); } while (error == EAGAIN || error == ERELOOKUP); return (error); } #ifndef _SYS_SYSPROTO_H_ struct fhreadlink_args { fhandle_t *fhp; char *buf; size_t bufsize; }; #endif int sys_fhreadlink(struct thread *td, struct fhreadlink_args *uap) { fhandle_t fh; struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); if (uap->bufsize > IOSIZE_MAX) return (EINVAL); error = copyin(uap->fhp, &fh, sizeof(fh)); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = kern_readlink_vp(vp, uap->buf, UIO_USERSPACE, uap->bufsize, td); vput(vp); return (error); } /* * syscall for the rpc.lockd to use to translate a NFS file handle into an * open descriptor. * * warning: do not remove the priv_check() call or this becomes one giant * security hole. */ #ifndef _SYS_SYSPROTO_H_ struct fhopen_args { const struct fhandle *u_fhp; int flags; }; #endif int sys_fhopen(struct thread *td, struct fhopen_args *uap) { return (kern_fhopen(td, uap->u_fhp, uap->flags)); } int kern_fhopen(struct thread *td, const struct fhandle *u_fhp, int flags) { struct mount *mp; struct vnode *vp; struct fhandle fhp; struct file *fp; int fmode, error; int indx; error = priv_check(td, PRIV_VFS_FHOPEN); if (error != 0) return (error); indx = -1; fmode = FFLAGS(flags); /* why not allow a non-read/write open for our lockd? */ if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT)) return (EINVAL); error = copyin(u_fhp, &fhp, sizeof(fhp)); if (error != 0) return(error); /* find the mount point */ mp = vfs_busyfs(&fhp.fh_fsid); if (mp == NULL) return (ESTALE); /* now give me my vnode, it gets returned to me locked */ error = VFS_FHTOVP(mp, &fhp.fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = falloc_noinstall(td, &fp); if (error != 0) { vput(vp); return (error); } /* * An extra reference on `fp' has been held for us by * falloc_noinstall(). */ #ifdef INVARIANTS td->td_dupfd = -1; #endif error = vn_open_vnode(vp, fmode, td->td_ucred, td, fp); if (error != 0) { KASSERT(fp->f_ops == &badfileops, ("VOP_OPEN in fhopen() set f_ops")); KASSERT(td->td_dupfd < 0, ("fhopen() encountered fdopen()")); vput(vp); goto bad; } #ifdef INVARIANTS td->td_dupfd = 0; #endif fp->f_vnode = vp; finit_vnode(fp, fmode, NULL, &vnops); VOP_UNLOCK(vp); if ((fmode & O_TRUNC) != 0) { error = fo_truncate(fp, 0, td->td_ucred, td); if (error != 0) goto bad; } error = finstall(td, fp, &indx, fmode, NULL); bad: fdrop(fp, td); td->td_retval[0] = indx; return (error); } /* * Stat an (NFS) file handle. */ #ifndef _SYS_SYSPROTO_H_ struct fhstat_args { struct fhandle *u_fhp; struct stat *sb; }; #endif int sys_fhstat(struct thread *td, struct fhstat_args *uap) { struct stat sb; struct fhandle fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fh)); if (error != 0) return (error); error = kern_fhstat(td, fh, &sb); if (error == 0) error = copyout(&sb, uap->sb, sizeof(sb)); return (error); } int kern_fhstat(struct thread *td, struct fhandle fh, struct stat *sb) { struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_FHSTAT); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = VOP_STAT(vp, sb, td->td_ucred, NOCRED); vput(vp); return (error); } /* * Implement fstatfs() for (NFS) file handles. */ #ifndef _SYS_SYSPROTO_H_ struct fhstatfs_args { struct fhandle *u_fhp; struct statfs *buf; }; #endif int sys_fhstatfs(struct thread *td, struct fhstatfs_args *uap) { struct statfs *sfp; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fhstatfs(td, fh, sfp); if (error == 0) error = copyout(sfp, uap->buf, sizeof(*sfp)); free(sfp, M_STATFS); return (error); } int kern_fhstatfs(struct thread *td, fhandle_t fh, struct statfs *buf) { struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_FHSTATFS); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp); if (error != 0) { vfs_unbusy(mp); return (error); } vput(vp); error = prison_canseemount(td->td_ucred, mp); if (error != 0) goto out; #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif error = VFS_STATFS(mp, buf); out: vfs_unbusy(mp); return (error); } /* * Unlike madvise(2), we do not make a best effort to remember every * possible caching hint. Instead, we remember the last setting with * the exception that we will allow POSIX_FADV_NORMAL to adjust the * region of any current setting. */ int kern_posix_fadvise(struct thread *td, int fd, off_t offset, off_t len, int advice) { struct fadvise_info *fa, *new; struct file *fp; struct vnode *vp; off_t end; int error; if (offset < 0 || len < 0 || offset > OFF_MAX - len) return (EINVAL); AUDIT_ARG_VALUE(advice); switch (advice) { case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_RANDOM: case POSIX_FADV_NOREUSE: new = malloc(sizeof(*fa), M_FADVISE, M_WAITOK); break; case POSIX_FADV_NORMAL: case POSIX_FADV_WILLNEED: case POSIX_FADV_DONTNEED: new = NULL; break; default: return (EINVAL); } /* XXX: CAP_POSIX_FADVISE? */ AUDIT_ARG_FD(fd); error = fget(td, fd, &cap_no_rights, &fp); if (error != 0) goto out; AUDIT_ARG_FILE(td->td_proc, fp); if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { error = ESPIPE; goto out; } if (fp->f_type != DTYPE_VNODE) { error = ENODEV; goto out; } vp = fp->f_vnode; if (vp->v_type != VREG) { error = ENODEV; goto out; } if (len == 0) end = OFF_MAX; else end = offset + len - 1; switch (advice) { case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_RANDOM: case POSIX_FADV_NOREUSE: /* * Try to merge any existing non-standard region with * this new region if possible, otherwise create a new * non-standard region for this request. */ mtx_pool_lock(mtxpool_sleep, fp); fa = fp->f_advice; if (fa != NULL && fa->fa_advice == advice && ((fa->fa_start <= end && fa->fa_end >= offset) || (end != OFF_MAX && fa->fa_start == end + 1) || (fa->fa_end != OFF_MAX && fa->fa_end + 1 == offset))) { if (offset < fa->fa_start) fa->fa_start = offset; if (end > fa->fa_end) fa->fa_end = end; } else { new->fa_advice = advice; new->fa_start = offset; new->fa_end = end; fp->f_advice = new; new = fa; } mtx_pool_unlock(mtxpool_sleep, fp); break; case POSIX_FADV_NORMAL: /* * If a the "normal" region overlaps with an existing * non-standard region, trim or remove the * non-standard region. */ mtx_pool_lock(mtxpool_sleep, fp); fa = fp->f_advice; if (fa != NULL) { if (offset <= fa->fa_start && end >= fa->fa_end) { new = fa; fp->f_advice = NULL; } else if (offset <= fa->fa_start && end >= fa->fa_start) fa->fa_start = end + 1; else if (offset <= fa->fa_end && end >= fa->fa_end) fa->fa_end = offset - 1; else if (offset >= fa->fa_start && end <= fa->fa_end) { /* * If the "normal" region is a middle * portion of the existing * non-standard region, just remove * the whole thing rather than picking * one side or the other to * preserve. */ new = fa; fp->f_advice = NULL; } } mtx_pool_unlock(mtxpool_sleep, fp); break; case POSIX_FADV_WILLNEED: case POSIX_FADV_DONTNEED: error = VOP_ADVISE(vp, offset, end, advice); break; } out: if (fp != NULL) fdrop(fp, td); free(new, M_FADVISE); return (error); } int sys_posix_fadvise(struct thread *td, struct posix_fadvise_args *uap) { int error; error = kern_posix_fadvise(td, uap->fd, uap->offset, uap->len, uap->advice); return (kern_posix_error(td, error)); } int kern_copy_file_range(struct thread *td, int infd, off_t *inoffp, int outfd, off_t *outoffp, size_t len, unsigned int flags) { struct file *infp, *outfp; struct vnode *invp, *outvp; int error; size_t retlen; void *rl_rcookie, *rl_wcookie; off_t savinoff, savoutoff; infp = outfp = NULL; rl_rcookie = rl_wcookie = NULL; savinoff = -1; error = 0; retlen = 0; if (flags != 0) { error = EINVAL; goto out; } if (len > SSIZE_MAX) /* * Although the len argument is size_t, the return argument * is ssize_t (which is signed). Therefore a size that won't * fit in ssize_t can't be returned. */ len = SSIZE_MAX; /* Get the file structures for the file descriptors. */ error = fget_read(td, infd, &cap_read_rights, &infp); if (error != 0) goto out; if (infp->f_ops == &badfileops) { error = EBADF; goto out; } if (infp->f_vnode == NULL) { error = EINVAL; goto out; } error = fget_write(td, outfd, &cap_write_rights, &outfp); if (error != 0) goto out; if (outfp->f_ops == &badfileops) { error = EBADF; goto out; } if (outfp->f_vnode == NULL) { error = EINVAL; goto out; } /* Set the offset pointers to the correct place. */ if (inoffp == NULL) inoffp = &infp->f_offset; if (outoffp == NULL) outoffp = &outfp->f_offset; savinoff = *inoffp; savoutoff = *outoffp; invp = infp->f_vnode; outvp = outfp->f_vnode; /* Sanity check the f_flag bits. */ if ((outfp->f_flag & (FWRITE | FAPPEND)) != FWRITE || (infp->f_flag & FREAD) == 0) { error = EBADF; goto out; } /* If len == 0, just return 0. */ if (len == 0) goto out; /* * If infp and outfp refer to the same file, the byte ranges cannot * overlap. */ if (invp == outvp && ((savinoff <= savoutoff && savinoff + len > savoutoff) || (savinoff > savoutoff && savoutoff + len > savinoff))) { error = EINVAL; goto out; } /* Range lock the byte ranges for both invp and outvp. */ for (;;) { rl_wcookie = vn_rangelock_wlock(outvp, *outoffp, *outoffp + len); rl_rcookie = vn_rangelock_tryrlock(invp, *inoffp, *inoffp + len); if (rl_rcookie != NULL) break; vn_rangelock_unlock(outvp, rl_wcookie); rl_rcookie = vn_rangelock_rlock(invp, *inoffp, *inoffp + len); vn_rangelock_unlock(invp, rl_rcookie); } retlen = len; error = vn_copy_file_range(invp, inoffp, outvp, outoffp, &retlen, flags, infp->f_cred, outfp->f_cred, td); out: if (rl_rcookie != NULL) vn_rangelock_unlock(invp, rl_rcookie); if (rl_wcookie != NULL) vn_rangelock_unlock(outvp, rl_wcookie); if (savinoff != -1 && (error == EINTR || error == ERESTART)) { *inoffp = savinoff; *outoffp = savoutoff; } if (outfp != NULL) fdrop(outfp, td); if (infp != NULL) fdrop(infp, td); td->td_retval[0] = retlen; return (error); } int sys_copy_file_range(struct thread *td, struct copy_file_range_args *uap) { off_t inoff, outoff, *inoffp, *outoffp; int error; inoffp = outoffp = NULL; if (uap->inoffp != NULL) { error = copyin(uap->inoffp, &inoff, sizeof(off_t)); if (error != 0) return (error); inoffp = &inoff; } if (uap->outoffp != NULL) { error = copyin(uap->outoffp, &outoff, sizeof(off_t)); if (error != 0) return (error); outoffp = &outoff; } error = kern_copy_file_range(td, uap->infd, inoffp, uap->outfd, outoffp, uap->len, uap->flags); if (error == 0 && uap->inoffp != NULL) error = copyout(inoffp, uap->inoffp, sizeof(off_t)); if (error == 0 && uap->outoffp != NULL) error = copyout(outoffp, uap->outoffp, sizeof(off_t)); return (error); } diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c index a7ca547f571f..04e8cb0037a2 100644 --- a/sys/kern/vfs_vnops.c +++ b/sys/kern/vfs_vnops.c @@ -1,3735 +1,3741 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Copyright (c) 2012 Konstantin Belousov * Copyright (c) 2013, 2014 The FreeBSD Foundation * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef HWPMC_HOOKS #include #endif static fo_rdwr_t vn_read; static fo_rdwr_t vn_write; static fo_rdwr_t vn_io_fault; static fo_truncate_t vn_truncate; static fo_ioctl_t vn_ioctl; static fo_poll_t vn_poll; static fo_kqfilter_t vn_kqfilter; static fo_close_t vn_closefile; static fo_mmap_t vn_mmap; static fo_fallocate_t vn_fallocate; static fo_fspacectl_t vn_fspacectl; struct fileops vnops = { .fo_read = vn_io_fault, .fo_write = vn_io_fault, .fo_truncate = vn_truncate, .fo_ioctl = vn_ioctl, .fo_poll = vn_poll, .fo_kqfilter = vn_kqfilter, .fo_stat = vn_statfile, .fo_close = vn_closefile, .fo_chmod = vn_chmod, .fo_chown = vn_chown, .fo_sendfile = vn_sendfile, .fo_seek = vn_seek, .fo_fill_kinfo = vn_fill_kinfo, .fo_mmap = vn_mmap, .fo_fallocate = vn_fallocate, .fo_fspacectl = vn_fspacectl, .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE }; const u_int io_hold_cnt = 16; static int vn_io_fault_enable = 1; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RWTUN, &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); static int vn_io_fault_prefault = 0; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RWTUN, &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting"); static int vn_io_pgcache_read_enable = 1; SYSCTL_INT(_debug, OID_AUTO, vn_io_pgcache_read_enable, CTLFLAG_RWTUN, &vn_io_pgcache_read_enable, 0, "Enable copying from page cache for reads, avoiding fs"); static u_long vn_io_faults_cnt; SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); static int vfs_allow_read_dir = 0; SYSCTL_INT(_security_bsd, OID_AUTO, allow_read_dir, CTLFLAG_RW, &vfs_allow_read_dir, 0, "Enable read(2) of directory by root for filesystems that support it"); /* * Returns true if vn_io_fault mode of handling the i/o request should * be used. */ static bool do_vn_io_fault(struct vnode *vp, struct uio *uio) { struct mount *mp; return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && (mp = vp->v_mount) != NULL && (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); } /* * Structure used to pass arguments to vn_io_fault1(), to do either * file- or vnode-based I/O calls. */ struct vn_io_fault_args { enum { VN_IO_FAULT_FOP, VN_IO_FAULT_VOP } kind; struct ucred *cred; int flags; union { struct fop_args_tag { struct file *fp; fo_rdwr_t *doio; } fop_args; struct vop_args_tag { struct vnode *vp; } vop_args; } args; }; static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td); int vn_open(struct nameidata *ndp, int *flagp, int cmode, struct file *fp) { struct thread *td = curthread; return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); } static uint64_t open2nameif(int fmode, u_int vn_open_flags) { uint64_t res; res = ISOPEN | LOCKLEAF; if ((fmode & O_RESOLVE_BENEATH) != 0) res |= RBENEATH; if ((fmode & O_EMPTY_PATH) != 0) res |= EMPTYPATH; if ((fmode & FREAD) != 0) res |= OPENREAD; if ((fmode & FWRITE) != 0) res |= OPENWRITE; if ((vn_open_flags & VN_OPEN_NOAUDIT) == 0) res |= AUDITVNODE1; if ((vn_open_flags & VN_OPEN_NOCAPCHECK) != 0) res |= NOCAPCHECK; if ((vn_open_flags & VN_OPEN_WANTIOCTLCAPS) != 0) res |= WANTIOCTLCAPS; return (res); } /* * Common code for vnode open operations via a name lookup. * Lookup the vnode and invoke VOP_CREATE if needed. * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. * * Note that this does NOT free nameidata for the successful case, * due to the NDINIT being done elsewhere. */ int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp) { struct vnode *vp; struct mount *mp; struct vattr vat; struct vattr *vap = &vat; int fmode, error; bool first_open; restart: first_open = false; fmode = *flagp; if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT | O_EXCL | O_DIRECTORY) || (fmode & (O_CREAT | O_EMPTY_PATH)) == (O_CREAT | O_EMPTY_PATH)) return (EINVAL); else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) { ndp->ni_cnd.cn_nameiop = CREATE; ndp->ni_cnd.cn_flags = open2nameif(fmode, vn_open_flags); /* * Set NOCACHE to avoid flushing the cache when * rolling in many files at once. * * Set NC_KEEPPOSENTRY to keep positive entries if they already * exist despite NOCACHE. */ ndp->ni_cnd.cn_flags |= LOCKPARENT | NOCACHE | NC_KEEPPOSENTRY; if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) ndp->ni_cnd.cn_flags |= FOLLOW; if ((vn_open_flags & VN_OPEN_INVFS) == 0) bwillwrite(); if ((error = namei(ndp)) != 0) return (error); if (ndp->ni_vp == NULL) { VATTR_NULL(vap); vap->va_type = VREG; vap->va_mode = cmode; if (fmode & O_EXCL) vap->va_vaflags |= VA_EXCLUSIVE; if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE_PNBUF(ndp); vput(ndp->ni_dvp); if ((error = vn_start_write(NULL, &mp, - V_XSLEEP | PCATCH)) != 0) + V_XSLEEP | V_PCATCH)) != 0) return (error); NDREINIT(ndp); goto restart; } if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) ndp->ni_cnd.cn_flags |= MAKEENTRY; #ifdef MAC error = mac_vnode_check_create(cred, ndp->ni_dvp, &ndp->ni_cnd, vap); if (error == 0) #endif error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, vap); vp = ndp->ni_vp; if (error == 0 && (fmode & O_EXCL) != 0 && (fmode & (O_EXLOCK | O_SHLOCK)) != 0) { VI_LOCK(vp); vp->v_iflag |= VI_FOPENING; VI_UNLOCK(vp); first_open = true; } VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &vp : NULL, false); vn_finished_write(mp); if (error) { NDFREE_PNBUF(ndp); if (error == ERELOOKUP) { NDREINIT(ndp); goto restart; } return (error); } fmode &= ~O_TRUNC; } else { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); ndp->ni_dvp = NULL; vp = ndp->ni_vp; if (fmode & O_EXCL) { error = EEXIST; goto bad; } if (vp->v_type == VDIR) { error = EISDIR; goto bad; } fmode &= ~O_CREAT; } } else { ndp->ni_cnd.cn_nameiop = LOOKUP; ndp->ni_cnd.cn_flags = open2nameif(fmode, vn_open_flags); ndp->ni_cnd.cn_flags |= (fmode & O_NOFOLLOW) != 0 ? NOFOLLOW : FOLLOW; if ((fmode & FWRITE) == 0) ndp->ni_cnd.cn_flags |= LOCKSHARED; if ((error = namei(ndp)) != 0) return (error); vp = ndp->ni_vp; } error = vn_open_vnode(vp, fmode, cred, curthread, fp); if (first_open) { VI_LOCK(vp); vp->v_iflag &= ~VI_FOPENING; wakeup(vp); VI_UNLOCK(vp); } if (error) goto bad; *flagp = fmode; return (0); bad: NDFREE_PNBUF(ndp); vput(vp); *flagp = fmode; ndp->ni_vp = NULL; return (error); } static int vn_open_vnode_advlock(struct vnode *vp, int fmode, struct file *fp) { struct flock lf; int error, lock_flags, type; ASSERT_VOP_LOCKED(vp, "vn_open_vnode_advlock"); if ((fmode & (O_EXLOCK | O_SHLOCK)) == 0) return (0); KASSERT(fp != NULL, ("open with flock requires fp")); if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) return (EOPNOTSUPP); lock_flags = VOP_ISLOCKED(vp); VOP_UNLOCK(vp); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = (fmode & O_EXLOCK) != 0 ? F_WRLCK : F_RDLCK; type = F_FLOCK; if ((fmode & FNONBLOCK) == 0) type |= F_WAIT; if ((fmode & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) type |= F_FIRSTOPEN; error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); if (error == 0) fp->f_flag |= FHASLOCK; vn_lock(vp, lock_flags | LK_RETRY); return (error); } /* * Common code for vnode open operations