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"
48 STATIC int xfs_mount_log_sb(xfs_mount_t *, __int64_t);
49 STATIC int xfs_uuid_mount(xfs_mount_t *);
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 }
126 * Free up the resources associated with a mount structure. Assume that
127 * the structure was initially zeroed, so we can tell which fields got
137 for (agno = 0; agno < mp->m_maxagi; agno++)
138 if (mp->m_perag[agno].pagb_list)
139 kmem_free(mp->m_perag[agno].pagb_list);
140 kmem_free(mp->m_perag);
143 spinlock_destroy(&mp->m_ail_lock);
144 spinlock_destroy(&mp->m_sb_lock);
145 mutex_destroy(&mp->m_ilock);
146 mutex_destroy(&mp->m_growlock);
152 * Check size of device based on the (data/realtime) block count.
153 * Note: this check is used by the growfs code as well as mount.
156 xfs_sb_validate_fsb_count(
160 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
161 ASSERT(sbp->sb_blocklog >= BBSHIFT);
163 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
164 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
166 #else /* Limited by UINT_MAX of sectors */
167 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
174 * Check the validity of the SB found.
177 xfs_mount_validate_sb(
183 * If the log device and data device have the
184 * same device number, the log is internal.
185 * Consequently, the sb_logstart should be non-zero. If
186 * we have a zero sb_logstart in this case, we may be trying to mount
187 * a volume filesystem in a non-volume manner.
189 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
190 xfs_fs_mount_cmn_err(flags, "bad magic number");
191 return XFS_ERROR(EWRONGFS);
194 if (!xfs_sb_good_version(sbp)) {
195 xfs_fs_mount_cmn_err(flags, "bad version");
196 return XFS_ERROR(EWRONGFS);
200 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
201 xfs_fs_mount_cmn_err(flags,
202 "filesystem is marked as having an external log; "
203 "specify logdev on the\nmount command line.");
204 return XFS_ERROR(EINVAL);
208 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
209 xfs_fs_mount_cmn_err(flags,
210 "filesystem is marked as having an internal log; "
211 "do not specify logdev on\nthe mount command line.");
212 return XFS_ERROR(EINVAL);
216 * More sanity checking. These were stolen directly from
220 sbp->sb_agcount <= 0 ||
221 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
222 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
223 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
224 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
225 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
226 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
227 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
228 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
229 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
230 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
231 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
232 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
233 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
234 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
235 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
236 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
237 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
238 return XFS_ERROR(EFSCORRUPTED);
242 * Sanity check AG count, size fields against data size field
245 sbp->sb_dblocks == 0 ||
247 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
248 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
249 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
250 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
251 return XFS_ERROR(EFSCORRUPTED);
255 * Until this is fixed only page-sized or smaller data blocks work.
257 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
258 xfs_fs_mount_cmn_err(flags,
259 "file system with blocksize %d bytes",
261 xfs_fs_mount_cmn_err(flags,
262 "only pagesize (%ld) or less will currently work.",
264 return XFS_ERROR(ENOSYS);
267 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
268 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
269 xfs_fs_mount_cmn_err(flags,
270 "file system too large to be mounted on this system.");
271 return XFS_ERROR(E2BIG);
274 if (unlikely(sbp->sb_inprogress)) {
275 xfs_fs_mount_cmn_err(flags, "file system busy");
276 return XFS_ERROR(EFSCORRUPTED);
280 * Version 1 directory format has never worked on Linux.
282 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
283 xfs_fs_mount_cmn_err(flags,
284 "file system using version 1 directory format");
285 return XFS_ERROR(ENOSYS);
292 xfs_initialize_perag_icache(
295 if (!pag->pag_ici_init) {
296 rwlock_init(&pag->pag_ici_lock);
297 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
298 pag->pag_ici_init = 1;
303 xfs_initialize_perag(
305 xfs_agnumber_t agcount)
307 xfs_agnumber_t index, max_metadata;
311 xfs_sb_t *sbp = &mp->m_sb;
312 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
314 /* Check to see if the filesystem can overflow 32 bit inodes */
315 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
316 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
318 /* Clear the mount flag if no inode can overflow 32 bits
319 * on this filesystem, or if specifically requested..
321 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
322 mp->m_flags |= XFS_MOUNT_32BITINODES;
324 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
327 /* If we can overflow then setup the ag headers accordingly */
328 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
329 /* Calculate how much should be reserved for inodes to
330 * meet the max inode percentage.
332 if (mp->m_maxicount) {
335 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
337 icount += sbp->sb_agblocks - 1;
338 do_div(icount, sbp->sb_agblocks);
339 max_metadata = icount;
341 max_metadata = agcount;
343 for (index = 0; index < agcount; index++) {
344 ino = XFS_AGINO_TO_INO(mp, index, agino);
345 if (ino > max_inum) {
350 /* This ag is preferred for inodes */
351 pag = &mp->m_perag[index];
352 pag->pagi_inodeok = 1;
353 if (index < max_metadata)
354 pag->pagf_metadata = 1;
355 xfs_initialize_perag_icache(pag);
358 /* Setup default behavior for smaller filesystems */
359 for (index = 0; index < agcount; index++) {
360 pag = &mp->m_perag[index];
361 pag->pagi_inodeok = 1;
362 xfs_initialize_perag_icache(pag);
373 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
374 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
375 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
376 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
377 to->sb_rextents = be64_to_cpu(from->sb_rextents);
378 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
379 to->sb_logstart = be64_to_cpu(from->sb_logstart);
380 to->sb_rootino = be64_to_cpu(from->sb_rootino);
381 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
382 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
383 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
384 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
385 to->sb_agcount = be32_to_cpu(from->sb_agcount);
386 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
387 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
388 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
389 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
390 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
391 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
392 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
393 to->sb_blocklog = from->sb_blocklog;
394 to->sb_sectlog = from->sb_sectlog;
395 to->sb_inodelog = from->sb_inodelog;
396 to->sb_inopblog = from->sb_inopblog;
397 to->sb_agblklog = from->sb_agblklog;
398 to->sb_rextslog = from->sb_rextslog;
399 to->sb_inprogress = from->sb_inprogress;
400 to->sb_imax_pct = from->sb_imax_pct;
401 to->sb_icount = be64_to_cpu(from->sb_icount);
402 to->sb_ifree = be64_to_cpu(from->sb_ifree);
403 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
404 to->sb_frextents = be64_to_cpu(from->sb_frextents);
405 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
406 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
407 to->sb_qflags = be16_to_cpu(from->sb_qflags);
408 to->sb_flags = from->sb_flags;
409 to->sb_shared_vn = from->sb_shared_vn;
410 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
411 to->sb_unit = be32_to_cpu(from->sb_unit);
412 to->sb_width = be32_to_cpu(from->sb_width);
413 to->sb_dirblklog = from->sb_dirblklog;
414 to->sb_logsectlog = from->sb_logsectlog;
415 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
416 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
417 to->sb_features2 = be32_to_cpu(from->sb_features2);
418 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
422 * Copy in core superblock to ondisk one.
