2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir_sf.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_btree.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_alloc.h"
42 #include "xfs_rtalloc.h"
44 #include "xfs_error.h"
46 #include "xfs_quota.h"
47 #include "xfs_fsops.h"
49 STATIC void xfs_mount_log_sbunit(xfs_mount_t *, __int64_t);
50 STATIC int xfs_uuid_mount(xfs_mount_t *);
51 STATIC void xfs_uuid_unmount(xfs_mount_t *mp);
52 STATIC void xfs_unmountfs_wait(xfs_mount_t *);
56 short type; /* 0 = integer
57 * 1 = binary / string (no translation)
60 { offsetof(xfs_sb_t, sb_magicnum), 0 },
61 { offsetof(xfs_sb_t, sb_blocksize), 0 },
62 { offsetof(xfs_sb_t, sb_dblocks), 0 },
63 { offsetof(xfs_sb_t, sb_rblocks), 0 },
64 { offsetof(xfs_sb_t, sb_rextents), 0 },
65 { offsetof(xfs_sb_t, sb_uuid), 1 },
66 { offsetof(xfs_sb_t, sb_logstart), 0 },
67 { offsetof(xfs_sb_t, sb_rootino), 0 },
68 { offsetof(xfs_sb_t, sb_rbmino), 0 },
69 { offsetof(xfs_sb_t, sb_rsumino), 0 },
70 { offsetof(xfs_sb_t, sb_rextsize), 0 },
71 { offsetof(xfs_sb_t, sb_agblocks), 0 },
72 { offsetof(xfs_sb_t, sb_agcount), 0 },
73 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
74 { offsetof(xfs_sb_t, sb_logblocks), 0 },
75 { offsetof(xfs_sb_t, sb_versionnum), 0 },
76 { offsetof(xfs_sb_t, sb_sectsize), 0 },
77 { offsetof(xfs_sb_t, sb_inodesize), 0 },
78 { offsetof(xfs_sb_t, sb_inopblock), 0 },
79 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
80 { offsetof(xfs_sb_t, sb_blocklog), 0 },
81 { offsetof(xfs_sb_t, sb_sectlog), 0 },
82 { offsetof(xfs_sb_t, sb_inodelog), 0 },
83 { offsetof(xfs_sb_t, sb_inopblog), 0 },
84 { offsetof(xfs_sb_t, sb_agblklog), 0 },
85 { offsetof(xfs_sb_t, sb_rextslog), 0 },
86 { offsetof(xfs_sb_t, sb_inprogress), 0 },
87 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
88 { offsetof(xfs_sb_t, sb_icount), 0 },
89 { offsetof(xfs_sb_t, sb_ifree), 0 },
90 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
91 { offsetof(xfs_sb_t, sb_frextents), 0 },
92 { offsetof(xfs_sb_t, sb_uquotino), 0 },
93 { offsetof(xfs_sb_t, sb_gquotino), 0 },
94 { offsetof(xfs_sb_t, sb_qflags), 0 },
95 { offsetof(xfs_sb_t, sb_flags), 0 },
96 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
97 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
98 { offsetof(xfs_sb_t, sb_unit), 0 },
99 { offsetof(xfs_sb_t, sb_width), 0 },
100 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
101 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
102 { offsetof(xfs_sb_t, sb_logsectsize),0 },
103 { offsetof(xfs_sb_t, sb_logsunit), 0 },
104 { offsetof(xfs_sb_t, sb_features2), 0 },
105 { sizeof(xfs_sb_t), 0 }
109 * Return a pointer to an initialized xfs_mount structure.
116 mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
118 AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail");
119 spinlock_init(&mp->m_sb_lock, "xfs_sb");
120 mutex_init(&mp->m_ilock);
121 initnsema(&mp->m_growlock, 1, "xfs_grow");
123 * Initialize the AIL.
