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"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_fsops.h"
46 #include "xfs_utils.h"
47 #include "xfs_trace.h"
50 STATIC void xfs_unmountfs_wait(xfs_mount_t *);
54 STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
56 STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
58 STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
60 STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
64 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
65 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
66 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
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 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
122 { sizeof(xfs_sb_t), 0 }
125 static DEFINE_MUTEX(xfs_uuid_table_mutex);
126 static int xfs_uuid_table_size;
127 static uuid_t *xfs_uuid_table;
130 * See if the UUID is unique among mounted XFS filesystems.
131 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
135 struct xfs_mount *mp)
137 uuid_t *uuid = &mp->m_sb.sb_uuid;
140 if (mp->m_flags & XFS_MOUNT_NOUUID)
143 if (uuid_is_nil(uuid)) {
145 "XFS: Filesystem %s has nil UUID - can't mount",
147 return XFS_ERROR(EINVAL);
150 mutex_lock(&xfs_uuid_table_mutex);
151 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
152 if (uuid_is_nil(&xfs_uuid_table[i])) {
156 if (uuid_equal(uuid, &xfs_uuid_table[i]))
161 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
162 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
163 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
165 hole = xfs_uuid_table_size++;
167 xfs_uuid_table[hole] = *uuid;
168 mutex_unlock(&xfs_uuid_table_mutex);
173 mutex_unlock(&xfs_uuid_table_mutex);
174 cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount",
176 return XFS_ERROR(EINVAL);
181 struct xfs_mount *mp)
183 uuid_t *uuid = &mp->m_sb.sb_uuid;
186 if (mp->m_flags & XFS_MOUNT_NOUUID)
189 mutex_lock(&xfs_uuid_table_mutex);
190 for (i = 0; i < xfs_uuid_table_size; i++) {
191 if (uuid_is_nil(&xfs_uuid_table[i]))
193 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
195 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
198 ASSERT(i < xfs_uuid_table_size);
199 mutex_unlock(&xfs_uuid_table_mutex);
204 * Free up the resources associated with a mount structure. Assume that
205 * the structure was initially zeroed, so we can tell which fields got
213 struct xfs_perag *pag;
215 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
216 spin_lock(&mp->m_perag_lock);
217 pag = radix_tree_delete(&mp->m_perag_tree, agno);
218 spin_unlock(&mp->m_perag_lock);
220 kmem_free(pag->pagb_list);
226 * Check size of device based on the (data/realtime) block count.
227 * Note: this check is used by the growfs code as well as mount.
230 xfs_sb_validate_fsb_count(
234 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
235 ASSERT(sbp->sb_blocklog >= BBSHIFT);
237 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
238 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
240 #else /* Limited by UINT_MAX of sectors */
241 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
248 * Check the validity of the SB found.
251 xfs_mount_validate_sb(
257 * If the log device and data device have the
258 * same device number, the log is internal.
259 * Consequently, the sb_logstart should be non-zero. If
260 * we have a zero sb_logstart in this case, we may be trying to mount
261 * a volume filesystem in a non-volume manner.
263 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
264 xfs_fs_mount_cmn_err(flags, "bad magic number");
265 return XFS_ERROR(EWRONGFS);
268 if (!xfs_sb_good_version(sbp)) {
269 xfs_fs_mount_cmn_err(flags, "bad version");
270 return XFS_ERROR(EWRONGFS);
274 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
275 xfs_fs_mount_cmn_err(flags,
276 "filesystem is marked as having an external log; "
277 "specify logdev on the\nmount command line.");
278 return XFS_ERROR(EINVAL);
282 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
283 xfs_fs_mount_cmn_err(flags,
284 "filesystem is marked as having an internal log; "
285 "do not specify logdev on\nthe mount command line.");
286 return XFS_ERROR(EINVAL);
290 * More sanity checking. These were stolen directly from
294 sbp->sb_agcount <= 0 ||
295 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
296 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
297 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
298 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
299 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
300 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
301 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
302 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
303 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
304 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
305 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
306 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
307 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
308 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
309 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
310 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
311 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
312 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
313 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
314 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
315 return XFS_ERROR(EFSCORRUPTED);
319 * Sanity check AG count, size fields against data size field
322 sbp->sb_dblocks == 0 ||
324 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
325 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
326 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
327 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
328 return XFS_ERROR(EFSCORRUPTED);
332 * Until this is fixed only page-sized or smaller data blocks work.
334 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
335 xfs_fs_mount_cmn_err(flags,
336 "file system with blocksize %d bytes",
338 xfs_fs_mount_cmn_err(flags,
339 "only pagesize (%ld) or less will currently work.",
341 return XFS_ERROR(ENOSYS);
345 * Currently only very few inode sizes are supported.
347 switch (sbp->sb_inodesize) {
354 xfs_fs_mount_cmn_err(flags,
355 "inode size of %d bytes not supported",
357 return XFS_ERROR(ENOSYS);
360 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
361 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
362 xfs_fs_mount_cmn_err(flags,
363 "file system too large to be mounted on this system.");
364 return XFS_ERROR(E2BIG);
367 if (unlikely(sbp->sb_inprogress)) {
368 xfs_fs_mount_cmn_err(flags, "file system busy");
369 return XFS_ERROR(EFSCORRUPTED);
373 * Version 1 directory format has never worked on Linux.
375 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
376 xfs_fs_mount_cmn_err(flags,
377 "file system using version 1 directory format");
378 return XFS_ERROR(ENOSYS);
385 xfs_initialize_perag_icache(
388 if (!pag->pag_ici_init) {
389 rwlock_init(&pag->pag_ici_lock);
390 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
391 pag->pag_ici_init = 1;
396 xfs_initialize_perag(
398 xfs_agnumber_t agcount,
399 xfs_agnumber_t *maxagi)
401 xfs_agnumber_t index, max_metadata;
405 xfs_sb_t *sbp = &mp->m_sb;
406 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
408 /* Check to see if the filesystem can overflow 32 bit inodes */
409 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
410 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
413 * Walk the current per-ag tree so we don't try to initialise AGs
414 * that already exist (growfs case). Allocate and insert all the
415 * AGs we don't find ready for initialisation.
417 for (index = 0; index < agcount; index++) {
418 pag = xfs_perag_get(mp, index);
423 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
426 if (radix_tree_preload(GFP_NOFS))
428 spin_lock(&mp->m_perag_lock);
429 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
431 spin_unlock(&mp->m_perag_lock);
435 spin_unlock(&mp->m_perag_lock);
436 radix_tree_preload_end();
439 /* Clear the mount flag if no inode can overflow 32 bits
440 * on this filesystem, or if specifically requested..
442 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
443 mp->m_flags |= XFS_MOUNT_32BITINODES;
445 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
448 /* If we can overflow then setup the ag headers accordingly */
449 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
450 /* Calculate how much should be reserved for inodes to
451 * meet the max inode percentage.
