2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_fsops.h"
46 #include "xfs_utils.h"
47 #include "xfs_trace.h"
50 STATIC void xfs_unmountfs_wait(xfs_mount_t *);
54 STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
56 STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
58 STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
60 STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
64 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
65 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
66 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
72 short type; /* 0 = integer
73 * 1 = binary / string (no translation)
76 { offsetof(xfs_sb_t, sb_magicnum), 0 },
77 { offsetof(xfs_sb_t, sb_blocksize), 0 },
78 { offsetof(xfs_sb_t, sb_dblocks), 0 },
79 { offsetof(xfs_sb_t, sb_rblocks), 0 },
80 { offsetof(xfs_sb_t, sb_rextents), 0 },
81 { offsetof(xfs_sb_t, sb_uuid), 1 },
82 { offsetof(xfs_sb_t, sb_logstart), 0 },
83 { offsetof(xfs_sb_t, sb_rootino), 0 },
84 { offsetof(xfs_sb_t, sb_rbmino), 0 },
85 { offsetof(xfs_sb_t, sb_rsumino), 0 },
86 { offsetof(xfs_sb_t, sb_rextsize), 0 },
87 { offsetof(xfs_sb_t, sb_agblocks), 0 },
88 { offsetof(xfs_sb_t, sb_agcount), 0 },
89 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
90 { offsetof(xfs_sb_t, sb_logblocks), 0 },
91 { offsetof(xfs_sb_t, sb_versionnum), 0 },
92 { offsetof(xfs_sb_t, sb_sectsize), 0 },
93 { offsetof(xfs_sb_t, sb_inodesize), 0 },
94 { offsetof(xfs_sb_t, sb_inopblock), 0 },
95 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
96 { offsetof(xfs_sb_t, sb_blocklog), 0 },
97 { offsetof(xfs_sb_t, sb_sectlog), 0 },
98 { offsetof(xfs_sb_t, sb_inodelog), 0 },
99 { offsetof(xfs_sb_t, sb_inopblog), 0 },
100 { offsetof(xfs_sb_t, sb_agblklog), 0 },
101 { offsetof(xfs_sb_t, sb_rextslog), 0 },
102 { offsetof(xfs_sb_t, sb_inprogress), 0 },
103 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
104 { offsetof(xfs_sb_t, sb_icount), 0 },
105 { offsetof(xfs_sb_t, sb_ifree), 0 },
106 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
107 { offsetof(xfs_sb_t, sb_frextents), 0 },
108 { offsetof(xfs_sb_t, sb_uquotino), 0 },
109 { offsetof(xfs_sb_t, sb_gquotino), 0 },
110 { offsetof(xfs_sb_t, sb_qflags), 0 },
111 { offsetof(xfs_sb_t, sb_flags), 0 },
112 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
113 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
114 { offsetof(xfs_sb_t, sb_unit), 0 },
115 { offsetof(xfs_sb_t, sb_width), 0 },
116 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
117 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
118 { offsetof(xfs_sb_t, sb_logsectsize),0 },
119 { offsetof(xfs_sb_t, sb_logsunit), 0 },
120 { offsetof(xfs_sb_t, sb_features2), 0 },
121 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
122 { sizeof(xfs_sb_t), 0 }
125 static DEFINE_MUTEX(xfs_uuid_table_mutex);
126 static int xfs_uuid_table_size;
127 static uuid_t *xfs_uuid_table;
130 * See if the UUID is unique among mounted XFS filesystems.
131 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
135 struct xfs_mount *mp)
137 uuid_t *uuid = &mp->m_sb.sb_uuid;
140 if (mp->m_flags & XFS_MOUNT_NOUUID)
143 if (uuid_is_nil(uuid)) {
145 "XFS: Filesystem %s has nil UUID - can't mount",
147 return XFS_ERROR(EINVAL);
150 mutex_lock(&xfs_uuid_table_mutex);
151 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
152 if (uuid_is_nil(&xfs_uuid_table[i])) {
156 if (uuid_equal(uuid, &xfs_uuid_table[i]))
161 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
162 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
163 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
165 hole = xfs_uuid_table_size++;
167 xfs_uuid_table[hole] = *uuid;
168 mutex_unlock(&xfs_uuid_table_mutex);
173 mutex_unlock(&xfs_uuid_table_mutex);
174 cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount",
176 return XFS_ERROR(EINVAL);
181 struct xfs_mount *mp)
183 uuid_t *uuid = &mp->m_sb.sb_uuid;
186 if (mp->m_flags & XFS_MOUNT_NOUUID)
189 mutex_lock(&xfs_uuid_table_mutex);
190 for (i = 0; i < xfs_uuid_table_size; i++) {
191 if (uuid_is_nil(&xfs_uuid_table[i]))
193 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
195 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
198 ASSERT(i < xfs_uuid_table_size);
199 mutex_unlock(&xfs_uuid_table_mutex);
204 * Free up the resources associated with a mount structure. Assume that
205 * the structure was initially zeroed, so we can tell which fields got
213 struct xfs_perag *pag;
215 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
216 spin_lock(&mp->m_perag_lock);
217 pag = radix_tree_delete(&mp->m_perag_tree, agno);
218 ASSERT(atomic_read(&pag->pag_ref) == 0);
219 spin_unlock(&mp->m_perag_lock);
221 kmem_free(pag->pagb_list);
227 * Check size of device based on the (data/realtime) block count.
228 * Note: this check is used by the growfs code as well as mount.
231 xfs_sb_validate_fsb_count(
235 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
236 ASSERT(sbp->sb_blocklog >= BBSHIFT);
238 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
239 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
241 #else /* Limited by UINT_MAX of sectors */
242 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
249 * Check the validity of the SB found.
252 xfs_mount_validate_sb(
258 * If the log device and data device have the
259 * same device number, the log is internal.
260 * Consequently, the sb_logstart should be non-zero. If
261 * we have a zero sb_logstart in this case, we may be trying to mount
262 * a volume filesystem in a non-volume manner.
264 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
265 xfs_fs_mount_cmn_err(flags, "bad magic number");
266 return XFS_ERROR(EWRONGFS);
269 if (!xfs_sb_good_version(sbp)) {
270 xfs_fs_mount_cmn_err(flags, "bad version");
271 return XFS_ERROR(EWRONGFS);
275 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
276 xfs_fs_mount_cmn_err(flags,
277 "filesystem is marked as having an external log; "
278 "specify logdev on the\nmount command line.");
279 return XFS_ERROR(EINVAL);
283 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
284 xfs_fs_mount_cmn_err(flags,
285 "filesystem is marked as having an internal log; "
286 "do not specify logdev on\nthe mount command line.");
287 return XFS_ERROR(EINVAL);
291 * More sanity checking. These were stolen directly from
295 sbp->sb_agcount <= 0 ||
296 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
297 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
298 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
299 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
300 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
301 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
302 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
303 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
304 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
305 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
306 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
307 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
308 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
309 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
310 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
311 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
312 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
313 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
314 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
315 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
316 return XFS_ERROR(EFSCORRUPTED);
320 * Sanity check AG count, size fields against data size field
323 sbp->sb_dblocks == 0 ||
325 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
326 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
327 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
328 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
329 return XFS_ERROR(EFSCORRUPTED);
333 * Until this is fixed only page-sized or smaller data blocks work.