125 xfs_trans_ail_init(mp);
127 atomic_set(&mp->m_active_trans, 0);
133 * Free up the resources associated with a mount structure. Assume that
134 * the structure was initially zeroed, so we can tell which fields got
150 for (agno = 0; agno < mp->m_maxagi; agno++)
151 if (mp->m_perag[agno].pagb_list)
152 kmem_free(mp->m_perag[agno].pagb_list,
153 sizeof(xfs_perag_busy_t) *
155 kmem_free(mp->m_perag,
156 sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
159 AIL_LOCK_DESTROY(&mp->m_ail_lock);
160 spinlock_destroy(&mp->m_sb_lock);
161 mutex_destroy(&mp->m_ilock);
162 freesema(&mp->m_growlock);
166 if (mp->m_fsname != NULL)
167 kmem_free(mp->m_fsname, mp->m_fsname_len);
168 if (mp->m_rtname != NULL)
169 kmem_free(mp->m_rtname, strlen(mp->m_rtname) + 1);
170 if (mp->m_logname != NULL)
171 kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
174 struct vfs *vfsp = XFS_MTOVFS(mp);
176 bhv_remove_all_vfsops(vfsp, 0);
177 VFS_REMOVEBHV(vfsp, &mp->m_bhv);
180 kmem_free(mp, sizeof(xfs_mount_t));
185 * Check the validity of the SB found.
188 xfs_mount_validate_sb(
193 * If the log device and data device have the
194 * same device number, the log is internal.
195 * Consequently, the sb_logstart should be non-zero. If
196 * we have a zero sb_logstart in this case, we may be trying to mount
197 * a volume filesystem in a non-volume manner.
199 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
200 cmn_err(CE_WARN, "XFS: bad magic number");
201 return XFS_ERROR(EWRONGFS);
204 if (!XFS_SB_GOOD_VERSION(sbp)) {
205 cmn_err(CE_WARN, "XFS: bad version");
206 return XFS_ERROR(EWRONGFS);
210 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
212 "XFS: filesystem is marked as having an external log; "
213 "specify logdev on the\nmount command line.");
214 XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(1)",
215 XFS_ERRLEVEL_HIGH, mp, sbp);
216 return XFS_ERROR(EFSCORRUPTED);
220 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
222 "XFS: filesystem is marked as having an internal log; "
223 "don't specify logdev on\nthe mount command line.");
224 XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(2)",
225 XFS_ERRLEVEL_HIGH, mp, sbp);
226 return XFS_ERROR(EFSCORRUPTED);
230 * More sanity checking. These were stolen directly from
234 sbp->sb_agcount <= 0 ||
235 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
236 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
237 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
238 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
239 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
240 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
241 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
242 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
243 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
244 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
245 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
246 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
247 sbp->sb_imax_pct > 100)) {
248 cmn_err(CE_WARN, "XFS: SB sanity check 1 failed");
249 XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(3)",
250 XFS_ERRLEVEL_LOW, mp, sbp);
251 return XFS_ERROR(EFSCORRUPTED);
255 * Sanity check AG count, size fields against data size field
258 sbp->sb_dblocks == 0 ||
260 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
261 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
262 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
263 cmn_err(CE_WARN, "XFS: SB sanity check 2 failed");
264 XFS_ERROR_REPORT("xfs_mount_validate_sb(4)",
265 XFS_ERRLEVEL_LOW, mp);
266 return XFS_ERROR(EFSCORRUPTED);
269 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
270 ASSERT(sbp->sb_blocklog >= BBSHIFT);
272 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
274 (sbp->sb_dblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX ||
275 (sbp->sb_rblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX)) {
276 #else /* Limited by UINT_MAX of sectors */
278 (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX ||
279 (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) {
282 "XFS: File system is too large to be mounted on this system.");
283 return XFS_ERROR(E2BIG);
286 if (unlikely(sbp->sb_inprogress)) {
287 cmn_err(CE_WARN, "XFS: file system busy");
288 XFS_ERROR_REPORT("xfs_mount_validate_sb(5)",
289 XFS_ERRLEVEL_LOW, mp);
290 return XFS_ERROR(EFSCORRUPTED);
294 * Version 1 directory format has never worked on Linux.
296 if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp))) {
298 "XFS: Attempted to mount file system using version 1 directory format");
299 return XFS_ERROR(ENOSYS);
303 * Until this is fixed only page-sized or smaller data blocks work.