453 if (mp->m_maxicount) {
456 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
458 icount += sbp->sb_agblocks - 1;
459 do_div(icount, sbp->sb_agblocks);
460 max_metadata = icount;
462 max_metadata = agcount;
464 for (index = 0; index < agcount; index++) {
465 ino = XFS_AGINO_TO_INO(mp, index, agino);
466 if (ino > max_inum) {
471 /* This ag is preferred for inodes */
472 pag = xfs_perag_get(mp, index);
473 pag->pagi_inodeok = 1;
474 if (index < max_metadata)
475 pag->pagf_metadata = 1;
476 xfs_initialize_perag_icache(pag);
480 /* Setup default behavior for smaller filesystems */
481 for (index = 0; index < agcount; index++) {
482 pag = xfs_perag_get(mp, index);
483 pag->pagi_inodeok = 1;
484 xfs_initialize_perag_icache(pag);
498 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
499 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
500 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
501 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
502 to->sb_rextents = be64_to_cpu(from->sb_rextents);
503 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
504 to->sb_logstart = be64_to_cpu(from->sb_logstart);
505 to->sb_rootino = be64_to_cpu(from->sb_rootino);
506 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
507 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
508 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
509 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
510 to->sb_agcount = be32_to_cpu(from->sb_agcount);
511 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
512 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
513 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
514 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
515 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
516 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
517 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
518 to->sb_blocklog = from->sb_blocklog;
519 to->sb_sectlog = from->sb_sectlog;
520 to->sb_inodelog = from->sb_inodelog;
521 to->sb_inopblog = from->sb_inopblog;
522 to->sb_agblklog = from->sb_agblklog;
523 to->sb_rextslog = from->sb_rextslog;
524 to->sb_inprogress = from->sb_inprogress;
525 to->sb_imax_pct = from->sb_imax_pct;
526 to->sb_icount = be64_to_cpu(from->sb_icount);
527 to->sb_ifree = be64_to_cpu(from->sb_ifree);
528 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
529 to->sb_frextents = be64_to_cpu(from->sb_frextents);
530 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
531 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
532 to->sb_qflags = be16_to_cpu(from->sb_qflags);
533 to->sb_flags = from->sb_flags;
534 to->sb_shared_vn = from->sb_shared_vn;
535 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
536 to->sb_unit = be32_to_cpu(from->sb_unit);
537 to->sb_width = be32_to_cpu(from->sb_width);
538 to->sb_dirblklog = from->sb_dirblklog;
539 to->sb_logsectlog = from->sb_logsectlog;
540 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
541 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
542 to->sb_features2 = be32_to_cpu(from->sb_features2);
543 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
547 * Copy in core superblock to ondisk one.
549 * The fields argument is mask of superblock fields to copy.
557 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
558 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
568 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
569 first = xfs_sb_info[f].offset;
570 size = xfs_sb_info[f + 1].offset - first;
572 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
574 if (size == 1 || xfs_sb_info[f].type == 1) {
575 memcpy(to_ptr + first, from_ptr + first, size);
579 *(__be16 *)(to_ptr + first) =
580 cpu_to_be16(*(__u16 *)(from_ptr + first));
583 *(__be32 *)(to_ptr + first) =
584 cpu_to_be32(*(__u32 *)(from_ptr + first));
587 *(__be64 *)(to_ptr + first) =
588 cpu_to_be64(*(__u64 *)(from_ptr + first));
595 fields &= ~(1LL << f);
602 * Does the initial read of the superblock.
605 xfs_readsb(xfs_mount_t *mp, int flags)
607 unsigned int sector_size;
608 unsigned int extra_flags;
612 ASSERT(mp->m_sb_bp == NULL);
613 ASSERT(mp->m_ddev_targp != NULL);
616 * Allocate a (locked) buffer to hold the superblock.
617 * This will be kept around at all times to optimize
618 * access to the superblock.
620 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
621 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
623 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR, BTOBB(sector_size),
625 if (!bp || XFS_BUF_ISERROR(bp)) {
626 xfs_fs_mount_cmn_err(flags, "SB read failed");
627 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
630 ASSERT(XFS_BUF_ISBUSY(bp));
631 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
634 * Initialize the mount structure from the superblock.
635 * But first do some basic consistency checking.
637 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
639 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
641 xfs_fs_mount_cmn_err(flags, "SB validate failed");
646 * We must be able to do sector-sized and sector-aligned IO.
648 if (sector_size > mp->m_sb.sb_sectsize) {
649 xfs_fs_mount_cmn_err(flags,
650 "device supports only %u byte sectors (not %u)",
651 sector_size, mp->m_sb.sb_sectsize);
657 * If device sector size is smaller than the superblock size,
658 * re-read the superblock so the buffer is correctly sized.
660 if (sector_size < mp->m_sb.sb_sectsize) {
661 XFS_BUF_UNMANAGE(bp);
663 sector_size = mp->m_sb.sb_sectsize;
664 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR,
665 BTOBB(sector_size), extra_flags);
666 if (!bp || XFS_BUF_ISERROR(bp)) {
667 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
668 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
671 ASSERT(XFS_BUF_ISBUSY(bp));
672 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
675 /* Initialize per-cpu counters */
676 xfs_icsb_reinit_counters(mp);
680 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
685 XFS_BUF_UNMANAGE(bp);
695 * Mount initialization code establishing various mount
696 * fields from the superblock associated with the given
700 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
702 mp->m_agfrotor = mp->m_agirotor = 0;
703 spin_lock_init(&mp->m_agirotor_lock);
704 mp->m_maxagi = mp->m_sb.sb_agcount;
705 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
706 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
707 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
708 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
709 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
710 mp->m_blockmask = sbp->sb_blocksize - 1;
711 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
712 mp->m_blockwmask = mp->m_blockwsize - 1;
714 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
715 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
716 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
717 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
719 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
720 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
721 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
722 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
724 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
725 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
726 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
727 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
729 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
730 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
732 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
736 * xfs_initialize_perag_data
738 * Read in each per-ag structure so we can count up the number of
739 * allocated inodes, free inodes and used filesystem blocks as this
740 * information is no longer persistent in the superblock. Once we have
741 * this information, write it into the in-core superblock structure.
744 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
746 xfs_agnumber_t index;
748 xfs_sb_t *sbp = &mp->m_sb;
752 uint64_t bfreelst = 0;
756 for (index = 0; index < agcount; index++) {
758 * read the agf, then the agi. This gets us
759 * all the information we need and populates the
760 * per-ag structures for us.
