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"
47 STATIC void xfs_mount_log_sb(xfs_mount_t *, __int64_t);
48 STATIC int xfs_uuid_mount(xfs_mount_t *);
49 STATIC void xfs_uuid_unmount(xfs_mount_t *mp);
50 STATIC void xfs_unmountfs_wait(xfs_mount_t *);
54 STATIC void xfs_icsb_destroy_counters(xfs_mount_t *);
55 STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
57 STATIC void xfs_icsb_sync_counters(xfs_mount_t *);
58 STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
60 STATIC int xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
64 #define xfs_icsb_destroy_counters(mp) do { } while (0)
65 #define xfs_icsb_balance_counter(mp, a, b, c) do { } while (0)
66 #define xfs_icsb_sync_counters(mp) do { } while (0)
67 #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
73 short type; /* 0 = integer
74 * 1 = binary / string (no translation)
77 { offsetof(xfs_sb_t, sb_magicnum), 0 },
78 { offsetof(xfs_sb_t, sb_blocksize), 0 },
79 { offsetof(xfs_sb_t, sb_dblocks), 0 },
80 { offsetof(xfs_sb_t, sb_rblocks), 0 },
81 { offsetof(xfs_sb_t, sb_rextents), 0 },
82 { offsetof(xfs_sb_t, sb_uuid), 1 },
83 { offsetof(xfs_sb_t, sb_logstart), 0 },
84 { offsetof(xfs_sb_t, sb_rootino), 0 },
85 { offsetof(xfs_sb_t, sb_rbmino), 0 },
86 { offsetof(xfs_sb_t, sb_rsumino), 0 },
87 { offsetof(xfs_sb_t, sb_rextsize), 0 },
88 { offsetof(xfs_sb_t, sb_agblocks), 0 },
89 { offsetof(xfs_sb_t, sb_agcount), 0 },
90 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
91 { offsetof(xfs_sb_t, sb_logblocks), 0 },
92 { offsetof(xfs_sb_t, sb_versionnum), 0 },
93 { offsetof(xfs_sb_t, sb_sectsize), 0 },
94 { offsetof(xfs_sb_t, sb_inodesize), 0 },
95 { offsetof(xfs_sb_t, sb_inopblock), 0 },
96 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
97 { offsetof(xfs_sb_t, sb_blocklog), 0 },
98 { offsetof(xfs_sb_t, sb_sectlog), 0 },
99 { offsetof(xfs_sb_t, sb_inodelog), 0 },
100 { offsetof(xfs_sb_t, sb_inopblog), 0 },
101 { offsetof(xfs_sb_t, sb_agblklog), 0 },
102 { offsetof(xfs_sb_t, sb_rextslog), 0 },
103 { offsetof(xfs_sb_t, sb_inprogress), 0 },
104 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
105 { offsetof(xfs_sb_t, sb_icount), 0 },
106 { offsetof(xfs_sb_t, sb_ifree), 0 },
107 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
108 { offsetof(xfs_sb_t, sb_frextents), 0 },
109 { offsetof(xfs_sb_t, sb_uquotino), 0 },
110 { offsetof(xfs_sb_t, sb_gquotino), 0 },
111 { offsetof(xfs_sb_t, sb_qflags), 0 },
112 { offsetof(xfs_sb_t, sb_flags), 0 },
113 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
114 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
115 { offsetof(xfs_sb_t, sb_unit), 0 },
116 { offsetof(xfs_sb_t, sb_width), 0 },
117 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
118 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
119 { offsetof(xfs_sb_t, sb_logsectsize),0 },
120 { offsetof(xfs_sb_t, sb_logsunit), 0 },
121 { offsetof(xfs_sb_t, sb_features2), 0 },
122 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
123 { sizeof(xfs_sb_t), 0 }
127 * Return a pointer to an initialized xfs_mount structure.
134 mp = kmem_zalloc(sizeof(xfs_mount_t), KM_SLEEP);
136 if (xfs_icsb_init_counters(mp)) {
137 mp->m_flags |= XFS_MOUNT_NO_PERCPU_SB;
140 spin_lock_init(&mp->m_sb_lock);
141 mutex_init(&mp->m_ilock);
142 mutex_init(&mp->m_growlock);
143 atomic_set(&mp->m_active_trans, 0);
149 * Free up the resources associated with a mount structure. Assume that
150 * the structure was initially zeroed, so we can tell which fields got
160 for (agno = 0; agno < mp->m_maxagi; agno++)
161 if (mp->m_perag[agno].pagb_list)
162 kmem_free(mp->m_perag[agno].pagb_list,
163 sizeof(xfs_perag_busy_t) *
165 kmem_free(mp->m_perag,
166 sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
169 spinlock_destroy(&mp->m_ail_lock);
170 spinlock_destroy(&mp->m_sb_lock);
171 mutex_destroy(&mp->m_ilock);
172 mutex_destroy(&mp->m_growlock);
176 if (mp->m_fsname != NULL)
177 kmem_free(mp->m_fsname, mp->m_fsname_len);
178 if (mp->m_rtname != NULL)
179 kmem_free(mp->m_rtname, strlen(mp->m_rtname) + 1);
180 if (mp->m_logname != NULL)
181 kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
183 xfs_icsb_destroy_counters(mp);
187 * Check size of device based on the (data/realtime) block count.
188 * Note: this check is used by the growfs code as well as mount.
191 xfs_sb_validate_fsb_count(
195 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
196 ASSERT(sbp->sb_blocklog >= BBSHIFT);
198 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
199 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
201 #else /* Limited by UINT_MAX of sectors */
202 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
209 * Check the validity of the SB found.
212 xfs_mount_validate_sb(
218 * If the log device and data device have the
219 * same device number, the log is internal.
220 * Consequently, the sb_logstart should be non-zero. If
221 * we have a zero sb_logstart in this case, we may be trying to mount
222 * a volume filesystem in a non-volume manner.