305 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
307 "XFS: Attempted to mount file system with blocksize %d bytes",
310 "XFS: Only page-sized (%ld) or less blocksizes currently work.",
312 return XFS_ERROR(ENOSYS);
319 xfs_initialize_perag(
322 xfs_agnumber_t agcount)
324 xfs_agnumber_t index, max_metadata;
328 xfs_sb_t *sbp = &mp->m_sb;
329 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
331 /* Check to see if the filesystem can overflow 32 bit inodes */
332 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
333 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
335 /* Clear the mount flag if no inode can overflow 32 bits
336 * on this filesystem, or if specifically requested..
338 if ((vfs->vfs_flag & VFS_32BITINODES) && ino > max_inum) {
339 mp->m_flags |= XFS_MOUNT_32BITINODES;
341 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
344 /* If we can overflow then setup the ag headers accordingly */
345 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
346 /* Calculate how much should be reserved for inodes to
347 * meet the max inode percentage.
349 if (mp->m_maxicount) {
352 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
354 icount += sbp->sb_agblocks - 1;
355 do_div(icount, sbp->sb_agblocks);
356 max_metadata = icount;
358 max_metadata = agcount;
360 for (index = 0; index < agcount; index++) {
361 ino = XFS_AGINO_TO_INO(mp, index, agino);
362 if (ino > max_inum) {
367 /* This ag is prefered for inodes */
368 pag = &mp->m_perag[index];
369 pag->pagi_inodeok = 1;
370 if (index < max_metadata)
371 pag->pagf_metadata = 1;
374 /* Setup default behavior for smaller filesystems */
375 for (index = 0; index < agcount; index++) {
376 pag = &mp->m_perag[index];
377 pag->pagi_inodeok = 1;
386 * data - on disk version of sb
388 * dir - conversion direction: <0 - convert sb to buf
389 * >0 - convert buf to sb
390 * fields - which fields to copy (bitmask)
411 buf_ptr = (xfs_caddr_t)data;
412 mem_ptr = (xfs_caddr_t)sb;
415 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
416 first = xfs_sb_info[f].offset;
417 size = xfs_sb_info[f + 1].offset - first;
419 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
421 if (size == 1 || xfs_sb_info[f].type == 1) {
423 memcpy(mem_ptr + first, buf_ptr + first, size);
425 memcpy(buf_ptr + first, mem_ptr + first, size);
430 INT_XLATE(*(__uint16_t*)(buf_ptr+first),
431 *(__uint16_t*)(mem_ptr+first),
435 INT_XLATE(*(__uint32_t*)(buf_ptr+first),
436 *(__uint32_t*)(mem_ptr+first),
440 INT_XLATE(*(__uint64_t*)(buf_ptr+first),
441 *(__uint64_t*)(mem_ptr+first), dir, ARCH_CONVERT);
448 fields &= ~(1LL << f);
455 * Does the initial read of the superblock.
458 xfs_readsb(xfs_mount_t *mp)
460 unsigned int sector_size;
461 unsigned int extra_flags;
466 ASSERT(mp->m_sb_bp == NULL);
467 ASSERT(mp->m_ddev_targp != NULL);
470 * Allocate a (locked) buffer to hold the superblock.
471 * This will be kept around at all times to optimize
472 * access to the superblock.
474 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
475 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
477 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
478 BTOBB(sector_size), extra_flags);
479 if (!bp || XFS_BUF_ISERROR(bp)) {
480 cmn_err(CE_WARN, "XFS: SB read failed");
481 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
484 ASSERT(XFS_BUF_ISBUSY(bp));
485 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
488 * Initialize the mount structure from the superblock.
489 * But first do some basic consistency checking.