762 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
766 error = xfs_ialloc_pagi_init(mp, NULL, index);
769 pag = xfs_perag_get(mp, index);
770 ifree += pag->pagi_freecount;
771 ialloc += pag->pagi_count;
772 bfree += pag->pagf_freeblks;
773 bfreelst += pag->pagf_flcount;
774 btree += pag->pagf_btreeblks;
778 * Overwrite incore superblock counters with just-read data
780 spin_lock(&mp->m_sb_lock);
781 sbp->sb_ifree = ifree;
782 sbp->sb_icount = ialloc;
783 sbp->sb_fdblocks = bfree + bfreelst + btree;
784 spin_unlock(&mp->m_sb_lock);
786 /* Fixup the per-cpu counters as well. */
787 xfs_icsb_reinit_counters(mp);
793 * Update alignment values based on mount options and sb values
796 xfs_update_alignment(xfs_mount_t *mp)
798 xfs_sb_t *sbp = &(mp->m_sb);
802 * If stripe unit and stripe width are not multiples
803 * of the fs blocksize turn off alignment.
805 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
806 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
807 if (mp->m_flags & XFS_MOUNT_RETERR) {
809 "XFS: alignment check 1 failed");
810 return XFS_ERROR(EINVAL);
812 mp->m_dalign = mp->m_swidth = 0;
815 * Convert the stripe unit and width to FSBs.
817 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
818 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
819 if (mp->m_flags & XFS_MOUNT_RETERR) {
820 return XFS_ERROR(EINVAL);
822 xfs_fs_cmn_err(CE_WARN, mp,
823 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
824 mp->m_dalign, mp->m_swidth,
829 } else if (mp->m_dalign) {
830 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
832 if (mp->m_flags & XFS_MOUNT_RETERR) {
833 xfs_fs_cmn_err(CE_WARN, mp,
834 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
837 return XFS_ERROR(EINVAL);
844 * Update superblock with new values
847 if (xfs_sb_version_hasdalign(sbp)) {
848 if (sbp->sb_unit != mp->m_dalign) {
849 sbp->sb_unit = mp->m_dalign;
850 mp->m_update_flags |= XFS_SB_UNIT;
852 if (sbp->sb_width != mp->m_swidth) {
853 sbp->sb_width = mp->m_swidth;
854 mp->m_update_flags |= XFS_SB_WIDTH;
857 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
858 xfs_sb_version_hasdalign(&mp->m_sb)) {
859 mp->m_dalign = sbp->sb_unit;
860 mp->m_swidth = sbp->sb_width;
867 * Set the maximum inode count for this filesystem
870 xfs_set_maxicount(xfs_mount_t *mp)
872 xfs_sb_t *sbp = &(mp->m_sb);
875 if (sbp->sb_imax_pct) {
877 * Make sure the maximum inode count is a multiple
878 * of the units we allocate inodes in.
880 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
882 do_div(icount, mp->m_ialloc_blks);
883 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
891 * Set the default minimum read and write sizes unless
892 * already specified in a mount option.
893 * We use smaller I/O sizes when the file system
894 * is being used for NFS service (wsync mount option).
897 xfs_set_rw_sizes(xfs_mount_t *mp)
899 xfs_sb_t *sbp = &(mp->m_sb);
900 int readio_log, writeio_log;
902 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
903 if (mp->m_flags & XFS_MOUNT_WSYNC) {
904 readio_log = XFS_WSYNC_READIO_LOG;
905 writeio_log = XFS_WSYNC_WRITEIO_LOG;
907 readio_log = XFS_READIO_LOG_LARGE;
908 writeio_log = XFS_WRITEIO_LOG_LARGE;
911 readio_log = mp->m_readio_log;
912 writeio_log = mp->m_writeio_log;
915 if (sbp->sb_blocklog > readio_log) {
916 mp->m_readio_log = sbp->sb_blocklog;
918 mp->m_readio_log = readio_log;
920 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
921 if (sbp->sb_blocklog > writeio_log) {
922 mp->m_writeio_log = sbp->sb_blocklog;
924 mp->m_writeio_log = writeio_log;
926 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
930 * Set whether we're using inode alignment.
933 xfs_set_inoalignment(xfs_mount_t *mp)
935 if (xfs_sb_version_hasalign(&mp->m_sb) &&
936 mp->m_sb.sb_inoalignmt >=
937 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
938 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
940 mp->m_inoalign_mask = 0;
942 * If we are using stripe alignment, check whether
943 * the stripe unit is a multiple of the inode alignment
945 if (mp->m_dalign && mp->m_inoalign_mask &&
946 !(mp->m_dalign & mp->m_inoalign_mask))
947 mp->m_sinoalign = mp->m_dalign;
953 * Check that the data (and log if separate) are an ok size.
956 xfs_check_sizes(xfs_mount_t *mp)
962 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
963 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
964 cmn_err(CE_WARN, "XFS: size check 1 failed");
965 return XFS_ERROR(E2BIG);
967 error = xfs_read_buf(mp, mp->m_ddev_targp,
968 d - XFS_FSS_TO_BB(mp, 1),
969 XFS_FSS_TO_BB(mp, 1), 0, &bp);
973 cmn_err(CE_WARN, "XFS: size check 2 failed");
975 error = XFS_ERROR(E2BIG);
979 if (mp->m_logdev_targp != mp->m_ddev_targp) {
980 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
981 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
982 cmn_err(CE_WARN, "XFS: size check 3 failed");
983 return XFS_ERROR(E2BIG);
985 error = xfs_read_buf(mp, mp->m_logdev_targp,
986 d - XFS_FSB_TO_BB(mp, 1),
987 XFS_FSB_TO_BB(mp, 1), 0, &bp);
991 cmn_err(CE_WARN, "XFS: size check 3 failed");
993 error = XFS_ERROR(E2BIG);
1001 * Clear the quotaflags in memory and in the superblock.
1004 xfs_mount_reset_sbqflags(
1005 struct xfs_mount *mp)
1008 struct xfs_trans *tp;
1013 * It is OK to look at sb_qflags here in mount path,
1014 * without m_sb_lock.
1016 if (mp->m_sb.sb_qflags == 0)
1018 spin_lock(&mp->m_sb_lock);
1019 mp->m_sb.sb_qflags = 0;
1020 spin_unlock(&mp->m_sb_lock);
1023 * If the fs is readonly, let the incore superblock run
1024 * with quotas off but don't flush the update out to disk
1026 if (mp->m_flags & XFS_MOUNT_RDONLY)
1030 xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes");
1033 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1034 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1035 XFS_DEFAULT_LOG_COUNT);
1037 xfs_trans_cancel(tp, 0);
1038 xfs_fs_cmn_err(CE_ALERT, mp,
1039 "xfs_mount_reset_sbqflags: Superblock update failed!");
1043 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1044 return xfs_trans_commit(tp, 0);
1048 * This function does the following on an initial mount of a file system:
1049 * - reads the superblock from disk and init the mount struct
1050 * - if we're a 32-bit kernel, do a size check on the superblock
1051 * so we don't mount terabyte filesystems
1052 * - init mount struct realtime fields
1053 * - allocate inode hash table for fs
1054 * - init directory manager
1055 * - perform recovery and init the log manager
1061 xfs_sb_t *sbp = &(mp->m_sb);
1064 uint quotamount = 0;
1065 uint quotaflags = 0;
1068 xfs_mount_common(mp, sbp);
1071 * Check for a mismatched features2 values. Older kernels
1072 * read & wrote into the wrong sb offset for sb_features2
1073 * on some platforms due to xfs_sb_t not being 64bit size aligned
1074 * when sb_features2 was added, which made older superblock
1075 * reading/writing routines swap it as a 64-bit value.