335 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
336 xfs_fs_mount_cmn_err(flags,
337 "file system with blocksize %d bytes",
339 xfs_fs_mount_cmn_err(flags,
340 "only pagesize (%ld) or less will currently work.",
342 return XFS_ERROR(ENOSYS);
346 * Currently only very few inode sizes are supported.
348 switch (sbp->sb_inodesize) {
355 xfs_fs_mount_cmn_err(flags,
356 "inode size of %d bytes not supported",
358 return XFS_ERROR(ENOSYS);
361 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
362 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
363 xfs_fs_mount_cmn_err(flags,
364 "file system too large to be mounted on this system.");
365 return XFS_ERROR(E2BIG);
368 if (unlikely(sbp->sb_inprogress)) {
369 xfs_fs_mount_cmn_err(flags, "file system busy");
370 return XFS_ERROR(EFSCORRUPTED);
374 * Version 1 directory format has never worked on Linux.
376 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
377 xfs_fs_mount_cmn_err(flags,
378 "file system using version 1 directory format");
379 return XFS_ERROR(ENOSYS);
386 xfs_initialize_perag_icache(
389 if (!pag->pag_ici_init) {
390 rwlock_init(&pag->pag_ici_lock);
391 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
392 pag->pag_ici_init = 1;
397 xfs_initialize_perag(
399 xfs_agnumber_t agcount,
400 xfs_agnumber_t *maxagi)
402 xfs_agnumber_t index, max_metadata;
406 xfs_sb_t *sbp = &mp->m_sb;
407 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
409 /* Check to see if the filesystem can overflow 32 bit inodes */
410 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
411 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
414 * Walk the current per-ag tree so we don't try to initialise AGs
415 * that already exist (growfs case). Allocate and insert all the
416 * AGs we don't find ready for initialisation.
418 for (index = 0; index < agcount; index++) {
419 pag = xfs_perag_get(mp, index);
424 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
427 if (radix_tree_preload(GFP_NOFS))
429 spin_lock(&mp->m_perag_lock);
430 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
432 spin_unlock(&mp->m_perag_lock);
436 spin_unlock(&mp->m_perag_lock);
437 radix_tree_preload_end();
440 /* Clear the mount flag if no inode can overflow 32 bits
441 * on this filesystem, or if specifically requested..
443 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
444 mp->m_flags |= XFS_MOUNT_32BITINODES;
446 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
449 /* If we can overflow then setup the ag headers accordingly */
450 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
451 /* Calculate how much should be reserved for inodes to
452 * meet the max inode percentage.
454 if (mp->m_maxicount) {
457 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
459 icount += sbp->sb_agblocks - 1;
460 do_div(icount, sbp->sb_agblocks);
461 max_metadata = icount;
463 max_metadata = agcount;
465 for (index = 0; index < agcount; index++) {
466 ino = XFS_AGINO_TO_INO(mp, index, agino);
467 if (ino > max_inum) {
472 /* This ag is preferred for inodes */
473 pag = xfs_perag_get(mp, index);
474 pag->pagi_inodeok = 1;
475 if (index < max_metadata)
476 pag->pagf_metadata = 1;
477 xfs_initialize_perag_icache(pag);
481 /* Setup default behavior for smaller filesystems */
482 for (index = 0; index < agcount; index++) {
483 pag = xfs_perag_get(mp, index);
484 pag->pagi_inodeok = 1;
485 xfs_initialize_perag_icache(pag);
499 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
500 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
501 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
502 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
503 to->sb_rextents = be64_to_cpu(from->sb_rextents);
504 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
505 to->sb_logstart = be64_to_cpu(from->sb_logstart);
506 to->sb_rootino = be64_to_cpu(from->sb_rootino);
507 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
508 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
509 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
510 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
511 to->sb_agcount = be32_to_cpu(from->sb_agcount);
512 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
513 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
514 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
515 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
516 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
517 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
518 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
519 to->sb_blocklog = from->sb_blocklog;
520 to->sb_sectlog = from->sb_sectlog;
521 to->sb_inodelog = from->sb_inodelog;
522 to->sb_inopblog = from->sb_inopblog;
523 to->sb_agblklog = from->sb_agblklog;
524 to->sb_rextslog = from->sb_rextslog;
525 to->sb_inprogress = from->sb_inprogress;
526 to->sb_imax_pct = from->sb_imax_pct;
527 to->sb_icount = be64_to_cpu(from->sb_icount);
528 to->sb_ifree = be64_to_cpu(from->sb_ifree);
529 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
530 to->sb_frextents = be64_to_cpu(from->sb_frextents);
531 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
532 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
533 to->sb_qflags = be16_to_cpu(from->sb_qflags);
534 to->sb_flags = from->sb_flags;
535 to->sb_shared_vn = from->sb_shared_vn;
536 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
537 to->sb_unit = be32_to_cpu(from->sb_unit);
538 to->sb_width = be32_to_cpu(from->sb_width);
539 to->sb_dirblklog = from->sb_dirblklog;
540 to->sb_logsectlog = from->sb_logsectlog;
541 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
542 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
543 to->sb_features2 = be32_to_cpu(from->sb_features2);
544 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
548 * Copy in core superblock to ondisk one.
550 * The fields argument is mask of superblock fields to copy.
558 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
559 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
569 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
570 first = xfs_sb_info[f].offset;
571 size = xfs_sb_info[f + 1].offset - first;
573 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
575 if (size == 1 || xfs_sb_info[f].type == 1) {
576 memcpy(to_ptr + first, from_ptr + first, size);
580 *(__be16 *)(to_ptr + first) =
581 cpu_to_be16(*(__u16 *)(from_ptr + first));
584 *(__be32 *)(to_ptr + first) =
585 cpu_to_be32(*(__u32 *)(from_ptr + first));
588 *(__be64 *)(to_ptr + first) =
589 cpu_to_be64(*(__u64 *)(from_ptr + first));
596 fields &= ~(1LL << f);
603 * Does the initial read of the superblock.
606 xfs_readsb(xfs_mount_t *mp, int flags)
608 unsigned int sector_size;
609 unsigned int extra_flags;
613 ASSERT(mp->m_sb_bp == NULL);
614 ASSERT(mp->m_ddev_targp != NULL);
617 * Allocate a (locked) buffer to hold the superblock.
618 * This will be kept around at all times to optimize
619 * access to the superblock.
621 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
622 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
624 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR, BTOBB(sector_size),
626 if (!bp || XFS_BUF_ISERROR(bp)) {
627 xfs_fs_mount_cmn_err(flags, "SB read failed");
628 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
631 ASSERT(XFS_BUF_ISBUSY(bp));
632 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
635 * Initialize the mount structure from the superblock.
636 * But first do some basic consistency checking.
638 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
640 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
642 xfs_fs_mount_cmn_err(flags, "SB validate failed");
647 * We must be able to do sector-sized and sector-aligned IO.