491 sbp = XFS_BUF_TO_SBP(bp);
492 xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1, 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) -
580 XFS_BMDR_SPACE_CALC(MINABTPTRS);
585 mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
590 ASSERT(mp->m_attroffset < XFS_LITINO(mp));
592 for (i = 0; i < 2; i++) {
593 mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
595 mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
598 for (i = 0; i < 2; i++) {
599 mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
601 mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
604 for (i = 0; i < 2; i++) {
605 mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
607 mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
611 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
612 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
614 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
619 * This function does the following on an initial mount of a file system:
620 * - reads the superblock from disk and init the mount struct
621 * - if we're a 32-bit kernel, do a size check on the superblock
622 * so we don't mount terabyte filesystems
623 * - init mount struct realtime fields
624 * - allocate inode hash table for fs
625 * - init directory manager
626 * - perform recovery and init the log manager
635 xfs_sb_t *sbp = &(mp->m_sb);
638 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);
685 xfs_fs_cmn_err(CE_WARN, mp,
686 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
687 mp->m_dalign, mp->m_swidth,
692 } else if (mp->m_dalign) {
693 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
695 if (mp->m_flags & XFS_MOUNT_RETERR) {
696 xfs_fs_cmn_err(CE_WARN, mp,
697 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
700 error = XFS_ERROR(EINVAL);
708 * Update superblock with new values
711 if (XFS_SB_VERSION_HASDALIGN(sbp)) {
712 if (sbp->sb_unit != mp->m_dalign) {
713 sbp->sb_unit = mp->m_dalign;
714 update_flags |= XFS_SB_UNIT;
716 if (sbp->sb_width != mp->m_swidth) {
717 sbp->sb_width = mp->m_swidth;
718 update_flags |= XFS_SB_WIDTH;
721 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
722 XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
723 mp->m_dalign = sbp->sb_unit;
724 mp->m_swidth = sbp->sb_width;
727 xfs_alloc_compute_maxlevels(mp);
728 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
729 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
730 xfs_ialloc_compute_maxlevels(mp);
732 if (sbp->sb_imax_pct) {
735 /* Make sure the maximum inode count is a multiple of the
736 * units we allocate inodes in.
739 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
741 do_div(icount, mp->m_ialloc_blks);
742 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
747 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
750 * XFS uses the uuid from the superblock as the unique
751 * identifier for fsid. We can not use the uuid from the volume
752 * since a single partition filesystem is identical to a single
753 * partition volume/filesystem.
755 if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
756 (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
757 if (xfs_uuid_mount(mp)) {
758 error = XFS_ERROR(EINVAL);
762 ret64 = uuid_hash64(&sbp->sb_uuid);
763 memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
767 * Set the default minimum read and write sizes unless
768 * already specified in a mount option.
769 * We use smaller I/O sizes when the file system
770 * is being used for NFS service (wsync mount option).
772 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
773 if (mp->m_flags & XFS_MOUNT_WSYNC) {
774 readio_log = XFS_WSYNC_READIO_LOG;
775 writeio_log = XFS_WSYNC_WRITEIO_LOG;
777 readio_log = XFS_READIO_LOG_LARGE;
778 writeio_log = XFS_WRITEIO_LOG_LARGE;
781 readio_log = mp->m_readio_log;
782 writeio_log = mp->m_writeio_log;
786 * Set the number of readahead buffers to use based on
787 * physical memory size.
789 if (xfs_physmem <= 4096) /* <= 16MB */
790 mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
791 else if (xfs_physmem <= 8192) /* <= 32MB */
792 mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
794 mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
795 if (sbp->sb_blocklog > readio_log) {
796 mp->m_readio_log = sbp->sb_blocklog;
798 mp->m_readio_log = readio_log;
800 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
801 if (sbp->sb_blocklog > writeio_log) {
802 mp->m_writeio_log = sbp->sb_blocklog;
804 mp->m_writeio_log = writeio_log;
806 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
809 * Set the inode cluster size based on the physical memory
810 * size. This may still be overridden by the file system
811 * block size if it is larger than the chosen cluster size.
813 if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
814 mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
816 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
819 * Set whether we're using inode alignment.
821 if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
822 mp->m_sb.sb_inoalignmt >=
823 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
824 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
826 mp->m_inoalign_mask = 0;
828 * If we are using stripe alignment, check whether
829 * the stripe unit is a multiple of the inode alignment
831 if (mp->m_dalign && mp->m_inoalign_mask &&
832 !(mp->m_dalign & mp->m_inoalign_mask))
833 mp->m_sinoalign = mp->m_dalign;
837 * Check that the data (and log if separate) are an ok size.