1077 * For backwards compatibility, we make both slots equal.
1079 * If we detect a mismatched field, we OR the set bits into the
1080 * existing features2 field in case it has already been modified; we
1081 * don't want to lose any features. We then update the bad location
1082 * with the ORed value so that older kernels will see any features2
1083 * flags, and mark the two fields as needing updates once the
1084 * transaction subsystem is online.
1086 if (xfs_sb_has_mismatched_features2(sbp)) {
1088 "XFS: correcting sb_features alignment problem");
1089 sbp->sb_features2 |= sbp->sb_bad_features2;
1090 sbp->sb_bad_features2 = sbp->sb_features2;
1091 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1094 * Re-check for ATTR2 in case it was found in bad_features2
1097 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1098 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1099 mp->m_flags |= XFS_MOUNT_ATTR2;
1102 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1103 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1104 xfs_sb_version_removeattr2(&mp->m_sb);
1105 mp->m_update_flags |= XFS_SB_FEATURES2;
1107 /* update sb_versionnum for the clearing of the morebits */
1108 if (!sbp->sb_features2)
1109 mp->m_update_flags |= XFS_SB_VERSIONNUM;
1113 * Check if sb_agblocks is aligned at stripe boundary
1114 * If sb_agblocks is NOT aligned turn off m_dalign since
1115 * allocator alignment is within an ag, therefore ag has
1116 * to be aligned at stripe boundary.
1118 error = xfs_update_alignment(mp);
1122 xfs_alloc_compute_maxlevels(mp);
1123 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1124 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1125 xfs_ialloc_compute_maxlevels(mp);
1127 xfs_set_maxicount(mp);
1129 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1131 error = xfs_uuid_mount(mp);
1136 * Set the minimum read and write sizes
1138 xfs_set_rw_sizes(mp);
1141 * Set the inode cluster size.
1142 * This may still be overridden by the file system
1143 * block size if it is larger than the chosen cluster size.
1145 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1148 * Set inode alignment fields
1150 xfs_set_inoalignment(mp);
1153 * Check that the data (and log if separate) are an ok size.
1155 error = xfs_check_sizes(mp);
1157 goto out_remove_uuid;
1160 * Initialize realtime fields in the mount structure
1162 error = xfs_rtmount_init(mp);
1164 cmn_err(CE_WARN, "XFS: RT mount failed");
1165 goto out_remove_uuid;
1169 * Copies the low order bits of the timestamp and the randomly
1170 * set "sequence" number out of a UUID.
1172 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1174 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1179 * Initialize the attribute manager's entries.
1181 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1184 * Initialize the precomputed transaction reservations values.
1189 * Allocate and initialize the per-ag data.
1191 spin_lock_init(&mp->m_perag_lock);
1192 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_NOFS);
1193 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1195 cmn_err(CE_WARN, "XFS: Failed per-ag init: %d", error);
1196 goto out_remove_uuid;
1199 if (!sbp->sb_logblocks) {
1200 cmn_err(CE_WARN, "XFS: no log defined");
1201 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1202 error = XFS_ERROR(EFSCORRUPTED);
1203 goto out_free_perag;
1207 * log's mount-time initialization. Perform 1st part recovery if needed
1209 error = xfs_log_mount(mp, mp->m_logdev_targp,
1210 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1211 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1213 cmn_err(CE_WARN, "XFS: log mount failed");
1214 goto out_free_perag;
1218 * Now the log is mounted, we know if it was an unclean shutdown or
1219 * not. If it was, with the first phase of recovery has completed, we
1220 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1221 * but they are recovered transactionally in the second recovery phase
1224 * Hence we can safely re-initialise incore superblock counters from
1225 * the per-ag data. These may not be correct if the filesystem was not
1226 * cleanly unmounted, so we need to wait for recovery to finish before
1229 * If the filesystem was cleanly unmounted, then we can trust the
1230 * values in the superblock to be correct and we don't need to do
1233 * If we are currently making the filesystem, the initialisation will
1234 * fail as the perag data is in an undefined state.
1236 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1237 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1238 !mp->m_sb.sb_inprogress) {
1239 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1241 goto out_free_perag;
1245 * Get and sanity-check the root inode.
1246 * Save the pointer to it in the mount structure.
1248 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
1250 cmn_err(CE_WARN, "XFS: failed to read root inode");
1251 goto out_log_dealloc;
1254 ASSERT(rip != NULL);
1256 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1257 cmn_err(CE_WARN, "XFS: corrupted root inode");
1258 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1259 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1260 (unsigned long long)rip->i_ino);
1261 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1262 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1264 error = XFS_ERROR(EFSCORRUPTED);
1267 mp->m_rootip = rip; /* save it */
1269 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1272 * Initialize realtime inode pointers in the mount structure
1274 error = xfs_rtmount_inodes(mp);
1277 * Free up the root inode.
1279 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1284 * If this is a read-only mount defer the superblock updates until
1285 * the next remount into writeable mode. Otherwise we would never
1286 * perform the update e.g. for the root filesystem.
1288 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1289 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1291 cmn_err(CE_WARN, "XFS: failed to write sb changes");
1297 * Initialise the XFS quota management subsystem for this mount
1299 if (XFS_IS_QUOTA_RUNNING(mp)) {
1300 error = xfs_qm_newmount(mp, "amount, "aflags);
1304 ASSERT(!XFS_IS_QUOTA_ON(mp));
1307 * If a file system had quotas running earlier, but decided to
1308 * mount without -o uquota/pquota/gquota options, revoke the
1309 * quotachecked license.
1311 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1313 "XFS: resetting qflags for filesystem %s",
1316 error = xfs_mount_reset_sbqflags(mp);
1323 * Finish recovering the file system. This part needed to be
1324 * delayed until after the root and real-time bitmap inodes
1325 * were consistently read in.
1327 error = xfs_log_mount_finish(mp);
1329 cmn_err(CE_WARN, "XFS: log mount finish failed");
1334 * Complete the quota initialisation, post-log-replay component.