once a vnode is located. * Check permissions, and call the VOP_OPEN routine. */ int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp) { accmode_t accmode; int error; if (vp->v_type == VLNK) { if ((fmode & O_PATH) == 0 || (fmode & FEXEC) != 0) return (EMLINK); } if (vp->v_type != VDIR && fmode & O_DIRECTORY) return (ENOTDIR); accmode = 0; if ((fmode & O_PATH) == 0) { if (vp->v_type == VSOCK) return (EOPNOTSUPP); if ((fmode & (FWRITE | O_TRUNC)) != 0) { if (vp->v_type == VDIR) return (EISDIR); accmode |= VWRITE; } if ((fmode & FREAD) != 0) accmode |= VREAD; if ((fmode & O_APPEND) && (fmode & FWRITE)) accmode |= VAPPEND; #ifdef MAC if ((fmode & O_CREAT) != 0) accmode |= VCREAT; #endif } if ((fmode & FEXEC) != 0) accmode |= VEXEC; #ifdef MAC if ((fmode & O_VERIFY) != 0) accmode |= VVERIFY; error = mac_vnode_check_open(cred, vp, accmode); if (error != 0) return (error); accmode &= ~(VCREAT | VVERIFY); #endif if ((fmode & O_CREAT) == 0 && accmode != 0) { error = VOP_ACCESS(vp, accmode, cred, td); if (error != 0) return (error); } if ((fmode & O_PATH) != 0) { if (vp->v_type != VFIFO && vp->v_type != VSOCK && VOP_ACCESS(vp, VREAD, cred, td) == 0) fp->f_flag |= FKQALLOWED; return (0); } if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vn_lock(vp, LK_UPGRADE | LK_RETRY); error = VOP_OPEN(vp, fmode, cred, td, fp); if (error != 0) return (error); error = vn_open_vnode_advlock(vp, fmode, fp); if (error == 0 && (fmode & FWRITE) != 0) { error = VOP_ADD_WRITECOUNT(vp, 1); if (error == 0) { CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); } } /* * Error from advlock or VOP_ADD_WRITECOUNT() still requires * calling VOP_CLOSE() to pair with earlier VOP_OPEN(). */ if (error != 0) { if (fp != NULL) { /* * Arrange the call by having fdrop() to use * vn_closefile(). This is to satisfy * filesystems like devfs or tmpfs, which * override fo_close(). */ fp->f_flag |= FOPENFAILED; fp->f_vnode = vp; if (fp->f_ops == &badfileops) { fp->f_type = DTYPE_VNODE; fp->f_ops = &vnops; } vref(vp); } else { /* * If there is no fp, due to kernel-mode open, * we can call VOP_CLOSE() now. */ if (vp->v_type != VFIFO && (fmode & FWRITE) != 0 && !MNT_EXTENDED_SHARED(vp->v_mount) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vn_lock(vp, LK_UPGRADE | LK_RETRY); (void)VOP_CLOSE(vp, fmode & (FREAD | FWRITE | FEXEC), cred, td); } } ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); return (error); } /* * Check for write permissions on the specified vnode. * Prototype text segments cannot be written. * It is racy. */ int vn_writechk(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, "vn_writechk"); /* * If there's shared text associated with * the vnode, try to free it up once. If * we fail, we can't allow writing. */ if (VOP_IS_TEXT(vp)) return (ETXTBSY); return (0); } /* * Vnode close call */ static int vn_close1(struct vnode *vp, int flags, struct ucred *file_cred, struct thread *td, bool keep_ref) { struct mount *mp; int error, lock_flags; if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && MNT_EXTENDED_SHARED(vp->v_mount)) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, lock_flags | LK_RETRY); AUDIT_ARG_VNODE1(vp); if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) { VOP_ADD_WRITECOUNT_CHECKED(vp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, vp, vp->v_writecount); } error = VOP_CLOSE(vp, flags, file_cred, td); if (keep_ref) VOP_UNLOCK(vp); else vput(vp); vn_finished_write(mp); return (error); } int vn_close(struct vnode *vp, int flags, struct ucred *file_cred, struct thread *td) { return (vn_close1(vp, flags, file_cred, td, false)); } /* * Heuristic to detect sequential operation. */ static int sequential_heuristic(struct uio *uio, struct file *fp) { enum uio_rw rw; ASSERT_VOP_LOCKED(fp->f_vnode, __func__); rw = uio->uio_rw; if (fp->f_flag & FRDAHEAD) return (fp->f_seqcount[rw] << IO_SEQSHIFT); /* * Offset 0 is handled specially. open() sets f_seqcount to 1 so * that the first I/O is normally considered to be slightly * sequential. Seeking to offset 0 doesn't change sequentiality * unless previous seeks have reduced f_seqcount to 0, in which * case offset 0 is not special. */ if ((uio->uio_offset == 0 && fp->f_seqcount[rw] > 0) || uio->uio_offset == fp->f_nextoff[rw]) { /* * f_seqcount is in units of fixed-size blocks so that it * depends mainly on the amount of sequential I/O and not * much on the number of sequential I/O's. The fixed size * of 16384 is hard-coded here since it is (not quite) just * a magic size that works well here. This size is more * closely related to the best I/O size for real disks than * to any block size used by software. */ if (uio->uio_resid >= IO_SEQMAX * 16384) fp->f_seqcount[rw] = IO_SEQMAX; else { fp->f_seqcount[rw] += howmany(uio->uio_resid, 16384); if (fp->f_seqcount[rw] > IO_SEQMAX) fp->f_seqcount[rw] = IO_SEQMAX; } return (fp->f_seqcount[rw] << IO_SEQSHIFT); } /* Not sequential. Quickly draw-down sequentiality. */ if (fp->f_seqcount[rw] > 1) fp->f_seqcount[rw] = 1; else fp->f_seqcount[rw] = 0; return (0); } /* * Package up an I/O request on a vnode into a uio and do it. */ int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td) { struct uio auio; struct iovec aiov; struct mount *mp; struct ucred *cred; void *rl_cookie; struct vn_io_fault_args args; int error, lock_flags; if (offset < 0 && vp->v_type != VCHR) return (EINVAL); auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = base; aiov.iov_len = len; auio.uio_resid = len; auio.uio_offset = offset; auio.uio_segflg = segflg; auio.uio_rw = rw; auio.uio_td = td; error = 0; if ((ioflg & IO_NODELOCKED) == 0) { if ((ioflg & IO_RANGELOCKED) == 0) { if (rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, offset, offset + len); } else if ((ioflg & IO_APPEND) != 0) { rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); } else { rl_cookie = vn_rangelock_wlock(vp, offset, offset + len); } } else rl_cookie = NULL; mp = NULL; if (rw == UIO_WRITE) { if (vp->v_type != VCHR && - (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) + (error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH)) != 0) goto out; lock_flags = vn_lktype_write(mp, vp); } else lock_flags = LK_SHARED; vn_lock(vp, lock_flags | LK_RETRY); } else rl_cookie = NULL; ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); #ifdef MAC if ((ioflg & IO_NOMACCHECK) == 0) { if (rw == UIO_READ) error = mac_vnode_check_read(active_cred, file_cred, vp); else error = mac_vnode_check_write(active_cred, file_cred, vp); } #endif if (error == 0) { if (file_cred != NULL) cred = file_cred; else cred = active_cred; if (do_vn_io_fault(vp, &auio)) { args.kind = VN_IO_FAULT_VOP; args.cred = cred; args.flags = ioflg; args.args.vop_args.vp = vp; error = vn_io_fault1(vp, &auio, &args, td); } else if (rw == UIO_READ) { error = VOP_READ(vp, &auio, ioflg, cred); } else /* if (rw == UIO_WRITE) */ { error = VOP_WRITE(vp, &auio, ioflg, cred); } } if (aresid) *aresid = auio.uio_resid; else if (auio.uio_resid && error == 0) error = EIO; if ((ioflg & IO_NODELOCKED) == 0) { VOP_UNLOCK(vp); if (mp != NULL) vn_finished_write(mp); } out: if (rl_cookie != NULL) vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * Package up an I/O request on a vnode into a uio and do it. The I/O * request is split up into smaller chunks and we try to avoid saturating * the buffer cache while potentially holding a vnode locked, so we * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() * to give other processes a chance to lock the vnode (either other processes * core'ing the same binary, or unrelated processes scanning the directory). */ int vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, size_t *aresid, struct thread *td) { int error = 0; ssize_t iaresid; do { int chunk; /* * Force `offset' to a multiple of MAXBSIZE except possibly * for the first chunk, so that filesystems only need to * write full blocks except possibly for the first and last * chunks. */ chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; if (chunk > len) chunk = len; if (rw != UIO_READ && vp->v_type == VREG) bwillwrite(); iaresid = 0; error = vn_rdwr(rw, vp, base, chunk, offset, segflg, ioflg, active_cred, file_cred, &iaresid, td); len -= chunk; /* aresid calc already includes length */ if (error) break; offset += chunk; base = (char *)base + chunk; kern_yield(PRI_USER); } while (len); if (aresid) *aresid = len + iaresid; return (error); } #if OFF_MAX <= LONG_MAX off_t foffset_lock(struct file *fp, int flags) { volatile short *flagsp; off_t res; short state; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); if ((flags & FOF_NOLOCK) != 0) return (atomic_load_long(&fp->f_offset)); /* * According to McKusick the vn lock was protecting f_offset here. * It is now protected by the FOFFSET_LOCKED flag. */ flagsp = &fp->f_vnread_flags; if (atomic_cmpset_acq_16(flagsp, 0, FOFFSET_LOCKED)) return (atomic_load_long(&fp->f_offset)); sleepq_lock(&fp->f_vnread_flags); state = atomic_load_16(flagsp); for (;;) { if ((state & FOFFSET_LOCKED) == 0) { if (!atomic_fcmpset_acq_16(flagsp, &state, FOFFSET_LOCKED)) continue; break; } if ((state & FOFFSET_LOCK_WAITING) == 0) { if (!atomic_fcmpset_acq_16(flagsp, &state, state | FOFFSET_LOCK_WAITING)) continue; } DROP_GIANT(); sleepq_add(&fp->f_vnread_flags, NULL, "vofflock", 0, 0); sleepq_wait(&fp->f_vnread_flags, PUSER -1); PICKUP_GIANT(); sleepq_lock(&fp->f_vnread_flags); state = atomic_load_16(flagsp); } res = atomic_load_long(&fp->f_offset); sleepq_release(&fp->f_vnread_flags); return (res); } void foffset_unlock(struct file *fp, off_t val, int flags) { volatile short *flagsp; short state; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); if ((flags & FOF_NOUPDATE) == 0) atomic_store_long(&fp->f_offset, val); if ((flags & FOF_NEXTOFF_R) != 0) fp->f_nextoff[UIO_READ] = val; if ((flags & FOF_NEXTOFF_W) != 0) fp->f_nextoff[UIO_WRITE] = val; if ((flags & FOF_NOLOCK) != 0) return; flagsp = &fp->f_vnread_flags; state = atomic_load_16(flagsp); if ((state & FOFFSET_LOCK_WAITING) == 0 && atomic_cmpset_rel_16(flagsp, state, 0)) return; sleepq_lock(&fp->f_vnread_flags); MPASS((fp->f_vnread_flags & FOFFSET_LOCKED) != 0); MPASS((fp->f_vnread_flags & FOFFSET_LOCK_WAITING) != 0); fp->f_vnread_flags = 0; sleepq_broadcast(&fp->f_vnread_flags, SLEEPQ_SLEEP, 0, 0); sleepq_release(&fp->f_vnread_flags); } #else off_t foffset_lock(struct file *fp, int flags) { struct mtx *mtxp; off_t res; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOLOCK) == 0) { while (fp->f_vnread_flags & FOFFSET_LOCKED) { fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; msleep(&fp->f_vnread_flags, mtxp, PUSER -1, "vofflock", 0); } fp->f_vnread_flags |= FOFFSET_LOCKED; } res = fp->f_offset; mtx_unlock(mtxp); return (res); } void foffset_unlock(struct file *fp, off_t val, int flags) { struct mtx *mtxp; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF_R) != 0) fp->f_nextoff[UIO_READ] = val; if ((flags & FOF_NEXTOFF_W) != 0) fp->f_nextoff[UIO_WRITE] = val; if ((flags & FOF_NOLOCK) == 0) { KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, ("Lost FOFFSET_LOCKED")); if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) wakeup(&fp->f_vnread_flags); fp->f_vnread_flags = 0; } mtx_unlock(mtxp); } #endif void foffset_lock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) uio->uio_offset = foffset_lock(fp, flags); } void foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) foffset_unlock(fp, uio->uio_offset, flags); } static int get_advice(struct file *fp, struct uio *uio) { struct mtx *mtxp; int ret; ret = POSIX_FADV_NORMAL; if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG) return (ret); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (fp->f_advice != NULL && uio->uio_offset >= fp->f_advice->fa_start && uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) ret = fp->f_advice->fa_advice; mtx_unlock(mtxp); return (ret); } static int get_write_ioflag(struct file *fp) { int ioflag; struct mount *mp; struct vnode *vp; ioflag = 0; vp = fp->f_vnode; mp = atomic_load_ptr(&vp->v_mount); if ((fp->f_flag & O_DIRECT) != 0) ioflag |= IO_DIRECT; if ((fp->f_flag & O_FSYNC) != 0 || (mp != NULL && (mp->mnt_flag & MNT_SYNCHRONOUS) != 0)) ioflag |= IO_SYNC; /* * For O_DSYNC we set both IO_SYNC and IO_DATASYNC, so that VOP_WRITE() * or VOP_DEALLOCATE() implementations that don't understand IO_DATASYNC * fall back to full O_SYNC behavior. */ if ((fp->f_flag & O_DSYNC) != 0) ioflag |= IO_SYNC | IO_DATASYNC; return (ioflag); } int vn_read_from_obj(struct vnode *vp, struct uio *uio) { vm_object_t obj; vm_page_t ma[io_hold_cnt + 2]; off_t off, vsz; ssize_t resid; int error, i, j; MPASS(uio->uio_resid <= ptoa(io_hold_cnt + 2)); obj = atomic_load_ptr(&vp->v_object); if (obj == NULL) return (EJUSTRETURN); /* * Depends on type stability of vm_objects. */ vm_object_pip_add(obj, 1); if ((obj->flags & OBJ_DEAD) != 0) { /* * Note that object might be already reused from the * vnode, and the OBJ_DEAD flag cleared. This is fine, * we recheck for DOOMED vnode state after all pages * are busied, and retract then. * * But we check for OBJ_DEAD to ensure that we do not * busy pages while vm_object_terminate_pages() * processes the queue. */ error = EJUSTRETURN; goto out_pip; } resid = uio->uio_resid; off = uio->uio_offset; for (i = 0; resid > 0; i++) { MPASS(i < io_hold_cnt + 2); ma[i] = vm_page_grab_unlocked(obj, atop(off), VM_ALLOC_NOCREAT | VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY | VM_ALLOC_NOWAIT); if (ma[i] == NULL) break; /* * Skip invalid pages. Valid mask can be partial only * at EOF, and we clip later. */ if (vm_page_none_valid(ma[i])) { vm_page_sunbusy(ma[i]); break; } resid -= PAGE_SIZE; off += PAGE_SIZE; } if (i == 0) { error = EJUSTRETURN; goto out_pip; } /* * Check VIRF_DOOMED after we busied our pages. Since * vgonel() terminates the vnode' vm_object, it cannot * process past pages busied by us. */ if (VN_IS_DOOMED(vp)) { error = EJUSTRETURN; goto out; } resid = PAGE_SIZE - (uio->uio_offset & PAGE_MASK) + ptoa(i - 1); if (resid > uio->uio_resid) resid = uio->uio_resid; /* * Unlocked read of vnp_size is safe because truncation cannot * pass busied page. But we load vnp_size into a local * variable so that possible concurrent extension does not * break calculation. */ #if defined(__powerpc__) && !defined(__powerpc64__) vsz = obj->un_pager.vnp.vnp_size; #else vsz = atomic_load_64(&obj->un_pager.vnp.vnp_size); #endif if (uio->uio_offset >= vsz) { error = EJUSTRETURN; goto out; } if (uio->uio_offset + resid > vsz) resid = vsz - uio->uio_offset; error = vn_io_fault_pgmove(ma, uio->uio_offset & PAGE_MASK, resid, uio); out: for (j = 0; j < i; j++) { if (error == 0) vm_page_reference(ma[j]); vm_page_sunbusy(ma[j]); } out_pip: vm_object_pip_wakeup(obj); if (error != 0) return (error); return (uio->uio_resid == 0 ? 0 : EJUSTRETURN); } /* * File table vnode read routine. */ static int vn_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct vnode *vp; off_t orig_offset; int error, ioflag; int advice; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; ioflag = 0; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; /* * Try to read from page cache. VIRF_DOOMED check is racy but * allows us to avoid unneeded work outright. */ if (vn_io_pgcache_read_enable && !mac_vnode_check_read_enabled() && (vn_irflag_read(vp) & (VIRF_DOOMED | VIRF_PGREAD)) == VIRF_PGREAD) { error = VOP_READ_PGCACHE(vp, uio, ioflag, fp->f_cred); if (error == 0) { fp->f_nextoff[UIO_READ] = uio->uio_offset; return (0); } if (error != EJUSTRETURN) return (error); } advice = get_advice(fp, uio); vn_lock(vp, LK_SHARED | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* Disable read-ahead for random I/O. */ break; } orig_offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_read(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_READ(vp, uio, ioflag, fp->f_cred); fp->f_nextoff[UIO_READ] = uio->uio_offset; VOP_UNLOCK(vp); if (error == 0 && advice == POSIX_FADV_NOREUSE && orig_offset != uio->uio_offset) /* * Use POSIX_FADV_DONTNEED to flush pages and buffers * for the backing file after a POSIX_FADV_NOREUSE * read(2). */ error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, POSIX_FADV_DONTNEED); return (error); } /* * File table vnode write routine. */ static int vn_write(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct vnode *vp; struct mount *mp; off_t orig_offset; int error, ioflag; int advice; bool need_finished_write; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; if (vp->v_type == VREG) bwillwrite(); ioflag = IO_UNIT; if (vp->v_type == VREG && (fp->f_flag & O_APPEND) != 0) ioflag |= IO_APPEND; if ((fp->f_flag & FNONBLOCK) != 0) ioflag |= IO_NDELAY; ioflag |= get_write_ioflag(fp); mp = NULL; need_finished_write = false; if (vp->v_type != VCHR) { - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error != 0) goto unlock; need_finished_write = true; } advice = get_advice(fp, uio); vn_lock(vp, vn_lktype_write(mp, vp) | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* XXX: Is this correct? */ break; } orig_offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); fp->f_nextoff[UIO_WRITE] = uio->uio_offset; VOP_UNLOCK(vp); if (need_finished_write) vn_finished_write(mp); if (error == 0 && advice == POSIX_FADV_NOREUSE && orig_offset != uio->uio_offset) /* * Use POSIX_FADV_DONTNEED to flush pages and buffers * for the backing file after a POSIX_FADV_NOREUSE * write(2). */ error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, POSIX_FADV_DONTNEED); unlock: return (error); } /* * The vn_io_fault() is a wrapper around vn_read() and vn_write() to * prevent the following deadlock: * * Assume that the thread A reads from the vnode vp1 into userspace * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is * currently not resident, then system ends up with the call chain * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) * which