224 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
225 xfs_fs_mount_cmn_err(flags, "bad magic number");
226 return XFS_ERROR(EWRONGFS);
229 if (!xfs_sb_good_version(sbp)) {
230 xfs_fs_mount_cmn_err(flags, "bad version");
231 return XFS_ERROR(EWRONGFS);
235 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
236 xfs_fs_mount_cmn_err(flags,
237 "filesystem is marked as having an external log; "
238 "specify logdev on the\nmount command line.");
239 return XFS_ERROR(EINVAL);
243 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
244 xfs_fs_mount_cmn_err(flags,
245 "filesystem is marked as having an internal log; "
246 "do not specify logdev on\nthe mount command line.");
247 return XFS_ERROR(EINVAL);
251 * More sanity checking. These were stolen directly from
255 sbp->sb_agcount <= 0 ||
256 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
257 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
258 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
259 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
260 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
261 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
262 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
263 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
264 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
265 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
266 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
267 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
268 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
269 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
270 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
271 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
272 xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
273 return XFS_ERROR(EFSCORRUPTED);
277 * Sanity check AG count, size fields against data size field
280 sbp->sb_dblocks == 0 ||
282 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
283 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
284 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
285 xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
286 return XFS_ERROR(EFSCORRUPTED);
289 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
290 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
291 xfs_fs_mount_cmn_err(flags,
292 "file system too large to be mounted on this system.");
293 return XFS_ERROR(E2BIG);
296 if (unlikely(sbp->sb_inprogress)) {
297 xfs_fs_mount_cmn_err(flags, "file system busy");
298 return XFS_ERROR(EFSCORRUPTED);
302 * Version 1 directory format has never worked on Linux.
304 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
305 xfs_fs_mount_cmn_err(flags,
306 "file system using version 1 directory format");
307 return XFS_ERROR(ENOSYS);
311 * Until this is fixed only page-sized or smaller data blocks work.
313 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
314 xfs_fs_mount_cmn_err(flags,
315 "file system with blocksize %d bytes",
317 xfs_fs_mount_cmn_err(flags,
318 "only pagesize (%ld) or less will currently work.",
320 return XFS_ERROR(ENOSYS);
327 xfs_initialize_perag_icache(
330 if (!pag->pag_ici_init) {
331 rwlock_init(&pag->pag_ici_lock);
332 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
333 pag->pag_ici_init = 1;
338 xfs_initialize_perag(
340 xfs_agnumber_t agcount)
342 xfs_agnumber_t index, max_metadata;
346 xfs_sb_t *sbp = &mp->m_sb;
347 xfs_ino_t max_inum = XFS_MAXINUMBER_32;
349 /* Check to see if the filesystem can overflow 32 bit inodes */
350 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
351 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
353 /* Clear the mount flag if no inode can overflow 32 bits
354 * on this filesystem, or if specifically requested..
356 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
357 mp->m_flags |= XFS_MOUNT_32BITINODES;
359 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
362 /* If we can overflow then setup the ag headers accordingly */
363 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
364 /* Calculate how much should be reserved for inodes to
365 * meet the max inode percentage.
367 if (mp->m_maxicount) {
370 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
372 icount += sbp->sb_agblocks - 1;
373 do_div(icount, sbp->sb_agblocks);
374 max_metadata = icount;
376 max_metadata = agcount;
378 for (index = 0; index < agcount; index++) {
379 ino = XFS_AGINO_TO_INO(mp, index, agino);
380 if (ino > max_inum) {
385 /* This ag is preferred for inodes */
386 pag = &mp->m_perag[index];
387 pag->pagi_inodeok = 1;
388 if (index < max_metadata)
389 pag->pagf_metadata = 1;
390 xfs_initialize_perag_icache(pag);
393 /* Setup default behavior for smaller filesystems */
394 for (index = 0; index < agcount; index++) {
395 pag = &mp->m_perag[index];
396 pag->pagi_inodeok = 1;
397 xfs_initialize_perag_icache(pag);
408 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
409 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
410 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
411 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
412 to->sb_rextents = be64_to_cpu(from->sb_rextents);
413 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
414 to->sb_logstart = be64_to_cpu(from->sb_logstart);
415 to->sb_rootino = be64_to_cpu(from->sb_rootino);
416 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
417 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
418 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
419 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
420 to->sb_agcount = be32_to_cpu(from->sb_agcount);
421 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
422 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
423 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
424 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
425 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
426 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
427 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
428 to->sb_blocklog = from->sb_blocklog;
429 to->sb_sectlog = from->sb_sectlog;
430 to->sb_inodelog = from->sb_inodelog;
431 to->sb_inopblog = from->sb_inopblog;
432 to->sb_agblklog = from->sb_agblklog;
433 to->sb_rextslog = from->sb_rextslog;
434 to->sb_inprogress = from->sb_inprogress;
435 to->sb_imax_pct = from->sb_imax_pct;
436 to->sb_icount = be64_to_cpu(from->sb_icount);
437 to->sb_ifree = be64_to_cpu(from->sb_ifree);
438 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
439 to->sb_frextents = be64_to_cpu(from->sb_frextents);
440 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
441 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
442 to->sb_qflags = be16_to_cpu(from->sb_qflags);
443 to->sb_flags = from->sb_flags;
444 to->sb_shared_vn = from->sb_shared_vn;
445 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
446 to->sb_unit = be32_to_cpu(from->sb_unit);
447 to->sb_width = be32_to_cpu(from->sb_width);
448 to->sb_dirblklog = from->sb_dirblklog;
449 to->sb_logsectlog = from->sb_logsectlog;
450 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
451 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
452 to->sb_features2 = be32_to_cpu(from->sb_features2);
453 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
457 * Copy in core superblock to ondisk one.
459 * The fields argument is mask of superblock fields to copy.