1337 ASSERT(mp->m_qflags == 0);
1338 mp->m_qflags = quotaflags;
1340 xfs_qm_mount_quotas(mp);
1343 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1344 if (XFS_IS_QUOTA_ON(mp))
1345 xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas turned on");
1347 xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas not turned on");
1351 * Now we are mounted, reserve a small amount of unused space for
1352 * privileged transactions. This is needed so that transaction
1353 * space required for critical operations can dip into this pool
1354 * when at ENOSPC. This is needed for operations like create with
1355 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1356 * are not allowed to use this reserved space.
1358 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1359 * This may drive us straight to ENOSPC on mount, but that implies
1360 * we were already there on the last unmount. Warn if this occurs.
1362 resblks = mp->m_sb.sb_dblocks;
1363 do_div(resblks, 20);
1364 resblks = min_t(__uint64_t, resblks, 1024);
1365 error = xfs_reserve_blocks(mp, &resblks, NULL);
1367 cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. "
1368 "Continuing without a reserve pool.");
1373 xfs_rtunmount_inodes(mp);
1377 xfs_log_unmount(mp);
1381 xfs_uuid_unmount(mp);
1387 * This flushes out the inodes,dquots and the superblock, unmounts the
1388 * log and makes sure that incore structures are freed.
1392 struct xfs_mount *mp)
1397 xfs_qm_unmount_quotas(mp);
1398 xfs_rtunmount_inodes(mp);
1399 IRELE(mp->m_rootip);
1402 * We can potentially deadlock here if we have an inode cluster
1403 * that has been freed has its buffer still pinned in memory because
1404 * the transaction is still sitting in a iclog. The stale inodes
1405 * on that buffer will have their flush locks held until the
1406 * transaction hits the disk and the callbacks run. the inode
1407 * flush takes the flush lock unconditionally and with nothing to
1408 * push out the iclog we will never get that unlocked. hence we
1409 * need to force the log first.
1411 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1412 xfs_reclaim_inodes(mp, XFS_IFLUSH_ASYNC);
1417 * Flush out the log synchronously so that we know for sure
1418 * that nothing is pinned. This is important because bflush()
1419 * will skip pinned buffers.
1421 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1423 xfs_binval(mp->m_ddev_targp);
1424 if (mp->m_rtdev_targp) {
1425 xfs_binval(mp->m_rtdev_targp);
1429 * Unreserve any blocks we have so that when we unmount we don't account
1430 * the reserved free space as used. This is really only necessary for
1431 * lazy superblock counting because it trusts the incore superblock
1432 * counters to be absolutely correct on clean unmount.
1434 * We don't bother correcting this elsewhere for lazy superblock
1435 * counting because on mount of an unclean filesystem we reconstruct the
1436 * correct counter value and this is irrelevant.
1438 * For non-lazy counter filesystems, this doesn't matter at all because
1439 * we only every apply deltas to the superblock and hence the incore
1440 * value does not matter....
1443 error = xfs_reserve_blocks(mp, &resblks, NULL);
1445 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1446 "Freespace may not be correct on next mount.");
1448 error = xfs_log_sbcount(mp, 1);
1450 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1451 "Freespace may not be correct on next mount.");
1452 xfs_unmountfs_writesb(mp);
1453 xfs_unmountfs_wait(mp); /* wait for async bufs */
1454 xfs_log_unmount_write(mp);
1455 xfs_log_unmount(mp);
1456 xfs_uuid_unmount(mp);
1459 xfs_errortag_clearall(mp, 0);
1465 xfs_unmountfs_wait(xfs_mount_t *mp)
1467 if (mp->m_logdev_targp != mp->m_ddev_targp)
1468 xfs_wait_buftarg(mp->m_logdev_targp);
1469 if (mp->m_rtdev_targp)
1470 xfs_wait_buftarg(mp->m_rtdev_targp);
1471 xfs_wait_buftarg(mp->m_ddev_targp);
1475 xfs_fs_writable(xfs_mount_t *mp)
1477 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1478 (mp->m_flags & XFS_MOUNT_RDONLY));
1484 * Called either periodically to keep the on disk superblock values
1485 * roughly up to date or from unmount to make sure the values are
1486 * correct on a clean unmount.
1488 * Note this code can be called during the process of freezing, so
1489 * we may need to use the transaction allocator which does not not
1490 * block when the transaction subsystem is in its frozen state.
1500 if (!xfs_fs_writable(mp))
1503 xfs_icsb_sync_counters(mp, 0);
1506 * we don't need to do this if we are updating the superblock
1507 * counters on every modification.
1509 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1512 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1513 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1514 XFS_DEFAULT_LOG_COUNT);
1516 xfs_trans_cancel(tp, 0);
1520 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1522 xfs_trans_set_sync(tp);
1523 error = xfs_trans_commit(tp, 0);
1528 xfs_unmountfs_writesb(xfs_mount_t *mp)
1534 * skip superblock write if fs is read-only, or
1535 * if we are doing a forced umount.
1537 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1538 XFS_FORCED_SHUTDOWN(mp))) {
1540 sbp = xfs_getsb(mp, 0);
1542 XFS_BUF_UNDONE(sbp);
1543 XFS_BUF_UNREAD(sbp);
1544 XFS_BUF_UNDELAYWRITE(sbp);
1546 XFS_BUF_UNASYNC(sbp);
1547 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1548 xfsbdstrat(mp, sbp);
1549 error = xfs_iowait(sbp);
1551 xfs_ioerror_alert("xfs_unmountfs_writesb",
1552 mp, sbp, XFS_BUF_ADDR(sbp));
1559 * xfs_mod_sb() can be used to copy arbitrary changes to the
1560 * in-core superblock into the superblock buffer to be logged.
1561 * It does not provide the higher level of locking that is
1562 * needed to protect the in-core superblock from concurrent
1566 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1578 bp = xfs_trans_getsb(tp, mp, 0);
1579 first = sizeof(xfs_sb_t);
1582 /* translate/copy */
1584 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1586 /* find modified range */
1588 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1589 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1590 first = xfs_sb_info[f].offset;
1592 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1593 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1594 last = xfs_sb_info[f + 1].offset - 1;
1596 xfs_trans_log_buf(tp, bp, first, last);
1601 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1602 * a delta to a specified field in the in-core superblock. Simply
1603 * switch on the field indicated and apply the delta to that field.
1604 * Fields are not allowed to dip below zero, so if the delta would
1605 * do this do not apply it and return EINVAL.
1607 * The m_sb_lock must be held when this routine is called.
1610 xfs_mod_incore_sb_unlocked(
1612 xfs_sb_field_t field,
1616 int scounter; /* short counter for 32 bit fields */
1617 long long lcounter; /* long counter for 64 bit fields */
1618 long long res_used, rem;
1621 * With the in-core superblock spin lock held, switch
1622 * on the indicated field. Apply the delta to the
1623 * proper field. If the fields value would dip below
1624 * 0, then do not apply the delta and return EINVAL.