839 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
840 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
841 cmn_err(CE_WARN, "XFS: size check 1 failed");
842 error = XFS_ERROR(E2BIG);
845 error = xfs_read_buf(mp, mp->m_ddev_targp,
846 d - XFS_FSS_TO_BB(mp, 1),
847 XFS_FSS_TO_BB(mp, 1), 0, &bp);
851 cmn_err(CE_WARN, "XFS: size check 2 failed");
852 if (error == ENOSPC) {
853 error = XFS_ERROR(E2BIG);
858 if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
859 mp->m_logdev_targp != mp->m_ddev_targp) {
860 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
861 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
862 cmn_err(CE_WARN, "XFS: size check 3 failed");
863 error = XFS_ERROR(E2BIG);
866 error = xfs_read_buf(mp, mp->m_logdev_targp,
867 d - XFS_FSB_TO_BB(mp, 1),
868 XFS_FSB_TO_BB(mp, 1), 0, &bp);
872 cmn_err(CE_WARN, "XFS: size check 3 failed");
873 if (error == ENOSPC) {
874 error = XFS_ERROR(E2BIG);
881 * Initialize realtime fields in the mount structure
883 if ((error = xfs_rtmount_init(mp))) {
884 cmn_err(CE_WARN, "XFS: RT mount failed");
889 * For client case we are done now
891 if (mfsi_flags & XFS_MFSI_CLIENT) {
896 * Copies the low order bits of the timestamp and the randomly
897 * set "sequence" number out of a UUID.
899 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
902 * The vfs structure needs to have a file system independent
903 * way of checking for the invariant file system ID. Since it
904 * can't look at mount structures it has a pointer to the data
905 * in the mount structure.
907 * File systems that don't support user level file handles (i.e.
908 * all of them except for XFS) will leave vfs_altfsid as NULL.
910 vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
911 mp->m_dmevmask = 0; /* not persistent; set after each mount */
914 * Select the right directory manager.
917 XFS_SB_VERSION_HASDIRV2(&mp->m_sb) ?
922 * Initialize directory manager's entries.
927 * Initialize the attribute manager's entries.
929 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
932 * Initialize the precomputed transaction reservations values.
937 * Allocate and initialize the inode hash table for this
944 * Allocate and initialize the per-ag data.
946 init_rwsem(&mp->m_peraglock);
948 kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
950 mp->m_maxagi = xfs_initialize_perag(vfsp, mp, sbp->sb_agcount);
953 * log's mount-time initialization. Perform 1st part recovery if needed
955 if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */
956 error = xfs_log_mount(mp, mp->m_logdev_targp,
957 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
958 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
960 cmn_err(CE_WARN, "XFS: log mount failed");
963 } else { /* No log has been defined */
964 cmn_err(CE_WARN, "XFS: no log defined");
965 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
966 error = XFS_ERROR(EFSCORRUPTED);
971 * Get and sanity-check the root inode.
972 * Save the pointer to it in the mount structure.
974 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
976 cmn_err(CE_WARN, "XFS: failed to read root inode");
983 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
984 cmn_err(CE_WARN, "XFS: corrupted root inode");
985 prdev("Root inode %llu is not a directory",
986 mp->m_ddev_targp, (unsigned long long)rip->i_ino);
987 xfs_iunlock(rip, XFS_ILOCK_EXCL);
988 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
990 error = XFS_ERROR(EFSCORRUPTED);
993 mp->m_rootip = rip; /* save it */
995 xfs_iunlock(rip, XFS_ILOCK_EXCL);
998 * Initialize realtime inode pointers in the mount structure
1000 if ((error = xfs_rtmount_inodes(mp))) {
1002 * Free up the root inode.
1004 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1009 * If fs is not mounted readonly, then update the superblock
1010 * unit and width changes.
1012 if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
1013 xfs_mount_log_sbunit(mp, update_flags);
1016 * Initialise the XFS quota management subsystem for this mount
1018 if ((error = XFS_QM_INIT(mp, "amount, "aflags)))
1022 * Finish recovering the file system. This part needed to be
1023 * delayed until after the root and real-time bitmap inodes
1024 * were consistently read in.
1026 error = xfs_log_mount_finish(mp, mfsi_flags);
1028 cmn_err(CE_WARN, "XFS: log mount finish failed");
1033 * Complete the quota initialisation, post-log-replay component.
