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);
145 * Check size of device based on the (data/realtime) block count.
146 * Note: this check is used by the growfs code as well as mount.
149 xfs_sb_validate_fsb_count(
153 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
154 ASSERT(sbp->sb_blocklog >= BBSHIFT);
156 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
157 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
159 #else /* Limited by UINT_MAX of sectors */
160 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
167 * Check the validity of the SB found.
170 xfs_mount_validate_sb(
176 * If the log device and data device have the
177 * same device number, the log is internal.
178 * Consequently, the sb_logstart should be non-zero. If
179 * we have a zero sb_logstart in this case, we may be trying to mount
180 * a volume filesystem in a non-volume manner.
182 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
183 xfs_fs_mount_cmn_err(flags, "bad magic number");
184 return XFS_ERROR(EWRONGFS);
187 if (!xfs_sb_good_version(sbp)) {
188 xfs_fs_mount_cmn_err(flags, "bad version");
189 return XFS_ERROR(EWRONGFS);
193 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
194 xfs_fs_mount_cmn_err(flags,
195 "filesystem is marked as having an external log; "
196 "specify logdev on the\nmount command line.");
197 return XFS_ERROR(EINVAL);
201 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
202 xfs_fs_mount_cmn_err(flags,
203 "filesystem is marked as having an internal log; "
204 "do not specify logdev on\nthe mount command line.");
205 return XFS_ERROR(EINVAL);
209 * More sanity checking. These were stolen directly from
213 sbp->sb_agcount <= 0 ||
214 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
215 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
216 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
217 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
218 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
219 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
220 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
221 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
222 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
223 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
224 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
225 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
226 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
227 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
228 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
229 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
230 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
231 return XFS_ERROR(EFSCORRUPTED);
235 * Sanity check AG count, size fields against data size field
238 sbp->sb_dblocks == 0 ||
240 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
241 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
242 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
243 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
244 return XFS_ERROR(EFSCORRUPTED);
248 * Until this is fixed only page-sized or smaller data blocks work.
250 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
251 xfs_fs_mount_cmn_err(flags,
252 "file system with blocksize %d bytes",
254 xfs_fs_mount_cmn_err(flags,
255 "only pagesize (%ld) or less will currently work.",
257 return XFS_ERROR(ENOSYS);
260 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
261 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
262 xfs_fs_mount_cmn_err(flags,
263 "file system too large to be mounted on this system.");
264 return XFS_ERROR(E2BIG);
267 if (unlikely(sbp->sb_inprogress)) {
268 xfs_fs_mount_cmn_err(flags, "file system busy");
269 return XFS_ERROR(EFSCORRUPTED);
273 * Version 1 directory format has never worked on Linux.
275 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
276 xfs_fs_mount_cmn_err(flags,
277 "file system using version 1 directory format");
278 return XFS_ERROR(ENOSYS);
285 xfs_initialize_perag_icache(
288 if (!pag->pag_ici_init) {
289 rwlock_init(&pag->pag_ici_lock);
290 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
291 pag->pag_ici_init = 1;
296 xfs_initialize_perag(
298 xfs_agnumber_t agcount)
300 xfs_agnumber_t index, max_metadata;
304 xfs_sb_t *sbp = &mp->m_sb;
305 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
307 /* Check to see if the filesystem can overflow 32 bit inodes */
308 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
309 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
311 /* Clear the mount flag if no inode can overflow 32 bits
312 * on this filesystem, or if specifically requested..
314 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
315 mp->m_flags |= XFS_MOUNT_32BITINODES;
317 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
320 /* If we can overflow then setup the ag headers accordingly */
321 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
322 /* Calculate how much should be reserved for inodes to
323 * meet the max inode percentage.
325 if (mp->m_maxicount) {
328 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
330 icount += sbp->sb_agblocks - 1;
331 do_div(icount, sbp->sb_agblocks);
332 max_metadata = icount;
334 max_metadata = agcount;
336 for (index = 0; index < agcount; index++) {
337 ino = XFS_AGINO_TO_INO(mp, index, agino);
338 if (ino > max_inum) {
343 /* This ag is preferred for inodes */
344 pag = &mp->m_perag[index];
345 pag->pagi_inodeok = 1;
346 if (index < max_metadata)
347 pag->pagf_metadata = 1;
348 xfs_initialize_perag_icache(pag);
351 /* Setup default behavior for smaller filesystems */
352 for (index = 0; index < agcount; index++) {
353 pag = &mp->m_perag[index];
354 pag->pagi_inodeok = 1;
355 xfs_initialize_perag_icache(pag);
366 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
367 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
368 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
369 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
370 to->sb_rextents = be64_to_cpu(from->sb_rextents);
371 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
372 to->sb_logstart = be64_to_cpu(from->sb_logstart);
373 to->sb_rootino = be64_to_cpu(from->sb_rootino);
374 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
375 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
376 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
377 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
378 to->sb_agcount = be32_to_cpu(from->sb_agcount);
379 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
380 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
381 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
382 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
383 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
384 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
385 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
386 to->sb_blocklog = from->sb_blocklog;
387 to->sb_sectlog = from->sb_sectlog;
388 to->sb_inodelog = from->sb_inodelog;
389 to->sb_inopblog = from->sb_inopblog;
390 to->sb_agblklog = from->sb_agblklog;
391 to->sb_rextslog = from->sb_rextslog;
392 to->sb_inprogress = from->sb_inprogress;
393 to->sb_imax_pct = from->sb_imax_pct;
394 to->sb_icount = be64_to_cpu(from->sb_icount);
395 to->sb_ifree = be64_to_cpu(from->sb_ifree);
396 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
397 to->sb_frextents = be64_to_cpu(from->sb_frextents);
398 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
399 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
400 to->sb_qflags = be16_to_cpu(from->sb_qflags);
401 to->sb_flags = from->sb_flags;
402 to->sb_shared_vn = from->sb_shared_vn;
403 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
404 to->sb_unit = be32_to_cpu(from->sb_unit);
405 to->sb_width = be32_to_cpu(from->sb_width);
406 to->sb_dirblklog = from->sb_dirblklog;
407 to->sb_logsectlog = from->sb_logsectlog;
408 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
409 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
410 to->sb_features2 = be32_to_cpu(from->sb_features2);
411 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
415 * Copy in core superblock to ondisk one.
417 * The fields argument is mask of superblock fields to copy.