424 * The fields argument is mask of superblock fields to copy.
432 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
433 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
443 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
444 first = xfs_sb_info[f].offset;
445 size = xfs_sb_info[f + 1].offset - first;
447 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
449 if (size == 1 || xfs_sb_info[f].type == 1) {
450 memcpy(to_ptr + first, from_ptr + first, size);
454 *(__be16 *)(to_ptr + first) =
455 cpu_to_be16(*(__u16 *)(from_ptr + first));
458 *(__be32 *)(to_ptr + first) =
459 cpu_to_be32(*(__u32 *)(from_ptr + first));
462 *(__be64 *)(to_ptr + first) =
463 cpu_to_be64(*(__u64 *)(from_ptr + first));
470 fields &= ~(1LL << f);
477 * Does the initial read of the superblock.
480 xfs_readsb(xfs_mount_t *mp, int flags)
482 unsigned int sector_size;
483 unsigned int extra_flags;
487 ASSERT(mp->m_sb_bp == NULL);
488 ASSERT(mp->m_ddev_targp != NULL);
491 * Allocate a (locked) buffer to hold the superblock.
492 * This will be kept around at all times to optimize
493 * access to the superblock.
495 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
496 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
498 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
499 BTOBB(sector_size), extra_flags);
500 if (!bp || XFS_BUF_ISERROR(bp)) {
501 xfs_fs_mount_cmn_err(flags, "SB read failed");
502 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
505 ASSERT(XFS_BUF_ISBUSY(bp));
506 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
509 * Initialize the mount structure from the superblock.
510 * But first do some basic consistency checking.
512 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
514 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
516 xfs_fs_mount_cmn_err(flags, "SB validate failed");
521 * We must be able to do sector-sized and sector-aligned IO.
523 if (sector_size > mp->m_sb.sb_sectsize) {
524 xfs_fs_mount_cmn_err(flags,
525 "device supports only %u byte sectors (not %u)",
526 sector_size, mp->m_sb.sb_sectsize);
532 * If device sector size is smaller than the superblock size,
533 * re-read the superblock so the buffer is correctly sized.
535 if (sector_size < mp->m_sb.sb_sectsize) {
536 XFS_BUF_UNMANAGE(bp);
538 sector_size = mp->m_sb.sb_sectsize;
539 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
540 BTOBB(sector_size), extra_flags);
541 if (!bp || XFS_BUF_ISERROR(bp)) {
542 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
543 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
546 ASSERT(XFS_BUF_ISBUSY(bp));
547 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
550 /* Initialize per-cpu counters */
551 xfs_icsb_reinit_counters(mp);
555 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
560 XFS_BUF_UNMANAGE(bp);
570 * Mount initialization code establishing various mount
571 * fields from the superblock associated with the given
575 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
579 mp->m_agfrotor = mp->m_agirotor = 0;
580 spin_lock_init(&mp->m_agirotor_lock);
581 mp->m_maxagi = mp->m_sb.sb_agcount;
582 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
583 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
584 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
585 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
586 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
587 mp->m_litino = sbp->sb_inodesize -
588 ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
589 mp->m_blockmask = sbp->sb_blocksize - 1;
590 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
591 mp->m_blockwmask = mp->m_blockwsize - 1;
592 INIT_LIST_HEAD(&mp->m_del_inodes);
595 * Setup for attributes, in case they get created.
596 * This value is for inodes getting attributes for the first time,
597 * the per-inode value is for old attribute values.
599 ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
600 switch (sbp->sb_inodesize) {
602 mp->m_attroffset = XFS_LITINO(mp) -
603 XFS_BMDR_SPACE_CALC(MINABTPTRS);
608 mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
613 ASSERT(mp->m_attroffset < XFS_LITINO(mp));
615 for (i = 0; i < 2; i++) {
616 mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
618 mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
621 for (i = 0; i < 2; i++) {
622 mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
624 mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
627 for (i = 0; i < 2; i++) {
628 mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
630 mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
634 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
635 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
637 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
641 * xfs_initialize_perag_data
643 * Read in each per-ag structure so we can count up the number of
644 * allocated inodes, free inodes and used filesystem blocks as this
645 * information is no longer persistent in the superblock. Once we have
646 * this information, write it into the in-core superblock structure.
649 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
651 xfs_agnumber_t index;
653 xfs_sb_t *sbp = &mp->m_sb;
657 uint64_t bfreelst = 0;
661 for (index = 0; index < agcount; index++) {
663 * read the agf, then the agi. This gets us
664 * all the inforamtion we need and populates the
665 * per-ag structures for us.
667 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
671 error = xfs_ialloc_pagi_init(mp, NULL, index);
674 pag = &mp->m_perag[index];
675 ifree += pag->pagi_freecount;
676 ialloc += pag->pagi_count;
677 bfree += pag->pagf_freeblks;
678 bfreelst += pag->pagf_flcount;
679 btree += pag->pagf_btreeblks;
682 * Overwrite incore superblock counters with just-read data
684 spin_lock(&mp->m_sb_lock);
685 sbp->sb_ifree = ifree;
686 sbp->sb_icount = ialloc;
687 sbp->sb_fdblocks = bfree + bfreelst + btree;
688 spin_unlock(&mp->m_sb_lock);
690 /* Fixup the per-cpu counters as well. */
691 xfs_icsb_reinit_counters(mp);
697 * Update alignment values based on mount options and sb values
700 xfs_update_alignment(xfs_mount_t *mp, int mfsi_flags, __uint64_t *update_flags)
702 xfs_sb_t *sbp = &(mp->m_sb);
704 if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
706 * If stripe unit and stripe width are not multiples
707 * of the fs blocksize turn off alignment.
709 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
710 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
711 if (mp->m_flags & XFS_MOUNT_RETERR) {
713 "XFS: alignment check 1 failed");
714 return XFS_ERROR(EINVAL);
716 mp->m_dalign = mp->m_swidth = 0;
719 * Convert the stripe unit and width to FSBs.