467 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
468 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
478 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
479 first = xfs_sb_info[f].offset;
480 size = xfs_sb_info[f + 1].offset - first;
482 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
484 if (size == 1 || xfs_sb_info[f].type == 1) {
485 memcpy(to_ptr + first, from_ptr + first, size);
489 *(__be16 *)(to_ptr + first) =
490 cpu_to_be16(*(__u16 *)(from_ptr + first));
493 *(__be32 *)(to_ptr + first) =
494 cpu_to_be32(*(__u32 *)(from_ptr + first));
497 *(__be64 *)(to_ptr + first) =
498 cpu_to_be64(*(__u64 *)(from_ptr + first));
505 fields &= ~(1LL << f);
512 * Does the initial read of the superblock.
515 xfs_readsb(xfs_mount_t *mp, int flags)
517 unsigned int sector_size;
518 unsigned int extra_flags;
522 ASSERT(mp->m_sb_bp == NULL);
523 ASSERT(mp->m_ddev_targp != NULL);
526 * Allocate a (locked) buffer to hold the superblock.
527 * This will be kept around at all times to optimize
528 * access to the superblock.
530 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
531 extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
533 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
534 BTOBB(sector_size), extra_flags);
535 if (!bp || XFS_BUF_ISERROR(bp)) {
536 xfs_fs_mount_cmn_err(flags, "SB read failed");
537 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
540 ASSERT(XFS_BUF_ISBUSY(bp));
541 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
544 * Initialize the mount structure from the superblock.
545 * But first do some basic consistency checking.
547 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
549 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
551 xfs_fs_mount_cmn_err(flags, "SB validate failed");
556 * We must be able to do sector-sized and sector-aligned IO.
558 if (sector_size > mp->m_sb.sb_sectsize) {
559 xfs_fs_mount_cmn_err(flags,
560 "device supports only %u byte sectors (not %u)",
561 sector_size, mp->m_sb.sb_sectsize);
567 * If device sector size is smaller than the superblock size,
568 * re-read the superblock so the buffer is correctly sized.
570 if (sector_size < mp->m_sb.sb_sectsize) {
571 XFS_BUF_UNMANAGE(bp);
573 sector_size = mp->m_sb.sb_sectsize;
574 bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
575 BTOBB(sector_size), extra_flags);
576 if (!bp || XFS_BUF_ISERROR(bp)) {
577 xfs_fs_mount_cmn_err(flags, "SB re-read failed");
578 error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
581 ASSERT(XFS_BUF_ISBUSY(bp));
582 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
585 /* Initialize per-cpu counters */
586 xfs_icsb_reinit_counters(mp);
590 ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
595 XFS_BUF_UNMANAGE(bp);
605 * Mount initialization code establishing various mount
606 * fields from the superblock associated with the given
610 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
614 mp->m_agfrotor = mp->m_agirotor = 0;
615 spin_lock_init(&mp->m_agirotor_lock);
616 mp->m_maxagi = mp->m_sb.sb_agcount;
617 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
618 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
619 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
620 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
621 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
622 mp->m_litino = sbp->sb_inodesize -
623 ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
624 mp->m_blockmask = sbp->sb_blocksize - 1;
625 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
626 mp->m_blockwmask = mp->m_blockwsize - 1;
627 INIT_LIST_HEAD(&mp->m_del_inodes);
630 * Setup for attributes, in case they get created.
631 * This value is for inodes getting attributes for the first time,
632 * the per-inode value is for old attribute values.
634 ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
635 switch (sbp->sb_inodesize) {
637 mp->m_attroffset = XFS_LITINO(mp) -
638 XFS_BMDR_SPACE_CALC(MINABTPTRS);
643 mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
648 ASSERT(mp->m_attroffset < XFS_LITINO(mp));
650 for (i = 0; i < 2; i++) {
651 mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
653 mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
656 for (i = 0; i < 2; i++) {
657 mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
659 mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
662 for (i = 0; i < 2; i++) {
663 mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
665 mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
669 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
670 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
672 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
676 * xfs_initialize_perag_data
678 * Read in each per-ag structure so we can count up the number of
679 * allocated inodes, free inodes and used filesystem blocks as this
680 * information is no longer persistent in the superblock. Once we have
681 * this information, write it into the in-core superblock structure.
684 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
686 xfs_agnumber_t index;
688 xfs_sb_t *sbp = &mp->m_sb;
692 uint64_t bfreelst = 0;
696 for (index = 0; index < agcount; index++) {
698 * read the agf, then the agi. This gets us
699 * all the inforamtion we need and populates the
700 * per-ag structures for us.
702 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
706 error = xfs_ialloc_pagi_init(mp, NULL, index);
709 pag = &mp->m_perag[index];
710 ifree += pag->pagi_freecount;
711 ialloc += pag->pagi_count;
712 bfree += pag->pagf_freeblks;
713 bfreelst += pag->pagf_flcount;
714 btree += pag->pagf_btreeblks;
717 * Overwrite incore superblock counters with just-read data
719 spin_lock(&mp->m_sb_lock);
720 sbp->sb_ifree = ifree;
721 sbp->sb_icount = ialloc;
722 sbp->sb_fdblocks = bfree + bfreelst + btree;
723 spin_unlock(&mp->m_sb_lock);
725 /* Fixup the per-cpu counters as well. */
726 xfs_icsb_reinit_counters(mp);
732 * Update alignment values based on mount options and sb values
735 xfs_update_alignment(xfs_mount_t *mp, int mfsi_flags, __uint64_t *update_flags)
737 xfs_sb_t *sbp = &(mp->m_sb);
739 if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
741 * If stripe unit and stripe width are not multiples
742 * of the fs blocksize turn off alignment.
744 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
745 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
746 if (mp->m_flags & XFS_MOUNT_RETERR) {
748 "XFS: alignment check 1 failed");
749 return XFS_ERROR(EINVAL);
751 mp->m_dalign = mp->m_swidth = 0;
754 * Convert the stripe unit and width to FSBs.
