diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c index a49070b6060d..1d90f76f9cd6 100644 --- a/sys/kern/vfs_vnops.c +++ b/sys/kern/vfs_vnops.c @@ -1,3827 +1,3837 @@ /*- * 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 | 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 | 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 | 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 | 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 vnode *vp; struct fiobmap2_arg *bmarg; off_t size; int error; vp = fp->f_vnode; switch (vp->v_type) { case VDIR: case VREG: switch (com) { case FIONREAD: error = vn_getsize(vp, &size, active_cred); if (error == 0) *(int *)data = 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 & 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 & 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); } static void vn_send_sigxfsz(struct proc *p) { PROC_LOCK(p); kern_psignal(p, SIGXFSZ); PROC_UNLOCK(p); } int vn_rlimit_trunc(u_quad_t size, struct thread *td) { if (size <= lim_cur(td, RLIMIT_FSIZE)) return (0); vn_send_sigxfsz(td->td_proc); return (EFBIG); } static int vn_rlimit_fsizex1(const struct vnode *vp, struct uio *uio, off_t maxfsz, bool adj, struct thread *td) { off_t lim; bool ktr_write; if (vp->v_type != VREG) return (0); /* * Handle file system maximum file size. */ if (maxfsz != 0 && uio->uio_offset + uio->uio_resid > maxfsz) { if (!adj || uio->uio_offset >= maxfsz) return (EFBIG); uio->uio_resid = maxfsz - uio->uio_offset; } /* * This is kernel write (e.g. vnode_pager) or accounting * write, ignore limit. */ if (td == NULL || (td->td_pflags2 & TDP2_ACCT) != 0) return (0); /* * Calculate file size limit. */ ktr_write = (td->td_pflags & TDP_INKTRACE) != 0; lim = __predict_false(ktr_write) ? td->td_ktr_io_lim : lim_cur(td, RLIMIT_FSIZE); /* * Is the limit reached? */ if (__predict_true((uoff_t)uio->uio_offset + uio->uio_resid <= lim)) return (0); /* * Prepared filesystems can handle writes truncated to the * file size limit. */ if (adj && (uoff_t)uio->uio_offset < lim) { uio->uio_resid = lim - (uoff_t)uio->uio_offset; return (0); } if (!ktr_write || ktr_filesize_limit_signal) vn_send_sigxfsz(td->td_proc); return (EFBIG); } /* * Helper for VOP_WRITE() implementations, the common code to * handle maximum supported file size on the filesystem, and * RLIMIT_FSIZE, except for special writes from accounting subsystem * and ktrace. * * For maximum file size (maxfsz argument): * - return EFBIG if uio_offset is beyond it * - otherwise, clamp uio_resid if write would extend file beyond maxfsz. * * For RLIMIT_FSIZE: * - return EFBIG and send SIGXFSZ if uio_offset is beyond the limit * - otherwise, clamp uio_resid if write would extend file beyond limit. * * If clamping occured, the adjustment for uio_resid is stored in * *resid_adj, to be re-applied by vn_rlimit_fsizex_res() on return * from the VOP. */ int vn_rlimit_fsizex(const struct vnode *vp, struct uio *uio, off_t maxfsz, ssize_t *resid_adj, struct thread *td) { ssize_t resid_orig; int error; bool adj; resid_orig = uio->uio_resid; adj = resid_adj != NULL; error = vn_rlimit_fsizex1(vp, uio, maxfsz, adj, td); if (adj) *resid_adj = resid_orig - uio->uio_resid; return (error); } void vn_rlimit_fsizex_res(struct uio *uio, ssize_t resid_adj) { uio->uio_resid += resid_adj; } int vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, struct thread *td) { return (vn_rlimit_fsizex(vp, __DECONST(struct uio *, uio), 0, NULL, td)); } 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) { + vm_object_t obj; off_t size; 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"); + ASSERT_VOP_ELOCKED(vp, "vn_bmap_seekhole_locked"); if (vp->v_type != VREG) { error = ENOTTY; goto out; } error = vn_getsize_locked(vp, &size, cred); if (error != 0) goto out; noff = *off; if (noff < 0 || noff >= size) { error = ENXIO; goto out; } + + /* See the comment in ufs_bmap_seekdata(). */ + obj = vp->v_object; + if (obj != NULL) { + VM_OBJECT_WLOCK(obj); + vm_object_page_clean(obj, 0, 0, OBJPC_SYNC); + VM_OBJECT_WUNLOCK(obj); + } + bsize = vp->v_mount->mnt_stat.f_iosize; for (bn = noff / bsize; noff < 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 > size) noff = size; /* noff == 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) + if (vn_lock(vp, LK_EXCLUSIVE) != 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; off_t foffset, fsize, 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: error = vn_getsize(vp, &fsize, cred); if (error != 0) break; /* * If the file references a disk device, then fetch * the media size and use that to determine the ending * offset. */ if (fsize == 0 && vp->v_type == VCHR && fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) fsize = size; if (noneg && offset > 0 && fsize > OFF_MAX - offset) { error = EOVERFLOW; break; } offset += fsize; 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 */ kif->kf_un.kf_file.kf_file_nlink = va.va_nlink; 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 mount *mp; off_t startoff, endoff, xfer, xfer2; u_long blksize; int error, interrupted; bool cantseek, readzeros, eof, lastblock, holetoeof; ssize_t aresid, r = 0; size_t copylen, len, savlen; off_t insize, outsize; 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 = vn_getsize_locked(invp, &insize, 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_fsizex() call, * now that outvp is locked. */ if (fsize_td != NULL) { struct uio io; io.uio_offset = *outoffp; io.uio_resid = len; error = vn_rlimit_fsizex(outvp, &io, 0, &r, fsize_td); len = savlen = io.uio_resid; /* * No need to call vn_rlimit_fsizex_res before return, * since the uio is local. */ } 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 size to decide if writing 0 * bytes is necessary in the loop below. */ if (error == 0) error = vn_getsize_locked(outvp, &outsize, outcred); if (error == 0 && outsize > *outoffp && outsize <= *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) outsize = *outoffp; } VOP_UNLOCK(outvp); } if (mp != NULL) vn_finished_write(mp); if (error != 0) goto out; if (holein == 0 && holeout > 0) { /* * For this special case, the input data will be scanned * for blocks of all 0 bytes. For these blocks, the * write can be skipped for the output file to create * an unallocated region. * Therefore, use the appropriate size for the output file. */ blksize = holeout; if (blksize <= 512) { /* * Use f_iosize, since ZFS reports a _PC_MIN_HOLE_SIZE * of 512, although it actually only creates * unallocated regions for blocks >= f_iosize. */ blksize = outvp->v_mount->mnt_stat.f_iosize; } } else { /* * Use the larger of the two f_iosize values. If they are * not the same size, one will normally be an exact multiple of * the other, since they are both likely to be a power of 2. */ blksize = MAX(invp->v_mount->mnt_stat.f_iosize, outvp->v_mount->mnt_stat.f_iosize); } /* Clip to sane limits. */ if (blksize < 4096) blksize = 4096; else if (blksize > maxphys) blksize = maxphys; 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 = insize; 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 < outsize) { /* Must write 0s to punch hole. */ xfer2 = MIN(outsize - *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 > outsize && (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 < outsize || lastblock || !readzeros) error = vn_write_outvp(outvp, dat, *outoffp, xfer, blksize, readzeros && lastblock && *outoffp >= outsize, 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 | 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 | 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/ufs/ufs/ufs_bmap.c b/sys/ufs/ufs/ufs_bmap.c index 4ac8ca149279..acdd334f6c7b 100644 --- a/sys/ufs/ufs/ufs_bmap.c +++ b/sys/ufs/ufs/ufs_bmap.c @@ -1,515 +1,533 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1991, 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. * * @(#)ufs_bmap.c 8.7 (Berkeley) 3/21/95 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include +#include #include #include #include #include #include +#include +#include + #include #include #include #include #include static ufs_lbn_t lbn_count(struct ufsmount *, int); static int readindir(struct vnode *, ufs_lbn_t, ufs2_daddr_t, struct buf **); /* * Bmap converts the logical block number of a file to its physical block * number on the disk. The conversion is done by using the logical block * number to index into the array of block pointers described by the dinode. */ int ufs_bmap( struct vop_bmap_args /* { struct vnode *a_vp; daddr_t a_bn; struct bufobj **a_bop; daddr_t *a_bnp; int *a_runp; int *a_runb; } */ *ap) { ufs2_daddr_t blkno; int error; /* * Check for underlying vnode requests and ensure that logical * to physical mapping is requested. */ if (ap->a_bop != NULL) *ap->a_bop = &VFSTOUFS(ap->a_vp->v_mount)->um_devvp->v_bufobj; if (ap->a_bnp == NULL) return (0); error = ufs_bmaparray(ap->a_vp, ap->a_bn, &blkno, NULL, ap->a_runp, ap->a_runb); *ap->a_bnp = blkno; return (error); } static int readindir(struct vnode *vp, ufs_lbn_t lbn, ufs2_daddr_t daddr, struct buf **bpp) { struct buf *bp; struct mount *mp; struct ufsmount *ump; int error; mp = vp->v_mount; ump = VFSTOUFS(mp); bp = getblk(vp, lbn, mp->mnt_stat.f_iosize, 0, 0, 0); if ((bp->b_flags & B_CACHE) == 0) { KASSERT(daddr != 0, ("readindir: indirect block not in cache")); bp->b_blkno = blkptrtodb(ump, daddr); bp->b_iocmd = BIO_READ; bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; vfs_busy_pages(bp, 0); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 0); PROC_UNLOCK(curproc); } #endif curthread->td_ru.ru_inblock++; error = bufwait(bp); if (error != 0) { brelse(bp); return (error); } } *bpp = bp; return (0); } /* * Indirect blocks are now on the vnode for the file. They are given negative * logical block numbers. Indirect blocks are addressed by the negative * address of the first data block to which they point. Double indirect blocks * are addressed by one less than the address of the first indirect block to * which they point. Triple indirect blocks are addressed by one less than * the address of the first double indirect block to which they point. * * ufs_bmaparray does the bmap conversion, and if requested returns the * array of logical blocks which must be traversed to get to a block. * Each entry contains the offset into that block that gets you to the * next block and the disk address of the block (if it is assigned). */ int ufs_bmaparray(struct vnode *vp, ufs2_daddr_t bn, ufs2_daddr_t *bnp, struct buf *nbp, int *runp, int *runb) { struct inode *ip; struct buf *bp; struct ufsmount *ump; struct mount *mp; struct indir a[UFS_NIADDR+1], *ap; ufs2_daddr_t daddr; ufs_lbn_t metalbn; int error, num, maxrun = 0; int *nump; ap = NULL; ip = VTOI(vp); mp = vp->v_mount; ump = VFSTOUFS(mp); if (runp) { maxrun = mp->mnt_iosize_max / mp->mnt_stat.f_iosize - 1; *runp = 0; } if (runb) { *runb = 0; } ap = a; nump = # error = ufs_getlbns(vp, bn, ap, nump); if (error) return (error); num = *nump; if (num == 0) { if (bn >= 0 && bn < UFS_NDADDR) { *bnp = blkptrtodb(ump, DIP(ip, i_db[bn])); } else if (bn < 0 && bn >= -UFS_NXADDR) { *bnp = blkptrtodb(ump, ip->i_din2->di_extb[-1 - bn]); if (*bnp == 0) *bnp = -1; if (nbp == NULL) { /* indirect block not found */ return (EINVAL); } nbp->b_xflags |= BX_ALTDATA; return (0); } else { /* blkno out of range */ return (EINVAL); } /* * Since this is FFS independent code, we are out of * scope for the definitions of BLK_NOCOPY and * BLK_SNAP, but we do know that they will fall in * the range 1..um_seqinc, so we use that test and * return a request for a zeroed out buffer if attempts * are made to read a BLK_NOCOPY or BLK_SNAP block. */ if (IS_SNAPSHOT(ip) && DIP(ip, i_db[bn]) > 0 && DIP(ip, i_db[bn]) < ump->um_seqinc) { *bnp = -1; } else if (*bnp == 0) { *bnp = IS_SNAPSHOT(ip) ? blkptrtodb(ump, bn * ump->um_seqinc) : -1; } else if (runp) { ufs2_daddr_t bnb = bn; for (++bn; bn < UFS_NDADDR && *runp < maxrun && is_sequential(ump, DIP(ip, i_db[bn - 1]), DIP(ip, i_db[bn])); ++bn, ++*runp); bn = bnb; if (runb && (bn > 0)) { for (--bn; (bn >= 0) && (*runb < maxrun) && is_sequential(ump, DIP(ip, i_db[bn]), DIP(ip, i_db[bn+1])); --bn, ++*runb); } } return (0); } /* Get disk address out of indirect block array */ daddr = DIP(ip, i_ib[ap->in_off]); for (bp = NULL, ++ap; --num; ++ap) { /* * Exit the loop if there is no disk address assigned yet and * the indirect block isn't in the cache, or if we were * looking for an indirect block and we've found it. */ metalbn = ap->in_lbn; if ((daddr == 0 && !incore(&vp->v_bufobj, metalbn)) || metalbn == bn) break; /* * If we get here, we've either got the block in the cache * or we have a disk address for it, go fetch it. */ if (bp) bqrelse(bp); error = readindir(vp, metalbn, daddr, &bp); if (error != 0) return (error); if (I_IS_UFS1(ip)) daddr = ((ufs1_daddr_t *)bp->b_data)[ap->in_off]; else daddr = ((ufs2_daddr_t *)bp->b_data)[ap->in_off]; if ((error = UFS_CHECK_BLKNO(mp, ip->i_number, daddr, mp->mnt_stat.f_iosize)) != 0) { bqrelse(bp); return (error); } if (I_IS_UFS1(ip)) { if (num == 1 && daddr && runp) { for (bn = ap->in_off + 1; bn < MNINDIR(ump) && *runp < maxrun && is_sequential(ump, ((ufs1_daddr_t *)bp->b_data)[bn - 1], ((ufs1_daddr_t *)bp->b_data)[bn]); ++bn, ++*runp); bn = ap->in_off; if (runb && bn) { for (--bn; bn >= 0 && *runb < maxrun && is_sequential(ump, ((ufs1_daddr_t *)bp->b_data)[bn], ((ufs1_daddr_t *)bp->b_data)[bn+1]); --bn, ++*runb); } } continue; } if (num == 1 && daddr && runp) { for (bn = ap->in_off + 1; bn < MNINDIR(ump) && *runp < maxrun && is_sequential(ump, ((ufs2_daddr_t *)bp->b_data)[bn - 1], ((ufs2_daddr_t *)bp->b_data)[bn]); ++bn, ++*runp); bn = ap->in_off; if (runb && bn) { for (--bn; bn >= 0 && *runb < maxrun && is_sequential(ump, ((ufs2_daddr_t *)bp->b_data)[bn], ((ufs2_daddr_t *)bp->b_data)[bn + 1]); --bn, ++*runb); } } } if (bp) bqrelse(bp); /* * Since this is FFS independent code, we are out of scope for the * definitions of BLK_NOCOPY and BLK_SNAP, but we do know that they * will fall in the range 1..um_seqinc, so we use that test and * return a request for a zeroed out buffer if attempts are made * to read a BLK_NOCOPY or BLK_SNAP block. */ if (IS_SNAPSHOT(ip) && daddr > 0 && daddr < ump->um_seqinc){ *bnp = -1; return (0); } *bnp = blkptrtodb(ump, daddr); if (*bnp == 0) { if (IS_SNAPSHOT(ip)) *bnp = blkptrtodb(ump, bn * ump->um_seqinc); else *bnp = -1; } return (0); } static ufs_lbn_t lbn_count(struct ufsmount *ump, int level) { ufs_lbn_t blockcnt; for (blockcnt = 1; level > 0; level--) blockcnt *= MNINDIR(ump); return (blockcnt); } int ufs_bmap_seekdata(struct vnode *vp, off_t *offp) { struct buf *bp; struct indir a[UFS_NIADDR + 1], *ap; struct inode *ip; struct mount *mp; struct ufsmount *ump; + vm_object_t obj; ufs2_daddr_t bn, daddr, nextbn; uint64_t bsize; off_t numblks; int error, num, num1, off; bp = NULL; error = 0; ip = VTOI(vp); mp = vp->v_mount; ump = VFSTOUFS(mp); if (vp->v_type != VREG || IS_SNAPSHOT(ip)) return (EINVAL); if (*offp < 0 || *offp >= ip->i_size) return (ENXIO); + /* + * We could have pages on the vnode' object queue which still + * do not have the data blocks allocated. Convert all dirty + * pages into buffer writes to ensure that we see all + * allocated data. + */ + obj = vp->v_object; + if (obj != NULL) { + VM_OBJECT_WLOCK(obj); + vm_object_page_clean(obj, 0, 0, OBJPC_SYNC); + VM_OBJECT_WUNLOCK(obj); + } + bsize = mp->mnt_stat.f_iosize; for (bn = *offp / bsize, numblks = howmany(ip->i_size, bsize); bn < numblks; bn = nextbn) { if (bn < UFS_NDADDR) { daddr = DIP(ip, i_db[bn]); if (daddr != 0) break; nextbn = bn + 1; continue; } ap = a; error = ufs_getlbns(vp, bn, ap, &num); if (error != 0) break; MPASS(num >= 2); daddr = DIP(ip, i_ib[ap->in_off]); ap++, num--; for (nextbn = UFS_NDADDR, num1 = num - 1; num1 > 0; num1--) nextbn += lbn_count(ump, num1); if (daddr == 0) { nextbn += lbn_count(ump, num); continue; } for (; daddr != 0 && num > 0; ap++, num--) { if (bp != NULL) bqrelse(bp); error = readindir(vp, ap->in_lbn, daddr, &bp); if (error != 0) return (error); /* * Scan the indirect block until we find a non-zero * pointer. */ off = ap->in_off; do { daddr = I_IS_UFS1(ip) ? ((ufs1_daddr_t *)bp->b_data)[off] : ((ufs2_daddr_t *)bp->b_data)[off]; } while (daddr == 0 && ++off < MNINDIR(ump)); nextbn += off * lbn_count(ump, num - 1); /* * We need to recompute the LBNs of indirect * blocks, so restart with the updated block offset. */ if (off != ap->in_off) break; } if (num == 0) { /* * We found a data block. */ bn = nextbn; break; } } if (bp != NULL) bqrelse(bp); if (bn >= numblks) error = ENXIO; if (error == 0 && *offp < bn * bsize) *offp = bn * bsize; return (error); } /* * Create an array of logical block number/offset pairs which represent the * path of indirect blocks required to access a data block. The first "pair" * contains the logical block number of the appropriate single, double or * triple indirect block and the offset into the inode indirect block array. * Note, the logical block number of the inode single/double/triple indirect * block appears twice in the array, once with the offset into the i_ib and * once with the offset into the page itself. */ int ufs_getlbns(struct vnode *vp, ufs2_daddr_t bn, struct indir *ap, int *nump) { ufs2_daddr_t blockcnt; ufs_lbn_t metalbn, realbn; struct ufsmount *ump; int i, numlevels, off; ump = VFSTOUFS(vp->v_mount); if (nump) *nump = 0; numlevels = 0; realbn = bn; if (bn < 0) bn = -bn; /* The first UFS_NDADDR blocks are direct blocks. */ if (bn < UFS_NDADDR) return (0); /* * Determine the number of levels of indirection. After this loop * is done, blockcnt indicates the number of data blocks possible * at the previous level of indirection, and UFS_NIADDR - i is the * number of levels of indirection needed to locate the requested block. */ for (blockcnt = 1, i = UFS_NIADDR, bn -= UFS_NDADDR; ; i--, bn -= blockcnt) { if (i == 0) return (EFBIG); blockcnt *= MNINDIR(ump); if (bn < blockcnt) break; } /* Calculate the address of the first meta-block. */ if (realbn >= 0) metalbn = -(realbn - bn + UFS_NIADDR - i); else metalbn = -(-realbn - bn + UFS_NIADDR - i); /* * At each iteration, off is the offset into the bap array which is * an array of disk addresses at the current level of indirection. * The logical block number and the offset in that block are stored * into the argument array. */ ap->in_lbn = metalbn; ap->in_off = off = UFS_NIADDR - i; ap++; for (++numlevels; i <= UFS_NIADDR; i++) { /* If searching for a meta-data block, quit when found. */ if (metalbn == realbn) break; blockcnt /= MNINDIR(ump); off = (bn / blockcnt) % MNINDIR(ump); ++numlevels; ap->in_lbn = metalbn; ap->in_off = off; ++ap; metalbn -= -1 + off * blockcnt; } if (nump) *nump = numlevels; return (0); } diff --git a/sys/ufs/ufs/ufs_vnops.c b/sys/ufs/ufs/ufs_vnops.c index 93a5b173b785..ae6d963920f3 100644 --- a/sys/ufs/ufs/ufs_vnops.c +++ b/sys/ufs/ufs/ufs_vnops.c @@ -1,3060 +1,3060 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1993, 1995 * 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. * * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_quota.h" #include "opt_suiddir.h" #include "opt_ufs.h" #include "opt_ffs.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* XXX */ #include #include #include #include #include #include #include #include #include #ifdef UFS_DIRHASH #include #endif #ifdef UFS_GJOURNAL #include FEATURE(ufs_gjournal, "Journaling support through GEOM for UFS"); #endif #ifdef QUOTA FEATURE(ufs_quota, "UFS disk quotas support"); FEATURE(ufs_quota64, "64bit UFS disk quotas support"); #endif #ifdef SUIDDIR FEATURE(suiddir, "Give all new files in directory the same ownership as the directory"); #endif VFS_SMR_DECLARE; #include static vop_accessx_t ufs_accessx; static vop_fplookup_vexec_t ufs_fplookup_vexec; static int ufs_chmod(struct vnode *, int, struct ucred *, struct thread *); static int ufs_chown(struct vnode *, uid_t, gid_t, struct ucred *, struct thread *); static vop_close_t ufs_close; static vop_create_t ufs_create; static vop_stat_t ufs_stat; static vop_getattr_t ufs_getattr; static vop_ioctl_t ufs_ioctl; static vop_link_t ufs_link; static int ufs_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *, const char *); static vop_mmapped_t ufs_mmapped; static vop_mkdir_t ufs_mkdir; static vop_mknod_t ufs_mknod; static vop_open_t ufs_open; static vop_pathconf_t ufs_pathconf; static vop_print_t ufs_print; static vop_readlink_t ufs_readlink; static vop_remove_t ufs_remove; static vop_rename_t ufs_rename; static vop_rmdir_t ufs_rmdir; static vop_setattr_t ufs_setattr; static vop_strategy_t ufs_strategy; static vop_symlink_t ufs_symlink; static vop_whiteout_t ufs_whiteout; static vop_close_t ufsfifo_close; SYSCTL_NODE(_vfs, OID_AUTO, ufs, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "UFS filesystem"); /* * A virgin directory (no blushing please). */ static struct dirtemplate mastertemplate = { 0, 12, DT_DIR, 1, ".", 0, DIRBLKSIZ - 12, DT_DIR, 2, ".." }; static struct odirtemplate omastertemplate = { 0, 12, 1, ".", 0, DIRBLKSIZ - 12, 2, ".." }; static void ufs_itimes_locked(struct vnode *vp) { struct inode *ip; struct timespec ts; ASSERT_VI_LOCKED(vp, __func__); ip = VTOI(vp); if (UFS_RDONLY(ip)) goto out; if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) return; if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp)) UFS_INODE_SET_FLAG(ip, IN_LAZYMOD); else if (((vp->v_mount->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND)) == 0) || (ip->i_flag & (IN_CHANGE | IN_UPDATE))) UFS_INODE_SET_FLAG(ip, IN_MODIFIED); else if (ip->i_flag & IN_ACCESS) UFS_INODE_SET_FLAG(ip, IN_LAZYACCESS); vfs_timestamp(&ts); if (ip->i_flag & IN_ACCESS) { DIP_SET(ip, i_atime, ts.tv_sec); DIP_SET(ip, i_atimensec, ts.tv_nsec); } if (ip->i_flag & IN_UPDATE) { DIP_SET(ip, i_mtime, ts.tv_sec); DIP_SET(ip, i_mtimensec, ts.tv_nsec); } if (ip->i_flag & IN_CHANGE) { DIP_SET(ip, i_ctime, ts.tv_sec); DIP_SET(ip, i_ctimensec, ts.tv_nsec); DIP_SET(ip, i_modrev, DIP(ip, i_modrev) + 1); } out: ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); } void ufs_itimes(struct vnode *vp) { struct inode *ip; ip = VTOI(vp); if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) return; VI_LOCK(vp); ufs_itimes_locked(vp); VI_UNLOCK(vp); } static int ufs_sync_nlink1(struct mount *mp) { int error; error = vfs_busy(mp, 0); if (error == 0) { VFS_SYNC(mp, MNT_WAIT); vfs_unbusy(mp); error = ERELOOKUP; } vfs_rel(mp); return (error); } static int ufs_sync_nlink(struct vnode *vp, struct vnode *vp1) { struct inode *ip; struct mount *mp; int error; ip = VTOI(vp); if (ip->i_nlink < UFS_LINK_MAX) return (0); if (!DOINGSOFTDEP(vp) || ip->i_effnlink >= UFS_LINK_MAX) return (EMLINK); mp = vp->v_mount; vfs_ref(mp); VOP_UNLOCK(vp); if (vp1 != NULL) VOP_UNLOCK(vp1); error = ufs_sync_nlink1(mp); vn_lock_pair(vp, false, vp1, false); return (error); } /* * Create a regular file */ static int ufs_create( struct vop_create_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap) { int error; error = ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), ap->a_dvp, ap->a_vpp, ap->a_cnp, "ufs_create"); if (error != 0) return (error); if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0) cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp); return (0); } /* * Mknod vnode call */ /* ARGSUSED */ static int ufs_mknod( struct vop_mknod_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap) { struct vattr *vap = ap->a_vap; struct vnode **vpp = ap->a_vpp; struct inode *ip; ino_t ino; int error; error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), ap->a_dvp, vpp, ap->a_cnp, "ufs_mknod"); if (error) return (error); ip = VTOI(*vpp); UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE); if (vap->va_rdev != VNOVAL) { /* * Want to be able to use this to make badblock * inodes, so don't truncate the dev number. */ DIP_SET(ip, i_rdev, vap->va_rdev); } /* * Remove inode, then reload it through VFS_VGET(). This is * needed to do further inode initialization, for instance * fifo, which was too early for VFS_VGET() done as part of * UFS_VALLOC(). */ (*vpp)->v_type = VNON; ino = ip->i_number; /* Save this before vgone() invalidates ip. */ vgone(*vpp); vput(*vpp); error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp); if (error) { *vpp = NULL; return (error); } return (0); } /* * Open called. */ /* ARGSUSED */ static int ufs_open(struct vop_open_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip; if (vp->v_type == VCHR || vp->v_type == VBLK) return (EOPNOTSUPP); ip = VTOI(vp); vnode_create_vobject(vp, DIP(ip, i_size), ap->a_td); if (vp->v_type == VREG && (vn_irflag_read(vp) & VIRF_PGREAD) == 0 && ip->i_ump->um_bsize >= PAGE_SIZE) { vn_irflag_set_cond(vp, VIRF_PGREAD); } /* * Files marked append-only must be opened for appending. */ if ((ip->i_flags & APPEND) && (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) return (EPERM); return (0); } /* * Close called. * * Update the times on the inode. */ /* ARGSUSED */ static int ufs_close( struct vop_close_args /* { struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; } */ *ap) { struct vnode *vp = ap->a_vp; ufs_itimes(vp); return (0); } static int ufs_accessx( struct vop_accessx_args /* { struct vnode *a_vp; accmode_t a_accmode; struct ucred *a_cred; struct thread *a_td; } */ *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); accmode_t accmode = ap->a_accmode; int error; #ifdef UFS_ACL struct acl *acl; acl_type_t type; #endif /* * Disallow write attempts on read-only filesystems; * unless the file is a socket, fifo, or a block or * character device resident on the filesystem. */ if (accmode & VMODIFY_PERMS) { switch (vp->v_type) { case VDIR: case VLNK: case VREG: if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); #ifdef QUOTA /* * Inode is accounted in the quotas only if struct * dquot is attached to it. VOP_ACCESS() is called * from vn_open_cred() and provides a convenient * point to call getinoquota(). The lock mode is * exclusive when the file is opening for write. */ if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { error = getinoquota(ip); if (error != 0) return (error); } #endif break; default: break; } } /* * If immutable bit set, nobody gets to write it. "& ~VADMIN_PERMS" * permits the owner of the file to remove the IMMUTABLE flag. */ if ((accmode & (VMODIFY_PERMS & ~VADMIN_PERMS)) && (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT))) return (EPERM); #ifdef UFS_ACL if ((vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) != 0) { if (vp->v_mount->mnt_flag & MNT_NFS4ACLS) type = ACL_TYPE_NFS4; else type = ACL_TYPE_ACCESS; acl = acl_alloc(M_WAITOK); if (type == ACL_TYPE_NFS4) error = ufs_getacl_nfs4_internal(vp, acl, ap->a_td); else error = VOP_GETACL(vp, type, acl, ap->a_cred, ap->a_td); switch (error) { case 0: if (type == ACL_TYPE_NFS4) { error = vaccess_acl_nfs4(vp->v_type, ip->i_uid, ip->i_gid, acl, accmode, ap->a_cred); } else { error = vfs_unixify_accmode(&accmode); if (error == 0) error = vaccess_acl_posix1e(vp->v_type, ip->i_uid, ip->i_gid, acl, accmode, ap->a_cred); } break; default: if (error != EOPNOTSUPP) printf( "ufs_accessx(): Error retrieving ACL on object (%d).\n", error); /* * XXX: Fall back until debugged. Should * eventually possibly log an error, and return * EPERM for safety. */ error = vfs_unixify_accmode(&accmode); if (error == 0) error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, accmode, ap->a_cred); } acl_free(acl); return (error); } #endif /* !UFS_ACL */ error = vfs_unixify_accmode(&accmode); if (error == 0) error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, accmode, ap->a_cred); return (error); } /* * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see * the comment above cache_fplookup for details. */ static int ufs_fplookup_vexec( struct vop_fplookup_vexec_args /* { struct vnode *a_vp; struct ucred *a_cred; struct thread *a_td; } */ *ap) { struct vnode *vp; struct inode *ip; struct ucred *cred; mode_t all_x, mode; vp = ap->a_vp; ip = VTOI_SMR(vp); if (__predict_false(ip == NULL)) return (EAGAIN); /* * XXX ACL race * * ACLs are not supported and UFS clears/sets this flag on mount and * remount. However, we may still be racing with seeing them and there * is no provision to make sure they were accounted for. This matches * the behavior of the locked case, since the lookup there is also * racy: mount takes no measures to block anyone from progressing. */ all_x = S_IXUSR | S_IXGRP | S_IXOTH; mode = atomic_load_short(&ip->i_mode); if (__predict_true((mode & all_x) == all_x)) return (0); cred = ap->a_cred; return (vaccess_vexec_smr(mode, ip->i_uid, ip->i_gid, cred)); } /* ARGSUSED */ static int ufs_stat(struct vop_stat_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct stat *sb = ap->a_sb; int error; error = vop_stat_helper_pre(ap); if (__predict_false(error)) return (error); VI_LOCK(vp); ufs_itimes_locked(vp); if (I_IS_UFS1(ip)) { sb->st_atim.tv_sec = ip->i_din1->di_atime; sb->st_atim.tv_nsec = ip->i_din1->di_atimensec; } else { sb->st_atim.tv_sec = ip->i_din2->di_atime; sb->st_atim.tv_nsec = ip->i_din2->di_atimensec; } VI_UNLOCK(vp); sb->st_dev = dev2udev(ITOUMP(ip)->um_dev); sb->st_ino = ip->i_number; sb->st_mode = (ip->i_mode & ~IFMT) | VTTOIF(vp->v_type); sb->st_nlink = ip->i_effnlink; sb->st_uid = ip->i_uid; sb->st_gid = ip->i_gid; if (I_IS_UFS1(ip)) { sb->st_rdev = ip->i_din1->di_rdev; sb->st_size = ip->i_din1->di_size; sb->st_mtim.tv_sec = ip->i_din1->di_mtime; sb->st_mtim.tv_nsec = ip->i_din1->di_mtimensec; sb->st_ctim.tv_sec = ip->i_din1->di_ctime; sb->st_ctim.tv_nsec = ip->i_din1->di_ctimensec; sb->st_birthtim.tv_sec = -1; sb->st_birthtim.tv_nsec = 0; sb->st_blocks = dbtob((u_quad_t)ip->i_din1->di_blocks) / S_BLKSIZE; } else { sb->st_rdev = ip->i_din2->di_rdev; sb->st_size = ip->i_din2->di_size; sb->st_mtim.tv_sec = ip->i_din2->di_mtime; sb->st_mtim.tv_nsec = ip->i_din2->di_mtimensec; sb->st_ctim.tv_sec = ip->i_din2->di_ctime; sb->st_ctim.tv_nsec = ip->i_din2->di_ctimensec; sb->st_birthtim.tv_sec = ip->i_din2->di_birthtime; sb->st_birthtim.tv_nsec = ip->i_din2->di_birthnsec; sb->st_blocks = dbtob((u_quad_t)ip->i_din2->di_blocks) / S_BLKSIZE; } sb->st_blksize = max(PAGE_SIZE, vp->v_mount->mnt_stat.f_iosize); sb->st_flags = ip->i_flags; sb->st_gen = ip->i_gen; return (vop_stat_helper_post(ap, error)); } /* ARGSUSED */ static int ufs_getattr( struct vop_getattr_args /* { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; } */ *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct vattr *vap = ap->a_vap; VI_LOCK(vp); ufs_itimes_locked(vp); if (I_IS_UFS1(ip)) { vap->va_atime.tv_sec = ip->i_din1->di_atime; vap->va_atime.tv_nsec = ip->i_din1->di_atimensec; } else { vap->va_atime.tv_sec = ip->i_din2->di_atime; vap->va_atime.tv_nsec = ip->i_din2->di_atimensec; } VI_UNLOCK(vp); /* * Copy from inode table */ vap->va_fsid = dev2udev(ITOUMP(ip)->um_dev); vap->va_fileid = ip->i_number; vap->va_mode = ip->i_mode & ~IFMT; vap->va_nlink = ip->i_effnlink; vap->va_uid = ip->i_uid; vap->va_gid = ip->i_gid; if (I_IS_UFS1(ip)) { vap->va_rdev = ip->i_din1->di_rdev; vap->va_size = ip->i_din1->di_size; vap->va_mtime.tv_sec = ip->i_din1->di_mtime; vap->va_mtime.tv_nsec = ip->i_din1->di_mtimensec; vap->va_ctime.tv_sec = ip->i_din1->di_ctime; vap->va_ctime.tv_nsec = ip->i_din1->di_ctimensec; vap->va_bytes = dbtob((u_quad_t)ip->i_din1->di_blocks); vap->va_filerev = ip->i_din1->di_modrev; } else { vap->va_rdev = ip->i_din2->di_rdev; vap->va_size = ip->i_din2->di_size; vap->va_mtime.tv_sec = ip->i_din2->di_mtime; vap->va_mtime.tv_nsec = ip->i_din2->di_mtimensec; vap->va_ctime.tv_sec = ip->i_din2->di_ctime; vap->va_ctime.tv_nsec = ip->i_din2->di_ctimensec; vap->va_birthtime.tv_sec = ip->i_din2->di_birthtime; vap->va_birthtime.tv_nsec = ip->i_din2->di_birthnsec; vap->va_bytes = dbtob((u_quad_t)ip->i_din2->di_blocks); vap->va_filerev = ip->i_din2->di_modrev; } vap->va_flags = ip->i_flags; vap->va_gen = ip->i_gen; vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; vap->va_type = IFTOVT(ip->i_mode); return (0); } /* * Set attribute vnode op. called from several syscalls */ static int ufs_setattr( struct vop_setattr_args /* { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; } */ *ap) { struct vattr *vap = ap->a_vap; struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct ucred *cred = ap->a_cred; struct thread *td = curthread; int error; /* * Check for unsettable attributes. */ if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { return (EINVAL); } if (vap->va_flags != VNOVAL) { if ((vap->va_flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | SF_SNAPSHOT | UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | UF_SPARSE | UF_SYSTEM)) != 0) return (EOPNOTSUPP); if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); /* * Callers may only modify the file flags on objects they * have VADMIN rights for. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) return (error); /* * Unprivileged processes are not permitted to unset system * flags, or modify flags if any system flags are set. * Privileged non-jail processes may not modify system flags * if securelevel > 0 and any existing system flags are set. * Privileged jail processes behave like privileged non-jail * processes if the PR_ALLOW_CHFLAGS permission bit is set; * otherwise, they behave like unprivileged processes. */ if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) { if (ip->i_flags & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { error = securelevel_gt(cred, 0); if (error) return (error); } /* The snapshot flag cannot be toggled. */ if ((vap->va_flags ^ ip->i_flags) & SF_SNAPSHOT) return (EPERM); } else { if (ip->i_flags & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE)) return (EPERM); } ip->i_flags = vap->va_flags; DIP_SET(ip, i_flags, vap->va_flags); UFS_INODE_SET_FLAG(ip, IN_CHANGE); error = UFS_UPDATE(vp, 0); if (ip->i_flags & (IMMUTABLE | APPEND)) return (error); } /* * If immutable or append, no one can change any of its attributes * except the ones already handled (in some cases, file flags * including the immutability flags themselves for the superuser). */ if (ip->i_flags & (IMMUTABLE | APPEND)) return (EPERM); /* * Go through the fields and update iff not VNOVAL. */ if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred, td)) != 0) return (error); } if (vap->va_size != VNOVAL) { /* * XXX most of the following special cases should be in * callers instead of in N filesystems. The VDIR check * mostly already is. */ switch (vp->v_type) { case VDIR: return (EISDIR); case VLNK: case VREG: /* * Truncation should have an effect in these cases. * Disallow it if the filesystem is read-only or * the file is being snapshotted. */ if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if (IS_SNAPSHOT(ip)) return (EPERM); break; default: /* * According to POSIX, the result is unspecified * for file types other than regular files, * directories and shared memory objects. We * don't support shared memory objects in the file * system, and have dubious support for truncating * symlinks. Just ignore the request in other cases. */ return (0); } error = vn_rlimit_trunc(vap->va_size, td); if (error != 0) return (error); if ((error = UFS_TRUNCATE(vp, vap->va_size, IO_NORMAL | ((vap->va_vaflags & VA_SYNC) != 0 ? IO_SYNC : 0), cred)) != 0) return (error); } if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL || vap->va_birthtime.tv_sec != VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if (IS_SNAPSHOT(ip)) return (EPERM); error = vn_utimes_perm(vp, vap, cred, td); if (error != 0) return (error); UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED); if (vap->va_atime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_ACCESS; DIP_SET(ip, i_atime, vap->va_atime.tv_sec); DIP_SET(ip, i_atimensec, vap->va_atime.tv_nsec); } if (vap->va_mtime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_UPDATE; DIP_SET(ip, i_mtime, vap->va_mtime.tv_sec); DIP_SET(ip, i_mtimensec, vap->va_mtime.tv_nsec); } if (vap->va_birthtime.tv_sec != VNOVAL && I_IS_UFS2(ip)) { ip->i_din2->di_birthtime = vap->va_birthtime.tv_sec; ip->i_din2->di_birthnsec = vap->va_birthtime.tv_nsec; } error = UFS_UPDATE(vp, 0); if (error) return (error); } error = 0; if (vap->va_mode != (mode_t)VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if (IS_SNAPSHOT(ip) && (vap->va_mode & (S_IXUSR | S_IWUSR | S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH)) != 0) return (EPERM); error = ufs_chmod(vp, (int)vap->va_mode, cred, td); } return (error); } #ifdef UFS_ACL static int ufs_update_nfs4_acl_after_mode_change(struct vnode *vp, int mode, int file_owner_id, struct ucred *cred, struct thread *td) { int error; struct acl *aclp; aclp = acl_alloc(M_WAITOK); error = ufs_getacl_nfs4_internal(vp, aclp, td); /* * We don't have to handle EOPNOTSUPP here, as the filesystem claims * it supports ACLs. */ if (error) goto out; acl_nfs4_sync_acl_from_mode(aclp, mode, file_owner_id); error = ufs_setacl_nfs4_internal(vp, aclp, td); out: acl_free(aclp); return (error); } #endif /* UFS_ACL */ static int ufs_mmapped( struct vop_mmapped_args /* { struct vnode *a_vp; } */ *ap) { struct vnode *vp; struct inode *ip; struct mount *mp; vp = ap->a_vp; ip = VTOI(vp); mp = vp->v_mount; if ((mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS); /* * XXXKIB No UFS_UPDATE(ap->a_vp, 0) there. */ return (0); } /* * Change the mode on a file. * Inode must be locked before calling. */ static int ufs_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td) { struct inode *ip = VTOI(vp); int newmode, error; /* * To modify the permissions on a file, must possess VADMIN * for that file. */ if ((error = VOP_ACCESSX(vp, VWRITE_ACL, cred, td))) return (error); /* * Privileged processes may set the sticky bit on non-directories, * as well as set the setgid bit on a file with a group that the * process is not a member of. Both of these are allowed in * jail(8). */ if (vp->v_type != VDIR && (mode & S_ISTXT)) { if (priv_check_cred(cred, PRIV_VFS_STICKYFILE)) return (EFTYPE); } if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) { error = priv_check_cred(cred, PRIV_VFS_SETGID); if (error) return (error); } /* * Deny setting setuid if we are not the file owner. */ if ((mode & ISUID) && ip->i_uid != cred->cr_uid) { error = priv_check_cred(cred, PRIV_VFS_ADMIN); if (error) return (error); } newmode = ip->i_mode & ~ALLPERMS; newmode |= (mode & ALLPERMS); UFS_INODE_SET_MODE(ip, newmode); DIP_SET(ip, i_mode, ip->i_mode); UFS_INODE_SET_FLAG(ip, IN_CHANGE); #ifdef UFS_ACL if ((vp->v_mount->mnt_flag & MNT_NFS4ACLS) != 0) error = ufs_update_nfs4_acl_after_mode_change(vp, mode, ip->i_uid, cred, td); #endif if (error == 0 && (ip->i_flag & IN_CHANGE) != 0) error = UFS_UPDATE(vp, 0); return (error); } /* * Perform chown operation on inode ip; * inode must be locked prior to call. */ static int ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, struct thread *td) { struct inode *ip = VTOI(vp); uid_t ouid; gid_t ogid; int error = 0; #ifdef QUOTA int i; ufs2_daddr_t change; #endif if (uid == (uid_t)VNOVAL) uid = ip->i_uid; if (gid == (gid_t)VNOVAL) gid = ip->i_gid; /* * To modify the ownership of a file, must possess VADMIN for that * file. */ if ((error = VOP_ACCESSX(vp, VWRITE_OWNER, cred, td))) return (error); /* * To change the owner of a file, or change the group of a file to a * group of which we are not a member, the caller must have * privilege. */ if (((uid != ip->i_uid && uid != cred->cr_uid) || (gid != ip->i_gid && !groupmember(gid, cred))) && (error = priv_check_cred(cred, PRIV_VFS_CHOWN))) return (error); ogid = ip->i_gid; ouid = ip->i_uid; #ifdef QUOTA if ((error = getinoquota(ip)) != 0) return (error); if (ouid == uid) { dqrele(vp, ip->i_dquot[USRQUOTA]); ip->i_dquot[USRQUOTA] = NODQUOT; } if (ogid == gid) { dqrele(vp, ip->i_dquot[GRPQUOTA]); ip->i_dquot[GRPQUOTA] = NODQUOT; } change = DIP(ip, i_blocks); (void) chkdq(ip, -change, cred, CHOWN|FORCE); (void) chkiq(ip, -1, cred, CHOWN|FORCE); for (i = 0; i < MAXQUOTAS; i++) { dqrele(vp, ip->i_dquot[i]); ip->i_dquot[i] = NODQUOT; } #endif ip->i_gid = gid; DIP_SET(ip, i_gid, gid); ip->i_uid = uid; DIP_SET(ip, i_uid, uid); #ifdef QUOTA if ((error = getinoquota(ip)) == 0) { if (ouid == uid) { dqrele(vp, ip->i_dquot[USRQUOTA]); ip->i_dquot[USRQUOTA] = NODQUOT; } if (ogid == gid) { dqrele(vp, ip->i_dquot[GRPQUOTA]); ip->i_dquot[GRPQUOTA] = NODQUOT; } if ((error = chkdq(ip, change, cred, CHOWN)) == 0) { if ((error = chkiq(ip, 1, cred, CHOWN)) == 0) goto good; else (void) chkdq(ip, -change, cred, CHOWN|FORCE); } for (i = 0; i < MAXQUOTAS; i++) { dqrele(vp, ip->i_dquot[i]); ip->i_dquot[i] = NODQUOT; } } ip->i_gid = ogid; DIP_SET(ip, i_gid, ogid); ip->i_uid = ouid; DIP_SET(ip, i_uid, ouid); if (getinoquota(ip) == 0) { if (ouid == uid) { dqrele(vp, ip->i_dquot[USRQUOTA]); ip->i_dquot[USRQUOTA] = NODQUOT; } if (ogid == gid) { dqrele(vp, ip->i_dquot[GRPQUOTA]); ip->i_dquot[GRPQUOTA] = NODQUOT; } (void) chkdq(ip, change, cred, FORCE|CHOWN); (void) chkiq(ip, 1, cred, FORCE|CHOWN); (void) getinoquota(ip); } return (error); good: if (getinoquota(ip)) panic("ufs_chown: lost quota"); #endif /* QUOTA */ UFS_INODE_SET_FLAG(ip, IN_CHANGE); if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) { UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID)); DIP_SET(ip, i_mode, ip->i_mode); } } error = UFS_UPDATE(vp, 0); return (error); } static int ufs_remove( struct vop_remove_args /* { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap) { struct inode *ip; struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; int error; struct thread *td; td = curthread; ip = VTOI(vp); if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(dvp)->i_flags & APPEND)) return (EPERM); if (DOINGSUJ(dvp)) { error = softdep_prelink(dvp, vp, ap->a_cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } #ifdef UFS_GJOURNAL ufs_gjournal_orphan(vp); #endif error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); if (ip->i_nlink <= 0) vp->v_vflag |= VV_NOSYNC; if (IS_SNAPSHOT(ip)) { /* * Avoid deadlock where another thread is trying to * update the inodeblock for dvp and is waiting on * snaplk. Temporary unlock the vnode lock for the * unlinked file and sync the directory. This should * allow vput() of the directory to not block later on * while holding the snapshot vnode locked, assuming * that the directory hasn't been unlinked too. */ VOP_UNLOCK(vp); (void) VOP_FSYNC(dvp, MNT_WAIT, td); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } return (error); } static void print_bad_link_count(const char *funcname, struct vnode *dvp) { struct inode *dip; dip = VTOI(dvp); uprintf("%s: Bad link count %d on parent inode %jd in file system %s\n", funcname, dip->i_effnlink, (intmax_t)dip->i_number, dvp->v_mount->mnt_stat.f_mntonname); } /* * link vnode call */ static int ufs_link( struct vop_link_args /* { struct vnode *a_tdvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap) { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct inode *ip; struct direct newdir; int error; if (DOINGSUJ(tdvp)) { error = softdep_prelink(tdvp, vp, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } if (VTOI(tdvp)->i_effnlink < 2) { print_bad_link_count("ufs_link", tdvp); error = EINVAL; goto out; } error = ufs_sync_nlink(vp, tdvp); if (error != 0) goto out; ip = VTOI(vp); /* * The file may have been removed after namei dropped the original * lock. */ if (ip->i_effnlink == 0) { error = ENOENT; goto out; } if (ip->i_flags & (IMMUTABLE | APPEND)) { error = EPERM; goto out; } ip->i_effnlink++; ip->i_nlink++; DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(vp)) softdep_setup_link(VTOI(tdvp), ip); error = UFS_UPDATE(vp, !DOINGSOFTDEP(vp) && !DOINGASYNC(vp)); if (!error) { ufs_makedirentry(ip, cnp, &newdir); error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL); } if (error) { ip->i_effnlink--; ip->i_nlink--; DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(vp)) softdep_revert_link(VTOI(tdvp), ip); } out: return (error); } /* * whiteout vnode call */ static int ufs_whiteout( struct vop_whiteout_args /* { struct vnode *a_dvp; struct componentname *a_cnp; int a_flags; } */ *ap) { struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct direct newdir; int error = 0; if (DOINGSUJ(dvp) && (ap->a_flags == CREATE || ap->a_flags == DELETE)) { error = softdep_prelink(dvp, NULL, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } switch (ap->a_flags) { case LOOKUP: /* 4.4 format directories support whiteout operations */ if (!OFSFMT(dvp)) return (0); return (EOPNOTSUPP); case CREATE: /* create a new directory whiteout */ #ifdef INVARIANTS if (OFSFMT(dvp)) panic("ufs_whiteout: old format filesystem"); #endif newdir.d_ino = UFS_WINO; newdir.d_namlen = cnp->cn_namelen; bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1); newdir.d_type = DT_WHT; error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL); break; case DELETE: /* remove an existing directory whiteout */ #ifdef INVARIANTS if (OFSFMT(dvp)) panic("ufs_whiteout: old format filesystem"); #endif cnp->cn_flags &= ~DOWHITEOUT; error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0); break; default: panic("ufs_whiteout: unknown op"); } return (error); } static volatile int rename_restarts; SYSCTL_INT(_vfs_ufs, OID_AUTO, rename_restarts, CTLFLAG_RD, __DEVOLATILE(int *, &rename_restarts), 0, "Times rename had to restart due to lock contention"); /* * Rename system call. * rename("foo", "bar"); * is essentially * unlink("bar"); * link("foo", "bar"); * unlink("foo"); * but ``atomically''. Can't do full commit without saving state in the * inode on disk which isn't feasible at this time. Best we can do is * always guarantee the target exists. * * Basic algorithm is: * * 1) Bump link count on source while we're linking it to the * target. This also ensure the inode won't be deleted out * from underneath us while we work (it may be truncated by * a concurrent `trunc' or `open' for creation). * 2) Link source to destination. If destination already exists, * delete it first. * 3) Unlink source reference to inode if still around. If a * directory was moved and the parent of the destination * is different from the source, patch the ".." entry in the * directory. */ static int ufs_rename( struct vop_rename_args /* { struct vnode *a_fdvp; struct vnode *a_fvp; struct componentname *a_fcnp; struct vnode *a_tdvp; struct vnode *a_tvp; struct componentname *a_tcnp; } */ *ap) { struct vnode *tvp = ap->a_tvp; struct vnode *tdvp = ap->a_tdvp; struct vnode *fvp = ap->a_fvp; struct vnode *fdvp = ap->a_fdvp; struct vnode *nvp; struct componentname *tcnp = ap->a_tcnp; struct componentname *fcnp = ap->a_fcnp; struct thread *td = curthread; struct inode *fip, *tip, *tdp, *fdp; struct direct newdir; off_t endoff; int doingdirectory, newparent; int error = 0; struct mount *mp; ino_t ino; seqc_t fdvp_s, fvp_s, tdvp_s, tvp_s; bool checkpath_locked, want_seqc_end; checkpath_locked = want_seqc_end = false; endoff = 0; mp = tdvp->v_mount; VOP_UNLOCK(tdvp); if (tvp && tvp != tdvp) VOP_UNLOCK(tvp); /* * Check for cross-device rename. */ if ((fvp->v_mount != tdvp->v_mount) || (tvp && (fvp->v_mount != tvp->v_mount))) { error = EXDEV; mp = NULL; goto releout; } fdvp_s = fvp_s = tdvp_s = tvp_s = SEQC_MOD; relock: /* * We need to acquire 2 to 4 locks depending on whether tvp is NULL * and fdvp and tdvp are the same directory. Subsequently we need * to double-check all paths and in the directory rename case we * need to verify that we are not creating a directory loop. To * handle this we acquire all but fdvp using non-blocking * acquisitions. If we fail to acquire any lock in the path we will * drop all held locks, acquire the new lock in a blocking fashion, * and then release it and restart the rename. This acquire/release * step ensures that we do not spin on a lock waiting for release. */ error = vn_lock(fdvp, LK_EXCLUSIVE); if (error) goto releout; if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { VOP_UNLOCK(fdvp); error = vn_lock(tdvp, LK_EXCLUSIVE); if (error) goto releout; VOP_UNLOCK(tdvp); atomic_add_int(&rename_restarts, 1); goto relock; } /* * Re-resolve fvp to be certain it still exists and fetch the * correct vnode. */ error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino); if (error) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); goto releout; } error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp); if (error) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); if (error != EBUSY) goto releout; error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp); if (error != 0) goto releout; VOP_UNLOCK(nvp); vrele(fvp); fvp = nvp; atomic_add_int(&rename_restarts, 1); goto relock; } vrele(fvp); fvp = nvp; /* * Re-resolve tvp and acquire the vnode lock if present. */ error = ufs_lookup_ino(tdvp, NULL, tcnp, &ino); if (error != 0 && error != EJUSTRETURN) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); VOP_UNLOCK(fvp); goto releout; } /* * If tvp disappeared we just carry on. */ if (error == EJUSTRETURN && tvp != NULL) { vrele(tvp); tvp = NULL; } /* * Get the tvp ino if the lookup succeeded. We may have to restart * if the non-blocking acquire fails. */ if (error == 0) { nvp = NULL; error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp); if (tvp) vrele(tvp); tvp = nvp; if (error) { VOP_UNLOCK(fdvp); VOP_UNLOCK(tdvp); VOP_UNLOCK(fvp); if (error != EBUSY) goto releout; error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp); if (error != 0) goto releout; vput(nvp); atomic_add_int(&rename_restarts, 1); goto relock; } } if (DOINGSUJ(fdvp) && (seqc_in_modify(fdvp_s) || !vn_seqc_consistent(fdvp, fdvp_s) || seqc_in_modify(fvp_s) || !vn_seqc_consistent(fvp, fvp_s) || seqc_in_modify(tdvp_s) || !vn_seqc_consistent(tdvp, tdvp_s) || (tvp != NULL && (seqc_in_modify(tvp_s) || !vn_seqc_consistent(tvp, tvp_s))))) { error = softdep_prerename(fdvp, fvp, tdvp, tvp); if (error != 0) goto releout; } fdp = VTOI(fdvp); fip = VTOI(fvp); tdp = VTOI(tdvp); tip = NULL; if (tvp) tip = VTOI(tvp); if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(tdvp)->i_flags & APPEND))) { error = EPERM; goto unlockout; } /* * Renaming a file to itself has no effect. The upper layers should * not call us in that case. However, things could change after * we drop the locks above. */ if (fvp == tvp) { error = 0; goto unlockout; } doingdirectory = 0; newparent = 0; ino = fip->i_number; if (fip->i_nlink >= UFS_LINK_MAX) { if (!DOINGSOFTDEP(fvp) || fip->i_effnlink >= UFS_LINK_MAX) { error = EMLINK; goto unlockout; } vfs_ref(mp); MPASS(!want_seqc_end); if (checkpath_locked) { sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock); checkpath_locked = false; } VOP_UNLOCK(fdvp); VOP_UNLOCK(fvp); vref(tdvp); if (tvp != NULL) vref(tvp); VOP_VPUT_PAIR(tdvp, &tvp, true); error = ufs_sync_nlink1(mp); vrele(fdvp); vrele(fvp); vrele(tdvp); if (tvp != NULL) vrele(tvp); return (error); } if ((fip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (fdp->i_flags & APPEND)) { error = EPERM; goto unlockout; } if ((fip->i_mode & IFMT) == IFDIR) { /* * Avoid ".", "..", and aliases of "." for obvious reasons. */ if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || fdp == fip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) { error = EINVAL; goto unlockout; } if (fdp->i_number != tdp->i_number) newparent = tdp->i_number; doingdirectory = 1; } if ((fvp->v_type == VDIR && fvp->v_mountedhere != NULL) || (tvp != NULL && tvp->v_type == VDIR && tvp->v_mountedhere != NULL)) { error = EXDEV; goto unlockout; } /* * If ".." must be changed (ie the directory gets a new * parent) then the source directory must not be in the * directory hierarchy above the target, as this would * orphan everything below the source directory. Also * the user must have write permission in the source so * as to be able to change "..". */ if (doingdirectory && newparent) { error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread); if (error) goto unlockout; sx_xlock(&VFSTOUFS(mp)->um_checkpath_lock); checkpath_locked = true; error = ufs_checkpath(ino, fdp->i_number, tdp, tcnp->cn_cred, &ino); /* * We encountered a lock that we have to wait for. Unlock * everything else and VGET before restarting. */ if (ino) { sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock); checkpath_locked = false; VOP_UNLOCK(fdvp); VOP_UNLOCK(fvp); VOP_UNLOCK(tdvp); if (tvp) VOP_UNLOCK(tvp); error = VFS_VGET(mp, ino, LK_SHARED, &nvp); if (error == 0) vput(nvp); atomic_add_int(&rename_restarts, 1); goto relock; } if (error) goto unlockout; } if (fip->i_effnlink == 0 || fdp->i_effnlink == 0 || tdp->i_effnlink == 0) panic("Bad effnlink fip %p, fdp %p, tdp %p", fip, fdp, tdp); if (tvp != NULL) vn_seqc_write_begin(tvp); vn_seqc_write_begin(tdvp); vn_seqc_write_begin(fvp); vn_seqc_write_begin(fdvp); want_seqc_end = true; /* * 1) Bump link count while we're moving stuff * around. If we crash somewhere before * completing our work, the link count * may be wrong, but correctable. */ fip->i_effnlink++; fip->i_nlink++; DIP_SET(fip, i_nlink, fip->i_nlink); UFS_INODE_SET_FLAG(fip, IN_CHANGE); if (DOINGSOFTDEP(fvp)) softdep_setup_link(tdp, fip); error = UFS_UPDATE(fvp, !DOINGSOFTDEP(fvp) && !DOINGASYNC(fvp)); if (error) goto bad; /* * 2) If target doesn't exist, link the target * to the source and unlink the source. * Otherwise, rewrite the target directory * entry to reference the source inode and * expunge the original entry's existence. */ if (tip == NULL) { if (ITODEV(tdp) != ITODEV(fip)) panic("ufs_rename: EXDEV"); if (doingdirectory && newparent) { /* * Account for ".." in new directory. * When source and destination have the same * parent we don't adjust the link count. The * actual link modification is completed when * .. is rewritten below. */ if (tdp->i_nlink >= UFS_LINK_MAX) { fip->i_effnlink--; fip->i_nlink--; DIP_SET(fip, i_nlink, fip->i_nlink); UFS_INODE_SET_FLAG(fip, IN_CHANGE); if (DOINGSOFTDEP(fvp)) softdep_revert_link(tdp, fip); if (!DOINGSOFTDEP(tdvp) || tdp->i_effnlink >= UFS_LINK_MAX) { error = EMLINK; goto unlockout; } MPASS(want_seqc_end); if (tvp != NULL) vn_seqc_write_end(tvp); vn_seqc_write_end(tdvp); vn_seqc_write_end(fvp); vn_seqc_write_end(fdvp); want_seqc_end = false; vfs_ref(mp); MPASS(checkpath_locked); sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock); checkpath_locked = false; VOP_UNLOCK(fdvp); VOP_UNLOCK(fvp); vref(tdvp); if (tvp != NULL) vref(tvp); VOP_VPUT_PAIR(tdvp, &tvp, true); error = ufs_sync_nlink1(mp); vrele(fdvp); vrele(fvp); vrele(tdvp); if (tvp != NULL) vrele(tvp); return (error); } } ufs_makedirentry(fip, tcnp, &newdir); error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL); if (error) goto bad; /* Setup tdvp for directory compaction if