establishes lock order vp1->vn_lock, then vp2->vn_lock. * If, at the same time, thread B reads from vnode vp2 into buffer buf2 * backed by the pages of vnode vp1, and some page in buf2 is not * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. * * To prevent the lock order reversal and deadlock, vn_io_fault() does * not allow page faults to happen during VOP_READ() or VOP_WRITE(). * Instead, it first tries to do the whole range i/o with pagefaults * disabled. If all pages in the i/o buffer are resident and mapped, * VOP will succeed (ignoring the genuine filesystem errors). * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do * i/o in chunks, with all pages in the chunk prefaulted and held * using vm_fault_quick_hold_pages(). * * Filesystems using this deadlock avoidance scheme should use the * array of the held pages from uio, saved in the curthread->td_ma, * instead of doing uiomove(). A helper function * vn_io_fault_uiomove() converts uiomove request into * uiomove_fromphys() over td_ma array. * * Since vnode locks do not cover the whole i/o anymore, rangelocks * make the current i/o request atomic with respect to other i/os and * truncations. */ /* * Decode vn_io_fault_args and perform the corresponding i/o. */ static int vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, struct thread *td) { int error, save; error = 0; save = vm_fault_disable_pagefaults(); switch (args->kind) { case VN_IO_FAULT_FOP: error = (args->args.fop_args.doio)(args->args.fop_args.fp, uio, args->cred, args->flags, td); break; case VN_IO_FAULT_VOP: if (uio->uio_rw == UIO_READ) { error = VOP_READ(args->args.vop_args.vp, uio, args->flags, args->cred); } else if (uio->uio_rw == UIO_WRITE) { error = VOP_WRITE(args->args.vop_args.vp, uio, args->flags, args->cred); } break; default: panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind, uio->uio_rw); } vm_fault_enable_pagefaults(save); return (error); } static int vn_io_fault_touch(char *base, const struct uio *uio) { int r; r = fubyte(base); if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1)) return (EFAULT); return (0); } static int vn_io_fault_prefault_user(const struct uio *uio) { char *base; const struct iovec *iov; size_t len; ssize_t resid; int error, i; KASSERT(uio->uio_segflg == UIO_USERSPACE, ("vn_io_fault_prefault userspace")); error = i = 0; iov = uio->uio_iov; resid = uio->uio_resid; base = iov->iov_base; len = iov->iov_len; while (resid > 0) { error = vn_io_fault_touch(base, uio); if (error != 0) break; if (len < PAGE_SIZE) { if (len != 0) { error = vn_io_fault_touch(base + len - 1, uio); if (error != 0) break; resid -= len; } if (++i >= uio->uio_iovcnt) break; iov = uio->uio_iov + i; base = iov->iov_base; len = iov->iov_len; } else { len -= PAGE_SIZE; base += PAGE_SIZE; resid -= PAGE_SIZE; } } return (error); } /* * Common code for vn_io_fault(), agnostic to the kind of i/o request. * Uses vn_io_fault_doio() to make the call to an actual i/o function. * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request * into args and call vn_io_fault1() to handle faults during the user * mode buffer accesses. */ static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td) { vm_page_t ma[io_hold_cnt + 2]; struct uio *uio_clone, short_uio; struct iovec short_iovec[1]; vm_page_t *prev_td_ma; vm_prot_t prot; vm_offset_t addr, end; size_t len, resid; ssize_t adv; int error, cnt, saveheld, prev_td_ma_cnt; if (vn_io_fault_prefault) { error = vn_io_fault_prefault_user(uio); if (error != 0) return (error); /* Or ignore ? */ } prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; /* * The UFS follows IO_UNIT directive and replays back both * uio_offset and uio_resid if an error is encountered during the * operation. But, since the iovec may be already advanced, * uio is still in an inconsistent state. * * Cache a copy of the original uio, which is advanced to the redo * point using UIO_NOCOPY below. */ uio_clone = cloneuio(uio); resid = uio->uio_resid; short_uio.uio_segflg = UIO_USERSPACE; short_uio.uio_rw = uio->uio_rw; short_uio.uio_td = uio->uio_td; error = vn_io_fault_doio(args, uio, td); if (error != EFAULT) goto out; atomic_add_long(&vn_io_faults_cnt, 1); uio_clone->uio_segflg = UIO_NOCOPY; uiomove(NULL, resid - uio->uio_resid, uio_clone); uio_clone->uio_segflg = uio->uio_segflg; saveheld = curthread_pflags_set(TDP_UIOHELD); prev_td_ma = td->td_ma; prev_td_ma_cnt = td->td_ma_cnt; while (uio_clone->uio_resid != 0) { len = uio_clone->uio_iov->iov_len; if (len == 0) { KASSERT(uio_clone->uio_iovcnt >= 1, ("iovcnt underflow")); uio_clone->uio_iov++; uio_clone->uio_iovcnt--; continue; } if (len > ptoa(io_hold_cnt)) len = ptoa(io_hold_cnt); addr = (uintptr_t)uio_clone->uio_iov->iov_base; end = round_page(addr + len); if (end < addr) { error = EFAULT; break; } /* * A perfectly misaligned address and length could cause * both the start and the end of the chunk to use partial * page. +2 accounts for such a situation. */ cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, addr, len, prot, ma, io_hold_cnt + 2); if (cnt == -1) { error = EFAULT; break; } short_uio.uio_iov = &short_iovec[0]; short_iovec[0].iov_base = (void *)addr; short_uio.uio_iovcnt = 1; short_uio.uio_resid = short_iovec[0].iov_len = len; short_uio.uio_offset = uio_clone->uio_offset; td->td_ma = ma; td->td_ma_cnt = cnt; error = vn_io_fault_doio(args, &short_uio, td); vm_page_unhold_pages(ma, cnt); adv = len - short_uio.uio_resid; uio_clone->uio_iov->iov_base = (char *)uio_clone->uio_iov->iov_base + adv; uio_clone->uio_iov->iov_len -= adv; uio_clone->uio_resid -= adv; uio_clone->uio_offset += adv; uio->uio_resid -= adv; uio->uio_offset += adv; if (error != 0 || adv == 0) break; } td->td_ma = prev_td_ma; td->td_ma_cnt = prev_td_ma_cnt; curthread_pflags_restore(saveheld); out: free(uio_clone, M_IOV); return (error); } static int vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { fo_rdwr_t *doio; struct vnode *vp; void *rl_cookie; struct vn_io_fault_args args; int error; doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; vp = fp->f_vnode; /* * The ability to read(2) on a directory has historically been * allowed for all users, but this can and has been the source of * at least one security issue in the past. As such, it is now hidden * away behind a sysctl for those that actually need it to use it, and * restricted to root when it's turned on to make it relatively safe to * leave on for longer sessions of need. */ if (vp->v_type == VDIR) { KASSERT(uio->uio_rw == UIO_READ, ("illegal write attempted on a directory")); if (!vfs_allow_read_dir) return (EISDIR); if ((error = priv_check(td, PRIV_VFS_READ_DIR)) != 0) return (EISDIR); } foffset_lock_uio(fp, uio, flags); if (do_vn_io_fault(vp, uio)) { args.kind = VN_IO_FAULT_FOP; args.args.fop_args.fp = fp; args.args.fop_args.doio = doio; args.cred = active_cred; args.flags = flags | FOF_OFFSET; if (uio->uio_rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } else if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0) { /* For appenders, punt and lock the whole range. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); } else { rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } error = vn_io_fault1(vp, uio, &args, td); vn_rangelock_unlock(vp, rl_cookie); } else { error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); } foffset_unlock_uio(fp, uio, flags); return (error); } /* * Helper function to perform the requested uiomove operation using * the held pages for io->uio_iov[0].iov_base buffer instead of * copyin/copyout. Access to the pages with uiomove_fromphys() * instead of iov_base prevents page faults that could occur due to * pmap_collect() invalidating the mapping created by * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or * object cleanup revoking the write access from page mappings. * * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() * instead of plain uiomove(). */ int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) { struct uio transp_uio; struct iovec transp_iov[1]; struct thread *td; size_t adv; int error, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove(data, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); transp_iov[0].iov_base = data; transp_uio.uio_iov = &transp_iov[0]; transp_uio.uio_iovcnt = 1; if (xfersize > uio->uio_resid) xfersize = uio->uio_resid; transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; transp_uio.uio_offset = 0; transp_uio.uio_segflg = UIO_SYSSPACE; /* * Since transp_iov points to data, and td_ma page array * corresponds to original uio->uio_iov, we need to invert the * direction of the i/o operation as passed to * uiomove_fromphys(). */ switch (uio->uio_rw) { case UIO_WRITE: transp_uio.uio_rw = UIO_READ; break; case UIO_READ: transp_uio.uio_rw = UIO_WRITE; break; } transp_uio.uio_td = uio->uio_td; error = uiomove_fromphys(td->td_ma, ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, xfersize, &transp_uio); adv = xfersize - transp_uio.uio_resid; pgadv = (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; uio->uio_iov->iov_len -= adv; uio->uio_resid -= adv; uio->uio_offset += adv; return (error); } int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio) { struct thread *td; vm_offset_t iov_base; int cnt, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove_fromphys(ma, offset, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; iov_base = (vm_offset_t)uio->uio_iov->iov_base; switch (uio->uio_rw) { case UIO_WRITE: pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, offset, cnt); break; case UIO_READ: pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, cnt); break; } pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)(iov_base + cnt); uio->uio_iov->iov_len -= cnt; uio->uio_resid -= cnt; uio->uio_offset += cnt; return (0); } /* * File table truncate routine. */ static int vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { struct mount *mp; struct vnode *vp; void *rl_cookie; int error; vp = fp->f_vnode; retry: /* * Lock the whole range for truncation. Otherwise split i/o * might happen partly before and partly after the truncation. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error) goto out1; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(vp); if (vp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error) goto out; #endif error = vn_truncate_locked(vp, length, (fp->f_flag & O_FSYNC) != 0, fp->f_cred); out: VOP_UNLOCK(vp); vn_finished_write(mp); out1: vn_rangelock_unlock(vp, rl_cookie); if (error == ERELOOKUP) goto retry; return (error); } /* * Truncate a file that is already locked. */ int vn_truncate_locked(struct vnode *vp, off_t length, bool sync, struct ucred *cred) { struct vattr vattr; int error; error = VOP_ADD_WRITECOUNT(vp, 1); if (error == 0) { VATTR_NULL(&vattr); vattr.va_size = length; if (sync) vattr.va_vaflags |= VA_SYNC; error = VOP_SETATTR(vp, &vattr, cred); VOP_ADD_WRITECOUNT_CHECKED(vp, -1); } return (error); } /* * File table vnode stat routine. */ int vn_statfile(struct file *fp, struct stat *sb, struct ucred *active_cred) { struct vnode *vp = fp->f_vnode; int error; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_STAT(vp, sb, active_cred, fp->f_cred); VOP_UNLOCK(vp); return (error); } /* * File table vnode ioctl routine. */ static int vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { struct vattr vattr; struct vnode *vp; struct fiobmap2_arg *bmarg; int error; vp = fp->f_vnode; switch (vp->v_type) { case VDIR: case VREG: switch (com) { case FIONREAD: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, active_cred); VOP_UNLOCK(vp); if (error == 0) *(int *)data = vattr.va_size - fp->f_offset; return (error); case FIOBMAP2: bmarg = (struct fiobmap2_arg *)data; vn_lock(vp, LK_SHARED | LK_RETRY); #ifdef MAC error = mac_vnode_check_read(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_BMAP(vp, bmarg->bn, NULL, &bmarg->bn, &bmarg->runp, &bmarg->runb); VOP_UNLOCK(vp); return (error); case FIONBIO: case FIOASYNC: return (0); default: return (VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td)); } break; case VCHR: return (VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td)); default: return (ENOTTY); } } /* * File table vnode poll routine. */ static int vn_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { struct vnode *vp; int error; vp = fp->f_vnode; #if defined(MAC) || defined(AUDIT) if (AUDITING_TD(td) || mac_vnode_check_poll_enabled()) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(vp); error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); VOP_UNLOCK(vp); if (error != 0) return (error); } #endif error = VOP_POLL(vp, events, fp->f_cred, td); return (error); } /* * Acquire the requested lock and then check for validity. LK_RETRY * permits vn_lock to return doomed vnodes. */ static int __noinline _vn_lock_fallback(struct vnode *vp, int flags, const char *file, int line, int error) { KASSERT((flags & LK_RETRY) == 0 || error == 0, ("vn_lock: error %d incompatible with flags %#x", error, flags)); if (error == 0) VNASSERT(VN_IS_DOOMED(vp), vp, ("vnode not doomed")); if ((flags & LK_RETRY) == 0) { if (error == 0) { VOP_UNLOCK(vp); error = ENOENT; } return (error); } /* * LK_RETRY case. * * Nothing to do if we got the lock. */ if (error == 0) return (0); /* * Interlock was dropped by the call in _vn_lock. */ flags &= ~LK_INTERLOCK; do { error = VOP_LOCK1(vp, flags, file, line); } while (error != 0); return (0); } int _vn_lock(struct vnode *vp, int flags, const char *file, int line) { int error; VNASSERT((flags & LK_TYPE_MASK) != 0, vp, ("vn_lock: no locktype (%d passed)", flags)); VNPASS(vp->v_holdcnt > 0, vp); error = VOP_LOCK1(vp, flags, file, line); if (__predict_false(error != 0 || VN_IS_DOOMED(vp))) return (_vn_lock_fallback(vp, flags, file, line, error)); return (0); } /* * File table vnode close routine. */ static int vn_closefile(struct file *fp, struct thread *td) { struct vnode *vp; struct flock lf; int error; bool ref; vp = fp->f_vnode; fp->f_ops = &badfileops; ref = (fp->f_flag & FHASLOCK) != 0; error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref); if (__predict_false(ref)) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); vrele(vp); } return (error); } /* * Preparing to start a filesystem write operation. If the operation is * permitted, then we bump the count of operations in progress and * proceed. If a suspend request is in progress, we wait until the * suspension is over, and then proceed. */ static int vn_start_write_refed(struct mount *mp, int flags, bool mplocked) { struct mount_pcpu *mpcpu; int error, mflags; if (__predict_true(!mplocked) && (flags & V_XSLEEP) == 0 && vfs_op_thread_enter(mp, mpcpu)) { MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) == 0); vfs_mp_count_add_pcpu(mpcpu, writeopcount, 1); vfs_op_thread_exit(mp, mpcpu); return (0); } if (mplocked) mtx_assert(MNT_MTX(mp), MA_OWNED); else MNT_ILOCK(mp); error = 0; /* * Check on status of suspension. */ if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || mp->mnt_susp_owner != curthread) { mflags = 0; if ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0) { - if (flags & PCATCH) + if (flags & V_PCATCH) mflags |= PCATCH; } mflags |= (PUSER - 1); while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { if (flags & V_NOWAIT) { error = EWOULDBLOCK; goto unlock; } error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, "suspfs", 0); if (error) goto unlock; } } if (flags & V_XSLEEP) goto unlock; mp->mnt_writeopcount++; unlock: if (error != 0 || (flags & V_XSLEEP) != 0) MNT_REL(mp); MNT_IUNLOCK(mp); return (error); } int vn_start_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error; + KASSERT((flags & ~V_VALID_FLAGS) == 0, + ("%s: invalid flags passed %d\n", __func__, flags)); + error = 0; /* * If a vnode is provided, get and return the mount point that * to which it will write. */ if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } if ((mp = *mpp) == NULL) return (0); /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ if (vp == NULL) vfs_ref(mp); return (vn_start_write_refed(mp, flags, false)); } /* * Secondary suspension. Used by operations such as vop_inactive * routines that are needed by the higher level functions. These * are allowed to proceed until all the higher level functions have * completed (indicated by mnt_writeopcount dropping to zero). At that * time, these operations are halted until the suspension is over. */ int vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error, mflags; + KASSERT((flags & ~V_VALID_FLAGS) == 0, + ("%s: invalid flags passed %d\n", __func__, flags)); + retry: if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } /* * If we are not suspended or have not yet reached suspended * mode, then let the operation proceed. */ if ((mp = *mpp) == NULL) return (0); /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL) MNT_REF(mp); if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { mp->mnt_secondary_writes++; mp->mnt_secondary_accwrites++; MNT_IUNLOCK(mp); return (0); } if (flags & V_NOWAIT) { MNT_REL(mp); MNT_IUNLOCK(mp); return (EWOULDBLOCK); } /* * Wait for the suspension to finish. */ mflags = 0; if ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0) { - if (flags & PCATCH) + if (flags & V_PCATCH) mflags |= PCATCH; } mflags |= (PUSER - 1) | PDROP; error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, "suspfs", 0); vfs_rel(mp); if (error == 0) goto retry; return (error); } /* * Filesystem write operation has completed. If we are suspending and this * operation is the last one, notify the suspender that the suspension is * now in effect. */ void vn_finished_write(struct mount *mp) { struct mount_pcpu *mpcpu; int c; if (mp == NULL) return; if (vfs_op_thread_enter(mp, mpcpu)) { vfs_mp_count_sub_pcpu(mpcpu, writeopcount, 1); vfs_mp_count_sub_pcpu(mpcpu, ref, 