2 * Copyright (c) 2000-2004 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
38 #include "xfs_trans.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_alloc_btree.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_btree.h"
49 #include "xfs_ialloc.h"
50 #include "xfs_attr_sf.h"
51 #include "xfs_dir_sf.h"
52 #include "xfs_dir2_sf.h"
53 #include "xfs_dinode.h"
54 #include "xfs_inode.h"
55 #include "xfs_alloc.h"
56 #include "xfs_rtalloc.h"
58 #include "xfs_error.h"
61 #include "xfs_quota.h"
62 #include "xfs_fsops.h"
64 STATIC void xfs_mount_log_sbunit(xfs_mount_t *, __int64_t);
65 STATIC int xfs_uuid_mount(xfs_mount_t *);
66 STATIC void xfs_uuid_unmount(xfs_mount_t *mp);
68 void xfs_xlatesb(void *, xfs_sb_t *, int, xfs_arch_t, __int64_t);
72 short type; /* 0 = integer
73 * 1 = binary / string (no translation)
76 { offsetof(xfs_sb_t, sb_magicnum), 0 },
77 { offsetof(xfs_sb_t, sb_blocksize), 0 },
78 { offsetof(xfs_sb_t, sb_dblocks), 0 },
79 { offsetof(xfs_sb_t, sb_rblocks), 0 },
80 { offsetof(xfs_sb_t, sb_rextents), 0 },
81 { offsetof(xfs_sb_t, sb_uuid), 1 },
82 { offsetof(xfs_sb_t, sb_logstart), 0 },
83 { offsetof(xfs_sb_t, sb_rootino), 0 },
84 { offsetof(xfs_sb_t, sb_rbmino), 0 },
85 { offsetof(xfs_sb_t, sb_rsumino), 0 },
86 { offsetof(xfs_sb_t, sb_rextsize), 0 },
87 { offsetof(xfs_sb_t, sb_agblocks), 0 },
88 { offsetof(xfs_sb_t, sb_agcount), 0 },
89 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
90 { offsetof(xfs_sb_t, sb_logblocks), 0 },
91 { offsetof(xfs_sb_t, sb_versionnum), 0 },
92 { offsetof(xfs_sb_t, sb_sectsize), 0 },
93 { offsetof(xfs_sb_t, sb_inodesize), 0 },
94 { offsetof(xfs_sb_t, sb_inopblock), 0 },
95 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
96 { offsetof(xfs_sb_t, sb_blocklog), 0 },
97 { offsetof(xfs_sb_t, sb_sectlog), 0 },
98 { offsetof(xfs_sb_t, sb_inodelog), 0 },
99 { offsetof(xfs_sb_t, sb_inopblog), 0 },
100 { offsetof(xfs_sb_t, sb_agblklog), 0 },
101 { offsetof(xfs_sb_t, sb_rextslog), 0 },
102 { offsetof(xfs_sb_t, sb_inprogress), 0 },
103 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
104 { offsetof(xfs_sb_t, sb_icount), 0 },
105 { offsetof(xfs_sb_t, sb_ifree), 0 },
106 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
107 { offsetof(xfs_sb_t, sb_frextents), 0 },
108 { offsetof(xfs_sb_t, sb_uquotino), 0 },
109 { offsetof(xfs_sb_t, sb_gquotino), 0 },
110 { offsetof(xfs_sb_t, sb_qflags), 0 },
111 { offsetof(xfs_sb_t, sb_flags), 0 },
112 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
113 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
114 { offsetof(xfs_sb_t, sb_unit), 0 },
115 { offsetof(xfs_sb_t, sb_width), 0 },
116 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
117 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
118 { offsetof(xfs_sb_t, sb_logsectsize),0 },
119 { offsetof(xfs_sb_t, sb_logsunit), 0 },
120 { offsetof(xfs_sb_t, sb_features2), 0 },
121 { sizeof(xfs_sb_t), 0 }
125 * Return a pointer to an initialized xfs_mount structure.
132 mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
134 AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail");
135 spinlock_init(&mp->m_sb_lock, "xfs_sb");
136 mutex_init(&mp->m_ilock, MUTEX_DEFAULT, "xfs_ilock");
137 initnsema(&mp->m_growlock, 1, "xfs_grow");
139 * Initialize the AIL.