721 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
722 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
723 if (mp->m_flags & XFS_MOUNT_RETERR) {
724 return XFS_ERROR(EINVAL);
726 xfs_fs_cmn_err(CE_WARN, mp,
727 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
728 mp->m_dalign, mp->m_swidth,
733 } else if (mp->m_dalign) {
734 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
736 if (mp->m_flags & XFS_MOUNT_RETERR) {
737 xfs_fs_cmn_err(CE_WARN, mp,
738 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
741 return XFS_ERROR(EINVAL);
748 * Update superblock with new values
751 if (xfs_sb_version_hasdalign(sbp)) {
752 if (sbp->sb_unit != mp->m_dalign) {
753 sbp->sb_unit = mp->m_dalign;
754 *update_flags |= XFS_SB_UNIT;
756 if (sbp->sb_width != mp->m_swidth) {
757 sbp->sb_width = mp->m_swidth;
758 *update_flags |= XFS_SB_WIDTH;
761 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
762 xfs_sb_version_hasdalign(&mp->m_sb)) {
763 mp->m_dalign = sbp->sb_unit;
764 mp->m_swidth = sbp->sb_width;
771 * Set the maximum inode count for this filesystem
774 xfs_set_maxicount(xfs_mount_t *mp)
776 xfs_sb_t *sbp = &(mp->m_sb);
779 if (sbp->sb_imax_pct) {
781 * Make sure the maximum inode count is a multiple
782 * of the units we allocate inodes in.
784 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
786 do_div(icount, mp->m_ialloc_blks);
787 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
795 * Set the default minimum read and write sizes unless
796 * already specified in a mount option.
797 * We use smaller I/O sizes when the file system
798 * is being used for NFS service (wsync mount option).
801 xfs_set_rw_sizes(xfs_mount_t *mp)
803 xfs_sb_t *sbp = &(mp->m_sb);
804 int readio_log, writeio_log;
806 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
807 if (mp->m_flags & XFS_MOUNT_WSYNC) {
808 readio_log = XFS_WSYNC_READIO_LOG;
809 writeio_log = XFS_WSYNC_WRITEIO_LOG;
811 readio_log = XFS_READIO_LOG_LARGE;
812 writeio_log = XFS_WRITEIO_LOG_LARGE;
815 readio_log = mp->m_readio_log;
816 writeio_log = mp->m_writeio_log;
819 if (sbp->sb_blocklog > readio_log) {
820 mp->m_readio_log = sbp->sb_blocklog;
822 mp->m_readio_log = readio_log;
824 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
825 if (sbp->sb_blocklog > writeio_log) {
826 mp->m_writeio_log = sbp->sb_blocklog;
828 mp->m_writeio_log = writeio_log;
830 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
834 * Set whether we're using inode alignment.
837 xfs_set_inoalignment(xfs_mount_t *mp)
839 if (xfs_sb_version_hasalign(&mp->m_sb) &&
840 mp->m_sb.sb_inoalignmt >=
841 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
842 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
844 mp->m_inoalign_mask = 0;
846 * If we are using stripe alignment, check whether
847 * the stripe unit is a multiple of the inode alignment
849 if (mp->m_dalign && mp->m_inoalign_mask &&
850 !(mp->m_dalign & mp->m_inoalign_mask))
851 mp->m_sinoalign = mp->m_dalign;
857 * Check that the data (and log if separate) are an ok size.
860 xfs_check_sizes(xfs_mount_t *mp, int mfsi_flags)
866 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
867 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
868 cmn_err(CE_WARN, "XFS: size check 1 failed");
869 return XFS_ERROR(E2BIG);
871 error = xfs_read_buf(mp, mp->m_ddev_targp,
872 d - XFS_FSS_TO_BB(mp, 1),
873 XFS_FSS_TO_BB(mp, 1), 0, &bp);
877 cmn_err(CE_WARN, "XFS: size check 2 failed");
879 error = XFS_ERROR(E2BIG);
883 if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
884 mp->m_logdev_targp != mp->m_ddev_targp) {
885 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
886 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
887 cmn_err(CE_WARN, "XFS: size check 3 failed");
888 return XFS_ERROR(E2BIG);
890 error = xfs_read_buf(mp, mp->m_logdev_targp,
891 d - XFS_FSB_TO_BB(mp, 1),
892 XFS_FSB_TO_BB(mp, 1), 0, &bp);
896 cmn_err(CE_WARN, "XFS: size check 3 failed");
898 error = XFS_ERROR(E2BIG);
908 * This function does the following on an initial mount of a file system:
909 * - reads the superblock from disk and init the mount struct
910 * - if we're a 32-bit kernel, do a size check on the superblock
911 * so we don't mount terabyte filesystems
912 * - init mount struct realtime fields
913 * - allocate inode hash table for fs
914 * - init directory manager
915 * - perform recovery and init the log manager
922 xfs_sb_t *sbp = &(mp->m_sb);
925 __int64_t update_flags = 0LL;
926 uint quotamount, quotaflags;
928 int uuid_mounted = 0;
931 xfs_mount_common(mp, sbp);
934 * Check for a mismatched features2 values. Older kernels
935 * read & wrote into the wrong sb offset for sb_features2
936 * on some platforms due to xfs_sb_t not being 64bit size aligned
937 * when sb_features2 was added, which made older superblock
938 * reading/writing routines swap it as a 64-bit value.
940 * For backwards compatibility, we make both slots equal.
942 * If we detect a mismatched field, we OR the set bits into the
943 * existing features2 field in case it has already been modified; we
944 * don't want to lose any features. We then update the bad location
945 * with the ORed value so that older kernels will see any features2
946 * flags, and mark the two fields as needing updates once the
947 * transaction subsystem is online.
949 if (xfs_sb_has_mismatched_features2(sbp)) {
951 "XFS: correcting sb_features alignment problem");
952 sbp->sb_features2 |= sbp->sb_bad_features2;
953 sbp->sb_bad_features2 = sbp->sb_features2;
954 update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
957 * Re-check for ATTR2 in case it was found in bad_features2
960 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
961 !(mp->m_flags & XFS_MOUNT_NOATTR2))
962 mp->m_flags |= XFS_MOUNT_ATTR2;
965 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
966 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
967 xfs_sb_version_removeattr2(&mp->m_sb);
968 update_flags |= XFS_SB_FEATURES2;
970 /* update sb_versionnum for the clearing of the morebits */
971 if (!sbp->sb_features2)
972 update_flags |= XFS_SB_VERSIONNUM;
976 * Check if sb_agblocks is aligned at stripe boundary
977 * If sb_agblocks is NOT aligned turn off m_dalign since
978 * allocator alignment is within an ag, therefore ag has
979 * to be aligned at stripe boundary.
