/* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2007 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */ /* All Rights Reserved */ /* * University Copyright- Copyright (c) 1982, 1986, 1988 * The Regents of the University of California * All Rights Reserved * * University Acknowledgment- Portions of this document are derived from * software developed by the University of California, Berkeley, and its * contributors. */ #pragma ident "%Z%%M% %I% %E% SMI" #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 #undef NFS #include #include #include #include #include #include #include "fs/fs_subr.h" #include #include #include #include #include #include #include /* * This is the loadable module wrapper. */ #include int ufsfstype; vfsops_t *ufs_vfsops; static int ufsinit(int, char *); static int mountfs(); extern int highbit(); extern struct instats ins; extern struct vnode *common_specvp(struct vnode *vp); extern vfs_t EIO_vfs; struct dquot *dquot, *dquotNDQUOT; /* * Cylinder group summary information handling tunable. * This defines when these deltas get logged. * If the number of cylinders in the file system is over the * tunable then we log csum updates. Otherwise the updates are only * done for performance on unmount. After a panic they can be * quickly constructed during mounting. See ufs_construct_si() * called from ufs_getsummaryinfo(). * * This performance feature can of course be disabled by setting * ufs_ncg_log to 0, and fully enabled by setting it to 0xffffffff. */ #define UFS_LOG_NCG_DEFAULT 10000 uint32_t ufs_ncg_log = UFS_LOG_NCG_DEFAULT; /* * ufs_clean_root indicates whether the root fs went down cleanly */ static int ufs_clean_root = 0; /* * UFS Mount options table */ static char *intr_cancel[] = { MNTOPT_NOINTR, NULL }; static char *nointr_cancel[] = { MNTOPT_INTR, NULL }; static char *forcedirectio_cancel[] = { MNTOPT_NOFORCEDIRECTIO, NULL }; static char *noforcedirectio_cancel[] = { MNTOPT_FORCEDIRECTIO, NULL }; static char *largefiles_cancel[] = { MNTOPT_NOLARGEFILES, NULL }; static char *nolargefiles_cancel[] = { MNTOPT_LARGEFILES, NULL }; static char *logging_cancel[] = { MNTOPT_NOLOGGING, NULL }; static char *nologging_cancel[] = { MNTOPT_LOGGING, NULL }; static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL }; static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL }; static char *quota_cancel[] = { MNTOPT_NOQUOTA, NULL }; static char *noquota_cancel[] = { MNTOPT_QUOTA, NULL }; static char *dfratime_cancel[] = { MNTOPT_NODFRATIME, NULL }; static char *nodfratime_cancel[] = { MNTOPT_DFRATIME, NULL }; static mntopt_t mntopts[] = { /* * option name cancel option default arg flags * ufs arg flag */ { MNTOPT_INTR, intr_cancel, NULL, MO_DEFAULT, (void *)0 }, { MNTOPT_NOINTR, nointr_cancel, NULL, 0, (void *)UFSMNT_NOINTR }, { MNTOPT_SYNCDIR, NULL, NULL, 0, (void *)UFSMNT_SYNCDIR }, { MNTOPT_FORCEDIRECTIO, forcedirectio_cancel, NULL, 0, (void *)UFSMNT_FORCEDIRECTIO }, { MNTOPT_NOFORCEDIRECTIO, noforcedirectio_cancel, NULL, 0, (void *)UFSMNT_NOFORCEDIRECTIO }, { MNTOPT_NOSETSEC, NULL, NULL, 0, (void *)UFSMNT_NOSETSEC }, { MNTOPT_LARGEFILES, largefiles_cancel, NULL, MO_DEFAULT, (void *)UFSMNT_LARGEFILES }, { MNTOPT_NOLARGEFILES, nolargefiles_cancel, NULL, 0, (void *)0 }, { MNTOPT_LOGGING, logging_cancel, NULL, MO_TAG, (void *)UFSMNT_LOGGING }, { MNTOPT_NOLOGGING, nologging_cancel, NULL, MO_NODISPLAY|MO_DEFAULT|MO_TAG, (void *)0 }, { MNTOPT_QUOTA, quota_cancel, NULL, MO_IGNORE, (void *)0 }, { MNTOPT_NOQUOTA, noquota_cancel, NULL, MO_NODISPLAY|MO_DEFAULT, (void *)0 }, { MNTOPT_GLOBAL, NULL, NULL, 0, (void *)0 }, { MNTOPT_XATTR, xattr_cancel, NULL, MO_DEFAULT, (void *)0 }, { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, (void *)0 }, { MNTOPT_NOATIME, NULL, NULL, 0, (void *)UFSMNT_NOATIME }, { MNTOPT_DFRATIME, dfratime_cancel, NULL, 0, (void *)0 }, { MNTOPT_NODFRATIME, nodfratime_cancel, NULL, MO_NODISPLAY|MO_DEFAULT, (void *)UFSMNT_NODFRATIME }, { MNTOPT_ONERROR, NULL, UFSMNT_ONERROR_PANIC_STR, MO_DEFAULT|MO_HASVALUE, (void *)0 }, }; static mntopts_t ufs_mntopts = { sizeof (mntopts) / sizeof (mntopt_t), mntopts }; static vfsdef_t vfw = { VFSDEF_VERSION, "ufs", ufsinit, VSW_HASPROTO|VSW_CANREMOUNT|VSW_STATS, &ufs_mntopts }; /* * Module linkage information for the kernel. */ extern struct mod_ops mod_fsops; static struct modlfs modlfs = { &mod_fsops, "filesystem for ufs", &vfw }; static struct modlinkage modlinkage = { MODREV_1, (void *)&modlfs, NULL }; /* * An attempt has been made to make this module unloadable. In order to * test it, we need a system in which the root fs is NOT ufs. THIS HAS NOT * BEEN DONE */ extern kstat_t *ufs_inode_kstat; extern uint_t ufs_lockfs_key; extern void ufs_lockfs_tsd_destructor(void *); extern uint_t bypass_snapshot_throttle_key; int _init(void) { /* * Create an index into the per thread array so that any thread doing * VOP will have a lockfs mark on it. */ tsd_create(&ufs_lockfs_key, ufs_lockfs_tsd_destructor); tsd_create(&bypass_snapshot_throttle_key, NULL); return (mod_install(&modlinkage)); } int _fini(void) { return (EBUSY); } int _info(struct modinfo *modinfop) { return (mod_info(&modlinkage, modinfop)); } extern struct vnode *makespecvp(dev_t dev, vtype_t type); extern kmutex_t ufs_scan_lock; static int mountfs(struct vfs *, enum whymountroot, struct vnode *, char *, struct cred *, int, void *, int); static int ufs_mount(struct vfs *vfsp, struct vnode *mvp, struct mounta *uap, struct cred *cr) { char *data = uap->dataptr; int datalen = uap->datalen; dev_t dev; struct vnode *bvp; struct pathname dpn; int error; enum whymountroot why = ROOT_INIT; struct ufs_args args; int oflag, aflag; int fromspace = (uap->flags & MS_SYSSPACE) ? UIO_SYSSPACE : UIO_USERSPACE; if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0) return (error); if (mvp->v_type != VDIR) return (ENOTDIR); mutex_enter(&mvp->v_lock); if ((uap->flags & MS_REMOUNT) == 0 && (uap->flags & MS_OVERLAY) == 0 && (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { mutex_exit(&mvp->v_lock); return (EBUSY); } mutex_exit(&mvp->v_lock); /* * Get arguments */ bzero(&args, sizeof (args)); if ((uap->flags & MS_DATA) && data != NULL && datalen != 0) { int copy_result = 0; if (datalen > sizeof (args)) return (EINVAL); if (uap->flags & MS_SYSSPACE) bcopy(data, &args, datalen); else copy_result = copyin(data, &args, datalen); if (copy_result) return (EFAULT); datalen = sizeof (struct ufs_args); } else { datalen = 0; } /* * Read in the mount point pathname * (so we can record the directory the file system was last mounted on). */ if (error = pn_get(uap->dir, fromspace, &dpn)) return (error); /* * Resolve path name of special file being mounted. */ if (error = lookupname(uap->spec, fromspace, FOLLOW, NULL, &bvp)) { pn_free(&dpn); return (error); } if (bvp->v_type != VBLK) { VN_RELE(bvp); pn_free(&dpn); return (ENOTBLK); } dev = bvp->v_rdev; if (getmajor(dev) >= devcnt) { pn_free(&dpn); VN_RELE(bvp); return (ENXIO); } if (uap->flags & MS_REMOUNT) why = ROOT_REMOUNT; /* * In SunCluster, requests to a global device are satisfied by * a local device. We substitute the global pxfs node with a * local spec node here. */ if (IS_PXFSVP(bvp)) { VN_RELE(bvp); bvp = makespecvp(dev, VBLK); } /* * Open block device mounted on. We need this to * check whether the caller has sufficient rights to * access the device in question. * When bio is fixed for vnodes this can all be vnode * operations. */ if ((vfsp->vfs_flag & VFS_RDONLY) != 0 || (uap->flags & MS_RDONLY) != 0) { oflag = FREAD; aflag = VREAD; } else { oflag = FREAD | FWRITE; aflag = VREAD | VWRITE; } if ((error = VOP_ACCESS(bvp, aflag, 0, cr, NULL)) != 0 || (error = secpolicy_spec_open(cr, bvp, oflag)) != 0) { pn_free(&dpn); VN_RELE(bvp); return (error); } /* * Ensure that this device isn't already mounted or in progress on a * mount unless this is a REMOUNT request or we are told to suppress * mount checks. Global mounts require special handling. */ if ((uap->flags & MS_NOCHECK) == 0) { if ((uap->flags & MS_GLOBAL) == 0 && vfs_devmounting(dev, vfsp)) { pn_free(&dpn); VN_RELE(bvp); return (EBUSY); } if (vfs_devismounted(dev)) { if ((uap->flags & MS_REMOUNT) == 0) { pn_free(&dpn); VN_RELE(bvp); return (EBUSY); } } } /* * If the device is a tape, mount it read only */ if (devopsp[getmajor(dev)]->devo_cb_ops->cb_flag & D_TAPE) { vfsp->vfs_flag |= VFS_RDONLY; vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); } if (uap->flags & MS_RDONLY) vfsp->vfs_flag |= VFS_RDONLY; /* * Mount the filesystem, free the device vnode on error. */ error = mountfs(vfsp, why, bvp, dpn.pn_path, cr, 0, &args, datalen); pn_free(&dpn); if (error) { VN_RELE(bvp); } if (error == 0) vfs_set_feature(vfsp, VFSFT_XVATTR); return (error); } /* * Mount root file system. * "why" is ROOT_INIT on initial call ROOT_REMOUNT if called to * remount the root file system, and ROOT_UNMOUNT if called to * unmount the root (e.g., as part of a system shutdown). * * XXX - this may be partially machine-dependent; it, along with the VFS_SWAPVP * operation, goes along with auto-configuration. A mechanism should be * provided by which machine-INdependent code in the kernel can say "get me the * right root file system" and "get me the right initial swap area", and have * that done in what may well be a machine-dependent fashion. * Unfortunately, it is also file-system-type dependent (NFS gets it via * bootparams calls, UFS gets it from various and sundry machine-dependent * mechanisms, as SPECFS does for swap). */ static int ufs_mountroot(struct vfs *vfsp, enum whymountroot why) { struct fs *fsp; int error; static int ufsrootdone = 0; dev_t rootdev; struct vnode *vp; struct vnode *devvp = 0; int ovflags; int doclkset; ufsvfs_t *ufsvfsp; if (why == ROOT_INIT) { if (ufsrootdone++) return (EBUSY); rootdev = getrootdev(); if (rootdev == (dev_t)NODEV) return (ENODEV); vfsp->vfs_dev = rootdev; vfsp->vfs_flag |= VFS_RDONLY; } else if (why == ROOT_REMOUNT) { vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp; (void) dnlc_purge_vfsp(vfsp, 0); vp = common_specvp(vp); (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_INVAL, CRED(), NULL); (void) bfinval(vfsp->vfs_dev, 0); fsp = getfs(vfsp); ovflags = vfsp->vfs_flag; vfsp->vfs_flag &= ~VFS_RDONLY; vfsp->vfs_flag |= VFS_REMOUNT; rootdev = vfsp->vfs_dev; } else if (why == ROOT_UNMOUNT) { if (vfs_lock(vfsp) == 0) { (void) ufs_flush(vfsp); /* * Mark the log as fully rolled */ ufsvfsp = (ufsvfs_t *)vfsp->vfs_data; fsp = ufsvfsp->vfs_fs; if (TRANS_ISTRANS(ufsvfsp) && !TRANS_ISERROR(ufsvfsp) && (fsp->fs_rolled == FS_NEED_ROLL)) { ml_unit_t *ul = ufsvfsp->vfs_log; error = ufs_putsummaryinfo(ul->un_dev, ufsvfsp, fsp); if (error == 0) { fsp->fs_rolled = FS_ALL_ROLLED; UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp); } } vfs_unlock(vfsp); } else { ufs_update(0); } vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp; (void) VOP_CLOSE(vp, FREAD|FWRITE, 1, (offset_t)0, CRED(), NULL); return (0); } error = vfs_lock(vfsp); if (error) return (error); devvp = makespecvp(rootdev, VBLK); /* If RO media, don't call clkset() (see below) */ doclkset = 1; if (why == ROOT_INIT) { error = VOP_OPEN(&devvp, FREAD|FWRITE, CRED(), NULL); if (error == 0) { (void) VOP_CLOSE(devvp, FREAD|FWRITE, 1, (offset_t)0, CRED(), NULL); } else { doclkset = 0; } } error = mountfs(vfsp, why, devvp, "/", CRED(), 1, NULL, 0); /* * XXX - assumes root device is not indirect, because we don't set * rootvp. Is rootvp used for anything? If so, make another arg * to mountfs. */ if (error) { vfs_unlock(vfsp); if (why == ROOT_REMOUNT) vfsp->vfs_flag = ovflags; if (rootvp) { VN_RELE(rootvp); rootvp = (struct vnode *)0; } VN_RELE(devvp); return (error); } if (why == ROOT_INIT) vfs_add((struct vnode *)0, vfsp, (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); vfs_unlock(vfsp); fsp = getfs(vfsp); clkset(doclkset ? fsp->fs_time : -1); ufsvfsp = (ufsvfs_t *)vfsp->vfs_data; if (ufsvfsp->vfs_log) { vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0); } return (0); } static int remountfs(struct vfs *vfsp, dev_t dev, void *raw_argsp, int args_len) { struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; struct ulockfs *ulp = &ufsvfsp->vfs_ulockfs; struct buf *bp = ufsvfsp->vfs_bufp; struct fs *fsp = (struct fs *)bp->b_un.b_addr; struct fs *fspt; struct buf *tpt = 0; int error = 0; int flags = 0; if (args_len == sizeof (struct ufs_args) && raw_argsp) flags = ((struct ufs_args *)raw_argsp)->flags; /* cannot remount to RDONLY */ if (vfsp->vfs_flag & VFS_RDONLY) return (ENOTSUP); /* whoops, wrong dev */ if (vfsp->vfs_dev != dev) return (EINVAL); /* * synchronize w/ufs ioctls */ mutex_enter(&ulp->ul_lock); atomic_add_long(&ufs_quiesce_pend, 1); /* * reset options */ ufsvfsp->vfs_nointr = flags & UFSMNT_NOINTR; ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR; ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC; ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME; if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime) ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME; else /* dfratime, default behavior */ ufsvfsp->vfs_dfritime |= UFS_DFRATIME; if (flags & UFSMNT_FORCEDIRECTIO) ufsvfsp->vfs_forcedirectio = 1; else /* default is no direct I/O */ ufsvfsp->vfs_forcedirectio = 0; ufsvfsp->vfs_iotstamp = lbolt; /* * set largefiles flag in ufsvfs equal to the * value passed in by the mount command. If * it is "nolargefiles", and the flag is set * in the superblock, the mount fails. */ if (!