1627 case XFS_SBS_ICOUNT:
1628 lcounter = (long long)mp->m_sb.sb_icount;
1632 return XFS_ERROR(EINVAL);
1634 mp->m_sb.sb_icount = lcounter;
1637 lcounter = (long long)mp->m_sb.sb_ifree;
1641 return XFS_ERROR(EINVAL);
1643 mp->m_sb.sb_ifree = lcounter;
1645 case XFS_SBS_FDBLOCKS:
1646 lcounter = (long long)
1647 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1648 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1650 if (delta > 0) { /* Putting blocks back */
1651 if (res_used > delta) {
1652 mp->m_resblks_avail += delta;
1654 rem = delta - res_used;
1655 mp->m_resblks_avail = mp->m_resblks;
1658 } else { /* Taking blocks away */
1663 * If were out of blocks, use any available reserved blocks if
1669 lcounter = (long long)mp->m_resblks_avail + delta;
1671 return XFS_ERROR(ENOSPC);
1673 mp->m_resblks_avail = lcounter;
1675 } else { /* not reserved */
1676 return XFS_ERROR(ENOSPC);
1681 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1683 case XFS_SBS_FREXTENTS:
1684 lcounter = (long long)mp->m_sb.sb_frextents;
1687 return XFS_ERROR(ENOSPC);
1689 mp->m_sb.sb_frextents = lcounter;
1691 case XFS_SBS_DBLOCKS:
1692 lcounter = (long long)mp->m_sb.sb_dblocks;
1696 return XFS_ERROR(EINVAL);
1698 mp->m_sb.sb_dblocks = lcounter;
1700 case XFS_SBS_AGCOUNT:
1701 scounter = mp->m_sb.sb_agcount;
1705 return XFS_ERROR(EINVAL);
1707 mp->m_sb.sb_agcount = scounter;
1709 case XFS_SBS_IMAX_PCT:
1710 scounter = mp->m_sb.sb_imax_pct;
1714 return XFS_ERROR(EINVAL);
1716 mp->m_sb.sb_imax_pct = scounter;
1718 case XFS_SBS_REXTSIZE:
1719 scounter = mp->m_sb.sb_rextsize;
1723 return XFS_ERROR(EINVAL);
1725 mp->m_sb.sb_rextsize = scounter;
1727 case XFS_SBS_RBMBLOCKS:
1728 scounter = mp->m_sb.sb_rbmblocks;
1732 return XFS_ERROR(EINVAL);
1734 mp->m_sb.sb_rbmblocks = scounter;
1736 case XFS_SBS_RBLOCKS:
1737 lcounter = (long long)mp->m_sb.sb_rblocks;
1741 return XFS_ERROR(EINVAL);
1743 mp->m_sb.sb_rblocks = lcounter;
1745 case XFS_SBS_REXTENTS:
1746 lcounter = (long long)mp->m_sb.sb_rextents;
1750 return XFS_ERROR(EINVAL);
1752 mp->m_sb.sb_rextents = lcounter;
1754 case XFS_SBS_REXTSLOG:
1755 scounter = mp->m_sb.sb_rextslog;
1759 return XFS_ERROR(EINVAL);
1761 mp->m_sb.sb_rextslog = scounter;
1765 return XFS_ERROR(EINVAL);
1770 * xfs_mod_incore_sb() is used to change a field in the in-core
1771 * superblock structure by the specified delta. This modification
1772 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1773 * routine to do the work.
1778 xfs_sb_field_t field,
1784 /* check for per-cpu counters */
1786 #ifdef HAVE_PERCPU_SB
1787 case XFS_SBS_ICOUNT:
1789 case XFS_SBS_FDBLOCKS:
1790 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1791 status = xfs_icsb_modify_counters(mp, field,
1798 spin_lock(&mp->m_sb_lock);
1799 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1800 spin_unlock(&mp->m_sb_lock);
1808 * xfs_mod_incore_sb_batch() is used to change more than one field
1809 * in the in-core superblock structure at a time. This modification
1810 * is protected by a lock internal to this module. The fields and
1811 * changes to those fields are specified in the array of xfs_mod_sb
1812 * structures passed in.
1814 * Either all of the specified deltas will be applied or none of
1815 * them will. If any modified field dips below 0, then all modifications
1816 * will be backed out and EINVAL will be returned.
1819 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1825 * Loop through the array of mod structures and apply each
1826 * individually. If any fail, then back out all those
1827 * which have already been applied. Do all of this within
1828 * the scope of the m_sb_lock so that all of the changes will
1831 spin_lock(&mp->m_sb_lock);
1833 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1835 * Apply the delta at index n. If it fails, break
1836 * from the loop so we'll fall into the undo loop
1839 switch (msbp->msb_field) {
1840 #ifdef HAVE_PERCPU_SB
1841 case XFS_SBS_ICOUNT:
1843 case XFS_SBS_FDBLOCKS:
1844 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1845 spin_unlock(&mp->m_sb_lock);
1846 status = xfs_icsb_modify_counters(mp,
1848 msbp->msb_delta, rsvd);
1849 spin_lock(&mp->m_sb_lock);
1855 status = xfs_mod_incore_sb_unlocked(mp,
1857 msbp->msb_delta, rsvd);
1867 * If we didn't complete the loop above, then back out
1868 * any changes made to the superblock. If you add code
1869 * between the loop above and here, make sure that you
1870 * preserve the value of status. Loop back until
1871 * we step below the beginning of the array. Make sure
1872 * we don't touch anything back there.
1876 while (msbp >= msb) {
1877 switch (msbp->msb_field) {
1878 #ifdef HAVE_PERCPU_SB
1879 case XFS_SBS_ICOUNT:
1881 case XFS_SBS_FDBLOCKS:
1882 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1883 spin_unlock(&mp->m_sb_lock);
1884 status = xfs_icsb_modify_counters(mp,
1888 spin_lock(&mp->m_sb_lock);
1894 status = xfs_mod_incore_sb_unlocked(mp,
1900 ASSERT(status == 0);
1904 spin_unlock(&mp->m_sb_lock);
1909 * xfs_getsb() is called to obtain the buffer for the superblock.
1910 * The buffer is returned locked and read in from disk.
1911 * The buffer should be released with a call to xfs_brelse().
1913 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1914 * the superblock buffer if it can be locked without sleeping.
1915 * If it can't then we'll return NULL.
1924 ASSERT(mp->m_sb_bp != NULL);
1926 if (flags & XFS_BUF_TRYLOCK) {
1927 if (!XFS_BUF_CPSEMA(bp)) {
1931 XFS_BUF_PSEMA(bp, PRIBIO);
1934 ASSERT(XFS_BUF_ISDONE(bp));
1939 * Used to free the superblock along various error paths.