1); vfs_op_thread_exit(mp, mpcpu); return; } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REL(mp); c = --mp->mnt_writeopcount; if (mp->mnt_vfs_ops == 0) { MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) == 0); MNT_IUNLOCK(mp); return; } if (c < 0) vfs_dump_mount_counters(mp); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && c == 0) wakeup(&mp->mnt_writeopcount); MNT_IUNLOCK(mp); } /* * Filesystem secondary write operation has completed. If we are * suspending and this operation is the last one, notify the suspender * that the suspension is now in effect. */ void vn_finished_secondary_write(struct mount *mp) { if (mp == NULL) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_secondary_writes--; if (mp->mnt_secondary_writes < 0) panic("vn_finished_secondary_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_secondary_writes <= 0) wakeup(&mp->mnt_secondary_writes); MNT_IUNLOCK(mp); } /* * Request a filesystem to suspend write operations. */ int vfs_write_suspend(struct mount *mp, int flags) { int error; vfs_op_enter(mp); MNT_ILOCK(mp); vfs_assert_mount_counters(mp); if (mp->mnt_susp_owner == curthread) { vfs_op_exit_locked(mp); MNT_IUNLOCK(mp); return (EALREADY); } while (mp->mnt_kern_flag & MNTK_SUSPEND) msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); /* * Unmount holds a write reference on the mount point. If we * own busy reference and drain for writers, we deadlock with * the reference draining in the unmount path. Callers of * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if * vfs_busy() reference is owned and caller is not in the * unmount context. */ if ((flags & VS_SKIP_UNMOUNT) != 0 && (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { vfs_op_exit_locked(mp); MNT_IUNLOCK(mp); return (EBUSY); } mp->mnt_kern_flag |= MNTK_SUSPEND; mp->mnt_susp_owner = curthread; if (mp->mnt_writeopcount > 0) (void) msleep(&mp->mnt_writeopcount, MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); else MNT_IUNLOCK(mp); if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) { vfs_write_resume(mp, 0); /* vfs_write_resume does vfs_op_exit() for us */ } return (error); } /* * Request a filesystem to resume write operations. */ void vfs_write_resume(struct mount *mp, int flags) { MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | MNTK_SUSPENDED); mp->mnt_susp_owner = NULL; wakeup(&mp->mnt_writeopcount); wakeup(&mp->mnt_flag); curthread->td_pflags &= ~TDP_IGNSUSP; if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); mp->mnt_writeopcount++; } MNT_IUNLOCK(mp); if ((flags & VR_NO_SUSPCLR) == 0) VFS_SUSP_CLEAN(mp); vfs_op_exit(mp); } else if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); vn_start_write_refed(mp, 0, true); } else { MNT_IUNLOCK(mp); } } /* * Helper loop around vfs_write_suspend() for filesystem unmount VFS * methods. */ int vfs_write_suspend_umnt(struct mount *mp) { int error; KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, ("vfs_write_suspend_umnt: recursed")); /* dounmount() already called vn_start_write(). */ for (;;) { vn_finished_write(mp); error = vfs_write_suspend(mp, 0); if (error != 0) { vn_start_write(NULL, &mp, V_WAIT); return (error); } MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) break; MNT_IUNLOCK(mp); vn_start_write(NULL, &mp, V_WAIT); } mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); wakeup(&mp->mnt_flag); MNT_IUNLOCK(mp); curthread->td_pflags |= TDP_IGNSUSP; return (0); } /* * Implement kqueues for files by translating it to vnode operation. */ static int vn_kqfilter(struct file *fp, struct knote *kn) { return (VOP_KQFILTER(fp->f_vnode, kn)); } int vn_kqfilter_opath(struct file *fp, struct knote *kn) { if ((fp->f_flag & FKQALLOWED) == 0) return (EBADF); return (vn_kqfilter(fp, kn)); } /* * Simplified in-kernel wrapper calls for extended attribute access. * Both calls pass in a NULL credential, authorizing as "kernel" access. * Set IO_NODELOCKED in ioflg if the vnode is already locked. */ int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; int error; iov.iov_len = *buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = *buflen; if ((ioflg & IO_NODELOCKED) == 0) vn_lock(vp, LK_SHARED | LK_RETRY); ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute retrieval as kernel */ error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) VOP_UNLOCK(vp); if (error == 0) { *buflen = *buflen - auio.uio_resid; } return (error); } /* * XXX failure mode if partially written? */ int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; struct mount *mp; int error; iov.iov_len = buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_WRITE; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = buflen; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute setting as kernel */ error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp); } return (error); } int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td) { struct mount *mp; int error; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute removal as kernel */ error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp); } return (error); } static int vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); } int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) { return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, lkflags, rvp)); } int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp) { struct mount *mp; int ltype, error; ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); mp = vp->v_mount; ltype = VOP_ISLOCKED(vp); KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, ("vn_vget_ino: vp not locked")); error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); VOP_UNLOCK(vp); error = vfs_busy(mp, 0); vn_lock(vp, ltype | LK_RETRY); vfs_rel(mp); if (error != 0) return (ENOENT); if (VN_IS_DOOMED(vp)) { vfs_unbusy(mp); return (ENOENT); } } VOP_UNLOCK(vp); error = alloc(mp, alloc_arg, lkflags, rvp); vfs_unbusy(mp); if (error != 0 || *rvp != vp) vn_lock(vp, ltype | LK_RETRY); if (VN_IS_DOOMED(vp)) { if (error == 0) { if (*rvp == vp) vunref(vp); else vput(*rvp); } error = ENOENT; } return (error); } int vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, struct thread *td) { off_t lim; bool ktr_write; if (td == NULL) return (0); /* * There are conditions where the limit is to be ignored. * However, since it is almost never reached, check it first. */ ktr_write = (td->td_pflags & TDP_INKTRACE) != 0; lim = lim_cur(td, RLIMIT_FSIZE); if (__predict_false(ktr_write)) lim = td->td_ktr_io_lim; if (__predict_true((uoff_t)uio->uio_offset + uio->uio_resid <= lim)) return (0); /* * The limit is reached. */ if (vp->v_type != VREG || (td->td_pflags2 & TDP2_ACCT) != 0) return (0); if (!ktr_write || ktr_filesize_limit_signal) { PROC_LOCK(td->td_proc); kern_psignal(td->td_proc, SIGXFSZ); PROC_UNLOCK(td->td_proc); } return (EFBIG); } int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp); #endif return (setfmode(td, active_cred, vp, mode)); } int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp); #endif return (setfown(td, active_cred, vp, uid, gid)); } /* * Remove pages in the range ["start", "end") from the vnode's VM object. If * "end" is 0, then the range extends to the end of the object. */ void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) { vm_object_t object; if ((object = vp->v_object) == NULL) return; VM_OBJECT_WLOCK(object); vm_object_page_remove(object, start, end, 0); VM_OBJECT_WUNLOCK(object); } /* * Like vn_pages_remove(), but skips invalid pages, which by definition are not * mapped into any process' address space. Filesystems may use this in * preference to vn_pages_remove() to avoid blocking on pages busied in * preparation for a VOP_GETPAGES. */ void vn_pages_remove_valid(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) { vm_object_t object; if ((object = vp->v_object) == NULL) return; VM_OBJECT_WLOCK(object); vm_object_page_remove(object, start, end, OBJPR_VALIDONLY); VM_OBJECT_WUNLOCK(object); } int vn_bmap_seekhole_locked(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) { struct vattr va; daddr_t bn, bnp; uint64_t bsize; off_t noff; int error; KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, ("%s: Wrong command %lu", __func__, cmd)); ASSERT_VOP_LOCKED(vp, "vn_bmap_seekhole_locked"); if (vp->v_type != VREG) { error = ENOTTY; goto out; } error = VOP_GETATTR(vp, &va, cred); if (error != 0) goto out; noff = *off; if (noff >= va.va_size) { error = ENXIO; goto out; } bsize = vp->v_mount->mnt_stat.f_iosize; for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize - noff % bsize) { error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); if (error == EOPNOTSUPP) { error = ENOTTY; goto out; } if ((bnp == -1 && cmd == FIOSEEKHOLE) || (bnp != -1 && cmd == FIOSEEKDATA)) { noff = bn * bsize; if (noff < *off) noff = *off; goto out; } } if (noff > va.va_size) noff = va.va_size; /* noff == va.va_size. There is an implicit hole at the end of file. */ if (cmd == FIOSEEKDATA) error = ENXIO; out: if (error == 0) *off = noff; return (error); } int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) { int error; KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, ("%s: Wrong command %lu", __func__, cmd)); if (vn_lock(vp, LK_SHARED) != 0) return (EBADF); error = vn_bmap_seekhole_locked(vp, cmd, off, cred); VOP_UNLOCK(vp); return (error); } int vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) { struct ucred *cred; struct vnode *vp; struct vattr vattr; off_t foffset, size; int error, noneg; cred = td->td_ucred; vp = fp->f_vnode; foffset = foffset_lock(fp, 0); noneg = (vp->v_type != VCHR); error = 0; switch (whence) { case L_INCR: if (noneg && (foffset < 0 || (offset > 0 && foffset > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += foffset; break; case L_XTND: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, cred); VOP_UNLOCK(vp); if (error) break; /* * If the file references a disk device, then fetch * the media size and use that to determine the ending * offset. */ if (vattr.va_size == 0 && vp->v_type == VCHR && fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) vattr.va_size = size; if (noneg && (vattr.va_size > OFF_MAX || (offset > 0 && vattr.va_size > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += vattr.va_size; break; case L_SET: break; case SEEK_DATA: error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); if (error == ENOTTY) error = EINVAL; break; case SEEK_HOLE: error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); if (error == ENOTTY) error = EINVAL; break; default: error = EINVAL; } if (error == 0 && noneg && offset < 0) error = EINVAL; if (error != 0) goto drop; VFS_KNOTE_UNLOCKED(vp, 0); td->td_uretoff.tdu_off = offset; drop: foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); return (error); } int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { int error; /* * Grant permission if the caller is the owner of the file, or * the super-user, or has ACL_WRITE_ATTRIBUTES permission on * on the file. If the time pointer is null, then write * permission on the file is also sufficient. * * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES * will be allowed to set the times [..] to the current * server time. */ error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) error = VOP_ACCESS(vp, VWRITE, cred, td); return (error); } int vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct vnode *vp; int error; if (fp->f_type == DTYPE_FIFO) kif->kf_type = KF_TYPE_FIFO; else kif->kf_type = KF_TYPE_VNODE; vp = fp->f_vnode; vref(vp); FILEDESC_SUNLOCK(fdp); error = vn_fill_kinfo_vnode(vp, kif); vrele(vp); FILEDESC_SLOCK(fdp); return (error); } static inline void vn_fill_junk(struct kinfo_file *kif) { size_t len, olen; /* * Simulate vn_fullpath returning changing values for a given * vp during e.g. coredump. */ len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1; olen = strlen(kif->kf_path); if (len < olen) strcpy(&kif->kf_path[len - 1], "$"); else for (; olen < len; olen++) strcpy(&kif->kf_path[olen], "A"); } int vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) { struct vattr va; char *fullpath, *freepath; int error; kif->kf_un.kf_file.kf_file_type = vntype_to_kinfo(vp->v_type); freepath = NULL; fullpath = "-"; error = vn_fullpath(vp, &fullpath, &freepath); if (error == 0) { strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); } if (freepath != NULL) free(freepath, M_TEMP); KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path, vn_fill_junk(kif); ); /* * Retrieve vnode attributes. */ va.va_fsid = VNOVAL; va.va_rdev = NODEV; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &va, curthread->td_ucred); VOP_UNLOCK(vp); if (error != 0) return (error); if (va.va_fsid != VNOVAL) kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; else kif->kf_un.kf_file.kf_file_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; kif->kf_un.kf_file.kf_file_fsid_freebsd11 = kif->kf_un.kf_file.kf_file_fsid; /* truncate */ kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); kif->kf_un.kf_file.kf_file_size = va.va_size; kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; kif->kf_un.kf_file.kf_file_rdev_freebsd11 = kif->kf_un.kf_file.kf_file_rdev; /* truncate */ return (0); } int vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { #ifdef HWPMC_HOOKS struct pmckern_map_in pkm; #endif struct mount *mp; struct vnode *vp; vm_object_t object; vm_prot_t maxprot; boolean_t writecounted; int error; #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) /* * POSIX shared-memory objects are defined to have * kernel persistence, and are not defined to support * read(2)/write(2) -- or even open(2). Thus, we can * use MAP_ASYNC to trade on-disk coherence for speed. * The shm_open(3) library routine turns on the FPOSIXSHM * flag to request this behavior. */ if ((fp->f_flag & FPOSIXSHM) != 0) flags |= MAP_NOSYNC; #endif vp = fp->f_vnode; /* * Ensure that file and memory protections are * compatible. Note that we only worry about * writability if mapping is shared; in this case, * current and max prot are dictated by the open file. * XXX use the vnode instead? Problem is: what * credentials do we use for determination? What if * proc does a setuid? */ mp = vp->v_mount; if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { maxprot = VM_PROT_NONE; if ((prot & VM_PROT_EXECUTE) != 0) return (EACCES); } else maxprot = VM_PROT_EXECUTE; if ((fp->f_flag & FREAD) != 0) maxprot |= VM_PROT_READ; else if ((prot & VM_PROT_READ) != 0) return (EACCES); /* * If we are sharing potential changes via MAP_SHARED and we * are trying to get write permission although we opened it * without asking for it, bail out. */ if ((flags & MAP_SHARED) != 0) { if ((fp->f_flag & FWRITE) != 0) maxprot |= VM_PROT_WRITE; else if ((prot & VM_PROT_WRITE) != 0) return (EACCES); } else { maxprot |= VM_PROT_WRITE; cap_maxprot |= VM_PROT_WRITE; } maxprot &= cap_maxprot; /* * For regular files and shared memory, POSIX requires that * the value of foff be a legitimate offset within the data * object. In particular, negative offsets are invalid. * Blocking negative offsets and overflows here avoids * possible wraparound or user-level access into reserved * ranges of the data object later. In contrast, POSIX does * not dictate how offsets are used by device drivers, so in * the case of a device mapping a negative offset is passed * on. */ if ( #ifdef _LP64 size > OFF_MAX || #endif foff > OFF_MAX - size) return (EINVAL); writecounted = FALSE; error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp, &foff, &object, &writecounted); if (error != 0) return (error); error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, writecounted, td); if (error != 0) { /* * If this mapping was accounted for in the vnode's * writecount, then undo that now. */ if (writecounted) vm_pager_release_writecount(object, 0, size); vm_object_deallocate(object); } #ifdef HWPMC_HOOKS /* Inform hwpmc(4) if an executable is being mapped. */ if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) { if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) { pkm.pm_file = vp; pkm.pm_address = (uintptr_t) *addr; PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_MMAP, (void *) &pkm); } } #endif return (error); } void vn_fsid(struct vnode *vp, struct vattr *va) { fsid_t *f; f = &vp->v_mount->mnt_stat.f_fsid; va->va_fsid = (uint32_t)f->val[1]; va->va_fsid <<= sizeof(f->val[1]) * NBBY; va->va_fsid += (uint32_t)f->val[0]; } int vn_fsync_buf(struct vnode *vp, int waitfor) { struct buf *bp, *nbp; struct bufobj *bo; struct mount *mp; int error, maxretry; error = 0; maxretry = 10000; /* large, arbitrarily chosen */ mp = NULL; if (vp->v_type == VCHR) { VI_LOCK(vp); mp = vp->v_rdev->si_mountpt; VI_UNLOCK(vp); } bo = &vp->v_bufobj; BO_LOCK(bo); loop1: /* * MARK/SCAN initialization to avoid infinite loops. */ TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) { bp->b_vflags &= ~BV_SCANNED; bp->b_error = 0; } /* * Flush all dirty buffers associated with a vnode. */ loop2: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if ((bp->b_vflags & BV_SCANNED) != 0) continue; bp->b_vflags |= BV_SCANNED; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { if (waitfor != MNT_WAIT) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL, BO_LOCKPTR(bo)) != 0) { BO_LOCK(bo); goto loop1; } BO_LOCK(bo); } BO_UNLOCK(bo); KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); if ((bp->b_flags & B_DELWRI) == 0) panic("fsync: not dirty"); if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) { vfs_bio_awrite(bp); } else { bremfree(bp); bawrite(bp); } if (maxretry < 1000) pause("dirty", hz < 1000 ? 