1035 if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags)))
1042 * Free up the root inode.
1046 xfs_log_unmount_dealloc(mp);
1050 for (agno = 0; agno < sbp->sb_agcount; agno++)
1051 if (mp->m_perag[agno].pagb_list)
1052 kmem_free(mp->m_perag[agno].pagb_list,
1053 sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1054 kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1059 xfs_uuid_unmount(mp);
1067 * This flushes out the inodes,dquots and the superblock, unmounts the
1068 * log and makes sure that incore structures are freed.
1071 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1073 struct vfs *vfsp = XFS_MTOVFS(mp);
1074 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1080 XFS_QM_DQPURGEALL(mp,
1081 XFS_QMOPT_UQUOTA | XFS_QMOPT_GQUOTA | XFS_QMOPT_UMOUNTING);
1084 * Flush out the log synchronously so that we know for sure
1085 * that nothing is pinned. This is important because bflush()
1086 * will skip pinned buffers.
1088 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1090 xfs_binval(mp->m_ddev_targp);
1091 if (mp->m_rtdev_targp) {
1092 xfs_binval(mp->m_rtdev_targp);
1095 xfs_unmountfs_writesb(mp);
1097 xfs_unmountfs_wait(mp); /* wait for async bufs */
1099 xfs_log_unmount(mp); /* Done! No more fs ops. */
1104 * All inodes from this mount point should be freed.
1106 ASSERT(mp->m_inodes == NULL);
1108 xfs_unmountfs_close(mp, cr);
1109 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1110 xfs_uuid_unmount(mp);
1112 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1114 * clear all error tags on this filesystem
1116 memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
1117 xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
1120 xfs_mount_free(mp, 1);
1125 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1127 if (mp->m_logdev_targp != mp->m_ddev_targp)
1128 xfs_free_buftarg(mp->m_logdev_targp, 1);
1129 if (mp->m_rtdev_targp)
1130 xfs_free_buftarg(mp->m_rtdev_targp, 1);
1131 xfs_free_buftarg(mp->m_ddev_targp, 0);
1135 xfs_unmountfs_wait(xfs_mount_t *mp)
1137 if (mp->m_logdev_targp != mp->m_ddev_targp)
1138 xfs_wait_buftarg(mp->m_logdev_targp);
1139 if (mp->m_rtdev_targp)
1140 xfs_wait_buftarg(mp->m_rtdev_targp);
1141 xfs_wait_buftarg(mp->m_ddev_targp);
1145 xfs_unmountfs_writesb(xfs_mount_t *mp)
1152 * skip superblock write if fs is read-only, or
1153 * if we are doing a forced umount.
1155 sbp = xfs_getsb(mp, 0);
1156 if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
1157 XFS_FORCED_SHUTDOWN(mp))) {
1159 * mark shared-readonly if desired
1161 sb = XFS_BUF_TO_SBP(sbp);
1162 if (mp->m_mk_sharedro) {
1163 if (!(sb->sb_flags & XFS_SBF_READONLY))
1164 sb->sb_flags |= XFS_SBF_READONLY;
1165 if (!XFS_SB_VERSION_HASSHARED(sb))
1166 XFS_SB_VERSION_ADDSHARED(sb);
1167 xfs_fs_cmn_err(CE_NOTE, mp,
1168 "Unmounting, marking shared read-only");
1170 XFS_BUF_UNDONE(sbp);
1171 XFS_BUF_UNREAD(sbp);
1172 XFS_BUF_UNDELAYWRITE(sbp);
1174 XFS_BUF_UNASYNC(sbp);
1175 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1176 xfsbdstrat(mp, sbp);
1177 /* Nevermind errors we might get here. */
1178 error = xfs_iowait(sbp);
1180 xfs_ioerror_alert("xfs_unmountfs_writesb",
1181 mp, sbp, XFS_BUF_ADDR(sbp));
1182 if (error && mp->m_mk_sharedro)
1183 xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1190 * xfs_mod_sb() can be used to copy arbitrary changes to the
1191 * in-core superblock into the superblock buffer to be logged.
