Index: head/sys/fs/ext2fs/ext2_vnops.c =================================================================== --- head/sys/fs/ext2fs/ext2_vnops.c (revision 342547) +++ head/sys/fs/ext2fs/ext2_vnops.c (revision 342548) @@ -1,2331 +1,2336 @@ /*- * modified for EXT2FS support in Lites 1.1 * * Aug 1995, Godmar Back (gback@cs.utah.edu) * University of Utah, Department of Computer Science */ /*- * 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. * * 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 8.7 (Berkeley) 2/3/94 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 * $FreeBSD$ */ #include "opt_suiddir.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 "opt_directio.h" #include #include #include #include #include #include #include #include #include #include #include static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *); static void ext2_itimes_locked(struct vnode *); static vop_access_t ext2_access; static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *); static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *, struct thread *); static vop_close_t ext2_close; static vop_create_t ext2_create; static vop_fsync_t ext2_fsync; static vop_getattr_t ext2_getattr; static vop_ioctl_t ext2_ioctl; static vop_link_t ext2_link; static vop_mkdir_t ext2_mkdir; static vop_mknod_t ext2_mknod; static vop_open_t ext2_open; static vop_pathconf_t ext2_pathconf; static vop_print_t ext2_print; static vop_read_t ext2_read; static vop_readlink_t ext2_readlink; static vop_remove_t ext2_remove; static vop_rename_t ext2_rename; static vop_rmdir_t ext2_rmdir; static vop_setattr_t ext2_setattr; static vop_strategy_t ext2_strategy; static vop_symlink_t ext2_symlink; static vop_write_t ext2_write; static vop_deleteextattr_t ext2_deleteextattr; static vop_getextattr_t ext2_getextattr; static vop_listextattr_t ext2_listextattr; static vop_setextattr_t ext2_setextattr; static vop_vptofh_t ext2_vptofh; static vop_close_t ext2fifo_close; static vop_kqfilter_t ext2fifo_kqfilter; /* Global vfs data structures for ext2. */ struct vop_vector ext2_vnodeops = { .vop_default = &default_vnodeops, .vop_access = ext2_access, .vop_bmap = ext2_bmap, .vop_cachedlookup = ext2_lookup, .vop_close = ext2_close, .vop_create = ext2_create, .vop_fsync = ext2_fsync, .vop_getpages = vnode_pager_local_getpages, .vop_getpages_async = vnode_pager_local_getpages_async, .vop_getattr = ext2_getattr, .vop_inactive = ext2_inactive, .vop_ioctl = ext2_ioctl, .vop_link = ext2_link, .vop_lookup = vfs_cache_lookup, .vop_mkdir = ext2_mkdir, .vop_mknod = ext2_mknod, .vop_open = ext2_open, .vop_pathconf = ext2_pathconf, .vop_poll = vop_stdpoll, .vop_print = ext2_print, .vop_read = ext2_read, .vop_readdir = ext2_readdir, .vop_readlink = ext2_readlink, .vop_reallocblks = ext2_reallocblks, .vop_reclaim = ext2_reclaim, .vop_remove = ext2_remove, .vop_rename = ext2_rename, .vop_rmdir = ext2_rmdir, .vop_setattr = ext2_setattr, .vop_strategy = ext2_strategy, .vop_symlink = ext2_symlink, .vop_write = ext2_write, .vop_deleteextattr = ext2_deleteextattr, .vop_getextattr = ext2_getextattr, .vop_listextattr = ext2_listextattr, .vop_setextattr = ext2_setextattr, #ifdef UFS_ACL .vop_getacl = ext2_getacl, .vop_setacl = ext2_setacl, .vop_aclcheck = ext2_aclcheck, #endif /* UFS_ACL */ .vop_vptofh = ext2_vptofh, }; struct vop_vector ext2_fifoops = { .vop_default = &fifo_specops, .vop_access = ext2_access, .vop_close = ext2fifo_close, .vop_fsync = ext2_fsync, .vop_getattr = ext2_getattr, .vop_inactive = ext2_inactive, .vop_kqfilter = ext2fifo_kqfilter, .vop_pathconf = ext2_pathconf, .vop_print = ext2_print, .vop_read = VOP_PANIC, .vop_reclaim = ext2_reclaim, .vop_setattr = ext2_setattr, .vop_write = VOP_PANIC, .vop_vptofh = ext2_vptofh, }; /* * A virgin directory (no blushing please). * Note that the type and namlen fields are reversed relative to ext2. * Also, we don't use `struct odirtemplate', since it would just cause * endianness problems. */ static struct dirtemplate mastertemplate = { 0, 12, 1, EXT2_FT_DIR, ".", 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".." }; static struct dirtemplate omastertemplate = { 0, 12, 1, EXT2_FT_UNKNOWN, ".", 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".." }; static void ext2_itimes_locked(struct vnode *vp) { struct inode *ip; struct timespec ts; ASSERT_VI_LOCKED(vp, __func__); ip = VTOI(vp); if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) return; if ((vp->v_type == VBLK || vp->v_type == VCHR)) ip->i_flag |= IN_LAZYMOD; else ip->i_flag |= IN_MODIFIED; if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { vfs_timestamp(&ts); if (ip->i_flag & IN_ACCESS) { ip->i_atime = ts.tv_sec; ip->i_atimensec = ts.tv_nsec; } if (ip->i_flag & IN_UPDATE) { ip->i_mtime = ts.tv_sec; ip->i_mtimensec = ts.tv_nsec; ip->i_modrev++; } if (ip->i_flag & IN_CHANGE) { ip->i_ctime = ts.tv_sec; ip->i_ctimensec = ts.tv_nsec; } } ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); } void ext2_itimes(struct vnode *vp) { VI_LOCK(vp); ext2_itimes_locked(vp); VI_UNLOCK(vp); } /* * Create a regular file */ static int ext2_create(struct vop_create_args *ap) { int error; error = ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), ap->a_dvp, ap->a_vpp, ap->a_cnp); 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); } static int ext2_open(struct vop_open_args *ap) { if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) return (EOPNOTSUPP); /* * Files marked append-only must be opened for appending. */ if ((VTOI(ap->a_vp)->i_flags & APPEND) && (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) return (EPERM); vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td); return (0); } /* * Close called. * * Update the times on the inode. */ static int ext2_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; VI_LOCK(vp); if (vp->v_usecount > 1) ext2_itimes_locked(vp); VI_UNLOCK(vp); return (0); } static int ext2_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); accmode_t accmode = ap->a_accmode; int error; if (vp->v_type == VBLK || vp->v_type == VCHR) return (EOPNOTSUPP); /* * Disallow write attempts on read-only file systems; * unless the file is a socket, fifo, or a block or * character device resident on the file system. */ if (accmode & VWRITE) { switch (vp->v_type) { case VDIR: case VLNK: case VREG: if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); break; default: break; } } /* If immutable bit set, nobody gets to write it. */ if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT))) return (EPERM); error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, ap->a_accmode, ap->a_cred, NULL); return (error); } static int ext2_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct vattr *vap = ap->a_vap; ext2_itimes(vp); /* * Copy from inode table */ vap->va_fsid = dev2udev(ip->i_devvp->v_rdev); vap->va_fileid = ip->i_number; vap->va_mode = ip->i_mode & ~IFMT; vap->va_nlink = ip->i_nlink; vap->va_uid = ip->i_uid; vap->va_gid = ip->i_gid; vap->va_rdev = ip->i_rdev; vap->va_size = ip->i_size; vap->va_atime.tv_sec = ip->i_atime; vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0; vap->va_mtime.tv_sec = ip->i_mtime; vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0; vap->va_ctime.tv_sec = ip->i_ctime; vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0; if E2DI_HAS_XTIME(ip) { vap->va_birthtime.tv_sec = ip->i_birthtime; vap->va_birthtime.tv_nsec = ip->i_birthnsec; } vap->va_flags = ip->i_flags; vap->va_gen = ip->i_gen; vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); vap->va_type = IFTOVT(ip->i_mode); vap->va_filerev = ip->i_modrev; return (0); } /* * Set attribute vnode op. called from several syscalls */ static int ext2_setattr(struct vop_setattr_args *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) { /* Disallow flags not supported by ext2fs. */ if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP)) 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 and privileged processes in * jail() 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. */ if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) { if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) { error = securelevel_gt(cred, 0); if (error) return (error); } } else { if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) || ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE)) return (EPERM); } ip->i_flags = vap->va_flags; ip->i_flag |= IN_CHANGE; if (ip->i_flags & (IMMUTABLE | APPEND)) return (0); } 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 = ext2_chown(vp, vap->va_uid, vap->va_gid, cred, td)) != 0) return (error); } if (vap->va_size != VNOVAL) { /* * Disallow write attempts on read-only file systems; * unless the file is a socket, fifo, or a block or * character device resident on the file system. */ switch (vp->v_type) { case VDIR: return (EISDIR); case VLNK: case VREG: if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); break; default: break; } if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0) return (error); } if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); /* * From utimes(2): * If times is NULL, ... The caller must be the owner of * the file, have permission to write the file, or be the * super-user. * If times is non-NULL, ... The caller must be the owner of * the file or be the super-user. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) && ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || (error = VOP_ACCESS(vp, VWRITE, cred, td)))) return (error); ip->i_flag |= IN_CHANGE | IN_MODIFIED; if (vap->va_atime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_ACCESS; ip->i_atime = vap->va_atime.tv_sec; ip->i_atimensec = vap->va_atime.tv_nsec; } if (vap->va_mtime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_UPDATE; ip->i_mtime = vap->va_mtime.tv_sec; ip->i_mtimensec = vap->va_mtime.tv_nsec; } ip->i_birthtime = vap->va_birthtime.tv_sec; ip->i_birthnsec = vap->va_birthtime.tv_nsec; error = ext2_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); error = ext2_chmod(vp, (int)vap->va_mode, cred, td); } return (error); } /* * Change the mode on a file. * Inode must be locked before calling. */ static int ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td) { struct inode *ip = VTOI(vp); int error; /* * To modify the permissions on a file, must possess VADMIN * for that file. */ if ((error = VOP_ACCESS(vp, VADMIN, 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. */ if (vp->v_type != VDIR && (mode & S_ISTXT)) { error = priv_check_cred(cred, PRIV_VFS_STICKYFILE); if (error) return (EFTYPE); } if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) { error = priv_check_cred(cred, PRIV_VFS_SETGID); if (error) return (error); } ip->i_mode &= ~ALLPERMS; ip->i_mode |= (mode & ALLPERMS); ip->i_flag |= IN_CHANGE; return (0); } /* * Perform chown operation on inode ip; * inode must be locked prior to call. */ static int ext2_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; 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_ACCESS(vp, VADMIN, 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 || (gid != ip->i_gid && !groupmember(gid, cred))) { error = priv_check_cred(cred, PRIV_VFS_CHOWN); if (error) return (error); } ogid = ip->i_gid; ouid = ip->i_uid; ip->i_gid = gid; ip->i_uid = uid; ip->i_flag |= IN_CHANGE; if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID) != 0) ip->i_mode &= ~(ISUID | ISGID); } return (0); } /* * Synch an open file. */ /* ARGSUSED */ static int ext2_fsync(struct vop_fsync_args *ap) { /* * Flush all dirty buffers associated with a vnode. */ vop_stdfsync(ap); return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT)); } /* * Mknod vnode call */ /* ARGSUSED */ static int ext2_mknod(struct vop_mknod_args *ap) { struct vattr *vap = ap->a_vap; struct vnode **vpp = ap->a_vpp; struct inode *ip; ino_t ino; int error; error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), ap->a_dvp, vpp, ap->a_cnp); if (error) return (error); ip = VTOI(*vpp); ip->i_flag |= 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. */ if (!(ip->i_flag & IN_E4EXTENTS)) ip->i_rdev = vap->va_rdev; } /* * Remove inode, then reload it through VFS_VGET so it is * checked to see if it is an alias of an existing entry in * the inode cache. XXX I don't believe this is necessary now. */ (*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); } static int ext2_remove(struct vop_remove_args *ap) { struct inode *ip; struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; int error; ip = VTOI(vp); if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(dvp)->i_flags & APPEND)) { error = EPERM; goto out; } error = ext2_dirremove(dvp, ap->a_cnp); if (error == 0) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; } out: return (error); } /* * link vnode call */ static int ext2_link(struct vop_link_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct inode *ip; int error; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_link: no name"); #endif ip = VTOI(vp); if ((nlink_t)ip->i_nlink >= EXT4_LINK_MAX) { error = EMLINK; goto out; } if (ip->i_flags & (IMMUTABLE | APPEND)) { error = EPERM; goto out; } ip->i_nlink++; ip->i_flag |= IN_CHANGE; error = ext2_update(vp, !DOINGASYNC(vp)); if (!error) error = ext2_direnter(ip, tdvp, cnp); if (error) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; } out: return (error); } static int ext2_inc_nlink(struct inode *ip) { ip->i_nlink++; if (S_ISDIR(ip->i_mode) && EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK) && ip->i_nlink > 1) { if (ip->i_nlink >= EXT4_LINK_MAX || ip->i_nlink == 2) ip->i_nlink = 1; } else if (ip->i_nlink > EXT4_LINK_MAX) { ip->i_nlink--; return (EMLINK); } return (0); } static void ext2_dec_nlink(struct inode *ip) { if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2) ip->i_nlink--; } /* * 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 ext2_rename(struct vop_rename_args *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 componentname *tcnp = ap->a_tcnp; struct componentname *fcnp = ap->a_fcnp; struct inode *ip, *xp, *dp; struct dirtemplate *dirbuf; int doingdirectory = 0, oldparent = 0, newparent = 0; int error = 0; u_char namlen; #ifdef INVARIANTS if ((tcnp->cn_flags & HASBUF) == 0 || (fcnp->cn_flags & HASBUF) == 0) panic("ext2_rename: no name"); #endif /* * Check for cross-device rename. */ if ((fvp->v_mount != tdvp->v_mount) || (tvp && (fvp->v_mount != tvp->v_mount))) { error = EXDEV; abortit: if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fdvp); vrele(fvp); return (error); } if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(tdvp)->i_flags & APPEND))) { error = EPERM; goto abortit; } /* * Renaming a file to itself has no effect. The upper layers should * not call us in that case. Temporarily just warn if they do. */ if (fvp == tvp) { printf("ext2_rename: fvp == tvp (can't happen)\n"); error = 0; goto abortit; } if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) goto abortit; dp = VTOI(fdvp); ip = VTOI(fvp); if (ip->i_nlink >= EXT4_LINK_MAX && !EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) { VOP_UNLOCK(fvp, 0); error = EMLINK; goto abortit; } if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (dp->i_flags & APPEND)) { VOP_UNLOCK(fvp, 0); error = EPERM; goto abortit; } if ((ip->i_mode & IFMT) == IFDIR) { /* * Avoid ".", "..", and aliases of "." for obvious reasons. */ if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT || (ip->i_flag & IN_RENAME)) { VOP_UNLOCK(fvp, 0); error = EINVAL; goto abortit; } ip->i_flag |= IN_RENAME; oldparent = dp->i_number; doingdirectory++; } vrele(fdvp); /* * When the target exists, both the directory * and target vnodes are returned locked. */ dp = VTOI(tdvp); xp = NULL; if (tvp) xp = VTOI(tvp); /* * 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. */ ext2_inc_nlink(ip); ip->i_flag |= IN_CHANGE; if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) { VOP_UNLOCK(fvp, 0); goto bad; } /* * 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 "..". We must repeat the call * to namei, as the parent directory is unlocked by the * call to checkpath(). */ error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); VOP_UNLOCK(fvp, 0); if (oldparent != dp->i_number) newparent = dp->i_number; if (doingdirectory && newparent) { if (error) /* write access check above */ goto bad; if (xp != NULL) vput(tvp); error = ext2_checkpath(ip, dp, tcnp->cn_cred); if (error) goto out; VREF(tdvp); error = relookup(tdvp, &tvp, tcnp); if (error) goto out; vrele(tdvp); dp = VTOI(tdvp); xp = NULL; if (tvp) xp = VTOI(tvp); } /* * 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 (xp == NULL) { if (dp->i_devvp != ip->i_devvp) panic("ext2_rename: EXDEV"); /* * Account for ".." in new directory. * When source and destination have the same * parent we don't fool with the link count. */ if (doingdirectory && newparent) { error = ext2_inc_nlink(dp); if (error) goto bad; dp->i_flag |= IN_CHANGE; error = ext2_update(tdvp, !DOINGASYNC(tdvp)); if (error) goto bad; } error = ext2_direnter(ip, tdvp, tcnp); if (error) { if (doingdirectory && newparent) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; (void)ext2_update(tdvp, 1); } goto bad; } vput(tdvp); } else { if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp) panic("ext2_rename: EXDEV"); /* * Short circuit rename(foo, foo). */ if (xp->i_number == ip->i_number) panic("ext2_rename: same file"); /* * If the parent directory is "sticky", then the user must * own the parent directory, or the destination of the rename, * otherwise the destination may not be changed (except by * root). This implements append-only directories. */ if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && tcnp->cn_cred->cr_uid != dp->i_uid && xp->i_uid != tcnp->cn_cred->cr_uid) { 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 ((xp->i_mode & IFMT) == IFDIR) { if (!ext2_dirempty(xp, dp->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; } error = ext2_dirrewrite(dp, ip, tcnp); if (error) goto bad; /* * If the target directory is in the same * directory as the source directory, * decrement the link count on the parent * of the target directory. */ if (doingdirectory && !newparent) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; } vput(tdvp); /* * Adjust the link count of the target to * reflect the dirrewrite above. If this is * a directory it is empty and there are * no links to it, so we can squash the inode and * any space associated with it. We disallowed * renaming over top of a directory with links to * it above, as the remaining link would point to * a directory without "." or ".." entries. */ ext2_dec_nlink(xp); if (doingdirectory) { if (--xp->i_nlink != 0) panic("ext2_rename: linked directory"); error = ext2_truncate(tvp, (off_t)0, IO_SYNC, tcnp->cn_cred, tcnp->cn_thread); } xp->i_flag |= IN_CHANGE; vput(tvp); xp = NULL; } /* * 3) Unlink the source. */ fcnp->cn_flags &= ~MODMASK; fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; VREF(fdvp); error = relookup(fdvp, &fvp, fcnp); if (error == 0) vrele(fdvp); if (fvp != NULL) { xp = VTOI(fvp); dp = VTOI(fdvp); } else { /* * From name has disappeared. IN_RENAME is not sufficient * to protect against directory races due to timing windows, * so we can't panic here. */ vrele(ap->a_fvp); return (0); } /* * Ensure that the directory entry still exists and has not * changed while the new name has been entered. If the source is * a file then the entry may have been unlinked or renamed. In * either case there is no further work to be done. If the source * is a directory then it cannot have been rmdir'ed; its link * count of three would cause a rmdir to fail with ENOTEMPTY. * The IN_RENAME flag ensures that it cannot be moved by another * rename. */ if (xp != ip) { /* * From name resolves to a different inode. IN_RENAME is * not sufficient protection against timing window races * so we can't panic here. */ } else { /* * 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) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO); if (!dirbuf) { error = ENOMEM; goto bad; } error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf, ip->i_e2fs->e2fs_bsize, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, tcnp->cn_cred, NOCRED, NULL, NULL); if (error == 0) { /* Like ufs little-endian: */ namlen = dirbuf->dotdot_type; if (namlen != 2 || dirbuf->dotdot_name[0] != '.' || dirbuf->dotdot_name[1] != '.') { ext2_dirbad(xp, (doff_t)12, "rename: mangled dir"); } else { dirbuf->dotdot_ino = newparent; /* * dirblock 0 could be htree root, * try both csum update functions. */ ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)dirbuf); ext2_dx_csum_set(ip, (struct ext2fs_direct_2 *)dirbuf); (void)vn_rdwr(UIO_WRITE, fvp, (caddr_t)dirbuf, ip->i_e2fs->e2fs_bsize, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, tcnp->cn_cred, NOCRED, NULL, NULL); cache_purge(fdvp); } } free(dirbuf, M_TEMP); } error = ext2_dirremove(fdvp, fcnp); if (!error) { ext2_dec_nlink(xp); xp->i_flag |= IN_CHANGE; } xp->i_flag &= ~IN_RENAME; } if (dp) vput(fdvp); if (xp) vput(fvp); vrele(ap->a_fvp); return (error); bad: if (xp) vput(ITOV(xp)); vput(ITOV(dp)); out: if (doingdirectory) ip->i_flag &= ~IN_RENAME; if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { ext2_dec_nlink(ip); ip->i_flag |= IN_CHANGE; ip->i_flag &= ~IN_RENAME; vput(fvp); } else vrele(fvp); return (error); } #ifdef UFS_ACL static int ext2_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); ip->i_mode = dmode; *dacl = *acl; ext2_sync_acl_from_inode(ip, acl); break; } /* FALLTHROUGH */ case EOPNOTSUPP: /* * Just use the mode as-is. */ ip->i_mode = dmode; error = 0; goto out; default: goto out; } 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. */ #ifdef DEBUG printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n"); #endif /* DEBUG */ break; default: goto out; } out: acl_free(acl); acl_free(dacl); return (error); } static int ext2_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); ip->i_mode = mode; ext2_sync_acl_from_inode(ip, acl); break; } /* FALLTHROUGH */ case EOPNOTSUPP: /* * Just use the mode as-is. */ ip->i_mode = mode; error = 0; goto out; default: goto out; } 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); } #endif /* UFS_ACL */ /* * Mkdir system call */ static int ext2_mkdir(struct vop_mkdir_args *ap) { struct m_ext2fs *fs; 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 dirtemplate dirtemplate, *dtp; char *buf = NULL; int error, dmode; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_mkdir: no name"); #endif dp = VTOI(dvp); if ((nlink_t)dp->i_nlink >= EXT4_LINK_MAX && !EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) { error = EMLINK; goto out; } dmode = vap->va_mode & 0777; dmode |= IFDIR; /* * Must simulate part of ext2_makeinode here to acquire the inode, * but not have it entered in the parent directory. The entry is * made later after writing "." and ".." entries. */ error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp); if (error) goto out; ip = VTOI(tvp); fs = ip->i_e2fs; ip->i_gid = dp->i_gid; #ifdef SUIDDIR { /* * if we are hacking owners here, (only do this where told to) * and we are not giving it TOO 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; } else { ip->i_uid = cnp->cn_cred->cr_uid; } } #else ip->i_uid = cnp->cn_cred->cr_uid; #endif ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = dmode; tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ ip->i_nlink = 2; if (cnp->cn_flags & ISWHITEOUT) ip->i_flags |= UF_OPAQUE; error = ext2_update(tvp, 1); /* * Bump link count in parent directory * to reflect work done below. Should * be done before reference is created * so reparation is possible if we crash. */ ext2_inc_nlink(dp); dp->i_flag |= IN_CHANGE; error = ext2_update(dvp, !DOINGASYNC(dvp)); if (error) goto bad; /* Initialize directory with "." and ".." from static template. */ if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, EXT2F_INCOMPAT_FTYPE)) dtp = &mastertemplate; else dtp = &omastertemplate; dirtemplate = *dtp; dirtemplate.dot_ino = ip->i_number; dirtemplate.dotdot_ino = dp->i_number; /* * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's * just redefine it - for this function only */ #undef DIRBLKSIZ #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize dirtemplate.dotdot_reclen = DIRBLKSIZ - 12; buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO); if (!buf) { error = ENOMEM; ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; goto bad; } if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) { dirtemplate.dotdot_reclen -= sizeof(struct ext2fs_direct_tail); ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ)); } memcpy(buf, &dirtemplate, sizeof(dirtemplate)); ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)buf); error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf, DIRBLKSIZ, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED, NULL, NULL); if (error) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; goto bad; } if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize) /* XXX should grow with balloc() */ panic("ext2_mkdir: blksize"); else { ip->i_size = DIRBLKSIZ; ip->i_flag |= IN_CHANGE; } #ifdef UFS_ACL if (dvp->v_mount->mnt_flag & MNT_ACLS) { error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode, cnp->cn_cred, cnp->cn_thread); if (error) goto bad; } #endif /* UFS_ACL */ /* Directory set up, now install its entry in the parent directory. */ error = ext2_direnter(ip, dvp, cnp); if (error) { ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; } bad: /* * No need to do an explicit VOP_TRUNCATE here, vrele will do this * for us because we set the link count to 0. */ if (error) { ip->i_nlink = 0; ip->i_flag |= IN_CHANGE; vput(tvp); } else *ap->a_vpp = tvp; out: free(buf, M_TEMP); return (error); #undef DIRBLKSIZ #define DIRBLKSIZ DEV_BSIZE } /* * Rmdir system call. */ static int ext2_rmdir(struct vop_rmdir_args *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); /* * Verify the directory is empty (and valid). * (Rmdir ".." won't be valid since * ".." will contain a reference to * the current directory and thus be * non-empty.) */ if (!ext2_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; } /* * Delete reference to directory before purging * inode. If we crash in between, the directory * will be reattached to lost+found, */ error = ext2_dirremove(dvp, cnp); if (error) goto out; ext2_dec_nlink(dp); dp->i_flag |= IN_CHANGE; cache_purge(dvp); VOP_UNLOCK(dvp, 0); /* * Truncate inode. The only stuff left * in the directory is "." and "..". */ ip->i_nlink = 0; error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred, cnp->cn_thread); cache_purge(ITOV(ip)); if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { VOP_UNLOCK(vp, 0); vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } out: return (error); } /* * symlink -- make a symbolic link */ static int ext2_symlink(struct vop_symlink_args *ap) { struct vnode *vp, **vpp = ap->a_vpp; struct inode *ip; int len, error; error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, vpp, ap->a_cnp); if (error) return (error); vp = *vpp; len = strlen(ap->a_target); if (len < vp->v_mount->mnt_maxsymlinklen) { ip = VTOI(vp); bcopy(ap->a_target, (char *)ip->i_shortlink, len); ip->i_size = len; ip->i_flag |= IN_CHANGE | IN_UPDATE; } 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); } /* * Return target name of a symbolic link */ static int ext2_readlink(struct vop_readlink_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); int isize; isize = ip->i_size; if (isize < vp->v_mount->mnt_maxsymlinklen) { uiomove((char *)ip->i_shortlink, isize, ap->a_uio); return (0); } 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 ext2_bmaparray() operation may not * deadlock on memory. See ext2_bmap() for details. */ static int ext2_strategy(struct vop_strategy_args *ap) { struct buf *bp = ap->a_bp; struct vnode *vp = ap->a_vp; struct bufobj *bo; daddr_t blkno; int error; if (vp->v_type == VBLK || vp->v_type == VCHR) panic("ext2_strategy: spec"); if (bp->b_blkno == bp->b_lblkno) { if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS) error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL); else error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, 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 = VFSTOEXT2(vp->v_mount)->um_bo; BO_STRATEGY(bo, bp); return (0); } /* * Print out the contents of an inode. */ static int ext2_print(struct vop_print_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number); 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 ext2fifo_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; VI_LOCK(vp); if (vp->v_usecount > 1) ext2_itimes_locked(vp); VI_UNLOCK(vp); return (fifo_specops.vop_close(ap)); } /* * Kqfilter wrapper for fifos. * * Fall through to ext2 kqfilter routines if needed */ static int ext2fifo_kqfilter(struct vop_kqfilter_args *ap) { int error; error = fifo_specops.vop_kqfilter(ap); if (error) error = vfs_kqfilter(ap); return (error); } /* * Return POSIX pathconf information applicable to ext2 filesystems. */ static int ext2_pathconf(struct vop_pathconf_args *ap) { int error = 0; switch (ap->a_name) { case _PC_LINK_MAX: if (EXT2_HAS_RO_COMPAT_FEATURE(VTOI(ap->a_vp)->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) *ap->a_retval = INT_MAX; else *ap->a_retval = EXT4_LINK_MAX; break; case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; 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_PATH_MAX: if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) *ap->a_retval = ACL_MAX_ENTRIES; else *ap->a_retval = 3; break; #endif /* UFS_ACL */ 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); } /* * Vnode operation to remove a named attribute. */ static int ext2_deleteextattr(struct vop_deleteextattr_args *ap) { struct inode *ip; struct m_ext2fs *fs; int error; ip = VTOI(ap->a_vp); fs = ip->i_e2fs; if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) return (EOPNOTSUPP); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error) return (error); error = ENOATTR; if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name); if (error != ENOATTR) return (error); } if (ip->i_facl) error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name); return (error); } /* * Vnode operation to retrieve a named extended attribute. */ static int ext2_getextattr(struct vop_getextattr_args *ap) { struct inode *ip; struct m_ext2fs *fs; int error; ip = VTOI(ap->a_vp); fs = ip->i_e2fs; if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) return (EOPNOTSUPP); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error) return (error); if (ap->a_size != NULL) *ap->a_size = 0; error = ENOATTR; if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { error = ext2_extattr_inode_get(ip, ap->a_attrnamespace, ap->a_name, ap->a_uio, ap->a_size); if (error != ENOATTR) return (error); } if (ip->i_facl) error = ext2_extattr_block_get(ip, ap->a_attrnamespace, ap->a_name, ap->a_uio, ap->a_size); return (error); } /* * Vnode operation to retrieve extended attributes on a vnode. */ static int ext2_listextattr(struct vop_listextattr_args *ap) { struct inode *ip; struct m_ext2fs *fs; int error; ip = VTOI(ap->a_vp); fs = ip->i_e2fs; if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) return (EOPNOTSUPP); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error) return (error); if (ap->a_size != NULL) *ap->a_size = 0; if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { error = ext2_extattr_inode_list(ip, ap->a_attrnamespace, ap->a_uio, ap->a_size); if (error) return (error); } if (ip->i_facl) error = ext2_extattr_block_list(ip, ap->a_attrnamespace, ap->a_uio, ap->a_size); return (error); } /* * Vnode operation to set a named attribute. */ static int ext2_setextattr(struct vop_setextattr_args *ap) { struct inode *ip; struct m_ext2fs *fs; int error; ip = VTOI(ap->a_vp); fs = ip->i_e2fs; if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR)) return (EOPNOTSUPP); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error) return (error); error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name); if (error) return (error); if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) { error = ext2_extattr_inode_set(ip, ap->a_attrnamespace, ap->a_name, ap->a_uio); if (error != ENOSPC) return (error); } error = ext2_extattr_block_set(ip, ap->a_attrnamespace, ap->a_name, ap->a_uio); return (error); } /* * Vnode pointer to File handle */ /* ARGSUSED */ static int ext2_vptofh(struct vop_vptofh_args *ap) { struct inode *ip; struct ufid *ufhp; ip = VTOI(ap->a_vp); ufhp = (struct ufid *)ap->a_fhp; ufhp->ufid_len = sizeof(struct ufid); ufhp->ufid_ino = ip->i_number; ufhp->ufid_gen = ip->i_gen; return (0); } /* * Initialize the vnode associated with a new inode, handle aliased * vnodes. */ int ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp) { struct inode *ip; struct vnode *vp; vp = *vpp; 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 == EXT2_ROOTINO) vp->v_vflag |= VV_ROOT; ip->i_modrev = init_va_filerev(); *vpp = vp; return (0); } /* * Allocate a new inode. */ static int ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) { struct inode *ip, *pdir; struct vnode *tvp; int error; pdir = VTOI(dvp); #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_makeinode: no name"); #endif *vpp = NULL; if ((mode & IFMT) == 0) mode |= IFREG; error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp); if (error) { return (error); } ip = VTOI(tvp); ip->i_gid = pdir->i_gid; #ifdef SUIDDIR { /* * 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 TOO 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; mode &= ~07111; } else { ip->i_uid = cnp->cn_cred->cr_uid; } } #else ip->i_uid = cnp->cn_cred->cr_uid; #endif ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = mode; tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ ip->i_nlink = 1; if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) { if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID)) ip->i_mode &= ~ISGID; } if (cnp->cn_flags & ISWHITEOUT) ip->i_flags |= UF_OPAQUE; /* * Make sure inode goes to disk before directory entry. */ error = ext2_update(tvp, !DOINGASYNC(tvp)); if (error) goto bad; #ifdef UFS_ACL if (dvp->v_mount->mnt_flag & MNT_ACLS) { error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode, cnp->cn_cred, cnp->cn_thread); if (error) goto bad; } #endif /* UFS_ACL */ error = ext2_direnter(ip, dvp, cnp); if (error) goto bad; *vpp = tvp; return (0); bad: /* * Write error occurred trying to update the inode * or the directory so must deallocate the inode. */ ip->i_nlink = 0; ip->i_flag |= IN_CHANGE; vput(tvp); return (error); } /* * Vnode op for reading. */ static int ext2_read(struct vop_read_args *ap) { struct vnode *vp; struct inode *ip; struct uio *uio; struct m_ext2fs *fs; struct buf *bp; daddr_t lbn, nextlbn; off_t bytesinfile; long size, xfersize, blkoffset; int error, orig_resid, seqcount; int ioflag; vp = ap->a_vp; uio = ap->a_uio; ioflag = ap->a_ioflag; seqcount = ap->a_ioflag >> IO_SEQSHIFT; ip = VTOI(vp); #ifdef INVARIANTS if (uio->uio_rw != UIO_READ) panic("%s: mode", "ext2_read"); if (vp->v_type == VLNK) { if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) panic("%s: short symlink", "ext2_read"); } else if (vp->v_type != VREG && vp->v_type != VDIR) panic("%s: type %d", "ext2_read", vp->v_type); #endif orig_resid = uio->uio_resid; KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0")); if (orig_resid == 0) return (0); KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0")); fs = ip->i_e2fs; if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize) return (EOVERFLOW); for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) break; lbn = lblkno(fs, uio->uio_offset); nextlbn = lbn + 1; size = blksize(fs, ip, lbn); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->e2fs_fsize - blkoffset; if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (bytesinfile < xfersize) xfersize = bytesinfile; if (lblktosize(fs, nextlbn) >= ip->i_size) error = bread(vp, lbn, size, NOCRED, &bp); else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { error = cluster_read(vp, ip->i_size, lbn, size, NOCRED, blkoffset + uio->uio_resid, seqcount, 0, &bp); } else if (seqcount > 1) { u_int nextsize = blksize(fs, ip, nextlbn); error = breadn(vp, lbn, size, &nextlbn, &nextsize, 1, NOCRED, &bp); } else error = bread(vp, lbn, size, NOCRED, &bp); if (error) { brelse(bp); bp = NULL; break; } /* * We should only get non-zero b_resid when an I/O error * has occurred, which should cause us to break above. * However, if the short read did not cause an error, * then we want to ensure that we do not uiomove bad * or uninitialized data. */ size -= bp->b_resid; if (size < xfersize) { if (size == 0) break; xfersize = size; } error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); if (error) break; vfs_bio_brelse(bp, ioflag); } /* * This can only happen in the case of an error because the loop * above resets bp to NULL on each iteration and on normal * completion has not set a new value into it. so it must have come * from a 'break' statement */ if (bp != NULL) vfs_bio_brelse(bp, ioflag); if ((error == 0 || uio->uio_resid != orig_resid) && (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) ip->i_flag |= IN_ACCESS; return (error); } static int ext2_ioctl(struct vop_ioctl_args *ap) { switch (ap->a_command) { case FIOSEEKDATA: case FIOSEEKHOLE: return (vn_bmap_seekhole(ap->a_vp, ap->a_command, (off_t *)ap->a_data, ap->a_cred)); default: return (ENOTTY); } } /* * Vnode op for writing. */ static int ext2_write(struct vop_write_args *ap) { struct vnode *vp; struct uio *uio; struct inode *ip; struct m_ext2fs *fs; struct buf *bp; daddr_t lbn; off_t osize; int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize; ioflag = ap->a_ioflag; uio = ap->a_uio; vp = ap->a_vp; seqcount = ioflag >> IO_SEQSHIFT; ip = VTOI(vp); #ifdef INVARIANTS if (uio->uio_rw != UIO_WRITE) panic("%s: mode", "ext2_write"); #endif switch (vp->v_type) { case VREG: if (ioflag & IO_APPEND) uio->uio_offset = ip->i_size; if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) return (EPERM); /* FALLTHROUGH */ case VLNK: break; case VDIR: /* XXX differs from ffs -- this is called from ext2_mkdir(). */ if ((ioflag & IO_SYNC) == 0) panic("ext2_write: nonsync dir write"); break; default: panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp, vp->v_type, (intmax_t)uio->uio_offset, (intmax_t)uio->uio_resid); } KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0")); KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0")); fs = ip->i_e2fs; if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize) return (EFBIG); /* * Maybe this should be above the vnode op call, but so long as * file servers have no limits, I don't think it matters. */ if (vn_rlimit_fsize(vp, uio, uio->uio_td)) return (EFBIG); resid = uio->uio_resid; osize = ip->i_size; if (seqcount > BA_SEQMAX) flags = BA_SEQMAX << BA_SEQSHIFT; else flags = seqcount << BA_SEQSHIFT; if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) flags |= IO_SYNC; for (error = 0; uio->uio_resid > 0;) { lbn = lblkno(fs, uio->uio_offset); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->e2fs_fsize - blkoffset; if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (uio->uio_offset + xfersize > ip->i_size) vnode_pager_setsize(vp, uio->uio_offset + xfersize); /* * We must perform a read-before-write if the transfer size * does not cover the entire buffer. */ if (fs->e2fs_bsize > xfersize) flags |= BA_CLRBUF; else flags &= ~BA_CLRBUF; error = ext2_balloc(ip, lbn, blkoffset + xfersize, ap->a_cred, &bp, flags); if (error != 0) break; if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL)) bp->b_flags |= B_NOCACHE; if (uio->uio_offset + xfersize > ip->i_size) ip->i_size = uio->uio_offset + xfersize; size = blksize(fs, ip, lbn) - bp->b_resid; if (size < xfersize) xfersize = size; error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); /* * If the buffer is not already filled and we encounter an * error while trying to fill it, we have to clear out any * garbage data from the pages instantiated for the buffer. * If we do not, a failed uiomove() during a write can leave * the prior contents of the pages exposed to a userland mmap. * * Note that we need only clear buffers with a transfer size * equal to the block size because buffers with a shorter * transfer size were cleared above by the call to ext2_balloc() * with the BA_CLRBUF flag set. * * If the source region for uiomove identically mmaps the * buffer, uiomove() performed the NOP copy, and the buffer * content remains valid because the page fault handler * validated the pages. */ if (error != 0 && (bp->b_flags & B_CACHE) == 0 && fs->e2fs_bsize == xfersize) vfs_bio_clrbuf(bp); vfs_bio_set_flags(bp, ioflag); /* * If IO_SYNC each buffer is written synchronously. Otherwise * if we have a severe page deficiency write the buffer * asynchronously. Otherwise try to cluster, and if that * doesn't do it then either do an async write (if O_DIRECT), * or a delayed write (if not). */ if (ioflag & IO_SYNC) { (void)bwrite(bp); } else if (vm_page_count_severe() || buf_dirty_count_severe() || (ioflag & IO_ASYNC)) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else if (xfersize + blkoffset == fs->e2fs_fsize) { if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { bp->b_flags |= B_CLUSTEROK; cluster_write(vp, bp, ip->i_size, seqcount, 0); } else { bawrite(bp); } } else if (ioflag & IO_DIRECT) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else { bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } if (error || xfersize == 0) break; } /* * If we successfully wrote any data, and we are not the superuser * we clear the setuid and setgid bits as a precaution against * tampering. */ if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ap->a_cred) { if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) ip->i_mode &= ~(ISUID | ISGID); } if (error) { if (ioflag & IO_UNIT) { (void)ext2_truncate(vp, osize, ioflag & IO_SYNC, ap->a_cred, uio->uio_td); uio->uio_offset -= resid - uio->uio_resid; uio->uio_resid = resid; } } if (uio->uio_resid != resid) { ip->i_flag |= IN_CHANGE | IN_UPDATE; if (ioflag & IO_SYNC) error = ext2_update(vp, 1); } return (error); } Index: head/sys/kern/vfs_subr.c =================================================================== --- head/sys/kern/vfs_subr.c (revision 342547) +++ head/sys/kern/vfs_subr.c (revision 342548) @@ -1,5564 +1,5564 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 */ /* * External virtual filesystem routines */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_watchdog.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 #include #include #include #ifdef DDB #include #endif static void delmntque(struct vnode *vp); static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo); static void syncer_shutdown(void *arg, int howto); static int vtryrecycle(struct vnode *vp); static void v_init_counters(struct vnode *); static void v_incr_usecount(struct vnode *); static void v_incr_usecount_locked(struct vnode *); static void v_incr_devcount(struct vnode *); static void v_decr_devcount(struct vnode *); static void vgonel(struct vnode *); static void vfs_knllock(void *arg); static void vfs_knlunlock(void *arg); static void vfs_knl_assert_locked(void *arg); static void vfs_knl_assert_unlocked(void *arg); static void vnlru_return_batches(struct vfsops *mnt_op); static void destroy_vpollinfo(struct vpollinfo *vi); /* * These fences are intended for cases where some synchronization is * needed between access of v_iflags and lockless vnode refcount (v_holdcnt * and v_usecount) updates. Access to v_iflags is generally synchronized * by the interlock, but we have some internal assertions that check vnode * flags without acquiring the lock. Thus, these fences are INVARIANTS-only * for now. */ #ifdef INVARIANTS #define VNODE_REFCOUNT_FENCE_ACQ() atomic_thread_fence_acq() #define VNODE_REFCOUNT_FENCE_REL() atomic_thread_fence_rel() #else #define VNODE_REFCOUNT_FENCE_ACQ() #define VNODE_REFCOUNT_FENCE_REL() #endif /* * Number of vnodes in existence. Increased whenever getnewvnode() * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode. */ static unsigned long numvnodes; SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "Number of vnodes in existence"); static counter_u64_t vnodes_created; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created, "Number of vnodes created by getnewvnode"); static u_long mnt_free_list_batch = 128; SYSCTL_ULONG(_vfs, OID_AUTO, mnt_free_list_batch, CTLFLAG_RW, &mnt_free_list_batch, 0, "Limit of vnodes held on mnt's free list"); /* * Conversion tables for conversion from vnode types to inode formats * and back. */ enum vtype iftovt_tab[16] = { VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, - VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, + VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON }; int vttoif_tab[10] = { 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT }; /* * List of vnodes that are ready for recycling. */ static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* * "Free" vnode target. Free vnodes are rarely completely free, but are * just ones that are cheap to recycle. Usually they are for files which * have been stat'd but not read; these usually have inode and namecache * data attached to them. This target is the preferred minimum size of a * sub-cache consisting mostly of such files. The system balances the size * of this sub-cache with its complement to try to prevent either from * thrashing while the other is relatively inactive. The targets express * a preference for the best balance. * * "Above" this target there are 2 further targets (watermarks) related * to recyling of free vnodes. In the best-operating case, the cache is * exactly full, the free list has size between vlowat and vhiwat above the * free target, and recycling from it and normal use maintains this state. * Sometimes the free list is below vlowat or even empty, but this state * is even better for immediate use provided the cache is not full. * Otherwise, vnlru_proc() runs to reclaim enough vnodes (usually non-free * ones) to reach one of these states. The watermarks are currently hard- * coded as 4% and 9% of the available space higher. These and the default * of 25% for wantfreevnodes are too large if the memory size is large. * E.g., 9% of 75% of MAXVNODES is more than 566000 vnodes to reclaim * whenever vnlru_proc() becomes active. */ static u_long wantfreevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Target for minimum number of \"free\" vnodes"); static u_long freevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "Number of \"free\" vnodes"); static counter_u64_t recycles_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, "Number of vnodes recycled to meet vnode cache targets"); /* * Various variables used for debugging the new implementation of * reassignbuf(). * XXX these are probably of (very) limited utility now. */ static int reassignbufcalls; SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "Number of calls to reassignbuf"); static counter_u64_t free_owe_inact; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, "Number of times free vnodes kept on active list due to VFS " "owing inactivation"); /* To keep more than one thread at a time from running vfs_getnewfsid */ static struct mtx mntid_mtx; /* * Lock for any access to the following: * vnode_free_list * numvnodes * freevnodes */ static struct mtx vnode_free_list_mtx; /* Publicly exported FS */ struct nfs_public nfs_pub; static uma_zone_t buf_trie_zone; /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ static uma_zone_t vnode_zone; static uma_zone_t vnodepoll_zone; /* * The workitem queue. * * It is useful to delay writes of file data and filesystem metadata * for tens of seconds so that quickly created and deleted files need * not waste disk bandwidth being created and removed. To realize this, * we append vnodes to a "workitem" queue. When running with a soft * updates implementation, most pending metadata dependencies should * not wait for more than a few seconds. Thus, mounted on block devices * are delayed only about a half the time that file data is delayed. * Similarly, directory updates are more critical, so are only delayed * about a third the time that file data is delayed. Thus, there are * SYNCER_MAXDELAY queues that are processed round-robin at a rate of * one each second (driven off the filesystem syncer process). The * syncer_delayno variable indicates the next queue that is to be processed. * Items that need to be processed soon are placed in this queue: * * syncer_workitem_pending[syncer_delayno] * * A delay of fifteen seconds is done by placing the request fifteen * entries later in the queue: * * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] * */ static int syncer_delayno; static long syncer_mask; LIST_HEAD(synclist, bufobj); static struct synclist *syncer_workitem_pending; /* * The sync_mtx protects: * bo->bo_synclist * sync_vnode_count * syncer_delayno * syncer_state * syncer_workitem_pending * syncer_worklist_len * rushjob */ static struct mtx sync_mtx; static struct cv sync_wakeup; #define SYNCER_MAXDELAY 32 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ static int syncdelay = 30; /* max time to delay syncing data */ static int filedelay = 30; /* time to delay syncing files */ SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "Time to delay syncing files (in seconds)"); static int dirdelay = 29; /* time to delay syncing directories */ SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "Time to delay syncing directories (in seconds)"); static int metadelay = 28; /* time to delay syncing metadata */ SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "Time to delay syncing metadata (in seconds)"); static int rushjob; /* number of slots to run ASAP */ static int stat_rush_requests; /* number of times I/O speeded up */ SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "Number of times I/O speeded up (rush requests)"); /* * When shutting down the syncer, run it at four times normal speed. */ #define SYNCER_SHUTDOWN_SPEEDUP 4 static int sync_vnode_count; static int syncer_worklist_len; static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } syncer_state; /* Target for maximum number of vnodes. */ int desiredvnodes; static int gapvnodes; /* gap between wanted and desired */ static int vhiwat; /* enough extras after expansion */ static int vlowat; /* minimal extras before expansion */ static int vstir; /* nonzero to stir non-free vnodes */ static volatile int vsmalltrigger = 8; /* pref to keep if > this many pages */ static int sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS) { int error, old_desiredvnodes; old_desiredvnodes = desiredvnodes; if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0) return (error); if (old_desiredvnodes != desiredvnodes) { wantfreevnodes = desiredvnodes / 4; /* XXX locking seems to be incomplete. */ vfs_hash_changesize(desiredvnodes); cache_changesize(desiredvnodes); } return (0); } SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0, sysctl_update_desiredvnodes, "I", "Target for maximum number of vnodes"); SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Old name for vfs.wantfreevnodes (legacy)"); static int vnlru_nowhere; SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */ static int vnsz2log; /* * Support for the bufobj clean & dirty pctrie. */ static void * buf_trie_alloc(struct pctrie *ptree) { return uma_zalloc(buf_trie_zone, M_NOWAIT); } static void buf_trie_free(struct pctrie *ptree, void *node) { uma_zfree(buf_trie_zone, node); } PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free); /* * Initialize the vnode management data structures. * * Reevaluate the following cap on the number of vnodes after the physical * memory size exceeds 512GB. In the limit, as the physical memory size * grows, the ratio of the memory size in KB to vnodes approaches 64:1. */ #ifndef MAXVNODES_MAX #define MAXVNODES_MAX (512 * 1024 * 1024 / 64) /* 8M */ #endif /* * Initialize a vnode as it first enters the zone. */ static int vnode_init(void *mem, int size, int flags) { struct vnode *vp; vp = mem; bzero(vp, size); /* * Setup locks. */ vp->v_vnlock = &vp->v_lock; mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); /* * By default, don't allow shared locks unless filesystems opt-in. */ lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE); /* * Initialize bufobj. */ bufobj_init(&vp->v_bufobj, vp); /* * Initialize namecache. */ LIST_INIT(&vp->v_cache_src); TAILQ_INIT(&vp->v_cache_dst); /* * Initialize rangelocks. */ rangelock_init(&vp->v_rl); return (0); } /* * Free a vnode when it is cleared from the zone. */ static void vnode_fini(void *mem, int size) { struct vnode *vp; struct bufobj *bo; vp = mem; rangelock_destroy(&vp->v_rl); lockdestroy(vp->v_vnlock); mtx_destroy(&vp->v_interlock); bo = &vp->v_bufobj; rw_destroy(BO_LOCKPTR(bo)); } /* * Provide the size of NFS nclnode and NFS fh for calculation of the * vnode memory consumption. The size is specified directly to * eliminate dependency on NFS-private header. * * Other filesystems may use bigger or smaller (like UFS and ZFS) * private inode data, but the NFS-based estimation is ample enough. * Still, we care about differences in the size between 64- and 32-bit * platforms. * * Namecache structure size is heuristically * sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1. */ #ifdef _LP64 #define NFS_NCLNODE_SZ (528 + 64) #define NC_SZ 148 #else #define NFS_NCLNODE_SZ (360 + 32) #define NC_SZ 92 #endif static void vntblinit(void *dummy __unused) { u_int i; int physvnodes, virtvnodes; /* * Desiredvnodes is a function of the physical memory size and the * kernel's heap size. Generally speaking, it scales with the * physical memory size. The ratio of desiredvnodes to the physical * memory size is 1:16 until desiredvnodes exceeds 98,304. * Thereafter, the * marginal ratio of desiredvnodes to the physical memory size is * 1:64. However, desiredvnodes is limited by the kernel's heap * size. The memory required by desiredvnodes vnodes and vm objects * must not exceed 1/10th of the kernel's heap size. */ physvnodes = maxproc + pgtok(vm_cnt.v_page_count) / 64 + 3 * min(98304 * 16, pgtok(vm_cnt.v_page_count)) / 64; virtvnodes = vm_kmem_size / (10 * (sizeof(struct vm_object) + sizeof(struct vnode) + NC_SZ * ncsizefactor + NFS_NCLNODE_SZ)); desiredvnodes = min(physvnodes, virtvnodes); if (desiredvnodes > MAXVNODES_MAX) { if (bootverbose) printf("Reducing kern.maxvnodes %d -> %d\n", desiredvnodes, MAXVNODES_MAX); desiredvnodes = MAXVNODES_MAX; } wantfreevnodes = desiredvnodes / 4; mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); TAILQ_INIT(&vnode_free_list); mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, vnode_init, vnode_fini, UMA_ALIGN_PTR, 0); vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); /* * Preallocate enough nodes to support one-per buf so that * we can not fail an insert. reassignbuf() callers can not * tolerate the insertion failure. */ buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(), NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM); uma_prealloc(buf_trie_zone, nbuf); vnodes_created = counter_u64_alloc(M_WAITOK); recycles_count = counter_u64_alloc(M_WAITOK); free_owe_inact = counter_u64_alloc(M_WAITOK); /* * Initialize the filesystem syncer. */ syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, &syncer_mask); syncer_maxdelay = syncer_mask + 1; mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); cv_init(&sync_wakeup, "syncer"); for (i = 1; i <= sizeof(struct vnode); i <<= 1) vnsz2log++; vnsz2log--; } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); /* * Mark a mount point as busy. Used to synchronize access and to delay * unmounting. Eventually, mountlist_mtx is not released on failure. * * vfs_busy() is a custom lock, it can block the caller. * vfs_busy() only sleeps if the unmount is active on the mount point. * For a mountpoint mp, vfs_busy-enforced lock is before lock of any * vnode belonging to mp. * * Lookup uses vfs_busy() to traverse mount points. * root fs var fs * / vnode lock A / vnode lock (/var) D * /var vnode lock B /log vnode lock(/var/log) E * vfs_busy lock C vfs_busy lock F * * Within each file system, the lock order is C->A->B and F->D->E. * * When traversing across mounts, the system follows that lock order: * * C->A->B * | * +->F->D->E * * The lookup() process for namei("/var") illustrates the process: * VOP_LOOKUP() obtains B while A is held * vfs_busy() obtains a shared lock on F while A and B are held * vput() releases lock on B * vput() releases lock on A * VFS_ROOT() obtains lock on D while shared lock on F is held * vfs_unbusy() releases shared lock on F * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A. * Attempt to lock A (instead of vp_crossmp) while D is held would * violate the global order, causing deadlocks. * * dounmount() locks B while F is drained. */ int vfs_busy(struct mount *mp, int flags) { MPASS((flags & ~MBF_MASK) == 0); CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); MNT_ILOCK(mp); MNT_REF(mp); /* * If mount point is currently being unmounted, sleep until the * mount point fate is decided. If thread doing the unmounting fails, * it will clear MNTK_UNMOUNT flag before waking us up, indicating * that this mount point has survived the unmount attempt and vfs_busy * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE * flag in addition to MNTK_UNMOUNT, indicating that mount point is * about to be really destroyed. vfs_busy needs to release its * reference on the mount point in this case and return with ENOENT, * telling the caller that mount mount it tried to busy is no longer * valid. */ while (mp->mnt_kern_flag & MNTK_UNMOUNT) { if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { MNT_REL(mp); MNT_IUNLOCK(mp); CTR1(KTR_VFS, "%s: failed busying before sleeping", __func__); return (ENOENT); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_kern_flag |= MNTK_MWAIT; msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0); if (flags & MBF_MNTLSTLOCK) mtx_lock(&mountlist_mtx); MNT_ILOCK(mp); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_lockref++; MNT_IUNLOCK(mp); return (0); } /* * Free a busy filesystem. */ void vfs_unbusy(struct mount *mp) { CTR2(KTR_VFS, "%s: mp %p", __func__, mp); MNT_ILOCK(mp); MNT_REL(mp); KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref")); mp->mnt_lockref--; if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); CTR1(KTR_VFS, "%s: waking up waiters", __func__); mp->mnt_kern_flag &= ~MNTK_DRAINING; wakeup(&mp->mnt_lockref); } MNT_IUNLOCK(mp); } /* * Lookup a mount point by filesystem identifier. */ struct mount * vfs_getvfs(fsid_t *fsid) { struct mount *mp; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { vfs_ref(mp); mtx_unlock(&mountlist_mtx); return (mp); } } mtx_unlock(&mountlist_mtx); CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); return ((struct mount *) 0); } /* * Lookup a mount point by filesystem identifier, busying it before * returning. * * To avoid congestion on mountlist_mtx, implement simple direct-mapped * cache for popular filesystem identifiers. The cache is lockess, using * the fact that struct mount's are never freed. In worst case we may * get pointer to unmounted or even different filesystem, so we have to * check what we got, and go slow way if so. */ struct mount * vfs_busyfs(fsid_t *fsid) { #define FSID_CACHE_SIZE 256 typedef struct mount * volatile vmp_t; static vmp_t cache[FSID_CACHE_SIZE]; struct mount *mp; int error; uint32_t hash; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); hash = fsid->val[0] ^ fsid->val[1]; hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1); mp = cache[hash]; if (mp == NULL || mp->mnt_stat.f_fsid.val[0] != fsid->val[0] || mp->mnt_stat.f_fsid.val[1] != fsid->val[1]) goto slow; if (vfs_busy(mp, 0) != 0) { cache[hash] = NULL; goto slow; } if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) return (mp); else vfs_unbusy(mp); slow: mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { error = vfs_busy(mp, MBF_MNTLSTLOCK); if (error) { cache[hash] = NULL; mtx_unlock(&mountlist_mtx); return (NULL); } cache[hash] = mp; return (mp); } } CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); mtx_unlock(&mountlist_mtx); return ((struct mount *) 0); } /* * Check if a user can access privileged mount options. */ int vfs_suser(struct mount *mp, struct thread *td) { int error; if (jailed(td->td_ucred)) { /* * If the jail of the calling thread lacks permission for * this type of file system, deny immediately. */ if (!prison_allow(td->td_ucred, mp->mnt_vfc->vfc_prison_flag)) return (EPERM); /* * If the file system was mounted outside the jail of the * calling thread, deny immediately. */ if (prison_check(td->td_ucred, mp->mnt_cred) != 0) return (EPERM); } /* * If file system supports delegated administration, we don't check * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified * by the file system itself. * If this is not the user that did original mount, we check for * the PRIV_VFS_MOUNT_OWNER privilege. */ if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) return (error); } return (0); } /* * Get a new unique fsid. Try to make its val[0] unique, since this value * will be used to create fake device numbers for stat(). Also try (but * not so hard) make its val[0] unique mod 2^16, since some emulators only * support 16-bit device numbers. We end up with unique val[0]'s for the * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. * * Keep in mind that several mounts may be running in parallel. Starting * the search one past where the previous search terminated is both a * micro-optimization and a defense against returning the same fsid to * different mounts. */ void vfs_getnewfsid(struct mount *mp) { static uint16_t mntid_base; struct mount *nmp; fsid_t tfsid; int mtype; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); mtx_lock(&mntid_mtx); mtype = mp->mnt_vfc->vfc_typenum; tfsid.val[1] = mtype; mtype = (mtype & 0xFF) << 24; for (;;) { tfsid.val[0] = makedev(255, mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); mntid_base++; if ((nmp = vfs_getvfs(&tfsid)) == NULL) break; vfs_rel(nmp); } mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; mtx_unlock(&mntid_mtx); } /* * Knob to control the precision of file timestamps: * * 0 = seconds only; nanoseconds zeroed. * 1 = seconds and nanoseconds, accurate within 1/HZ. * 2 = seconds and nanoseconds, truncated to microseconds. * >=3 = seconds and nanoseconds, maximum precision. */ enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; static