1 : hz / 1000); BO_LOCK(bo); goto loop2; } /* * If synchronous the caller expects us to completely resolve all * dirty buffers in the system. Wait for in-progress I/O to * complete (which could include background bitmap writes), then * retry if dirty blocks still exist. */ if (waitfor == MNT_WAIT) { bufobj_wwait(bo, 0, 0); if (bo->bo_dirty.bv_cnt > 0) { /* * If we are unable to write any of these buffers * then we fail now rather than trying endlessly * to write them out. */ TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) if ((error = bp->b_error) != 0) break; if ((mp != NULL && mp->mnt_secondary_writes > 0) || (error == 0 && --maxretry >= 0)) goto loop1; if (error == 0) error = EAGAIN; } } BO_UNLOCK(bo); if (error != 0) vn_printf(vp, "fsync: giving up on dirty (error = %d) ", error); return (error); } /* * Copies a byte range from invp to outvp. Calls VOP_COPY_FILE_RANGE() * or vn_generic_copy_file_range() after rangelocking the byte ranges, * to do the actual copy. * vn_generic_copy_file_range() is factored out, so it can be called * from a VOP_COPY_FILE_RANGE() call as well, but handles vnodes from * different file systems. */ int vn_copy_file_range(struct vnode *invp, off_t *inoffp, struct vnode *outvp, off_t *outoffp, size_t *lenp, unsigned int flags, struct ucred *incred, struct ucred *outcred, struct thread *fsize_td) { int error; size_t len; uint64_t uval; len = *lenp; *lenp = 0; /* For error returns. */ error = 0; /* Do some sanity checks on the arguments. */ if (invp->v_type == VDIR || outvp->v_type == VDIR) error = EISDIR; else if (*inoffp < 0 || *outoffp < 0 || invp->v_type != VREG || outvp->v_type != VREG) error = EINVAL; if (error != 0) goto out; /* Ensure offset + len does not wrap around. */ uval = *inoffp; uval += len; if (uval > INT64_MAX) len = INT64_MAX - *inoffp; uval = *outoffp; uval += len; if (uval > INT64_MAX) len = INT64_MAX - *outoffp; if (len == 0) goto out; /* * If the two vnode are for the same file system, call * VOP_COPY_FILE_RANGE(), otherwise call vn_generic_copy_file_range() * which can handle copies across multiple file systems. */ *lenp = len; if (invp->v_mount == outvp->v_mount) error = VOP_COPY_FILE_RANGE(invp, inoffp, outvp, outoffp, lenp, flags, incred, outcred, fsize_td); else error = vn_generic_copy_file_range(invp, inoffp, outvp, outoffp, lenp, flags, incred, outcred, fsize_td); out: return (error); } /* * Test len bytes of data starting at dat for all bytes == 0. * Return true if all bytes are zero, false otherwise. * Expects dat to be well aligned. */ static bool mem_iszero(void *dat, int len) { int i; const u_int *p; const char *cp; for (p = dat; len > 0; len -= sizeof(*p), p++) { if (len >= sizeof(*p)) { if (*p != 0) return (false); } else { cp = (const char *)p; for (i = 0; i < len; i++, cp++) if (*cp != '\0') return (false); } } return (true); } /* * Look for a hole in the output file and, if found, adjust *outoffp * and *xferp to skip past the hole. * *xferp is the entire hole length to be written and xfer2 is how many bytes * to be written as 0's upon return. */ static off_t vn_skip_hole(struct vnode *outvp, off_t xfer2, off_t *outoffp, off_t *xferp, off_t *dataoffp, off_t *holeoffp, struct ucred *cred) { int error; off_t delta; if (*holeoffp == 0 || *holeoffp <= *outoffp) { *dataoffp = *outoffp; error = VOP_IOCTL(outvp, FIOSEEKDATA, dataoffp, 0, cred, curthread); if (error == 0) { *holeoffp = *dataoffp; error = VOP_IOCTL(outvp, FIOSEEKHOLE, holeoffp, 0, cred, curthread); } if (error != 0 || *holeoffp == *dataoffp) { /* * Since outvp is unlocked, it may be possible for * another thread to do a truncate(), lseek(), write() * creating a hole at startoff between the above * VOP_IOCTL() calls, if the other thread does not do * rangelocking. * If that happens, *holeoffp == *dataoffp and finding * the hole has failed, so disable vn_skip_hole(). */ *holeoffp = -1; /* Disable use of vn_skip_hole(). */ return (xfer2); } KASSERT(*dataoffp >= *outoffp, ("vn_skip_hole: dataoff=%jd < outoff=%jd", (intmax_t)*dataoffp, (intmax_t)*outoffp)); KASSERT(*holeoffp > *dataoffp, ("vn_skip_hole: holeoff=%jd <= dataoff=%jd", (intmax_t)*holeoffp, (intmax_t)*dataoffp)); } /* * If there is a hole before the data starts, advance *outoffp and * *xferp past the hole. */ if (*dataoffp > *outoffp) { delta = *dataoffp - *outoffp; if (delta >= *xferp) { /* Entire *xferp is a hole. */ *outoffp += *xferp; *xferp = 0; return (0); } *xferp -= delta; *outoffp += delta; xfer2 = MIN(xfer2, *xferp); } /* * If a hole starts before the end of this xfer2, reduce this xfer2 so * that the write ends at the start of the hole. * *holeoffp should always be greater than *outoffp, but for the * non-INVARIANTS case, check this to make sure xfer2 remains a sane * value. */ if (*holeoffp > *outoffp && *holeoffp < *outoffp + xfer2) xfer2 = *holeoffp - *outoffp; return (xfer2); } /* * Write an xfer sized chunk to outvp in blksize blocks from dat. * dat is a maximum of blksize in length and can be written repeatedly in * the chunk. * If growfile == true, just grow the file via vn_truncate_locked() instead * of doing actual writes. * If checkhole == true, a hole is being punched, so skip over any hole * already in the output file. */ static int vn_write_outvp(struct vnode *outvp, char *dat, off_t outoff, off_t xfer, u_long blksize, bool growfile, bool checkhole, struct ucred *cred) { struct mount *mp; off_t dataoff, holeoff, xfer2; int error; /* * Loop around doing writes of blksize until write has been completed. * Lock/unlock on each loop iteration so that a bwillwrite() can be * done for each iteration, since the xfer argument can be very * large if there is a large hole to punch in the output file. */ error = 0; holeoff = 0; do { xfer2 = MIN(xfer, blksize); if (checkhole) { /* * Punching a hole. Skip writing if there is * already a hole in the output file. */ xfer2 = vn_skip_hole(outvp, xfer2, &outoff, &xfer, &dataoff, &holeoff, cred); if (xfer == 0) break; if (holeoff < 0) checkhole = false; KASSERT(xfer2 > 0, ("vn_write_outvp: xfer2=%jd", (intmax_t)xfer2)); } bwillwrite(); mp = NULL; error = vn_start_write(outvp, &mp, V_WAIT); if (error != 0) break; if (growfile) { error = vn_lock(outvp, LK_EXCLUSIVE); if (error == 0) { error = vn_truncate_locked(outvp, outoff + xfer, false, cred); VOP_UNLOCK(outvp); } } else { error = vn_lock(outvp, vn_lktype_write(mp, outvp)); if (error == 0) { error = vn_rdwr(UIO_WRITE, outvp, dat, xfer2, outoff, UIO_SYSSPACE, IO_NODELOCKED, curthread->td_ucred, cred, NULL, curthread); outoff += xfer2; xfer -= xfer2; VOP_UNLOCK(outvp); } } if (mp != NULL) vn_finished_write(mp); } while (!growfile && xfer > 0 && error == 0); return (error); } /* * Copy a byte range of one file to another. This function can handle the * case where invp and outvp are on different file systems. * It can also be called by a VOP_COPY_FILE_RANGE() to do the work, if there * is no better file system specific way to do it. */ int vn_generic_copy_file_range(struct vnode *invp, off_t *inoffp, struct vnode *outvp, off_t *outoffp, size_t *lenp, unsigned int flags, struct ucred *incred, struct ucred *outcred, struct thread *fsize_td) { struct vattr va, inva; struct mount *mp; struct uio io; off_t startoff, endoff, xfer, xfer2; u_long blksize; int error, interrupted; bool cantseek, readzeros, eof, lastblock, holetoeof; ssize_t aresid; size_t copylen, len, rem, savlen; char *dat; long holein, holeout; struct timespec curts, endts; holein = holeout = 0; savlen = len = *lenp; error = 0; interrupted = 0; dat = NULL; error = vn_lock(invp, LK_SHARED); if (error != 0) goto out; if (VOP_PATHCONF(invp, _PC_MIN_HOLE_SIZE, &holein) != 0) holein = 0; if (holein > 0) error = VOP_GETATTR(invp, &inva, incred); VOP_UNLOCK(invp); if (error != 0) goto out; mp = NULL; error = vn_start_write(outvp, &mp, V_WAIT); if (error == 0) error = vn_lock(outvp, LK_EXCLUSIVE); if (error == 0) { /* * If fsize_td != NULL, do a vn_rlimit_fsize() call, * now that outvp is locked. */ if (fsize_td != NULL) { io.uio_offset = *outoffp; io.uio_resid = len; error = vn_rlimit_fsize(outvp, &io, fsize_td); if (error != 0) error = EFBIG; } if (VOP_PATHCONF(outvp, _PC_MIN_HOLE_SIZE, &holeout) != 0) holeout = 0; /* * Holes that are past EOF do not need to be written as a block * of zero bytes. So, truncate the output file as far as * possible and then use va.va_size to decide if writing 0 * bytes is necessary in the loop below. */ if (error == 0) error = VOP_GETATTR(outvp, &va, outcred); if (error == 0 && va.va_size > *outoffp && va.va_size <= *outoffp + len) { #ifdef MAC error = mac_vnode_check_write(curthread->td_ucred, outcred, outvp); if (error == 0) #endif error = vn_truncate_locked(outvp, *outoffp, false, outcred); if (error == 0) va.va_size = *outoffp; } VOP_UNLOCK(outvp); } if (mp != NULL) vn_finished_write(mp); if (error != 0) goto out; /* * Set the blksize to the larger of the hole sizes for invp and outvp. * If hole sizes aren't available, set the blksize to the larger * f_iosize of invp and outvp. * This code expects the hole sizes and f_iosizes to be powers of 2. * This value is clipped at 4Kbytes and 1Mbyte. */ blksize = MAX(holein, holeout); /* Clip len to end at an exact multiple of hole size. */ if (blksize > 1) { rem = *inoffp % blksize; if (rem > 0) rem = blksize - rem; if (len > rem && len - rem > blksize) len = savlen = rounddown(len - rem, blksize) + rem; } if (blksize <= 1) blksize = MAX(invp->v_mount->mnt_stat.f_iosize, outvp->v_mount->mnt_stat.f_iosize); if (blksize < 4096) blksize = 4096; else if (blksize > 1024 * 1024) blksize = 1024 * 1024; dat = malloc(blksize, M_TEMP, M_WAITOK); /* * If VOP_IOCTL(FIOSEEKHOLE) works for invp, use it and FIOSEEKDATA * to find holes. Otherwise, just scan the read block for all 0s * in the inner loop where the data copying is done. * Note that some file systems such as NFSv3, NFSv4.0 and NFSv4.1 may * support holes on the server, but do not support FIOSEEKHOLE. * The kernel flag COPY_FILE_RANGE_TIMEO1SEC is used to indicate * that this function should return after 1second with a partial * completion. */ if ((flags & COPY_FILE_RANGE_TIMEO1SEC) != 0) { getnanouptime(&endts); endts.tv_sec++; } else timespecclear(&endts); holetoeof = eof = false; while (len > 0 && error == 0 && !eof && interrupted == 0) { endoff = 0; /* To shut up compilers. */ cantseek = true; startoff = *inoffp; copylen = len; /* * Find the next data area. If there is just a hole to EOF, * FIOSEEKDATA should fail with ENXIO. * (I do not know if any file system will report a hole to * EOF via FIOSEEKHOLE, but I am pretty sure FIOSEEKDATA * will fail for those file systems.) * * For input files that don't support FIOSEEKDATA/FIOSEEKHOLE, * the code just falls through to the inner copy loop. */ error = EINVAL; if (holein > 0) { error = VOP_IOCTL(invp, FIOSEEKDATA, &startoff, 0, incred, curthread); if (error == ENXIO) { startoff = endoff = inva.va_size; eof = holetoeof = true; error = 0; } } if (error == 0 && !holetoeof) { endoff = startoff; error = VOP_IOCTL(invp, FIOSEEKHOLE, &endoff, 0, incred, curthread); /* * Since invp is unlocked, it may be possible for * another thread to do a truncate(), lseek(), write() * creating a hole at startoff between the above * VOP_IOCTL() calls, if the other thread does not do * rangelocking. * If that happens, startoff == endoff and finding * the hole has failed, so set an error. */ if (error == 0 && startoff == endoff) error = EINVAL; /* Any error. Reset to 0. */ } if (error == 0) { if (startoff > *inoffp) { /* Found hole before data block. */ xfer = MIN(startoff - *inoffp, len); if (*outoffp < va.va_size) { /* Must write 0s to punch hole. */ xfer2 = MIN(va.va_size - *outoffp, xfer); memset(dat, 0, MIN(xfer2, blksize)); error = vn_write_outvp(outvp, dat, *outoffp, xfer2, blksize, false, holeout > 0, outcred); } if (error == 0 && *outoffp + xfer > va.va_size && (xfer == len || holetoeof)) { /* Grow output file (hole at end). */ error = vn_write_outvp(outvp, dat, *outoffp, xfer, blksize, true, false, outcred); } if (error == 0) { *inoffp += xfer; *outoffp += xfer; len -= xfer; if (len < savlen) { interrupted = sig_intr(); if (timespecisset(&endts) && interrupted == 0) { getnanouptime(&curts); if (timespeccmp(&curts, &endts, >=)) interrupted = EINTR; } } } } copylen = MIN(len, endoff - startoff); cantseek = false; } else { cantseek = true; startoff = *inoffp; copylen = len; error = 0; } xfer = blksize; if (cantseek) { /* * Set first xfer to end at a block boundary, so that * holes are more likely detected in the loop below via * the for all bytes 0 method. */ xfer -= (*inoffp % blksize); } /* Loop copying the data block. */ while (copylen > 0 && error == 0 && !eof && interrupted == 0) { if (copylen < xfer) xfer = copylen; error = vn_lock(invp, LK_SHARED); if (error != 0) goto out; error = vn_rdwr(UIO_READ, invp, dat, xfer, startoff, UIO_SYSSPACE, IO_NODELOCKED, curthread->td_ucred, incred, &aresid, curthread); VOP_UNLOCK(invp); lastblock = false; if (error == 0 && aresid > 0) { /* Stop the copy at EOF on the input file. */ xfer -= aresid; eof = true; lastblock = true; } if (error == 0) { /* * Skip the write for holes past the initial EOF * of the output file, unless this is the last * write of the output file at EOF. */ readzeros = cantseek ? mem_iszero(dat, xfer) : false; if (xfer == len) lastblock = true; if (!cantseek || *outoffp < va.va_size || lastblock || !readzeros) error = vn_write_outvp(outvp, dat, *outoffp, xfer, blksize, readzeros && lastblock && *outoffp >= va.va_size, false, outcred); if (error == 0) { *inoffp += xfer; startoff += xfer; *outoffp += xfer; copylen -= xfer; len -= xfer; if (len < savlen) { interrupted = sig_intr(); if (timespecisset(&endts) && interrupted == 0) { getnanouptime(&curts); if (timespeccmp(&curts, &endts, >=)) interrupted = EINTR; } } } } xfer = blksize; } } out: *lenp = savlen - len; free(dat, M_TEMP); return (error); } static int vn_fallocate(struct file *fp, off_t offset, off_t len, struct thread *td) { struct mount *mp; struct vnode *vp; off_t olen, ooffset; int error; #ifdef AUDIT int audited_vnode1 = 0; #endif vp = fp->f_vnode; if (vp->v_type != VREG) return (ENODEV); /* Allocating blocks may take a long time, so iterate. */ for (;;) { olen = len; ooffset = offset; bwillwrite(); mp = NULL; - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error != 0) break; error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) { vn_finished_write(mp); break; } #ifdef AUDIT if (!audited_vnode1) { AUDIT_ARG_VNODE1(vp); audited_vnode1 = 1; } #endif #ifdef MAC error = mac_vnode_check_write(td->td_ucred, fp->f_cred, vp); if (error == 0) #endif error = VOP_ALLOCATE(vp, &offset, &len, 0, td->td_ucred); VOP_UNLOCK(vp); vn_finished_write(mp); if (olen + ooffset != offset + len) { panic("offset + len changed from %jx/%jx to %jx/%jx", ooffset, olen, offset, len); } if (error != 0 || len == 0) break; KASSERT(olen > len, ("Iteration did not make progress?")); maybe_yield(); } return (error); } static int vn_deallocate_impl(struct vnode *vp, off_t *offset, off_t *length, int flags, int ioflag, struct ucred *cred, struct ucred *active_cred, struct ucred *file_cred) { struct mount *mp; void *rl_cookie; off_t off, len; int error; #ifdef AUDIT bool audited_vnode1 = false; #endif rl_cookie = NULL; error = 0; mp = NULL; off = *offset; len = *length; if ((ioflag & (IO_NODELOCKED | IO_RANGELOCKED)) == 0) rl_cookie = vn_rangelock_wlock(vp, off, off + len); while (len > 0 && error == 0) { /* * Try to deallocate the longest range in one pass. * In case a pass takes too long to be executed, it returns * partial result. The residue will be proceeded in the next * pass. */ if ((ioflag & IO_NODELOCKED) == 0) { bwillwrite(); if ((error = vn_start_write(vp, &mp, - V_WAIT | PCATCH)) != 0) + V_WAIT | V_PCATCH)) != 0) goto out; vn_lock(vp, vn_lktype_write(mp, vp) | LK_RETRY); } #ifdef AUDIT if (!audited_vnode1) { AUDIT_ARG_VNODE1(vp); audited_vnode1 = true; } #endif #ifdef MAC if ((ioflag & IO_NOMACCHECK) == 0) error = mac_vnode_check_write(active_cred, file_cred, vp); #endif if (error == 0) error = VOP_DEALLOCATE(vp, &off, &len, flags, ioflag, cred); if ((ioflag & IO_NODELOCKED) == 0) { VOP_UNLOCK(vp); if (mp != NULL) { vn_finished_write(mp); mp = NULL; } } if (error == 0 && len != 0) maybe_yield(); } out: if (rl_cookie != NULL) vn_rangelock_unlock(vp, rl_cookie); *offset = off; *length = len; return (error); } /* * This function is supposed to be used in the situations where the deallocation * is not triggered by a user request. */ int vn_deallocate(struct vnode *vp, off_t *offset, off_t *length, int flags, int ioflag, struct ucred *active_cred, struct ucred *file_cred) { struct ucred *cred; if (*offset < 0 || *length <= 0 || *length > OFF_MAX - *offset || flags != 0) return (EINVAL); if (vp->v_type != VREG) return (ENODEV); cred = file_cred != NOCRED ? file_cred : active_cred; return (vn_deallocate_impl(vp, offset, length, flags, ioflag, cred, active_cred, file_cred)); } static int vn_fspacectl(struct file *fp, int cmd, off_t *offset, off_t *length, int flags, struct ucred *active_cred, struct thread *td) { int error; struct vnode *vp; int ioflag; KASSERT(cmd == SPACECTL_DEALLOC, ("vn_fspacectl: Invalid cmd")); KASSERT((flags & ~SPACECTL_F_SUPPORTED) == 0, ("vn_fspacectl: non-zero flags")); KASSERT(*offset >= 0 && *length > 0 && *length <= OFF_MAX - *offset, ("vn_fspacectl: offset/length overflow or underflow")); vp = fp->f_vnode; if (vp->v_type != VREG) return (ENODEV); ioflag = get_write_ioflag(fp); switch (cmd) { case SPACECTL_DEALLOC: error = vn_deallocate_impl(vp, offset, length, flags, ioflag, active_cred, active_cred, fp->f_cred); break; default: panic("vn_fspacectl: unknown cmd %d", cmd); } return (error); } static u_long vn_lock_pair_pause_cnt; SYSCTL_ULONG(_debug, OID_AUTO, vn_lock_pair_pause, CTLFLAG_RD, &vn_lock_pair_pause_cnt, 0, "Count of vn_lock_pair deadlocks"); u_int vn_lock_pair_pause_max; SYSCTL_UINT(_debug, OID_AUTO, vn_lock_pair_pause_max, CTLFLAG_RW, &vn_lock_pair_pause_max, 0, "Max ticks for vn_lock_pair deadlock avoidance sleep"); static