756 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
757 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
758 if (mp->m_flags & XFS_MOUNT_RETERR) {
759 return XFS_ERROR(EINVAL);
761 xfs_fs_cmn_err(CE_WARN, mp,
762 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
763 mp->m_dalign, mp->m_swidth,
768 } else if (mp->m_dalign) {
769 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
771 if (mp->m_flags & XFS_MOUNT_RETERR) {
772 xfs_fs_cmn_err(CE_WARN, mp,
773 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
776 return XFS_ERROR(EINVAL);
783 * Update superblock with new values
786 if (xfs_sb_version_hasdalign(sbp)) {
787 if (sbp->sb_unit != mp->m_dalign) {
788 sbp->sb_unit = mp->m_dalign;
789 *update_flags |= XFS_SB_UNIT;
791 if (sbp->sb_width != mp->m_swidth) {
792 sbp->sb_width = mp->m_swidth;
793 *update_flags |= XFS_SB_WIDTH;
796 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
797 xfs_sb_version_hasdalign(&mp->m_sb)) {
798 mp->m_dalign = sbp->sb_unit;
799 mp->m_swidth = sbp->sb_width;
806 * Set the maximum inode count for this filesystem
809 xfs_set_maxicount(xfs_mount_t *mp)
811 xfs_sb_t *sbp = &(mp->m_sb);
814 if (sbp->sb_imax_pct) {
816 * Make sure the maximum inode count is a multiple
817 * of the units we allocate inodes in.
819 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
821 do_div(icount, mp->m_ialloc_blks);
822 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
830 * Set the default minimum read and write sizes unless
831 * already specified in a mount option.
832 * We use smaller I/O sizes when the file system
833 * is being used for NFS service (wsync mount option).
836 xfs_set_rw_sizes(xfs_mount_t *mp)
838 xfs_sb_t *sbp = &(mp->m_sb);
839 int readio_log, writeio_log;
841 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
842 if (mp->m_flags & XFS_MOUNT_WSYNC) {
843 readio_log = XFS_WSYNC_READIO_LOG;
844 writeio_log = XFS_WSYNC_WRITEIO_LOG;
846 readio_log = XFS_READIO_LOG_LARGE;
847 writeio_log = XFS_WRITEIO_LOG_LARGE;
850 readio_log = mp->m_readio_log;
851 writeio_log = mp->m_writeio_log;
854 if (sbp->sb_blocklog > readio_log) {
855 mp->m_readio_log = sbp->sb_blocklog;
857 mp->m_readio_log = readio_log;
859 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
860 if (sbp->sb_blocklog > writeio_log) {
861 mp->m_writeio_log = sbp->sb_blocklog;
863 mp->m_writeio_log = writeio_log;
865 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
869 * Set whether we're using inode alignment.
872 xfs_set_inoalignment(xfs_mount_t *mp)
874 if (xfs_sb_version_hasalign(&mp->m_sb) &&
875 mp->m_sb.sb_inoalignmt >=
876 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
877 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
879 mp->m_inoalign_mask = 0;
881 * If we are using stripe alignment, check whether
882 * the stripe unit is a multiple of the inode alignment
884 if (mp->m_dalign && mp->m_inoalign_mask &&
885 !(mp->m_dalign & mp->m_inoalign_mask))
886 mp->m_sinoalign = mp->m_dalign;
892 * Check that the data (and log if separate) are an ok size.
895 xfs_check_sizes(xfs_mount_t *mp, int mfsi_flags)
901 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
902 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
903 cmn_err(CE_WARN, "XFS: size check 1 failed");
904 return XFS_ERROR(E2BIG);
906 error = xfs_read_buf(mp, mp->m_ddev_targp,
907 d - XFS_FSS_TO_BB(mp, 1),
908 XFS_FSS_TO_BB(mp, 1), 0, &bp);
912 cmn_err(CE_WARN, "XFS: size check 2 failed");
914 error = XFS_ERROR(E2BIG);
918 if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
919 mp->m_logdev_targp != mp->m_ddev_targp) {
920 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
921 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
922 cmn_err(CE_WARN, "XFS: size check 3 failed");
923 return XFS_ERROR(E2BIG);
925 error = xfs_read_buf(mp, mp->m_logdev_targp,
926 d - XFS_FSB_TO_BB(mp, 1),
927 XFS_FSB_TO_BB(mp, 1), 0, &bp);
931 cmn_err(CE_WARN, "XFS: size check 3 failed");
933 error = XFS_ERROR(E2BIG);
943 * This function does the following on an initial mount of a file system:
944 * - reads the superblock from disk and init the mount struct
945 * - if we're a 32-bit kernel, do a size check on the superblock
946 * so we don't mount terabyte filesystems
947 * - init mount struct realtime fields
948 * - allocate inode hash table for fs
949 * - init directory manager
950 * - perform recovery and init the log manager
957 xfs_sb_t *sbp = &(mp->m_sb);
959 bhv_vnode_t *rvp = NULL;
961 __int64_t update_flags = 0LL;
962 uint quotamount, quotaflags;
964 int uuid_mounted = 0;
967 xfs_mount_common(mp, sbp);
970 * Check for a mismatched features2 values. Older kernels
971 * read & wrote into the wrong sb offset for sb_features2
972 * on some platforms due to xfs_sb_t not being 64bit size aligned
973 * when sb_features2 was added, which made older superblock
974 * reading/writing routines swap it as a 64-bit value.
976 * For backwards compatibility, we make both slots equal.
978 * If we detect a mismatched field, we OR the set bits into the
979 * existing features2 field in case it has already been modified; we
980 * don't want to lose any features. We then update the bad location
981 * with the ORed value so that older kernels will see any features2
982 * flags, and mark the two fields as needing updates once the
983 * transaction subsystem is online.