1948 * Use xfs_getsb() so that the buffer will be locked
1949 * when we call xfs_buf_relse().
1951 bp = xfs_getsb(mp, 0);
1952 XFS_BUF_UNMANAGE(bp);
1958 * Used to log changes to the superblock unit and width fields which could
1959 * be altered by the mount options, as well as any potential sb_features2
1960 * fixup. Only the first superblock is updated.
1970 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1971 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1972 XFS_SB_VERSIONNUM));
1974 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1975 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1976 XFS_DEFAULT_LOG_COUNT);
1978 xfs_trans_cancel(tp, 0);
1981 xfs_mod_sb(tp, fields);
1982 error = xfs_trans_commit(tp, 0);
1987 #ifdef HAVE_PERCPU_SB
1989 * Per-cpu incore superblock counters
1991 * Simple concept, difficult implementation
1993 * Basically, replace the incore superblock counters with a distributed per cpu
1994 * counter for contended fields (e.g. free block count).
1996 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1997 * hence needs to be accurately read when we are running low on space. Hence
1998 * there is a method to enable and disable the per-cpu counters based on how
1999 * much "stuff" is available in them.
2001 * Basically, a counter is enabled if there is enough free resource to justify
2002 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2003 * ENOSPC), then we disable the counters to synchronise all callers and
2004 * re-distribute the available resources.
2006 * If, once we redistributed the available resources, we still get a failure,
2007 * we disable the per-cpu counter and go through the slow path.
2009 * The slow path is the current xfs_mod_incore_sb() function. This means that
2010 * when we disable a per-cpu counter, we need to drain its resources back to
2011 * the global superblock. We do this after disabling the counter to prevent
2012 * more threads from queueing up on the counter.
2014 * Essentially, this means that we still need a lock in the fast path to enable
2015 * synchronisation between the global counters and the per-cpu counters. This
2016 * is not a problem because the lock will be local to a CPU almost all the time
2017 * and have little contention except when we get to ENOSPC conditions.
2019 * Basically, this lock becomes a barrier that enables us to lock out the fast
2020 * path while we do things like enabling and disabling counters and
2021 * synchronising the counters.
2025 * 1. m_sb_lock before picking up per-cpu locks
2026 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2027 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2028 * 4. modifying per-cpu counters requires holding per-cpu lock
2029 * 5. modifying global counters requires holding m_sb_lock
2030 * 6. enabling or disabling a counter requires holding the m_sb_lock
2031 * and _none_ of the per-cpu locks.
2033 * Disabled counters are only ever re-enabled by a balance operation
2034 * that results in more free resources per CPU than a given threshold.
2035 * To ensure counters don't remain disabled, they are rebalanced when
2036 * the global resource goes above a higher threshold (i.e. some hysteresis
2037 * is present to prevent thrashing).
2040 #ifdef CONFIG_HOTPLUG_CPU
2042 * hot-plug CPU notifier support.
2044 * We need a notifier per filesystem as we need to be able to identify
2045 * the filesystem to balance the counters out. This is achieved by
2046 * having a notifier block embedded in the xfs_mount_t and doing pointer
2047 * magic to get the mount pointer from the notifier block address.
2050 xfs_icsb_cpu_notify(
2051 struct notifier_block *nfb,
2052 unsigned long action,
2055 xfs_icsb_cnts_t *cntp;
2058 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2059 cntp = (xfs_icsb_cnts_t *)
2060 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2062 case CPU_UP_PREPARE:
2063 case CPU_UP_PREPARE_FROZEN:
2064 /* Easy Case - initialize the area and locks, and
2065 * then rebalance when online does everything else for us. */
2066 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2069 case CPU_ONLINE_FROZEN:
2071 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2072 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2073 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2074 xfs_icsb_unlock(mp);
2077 case CPU_DEAD_FROZEN:
2078 /* Disable all the counters, then fold the dead cpu's
2079 * count into the total on the global superblock and
2080 * re-enable the counters. */
2082 spin_lock(&mp->m_sb_lock);
2083 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2084 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2085 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2087 mp->m_sb.sb_icount += cntp->icsb_icount;
2088 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2089 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2091 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2093 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2094 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2095 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2096 spin_unlock(&mp->m_sb_lock);
2097 xfs_icsb_unlock(mp);
2103 #endif /* CONFIG_HOTPLUG_CPU */
2106 xfs_icsb_init_counters(
2109 xfs_icsb_cnts_t *cntp;
2112 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2113 if (mp->m_sb_cnts == NULL)
2116 #ifdef CONFIG_HOTPLUG_CPU
2117 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2118 mp->m_icsb_notifier.priority = 0;
2119 register_hotcpu_notifier(&mp->m_icsb_notifier);
2120 #endif /* CONFIG_HOTPLUG_CPU */
2122 for_each_online_cpu(i) {
2123 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2124 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2127 mutex_init(&mp->m_icsb_mutex);
2130 * start with all counters disabled so that the
2131 * initial balance kicks us off correctly
2133 mp->m_icsb_counters = -1;
2138 xfs_icsb_reinit_counters(
2143 * start with all counters disabled so that the
2144 * initial balance kicks us off correctly
2146 mp->m_icsb_counters = -1;
2147 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2148 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2149 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2150 xfs_icsb_unlock(mp);
2154 xfs_icsb_destroy_counters(
2157 if (mp->m_sb_cnts) {
2158 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2159 free_percpu(mp->m_sb_cnts);
2161 mutex_destroy(&mp->m_icsb_mutex);
2166 xfs_icsb_cnts_t *icsbp)
2168 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2174 xfs_icsb_unlock_cntr(
2175 xfs_icsb_cnts_t *icsbp)
2177 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2182 xfs_icsb_lock_all_counters(
2185 xfs_icsb_cnts_t *cntp;
2188 for_each_online_cpu(i) {
2189 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2190 xfs_icsb_lock_cntr(cntp);
2195 xfs_icsb_unlock_all_counters(
2198 xfs_icsb_cnts_t *cntp;
2201 for_each_online_cpu(i) {
2202 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2203 xfs_icsb_unlock_cntr(cntp);
2210 xfs_icsb_cnts_t *cnt,
2213 xfs_icsb_cnts_t *cntp;
2216 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2218 if (!(flags & XFS_ICSB_LAZY_COUNT))
2219 xfs_icsb_lock_all_counters(mp);
2221 for_each_online_cpu(i) {
2222 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2223 cnt->icsb_icount += cntp->icsb_icount;