649 if (sector_size > mp->m_sb.sb_sectsize) {
650 xfs_fs_mount_cmn_err(flags,
651 "device supports only %u byte sectors (not %u)",
652 sector_size, mp->m_sb.sb_sectsize);
658 * If device sector size is smaller than the superblock size,
659 * re-read the superblock so the buffer is correctly sized.
661 if (sector_size < mp->m_sb.sb_sectsize) {
662 XFS_BUF_UNMANAGE(bp);
664 sector_size = mp->m_sb.sb_sectsize;
665 bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR,
666 BTOBB(sector_size), extra_flags);
667 if (!bp || XFS_BUF_ISERROR(bp)) {
668 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
669 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
672 ASSERT(XFS_BUF_ISBUSY(bp));
673 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
676 /* Initialize per-cpu counters */
677 xfs_icsb_reinit_counters(mp);
681 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
686 XFS_BUF_UNMANAGE(bp);
696 * Mount initialization code establishing various mount
697 * fields from the superblock associated with the given
701 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
703 mp->m_agfrotor = mp->m_agirotor = 0;
704 spin_lock_init(&mp->m_agirotor_lock);
705 mp->m_maxagi = mp->m_sb.sb_agcount;
706 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
707 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
708 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
709 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
710 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
711 mp->m_blockmask = sbp->sb_blocksize - 1;
712 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
713 mp->m_blockwmask = mp->m_blockwsize - 1;
715 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
716 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
717 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
718 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
720 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
721 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
722 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
723 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
725 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
726 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
727 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
728 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
730 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
731 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
733 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
737 * xfs_initialize_perag_data
739 * Read in each per-ag structure so we can count up the number of
740 * allocated inodes, free inodes and used filesystem blocks as this
741 * information is no longer persistent in the superblock. Once we have
742 * this information, write it into the in-core superblock structure.
745 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
747 xfs_agnumber_t index;
749 xfs_sb_t *sbp = &mp->m_sb;
753 uint64_t bfreelst = 0;
757 for (index = 0; index < agcount; index++) {
759 * read the agf, then the agi. This gets us
760 * all the information we need and populates the
761 * per-ag structures for us.
763 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
767 error = xfs_ialloc_pagi_init(mp, NULL, index);
770 pag = xfs_perag_get(mp, index);
771 ifree += pag->pagi_freecount;
772 ialloc += pag->pagi_count;
773 bfree += pag->pagf_freeblks;
774 bfreelst += pag->pagf_flcount;
775 btree += pag->pagf_btreeblks;
779 * Overwrite incore superblock counters with just-read data
781 spin_lock(&mp->m_sb_lock);
782 sbp->sb_ifree = ifree;
783 sbp->sb_icount = ialloc;
784 sbp->sb_fdblocks = bfree + bfreelst + btree;
785 spin_unlock(&mp->m_sb_lock);
787 /* Fixup the per-cpu counters as well. */
788 xfs_icsb_reinit_counters(mp);
794 * Update alignment values based on mount options and sb values
797 xfs_update_alignment(xfs_mount_t *mp)
799 xfs_sb_t *sbp = &(mp->m_sb);
803 * If stripe unit and stripe width are not multiples
804 * of the fs blocksize turn off alignment.
806 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
807 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
808 if (mp->m_flags & XFS_MOUNT_RETERR) {
810 "XFS: alignment check 1 failed");
811 return XFS_ERROR(EINVAL);
813 mp->m_dalign = mp->m_swidth = 0;
816 * Convert the stripe unit and width to FSBs.
818 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
819 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
820 if (mp->m_flags & XFS_MOUNT_RETERR) {
821 return XFS_ERROR(EINVAL);
823 xfs_fs_cmn_err(CE_WARN, mp,
824 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
825 mp->m_dalign, mp->m_swidth,
830 } else if (mp->m_dalign) {
831 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
833 if (mp->m_flags & XFS_MOUNT_RETERR) {
834 xfs_fs_cmn_err(CE_WARN, mp,
835 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
838 return XFS_ERROR(EINVAL);
845 * Update superblock with new values
848 if (xfs_sb_version_hasdalign(sbp)) {
849 if (sbp->sb_unit != mp->m_dalign) {
850 sbp->sb_unit = mp->m_dalign;
851 mp->m_update_flags |= XFS_SB_UNIT;
853 if (sbp->sb_width != mp->m_swidth) {
854 sbp->sb_width = mp->m_swidth;
855 mp->m_update_flags |= XFS_SB_WIDTH;
858 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
859 xfs_sb_version_hasdalign(&mp->m_sb)) {
860 mp->m_dalign = sbp->sb_unit;
861 mp->m_swidth = sbp->sb_width;
868 * Set the maximum inode count for this filesystem
871 xfs_set_maxicount(xfs_mount_t *mp)
873 xfs_sb_t *sbp = &(mp->m_sb);
876 if (sbp->sb_imax_pct) {
878 * Make sure the maximum inode count is a multiple
879 * of the units we allocate inodes in.
881 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
883 do_div(icount, mp->m_ialloc_blks);
884 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
892 * Set the default minimum read and write sizes unless
893 * already specified in a mount option.
894 * We use smaller I/O sizes when the file system
895 * is being used for NFS service (wsync mount option).
898 xfs_set_rw_sizes(xfs_mount_t *mp)
900 xfs_sb_t *sbp = &(mp->m_sb);
901 int readio_log, writeio_log;
903 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
904 if (mp->m_flags & XFS_MOUNT_WSYNC) {
905 readio_log = XFS_WSYNC_READIO_LOG;
906 writeio_log = XFS_WSYNC_WRITEIO_LOG;
908 readio_log = XFS_READIO_LOG_LARGE;
909 writeio_log = XFS_WRITEIO_LOG_LARGE;
912 readio_log = mp->m_readio_log;
913 writeio_log = mp->m_writeio_log;
916 if (sbp->sb_blocklog > readio_log) {
917 mp->m_readio_log = sbp->sb_blocklog;
919 mp->m_readio_log = readio_log;
921 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
922 if (sbp->sb_blocklog > writeio_log) {
923 mp->m_writeio_log = sbp->sb_blocklog;
925 mp->m_writeio_log = writeio_log;
927 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
931 * Set whether we're using inode alignment.
934 xfs_set_inoalignment(xfs_mount_t *mp)
936 if (xfs_sb_version_hasalign(&mp->m_sb) &&
937 mp->m_sb.sb_inoalignmt >=
938 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
939 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
941 mp->m_inoalign_mask = 0;
943 * If we are using stripe alignment, check whether
944 * the stripe unit is a multiple of the inode alignment
946 if (mp->m_dalign && mp->m_inoalign_mask &&
947 !(mp->m_dalign & mp->m_inoalign_mask))
948 mp->m_sinoalign = mp->m_dalign;
954 * Check that the data (and log if separate) are an ok size.