985 if (xfs_sb_has_mismatched_features2(sbp)) {
987 "XFS: correcting sb_features alignment problem");
988 sbp->sb_features2 |= sbp->sb_bad_features2;
989 sbp->sb_bad_features2 = sbp->sb_features2;
990 update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
993 * Re-check for ATTR2 in case it was found in bad_features2
996 if (xfs_sb_version_hasattr2(&mp->m_sb))
997 mp->m_flags |= XFS_MOUNT_ATTR2;
1002 * Check if sb_agblocks is aligned at stripe boundary
1003 * If sb_agblocks is NOT aligned turn off m_dalign since
1004 * allocator alignment is within an ag, therefore ag has
1005 * to be aligned at stripe boundary.
1007 error = xfs_update_alignment(mp, mfsi_flags, &update_flags);
1011 xfs_alloc_compute_maxlevels(mp);
1012 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1013 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1014 xfs_ialloc_compute_maxlevels(mp);
1016 xfs_set_maxicount(mp);
1018 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1021 * XFS uses the uuid from the superblock as the unique
1022 * identifier for fsid. We can not use the uuid from the volume
1023 * since a single partition filesystem is identical to a single
1024 * partition volume/filesystem.
1026 if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
1027 (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
1028 if (xfs_uuid_mount(mp)) {
1029 error = XFS_ERROR(EINVAL);
1036 * Set the minimum read and write sizes
1038 xfs_set_rw_sizes(mp);
1041 * Set the inode cluster size.
1042 * This may still be overridden by the file system
1043 * block size if it is larger than the chosen cluster size.
1045 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1048 * Set inode alignment fields
1050 xfs_set_inoalignment(mp);
1053 * Check that the data (and log if separate) are an ok size.
1055 error = xfs_check_sizes(mp, mfsi_flags);
1060 * Initialize realtime fields in the mount structure
1062 error = xfs_rtmount_init(mp);
1064 cmn_err(CE_WARN, "XFS: RT mount failed");
1069 * For client case we are done now
1071 if (mfsi_flags & XFS_MFSI_CLIENT) {
1076 * Copies the low order bits of the timestamp and the randomly
1077 * set "sequence" number out of a UUID.
1079 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1081 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1086 * Initialize the attribute manager's entries.
1088 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1091 * Initialize the precomputed transaction reservations values.
1096 * Allocate and initialize the per-ag data.
1098 init_rwsem(&mp->m_peraglock);
1100 kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
1102 mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
1105 * log's mount-time initialization. Perform 1st part recovery if needed
1107 if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */
1108 error = xfs_log_mount(mp, mp->m_logdev_targp,
1109 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1110 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1112 cmn_err(CE_WARN, "XFS: log mount failed");
1115 } else { /* No log has been defined */
1116 cmn_err(CE_WARN, "XFS: no log defined");
1117 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
1118 error = XFS_ERROR(EFSCORRUPTED);
1123 * Now the log is mounted, we know if it was an unclean shutdown or
1124 * not. If it was, with the first phase of recovery has completed, we
1125 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1126 * but they are recovered transactionally in the second recovery phase
1129 * Hence we can safely re-initialise incore superblock counters from
1130 * the per-ag data. These may not be correct if the filesystem was not
1131 * cleanly unmounted, so we need to wait for recovery to finish before
1134 * If the filesystem was cleanly unmounted, then we can trust the
1135 * values in the superblock to be correct and we don't need to do
1138 * If we are currently making the filesystem, the initialisation will
1139 * fail as the perag data is in an undefined state.
1142 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1143 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1144 !mp->m_sb.sb_inprogress) {
1145 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1151 * Get and sanity-check the root inode.
1152 * Save the pointer to it in the mount structure.
1154 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
1156 cmn_err(CE_WARN, "XFS: failed to read root inode");
1160 ASSERT(rip != NULL);
1161 rvp = XFS_ITOV(rip);
1163 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1164 cmn_err(CE_WARN, "XFS: corrupted root inode");
1165 cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1166 XFS_BUFTARG_NAME(mp->m_ddev_targp),
1167 (unsigned long long)rip->i_ino);
1168 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1169 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1171 error = XFS_ERROR(EFSCORRUPTED);
1174 mp->m_rootip = rip; /* save it */
1176 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1179 * Initialize realtime inode pointers in the mount structure
1181 error = xfs_rtmount_inodes(mp);
1184 * Free up the root inode.
1186 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1191 * If fs is not mounted readonly, then update the superblock changes.
1193 if (update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY))
1194 xfs_mount_log_sb(mp, update_flags);
1197 * Initialise the XFS quota management subsystem for this mount
1199 error = XFS_QM_INIT(mp, "amount, "aflags);
1204 * Finish recovering the file system. This part needed to be
1205 * delayed until after the root and real-time bitmap inodes
1206 * were consistently read in.
1208 error = xfs_log_mount_finish(mp, mfsi_flags);
1210 cmn_err(CE_WARN, "XFS: log mount finish failed");
1215 * Complete the quota initialisation, post-log-replay component.
1217 error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags);
1222 * Now we are mounted, reserve a small amount of unused space for
1223 * privileged transactions. This is needed so that transaction
1224 * space required for critical operations can dip into this pool
1225 * when at ENOSPC. This is needed for operations like create with
1226 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1227 * are not allowed to use this reserved space.
1229 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1230 * This may drive us straight to ENOSPC on mount, but that implies
1231 * we were already there on the last unmount.
1233 resblks = mp->m_sb.sb_dblocks;
1234 do_div(resblks, 20);
1235 resblks = min_t(__uint64_t, resblks, 1024);
1236 xfs_reserve_blocks(mp, &resblks, NULL);
1242 * Free up the root inode.
1246 xfs_log_unmount_dealloc(mp);
1248 for (agno = 0; agno < sbp->sb_agcount; agno++)
1249 if (mp->m_perag[agno].pagb_list)
1250 kmem_free(mp->m_perag[agno].pagb_list,
1251 sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1252 kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1257 xfs_uuid_unmount(mp);
1265 * This flushes out the inodes,dquots and the superblock, unmounts the
1266 * log and makes sure that incore structures are freed.