425 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
426 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
436 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
437 first = xfs_sb_info[f].offset;
438 size = xfs_sb_info[f + 1].offset - first;
440 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
442 if (size == 1 || xfs_sb_info[f].type == 1) {
443 memcpy(to_ptr + first, from_ptr + first, size);
447 *(__be16 *)(to_ptr + first) =
448 cpu_to_be16(*(__u16 *)(from_ptr + first));
451 *(__be32 *)(to_ptr + first) =
452 cpu_to_be32(*(__u32 *)(from_ptr + first));
455 *(__be64 *)(to_ptr + first) =
456 cpu_to_be64(*(__u64 *)(from_ptr + first));
463 fields &= ~(1LL << f);
470 * Does the initial read of the superblock.
473 xfs_readsb(xfs_mount_t *mp, int flags)
475 unsigned int sector_size;
476 unsigned int extra_flags;
480 ASSERT(mp->m_sb_bp == NULL);
481 ASSERT(mp->m_ddev_targp != NULL);
484 * Allocate a (locked) buffer to hold the superblock.
485 * This will be kept around at all times to optimize
486 * access to the superblock.
488 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
489 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
491 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
492 BTOBB(sector_size), extra_flags);
493 if (!bp || XFS_BUF_ISERROR(bp)) {
494 xfs_fs_mount_cmn_err(flags, "SB read failed");
495 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
498 ASSERT(XFS_BUF_ISBUSY(bp));
499 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
502 * Initialize the mount structure from the superblock.
503 * But first do some basic consistency checking.
505 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
507 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
509 xfs_fs_mount_cmn_err(flags, "SB validate failed");
514 * We must be able to do sector-sized and sector-aligned IO.
516 if (sector_size > mp->m_sb.sb_sectsize) {
517 xfs_fs_mount_cmn_err(flags,
518 "device supports only %u byte sectors (not %u)",
519 sector_size, mp->m_sb.sb_sectsize);
525 * If device sector size is smaller than the superblock size,
526 * re-read the superblock so the buffer is correctly sized.
528 if (sector_size < mp->m_sb.sb_sectsize) {
529 XFS_BUF_UNMANAGE(bp);
531 sector_size = mp->m_sb.sb_sectsize;
532 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
533 BTOBB(sector_size), extra_flags);
534 if (!bp || XFS_BUF_ISERROR(bp)) {
535 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
536 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
539 ASSERT(XFS_BUF_ISBUSY(bp));
540 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
543 /* Initialize per-cpu counters */
544 xfs_icsb_reinit_counters(mp);
548 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
553 XFS_BUF_UNMANAGE(bp);
563 * Mount initialization code establishing various mount
564 * fields from the superblock associated with the given
568 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
570 mp->m_agfrotor = mp->m_agirotor = 0;
571 spin_lock_init(&mp->m_agirotor_lock);
572 mp->m_maxagi = mp->m_sb.sb_agcount;
573 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
574 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
575 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
576 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
577 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
578 mp->m_litino = sbp->sb_inodesize -
579 ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
580 mp->m_blockmask = sbp->sb_blocksize - 1;
581 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
582 mp->m_blockwmask = mp->m_blockwsize - 1;
583 INIT_LIST_HEAD(&mp->m_del_inodes);
586 * Setup for attributes, in case they get created.
587 * This value is for inodes getting attributes for the first time,
588 * the per-inode value is for old attribute values.
590 ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
591 switch (sbp->sb_inodesize) {
593 mp->m_attroffset = XFS_LITINO(mp) -
594 XFS_BMDR_SPACE_CALC(MINABTPTRS);
599 mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
604 ASSERT(mp->m_attroffset < XFS_LITINO(mp));
606 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
607 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
608 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
609 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
611 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
612 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
613 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
614 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
616 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
617 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
618 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
619 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
621 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
622 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
624 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
628 * xfs_initialize_perag_data
630 * Read in each per-ag structure so we can count up the number of
631 * allocated inodes, free inodes and used filesystem blocks as this
632 * information is no longer persistent in the superblock. Once we have
633 * this information, write it into the in-core superblock structure.
636 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
638 xfs_agnumber_t index;
640 xfs_sb_t *sbp = &mp->m_sb;
644 uint64_t bfreelst = 0;
648 for (index = 0; index < agcount; index++) {
650 * read the agf, then the agi. This gets us
651 * all the inforamtion we need and populates the
652 * per-ag structures for us.
654 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
658 error = xfs_ialloc_pagi_init(mp, NULL, index);
661 pag = &mp->m_perag[index];
662 ifree += pag->pagi_freecount;
663 ialloc += pag->pagi_count;
664 bfree += pag->pagf_freeblks;
665 bfreelst += pag->pagf_flcount;
666 btree += pag->pagf_btreeblks;
669 * Overwrite incore superblock counters with just-read data
671 spin_lock(&mp->m_sb_lock);
672 sbp->sb_ifree = ifree;
673 sbp->sb_icount = ialloc;
674 sbp->sb_fdblocks = bfree + bfreelst + btree;
675 spin_unlock(&mp->m_sb_lock);
677 /* Fixup the per-cpu counters as well. */
678 xfs_icsb_reinit_counters(mp);
684 * Update alignment values based on mount options and sb values
687 xfs_update_alignment(xfs_mount_t *mp, __uint64_t *update_flags)
689 xfs_sb_t *sbp = &(mp->m_sb);
693 * If stripe unit and stripe width are not multiples
694 * of the fs blocksize turn off alignment.
696 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
697 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
698 if (mp->m_flags & XFS_MOUNT_RETERR) {
700 "XFS: alignment check 1 failed");
701 return XFS_ERROR(EINVAL);
703 mp->m_dalign = mp->m_swidth = 0;
706 * Convert the stripe unit and width to FSBs.