141 xfs_trans_ail_init(mp);
143 atomic_set(&mp->m_active_trans, 0);
149 * Free up the resources associated with a mount structure. Assume that
150 * the structure was initially zeroed, so we can tell which fields got
166 for (agno = 0; agno < mp->m_maxagi; agno++)
167 if (mp->m_perag[agno].pagb_list)
168 kmem_free(mp->m_perag[agno].pagb_list,
169 sizeof(xfs_perag_busy_t) *
171 kmem_free(mp->m_perag,
172 sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
175 AIL_LOCK_DESTROY(&mp->m_ail_lock);
176 spinlock_destroy(&mp->m_sb_lock);
177 mutex_destroy(&mp->m_ilock);
178 freesema(&mp->m_growlock);
182 if (mp->m_fsname != NULL)
183 kmem_free(mp->m_fsname, mp->m_fsname_len);
186 struct vfs *vfsp = XFS_MTOVFS(mp);
188 bhv_remove_all_vfsops(vfsp, 0);
189 VFS_REMOVEBHV(vfsp, &mp->m_bhv);
192 kmem_free(mp, sizeof(xfs_mount_t));
197 * Check the validity of the SB found.
200 xfs_mount_validate_sb(
205 * If the log device and data device have the
206 * same device number, the log is internal.
207 * Consequently, the sb_logstart should be non-zero. If
208 * we have a zero sb_logstart in this case, we may be trying to mount
209 * a volume filesystem in a non-volume manner.
211 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
212 cmn_err(CE_WARN, "XFS: bad magic number");
213 return XFS_ERROR(EWRONGFS);
216 if (!XFS_SB_GOOD_VERSION(sbp)) {
217 cmn_err(CE_WARN, "XFS: bad version");
218 return XFS_ERROR(EWRONGFS);
222 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
224 "XFS: filesystem is marked as having an external log; "
225 "specify logdev on the\nmount command line.");
226 XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(1)",
227 XFS_ERRLEVEL_HIGH, mp, sbp);
228 return XFS_ERROR(EFSCORRUPTED);
232 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
234 "XFS: filesystem is marked as having an internal log; "
235 "don't specify logdev on\nthe mount command line.");
236 XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(2)",
237 XFS_ERRLEVEL_HIGH, mp, sbp);
238 return XFS_ERROR(EFSCORRUPTED);
242 * More sanity checking. These were stolen directly from
246 sbp->sb_agcount <= 0 ||
247 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
248 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
249 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
250 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
251 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
252 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
253 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
254 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
255 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
256 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
257 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
258 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
259 sbp->sb_imax_pct > 100)) {
260 cmn_err(CE_WARN, "XFS: SB sanity check 1 failed");
261 XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(3)",
262 XFS_ERRLEVEL_LOW, mp, sbp);
263 return XFS_ERROR(EFSCORRUPTED);
267 * Sanity check AG count, size fields against data size field
270 sbp->sb_dblocks == 0 ||
272 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
273 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
274 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
275 cmn_err(CE_WARN, "XFS: SB sanity check 2 failed");
276 XFS_ERROR_REPORT("xfs_mount_validate_sb(4)",
277 XFS_ERRLEVEL_LOW, mp);
278 return XFS_ERROR(EFSCORRUPTED);
283 (sbp->sb_dblocks << (__uint64_t)(sbp->sb_blocklog - BBSHIFT))
285 (sbp->sb_rblocks << (__uint64_t)(sbp->sb_blocklog - BBSHIFT))
288 "XFS: File system is too large to be mounted on this system.");
289 return XFS_ERROR(E2BIG);
293 if (unlikely(sbp->sb_inprogress)) {
294 cmn_err(CE_WARN, "XFS: file system busy");
295 XFS_ERROR_REPORT("xfs_mount_validate_sb(5)",
296 XFS_ERRLEVEL_LOW, mp);
297 return XFS_ERROR(EFSCORRUPTED);
301 * Until this is fixed only page-sized or smaller data blocks work.
303 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
305 "XFS: Attempted to mount file system with blocksize %d bytes",
308 "XFS: Only page-sized (%d) or less blocksizes currently work.",
310 return XFS_ERROR(ENOSYS);
317 xfs_initialize_perag(xfs_mount_t *mp, int agcount)
319 int index, max_metadata;
323 xfs_sb_t *sbp = &mp->m_sb;
324 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
326 /* Check to see if the filesystem can overflow 32 bit inodes */
327 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
328 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
330 /* Clear the mount flag if no inode can overflow 32 bits
331 * on this filesystem, or if specifically requested..
333 if ((mp->m_flags & XFS_MOUNT_32BITINOOPT) && ino > max_inum) {
334 mp->m_flags |= XFS_MOUNT_32BITINODES;
336 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
339 /* If we can overflow then setup the ag headers accordingly */
340 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
341 /* Calculate how much should be reserved for inodes to
342 * meet the max inode percentage.