1269 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1274 * We can potentially deadlock here if we have an inode cluster
1275 * that has been freed has it's buffer still pinned in memory because
1276 * the transaction is still sitting in a iclog. The stale inodes
1277 * on that buffer will have their flush locks held until the
1278 * transaction hits the disk and the callbacks run. the inode
1279 * flush takes the flush lock unconditionally and with nothing to
1280 * push out the iclog we will never get that unlocked. hence we
1281 * need to force the log first.
1283 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1286 XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
1289 * Flush out the log synchronously so that we know for sure
1290 * that nothing is pinned. This is important because bflush()
1291 * will skip pinned buffers.
1293 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1295 xfs_binval(mp->m_ddev_targp);
1296 if (mp->m_rtdev_targp) {
1297 xfs_binval(mp->m_rtdev_targp);
1301 * Unreserve any blocks we have so that when we unmount we don't account
1302 * the reserved free space as used. This is really only necessary for
1303 * lazy superblock counting because it trusts the incore superblock
1304 * counters to be aboslutely correct on clean unmount.
1306 * We don't bother correcting this elsewhere for lazy superblock
1307 * counting because on mount of an unclean filesystem we reconstruct the
1308 * correct counter value and this is irrelevant.
1310 * For non-lazy counter filesystems, this doesn't matter at all because
1311 * we only every apply deltas to the superblock and hence the incore
1312 * value does not matter....
1315 xfs_reserve_blocks(mp, &resblks, NULL);
1317 xfs_log_sbcount(mp, 1);
1318 xfs_unmountfs_writesb(mp);
1319 xfs_unmountfs_wait(mp); /* wait for async bufs */
1320 xfs_log_unmount(mp); /* Done! No more fs ops. */
1325 * All inodes from this mount point should be freed.
1327 ASSERT(mp->m_inodes == NULL);
1329 xfs_unmountfs_close(mp, cr);
1330 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1331 xfs_uuid_unmount(mp);
1333 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1334 xfs_errortag_clearall(mp, 0);
1341 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1343 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1344 xfs_free_buftarg(mp->m_logdev_targp, 1);
1345 if (mp->m_rtdev_targp)
1346 xfs_free_buftarg(mp->m_rtdev_targp, 1);
1347 xfs_free_buftarg(mp->m_ddev_targp, 0);
1351 xfs_unmountfs_wait(xfs_mount_t *mp)
1353 if (mp->m_logdev_targp != mp->m_ddev_targp)
1354 xfs_wait_buftarg(mp->m_logdev_targp);
1355 if (mp->m_rtdev_targp)
1356 xfs_wait_buftarg(mp->m_rtdev_targp);
1357 xfs_wait_buftarg(mp->m_ddev_targp);
1361 xfs_fs_writable(xfs_mount_t *mp)
1363 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1364 (mp->m_flags & XFS_MOUNT_RDONLY));
1370 * Called either periodically to keep the on disk superblock values
1371 * roughly up to date or from unmount to make sure the values are
1372 * correct on a clean unmount.
1374 * Note this code can be called during the process of freezing, so
1375 * we may need to use the transaction allocator which does not not
1376 * block when the transaction subsystem is in its frozen state.
1386 if (!xfs_fs_writable(mp))
1389 xfs_icsb_sync_counters(mp);
1392 * we don't need to do this if we are updating the superblock
1393 * counters on every modification.
1395 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1398 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT);
1399 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1400 XFS_DEFAULT_LOG_COUNT);
1402 xfs_trans_cancel(tp, 0);
1406 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1408 xfs_trans_set_sync(tp);
1409 xfs_trans_commit(tp, 0);
1419 xfs_dsb_t *sb = XFS_BUF_TO_SBP(bp);
1422 if (!(sb->sb_flags & XFS_SBF_READONLY))
1423 sb->sb_flags |= XFS_SBF_READONLY;
1425 version = be16_to_cpu(sb->sb_versionnum);
1426 if ((version & XFS_SB_VERSION_NUMBITS) != XFS_SB_VERSION_4 ||
1427 !(version & XFS_SB_VERSION_SHAREDBIT))
1428 version |= XFS_SB_VERSION_SHAREDBIT;
1429 sb->sb_versionnum = cpu_to_be16(version);
1433 xfs_unmountfs_writesb(xfs_mount_t *mp)
1439 * skip superblock write if fs is read-only, or
1440 * if we are doing a forced umount.
1442 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1443 XFS_FORCED_SHUTDOWN(mp))) {
1445 sbp = xfs_getsb(mp, 0);
1448 * mark shared-readonly if desired
1450 if (mp->m_mk_sharedro)
1451 xfs_mark_shared_ro(mp, sbp);
1453 XFS_BUF_UNDONE(sbp);
1454 XFS_BUF_UNREAD(sbp);
1455 XFS_BUF_UNDELAYWRITE(sbp);
1457 XFS_BUF_UNASYNC(sbp);
1458 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1459 xfsbdstrat(mp, sbp);
1460 /* Nevermind errors we might get here. */
1461 error = xfs_iowait(sbp);
1463 xfs_ioerror_alert("xfs_unmountfs_writesb",
1464 mp, sbp, XFS_BUF_ADDR(sbp));
1465 if (error && mp->m_mk_sharedro)
1466 xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1473 * xfs_mod_sb() can be used to copy arbitrary changes to the
1474 * in-core superblock into the superblock buffer to be logged.
1475 * It does not provide the higher level of locking that is
1476 * needed to protect the in-core superblock from concurrent
1480 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1492 bp = xfs_trans_getsb(tp, mp, 0);
1493 first = sizeof(xfs_sb_t);
1496 /* translate/copy */
1498 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1500 /* find modified range */
1502 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1503 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1504 first = xfs_sb_info[f].offset;
1506 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1507 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1508 last = xfs_sb_info[f + 1].offset - 1;
1510 xfs_trans_log_buf(tp, bp, first, last);
1515 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1516 * a delta to a specified field in the in-core superblock. Simply
1517 * switch on the field indicated and apply the delta to that field.