708 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
709 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
710 if (mp->m_flags & XFS_MOUNT_RETERR) {
711 return XFS_ERROR(EINVAL);
713 xfs_fs_cmn_err(CE_WARN, mp,
714 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
715 mp->m_dalign, mp->m_swidth,
720 } else if (mp->m_dalign) {
721 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
723 if (mp->m_flags & XFS_MOUNT_RETERR) {
724 xfs_fs_cmn_err(CE_WARN, mp,
725 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
728 return XFS_ERROR(EINVAL);
735 * Update superblock with new values
738 if (xfs_sb_version_hasdalign(sbp)) {
739 if (sbp->sb_unit != mp->m_dalign) {
740 sbp->sb_unit = mp->m_dalign;
741 *update_flags |= XFS_SB_UNIT;
743 if (sbp->sb_width != mp->m_swidth) {
744 sbp->sb_width = mp->m_swidth;
745 *update_flags |= XFS_SB_WIDTH;
748 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
749 xfs_sb_version_hasdalign(&mp->m_sb)) {
750 mp->m_dalign = sbp->sb_unit;
751 mp->m_swidth = sbp->sb_width;
758 * Set the maximum inode count for this filesystem
761 xfs_set_maxicount(xfs_mount_t *mp)
763 xfs_sb_t *sbp = &(mp->m_sb);
766 if (sbp->sb_imax_pct) {
768 * Make sure the maximum inode count is a multiple
769 * of the units we allocate inodes in.
771 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
773 do_div(icount, mp->m_ialloc_blks);
774 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
782 * Set the default minimum read and write sizes unless
783 * already specified in a mount option.
784 * We use smaller I/O sizes when the file system
785 * is being used for NFS service (wsync mount option).
788 xfs_set_rw_sizes(xfs_mount_t *mp)
790 xfs_sb_t *sbp = &(mp->m_sb);
791 int readio_log, writeio_log;
793 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
794 if (mp->m_flags & XFS_MOUNT_WSYNC) {
795 readio_log = XFS_WSYNC_READIO_LOG;
796 writeio_log = XFS_WSYNC_WRITEIO_LOG;
798 readio_log = XFS_READIO_LOG_LARGE;
799 writeio_log = XFS_WRITEIO_LOG_LARGE;
802 readio_log = mp->m_readio_log;
803 writeio_log = mp->m_writeio_log;
806 if (sbp->sb_blocklog > readio_log) {
807 mp->m_readio_log = sbp->sb_blocklog;
809 mp->m_readio_log = readio_log;
811 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
812 if (sbp->sb_blocklog > writeio_log) {
813 mp->m_writeio_log = sbp->sb_blocklog;
815 mp->m_writeio_log = writeio_log;
817 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
821 * Set whether we're using inode alignment.
824 xfs_set_inoalignment(xfs_mount_t *mp)
826 if (xfs_sb_version_hasalign(&mp->m_sb) &&
827 mp->m_sb.sb_inoalignmt >=
828 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
829 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
831 mp->m_inoalign_mask = 0;
833 * If we are using stripe alignment, check whether
834 * the stripe unit is a multiple of the inode alignment
836 if (mp->m_dalign && mp->m_inoalign_mask &&
837 !(mp->m_dalign & mp->m_inoalign_mask))
838 mp->m_sinoalign = mp->m_dalign;
844 * Check that the data (and log if separate) are an ok size.
847 xfs_check_sizes(xfs_mount_t *mp)
853 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
854 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
855 cmn_err(CE_WARN, "XFS: size check 1 failed");
856 return XFS_ERROR(E2BIG);
858 error = xfs_read_buf(mp, mp->m_ddev_targp,
859 d - XFS_FSS_TO_BB(mp, 1),
860 XFS_FSS_TO_BB(mp, 1), 0, &bp);
864 cmn_err(CE_WARN, "XFS: size check 2 failed");
866 error = XFS_ERROR(E2BIG);
870 if (mp->m_logdev_targp != mp->m_ddev_targp) {
871 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
872 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
873 cmn_err(CE_WARN, "XFS: size check 3 failed");
874 return XFS_ERROR(E2BIG);
876 error = xfs_read_buf(mp, mp->m_logdev_targp,
877 d - XFS_FSB_TO_BB(mp, 1),
878 XFS_FSB_TO_BB(mp, 1), 0, &bp);
882 cmn_err(CE_WARN, "XFS: size check 3 failed");
884 error = XFS_ERROR(E2BIG);
894 * This function does the following on an initial mount of a file system:
895 * - reads the superblock from disk and init the mount struct
896 * - if we're a 32-bit kernel, do a size check on the superblock
897 * so we don't mount terabyte filesystems
898 * - init mount struct realtime fields
899 * - allocate inode hash table for fs
900 * - init directory manager
901 * - perform recovery and init the log manager
907 xfs_sb_t *sbp = &(mp->m_sb);
910 __int64_t update_flags = 0LL;
911 uint quotamount, quotaflags;
912 int uuid_mounted = 0;
915 xfs_mount_common(mp, sbp);
918 * Check for a mismatched features2 values. Older kernels
919 * read & wrote into the wrong sb offset for sb_features2
920 * on some platforms due to xfs_sb_t not being 64bit size aligned
921 * when sb_features2 was added, which made older superblock
922 * reading/writing routines swap it as a 64-bit value.
924 * For backwards compatibility, we make both slots equal.
926 * If we detect a mismatched field, we OR the set bits into the
927 * existing features2 field in case it has already been modified; we
928 * don't want to lose any features. We then update the bad location
929 * with the ORed value so that older kernels will see any features2
930 * flags, and mark the two fields as needing updates once the
931 * transaction subsystem is online.
933 if (xfs_sb_has_mismatched_features2(sbp)) {
935 "XFS: correcting sb_features alignment problem");
936 sbp->sb_features2 |= sbp->sb_bad_features2;
937 sbp->sb_bad_features2 = sbp->sb_features2;
938 update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
941 * Re-check for ATTR2 in case it was found in bad_features2
944 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
945 !(mp->m_flags & XFS_MOUNT_NOATTR2))
946 mp->m_flags |= XFS_MOUNT_ATTR2;
949 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
950 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
951 xfs_sb_version_removeattr2(&mp->m_sb);
952 update_flags |= XFS_SB_FEATURES2;
954 /* update sb_versionnum for the clearing of the morebits */
955 if (!sbp->sb_features2)
956 update_flags |= XFS_SB_VERSIONNUM;
960 * Check if sb_agblocks is aligned at stripe boundary
961 * If sb_agblocks is NOT aligned turn off m_dalign since
962 * allocator alignment is within an ag, therefore ag has
963 * to be aligned at stripe boundary.
965 error = xfs_update_alignment(mp, &update_flags);
969 xfs_alloc_compute_maxlevels(mp);
970 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
971 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
972 xfs_ialloc_compute_maxlevels(mp);
974 xfs_set_maxicount(mp);
976 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
979 * XFS uses the uuid from the superblock as the unique
980 * identifier for fsid. We can not use the uuid from the volume
981 * since a single partition filesystem is identical to a single
982 * partition volume/filesystem.