981 error = xfs_update_alignment(mp, mfsi_flags, &update_flags);
985 xfs_alloc_compute_maxlevels(mp);
986 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
987 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
988 xfs_ialloc_compute_maxlevels(mp);
990 xfs_set_maxicount(mp);
992 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
995 * XFS uses the uuid from the superblock as the unique
996 * identifier for fsid. We can not use the uuid from the volume
997 * since a single partition filesystem is identical to a single
998 * partition volume/filesystem.
1000 if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
1001 (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
1002 if (xfs_uuid_mount(mp)) {
1003 error = XFS_ERROR(EINVAL);
1010 * Set the minimum read and write sizes
1012 xfs_set_rw_sizes(mp);
1015 * Set the inode cluster size.
1016 * This may still be overridden by the file system
1017 * block size if it is larger than the chosen cluster size.
1019 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1022 * Set inode alignment fields
1024 xfs_set_inoalignment(mp);
1027 * Check that the data (and log if separate) are an ok size.
1029 error = xfs_check_sizes(mp, mfsi_flags);
1034 * Initialize realtime fields in the mount structure
1036 error = xfs_rtmount_init(mp);
1038 cmn_err(CE_WARN, "XFS: RT mount failed");
1043 * For client case we are done now
1045 if (mfsi_flags & XFS_MFSI_CLIENT) {
1050 * Copies the low order bits of the timestamp and the randomly
1051 * set "sequence" number out of a UUID.
1053 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1055 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1060 * Initialize the attribute manager's entries.
1062 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1065 * Initialize the precomputed transaction reservations values.
1070 * Allocate and initialize the per-ag data.
1072 init_rwsem(&mp->m_peraglock);
1074 kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
1076 mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
1079 * log's mount-time initialization. Perform 1st part recovery if needed
1081 if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */
1082 error = xfs_log_mount(mp, mp->m_logdev_targp,
1083 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1084 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1086 cmn_err(CE_WARN, "XFS: log mount failed");
1089 } else { /* No log has been defined */
1090 cmn_err(CE_WARN, "XFS: no log defined");
1091 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
1092 error = XFS_ERROR(EFSCORRUPTED);
1097 * Now the log is mounted, we know if it was an unclean shutdown or
1098 * not. If it was, with the first phase of recovery has completed, we
1099 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1100 * but they are recovered transactionally in the second recovery phase
1103 * Hence we can safely re-initialise incore superblock counters from
1104 * the per-ag data. These may not be correct if the filesystem was not
1105 * cleanly unmounted, so we need to wait for recovery to finish before
1108 * If the filesystem was cleanly unmounted, then we can trust the
1109 * values in the superblock to be correct and we don't need to do
1112 * If we are currently making the filesystem, the initialisation will
1113 * fail as the perag data is in an undefined state.
1116 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1117 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1118 !mp->m_sb.sb_inprogress) {
1119 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1125 * Get and sanity-check the root inode.
1126 * Save the pointer to it in the mount structure.
1128 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
1130 cmn_err(CE_WARN, "XFS: failed to read root inode");
1134 ASSERT(rip != NULL);
1136 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1137 cmn_err(CE_WARN, "XFS: corrupted root inode");
1138 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1139 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1140 (unsigned long long)rip->i_ino);
1141 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1142 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1144 error = XFS_ERROR(EFSCORRUPTED);
1147 mp->m_rootip = rip; /* save it */
1149 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1152 * Initialize realtime inode pointers in the mount structure
1154 error = xfs_rtmount_inodes(mp);
1157 * Free up the root inode.
1159 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1164 * If fs is not mounted readonly, then update the superblock changes.
1166 if (update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1167 error = xfs_mount_log_sb(mp, update_flags);
1169 cmn_err(CE_WARN, "XFS: failed to write sb changes");
1175 * Initialise the XFS quota management subsystem for this mount
1177 error = XFS_QM_INIT(mp, "amount, "aflags);
1182 * Finish recovering the file system. This part needed to be
1183 * delayed until after the root and real-time bitmap inodes
1184 * were consistently read in.
1186 error = xfs_log_mount_finish(mp, mfsi_flags);
1188 cmn_err(CE_WARN, "XFS: log mount finish failed");
1193 * Complete the quota initialisation, post-log-replay component.
1195 error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags);
1200 * Now we are mounted, reserve a small amount of unused space for
1201 * privileged transactions. This is needed so that transaction
1202 * space required for critical operations can dip into this pool
1203 * when at ENOSPC. This is needed for operations like create with
1204 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1205 * are not allowed to use this reserved space.
1207 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1208 * This may drive us straight to ENOSPC on mount, but that implies
1209 * we were already there on the last unmount. Warn if this occurs.
1211 resblks = mp->m_sb.sb_dblocks;
1212 do_div(resblks, 20);
1213 resblks = min_t(__uint64_t, resblks, 1024);
1214 error = xfs_reserve_blocks(mp, &resblks, NULL);
1216 cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. "
1217 "Continuing without a reserve pool.");
1223 * Free up the root inode.
1227 xfs_log_unmount_dealloc(mp);
1229 for (agno = 0; agno < sbp->sb_agcount; agno++)
1230 if (mp->m_perag[agno].pagb_list)
1231 kmem_free(mp->m_perag[agno].pagb_list);
1232 kmem_free(mp->m_perag);
1237 uuid_table_remove(&mp->m_sb.sb_uuid);
1244 * This flushes out the inodes,dquots and the superblock, unmounts the
1245 * log and makes sure that incore structures are freed.
1248 xfs_unmountfs(xfs_mount_t *mp)
1254 * We can potentially deadlock here if we have an inode cluster
1255 * that has been freed has it's buffer still pinned in memory because
1256 * the transaction is still sitting in a iclog. The stale inodes
1257 * on that buffer will have their flush locks held until the
1258 * transaction hits the disk and the callbacks run. the inode
1259 * flush takes the flush lock unconditionally and with nothing to
1260 * push out the iclog we will never get that unlocked. hence we
1261 * need to force the log first.
1263 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1266 XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
1269 * Flush out the log synchronously so that we know for sure
1270 * that nothing is pinned. This is important because bflush()
1271 * will skip pinned buffers.