2224 cnt->icsb_ifree += cntp->icsb_ifree;
2225 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2228 if (!(flags & XFS_ICSB_LAZY_COUNT))
2229 xfs_icsb_unlock_all_counters(mp);
2233 xfs_icsb_counter_disabled(
2235 xfs_sb_field_t field)
2237 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2238 return test_bit(field, &mp->m_icsb_counters);
2242 xfs_icsb_disable_counter(
2244 xfs_sb_field_t field)
2246 xfs_icsb_cnts_t cnt;
2248 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2251 * If we are already disabled, then there is nothing to do
2252 * here. We check before locking all the counters to avoid
2253 * the expensive lock operation when being called in the
2254 * slow path and the counter is already disabled. This is
2255 * safe because the only time we set or clear this state is under
2258 if (xfs_icsb_counter_disabled(mp, field))
2261 xfs_icsb_lock_all_counters(mp);
2262 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2263 /* drain back to superblock */
2265 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2267 case XFS_SBS_ICOUNT:
2268 mp->m_sb.sb_icount = cnt.icsb_icount;
2271 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2273 case XFS_SBS_FDBLOCKS:
2274 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2281 xfs_icsb_unlock_all_counters(mp);
2285 xfs_icsb_enable_counter(
2287 xfs_sb_field_t field,
2291 xfs_icsb_cnts_t *cntp;
2294 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2296 xfs_icsb_lock_all_counters(mp);
2297 for_each_online_cpu(i) {
2298 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2300 case XFS_SBS_ICOUNT:
2301 cntp->icsb_icount = count + resid;
2304 cntp->icsb_ifree = count + resid;
2306 case XFS_SBS_FDBLOCKS:
2307 cntp->icsb_fdblocks = count + resid;
2315 clear_bit(field, &mp->m_icsb_counters);
2316 xfs_icsb_unlock_all_counters(mp);
2320 xfs_icsb_sync_counters_locked(
2324 xfs_icsb_cnts_t cnt;
2326 xfs_icsb_count(mp, &cnt, flags);
2328 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2329 mp->m_sb.sb_icount = cnt.icsb_icount;
2330 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2331 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2332 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2333 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2337 * Accurate update of per-cpu counters to incore superblock
2340 xfs_icsb_sync_counters(
2344 spin_lock(&mp->m_sb_lock);
2345 xfs_icsb_sync_counters_locked(mp, flags);
2346 spin_unlock(&mp->m_sb_lock);
2350 * Balance and enable/disable counters as necessary.
2352 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2353 * chosen to be the same number as single on disk allocation chunk per CPU, and
2354 * free blocks is something far enough zero that we aren't going thrash when we
2355 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2356 * prevent looping endlessly when xfs_alloc_space asks for more than will
2357 * be distributed to a single CPU but each CPU has enough blocks to be
2360 * Note that we can be called when counters are already disabled.
2361 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2362 * prevent locking every per-cpu counter needlessly.
2365 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2366 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2367 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2369 xfs_icsb_balance_counter_locked(
2371 xfs_sb_field_t field,
2374 uint64_t count, resid;
2375 int weight = num_online_cpus();
2376 uint64_t min = (uint64_t)min_per_cpu;
2378 /* disable counter and sync counter */
2379 xfs_icsb_disable_counter(mp, field);
2381 /* update counters - first CPU gets residual*/
2383 case XFS_SBS_ICOUNT:
2384 count = mp->m_sb.sb_icount;
2385 resid = do_div(count, weight);
2386 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2390 count = mp->m_sb.sb_ifree;
2391 resid = do_div(count, weight);
2392 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2395 case XFS_SBS_FDBLOCKS:
2396 count = mp->m_sb.sb_fdblocks;
2397 resid = do_div(count, weight);
2398 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2403 count = resid = 0; /* quiet, gcc */
2407 xfs_icsb_enable_counter(mp, field, count, resid);
2411 xfs_icsb_balance_counter(
2413 xfs_sb_field_t fields,
2416 spin_lock(&mp->m_sb_lock);
2417 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2418 spin_unlock(&mp->m_sb_lock);
2422 xfs_icsb_modify_counters(
2424 xfs_sb_field_t field,
2428 xfs_icsb_cnts_t *icsbp;
2429 long long lcounter; /* long counter for 64 bit fields */
2435 icsbp = this_cpu_ptr(mp->m_sb_cnts);
2438 * if the counter is disabled, go to slow path
2440 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2442 xfs_icsb_lock_cntr(icsbp);
2443 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2444 xfs_icsb_unlock_cntr(icsbp);
2449 case XFS_SBS_ICOUNT:
2450 lcounter = icsbp->icsb_icount;
2452 if (unlikely(lcounter < 0))
2453 goto balance_counter;
2454 icsbp->icsb_icount = lcounter;
2458 lcounter = icsbp->icsb_ifree;
2460 if (unlikely(lcounter < 0))
2461 goto balance_counter;
2462 icsbp->icsb_ifree = lcounter;
2465 case XFS_SBS_FDBLOCKS:
2466 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2468 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2470 if (unlikely(lcounter < 0))
2471 goto balance_counter;
2472 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2478 xfs_icsb_unlock_cntr(icsbp);
2486 * serialise with a mutex so we don't burn lots of cpu on
2487 * the superblock lock. We still need to hold the superblock
2488 * lock, however, when we modify the global structures.
2493 * Now running atomically.
2495 * If the counter is enabled, someone has beaten us to rebalancing.
2496 * Drop the lock and try again in the fast path....
2498 if (!(xfs_icsb_counter_disabled(mp, field))) {
2499 xfs_icsb_unlock(mp);
2504 * The counter is currently disabled. Because we are
2505 * running atomically here, we know a rebalance cannot
2506 * be in progress. Hence we can go straight to operating
2507 * on the global superblock. We do not call xfs_mod_incore_sb()
2508 * here even though we need to get the m_sb_lock. Doing so
2509 * will cause us to re-enter this function and deadlock.
2510 * Hence we get the m_sb_lock ourselves and then call
2511 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2512 * directly on the global counters.
2514 spin_lock(&mp->m_sb_lock);
2515 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2516 spin_unlock(&mp->m_sb_lock);
2519 * Now that we've modified the global superblock, we
2520 * may be able to re-enable the distributed counters
2521 * (e.g. lots of space just got freed). After that
2525 xfs_icsb_balance_counter(mp, field, 0);
2526 xfs_icsb_unlock(mp);
2530 xfs_icsb_unlock_cntr(icsbp);
2534 * We may have multiple threads here if multiple per-cpu
2535 * counters run dry at the same time. This will mean we can
2536 * do more balances than strictly necessary but it is not
2537 * the common slowpath case.
2542 * running atomically.
2544 * This will leave the counter in the correct state for future
2545 * accesses. After the rebalance, we simply try again and our retry
2546 * will either succeed through the fast path or slow path without
2547 * another balance operation being required.
2549 xfs_icsb_balance_counter(mp, field, delta);
2550 xfs_icsb_unlock(mp);