984 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
985 if (xfs_uuid_mount(mp)) {
986 error = XFS_ERROR(EINVAL);
993 * Set the minimum read and write sizes
995 xfs_set_rw_sizes(mp);
998 * Set the inode cluster size.
999 * This may still be overridden by the file system
1000 * block size if it is larger than the chosen cluster size.
1002 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1005 * Set inode alignment fields
1007 xfs_set_inoalignment(mp);
1010 * Check that the data (and log if separate) are an ok size.
1012 error = xfs_check_sizes(mp);
1017 * Initialize realtime fields in the mount structure
1019 error = xfs_rtmount_init(mp);
1021 cmn_err(CE_WARN, "XFS: RT mount failed");
1026 * Copies the low order bits of the timestamp and the randomly
1027 * set "sequence" number out of a UUID.
1029 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1031 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1036 * Initialize the attribute manager's entries.
1038 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1041 * Initialize the precomputed transaction reservations values.
1046 * Allocate and initialize the per-ag data.
1048 init_rwsem(&mp->m_peraglock);
1049 mp->m_perag = kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t),
1054 mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
1057 * log's mount-time initialization. Perform 1st part recovery if needed
1059 if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */
1060 error = xfs_log_mount(mp, mp->m_logdev_targp,
1061 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1062 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1064 cmn_err(CE_WARN, "XFS: log mount failed");
1067 } else { /* No log has been defined */
1068 cmn_err(CE_WARN, "XFS: no log defined");
1069 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
1070 error = XFS_ERROR(EFSCORRUPTED);
1075 * Now the log is mounted, we know if it was an unclean shutdown or
1076 * not. If it was, with the first phase of recovery has completed, we
1077 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1078 * but they are recovered transactionally in the second recovery phase
1081 * Hence we can safely re-initialise incore superblock counters from
1082 * the per-ag data. These may not be correct if the filesystem was not
1083 * cleanly unmounted, so we need to wait for recovery to finish before
1086 * If the filesystem was cleanly unmounted, then we can trust the
1087 * values in the superblock to be correct and we don't need to do
1090 * If we are currently making the filesystem, the initialisation will
1091 * fail as the perag data is in an undefined state.
1094 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1095 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1096 !mp->m_sb.sb_inprogress) {
1097 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1103 * Get and sanity-check the root inode.
1104 * Save the pointer to it in the mount structure.
1106 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
1108 cmn_err(CE_WARN, "XFS: failed to read root inode");
1112 ASSERT(rip != NULL);
1114 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1115 cmn_err(CE_WARN, "XFS: corrupted root inode");
1116 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1117 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1118 (unsigned long long)rip->i_ino);
1119 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1120 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1122 error = XFS_ERROR(EFSCORRUPTED);
1125 mp->m_rootip = rip; /* save it */
1127 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1130 * Initialize realtime inode pointers in the mount structure
1132 error = xfs_rtmount_inodes(mp);
1135 * Free up the root inode.
1137 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1142 * If fs is not mounted readonly, then update the superblock changes.
1144 if (update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1145 error = xfs_mount_log_sb(mp, update_flags);
1147 cmn_err(CE_WARN, "XFS: failed to write sb changes");
1153 * Initialise the XFS quota management subsystem for this mount
1155 error = XFS_QM_INIT(mp, "amount, "aflags);
1160 * Finish recovering the file system. This part needed to be
1161 * delayed until after the root and real-time bitmap inodes
1162 * were consistently read in.
1164 error = xfs_log_mount_finish(mp);
1166 cmn_err(CE_WARN, "XFS: log mount finish failed");
1171 * Complete the quota initialisation, post-log-replay component.
1173 error = XFS_QM_MOUNT(mp, quotamount, quotaflags);
1178 * Now we are mounted, reserve a small amount of unused space for
1179 * privileged transactions. This is needed so that transaction
1180 * space required for critical operations can dip into this pool
1181 * when at ENOSPC. This is needed for operations like create with
1182 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1183 * are not allowed to use this reserved space.
1185 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1186 * This may drive us straight to ENOSPC on mount, but that implies
1187 * we were already there on the last unmount. Warn if this occurs.
1189 resblks = mp->m_sb.sb_dblocks;
1190 do_div(resblks, 20);
1191 resblks = min_t(__uint64_t, resblks, 1024);
1192 error = xfs_reserve_blocks(mp, &resblks, NULL);
1194 cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. "
1195 "Continuing without a reserve pool.");
1201 * Free up the root inode.
1205 xfs_log_unmount_dealloc(mp);
1210 uuid_table_remove(&mp->m_sb.sb_uuid);
1215 * This flushes out the inodes,dquots and the superblock, unmounts the
1216 * log and makes sure that incore structures are freed.
1220 struct xfs_mount *mp)
1225 IRELE(mp->m_rootip);
1228 * We can potentially deadlock here if we have an inode cluster
1229 * that has been freed has it's buffer still pinned in memory because
1230 * the transaction is still sitting in a iclog. The stale inodes
1231 * on that buffer will have their flush locks held until the
1232 * transaction hits the disk and the callbacks run. the inode
1233 * flush takes the flush lock unconditionally and with nothing to
1234 * push out the iclog we will never get that unlocked. hence we
1235 * need to force the log first.
1237 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1238 xfs_reclaim_inodes(mp, 0, XFS_IFLUSH_ASYNC);
1240 XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
1242 if (mp->m_quotainfo)
1246 * Flush out the log synchronously so that we know for sure
1247 * that nothing is pinned. This is important because bflush()
1248 * will skip pinned buffers.
1250 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1252 xfs_binval(mp->m_ddev_targp);
1253 if (mp->m_rtdev_targp) {
1254 xfs_binval(mp->m_rtdev_targp);
1258 * Unreserve any blocks we have so that when we unmount we don't account
1259 * the reserved free space as used. This is really only necessary for
1260 * lazy superblock counting because it trusts the incore superblock
1261 * counters to be aboslutely correct on clean unmount.
1263 * We don't bother correcting this elsewhere for lazy superblock
1264 * counting because on mount of an unclean filesystem we reconstruct the
1265 * correct counter value and this is irrelevant.
1267 * For non-lazy counter filesystems, this doesn't matter at all because
1268 * we only every apply deltas to the superblock and hence the incore
1269 * value does not matter....