957 xfs_check_sizes(xfs_mount_t *mp)
963 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
964 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
965 cmn_err(CE_WARN, "XFS: size check 1 failed");
966 return XFS_ERROR(E2BIG);
968 error = xfs_read_buf(mp, mp->m_ddev_targp,
969 d - XFS_FSS_TO_BB(mp, 1),
970 XFS_FSS_TO_BB(mp, 1), 0, &bp);
974 cmn_err(CE_WARN, "XFS: size check 2 failed");
976 error = XFS_ERROR(E2BIG);
980 if (mp->m_logdev_targp != mp->m_ddev_targp) {
981 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
982 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
983 cmn_err(CE_WARN, "XFS: size check 3 failed");
984 return XFS_ERROR(E2BIG);
986 error = xfs_read_buf(mp, mp->m_logdev_targp,
987 d - XFS_FSB_TO_BB(mp, 1),
988 XFS_FSB_TO_BB(mp, 1), 0, &bp);
992 cmn_err(CE_WARN, "XFS: size check 3 failed");
994 error = XFS_ERROR(E2BIG);
1002 * Clear the quotaflags in memory and in the superblock.
1005 xfs_mount_reset_sbqflags(
1006 struct xfs_mount *mp)
1009 struct xfs_trans *tp;
1014 * It is OK to look at sb_qflags here in mount path,
1015 * without m_sb_lock.
1017 if (mp->m_sb.sb_qflags == 0)
1019 spin_lock(&mp->m_sb_lock);
1020 mp->m_sb.sb_qflags = 0;
1021 spin_unlock(&mp->m_sb_lock);
1024 * If the fs is readonly, let the incore superblock run
1025 * with quotas off but don't flush the update out to disk
1027 if (mp->m_flags & XFS_MOUNT_RDONLY)
1031 xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes");
1034 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1035 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1036 XFS_DEFAULT_LOG_COUNT);
1038 xfs_trans_cancel(tp, 0);
1039 xfs_fs_cmn_err(CE_ALERT, mp,
1040 "xfs_mount_reset_sbqflags: Superblock update failed!");
1044 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1045 return xfs_trans_commit(tp, 0);
1049 * This function does the following on an initial mount of a file system:
1050 * - reads the superblock from disk and init the mount struct
1051 * - if we're a 32-bit kernel, do a size check on the superblock
1052 * so we don't mount terabyte filesystems
1053 * - init mount struct realtime fields
1054 * - allocate inode hash table for fs
1055 * - init directory manager
1056 * - perform recovery and init the log manager
1062 xfs_sb_t *sbp = &(mp->m_sb);
1065 uint quotamount = 0;
1066 uint quotaflags = 0;
1069 xfs_mount_common(mp, sbp);
1072 * Check for a mismatched features2 values. Older kernels
1073 * read & wrote into the wrong sb offset for sb_features2
1074 * on some platforms due to xfs_sb_t not being 64bit size aligned
1075 * when sb_features2 was added, which made older superblock
1076 * reading/writing routines swap it as a 64-bit value.
1078 * For backwards compatibility, we make both slots equal.
1080 * If we detect a mismatched field, we OR the set bits into the
1081 * existing features2 field in case it has already been modified; we
1082 * don't want to lose any features. We then update the bad location
1083 * with the ORed value so that older kernels will see any features2
1084 * flags, and mark the two fields as needing updates once the
1085 * transaction subsystem is online.
1087 if (xfs_sb_has_mismatched_features2(sbp)) {
1089 "XFS: correcting sb_features alignment problem");
1090 sbp->sb_features2 |= sbp->sb_bad_features2;
1091 sbp->sb_bad_features2 = sbp->sb_features2;
1092 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1095 * Re-check for ATTR2 in case it was found in bad_features2
1098 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1099 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1100 mp->m_flags |= XFS_MOUNT_ATTR2;
1103 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1104 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1105 xfs_sb_version_removeattr2(&mp->m_sb);
1106 mp->m_update_flags |= XFS_SB_FEATURES2;
1108 /* update sb_versionnum for the clearing of the morebits */
1109 if (!sbp->sb_features2)
1110 mp->m_update_flags |= XFS_SB_VERSIONNUM;
1114 * Check if sb_agblocks is aligned at stripe boundary
1115 * If sb_agblocks is NOT aligned turn off m_dalign since
1116 * allocator alignment is within an ag, therefore ag has
1117 * to be aligned at stripe boundary.
1119 error = xfs_update_alignment(mp);
1123 xfs_alloc_compute_maxlevels(mp);
1124 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1125 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1126 xfs_ialloc_compute_maxlevels(mp);
1128 xfs_set_maxicount(mp);
1130 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1132 error = xfs_uuid_mount(mp);
1137 * Set the minimum read and write sizes
1139 xfs_set_rw_sizes(mp);
1142 * Set the inode cluster size.
1143 * This may still be overridden by the file system
1144 * block size if it is larger than the chosen cluster size.
1146 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1149 * Set inode alignment fields
1151 xfs_set_inoalignment(mp);
1154 * Check that the data (and log if separate) are an ok size.
1156 error = xfs_check_sizes(mp);
1158 goto out_remove_uuid;
1161 * Initialize realtime fields in the mount structure
1163 error = xfs_rtmount_init(mp);
1165 cmn_err(CE_WARN, "XFS: RT mount failed");
1166 goto out_remove_uuid;
1170 * Copies the low order bits of the timestamp and the randomly
1171 * set "sequence" number out of a UUID.
1173 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1175 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1180 * Initialize the attribute manager's entries.
1182 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1185 * Initialize the precomputed transaction reservations values.
1190 * Allocate and initialize the per-ag data.
1192 spin_lock_init(&mp->m_perag_lock);
1193 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_NOFS);
1194 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1196 cmn_err(CE_WARN, "XFS: Failed per-ag init: %d", error);
1197 goto out_remove_uuid;
1200 if (!sbp->sb_logblocks) {
1201 cmn_err(CE_WARN, "XFS: no log defined");
1202 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1203 error = XFS_ERROR(EFSCORRUPTED);
1204 goto out_free_perag;
1208 * log's mount-time initialization. Perform 1st part recovery if needed
1210 error = xfs_log_mount(mp, mp->m_logdev_targp,
1211 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1212 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1214 cmn_err(CE_WARN, "XFS: log mount failed");
1215 goto out_free_perag;
1219 * Now the log is mounted, we know if it was an unclean shutdown or
1220 * not. If it was, with the first phase of recovery has completed, we
1221 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1222 * but they are recovered transactionally in the second recovery phase
1225 * Hence we can safely re-initialise incore superblock counters from
1226 * the per-ag data. These may not be correct if the filesystem was not
1227 * cleanly unmounted, so we need to wait for recovery to finish before
1230 * If the filesystem was cleanly unmounted, then we can trust the
1231 * values in the superblock to be correct and we don't need to do
1234 * If we are currently making the filesystem, the initialisation will
1235 * fail as the perag data is in an undefined state.
1237 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1238 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1239 !mp->m_sb.sb_inprogress) {
1240 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1242 goto out_free_perag;
1246 * Get and sanity-check the root inode.
1247 * Save the pointer to it in the mount structure.
1249 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
1251 cmn_err(CE_WARN, "XFS: failed to read root inode");
1252 goto out_log_dealloc;
1255 ASSERT(rip != NULL);
1257 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1258 cmn_err(CE_WARN, "XFS: corrupted root inode");
1259 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1260 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1261 (unsigned long long)rip->i_ino);
1262 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1263 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1265 error = XFS_ERROR(EFSCORRUPTED);
1268 mp->m_rootip = rip; /* save it */
1270 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1273 * Initialize realtime inode pointers in the mount structure
1275 error = xfs_rtmount_inodes(mp);
1278 * Free up the root inode.