344 if (mp->m_maxicount) {
347 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
349 icount += sbp->sb_agblocks - 1;
350 do_div(icount, mp->m_ialloc_blks);
351 max_metadata = icount;
353 max_metadata = agcount;
355 for (index = 0; index < agcount; index++) {
356 ino = XFS_AGINO_TO_INO(mp, index, agino);
357 if (ino > max_inum) {
362 /* This ag is prefered for inodes */
363 pag = &mp->m_perag[index];
364 pag->pagi_inodeok = 1;
365 if (index < max_metadata)
366 pag->pagf_metadata = 1;
369 /* Setup default behavior for smaller filesystems */
370 for (index = 0; index < agcount; index++) {
371 pag = &mp->m_perag[index];
372 pag->pagi_inodeok = 1;
375 mp->m_maxagi = index;
381 * data - on disk version of sb
383 * dir - conversion direction: <0 - convert sb to buf
384 * >0 - convert buf to sb
385 * arch - architecture to read/write from/to buf
386 * fields - which fields to copy (bitmask)
408 buf_ptr = (xfs_caddr_t)data;
409 mem_ptr = (xfs_caddr_t)sb;
412 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
413 first = xfs_sb_info[f].offset;
414 size = xfs_sb_info[f + 1].offset - first;
416 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
418 if (arch == ARCH_NOCONVERT ||
420 xfs_sb_info[f].type == 1) {
422 memcpy(mem_ptr + first, buf_ptr + first, size);
424 memcpy(buf_ptr + first, mem_ptr + first, size);
429 INT_XLATE(*(__uint16_t*)(buf_ptr+first),
430 *(__uint16_t*)(mem_ptr+first),
434 INT_XLATE(*(__uint32_t*)(buf_ptr+first),
435 *(__uint32_t*)(mem_ptr+first),
439 INT_XLATE(*(__uint64_t*)(buf_ptr+first),
440 *(__uint64_t*)(mem_ptr+first), dir, arch);
447 fields &= ~(1LL << f);
454 * Does the initial read of the superblock.
457 xfs_readsb(xfs_mount_t *mp)
459 unsigned int sector_size;
460 unsigned int extra_flags;
465 ASSERT(mp->m_sb_bp == NULL);
466 ASSERT(mp->m_ddev_targp != NULL);
469 * Allocate a (locked) buffer to hold the superblock.
470 * This will be kept around at all times to optimize
471 * access to the superblock.
473 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
474 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
476 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
477 BTOBB(sector_size), extra_flags);
478 if (!bp || XFS_BUF_ISERROR(bp)) {
479 cmn_err(CE_WARN, "XFS: SB read failed");
480 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
483 ASSERT(XFS_BUF_ISBUSY(bp));
484 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
487 * Initialize the mount structure from the superblock.
488 * But first do some basic consistency checking.
490 sbp = XFS_BUF_TO_SBP(bp);
491 xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1,
492 ARCH_CONVERT, XFS_SB_ALL_BITS);
494 error = xfs_mount_validate_sb(mp, &(mp->m_sb));
496 cmn_err(CE_WARN, "XFS: SB validate failed");
501 * We must be able to do sector-sized and sector-aligned IO.
503 if (sector_size > mp->m_sb.sb_sectsize) {
505 "XFS: device supports only %u byte sectors (not %u)",
506 sector_size, mp->m_sb.sb_sectsize);
512 * If device sector size is smaller than the superblock size,
513 * re-read the superblock so the buffer is correctly sized.
515 if (sector_size < mp->m_sb.sb_sectsize) {
516 XFS_BUF_UNMANAGE(bp);
518 sector_size = mp->m_sb.sb_sectsize;
519 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
520 BTOBB(sector_size), extra_flags);
521 if (!bp || XFS_BUF_ISERROR(bp)) {
522 cmn_err(CE_WARN, "XFS: SB re-read failed");
523 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
526 ASSERT(XFS_BUF_ISBUSY(bp));
527 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
532 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
537 XFS_BUF_UNMANAGE(bp);
547 * Mount initialization code establishing various mount
548 * fields from the superblock associated with the given
552 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
556 mp->m_agfrotor = mp->m_agirotor = 0;
557 spinlock_init(&mp->m_agirotor_lock, "m_agirotor_lock");
558 mp->m_maxagi = mp->m_sb.sb_agcount;
559 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
560 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
561 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
562 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
563 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
564 mp->m_litino = sbp->sb_inodesize -
565 ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
566 mp->m_blockmask = sbp->sb_blocksize - 1;
567 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
568 mp->m_blockwmask = mp->m_blockwsize - 1;
569 INIT_LIST_HEAD(&mp->m_del_inodes);
572 * Setup for attributes, in case they get created.