(flags & UFSMNT_LARGEFILES)) { /* "nolargefiles" */ if (fsp->fs_flags & FSLARGEFILES) { error = EFBIG; goto remounterr; } ufsvfsp->vfs_lfflags &= ~UFS_LARGEFILES; } else /* "largefiles" */ ufsvfsp->vfs_lfflags |= UFS_LARGEFILES; /* * read/write to read/write; all done */ if (fsp->fs_ronly == 0) goto remounterr; /* * fix-on-panic assumes RO->RW remount implies system-critical fs * if it is shortly after boot; so, don't attempt to lock and fix * (unless the user explicitly asked for another action on error) * XXX UFSMNT_ONERROR_RDONLY rather than UFSMNT_ONERROR_PANIC */ #define BOOT_TIME_LIMIT (180*hz) if (!(flags & UFSMNT_ONERROR_FLGMASK) && lbolt < BOOT_TIME_LIMIT) { cmn_err(CE_WARN, "%s is required to be mounted onerror=%s", ufsvfsp->vfs_fs->fs_fsmnt, UFSMNT_ONERROR_PANIC_STR); flags |= UFSMNT_ONERROR_PANIC; } if ((error = ufsfx_mount(ufsvfsp, flags)) != 0) goto remounterr; /* * quiesce the file system */ error = ufs_quiesce(ulp); if (error) goto remounterr; tpt = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, SBLOCK, SBSIZE); if (tpt->b_flags & B_ERROR) { error = EIO; goto remounterr; } fspt = (struct fs *)tpt->b_un.b_addr; if (((fspt->fs_magic != FS_MAGIC) && (fspt->fs_magic != MTB_UFS_MAGIC)) || (fspt->fs_magic == FS_MAGIC && (fspt->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 && fspt->fs_version != UFS_VERSION_MIN)) || (fspt->fs_magic == MTB_UFS_MAGIC && (fspt->fs_version > MTB_UFS_VERSION_1 || fspt->fs_version < MTB_UFS_VERSION_MIN)) || fspt->fs_bsize > MAXBSIZE || fspt->fs_frag > MAXFRAG || fspt->fs_bsize < sizeof (struct fs) || fspt->fs_bsize < PAGESIZE) { tpt->b_flags |= B_STALE | B_AGE; error = EINVAL; goto remounterr; } if (ufsvfsp->vfs_log && (ufsvfsp->vfs_log->un_flags & LDL_NOROLL)) { ufsvfsp->vfs_log->un_flags &= ~LDL_NOROLL; logmap_start_roll(ufsvfsp->vfs_log); } if (TRANS_ISERROR(ufsvfsp)) goto remounterr; TRANS_DOMATAMAP(ufsvfsp); if ((fspt->fs_state + fspt->fs_time == FSOKAY) && fspt->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp)) { ufsvfsp->vfs_log = NULL; ufsvfsp->vfs_domatamap = 0; error = ENOSPC; goto remounterr; } if (fspt->fs_state + fspt->fs_time == FSOKAY && (fspt->fs_clean == FSCLEAN || fspt->fs_clean == FSSTABLE || fspt->fs_clean == FSLOG)) { /* * Ensure that ufs_getsummaryinfo doesn't reconstruct * the summary info. */ error = ufs_getsummaryinfo(vfsp->vfs_dev, ufsvfsp, fspt); if (error) goto remounterr; /* preserve mount name */ (void) strncpy(fspt->fs_fsmnt, fsp->fs_fsmnt, MAXMNTLEN); /* free the old cg space */ kmem_free(fsp->fs_u.fs_csp, fsp->fs_cssize); /* switch in the new superblock */ fspt->fs_rolled = FS_NEED_ROLL; bcopy(tpt->b_un.b_addr, bp->b_un.b_addr, fspt->fs_sbsize); fsp->fs_clean = FSSTABLE; } /* superblock updated in memory */ tpt->b_flags |= B_STALE | B_AGE; brelse(tpt); tpt = 0; if (fsp->fs_clean != FSSTABLE) { error = ENOSPC; goto remounterr; } if (TRANS_ISTRANS(ufsvfsp)) { fsp->fs_clean = FSLOG; ufsvfsp->vfs_dio = 0; } else if (ufsvfsp->vfs_dio) fsp->fs_clean = FSSUSPEND; TRANS_MATA_MOUNT(ufsvfsp); fsp->fs_fmod = 0; fsp->fs_ronly = 0; atomic_add_long(&ufs_quiesce_pend, -1); cv_broadcast(&ulp->ul_cv); mutex_exit(&ulp->ul_lock); if (TRANS_ISTRANS(ufsvfsp)) { /* * start the delete thread */ ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp); /* * start the reclaim thread */ if (fsp->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING)) { fsp->fs_reclaim &= ~FS_RECLAIM; fsp->fs_reclaim |= FS_RECLAIMING; ufs_thread_start(&ufsvfsp->vfs_reclaim, ufs_thread_reclaim, vfsp); } } TRANS_SBWRITE(ufsvfsp, TOP_MOUNT); return (0); remounterr: if (tpt) brelse(tpt); atomic_add_long(&ufs_quiesce_pend, -1); cv_broadcast(&ulp->ul_cv); mutex_exit(&ulp->ul_lock); return (error); } /* * If the device maxtransfer size is not available, we use ufs_maxmaxphys * along with the system value for maxphys to determine the value for * maxtransfer. */ int ufs_maxmaxphys = (1024 * 1024); #include /* for delay(9f) */ int ufs_mount_error_delay = 20; /* default to 20ms */ int ufs_mount_timeout = 60000; /* default to 1 minute */ static int mountfs(struct vfs *vfsp, enum whymountroot why, struct vnode *devvp, char *path, cred_t *cr, int isroot, void *raw_argsp, int args_len) { dev_t dev = devvp->v_rdev; struct fs *fsp; struct ufsvfs *ufsvfsp = 0; struct buf *bp = 0; struct buf *tp = 0; struct dk_cinfo ci; int error = 0; size_t len; int needclose = 0; int needtrans = 0; struct inode *rip; struct vnode *rvp = NULL; int flags = 0; kmutex_t *ihm; int elapsed; int status; extern int maxphys; if (args_len == sizeof (struct ufs_args) && raw_argsp) flags = ((struct ufs_args *)raw_argsp)->flags; ASSERT(vfs_lock_held(vfsp)); if (why == ROOT_INIT) { /* * Open block device mounted on. * When bio is fixed for vnodes this can all be vnode * operations. */ error = VOP_OPEN(&devvp, (vfsp->vfs_flag & VFS_RDONLY) ? FREAD : FREAD|FWRITE, cr, NULL); if (error) goto out; needclose = 1; /* * Refuse to go any further if this * device is being used for swapping. */ if (IS_SWAPVP(devvp)) { error = EBUSY; goto out; } } /* * check for dev already mounted on */ if (vfsp->vfs_flag & VFS_REMOUNT) { error = remountfs(vfsp, dev, raw_argsp, args_len); if (error == 0) VN_RELE(devvp); return (error); } ASSERT(devvp != 0); /* * Flush back any dirty pages on the block device to * try and keep the buffer cache in sync with the page * cache if someone is trying to use block devices when * they really should be using the raw device. */ (void) VOP_PUTPAGE(common_specvp(devvp), (offset_t)0, (size_t)0, B_INVAL, cr, NULL); /* * read in superblock */ ufsvfsp = kmem_zalloc(sizeof (struct ufsvfs), KM_SLEEP); tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE); if (tp->b_flags & B_ERROR) goto out; fsp = (struct fs *)tp->b_un.b_addr; if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC)) { cmn_err(CE_NOTE, "mount: not a UFS magic number (0x%x)", fsp->fs_magic); error = EINVAL; goto out; } if ((fsp->fs_magic == FS_MAGIC) && (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 && fsp->fs_version != UFS_VERSION_MIN)) { cmn_err(CE_NOTE, "mount: unrecognized version of UFS on-disk format: %d", fsp->fs_version); error = EINVAL; goto out; } if ((fsp->fs_magic == MTB_UFS_MAGIC) && (fsp->fs_version > MTB_UFS_VERSION_1 || fsp->fs_version < MTB_UFS_VERSION_MIN)) { cmn_err(CE_NOTE, "mount: unrecognized version of UFS on-disk format: %d", fsp->fs_version); error = EINVAL; goto out; } #ifndef _LP64 if (fsp->fs_magic == MTB_UFS_MAGIC) { /* * Find the size of the device in sectors. If the * the size in sectors is greater than INT_MAX, it's * a multi-terabyte file system, which can't be * mounted by a 32-bit kernel. We can't use the * fsbtodb() macro in the next line because the macro * casts the intermediate values to daddr_t, which is * a 