1273 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1275 xfs_binval(mp->m_ddev_targp);
1276 if (mp->m_rtdev_targp) {
1277 xfs_binval(mp->m_rtdev_targp);
1281 * Unreserve any blocks we have so that when we unmount we don't account
1282 * the reserved free space as used. This is really only necessary for
1283 * lazy superblock counting because it trusts the incore superblock
1284 * counters to be aboslutely correct on clean unmount.
1286 * We don't bother correcting this elsewhere for lazy superblock
1287 * counting because on mount of an unclean filesystem we reconstruct the
1288 * correct counter value and this is irrelevant.
1290 * For non-lazy counter filesystems, this doesn't matter at all because
1291 * we only every apply deltas to the superblock and hence the incore
1292 * value does not matter....
1295 error = xfs_reserve_blocks(mp, &resblks, NULL);
1297 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1298 "Freespace may not be correct on next mount.");
1300 error = xfs_log_sbcount(mp, 1);
1302 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1303 "Freespace may not be correct on next mount.");
1304 xfs_unmountfs_writesb(mp);
1305 xfs_unmountfs_wait(mp); /* wait for async bufs */
1306 xfs_log_unmount(mp); /* Done! No more fs ops. */
1311 * All inodes from this mount point should be freed.
1313 ASSERT(mp->m_inodes == NULL);
1315 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1316 uuid_table_remove(&mp->m_sb.sb_uuid);
1319 xfs_errortag_clearall(mp, 0);
1326 xfs_unmountfs_wait(xfs_mount_t *mp)
1328 if (mp->m_logdev_targp != mp->m_ddev_targp)
1329 xfs_wait_buftarg(mp->m_logdev_targp);
1330 if (mp->m_rtdev_targp)
1331 xfs_wait_buftarg(mp->m_rtdev_targp);
1332 xfs_wait_buftarg(mp->m_ddev_targp);
1336 xfs_fs_writable(xfs_mount_t *mp)
1338 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1339 (mp->m_flags & XFS_MOUNT_RDONLY));
1345 * Called either periodically to keep the on disk superblock values
1346 * roughly up to date or from unmount to make sure the values are
1347 * correct on a clean unmount.
1349 * Note this code can be called during the process of freezing, so
1350 * we may need to use the transaction allocator which does not not
1351 * block when the transaction subsystem is in its frozen state.
1361 if (!xfs_fs_writable(mp))
1364 xfs_icsb_sync_counters(mp, 0);
1367 * we don't need to do this if we are updating the superblock
1368 * counters on every modification.
1370 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1373 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT);
1374 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1375 XFS_DEFAULT_LOG_COUNT);
1377 xfs_trans_cancel(tp, 0);
1381 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1383 xfs_trans_set_sync(tp);
1384 error = xfs_trans_commit(tp, 0);
1393 xfs_dsb_t *sb = XFS_BUF_TO_SBP(bp);
1396 if (!(sb->sb_flags & XFS_SBF_READONLY))
1397 sb->sb_flags |= XFS_SBF_READONLY;
1399 version = be16_to_cpu(sb->sb_versionnum);
1400 if ((version & XFS_SB_VERSION_NUMBITS) != XFS_SB_VERSION_4 ||
1401 !(version & XFS_SB_VERSION_SHAREDBIT))
1402 version |= XFS_SB_VERSION_SHAREDBIT;
1403 sb->sb_versionnum = cpu_to_be16(version);
1407 xfs_unmountfs_writesb(xfs_mount_t *mp)
1413 * skip superblock write if fs is read-only, or
1414 * if we are doing a forced umount.
1416 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1417 XFS_FORCED_SHUTDOWN(mp))) {
1419 sbp = xfs_getsb(mp, 0);
1422 * mark shared-readonly if desired
1424 if (mp->m_mk_sharedro)
1425 xfs_mark_shared_ro(mp, sbp);
1427 XFS_BUF_UNDONE(sbp);
1428 XFS_BUF_UNREAD(sbp);
1429 XFS_BUF_UNDELAYWRITE(sbp);
1431 XFS_BUF_UNASYNC(sbp);
1432 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1433 xfsbdstrat(mp, sbp);
1434 error = xfs_iowait(sbp);
1436 xfs_ioerror_alert("xfs_unmountfs_writesb",
1437 mp, sbp, XFS_BUF_ADDR(sbp));
1438 if (error && mp->m_mk_sharedro)
1439 xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1446 * xfs_mod_sb() can be used to copy arbitrary changes to the
1447 * in-core superblock into the superblock buffer to be logged.
1448 * It does not provide the higher level of locking that is
1449 * needed to protect the in-core superblock from concurrent
1453 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1465 bp = xfs_trans_getsb(tp, mp, 0);
1466 first = sizeof(xfs_sb_t);
1469 /* translate/copy */
1471 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1473 /* find modified range */
1475 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1476 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1477 first = xfs_sb_info[f].offset;
1479 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1480 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1481 last = xfs_sb_info[f + 1].offset - 1;
1483 xfs_trans_log_buf(tp, bp, first, last);
1488 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1489 * a delta to a specified field in the in-core superblock. Simply
1490 * switch on the field indicated and apply the delta to that field.
1491 * Fields are not allowed to dip below zero, so if the delta would
1492 * do this do not apply it and return EINVAL.
1494 * The m_sb_lock must be held when this routine is called.
1497 xfs_mod_incore_sb_unlocked(
1499 xfs_sb_field_t field,
1503 int scounter; /* short counter for 32 bit fields */
1504 long long lcounter; /* long counter for 64 bit fields */
1505 long long res_used, rem;
1508 * With the in-core superblock spin lock held, switch
1509 * on the indicated field. Apply the delta to the
1510 * proper field. If the fields value would dip below
1511 * 0, then do not apply the delta and return EINVAL.