573 * This value is for inodes getting attributes for the first time,
574 * the per-inode value is for old attribute values.
576 ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
577 switch (sbp->sb_inodesize) {
579 mp->m_attroffset = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(2);
584 mp->m_attroffset = XFS_BMDR_SPACE_CALC(12);
589 ASSERT(mp->m_attroffset < XFS_LITINO(mp));
591 for (i = 0; i < 2; i++) {
592 mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
594 mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
597 for (i = 0; i < 2; i++) {
598 mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
600 mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
603 for (i = 0; i < 2; i++) {
604 mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
606 mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
610 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
611 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
613 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
618 * This function does the following on an initial mount of a file system:
619 * - reads the superblock from disk and init the mount struct
620 * - if we're a 32-bit kernel, do a size check on the superblock
621 * so we don't mount terabyte filesystems
622 * - init mount struct realtime fields
623 * - allocate inode hash table for fs
624 * - init directory manager
625 * - perform recovery and init the log manager
634 xfs_sb_t *sbp = &(mp->m_sb);
637 int readio_log, writeio_log;
641 __int64_t update_flags;
642 uint quotamount, quotaflags;
644 int uuid_mounted = 0;
647 if (mp->m_sb_bp == NULL) {
648 if ((error = xfs_readsb(mp))) {
652 xfs_mount_common(mp, sbp);
655 * Check if sb_agblocks is aligned at stripe boundary
656 * If sb_agblocks is NOT aligned turn off m_dalign since
657 * allocator alignment is within an ag, therefore ag has
658 * to be aligned at stripe boundary.
661 if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
663 * If stripe unit and stripe width are not multiples
664 * of the fs blocksize turn off alignment.
666 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
667 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
668 if (mp->m_flags & XFS_MOUNT_RETERR) {
670 "XFS: alignment check 1 failed");
671 error = XFS_ERROR(EINVAL);
674 mp->m_dalign = mp->m_swidth = 0;
677 * Convert the stripe unit and width to FSBs.
679 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
680 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
681 if (mp->m_flags & XFS_MOUNT_RETERR) {
682 error = XFS_ERROR(EINVAL);
687 } else if (mp->m_dalign) {
688 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
690 if (mp->m_flags & XFS_MOUNT_RETERR) {
692 "XFS: alignment check 3 failed");
693 error = XFS_ERROR(EINVAL);
701 * Update superblock with new values
704 if (XFS_SB_VERSION_HASDALIGN(sbp)) {
705 if (sbp->sb_unit != mp->m_dalign) {
706 sbp->sb_unit = mp->m_dalign;
707 update_flags |= XFS_SB_UNIT;
709 if (sbp->sb_width != mp->m_swidth) {
710 sbp->sb_width = mp->m_swidth;
711 update_flags |= XFS_SB_WIDTH;
714 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
715 XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
716 mp->m_dalign = sbp->sb_unit;
717 mp->m_swidth = sbp->sb_width;
720 xfs_alloc_compute_maxlevels(mp);
721 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
722 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
723 xfs_ialloc_compute_maxlevels(mp);
725 if (sbp->sb_imax_pct) {
728 /* Make sure the maximum inode count is a multiple of the
729 * units we allocate inodes in.
732 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
734 do_div(icount, mp->m_ialloc_blks);
735 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
740 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
743 * XFS uses the uuid from the superblock as the unique
744 * identifier for fsid. We can not use the uuid from the volume
745 * since a single partition filesystem is identical to a single
746 * partition volume/filesystem.
748 if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
749 (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
750 if (xfs_uuid_mount(mp)) {
751 error = XFS_ERROR(EINVAL);
755 ret64 = uuid_hash64(&sbp->sb_uuid);
756 memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
760 * Set the default minimum read and write sizes unless
761 * already specified in a mount option.
762 * We use smaller I/O sizes when the file system
763 * is being used for NFS service (wsync mount option).
765 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
766 if (mp->m_flags & XFS_MOUNT_WSYNC) {
767 readio_log = XFS_WSYNC_READIO_LOG;
768 writeio_log = XFS_WSYNC_WRITEIO_LOG;
770 readio_log = XFS_READIO_LOG_LARGE;
771 writeio_log = XFS_WRITEIO_LOG_LARGE;
774 readio_log = mp->m_readio_log;
775 writeio_log = mp->m_writeio_log;
779 * Set the number of readahead buffers to use based on
780 * physical memory size.