1280 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1285 * If this is a read-only mount defer the superblock updates until
1286 * the next remount into writeable mode. Otherwise we would never
1287 * perform the update e.g. for the root filesystem.
1289 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1290 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1292 cmn_err(CE_WARN, "XFS: failed to write sb changes");
1298 * Initialise the XFS quota management subsystem for this mount
1300 if (XFS_IS_QUOTA_RUNNING(mp)) {
1301 error = xfs_qm_newmount(mp, "amount, "aflags);
1305 ASSERT(!XFS_IS_QUOTA_ON(mp));
1308 * If a file system had quotas running earlier, but decided to
1309 * mount without -o uquota/pquota/gquota options, revoke the
1310 * quotachecked license.
1312 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1314 "XFS: resetting qflags for filesystem %s",
1317 error = xfs_mount_reset_sbqflags(mp);
1324 * Finish recovering the file system. This part needed to be
1325 * delayed until after the root and real-time bitmap inodes
1326 * were consistently read in.
1328 error = xfs_log_mount_finish(mp);
1330 cmn_err(CE_WARN, "XFS: log mount finish failed");
1335 * Complete the quota initialisation, post-log-replay component.
1338 ASSERT(mp->m_qflags == 0);
1339 mp->m_qflags = quotaflags;
1341 xfs_qm_mount_quotas(mp);
1344 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1345 if (XFS_IS_QUOTA_ON(mp))
1346 xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas turned on");
1348 xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas not turned on");
1352 * Now we are mounted, reserve a small amount of unused space for
1353 * privileged transactions. This is needed so that transaction
1354 * space required for critical operations can dip into this pool
1355 * when at ENOSPC. This is needed for operations like create with
1356 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1357 * are not allowed to use this reserved space.
1359 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1360 * This may drive us straight to ENOSPC on mount, but that implies
1361 * we were already there on the last unmount. Warn if this occurs.
1363 resblks = mp->m_sb.sb_dblocks;
1364 do_div(resblks, 20);
1365 resblks = min_t(__uint64_t, resblks, 1024);
1366 error = xfs_reserve_blocks(mp, &resblks, NULL);
1368 cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. "
1369 "Continuing without a reserve pool.");
1374 xfs_rtunmount_inodes(mp);
1378 xfs_log_unmount(mp);
1382 xfs_uuid_unmount(mp);
1388 * This flushes out the inodes,dquots and the superblock, unmounts the
1389 * log and makes sure that incore structures are freed.
1393 struct xfs_mount *mp)
1398 xfs_qm_unmount_quotas(mp);
1399 xfs_rtunmount_inodes(mp);
1400 IRELE(mp->m_rootip);
1403 * We can potentially deadlock here if we have an inode cluster
1404 * that has been freed has its buffer still pinned in memory because
1405 * the transaction is still sitting in a iclog. The stale inodes
1406 * on that buffer will have their flush locks held until the
1407 * transaction hits the disk and the callbacks run. the inode
1408 * flush takes the flush lock unconditionally and with nothing to
1409 * push out the iclog we will never get that unlocked. hence we
1410 * need to force the log first.
1412 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1413 xfs_reclaim_inodes(mp, XFS_IFLUSH_ASYNC);
1418 * Flush out the log synchronously so that we know for sure
1419 * that nothing is pinned. This is important because bflush()
1420 * will skip pinned buffers.
1422 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1424 xfs_binval(mp->m_ddev_targp);
1425 if (mp->m_rtdev_targp) {
1426 xfs_binval(mp->m_rtdev_targp);
1430 * Unreserve any blocks we have so that when we unmount we don't account
1431 * the reserved free space as used. This is really only necessary for
1432 * lazy superblock counting because it trusts the incore superblock
1433 * counters to be absolutely correct on clean unmount.
1435 * We don't bother correcting this elsewhere for lazy superblock
1436 * counting because on mount of an unclean filesystem we reconstruct the
1437 * correct counter value and this is irrelevant.
1439 * For non-lazy counter filesystems, this doesn't matter at all because
1440 * we only every apply deltas to the superblock and hence the incore
1441 * value does not matter....
1444 error = xfs_reserve_blocks(mp, &resblks, NULL);
1446 cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1447 "Freespace may not be correct on next mount.");
1449 error = xfs_log_sbcount(mp, 1);
1451 cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1452 "Freespace may not be correct on next mount.");
1453 xfs_unmountfs_writesb(mp);
1454 xfs_unmountfs_wait(mp); /* wait for async bufs */
1455 xfs_log_unmount_write(mp);
1456 xfs_log_unmount(mp);
1457 xfs_uuid_unmount(mp);
1460 xfs_errortag_clearall(mp, 0);
1466 xfs_unmountfs_wait(xfs_mount_t *mp)
1468 if (mp->m_logdev_targp != mp->m_ddev_targp)
1469 xfs_wait_buftarg(mp->m_logdev_targp);
1470 if (mp->m_rtdev_targp)
1471 xfs_wait_buftarg(mp->m_rtdev_targp);
1472 xfs_wait_buftarg(mp->m_ddev_targp);
1476 xfs_fs_writable(xfs_mount_t *mp)
1478 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1479 (mp->m_flags & XFS_MOUNT_RDONLY));
1485 * Called either periodically to keep the on disk superblock values
1486 * roughly up to date or from unmount to make sure the values are
1487 * correct on a clean unmount.
1489 * Note this code can be called during the process of freezing, so
1490 * we may need to use the transaction allocator which does not not
1491 * block when the transaction subsystem is in its frozen state.
1501 if (!xfs_fs_writable(mp))
1504 xfs_icsb_sync_counters(mp, 0);
1507 * we don't need to do this if we are updating the superblock
1508 * counters on every modification.
1510 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1513 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1514 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1515 XFS_DEFAULT_LOG_COUNT);
1517 xfs_trans_cancel(tp, 0);
1521 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1523 xfs_trans_set_sync(tp);
1524 error = xfs_trans_commit(tp, 0);
1529 xfs_unmountfs_writesb(xfs_mount_t *mp)
1535 * skip superblock write if fs is read-only, or
1536 * if we are doing a forced umount.
1538 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1539 XFS_FORCED_SHUTDOWN(mp))) {
1541 sbp = xfs_getsb(mp, 0);
1543 XFS_BUF_UNDONE(sbp);
1544 XFS_BUF_UNREAD(sbp);
1545 XFS_BUF_UNDELAYWRITE(sbp);
1547 XFS_BUF_UNASYNC(sbp);
1548 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1549 xfsbdstrat(mp, sbp);
1550 error = xfs_iowait(sbp);
1552 xfs_ioerror_alert("xfs_unmountfs_writesb",
1553 mp, sbp, XFS_BUF_ADDR(sbp));
1560 * xfs_mod_sb() can be used to copy arbitrary changes to the
1561 * in-core superblock into the superblock buffer to be logged.