1518 * Fields are not allowed to dip below zero, so if the delta would
1519 * do this do not apply it and return EINVAL.
1521 * The m_sb_lock must be held when this routine is called.
1524 xfs_mod_incore_sb_unlocked(
1526 xfs_sb_field_t field,
1530 int scounter; /* short counter for 32 bit fields */
1531 long long lcounter; /* long counter for 64 bit fields */
1532 long long res_used, rem;
1535 * With the in-core superblock spin lock held, switch
1536 * on the indicated field. Apply the delta to the
1537 * proper field. If the fields value would dip below
1538 * 0, then do not apply the delta and return EINVAL.
1541 case XFS_SBS_ICOUNT:
1542 lcounter = (long long)mp->m_sb.sb_icount;
1546 return XFS_ERROR(EINVAL);
1548 mp->m_sb.sb_icount = lcounter;
1551 lcounter = (long long)mp->m_sb.sb_ifree;
1555 return XFS_ERROR(EINVAL);
1557 mp->m_sb.sb_ifree = lcounter;
1559 case XFS_SBS_FDBLOCKS:
1560 lcounter = (long long)
1561 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1562 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1564 if (delta > 0) { /* Putting blocks back */
1565 if (res_used > delta) {
1566 mp->m_resblks_avail += delta;
1568 rem = delta - res_used;
1569 mp->m_resblks_avail = mp->m_resblks;
1572 } else { /* Taking blocks away */
1577 * If were out of blocks, use any available reserved blocks if
1583 lcounter = (long long)mp->m_resblks_avail + delta;
1585 return XFS_ERROR(ENOSPC);
1587 mp->m_resblks_avail = lcounter;
1589 } else { /* not reserved */
1590 return XFS_ERROR(ENOSPC);
1595 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1597 case XFS_SBS_FREXTENTS:
1598 lcounter = (long long)mp->m_sb.sb_frextents;
1601 return XFS_ERROR(ENOSPC);
1603 mp->m_sb.sb_frextents = lcounter;
1605 case XFS_SBS_DBLOCKS:
1606 lcounter = (long long)mp->m_sb.sb_dblocks;
1610 return XFS_ERROR(EINVAL);
1612 mp->m_sb.sb_dblocks = lcounter;
1614 case XFS_SBS_AGCOUNT:
1615 scounter = mp->m_sb.sb_agcount;
1619 return XFS_ERROR(EINVAL);
1621 mp->m_sb.sb_agcount = scounter;
1623 case XFS_SBS_IMAX_PCT:
1624 scounter = mp->m_sb.sb_imax_pct;
1628 return XFS_ERROR(EINVAL);
1630 mp->m_sb.sb_imax_pct = scounter;
1632 case XFS_SBS_REXTSIZE:
1633 scounter = mp->m_sb.sb_rextsize;
1637 return XFS_ERROR(EINVAL);
1639 mp->m_sb.sb_rextsize = scounter;
1641 case XFS_SBS_RBMBLOCKS:
1642 scounter = mp->m_sb.sb_rbmblocks;
1646 return XFS_ERROR(EINVAL);
1648 mp->m_sb.sb_rbmblocks = scounter;
1650 case XFS_SBS_RBLOCKS:
1651 lcounter = (long long)mp->m_sb.sb_rblocks;
1655 return XFS_ERROR(EINVAL);
1657 mp->m_sb.sb_rblocks = lcounter;
1659 case XFS_SBS_REXTENTS:
1660 lcounter = (long long)mp->m_sb.sb_rextents;
1664 return XFS_ERROR(EINVAL);
1666 mp->m_sb.sb_rextents = lcounter;
1668 case XFS_SBS_REXTSLOG:
1669 scounter = mp->m_sb.sb_rextslog;
1673 return XFS_ERROR(EINVAL);
1675 mp->m_sb.sb_rextslog = scounter;
1679 return XFS_ERROR(EINVAL);
1684 * xfs_mod_incore_sb() is used to change a field in the in-core
1685 * superblock structure by the specified delta. This modification
1686 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1687 * routine to do the work.
1692 xfs_sb_field_t field,
1698 /* check for per-cpu counters */
1700 #ifdef HAVE_PERCPU_SB
1701 case XFS_SBS_ICOUNT:
1703 case XFS_SBS_FDBLOCKS:
1704 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1705 status = xfs_icsb_modify_counters(mp, field,
1712 spin_lock(&mp->m_sb_lock);
1713 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1714 spin_unlock(&mp->m_sb_lock);
1722 * xfs_mod_incore_sb_batch() is used to change more than one field
1723 * in the in-core superblock structure at a time. This modification
1724 * is protected by a lock internal to this module. The fields and
1725 * changes to those fields are specified in the array of xfs_mod_sb
1726 * structures passed in.
1728 * Either all of the specified deltas will be applied or none of
1729 * them will. If any modified field dips below 0, then all modifications
1730 * will be backed out and EINVAL will be returned.
1733 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1739 * Loop through the array of mod structures and apply each
1740 * individually. If any fail, then back out all those
1741 * which have already been applied. Do all of this within
1742 * the scope of the m_sb_lock so that all of the changes will
1745 spin_lock(&mp->m_sb_lock);
1747 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1749 * Apply the delta at index n. If it fails, break
1750 * from the loop so we'll fall into the undo loop
1753 switch (msbp->msb_field) {
1754 #ifdef HAVE_PERCPU_SB
1755 case XFS_SBS_ICOUNT:
1757 case XFS_SBS_FDBLOCKS:
1758 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1759 spin_unlock(&mp->m_sb_lock);
1760 status = xfs_icsb_modify_counters(mp,
1762 msbp->msb_delta, rsvd);
1763 spin_lock(&mp->m_sb_lock);
1769 status = xfs_mod_incore_sb_unlocked(mp,
1771 msbp->msb_delta, rsvd);
1781 * If we didn't complete the loop above, then back out
1782 * any changes made to the superblock. If you add code
1783 * between the loop above and here, make sure that you
1784 * preserve the value of status. Loop back until
1785 * we step below the beginning of the array. Make sure
1786 * we don't touch anything back there.