782 if (xfs_physmem <= 4096) /* <= 16MB */
783 mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
784 else if (xfs_physmem <= 8192) /* <= 32MB */
785 mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
787 mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
788 if (sbp->sb_blocklog > readio_log) {
789 mp->m_readio_log = sbp->sb_blocklog;
791 mp->m_readio_log = readio_log;
793 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
794 if (sbp->sb_blocklog > writeio_log) {
795 mp->m_writeio_log = sbp->sb_blocklog;
797 mp->m_writeio_log = writeio_log;
799 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
802 * Set the inode cluster size based on the physical memory
803 * size. This may still be overridden by the file system
804 * block size if it is larger than the chosen cluster size.
806 if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
807 mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
809 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
812 * Set whether we're using inode alignment.
814 if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
815 mp->m_sb.sb_inoalignmt >=
816 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
817 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
819 mp->m_inoalign_mask = 0;
821 * If we are using stripe alignment, check whether
822 * the stripe unit is a multiple of the inode alignment
824 if (mp->m_dalign && mp->m_inoalign_mask &&
825 !(mp->m_dalign & mp->m_inoalign_mask))
826 mp->m_sinoalign = mp->m_dalign;
830 * Check that the data (and log if separate) are an ok size.
832 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
833 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
834 cmn_err(CE_WARN, "XFS: size check 1 failed");
835 error = XFS_ERROR(E2BIG);
838 error = xfs_read_buf(mp, mp->m_ddev_targp,
839 d - XFS_FSS_TO_BB(mp, 1),
840 XFS_FSS_TO_BB(mp, 1), 0, &bp);
844 cmn_err(CE_WARN, "XFS: size check 2 failed");
845 if (error == ENOSPC) {
846 error = XFS_ERROR(E2BIG);
851 if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
852 mp->m_logdev_targp != mp->m_ddev_targp) {
853 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
854 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
855 cmn_err(CE_WARN, "XFS: size check 3 failed");
856 error = XFS_ERROR(E2BIG);
859 error = xfs_read_buf(mp, mp->m_logdev_targp,
860 d - XFS_FSB_TO_BB(mp, 1),
861 XFS_FSB_TO_BB(mp, 1), 0, &bp);
865 cmn_err(CE_WARN, "XFS: size check 3 failed");
866 if (error == ENOSPC) {
867 error = XFS_ERROR(E2BIG);
874 * Initialize realtime fields in the mount structure
876 if ((error = xfs_rtmount_init(mp))) {
877 cmn_err(CE_WARN, "XFS: RT mount failed");
882 * For client case we are done now
884 if (mfsi_flags & XFS_MFSI_CLIENT) {
889 * Copies the low order bits of the timestamp and the randomly
890 * set "sequence" number out of a UUID.
892 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
895 * The vfs structure needs to have a file system independent
896 * way of checking for the invariant file system ID. Since it
897 * can't look at mount structures it has a pointer to the data
898 * in the mount structure.
900 * File systems that don't support user level file handles (i.e.
901 * all of them except for XFS) will leave vfs_altfsid as NULL.
903 vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
904 mp->m_dmevmask = 0; /* not persistent; set after each mount */
907 * Select the right directory manager.
910 XFS_SB_VERSION_HASDIRV2(&mp->m_sb) ?
915 * Initialize directory manager's entries.
920 * Initialize the attribute manager's entries.
922 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
925 * Initialize the precomputed transaction reservations values.
930 * Allocate and initialize the inode hash table for this
937 * Allocate and initialize the per-ag data.
939 init_rwsem(&mp->m_peraglock);
941 kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
943 xfs_initialize_perag(mp, sbp->sb_agcount);
946 * log's mount-time initialization. Perform 1st part recovery if needed
948 if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */
949 error = xfs_log_mount(mp, mp->m_logdev_targp,
950 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
951 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
953 cmn_err(CE_WARN, "XFS: log mount failed");
956 } else { /* No log has been defined */
957 cmn_err(CE_WARN, "XFS: no log defined");
958 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
959 error = XFS_ERROR(EFSCORRUPTED);
964 * Get and sanity-check the root inode.
965 * Save the pointer to it in the mount structure.