1192 * It does not provide the higher level of locking that is
1193 * needed to protect the in-core superblock from concurrent
1197 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1210 bp = xfs_trans_getsb(tp, mp, 0);
1211 sbp = XFS_BUF_TO_SBP(bp);
1212 first = sizeof(xfs_sb_t);
1215 /* translate/copy */
1217 xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, fields);
1219 /* find modified range */
1221 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1222 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1223 first = xfs_sb_info[f].offset;
1225 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1226 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1227 last = xfs_sb_info[f + 1].offset - 1;
1229 xfs_trans_log_buf(tp, bp, first, last);
1233 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1234 * a delta to a specified field in the in-core superblock. Simply
1235 * switch on the field indicated and apply the delta to that field.
1236 * Fields are not allowed to dip below zero, so if the delta would
1237 * do this do not apply it and return EINVAL.
1239 * The SB_LOCK must be held when this routine is called.
1242 xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field,
1243 int delta, int rsvd)
1245 int scounter; /* short counter for 32 bit fields */
1246 long long lcounter; /* long counter for 64 bit fields */
1247 long long res_used, rem;
1250 * With the in-core superblock spin lock held, switch
1251 * on the indicated field. Apply the delta to the
1252 * proper field. If the fields value would dip below
1253 * 0, then do not apply the delta and return EINVAL.
1256 case XFS_SBS_ICOUNT:
1257 lcounter = (long long)mp->m_sb.sb_icount;
1261 return (XFS_ERROR(EINVAL));
1263 mp->m_sb.sb_icount = lcounter;
1266 lcounter = (long long)mp->m_sb.sb_ifree;
1270 return (XFS_ERROR(EINVAL));
1272 mp->m_sb.sb_ifree = lcounter;
1274 case XFS_SBS_FDBLOCKS:
1276 lcounter = (long long)mp->m_sb.sb_fdblocks;
1277 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1279 if (delta > 0) { /* Putting blocks back */
1280 if (res_used > delta) {
1281 mp->m_resblks_avail += delta;
1283 rem = delta - res_used;
1284 mp->m_resblks_avail = mp->m_resblks;
1287 } else { /* Taking blocks away */
1292 * If were out of blocks, use any available reserved blocks if
1298 lcounter = (long long)mp->m_resblks_avail + delta;
1300 return (XFS_ERROR(ENOSPC));
1302 mp->m_resblks_avail = lcounter;
1304 } else { /* not reserved */
1305 return (XFS_ERROR(ENOSPC));
1310 mp->m_sb.sb_fdblocks = lcounter;
1312 case XFS_SBS_FREXTENTS:
1313 lcounter = (long long)mp->m_sb.sb_frextents;
1316 return (XFS_ERROR(ENOSPC));
1318 mp->m_sb.sb_frextents = lcounter;
1320 case XFS_SBS_DBLOCKS:
1321 lcounter = (long long)mp->m_sb.sb_dblocks;
1325 return (XFS_ERROR(EINVAL));
1327 mp->m_sb.sb_dblocks = lcounter;
1329 case XFS_SBS_AGCOUNT:
1330 scounter = mp->m_sb.sb_agcount;
1334 return (XFS_ERROR(EINVAL));
1336 mp->m_sb.sb_agcount = scounter;
1338 case XFS_SBS_IMAX_PCT:
1339 scounter = mp->m_sb.sb_imax_pct;
1343 return (XFS_ERROR(EINVAL));
1345 mp->m_sb.sb_imax_pct = scounter;
1347 case XFS_SBS_REXTSIZE:
1348 scounter = mp->m_sb.sb_rextsize;
1352 return (XFS_ERROR(EINVAL));
1354 mp->m_sb.sb_rextsize = scounter;
1356 case XFS_SBS_RBMBLOCKS:
1357 scounter = mp->m_sb.sb_rbmblocks;
1361 return (XFS_ERROR(EINVAL));
1363 mp->m_sb.sb_rbmblocks = scounter;
1365 case XFS_SBS_RBLOCKS:
1366 lcounter = (long long)mp->m_sb.sb_rblocks;
1370 return (XFS_ERROR(EINVAL));
1372 mp->m_sb.sb_rblocks = lcounter;
1374 case XFS_SBS_REXTENTS:
1375 lcounter = (long long)mp->m_sb.sb_rextents;
1379 return (XFS_ERROR(EINVAL));
1381 mp->m_sb.sb_rextents = lcounter;
1383 case XFS_SBS_REXTSLOG:
1384 scounter = mp->m_sb.sb_rextslog;
1388 return (XFS_ERROR(EINVAL));
1390 mp->m_sb.sb_rextslog = scounter;
1394 return (XFS_ERROR(EINVAL));
1399 * xfs_mod_incore_sb() is used to change a field in the in-core
1400 * superblock structure by the specified delta. This modification
1401 * is protected by the SB_LOCK. Just use the xfs_mod_incore_sb_unlocked()
1402 * routine to do the work.