void vn_lock_pair_pause(const char *wmesg) { atomic_add_long(&vn_lock_pair_pause_cnt, 1); pause(wmesg, prng32_bounded(vn_lock_pair_pause_max)); } /* * Lock pair of vnodes vp1, vp2, avoiding lock order reversal. * vp1_locked indicates whether vp1 is exclusively locked; if not, vp1 * must be unlocked. Same for vp2 and vp2_locked. One of the vnodes * can be NULL. * * The function returns with both vnodes exclusively locked, and * guarantees that it does not create lock order reversal with other * threads during its execution. Both vnodes could be unlocked * temporary (and reclaimed). */ void vn_lock_pair(struct vnode *vp1, bool vp1_locked, struct vnode *vp2, bool vp2_locked) { int error; if (vp1 == NULL && vp2 == NULL) return; if (vp1 != NULL) { if (vp1_locked) ASSERT_VOP_ELOCKED(vp1, "vp1"); else ASSERT_VOP_UNLOCKED(vp1, "vp1"); } else { vp1_locked = true; } if (vp2 != NULL) { if (vp2_locked) ASSERT_VOP_ELOCKED(vp2, "vp2"); else ASSERT_VOP_UNLOCKED(vp2, "vp2"); } else { vp2_locked = true; } if (!vp1_locked && !vp2_locked) { vn_lock(vp1, LK_EXCLUSIVE | LK_RETRY); vp1_locked = true; } for (;;) { if (vp1_locked && vp2_locked) break; if (vp1_locked && vp2 != NULL) { if (vp1 != NULL) { error = VOP_LOCK1(vp2, LK_EXCLUSIVE | LK_NOWAIT, __FILE__, __LINE__); if (error == 0) break; VOP_UNLOCK(vp1); vp1_locked = false; vn_lock_pair_pause("vlp1"); } vn_lock(vp2, LK_EXCLUSIVE | LK_RETRY); vp2_locked = true; } if (vp2_locked && vp1 != NULL) { if (vp2 != NULL) { error = VOP_LOCK1(vp1, LK_EXCLUSIVE | LK_NOWAIT, __FILE__, __LINE__); if (error == 0) break; VOP_UNLOCK(vp2); vp2_locked = false; vn_lock_pair_pause("vlp2"); } vn_lock(vp1, LK_EXCLUSIVE | LK_RETRY); vp1_locked = true; } } if (vp1 != NULL) ASSERT_VOP_ELOCKED(vp1, "vp1 ret"); if (vp2 != NULL) ASSERT_VOP_ELOCKED(vp2, "vp2 ret"); } int vn_lktype_write(struct mount *mp, struct vnode *vp) { if (MNT_SHARED_WRITES(mp) || (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) return (LK_SHARED); return (LK_EXCLUSIVE); } diff --git a/sys/security/mac/mac_syscalls.c b/sys/security/mac/mac_syscalls.c index d8979a70b72d..9bc334686aa4 100644 --- a/sys/security/mac/mac_syscalls.c +++ b/sys/security/mac/mac_syscalls.c @@ -1,662 +1,662 @@ /*- * Copyright (c) 1999-2002, 2006, 2009 Robert N. M. Watson * Copyright (c) 2001 Ilmar S. Habibulin * Copyright (c) 2001-2005 Networks Associates Technology, Inc. * Copyright (c) 2005-2006 SPARTA, Inc. * Copyright (c) 2008 Apple Inc. * All rights reserved. * * This software was developed by Robert Watson and Ilmar Habibulin for the * TrustedBSD Project. * * This software was developed for the FreeBSD Project in part by Network * Associates Laboratories, the Security Research Division of Network * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), * as part of the DARPA CHATS research program. * * This software was enhanced by SPARTA ISSO under SPAWAR contract * N66001-04-C-6019 ("SEFOS"). * * This software was developed at the University of Cambridge Computer * Laboratory with support from a grant from Google, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_mac.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef MAC FEATURE(security_mac, "Mandatory Access Control Framework support"); static int kern___mac_get_path(struct thread *td, const char *path_p, struct mac *mac_p, int follow); static int kern___mac_set_path(struct thread *td, const char *path_p, struct mac *mac_p, int follow); int sys___mac_get_pid(struct thread *td, struct __mac_get_pid_args *uap) { char *elements, *buffer; struct mac mac; struct proc *tproc; struct ucred *tcred; int error; error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); tproc = pfind(uap->pid); if (tproc == NULL) return (ESRCH); tcred = NULL; /* Satisfy gcc. */ error = p_cansee(td, tproc); if (error == 0) tcred = crhold(tproc->p_ucred); PROC_UNLOCK(tproc); if (error) return (error); elements = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, elements, mac.m_buflen, NULL); if (error) { free(elements, M_MACTEMP); crfree(tcred); return (error); } buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK | M_ZERO); error = mac_cred_externalize_label(tcred->cr_label, elements, buffer, mac.m_buflen); if (error == 0) error = copyout(buffer, mac.m_string, strlen(buffer)+1); free(buffer, M_MACTEMP); free(elements, M_MACTEMP); crfree(tcred); return (error); } int sys___mac_get_proc(struct thread *td, struct __mac_get_proc_args *uap) { char *elements, *buffer; struct mac mac; int error; error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); elements = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, elements, mac.m_buflen, NULL); if (error) { free(elements, M_MACTEMP); return (error); } buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK | M_ZERO); error = mac_cred_externalize_label(td->td_ucred->cr_label, elements, buffer, mac.m_buflen); if (error == 0) error = copyout(buffer, mac.m_string, strlen(buffer)+1); free(buffer, M_MACTEMP); free(elements, M_MACTEMP); return (error); } int sys___mac_set_proc(struct thread *td, struct __mac_set_proc_args *uap) { struct ucred *newcred, *oldcred; struct label *intlabel; struct proc *p; struct mac mac; char *buffer; int error; if (!(mac_labeled & MPC_OBJECT_CRED)) return (EINVAL); error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, buffer, mac.m_buflen, NULL); if (error) { free(buffer, M_MACTEMP); return (error); } intlabel = mac_cred_label_alloc(); error = mac_cred_internalize_label(intlabel, buffer); free(buffer, M_MACTEMP); if (error) goto out; newcred = crget(); p = td->td_proc; PROC_LOCK(p); oldcred = p->p_ucred; error = mac_cred_check_relabel(oldcred, intlabel); if (error) { PROC_UNLOCK(p); crfree(newcred); goto out; } setsugid(p); crcopy(newcred, oldcred); mac_cred_relabel(newcred, intlabel); proc_set_cred(p, newcred); PROC_UNLOCK(p); crfree(oldcred); mac_proc_vm_revoke(td); out: mac_cred_label_free(intlabel); return (error); } int sys___mac_get_fd(struct thread *td, struct __mac_get_fd_args *uap) { char *elements, *buffer; struct label *intlabel; struct file *fp; struct mac mac; struct vnode *vp; struct pipe *pipe; struct socket *so; cap_rights_t rights; int error; error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); elements = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, elements, mac.m_buflen, NULL); if (error) { free(elements, M_MACTEMP); return (error); } buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK | M_ZERO); error = fget(td, uap->fd, cap_rights_init_one(&rights, CAP_MAC_GET), &fp); if (error) goto out; switch (fp->f_type) { case DTYPE_FIFO: case DTYPE_VNODE: if (!(mac_labeled & MPC_OBJECT_VNODE)) { error = EINVAL; goto out_fdrop; } vp = fp->f_vnode; intlabel = mac_vnode_label_alloc(); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); mac_vnode_copy_label(vp->v_label, intlabel); VOP_UNLOCK(vp); error = mac_vnode_externalize_label(intlabel, elements, buffer, mac.m_buflen); mac_vnode_label_free(intlabel); break; case DTYPE_PIPE: if (!(mac_labeled & MPC_OBJECT_PIPE)) { error = EINVAL; goto out_fdrop; } pipe = fp->f_data; intlabel = mac_pipe_label_alloc(); PIPE_LOCK(pipe); mac_pipe_copy_label(pipe->pipe_pair->pp_label, intlabel); PIPE_UNLOCK(pipe); error = mac_pipe_externalize_label(intlabel, elements, buffer, mac.m_buflen); mac_pipe_label_free(intlabel); break; case DTYPE_SOCKET: if (!(mac_labeled & MPC_OBJECT_SOCKET)) { error = EINVAL; goto out_fdrop; } so = fp->f_data; intlabel = mac_socket_label_alloc(M_WAITOK); SOCK_LOCK(so); mac_socket_copy_label(so->so_label, intlabel); SOCK_UNLOCK(so); error = mac_socket_externalize_label(intlabel, elements, buffer, mac.m_buflen); mac_socket_label_free(intlabel); break; default: error = EINVAL; } if (error == 0) error = copyout(buffer, mac.m_string, strlen(buffer)+1); out_fdrop: fdrop(fp, td); out: free(buffer, M_MACTEMP); free(elements, M_MACTEMP); return (error); } int sys___mac_get_file(struct thread *td, struct __mac_get_file_args *uap) { return (kern___mac_get_path(td, uap->path_p, uap->mac_p, FOLLOW)); } int sys___mac_get_link(struct thread *td, struct __mac_get_link_args *uap) { return (kern___mac_get_path(td, uap->path_p, uap->mac_p, NOFOLLOW)); } static int kern___mac_get_path(struct thread *td, const char *path_p, struct mac *mac_p, int follow) { char *elements, *buffer; struct nameidata nd; struct label *intlabel; struct mac mac; int error; if (!(mac_labeled & MPC_OBJECT_VNODE)) return (EINVAL); error = copyin(mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); elements = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, elements, mac.m_buflen, NULL); if (error) { free(elements, M_MACTEMP); return (error); } buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK | M_ZERO); NDINIT(&nd, LOOKUP, LOCKLEAF | follow, UIO_USERSPACE, path_p); error = namei(&nd); if (error) goto out; intlabel = mac_vnode_label_alloc(); mac_vnode_copy_label(nd.ni_vp->v_label, intlabel); error = mac_vnode_externalize_label(intlabel, elements, buffer, mac.m_buflen); NDFREE(&nd, 0); mac_vnode_label_free(intlabel); if (error == 0) error = copyout(buffer, mac.m_string, strlen(buffer)+1); out: free(buffer, M_MACTEMP); free(elements, M_MACTEMP); return (error); } int sys___mac_set_fd(struct thread *td, struct __mac_set_fd_args *uap) { struct label *intlabel; struct pipe *pipe; struct socket *so; struct file *fp; struct mount *mp; struct vnode *vp; struct mac mac; cap_rights_t rights; char *buffer; int error; error = copyin(uap->mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, buffer, mac.m_buflen, NULL); if (error) { free(buffer, M_MACTEMP); return (error); } error = fget(td, uap->fd, cap_rights_init_one(&rights, CAP_MAC_SET), &fp); if (error) goto out; switch (fp->f_type) { case DTYPE_FIFO: case DTYPE_VNODE: if (!(mac_labeled & MPC_OBJECT_VNODE)) { error = EINVAL; goto out_fdrop; } intlabel = mac_vnode_label_alloc(); error = mac_vnode_internalize_label(intlabel, buffer); if (error) { mac_vnode_label_free(intlabel); break; } vp = fp->f_vnode; - error = vn_start_write(vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(vp, &mp, V_WAIT | V_PCATCH); if (error != 0) { mac_vnode_label_free(intlabel); break; } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = vn_setlabel(vp, intlabel, td->td_ucred); VOP_UNLOCK(vp); vn_finished_write(mp); mac_vnode_label_free(intlabel); break; case DTYPE_PIPE: if (!(mac_labeled & MPC_OBJECT_PIPE)) { error = EINVAL; goto out_fdrop; } intlabel = mac_pipe_label_alloc(); error = mac_pipe_internalize_label(intlabel, buffer); if (error == 0) { pipe = fp->f_data; PIPE_LOCK(pipe); error = mac_pipe_label_set(td->td_ucred, pipe->pipe_pair, intlabel); PIPE_UNLOCK(pipe); } mac_pipe_label_free(intlabel); break; case DTYPE_SOCKET: if (!(mac_labeled & MPC_OBJECT_SOCKET)) { error = EINVAL; goto out_fdrop; } intlabel = mac_socket_label_alloc(M_WAITOK); error = mac_socket_internalize_label(intlabel, buffer); if (error == 0) { so = fp->f_data; error = mac_socket_label_set(td->td_ucred, so, intlabel); } mac_socket_label_free(intlabel); break; default: error = EINVAL; } out_fdrop: fdrop(fp, td); out: free(buffer, M_MACTEMP); return (error); } int sys___mac_set_file(struct thread *td, struct __mac_set_file_args *uap) { return (kern___mac_set_path(td, uap->path_p, uap->mac_p, FOLLOW)); } int sys___mac_set_link(struct thread *td, struct __mac_set_link_args *uap) { return (kern___mac_set_path(td, uap->path_p, uap->mac_p, NOFOLLOW)); } static int kern___mac_set_path(struct thread *td, const char *path_p, struct mac *mac_p, int follow) { struct label *intlabel; struct nameidata nd; struct mount *mp; struct mac mac; char *buffer; int error; if (!(mac_labeled & MPC_OBJECT_VNODE)) return (EINVAL); error = copyin(mac_p, &mac, sizeof(mac)); if (error) return (error); error = mac_check_structmac_consistent(&mac); if (error) return (error); buffer = malloc(mac.m_buflen, M_MACTEMP, M_WAITOK); error = copyinstr(mac.m_string, buffer, mac.m_buflen, NULL); if (error) { free(buffer, M_MACTEMP); return (error); } intlabel = mac_vnode_label_alloc(); error = mac_vnode_internalize_label(intlabel, buffer); free(buffer, M_MACTEMP); if (error) goto out; NDINIT(&nd, LOOKUP, LOCKLEAF | follow, UIO_USERSPACE, path_p); error = namei(&nd); if (error == 0) { - error = vn_start_write(nd.ni_vp, &mp, V_WAIT | PCATCH); + error = vn_start_write(nd.ni_vp, &mp, V_WAIT | V_PCATCH); if (error == 0) { error = vn_setlabel(nd.ni_vp, intlabel, td->td_ucred); vn_finished_write(mp); } } NDFREE(&nd, 0); out: mac_vnode_label_free(intlabel); return (error); } int sys_mac_syscall(struct thread *td, struct mac_syscall_args *uap) { struct mac_policy_conf *mpc; char target[MAC_MAX_POLICY_NAME]; int error; error = copyinstr(uap->policy, target, sizeof(target), NULL); if (error) return (error); error = ENOSYS; LIST_FOREACH(mpc, &mac_static_policy_list, mpc_list) { if (strcmp(mpc->mpc_name, target) == 0 && mpc->mpc_ops->mpo_syscall != NULL) { error = mpc->mpc_ops->mpo_syscall(td, uap->call, uap->arg); goto out; } } if (!LIST_EMPTY(&mac_policy_list)) { mac_policy_slock_sleep(); LIST_FOREACH(mpc, &mac_policy_list, mpc_list) { if (strcmp(mpc->mpc_name, target) == 0 && mpc->mpc_ops->mpo_syscall != NULL) { error = mpc->mpc_ops->mpo_syscall(td, uap->call, uap->arg); break; } } mac_policy_sunlock_sleep(); } out: return (error); } #else /* !MAC */ int sys___mac_get_pid(struct thread *td, struct __mac_get_pid_args *uap) { return (ENOSYS); } int sys___mac_get_proc(struct thread *td, struct __mac_get_proc_args *uap) { return (ENOSYS); } int sys___mac_set_proc(struct thread *td, struct __mac_set_proc_args *uap) { return (ENOSYS); } int sys___mac_get_fd(struct thread *td, struct __mac_get_fd_args *uap) { return (ENOSYS); } int sys___mac_get_file(struct thread *td, struct __mac_get_file_args *uap) { return (ENOSYS); } int sys___mac_get_link(struct thread *td, struct __mac_get_link_args *uap) { return (ENOSYS); } int sys___mac_set_fd(struct thread *td, struct __mac_set_fd_args *uap) { return (ENOSYS); } int sys___mac_set_file(struct thread *td, struct __mac_set_file_args *uap) { return (ENOSYS); } int sys___mac_set_link(struct thread *td, struct __mac_set_link_args *uap) { return (ENOSYS); } int sys_mac_syscall(struct thread *td, struct mac_syscall_args *uap) { return (ENOSYS); } #endif /* !MAC */ diff --git a/sys/sys/vnode.h b/sys/sys/vnode.h index 30b74fef0902..4ae3dd22a390 100644 --- a/sys/sys/vnode.h +++ b/sys/sys/vnode.h @@ -1,1162 +1,1164 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vnode.h 8.7 (Berkeley) 2/4/94 * $FreeBSD$ */ #ifndef _SYS_VNODE_H_ #define _SYS_VNODE_H_ #include #include #include #include #include #include #include #include #include #include #include /* * The vnode is the focus of all file activity in UNIX. There is a * unique vnode allocated for each active file, each current directory, * each mounted-on file, text file, and the root. */ /* * Vnode types. VNON means no type. */ enum vtype { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO, VBAD, VMARKER }; enum vgetstate { VGET_NONE, VGET_HOLDCNT, VGET_USECOUNT }; /* * Each underlying filesystem allocates its own private area and hangs * it from v_data. If non-null, this area is freed in getnewvnode(). */ struct namecache; struct cache_fpl; struct vpollinfo { struct mtx vpi_lock; /* lock to protect below */ struct selinfo vpi_selinfo; /* identity of poller(s) */ short vpi_events; /* what they are looking for */ short vpi_revents; /* what has happened */ }; /* * Reading or writing any of these items requires holding the appropriate lock. * * Lock reference: * c - namecache mutex * i - interlock * l - mp mnt_listmtx or freelist mutex * I - updated with atomics, 0->1 and 1->0 transitions with interlock held * m - mount point interlock * p - pollinfo lock * u - Only a reference to the vnode is needed to read. * v - vnode lock * * Vnodes may be found on many lists. The general way to deal with operating * on a vnode that is on a list is: * 1) Lock the list and find the vnode. * 2) Lock interlock so that the vnode does not go away. * 3) Unlock the list to avoid lock order reversals. * 4) vget with LK_INTERLOCK and check for ENOENT, or * 5) Check for DOOMED if the vnode lock is not required. * 6) Perform your operation, then vput(). */ #if defined(_KERNEL) || defined(_KVM_VNODE) struct vnode { /* * Fields which define the identity of the vnode. These fields are * owned by the filesystem (XXX: and vgone() ?) */ enum vtype v_type:8; /* u vnode type */ short v_irflag; /* i frequently read flags */ seqc_t v_seqc; /* i modification count */ uint32_t v_nchash; /* u namecache hash */ u_int v_hash; struct vop_vector *v_op; /* u vnode operations vector */ void *v_data; /* u private data for fs */ /* * Filesystem instance stuff */ struct mount *v_mount; /* u ptr to vfs we are in */ TAILQ_ENTRY(vnode) v_nmntvnodes; /* m vnodes for mount point */ /* * Type specific fields, only one applies to any given vnode. */ union { struct mount *v_mountedhere; /* v ptr to mountpoint (VDIR) */ struct unpcb *v_unpcb; /* v unix domain net (VSOCK) */ struct cdev *v_rdev; /* v device (VCHR, VBLK) */ struct fifoinfo *v_fifoinfo; /* v fifo (VFIFO) */ }; /* * vfs_hash: (mount + inode) -> vnode hash. The hash value * itself is grouped with other int fields, to avoid padding. */ LIST_ENTRY(vnode) v_hashlist; /* * VFS_namecache stuff */ LIST_HEAD(, namecache) v_cache_src; /* c Cache entries from us */ TAILQ_HEAD(, namecache) v_cache_dst; /* c Cache entries to us */ struct namecache *v_cache_dd; /* c Cache entry for .. vnode */ /* * Locking */ struct lock v_lock; /* u (if fs don't have one) */ struct mtx v_interlock; /* lock for "i" things */ struct lock *v_vnlock; /* u pointer to vnode lock */ /* * The machinery of being a vnode */ TAILQ_ENTRY(vnode) v_vnodelist; /* l vnode lists */ TAILQ_ENTRY(vnode) v_lazylist; /* l vnode lazy list */ struct bufobj v_bufobj; /* * Buffer cache