967 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_ILOCK_EXCL, &rip, 0);
969 cmn_err(CE_WARN, "XFS: failed to read root inode");
977 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
978 cmn_err(CE_WARN, "XFS: corrupted root inode");
979 prdev("Root inode %llu is not a directory",
980 mp->m_ddev_targp, (unsigned long long)rip->i_ino);
981 xfs_iunlock(rip, XFS_ILOCK_EXCL);
982 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
984 error = XFS_ERROR(EFSCORRUPTED);
987 mp->m_rootip = rip; /* save it */
989 xfs_iunlock(rip, XFS_ILOCK_EXCL);
992 * Initialize realtime inode pointers in the mount structure
994 if ((error = xfs_rtmount_inodes(mp))) {
996 * Free up the root inode.
998 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1003 * If fs is not mounted readonly, then update the superblock
1004 * unit and width changes.
1006 if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
1007 xfs_mount_log_sbunit(mp, update_flags);
1010 * Initialise the XFS quota management subsystem for this mount
1012 if ((error = XFS_QM_INIT(mp, "amount, "aflags)))
1016 * Finish recovering the file system. This part needed to be
1017 * delayed until after the root and real-time bitmap inodes
1018 * were consistently read in.
1020 error = xfs_log_mount_finish(mp, mfsi_flags);
1022 cmn_err(CE_WARN, "XFS: log mount finish failed");
1027 * Complete the quota initialisation, post-log-replay component.
1029 if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags)))
1036 * Free up the root inode.
1039 vn_purge(rvp, &vmap);
1041 xfs_log_unmount_dealloc(mp);
1045 for (agno = 0; agno < sbp->sb_agcount; agno++)
1046 if (mp->m_perag[agno].pagb_list)
1047 kmem_free(mp->m_perag[agno].pagb_list,
1048 sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1049 kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1054 xfs_uuid_unmount(mp);
1062 * This flushes out the inodes,dquots and the superblock, unmounts the
1063 * log and makes sure that incore structures are freed.
1066 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1068 struct vfs *vfsp = XFS_MTOVFS(mp);
1069 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1073 xfs_iflush_all(mp, XFS_FLUSH_ALL);
1075 XFS_QM_DQPURGEALL(mp,
1076 XFS_QMOPT_UQUOTA | XFS_QMOPT_GQUOTA | XFS_QMOPT_UMOUNTING);
1079 * Flush out the log synchronously so that we know for sure
1080 * that nothing is pinned. This is important because bflush()
1081 * will skip pinned buffers.
1083 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1085 xfs_binval(mp->m_ddev_targp);
1086 if (mp->m_rtdev_targp) {
1087 xfs_binval(mp->m_rtdev_targp);
1090 xfs_unmountfs_writesb(mp);
1092 xfs_log_unmount(mp); /* Done! No more fs ops. */
1097 * All inodes from this mount point should be freed.
1099 ASSERT(mp->m_inodes == NULL);
1102 * We may have bufs that are in the process of getting written still.
1103 * We must wait for the I/O completion of those. The sync flag here
1104 * does a two pass iteration thru the bufcache.
1106 if (XFS_FORCED_SHUTDOWN(mp)) {
1107 xfs_incore_relse(mp->m_ddev_targp, 0, 1); /* synchronous */
1110 xfs_unmountfs_close(mp, cr);
1111 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1112 xfs_uuid_unmount(mp);
1114 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1116 * clear all error tags on this filesystem
1118 memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
1119 xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
1122 xfs_mount_free(mp, 1);
1127 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1129 if (mp->m_logdev_targp != mp->m_ddev_targp)
1130 xfs_free_buftarg(mp->m_logdev_targp, 1);
1131 if (mp->m_rtdev_targp)
1132 xfs_free_buftarg(mp->m_rtdev_targp, 1);
1133 xfs_free_buftarg(mp->m_ddev_targp, 0);
1137 xfs_unmountfs_writesb(xfs_mount_t *mp)
1144 * skip superblock write if fs is read-only, or
1145 * if we are doing a forced umount.
1147 sbp = xfs_getsb(mp, 0);
1148 if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
1149 XFS_FORCED_SHUTDOWN(mp))) {
1151 * mark shared-readonly if desired
1153 sb = XFS_BUF_TO_SBP(sbp);
1154 if (mp->m_mk_sharedro) {
1155 if (!(sb->sb_flags & XFS_SBF_READONLY))
1156 sb->sb_flags |= XFS_SBF_READONLY;
1157 if (!XFS_SB_VERSION_HASSHARED(sb))
1158 XFS_SB_VERSION_ADDSHARED(sb);
1159 xfs_fs_cmn_err(CE_NOTE, mp,
1160 "Unmounting, marking shared read-only");
1162 XFS_BUF_UNDONE(sbp);
1163 XFS_BUF_UNREAD(sbp);
1164 XFS_BUF_UNDELAYWRITE(sbp);
1166 XFS_BUF_UNASYNC(sbp);
1167 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1168 xfsbdstrat(mp, sbp);
1169 /* Nevermind errors we might get here. */
1170 error = xfs_iowait(sbp);
1172 xfs_ioerror_alert("xfs_unmountfs_writesb",
1173 mp, sbp, XFS_BUF_ADDR(sbp));
1174 if (error && mp->m_mk_sharedro)
1175 xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1182 * xfs_mod_sb() can be used to copy arbitrary changes to the
1183 * in-core superblock into the superblock buffer to be logged.