1562 * It does not provide the higher level of locking that is
1563 * needed to protect the in-core superblock from concurrent
1567 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1579 bp = xfs_trans_getsb(tp, mp, 0);
1580 first = sizeof(xfs_sb_t);
1583 /* translate/copy */
1585 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1587 /* find modified range */
1589 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1590 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1591 first = xfs_sb_info[f].offset;
1593 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1594 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1595 last = xfs_sb_info[f + 1].offset - 1;
1597 xfs_trans_log_buf(tp, bp, first, last);
1602 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1603 * a delta to a specified field in the in-core superblock. Simply
1604 * switch on the field indicated and apply the delta to that field.
1605 * Fields are not allowed to dip below zero, so if the delta would
1606 * do this do not apply it and return EINVAL.
1608 * The m_sb_lock must be held when this routine is called.
1611 xfs_mod_incore_sb_unlocked(
1613 xfs_sb_field_t field,
1617 int scounter; /* short counter for 32 bit fields */
1618 long long lcounter; /* long counter for 64 bit fields */
1619 long long res_used, rem;
1622 * With the in-core superblock spin lock held, switch
1623 * on the indicated field. Apply the delta to the
1624 * proper field. If the fields value would dip below
1625 * 0, then do not apply the delta and return EINVAL.
1628 case XFS_SBS_ICOUNT:
1629 lcounter = (long long)mp->m_sb.sb_icount;
1633 return XFS_ERROR(EINVAL);
1635 mp->m_sb.sb_icount = lcounter;
1638 lcounter = (long long)mp->m_sb.sb_ifree;
1642 return XFS_ERROR(EINVAL);
1644 mp->m_sb.sb_ifree = lcounter;
1646 case XFS_SBS_FDBLOCKS:
1647 lcounter = (long long)
1648 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1649 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1651 if (delta > 0) { /* Putting blocks back */
1652 if (res_used > delta) {
1653 mp->m_resblks_avail += delta;
1655 rem = delta - res_used;
1656 mp->m_resblks_avail = mp->m_resblks;
1659 } else { /* Taking blocks away */
1664 * If were out of blocks, use any available reserved blocks if
1670 lcounter = (long long)mp->m_resblks_avail + delta;
1672 return XFS_ERROR(ENOSPC);
1674 mp->m_resblks_avail = lcounter;
1676 } else { /* not reserved */
1677 return XFS_ERROR(ENOSPC);
1682 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1684 case XFS_SBS_FREXTENTS:
1685 lcounter = (long long)mp->m_sb.sb_frextents;
1688 return XFS_ERROR(ENOSPC);
1690 mp->m_sb.sb_frextents = lcounter;
1692 case XFS_SBS_DBLOCKS:
1693 lcounter = (long long)mp->m_sb.sb_dblocks;
1697 return XFS_ERROR(EINVAL);
1699 mp->m_sb.sb_dblocks = lcounter;
1701 case XFS_SBS_AGCOUNT:
1702 scounter = mp->m_sb.sb_agcount;
1706 return XFS_ERROR(EINVAL);
1708 mp->m_sb.sb_agcount = scounter;
1710 case XFS_SBS_IMAX_PCT:
1711 scounter = mp->m_sb.sb_imax_pct;
1715 return XFS_ERROR(EINVAL);
1717 mp->m_sb.sb_imax_pct = scounter;
1719 case XFS_SBS_REXTSIZE:
1720 scounter = mp->m_sb.sb_rextsize;
1724 return XFS_ERROR(EINVAL);
1726 mp->m_sb.sb_rextsize = scounter;
1728 case XFS_SBS_RBMBLOCKS:
1729 scounter = mp->m_sb.sb_rbmblocks;
1733 return XFS_ERROR(EINVAL);
1735 mp->m_sb.sb_rbmblocks = scounter;
1737 case XFS_SBS_RBLOCKS:
1738 lcounter = (long long)mp->m_sb.sb_rblocks;
1742 return XFS_ERROR(EINVAL);
1744 mp->m_sb.sb_rblocks = lcounter;
1746 case XFS_SBS_REXTENTS:
1747 lcounter = (long long)mp->m_sb.sb_rextents;
1751 return XFS_ERROR(EINVAL);
1753 mp->m_sb.sb_rextents = lcounter;
1755 case XFS_SBS_REXTSLOG:
1756 scounter = mp->m_sb.sb_rextslog;
1760 return XFS_ERROR(EINVAL);
1762 mp->m_sb.sb_rextslog = scounter;
1766 return XFS_ERROR(EINVAL);
1771 * xfs_mod_incore_sb() is used to change a field in the in-core
1772 * superblock structure by the specified delta. This modification
1773 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1774 * routine to do the work.
1779 xfs_sb_field_t field,
1785 /* check for per-cpu counters */
1787 #ifdef HAVE_PERCPU_SB
1788 case XFS_SBS_ICOUNT:
1790 case XFS_SBS_FDBLOCKS:
1791 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1792 status = xfs_icsb_modify_counters(mp, field,
1799 spin_lock(&mp->m_sb_lock);
1800 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1801 spin_unlock(&mp->m_sb_lock);
1809 * xfs_mod_incore_sb_batch() is used to change more than one field
1810 * in the in-core superblock structure at a time. This modification
1811 * is protected by a lock internal to this module. The fields and
1812 * changes to those fields are specified in the array of xfs_mod_sb
1813 * structures passed in.
1815 * Either all of the specified deltas will be applied or none of
1816 * them will. If any modified field dips below 0, then all modifications
1817 * will be backed out and EINVAL will be returned.
1820 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1826 * Loop through the array of mod structures and apply each
1827 * individually. If any fail, then back out all those
1828 * which have already been applied. Do all of this within
1829 * the scope of the m_sb_lock so that all of the changes will
1832 spin_lock(&mp->m_sb_lock);
1834 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1836 * Apply the delta at index n. If it fails, break
1837 * from the loop so we'll fall into the undo loop
1840 switch (msbp->msb_field) {
1841 #ifdef HAVE_PERCPU_SB
1842 case XFS_SBS_ICOUNT:
1844 case XFS_SBS_FDBLOCKS:
1845 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1846 spin_unlock(&mp->m_sb_lock);
1847 status = xfs_icsb_modify_counters(mp,
1849 msbp->msb_delta, rsvd);
1850 spin_lock(&mp->m_sb_lock);
1856 status = xfs_mod_incore_sb_unlocked(mp,
1858 msbp->msb_delta, rsvd);
1868 * If we didn't complete the loop above, then back out
1869 * any changes made to the superblock. If you add code
1870 * between the loop above and here, make sure that you
1871 * preserve the value of status. Loop back until
1872 * we step below the beginning of the array. Make sure
1873 * we don't touch anything back there.
1877 while (msbp >= msb) {
1878 switch (msbp->msb_field) {
1879 #ifdef HAVE_PERCPU_SB
1880 case XFS_SBS_ICOUNT:
1882 case XFS_SBS_FDBLOCKS:
1883 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1884 spin_unlock(&mp->m_sb_lock);
1885 status = xfs_icsb_modify_counters(mp,
1889 spin_lock(&mp->m_sb_lock);
1895 status = xfs_mod_incore_sb_unlocked(mp,
1901 ASSERT(status == 0);
1905 spin_unlock(&mp->m_sb_lock);
1910 * xfs_getsb() is called to obtain the buffer for the superblock.