1272 error = xfs_reserve_blocks(mp, &resblks, NULL);
1274 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1275 "Freespace may not be correct on next mount.");
1277 error = xfs_log_sbcount(mp, 1);
1279 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1280 "Freespace may not be correct on next mount.");
1281 xfs_unmountfs_writesb(mp);
1282 xfs_unmountfs_wait(mp); /* wait for async bufs */
1283 xfs_log_unmount(mp); /* Done! No more fs ops. */
1285 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1286 uuid_table_remove(&mp->m_sb.sb_uuid);
1289 xfs_errortag_clearall(mp, 0);
1295 xfs_unmountfs_wait(xfs_mount_t *mp)
1297 if (mp->m_logdev_targp != mp->m_ddev_targp)
1298 xfs_wait_buftarg(mp->m_logdev_targp);
1299 if (mp->m_rtdev_targp)
1300 xfs_wait_buftarg(mp->m_rtdev_targp);
1301 xfs_wait_buftarg(mp->m_ddev_targp);
1305 xfs_fs_writable(xfs_mount_t *mp)
1307 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1308 (mp->m_flags & XFS_MOUNT_RDONLY));
1314 * Called either periodically to keep the on disk superblock values
1315 * roughly up to date or from unmount to make sure the values are
1316 * correct on a clean unmount.
1318 * Note this code can be called during the process of freezing, so
1319 * we may need to use the transaction allocator which does not not
1320 * block when the transaction subsystem is in its frozen state.
1330 if (!xfs_fs_writable(mp))
1333 xfs_icsb_sync_counters(mp, 0);
1336 * we don't need to do this if we are updating the superblock
1337 * counters on every modification.
1339 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1342 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT);
1343 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1344 XFS_DEFAULT_LOG_COUNT);
1346 xfs_trans_cancel(tp, 0);
1350 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1352 xfs_trans_set_sync(tp);
1353 error = xfs_trans_commit(tp, 0);
1362 xfs_dsb_t *sb = XFS_BUF_TO_SBP(bp);
1365 if (!(sb->sb_flags & XFS_SBF_READONLY))
1366 sb->sb_flags |= XFS_SBF_READONLY;
1368 version = be16_to_cpu(sb->sb_versionnum);
1369 if ((version & XFS_SB_VERSION_NUMBITS) != XFS_SB_VERSION_4 ||
1370 !(version & XFS_SB_VERSION_SHAREDBIT))
1371 version |= XFS_SB_VERSION_SHAREDBIT;
1372 sb->sb_versionnum = cpu_to_be16(version);
1376 xfs_unmountfs_writesb(xfs_mount_t *mp)
1382 * skip superblock write if fs is read-only, or
1383 * if we are doing a forced umount.
1385 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1386 XFS_FORCED_SHUTDOWN(mp))) {
1388 sbp = xfs_getsb(mp, 0);
1391 * mark shared-readonly if desired
1393 if (mp->m_mk_sharedro)
1394 xfs_mark_shared_ro(mp, sbp);
1396 XFS_BUF_UNDONE(sbp);
1397 XFS_BUF_UNREAD(sbp);
1398 XFS_BUF_UNDELAYWRITE(sbp);
1400 XFS_BUF_UNASYNC(sbp);
1401 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1402 xfsbdstrat(mp, sbp);
1403 error = xfs_iowait(sbp);
1405 xfs_ioerror_alert("xfs_unmountfs_writesb",
1406 mp, sbp, XFS_BUF_ADDR(sbp));
1407 if (error && mp->m_mk_sharedro)
1408 xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1415 * xfs_mod_sb() can be used to copy arbitrary changes to the
1416 * in-core superblock into the superblock buffer to be logged.
1417 * It does not provide the higher level of locking that is
1418 * needed to protect the in-core superblock from concurrent
1422 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1434 bp = xfs_trans_getsb(tp, mp, 0);
1435 first = sizeof(xfs_sb_t);
1438 /* translate/copy */
1440 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1442 /* find modified range */
1444 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1445 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1446 first = xfs_sb_info[f].offset;
1448 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1449 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1450 last = xfs_sb_info[f + 1].offset - 1;
1452 xfs_trans_log_buf(tp, bp, first, last);
1457 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1458 * a delta to a specified field in the in-core superblock. Simply
1459 * switch on the field indicated and apply the delta to that field.
1460 * Fields are not allowed to dip below zero, so if the delta would
1461 * do this do not apply it and return EINVAL.
1463 * The m_sb_lock must be held when this routine is called.
1466 xfs_mod_incore_sb_unlocked(
1468 xfs_sb_field_t field,
1472 int scounter; /* short counter for 32 bit fields */
1473 long long lcounter; /* long counter for 64 bit fields */
1474 long long res_used, rem;
1477 * With the in-core superblock spin lock held, switch
1478 * on the indicated field. Apply the delta to the
1479 * proper field. If the fields value would dip below
1480 * 0, then do not apply the delta and return EINVAL.