1184 * It does not provide the higher level of locking that is
1185 * needed to protect the in-core superblock from concurrent
1189 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1202 bp = xfs_trans_getsb(tp, mp, 0);
1203 sbp = XFS_BUF_TO_SBP(bp);
1204 first = sizeof(xfs_sb_t);
1207 /* translate/copy */
1209 xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, ARCH_CONVERT, fields);
1211 /* find modified range */
1213 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1214 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1215 first = xfs_sb_info[f].offset;
1217 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1218 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1219 last = xfs_sb_info[f + 1].offset - 1;
1221 xfs_trans_log_buf(tp, bp, first, last);
1225 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1226 * a delta to a specified field in the in-core superblock. Simply
1227 * switch on the field indicated and apply the delta to that field.
1228 * Fields are not allowed to dip below zero, so if the delta would
1229 * do this do not apply it and return EINVAL.
1231 * The SB_LOCK must be held when this routine is called.
1234 xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field,
1235 int delta, int rsvd)
1237 int scounter; /* short counter for 32 bit fields */
1238 long long lcounter; /* long counter for 64 bit fields */
1239 long long res_used, rem;
1242 * With the in-core superblock spin lock held, switch
1243 * on the indicated field. Apply the delta to the
1244 * proper field. If the fields value would dip below
1245 * 0, then do not apply the delta and return EINVAL.
1248 case XFS_SBS_ICOUNT:
1249 lcounter = (long long)mp->m_sb.sb_icount;
1253 return (XFS_ERROR(EINVAL));
1255 mp->m_sb.sb_icount = lcounter;
1258 lcounter = (long long)mp->m_sb.sb_ifree;
1262 return (XFS_ERROR(EINVAL));
1264 mp->m_sb.sb_ifree = lcounter;
1266 case XFS_SBS_FDBLOCKS:
1268 lcounter = (long long)mp->m_sb.sb_fdblocks;
1269 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1271 if (delta > 0) { /* Putting blocks back */
1272 if (res_used > delta) {
1273 mp->m_resblks_avail += delta;
1275 rem = delta - res_used;
1276 mp->m_resblks_avail = mp->m_resblks;
1279 } else { /* Taking blocks away */
1284 * If were out of blocks, use any available reserved blocks if
1290 lcounter = (long long)mp->m_resblks_avail + delta;
1292 return (XFS_ERROR(ENOSPC));
1294 mp->m_resblks_avail = lcounter;
1296 } else { /* not reserved */
1297 return (XFS_ERROR(ENOSPC));
1302 mp->m_sb.sb_fdblocks = lcounter;
1304 case XFS_SBS_FREXTENTS:
1305 lcounter = (long long)mp->m_sb.sb_frextents;
1308 return (XFS_ERROR(ENOSPC));
1310 mp->m_sb.sb_frextents = lcounter;
1312 case XFS_SBS_DBLOCKS:
1313 lcounter = (long long)mp->m_sb.sb_dblocks;
1317 return (XFS_ERROR(EINVAL));
1319 mp->m_sb.sb_dblocks = lcounter;
1321 case XFS_SBS_AGCOUNT:
1322 scounter = mp->m_sb.sb_agcount;
1326 return (XFS_ERROR(EINVAL));
1328 mp->m_sb.sb_agcount = scounter;
1330 case XFS_SBS_IMAX_PCT:
1331 scounter = mp->m_sb.sb_imax_pct;
1335 return (XFS_ERROR(EINVAL));
1337 mp->m_sb.sb_imax_pct = scounter;
1339 case XFS_SBS_REXTSIZE:
1340 scounter = mp->m_sb.sb_rextsize;
1344 return (XFS_ERROR(EINVAL));
1346 mp->m_sb.sb_rextsize = scounter;
1348 case XFS_SBS_RBMBLOCKS:
1349 scounter = mp->m_sb.sb_rbmblocks;
1353 return (XFS_ERROR(EINVAL));
1355 mp->m_sb.sb_rbmblocks = scounter;
1357 case XFS_SBS_RBLOCKS:
1358 lcounter = (long long)mp->m_sb.sb_rblocks;
1362 return (XFS_ERROR(EINVAL));
1364 mp->m_sb.sb_rblocks = lcounter;
1366 case XFS_SBS_REXTENTS:
1367 lcounter = (long long)mp->m_sb.sb_rextents;
1371 return (XFS_ERROR(EINVAL));
1373 mp->m_sb.sb_rextents = lcounter;
1375 case XFS_SBS_REXTSLOG:
1376 scounter = mp->m_sb.sb_rextslog;
1380 return (XFS_ERROR(EINVAL));
1382 mp->m_sb.sb_rextslog = scounter;
1386 return (XFS_ERROR(EINVAL));
1391 * xfs_mod_incore_sb() is used to change a field in the in-core
1392 * superblock structure by the specified delta. This modification
1393 * is protected by the SB_LOCK. Just use the xfs_mod_incore_sb_unlocked()
1394 * routine to do the work.