1911 * The buffer is returned locked and read in from disk.
1912 * The buffer should be released with a call to xfs_brelse().
1914 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1915 * the superblock buffer if it can be locked without sleeping.
1916 * If it can't then we'll return NULL.
1925 ASSERT(mp->m_sb_bp != NULL);
1927 if (flags & XFS_BUF_TRYLOCK) {
1928 if (!XFS_BUF_CPSEMA(bp)) {
1932 XFS_BUF_PSEMA(bp, PRIBIO);
1935 ASSERT(XFS_BUF_ISDONE(bp));
1940 * Used to free the superblock along various error paths.
1949 * Use xfs_getsb() so that the buffer will be locked
1950 * when we call xfs_buf_relse().
1952 bp = xfs_getsb(mp, 0);
1953 XFS_BUF_UNMANAGE(bp);
1959 * Used to log changes to the superblock unit and width fields which could
1960 * be altered by the mount options, as well as any potential sb_features2
1961 * fixup. Only the first superblock is updated.
1971 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1972 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1973 XFS_SB_VERSIONNUM));
1975 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1976 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1977 XFS_DEFAULT_LOG_COUNT);
1979 xfs_trans_cancel(tp, 0);
1982 xfs_mod_sb(tp, fields);
1983 error = xfs_trans_commit(tp, 0);
1988 #ifdef HAVE_PERCPU_SB
1990 * Per-cpu incore superblock counters
1992 * Simple concept, difficult implementation
1994 * Basically, replace the incore superblock counters with a distributed per cpu
1995 * counter for contended fields (e.g. free block count).
1997 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1998 * hence needs to be accurately read when we are running low on space. Hence
1999 * there is a method to enable and disable the per-cpu counters based on how
2000 * much "stuff" is available in them.
2002 * Basically, a counter is enabled if there is enough free resource to justify
2003 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2004 * ENOSPC), then we disable the counters to synchronise all callers and
2005 * re-distribute the available resources.
2007 * If, once we redistributed the available resources, we still get a failure,
2008 * we disable the per-cpu counter and go through the slow path.
2010 * The slow path is the current xfs_mod_incore_sb() function. This means that
2011 * when we disable a per-cpu counter, we need to drain its resources back to
2012 * the global superblock. We do this after disabling the counter to prevent
2013 * more threads from queueing up on the counter.
2015 * Essentially, this means that we still need a lock in the fast path to enable
2016 * synchronisation between the global counters and the per-cpu counters. This
2017 * is not a problem because the lock will be local to a CPU almost all the time
2018 * and have little contention except when we get to ENOSPC conditions.
2020 * Basically, this lock becomes a barrier that enables us to lock out the fast
2021 * path while we do things like enabling and disabling counters and
2022 * synchronising the counters.
2026 * 1. m_sb_lock before picking up per-cpu locks
2027 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2028 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2029 * 4. modifying per-cpu counters requires holding per-cpu lock
2030 * 5. modifying global counters requires holding m_sb_lock
2031 * 6. enabling or disabling a counter requires holding the m_sb_lock
2032 * and _none_ of the per-cpu locks.
2034 * Disabled counters are only ever re-enabled by a balance operation
2035 * that results in more free resources per CPU than a given threshold.
2036 * To ensure counters don't remain disabled, they are rebalanced when
2037 * the global resource goes above a higher threshold (i.e. some hysteresis
2038 * is present to prevent thrashing).
2041 #ifdef CONFIG_HOTPLUG_CPU
2043 * hot-plug CPU notifier support.
2045 * We need a notifier per filesystem as we need to be able to identify
2046 * the filesystem to balance the counters out. This is achieved by
2047 * having a notifier block embedded in the xfs_mount_t and doing pointer
2048 * magic to get the mount pointer from the notifier block address.
2051 xfs_icsb_cpu_notify(
2052 struct notifier_block *nfb,
2053 unsigned long action,
2056 xfs_icsb_cnts_t *cntp;
2059 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2060 cntp = (xfs_icsb_cnts_t *)
2061 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2063 case CPU_UP_PREPARE:
2064 case CPU_UP_PREPARE_FROZEN:
2065 /* Easy Case - initialize the area and locks, and
2066 * then rebalance when online does everything else for us. */
2067 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2070 case CPU_ONLINE_FROZEN:
2072 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2073 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2074 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2075 xfs_icsb_unlock(mp);
2078 case CPU_DEAD_FROZEN:
2079 /* Disable all the counters, then fold the dead cpu's
2080 * count into the total on the global superblock and
2081 * re-enable the counters. */
2083 spin_lock(&mp->m_sb_lock);
2084 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2085 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2086 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2088 mp->m_sb.sb_icount += cntp->icsb_icount;
2089 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2090 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2092 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2094 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2095 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2096 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2097 spin_unlock(&mp->m_sb_lock);
2098 xfs_icsb_unlock(mp);
2104 #endif /* CONFIG_HOTPLUG_CPU */
2107 xfs_icsb_init_counters(
2110 xfs_icsb_cnts_t *cntp;
2113 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2114 if (mp->m_sb_cnts == NULL)
2117 #ifdef CONFIG_HOTPLUG_CPU
2118 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2119 mp->m_icsb_notifier.priority = 0;
2120 register_hotcpu_notifier(&mp->m_icsb_notifier);
2121 #endif /* CONFIG_HOTPLUG_CPU */
2123 for_each_online_cpu(i) {
2124 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2125 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2128 mutex_init(&mp->m_icsb_mutex);
2131 * start with all counters disabled so that the
2132 * initial balance kicks us off correctly
2134 mp->m_icsb_counters = -1;
2139 xfs_icsb_reinit_counters(
2144 * start with all counters disabled so that the
2145 * initial balance kicks us off correctly
2147 mp->m_icsb_counters = -1;
2148 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2149 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2150 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2151 xfs_icsb_unlock(mp);
2155 xfs_icsb_destroy_counters(
2158 if (mp->m_sb_cnts) {
2159 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2160 free_percpu(mp->m_sb_cnts);
2162 mutex_destroy(&mp->m_icsb_mutex);
2167 xfs_icsb_cnts_t *icsbp)
2169 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2175 xfs_icsb_unlock_cntr(
2176 xfs_icsb_cnts_t *icsbp)
2178 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2183 xfs_icsb_lock_all_counters(
2186 xfs_icsb_cnts_t *cntp;
2189 for_each_online_cpu(i) {
2190 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2191 xfs_icsb_lock_cntr(cntp);
2196 xfs_icsb_unlock_all_counters(
2199 xfs_icsb_cnts_t *cntp;
2202 for_each_online_cpu(i) {
2203 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2204 xfs_icsb_unlock_cntr(cntp);
2211 xfs_icsb_cnts_t *cnt,