1483 case XFS_SBS_ICOUNT:
1484 lcounter = (long long)mp->m_sb.sb_icount;
1488 return XFS_ERROR(EINVAL);
1490 mp->m_sb.sb_icount = lcounter;
1493 lcounter = (long long)mp->m_sb.sb_ifree;
1497 return XFS_ERROR(EINVAL);
1499 mp->m_sb.sb_ifree = lcounter;
1501 case XFS_SBS_FDBLOCKS:
1502 lcounter = (long long)
1503 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1504 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1506 if (delta > 0) { /* Putting blocks back */
1507 if (res_used > delta) {
1508 mp->m_resblks_avail += delta;
1510 rem = delta - res_used;
1511 mp->m_resblks_avail = mp->m_resblks;
1514 } else { /* Taking blocks away */
1519 * If were out of blocks, use any available reserved blocks if
1525 lcounter = (long long)mp->m_resblks_avail + delta;
1527 return XFS_ERROR(ENOSPC);
1529 mp->m_resblks_avail = lcounter;
1531 } else { /* not reserved */
1532 return XFS_ERROR(ENOSPC);
1537 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1539 case XFS_SBS_FREXTENTS:
1540 lcounter = (long long)mp->m_sb.sb_frextents;
1543 return XFS_ERROR(ENOSPC);
1545 mp->m_sb.sb_frextents = lcounter;
1547 case XFS_SBS_DBLOCKS:
1548 lcounter = (long long)mp->m_sb.sb_dblocks;
1552 return XFS_ERROR(EINVAL);
1554 mp->m_sb.sb_dblocks = lcounter;
1556 case XFS_SBS_AGCOUNT:
1557 scounter = mp->m_sb.sb_agcount;
1561 return XFS_ERROR(EINVAL);
1563 mp->m_sb.sb_agcount = scounter;
1565 case XFS_SBS_IMAX_PCT:
1566 scounter = mp->m_sb.sb_imax_pct;
1570 return XFS_ERROR(EINVAL);
1572 mp->m_sb.sb_imax_pct = scounter;
1574 case XFS_SBS_REXTSIZE:
1575 scounter = mp->m_sb.sb_rextsize;
1579 return XFS_ERROR(EINVAL);
1581 mp->m_sb.sb_rextsize = scounter;
1583 case XFS_SBS_RBMBLOCKS:
1584 scounter = mp->m_sb.sb_rbmblocks;
1588 return XFS_ERROR(EINVAL);
1590 mp->m_sb.sb_rbmblocks = scounter;
1592 case XFS_SBS_RBLOCKS:
1593 lcounter = (long long)mp->m_sb.sb_rblocks;
1597 return XFS_ERROR(EINVAL);
1599 mp->m_sb.sb_rblocks = lcounter;
1601 case XFS_SBS_REXTENTS:
1602 lcounter = (long long)mp->m_sb.sb_rextents;
1606 return XFS_ERROR(EINVAL);
1608 mp->m_sb.sb_rextents = lcounter;
1610 case XFS_SBS_REXTSLOG:
1611 scounter = mp->m_sb.sb_rextslog;
1615 return XFS_ERROR(EINVAL);
1617 mp->m_sb.sb_rextslog = scounter;
1621 return XFS_ERROR(EINVAL);
1626 * xfs_mod_incore_sb() is used to change a field in the in-core
1627 * superblock structure by the specified delta. This modification
1628 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1629 * routine to do the work.
1634 xfs_sb_field_t field,
1640 /* check for per-cpu counters */
1642 #ifdef HAVE_PERCPU_SB
1643 case XFS_SBS_ICOUNT:
1645 case XFS_SBS_FDBLOCKS:
1646 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1647 status = xfs_icsb_modify_counters(mp, field,
1654 spin_lock(&mp->m_sb_lock);
1655 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1656 spin_unlock(&mp->m_sb_lock);
1664 * xfs_mod_incore_sb_batch() is used to change more than one field
1665 * in the in-core superblock structure at a time. This modification
1666 * is protected by a lock internal to this module. The fields and
1667 * changes to those fields are specified in the array of xfs_mod_sb
1668 * structures passed in.
1670 * Either all of the specified deltas will be applied or none of
1671 * them will. If any modified field dips below 0, then all modifications
1672 * will be backed out and EINVAL will be returned.
1675 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1681 * Loop through the array of mod structures and apply each
1682 * individually. If any fail, then back out all those
1683 * which have already been applied. Do all of this within
1684 * the scope of the m_sb_lock so that all of the changes will
1687 spin_lock(&mp->m_sb_lock);
1689 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1691 * Apply the delta at index n. If it fails, break
1692 * from the loop so we'll fall into the undo loop
1695 switch (msbp->msb_field) {
1696 #ifdef HAVE_PERCPU_SB
1697 case XFS_SBS_ICOUNT:
1699 case XFS_SBS_FDBLOCKS:
1700 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1701 spin_unlock(&mp->m_sb_lock);
1702 status = xfs_icsb_modify_counters(mp,
1704 msbp->msb_delta, rsvd);
1705 spin_lock(&mp->m_sb_lock);
1711 status = xfs_mod_incore_sb_unlocked(mp,
1713 msbp->msb_delta, rsvd);
1723 * If we didn't complete the loop above, then back out
1724 * any changes made to the superblock. If you add code
1725 * between the loop above and here, make sure that you
1726 * preserve the value of status. Loop back until
1727 * we step below the beginning of the array. Make sure
1728 * we don't touch anything back there.
1732 while (msbp >= msb) {
1733 switch (msbp->msb_field) {
1734 #ifdef HAVE_PERCPU_SB
1735 case XFS_SBS_ICOUNT:
1737 case XFS_SBS_FDBLOCKS:
1738 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1739 spin_unlock(&mp->m_sb_lock);
1740 status = xfs_icsb_modify_counters(mp,
1744 spin_lock(&mp->m_sb_lock);
1750 status = xfs_mod_incore_sb_unlocked(mp,
1756 ASSERT(status == 0);
1760 spin_unlock(&mp->m_sb_lock);
1765 * xfs_getsb() is called to obtain the buffer for the superblock.
1766 * The buffer is returned locked and read in from disk.
1767 * The buffer should be released with a call to xfs_brelse().
1769 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1770 * the superblock buffer if it can be locked without sleeping.
1771 * If it can't then we'll return NULL.
1780 ASSERT(mp->m_sb_bp != NULL);
1782 if (flags & XFS_BUF_TRYLOCK) {
1783 if (!XFS_BUF_CPSEMA(bp)) {
1787 XFS_BUF_PSEMA(bp, PRIBIO);
1790 ASSERT(XFS_BUF_ISDONE(bp));
1795 * Used to free the superblock along various error paths.
1804 * Use xfs_getsb() so that the buffer will be locked
1805 * when we call xfs_buf_relse().
1807 bp = xfs_getsb(mp, 0);
1808 XFS_BUF_UNMANAGE(bp);
1814 * See if the UUID is unique among mounted XFS filesystems.
1815 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1821 if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1823 "XFS: Filesystem %s has nil UUID - can't mount",
1827 if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1829 "XFS: Filesystem %s has duplicate UUID - can't mount",
1837 * Used to log changes to the superblock unit and width fields which could
1838 * be altered by the mount options, as well as any potential sb_features2
1839 * fixup. Only the first superblock is updated.
1849 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1850 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1851 XFS_SB_VERSIONNUM));
1853 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1854 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1855 XFS_DEFAULT_LOG_COUNT);
1857 xfs_trans_cancel(tp, 0);
1860 xfs_mod_sb(tp, fields);
1861 error = xfs_trans_commit(tp, 0);
1866 #ifdef HAVE_PERCPU_SB
1868 * Per-cpu incore superblock counters
1870 * Simple concept, difficult implementation
1872 * Basically, replace the incore superblock counters with a distributed per cpu
1873 * counter for contended fields (e.g. free block count).