1397 xfs_mod_incore_sb(xfs_mount_t *mp, xfs_sb_field_t field, int delta, int rsvd)
1402 s = XFS_SB_LOCK(mp);
1403 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1404 XFS_SB_UNLOCK(mp, s);
1409 * xfs_mod_incore_sb_batch() is used to change more than one field
1410 * in the in-core superblock structure at a time. This modification
1411 * is protected by a lock internal to this module. The fields and
1412 * changes to those fields are specified in the array of xfs_mod_sb
1413 * structures passed in.
1415 * Either all of the specified deltas will be applied or none of
1416 * them will. If any modified field dips below 0, then all modifications
1417 * will be backed out and EINVAL will be returned.
1420 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1427 * Loop through the array of mod structures and apply each
1428 * individually. If any fail, then back out all those
1429 * which have already been applied. Do all of this within
1430 * the scope of the SB_LOCK so that all of the changes will
1433 s = XFS_SB_LOCK(mp);
1435 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1437 * Apply the delta at index n. If it fails, break
1438 * from the loop so we'll fall into the undo loop
1441 status = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1442 msbp->msb_delta, rsvd);
1449 * If we didn't complete the loop above, then back out
1450 * any changes made to the superblock. If you add code
1451 * between the loop above and here, make sure that you
1452 * preserve the value of status. Loop back until
1453 * we step below the beginning of the array. Make sure
1454 * we don't touch anything back there.
1458 while (msbp >= msb) {
1459 status = xfs_mod_incore_sb_unlocked(mp,
1460 msbp->msb_field, -(msbp->msb_delta), rsvd);
1461 ASSERT(status == 0);
1465 XFS_SB_UNLOCK(mp, s);
1470 * xfs_getsb() is called to obtain the buffer for the superblock.
1471 * The buffer is returned locked and read in from disk.
1472 * The buffer should be released with a call to xfs_brelse().
1474 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1475 * the superblock buffer if it can be locked without sleeping.
1476 * If it can't then we'll return NULL.
1485 ASSERT(mp->m_sb_bp != NULL);
1487 if (flags & XFS_BUF_TRYLOCK) {
1488 if (!XFS_BUF_CPSEMA(bp)) {
1492 XFS_BUF_PSEMA(bp, PRIBIO);
1495 ASSERT(XFS_BUF_ISDONE(bp));
1500 * Used to free the superblock along various error paths.
1509 * Use xfs_getsb() so that the buffer will be locked
1510 * when we call xfs_buf_relse().
1512 bp = xfs_getsb(mp, 0);
1513 XFS_BUF_UNMANAGE(bp);
1519 * See if the UUID is unique among mounted XFS filesystems.
1520 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1526 if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1528 "XFS: Filesystem %s has nil UUID - can't mount",
1532 if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1534 "XFS: Filesystem %s has duplicate UUID - can't mount",
1542 * Remove filesystem from the UUID table.
1548 uuid_table_remove(&mp->m_sb.sb_uuid);
1552 * Used to log changes to the superblock unit and width fields which could
1553 * be altered by the mount options. Only the first superblock is updated.
1556 xfs_mount_log_sbunit(
1562 ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
1564 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1565 if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1566 XFS_DEFAULT_LOG_COUNT)) {
1567 xfs_trans_cancel(tp, 0);
1570 xfs_mod_sb(tp, fields);
1571 xfs_trans_commit(tp, 0, NULL);