object */ /* * Hooks for various subsystems and features. */ struct vpollinfo *v_pollinfo; /* i Poll events, p for *v_pi */ struct label *v_label; /* MAC label for vnode */ struct lockf *v_lockf; /* Byte-level advisory lock list */ struct rangelock v_rl; /* Byte-range lock */ u_int v_holdcnt; /* I prevents recycling. */ u_int v_usecount; /* I ref count of users */ u_short v_iflag; /* i vnode flags (see below) */ u_short v_vflag; /* v vnode flags */ u_short v_mflag; /* l mnt-specific vnode flags */ short v_dbatchcpu; /* i LRU requeue deferral batch */ int v_writecount; /* I ref count of writers or (negative) text users */ int v_seqc_users; /* i modifications pending */ }; #ifndef DEBUG_LOCKS #ifdef _LP64 /* * Not crossing 448 bytes fits 9 vnodes per page. If you have to add fields * to the structure and there is nothing which can be done to prevent growth * then so be it. But don't grow it without a good reason. */ _Static_assert(sizeof(struct vnode) <= 448, "vnode size crosses 448 bytes"); #endif #endif #endif /* defined(_KERNEL) || defined(_KVM_VNODE) */ #define bo2vnode(bo) __containerof((bo), struct vnode, v_bufobj) /* XXX: These are temporary to avoid a source sweep at this time */ #define v_object v_bufobj.bo_object /* * Userland version of struct vnode, for sysctl. */ struct xvnode { size_t xv_size; /* sizeof(struct xvnode) */ void *xv_vnode; /* address of real vnode */ u_long xv_flag; /* vnode vflags */ int xv_usecount; /* reference count of users */ int xv_writecount; /* reference count of writers */ int xv_holdcnt; /* page & buffer references */ u_long xv_id; /* capability identifier */ void *xv_mount; /* address of parent mount */ long xv_numoutput; /* num of writes in progress */ enum vtype xv_type; /* vnode type */ union { void *xvu_socket; /* unpcb, if VSOCK */ void *xvu_fifo; /* fifo, if VFIFO */ dev_t xvu_rdev; /* maj/min, if VBLK/VCHR */ struct { dev_t xvu_dev; /* device, if VDIR/VREG/VLNK */ ino_t xvu_ino; /* id, if VDIR/VREG/VLNK */ } xv_uns; } xv_un; }; #define xv_socket xv_un.xvu_socket #define xv_fifo xv_un.xvu_fifo #define xv_rdev xv_un.xvu_rdev #define xv_dev xv_un.xv_uns.xvu_dev #define xv_ino xv_un.xv_uns.xvu_ino /* We don't need to lock the knlist */ #define VN_KNLIST_EMPTY(vp) ((vp)->v_pollinfo == NULL || \ KNLIST_EMPTY(&(vp)->v_pollinfo->vpi_selinfo.si_note)) #define VN_KNOTE(vp, b, a) \ do { \ if (!VN_KNLIST_EMPTY(vp)) \ KNOTE(&vp->v_pollinfo->vpi_selinfo.si_note, (b), \ (a) | KNF_NOKQLOCK); \ } while (0) #define VN_KNOTE_LOCKED(vp, b) VN_KNOTE(vp, b, KNF_LISTLOCKED) #define VN_KNOTE_UNLOCKED(vp, b) VN_KNOTE(vp, b, 0) /* * Vnode flags. * VI flags are protected by interlock and live in v_iflag * VV flags are protected by the vnode lock and live in v_vflag * * VIRF_DOOMED is doubly protected by the interlock and vnode lock. Both * are required for writing but the status may be checked with either. */ #define VHOLD_NO_SMR (1<<29) /* Disable vhold_smr */ #define VHOLD_ALL_FLAGS (VHOLD_NO_SMR) #define VIRF_DOOMED 0x0001 /* This vnode is being recycled */ #define VIRF_PGREAD 0x0002 /* Direct reads from the page cache are permitted, never cleared once set */ #define VIRF_MOUNTPOINT 0x0004 /* This vnode is mounted on */ #define VIRF_TEXT_REF 0x0008 /* Executable mappings ref the vnode */ #define VI_UNUSED0 0x0001 /* unused */ #define VI_MOUNT 0x0002 /* Mount in progress */ #define VI_DOINGINACT 0x0004 /* VOP_INACTIVE is in progress */ #define VI_OWEINACT 0x0008 /* Need to call inactive */ #define VI_DEFINACT 0x0010 /* deferred inactive */ #define VI_FOPENING 0x0020 /* In open, with opening process having the first right to advlock file */ #define VV_ROOT 0x0001 /* root of its filesystem */ #define VV_ISTTY 0x0002 /* vnode represents a tty */ #define VV_NOSYNC 0x0004 /* unlinked, stop syncing */ #define VV_ETERNALDEV 0x0008 /* device that is never destroyed */ #define VV_CACHEDLABEL 0x0010 /* Vnode has valid cached MAC label */ #define VV_VMSIZEVNLOCK 0x0020 /* object size check requires vnode lock */ #define VV_COPYONWRITE 0x0040 /* vnode is doing copy-on-write */ #define VV_SYSTEM 0x0080 /* vnode being used by kernel */ #define VV_PROCDEP 0x0100 /* vnode is process dependent */ #define VV_UNLINKED 0x0200 /* unlinked but stil open directory */ #define VV_DELETED 0x0400 /* should be removed */ #define VV_MD 0x0800 /* vnode backs the md device */ #define VV_FORCEINSMQ 0x1000 /* force the insmntque to succeed */ #define VV_READLINK 0x2000 /* fdescfs linux vnode */ #define VV_UNREF 0x4000 /* vunref, do not drop lock in inactive() */ #define VMP_LAZYLIST 0x0001 /* Vnode is on mnt's lazy list */ /* * Vnode attributes. A field value of VNOVAL represents a field whose value * is unavailable (getattr) or which is not to be changed (setattr). */ struct vattr { enum vtype va_type; /* vnode type (for create) */ u_short va_mode; /* files access mode and type */ u_short va_padding0; uid_t va_uid; /* owner user id */ gid_t va_gid; /* owner group id */ nlink_t va_nlink; /* number of references to file */ dev_t va_fsid; /* filesystem id */ ino_t va_fileid; /* file id */ u_quad_t va_size; /* file size in bytes */ long va_blocksize; /* blocksize preferred for i/o */ struct timespec va_atime; /* time of last access */ struct timespec va_mtime; /* time of last modification */ struct timespec va_ctime; /* time file changed */ struct timespec va_birthtime; /* time file created */ u_long va_gen; /* generation number of file */ u_long va_flags; /* flags defined for file */ dev_t va_rdev; /* device the special file represents */ u_quad_t va_bytes; /* bytes of disk space held by file */ u_quad_t va_filerev; /* file modification number */ u_int va_vaflags; /* operations flags, see below */ long va_spare; /* remain quad aligned */ }; /* * Flags for va_vaflags. */ #define VA_UTIMES_NULL 0x01 /* utimes argument was NULL */ #define VA_EXCLUSIVE 0x02 /* exclusive create request */ #define VA_SYNC 0x04 /* O_SYNC truncation */ /* * Flags for ioflag. (high 16 bits used to ask for read-ahead and * help with write clustering) * NB: IO_NDELAY and IO_DIRECT are linked to fcntl.h */ #define IO_UNIT 0x0001 /* do I/O as atomic unit */ #define IO_APPEND 0x0002 /* append write to end */ #define IO_NDELAY 0x0004 /* FNDELAY flag set in file table */ #define IO_NODELOCKED 0x0008 /* underlying node already locked */ #define IO_ASYNC 0x0010 /* bawrite rather then bdwrite */ #define IO_VMIO 0x0020 /* data already in VMIO space */ #define IO_INVAL 0x0040 /* invalidate after I/O */ #define IO_SYNC 0x0080 /* do I/O synchronously */ #define IO_DIRECT 0x0100 /* attempt to bypass buffer cache */ #define IO_NOREUSE 0x0200 /* VMIO data won't be reused */ #define IO_EXT 0x0400 /* operate on external attributes */ #define IO_NORMAL 0x0800 /* operate on regular data */ #define IO_NOMACCHECK 0x1000 /* MAC checks unnecessary */ #define IO_BUFLOCKED 0x2000 /* ffs flag; indir buf is locked */ #define IO_RANGELOCKED 0x4000 /* range locked */ #define IO_DATASYNC 0x8000 /* do only data I/O synchronously */ #define IO_SEQMAX 0x7F /* seq heuristic max value */ #define IO_SEQSHIFT 16 /* seq heuristic in upper 16 bits */ /* * Flags for accmode_t. */ #define VEXEC 000000000100 /* execute/search permission */ #define VWRITE 000000000200 /* write permission */ #define VREAD 000000000400 /* read permission */ #define VADMIN 000000010000 /* being the file owner */ #define VAPPEND 000000040000 /* permission to write/append */ /* * VEXPLICIT_DENY makes VOP_ACCESSX(9) return EPERM or EACCES only * if permission was denied explicitly, by a "deny" rule in NFSv4 ACL, * and 0 otherwise. This never happens with ordinary unix access rights * or POSIX.1e ACLs. Obviously, VEXPLICIT_DENY must be OR-ed with * some other V* constant. */ #define VEXPLICIT_DENY 000000100000 #define VREAD_NAMED_ATTRS 000000200000 /* not used */ #define VWRITE_NAMED_ATTRS 000000400000 /* not used */ #define VDELETE_CHILD 000001000000 #define VREAD_ATTRIBUTES 000002000000 /* permission to stat(2) */ #define VWRITE_ATTRIBUTES 000004000000 /* change {m,c,a}time */ #define VDELETE 000010000000 #define VREAD_ACL 000020000000 /* read ACL and file mode */ #define VWRITE_ACL 000040000000 /* change ACL and/or file mode */ #define VWRITE_OWNER 000100000000 /* change file owner */ #define VSYNCHRONIZE 000200000000 /* not used */ #define VCREAT 000400000000 /* creating new file */ #define VVERIFY 001000000000 /* verification required */ /* * Permissions that were traditionally granted only to the file owner. */ #define VADMIN_PERMS (VADMIN | VWRITE_ATTRIBUTES | VWRITE_ACL | \ VWRITE_OWNER) /* * Permissions that were traditionally granted to everyone. */ #define VSTAT_PERMS (VREAD_ATTRIBUTES | VREAD_ACL) /* * Permissions that allow to change the state of the file in any way. */ #define VMODIFY_PERMS (VWRITE | VAPPEND | VADMIN_PERMS | VDELETE_CHILD | \ VDELETE) /* * Token indicating no attribute value yet assigned. */ #define VNOVAL (-1) /* * LK_TIMELOCK timeout for vnode locks (used mainly by the pageout daemon) */ #define VLKTIMEOUT (hz / 20 + 1) #ifdef _KERNEL #ifdef MALLOC_DECLARE MALLOC_DECLARE(M_VNODE); #endif extern u_int ncsizefactor; extern const u_int io_hold_cnt; /* * Convert between vnode types and inode formats (since POSIX.1 * defines mode word of stat structure in terms of inode formats). */ extern enum vtype iftovt_tab[]; extern int vttoif_tab[]; #define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12]) #define VTTOIF(indx) (vttoif_tab[(int)(indx)]) #define MAKEIMODE(indx, mode) (int)(VTTOIF(indx) | (mode)) /* * Flags to various vnode functions. */ #define SKIPSYSTEM 0x0001 /* vflush: skip vnodes marked VSYSTEM */ #define FORCECLOSE 0x0002 /* vflush: force file closure */ #define WRITECLOSE 0x0004 /* vflush: only close writable files */ #define EARLYFLUSH 0x0008 /* vflush: early call for ffs_flushfiles */ #define V_SAVE 0x0001 /* vinvalbuf: sync file first */ #define V_ALT 0x0002 /* vinvalbuf: invalidate only alternate bufs */ #define V_NORMAL 0x0004 /* vinvalbuf: invalidate only regular bufs */ #define V_CLEANONLY 0x0008 /* vinvalbuf: invalidate only clean bufs */ #define V_VMIO 0x0010 /* vinvalbuf: called during pageout */ #define V_ALLOWCLEAN 0x0020 /* vinvalbuf: allow clean buffers after flush */ #define REVOKEALL 0x0001 /* vop_revoke: revoke all aliases */ #define V_WAIT 0x0001 /* vn_start_write: sleep for suspend */ #define V_NOWAIT 0x0002 /* vn_start_write: don't sleep for suspend */ #define V_XSLEEP 0x0004 /* vn_start_write: just return after sleep */ +#define V_PCATCH 0x0008 /* vn_start_write: make the sleep interruptible */ +#define V_VALID_FLAGS (V_WAIT | V_NOWAIT | V_XSLEEP | V_PCATCH) #define VR_START_WRITE 0x0001 /* vfs_write_resume: start write atomically */ #define VR_NO_SUSPCLR 0x0002 /* vfs_write_resume: do not clear suspension */ #define VS_SKIP_UNMOUNT 0x0001 /* vfs_write_suspend: fail if the filesystem is being unmounted */ #define VREF(vp) vref(vp) #ifdef DIAGNOSTIC #define VATTR_NULL(vap) vattr_null(vap) #else #define VATTR_NULL(vap) (*(vap) = va_null) /* initialize a vattr */ #endif /* DIAGNOSTIC */ #define NULLVP ((struct vnode *)NULL) /* * Global vnode data. */ extern struct vnode *rootvnode; /* root (i.e. "/") vnode */ extern struct mount *rootdevmp; /* "/dev" mount */ extern u_long desiredvnodes; /* number of vnodes desired */ extern struct uma_zone *namei_zone; extern struct vattr va_null; /* predefined null vattr structure */ extern u_int vn_lock_pair_pause_max; #define VI_LOCK(vp) mtx_lock(&(vp)->v_interlock) #define VI_LOCK_FLAGS(vp, flags) mtx_lock_flags(&(vp)->v_interlock, (flags)) #define VI_TRYLOCK(vp) mtx_trylock(&(vp)->v_interlock) #define VI_UNLOCK(vp) mtx_unlock(&(vp)->v_interlock) #define VI_MTX(vp) (&(vp)->v_interlock) #define VN_LOCK_AREC(vp) lockallowrecurse((vp)->v_vnlock) #define VN_LOCK_ASHARE(vp) lockallowshare((vp)->v_vnlock) #define VN_LOCK_DSHARE(vp) lockdisableshare((vp)->v_vnlock) #endif /* _KERNEL */ /* * Mods for extensibility. */ /* * Flags for vdesc_flags: */ #define VDESC_MAX_VPS 16 /* Low order 16 flag bits are reserved for willrele flags for vp arguments. */ #define VDESC_VP0_WILLRELE 0x0001 #define VDESC_VP1_WILLRELE 0x0002 #define VDESC_VP2_WILLRELE 0x0004 #define VDESC_VP3_WILLRELE 0x0008 /* * A generic structure. * This can be used by bypass routines to identify generic arguments. */ struct vop_generic_args { struct vnodeop_desc *a_desc; /* other random data follows, presumably */ }; typedef int vop_bypass_t(struct vop_generic_args *); /* * VDESC_NO_OFFSET is used to identify the end of the offset list * and in places where no such field exists. */ #define VDESC_NO_OFFSET -1 /* * This structure describes the vnode operation taking place. */ struct vnodeop_desc { char *vdesc_name; /* a readable name for debugging */ int vdesc_flags; /* VDESC_* flags */ int vdesc_vop_offset; vop_bypass_t *vdesc_call; /* Function to call */ /* * These ops are used by bypass routines to map and locate arguments. * Creds and procs are not needed in bypass routines, but sometimes * they are useful to (for example) transport layers. * Nameidata is useful because it has a cred in it. */ int *vdesc_vp_offsets; /* list ended by VDESC_NO_OFFSET */ int vdesc_vpp_offset; /* return vpp location */ int vdesc_cred_offset; /* cred location, if any */ int vdesc_thread_offset; /* thread location, if any */ int vdesc_componentname_offset; /* if any */ }; #ifdef _KERNEL /* * A list of all the operation descs. */ extern struct vnodeop_desc *vnodeop_descs[]; #define VOPARG_OFFSETOF(s_type, field) __offsetof(s_type, field) #define VOPARG_OFFSETTO(s_type, s_offset, struct_p) \ ((s_type)(((char*)(struct_p)) + (s_offset))) #ifdef DEBUG_VFS_LOCKS /* * Support code to aid in debugging VFS locking problems. Not totally * reliable since if the thread sleeps between changing the lock * state and checking it with the assert, some other thread could * change the state. They are good enough for debugging a single * filesystem using a single-threaded test. Note that the unreliability is * limited to false negatives; efforts were made to ensure that false * positives cannot occur. */ void assert_vi_locked(struct vnode *vp, const char *str); void assert_vi_unlocked(struct vnode *vp, const char *str); void assert_vop_elocked(struct vnode *vp, const char *str); void assert_vop_locked(struct vnode *vp, const char *str); void assert_vop_unlocked(struct vnode *vp, const char *str); #define ASSERT_VI_LOCKED(vp, str) assert_vi_locked((vp), (str)) #define ASSERT_VI_UNLOCKED(vp, str) assert_vi_unlocked((vp), (str)) #define ASSERT_VOP_ELOCKED(vp, str) assert_vop_elocked((vp), (str)) #define ASSERT_VOP_LOCKED(vp, str) assert_vop_locked((vp), (str)) #define ASSERT_VOP_UNLOCKED(vp, str) assert_vop_unlocked((vp), (str)) #define ASSERT_VOP_IN_SEQC(vp) do { \ struct vnode *_vp = (vp); \ \ VNPASS(seqc_in_modify(_vp->v_seqc), _vp); \ } while (0) #define ASSERT_VOP_NOT_IN_SEQC(vp) do { \ struct vnode *_vp = (vp); \ \ VNPASS(!seqc_in_modify(_vp->v_seqc), _vp); \ } while (0) #else /* !DEBUG_VFS_LOCKS */ #define ASSERT_VI_LOCKED(vp, str) ((void)0) #define ASSERT_VI_UNLOCKED(vp, str) ((void)0) #define ASSERT_VOP_ELOCKED(vp, str) ((void)0) #define ASSERT_VOP_LOCKED(vp, str) ((void)0) #define ASSERT_VOP_UNLOCKED(vp, str) ((void)0) #define ASSERT_VOP_IN_SEQC(vp) ((void)0) #define ASSERT_VOP_NOT_IN_SEQC(vp) ((void)0) #endif /* DEBUG_VFS_LOCKS */ /* * This call works for vnodes in the kernel. */ #define VCALL(c) ((c)->a_desc->vdesc_call(c)) #define DOINGASYNC(vp) \ (((vp)->v_mount->mnt_kern_flag & MNTK_ASYNC) != 0 && \ ((curthread->td_pflags & TDP_SYNCIO) == 0)) /* * VMIO support inline */ extern int vmiodirenable; static __inline int vn_canvmio(struct vnode *vp) { if (vp && (vp->v_type == VREG || (vmiodirenable && vp->v_type == VDIR))) return(TRUE); return(FALSE); } /* * Finally, include the default set of vnode operations. */ typedef void vop_getpages_iodone_t(void *, vm_page_t *, int, int); #include "vnode_if.h" /* vn_open_flags */ #define VN_OPEN_NOAUDIT 0x00000001 #define VN_OPEN_NOCAPCHECK 0x00000002 #define VN_OPEN_NAMECACHE 0x00000004 #define VN_OPEN_INVFS 0x00000008 #define VN_OPEN_WANTIOCTLCAPS 0x00000010 /* copy_file_range kernel flags */ #define COPY_FILE_RANGE_KFLAGS 0xff000000 #define COPY_FILE_RANGE_TIMEO1SEC 0x01000000 /* Return after 1sec. */ /* * Public vnode manipulation functions. */ struct componentname; struct file; struct mount; struct nameidata; struct ostat; struct freebsd11_stat; struct thread; struct proc; struct stat; struct nstat; struct ucred; struct uio; struct vattr; struct vfsops; struct vnode; typedef int (*vn_get_ino_t)(struct mount *, void *, int, struct vnode **); int bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn); /* cache_* may belong in namei.h. */ void cache_changesize(u_long newhashsize); #define VFS_CACHE_DROPOLD 0x1 void cache_enter_time_flags(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp, int flags); #define cache_enter(dvp, vp, cnp) \ cache_enter_time(dvp, vp, cnp, NULL, NULL) void cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp); int cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp); void cache_vnode_init(struct vnode *vp); void cache_purge(struct vnode *vp); void cache_purge_vgone(struct vnode *vp); void cache_purge_negative(struct vnode *vp); void cache_purgevfs(struct mount *mp); char *cache_symlink_alloc(size_t size, int flags); void cache_symlink_free(char *string, size_t size); int cache_symlink_resolve(struct cache_fpl *fpl, const char *string, size_t len); void cache_vop_rename(struct vnode *fdvp, struct vnode *fvp, struct vnode *tdvp, struct vnode *tvp, struct componentname *fcnp, struct componentname *tcnp); void cache_vop_rmdir(struct vnode *dvp, struct vnode *vp); #ifdef INVARIANTS void cache_validate(struct vnode *dvp, struct vnode *vp, struct componentname *cnp); #else static inline void cache_validate(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) { } #endif