1875 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1876 * hence needs to be accurately read when we are running low on space. Hence
1877 * there is a method to enable and disable the per-cpu counters based on how
1878 * much "stuff" is available in them.
1880 * Basically, a counter is enabled if there is enough free resource to justify
1881 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1882 * ENOSPC), then we disable the counters to synchronise all callers and
1883 * re-distribute the available resources.
1885 * If, once we redistributed the available resources, we still get a failure,
1886 * we disable the per-cpu counter and go through the slow path.
1888 * The slow path is the current xfs_mod_incore_sb() function. This means that
1889 * when we disable a per-cpu counter, we need to drain it's resources back to
1890 * the global superblock. We do this after disabling the counter to prevent
1891 * more threads from queueing up on the counter.
1893 * Essentially, this means that we still need a lock in the fast path to enable
1894 * synchronisation between the global counters and the per-cpu counters. This
1895 * is not a problem because the lock will be local to a CPU almost all the time
1896 * and have little contention except when we get to ENOSPC conditions.
1898 * Basically, this lock becomes a barrier that enables us to lock out the fast
1899 * path while we do things like enabling and disabling counters and
1900 * synchronising the counters.
1904 * 1. m_sb_lock before picking up per-cpu locks
1905 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1906 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1907 * 4. modifying per-cpu counters requires holding per-cpu lock
1908 * 5. modifying global counters requires holding m_sb_lock
1909 * 6. enabling or disabling a counter requires holding the m_sb_lock
1910 * and _none_ of the per-cpu locks.
1912 * Disabled counters are only ever re-enabled by a balance operation
1913 * that results in more free resources per CPU than a given threshold.
1914 * To ensure counters don't remain disabled, they are rebalanced when
1915 * the global resource goes above a higher threshold (i.e. some hysteresis
1916 * is present to prevent thrashing).
1919 #ifdef CONFIG_HOTPLUG_CPU
1921 * hot-plug CPU notifier support.
1923 * We need a notifier per filesystem as we need to be able to identify
1924 * the filesystem to balance the counters out. This is achieved by
1925 * having a notifier block embedded in the xfs_mount_t and doing pointer
1926 * magic to get the mount pointer from the notifier block address.
1929 xfs_icsb_cpu_notify(
1930 struct notifier_block *nfb,
1931 unsigned long action,
1934 xfs_icsb_cnts_t *cntp;
1937 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1938 cntp = (xfs_icsb_cnts_t *)
1939 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1941 case CPU_UP_PREPARE:
1942 case CPU_UP_PREPARE_FROZEN:
1943 /* Easy Case - initialize the area and locks, and
1944 * then rebalance when online does everything else for us. */
1945 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1948 case CPU_ONLINE_FROZEN:
1950 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
1951 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
1952 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
1953 xfs_icsb_unlock(mp);
1956 case CPU_DEAD_FROZEN:
1957 /* Disable all the counters, then fold the dead cpu's
1958 * count into the total on the global superblock and
1959 * re-enable the counters. */
1961 spin_lock(&mp->m_sb_lock);
1962 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1963 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1964 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1966 mp->m_sb.sb_icount += cntp->icsb_icount;
1967 mp->m_sb.sb_ifree += cntp->icsb_ifree;
1968 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1970 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1972 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
1973 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
1974 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
1975 spin_unlock(&mp->m_sb_lock);
1976 xfs_icsb_unlock(mp);
1982 #endif /* CONFIG_HOTPLUG_CPU */
1985 xfs_icsb_init_counters(
1988 xfs_icsb_cnts_t *cntp;
1991 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1992 if (mp->m_sb_cnts == NULL)
1995 #ifdef CONFIG_HOTPLUG_CPU
1996 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1997 mp->m_icsb_notifier.priority = 0;
1998 register_hotcpu_notifier(&mp->m_icsb_notifier);
1999 #endif /* CONFIG_HOTPLUG_CPU */
2001 for_each_online_cpu(i) {
2002 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2003 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2006 mutex_init(&mp->m_icsb_mutex);
2009 * start with all counters disabled so that the
2010 * initial balance kicks us off correctly
2012 mp->m_icsb_counters = -1;
2017 xfs_icsb_reinit_counters(
2022 * start with all counters disabled so that the
2023 * initial balance kicks us off correctly
2025 mp->m_icsb_counters = -1;
2026 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2027 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2028 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2029 xfs_icsb_unlock(mp);
2033 xfs_icsb_destroy_counters(
2036 if (mp->m_sb_cnts) {
2037 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2038 free_percpu(mp->m_sb_cnts);
2040 mutex_destroy(&mp->m_icsb_mutex);
2045 xfs_icsb_cnts_t *icsbp)
2047 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2053 xfs_icsb_unlock_cntr(
2054 xfs_icsb_cnts_t *icsbp)
2056 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2061 xfs_icsb_lock_all_counters(
2064 xfs_icsb_cnts_t *cntp;
2067 for_each_online_cpu(i) {
2068 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2069 xfs_icsb_lock_cntr(cntp);
2074 xfs_icsb_unlock_all_counters(
2077 xfs_icsb_cnts_t *cntp;
2080 for_each_online_cpu(i) {
2081 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2082 xfs_icsb_unlock_cntr(cntp);
2089 xfs_icsb_cnts_t *cnt,
2092 xfs_icsb_cnts_t *cntp;
2095 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2097 if (!(flags & XFS_ICSB_LAZY_COUNT))
2098 xfs_icsb_lock_all_counters(mp);
2100 for_each_online_cpu(i) {
2101 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2102 cnt->icsb_icount += cntp->icsb_icount;
2103 cnt->icsb_ifree += cntp->icsb_ifree;
2104 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2107 if (!(flags & XFS_ICSB_LAZY_COUNT))
2108 xfs_icsb_unlock_all_counters(mp);
2112 xfs_icsb_counter_disabled(
2114 xfs_sb_field_t field)
2116 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2117 return test_bit(field, &mp->m_icsb_counters);
2121 xfs_icsb_disable_counter(
2123 xfs_sb_field_t field)
2125 xfs_icsb_cnts_t cnt;
2127 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2130 * If we are already disabled, then there is nothing to do
2131 * here. We check before locking all the counters to avoid
2132 * the expensive lock operation when being called in the
2133 * slow path and the counter is already disabled. This is
2134 * safe because the only time we set or clear this state is under
2137 if (xfs_icsb_counter_disabled(mp, field))
2140 xfs_icsb_lock_all_counters(mp);
2141 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2142 /* drain back to superblock */
2144 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2146 case XFS_SBS_ICOUNT:
2147 mp->m_sb.sb_icount = cnt.icsb_icount;
2150 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2152 case XFS_SBS_FDBLOCKS:
2153 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2160 xfs_icsb_unlock_all_counters(mp);
2164 xfs_icsb_enable_counter(
2166 xfs_sb_field_t field,
2170 xfs_icsb_cnts_t *cntp;
2173 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2175 xfs_icsb_lock_all_counters(mp);
2176 for_each_online_cpu(i) {
2177 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2179 case XFS_SBS_ICOUNT:
2180 cntp->icsb_icount = count + resid;
2183 cntp->icsb_ifree = count + resid;
2185 case XFS_SBS_FDBLOCKS:
2186 cntp->icsb_fdblocks = count + resid;
2194 clear_bit(field, &mp->m_icsb_counters);
2195 xfs_icsb_unlock_all_counters(mp);
2199 xfs_icsb_sync_counters_locked(
2203 xfs_icsb_cnts_t cnt;
2205 xfs_icsb_count(mp, &cnt, flags);
2207 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2208 mp->m_sb.sb_icount = cnt.icsb_icount;
2209 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2210 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2211 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2212 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2216 * Accurate update of per-cpu counters to incore superblock
2219 xfs_icsb_sync_counters(
2223 spin_lock(&mp->m_sb_lock);
2224 xfs_icsb_sync_counters_locked(mp, flags);
2225 spin_unlock(&mp->m_sb_lock);
2229 * Balance and enable/disable counters as necessary.