1514 case XFS_SBS_ICOUNT:
1515 lcounter = (long long)mp->m_sb.sb_icount;
1519 return XFS_ERROR(EINVAL);
1521 mp->m_sb.sb_icount = lcounter;
1524 lcounter = (long long)mp->m_sb.sb_ifree;
1528 return XFS_ERROR(EINVAL);
1530 mp->m_sb.sb_ifree = lcounter;
1532 case XFS_SBS_FDBLOCKS:
1533 lcounter = (long long)
1534 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1535 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1537 if (delta > 0) { /* Putting blocks back */
1538 if (res_used > delta) {
1539 mp->m_resblks_avail += delta;
1541 rem = delta - res_used;
1542 mp->m_resblks_avail = mp->m_resblks;
1545 } else { /* Taking blocks away */
1550 * If were out of blocks, use any available reserved blocks if
1556 lcounter = (long long)mp->m_resblks_avail + delta;
1558 return XFS_ERROR(ENOSPC);
1560 mp->m_resblks_avail = lcounter;
1562 } else { /* not reserved */
1563 return XFS_ERROR(ENOSPC);
1568 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1570 case XFS_SBS_FREXTENTS:
1571 lcounter = (long long)mp->m_sb.sb_frextents;
1574 return XFS_ERROR(ENOSPC);
1576 mp->m_sb.sb_frextents = lcounter;
1578 case XFS_SBS_DBLOCKS:
1579 lcounter = (long long)mp->m_sb.sb_dblocks;
1583 return XFS_ERROR(EINVAL);
1585 mp->m_sb.sb_dblocks = lcounter;
1587 case XFS_SBS_AGCOUNT:
1588 scounter = mp->m_sb.sb_agcount;
1592 return XFS_ERROR(EINVAL);
1594 mp->m_sb.sb_agcount = scounter;
1596 case XFS_SBS_IMAX_PCT:
1597 scounter = mp->m_sb.sb_imax_pct;
1601 return XFS_ERROR(EINVAL);
1603 mp->m_sb.sb_imax_pct = scounter;
1605 case XFS_SBS_REXTSIZE:
1606 scounter = mp->m_sb.sb_rextsize;
1610 return XFS_ERROR(EINVAL);
1612 mp->m_sb.sb_rextsize = scounter;
1614 case XFS_SBS_RBMBLOCKS:
1615 scounter = mp->m_sb.sb_rbmblocks;
1619 return XFS_ERROR(EINVAL);
1621 mp->m_sb.sb_rbmblocks = scounter;
1623 case XFS_SBS_RBLOCKS:
1624 lcounter = (long long)mp->m_sb.sb_rblocks;
1628 return XFS_ERROR(EINVAL);
1630 mp->m_sb.sb_rblocks = lcounter;
1632 case XFS_SBS_REXTENTS:
1633 lcounter = (long long)mp->m_sb.sb_rextents;
1637 return XFS_ERROR(EINVAL);
1639 mp->m_sb.sb_rextents = lcounter;
1641 case XFS_SBS_REXTSLOG:
1642 scounter = mp->m_sb.sb_rextslog;
1646 return XFS_ERROR(EINVAL);
1648 mp->m_sb.sb_rextslog = scounter;
1652 return XFS_ERROR(EINVAL);
1657 * xfs_mod_incore_sb() is used to change a field in the in-core
1658 * superblock structure by the specified delta. This modification
1659 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1660 * routine to do the work.
1665 xfs_sb_field_t field,
1671 /* check for per-cpu counters */
1673 #ifdef HAVE_PERCPU_SB
1674 case XFS_SBS_ICOUNT:
1676 case XFS_SBS_FDBLOCKS:
1677 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1678 status = xfs_icsb_modify_counters(mp, field,
1685 spin_lock(&mp->m_sb_lock);
1686 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1687 spin_unlock(&mp->m_sb_lock);
1695 * xfs_mod_incore_sb_batch() is used to change more than one field
1696 * in the in-core superblock structure at a time. This modification
1697 * is protected by a lock internal to this module. The fields and
1698 * changes to those fields are specified in the array of xfs_mod_sb
1699 * structures passed in.
1701 * Either all of the specified deltas will be applied or none of
1702 * them will. If any modified field dips below 0, then all modifications
1703 * will be backed out and EINVAL will be returned.
1706 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1712 * Loop through the array of mod structures and apply each
1713 * individually. If any fail, then back out all those
1714 * which have already been applied. Do all of this within
1715 * the scope of the m_sb_lock so that all of the changes will
1718 spin_lock(&mp->m_sb_lock);
1720 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1722 * Apply the delta at index n. If it fails, break
1723 * from the loop so we'll fall into the undo loop
1726 switch (msbp->msb_field) {
1727 #ifdef HAVE_PERCPU_SB
1728 case XFS_SBS_ICOUNT:
1730 case XFS_SBS_FDBLOCKS:
1731 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1732 spin_unlock(&mp->m_sb_lock);
1733 status = xfs_icsb_modify_counters(mp,
1735 msbp->msb_delta, rsvd);
1736 spin_lock(&mp->m_sb_lock);
1742 status = xfs_mod_incore_sb_unlocked(mp,
1744 msbp->msb_delta, rsvd);
1754 * If we didn't complete the loop above, then back out
1755 * any changes made to the superblock. If you add code
1756 * between the loop above and here, make sure that you
1757 * preserve the value of status. Loop back until
1758 * we step below the beginning of the array. Make sure
1759 * we don't touch anything back there.
1763 while (msbp >= msb) {
1764 switch (msbp->msb_field) {
1765 #ifdef HAVE_PERCPU_SB
1766 case XFS_SBS_ICOUNT:
1768 case XFS_SBS_FDBLOCKS:
1769 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1770 spin_unlock(&mp->m_sb_lock);
1771 status = xfs_icsb_modify_counters(mp,
1775 spin_lock(&mp->m_sb_lock);
1781 status = xfs_mod_incore_sb_unlocked(mp,
1787 ASSERT(status == 0);
1791 spin_unlock(&mp->m_sb_lock);
1796 * xfs_getsb() is called to obtain the buffer for the superblock.
1797 * The buffer is returned locked and read in from disk.
1798 * The buffer should be released with a call to xfs_brelse().
1800 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1801 * the superblock buffer if it can be locked without sleeping.
1802 * If it can't then we'll return NULL.
1811 ASSERT(mp->m_sb_bp != NULL);
1813 if (flags & XFS_BUF_TRYLOCK) {
1814 if (!XFS_BUF_CPSEMA(bp)) {
1818 XFS_BUF_PSEMA(bp, PRIBIO);
1821 ASSERT(XFS_BUF_ISDONE(bp));
1826 * Used to free the superblock along various error paths.
1835 * Use xfs_getsb() so that the buffer will be locked
1836 * when we call xfs_buf_relse().
1838 bp = xfs_getsb(mp, 0);
1839 XFS_BUF_UNMANAGE(bp);
1845 * See if the UUID is unique among mounted XFS filesystems.
1846 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1852 if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1854 "XFS: Filesystem %s has nil UUID - can't mount",
1858 if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1860 "XFS: Filesystem %s has duplicate UUID - can't mount",
1868 * Used to log changes to the superblock unit and width fields which could
1869 * be altered by the mount options, as well as any potential sb_features2
1870 * fixup. Only the first superblock is updated.