1405 xfs_mod_incore_sb(xfs_mount_t *mp, xfs_sb_field_t field, int delta, int rsvd)
1410 s = XFS_SB_LOCK(mp);
1411 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1412 XFS_SB_UNLOCK(mp, s);
1417 * xfs_mod_incore_sb_batch() is used to change more than one field
1418 * in the in-core superblock structure at a time. This modification
1419 * is protected by a lock internal to this module. The fields and
1420 * changes to those fields are specified in the array of xfs_mod_sb
1421 * structures passed in.
1423 * Either all of the specified deltas will be applied or none of
1424 * them will. If any modified field dips below 0, then all modifications
1425 * will be backed out and EINVAL will be returned.
1428 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1435 * Loop through the array of mod structures and apply each
1436 * individually. If any fail, then back out all those
1437 * which have already been applied. Do all of this within
1438 * the scope of the SB_LOCK so that all of the changes will
1441 s = XFS_SB_LOCK(mp);
1443 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1445 * Apply the delta at index n. If it fails, break
1446 * from the loop so we'll fall into the undo loop
1449 status = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1450 msbp->msb_delta, rsvd);
1457 * If we didn't complete the loop above, then back out
1458 * any changes made to the superblock. If you add code
1459 * between the loop above and here, make sure that you
1460 * preserve the value of status. Loop back until
1461 * we step below the beginning of the array. Make sure
1462 * we don't touch anything back there.
1466 while (msbp >= msb) {
1467 status = xfs_mod_incore_sb_unlocked(mp,
1468 msbp->msb_field, -(msbp->msb_delta), rsvd);
1469 ASSERT(status == 0);
1473 XFS_SB_UNLOCK(mp, s);
1478 * xfs_getsb() is called to obtain the buffer for the superblock.
1479 * The buffer is returned locked and read in from disk.
1480 * The buffer should be released with a call to xfs_brelse().
1482 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1483 * the superblock buffer if it can be locked without sleeping.
1484 * If it can't then we'll return NULL.
1493 ASSERT(mp->m_sb_bp != NULL);
1495 if (flags & XFS_BUF_TRYLOCK) {
1496 if (!XFS_BUF_CPSEMA(bp)) {
1500 XFS_BUF_PSEMA(bp, PRIBIO);
1503 ASSERT(XFS_BUF_ISDONE(bp));
1508 * Used to free the superblock along various error paths.
1517 * Use xfs_getsb() so that the buffer will be locked
1518 * when we call xfs_buf_relse().
1520 bp = xfs_getsb(mp, 0);
1521 XFS_BUF_UNMANAGE(bp);
1527 * See if the UUID is unique among mounted XFS filesystems.
1528 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1534 if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1536 "XFS: Filesystem %s has nil UUID - can't mount",
1540 if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1542 "XFS: Filesystem %s has duplicate UUID - can't mount",
1550 * Remove filesystem from the UUID table.
1556 uuid_table_remove(&mp->m_sb.sb_uuid);
1560 * Used to log changes to the superblock unit and width fields which could
1561 * be altered by the mount options. Only the first superblock is updated.
1564 xfs_mount_log_sbunit(
1570 ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
1572 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1573 if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1574 XFS_DEFAULT_LOG_COUNT)) {
1575 xfs_trans_cancel(tp, 0);
1578 xfs_mod_sb(tp, fields);
1579 xfs_trans_commit(tp, 0, NULL);