int timestamp_precision = TSP_USEC; SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, ×tamp_precision, 0, "File timestamp precision (0: seconds, " "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to us, " "3+: sec + ns (max. precision))"); /* * Get a current timestamp. */ void vfs_timestamp(struct timespec *tsp) { struct timeval tv; switch (timestamp_precision) { case TSP_SEC: tsp->tv_sec = time_second; tsp->tv_nsec = 0; break; case TSP_HZ: getnanotime(tsp); break; case TSP_USEC: microtime(&tv); TIMEVAL_TO_TIMESPEC(&tv, tsp); break; case TSP_NSEC: default: nanotime(tsp); break; } } /* * Set vnode attributes to VNOVAL */ void vattr_null(struct vattr *vap) { vap->va_type = VNON; vap->va_size = VNOVAL; vap->va_bytes = VNOVAL; vap->va_mode = VNOVAL; vap->va_nlink = VNOVAL; vap->va_uid = VNOVAL; vap->va_gid = VNOVAL; vap->va_fsid = VNOVAL; vap->va_fileid = VNOVAL; vap->va_blocksize = VNOVAL; vap->va_rdev = VNOVAL; vap->va_atime.tv_sec = VNOVAL; vap->va_atime.tv_nsec = VNOVAL; vap->va_mtime.tv_sec = VNOVAL; vap->va_mtime.tv_nsec = VNOVAL; vap->va_ctime.tv_sec = VNOVAL; vap->va_ctime.tv_nsec = VNOVAL; vap->va_birthtime.tv_sec = VNOVAL; vap->va_birthtime.tv_nsec = VNOVAL; vap->va_flags = VNOVAL; vap->va_gen = VNOVAL; vap->va_vaflags = 0; } /* * This routine is called when we have too many vnodes. It attempts * to free vnodes and will potentially free vnodes that still * have VM backing store (VM backing store is typically the cause * of a vnode blowout so we want to do this). Therefore, this operation * is not considered cheap. * * A number of conditions may prevent a vnode from being reclaimed. * the buffer cache may have references on the vnode, a directory * vnode may still have references due to the namei cache representing * underlying files, or the vnode may be in active use. It is not * desirable to reuse such vnodes. These conditions may cause the * number of vnodes to reach some minimum value regardless of what * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. */ static int vlrureclaim(struct mount *mp, int reclaim_nc_src, int trigger) { struct vnode *vp; int count, done, target; done = 0; vn_start_write(NULL, &mp, V_WAIT); MNT_ILOCK(mp); count = mp->mnt_nvnodelistsize; target = count * (int64_t)gapvnodes / imax(desiredvnodes, 1); target = target / 10 + 1; while (count != 0 && done < target) { vp = TAILQ_FIRST(&mp->mnt_nvnodelist); while (vp != NULL && vp->v_type == VMARKER) vp = TAILQ_NEXT(vp, v_nmntvnodes); if (vp == NULL) break; /* * XXX LRU is completely broken for non-free vnodes. First * by calling here in mountpoint order, then by moving * unselected vnodes to the end here, and most grossly by * removing the vlruvp() function that was supposed to * maintain the order. (This function was born broken * since syncer problems prevented it doing anything.) The * order is closer to LRC (C = Created). * * LRU reclaiming of vnodes seems to have last worked in * FreeBSD-3 where LRU wasn't mentioned under any spelling. * Then there was no hold count, and inactive vnodes were * simply put on the free list in LRU order. The separate * lists also break LRU. We prefer to reclaim from the * free list for technical reasons. This tends to thrash * the free list to keep very unrecently used held vnodes. * The problem is mitigated by keeping the free list large. */ TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); --count; if (!VI_TRYLOCK(vp)) goto next_iter; /* * If it's been deconstructed already, it's still * referenced, or it exceeds the trigger, skip it. * Also skip free vnodes. We are trying to make space * to expand the free list, not reduce it. */ if (vp->v_usecount || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || ((vp->v_iflag & VI_FREE) != 0) || (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && vp->v_object->resident_page_count > trigger)) { VI_UNLOCK(vp); goto next_iter; } MNT_IUNLOCK(mp); vholdl(vp); if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { vdrop(vp); goto next_iter_mntunlocked; } VI_LOCK(vp); /* * v_usecount may have been bumped after VOP_LOCK() dropped * the vnode interlock and before it was locked again. * * It is not necessary to recheck VI_DOOMED because it can * only be set by another thread that holds both the vnode * lock and vnode interlock. If another thread has the * vnode lock before we get to VOP_LOCK() and obtains the * vnode interlock after VOP_LOCK() drops the vnode * interlock, the other thread will be unable to drop the * vnode lock before our VOP_LOCK() call fails. */ if (vp->v_usecount || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || (vp->v_iflag & VI_FREE) != 0 || (vp->v_object != NULL && vp->v_object->resident_page_count > trigger)) { VOP_UNLOCK(vp, LK_INTERLOCK); vdrop(vp); goto next_iter_mntunlocked; } KASSERT((vp->v_iflag & VI_DOOMED) == 0, ("VI_DOOMED unexpectedly detected in vlrureclaim()")); counter_u64_add(recycles_count, 1); vgonel(vp); VOP_UNLOCK(vp, 0); vdropl(vp); done++; next_iter_mntunlocked: if (!should_yield()) goto relock_mnt; goto yield; next_iter: if (!should_yield()) continue; MNT_IUNLOCK(mp); yield: kern_yield(PRI_USER); relock_mnt: MNT_ILOCK(mp); } MNT_IUNLOCK(mp); vn_finished_write(mp); return done; } static int max_vnlru_free = 10000; /* limit on vnode free requests per call */ SYSCTL_INT(_debug, OID_AUTO, max_vnlru_free, CTLFLAG_RW, &max_vnlru_free, 0, "limit on vnode free requests per call to the vnlru_free routine"); /* * Attempt to reduce the free list by the requested amount. */ static void vnlru_free_locked(int count, struct vfsops *mnt_op) { struct vnode *vp; struct mount *mp; bool tried_batches; tried_batches = false; mtx_assert(&vnode_free_list_mtx, MA_OWNED); if (count > max_vnlru_free) count = max_vnlru_free; for (; count > 0; count--) { vp = TAILQ_FIRST(&vnode_free_list); /* * The list can be modified while the free_list_mtx * has been dropped and vp could be NULL here. */ if (vp == NULL) { if (tried_batches) break; mtx_unlock(&vnode_free_list_mtx); vnlru_return_batches(mnt_op); tried_batches = true; mtx_lock(&vnode_free_list_mtx); continue; } VNASSERT(vp->v_op != NULL, vp, ("vnlru_free: vnode already reclaimed.")); KASSERT((vp->v_iflag & VI_FREE) != 0, ("Removing vnode not on freelist")); KASSERT((vp->v_iflag & VI_ACTIVE) == 0, ("Mangling active vnode")); TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); /* * Don't recycle if our vnode is from different type * of mount point. Note that mp is type-safe, the * check does not reach unmapped address even if * vnode is reclaimed. * Don't recycle if we can't get the interlock without * blocking. */ if ((mnt_op != NULL && (mp = vp->v_mount) != NULL && mp->mnt_op != mnt_op) || !VI_TRYLOCK(vp)) { TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); continue; } VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0, vp, ("vp inconsistent on freelist")); /* * The clear of VI_FREE prevents activation of the * vnode. There is no sense in putting the vnode on * the mount point active list, only to remove it * later during recycling. Inline the relevant part * of vholdl(), to avoid triggering assertions or * activating. */ freevnodes--; vp->v_iflag &= ~VI_FREE; VNODE_REFCOUNT_FENCE_REL(); refcount_acquire(&vp->v_holdcnt); mtx_unlock(&vnode_free_list_mtx); VI_UNLOCK(vp); vtryrecycle(vp); /* * If the recycled succeeded this vdrop will actually free * the vnode. If not it will simply place it back on * the free list. */ vdrop(vp); mtx_lock(&vnode_free_list_mtx); } } void vnlru_free(int count, struct vfsops *mnt_op) { mtx_lock(&vnode_free_list_mtx); vnlru_free_locked(count, mnt_op); mtx_unlock(&vnode_free_list_mtx); } /* XXX some names and initialization are bad for limits and watermarks. */ static int vspace(void) { int space; gapvnodes = imax(desiredvnodes - wantfreevnodes, 100); vhiwat = gapvnodes / 11; /* 9% -- just under the 10% in vlrureclaim() */ vlowat = vhiwat / 2; if (numvnodes > desiredvnodes) return (0); space = desiredvnodes - numvnodes; if (freevnodes > wantfreevnodes) space += freevnodes - wantfreevnodes; return (space); } static void vnlru_return_batch_locked(struct mount *mp) { struct vnode *vp; mtx_assert(&mp->mnt_listmtx, MA_OWNED); if (mp->mnt_tmpfreevnodelistsize == 0) return; TAILQ_FOREACH(vp, &mp->mnt_tmpfreevnodelist, v_actfreelist) { VNASSERT((vp->v_mflag & VMP_TMPMNTFREELIST) != 0, vp, ("vnode without VMP_TMPMNTFREELIST on mnt_tmpfreevnodelist")); vp->v_mflag &= ~VMP_TMPMNTFREELIST; } mtx_lock(&vnode_free_list_mtx); TAILQ_CONCAT(&vnode_free_list, &mp->mnt_tmpfreevnodelist, v_actfreelist); freevnodes += mp->mnt_tmpfreevnodelistsize; mtx_unlock(&vnode_free_list_mtx); mp->mnt_tmpfreevnodelistsize = 0; } static void vnlru_return_batch(struct mount *mp) { mtx_lock(&mp->mnt_listmtx); vnlru_return_batch_locked(mp); mtx_unlock(&mp->mnt_listmtx); } static void vnlru_return_batches(struct vfsops *mnt_op) { struct mount *mp, *nmp; bool need_unbusy; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { need_unbusy = false; if (mnt_op != NULL && mp->mnt_op != mnt_op) goto next; if (mp->mnt_tmpfreevnodelistsize == 0) goto next; if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) == 0) { vnlru_return_batch(mp); need_unbusy = true; mtx_lock(&mountlist_mtx); } next: nmp = TAILQ_NEXT(mp, mnt_list); if (need_unbusy) vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); } /* * Attempt to recycle vnodes in a context that is always safe to block. * Calling vlrurecycle() from the bowels of filesystem code has some * interesting deadlock problems. */ static struct proc *vnlruproc; static int vnlruproc_sig; static void vnlru_proc(void) { struct mount *mp, *nmp; unsigned long onumvnodes; int done, force, reclaim_nc_src, trigger, usevnodes; EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, vnlruproc, SHUTDOWN_PRI_FIRST); force = 0; for (;;) { kproc_suspend_check(vnlruproc); mtx_lock(&vnode_free_list_mtx); /* * If numvnodes is too large (due to desiredvnodes being * adjusted using its sysctl, or emergency growth), first * try to reduce it by discarding from the free list. */ if (numvnodes > desiredvnodes) vnlru_free_locked(numvnodes - desiredvnodes, NULL); /* * Sleep if the vnode cache is in a good state. This is * when it is not over-full and has space for about a 4% * or 9% expansion (by growing its size or inexcessively * reducing its free list). Otherwise, try to reclaim * space for a 10% expansion. */ if (vstir && force == 0) { force = 1; vstir = 0; } if (vspace() >= vlowat && force == 0) { vnlruproc_sig = 0; wakeup(&vnlruproc_sig); msleep(vnlruproc, &vnode_free_list_mtx, PVFS|PDROP, "vlruwt", hz); continue; } mtx_unlock(&vnode_free_list_mtx); done = 0; onumvnodes = numvnodes; /* * Calculate parameters for recycling. These are the same * throughout the loop to give some semblance of fairness. * The trigger point is to avoid recycling vnodes with lots * of resident pages. We aren't trying to free memory; we * are trying to recycle or at least free vnodes. */ if (numvnodes <= desiredvnodes) usevnodes = numvnodes - freevnodes; else usevnodes = numvnodes; if (usevnodes <= 0) usevnodes = 1; /* * The trigger value is is chosen to give a conservatively * large value to ensure that it alone doesn't prevent * making progress. The value can easily be so large that * it is effectively infinite in some congested and * misconfigured cases, and this is necessary. Normally * it is about 8 to 100 (pages), which is quite large. */ trigger = vm_cnt.v_page_count * 2 / usevnodes; if (force < 2) trigger = vsmalltrigger; reclaim_nc_src = force >= 3; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } done += vlrureclaim(mp, reclaim_nc_src, trigger); mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); if (onumvnodes > desiredvnodes && numvnodes <= desiredvnodes) uma_reclaim(); if (done == 0) { if (force == 0 || force == 1) { force = 2; continue; } if (force == 2) { force = 3; continue; } force = 0; vnlru_nowhere++; tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); } else kern_yield(PRI_USER); /* * After becoming active to expand above low water, keep * active until above high water. */ force = vspace() < vhiwat; } } static struct kproc_desc vnlru_kp = { "vnlru", vnlru_proc, &vnlruproc }; SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); /* * Routines having to do with the management of the vnode table. */ /* * Try to recycle a freed vnode. We abort if anyone picks up a reference * before we actually vgone(). This function must be called with the vnode * held to prevent the vnode from being returned to the free list midway * through vgone(). */ static int vtryrecycle(struct vnode *vp) { struct mount *vnmp; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNASSERT(vp->v_holdcnt, vp, ("vtryrecycle: Recycling vp %p without a reference.", vp)); /* * This vnode may found and locked via some other list, if so we * can't recycle it yet. */ if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { CTR2(KTR_VFS, "%s: impossible to recycle, vp %p lock is already held", __func__, vp); return (EWOULDBLOCK); } /* * Don't recycle if its filesystem is being suspended. */ if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { VOP_UNLOCK(vp, 0); CTR2(KTR_VFS, "%s: impossible to recycle, cannot start the write for %p", __func__, vp); return (EBUSY); } /* * If we got this far, we need to acquire the interlock and see if * anyone picked up this vnode from another list. If not, we will * mark it with DOOMED via vgonel() so that anyone who does find it * will skip over it. */ VI_LOCK(vp); if (vp->v_usecount) { VOP_UNLOCK(vp, LK_INTERLOCK); vn_finished_write(vnmp); CTR2(KTR_VFS, "%s: impossible to recycle, %p is already referenced", __func__, vp); return (EBUSY); } if ((vp->v_iflag & VI_DOOMED) == 0) { counter_u64_add(recycles_count, 1); vgonel(vp); } VOP_UNLOCK(vp, LK_INTERLOCK); vn_finished_write(vnmp); return (0); } static void vcheckspace(void) { if (vspace() < vlowat && vnlruproc_sig == 0) { vnlruproc_sig = 1; wakeup(vnlruproc); } } /* * Wait if necessary for space for a new vnode. */ static int getnewvnode_wait(int suspended) { mtx_assert(&vnode_free_list_mtx, MA_OWNED); if (numvnodes >= desiredvnodes) { if (suspended) { /* * The file system is being suspended. We cannot * risk a deadlock here, so allow allocation of * another vnode even if this would give too many. */ return (0); } if (vnlruproc_sig == 0) { vnlruproc_sig = 1; /* avoid unnecessary wakeups */ wakeup(vnlruproc); } msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, "vlruwk", hz); } /* Post-adjust like the pre-adjust in getnewvnode(). */ if (numvnodes + 1 > desiredvnodes && freevnodes > 1) vnlru_free_locked(1, NULL); return (numvnodes >= desiredvnodes ? ENFILE : 0); } /* * This hack is fragile, and probably not needed any more now that the * watermark handling works. */ void getnewvnode_reserve(u_int count) { struct thread *td; /* Pre-adjust like the pre-adjust in getnewvnode(), with any count. */ /* XXX no longer so quick, but this part is not racy. */ mtx_lock(&vnode_free_list_mtx); if (numvnodes + count > desiredvnodes && freevnodes > wantfreevnodes) vnlru_free_locked(ulmin(numvnodes + count - desiredvnodes, freevnodes - wantfreevnodes), NULL); mtx_unlock(&vnode_free_list_mtx); td = curthread; /* First try to be quick and racy. */ if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) { td->td_vp_reserv += count; vcheckspace(); /* XXX no longer so quick, but more racy */ return; } else atomic_subtract_long(&numvnodes, count); mtx_lock(&vnode_free_list_mtx); while (count > 0) { if (getnewvnode_wait(0) == 0) { count--; td->td_vp_reserv++; atomic_add_long(&numvnodes, 1); } } vcheckspace(); mtx_unlock(&vnode_free_list_mtx); } /* * This hack is fragile, especially if desiredvnodes or wantvnodes are * misconfgured or changed significantly. Reducing desiredvnodes below * the reserved amount should cause bizarre behaviour like reducing it * below the number of active vnodes -- the system will try to reduce * numvnodes to match, but should fail, so the subtraction below should * not overflow. */ void getnewvnode_drop_reserve(void) { struct thread *td; td = curthread; atomic_subtract_long(&numvnodes, td->td_vp_reserv); td->td_vp_reserv = 0; } /* * Return the next vnode from the free list. */ int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp) { struct vnode *vp; struct thread *td; struct lock_object *lo; static int cyclecount; int error __unused; CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); vp = NULL; td = curthread; if (td->td_vp_reserv > 0) { td->td_vp_reserv -= 1; goto alloc; } mtx_lock(&vnode_free_list_mtx); if (numvnodes < desiredvnodes) cyclecount = 0; else if (cyclecount++ >= freevnodes) { cyclecount = 0; vstir = 1; } /* * Grow the vnode cache if it will not be above its target max * after growing. Otherwise, if the free list is nonempty, try * to reclaim 1 item from it before growing the cache (possibly * above its target max if the reclamation failed or is delayed). * Otherwise, wait for some space. In all cases, schedule * vnlru_proc() if we are getting short of space. The watermarks * should be chosen so that we never wait or even reclaim from * the free list to below its target minimum. */ if (numvnodes + 1 <= desiredvnodes) ; else if (freevnodes > 0) vnlru_free_locked(1, NULL); else { error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)); #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ if (error != 0) { mtx_unlock(&vnode_free_list_mtx); return (error); } #endif } vcheckspace(); atomic_add_long(&numvnodes, 1); mtx_unlock(&vnode_free_list_mtx); alloc: counter_u64_add(vnodes_created, 1); vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK); /* * Locks are given the generic name "vnode" when created. * Follow the historic practice of using the filesystem * name when they allocated, e.g., "zfs", "ufs", "nfs, etc. * * Locks live in a witness group keyed on their name. Thus, * when a lock is renamed, it must also move from the witness * group of its old name to the witness group of its new name. * * The change only needs to be made when the vnode moves * from one filesystem type to another. We ensure that each * filesystem use a single static name pointer for its tag so * that we can compare pointers rather than doing a strcmp(). */ lo = &vp->v_vnlock->lock_object; if (lo->lo_name != tag) { lo->lo_name = tag; WITNESS_DESTROY(lo); WITNESS_INIT(lo, tag); } /* * By default, don't allow shared locks unless filesystems opt-in. */ vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE; /* * Finalize various vnode identity bits. */ KASSERT(vp->v_object == NULL, ("stale v_object %p", vp)); KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp)); KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp)); vp->v_type = VNON; vp->v_tag = tag; vp->v_op = vops; v_init_counters(vp); vp->v_bufobj.bo_ops = &buf_ops_bio; #ifdef DIAGNOSTIC if (mp == NULL && vops != &dead_vnodeops) printf("NULL mp in getnewvnode(9), tag %s\n", tag); #endif #ifdef MAC mac_vnode_init(vp); if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) mac_vnode_associate_singlelabel(mp, vp); #endif if (mp != NULL) { vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize; if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) vp->v_vflag |= VV_NOKNOTE; } /* * For the filesystems which do not use vfs_hash_insert(), * still initialize v_hash to have vfs_hash_index() useful. * E.g., nullfs uses vfs_hash_index() on the lower vnode for * its own hashing. */ vp->v_hash = (uintptr_t)vp >> vnsz2log; *vpp = vp; return (0); } /* * Delete from old mount point vnode list, if on one. */ static void delmntque(struct vnode *vp) { struct mount *mp; int active; mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); VI_LOCK(vp); KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize, ("Active vnode list size %d > Vnode list size %d", mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize)); active = vp->v_iflag & VI_ACTIVE; vp->v_iflag &= ~VI_ACTIVE; if (active) { mtx_lock(&mp->mnt_listmtx); TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize--; mtx_unlock(&mp->mnt_listmtx); } vp->v_mount = NULL; VI_UNLOCK(vp); VNASSERT(mp->mnt_nvnodelistsize > 0, vp, ("bad mount point vnode list size")); TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); mp->mnt_nvnodelistsize--; MNT_REL(mp); MNT_IUNLOCK(mp); } static void insmntque_stddtr(struct vnode *vp, void *dtr_arg) { vp->v_data = NULL; vp->v_op = &dead_vnodeops; vgone(vp); vput(vp); } /* * Insert into list of vnodes for the new mount point, if available. */ int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg) { KASSERT(vp->v_mount == NULL, ("insmntque: vnode already on per mount vnode list")); VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp"); /* * We acquire the vnode interlock early to ensure that the * vnode cannot be recycled by another process releasing a * holdcnt on it before we get it on both the vnode list * and the active vnode list. The mount mutex protects only * manipulation of the vnode list and the vnode freelist * mutex protects only manipulation of the active vnode list. * Hence the need to hold the vnode interlock throughout. */ MNT_ILOCK(mp); VI_LOCK(vp); if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || mp->mnt_nvnodelistsize == 0)) && (vp->v_vflag & VV_FORCEINSMQ) == 0) { VI_UNLOCK(vp); MNT_IUNLOCK(mp); if (dtr != NULL) dtr(vp, dtr_arg); return (EBUSY); } vp->v_mount = mp; MNT_REF(mp); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, ("neg mount point vnode list size")); mp->mnt_nvnodelistsize++; KASSERT((vp->v_iflag & VI_ACTIVE) == 0, ("Activating already active vnode")); vp->v_iflag |= VI_ACTIVE; mtx_lock(&mp->mnt_listmtx); TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize++; mtx_unlock(&mp->mnt_listmtx); VI_UNLOCK(vp); MNT_IUNLOCK(mp); return (0); } int insmntque(struct vnode *vp, struct mount *mp) { return (insmntque1(vp, mp, insmntque_stddtr, NULL)); } /* * Flush out and invalidate all buffers associated with a bufobj * Called with the underlying object locked. */ int bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) { int error; BO_LOCK(bo); if (flags & V_SAVE) { error = bufobj_wwait(bo, slpflag, slptimeo); if (error) { BO_UNLOCK(bo); return (error); } if (bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) return (error); /* * XXX We could save a lock/unlock if this was only * enabled under INVARIANTS */ BO_LOCK(bo); if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: dirty bufs"); } } /* * If you alter this loop please notice that interlock is dropped and * reacquired in flushbuflist. Special care is needed to ensure that * no race conditions occur from this. */ do { error = flushbuflist(&bo->bo_clean, flags, bo, slpflag, slptimeo); if (error == 0 && !(flags & V_CLEANONLY)) error = flushbuflist(&bo->bo_dirty, flags, bo, slpflag, slptimeo); if (error != 0 && error != EAGAIN) { BO_UNLOCK(bo); return (error); } } while (error != 0); /* * Wait for I/O to complete. XXX needs cleaning up. The vnode can * have write I/O in-progress but if there is a VM object then the * VM object can also have read-I/O in-progress. */ do { bufobj_wwait(bo, 0, 0); if ((flags & V_VMIO) == 0) { BO_UNLOCK(bo); if (bo->bo_object != NULL) { VM_OBJECT_WLOCK(bo->bo_object); vm_object_pip_wait(bo->bo_object, "bovlbx"); VM_OBJECT_WUNLOCK(bo->bo_object); } BO_LOCK(bo); } } while (bo->bo_numoutput > 0); BO_UNLOCK(bo); /* * Destroy the copy in the VM cache, too. */ if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0) { VM_OBJECT_WLOCK(bo->bo_object); vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ? OBJPR_CLEANONLY : 0); VM_OBJECT_WUNLOCK(bo->bo_object); } #ifdef INVARIANTS BO_LOCK(bo); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO | V_ALLOWCLEAN)) == 0 && (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) panic("vinvalbuf: flush failed"); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0 && bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: flush dirty failed"); BO_UNLOCK(bo); #endif return (0); } /* * Flush out and invalidate all buffers associated with a vnode. * Called with the underlying object locked. */ int vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) { CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); ASSERT_VOP_LOCKED(vp, "vinvalbuf"); if (vp->v_object != NULL && vp->v_object->handle != vp) return (0); return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); } /* * Flush out buffers on the specified list. * */ static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo) { struct buf *bp, *nbp; int retval, error; daddr_t lblkno; b_xflags_t xflags; ASSERT_BO_WLOCKED(bo); retval = 0; TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { continue; } if (nbp != NULL) { lblkno = nbp->b_lblkno; xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN); } retval = EAGAIN; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "flushbuf", slpflag, slptimeo); if (error) { BO_LOCK(bo); return (error != ENOLCK ? error : EAGAIN); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); /* * XXX Since there are no node locks for NFS, I * believe there is a slight chance that a delayed * write will occur while sleeping just above, so * check for it. */ if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && (flags & V_SAVE)) { bremfree(bp); bp->b_flags |= B_ASYNC; bwrite(bp); BO_LOCK(bo); return (EAGAIN); /* XXX: why not loop ? */ } bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); BO_LOCK(bo); if (nbp == NULL) break; nbp = gbincore(bo, lblkno); if (nbp == NULL || (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags) break; /* nbp invalid */ } return (retval); } int bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn) { struct buf *bp; int error; daddr_t lblkno; ASSERT_BO_LOCKED(bo); for (lblkno = startn;;) { again: bp = BUF_PCTRIE_LOOKUP_GE(&bufv->bv_root, lblkno); if (bp == NULL || bp->b_lblkno >= endn || bp->b_lblkno < startn) break; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "brlsfl", 0, 0); if (error != 0) { BO_RLOCK(bo); if (error == ENOLCK) goto again; return (error); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); lblkno = bp->b_lblkno + 1; if ((bp->b_flags & B_MANAGED) == 0) bremfree(bp); bp->b_flags |= B_RELBUF; /* * In the VMIO case, use the B_NOREUSE flag to hint that the * pages backing each buffer in the range are unlikely to be * reused. Dirty buffers will have the hint applied once * they've been written. */ if ((bp->b_flags & B_VMIO) != 0) bp->b_flags |= B_NOREUSE; brelse(bp); BO_RLOCK(bo); } return (0); } /* * Truncate a file's buffer and pages to a specified length. This * is in lieu of the old vinvalbuf mechanism, which performed unneeded * sync activity. */ int vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize) { struct buf *bp, *nbp; int anyfreed; int trunclbn; struct bufobj *bo; CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__, vp, cred, blksize, (uintmax_t)length); /* * Round up to the *next* lbn. */ trunclbn = howmany(length, blksize); ASSERT_VOP_LOCKED(vp, "vtruncbuf"); restart: bo = &vp->v_bufobj; BO_LOCK(bo); anyfreed = 1; for (;anyfreed;) { anyfreed = 0; TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < trunclbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) goto restart; bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = 1; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNCLEAN) == 0) || (nbp->b_vp != vp) || (nbp->b_flags & B_DELWRI))) { BO_UNLOCK(bo); goto restart; } } TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < trunclbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) goto restart; bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = 1; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNDIRTY) == 0) || (nbp->b_vp != vp) || (nbp->b_flags & B_DELWRI) == 0)) { BO_UNLOCK(bo); goto restart; } } } if (length > 0) { restartsync: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno > 0) continue; /* * Since we hold the vnode lock this should only * fail if we're racing with the buf daemon. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { goto restart; } VNASSERT((bp->b_flags & B_DELWRI), vp, ("buf(%p) on dirty queue without DELWRI", bp)); bremfree(bp); bawrite(bp); BO_LOCK(bo); goto restartsync; } } bufobj_wwait(bo, 0, 0); BO_UNLOCK(bo); vnode_pager_setsize(vp, length); return (0); } static void buf_vlist_remove(struct buf *bp) { struct bufv *bv; KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); ASSERT_BO_WLOCKED(bp->b_bufobj); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != (BX_VNDIRTY|BX_VNCLEAN), ("buf_vlist_remove: Buf %p is on two lists", bp)); if (bp->b_xflags & BX_VNDIRTY) bv = &bp->b_bufobj->bo_dirty; else bv = &bp->b_bufobj->bo_clean; BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno); TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); bv->bv_cnt--; bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); } /* * Add the buffer to the sorted clean or dirty block list. * * NOTE: xflags is passed as a constant, optimizing this inline function! */ static void buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) { struct bufv *bv; struct buf *n; int error; ASSERT_BO_WLOCKED(bo); KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo)); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); bp->b_xflags |= xflags; if (xflags & BX_VNDIRTY) bv = &bo->bo_dirty; else bv = &bo->bo_clean; /* * Keep the list ordered. Optimize empty list insertion. Assume * we tend to grow at the tail so lookup_le should usually be cheaper * than _ge. */ if (bv->bv_cnt == 0 || bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno) TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL) TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs); else TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs); error = BUF_PCTRIE_INSERT(&bv->bv_root, bp); if (error) panic("buf_vlist_add: Preallocated nodes insufficient."); bv->bv_cnt++; } /* * Look up a buffer using the buffer tries. */ struct buf * gbincore(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_LOCKED(bo); bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno); } /* * Associate a buffer with a vnode. */ void bgetvp(struct vnode *vp, struct buf *bp) { struct bufobj *bo; bo = &vp->v_bufobj; ASSERT_BO_WLOCKED(bo); VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, ("bgetvp: bp already attached! %p", bp)); vhold(vp); bp->b_vp = vp; bp->b_bufobj = bo; /* * Insert onto list for new vnode. */ buf_vlist_add(bp, bo, BX_VNCLEAN); } /* * Disassociate a buffer from a vnode. */ void brelvp(struct buf *bp) { struct bufobj *bo; struct vnode *vp; CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); /* * Delete from old vnode list, if on one. */ vp = bp->b_vp; /* XXX */ bo = bp->b_bufobj; BO_LOCK(bo); if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) buf_vlist_remove(bp); else panic("brelvp: Buffer %p not on queue.", bp); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { bo->bo_flag &= ~BO_ONWORKLST; mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); } bp->b_vp = NULL; bp->b_bufobj = NULL; BO_UNLOCK(bo); vdrop(vp); } /* * Add an item to the syncer work queue. */ static void vn_syncer_add_to_worklist(struct bufobj *bo, int delay) { int slot; ASSERT_BO_WLOCKED(bo); mtx_lock(&sync_mtx); if (bo->bo_flag & BO_ONWORKLST) LIST_REMOVE(bo, bo_synclist); else { bo->bo_flag |= BO_ONWORKLST; syncer_worklist_len++; } if (delay > syncer_maxdelay - 2) delay = syncer_maxdelay - 2; slot = (syncer_delayno + delay) & syncer_mask; LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist); mtx_unlock(&sync_mtx); } static int sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) { int error, len; mtx_lock(&sync_mtx); len = syncer_worklist_len - sync_vnode_count; mtx_unlock(&sync_mtx); error = SYSCTL_OUT(req, &len, sizeof(len)); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); static struct proc *updateproc; static void sched_sync(void); static struct kproc_desc up_kp = { "syncer", sched_sync, &updateproc }; SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); static int sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) { struct vnode *vp; struct mount *mp; *bo = LIST_FIRST(slp); if (*bo == NULL) return (0); vp = bo2vnode(*bo); if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) return (1); /* * We use vhold in case the vnode does not * successfully sync. vhold prevents the vnode from * going away when we unlock the sync_mtx so that * we can acquire the vnode interlock. */ vholdl(vp); mtx_unlock(&sync_mtx); VI_UNLOCK(vp); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); mtx_lock(&sync_mtx); return (*bo == LIST_FIRST(slp)); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); (void) VOP_FSYNC(vp, MNT_LAZY, td); VOP_UNLOCK(vp, 0); vn_finished_write(mp); BO_LOCK(*bo); if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { /* * Put us back on the worklist. The worklist * routine will remove us from our current * position and then add us back in at a later * position. */ vn_syncer_add_to_worklist(*bo, syncdelay); } BO_UNLOCK(*bo); vdrop(vp); mtx_lock(&sync_mtx); return (0); } static int first_printf = 1; /* * System filesystem synchronizer daemon. */ static void sched_sync(void) { struct synclist *next, *slp; struct bufobj *bo; long starttime; struct thread *td = curthread; int last_work_seen; int net_worklist_len; int syncer_final_iter; int error; last_work_seen = 0; syncer_final_iter = 0; syncer_state = SYNCER_RUNNING; starttime = time_uptime; td->td_pflags |= TDP_NORUNNINGBUF; EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, SHUTDOWN_PRI_LAST); mtx_lock(&sync_mtx); for (;;) { if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter == 0) { mtx_unlock(&sync_mtx); kproc_suspend_check(td->td_proc); mtx_lock(&sync_mtx); } net_worklist_len = syncer_worklist_len - sync_vnode_count; if (syncer_state != SYNCER_RUNNING && starttime != time_uptime) { if (first_printf) { printf("\nSyncing disks, vnodes remaining... "); first_printf = 0; } printf("%d ", net_worklist_len); } starttime = time_uptime; /* * Push files whose dirty time has expired. Be careful * of interrupt race on slp queue. * * Skip over empty worklist slots when shutting down. */ do { slp = &syncer_workitem_pending[syncer_delayno]; syncer_delayno += 1; if (syncer_delayno == syncer_maxdelay) syncer_delayno = 0; next = &syncer_workitem_pending[syncer_delayno]; /* * If the worklist has wrapped since the * it was emptied of all but syncer vnodes, * switch to the FINAL_DELAY state and run * for one more second. */ if (syncer_state == SYNCER_SHUTTING_DOWN && net_worklist_len == 0 && last_work_seen == syncer_delayno) { syncer_state = SYNCER_FINAL_DELAY; syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; } } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && syncer_worklist_len > 0); /* * Keep track of the last time there was anything * on the worklist other than syncer vnodes. * Return to the SHUTTING_DOWN state if any * new work appears. */ if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) last_work_seen = syncer_delayno; if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) syncer_state = SYNCER_SHUTTING_DOWN; while (!LIST_EMPTY(slp)) { error = sync_vnode(slp, &bo, td); if (error == 1) { LIST_REMOVE(bo, bo_synclist); LIST_INSERT_HEAD(next, bo, bo_synclist); continue; } if (first_printf == 0) { /* * Drop the sync mutex, because some watchdog * drivers need to sleep while patting */ mtx_unlock(&sync_mtx); wdog_kern_pat(WD_LASTVAL); mtx_lock(&sync_mtx); } } if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) syncer_final_iter--; /* * The variable rushjob allows the kernel to speed up the * processing of the filesystem syncer process. A rushjob * value of N tells the filesystem syncer to process the next * N seconds worth of work on its queue ASAP. Currently rushjob * is used by the soft update code to speed up the filesystem * syncer process when the incore state is getting so far * ahead of the disk that the kernel memory pool is being * threatened with exhaustion. */ if (rushjob > 0) { rushjob -= 1; continue; } /* * Just sleep for a short period of time between * iterations when shutting down to allow some I/O * to happen. * * If it has taken us less than a second to process the * current work, then wait. Otherwise start right over * again. We can still lose time if any single round * takes more than two seconds, but it does not really * matter as we are just trying to generally pace the * filesystem activity. */ if (syncer_state != SYNCER_RUNNING || time_uptime == starttime) { thread_lock(td); sched_prio(td, PPAUSE); thread_unlock(td); } if (syncer_state != SYNCER_RUNNING) cv_timedwait(&sync_wakeup, &sync_mtx, hz / SYNCER_SHUTDOWN_SPEEDUP); else if (time_uptime == starttime) cv_timedwait(&sync_wakeup, &sync_mtx, hz); } } /* * Request the syncer daemon to speed up its work. * We never push it to speed up more than half of its * normal turn time, otherwise it could take over the cpu. */ int speedup_syncer(void) { int ret = 0; mtx_lock(&sync_mtx); if (rushjob < syncdelay / 2) { rushjob += 1; stat_rush_requests += 1; ret = 1; } mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); return (ret); } /* * Tell the syncer to speed up its work and run though its work * list several times, then tell it to shut down. */ static void syncer_shutdown(void *arg, int howto) { if (howto & RB_NOSYNC) return; mtx_lock(&sync_mtx); syncer_state = SYNCER_SHUTTING_DOWN; rushjob = 0; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_shutdown(arg, howto); } void syncer_suspend(void) { syncer_shutdown(updateproc, 0); } void syncer_resume(void) { mtx_lock(&sync_mtx); first_printf = 1; syncer_state = SYNCER_RUNNING; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_resume(updateproc); } /* * Reassign a buffer from one vnode to another. * Used to assign file specific control information * (indirect blocks) to the vnode to which they belong. */ void reassignbuf(struct buf *bp) { struct vnode *vp; struct bufobj *bo; int delay; #ifdef INVARIANTS struct bufv *bv; #endif vp = bp->b_vp; bo = bp->b_bufobj; ++reassignbufcalls; CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); /* * B_PAGING flagged buffers cannot be reassigned because their vp * is not fully linked in. */ if (bp->b_flags & B_PAGING) panic("cannot reassign paging buffer"); /* * Delete from old vnode list, if on one. */ BO_LOCK(bo); if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) buf_vlist_remove(bp); else panic("reassignbuf: Buffer %p not on queue.", bp); /* * If dirty, put on list of dirty buffers; otherwise insert onto list * of clean buffers. */ if (bp->b_flags & B_DELWRI) { if ((bo->bo_flag & BO_ONWORKLST) == 0) { switch (vp->v_type) { case VDIR: delay = dirdelay; break; case VCHR: delay = metadelay; break; default: delay = filedelay; } vn_syncer_add_to_worklist(bo, delay); } buf_vlist_add(bp, bo, BX_VNDIRTY); } else { buf_vlist_add(bp, bo, BX_VNCLEAN); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); bo->bo_flag &= ~BO_ONWORKLST; } } #ifdef INVARIANTS bv = &bo->bo_clean; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bv = &bo->bo_dirty; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); #endif BO_UNLOCK(bo); } static void v_init_counters(struct vnode *vp) { VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0, vp, ("%s called for an initialized vnode", __FUNCTION__)); ASSERT_VI_UNLOCKED(vp, __FUNCTION__); refcount_init(&vp->v_holdcnt, 1); refcount_init(&vp->v_usecount, 1); } static void v_incr_usecount_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); if ((vp->v_iflag & VI_OWEINACT) != 0) { VNASSERT(vp->v_usecount == 0, vp, ("vnode with usecount and VI_OWEINACT set")); vp->v_iflag &= ~VI_OWEINACT; } refcount_acquire(&vp->v_usecount); v_incr_devcount(vp); } /* * Increment the use count on the vnode, taking care to reference * the driver's usecount if this is a chardev. */ static void v_incr_usecount(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (vp->v_type != VCHR && refcount_acquire_if_not_zero(&vp->v_usecount)) { VNODE_REFCOUNT_FENCE_ACQ(); VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, ("vnode with usecount and VI_OWEINACT set")); } else { VI_LOCK(vp); v_incr_usecount_locked(vp); VI_UNLOCK(vp); } } /* * Increment si_usecount of the associated device, if any. */ static void v_incr_devcount(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __FUNCTION__); if (vp->v_type == VCHR && vp->v_rdev != NULL) { dev_lock(); vp->v_rdev->si_usecount++; dev_unlock(); } } /* * Decrement si_usecount of the associated device, if any. */ static void v_decr_devcount(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __FUNCTION__); if (vp->v_type == VCHR && vp->v_rdev != NULL) { dev_lock(); vp->v_rdev->si_usecount--; dev_unlock(); } } /* * Grab a particular vnode from the free list, increment its * reference count and lock it. VI_DOOMED is set if the vnode * is being destroyed. Only callers who specify LK_RETRY will * see doomed vnodes. If inactive processing was delayed in * vput try to do it here. * * Notes on lockless counter manipulation: * _vhold, vputx and other routines make various decisions based * on either holdcnt or usecount being 0. As long as either counter * is not transitioning 0->1 nor 1->0, the manipulation can be done * with atomic operations. Otherwise the interlock is taken covering * both the atomic and additional actions. */ int vget(struct vnode *vp, int flags, struct thread *td) { int error, oweinact; VNASSERT((flags & LK_TYPE_MASK) != 0, vp, ("vget: invalid lock operation")); if ((flags & LK_INTERLOCK) != 0) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); if ((flags & LK_VNHELD) != 0) VNASSERT((vp->v_holdcnt > 0), vp, ("vget: LK_VNHELD passed but vnode not held")); CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); if ((flags & LK_VNHELD) == 0) _vhold(vp, (flags & LK_INTERLOCK) != 0); if ((error = vn_lock(vp, flags)) != 0) { vdrop(vp); CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, vp); return (error); } if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) panic("vget: vn_lock failed to return ENOENT\n"); /* * We don't guarantee that any particular close will * trigger inactive processing so just make a best effort * here at preventing a reference to a removed file. If * we don't succeed no harm is done. * * Upgrade our holdcnt to a usecount. */ if (vp->v_type == VCHR || !refcount_acquire_if_not_zero(&vp->v_usecount)) { VI_LOCK(vp); if ((vp->v_iflag & VI_OWEINACT) == 0) { oweinact = 0; } else { oweinact = 1; vp->v_iflag &= ~VI_OWEINACT; VNODE_REFCOUNT_FENCE_REL(); } refcount_acquire(&vp->v_usecount); v_incr_devcount(vp); if (oweinact && VOP_ISLOCKED(vp) == LK_EXCLUSIVE && (flags & LK_NOWAIT) == 0) vinactive(vp, td); VI_UNLOCK(vp); } return (0); } /* * Increase the reference (use) and hold count of a vnode. * This will also remove the vnode from the free list if it is presently free. */ void vref(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); _vhold(vp, false); v_incr_usecount(vp); } void vrefl(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); _vhold(vp, true); v_incr_usecount_locked(vp); } void vrefact(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (__predict_false(vp->v_type == VCHR)) { VNASSERT(vp->v_holdcnt > 0 && vp->v_usecount > 0, vp, ("%s: wrong ref counts", __func__)); vref(vp); return; } #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old > 0, vp, ("%s: wrong hold count", __func__)); old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old > 0, vp, ("%s: wrong use count", __func__)); #else refcount_acquire(&vp->v_holdcnt); refcount_acquire(&vp->v_usecount); #endif } /* * Return reference count of a vnode. * * The results of this call are only guaranteed when some mechanism is used to * stop other processes from gaining references to the vnode. This may be the * case if the caller holds the only reference. This is also useful when stale * data is acceptable as race conditions may be accounted for by some other * means. */ int vrefcnt(struct vnode *vp) { return (vp->v_usecount); } #define VPUTX_VRELE 1 #define VPUTX_VPUT 2 #define VPUTX_VUNREF 3 /* * Decrement the use and hold counts for a vnode. * * See an explanation near vget() as to why atomic operation is safe. */ static void vputx(struct vnode *vp, int func) { int error; KASSERT(vp != NULL, ("vputx: null vp")); if (func == VPUTX_VUNREF) ASSERT_VOP_LOCKED(vp, "vunref"); else if (func == VPUTX_VPUT) ASSERT_VOP_LOCKED(vp, "vput"); else KASSERT(func == VPUTX_VRELE, ("vputx: wrong func")); ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (vp->v_type != VCHR && refcount_release_if_not_last(&vp->v_usecount)) { if (func == VPUTX_VPUT) VOP_UNLOCK(vp, 0); vdrop(vp); return; } VI_LOCK(vp); /* * We want to hold the vnode until the inactive finishes to * prevent vgone() races. We drop the use count here and the * hold count below when we're done. */ if (!refcount_release(&vp->v_usecount) || (vp->v_iflag & VI_DOINGINACT)) { if (func == VPUTX_VPUT) VOP_UNLOCK(vp, 0); v_decr_devcount(vp); vdropl(vp); return; } v_decr_devcount(vp); error = 0; if (vp->v_usecount != 0) { vn_printf(vp, "vputx: usecount not zero for vnode "); panic("vputx: usecount not zero"); } CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp); /* * We must call VOP_INACTIVE with the node locked. Mark * as VI_DOINGINACT to avoid recursion. */ vp->v_iflag |= VI_OWEINACT; switch (func) { case VPUTX_VRELE: error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); VI_LOCK(vp); break; case VPUTX_VPUT: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT); VI_LOCK(vp); } break; case VPUTX_VUNREF: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK); VI_LOCK(vp); } break; } VNASSERT(vp->v_usecount == 0 || (vp->v_iflag & VI_OWEINACT) == 0, vp, ("vnode with usecount and VI_OWEINACT set")); if (error == 0) { if (vp->v_iflag & VI_OWEINACT) vinactive(vp, curthread); if (func != VPUTX_VUNREF) VOP_UNLOCK(vp, 0); } vdropl(vp); } /* * Vnode put/release. * If count drops to zero, call inactive routine and return to freelist. */ void vrele(struct vnode *vp) { vputx(vp, VPUTX_VRELE); } /* * Release an already locked vnode. This give the same effects as * unlock+vrele(), but takes less time and avoids releasing and * re-aquiring the lock (as vrele() acquires the lock internally.) */ void vput(struct vnode *vp) { vputx(vp, VPUTX_VPUT); } /* * Release an exclusively locked vnode. Do not unlock the vnode lock. */ void vunref(struct vnode *vp) { vputx(vp, VPUTX_VUNREF); } /* * Increase the hold count and activate if this is the first reference. */ void _vhold(struct vnode *vp, bool locked) { struct mount *mp; if (locked) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (!locked) { if (refcount_acquire_if_not_zero(&vp->v_holdcnt)) { VNODE_REFCOUNT_FENCE_ACQ(); VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("_vhold: vnode with holdcnt is free")); return; } VI_LOCK(vp); } if ((vp->v_iflag & VI_FREE) == 0) { refcount_acquire(&vp->v_holdcnt); if (!locked) VI_UNLOCK(vp); return; } VNASSERT(vp->v_holdcnt == 0, vp, ("%s: wrong hold count", __func__)); VNASSERT(vp->v_op != NULL, vp, ("%s: vnode already reclaimed.", __func__)); /* * Remove a vnode from the free list, mark it as in use, * and put it on the active list. */ VNASSERT(vp->v_mount != NULL, vp, ("_vhold: vnode not on per mount vnode list")); mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); if ((vp->v_mflag & VMP_TMPMNTFREELIST) != 0) { TAILQ_REMOVE(&mp->mnt_tmpfreevnodelist, vp, v_actfreelist); mp->mnt_tmpfreevnodelistsize--; vp->v_mflag &= ~VMP_TMPMNTFREELIST; } else { mtx_lock(&vnode_free_list_mtx); TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); freevnodes--; mtx_unlock(&vnode_free_list_mtx); } KASSERT((vp->v_iflag & VI_ACTIVE) == 0, ("Activating already active vnode")); vp->v_iflag &= ~VI_FREE; vp->v_iflag |= VI_ACTIVE; TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize++; mtx_unlock(&mp->mnt_listmtx); refcount_acquire(&vp->v_holdcnt); if (!locked) VI_UNLOCK(vp); } /* * Drop the hold count of the vnode. If this is the last reference to * the vnode we place it on the free list unless it has been vgone'd * (marked VI_DOOMED) in which case we will free it. * * Because the vnode vm object keeps a hold reference on the vnode if * there is at least one resident non-cached page, the vnode cannot * leave the active list without the page cleanup done. */ void _vdrop(struct vnode *vp, bool locked) { struct bufobj *bo; struct mount *mp; int active; if (locked) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if ((int)vp->v_holdcnt <= 0) panic("vdrop: holdcnt %d", vp->v_holdcnt); if (!locked) { if (refcount_release_if_not_last(&vp->v_holdcnt)) return; VI_LOCK(vp); } if (refcount_release(&vp->v_holdcnt) == 0) { VI_UNLOCK(vp); return; } if ((vp->v_iflag & VI_DOOMED) == 0) { /* * Mark a vnode as free: remove it from its active list * and put it up for recycling on the freelist. */ VNASSERT(vp->v_op != NULL, vp, ("vdropl: vnode already reclaimed.")); VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free")); VNASSERT(vp->v_holdcnt == 0, vp, ("vdropl: freeing when we shouldn't")); active = vp->v_iflag & VI_ACTIVE; if ((vp->v_iflag & VI_OWEINACT) == 0) { vp->v_iflag &= ~VI_ACTIVE; mp = vp->v_mount; if (mp != NULL) { mtx_lock(&mp->mnt_listmtx); if (active) { TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize--; } TAILQ_INSERT_TAIL(&mp->mnt_tmpfreevnodelist, vp, v_actfreelist); mp->mnt_tmpfreevnodelistsize++; vp->v_iflag |= VI_FREE; vp->v_mflag |= VMP_TMPMNTFREELIST; VI_UNLOCK(vp); if (mp->mnt_tmpfreevnodelistsize >= mnt_free_list_batch) vnlru_return_batch_locked(mp); mtx_unlock(&mp->mnt_listmtx); } else { VNASSERT(active == 0, vp, ("vdropl: active vnode not on per mount " "vnode list")); mtx_lock(&vnode_free_list_mtx); TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); freevnodes++; vp->v_iflag |= VI_FREE; VI_UNLOCK(vp); mtx_unlock(&vnode_free_list_mtx); } } else { VI_UNLOCK(vp); counter_u64_add(free_owe_inact, 1); } return; } /* * The vnode has been marked for destruction, so free it. * * The vnode will be returned to the zone where it will * normally remain until it is needed for another vnode. We * need to cleanup (or verify that the cleanup has already * been done) any residual data left from its current use * so as not to contaminate the freshly allocated vnode. */ CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp); atomic_subtract_long(&numvnodes, 1); bo = &vp->v_bufobj; VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("cleaned vnode still on the free list.")); VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp, ("clean blk trie not empty")); VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp, ("dirty blk trie not empty")); VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp, ("Dangling rangelock waiters")); VI_UNLOCK(vp); #ifdef MAC mac_vnode_destroy(vp); #endif if (vp->v_pollinfo != NULL) { destroy_vpollinfo(vp->v_pollinfo); vp->v_pollinfo = NULL; } #ifdef INVARIANTS /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */ vp->v_op = NULL; #endif vp->v_mountedhere = NULL; vp->v_unpcb = NULL; vp->v_rdev = NULL; vp->v_fifoinfo = NULL; vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; vp->v_iflag = 0; vp->v_vflag = 0; bo->bo_flag = 0; uma_zfree(vnode_zone, vp); } /* * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT * flags. DOINGINACT prevents us from recursing in calls to vinactive. * OWEINACT tracks whether a vnode missed a call to inactive due to a * failed lock upgrade. */ void vinactive(struct vnode *vp, struct thread *td) { struct vm_object *obj; ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, ("vinactive: recursed on VI_DOINGINACT")); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vp->v_iflag |= VI_DOINGINACT; vp->v_iflag &= ~VI_OWEINACT; VI_UNLOCK(vp); /* * Before moving off the active list, we must be sure that any * modified pages are converted into the vnode's dirty * buffers, since these will no longer be checked once the * vnode is on the inactive list. * * The write-out of the dirty pages is asynchronous. At the * point that VOP_INACTIVE() is called, there could still be * pending I/O and dirty pages in the object. */ if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, 0); VM_OBJECT_WUNLOCK(obj); } VOP_INACTIVE(vp, td); VI_LOCK(vp); VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, ("vinactive: lost VI_DOINGINACT")); vp->v_iflag &= ~VI_DOINGINACT; } /* * Remove any vnodes in the vnode table belonging to mount point mp. * * If FORCECLOSE is not specified, there should not be any active ones, * return error if any are found (nb: this is a user error, not a * system error). If FORCECLOSE is specified, detach any active vnodes * that are found. * * If WRITECLOSE is set, only flush out regular file vnodes open for * writing. * * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. * * `rootrefs' specifies the base reference count for the root vnode * of this filesystem. The root vnode is considered busy if its * v_usecount exceeds this value. On a successful return, vflush(, td) * will call vrele() on the root vnode exactly rootrefs times. * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must * be zero. */ #ifdef DIAGNOSTIC static int busyprt = 0; /* print out busy vnodes */ SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); #endif int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) { struct vnode *vp, *mvp, *rootvp = NULL; struct vattr vattr; int busy = 0, error; CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, rootrefs, flags); if (rootrefs > 0) { KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, ("vflush: bad args")); /* * Get the filesystem root vnode. We can vput() it * immediately, since with rootrefs > 0, it won't go away. */ if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", __func__, error); return (error); } vput(rootvp); } loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { vholdl(vp); error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); if (error) { vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } /* * Skip over a vnodes marked VV_SYSTEM. */ if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { VOP_UNLOCK(vp, 0); vdrop(vp); continue; } /* * If WRITECLOSE is set, flush out unlinked but still open * files (even if open only for reading) and regular file * vnodes open for writing. */ if (flags & WRITECLOSE) { if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } error = VOP_FSYNC(vp, MNT_WAIT, td); if (error != 0) { VOP_UNLOCK(vp, 0); vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } error = VOP_GETATTR(vp, &vattr, td->td_ucred); VI_LOCK(vp); if ((vp->v_type == VNON || (error == 0 && vattr.va_nlink > 0)) && (vp->v_writecount == 0 || vp->v_type != VREG)) { VOP_UNLOCK(vp, 0); vdropl(vp); continue; } } else VI_LOCK(vp); /* * With v_usecount == 0, all we need to do is clear out the * vnode data structures and we are done. * * If FORCECLOSE is set, forcibly close the vnode. */ if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { vgonel(vp); } else { busy++; #ifdef DIAGNOSTIC if (busyprt) vn_printf(vp, "vflush: busy vnode "); #endif } VOP_UNLOCK(vp, 0); vdropl(vp); } if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { /* * If just the root vnode is busy, and if its refcount * is equal to `rootrefs', then go ahead and kill it. */ VI_LOCK(rootvp); KASSERT(busy > 0, ("vflush: not busy")); VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, ("vflush: usecount %d < rootrefs %d", rootvp->v_usecount, rootrefs)); if (busy == 1 && rootvp->v_usecount == rootrefs) { VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); vgone(rootvp); VOP_UNLOCK(rootvp, 0); busy = 0; } else VI_UNLOCK(rootvp); } if (busy) { CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, busy); return (EBUSY); } for (; rootrefs > 0; rootrefs--) vrele(rootvp); return (0); } /* * Recycle an unused vnode to the front of the free list. */ int vrecycle(struct vnode *vp) { int recycled; VI_LOCK(vp); recycled = vrecyclel(vp); VI_UNLOCK(vp); return (recycled); } /* * vrecycle, with the vp interlock held. */ int vrecyclel(struct vnode *vp) { int recycled; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); recycled = 0; if (vp->v_usecount == 0) { recycled = 1; vgonel(vp); } return (recycled); } /* * Eliminate all activity associated with a vnode * in preparation for reuse. */ void vgone(struct vnode *vp) { VI_LOCK(vp); vgonel(vp); VI_UNLOCK(vp); } static void notify_lowervp_vfs_dummy(struct mount *mp __unused, struct vnode *lowervp __unused) { } /* * Notify upper mounts about reclaimed or unlinked vnode. */ void vfs_notify_upper(struct vnode *vp, int event) { static struct vfsops vgonel_vfsops = { .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy, .vfs_unlink_lowervp = notify_lowervp_vfs_dummy, }; struct mount *mp, *ump, *mmp; mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); if (TAILQ_EMPTY(&mp->mnt_uppers)) goto unlock; MNT_IUNLOCK(mp); mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO); mmp->mnt_op = &vgonel_vfsops; mmp->mnt_kern_flag |= MNTK_MARKER; MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_VGONE_UPPER; for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) { if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) { ump = TAILQ_NEXT(ump, mnt_upper_link); continue; } TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link); MNT_IUNLOCK(mp); switch (event) { case VFS_NOTIFY_UPPER_RECLAIM: VFS_RECLAIM_LOWERVP(ump, vp); break; case VFS_NOTIFY_UPPER_UNLINK: VFS_UNLINK_LOWERVP(ump, vp); break; default: KASSERT(0, ("invalid event %d", event)); break; } MNT_ILOCK(mp); ump = TAILQ_NEXT(mmp, mnt_upper_link); TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link); } free(mmp, M_TEMP); mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER; if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) { mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER; wakeup(&mp->mnt_uppers); } unlock: MNT_IUNLOCK(mp); } /* * vgone, with the vp interlock held. */ static void vgonel(struct vnode *vp) { struct thread *td; int oweinact; int active; struct mount *mp; ASSERT_VOP_ELOCKED(vp, "vgonel"); ASSERT_VI_LOCKED(vp, "vgonel"); VNASSERT(vp->v_holdcnt, vp, ("vgonel: vp %p has no reference.", vp)); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); td = curthread; /* * Don't vgonel if we're already doomed. */ if (vp->v_iflag & VI_DOOMED) return; vp->v_iflag |= VI_DOOMED; /* * Check to see if the vnode is in use. If so, we have to call * VOP_CLOSE() and VOP_INACTIVE(). */ active = vp->v_usecount; oweinact = (vp->v_iflag & VI_OWEINACT); VI_UNLOCK(vp); vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM); /* * If purging an active vnode, it must be closed and * deactivated before being reclaimed. */ if (active) VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); if (oweinact || active) { VI_LOCK(vp); if ((vp->v_iflag & VI_DOINGINACT) == 0) vinactive(vp, td); VI_UNLOCK(vp); } if (vp->v_type == VSOCK) vfs_unp_reclaim(vp); /* * Clean out any buffers associated with the vnode. * If the flush fails, just toss the buffers. */ mp = NULL; if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) (void) vn_start_secondary_write(vp, &mp, V_WAIT); if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) { while (vinvalbuf(vp, 0, 0, 0) != 0) ; } BO_LOCK(&vp->v_bufobj); KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) && vp->v_bufobj.bo_dirty.bv_cnt == 0 && TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) && vp->v_bufobj.bo_clean.bv_cnt == 0, ("vp %p bufobj not invalidated", vp)); /* * For VMIO bufobj, BO_DEAD is set in vm_object_terminate() * after the object's page queue is flushed. */ if (vp->v_bufobj.bo_object == NULL) vp->v_bufobj.bo_flag |= BO_DEAD; BO_UNLOCK(&vp->v_bufobj); /* * Reclaim the vnode. */ if (VOP_RECLAIM(vp, td)) panic("vgone: cannot reclaim"); if (mp != NULL) vn_finished_secondary_write(mp); VNASSERT(vp->v_object == NULL, vp, ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); /* * Clear the advisory locks and wake up waiting threads. */ (void)VOP_ADVLOCKPURGE(vp); vp->v_lockf = NULL; /* * Delete from old mount point vnode list. */ delmntque(vp); cache_purge(vp); /* * Done with purge, reset to the standard lock and invalidate * the vnode. */ VI_LOCK(vp); vp->v_vnlock = &vp->v_lock; vp->v_op = &dead_vnodeops; vp->v_tag = "none"; vp->v_type = VBAD; } /* * Calculate the total number of references to a special device. */ int vcount(struct vnode *vp) { int count; dev_lock(); count = vp->v_rdev->si_usecount; dev_unlock(); return (count); } /* * Same as above, but using the struct cdev *as argument */ int count_dev(struct cdev *dev) { int count; dev_lock(); count = dev->si_usecount; dev_unlock(); return(count); } /* * Print out a description of a vnode. */ static char *typename[] = {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", "VMARKER"}; void vn_printf(struct vnode *vp, const char *fmt, ...) { va_list ap; char buf[256], buf2[16]; u_long flags; va_start(ap, fmt); vprintf(fmt, ap); va_end(ap); printf("%p: ", (void *)vp); printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); printf(" usecount %d, writecount %d, refcount %d", vp->v_usecount, vp->v_writecount, vp->v_holdcnt); switch (vp->v_type) { case VDIR: printf(" mountedhere %p\n", vp->v_mountedhere); break; case VCHR: printf(" rdev %p\n", vp->v_rdev); break; case VSOCK: printf(" socket %p\n", vp->v_unpcb); break; case VFIFO: printf(" fifoinfo %p\n", vp->v_fifoinfo); break; default: printf("\n"); break; } buf[0] = '\0'; buf[1] = '\0'; if (vp->v_vflag & VV_ROOT) strlcat(buf, "|VV_ROOT", sizeof(buf)); if (vp->v_vflag & VV_ISTTY) strlcat(buf, "|VV_ISTTY", sizeof(buf)); if (vp->v_vflag & VV_NOSYNC) strlcat(buf, "|VV_NOSYNC", sizeof(buf)); if (vp->v_vflag & VV_ETERNALDEV) strlcat(buf, "|VV_ETERNALDEV", sizeof(buf)); if (vp->v_vflag & VV_CACHEDLABEL) strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); if (vp->v_vflag & VV_TEXT) strlcat(buf, "|VV_TEXT", sizeof(buf)); if (vp->v_vflag & VV_COPYONWRITE) strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); if (vp->v_vflag & VV_SYSTEM) strlcat(buf, "|VV_SYSTEM", sizeof(buf)); if (vp->v_vflag & VV_PROCDEP) strlcat(buf, "|VV_PROCDEP", sizeof(buf)); if (vp->v_vflag & VV_NOKNOTE) strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); if (vp->v_vflag & VV_DELETED) strlcat(buf, "|VV_DELETED", sizeof(buf)); if (vp->v_vflag & VV_MD) strlcat(buf, "|VV_MD", sizeof(buf)); if (vp->v_vflag & VV_FORCEINSMQ) strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf)); flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV | VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_iflag & VI_MOUNT) strlcat(buf, "|VI_MOUNT", sizeof(buf)); if (vp->v_iflag & VI_DOOMED) strlcat(buf, "|VI_DOOMED", sizeof(buf)); if (vp->v_iflag & VI_FREE) strlcat(buf, "|VI_FREE", sizeof(buf)); if (vp->v_iflag & VI_ACTIVE) strlcat(buf, "|VI_ACTIVE", sizeof(buf)); if (vp->v_iflag & VI_DOINGINACT) strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); if (vp->v_iflag & VI_OWEINACT) strlcat(buf, "|VI_OWEINACT", sizeof(buf)); flags = vp->v_iflag & ~(VI_MOUNT | VI_DOOMED | VI_FREE | VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } printf(" flags (%s)\n", buf + 1); if (mtx_owned(VI_MTX(vp))) printf(" VI_LOCKed"); if (vp->v_object != NULL) printf(" v_object %p ref %d pages %d " "cleanbuf %d dirtybuf %d\n", vp->v_object, vp->v_object->ref_count, vp->v_object->resident_page_count, vp->v_bufobj.bo_clean.bv_cnt, vp->v_bufobj.bo_dirty.bv_cnt); printf(" "); lockmgr_printinfo(vp->v_vnlock); if (vp->v_data != NULL) VOP_PRINT(vp); } #ifdef DDB /* * List all of the locked vnodes in the system. * Called when debugging the kernel. */ DB_SHOW_COMMAND(lockedvnods, lockedvnodes) { struct mount *mp; struct vnode *vp; /* * Note: because this is DDB, we can't obey the locking semantics * for these structures, which means we could catch an inconsistent * state and dereference a nasty pointer. Not much to be done * about that. */ db_printf("Locked vnodes\n"); TAILQ_FOREACH(mp, &mountlist, mnt_list) { TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && VOP_ISLOCKED(vp)) vn_printf(vp, "vnode "); } } } /* * Show details about the given vnode. */ DB_SHOW_COMMAND(vnode, db_show_vnode) { struct vnode *vp; if (!have_addr) return; vp = (struct vnode *)addr; vn_printf(vp, "vnode "); } /* * Show details about the given mount point. */ DB_SHOW_COMMAND(mount, db_show_mount) { struct mount *mp; struct vfsopt *opt; struct statfs *sp; struct vnode *vp; char buf[512]; uint64_t mflags; u_int flags; if (!have_addr) { /* No address given, print short info about all mount points. */ TAILQ_FOREACH(mp, &mountlist, mnt_list) { db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); if (db_pager_quit) break; } db_printf("\nMore info: show mount \n"); return; } mp = (struct mount *)addr; db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); buf[0] = '\0'; mflags = mp->mnt_flag; #define MNT_FLAG(flag) do { \ if (mflags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 4, sizeof(buf)); \ mflags &= ~(flag); \ } \ } while (0) MNT_FLAG(MNT_RDONLY); MNT_FLAG(MNT_SYNCHRONOUS); MNT_FLAG(MNT_NOEXEC); MNT_FLAG(MNT_NOSUID); MNT_FLAG(MNT_NFS4ACLS); MNT_FLAG(MNT_UNION); MNT_FLAG(MNT_ASYNC); MNT_FLAG(MNT_SUIDDIR); MNT_FLAG(MNT_SOFTDEP); MNT_FLAG(MNT_NOSYMFOLLOW); MNT_FLAG(MNT_GJOURNAL); MNT_FLAG(MNT_MULTILABEL); MNT_FLAG(MNT_ACLS); MNT_FLAG(MNT_NOATIME); MNT_FLAG(MNT_NOCLUSTERR); MNT_FLAG(MNT_NOCLUSTERW); MNT_FLAG(MNT_SUJ); MNT_FLAG(MNT_EXRDONLY); MNT_FLAG(MNT_EXPORTED); MNT_FLAG(MNT_DEFEXPORTED); MNT_FLAG(MNT_EXPORTANON); MNT_FLAG(MNT_EXKERB); MNT_FLAG(MNT_EXPUBLIC); MNT_FLAG(MNT_LOCAL); MNT_FLAG(MNT_QUOTA); MNT_FLAG(MNT_ROOTFS); MNT_FLAG(MNT_USER); MNT_FLAG(MNT_IGNORE); MNT_FLAG(MNT_UPDATE); MNT_FLAG(MNT_DELEXPORT); MNT_FLAG(MNT_RELOAD); MNT_FLAG(MNT_FORCE); MNT_FLAG(MNT_SNAPSHOT); MNT_FLAG(MNT_BYFSID); #undef MNT_FLAG if (mflags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%016jx", mflags); } db_printf(" mnt_flag = %s\n", buf); buf[0] = '\0'; flags = mp->mnt_kern_flag; #define MNT_KERN_FLAG(flag) do { \ if (flags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 5, sizeof(buf)); \ flags &= ~(flag); \ } \ } while (0) MNT_KERN_FLAG(MNTK_UNMOUNTF); MNT_KERN_FLAG(MNTK_ASYNC); MNT_KERN_FLAG(MNTK_SOFTDEP); MNT_KERN_FLAG(MNTK_NOINSMNTQ); MNT_KERN_FLAG(MNTK_DRAINING); MNT_KERN_FLAG(MNTK_REFEXPIRE); MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); MNT_KERN_FLAG(MNTK_SHARED_WRITES); MNT_KERN_FLAG(MNTK_NO_IOPF); MNT_KERN_FLAG(MNTK_VGONE_UPPER); MNT_KERN_FLAG(MNTK_VGONE_WAITER); MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT); MNT_KERN_FLAG(MNTK_MARKER); MNT_KERN_FLAG(MNTK_USES_BCACHE); MNT_KERN_FLAG(MNTK_NOASYNC); MNT_KERN_FLAG(MNTK_UNMOUNT); MNT_KERN_FLAG(MNTK_MWAIT); MNT_KERN_FLAG(MNTK_SUSPEND); MNT_KERN_FLAG(MNTK_SUSPEND2); MNT_KERN_FLAG(MNTK_SUSPENDED); MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); MNT_KERN_FLAG(MNTK_NOKNOTE); #undef MNT_KERN_FLAG if (flags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%08x", flags); } db_printf(" mnt_kern_flag = %s\n", buf); db_printf(" mnt_opt = "); opt = TAILQ_FIRST(mp->mnt_opt); if (opt != NULL) { db_printf("%s", opt->name); opt = TAILQ_NEXT(opt, link); while (opt != NULL) { db_printf(", %s", opt->name); opt = TAILQ_NEXT(opt, link); } } db_printf("\n"); sp = &mp->mnt_stat; db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); db_printf(" mnt_cred = { uid=%u ruid=%u", (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); if (jailed(mp->mnt_cred)) db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); db_printf(" }\n"); db_printf(" mnt_ref = %d\n", mp->mnt_ref); db_printf(" mnt_gen = %d\n", mp->mnt_gen); db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); db_printf(" mnt_activevnodelistsize = %d\n", mp->mnt_activevnodelistsize); db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); db_printf(" mnt_lockref = %d\n", mp->mnt_lockref); db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); db_printf(" mnt_secondary_accwrites = %d\n", mp->mnt_secondary_accwrites); db_printf(" mnt_gjprovider = %s\n", mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); db_printf("\n\nList of active vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) { if (vp->v_type != VMARKER) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } db_printf("\n\nList of inactive vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } } #endif /* DDB */ /* * Fill in a struct xvfsconf based on a struct vfsconf. */ static int vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; /* * These are unused in userland, we keep them * to not break binary compatibility. */ xvfsp.vfc_vfsops = NULL; xvfsp.vfc_next = NULL; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #ifdef COMPAT_FREEBSD32 struct xvfsconf32 { uint32_t vfc_vfsops; char vfc_name[MFSNAMELEN]; int32_t vfc_typenum; int32_t vfc_refcount; int32_t vfc_flags; uint32_t vfc_next; }; static int vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf32 xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #endif /* * Top level filesystem related information gathering. */ static int sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) { struct vfsconf *vfsp; int error; error = 0; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) error = vfsconf2x32(req, vfsp); else #endif error = vfsconf2x(req, vfsp); if (error) break; } vfsconf_sunlock(); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist, "S,xvfsconf", "List of all configured filesystems"); #ifndef BURN_BRIDGES static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); static int vfs_sysctl(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1 - 1; /* XXX */ u_int namelen = arg2 + 1; /* XXX */ struct vfsconf *vfsp; log(LOG_WARNING, "userland calling deprecated sysctl, " "please rebuild world\n"); #if 1 || defined(COMPAT_PRELITE2) /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ if (namelen == 1) return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); #endif switch (name[1]) { case VFS_MAXTYPENUM: if (namelen != 2) return (ENOTDIR); return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); case VFS_CONF: if (namelen != 3) return (ENOTDIR); /* overloaded */ vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { if (vfsp->vfc_typenum == name[2]) break; } vfsconf_sunlock(); if (vfsp == NULL) return (EOPNOTSUPP); #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) return (vfsconf2x32(req, vfsp)); else #endif return (vfsconf2x(req, vfsp)); } return (EOPNOTSUPP); } static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_MPSAFE, vfs_sysctl, "Generic filesystem"); #if 1 || defined(COMPAT_PRELITE2) static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) { int error; struct vfsconf *vfsp; struct ovfsconf ovfs; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { bzero(&ovfs, sizeof(ovfs)); ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ strcpy(ovfs.vfc_name, vfsp->vfc_name); ovfs.vfc_index = vfsp->vfc_typenum; ovfs.vfc_refcount = vfsp->vfc_refcount; ovfs.vfc_flags = vfsp->vfc_flags; error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); if (error != 0) { vfsconf_sunlock(); return (error); } } vfsconf_sunlock(); return (0); } #endif /* 1 || COMPAT_PRELITE2 */ #endif /* !BURN_BRIDGES */ #define KINFO_VNODESLOP 10 #ifdef notyet /* * Dump vnode list (via sysctl). */ /* ARGSUSED */ static int sysctl_vnode(SYSCTL_HANDLER_ARGS) { struct xvnode *xvn; struct mount *mp; struct vnode *vp; int error, len, n; /* * Stale numvnodes access is not fatal here. */ req->lock = 0; len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; if (!req->oldptr) /* Make an estimate */ return (SYSCTL_OUT(req, 0, len)); error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); n = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) continue; MNT_ILOCK(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (n == len) break; vref(vp); xvn[n].xv_size = sizeof *xvn; xvn[n].xv_vnode = vp; xvn[n].xv_id = 0; /* XXX compat */ #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field XV_COPY(usecount); XV_COPY(writecount); XV_COPY(holdcnt); XV_COPY(mount); XV_COPY(numoutput); XV_COPY(type); #undef XV_COPY xvn[n].xv_flag = vp->v_vflag; switch (vp->v_type) { case VREG: case VDIR: case VLNK: break; case VBLK: case VCHR: if (vp->v_rdev == NULL) { vrele(vp); continue; } xvn[n].xv_dev = dev2udev(vp->v_rdev); break; case VSOCK: xvn[n].xv_socket = vp->v_socket; break; case VFIFO: xvn[n].xv_fifo = vp->v_fifoinfo; break; case VNON: case VBAD: default: /* shouldn't happen? */ vrele(vp); continue; } vrele(vp); ++n; } MNT_IUNLOCK(mp); mtx_lock(&mountlist_mtx); vfs_unbusy(mp); if (n == len) break; } mtx_unlock(&mountlist_mtx); error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); free(xvn, M_TEMP); return (error); } SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode", ""); #endif static void unmount_or_warn(struct mount *mp) { int error; error = dounmount(mp, MNT_FORCE, curthread); if (error != 0) { printf("unmount of %s failed (", mp->mnt_stat.f_mntonname); if (error == EBUSY) printf("BUSY)\n"); else printf("%d)\n", error); } } /* * Unmount all filesystems. The list is traversed in reverse order * of mounting to avoid dependencies. */ void vfs_unmountall(void) { struct mount *mp, *tmp; CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); /* * Since this only runs when rebooting, it is not interlocked. */ TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) { vfs_ref(mp); /* * Forcibly unmounting "/dev" before "/" would prevent clean * unmount of the latter. */ if (mp == rootdevmp) continue; unmount_or_warn(mp); } if (rootdevmp != NULL) unmount_or_warn(rootdevmp); } /* * perform msync on all vnodes under a mount point * the mount point must be locked. */ void vfs_msync(struct mount *mp, int flags) { struct vnode *vp, *mvp; struct vm_object *obj; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); vnlru_return_batch(mp); MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) { obj = vp->v_object; if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 && (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { if (!vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, curthread)) { if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ vput(vp); continue; } obj = vp->v_object; if (obj != NULL) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); VM_OBJECT_WUNLOCK(obj); } vput(vp); } } else VI_UNLOCK(vp); } } static void destroy_vpollinfo_free(struct vpollinfo *vi) { knlist_destroy(&vi->vpi_selinfo.si_note); mtx_destroy(&vi->vpi_lock); uma_zfree(vnodepoll_zone, vi); } static void destroy_vpollinfo(struct vpollinfo *vi) { knlist_clear(&vi->vpi_selinfo.si_note, 1); seldrain(&vi->vpi_selinfo); destroy_vpollinfo_free(vi); } /* * Initialize per-vnode helper structure to hold poll-related state. */ void v_addpollinfo(struct vnode *vp) { struct vpollinfo *vi; if (vp->v_pollinfo != NULL) return; vi = uma_zalloc(vnodepoll_zone, M_WAITOK | M_ZERO); mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); VI_LOCK(vp); if (vp->v_pollinfo != NULL) { VI_UNLOCK(vp); destroy_vpollinfo_free(vi); return; } vp->v_pollinfo = vi; VI_UNLOCK(vp); } /* * Record a process's interest in events which might happen to * a vnode. Because poll uses the historic select-style interface * internally, this routine serves as both the ``check for any * pending events'' and the ``record my interest in future events'' * functions. (These are done together, while the lock is held, * to avoid race conditions.) */ int vn_pollrecord(struct vnode *vp, struct thread *td, int events) { v_addpollinfo(vp); mtx_lock(&vp->v_pollinfo->vpi_lock); if (vp->v_pollinfo->vpi_revents & events) { /* * This leaves events we are not interested * in available for the other process which * which presumably had requested them * (otherwise they would never have been * recorded). */ events &= vp->v_pollinfo->vpi_revents; vp->v_pollinfo->vpi_revents &= ~events; mtx_unlock(&vp->v_pollinfo->vpi_lock); return (events); } vp->v_pollinfo->vpi_events |= events; selrecord(td, &vp->v_pollinfo->vpi_selinfo); mtx_unlock(&vp->v_pollinfo->vpi_lock); return (0); } /* * Routine to create and manage a filesystem syncer vnode. */ #define sync_close ((int (*)(struct vop_close_args *))nullop) static int sync_fsync(struct vop_fsync_args *); static int sync_inactive(struct vop_inactive_args *); static int sync_reclaim(struct vop_reclaim_args *); static struct vop_vector sync_vnodeops = { .vop_bypass = VOP_EOPNOTSUPP, .vop_close = sync_close, /* close */ .vop_fsync = sync_fsync, /* fsync */ .vop_inactive = sync_inactive, /* inactive */ .vop_reclaim = sync_reclaim, /* reclaim */ .vop_lock1 = vop_stdlock, /* lock */ .vop_unlock = vop_stdunlock, /* unlock */ .vop_islocked = vop_stdislocked, /* islocked */ }; /* * Create a new filesystem syncer vnode for the specified mount point. */ void vfs_allocate_syncvnode(struct mount *mp) { struct vnode *vp; struct bufobj *bo; static long start, incr, next; int error; /* Allocate a new vnode */ error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); if (error != 0) panic("vfs_allocate_syncvnode: getnewvnode() failed"); vp->v_type = VNON; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_vflag |= VV_FORCEINSMQ; error = insmntque(vp, mp); if (error != 0) panic("vfs_allocate_syncvnode: insmntque() failed"); vp->v_vflag &= ~VV_FORCEINSMQ; VOP_UNLOCK(vp, 0); /* * Place the vnode onto the syncer worklist. We attempt to * scatter them about on the list so that they will go off * at evenly distributed times even if all the filesystems * are mounted at once. */ next += incr; if (next == 0 || next > syncer_maxdelay) { start /= 2; incr /= 2; if (start == 0) { start = syncer_maxdelay / 2; incr = syncer_maxdelay; } next = start; } bo = &vp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ mtx_lock(&sync_mtx); sync_vnode_count++; if (mp->mnt_syncer == NULL) { mp->mnt_syncer = vp; vp = NULL; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); if (vp != NULL) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vgone(vp); vput(vp); } } void vfs_deallocate_syncvnode(struct mount *mp) { struct vnode *vp; mtx_lock(&sync_mtx); vp = mp->mnt_syncer; if (vp != NULL) mp->mnt_syncer = NULL; mtx_unlock(&sync_mtx); if (vp != NULL) vrele(vp); } /* * Do a lazy sync of the filesystem. */ static int sync_fsync(struct vop_fsync_args *ap) { struct vnode *syncvp = ap->a_vp; struct mount *mp = syncvp->v_mount; int error, save; struct bufobj *bo; /* * We only need to do something if this is a lazy evaluation. */ if (ap->a_waitfor != MNT_LAZY) return (0); /* * Move ourselves to the back of the sync list. */ bo = &syncvp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay); BO_UNLOCK(bo); /* * Walk the list of vnodes pushing all that are dirty and * not already on the sync list. */ if (vfs_busy(mp, MBF_NOWAIT) != 0) return (0); if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { vfs_unbusy(mp); return (0); } save = curthread_pflags_set(TDP_SYNCIO); vfs_msync(mp, MNT_NOWAIT); error = VFS_SYNC(mp, MNT_LAZY); curthread_pflags_restore(save); vn_finished_write(mp); vfs_unbusy(mp); return (error); } /* * The syncer vnode is no referenced. */ static int sync_inactive(struct vop_inactive_args *ap) { vgone(ap->a_vp); return (0); } /* * The syncer vnode is no longer needed and is being decommissioned. * * Modifications to the worklist must be protected by sync_mtx. */ static int sync_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp = ap->a_vp; struct bufobj *bo; bo = &vp->v_bufobj; BO_LOCK(bo); mtx_lock(&sync_mtx); if (vp->v_mount->mnt_syncer == vp) vp->v_mount->mnt_syncer = NULL; if (bo->bo_flag & BO_ONWORKLST) { LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; sync_vnode_count--; bo->bo_flag &= ~BO_ONWORKLST; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); return (0); } /* * Check if vnode represents a disk device */ int vn_isdisk(struct vnode *vp, int *errp) { int error; if (vp->v_type != VCHR) { error = ENOTBLK; goto out; } error = 0; dev_lock(); if (vp->v_rdev == NULL) error = ENXIO; else if (vp->v_rdev->si_devsw == NULL) error = ENXIO; else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) error = ENOTBLK; dev_unlock(); out: if (errp != NULL) *errp = error; return (error == 0); } /* * Common filesystem object access control check routine. Accepts a * vnode's type, "mode", uid and gid, requested access mode, credentials, * and optional call-by-reference privused argument allowing vaccess() * to indicate to the caller whether privilege was used to satisfy the * request (obsoleted). Returns 0 on success, or an errno on failure. */ int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred, int *privused) { accmode_t dac_granted; accmode_t priv_granted; KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, ("invalid bit in accmode")); KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), ("VAPPEND without VWRITE")); /* * Look for a normal, non-privileged way to access the file/directory * as requested. If it exists, go with that. */ if (privused != NULL) *privused = 0; dac_granted = 0; /* Check the owner. */ if (cred->cr_uid == file_uid) { dac_granted |= VADMIN; if (file_mode & S_IXUSR) dac_granted |= VEXEC; if (file_mode & S_IRUSR) dac_granted |= VREAD; if (file_mode & S_IWUSR) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) dac_granted |= VEXEC; if (file_mode & S_IRGRP) dac_granted |= VREAD; if (file_mode & S_IWGRP) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) dac_granted |= VEXEC; if (file_mode & S_IROTH) dac_granted |= VREAD; if (file_mode & S_IWOTH) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); privcheck: /* * Build a privilege mask to determine if the set of privileges * satisfies the requirements when combined with the granted mask * from above. For each privilege, if the privilege is required, * bitwise or the request type onto the priv_granted mask. */ priv_granted = 0; if (type == VDIR) { /* * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC * requests, instead of PRIV_VFS_EXEC. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && !priv_check_cred(cred, PRIV_VFS_LOOKUP)) priv_granted |= VEXEC; } else { /* * Ensure that at least one execute bit is on. Otherwise, * a privileged user will always succeed, and we don't want * this to happen unless the file really is executable. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && !priv_check_cred(cred, PRIV_VFS_EXEC)) priv_granted |= VEXEC; } if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && !priv_check_cred(cred, PRIV_VFS_READ)) priv_granted |= VREAD; if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && !priv_check_cred(cred, PRIV_VFS_WRITE)) priv_granted |= (VWRITE | VAPPEND); if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && !priv_check_cred(cred, PRIV_VFS_ADMIN)) priv_granted |= VADMIN; if ((accmode & (priv_granted | dac_granted)) == accmode) { /* XXX audit: privilege used */ if (privused != NULL) *privused = 1; return (0); } return ((accmode & VADMIN) ? EPERM : EACCES); } /* * Credential check based on process requesting service, and per-attribute * permissions. */ int extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, struct thread *td, accmode_t accmode) { /* * Kernel-invoked always succeeds. */ if (cred == NOCRED) return (0); /* * Do not allow privileged processes in jail to directly manipulate * system attributes. */ switch (attrnamespace) { case EXTATTR_NAMESPACE_SYSTEM: /* Potentially should be: return (EPERM); */ return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); case EXTATTR_NAMESPACE_USER: return (VOP_ACCESS(vp, accmode, cred, td)); default: return (EPERM); } } #ifdef DEBUG_VFS_LOCKS /* * This only exists to suppress warnings from unlocked specfs accesses. It is * no longer ok to have an unlocked VFS. */ #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ (vp)->v_type == VCHR || (vp)->v_type == VBAD) int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "Drop into debugger on lock violation"); int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "Check for interlock across VOPs"); int vfs_badlock_print = 1; /* Print lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "Print lock violations"); int vfs_badlock_vnode = 1; /* Print vnode details on lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_vnode, CTLFLAG_RW, &vfs_badlock_vnode, 0, "Print vnode details on lock violations"); #ifdef KDB int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); #endif static void vfs_badlock(const char *msg, const char *str, struct vnode *vp) { #ifdef KDB if (vfs_badlock_backtrace) kdb_backtrace(); #endif if (vfs_badlock_vnode) vn_printf(vp, "vnode "); if (vfs_badlock_print) printf("%s: %p %s\n", str, (void *)vp, msg); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } void assert_vi_locked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is not locked but should be", str, vp); } void assert_vi_unlocked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is locked but should not be", str, vp); } void assert_vop_locked(struct vnode *vp, const char *str) { int locked; if (!IGNORE_LOCK(vp)) { locked = VOP_ISLOCKED(vp); if (locked == 0 || locked == LK_EXCLOTHER) vfs_badlock("is not locked but should be", str, vp); } } void assert_vop_unlocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) vfs_badlock("is locked but should not be", str, vp); } void assert_vop_elocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vfs_badlock("is not exclusive locked but should be", str, vp); } #endif /* DEBUG_VFS_LOCKS */ void vop_rename_fail(struct vop_rename_args *ap) { if (ap->a_tvp != NULL) vput(ap->a_tvp); if (ap->a_tdvp == ap->a_tvp) vrele(ap->a_tdvp); else vput(ap->a_tdvp); vrele(ap->a_fdvp); vrele(ap->a_fvp); } void vop_rename_pre(void *ap) { struct vop_rename_args *a = ap; #ifdef DEBUG_VFS_LOCKS if (a->a_tvp) ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); /* Check the source (from). */ if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); /* Check the target. */ if (a->a_tvp) ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); #endif if (a->a_tdvp != a->a_fdvp) vhold(a->a_fdvp); if (a->a_tvp != a->a_fvp) vhold(a->a_fvp); vhold(a->a_tdvp); if (a->a_tvp) vhold(a->a_tvp); } #ifdef DEBUG_VFS_LOCKS void vop_strategy_pre(void *ap) { struct vop_strategy_args *a; struct buf *bp; a = ap; bp = a->a_bp; /* * Cluster ops lock their component buffers but not the IO container. */ if ((bp->b_flags & B_CLUSTER) != 0) return; if (panicstr == NULL && !BUF_ISLOCKED(bp)) { if (vfs_badlock_print) printf( "VOP_STRATEGY: bp is not locked but should be\n"); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } } void vop_lock_pre(void *ap) { struct vop_lock1_args *a = ap; if ((a->a_flags & LK_INTERLOCK) == 0) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); else ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_lock_post(void *ap, int rc) { struct vop_lock1_args *a = ap; ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0) ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_unlock_pre(void *ap) { struct vop_unlock_args *a = ap; if (a->a_flags & LK_INTERLOCK) ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); } void vop_unlock_post(void *ap, int rc) { struct vop_unlock_args *a = ap; if (a->a_flags & LK_INTERLOCK) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); } #endif void vop_create_post(void *ap, int rc) { struct vop_create_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_deleteextattr_post(void *ap, int rc) { struct vop_deleteextattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_link_post(void *ap, int rc) { struct vop_link_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); } } void vop_mkdir_post(void *ap, int rc) { struct vop_mkdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); } void vop_mknod_post(void *ap, int rc) { struct vop_mknod_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_reclaim_post(void *ap, int rc) { struct vop_reclaim_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_REVOKE); } void vop_remove_post(void *ap, int rc) { struct vop_remove_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); } } void vop_rename_post(void *ap, int rc) { struct vop_rename_args *a = ap; long hint; if (!rc) { hint = NOTE_WRITE; if (a->a_fdvp == a->a_tdvp) { if (a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } else { hint |= NOTE_EXTEND; if (a->a_fvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); if (a->a_fvp->v_type == VDIR && a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint &= ~NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); if (a->a_tvp) VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); } if (a->a_tdvp != a->a_fdvp) vdrop(a->a_fdvp); if (a->a_tvp != a->a_fvp) vdrop(a->a_fvp); vdrop(a->a_tdvp); if (a->a_tvp) vdrop(a->a_tvp); } void vop_rmdir_post(void *ap, int rc) { struct vop_rmdir_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); } } void vop_setattr_post(void *ap, int rc) { struct vop_setattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_setextattr_post(void *ap, int rc) { struct vop_setextattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_symlink_post(void *ap, int rc) { struct vop_symlink_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_open_post(void *ap, int rc) { struct vop_open_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_OPEN); } void vop_close_post(void *ap, int rc) { struct vop_close_args *a = ap; if (!rc && (a->a_cred != NOCRED || /* filter out revokes */ (a->a_vp->v_iflag & VI_DOOMED) == 0)) { VFS_KNOTE_LOCKED(a->a_vp, (a->a_fflag & FWRITE) != 0 ? NOTE_CLOSE_WRITE : NOTE_CLOSE); } } void vop_read_post(void *ap, int rc) { struct vop_read_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } void vop_readdir_post(void *ap, int rc) { struct vop_readdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } static struct knlist fs_knlist; static void vfs_event_init(void *arg) { knlist_init_mtx(&fs_knlist, NULL); } /* XXX - correct order? */ SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); void vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) { KNOTE_UNLOCKED(&fs_knlist, event); } static int filt_fsattach(struct knote *kn); static void filt_fsdetach(struct knote *kn); static int filt_fsevent(struct knote *kn, long hint); struct filterops fs_filtops = { .f_isfd = 0, .f_attach = filt_fsattach, .f_detach = filt_fsdetach, .f_event = filt_fsevent }; static int filt_fsattach(struct knote *kn) { kn->kn_flags |= EV_CLEAR; knlist_add(&fs_knlist, kn, 0); return (0); } static void filt_fsdetach(struct knote *kn) { knlist_remove(&fs_knlist, kn, 0); } static int filt_fsevent(struct knote *kn, long hint) { kn->kn_fflags |= hint; return (kn->kn_fflags != 0); } static int sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) { struct vfsidctl vc; int error; struct mount *mp; error = SYSCTL_IN(req, &vc, sizeof(vc)); if (error) return (error); if (vc.vc_vers != VFS_CTL_VERS1) return (EINVAL); mp = vfs_getvfs(&vc.vc_fsid); if (mp == NULL) return (ENOENT); /* ensure that a specific sysctl goes to the right filesystem. */ if (strcmp(vc.vc_fstypename, "*") != 0 && strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { vfs_rel(mp); return (EINVAL); } VCTLTOREQ(&vc, req); error = VFS_SYSCTL(mp, vc.vc_op, req); vfs_rel(mp); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid"); /* * Function to initialize a va_filerev field sensibly. * XXX: Wouldn't a random number make a lot more sense ?? */ u_quad_t init_va_filerev(void) { struct bintime bt; getbinuptime(&bt); return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); } static int filt_vfsread(struct knote *kn, long hint); static int filt_vfswrite(struct knote *kn, long hint); static int filt_vfsvnode(struct knote *kn, long hint); static void filt_vfsdetach(struct knote *kn); static struct filterops vfsread_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsread }; static struct filterops vfswrite_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfswrite }; static struct filterops vfsvnode_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsvnode }; static void vfs_knllock(void *arg) { struct vnode *vp = arg; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } static void vfs_knlunlock(void *arg) { struct vnode *vp = arg; VOP_UNLOCK(vp, 0); } static void vfs_knl_assert_locked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); #endif } static void vfs_knl_assert_unlocked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); #endif } int vfs_kqfilter(struct vop_kqfilter_args *ap) { struct vnode *vp = ap->a_vp; struct knote *kn = ap->a_kn; struct knlist *knl; switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &vfsread_filtops; break; case EVFILT_WRITE: kn->kn_fop = &vfswrite_filtops; break; case EVFILT_VNODE: kn->kn_fop = &vfsvnode_filtops; break; default: return (EINVAL); } kn->kn_hook = (caddr_t)vp; v_addpollinfo(vp); if (vp->v_pollinfo == NULL) return (ENOMEM); knl = &vp->v_pollinfo->vpi_selinfo.si_note; vhold(vp); knlist_add(knl, kn, 0); return (0); } /* * Detach knote from vnode */ static void filt_vfsdetach(struct knote *kn) { struct vnode *vp = (struct vnode *)kn->kn_hook; KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); vdrop(vp); } /*ARGSUSED*/ static int filt_vfsread(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; struct vattr va; int res; /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { VI_LOCK(vp); kn->kn_flags |= (EV_EOF | EV_ONESHOT); VI_UNLOCK(vp); return (1); } if (VOP_GETATTR(vp, &va, curthread->td_ucred)) return (0); VI_LOCK(vp); kn->kn_data = va.va_size - kn->kn_fp->f_offset; res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0; VI_UNLOCK(vp); return (res); } /*ARGSUSED*/ static int filt_vfswrite(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; VI_LOCK(vp); /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) kn->kn_flags |= (EV_EOF | EV_ONESHOT); kn->kn_data = 0; VI_UNLOCK(vp); return (1); } static int filt_vfsvnode(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; int res; VI_LOCK(vp); if (kn->kn_sfflags & hint) kn->kn_fflags |= hint; if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { kn->kn_flags |= EV_EOF; VI_UNLOCK(vp); return (1); } res = (kn->kn_fflags != 0); VI_UNLOCK(vp); return (res); } int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) { int error; if (dp->d_reclen > ap->a_uio->uio_resid) return (ENAMETOOLONG); error = uiomove(dp, dp->d_reclen, ap->a_uio); if (error) { if (ap->a_ncookies != NULL) { if (ap->a_cookies != NULL) free(ap->a_cookies, M_TEMP); ap->a_cookies = NULL; *ap->a_ncookies = 0; } return (error); } if (ap->a_ncookies == NULL) return (0); KASSERT(ap->a_cookies, ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); *ap->a_cookies = realloc(*ap->a_cookies, (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); (*ap->a_cookies)[*ap->a_ncookies] = off; *ap->a_ncookies += 1; return (0); } /* * Mark for update the access time of the file if the filesystem * supports VOP_MARKATIME. This functionality is used by execve and * mmap, so we want to avoid the I/O implied by directly setting * va_atime for the sake of efficiency. */ void vfs_mark_atime(struct vnode *vp, struct ucred *cred) { struct mount *mp; mp = vp->v_mount; ASSERT_VOP_LOCKED(vp, "vfs_mark_atime"); if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) (void)VOP_MARKATIME(vp); } /* * The purpose of this routine is to remove granularity from accmode_t, * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, * VADMIN and VAPPEND. * * If it returns 0, the caller is supposed to continue with the usual * access checks using 'accmode' as modified by this routine. If it * returns nonzero value, the caller is supposed to return that value * as errno. * * Note that after this routine runs, accmode may be zero. */ int vfs_unixify_accmode(accmode_t *accmode) { /* * There is no way to specify explicit "deny" rule using * file mode or POSIX.1e ACLs. */ if (*accmode & VEXPLICIT_DENY) { *accmode = 0; return (0); } /* * None of these can be translated into usual access bits. * Also, the common case for NFSv4 ACLs is to not contain * either of these bits. Caller should check for VWRITE * on the containing directory instead. */ if (*accmode & (VDELETE_CHILD | VDELETE)) return (EPERM); if (*accmode & VADMIN_PERMS) { *accmode &= ~VADMIN_PERMS; *accmode |= VADMIN; } /* * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL * or VSYNCHRONIZE using file mode or POSIX.1e ACL. */ *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); return (0); } /* * These are helper functions for filesystems to traverse all * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h. * * This interface replaces MNT_VNODE_FOREACH. */ MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); struct vnode * __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; if (should_yield()) kern_yield(PRI_USER); MNT_ILOCK(mp); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); for (vp = TAILQ_NEXT(*mvp, v_nmntvnodes); vp != NULL; vp = TAILQ_NEXT(vp, v_nmntvnodes)) { /* Allow a racy peek at VI_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0) continue; VI_LOCK(vp); if ((vp->v_iflag & VI_DOOMED) != 0) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { __mnt_vnode_markerfree_all(mvp, mp); /* MNT_IUNLOCK(mp); -- done in above function */ mtx_assert(MNT_MTX(mp), MA_NOTOWNED); return (NULL); } TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } struct vnode * __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); MNT_ILOCK(mp); MNT_REF(mp); (*mvp)->v_mount = mp; (*mvp)->v_type = VMARKER; TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { /* Allow a racy peek at VI_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0) continue; VI_LOCK(vp); if ((vp->v_iflag & VI_DOOMED) != 0) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { MNT_REL(mp); MNT_IUNLOCK(mp); free(*mvp, M_VNODE_MARKER); *mvp = NULL; return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } void __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) { MNT_IUNLOCK(mp); return; } mtx_assert(MNT_MTX(mp), MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); MNT_REL(mp); MNT_IUNLOCK(mp); free(*mvp, M_VNODE_MARKER); *mvp = NULL; } /* * These are helper functions for filesystems to traverse their * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h */ static void mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp) { KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); MNT_ILOCK(mp); MNT_REL(mp); MNT_IUNLOCK(mp); free(*mvp, M_VNODE_MARKER); *mvp = NULL; } /* * Relock the mp mount vnode list lock with the vp vnode interlock in the * conventional lock order during mnt_vnode_next_active iteration. * * On entry, the mount vnode list lock is held and the vnode interlock is not. * The list lock is dropped and reacquired. On success, both locks are held. * On failure, the mount vnode list lock is held but the vnode interlock is * not, and the procedure may have yielded. */ static bool mnt_vnode_next_active_relock(struct vnode *mvp, struct mount *mp, struct vnode *vp) { const struct vnode *tmp; bool held, ret; VNASSERT(mvp->v_mount == mp && mvp->v_type == VMARKER && TAILQ_NEXT(mvp, v_actfreelist) != NULL, mvp, ("%s: bad marker", __func__)); VNASSERT(vp->v_mount == mp && vp->v_type != VMARKER, vp, ("%s: inappropriate vnode", __func__)); ASSERT_VI_UNLOCKED(vp, __func__); mtx_assert(&mp->mnt_listmtx, MA_OWNED); ret = false; TAILQ_REMOVE(&mp->mnt_activevnodelist, mvp, v_actfreelist); TAILQ_INSERT_BEFORE(vp, mvp, v_actfreelist); /* * Use a hold to prevent vp from disappearing while the mount vnode * list lock is dropped and reacquired. Normally a hold would be * acquired with vhold(), but that might try to acquire the vnode * interlock, which would be a LOR with the mount vnode list lock. */ held = refcount_acquire_if_not_zero(&vp->v_holdcnt); mtx_unlock(&mp->mnt_listmtx); if (!held) goto abort; VI_LOCK(vp); if (!refcount_release_if_not_last(&vp->v_holdcnt)) { vdropl(vp); goto abort; } mtx_lock(&mp->mnt_listmtx); /* * Determine whether the vnode is still the next one after the marker, * excepting any other markers. If the vnode has not been doomed by * vgone() then the hold should have ensured that it remained on the * active list. If it has been doomed but is still on the active list, * don't abort, but rather skip over it (avoid spinning on doomed * vnodes). */ tmp = mvp; do { tmp = TAILQ_NEXT(tmp, v_actfreelist); } while (tmp != NULL && tmp->v_type == VMARKER); if (tmp != vp) { mtx_unlock(&mp->mnt_listmtx); VI_UNLOCK(vp); goto abort; } ret = true; goto out; abort: maybe_yield(); mtx_lock(&mp->mnt_listmtx); out: if (ret) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); mtx_assert(&mp->mnt_listmtx, MA_OWNED); return (ret); } static struct vnode * mnt_vnode_next_active(struct vnode **mvp, struct mount *mp) { struct vnode *vp, *nvp; mtx_assert(&mp->mnt_listmtx, MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); restart: vp = TAILQ_NEXT(*mvp, v_actfreelist); while (vp != NULL) { if (vp->v_type == VMARKER) { vp = TAILQ_NEXT(vp, v_actfreelist); continue; } /* * Try-lock because this is the wrong lock order. If that does * not succeed, drop the mount vnode list lock and try to * reacquire it and the vnode interlock in the right order. */ if (!VI_TRYLOCK(vp) && !mnt_vnode_next_active_relock(*mvp, mp, vp)) goto restart; KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp)); KASSERT(vp->v_mount == mp || vp->v_mount == NULL, ("alien vnode on the active list %p %p", vp, mp)); if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0) break; nvp = TAILQ_NEXT(vp, v_actfreelist); VI_UNLOCK(vp); vp = nvp; } TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); /* Check if we are done */ if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_active(mvp, mp); return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist); mtx_unlock(&mp->mnt_listmtx); ASSERT_VI_LOCKED(vp, "active iter"); KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp)); return (vp); } struct vnode * __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp) { if (should_yield()) kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); return (mnt_vnode_next_active(mvp, mp)); } struct vnode * __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); MNT_ILOCK(mp); MNT_REF(mp); MNT_IUNLOCK(mp); (*mvp)->v_type = VMARKER; (*mvp)->v_mount = mp; mtx_lock(&mp->mnt_listmtx); vp = TAILQ_FIRST(&mp->mnt_activevnodelist); if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_active(mvp, mp); return (NULL); } TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist); return (mnt_vnode_next_active(mvp, mp)); } void __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) return; mtx_lock(&mp->mnt_listmtx); TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_active(mvp, mp); } Index: head/sys/ufs/ffs/ffs_snapshot.c =================================================================== --- head/sys/ufs/ffs/ffs_snapshot.c (revision 342547) +++ head/sys/ufs/ffs/ffs_snapshot.c (revision 342548) @@ -1,2710 +1,2714 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright 2000 Marshall Kirk McKusick. All Rights Reserved. * * Further information about snapshots can be obtained from: * * Marshall Kirk McKusick http://www.mckusick.com/softdep/ * 1614 Oxford Street mckusick@mckusick.com * Berkeley, CA 94709-1608 +1-510-843-9542 * USA * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``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 MARSHALL KIRK MCKUSICK 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. * * @(#)ffs_snapshot.c 8.11 (McKusick) 7/23/00 */ #include __FBSDID("$FreeBSD$"); #include "opt_quota.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 #define KERNCRED thread0.td_ucred #define DEBUG 1 #include "opt_ffs.h" #ifdef NO_FFS_SNAPSHOT int ffs_snapshot(mp, snapfile) struct mount *mp; char *snapfile; { return (EINVAL); } int ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd) struct fs *fs; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; enum vtype vtype; struct workhead *wkhd; { return (EINVAL); } void ffs_snapremove(vp) struct vnode *vp; { } void ffs_snapshot_mount(mp) struct mount *mp; { } void ffs_snapshot_unmount(mp) struct mount *mp; { } void ffs_snapgone(ip) struct inode *ip; { } int ffs_copyonwrite(devvp, bp) struct vnode *devvp; struct buf *bp; { return (EINVAL); } void ffs_sync_snap(mp, waitfor) struct mount *mp; int waitfor; { } #else FEATURE(ffs_snapshot, "FFS snapshot support"); LIST_HEAD(, snapdata) snapfree; static struct mtx snapfree_lock; MTX_SYSINIT(ffs_snapfree, &snapfree_lock, "snapdata free list", MTX_DEF); static int cgaccount(int, struct vnode *, struct buf *, int); static int expunge_ufs1(struct vnode *, struct inode *, struct fs *, int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, ufs_lbn_t, int), int, int); static int indiracct_ufs1(struct vnode *, struct vnode *, int, ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *, int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, ufs_lbn_t, int), int); static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, ufs_lbn_t, int); static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, ufs_lbn_t, int); static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, ufs_lbn_t, int); static int expunge_ufs2(struct vnode *, struct inode *, struct fs *, int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, ufs_lbn_t, int), int, int); static int indiracct_ufs2(struct vnode *, struct vnode *, int, ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *, int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, ufs_lbn_t, int), int); static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, ufs_lbn_t, int); static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, ufs_lbn_t, int); static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, ufs_lbn_t, int); static int readblock(struct vnode *vp, struct buf *, ufs2_daddr_t); static void try_free_snapdata(struct vnode *devvp); static struct snapdata *ffs_snapdata_acquire(struct vnode *devvp); static int ffs_bp_snapblk(struct vnode *, struct buf *); /* * To ensure the consistency of snapshots across crashes, we must * synchronously write out copied blocks before allowing the * originals to be modified. Because of the rather severe speed * penalty that this imposes, the code normally only ensures * persistence for the filesystem metadata contained within a * snapshot. Setting the following flag allows this crash * persistence to be enabled for file contents. */ int dopersistence = 0; #ifdef DEBUG #include SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, ""); static int snapdebug = 0; SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, ""); int collectsnapstats = 0; SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats, 0, ""); #endif /* DEBUG */ /* * Create a snapshot file and initialize it for the filesystem. */ int ffs_snapshot(mp, snapfile) struct mount *mp; char *snapfile; { ufs2_daddr_t numblks, blkno, *blkp, *snapblklist; int error, cg, snaploc; int i, size, len, loc; ufs2_daddr_t blockno; uint64_t flag; struct timespec starttime = {0, 0}, endtime; char saved_nice = 0; long redo = 0, snaplistsize = 0; int32_t *lp; void *space; struct fs *copy_fs = NULL, *fs; struct thread *td = curthread; struct inode *ip, *xp; struct buf *bp, *nbp, *ibp; struct nameidata nd; struct mount *wrtmp; struct vattr vat; struct vnode *vp, *xvp, *mvp, *devvp; struct uio auio; struct iovec aiov; struct snapdata *sn; struct ufsmount *ump; ump = VFSTOUFS(mp); fs = ump->um_fs; sn = NULL; /* * At the moment, journaled soft updates cannot support * taking snapshots. */ if (MOUNTEDSUJ(mp)) { vfs_mount_error(mp, "%s: Snapshots are not yet supported when " "running with journaled soft updates", fs->fs_fsmnt); return (EOPNOTSUPP); } MNT_ILOCK(mp); flag = mp->mnt_flag; MNT_IUNLOCK(mp); /* * Need to serialize access to snapshot code per filesystem. */ /* * Assign a snapshot slot in the superblock. */ UFS_LOCK(ump); for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) if (fs->fs_snapinum[snaploc] == 0) break; UFS_UNLOCK(ump); if (snaploc == FSMAXSNAP) return (ENOSPC); /* * Create the snapshot file. */ restart: NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF | NOCACHE, UIO_SYSSPACE, snapfile, td); if ((error = namei(&nd)) != 0) return (error); if (nd.ni_vp != NULL) { vput(nd.ni_vp); error = EEXIST; } if (nd.ni_dvp->v_mount != mp) error = EXDEV; if (error) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_dvp == nd.ni_vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); return (error); } VATTR_NULL(&vat); vat.va_type = VREG; vat.va_mode = S_IRUSR; vat.va_vaflags |= VA_EXCLUSIVE; if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp)) wrtmp = NULL; if (wrtmp != mp) panic("ffs_snapshot: mount mismatch"); vfs_rel(wrtmp); if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &wrtmp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat); VOP_UNLOCK(nd.ni_dvp, 0); if (error) { NDFREE(&nd, NDF_ONLY_PNBUF); vn_finished_write(wrtmp); vrele(nd.ni_dvp); return (error); } vp = nd.ni_vp; vnode_create_vobject(nd.ni_vp, fs->fs_size, td); vp->v_vflag |= VV_SYSTEM; ip = VTOI(vp); devvp = ITODEVVP(ip); /* * Allocate and copy the last block contents so as to be able * to set size to that of the filesystem. */ numblks = howmany(fs->fs_size, fs->fs_frag); error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)), fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); if (error) goto out; ip->i_size = lblktosize(fs, (off_t)numblks); DIP_SET(ip, i_size, ip->i_size); ip->i_flag |= IN_CHANGE | IN_UPDATE; error = readblock(vp, bp, numblks - 1); bawrite(bp); if (error != 0) goto out; /* * Preallocate critical data structures so that we can copy * them in without further allocation after we suspend all * operations on the filesystem. We would like to just release * the allocated buffers without writing them since they will * be filled in below once we are ready to go, but this upsets * the soft update code, so we go ahead and write the new buffers. * * Allocate all indirect blocks and mark all of them as not * needing to be copied. */ for (blkno = UFS_NDADDR; blkno < numblks; blkno += NINDIR(fs)) { error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp); if (error) goto out; bawrite(ibp); } /* * Allocate copies for the superblock and its summary information. */ error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED, 0, &nbp); if (error) goto out; bawrite(nbp); blkno = fragstoblks(fs, fs->fs_csaddr); len = howmany(fs->fs_cssize, fs->fs_bsize); for (loc = 0; loc < len; loc++) { error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)), fs->fs_bsize, KERNCRED, 0, &nbp); if (error) goto out; bawrite(nbp); } /* * Allocate all cylinder group blocks. */ for (cg = 0; cg < fs->fs_ncg; cg++) { error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), fs->fs_bsize, KERNCRED, 0, &nbp); if (error) goto out; bawrite(nbp); if (cg % 10 == 0) ffs_syncvnode(vp, MNT_WAIT, 0); } /* * Copy all the cylinder group maps. Although the * filesystem is still active, we hope that only a few * cylinder groups will change between now and when we * suspend operations. Thus, we will be able to quickly * touch up the few cylinder groups that changed during * the suspension period. */ len = howmany(fs->fs_ncg, NBBY); space = malloc(len, M_DEVBUF, M_WAITOK|M_ZERO); UFS_LOCK(ump); fs->fs_active = space; UFS_UNLOCK(ump); for (cg = 0; cg < fs->fs_ncg; cg++) { error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), fs->fs_bsize, KERNCRED, 0, &nbp); if (error) goto out; error = cgaccount(cg, vp, nbp, 1); bawrite(nbp); if (cg % 10 == 0) ffs_syncvnode(vp, MNT_WAIT, 0); if (error) goto out; } /* * Change inode to snapshot type file. */ ip->i_flags |= SF_SNAPSHOT; DIP_SET(ip, i_flags, ip->i_flags); ip->i_flag |= IN_CHANGE | IN_UPDATE; /* * Ensure that the snapshot is completely on disk. * Since we have marked it as a snapshot it is safe to * unlock it as no process will be allowed to write to it. */ if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) goto out; VOP_UNLOCK(vp, 0); /* * All allocations are done, so we can now snapshot the system. * * Recind nice scheduling while running with the filesystem suspended. */ if (td->td_proc->p_nice > 0) { struct proc *p; p = td->td_proc; PROC_LOCK(p); saved_nice = p->p_nice; sched_nice(p, 0); PROC_UNLOCK(p); } /* * Suspend operation on filesystem. */ for (;;) { vn_finished_write(wrtmp); if ((error = vfs_write_suspend(vp->v_mount, 0)) != 0) { vn_start_write(NULL, &wrtmp, V_WAIT); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); goto out; } if (mp->mnt_kern_flag & MNTK_SUSPENDED) break; vn_start_write(NULL, &wrtmp, V_WAIT); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (ip->i_effnlink == 0) { error = ENOENT; /* Snapshot file unlinked */ goto out1; } if (collectsnapstats) nanotime(&starttime); /* The last block might have changed. Copy it again to be sure. */ error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)), fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); if (error != 0) goto out1; error = readblock(vp, bp, numblks - 1); bp->b_flags |= B_VALIDSUSPWRT; bawrite(bp); if (error != 0) goto out1; /* * First, copy all the cylinder group maps that have changed. */ for (cg = 0; cg < fs->fs_ncg; cg++) { if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0) continue; redo++; error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)), fs->fs_bsize, KERNCRED, 0, &nbp); if (error) goto out1; error = cgaccount(cg, vp, nbp, 2); bawrite(nbp); if (error) goto out1; } /* * Grab a copy of the superblock and its summary information. * We delay writing it until the suspension is released below. */ copy_fs = malloc((u_long)fs->fs_bsize, M_UFSMNT, M_WAITOK); bcopy(fs, copy_fs, fs->fs_sbsize); if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) copy_fs->fs_clean = 1; size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE; if (fs->fs_sbsize < size) bzero(&((char *)copy_fs)[fs->fs_sbsize], size - fs->fs_sbsize); size = blkroundup(fs, fs->fs_cssize); if (fs->fs_contigsumsize > 0) size += fs->fs_ncg * sizeof(int32_t); space = malloc((u_long)size, M_UFSMNT, M_WAITOK); copy_fs->fs_csp = space; bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize); space = (char *)space + fs->fs_cssize; loc = howmany(fs->fs_cssize, fs->fs_fsize); i = fs->fs_frag - loc % fs->fs_frag; len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize; if (len > 0) { if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc), len, KERNCRED, &bp)) != 0) { brelse(bp); free(copy_fs->fs_csp, M_UFSMNT); free(copy_fs, M_UFSMNT); copy_fs = NULL; goto out1; } bcopy(bp->b_data, space, (u_int)len); space = (char *)space + len; bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); } if (fs->fs_contigsumsize > 0) { copy_fs->fs_maxcluster = lp = space; for (i = 0; i < fs->fs_ncg; i++) *lp++ = fs->fs_contigsumsize; } /* * We must check for active files that have been unlinked * (e.g., with a zero link count). We have to expunge all * trace of these files from the snapshot so that they are * not reclaimed prematurely by fsck or unnecessarily dumped. * We turn off the MNTK_SUSPENDED flag to avoid a panic from * spec_strategy about writing on a suspended filesystem. * Note that we skip unlinked snapshot files as they will * be handled separately below. * * We also calculate the needed size for the snapshot list. */ snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) + FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */; MNT_ILOCK(mp); mp->mnt_kern_flag &= ~MNTK_SUSPENDED; MNT_IUNLOCK(mp); loop: MNT_VNODE_FOREACH_ALL(xvp, mp, mvp) { if ((xvp->v_usecount == 0 && (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) || xvp->v_type == VNON || IS_SNAPSHOT(VTOI(xvp))) { VI_UNLOCK(xvp); continue; } /* * We can skip parent directory vnode because it must have * this snapshot file in it. */ if (xvp == nd.ni_dvp) { VI_UNLOCK(xvp); continue; } vholdl(xvp); if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK) != 0) { MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); vdrop(xvp); goto loop; } VI_LOCK(xvp); if (xvp->v_usecount == 0 && (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) { VI_UNLOCK(xvp); VOP_UNLOCK(xvp, 0); vdrop(xvp); continue; } VI_UNLOCK(xvp); if (snapdebug) vn_printf(xvp, "ffs_snapshot: busy vnode "); if (VOP_GETATTR(xvp, &vat, td->td_ucred) == 0 && vat.va_nlink > 0) { VOP_UNLOCK(xvp, 0); vdrop(xvp); continue; } xp = VTOI(xvp); if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) { VOP_UNLOCK(xvp, 0); vdrop(xvp); continue; } /* * If there is a fragment, clear it here. */ blkno = 0; loc = howmany(xp->i_size, fs->fs_bsize) - 1; if (loc < UFS_NDADDR) { len = fragroundup(fs, blkoff(fs, xp->i_size)); if (len != 0 && len < fs->fs_bsize) { ffs_blkfree(ump, copy_fs, vp, DIP(xp, i_db[loc]), len, xp->i_number, xvp->v_type, NULL, SINGLETON_KEY); blkno = DIP(xp, i_db[loc]); DIP_SET(xp, i_db[loc], 0); } } snaplistsize += 1; if (I_IS_UFS1(xp)) error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1, BLK_NOCOPY, 1); else error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2, BLK_NOCOPY, 1); if (blkno) DIP_SET(xp, i_db[loc], blkno); if (!error) error = ffs_freefile(ump, copy_fs, vp, xp->i_number, xp->i_mode, NULL); VOP_UNLOCK(xvp, 0); vdrop(xvp); if (error) { free(copy_fs->fs_csp, M_UFSMNT); free(copy_fs, M_UFSMNT); copy_fs = NULL; MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto out1; } } /* * Erase the journal file from the snapshot. */ if (fs->fs_flags & FS_SUJ) { error = softdep_journal_lookup(mp, &xvp); if (error) { free(copy_fs->fs_csp, M_UFSMNT); free(copy_fs, M_UFSMNT); copy_fs = NULL; goto out1; } xp = VTOI(xvp); if (I_IS_UFS1(xp)) error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1, BLK_NOCOPY, 0); else error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2, BLK_NOCOPY, 0); vput(xvp); } /* * Acquire a lock on the snapdata structure, creating it if necessary. */ sn = ffs_snapdata_acquire(devvp); /* * Change vnode to use shared snapshot lock instead of the original * private lock. */ vp->v_vnlock = &sn->sn_lock; lockmgr(&vp->v_lock, LK_RELEASE, NULL); xp = TAILQ_FIRST(&sn->sn_head); /* * If this is the first snapshot on this filesystem, then we need * to allocate the space for the list of preallocated snapshot blocks. * This list will be refined below, but this preliminary one will * keep us out of deadlock until the full one is ready. */ if (xp == NULL) { snapblklist = malloc(snaplistsize * sizeof(daddr_t), M_UFSMNT, M_WAITOK); blkp = &snapblklist[1]; *blkp++ = lblkno(fs, fs->fs_sblockloc); blkno = fragstoblks(fs, fs->fs_csaddr); for (cg = 0; cg < fs->fs_ncg; cg++) { if (fragstoblks(fs, cgtod(fs, cg) > blkno)) break; *blkp++ = fragstoblks(fs, cgtod(fs, cg)); } len = howmany(fs->fs_cssize, fs->fs_bsize); for (loc = 0; loc < len; loc++) *blkp++ = blkno + loc; for (; cg < fs->fs_ncg; cg++) *blkp++ = fragstoblks(fs, cgtod(fs, cg)); snapblklist[0] = blkp - snapblklist; VI_LOCK(devvp); if (sn->sn_blklist != NULL) panic("ffs_snapshot: non-empty list"); sn->sn_blklist = snapblklist; sn->sn_listsize = blkp - snapblklist; VI_UNLOCK(devvp); } /* * Record snapshot inode. Since this is the newest snapshot, * it must be placed at the end of the list. */ VI_LOCK(devvp); fs->fs_snapinum[snaploc] = ip->i_number; if (ip->i_nextsnap.tqe_prev != 0) panic("ffs_snapshot: %ju already on list", (uintmax_t)ip->i_number); TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap); devvp->v_vflag |= VV_COPYONWRITE; VI_UNLOCK(devvp); ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp"); out1: KASSERT((sn != NULL && copy_fs != NULL && error == 0) || (sn == NULL && copy_fs == NULL && error != 0), ("email phk@ and mckusick@")); /* * Resume operation on filesystem. */ vfs_write_resume(vp->v_mount, VR_START_WRITE | VR_NO_SUSPCLR); if (collectsnapstats && starttime.tv_sec > 0) { nanotime(&endtime); timespecsub(&endtime, &starttime, &endtime); printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n", vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec, endtime.tv_nsec / 1000000, redo, fs->fs_ncg); } if (copy_fs == NULL) goto out; /* * Copy allocation information from all the snapshots in * this snapshot and then expunge them from its view. */ TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) { if (xp == ip) break; if (I_IS_UFS1(xp)) error = expunge_ufs1(vp, xp, fs, snapacct_ufs1, BLK_SNAP, 0); else error = expunge_ufs2(vp, xp, fs, snapacct_ufs2, BLK_SNAP, 0); if (error == 0 && xp->i_effnlink == 0) { error = ffs_freefile(ump, copy_fs, vp, xp->i_number, xp->i_mode, NULL); } if (error) { fs->fs_snapinum[snaploc] = 0; goto done; } } /* * Allocate space for the full list of preallocated snapshot blocks. */ snapblklist = malloc(snaplistsize * sizeof(daddr_t), M_UFSMNT, M_WAITOK); ip->i_snapblklist = &snapblklist[1]; /* * Expunge the blocks used by the snapshots from the set of * blocks marked as used in the snapshot bitmaps. Also, collect * the list of allocated blocks in i_snapblklist. */ if (I_IS_UFS1(ip)) error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1, BLK_SNAP, 0); else error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2, BLK_SNAP, 0); if (error) { fs->fs_snapinum[snaploc] = 0; free(snapblklist, M_UFSMNT); goto done; } if (snaplistsize < ip->i_snapblklist - snapblklist) panic("ffs_snapshot: list too small"); snaplistsize = ip->i_snapblklist - snapblklist; snapblklist[0] = snaplistsize; ip->i_snapblklist = 0; /* * Write out the list of allocated blocks to the end of the snapshot. */ auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = (void *)snapblklist; aiov.iov_len = snaplistsize * sizeof(daddr_t); auio.uio_resid = aiov.iov_len; auio.uio_offset = ip->i_size; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_WRITE; auio.uio_td = td; if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { fs->fs_snapinum[snaploc] = 0; free(snapblklist, M_UFSMNT); goto done; } /* * Write the superblock and its summary information * to the snapshot. */ blkno = fragstoblks(fs, fs->fs_csaddr); len = howmany(fs->fs_cssize, fs->fs_bsize); space = copy_fs->fs_csp; for (loc = 0; loc < len; loc++) { error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp); if (error) { brelse(nbp); fs->fs_snapinum[snaploc] = 0; free(snapblklist, M_UFSMNT); goto done; } bcopy(space, nbp->b_data, fs->fs_bsize); space = (char *)space + fs->fs_bsize; bawrite(nbp); } error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize, KERNCRED, &nbp); if (error) { brelse(nbp); } else { loc = blkoff(fs, fs->fs_sblockloc); copy_fs->fs_ckhash = ffs_calc_sbhash(copy_fs); bcopy((char *)copy_fs, &nbp->b_data[loc], (u_int)fs->fs_sbsize); bawrite(nbp); } /* * As this is the newest list, it is the most inclusive, so * should replace the previous list. */ VI_LOCK(devvp); space = sn->sn_blklist; sn->sn_blklist = snapblklist; sn->sn_listsize = snaplistsize; VI_UNLOCK(devvp); if (space != NULL) free(space, M_UFSMNT); /* * Preallocate all the direct blocks in the snapshot inode so * that we never have to write the inode itself to commit an * update to the contents of the snapshot. Note that once * created, the size of the snapshot will never change, so * there will never be a need to write the inode except to * update the non-integrity-critical time fields and * allocated-block count. */ for (blockno = 0; blockno < UFS_NDADDR; blockno++) { if (DIP(ip, i_db[blockno]) != 0) continue; error = UFS_BALLOC(vp, lblktosize(fs, blockno), fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp); if (error) break; error = readblock(vp, bp, blockno); bawrite(bp); if (error != 0) break; } done: free(copy_fs->fs_csp, M_UFSMNT); free(copy_fs, M_UFSMNT); copy_fs = NULL; out: NDFREE(&nd, NDF_ONLY_PNBUF); if (saved_nice > 0) { struct proc *p; p = td->td_proc; PROC_LOCK(p); sched_nice(td->td_proc, saved_nice); PROC_UNLOCK(td->td_proc); } UFS_LOCK(ump); if (fs->fs_active != 0) { free(fs->fs_active, M_DEVBUF); fs->fs_active = 0; } UFS_UNLOCK(ump); MNT_ILOCK(mp); mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); MNT_IUNLOCK(mp); if (error) (void) ffs_truncate(vp, (off_t)0, 0, NOCRED); (void) ffs_syncvnode(vp, MNT_WAIT, 0); if (error) vput(vp); else VOP_UNLOCK(vp, 0); vrele(nd.ni_dvp); vn_finished_write(wrtmp); process_deferred_inactive(mp); return (error); } /* * Copy a cylinder group map. All the unallocated blocks are marked * BLK_NOCOPY so that the snapshot knows that it need not copy them * if they are later written. If passno is one, then this is a first * pass, so only setting needs to be done. If passno is 2, then this * is a revision to a previous pass which must be undone as the * replacement pass is done. */ static int cgaccount(cg, vp, nbp, passno) int cg; struct vnode *vp; struct buf *nbp; int passno; { struct buf *bp, *ibp; struct inode *ip; struct cg *cgp; struct fs *fs; ufs2_daddr_t base, numblks; int error, len, loc, indiroff; ip = VTOI(vp); fs = ITOFS(ip); if ((error = ffs_getcg(fs, ITODEVVP(ip), cg, &bp, &cgp)) != 0) return (error); UFS_LOCK(ITOUMP(ip)); ACTIVESET(fs, cg); /* * Recomputation of summary information might not have been performed * at mount time. Sync up summary information for current cylinder * group while data is in memory to ensure that result of background * fsck is slightly more consistent. */ fs->fs_cs(fs, cg) = cgp->cg_cs; UFS_UNLOCK(ITOUMP(ip)); bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize); if (fs->fs_cgsize < fs->fs_bsize) bzero(&nbp->b_data[fs->fs_cgsize], fs->fs_bsize - fs->fs_cgsize); cgp = (struct cg *)nbp->b_data; bqrelse(bp); if (passno == 2) nbp->b_flags |= B_VALIDSUSPWRT; numblks = howmany(fs->fs_size, fs->fs_frag); len = howmany(fs->fs_fpg, fs->fs_frag); base = cgbase(fs, cg) / fs->fs_frag; if (base + len >= numblks) len = numblks - base - 1; loc = 0; if (base < UFS_NDADDR) { for ( ; loc < UFS_NDADDR; loc++) { if (ffs_isblock(fs, cg_blksfree(cgp), loc)) DIP_SET(ip, i_db[loc], BLK_NOCOPY); else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY) DIP_SET(ip, i_db[loc], 0); else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY) panic("ffs_snapshot: lost direct block"); } } error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)), fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); if (error) { goto out; } indiroff = (base + loc - UFS_NDADDR) % NINDIR(fs); for ( ; loc < len; loc++, indiroff++) { if (indiroff >= NINDIR(fs)) { if (passno == 2) ibp->b_flags |= B_VALIDSUSPWRT; bawrite(ibp); error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)), fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); if (error) { goto out; } indiroff = 0; } if (I_IS_UFS1(ip)) { if (ffs_isblock(fs, cg_blksfree(cgp), loc)) ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY; else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0; else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) panic("ffs_snapshot: lost indirect block"); continue; } if (ffs_isblock(fs, cg_blksfree(cgp), loc)) ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY; else if (passno == 2 && ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0; else if (passno == 1 && ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY) panic("ffs_snapshot: lost indirect block"); } if (passno == 2) ibp->b_flags |= B_VALIDSUSPWRT; bdwrite(ibp); out: /* * We have to calculate the crc32c here rather than just setting the * BX_CYLGRP b_xflags because the allocation of the block for the * the cylinder group map will always be a full size block (fs_bsize) * even though the cylinder group may be smaller (fs_cgsize). The * crc32c must be computed only over fs_cgsize whereas the BX_CYLGRP * flag causes it to be computed over the size of the buffer. */ if ((fs->fs_metackhash & CK_CYLGRP) != 0) { ((struct cg *)nbp->b_data)->cg_ckhash = 0; ((struct cg *)nbp->b_data)->cg_ckhash = calculate_crc32c(~0L, nbp->b_data, fs->fs_cgsize); } return (error); } /* * Before expunging a snapshot inode, note all the * blocks that it claims with BLK_SNAP so that fsck will * be able to account for those blocks properly and so * that this snapshot knows that it need not copy them * if the other snapshot holding them is freed. This code * is reproduced once each for UFS1 and UFS2. */ static int expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype, clearmode) struct vnode *snapvp; struct inode *cancelip; struct fs *fs; int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, ufs_lbn_t, int); int expungetype; int clearmode; { int i, error, indiroff; ufs_lbn_t lbn, rlbn; ufs2_daddr_t len, blkno, numblks, blksperindir; struct ufs1_dinode *dip; struct thread *td = curthread; struct buf *bp; /* * Prepare to expunge the inode. If its inode block has not * yet been copied, then allocate and fill the copy. */ lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); blkno = 0; if (lbn < UFS_NDADDR) { blkno = VTOI(snapvp)->i_din1->di_db[lbn]; } else { if (DOINGSOFTDEP(snapvp)) softdep_prealloc(snapvp, MNT_WAIT); td->td_pflags |= TDP_COWINPROGRESS; error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); td->td_pflags &= ~TDP_COWINPROGRESS; if (error) return (error); indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff]; bqrelse(bp); } if (blkno != 0) { if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) return (error); } else { error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, 0, &bp); if (error) return (error); if ((error = readblock(snapvp, bp, lbn)) != 0) return (error); } /* * Set a snapshot inode to be a zero length file, regular files * or unlinked snapshots to be completely unallocated. */ dip = (struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, cancelip->i_number); if (clearmode || cancelip->i_effnlink == 0) dip->di_mode = 0; dip->di_size = 0; dip->di_blocks = 0; dip->di_flags &= ~SF_SNAPSHOT; bzero(&dip->di_db[0], (UFS_NDADDR + UFS_NIADDR) * sizeof(ufs1_daddr_t)); bdwrite(bp); /* * Now go through and expunge all the blocks in the file * using the function requested. */ numblks = howmany(cancelip->i_size, fs->fs_bsize); if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0], &cancelip->i_din1->di_db[UFS_NDADDR], fs, 0, expungetype))) return (error); if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0], &cancelip->i_din1->di_ib[UFS_NIADDR], fs, -1, expungetype))) return (error); blksperindir = 1; lbn = -UFS_NDADDR; len = numblks - UFS_NDADDR; rlbn = UFS_NDADDR; for (i = 0; len > 0 && i < UFS_NIADDR; i++) { error = indiracct_ufs1(snapvp, ITOV(cancelip), i, cancelip->i_din1->di_ib[i], lbn, rlbn, len, blksperindir, fs, acctfunc, expungetype); if (error) return (error); blksperindir *= NINDIR(fs); lbn -= blksperindir + 1; len -= blksperindir; rlbn += blksperindir; } return (0); } /* * Descend an indirect block chain for vnode cancelvp accounting for all * its indirect blocks in snapvp. */ static int indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, blksperindir, fs, acctfunc, expungetype) struct vnode *snapvp; struct vnode *cancelvp; int level; ufs1_daddr_t blkno; ufs_lbn_t lbn; ufs_lbn_t rlbn; ufs_lbn_t remblks; ufs_lbn_t blksperindir; struct fs *fs; int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *, ufs_lbn_t, int); int expungetype; { int error, num, i; ufs_lbn_t subblksperindir; struct indir indirs[UFS_NIADDR + 2]; ufs1_daddr_t last, *bap; struct buf *bp; if (blkno == 0) { if (expungetype == BLK_NOCOPY) return (0); panic("indiracct_ufs1: missing indir"); } if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) return (error); if (lbn != indirs[num - 1 - level].in_lbn || num < 2) panic("indiracct_ufs1: botched params"); /* * We have to expand bread here since it will deadlock looking * up the block number for any blocks that are not in the cache. */ bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0); bp->b_blkno = fsbtodb(fs, blkno); if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 && (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) { brelse(bp); return (error); } /* * Account for the block pointers in this indirect block. */ last = howmany(remblks, blksperindir); if (last > NINDIR(fs)) last = NINDIR(fs); bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK); bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); bqrelse(bp); error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, level == 0 ? rlbn : -1, expungetype); if (error || level == 0) goto out; /* * Account for the block pointers in each of the indirect blocks * in the levels below us. */ subblksperindir = blksperindir / NINDIR(fs); for (lbn++, level--, i = 0; i < last; i++) { error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn, rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); if (error) goto out; rlbn += blksperindir; lbn -= blksperindir; remblks -= blksperindir; } out: free(bap, M_DEVBUF); return (error); } /* * Do both snap accounting and map accounting. */ static int fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype) struct vnode *vp; ufs1_daddr_t *oldblkp, *lastblkp; struct fs *fs; ufs_lbn_t lblkno; int exptype; /* BLK_SNAP or BLK_NOCOPY */ { int error; if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype))) return (error); return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)); } /* * Identify a set of blocks allocated in a snapshot inode. */ static int snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) struct vnode *vp; ufs1_daddr_t *oldblkp, *lastblkp; struct fs *fs; ufs_lbn_t lblkno; int expungetype; /* BLK_SNAP or BLK_NOCOPY */ { struct inode *ip = VTOI(vp); ufs1_daddr_t blkno, *blkp; ufs_lbn_t lbn; struct buf *ibp; int error; for ( ; oldblkp < lastblkp; oldblkp++) { blkno = *oldblkp; if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) continue; lbn = fragstoblks(fs, blkno); if (lbn < UFS_NDADDR) { blkp = &ip->i_din1->di_db[lbn]; ip->i_flag |= IN_CHANGE | IN_UPDATE; } else { error = ffs_balloc_ufs1(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); if (error) return (error); blkp = &((ufs1_daddr_t *)(ibp->b_data)) [(lbn - UFS_NDADDR) % NINDIR(fs)]; } /* * If we are expunging a snapshot vnode and we * find a block marked BLK_NOCOPY, then it is * one that has been allocated to this snapshot after * we took our current snapshot and can be ignored. */ if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { if (lbn >= UFS_NDADDR) brelse(ibp); } else { if (*blkp != 0) panic("snapacct_ufs1: bad block"); *blkp = expungetype; if (lbn >= UFS_NDADDR) bdwrite(ibp); } } return (0); } /* * Account for a set of blocks allocated in a snapshot inode. */ static int mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype) struct vnode *vp; ufs1_daddr_t *oldblkp, *lastblkp; struct fs *fs; ufs_lbn_t lblkno; int expungetype; { ufs1_daddr_t blkno; struct inode *ip; ino_t inum; int acctit; ip = VTOI(vp); inum = ip->i_number; if (lblkno == -1) acctit = 0; else acctit = 1; for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { blkno = *oldblkp; if (blkno == 0 || blkno == BLK_NOCOPY) continue; if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP) *ip->i_snapblklist++ = lblkno; if (blkno == BLK_SNAP) blkno = blkstofrags(fs, lblkno); ffs_blkfree(ITOUMP(ip), fs, vp, blkno, fs->fs_bsize, inum, vp->v_type, NULL, SINGLETON_KEY); } return (0); } /* * Before expunging a snapshot inode, note all the * blocks that it claims with BLK_SNAP so that fsck will * be able to account for those blocks properly and so * that this snapshot knows that it need not copy them * if the other snapshot holding them is freed. This code * is reproduced once each for UFS1 and UFS2. */ static int expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype, clearmode) struct vnode *snapvp; struct inode *cancelip; struct fs *fs; int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, ufs_lbn_t, int); int expungetype; int clearmode; { int i, error, indiroff; ufs_lbn_t lbn, rlbn; ufs2_daddr_t len, blkno, numblks, blksperindir; struct ufs2_dinode *dip; struct thread *td = curthread; struct buf *bp; /* * Prepare to expunge the inode. If its inode block has not * yet been copied, then allocate and fill the copy. */ lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number)); blkno = 0; if (lbn < UFS_NDADDR) { blkno = VTOI(snapvp)->i_din2->di_db[lbn]; } else { if (DOINGSOFTDEP(snapvp)) softdep_prealloc(snapvp, MNT_WAIT); td->td_pflags |= TDP_COWINPROGRESS; error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, BA_METAONLY, &bp); td->td_pflags &= ~TDP_COWINPROGRESS; if (error) return (error); indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff]; bqrelse(bp); } if (blkno != 0) { if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp))) return (error); } else { error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, 0, &bp); if (error) return (error); if ((error = readblock(snapvp, bp, lbn)) != 0) return (error); } /* * Set a snapshot inode to be a zero length file, regular files * to be completely unallocated. */ dip = (struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, cancelip->i_number); dip->di_size = 0; dip->di_blocks = 0; dip->di_flags &= ~SF_SNAPSHOT; bzero(&dip->di_db[0], (UFS_NDADDR + UFS_NIADDR) * sizeof(ufs2_daddr_t)); if (clearmode || cancelip->i_effnlink == 0) dip->di_mode = 0; else ffs_update_dinode_ckhash(fs, dip); bdwrite(bp); /* * Now go through and expunge all the blocks in the file * using the function requested. */ numblks = howmany(cancelip->i_size, fs->fs_bsize); if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0], &cancelip->i_din2->di_db[UFS_NDADDR], fs, 0, expungetype))) return (error); if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0], &cancelip->i_din2->di_ib[UFS_NIADDR], fs, -1, expungetype))) return (error); blksperindir = 1; lbn = -UFS_NDADDR; len = numblks - UFS_NDADDR; rlbn = UFS_NDADDR; for (i = 0; len > 0 && i < UFS_NIADDR; i++) { error = indiracct_ufs2(snapvp, ITOV(cancelip), i, cancelip->i_din2->di_ib[i], lbn, rlbn, len, blksperindir, fs, acctfunc, expungetype); if (error) return (error); blksperindir *= NINDIR(fs); lbn -= blksperindir + 1; len -= blksperindir; rlbn += blksperindir; } return (0); } /* * Descend an indirect block chain for vnode cancelvp accounting for all * its indirect blocks in snapvp. */ static int indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, blksperindir, fs, acctfunc, expungetype) struct vnode *snapvp; struct vnode *cancelvp; int level; ufs2_daddr_t blkno; ufs_lbn_t lbn; ufs_lbn_t rlbn; ufs_lbn_t remblks; ufs_lbn_t blksperindir; struct fs *fs; int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *, ufs_lbn_t, int); int expungetype; { int error, num, i; ufs_lbn_t subblksperindir; struct indir indirs[UFS_NIADDR + 2]; ufs2_daddr_t last, *bap; struct buf *bp; if (blkno == 0) { if (expungetype == BLK_NOCOPY) return (0); panic("indiracct_ufs2: missing indir"); } if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0) return (error); if (lbn != indirs[num - 1 - level].in_lbn || num < 2) panic("indiracct_ufs2: botched params"); /* * We have to expand bread here since it will deadlock looking * up the block number for any blocks that are not in the cache. */ bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0); bp->b_blkno = fsbtodb(fs, blkno); if ((bp->b_flags & B_CACHE) == 0 && (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) { brelse(bp); return (error); } /* * Account for the block pointers in this indirect block. */ last = howmany(remblks, blksperindir); if (last > NINDIR(fs)) last = NINDIR(fs); bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK); bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize); bqrelse(bp); error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs, level == 0 ? rlbn : -1, expungetype); if (error || level == 0) goto out; /* * Account for the block pointers in each of the indirect blocks * in the levels below us. */ subblksperindir = blksperindir / NINDIR(fs); for (lbn++, level--, i = 0; i < last; i++) { error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn, rlbn, remblks, subblksperindir, fs, acctfunc, expungetype); if (error) goto out; rlbn += blksperindir; lbn -= blksperindir; remblks -= blksperindir; } out: free(bap, M_DEVBUF); return (error); } /* * Do both snap accounting and map accounting. */ static int fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype) struct vnode *vp; ufs2_daddr_t *oldblkp, *lastblkp; struct fs *fs; ufs_lbn_t lblkno; int exptype; /* BLK_SNAP or BLK_NOCOPY */ { int error; if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype))) return (error); return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)); } /* * Identify a set of blocks allocated in a snapshot inode. */ static int snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) struct vnode *vp; ufs2_daddr_t *oldblkp, *lastblkp; struct fs *fs; ufs_lbn_t lblkno; int expungetype; /* BLK_SNAP or BLK_NOCOPY */ { struct inode *ip = VTOI(vp); ufs2_daddr_t blkno, *blkp; ufs_lbn_t lbn; struct buf *ibp; int error; for ( ; oldblkp < lastblkp; oldblkp++) { blkno = *oldblkp; if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP) continue; lbn = fragstoblks(fs, blkno); if (lbn < UFS_NDADDR) { blkp = &ip->i_din2->di_db[lbn]; ip->i_flag |= IN_CHANGE | IN_UPDATE; } else { error = ffs_balloc_ufs2(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); if (error) return (error); blkp = &((ufs2_daddr_t *)(ibp->b_data)) [(lbn - UFS_NDADDR) % NINDIR(fs)]; } /* * If we are expunging a snapshot vnode and we * find a block marked BLK_NOCOPY, then it is * one that has been allocated to this snapshot after * we took our current snapshot and can be ignored. */ if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) { if (lbn >= UFS_NDADDR) brelse(ibp); } else { if (*blkp != 0) panic("snapacct_ufs2: bad block"); *blkp = expungetype; if (lbn >= UFS_NDADDR) bdwrite(ibp); } } return (0); } /* * Account for a set of blocks allocated in a snapshot inode. */ static int mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype) struct vnode *vp; ufs2_daddr_t *oldblkp, *lastblkp; struct fs *fs; ufs_lbn_t lblkno; int expungetype; { ufs2_daddr_t blkno; struct inode *ip; ino_t inum; int acctit; ip = VTOI(vp); inum = ip->i_number; if (lblkno == -1) acctit = 0; else acctit = 1; for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) { blkno = *oldblkp; if (blkno == 0 || blkno == BLK_NOCOPY) continue; if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP) *ip->i_snapblklist++ = lblkno; if (blkno == BLK_SNAP) blkno = blkstofrags(fs, lblkno); ffs_blkfree(ITOUMP(ip), fs, vp, blkno, fs->fs_bsize, inum, vp->v_type, NULL, SINGLETON_KEY); } return (0); } /* * Decrement extra reference on snapshot when last name is removed. * It will not be freed until the last open reference goes away. */ void ffs_snapgone(ip) struct inode *ip; { struct inode *xp; struct fs *fs; int snaploc; struct snapdata *sn; struct ufsmount *ump; /* * Find snapshot in incore list. */ xp = NULL; sn = ITODEVVP(ip)->v_rdev->si_snapdata; if (sn != NULL) TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) if (xp == ip) break; if (xp != NULL) vrele(ITOV(ip)); else if (snapdebug) printf("ffs_snapgone: lost snapshot vnode %ju\n", (uintmax_t)ip->i_number); /* * Delete snapshot inode from superblock. Keep list dense. */ ump = ITOUMP(ip); fs = ump->um_fs; UFS_LOCK(ump); for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) if (fs->fs_snapinum[snaploc] == ip->i_number) break; if (snaploc < FSMAXSNAP) { for (snaploc++; snaploc < FSMAXSNAP; snaploc++) { if (fs->fs_snapinum[snaploc] == 0) break; fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc]; } fs->fs_snapinum[snaploc - 1] = 0; } UFS_UNLOCK(ump); } /* * Prepare a snapshot file for being removed. */ void ffs_snapremove(vp) struct vnode *vp; { struct inode *ip; struct vnode *devvp; struct buf *ibp; struct fs *fs; ufs2_daddr_t numblks, blkno, dblk; int error, i, last, loc; struct snapdata *sn; ip = VTOI(vp); fs = ITOFS(ip); devvp = ITODEVVP(ip); /* * If active, delete from incore list (this snapshot may * already have been in the process of being deleted, so * would not have been active). * * Clear copy-on-write flag if last snapshot. */ VI_LOCK(devvp); if (ip->i_nextsnap.tqe_prev != 0) { sn = devvp->v_rdev->si_snapdata; TAILQ_REMOVE(&sn->sn_head, ip, i_nextsnap); ip->i_nextsnap.tqe_prev = 0; VI_UNLOCK(devvp); lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); for (i = 0; i < sn->sn_lock.lk_recurse; i++) lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); KASSERT(vp->v_vnlock == &sn->sn_lock, ("ffs_snapremove: lost lock mutation")); vp->v_vnlock = &vp->v_lock; VI_LOCK(devvp); while (sn->sn_lock.lk_recurse > 0) lockmgr(&sn->sn_lock, LK_RELEASE, NULL); lockmgr(&sn->sn_lock, LK_RELEASE, NULL); try_free_snapdata(devvp); } else VI_UNLOCK(devvp); /* * Clear all BLK_NOCOPY fields. Pass any block claims to other * snapshots that want them (see ffs_snapblkfree below). */ for (blkno = 1; blkno < UFS_NDADDR; blkno++) { dblk = DIP(ip, i_db[blkno]); if (dblk == 0) continue; if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) DIP_SET(ip, i_db[blkno], 0); else if ((dblk == blkstofrags(fs, blkno) && ffs_snapblkfree(fs, ITODEVVP(ip), dblk, fs->fs_bsize, ip->i_number, vp->v_type, NULL))) { DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - btodb(fs->fs_bsize)); DIP_SET(ip, i_db[blkno], 0); } } numblks = howmany(ip->i_size, fs->fs_bsize); for (blkno = UFS_NDADDR; blkno < numblks; blkno += NINDIR(fs)) { error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno), fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); if (error) continue; if (fs->fs_size - blkno > NINDIR(fs)) last = NINDIR(fs); else last = fs->fs_size - blkno; for (loc = 0; loc < last; loc++) { if (I_IS_UFS1(ip)) { dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc]; if (dblk == 0) continue; if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; else if ((dblk == blkstofrags(fs, blkno) && ffs_snapblkfree(fs, ITODEVVP(ip), dblk, fs->fs_bsize, ip->i_number, vp->v_type, NULL))) { ip->i_din1->di_blocks -= btodb(fs->fs_bsize); ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0; } continue; } dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc]; if (dblk == 0) continue; if (dblk == BLK_NOCOPY || dblk == BLK_SNAP) ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; else if ((dblk == blkstofrags(fs, blkno) && ffs_snapblkfree(fs, ITODEVVP(ip), dblk, fs->fs_bsize, ip->i_number, vp->v_type, NULL))) { ip->i_din2->di_blocks -= btodb(fs->fs_bsize); ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0; } } bawrite(ibp); } /* * Clear snapshot flag and drop reference. */ ip->i_flags &= ~SF_SNAPSHOT; DIP_SET(ip, i_flags, ip->i_flags); ip->i_flag |= IN_CHANGE | IN_UPDATE; /* * The dirtied indirects must be written out before * softdep_setup_freeblocks() is called. Otherwise indir_trunc() * may find indirect pointers using the magic BLK_* values. */ if (DOINGSOFTDEP(vp)) ffs_syncvnode(vp, MNT_WAIT, 0); #ifdef QUOTA /* * Reenable disk quotas for ex-snapshot file. */ if (!getinoquota(ip)) (void) chkdq(ip, DIP(ip, i_blocks), KERNCRED, FORCE); #endif } /* * Notification that a block is being freed. Return zero if the free * should be allowed to proceed. Return non-zero if the snapshot file * wants to claim the block. The block will be claimed if it is an * uncopied part of one of the snapshots. It will be freed if it is * either a BLK_NOCOPY or has already been copied in all of the snapshots. * If a fragment is being freed, then all snapshots that care about * it must make a copy since a snapshot file can only claim full sized * blocks. Note that if more than one snapshot file maps the block, * we can pick one at random to claim it. Since none of the snapshots * can change, we are assurred that they will all see the same unmodified * image. When deleting a snapshot file (see ffs_snapremove above), we * must push any of these claimed blocks to one of the other snapshots * that maps it. These claimed blocks are easily identified as they will * have a block number equal to their logical block number within the * snapshot. A copied block can never have this property because they * must always have been allocated from a BLK_NOCOPY location. */ int ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd) struct fs *fs; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; enum vtype vtype; struct workhead *wkhd; { struct buf *ibp, *cbp, *savedcbp = NULL; struct thread *td = curthread; struct inode *ip; struct vnode *vp = NULL; ufs_lbn_t lbn; ufs2_daddr_t blkno; int indiroff = 0, error = 0, claimedblk = 0; struct snapdata *sn; lbn = fragstoblks(fs, bno); retry: VI_LOCK(devvp); sn = devvp->v_rdev->si_snapdata; if (sn == NULL) { VI_UNLOCK(devvp); return (0); } if (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, VI_MTX(devvp)) != 0) goto retry; TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { vp = ITOV(ip); if (DOINGSOFTDEP(vp)) softdep_prealloc(vp, MNT_WAIT); /* * Lookup block being written. */ if (lbn < UFS_NDADDR) { blkno = DIP(ip, i_db[lbn]); } else { td->td_pflags |= TDP_COWINPROGRESS; error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); td->td_pflags &= ~TDP_COWINPROGRESS; if (error) break; indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); if (I_IS_UFS1(ip)) blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; else blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; } /* * Check to see if block needs to be copied. */ if (blkno == 0) { /* * A block that we map is being freed. If it has not * been claimed yet, we will claim or copy it (below). */ claimedblk = 1; } else if (blkno == BLK_SNAP) { /* * No previous snapshot claimed the block, * so it will be freed and become a BLK_NOCOPY * (don't care) for us. */ if (claimedblk) panic("snapblkfree: inconsistent block type"); if (lbn < UFS_NDADDR) { DIP_SET(ip, i_db[lbn], BLK_NOCOPY); ip->i_flag |= IN_CHANGE | IN_UPDATE; } else if (I_IS_UFS1(ip)) { ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY; bdwrite(ibp); } else { ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY; bdwrite(ibp); } continue; } else /* BLK_NOCOPY or default */ { /* * If the snapshot has already copied the block * (default), or does not care about the block, * it is not needed. */ if (lbn >= UFS_NDADDR) bqrelse(ibp); continue; } /* * If this is a full size block, we will just grab it * and assign it to the snapshot inode. Otherwise we * will proceed to copy it. See explanation for this * routine as to why only a single snapshot needs to * claim this block. */ if (size == fs->fs_bsize) { #ifdef DEBUG if (snapdebug) printf("%s %ju lbn %jd from inum %ju\n", "Grabonremove: snapino", (uintmax_t)ip->i_number, (intmax_t)lbn, (uintmax_t)inum); #endif /* * If journaling is tracking this write we must add * the work to the inode or indirect being written. */ if (wkhd != NULL) { if (lbn < UFS_NDADDR) softdep_inode_append(ip, curthread->td_ucred, wkhd); else softdep_buf_append(ibp, wkhd); } if (lbn < UFS_NDADDR) { DIP_SET(ip, i_db[lbn], bno); } else if (I_IS_UFS1(ip)) { ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno; bdwrite(ibp); } else { ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno; bdwrite(ibp); } DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size)); ip->i_flag |= IN_CHANGE | IN_UPDATE; lockmgr(vp->v_vnlock, LK_RELEASE, NULL); return (1); } if (lbn >= UFS_NDADDR) bqrelse(ibp); /* * Allocate the block into which to do the copy. Note that this * allocation will never require any additional allocations for * the snapshot inode. */ td->td_pflags |= TDP_COWINPROGRESS; error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, 0, &cbp); td->td_pflags &= ~TDP_COWINPROGRESS; if (error) break; #ifdef DEBUG if (snapdebug) printf("%s%ju lbn %jd %s %ju size %ld to blkno %jd\n", "Copyonremove: snapino ", (uintmax_t)ip->i_number, (intmax_t)lbn, "for inum", (uintmax_t)inum, size, (intmax_t)cbp->b_blkno); #endif /* * If we have already read the old block contents, then * simply copy them to the new block. Note that we need * to synchronously write snapshots that have not been * unlinked, and hence will be visible after a crash, * to ensure their integrity. At a minimum we ensure the * integrity of the filesystem metadata, but use the * dopersistence sysctl-setable flag to decide on the * persistence needed for file content data. */ if (savedcbp != NULL) { bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); bawrite(cbp); if ((vtype == VDIR || dopersistence) && ip->i_effnlink > 0) (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); continue; } /* * Otherwise, read the old block contents into the buffer. */ if ((error = readblock(vp, cbp, lbn)) != 0) { bzero(cbp->b_data, fs->fs_bsize); bawrite(cbp); if ((vtype == VDIR || dopersistence) && ip->i_effnlink > 0) (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); break; } savedcbp = cbp; } /* * Note that we need to synchronously write snapshots that * have not been unlinked, and hence will be visible after * a crash, to ensure their integrity. At a minimum we * ensure the integrity of the filesystem metadata, but * use the dopersistence sysctl-setable flag to decide on * the persistence needed for file content data. */ if (savedcbp) { vp = savedcbp->b_vp; bawrite(savedcbp); if ((vtype == VDIR || dopersistence) && VTOI(vp)->i_effnlink > 0) (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); } /* * If we have been unable to allocate a block in which to do * the copy, then return non-zero so that the fragment will * not be freed. Although space will be lost, the snapshot * will stay consistent. */ if (error != 0 && wkhd != NULL) softdep_freework(wkhd); lockmgr(&sn->sn_lock, LK_RELEASE, NULL); return (error); } /* * Associate snapshot files when mounting. */ void ffs_snapshot_mount(mp) struct mount *mp; { struct ufsmount *ump = VFSTOUFS(mp); struct vnode *devvp = ump->um_devvp; struct fs *fs = ump->um_fs; struct thread *td = curthread; struct snapdata *sn; struct vnode *vp; struct vnode *lastvp; struct inode *ip; struct uio auio; struct iovec aiov; void *snapblklist; char *reason; daddr_t snaplistsize; int error, snaploc, loc; /* * XXX The following needs to be set before ffs_truncate or * VOP_READ can be called. */ mp->mnt_stat.f_iosize = fs->fs_bsize; /* * Process each snapshot listed in the superblock. */ vp = NULL; lastvp = NULL; sn = NULL; for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) { if (fs->fs_snapinum[snaploc] == 0) break; if ((error = ffs_vget(mp, fs->fs_snapinum[snaploc], LK_EXCLUSIVE, &vp)) != 0){ printf("ffs_snapshot_mount: vget failed %d\n", error); continue; } ip = VTOI(vp); - if (!IS_SNAPSHOT(ip) || ip->i_size == + if (vp->v_type != VREG) { + reason = "non-file snapshot"; + } else if (!IS_SNAPSHOT(ip)) { + reason = "non-snapshot"; + } else if (ip->i_size == lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) { - if (!IS_SNAPSHOT(ip)) { - reason = "non-snapshot"; - } else { - reason = "old format snapshot"; - (void)ffs_truncate(vp, (off_t)0, 0, NOCRED); - (void)ffs_syncvnode(vp, MNT_WAIT, 0); - } + reason = "old format snapshot"; + (void)ffs_truncate(vp, (off_t)0, 0, NOCRED); + (void)ffs_syncvnode(vp, MNT_WAIT, 0); + } else { + reason = NULL; + } + if (reason != NULL) { printf("ffs_snapshot_mount: %s inode %d\n", reason, fs->fs_snapinum[snaploc]); vput(vp); vp = NULL; for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) { if (fs->fs_snapinum[loc] == 0) break; fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc]; } fs->fs_snapinum[loc - 1] = 0; snaploc--; continue; } /* * Acquire a lock on the snapdata structure, creating it if * necessary. */ sn = ffs_snapdata_acquire(devvp); /* * Change vnode to use shared snapshot lock instead of the * original private lock. */ vp->v_vnlock = &sn->sn_lock; lockmgr(&vp->v_lock, LK_RELEASE, NULL); /* * Link it onto the active snapshot list. */ VI_LOCK(devvp); if (ip->i_nextsnap.tqe_prev != 0) panic("ffs_snapshot_mount: %ju already on list", (uintmax_t)ip->i_number); else TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap); vp->v_vflag |= VV_SYSTEM; VI_UNLOCK(devvp); VOP_UNLOCK(vp, 0); lastvp = vp; } vp = lastvp; /* * No usable snapshots found. */ if (sn == NULL || vp == NULL) return; /* * Allocate the space for the block hints list. We always want to * use the list from the newest snapshot. */ auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = (void *)&snaplistsize; aiov.iov_len = sizeof(snaplistsize); auio.uio_resid = aiov.iov_len; auio.uio_offset = lblktosize(fs, howmany(fs->fs_size, fs->fs_frag)); auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_READ; auio.uio_td = td; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { printf("ffs_snapshot_mount: read_1 failed %d\n", error); VOP_UNLOCK(vp, 0); return; } snapblklist = malloc(snaplistsize * sizeof(daddr_t), M_UFSMNT, M_WAITOK); auio.uio_iovcnt = 1; aiov.iov_base = snapblklist; aiov.iov_len = snaplistsize * sizeof (daddr_t); auio.uio_resid = aiov.iov_len; auio.uio_offset -= sizeof(snaplistsize); if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) { printf("ffs_snapshot_mount: read_2 failed %d\n", error); VOP_UNLOCK(vp, 0); free(snapblklist, M_UFSMNT); return; } VOP_UNLOCK(vp, 0); VI_LOCK(devvp); ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount"); sn->sn_listsize = snaplistsize; sn->sn_blklist = (daddr_t *)snapblklist; devvp->v_vflag |= VV_COPYONWRITE; VI_UNLOCK(devvp); } /* * Disassociate snapshot files when unmounting. */ void ffs_snapshot_unmount(mp) struct mount *mp; { struct vnode *devvp = VFSTOUFS(mp)->um_devvp; struct snapdata *sn; struct inode *xp; struct vnode *vp; VI_LOCK(devvp); sn = devvp->v_rdev->si_snapdata; while (sn != NULL && (xp = TAILQ_FIRST(&sn->sn_head)) != NULL) { vp = ITOV(xp); TAILQ_REMOVE(&sn->sn_head, xp, i_nextsnap); xp->i_nextsnap.tqe_prev = 0; lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE, VI_MTX(devvp)); lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL); KASSERT(vp->v_vnlock == &sn->sn_lock, ("ffs_snapshot_unmount: lost lock mutation")); vp->v_vnlock = &vp->v_lock; lockmgr(&vp->v_lock, LK_RELEASE, NULL); lockmgr(&sn->sn_lock, LK_RELEASE, NULL); if (xp->i_effnlink > 0) vrele(vp); VI_LOCK(devvp); sn = devvp->v_rdev->si_snapdata; } try_free_snapdata(devvp); ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount"); } /* * Check the buffer block to be belong to device buffer that shall be * locked after snaplk. devvp shall be locked on entry, and will be * leaved locked upon exit. */ static int ffs_bp_snapblk(devvp, bp) struct vnode *devvp; struct buf *bp; { struct snapdata *sn; struct fs *fs; ufs2_daddr_t lbn, *snapblklist; int lower, upper, mid; ASSERT_VI_LOCKED(devvp, "ffs_bp_snapblk"); KASSERT(devvp->v_type == VCHR, ("Not a device %p", devvp)); sn = devvp->v_rdev->si_snapdata; if (sn == NULL || TAILQ_FIRST(&sn->sn_head) == NULL) return (0); fs = ITOFS(TAILQ_FIRST(&sn->sn_head)); lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno)); snapblklist = sn->sn_blklist; upper = sn->sn_listsize - 1; lower = 1; while (lower <= upper) { mid = (lower + upper) / 2; if (snapblklist[mid] == lbn) break; if (snapblklist[mid] < lbn) lower = mid + 1; else upper = mid - 1; } if (lower <= upper) return (1); return (0); } void ffs_bdflush(bo, bp) struct bufobj *bo; struct buf *bp; { struct thread *td; struct vnode *vp, *devvp; struct buf *nbp; int bp_bdskip; if (bo->bo_dirty.bv_cnt <= dirtybufthresh) return; td = curthread; vp = bp->b_vp; devvp = bo2vnode(bo); KASSERT(vp == devvp, ("devvp != vp %p %p", bo, bp)); VI_LOCK(devvp); bp_bdskip = ffs_bp_snapblk(devvp, bp); if (bp_bdskip) bdwriteskip++; VI_UNLOCK(devvp); if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10 && !bp_bdskip) { (void) VOP_FSYNC(vp, MNT_NOWAIT, td); altbufferflushes++; } else { BO_LOCK(bo); /* * Try to find a buffer to flush. */ TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) { if ((nbp->b_vflags & BV_BKGRDINPROG) || BUF_LOCK(nbp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) continue; if (bp == nbp) panic("bdwrite: found ourselves"); BO_UNLOCK(bo); /* * Don't countdeps with the bo lock * held. */ if (buf_countdeps(nbp, 0)) { BO_LOCK(bo); BUF_UNLOCK(nbp); continue; } if (bp_bdskip) { VI_LOCK(devvp); if (!ffs_bp_snapblk(vp, nbp)) { VI_UNLOCK(devvp); BO_LOCK(bo); BUF_UNLOCK(nbp); continue; } VI_UNLOCK(devvp); } if (nbp->b_flags & B_CLUSTEROK) { vfs_bio_awrite(nbp); } else { bremfree(nbp); bawrite(nbp); } dirtybufferflushes++; break; } if (nbp == NULL) BO_UNLOCK(bo); } } /* * Check for need to copy block that is about to be written, * copying the block if necessary. */ int ffs_copyonwrite(devvp, bp) struct vnode *devvp; struct buf *bp; { struct snapdata *sn; struct buf *ibp, *cbp, *savedcbp = NULL; struct thread *td = curthread; struct fs *fs; struct inode *ip; struct vnode *vp = NULL; ufs2_daddr_t lbn, blkno, *snapblklist; int lower, upper, mid, indiroff, error = 0; int launched_async_io, prev_norunningbuf; long saved_runningbufspace; if (devvp != bp->b_vp && IS_SNAPSHOT(VTOI(bp->b_vp))) return (0); /* Update on a snapshot file */ if (td->td_pflags & TDP_COWINPROGRESS) panic("ffs_copyonwrite: recursive call"); /* * First check to see if it is in the preallocated list. * By doing this check we avoid several potential deadlocks. */ VI_LOCK(devvp); sn = devvp->v_rdev->si_snapdata; if (sn == NULL || TAILQ_EMPTY(&sn->sn_head)) { VI_UNLOCK(devvp); return (0); /* No snapshot */ } ip = TAILQ_FIRST(&sn->sn_head); fs = ITOFS(ip); lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno)); snapblklist = sn->sn_blklist; upper = sn->sn_listsize - 1; lower = 1; while (lower <= upper) { mid = (lower + upper) / 2; if (snapblklist[mid] == lbn) break; if (snapblklist[mid] < lbn) lower = mid + 1; else upper = mid - 1; } if (lower <= upper) { VI_UNLOCK(devvp); return (0); } launched_async_io = 0; prev_norunningbuf = td->td_pflags & TDP_NORUNNINGBUF; /* * Since I/O on bp isn't yet in progress and it may be blocked * for a long time waiting on snaplk, back it out of * runningbufspace, possibly waking other threads waiting for space. */ saved_runningbufspace = bp->b_runningbufspace; if (saved_runningbufspace != 0) runningbufwakeup(bp); /* * Not in the precomputed list, so check the snapshots. */ while (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, VI_MTX(devvp)) != 0) { VI_LOCK(devvp); sn = devvp->v_rdev->si_snapdata; if (sn == NULL || TAILQ_EMPTY(&sn->sn_head)) { VI_UNLOCK(devvp); if (saved_runningbufspace != 0) { bp->b_runningbufspace = saved_runningbufspace; atomic_add_long(&runningbufspace, bp->b_runningbufspace); } return (0); /* Snapshot gone */ } } TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { vp = ITOV(ip); if (DOINGSOFTDEP(vp)) softdep_prealloc(vp, MNT_WAIT); /* * We ensure that everything of our own that needs to be * copied will be done at the time that ffs_snapshot is * called. Thus we can skip the check here which can * deadlock in doing the lookup in UFS_BALLOC. */ if (bp->b_vp == vp) continue; /* * Check to see if block needs to be copied. We do not have * to hold the snapshot lock while doing this lookup as it * will never require any additional allocations for the * snapshot inode. */ if (lbn < UFS_NDADDR) { blkno = DIP(ip, i_db[lbn]); } else { td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF; error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp); td->td_pflags &= ~TDP_COWINPROGRESS; if (error) break; indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); if (I_IS_UFS1(ip)) blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff]; else blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff]; bqrelse(ibp); } #ifdef INVARIANTS if (blkno == BLK_SNAP && bp->b_lblkno >= 0) panic("ffs_copyonwrite: bad copy block"); #endif if (blkno != 0) continue; /* * Allocate the block into which to do the copy. Since * multiple processes may all try to copy the same block, * we have to recheck our need to do a copy if we sleep * waiting for the lock. * * Because all snapshots on a filesystem share a single * lock, we ensure that we will never be in competition * with another process to allocate a block. */ td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF; error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, KERNCRED, 0, &cbp); td->td_pflags &= ~TDP_COWINPROGRESS; if (error) break; #ifdef DEBUG if (snapdebug) { printf("Copyonwrite: snapino %ju lbn %jd for ", (uintmax_t)ip->i_number, (intmax_t)lbn); if (bp->b_vp == devvp) printf("fs metadata"); else printf("inum %ju", (uintmax_t)VTOI(bp->b_vp)->i_number); printf(" lblkno %jd to blkno %jd\n", (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno); } #endif /* * If we have already read the old block contents, then * simply copy them to the new block. Note that we need * to synchronously write snapshots that have not been * unlinked, and hence will be visible after a crash, * to ensure their integrity. At a minimum we ensure the * integrity of the filesystem metadata, but use the * dopersistence sysctl-setable flag to decide on the * persistence needed for file content data. */ if (savedcbp != NULL) { bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize); bawrite(cbp); if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR || dopersistence) && ip->i_effnlink > 0) (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); else launched_async_io = 1; continue; } /* * Otherwise, read the old block contents into the buffer. */ if ((error = readblock(vp, cbp, lbn)) != 0) { bzero(cbp->b_data, fs->fs_bsize); bawrite(cbp); if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR || dopersistence) && ip->i_effnlink > 0) (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); else launched_async_io = 1; break; } savedcbp = cbp; } /* * Note that we need to synchronously write snapshots that * have not been unlinked, and hence will be visible after * a crash, to ensure their integrity. At a minimum we * ensure the integrity of the filesystem metadata, but * use the dopersistence sysctl-setable flag to decide on * the persistence needed for file content data. */ if (savedcbp) { vp = savedcbp->b_vp; bawrite(savedcbp); if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR || dopersistence) && VTOI(vp)->i_effnlink > 0) (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT); else launched_async_io = 1; } lockmgr(vp->v_vnlock, LK_RELEASE, NULL); td->td_pflags = (td->td_pflags & ~TDP_NORUNNINGBUF) | prev_norunningbuf; if (launched_async_io && (td->td_pflags & TDP_NORUNNINGBUF) == 0) waitrunningbufspace(); /* * I/O on bp will now be started, so count it in runningbufspace. */ if (saved_runningbufspace != 0) { bp->b_runningbufspace = saved_runningbufspace; atomic_add_long(&runningbufspace, bp->b_runningbufspace); } return (error); } /* * sync snapshots to force freework records waiting on snapshots to claim * blocks to free. */ void ffs_sync_snap(mp, waitfor) struct mount *mp; int waitfor; { struct snapdata *sn; struct vnode *devvp; struct vnode *vp; struct inode *ip; devvp = VFSTOUFS(mp)->um_devvp; if ((devvp->v_vflag & VV_COPYONWRITE) == 0) return; for (;;) { VI_LOCK(devvp); sn = devvp->v_rdev->si_snapdata; if (sn == NULL) { VI_UNLOCK(devvp); return; } if (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, VI_MTX(devvp)) == 0) break; } TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) { vp = ITOV(ip); ffs_syncvnode(vp, waitfor, NO_INO_UPDT); } lockmgr(&sn->sn_lock, LK_RELEASE, NULL); } /* * Read the specified block into the given buffer. * Much of this boiler-plate comes from bwrite(). */ static int readblock(vp, bp, lbn) struct vnode *vp; struct buf *bp; ufs2_daddr_t lbn; { struct inode *ip; struct bio *bip; struct fs *fs; ip = VTOI(vp); fs = ITOFS(ip); bip = g_alloc_bio(); bip->bio_cmd = BIO_READ; bip->bio_offset = dbtob(fsbtodb(fs, blkstofrags(fs, lbn))); bip->bio_data = bp->b_data; bip->bio_length = bp->b_bcount; bip->bio_done = NULL; g_io_request(bip, ITODEVVP(ip)->v_bufobj.bo_private); bp->b_error = biowait(bip, "snaprdb"); g_destroy_bio(bip); return (bp->b_error); } #endif /* * Process file deletes that were deferred by ufs_inactive() due to * the file system being suspended. Transfer IN_LAZYACCESS into * IN_MODIFIED for vnodes that were accessed during suspension. */ void process_deferred_inactive(struct mount *mp) { struct vnode *vp, *mvp; struct inode *ip; struct thread *td; int error; td = curthread; (void) vn_start_secondary_write(NULL, &mp, V_WAIT); loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { /* * IN_LAZYACCESS is checked here without holding any * vnode lock, but this flag is set only while holding * vnode interlock. */ if (vp->v_type == VNON || ((VTOI(vp)->i_flag & IN_LAZYACCESS) == 0 && ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0))) { VI_UNLOCK(vp); continue; } vholdl(vp); error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); if (error != 0) { vdrop(vp); if (error == ENOENT) continue; /* vnode recycled */ MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } ip = VTOI(vp); if ((ip->i_flag & IN_LAZYACCESS) != 0) { ip->i_flag &= ~IN_LAZYACCESS; ip->i_flag |= IN_MODIFIED; } VI_LOCK(vp); if ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0) { VI_UNLOCK(vp); VOP_UNLOCK(vp, 0); vdrop(vp); continue; } vinactive(vp, td); VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, ("process_deferred_inactive: got VI_OWEINACT")); VI_UNLOCK(vp); VOP_UNLOCK(vp, 0); vdrop(vp); } vn_finished_secondary_write(mp); } #ifndef NO_FFS_SNAPSHOT static struct snapdata * ffs_snapdata_alloc(void) { struct snapdata *sn; /* * Fetch a snapdata from the free list if there is one available. */ mtx_lock(&snapfree_lock); sn = LIST_FIRST(&snapfree); if (sn != NULL) LIST_REMOVE(sn, sn_link); mtx_unlock(&snapfree_lock); if (sn != NULL) return (sn); /* * If there were no free snapdatas allocate one. */ sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO); TAILQ_INIT(&sn->sn_head); lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT, LK_CANRECURSE | LK_NOSHARE); return (sn); } /* * The snapdata is never freed because we can not be certain that * there are no threads sleeping on the snap lock. Persisting * them permanently avoids costly synchronization in ffs_lock(). */ static void ffs_snapdata_free(struct snapdata *sn) { mtx_lock(&snapfree_lock); LIST_INSERT_HEAD(&snapfree, sn, sn_link); mtx_unlock(&snapfree_lock); } /* Try to free snapdata associated with devvp */ static void try_free_snapdata(struct vnode *devvp) { struct snapdata *sn; ufs2_daddr_t *snapblklist; ASSERT_VI_LOCKED(devvp, "try_free_snapdata"); sn = devvp->v_rdev->si_snapdata; if (sn == NULL || TAILQ_FIRST(&sn->sn_head) != NULL || (devvp->v_vflag & VV_COPYONWRITE) == 0) { VI_UNLOCK(devvp); return; } devvp->v_rdev->si_snapdata = NULL; devvp->v_vflag &= ~VV_COPYONWRITE; lockmgr(&sn->sn_lock, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp)); snapblklist = sn->sn_blklist; sn->sn_blklist = NULL; sn->sn_listsize = 0; lockmgr(&sn->sn_lock, LK_RELEASE, NULL); if (snapblklist != NULL) free(snapblklist, M_UFSMNT); ffs_snapdata_free(sn); } static struct snapdata * ffs_snapdata_acquire(struct vnode *devvp) { struct snapdata *nsn, *sn; int error; /* * Allocate a free snapdata. This is done before acquiring the * devvp lock to avoid allocation while the devvp interlock is * held. */ nsn = ffs_snapdata_alloc(); for (;;) { VI_LOCK(devvp); sn = devvp->v_rdev->si_snapdata; if (sn == NULL) { /* * This is the first snapshot on this * filesystem and we use our pre-allocated * snapdata. Publish sn with the sn_lock * owned by us, to avoid the race. */ error = lockmgr(&nsn->sn_lock, LK_EXCLUSIVE | LK_NOWAIT, NULL); if (error != 0) panic("leaked sn, lockmgr error %d", error); sn = devvp->v_rdev->si_snapdata = nsn; VI_UNLOCK(devvp); nsn = NULL; break; } /* * There is a snapshots which already exists on this * filesystem, grab a reference to the common lock. */ error = lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL, VI_MTX(devvp)); if (error == 0) break; } /* * Free any unused snapdata. */ if (nsn != NULL) ffs_snapdata_free(nsn); return (sn); } #endif Index: head/sys/ufs/ufs/ufs_vnops.c =================================================================== --- head/sys/ufs/ufs/ufs_vnops.c (revision 342547) +++ head/sys/ufs/ufs/ufs_vnops.c (revision 342548) @@ -1,2790 +1,2795 @@ /*- * 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 /* 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 #include static vop_accessx_t ufs_accessx; 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_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_markatime_t ufs_markatime; 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; static vop_kqfilter_t ufsfifo_kqfilter; SYSCTL_NODE(_vfs, OID_AUTO, ufs, CTLFLAG_RD, 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)) ip->i_flag |= IN_LAZYMOD; else if (((vp->v_mount->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND)) == 0) || (ip->i_flag & (IN_CHANGE | IN_UPDATE))) ip->i_flag |= IN_MODIFIED; else if (ip->i_flag & IN_ACCESS) ip->i_flag |= 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) { VI_LOCK(vp); ufs_itimes_locked(vp); VI_UNLOCK(vp); } /* * Create a regular file */ static int ufs_create(ap) 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(ap) 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); ip->i_flag |= 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 so it is * checked to see if it is an alias of an existing entry in * the inode cache. XXX I don't believe this is necessary now. */ (*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); /* * Files marked append-only must be opened for appending. */ if ((ip->i_flags & APPEND) && (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) return (EPERM); vnode_create_vobject(vp, DIP(ip, i_size), ap->a_td); return (0); } /* * Close called. * * Update the times on the inode. */ /* ARGSUSED */ static int ufs_close(ap) 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; int usecount; VI_LOCK(vp); usecount = vp->v_usecount; if (usecount > 1) ufs_itimes_locked(vp); VI_UNLOCK(vp); return (0); } static int ufs_accessx(ap) 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, NULL); } 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, NULL); } 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, NULL); } 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, NULL); return (error); } /* ARGSUSED */ static int ufs_getattr(ap) 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(ap) 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); ip->i_flag |= 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 ((ip->i_flags & SF_SNAPSHOT) != 0) 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); } 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 ((ip->i_flags & SF_SNAPSHOT) != 0) return (EPERM); error = vn_utimes_perm(vp, vap, cred, td); if (error != 0) return (error); ip->i_flag |= 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 ((ip->i_flags & SF_SNAPSHOT) != 0 && (vap->va_mode & (S_IXUSR | S_IWUSR | S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH))) 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 */ /* * Mark this file's access time for update for vfs_mark_atime(). This * is called from execve() and mmap(). */ static int ufs_markatime(ap) struct vop_markatime_args /* { struct vnode *a_vp; } */ *ap; { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); VI_LOCK(vp); ip->i_flag |= IN_ACCESS; VI_UNLOCK(vp); /* * 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(vp, mode, cred, td) struct vnode *vp; int mode; struct ucred *cred; struct thread *td; { struct inode *ip = VTOI(vp); int 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); } ip->i_mode &= ~ALLPERMS; ip->i_mode |= (mode & ALLPERMS); DIP_SET(ip, i_mode, ip->i_mode); ip->i_flag |= 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(vp, uid, gid, cred, td) 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); (void) chkiq(ip, -1, cred, CHOWN); 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 */ ip->i_flag |= IN_CHANGE; if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) { ip->i_mode &= ~(ISUID | ISGID); DIP_SET(ip, i_mode, ip->i_mode); } } error = UFS_UPDATE(vp, 0); return (error); } static int ufs_remove(ap) 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)) { error = EPERM; goto out; } #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 ((ip->i_flags & SF_SNAPSHOT) != 0) { /* * 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, 0); (void) VOP_FSYNC(dvp, MNT_WAIT, td); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } out: 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(ap) 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; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ufs_link: no name"); #endif if (VTOI(tdvp)->i_effnlink < 2) { print_bad_link_count("ufs_link", tdvp); error = EINVAL; goto out; } ip = VTOI(vp); if (ip->i_nlink >= UFS_LINK_MAX) { error = EMLINK; goto out; } /* * The file may have been removed after namei droped 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); ip->i_flag |= 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, 0); } if (error) { ip->i_effnlink--; ip->i_nlink--; DIP_SET(ip, i_nlink, ip->i_nlink); ip->i_flag |= IN_CHANGE; if (DOINGSOFTDEP(vp)) softdep_revert_link(VTOI(tdvp), ip); } out: return (error); } /* * whiteout vnode call */ static int ufs_whiteout(ap) 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; switch (ap->a_flags) { case LOOKUP: /* 4.4 format directories support whiteout operations */ if (dvp->v_mount->mnt_maxsymlinklen > 0) return (0); return (EOPNOTSUPP); case CREATE: /* create a new directory whiteout */ #ifdef INVARIANTS if ((cnp->cn_flags & SAVENAME) == 0) panic("ufs_whiteout: missing name"); if (dvp->v_mount->mnt_maxsymlinklen <= 0) 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, 0); break; case DELETE: /* remove an existing directory whiteout */ #ifdef INVARIANTS if (dvp->v_mount->mnt_maxsymlinklen <= 0) 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(ap) 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 = fcnp->cn_thread; struct inode *fip, *tip, *tdp, *fdp; struct direct newdir; off_t endoff; int doingdirectory, newparent; int error = 0; struct mount *mp; ino_t ino; #ifdef INVARIANTS if ((tcnp->cn_flags & HASBUF) == 0 || (fcnp->cn_flags & HASBUF) == 0) panic("ufs_rename: no name"); #endif endoff = 0; mp = tdvp->v_mount; VOP_UNLOCK(tdvp, 0); if (tvp && tvp != tdvp) VOP_UNLOCK(tvp, 0); /* * 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; } 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, 0); error = vn_lock(tdvp, LK_EXCLUSIVE); if (error) goto releout; VOP_UNLOCK(tdvp, 0); 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, 0); VOP_UNLOCK(tdvp, 0); goto releout; } error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp); if (error) { VOP_UNLOCK(fdvp, 0); VOP_UNLOCK(tdvp, 0); if (error != EBUSY) goto releout; error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp); if (error != 0) goto releout; VOP_UNLOCK(nvp, 0); 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, 0); VOP_UNLOCK(tdvp, 0); VOP_UNLOCK(fvp, 0); 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, 0); VOP_UNLOCK(tdvp, 0); VOP_UNLOCK(fvp, 0); 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; } } 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) { error = EMLINK; goto unlockout; } 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, tcnp->cn_thread); if (error) goto unlockout; 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) { VOP_UNLOCK(fdvp, 0); VOP_UNLOCK(fvp, 0); VOP_UNLOCK(tdvp, 0); if (tvp) VOP_UNLOCK(tvp, 0); 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 ((tcnp->cn_flags & SAVESTART) == 0) panic("ufs_rename: lost to startdir"); } 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); /* * 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); fip->i_flag |= 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) { error = EMLINK; goto bad; } } ufs_makedirentry(fip, tcnp, &newdir); error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL, 1); if (error) goto bad; /* Setup tdvp for directory compaction if needed. */ if (tdp->i_count && tdp->i_endoff && tdp->i_endoff < tdp->i_size) endoff = tdp->i_endoff; } 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); } } 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); tdp->i_flag |= IN_CHANGE; } tip->i_nlink--; DIP_SET(tip, i_nlink, tip->i_nlink); tip->i_flag |= 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); tdp->i_flag |= 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); fip->i_offset = 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_purge(fvp); if (tvp) cache_purge(tvp); cache_purge_negative(tdvp); unlockout: vput(fdvp); vput(fvp); if (tvp) vput(tvp); /* * If compaction or fsync was requested do it now that other locks * are no longer needed. */ if (error == 0 && endoff != 0) { error = UFS_TRUNCATE(tdvp, endoff, IO_NORMAL | (DOINGASYNC(tdvp) ? 0 : IO_SYNC), tcnp->cn_cred); if (error != 0) vn_printf(tdvp, "ufs_rename: failed to truncate, error %d\n", error); #ifdef UFS_DIRHASH else if (tdp->i_dirhash != NULL) ufsdirhash_dirtrunc(tdp, endoff); #endif /* * Even if the directory compaction failed, rename was * succesful. Do not propagate a UFS_TRUNCATE() error * to the caller. */ error = 0; } if (error == 0 && tdp->i_flag & IN_NEEDSYNC) error = VOP_FSYNC(tdvp, MNT_WAIT, td); vput(tdvp); return (error); bad: fip->i_effnlink--; fip->i_nlink--; DIP_SET(fip, i_nlink, fip->i_nlink); fip->i_flag |= IN_CHANGE; if (DOINGSOFTDEP(fvp)) softdep_revert_link(tdp, fip); goto unlockout; releout: 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); ip->i_mode = 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. */ ip->i_mode = 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); ip->i_mode = mode; DIP_SET(ip, i_mode, mode); ufs_sync_acl_from_inode(ip, acl); break; } /* FALLTHROUGH */ case EOPNOTSUPP: /* * Just use the mode as-is. */ ip->i_mode = 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(ap) 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; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ufs_mkdir: no name"); #endif dp = VTOI(dvp); if (dp->i_nlink >= UFS_LINK_MAX) { error = EMLINK; 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; } error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp); if (error) goto out; 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); 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); vput(tvp); return (error); } #endif #endif /* !SUIDDIR */ ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = 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); dp->i_flag |= 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, cnp->cn_thread); 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, cnp->cn_thread); if (error) goto bad; } #endif /* !UFS_ACL */ /* * Initialize directory with "." and ".." from static template. */ if (dvp->v_mount->mnt_maxsymlinklen > 0) 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); ip->i_flag |= 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, 0); bad: if (error == 0) { *ap->a_vpp = tvp; } else { dp->i_effnlink--; dp->i_nlink--; DIP_SET(dp, i_nlink, dp->i_nlink); dp->i_flag |= 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); ip->i_flag |= IN_CHANGE; if (DOINGSOFTDEP(tvp)) softdep_revert_mkdir(dp, ip); vput(tvp); } out: return (error); } /* * Rmdir system call. */ static int ufs_rmdir(ap) 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; } #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; } cache_purge(dvp); /* * 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); dp->i_flag |= IN_CHANGE; error = UFS_UPDATE(dvp, 0); ip->i_nlink--; DIP_SET(ip, i_nlink, ip->i_nlink); ip->i_flag |= IN_CHANGE; } cache_purge(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(ap) 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 < vp->v_mount->mnt_maxsymlinklen) { ip = VTOI(vp); bcopy(ap->a_target, SHORTLINK(ip), len); ip->i_size = len; DIP_SET(ip, i_size, len); ip->i_flag |= 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(ap) struct vop_readdir_args /* { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; int *a_eofflag; int *a_ncookies; u_long **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; u_long *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 = ffs_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 (vp->v_mount->mnt_maxsymlinklen <= 0) { 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(ap) 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 < vp->v_mount->mnt_maxsymlinklen) || DIP(ip, i_blocks) == 0) { /* XXX - for old fastlink support */ return (uiomove(SHORTLINK(ip), 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(ap) 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(ap) struct vop_print_args /* { struct vnode *a_vp; } */ *ap; { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); 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(ap) 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; int usecount; VI_LOCK(vp); usecount = vp->v_usecount; if (usecount > 1) ufs_itimes_locked(vp); VI_UNLOCK(vp); return (fifo_specops.vop_close(ap)); } /* * Kqfilter wrapper for fifos. * * Fall through to ufs kqfilter routines if needed */ static int ufsfifo_kqfilter(ap) struct vop_kqfilter_args *ap; { int error; error = fifo_specops.vop_kqfilter(ap); if (error) error = vfs_kqfilter(ap); return (error); } /* * Return POSIX pathconf information applicable to ufs filesystems. */ static int ufs_pathconf(ap) 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; case _PC_ACL_EXTENDED: #ifdef UFS_ACL if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS) *ap->a_retval = 1; else *ap->a_retval = 0; #else *ap->a_retval = 0; #endif break; case _PC_ACL_NFS4: #ifdef UFS_ACL if (ap->a_vp->v_mount->mnt_flag & MNT_NFS4ACLS) *ap->a_retval = 1; else *ap->a_retval = 0; #else *ap->a_retval = 0; #endif break; 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; case _PC_MAC_PRESENT: #ifdef MAC if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL) *ap->a_retval = 1; else *ap->a_retval = 0; #else *ap->a_retval = 0; #endif break; 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(mntp, fifoops, vpp) struct mount *mntp; struct vop_vector *fifoops; struct vnode **vpp; { struct inode *ip; struct vnode *vp; vp = *vpp; 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; ASSERT_VOP_LOCKED(vp, "ufs_vinit"); 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(mode, dvp, vpp, cnp, callfunc) 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); #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("%s: no name", callfunc); #endif *vpp = NULL; if ((mode & IFMT) == 0) mode |= IFREG; if (pdir->i_effnlink < 2) { print_bad_link_count(callfunc, dvp); return (EINVAL); } 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); 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); vput(tvp); return (error); } #endif #endif /* !SUIDDIR */ ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = 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)) { 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, cnp->cn_thread); 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, cnp->cn_thread); if (error) goto bad; } #endif /* !UFS_ACL */ ufs_makedirentry(ip, cnp, &newdir); error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL, 0); if (error) goto bad; *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); ip->i_flag |= IN_CHANGE; if (DOINGSOFTDEP(tvp)) softdep_revert_create(VTOI(dvp), ip); vput(tvp); return (error); } static int ufs_ioctl(struct vop_ioctl_args *ap) { switch (ap->a_command) { case FIOSEEKDATA: case FIOSEEKHOLE: return (vn_bmap_seekhole(ap->a_vp, ap->a_command, (off_t *)ap->a_data, ap->a_cred)); default: return (ENOTTY); } } /* 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_cachedlookup = ufs_lookup, .vop_close = ufs_close, .vop_create = ufs_create, .vop_getattr = ufs_getattr, .vop_inactive = ufs_inactive, .vop_ioctl = ufs_ioctl, .vop_link = ufs_link, .vop_lookup = vfs_cache_lookup, .vop_markatime = ufs_markatime, .vop_mkdir = ufs_mkdir, .vop_mknod = ufs_mknod, .vop_open = ufs_open, .vop_pathconf = ufs_pathconf, .vop_poll = vop_stdpoll, .vop_print = ufs_print, .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 }; 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_kqfilter = ufsfifo_kqfilter, .vop_markatime = ufs_markatime, .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 };