1880 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1881 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1882 XFS_SB_VERSIONNUM));
1884 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1885 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1886 XFS_DEFAULT_LOG_COUNT);
1888 xfs_trans_cancel(tp, 0);
1891 xfs_mod_sb(tp, fields);
1892 error = xfs_trans_commit(tp, 0);
1897 #ifdef HAVE_PERCPU_SB
1899 * Per-cpu incore superblock counters
1901 * Simple concept, difficult implementation
1903 * Basically, replace the incore superblock counters with a distributed per cpu
1904 * counter for contended fields (e.g. free block count).
1906 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1907 * hence needs to be accurately read when we are running low on space. Hence
1908 * there is a method to enable and disable the per-cpu counters based on how
1909 * much "stuff" is available in them.
1911 * Basically, a counter is enabled if there is enough free resource to justify
1912 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1913 * ENOSPC), then we disable the counters to synchronise all callers and
1914 * re-distribute the available resources.
1916 * If, once we redistributed the available resources, we still get a failure,
1917 * we disable the per-cpu counter and go through the slow path.
1919 * The slow path is the current xfs_mod_incore_sb() function. This means that
1920 * when we disable a per-cpu counter, we need to drain it's resources back to
1921 * the global superblock. We do this after disabling the counter to prevent
1922 * more threads from queueing up on the counter.
1924 * Essentially, this means that we still need a lock in the fast path to enable
1925 * synchronisation between the global counters and the per-cpu counters. This
1926 * is not a problem because the lock will be local to a CPU almost all the time
1927 * and have little contention except when we get to ENOSPC conditions.
1929 * Basically, this lock becomes a barrier that enables us to lock out the fast
1930 * path while we do things like enabling and disabling counters and
1931 * synchronising the counters.
1935 * 1. m_sb_lock before picking up per-cpu locks
1936 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1937 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1938 * 4. modifying per-cpu counters requires holding per-cpu lock
1939 * 5. modifying global counters requires holding m_sb_lock
1940 * 6. enabling or disabling a counter requires holding the m_sb_lock
1941 * and _none_ of the per-cpu locks.
1943 * Disabled counters are only ever re-enabled by a balance operation
1944 * that results in more free resources per CPU than a given threshold.
1945 * To ensure counters don't remain disabled, they are rebalanced when
1946 * the global resource goes above a higher threshold (i.e. some hysteresis
1947 * is present to prevent thrashing).
1950 #ifdef CONFIG_HOTPLUG_CPU
1952 * hot-plug CPU notifier support.
1954 * We need a notifier per filesystem as we need to be able to identify
1955 * the filesystem to balance the counters out. This is achieved by
1956 * having a notifier block embedded in the xfs_mount_t and doing pointer
1957 * magic to get the mount pointer from the notifier block address.
1960 xfs_icsb_cpu_notify(
1961 struct notifier_block *nfb,
1962 unsigned long action,
1965 xfs_icsb_cnts_t *cntp;
1968 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1969 cntp = (xfs_icsb_cnts_t *)
1970 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1972 case CPU_UP_PREPARE:
1973 case CPU_UP_PREPARE_FROZEN:
1974 /* Easy Case - initialize the area and locks, and
1975 * then rebalance when online does everything else for us. */
1976 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1979 case CPU_ONLINE_FROZEN:
1981 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1982 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1983 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1984 xfs_icsb_unlock(mp);
1987 case CPU_DEAD_FROZEN:
1988 /* Disable all the counters, then fold the dead cpu's
1989 * count into the total on the global superblock and
1990 * re-enable the counters. */
1992 spin_lock(&mp->m_sb_lock);
1993 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1994 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1995 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1997 mp->m_sb.sb_icount += cntp->icsb_icount;
1998 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1999 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2001 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2003 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2004 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2005 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2006 spin_unlock(&mp->m_sb_lock);
2007 xfs_icsb_unlock(mp);
2013 #endif /* CONFIG_HOTPLUG_CPU */
2016 xfs_icsb_init_counters(
2019 xfs_icsb_cnts_t *cntp;
2022 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2023 if (mp->m_sb_cnts == NULL)
2026 #ifdef CONFIG_HOTPLUG_CPU
2027 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2028 mp->m_icsb_notifier.priority = 0;
2029 register_hotcpu_notifier(&mp->m_icsb_notifier);
2030 #endif /* CONFIG_HOTPLUG_CPU */
2032 for_each_online_cpu(i) {
2033 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2034 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2037 mutex_init(&mp->m_icsb_mutex);
2040 * start with all counters disabled so that the
2041 * initial balance kicks us off correctly
2043 mp->m_icsb_counters = -1;
2048 xfs_icsb_reinit_counters(
2053 * start with all counters disabled so that the
2054 * initial balance kicks us off correctly
2056 mp->m_icsb_counters = -1;
2057 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2058 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2059 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2060 xfs_icsb_unlock(mp);
2064 xfs_icsb_destroy_counters(
2067 if (mp->m_sb_cnts) {
2068 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2069 free_percpu(mp->m_sb_cnts);
2071 mutex_destroy(&mp->m_icsb_mutex);
2076 xfs_icsb_cnts_t *icsbp)
2078 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2084 xfs_icsb_unlock_cntr(
2085 xfs_icsb_cnts_t *icsbp)
2087 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2092 xfs_icsb_lock_all_counters(
2095 xfs_icsb_cnts_t *cntp;
2098 for_each_online_cpu(i) {
2099 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2100 xfs_icsb_lock_cntr(cntp);
2105 xfs_icsb_unlock_all_counters(
2108 xfs_icsb_cnts_t *cntp;
2111 for_each_online_cpu(i) {
2112 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2113 xfs_icsb_unlock_cntr(cntp);
2120 xfs_icsb_cnts_t *cnt,
2123 xfs_icsb_cnts_t *cntp;
2126 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2128 if (!(flags & XFS_ICSB_LAZY_COUNT))
2129 xfs_icsb_lock_all_counters(mp);
2131 for_each_online_cpu(i) {
2132 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2133 cnt->icsb_icount += cntp->icsb_icount;
2134 cnt->icsb_ifree += cntp->icsb_ifree;
2135 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2138 if (!(flags & XFS_ICSB_LAZY_COUNT))
2139 xfs_icsb_unlock_all_counters(mp);