2231 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2232 * chosen to be the same number as single on disk allocation chunk per CPU, and
2233 * free blocks is something far enough zero that we aren't going thrash when we
2234 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2235 * prevent looping endlessly when xfs_alloc_space asks for more than will
2236 * be distributed to a single CPU but each CPU has enough blocks to be
2239 * Note that we can be called when counters are already disabled.
2240 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2241 * prevent locking every per-cpu counter needlessly.
2244 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2245 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2246 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2248 xfs_icsb_balance_counter_locked(
2250 xfs_sb_field_t field,
2253 uint64_t count, resid;
2254 int weight = num_online_cpus();
2255 uint64_t min = (uint64_t)min_per_cpu;
2257 /* disable counter and sync counter */
2258 xfs_icsb_disable_counter(mp, field);
2260 /* update counters - first CPU gets residual*/
2262 case XFS_SBS_ICOUNT:
2263 count = mp->m_sb.sb_icount;
2264 resid = do_div(count, weight);
2265 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2269 count = mp->m_sb.sb_ifree;
2270 resid = do_div(count, weight);
2271 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2274 case XFS_SBS_FDBLOCKS:
2275 count = mp->m_sb.sb_fdblocks;
2276 resid = do_div(count, weight);
2277 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2282 count = resid = 0; /* quiet, gcc */
2286 xfs_icsb_enable_counter(mp, field, count, resid);
2290 xfs_icsb_balance_counter(
2292 xfs_sb_field_t fields,
2295 spin_lock(&mp->m_sb_lock);
2296 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2297 spin_unlock(&mp->m_sb_lock);
2301 xfs_icsb_modify_counters(
2303 xfs_sb_field_t field,
2307 xfs_icsb_cnts_t *icsbp;
2308 long long lcounter; /* long counter for 64 bit fields */
2314 icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2317 * if the counter is disabled, go to slow path
2319 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2321 xfs_icsb_lock_cntr(icsbp);
2322 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2323 xfs_icsb_unlock_cntr(icsbp);
2328 case XFS_SBS_ICOUNT:
2329 lcounter = icsbp->icsb_icount;
2331 if (unlikely(lcounter < 0))
2332 goto balance_counter;
2333 icsbp->icsb_icount = lcounter;
2337 lcounter = icsbp->icsb_ifree;
2339 if (unlikely(lcounter < 0))
2340 goto balance_counter;
2341 icsbp->icsb_ifree = lcounter;
2344 case XFS_SBS_FDBLOCKS:
2345 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2347 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2349 if (unlikely(lcounter < 0))
2350 goto balance_counter;
2351 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2357 xfs_icsb_unlock_cntr(icsbp);
2365 * serialise with a mutex so we don't burn lots of cpu on
2366 * the superblock lock. We still need to hold the superblock
2367 * lock, however, when we modify the global structures.
2372 * Now running atomically.
2374 * If the counter is enabled, someone has beaten us to rebalancing.
2375 * Drop the lock and try again in the fast path....
2377 if (!(xfs_icsb_counter_disabled(mp, field))) {
2378 xfs_icsb_unlock(mp);
2383 * The counter is currently disabled. Because we are
2384 * running atomically here, we know a rebalance cannot
2385 * be in progress. Hence we can go straight to operating
2386 * on the global superblock. We do not call xfs_mod_incore_sb()
2387 * here even though we need to get the m_sb_lock. Doing so
2388 * will cause us to re-enter this function and deadlock.
2389 * Hence we get the m_sb_lock ourselves and then call
2390 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2391 * directly on the global counters.
2393 spin_lock(&mp->m_sb_lock);
2394 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2395 spin_unlock(&mp->m_sb_lock);
2398 * Now that we've modified the global superblock, we
2399 * may be able to re-enable the distributed counters
2400 * (e.g. lots of space just got freed). After that
2404 xfs_icsb_balance_counter(mp, field, 0);
2405 xfs_icsb_unlock(mp);
2409 xfs_icsb_unlock_cntr(icsbp);
2413 * We may have multiple threads here if multiple per-cpu
2414 * counters run dry at the same time. This will mean we can
2415 * do more balances than strictly necessary but it is not
2416 * the common slowpath case.
2421 * running atomically.
2423 * This will leave the counter in the correct state for future
2424 * accesses. After the rebalance, we simply try again and our retry
2425 * will either succeed through the fast path or slow path without
2426 * another balance operation being required.
2428 xfs_icsb_balance_counter(mp, field, delta);
2429 xfs_icsb_unlock(mp);