2214 xfs_icsb_cnts_t *cntp;
2217 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2219 if (!(flags & XFS_ICSB_LAZY_COUNT))
2220 xfs_icsb_lock_all_counters(mp);
2222 for_each_online_cpu(i) {
2223 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2224 cnt->icsb_icount += cntp->icsb_icount;
2225 cnt->icsb_ifree += cntp->icsb_ifree;
2226 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2229 if (!(flags & XFS_ICSB_LAZY_COUNT))
2230 xfs_icsb_unlock_all_counters(mp);
2234 xfs_icsb_counter_disabled(
2236 xfs_sb_field_t field)
2238 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2239 return test_bit(field, &mp->m_icsb_counters);
2243 xfs_icsb_disable_counter(
2245 xfs_sb_field_t field)
2247 xfs_icsb_cnts_t cnt;
2249 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2252 * If we are already disabled, then there is nothing to do
2253 * here. We check before locking all the counters to avoid
2254 * the expensive lock operation when being called in the
2255 * slow path and the counter is already disabled. This is
2256 * safe because the only time we set or clear this state is under
2259 if (xfs_icsb_counter_disabled(mp, field))
2262 xfs_icsb_lock_all_counters(mp);
2263 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2264 /* drain back to superblock */
2266 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2268 case XFS_SBS_ICOUNT:
2269 mp->m_sb.sb_icount = cnt.icsb_icount;
2272 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2274 case XFS_SBS_FDBLOCKS:
2275 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2282 xfs_icsb_unlock_all_counters(mp);
2286 xfs_icsb_enable_counter(
2288 xfs_sb_field_t field,
2292 xfs_icsb_cnts_t *cntp;
2295 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2297 xfs_icsb_lock_all_counters(mp);
2298 for_each_online_cpu(i) {
2299 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2301 case XFS_SBS_ICOUNT:
2302 cntp->icsb_icount = count + resid;
2305 cntp->icsb_ifree = count + resid;
2307 case XFS_SBS_FDBLOCKS:
2308 cntp->icsb_fdblocks = count + resid;
2316 clear_bit(field, &mp->m_icsb_counters);
2317 xfs_icsb_unlock_all_counters(mp);
2321 xfs_icsb_sync_counters_locked(
2325 xfs_icsb_cnts_t cnt;
2327 xfs_icsb_count(mp, &cnt, flags);
2329 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2330 mp->m_sb.sb_icount = cnt.icsb_icount;
2331 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2332 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2333 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2334 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2338 * Accurate update of per-cpu counters to incore superblock
2341 xfs_icsb_sync_counters(
2345 spin_lock(&mp->m_sb_lock);
2346 xfs_icsb_sync_counters_locked(mp, flags);
2347 spin_unlock(&mp->m_sb_lock);
2351 * Balance and enable/disable counters as necessary.
2353 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2354 * chosen to be the same number as single on disk allocation chunk per CPU, and
2355 * free blocks is something far enough zero that we aren't going thrash when we
2356 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2357 * prevent looping endlessly when xfs_alloc_space asks for more than will
2358 * be distributed to a single CPU but each CPU has enough blocks to be
2361 * Note that we can be called when counters are already disabled.
2362 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2363 * prevent locking every per-cpu counter needlessly.
2366 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2367 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2368 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2370 xfs_icsb_balance_counter_locked(
2372 xfs_sb_field_t field,
2375 uint64_t count, resid;
2376 int weight = num_online_cpus();
2377 uint64_t min = (uint64_t)min_per_cpu;
2379 /* disable counter and sync counter */
2380 xfs_icsb_disable_counter(mp, field);
2382 /* update counters - first CPU gets residual*/
2384 case XFS_SBS_ICOUNT:
2385 count = mp->m_sb.sb_icount;
2386 resid = do_div(count, weight);
2387 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2391 count = mp->m_sb.sb_ifree;
2392 resid = do_div(count, weight);
2393 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2396 case XFS_SBS_FDBLOCKS:
2397 count = mp->m_sb.sb_fdblocks;
2398 resid = do_div(count, weight);
2399 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2404 count = resid = 0; /* quiet, gcc */
2408 xfs_icsb_enable_counter(mp, field, count, resid);
2412 xfs_icsb_balance_counter(
2414 xfs_sb_field_t fields,
2417 spin_lock(&mp->m_sb_lock);
2418 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2419 spin_unlock(&mp->m_sb_lock);
2423 xfs_icsb_modify_counters(
2425 xfs_sb_field_t field,
2429 xfs_icsb_cnts_t *icsbp;
2430 long long lcounter; /* long counter for 64 bit fields */
2436 icsbp = this_cpu_ptr(mp->m_sb_cnts);
2439 * if the counter is disabled, go to slow path
2441 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2443 xfs_icsb_lock_cntr(icsbp);
2444 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2445 xfs_icsb_unlock_cntr(icsbp);
2450 case XFS_SBS_ICOUNT:
2451 lcounter = icsbp->icsb_icount;
2453 if (unlikely(lcounter < 0))
2454 goto balance_counter;
2455 icsbp->icsb_icount = lcounter;
2459 lcounter = icsbp->icsb_ifree;
2461 if (unlikely(lcounter < 0))
2462 goto balance_counter;
2463 icsbp->icsb_ifree = lcounter;
2466 case XFS_SBS_FDBLOCKS:
2467 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2469 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2471 if (unlikely(lcounter < 0))
2472 goto balance_counter;
2473 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2479 xfs_icsb_unlock_cntr(icsbp);
2487 * serialise with a mutex so we don't burn lots of cpu on
2488 * the superblock lock. We still need to hold the superblock
2489 * lock, however, when we modify the global structures.
2494 * Now running atomically.
2496 * If the counter is enabled, someone has beaten us to rebalancing.
2497 * Drop the lock and try again in the fast path....
2499 if (!(xfs_icsb_counter_disabled(mp, field))) {
2500 xfs_icsb_unlock(mp);
2505 * The counter is currently disabled. Because we are
2506 * running atomically here, we know a rebalance cannot
2507 * be in progress. Hence we can go straight to operating
2508 * on the global superblock. We do not call xfs_mod_incore_sb()
2509 * here even though we need to get the m_sb_lock. Doing so
2510 * will cause us to re-enter this function and deadlock.
2511 * Hence we get the m_sb_lock ourselves and then call
2512 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2513 * directly on the global counters.
2515 spin_lock(&mp->m_sb_lock);
2516 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2517 spin_unlock(&mp->m_sb_lock);
2520 * Now that we've modified the global superblock, we
2521 * may be able to re-enable the distributed counters
2522 * (e.g. lots of space just got freed). After that
2526 xfs_icsb_balance_counter(mp, field, 0);
2527 xfs_icsb_unlock(mp);
2531 xfs_icsb_unlock_cntr(icsbp);
2535 * We may have multiple threads here if multiple per-cpu
2536 * counters run dry at the same time. This will mean we can
2537 * do more balances than strictly necessary but it is not
2538 * the common slowpath case.
2543 * running atomically.
2545 * This will leave the counter in the correct state for future
2546 * accesses. After the rebalance, we simply try again and our retry
2547 * will either succeed through the fast path or slow path without
2548 * another balance operation being required.
2550 xfs_icsb_balance_counter(mp, field, delta);
2551 xfs_icsb_unlock(mp);