void cache_fast_lookup_enabled_recalc(void); int change_dir(struct vnode *vp, struct thread *td); void cvtstat(struct stat *st, struct ostat *ost); int freebsd11_cvtnstat(struct stat *sb, struct nstat *nsb); int freebsd11_cvtstat(struct stat *st, struct freebsd11_stat *ost); int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp); void getnewvnode_reserve(void); void getnewvnode_drop_reserve(void); int insmntque(struct vnode *vp, struct mount *mp); int insmntque1(struct vnode *vp, struct mount *mp); u_quad_t init_va_filerev(void); int speedup_syncer(void); int vn_vptocnp(struct vnode **vp, char *buf, size_t *buflen); int vn_getcwd(char *buf, char **retbuf, size_t *buflen); int vn_fullpath(struct vnode *vp, char **retbuf, char **freebuf); int vn_fullpath_global(struct vnode *vp, char **retbuf, char **freebuf); int vn_fullpath_hardlink(struct vnode *vp, struct vnode *dvp, const char *hdrl_name, size_t hrdl_name_length, char **retbuf, char **freebuf, size_t *buflen); struct vnode * vn_dir_dd_ino(struct vnode *vp); int vn_commname(struct vnode *vn, char *buf, u_int buflen); int vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, u_int pathlen); int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred); int vaccess_vexec_smr(mode_t file_mode, uid_t file_uid, gid_t file_gid, struct ucred *cred); int vaccess_acl_nfs4(enum vtype type, uid_t file_uid, gid_t file_gid, struct acl *aclp, accmode_t accmode, struct ucred *cred); int vaccess_acl_posix1e(enum vtype type, uid_t file_uid, gid_t file_gid, struct acl *acl, accmode_t accmode, struct ucred *cred); void vattr_null(struct vattr *vap); void vlazy(struct vnode *); void vdrop(struct vnode *); void vdropl(struct vnode *); int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td); int vget(struct vnode *vp, int flags); enum vgetstate vget_prep_smr(struct vnode *vp); enum vgetstate vget_prep(struct vnode *vp); int vget_finish(struct vnode *vp, int flags, enum vgetstate vs); void vget_finish_ref(struct vnode *vp, enum vgetstate vs); void vget_abort(struct vnode *vp, enum vgetstate vs); void vgone(struct vnode *vp); void vhold(struct vnode *); void vholdnz(struct vnode *); bool vhold_smr(struct vnode *); int vinactive(struct vnode *vp); int vinvalbuf(struct vnode *vp, int save, int slpflag, int slptimeo); int vtruncbuf(struct vnode *vp, off_t length, int blksize); void v_inval_buf_range(struct vnode *vp, daddr_t startlbn, daddr_t endlbn, int blksize); void vunref(struct vnode *); void vn_printf(struct vnode *vp, const char *fmt, ...) __printflike(2,3); int vrecycle(struct vnode *vp); int vrecyclel(struct vnode *vp); int vn_bmap_seekhole_locked(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred); int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred); int vn_close(struct vnode *vp, int flags, struct ucred *file_cred, struct thread *td); int vn_copy_file_range(struct vnode *invp, off_t *inoffp, struct vnode *outvp, off_t *outoffp, size_t *lenp, unsigned int flags, struct ucred *incred, struct ucred *outcred, struct thread *fsize_td); int vn_deallocate(struct vnode *vp, off_t *offset, off_t *length, int flags, int ioflg, struct ucred *active_cred, struct ucred *file_cred); void vn_finished_write(struct mount *mp); void vn_finished_secondary_write(struct mount *mp); int vn_fsync_buf(struct vnode *vp, int waitfor); int vn_generic_copy_file_range(struct vnode *invp, off_t *inoffp, struct vnode *outvp, off_t *outoffp, size_t *lenp, unsigned int flags, struct ucred *incred, struct ucred *outcred, struct thread *fsize_td); int vn_need_pageq_flush(struct vnode *vp); bool vn_isdisk_error(struct vnode *vp, int *errp); bool vn_isdisk(struct vnode *vp); int _vn_lock(struct vnode *vp, int flags, const char *file, int line); #define vn_lock(vp, flags) _vn_lock(vp, flags, __FILE__, __LINE__) void vn_lock_pair(struct vnode *vp1, bool vp1_locked, struct vnode *vp2, bool vp2_locked); int vn_open(struct nameidata *ndp, int *flagp, int cmode, struct file *fp); int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp); int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp); void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end); void vn_pages_remove_valid(struct vnode *vp, vm_pindex_t start, vm_pindex_t end); int vn_pollrecord(struct vnode *vp, struct thread *p, int events); int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td); int vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, size_t *aresid, struct thread *td); int vn_read_from_obj(struct vnode *vp, struct uio *uio); int vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, struct thread *td); int vn_start_write(struct vnode *vp, struct mount **mpp, int flags); int vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags); int vn_truncate_locked(struct vnode *vp, off_t length, bool sync, struct ucred *cred); int vn_writechk(struct vnode *vp); int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td); int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td); int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td); int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp); int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp); int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td); int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio); int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio); void vn_seqc_write_begin_locked(struct vnode *vp); void vn_seqc_write_begin(struct vnode *vp); void vn_seqc_write_end_locked(struct vnode *vp); void vn_seqc_write_end(struct vnode *vp); #define vn_seqc_read_any(vp) seqc_read_any(&(vp)->v_seqc) #define vn_seqc_read_notmodify(vp) seqc_read_notmodify(&(vp)->v_seqc) #define vn_seqc_consistent(vp, seq) seqc_consistent(&(vp)->v_seqc, seq) #define vn_rangelock_unlock(vp, cookie) \ rangelock_unlock(&(vp)->v_rl, (cookie), VI_MTX(vp)) #define vn_rangelock_unlock_range(vp, cookie, start, end) \ rangelock_unlock_range(&(vp)->v_rl, (cookie), (start), (end), \ VI_MTX(vp)) #define vn_rangelock_rlock(vp, start, end) \ rangelock_rlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) #define vn_rangelock_tryrlock(vp, start, end) \ rangelock_tryrlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) #define vn_rangelock_wlock(vp, start, end) \ rangelock_wlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) #define vn_rangelock_trywlock(vp, start, end) \ rangelock_trywlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) #define vn_irflag_read(vp) atomic_load_short(&(vp)->v_irflag) void vn_irflag_set_locked(struct vnode *vp, short toset); void vn_irflag_set(struct vnode *vp, short toset); void vn_irflag_set_cond_locked(struct vnode *vp, short toset); void vn_irflag_set_cond(struct vnode *vp, short toset); void vn_irflag_unset_locked(struct vnode *vp, short tounset); void vn_irflag_unset(struct vnode *vp, short tounset); int vfs_cache_lookup(struct vop_lookup_args *ap); int vfs_cache_root(struct mount *mp, int flags, struct vnode **vpp); void vfs_timestamp(struct timespec *); void vfs_write_resume(struct mount *mp, int flags); int vfs_write_suspend(struct mount *mp, int flags); int vfs_write_suspend_umnt(struct mount *mp); struct vnode *vnlru_alloc_marker(void); void vnlru_free_marker(struct vnode *); void vnlru_free_vfsops(int, struct vfsops *, struct vnode *); int vop_stdbmap(struct vop_bmap_args *); int vop_stdfdatasync_buf(struct vop_fdatasync_args *); int vop_stdfsync(struct vop_fsync_args *); int vop_stdgetwritemount(struct vop_getwritemount_args *); int vop_stdgetpages(struct vop_getpages_args *); int vop_stdinactive(struct vop_inactive_args *); int vop_stdioctl(struct vop_ioctl_args *); int vop_stdneed_inactive(struct vop_need_inactive_args *); int vop_stdkqfilter(struct vop_kqfilter_args *); int vop_stdlock(struct vop_lock1_args *); int vop_stdunlock(struct vop_unlock_args *); int vop_stdislocked(struct vop_islocked_args *); int vop_lock(struct vop_lock1_args *); int vop_unlock(struct vop_unlock_args *); int vop_islocked(struct vop_islocked_args *); int vop_stdputpages(struct vop_putpages_args *); int vop_nopoll(struct vop_poll_args *); int vop_stdaccess(struct vop_access_args *ap); int vop_stdaccessx(struct vop_accessx_args *ap); int vop_stdadvise(struct vop_advise_args *ap); int vop_stdadvlock(struct vop_advlock_args *ap); int vop_stdadvlockasync(struct vop_advlockasync_args *ap); int vop_stdadvlockpurge(struct vop_advlockpurge_args *ap); int vop_stdallocate(struct vop_allocate_args *ap); int vop_stddeallocate(struct vop_deallocate_args *ap); int vop_stdset_text(struct vop_set_text_args *ap); int vop_stdpathconf(struct vop_pathconf_args *); int vop_stdpoll(struct vop_poll_args *); int vop_stdvptocnp(struct vop_vptocnp_args *ap); int vop_stdvptofh(struct vop_vptofh_args *ap); int vop_stdunp_bind(struct vop_unp_bind_args *ap); int vop_stdunp_connect(struct vop_unp_connect_args *ap); int vop_stdunp_detach(struct vop_unp_detach_args *ap); int vop_stdadd_writecount_nomsync(struct vop_add_writecount_args *ap); int vop_eopnotsupp(struct vop_generic_args *ap); int vop_ebadf(struct vop_generic_args *ap); int vop_einval(struct vop_generic_args *ap); int vop_enoent(struct vop_generic_args *ap); int vop_enotty(struct vop_generic_args *ap); int vop_eagain(struct vop_generic_args *ap); int vop_null(struct vop_generic_args *ap); int vop_panic(struct vop_generic_args *ap); int dead_poll(struct vop_poll_args *ap); int dead_read(struct vop_read_args *ap); int dead_write(struct vop_write_args *ap); /* These are called from within the actual VOPS. */ void vop_close_post(void *a, int rc); void vop_create_pre(void *a); void vop_create_post(void *a, int rc); void vop_whiteout_pre(void *a); void vop_whiteout_post(void *a, int rc); void vop_deleteextattr_pre(void *a); void vop_deleteextattr_post(void *a, int rc); void vop_link_pre(void *a); void vop_link_post(void *a, int rc); void vop_lookup_post(void *a, int rc); void vop_lookup_pre(void *a); void vop_mkdir_pre(void *a); void vop_mkdir_post(void *a, int rc); void vop_mknod_pre(void *a); void vop_mknod_post(void *a, int rc); void vop_open_post(void *a, int rc); void vop_read_post(void *a, int rc); void vop_read_pgcache_post(void *ap, int rc); void vop_readdir_post(void *a, int rc); void vop_reclaim_post(void *a, int rc); void vop_remove_pre(void *a); void vop_remove_post(void *a, int rc); void vop_rename_post(void *a, int rc); void vop_rename_pre(void *a); void vop_rmdir_pre(void *a); void vop_rmdir_post(void *a, int rc); void vop_setattr_pre(void *a); void vop_setattr_post(void *a, int rc); void vop_setacl_pre(void *a); void vop_setacl_post(void *a, int rc); void vop_setextattr_pre(void *a); void vop_setextattr_post(void *a, int rc); void vop_symlink_pre(void *a); void vop_symlink_post(void *a, int rc); int vop_sigdefer(struct vop_vector *vop, struct vop_generic_args *a); #ifdef DEBUG_VFS_LOCKS void vop_fdatasync_debugpre(void *a); void vop_fdatasync_debugpost(void *a, int rc); void vop_fplookup_vexec_debugpre(void *a); void vop_fplookup_vexec_debugpost(void *a, int rc); void vop_fplookup_symlink_debugpre(void *a); void vop_fplookup_symlink_debugpost(void *a, int rc); void vop_fsync_debugpre(void *a); void vop_fsync_debugpost(void *a, int rc); void vop_strategy_debugpre(void *a); void vop_lock_debugpre(void *a); void vop_lock_debugpost(void *a, int rc); void vop_unlock_debugpre(void *a); void vop_need_inactive_debugpre(void *a); void vop_need_inactive_debugpost(void *a, int rc); void vop_mkdir_debugpost(void *a, int rc); #else #define vop_fdatasync_debugpre(x) do { } while (0) #define vop_fdatasync_debugpost(x, y) do { } while (0) #define vop_fplookup_vexec_debugpre(x) do { } while (0) #define vop_fplookup_vexec_debugpost(x, y) do { } while (0) #define vop_fplookup_symlink_debugpre(x) do { } while (0) #define vop_fplookup_symlink_debugpost(x, y) do { } while (0) #define vop_fsync_debugpre(x) do { } while (0) #define vop_fsync_debugpost(x, y) do { } while (0) #define vop_strategy_debugpre(x) do { } while (0) #define vop_lock_debugpre(x) do { } while (0) #define vop_lock_debugpost(x, y) do { } while (0) #define vop_unlock_debugpre(x) do { } while (0) #define vop_need_inactive_debugpre(x) do { } while (0) #define vop_need_inactive_debugpost(x, y) do { } while (0) #define vop_mkdir_debugpost(x, y) do { } while (0) #endif void vop_rename_fail(struct vop_rename_args *ap); #define vop_stat_helper_pre(ap) ({ \ struct vop_stat_args *_ap = (ap); \ int _error; \ AUDIT_ARG_VNODE1(ap->a_vp); \ _error = mac_vnode_check_stat(_ap->a_active_cred, _ap->a_file_cred, _ap->a_vp);\ if (__predict_true(_error == 0)) { \ ap->a_sb->st_padding0 = 0; \ ap->a_sb->st_padding1 = 0; \ bzero(_ap->a_sb->st_spare, sizeof(_ap->a_sb->st_spare)); \ } \ _error; \ }) #define vop_stat_helper_post(ap, error) ({ \ struct vop_stat_args *_ap = (ap); \ int _error = (error); \ if (priv_check_cred_vfs_generation(_ap->a_active_cred)) \ _ap->a_sb->st_gen = 0; \ _error; \ }) #define VOP_WRITE_PRE(ap) \ struct vattr va; \ int error; \ off_t osize, ooffset, noffset; \ \ osize = ooffset = noffset = 0; \ if (!VN_KNLIST_EMPTY((ap)->a_vp)) { \ error = VOP_GETATTR((ap)->a_vp, &va, (ap)->a_cred); \ if (error) \ return (error); \ ooffset = (ap)->a_uio->uio_offset; \ osize = (off_t)va.va_size; \ } #define VOP_WRITE_POST(ap, ret) \ noffset = (ap)->a_uio->uio_offset; \ if (noffset > ooffset && !VN_KNLIST_EMPTY((ap)->a_vp)) { \ VFS_KNOTE_LOCKED((ap)->a_vp, NOTE_WRITE \ | (noffset > osize ? NOTE_EXTEND : 0)); \ } #define VOP_LOCK(vp, flags) VOP_LOCK1(vp, flags, __FILE__, __LINE__) #ifdef INVARIANTS #define VOP_ADD_WRITECOUNT_CHECKED(vp, cnt) \ do { \ int error_; \ \ error_ = VOP_ADD_WRITECOUNT((vp), (cnt)); \ VNASSERT(error_ == 0, (vp), ("VOP_ADD_WRITECOUNT returned %d", \ error_)); \ } while (0) #define VOP_SET_TEXT_CHECKED(vp) \ do { \ int error_; \ \ error_ = VOP_SET_TEXT((vp)); \ VNASSERT(error_ == 0, (vp), ("VOP_SET_TEXT returned %d", \ error_)); \ } while (0) #define VOP_UNSET_TEXT_CHECKED(vp) \ do { \ int error_; \ \ error_ = VOP_UNSET_TEXT((vp)); \ VNASSERT(error_ == 0, (vp), ("VOP_UNSET_TEXT returned %d", \ error_)); \ } while (0) #else #define VOP_ADD_WRITECOUNT_CHECKED(vp, cnt) VOP_ADD_WRITECOUNT((vp), (cnt)) #define VOP_SET_TEXT_CHECKED(vp) VOP_SET_TEXT((vp)) #define VOP_UNSET_TEXT_CHECKED(vp) VOP_UNSET_TEXT((vp)) #endif #define VN_IS_DOOMED(vp) __predict_false((vn_irflag_read(vp) & VIRF_DOOMED) != 0) void vput(struct vnode *vp); void vrele(struct vnode *vp); void vref(struct vnode *vp); void vrefact(struct vnode *vp); void v_addpollinfo(struct vnode *vp); static __inline int vrefcnt(struct vnode *vp) { return (vp->v_usecount); } #define vholdl(vp) do { \ ASSERT_VI_LOCKED(vp, __func__); \ vhold(vp); \ } while (0) #define vrefl(vp) do { \ ASSERT_VI_LOCKED(vp, __func__); \ vref(vp); \ } while (0) int vnode_create_vobject(struct vnode *vp, off_t size, struct thread *td); void vnode_destroy_vobject(struct vnode *vp); extern struct vop_vector fifo_specops; extern struct vop_vector dead_vnodeops; extern struct vop_vector default_vnodeops; #define VOP_PANIC ((void*)(uintptr_t)vop_panic) #define VOP_NULL ((void*)(uintptr_t)vop_null) #define VOP_EBADF ((void*)(uintptr_t)vop_ebadf) #define VOP_ENOTTY ((void*)(uintptr_t)vop_enotty) #define VOP_EINVAL ((void*)(uintptr_t)vop_einval) #define VOP_ENOENT ((void*)(uintptr_t)vop_enoent) #define VOP_EOPNOTSUPP ((void*)(uintptr_t)vop_eopnotsupp) #define VOP_EAGAIN ((void*)(uintptr_t)vop_eagain) /* fifo_vnops.c */ int fifo_printinfo(struct vnode *); /* vfs_hash.c */ typedef int vfs_hash_cmp_t(struct vnode *vp, void *arg); void vfs_hash_changesize(u_long newhashsize); int vfs_hash_get(const struct mount *mp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); u_int vfs_hash_index(struct vnode *vp); int vfs_hash_insert(struct vnode *vp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); void vfs_hash_ref(const struct mount *mp, u_int hash, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); void vfs_hash_rehash(struct vnode *vp, u_int hash); void vfs_hash_remove(struct vnode *vp); int vfs_kqfilter(struct vop_kqfilter_args *); struct dirent; int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off); int vfs_emptydir(struct vnode *vp); int vfs_unixify_accmode(accmode_t *accmode); void vfs_unp_reclaim(struct vnode *vp); int setfmode(struct thread *td, struct ucred *cred, struct vnode *vp, int mode); int setfown(struct thread *td, struct ucred *cred, struct vnode *vp, uid_t uid, gid_t gid); int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td); int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td); void vn_fsid(struct vnode *vp, struct vattr *va); int vn_dir_check_exec(struct vnode *vp, struct componentname *cnp); int vn_lktype_write(struct mount *mp, struct vnode *vp); #define VOP_UNLOCK_FLAGS(vp, flags) ({ \ struct vnode *_vp = (vp); \ int _flags = (flags); \ int _error; \ \ if ((_flags & ~(LK_INTERLOCK | LK_RELEASE)) != 0) \ panic("%s: unsupported flags %x\n", __func__, flags); \ _error = VOP_UNLOCK(_vp); \ if (_flags & LK_INTERLOCK) \ VI_UNLOCK(_vp); \ _error; \ }) #include #define VFS_VOP_VECTOR_REGISTER(vnodeops) \ SYSINIT(vfs_vector_##vnodeops##_f, SI_SUB_VFS, SI_ORDER_ANY, \ vfs_vector_op_register, &vnodeops) #define VFS_SMR_DECLARE \ extern smr_t vfs_smr #define VFS_SMR() vfs_smr #define vfs_smr_enter() smr_enter(VFS_SMR()) #define vfs_smr_exit() smr_exit(VFS_SMR()) #define vfs_smr_synchronize() smr_synchronize(VFS_SMR()) #define vfs_smr_entered_load(ptr) smr_entered_load((ptr), VFS_SMR()) #define VFS_SMR_ASSERT_ENTERED() SMR_ASSERT_ENTERED(VFS_SMR()) #define VFS_SMR_ASSERT_NOT_ENTERED() SMR_ASSERT_NOT_ENTERED(VFS_SMR()) #define VFS_SMR_ZONE_SET(zone) uma_zone_set_smr((zone), VFS_SMR()) #define vn_load_v_data_smr(vp) ({ \ struct vnode *_vp = (vp); \ \ VFS_SMR_ASSERT_ENTERED(); \ atomic_load_consume_ptr(&(_vp)->v_data);\ }) #endif /* _KERNEL */ #endif /* !_SYS_VNODE_H_ */