needed. */ if (I_COUNT(tdp) != 0 && I_ENDOFF(tdp) != 0 && I_ENDOFF(tdp) < tdp->i_size) endoff = I_ENDOFF(tdp); } else { if (ITODEV(tip) != ITODEV(tdp) || ITODEV(tip) != ITODEV(fip)) panic("ufs_rename: EXDEV"); /* * Short circuit rename(foo, foo). */ if (tip->i_number == fip->i_number) panic("ufs_rename: same file"); /* * If the parent directory is "sticky", then the caller * must possess VADMIN for the parent directory, or the * destination of the rename. This implements append-only * directories. */ if ((tdp->i_mode & S_ISTXT) && VOP_ACCESS(tdvp, VADMIN, tcnp->cn_cred, td) && VOP_ACCESS(tvp, VADMIN, tcnp->cn_cred, td)) { error = EPERM; goto bad; } /* * Target must be empty if a directory and have no links * to it. Also, ensure source and target are compatible * (both directories, or both not directories). */ if ((tip->i_mode & IFMT) == IFDIR) { if ((tip->i_effnlink > 2) || !ufs_dirempty(tip, tdp->i_number, tcnp->cn_cred)) { error = ENOTEMPTY; goto bad; } if (!doingdirectory) { error = ENOTDIR; goto bad; } cache_purge(tdvp); } else if (doingdirectory) { error = EISDIR; goto bad; } if (doingdirectory) { if (!newparent) { tdp->i_effnlink--; if (DOINGSOFTDEP(tdvp)) softdep_change_linkcnt(tdp); } tip->i_effnlink--; if (DOINGSOFTDEP(tvp)) softdep_change_linkcnt(tip); } error = ufs_dirrewrite(tdp, tip, fip->i_number, IFTODT(fip->i_mode), (doingdirectory && newparent) ? newparent : doingdirectory); if (error) { if (doingdirectory) { if (!newparent) { tdp->i_effnlink++; if (DOINGSOFTDEP(tdvp)) softdep_change_linkcnt(tdp); } tip->i_effnlink++; if (DOINGSOFTDEP(tvp)) softdep_change_linkcnt(tip); } goto bad; } if (doingdirectory && !DOINGSOFTDEP(tvp)) { /* * The only stuff left in the directory is "." * and "..". The "." reference is inconsequential * since we are quashing it. We have removed the "." * reference and the reference in the parent directory, * but there may be other hard links. The soft * dependency code will arrange to do these operations * after the parent directory entry has been deleted on * disk, so when running with that code we avoid doing * them now. */ if (!newparent) { tdp->i_nlink--; DIP_SET(tdp, i_nlink, tdp->i_nlink); UFS_INODE_SET_FLAG(tdp, IN_CHANGE); } tip->i_nlink--; DIP_SET(tip, i_nlink, tip->i_nlink); UFS_INODE_SET_FLAG(tip, IN_CHANGE); } } /* * 3) Unlink the source. We have to resolve the path again to * fixup the directory offset and count for ufs_dirremove. */ if (fdvp == tdvp) { error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino); if (error) panic("ufs_rename: from entry went away!"); if (ino != fip->i_number) panic("ufs_rename: ino mismatch %ju != %ju\n", (uintmax_t)ino, (uintmax_t)fip->i_number); } /* * If the source is a directory with a * new parent, the link count of the old * parent directory must be decremented * and ".." set to point to the new parent. */ if (doingdirectory && newparent) { /* * If tip exists we simply use its link, otherwise we must * add a new one. */ if (tip == NULL) { tdp->i_effnlink++; tdp->i_nlink++; DIP_SET(tdp, i_nlink, tdp->i_nlink); UFS_INODE_SET_FLAG(tdp, IN_CHANGE); if (DOINGSOFTDEP(tdvp)) softdep_setup_dotdot_link(tdp, fip); error = UFS_UPDATE(tdvp, !DOINGSOFTDEP(tdvp) && !DOINGASYNC(tdvp)); /* Don't go to bad here as the new link exists. */ if (error) goto unlockout; } else if (DOINGSUJ(tdvp)) /* Journal must account for each new link. */ softdep_setup_dotdot_link(tdp, fip); SET_I_OFFSET(fip, mastertemplate.dot_reclen); ufs_dirrewrite(fip, fdp, newparent, DT_DIR, 0); cache_purge(fdvp); } error = ufs_dirremove(fdvp, fip, fcnp->cn_flags, 0); /* * The kern_renameat() looks up the fvp using the DELETE flag, which * causes the removal of the name cache entry for fvp. * As the relookup of the fvp is done in two steps: * ufs_lookup_ino() and then VFS_VGET(), another thread might do a * normal lookup of the from name just before the VFS_VGET() call, * causing the cache entry to be re-instantiated. * * The same issue also applies to tvp if it exists as * otherwise we may have a stale name cache entry for the new * name that references the old i-node if it has other links * or open file descriptors. */ cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp); unlockout: if (want_seqc_end) { if (tvp != NULL) vn_seqc_write_end(tvp); vn_seqc_write_end(tdvp); vn_seqc_write_end(fvp); vn_seqc_write_end(fdvp); } if (checkpath_locked) sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock); vput(fdvp); vput(fvp); /* * If compaction or fsync was requested do it in * ffs_vput_pair() now that other locks are no longer needed. */ if (error == 0 && endoff != 0) { UFS_INODE_SET_FLAG(tdp, IN_ENDOFF); SET_I_ENDOFF(tdp, endoff); } VOP_VPUT_PAIR(tdvp, &tvp, true); return (error); bad: fip->i_effnlink--; fip->i_nlink--; DIP_SET(fip, i_nlink, fip->i_nlink); UFS_INODE_SET_FLAG(fip, IN_CHANGE); if (DOINGSOFTDEP(fvp)) softdep_revert_link(tdp, fip); goto unlockout; releout: if (want_seqc_end) { if (tvp != NULL) vn_seqc_write_end(tvp); vn_seqc_write_end(tdvp); vn_seqc_write_end(fvp); vn_seqc_write_end(fdvp); } vrele(fdvp); vrele(fvp); vrele(tdvp); if (tvp) vrele(tvp); return (error); } #ifdef UFS_ACL static int ufs_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp, mode_t dmode, struct ucred *cred, struct thread *td) { int error; struct inode *ip = VTOI(tvp); struct acl *dacl, *acl; acl = acl_alloc(M_WAITOK); dacl = acl_alloc(M_WAITOK); /* * Retrieve default ACL from parent, if any. */ error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); switch (error) { case 0: /* * Retrieved a default ACL, so merge mode and ACL if * necessary. If the ACL is empty, fall through to * the "not defined or available" case. */ if (acl->acl_cnt != 0) { dmode = acl_posix1e_newfilemode(dmode, acl); UFS_INODE_SET_MODE(ip, dmode); DIP_SET(ip, i_mode, dmode); *dacl = *acl; ufs_sync_acl_from_inode(ip, acl); break; } /* FALLTHROUGH */ case EOPNOTSUPP: /* * Just use the mode as-is. */ UFS_INODE_SET_MODE(ip, dmode); DIP_SET(ip, i_mode, dmode); error = 0; goto out; default: goto out; } /* * XXX: If we abort now, will Soft Updates notify the extattr * code that the EAs for the file need to be released? */ error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); if (error == 0) error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td); switch (error) { case 0: break; case EOPNOTSUPP: /* * XXX: This should not happen, as EOPNOTSUPP above * was supposed to free acl. */ printf("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()\n"); /* panic("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()"); */ break; default: goto out; } out: acl_free(acl); acl_free(dacl); return (error); } static int ufs_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp, mode_t mode, struct ucred *cred, struct thread *td) { int error; struct inode *ip = VTOI(tvp); struct acl *acl; acl = acl_alloc(M_WAITOK); /* * Retrieve default ACL for parent, if any. */ error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td); switch (error) { case 0: /* * Retrieved a default ACL, so merge mode and ACL if * necessary. */ if (acl->acl_cnt != 0) { /* * Two possible ways for default ACL to not * be present. First, the EA can be * undefined, or second, the default ACL can * be blank. If it's blank, fall through to * the it's not defined case. */ mode = acl_posix1e_newfilemode(mode, acl); UFS_INODE_SET_MODE(ip, mode); DIP_SET(ip, i_mode, mode); ufs_sync_acl_from_inode(ip, acl); break; } /* FALLTHROUGH */ case EOPNOTSUPP: /* * Just use the mode as-is. */ UFS_INODE_SET_MODE(ip, mode); DIP_SET(ip, i_mode, mode); error = 0; goto out; default: goto out; } /* * XXX: If we abort now, will Soft Updates notify the extattr * code that the EAs for the file need to be released? */ error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td); switch (error) { case 0: break; case EOPNOTSUPP: /* * XXX: This should not happen, as EOPNOTSUPP above was * supposed to free acl. */ printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " "but no VOP_SETACL()\n"); /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() " "but no VOP_SETACL()"); */ break; default: goto out; } out: acl_free(acl); return (error); } static int ufs_do_nfs4_acl_inheritance(struct vnode *dvp, struct vnode *tvp, mode_t child_mode, struct ucred *cred, struct thread *td) { int error; struct acl *parent_aclp, *child_aclp; parent_aclp = acl_alloc(M_WAITOK); child_aclp = acl_alloc(M_WAITOK | M_ZERO); error = ufs_getacl_nfs4_internal(dvp, parent_aclp, td); if (error) goto out; acl_nfs4_compute_inherited_acl(parent_aclp, child_aclp, child_mode, VTOI(tvp)->i_uid, tvp->v_type == VDIR); error = ufs_setacl_nfs4_internal(tvp, child_aclp, td); if (error) goto out; out: acl_free(parent_aclp); acl_free(child_aclp); return (error); } #endif /* * Mkdir system call */ static int ufs_mkdir( struct vop_mkdir_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; } */ *ap) { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; struct vnode *tvp; struct buf *bp; struct dirtemplate dirtemplate, *dtp; struct direct newdir; int error, dmode; long blkoff; dp = VTOI(dvp); error = ufs_sync_nlink(dvp, NULL); if (error != 0) goto out; dmode = vap->va_mode & 0777; dmode |= IFDIR; /* * Must simulate part of ufs_makeinode here to acquire the inode, * but not have it entered in the parent directory. The entry is * made later after writing "." and ".." entries. */ if (dp->i_effnlink < 2) { print_bad_link_count("ufs_mkdir", dvp); error = EINVAL; goto out; } if (DOINGSUJ(dvp)) { error = softdep_prelink(dvp, NULL, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp); if (error) goto out; vn_seqc_write_begin(tvp); ip = VTOI(tvp); ip->i_gid = dp->i_gid; DIP_SET(ip, i_gid, dp->i_gid); #ifdef SUIDDIR { #ifdef QUOTA struct ucred ucred, *ucp; gid_t ucred_group; ucp = cnp->cn_cred; #endif /* * If we are hacking owners here, (only do this where told to) * and we are not giving it TO root, (would subvert quotas) * then go ahead and give it to the other user. * The new directory also inherits the SUID bit. * If user's UID and dir UID are the same, * 'give it away' so that the SUID is still forced on. */ if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (dp->i_mode & ISUID) && dp->i_uid) { dmode |= ISUID; ip->i_uid = dp->i_uid; DIP_SET(ip, i_uid, dp->i_uid); #ifdef QUOTA if (dp->i_uid != cnp->cn_cred->cr_uid) { /* * Make sure the correct user gets charged * for the space. * Make a dummy credential for the victim. * XXX This seems to never be accessed out of * our context so a stack variable is ok. */ refcount_init(&ucred.cr_ref, 1); ucred.cr_uid = ip->i_uid; ucred.cr_ngroups = 1; ucred.cr_groups = &ucred_group; ucred.cr_groups[0] = dp->i_gid; ucp = &ucred; } #endif } else { ip->i_uid = cnp->cn_cred->cr_uid; DIP_SET(ip, i_uid, ip->i_uid); } #ifdef QUOTA if ((error = getinoquota(ip)) || (error = chkiq(ip, 1, ucp, 0))) { if (DOINGSOFTDEP(tvp)) softdep_revert_link(dp, ip); UFS_VFREE(tvp, ip->i_number, dmode); vn_seqc_write_end(tvp); vgone(tvp); vput(tvp); return (error); } #endif } #else /* !SUIDDIR */ ip->i_uid = cnp->cn_cred->cr_uid; DIP_SET(ip, i_uid, ip->i_uid); #ifdef QUOTA if ((error = getinoquota(ip)) || (error = chkiq(ip, 1, cnp->cn_cred, 0))) { if (DOINGSOFTDEP(tvp)) softdep_revert_link(dp, ip); UFS_VFREE(tvp, ip->i_number, dmode); vn_seqc_write_end(tvp); vgone(tvp); vput(tvp); return (error); } #endif #endif /* !SUIDDIR */ UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE); UFS_INODE_SET_MODE(ip, dmode); DIP_SET(ip, i_mode, dmode); tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ ip->i_effnlink = 2; ip->i_nlink = 2; DIP_SET(ip, i_nlink, 2); if (cnp->cn_flags & ISWHITEOUT) { ip->i_flags |= UF_OPAQUE; DIP_SET(ip, i_flags, ip->i_flags); } /* * Bump link count in parent directory to reflect work done below. * Should be done before reference is created so cleanup is * possible if we crash. */ dp->i_effnlink++; dp->i_nlink++; DIP_SET(dp, i_nlink, dp->i_nlink); UFS_INODE_SET_FLAG(dp, IN_CHANGE); if (DOINGSOFTDEP(dvp)) softdep_setup_mkdir(dp, ip); error = UFS_UPDATE(dvp, !DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp)); if (error) goto bad; #ifdef MAC if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) { error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount, dvp, tvp, cnp); if (error) goto bad; } #endif #ifdef UFS_ACL if (dvp->v_mount->mnt_flag & MNT_ACLS) { error = ufs_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode, cnp->cn_cred, curthread); if (error) goto bad; } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) { error = ufs_do_nfs4_acl_inheritance(dvp, tvp, dmode, cnp->cn_cred, curthread); if (error) goto bad; } #endif /* !UFS_ACL */ /* * Initialize directory with "." and ".." from static template. */ if (!OFSFMT(dvp)) dtp = &mastertemplate; else dtp = (struct dirtemplate *)&omastertemplate; dirtemplate = *dtp; dirtemplate.dot_ino = ip->i_number; dirtemplate.dotdot_ino = dp->i_number; vnode_pager_setsize(tvp, DIRBLKSIZ); if ((error = UFS_BALLOC(tvp, (off_t)0, DIRBLKSIZ, cnp->cn_cred, BA_CLRBUF, &bp)) != 0) goto bad; ip->i_size = DIRBLKSIZ; DIP_SET(ip, i_size, DIRBLKSIZ); UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate); if (DOINGSOFTDEP(tvp)) { /* * Ensure that the entire newly allocated block is a * valid directory so that future growth within the * block does not have to ensure that the block is * written before the inode. */ blkoff = DIRBLKSIZ; while (blkoff < bp->b_bcount) { ((struct direct *) (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ; blkoff += DIRBLKSIZ; } } if ((error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) && !DOINGASYNC(tvp))) != 0) { (void)bwrite(bp); goto bad; } /* * Directory set up, now install its entry in the parent directory. * * If we are not doing soft dependencies, then we must write out the * buffer containing the new directory body before entering the new * name in the parent. If we are doing soft dependencies, then the * buffer containing the new directory body will be passed to and * released in the soft dependency code after the code has attached * an appropriate ordering dependency to the buffer which ensures that * the buffer is written before the new name is written in the parent. */ if (DOINGASYNC(dvp)) bdwrite(bp); else if (!DOINGSOFTDEP(dvp) && ((error = bwrite(bp)))) goto bad; ufs_makedirentry(ip, cnp, &newdir); error = ufs_direnter(dvp, tvp, &newdir, cnp, bp); bad: if (error == 0) { *ap->a_vpp = tvp; vn_seqc_write_end(tvp); } else { dp->i_effnlink--; dp->i_nlink--; DIP_SET(dp, i_nlink, dp->i_nlink); UFS_INODE_SET_FLAG(dp, IN_CHANGE); /* * No need to do an explicit VOP_TRUNCATE here, vrele will * do this for us because we set the link count to 0. */ ip->i_effnlink = 0; ip->i_nlink = 0; DIP_SET(ip, i_nlink, 0); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(tvp)) softdep_revert_mkdir(dp, ip); vn_seqc_write_end(tvp); vgone(tvp); vput(tvp); } out: return (error); } /* * Rmdir system call. */ static int ufs_rmdir( struct vop_rmdir_args /* { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; } */ *ap) { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; int error; ip = VTOI(vp); dp = VTOI(dvp); /* * Do not remove a directory that is in the process of being renamed. * Verify the directory is empty (and valid). Rmdir ".." will not be * valid since ".." will contain a reference to the current directory * and thus be non-empty. Do not allow the removal of mounted on * directories (this can happen when an NFS exported filesystem * tries to remove a locally mounted on directory). */ error = 0; if (dp->i_effnlink <= 2) { if (dp->i_effnlink == 2) print_bad_link_count("ufs_rmdir", dvp); error = EINVAL; goto out; } if (!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { error = ENOTEMPTY; goto out; } if ((dp->i_flags & APPEND) || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { error = EPERM; goto out; } if (vp->v_mountedhere != 0) { error = EINVAL; goto out; } if (DOINGSUJ(dvp)) { error = softdep_prelink(dvp, vp, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } #ifdef UFS_GJOURNAL ufs_gjournal_orphan(vp); #endif /* * Delete reference to directory before purging * inode. If we crash in between, the directory * will be reattached to lost+found, */ dp->i_effnlink--; ip->i_effnlink--; if (DOINGSOFTDEP(vp)) softdep_setup_rmdir(dp, ip); error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1); if (error) { dp->i_effnlink++; ip->i_effnlink++; if (DOINGSOFTDEP(vp)) softdep_revert_rmdir(dp, ip); goto out; } /* * The only stuff left in the directory is "." and "..". The "." * reference is inconsequential since we are quashing it. The soft * dependency code will arrange to do these operations after * the parent directory entry has been deleted on disk, so * when running with that code we avoid doing them now. */ if (!DOINGSOFTDEP(vp)) { dp->i_nlink--; DIP_SET(dp, i_nlink, dp->i_nlink); UFS_INODE_SET_FLAG(dp, IN_CHANGE); error = UFS_UPDATE(dvp, 0); ip->i_nlink--; DIP_SET(ip, i_nlink, ip->i_nlink); UFS_INODE_SET_FLAG(ip, IN_CHANGE); } cache_vop_rmdir(dvp, vp); #ifdef UFS_DIRHASH /* Kill any active hash; i_effnlink == 0, so it will not come back. */ if (ip->i_dirhash != NULL) ufsdirhash_free(ip); #endif out: return (error); } /* * symlink -- make a symbolic link */ static int ufs_symlink( struct vop_symlink_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; const char *a_target; } */ *ap) { struct vnode *vp, **vpp = ap->a_vpp; struct inode *ip; int len, error; error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, vpp, ap->a_cnp, "ufs_symlink"); if (error) return (error); vp = *vpp; len = strlen(ap->a_target); if (len < VFSTOUFS(vp->v_mount)->um_maxsymlinklen) { ip = VTOI(vp); bcopy(ap->a_target, DIP(ip, i_shortlink), len); ip->i_size = len; DIP_SET(ip, i_size, len); UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE); error = UFS_UPDATE(vp, 0); } else error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target), len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, ap->a_cnp->cn_cred, NOCRED, NULL, NULL); if (error) vput(vp); return (error); } /* * Vnode op for reading directories. */ int ufs_readdir( struct vop_readdir_args /* { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; int *a_eofflag; int *a_ncookies; uint64_t **a_cookies; } */ *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct buf *bp; struct inode *ip; struct direct *dp, *edp; uint64_t *cookies; struct dirent dstdp; off_t offset, startoffset; size_t readcnt, skipcnt; ssize_t startresid; u_int ncookies; int error; if (uio->uio_offset < 0) return (EINVAL); ip = VTOI(vp); if (ip->i_effnlink == 0) return (0); if (ap->a_ncookies != NULL) { if (uio->uio_resid < 0) ncookies = 0; else ncookies = uio->uio_resid; if (uio->uio_offset >= ip->i_size) ncookies = 0; else if (ip->i_size - uio->uio_offset < ncookies) ncookies = ip->i_size - uio->uio_offset; ncookies = ncookies / (offsetof(struct direct, d_name) + 4) + 1; cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); *ap->a_ncookies = ncookies; *ap->a_cookies = cookies; } else { ncookies = 0; cookies = NULL; } offset = startoffset = uio->uio_offset; startresid = uio->uio_resid; error = 0; while (error == 0 && uio->uio_resid > 0 && uio->uio_offset < ip->i_size) { error = UFS_BLKATOFF(vp, uio->uio_offset, NULL, &bp); if (error) break; if (bp->b_offset + bp->b_bcount > ip->i_size) readcnt = ip->i_size - bp->b_offset; else readcnt = bp->b_bcount; skipcnt = (size_t)(uio->uio_offset - bp->b_offset) & ~(size_t)(DIRBLKSIZ - 1); offset = bp->b_offset + skipcnt; dp = (struct direct *)&bp->b_data[skipcnt]; edp = (struct direct *)&bp->b_data[readcnt]; while (error == 0 && uio->uio_resid > 0 && dp < edp) { if (dp->d_reclen <= offsetof(struct direct, d_name) || (caddr_t)dp + dp->d_reclen > (caddr_t)edp) { error = EIO; break; } #if BYTE_ORDER == LITTLE_ENDIAN /* Old filesystem format. */ if (OFSFMT(vp)) { dstdp.d_namlen = dp->d_type; dstdp.d_type = dp->d_namlen; } else #endif { dstdp.d_namlen = dp->d_namlen; dstdp.d_type = dp->d_type; } if (offsetof(struct direct, d_name) + dstdp.d_namlen > dp->d_reclen) { error = EIO; break; } if (offset < startoffset || dp->d_ino == 0) goto nextentry; dstdp.d_fileno = dp->d_ino; dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp); bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen); /* NOTE: d_off is the offset of the *next* entry. */ dstdp.d_off = offset + dp->d_reclen; dirent_terminate(&dstdp); if (dstdp.d_reclen > uio->uio_resid) { if (uio->uio_resid == startresid) error = EINVAL; else error = EJUSTRETURN; break; } /* Advance dp. */ error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio); if (error) break; if (cookies != NULL) { KASSERT(ncookies > 0, ("ufs_readdir: cookies buffer too small")); *cookies = offset + dp->d_reclen; cookies++; ncookies--; } nextentry: offset += dp->d_reclen; dp = (struct direct *)((caddr_t)dp + dp->d_reclen); } bqrelse(bp); uio->uio_offset = offset; } /* We need to correct uio_offset. */ uio->uio_offset = offset; if (error == EJUSTRETURN) error = 0; if (ap->a_ncookies != NULL) { if (error == 0) { *ap->a_ncookies -= ncookies; } else { free(*ap->a_cookies, M_TEMP); *ap->a_ncookies = 0; *ap->a_cookies = NULL; } } if (error == 0 && ap->a_eofflag) *ap->a_eofflag = ip->i_size <= uio->uio_offset; return (error); } /* * Return target name of a symbolic link */ static int ufs_readlink( struct vop_readlink_args /* { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; } */ *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); doff_t isize; isize = ip->i_size; if (isize < VFSTOUFS(vp->v_mount)->um_maxsymlinklen) return (uiomove(DIP(ip, i_shortlink), isize, ap->a_uio)); return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); } /* * Calculate the logical to physical mapping if not done already, * then call the device strategy routine. * * In order to be able to swap to a file, the ufs_bmaparray() operation may not * deadlock on memory. See ufs_bmap() for details. */ static int ufs_strategy( struct vop_strategy_args /* { struct vnode *a_vp; struct buf *a_bp; } */ *ap) { struct buf *bp = ap->a_bp; struct vnode *vp = ap->a_vp; ufs2_daddr_t blkno; int error; if (bp->b_blkno == bp->b_lblkno) { error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, bp, NULL, NULL); bp->b_blkno = blkno; if (error) { bp->b_error = error; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return (0); } if ((long)bp->b_blkno == -1) vfs_bio_clrbuf(bp); } if ((long)bp->b_blkno == -1) { bufdone(bp); return (0); } bp->b_iooffset = dbtob(bp->b_blkno); BO_STRATEGY(VFSTOUFS(vp->v_mount)->um_bo, bp); return (0); } /* * Print out the contents of an inode. */ static int ufs_print( struct vop_print_args /* { struct vnode *a_vp; } */ *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); printf("\tnlink=%d, effnlink=%d, size=%jd", ip->i_nlink, ip->i_effnlink, (intmax_t)ip->i_size); if (I_IS_UFS2(ip)) printf(", extsize %d", ip->i_din2->di_extsize); printf("\n\tgeneration=%jx, uid=%d, gid=%d, flags=0x%b\n", (uintmax_t)ip->i_gen, ip->i_uid, ip->i_gid, (u_int)ip->i_flags, PRINT_INODE_FLAGS); printf("\tino %lu, on dev %s", (u_long)ip->i_number, devtoname(ITODEV(ip))); if (vp->v_type == VFIFO) fifo_printinfo(vp); printf("\n"); return (0); } /* * Close wrapper for fifos. * * Update the times on the inode then do device close. */ static int ufsfifo_close( struct vop_close_args /* { struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; } */ *ap) { ufs_close(ap); return (fifo_specops.vop_close(ap)); } /* * Return POSIX pathconf information applicable to ufs filesystems. */ static int ufs_pathconf( struct vop_pathconf_args /* { struct vnode *a_vp; int a_name; int *a_retval; } */ *ap) { int error; error = 0; switch (ap->a_name) { case _PC_LINK_MAX: *ap->a_retval = UFS_LINK_MAX; break; case _PC_NAME_MAX: *ap->a_retval = UFS_MAXNAMLEN; break; case _PC_PIPE_BUF: if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) *ap->a_retval = PIPE_BUF; else error = EINVAL; break; case _PC_CHOWN_RESTRICTED: *ap->a_retval = 1; break; case _PC_NO_TRUNC: *ap->a_retval = 1; break; #ifdef UFS_ACL case _PC_ACL_EXTENDED: if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) *ap->a_retval = 1; else *ap->a_retval = 0; break; case _PC_ACL_NFS4: if (ap->a_vp->v_mount->mnt_flag & MNT_NFS4ACLS) *ap->a_retval = 1; else *ap->a_retval = 0; break; #endif case _PC_ACL_PATH_MAX: #ifdef UFS_ACL if (ap->a_vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) *ap->a_retval = ACL_MAX_ENTRIES; else *ap->a_retval = 3; #else *ap->a_retval = 3; #endif break; #ifdef MAC case _PC_MAC_PRESENT: if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL) *ap->a_retval = 1; else *ap->a_retval = 0; break; #endif case _PC_MIN_HOLE_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_PRIO_IO: *ap->a_retval = 0; break; case _PC_SYNC_IO: *ap->a_retval = 0; break; case _PC_ALLOC_SIZE_MIN: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize; break; case _PC_FILESIZEBITS: *ap->a_retval = 64; break; case _PC_REC_INCR_XFER_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_MAX_XFER_SIZE: *ap->a_retval = -1; /* means ``unlimited'' */ break; case _PC_REC_MIN_XFER_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_XFER_ALIGN: *ap->a_retval = PAGE_SIZE; break; case _PC_SYMLINK_MAX: *ap->a_retval = MAXPATHLEN; break; default: error = vop_stdpathconf(ap); break; } return (error); } /* * Initialize the vnode associated with a new inode, handle aliased * vnodes. */ int ufs_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp) { struct inode *ip; struct vnode *vp; vp = *vpp; ASSERT_VOP_LOCKED(vp, "ufs_vinit"); ip = VTOI(vp); vp->v_type = IFTOVT(ip->i_mode); /* * Only unallocated inodes should be of type VNON. */ if (ip->i_mode != 0 && vp->v_type == VNON) return (EINVAL); if (vp->v_type == VFIFO) vp->v_op = fifoops; if (ip->i_number == UFS_ROOTINO) vp->v_vflag |= VV_ROOT; *vpp = vp; return (0); } /* * Allocate a new inode. * Vnode dvp must be locked. */ static int ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, const char *callfunc) { struct inode *ip, *pdir; struct direct newdir; struct vnode *tvp; int error; pdir = VTOI(dvp); *vpp = NULL; if ((mode & IFMT) == 0) mode |= IFREG; if (pdir->i_effnlink < 2) { print_bad_link_count(callfunc, dvp); return (EINVAL); } if (DOINGSUJ(dvp)) { error = softdep_prelink(dvp, NULL, cnp); if (error != 0) { MPASS(error == ERELOOKUP); return (error); } } error = UFS_VALLOC(dvp, mode, cnp->cn_cred, &tvp); if (error) return (error); ip = VTOI(tvp); ip->i_gid = pdir->i_gid; DIP_SET(ip, i_gid, pdir->i_gid); #ifdef SUIDDIR { #ifdef QUOTA struct ucred ucred, *ucp; gid_t ucred_group; ucp = cnp->cn_cred; #endif /* * If we are not the owner of the directory, * and we are hacking owners here, (only do this where told to) * and we are not giving it TO root, (would subvert quotas) * then go ahead and give it to the other user. * Note that this drops off the execute bits for security. */ if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (pdir->i_mode & ISUID) && (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { ip->i_uid = pdir->i_uid; DIP_SET(ip, i_uid, ip->i_uid); mode &= ~07111; #ifdef QUOTA /* * Make sure the correct user gets charged * for the space. * Quickly knock up a dummy credential for the victim. * XXX This seems to never be accessed out of our * context so a stack variable is ok. */ refcount_init(&ucred.cr_ref, 1); ucred.cr_uid = ip->i_uid; ucred.cr_ngroups = 1; ucred.cr_groups = &ucred_group; ucred.cr_groups[0] = pdir->i_gid; ucp = &ucred; #endif } else { ip->i_uid = cnp->cn_cred->cr_uid; DIP_SET(ip, i_uid, ip->i_uid); } #ifdef QUOTA if ((error = getinoquota(ip)) || (error = chkiq(ip, 1, ucp, 0))) { if (DOINGSOFTDEP(tvp)) softdep_revert_link(pdir, ip); UFS_VFREE(tvp, ip->i_number, mode); vgone(tvp); vput(tvp); return (error); } #endif } #else /* !SUIDDIR */ ip->i_uid = cnp->cn_cred->cr_uid; DIP_SET(ip, i_uid, ip->i_uid); #ifdef QUOTA if ((error = getinoquota(ip)) || (error = chkiq(ip, 1, cnp->cn_cred, 0))) { if (DOINGSOFTDEP(tvp)) softdep_revert_link(pdir, ip); UFS_VFREE(tvp, ip->i_number, mode); vgone(tvp); vput(tvp); return (error); } #endif #endif /* !SUIDDIR */ vn_seqc_write_begin(tvp); /* Mostly to cover asserts */ UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE); UFS_INODE_SET_MODE(ip, mode); DIP_SET(ip, i_mode, mode); tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ ip->i_effnlink = 1; ip->i_nlink = 1; DIP_SET(ip, i_nlink, 1); if (DOINGSOFTDEP(tvp)) softdep_setup_create(VTOI(dvp), ip); if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) && priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID)) { UFS_INODE_SET_MODE(ip, ip->i_mode & ~ISGID); DIP_SET(ip, i_mode, ip->i_mode); } if (cnp->cn_flags & ISWHITEOUT) { ip->i_flags |= UF_OPAQUE; DIP_SET(ip, i_flags, ip->i_flags); } /* * Make sure inode goes to disk before directory entry. */ error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) && !DOINGASYNC(tvp)); if (error) goto bad; #ifdef MAC if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) { error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount, dvp, tvp, cnp); if (error) goto bad; } #endif #ifdef UFS_ACL if (dvp->v_mount->mnt_flag & MNT_ACLS) { error = ufs_do_posix1e_acl_inheritance_file(dvp, tvp, mode, cnp->cn_cred, curthread); if (error) goto bad; } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) { error = ufs_do_nfs4_acl_inheritance(dvp, tvp, mode, cnp->cn_cred, curthread); if (error) goto bad; } #endif /* !UFS_ACL */ ufs_makedirentry(ip, cnp, &newdir); error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL); if (error) goto bad; vn_seqc_write_end(tvp); *vpp = tvp; return (0); bad: /* * Write error occurred trying to update the inode * or the directory so must deallocate the inode. */ ip->i_effnlink = 0; ip->i_nlink = 0; DIP_SET(ip, i_nlink, 0); UFS_INODE_SET_FLAG(ip, IN_CHANGE); if (DOINGSOFTDEP(tvp)) softdep_revert_create(VTOI(dvp), ip); vn_seqc_write_end(tvp); vgone(tvp); vput(tvp); return (error); } static int ufs_ioctl(struct vop_ioctl_args *ap) { struct vnode *vp; int error; vp = ap->a_vp; switch (ap->a_command) { case FIOSEEKDATA: - error = vn_lock(vp, LK_SHARED); + error = vn_lock(vp, LK_EXCLUSIVE); if (error == 0) { error = ufs_bmap_seekdata(vp, (off_t *)ap->a_data); VOP_UNLOCK(vp); } else error = EBADF; return (error); case FIOSEEKHOLE: return (vn_bmap_seekhole(vp, ap->a_command, (off_t *)ap->a_data, ap->a_cred)); default: return (ENOTTY); } } static int ufs_read_pgcache(struct vop_read_pgcache_args *ap) { struct uio *uio; struct vnode *vp; uio = ap->a_uio; vp = ap->a_vp; VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp); if (uio->uio_resid > ptoa(io_hold_cnt) || uio->uio_offset < 0 || (ap->a_ioflag & IO_DIRECT) != 0) return (EJUSTRETURN); return (vn_read_from_obj(vp, uio)); } /* Global vfs data structures for ufs. */ struct vop_vector ufs_vnodeops = { .vop_default = &default_vnodeops, .vop_fsync = VOP_PANIC, .vop_read = VOP_PANIC, .vop_reallocblks = VOP_PANIC, .vop_write = VOP_PANIC, .vop_accessx = ufs_accessx, .vop_bmap = ufs_bmap, .vop_fplookup_vexec = ufs_fplookup_vexec, .vop_fplookup_symlink = VOP_EAGAIN, .vop_cachedlookup = ufs_lookup, .vop_close = ufs_close, .vop_create = ufs_create, .vop_stat = ufs_stat, .vop_getattr = ufs_getattr, .vop_inactive = ufs_inactive, .vop_ioctl = ufs_ioctl, .vop_link = ufs_link, .vop_lookup = vfs_cache_lookup, .vop_mmapped = ufs_mmapped, .vop_mkdir = ufs_mkdir, .vop_mknod = ufs_mknod, .vop_need_inactive = ufs_need_inactive, .vop_open = ufs_open, .vop_pathconf = ufs_pathconf, .vop_poll = vop_stdpoll, .vop_print = ufs_print, .vop_read_pgcache = ufs_read_pgcache, .vop_readdir = ufs_readdir, .vop_readlink = ufs_readlink, .vop_reclaim = ufs_reclaim, .vop_remove = ufs_remove, .vop_rename = ufs_rename, .vop_rmdir = ufs_rmdir, .vop_setattr = ufs_setattr, #ifdef MAC .vop_setlabel = vop_stdsetlabel_ea, #endif .vop_strategy = ufs_strategy, .vop_symlink = ufs_symlink, .vop_whiteout = ufs_whiteout, #ifdef UFS_EXTATTR .vop_getextattr = ufs_getextattr, .vop_deleteextattr = ufs_deleteextattr, .vop_setextattr = ufs_setextattr, #endif #ifdef UFS_ACL .vop_getacl = ufs_getacl, .vop_setacl = ufs_setacl, .vop_aclcheck = ufs_aclcheck, #endif }; VFS_VOP_VECTOR_REGISTER(ufs_vnodeops); struct vop_vector ufs_fifoops = { .vop_default = &fifo_specops, .vop_fsync = VOP_PANIC, .vop_accessx = ufs_accessx, .vop_close = ufsfifo_close, .vop_getattr = ufs_getattr, .vop_inactive = ufs_inactive, .vop_pathconf = ufs_pathconf, .vop_print = ufs_print, .vop_read = VOP_PANIC, .vop_reclaim = ufs_reclaim, .vop_setattr = ufs_setattr, #ifdef MAC .vop_setlabel = vop_stdsetlabel_ea, #endif .vop_write = VOP_PANIC, #ifdef UFS_EXTATTR .vop_getextattr = ufs_getextattr, .vop_deleteextattr = ufs_deleteextattr, .vop_setextattr = ufs_setextattr, #endif #ifdef UFS_ACL .vop_getacl = ufs_getacl, .vop_setacl = ufs_setacl, .vop_aclcheck = ufs_aclcheck, #endif }; VFS_VOP_VECTOR_REGISTER(ufs_fifoops);