2143 xfs_icsb_counter_disabled(
2145 xfs_sb_field_t field)
2147 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2148 return test_bit(field, &mp->m_icsb_counters);
2152 xfs_icsb_disable_counter(
2154 xfs_sb_field_t field)
2156 xfs_icsb_cnts_t cnt;
2158 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2161 * If we are already disabled, then there is nothing to do
2162 * here. We check before locking all the counters to avoid
2163 * the expensive lock operation when being called in the
2164 * slow path and the counter is already disabled. This is
2165 * safe because the only time we set or clear this state is under
2168 if (xfs_icsb_counter_disabled(mp, field))
2171 xfs_icsb_lock_all_counters(mp);
2172 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2173 /* drain back to superblock */
2175 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2177 case XFS_SBS_ICOUNT:
2178 mp->m_sb.sb_icount = cnt.icsb_icount;
2181 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2183 case XFS_SBS_FDBLOCKS:
2184 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2191 xfs_icsb_unlock_all_counters(mp);
2195 xfs_icsb_enable_counter(
2197 xfs_sb_field_t field,
2201 xfs_icsb_cnts_t *cntp;
2204 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2206 xfs_icsb_lock_all_counters(mp);
2207 for_each_online_cpu(i) {
2208 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2210 case XFS_SBS_ICOUNT:
2211 cntp->icsb_icount = count + resid;
2214 cntp->icsb_ifree = count + resid;
2216 case XFS_SBS_FDBLOCKS:
2217 cntp->icsb_fdblocks = count + resid;
2225 clear_bit(field, &mp->m_icsb_counters);
2226 xfs_icsb_unlock_all_counters(mp);
2230 xfs_icsb_sync_counters_locked(
2234 xfs_icsb_cnts_t cnt;
2236 xfs_icsb_count(mp, &cnt, flags);
2238 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2239 mp->m_sb.sb_icount = cnt.icsb_icount;
2240 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2241 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2242 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2243 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2247 * Accurate update of per-cpu counters to incore superblock
2250 xfs_icsb_sync_counters(
2254 spin_lock(&mp->m_sb_lock);
2255 xfs_icsb_sync_counters_locked(mp, flags);
2256 spin_unlock(&mp->m_sb_lock);
2260 * Balance and enable/disable counters as necessary.
2262 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2263 * chosen to be the same number as single on disk allocation chunk per CPU, and
2264 * free blocks is something far enough zero that we aren't going thrash when we
2265 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2266 * prevent looping endlessly when xfs_alloc_space asks for more than will
2267 * be distributed to a single CPU but each CPU has enough blocks to be
2270 * Note that we can be called when counters are already disabled.
2271 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2272 * prevent locking every per-cpu counter needlessly.
2275 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2276 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2277 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2279 xfs_icsb_balance_counter_locked(
2281 xfs_sb_field_t field,
2284 uint64_t count, resid;
2285 int weight = num_online_cpus();
2286 uint64_t min = (uint64_t)min_per_cpu;
2288 /* disable counter and sync counter */
2289 xfs_icsb_disable_counter(mp, field);
2291 /* update counters - first CPU gets residual*/
2293 case XFS_SBS_ICOUNT:
2294 count = mp->m_sb.sb_icount;
2295 resid = do_div(count, weight);
2296 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2300 count = mp->m_sb.sb_ifree;
2301 resid = do_div(count, weight);
2302 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2305 case XFS_SBS_FDBLOCKS:
2306 count = mp->m_sb.sb_fdblocks;
2307 resid = do_div(count, weight);
2308 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2313 count = resid = 0; /* quiet, gcc */
2317 xfs_icsb_enable_counter(mp, field, count, resid);
2321 xfs_icsb_balance_counter(
2323 xfs_sb_field_t fields,
2326 spin_lock(&mp->m_sb_lock);
2327 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2328 spin_unlock(&mp->m_sb_lock);
2332 xfs_icsb_modify_counters(
2334 xfs_sb_field_t field,
2338 xfs_icsb_cnts_t *icsbp;
2339 long long lcounter; /* long counter for 64 bit fields */
2345 icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2348 * if the counter is disabled, go to slow path
2350 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2352 xfs_icsb_lock_cntr(icsbp);
2353 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2354 xfs_icsb_unlock_cntr(icsbp);
2359 case XFS_SBS_ICOUNT:
2360 lcounter = icsbp->icsb_icount;
2362 if (unlikely(lcounter < 0))
2363 goto balance_counter;
2364 icsbp->icsb_icount = lcounter;
2368 lcounter = icsbp->icsb_ifree;
2370 if (unlikely(lcounter < 0))
2371 goto balance_counter;
2372 icsbp->icsb_ifree = lcounter;
2375 case XFS_SBS_FDBLOCKS:
2376 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2378 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2380 if (unlikely(lcounter < 0))
2381 goto balance_counter;
2382 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2388 xfs_icsb_unlock_cntr(icsbp);
2396 * serialise with a mutex so we don't burn lots of cpu on
2397 * the superblock lock. We still need to hold the superblock
2398 * lock, however, when we modify the global structures.
2403 * Now running atomically.
2405 * If the counter is enabled, someone has beaten us to rebalancing.
2406 * Drop the lock and try again in the fast path....
2408 if (!(xfs_icsb_counter_disabled(mp, field))) {
2409 xfs_icsb_unlock(mp);
2414 * The counter is currently disabled. Because we are
2415 * running atomically here, we know a rebalance cannot
2416 * be in progress. Hence we can go straight to operating
2417 * on the global superblock. We do not call xfs_mod_incore_sb()
2418 * here even though we need to get the m_sb_lock. Doing so
2419 * will cause us to re-enter this function and deadlock.
2420 * Hence we get the m_sb_lock ourselves and then call
2421 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2422 * directly on the global counters.
2424 spin_lock(&mp->m_sb_lock);
2425 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2426 spin_unlock(&mp->m_sb_lock);
2429 * Now that we've modified the global superblock, we
2430 * may be able to re-enable the distributed counters
2431 * (e.g. lots of space just got freed). After that
2435 xfs_icsb_balance_counter(mp, field, 0);
2436 xfs_icsb_unlock(mp);
2440 xfs_icsb_unlock_cntr(icsbp);
2444 * We may have multiple threads here if multiple per-cpu
2445 * counters run dry at the same time. This will mean we can
2446 * do more balances than strictly necessary but it is not
2447 * the common slowpath case.
2452 * running atomically.
2454 * This will leave the counter in the correct state for future
2455 * accesses. After the rebalance, we simply try again and our retry
2456 * will either succeed through the fast path or slow path without
2457 * another balance operation being required.
2459 xfs_icsb_balance_counter(mp, field, delta);
2460 xfs_icsb_unlock(mp);