32-bit quantity in a 32-bit kernel. Here we * really do need the intermediate values to be held * in 64-bit quantities because we're checking for * overflow of a 32-bit field. */ if ((((diskaddr_t)(fsp->fs_size)) << fsp->fs_fsbtodb) > INT_MAX) { cmn_err(CE_NOTE, "mount: multi-terabyte UFS cannot be" " mounted by a 32-bit kernel"); error = EINVAL; goto out; } } #endif if (fsp->fs_bsize > MAXBSIZE || fsp->fs_frag > MAXFRAG || fsp->fs_bsize < sizeof (struct fs) || fsp->fs_bsize < PAGESIZE) { error = EINVAL; /* also needs translation */ goto out; } /* * Allocate VFS private data. */ vfsp->vfs_bcount = 0; vfsp->vfs_data = (caddr_t)ufsvfsp; vfsp->vfs_fstype = ufsfstype; vfsp->vfs_dev = dev; vfsp->vfs_flag |= VFS_NOTRUNC; vfs_make_fsid(&vfsp->vfs_fsid, dev, ufsfstype); ufsvfsp->vfs_devvp = devvp; /* * Cross-link with vfs and add to instance list. */ ufsvfsp->vfs_vfs = vfsp; ufs_vfs_add(ufsvfsp); ufsvfsp->vfs_dev = dev; ufsvfsp->vfs_bufp = tp; ufsvfsp->vfs_dirsize = INODESIZE + (4 * ALLOCSIZE) + fsp->fs_fsize; ufsvfsp->vfs_minfrags = (int)((int64_t)fsp->fs_dsize * fsp->fs_minfree / 100); /* * if mount allows largefiles, indicate so in ufsvfs */ if (flags & UFSMNT_LARGEFILES) ufsvfsp->vfs_lfflags |= UFS_LARGEFILES; /* * Initialize threads */ ufs_delete_init(ufsvfsp, 1); ufs_thread_init(&ufsvfsp->vfs_reclaim, 0); /* * Chicken and egg problem. The superblock may have deltas * in the log. So after the log is scanned we reread the * superblock. We guarantee that the fields needed to * scan the log will not be in the log. */ if (fsp->fs_logbno && fsp->fs_clean == FSLOG && (fsp->fs_state + fsp->fs_time == FSOKAY)) { error = lufs_snarf(ufsvfsp, fsp, (vfsp->vfs_flag & VFS_RDONLY)); if (error) { /* * Allow a ro mount to continue even if the * log cannot be processed - yet. */ if (!(vfsp->vfs_flag & VFS_RDONLY)) { cmn_err(CE_WARN, "Error accessing ufs " "log for %s; Please run fsck(1M)", path); goto out; } } tp->b_flags |= (B_AGE | B_STALE); brelse(tp); tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE); fsp = (struct fs *)tp->b_un.b_addr; ufsvfsp->vfs_bufp = tp; if (tp->b_flags & B_ERROR) goto out; } /* * Set logging mounted flag used by lockfs */ ufsvfsp->vfs_validfs = UT_MOUNTED; /* * Copy the super block into a buffer in its native size. * Use ngeteblk to allocate the buffer */ bp = ngeteblk(fsp->fs_bsize); ufsvfsp->vfs_bufp = bp; bp->b_edev = dev; bp->b_dev = cmpdev(dev); bp->b_blkno = SBLOCK; bp->b_bcount = fsp->fs_sbsize; bcopy(tp->b_un.b_addr, bp->b_un.b_addr, fsp->fs_sbsize); tp->b_flags |= B_STALE | B_AGE; brelse(tp); tp = 0; fsp = (struct fs *)bp->b_un.b_addr; /* * Mount fails if superblock flag indicates presence of large * files and filesystem is attempted to be mounted 'nolargefiles'. * The exception is for a read only mount of root, which we * always want to succeed, so fsck can fix potential problems. * The assumption is that we will remount root at some point, * and the remount will enforce the mount option. */ if (!(isroot & (vfsp->vfs_flag & VFS_RDONLY)) && (fsp->fs_flags & FSLARGEFILES) && !(flags & UFSMNT_LARGEFILES)) { error = EFBIG; goto out; } if (vfsp->vfs_flag & VFS_RDONLY) { fsp->fs_ronly = 1; fsp->fs_fmod = 0; if (((fsp->fs_state + fsp->fs_time) == FSOKAY) && ((fsp->fs_clean == FSCLEAN) || (fsp->fs_clean == FSSTABLE) || (fsp->fs_clean == FSLOG))) { if (isroot) { if (fsp->fs_clean == FSLOG) { if (fsp->fs_rolled == FS_ALL_ROLLED) { ufs_clean_root = 1; } } else { ufs_clean_root = 1; } } fsp->fs_clean = FSSTABLE; } else { fsp->fs_clean = FSBAD; } } else { fsp->fs_fmod = 0; fsp->fs_ronly = 0; TRANS_DOMATAMAP(ufsvfsp); if ((TRANS_ISERROR(ufsvfsp)) || (((fsp->fs_state + fsp->fs_time) == FSOKAY) && fsp->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp))) { ufsvfsp->vfs_log = NULL; ufsvfsp->vfs_domatamap = 0; error = ENOSPC; goto out; } if (((fsp->fs_state + fsp->fs_time) == FSOKAY) && (fsp->fs_clean == FSCLEAN || fsp->fs_clean == FSSTABLE || fsp->fs_clean == FSLOG)) fsp->fs_clean = FSSTABLE; else { if (isroot) { /* * allow root partition to be mounted even * when fs_state is not ok * will be fixed later by a remount root */ fsp->fs_clean = FSBAD; ufsvfsp->vfs_log = NULL; ufsvfsp->vfs_domatamap = 0; } else { error = ENOSPC; goto out; } } if (fsp->fs_clean == FSSTABLE && TRANS_ISTRANS(ufsvfsp)) fsp->fs_clean = FSLOG; } TRANS_MATA_MOUNT(ufsvfsp); needtrans = 1; vfsp->vfs_bsize = fsp->fs_bsize; /* * Read in summary info */ if (error = ufs_getsummaryinfo(dev, ufsvfsp, fsp)) goto out; /* * lastwhinetime is set to zero rather than lbolt, so that after * mounting if the filesystem is found to be full, then immediately the * "file system message" will be logged. */ ufsvfsp->vfs_lastwhinetime = 0L; mutex_init(&ufsvfsp->vfs_lock, NULL, MUTEX_DEFAULT, NULL); (void) copystr(path, fsp->fs_fsmnt, sizeof (fsp->fs_fsmnt) - 1, &len); bzero(fsp->fs_fsmnt + len, sizeof (fsp->fs_fsmnt) - len); /* * Sanity checks for old file systems */ if (fsp->fs_postblformat == FS_42POSTBLFMT) ufsvfsp->vfs_nrpos = 8; else ufsvfsp->vfs_nrpos = fsp->fs_nrpos; /* * Initialize lockfs structure to support file system locking */ bzero(&ufsvfsp->vfs_ulockfs.ul_lockfs, sizeof (struct lockfs)); ufsvfsp->vfs_ulockfs.ul_fs_lock = ULOCKFS_ULOCK; mutex_init(&ufsvfsp->vfs_ulockfs.ul_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&ufsvfsp->vfs_ulockfs.ul_cv, NULL, CV_DEFAULT, NULL); /* * We don't need to grab vfs_dqrwlock for this ufs_iget() call. * We are in the process of mounting the file system so there * is no need to grab the quota lock. If a quota applies to the * root inode, then it will be updated when quotas are enabled. * * However, we have an ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock)) * in getinoquota() that we want to keep so grab it anyway. */ rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); error = ufs_iget_alloced(vfsp, UFSROOTINO, &rip, cr); rw_exit(&ufsvfsp->vfs_dqrwlock); if (error) goto out; /* * make sure root inode is a directory. Returning ENOTDIR might * be confused with the mount point not being a directory, so * we use EIO instead. */ if ((rip->i_mode & IFMT) != IFDIR) { /* * Mark this inode as subject for cleanup * to avoid stray inodes in the cache. */ rvp = ITOV(rip); error = EIO; goto out; } rvp = ITOV(rip); mutex_enter(&rvp->v_lock); rvp->v_flag |= VROOT; mutex_exit(&rvp->v_lock); ufsvfsp->vfs_root = rvp; /* The buffer for the root inode does not contain a valid b_vp */ (void) bfinval(dev, 0); /* options */ ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC; ufsvfsp->vfs_nointr = flags & UFSMNT_NOINTR; ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR; ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME; if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime) ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME; else /* dfratime, default behavior */ ufsvfsp->vfs_dfritime |= UFS_DFRATIME; if (flags & UFSMNT_FORCEDIRECTIO) ufsvfsp->vfs_forcedirectio = 1; else if (flags & UFSMNT_NOFORCEDIRECTIO) ufsvfsp->vfs_forcedirectio = 0; ufsvfsp->vfs_iotstamp = lbolt; ufsvfsp->vfs_nindiroffset = fsp->fs_nindir - 1; ufsvfsp->vfs_nindirshift = highbit(ufsvfsp->vfs_nindiroffset); ufsvfsp->vfs_ioclustsz = fsp->fs_bsize * fsp->fs_maxcontig; if (cdev_ioctl(dev, DKIOCINFO, (intptr_t)&ci, FKIOCTL|FNATIVE|FREAD, CRED(), &status) == 0) { ufsvfsp->vfs_iotransz = ci.dki_maxtransfer * DEV_BSIZE; } else { ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys); } if (ufsvfsp->vfs_iotransz <= 0) { ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys); } /* * When logging, used to reserve log space for writes and truncs */ ufsvfsp->vfs_avgbfree = fsp->fs_cstotal.cs_nbfree / fsp->fs_ncg; /* * Determine whether to log cylinder group summary info. */ ufsvfsp->vfs_nolog_si = (fsp->fs_ncg < ufs_ncg_log); if (TRANS_ISTRANS(ufsvfsp)) { /* * start the delete thread */ ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp); /* * start reclaim thread if the filesystem was not mounted * read only. */ if (!fsp->fs_ronly && (fsp->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING))) { fsp->fs_reclaim &= ~FS_RECLAIM; fsp->fs_reclaim |= FS_RECLAIMING; ufs_thread_start(&ufsvfsp->vfs_reclaim, ufs_thread_reclaim, vfsp); } /* Mark the fs as unrolled */ fsp->fs_rolled = FS_NEED_ROLL; } else if (!fsp->fs_ronly && (fsp->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING))) { /* * If a file system that is mounted nologging, after * having previously been mounted logging, becomes * unmounted whilst the reclaim thread is in the throes * of reclaiming open/deleted inodes, a subsequent mount * of such a file system with logging disabled could lead * to inodes becoming lost. So, start reclaim now, even * though logging was disabled for the previous mount, to * tidy things up. */ fsp->fs_reclaim &= ~FS_RECLAIM; fsp->fs_reclaim |= FS_RECLAIMING; ufs_thread_start(&ufsvfsp->vfs_reclaim, ufs_thread_reclaim, vfsp); } if (!fsp->fs_ronly) { TRANS_SBWRITE(ufsvfsp, TOP_MOUNT); if (error = geterror(ufsvfsp->vfs_bufp)) goto out; } /* fix-on-panic initialization */ if (isroot && !(flags & UFSMNT_ONERROR_FLGMASK)) flags |= UFSMNT_ONERROR_PANIC; /* XXX ..._RDONLY */ if ((error = ufsfx_mount(ufsvfsp, flags)) != 0) goto out; if (why == ROOT_INIT && isroot) rootvp = devvp; return (0); out: if (error == 0) error = EIO; if (rvp) { /* the following sequence is similar to ufs_unmount() */ /* * There's a problem that ufs_iget() puts inodes into * the inode cache before it returns them. If someone * traverses that cache and gets a reference to our * inode, there's a chance they'll still be using it * after we've destroyed it. This is a hard race to * hit, but it's happened (putting in a medium delay * here, and a large delay in ufs_scan_inodes() for * inodes on the device we're bailing out on, makes * the race easy to demonstrate). The symptom is some * other part of UFS faulting on bad inode contents, * or when grabbing one of the locks inside the inode, * etc. The usual victim is ufs_scan_inodes() or * someone called by it. */ /* * First, isolate it so that no new references can be * gotten via the inode cache. */ ihm = &ih_lock[INOHASH(UFSROOTINO)]; mutex_enter(ihm); remque(rip); mutex_exit(ihm); /* * Now wait for all outstanding references except our * own to drain. This could, in theory, take forever, * so don't wait *too* long. If we time out, mark * it stale and leak it, so we don't hit the problem * described above. * * Note that v_count is an int, which means we can read * it in one operation. Thus, there's no need to lock * around our tests. */ elapsed = 0; while ((rvp->v_count > 1) && (elapsed < ufs_mount_timeout)) { delay(ufs_mount_error_delay * drv_usectohz(1000)); elapsed += ufs_mount_error_delay; } if (rvp->v_count > 1) { mutex_enter(&rip->i_tlock); rip->i_flag |= ISTALE; mutex_exit(&rip->i_tlock); cmn_err(CE_WARN, "Timed out while cleaning up after " "failed mount of %s", path); } else { /* * Now we're the only one with a handle left, so tear * it down the rest of the way. */ if (ufs_rmidle(rip)) VN_RELE(rvp); ufs_si_del(rip); rip->i_ufsvfs = NULL; rvp->v_vfsp = NULL; rvp->v_type = VBAD; VN_RELE(rvp); } } if (needtrans) { TRANS_MATA_UMOUNT(ufsvfsp); } if (ufsvfsp) { ufs_vfs_remove(ufsvfsp); ufs_thread_exit(&ufsvfsp->vfs_delete); ufs_thread_exit(&ufsvfsp->vfs_reclaim); mutex_destroy(&ufsvfsp->vfs_lock); if (ufsvfsp->vfs_log) { lufs_unsnarf(ufsvfsp); } kmem_free(ufsvfsp, sizeof (struct ufsvfs)); } if (bp) { bp->b_flags |= (B_STALE|B_AGE); brelse(bp); } if (tp) { tp->b_flags |= (B_STALE|B_AGE); brelse(tp); } if (needclose) { (void) VOP_CLOSE(devvp, (vfsp->vfs_flag & VFS_RDONLY) ? FREAD : FREAD|FWRITE, 1, (offset_t)0, cr, NULL); bflush(dev); (void) bfinval(dev, 1); } return (error); } /* * vfs operations */ static int ufs_unmount(struct vfs *vfsp, int fflag, struct cred *cr) { dev_t dev = vfsp->vfs_dev; struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; struct fs *fs = ufsvfsp->vfs_fs; struct ulockfs *ulp = &ufsvfsp->vfs_ulockfs; struct vnode *bvp, *vp; struct buf *bp; struct inode *ip, *inext, *rip; union ihead *ih; int error, flag, i; struct lockfs lockfs; int poll_events = POLLPRI; extern struct pollhead ufs_pollhd; refstr_t *mountpoint; ASSERT(vfs_lock_held(vfsp)); if (secpolicy_fs_unmount(cr, vfsp) != 0) return (EPERM); /* * Forced unmount is now supported through the * lockfs protocol. */ if (fflag & MS_FORCE) { /* * Mark the filesystem as being unmounted now in * case of a forcible umount before we take any * locks inside UFS to prevent racing with a VFS_VGET() * request. Throw these VFS_VGET() requests away for * the duration of the forcible umount so they won't * use stale or even freed data later on when we're done. * It may happen that the VFS has had a additional hold * placed on it by someone other than UFS and thus will * not get freed immediately once we're done with the * umount by dounmount() - use VFS_UNMOUNTED to inform * users of this still-alive VFS that its corresponding * filesystem being gone so they can detect that and error * out. */ vfsp->vfs_flag |= VFS_UNMOUNTED; ufs_thread_suspend(&ufsvfsp->vfs_delete); mutex_enter(&ulp->ul_lock); /* * If file system is already hard locked, * unmount the file system, otherwise * hard lock it before unmounting. */ if (!ULOCKFS_IS_HLOCK(ulp)) { atomic_add_long(&ufs_quiesce_pend, 1); lockfs.lf_lock = LOCKFS_HLOCK; lockfs.lf_flags = 0; lockfs.lf_key = ulp->ul_lockfs.lf_key + 1; lockfs.lf_comlen = 0; lockfs.lf_comment = NULL; ufs_freeze(ulp, &lockfs); ULOCKFS_SET_BUSY(ulp); LOCKFS_SET_BUSY(&ulp->ul_lockfs); (void) ufs_quiesce(ulp); (void) ufs_flush(vfsp); (void) ufs_thaw(vfsp, ufsvfsp, ulp); atomic_add_long(&ufs_quiesce_pend, -1); ULOCKFS_CLR_BUSY(ulp); LOCKFS_CLR_BUSY(&ulp->ul_lockfs); poll_events |= POLLERR; pollwakeup(&ufs_pollhd, poll_events); } ufs_thread_continue(&ufsvfsp->vfs_delete); mutex_exit(&ulp->ul_lock); } /* let all types of writes go through */ ufsvfsp->vfs_iotstamp = lbolt; /* coordinate with global hlock thread */ if (TRANS_ISTRANS(ufsvfsp) && (ufsvfsp->vfs_validfs == UT_HLOCKING)) { /* * last possibility for a forced umount to fail hence clear * VFS_UNMOUNTED if appropriate. */ if (fflag & MS_FORCE) vfsp->vfs_flag &= ~VFS_UNMOUNTED; return (EAGAIN); } ufsvfsp->vfs_validfs = UT_UNMOUNTED; /* kill the reclaim thread */ ufs_thread_exit(&ufsvfsp->vfs_reclaim); /* suspend the delete thread */ ufs_thread_suspend(&ufsvfsp->vfs_delete); /* * drain the delete and idle queues */ ufs_delete_drain(vfsp, -1, 1); ufs_idle_drain(vfsp); /* * use the lockfs protocol to prevent new ops from starting * a forcible umount can not fail beyond this point as * we hard-locked the filesystem and drained all current consumers * before. */ mutex_enter(&ulp->ul_lock); /* * if the file system is busy; return EBUSY */ if (ulp->ul_vnops_cnt || ulp->ul_falloc_cnt || ULOCKFS_IS_SLOCK(ulp)) { error = EBUSY; goto out; } /* * if this is not a forced unmount (!hard/error locked), then * get rid of every inode except the root and quota inodes * also, commit any outstanding transactions */ if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp)) if (error = ufs_flush(vfsp)) goto out; /* * ignore inodes in the cache if fs is hard locked or error locked */ rip = VTOI(ufsvfsp->vfs_root); if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp)) { /* * Otherwise, only the quota and root inodes are in the cache. * * Avoid racing with ufs_update() and ufs_sync(). */ mutex_enter(&ufs_scan_lock); for (i = 0, ih = ihead; i < inohsz; i++, ih++) { mutex_enter(&ih_lock[i]); for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw) { if (ip->i_ufsvfs != ufsvfsp) continue; if (ip == ufsvfsp->vfs_qinod) continue; if (ip == rip && ITOV(ip)->v_count == 1) continue; mutex_exit(&ih_lock[i]); mutex_exit(&ufs_scan_lock); error = EBUSY; goto out; } mutex_exit(&ih_lock[i]); } mutex_exit(&ufs_scan_lock); } /* * if a snapshot exists and this is a forced unmount, then delete * the snapshot. Otherwise return EBUSY. This will insure the * snapshot always belongs to a valid file system. */ if (ufsvfsp->vfs_snapshot) { if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) { (void) fssnap_delete(&ufsvfsp->vfs_snapshot); } else { error = EBUSY; goto out; } } /* * Close the quota file and invalidate anything left in the quota * cache for this file system. Pass kcred to allow all quota * manipulations. */ (void) closedq(ufsvfsp, kcred); invalidatedq(ufsvfsp); /* * drain the delete and idle queues */ ufs_delete_drain(vfsp, -1, 0); ufs_idle_drain(vfsp); /* * discard the inodes for this fs (including root, shadow, and quota) */ for (i = 0, ih = ihead; i < inohsz; i++, ih++) { mutex_enter(&ih_lock[i]); for (inext = 0, ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = inext) { inext = ip->i_forw; if (ip->i_ufsvfs != ufsvfsp) continue; /* * We've found the inode in the cache and as we * hold the hash mutex the inode can not * disappear from underneath us. * We also know it must have at least a vnode * reference count of 1. * We perform an additional VN_HOLD so the VN_RELE * in case we take the inode off the idle queue * can not be the last one. * It is safe to grab the writer contents lock here * to prevent a race with ufs_iinactive() putting * inodes into the idle queue while we operate on * this inode. */ rw_enter(&ip->i_contents, RW_WRITER); vp = ITOV(ip); VN_HOLD(vp) remque(ip); if (ufs_rmidle(ip)) VN_RELE(vp); ufs_si_del(ip); /* * rip->i_ufsvfsp is needed by bflush() */ if (ip != rip) ip->i_ufsvfs = NULL; /* * Set vnode's vfsops to dummy ops, which return * EIO. This is needed to forced unmounts to work * with lofs/nfs properly. */ if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) vp->v_vfsp = &EIO_vfs; else vp->v_vfsp = NULL; vp->v_type = VBAD; rw_exit(&ip->i_contents); VN_RELE(vp); } mutex_exit(&ih_lock[i]); } ufs_si_cache_flush(dev); /* * kill the delete thread and drain the idle queue */ ufs_thread_exit(&ufsvfsp->vfs_delete); ufs_idle_drain(vfsp); bp = ufsvfsp->vfs_bufp; bvp = ufsvfsp->vfs_devvp; flag = !fs->fs_ronly; if (flag) { bflush(dev); if (fs->fs_clean != FSBAD) { if (fs->fs_clean == FSSTABLE) fs->fs_clean = FSCLEAN; fs->fs_reclaim &= ~FS_RECLAIM; } if (TRANS_ISTRANS(ufsvfsp) && !TRANS_ISERROR(ufsvfsp) && !ULOCKFS_IS_HLOCK(ulp) && (fs->fs_rolled == FS_NEED_ROLL)) { /* * ufs_flush() above has flushed the last Moby. * This is needed to ensure the following superblock * update really is the last metadata update */ error = ufs_putsummaryinfo(dev, ufsvfsp, fs); if (error == 0) { fs->fs_rolled = FS_ALL_ROLLED; } } TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UNMOUNT); /* * push this last transaction */ curthread->t_flag |= T_DONTBLOCK; TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UNMOUNT, TOP_COMMIT_SIZE, error); if (!error) TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UNMOUNT, TOP_COMMIT_SIZE); curthread->t_flag &= ~T_DONTBLOCK; } TRANS_MATA_UMOUNT(ufsvfsp); lufs_unsnarf(ufsvfsp); /* Release the in-memory structs */ ufsfx_unmount(ufsvfsp); /* fix-on-panic bookkeeping */ kmem_free(fs->fs_u.fs_csp, fs->fs_cssize); bp->b_flags |= B_STALE|B_AGE; ufsvfsp->vfs_bufp = NULL; /* don't point at freed buf */ brelse(bp); /* free the superblock buf */ (void) VOP_PUTPAGE(common_specvp(bvp), (offset_t)0, (size_t)0, B_INVAL, cr, NULL); (void) VOP_CLOSE(bvp, flag, 1, (offset_t)0, cr, NULL); bflush(dev); (void) bfinval(dev, 1); VN_RELE(bvp); /* * It is now safe to NULL out the ufsvfs pointer and discard * the root inode. */ rip->i_ufsvfs = NULL; VN_RELE(ITOV(rip)); /* free up lockfs comment structure, if any */ if (ulp->ul_lockfs.lf_comlen && ulp->ul_lockfs.lf_comment) kmem_free(ulp->ul_lockfs.lf_comment, ulp->ul_lockfs.lf_comlen); /* * Remove from instance list. */ ufs_vfs_remove(ufsvfsp); /* * For a forcible unmount, threads may be asleep in * ufs_lockfs_begin/ufs_check_lockfs. These threads will need * the ufsvfs structure so we don't free it, yet. ufs_update * will free it up after awhile. */ if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) { extern kmutex_t ufsvfs_mutex; extern struct ufsvfs *ufsvfslist; mutex_enter(&ufsvfs_mutex); ufsvfsp->vfs_dontblock = 1; ufsvfsp->vfs_next = ufsvfslist; ufsvfslist = ufsvfsp; mutex_exit(&ufsvfs_mutex); /* wakeup any suspended threads */ cv_broadcast(&ulp->ul_cv); mutex_exit(&ulp->ul_lock); } else { mutex_destroy(&ufsvfsp->vfs_lock); kmem_free(ufsvfsp, sizeof (struct ufsvfs)); } /* * Now mark the filesystem as unmounted since we're done with it. */ vfsp->vfs_flag |= VFS_UNMOUNTED; return (0); out: /* open the fs to new ops */ cv_broadcast(&ulp->ul_cv); mutex_exit(&ulp->ul_lock); if (TRANS_ISTRANS(ufsvfsp)) { /* allow the delete thread to continue */ ufs_thread_continue(&ufsvfsp->vfs_delete); /* restart the reclaim thread */ ufs_thread_start(&ufsvfsp->vfs_reclaim, ufs_thread_reclaim, vfsp); /* coordinate with global hlock thread */ ufsvfsp->vfs_validfs = UT_MOUNTED; /* check for trans errors during umount */ ufs_trans_onerror(); /* * if we have a separate /usr it will never unmount * when halting. In order to not re-read all the * cylinder group summary info on mounting after * reboot the logging of summary info is re-enabled * and the super block written out. */ mountpoint = vfs_getmntpoint(vfsp); if ((fs->fs_si == FS_SI_OK) && (strcmp("/usr", refstr_value(mountpoint)) == 0)) { ufsvfsp->vfs_nolog_si = 0; UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp); } refstr_rele(mountpoint); } return (error); } static int ufs_root(struct vfs *vfsp, struct vnode **vpp) { struct ufsvfs *ufsvfsp; struct vnode *vp; if (!vfsp) return (EIO); ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; if (!ufsvfsp || !ufsvfsp->vfs_root) return (EIO); /* forced unmount */ vp = ufsvfsp->vfs_root; VN_HOLD(vp); *vpp = vp; return (0); } /* * Get file system statistics. */ static int ufs_statvfs(struct vfs *vfsp, struct statvfs64 *sp) { struct fs *fsp; struct ufsvfs *ufsvfsp; int blk, i; long max_avail, used; dev32_t d32; if (vfsp->vfs_flag & VFS_UNMOUNTED) return (EIO); ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; fsp = ufsvfsp->vfs_fs; if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC)) return (EINVAL); if (fsp->fs_magic == FS_MAGIC && (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 && fsp->fs_version != UFS_VERSION_MIN)) return (EINVAL); if (fsp->fs_magic == MTB_UFS_MAGIC && (fsp->fs_version > MTB_UFS_VERSION_1 || fsp->fs_version < MTB_UFS_VERSION_MIN)) return (EINVAL); /* * get the basic numbers */ (void) bzero(sp, sizeof (*sp)); sp->f_bsize = fsp->fs_bsize; sp->f_frsize = fsp->fs_fsize; sp->f_blocks = (fsblkcnt64_t)fsp->fs_dsize; sp->f_bfree = (fsblkcnt64_t)fsp->fs_cstotal.cs_nbfree * fsp->fs_frag + fsp->fs_cstotal.cs_nffree; sp->f_files = (fsfilcnt64_t)fsp->fs_ncg * fsp->fs_ipg; sp->f_ffree = (fsfilcnt64_t)fsp->fs_cstotal.cs_nifree; /* * Adjust the numbers based on things waiting to be deleted. * modifies f_bfree and f_ffree. Afterwards, everything we * come up with will be self-consistent. By definition, this * is a point-in-time snapshot, so the fact that the delete * thread's probably already invalidated the results is not a * problem. Note that if the delete thread is ever extended to * non-logging ufs, this adjustment must always be made. */ if (TRANS_ISTRANS(ufsvfsp)) ufs_delete_adjust_stats(ufsvfsp, sp); /* * avail = MAX(max_avail - used, 0) */ max_avail = fsp->fs_dsize - ufsvfsp->vfs_minfrags; used = (fsp->fs_dsize - sp->f_bfree); if (max_avail > used) sp->f_bavail = (fsblkcnt64_t)max_avail - used; else sp->f_bavail = (fsblkcnt64_t)0; sp->f_favail = sp->f_ffree; (void) cmpldev(&d32, vfsp->vfs_dev); sp->f_fsid = d32; (void) strcpy(sp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name); sp->f_flag = vf_to_stf(vfsp->vfs_flag); /* keep coordinated with ufs_l_pathconf() */ sp->f_namemax = MAXNAMLEN; if (fsp->fs_cpc == 0) { bzero(sp->f_fstr, 14); return (0); } blk = fsp->fs_spc * fsp->fs_cpc / NSPF(fsp); for (i = 0; i < blk; i += fsp->fs_frag) /* CSTYLED */ /* void */; i -= fsp->fs_frag; blk = i / fsp->fs_frag; bcopy(&(fs_rotbl(fsp)[blk]), sp->f_fstr, 14); return (0); } /* * Flush any pending I/O to file system vfsp. * The ufs_update() routine will only flush *all* ufs files. * If vfsp is non-NULL, only sync this ufs (in preparation * for a umount). */ /*ARGSUSED*/ static int ufs_sync(struct vfs *vfsp, short flag, struct cred *cr) { struct ufsvfs *ufsvfsp; struct fs *fs; int cheap = flag & SYNC_ATTR; int error; /* * SYNC_CLOSE means we're rebooting. Toss everything * on the idle queue so we don't have to slog through * a bunch of uninteresting inodes over and over again. */ if (flag & SYNC_CLOSE) ufs_idle_drain(NULL); if (vfsp == NULL) { ufs_update(flag); return (0); } /* Flush a single ufs */ if (!vfs_matchops(vfsp, ufs_vfsops) || vfs_lock(vfsp) != 0) return (0); ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; if (!ufsvfsp) return (EIO); fs = ufsvfsp->vfs_fs; mutex_enter(&ufsvfsp->vfs_lock); if (ufsvfsp->vfs_dio && fs->fs_ronly == 0 && fs->fs_clean != FSBAD && fs->fs_clean != FSLOG) { /* turn off fast-io on unmount, so no fsck needed (4029401) */ ufsvfsp->vfs_dio = 0; fs->fs_clean = FSACTIVE; fs->fs_fmod = 1; } /* Write back modified superblock */ if (fs->fs_fmod == 0) { mutex_exit(&ufsvfsp->vfs_lock); } else { if (fs->fs_ronly != 0) { mutex_exit(&ufsvfsp->vfs_lock); vfs_unlock(vfsp); return (ufs_fault(ufsvfsp->vfs_root, "fs = %s update: ro fs mod\n", fs->fs_fsmnt)); } fs->fs_fmod = 0; mutex_exit(&ufsvfsp->vfs_lock); TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UPDATE); } vfs_unlock(vfsp); /* * Avoid racing with ufs_update() and ufs_unmount(). * */ mutex_enter(&ufs_scan_lock); (void) ufs_scan_inodes(1, ufs_sync_inode, (void *)(uintptr_t)cheap, ufsvfsp); mutex_exit(&ufs_scan_lock); bflush((dev_t)vfsp->vfs_dev); /* * commit any outstanding async transactions */ curthread->t_flag |= T_DONTBLOCK; TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UPDATE, TOP_COMMIT_SIZE, error); if (!error) { TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UPDATE, TOP_COMMIT_SIZE); } curthread->t_flag &= ~T_DONTBLOCK; return (0); } void sbupdate(struct vfs *vfsp) { struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; struct fs *fs = ufsvfsp->vfs_fs; struct buf *bp; int blks; caddr_t space; int i; size_t size; /* * for ulockfs processing, limit the superblock writes */ if ((ufsvfsp->vfs_ulockfs.ul_sbowner) && (curthread != ufsvfsp->vfs_ulockfs.ul_sbowner)) { /* process later */ fs->fs_fmod = 1; return; } ULOCKFS_SET_MOD((&ufsvfsp->vfs_ulockfs)); if (TRANS_ISTRANS(ufsvfsp)) { mutex_enter(&ufsvfsp->vfs_lock); ufs_sbwrite(ufsvfsp); mutex_exit(&ufsvfsp->vfs_lock); return; } blks = howmany(fs->fs_cssize, fs->fs_fsize); space = (caddr_t)fs->fs_u.fs_csp; for (i = 0; i < blks; i += fs->fs_frag) { size = fs->fs_bsize; if (i + fs->fs_frag > blks) size = (blks - i) * fs->fs_fsize; bp = UFS_GETBLK(ufsvfsp, ufsvfsp->vfs_dev, (daddr_t)(fsbtodb(fs, fs->fs_csaddr + i)), fs->fs_bsize); bcopy(space, bp->b_un.b_addr, size); space += size; bp->b_bcount = size; UFS_BRWRITE(ufsvfsp, bp); } mutex_enter(&ufsvfsp->vfs_lock); ufs_sbwrite(ufsvfsp); mutex_exit(&ufsvfsp->vfs_lock); } int ufs_vget_idle_count = 2; /* Number of inodes to idle each time */ static int ufs_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp) { int error = 0; struct ufid *ufid; struct inode *ip; struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; struct ulockfs *ulp; /* * Check for unmounted filesystem. */ if (vfsp->vfs_flag & VFS_UNMOUNTED) { error = EIO; goto errout; } /* * Keep the idle queue from getting too long by * idling an inode before attempting to allocate another. * This operation must be performed before entering * lockfs or a transaction. */ if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat) if ((curthread->t_flag & T_DONTBLOCK) == 0) { ins.in_vidles.value.ul += ufs_vget_idle_count; ufs_idle_some(ufs_vget_idle_count); } ufid = (struct ufid *)fidp; if (error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_VGET_MASK)) goto errout; rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); error = ufs_iget(vfsp, ufid->ufid_ino, &ip, CRED()); rw_exit(&ufsvfsp->vfs_dqrwlock); ufs_lockfs_end(ulp); if (error) goto errout; /* * Check if the inode has been deleted or freed or is in transient state * since the last VFS_VGET() request for it, release it and don't return * it to the caller, presumably NFS, as it's no longer valid. */ if (ip->i_gen != ufid->ufid_gen || ip->i_mode == 0 || (ip->i_flag & IDEL)) { VN_RELE(ITOV(ip)); error = EINVAL; goto errout; } *vpp = ITOV(ip); return (0); errout: *vpp = NULL; return (error); } static int ufsinit(int fstype, char *name) { static const fs_operation_def_t ufs_vfsops_template[] = { VFSNAME_MOUNT, { .vfs_mount = ufs_mount }, VFSNAME_UNMOUNT, { .vfs_unmount = ufs_unmount }, VFSNAME_ROOT, { .vfs_root = ufs_root }, VFSNAME_STATVFS, { .vfs_statvfs = ufs_statvfs }, VFSNAME_SYNC, { .vfs_sync = ufs_sync }, VFSNAME_VGET, { .vfs_vget = ufs_vget }, VFSNAME_MOUNTROOT, { .vfs_mountroot = ufs_mountroot }, NULL, NULL }; int error; ufsfstype = fstype; error = vfs_setfsops(fstype, ufs_vfsops_template, &ufs_vfsops); if (error != 0) { cmn_err(CE_WARN, "ufsinit: bad vfs ops template"); return (error); } error = vn_make_ops(name, ufs_vnodeops_template, &ufs_vnodeops); if (error != 0) { (void) vfs_freevfsops_by_type(fstype); cmn_err(CE_WARN, "ufsinit: bad vnode ops template"); return (error); } ufs_iinit(); return (0); } #ifdef __sparc /* * Mounting a mirrored SVM volume is only supported on ufs, * this is special-case boot code to support that configuration. * At this point, we have booted and mounted root on a * single component of the mirror. Complete the boot * by configuring SVM and converting the root to the * dev_t of the mirrored root device. This dev_t conversion * only works because the underlying device doesn't change. */ int ufs_remountroot(struct vfs *vfsp) { struct ufsvfs *ufsvfsp; struct ulockfs *ulp; dev_t new_rootdev; dev_t old_rootdev; struct vnode *old_rootvp; struct vnode *new_rootvp; int error, sberror = 0; struct inode *ip; union ihead *ih; struct buf *bp; int i; old_rootdev = rootdev; old_rootvp = rootvp; new_rootdev = getrootdev(); if (new_rootdev == (dev_t)NODEV) { return (ENODEV); } new_rootvp = makespecvp(new_rootdev, VBLK); error = VOP_OPEN(&new_rootvp, (vfsp->vfs_flag & VFS_RDONLY) ? FREAD : FREAD|FWRITE, CRED(), NULL); if (error) { cmn_err(CE_CONT, "Cannot open mirrored root device, error %d\n", error); return (error); } if (vfs_lock(vfsp) != 0) { return (EBUSY); } ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; ulp = &ufsvfsp->vfs_ulockfs; mutex_enter(&ulp->ul_lock); atomic_add_long(&ufs_quiesce_pend, 1); (void) ufs_quiesce(ulp); (void) ufs_flush(vfsp); /* * Convert root vfs to new dev_t, including vfs hash * table and fs id. */ vfs_root_redev(vfsp, new_rootdev, ufsfstype); ufsvfsp->vfs_devvp = new_rootvp; ufsvfsp->vfs_dev = new_rootdev; bp = ufsvfsp->vfs_bufp; bp->b_edev = new_rootdev; bp->b_dev = cmpdev(new_rootdev); /* * The buffer for the root inode does not contain a valid b_vp */ (void) bfinval(new_rootdev, 0); /* * Here we hand-craft inodes with old root device * references to refer to the new device instead. */ mutex_enter(&ufs_scan_lock); for (i = 0, ih = ihead; i < inohsz; i++, ih++) { mutex_enter(&ih_lock[i]); for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw) { if (ip->i_ufsvfs != ufsvfsp) continue; if (ip == ufsvfsp->vfs_qinod) continue; if (ip->i_dev == old_rootdev) { ip->i_dev = new_rootdev; } if (ip->i_devvp == old_rootvp) { ip->i_devvp = new_rootvp; } } mutex_exit(&ih_lock[i]); } mutex_exit(&ufs_scan_lock); /* * Make Sure logging structures are using the new device * if logging is enabled. Also start any logging thread that * needs to write to the device and couldn't earlier. */ if (ufsvfsp->vfs_log) { buf_t *bp, *tbp; ml_unit_t *ul = ufsvfsp->vfs_log; struct fs *fsp = ufsvfsp->vfs_fs; /* * Update the main logging structure. */ ul->un_dev = new_rootdev; /* * Get a new bp for the on disk structures. */ bp = ul->un_bp; tbp = ngeteblk(dbtob(LS_SECTORS)); tbp->b_edev = new_rootdev; tbp->b_dev = cmpdev(new_rootdev); tbp->b_blkno = bp->b_blkno; bcopy(bp->b_un.b_addr, tbp->b_un.b_addr, DEV_BSIZE); bcopy(bp->b_un.b_addr, tbp->b_un.b_addr + DEV_BSIZE, DEV_BSIZE); bp->b_flags |= (B_STALE | B_AGE); brelse(bp); ul->un_bp = tbp; /* * Allocate new circular buffers. */ alloc_rdbuf(&ul->un_rdbuf, MAPBLOCKSIZE, MAPBLOCKSIZE); alloc_wrbuf(&ul->un_wrbuf, ldl_bufsize(ul)); /* * Clear the noroll bit which indicates that logging * can't roll the log yet and start the logmap roll thread * unless the filesystem is still read-only in which case * remountfs() will do it when going to read-write. */ ASSERT(ul->un_flags & LDL_NOROLL); if (!fsp->fs_ronly) { ul->un_flags &= ~LDL_NOROLL; logmap_start_roll(ul); } /* * Start the reclaim thread if needed. */ if (!fsp->fs_ronly && (fsp->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING))) { fsp->fs_reclaim &= ~FS_RECLAIM; fsp->fs_reclaim |= FS_RECLAIMING; ufs_thread_start(&ufsvfsp->vfs_reclaim, ufs_thread_reclaim, vfsp); TRANS_SBWRITE(ufsvfsp, TOP_SBUPDATE_UPDATE); if (sberror = geterror(ufsvfsp->vfs_bufp)) { refstr_t *mntpt; mntpt = vfs_getmntpoint(vfsp); cmn_err(CE_WARN, "Remountroot failed to update Reclaim" "state for filesystem %s " "Error writing SuperBlock %d", refstr_value(mntpt), error); refstr_rele(mntpt); } } } rootdev = new_rootdev; rootvp = new_rootvp; atomic_add_long(&ufs_quiesce_pend, -1); cv_broadcast(&ulp->ul_cv); mutex_exit(&ulp->ul_lock); vfs_unlock(vfsp); error = VOP_CLOSE(old_rootvp, FREAD, 1, (offset_t)0, CRED(), NULL); if (error) { cmn_err(CE_CONT, "close of root device component failed, error %d\n", error); } VN_RELE(old_rootvp); return (sberror ? sberror : error); } #endif /* __sparc */