1790 while (msbp >= msb) {
1791 switch (msbp->msb_field) {
1792 #ifdef HAVE_PERCPU_SB
1793 case XFS_SBS_ICOUNT:
1795 case XFS_SBS_FDBLOCKS:
1796 if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1797 spin_unlock(&mp->m_sb_lock);
1798 status = xfs_icsb_modify_counters(mp,
1802 spin_lock(&mp->m_sb_lock);
1808 status = xfs_mod_incore_sb_unlocked(mp,
1814 ASSERT(status == 0);
1818 spin_unlock(&mp->m_sb_lock);
1823 * xfs_getsb() is called to obtain the buffer for the superblock.
1824 * The buffer is returned locked and read in from disk.
1825 * The buffer should be released with a call to xfs_brelse().
1827 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1828 * the superblock buffer if it can be locked without sleeping.
1829 * If it can't then we'll return NULL.
1838 ASSERT(mp->m_sb_bp != NULL);
1840 if (flags & XFS_BUF_TRYLOCK) {
1841 if (!XFS_BUF_CPSEMA(bp)) {
1845 XFS_BUF_PSEMA(bp, PRIBIO);
1848 ASSERT(XFS_BUF_ISDONE(bp));
1853 * Used to free the superblock along various error paths.
1862 * Use xfs_getsb() so that the buffer will be locked
1863 * when we call xfs_buf_relse().
1865 bp = xfs_getsb(mp, 0);
1866 XFS_BUF_UNMANAGE(bp);
1872 * See if the UUID is unique among mounted XFS filesystems.
1873 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1879 if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1881 "XFS: Filesystem %s has nil UUID - can't mount",
1885 if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1887 "XFS: Filesystem %s has duplicate UUID - can't mount",
1895 * Remove filesystem from the UUID table.
1901 uuid_table_remove(&mp->m_sb.sb_uuid);
1905 * Used to log changes to the superblock unit and width fields which could
1906 * be altered by the mount options, as well as any potential sb_features2
1907 * fixup. Only the first superblock is updated.
1916 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1917 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2));
1919 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1920 if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1921 XFS_DEFAULT_LOG_COUNT)) {
1922 xfs_trans_cancel(tp, 0);
1925 xfs_mod_sb(tp, fields);
1926 xfs_trans_commit(tp, 0);
1930 #ifdef HAVE_PERCPU_SB
1932 * Per-cpu incore superblock counters
1934 * Simple concept, difficult implementation
1936 * Basically, replace the incore superblock counters with a distributed per cpu
1937 * counter for contended fields (e.g. free block count).
1939 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1940 * hence needs to be accurately read when we are running low on space. Hence
1941 * there is a method to enable and disable the per-cpu counters based on how
1942 * much "stuff" is available in them.
1944 * Basically, a counter is enabled if there is enough free resource to justify
1945 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1946 * ENOSPC), then we disable the counters to synchronise all callers and
1947 * re-distribute the available resources.
1949 * If, once we redistributed the available resources, we still get a failure,
1950 * we disable the per-cpu counter and go through the slow path.
1952 * The slow path is the current xfs_mod_incore_sb() function. This means that
1953 * when we disable a per-cpu counter, we need to drain it's resources back to
1954 * the global superblock. We do this after disabling the counter to prevent
1955 * more threads from queueing up on the counter.
1957 * Essentially, this means that we still need a lock in the fast path to enable
1958 * synchronisation between the global counters and the per-cpu counters. This
1959 * is not a problem because the lock will be local to a CPU almost all the time
1960 * and have little contention except when we get to ENOSPC conditions.
1962 * Basically, this lock becomes a barrier that enables us to lock out the fast
1963 * path while we do things like enabling and disabling counters and
1964 * synchronising the counters.
1968 * 1. m_sb_lock before picking up per-cpu locks
1969 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1970 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1971 * 4. modifying per-cpu counters requires holding per-cpu lock
1972 * 5. modifying global counters requires holding m_sb_lock
1973 * 6. enabling or disabling a counter requires holding the m_sb_lock
1974 * and _none_ of the per-cpu locks.
1976 * Disabled counters are only ever re-enabled by a balance operation
1977 * that results in more free resources per CPU than a given threshold.
1978 * To ensure counters don't remain disabled, they are rebalanced when
1979 * the global resource goes above a higher threshold (i.e. some hysteresis
1980 * is present to prevent thrashing).
1983 #ifdef CONFIG_HOTPLUG_CPU
1985 * hot-plug CPU notifier support.
1987 * We need a notifier per filesystem as we need to be able to identify
1988 * the filesystem to balance the counters out. This is achieved by
1989 * having a notifier block embedded in the xfs_mount_t and doing pointer
1990 * magic to get the mount pointer from the notifier block address.
1993 xfs_icsb_cpu_notify(
1994 struct notifier_block *nfb,
1995 unsigned long action,
1998 xfs_icsb_cnts_t *cntp;
2001 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2002 cntp = (xfs_icsb_cnts_t *)
2003 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2005 case CPU_UP_PREPARE:
2006 case CPU_UP_PREPARE_FROZEN:
2007 /* Easy Case - initialize the area and locks, and
2008 * then rebalance when online does everything else for us. */
2009 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2012 case CPU_ONLINE_FROZEN:
2014 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
2015 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
2016 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
2017 xfs_icsb_unlock(mp);
2020 case CPU_DEAD_FROZEN:
2021 /* Disable all the counters, then fold the dead cpu's
2022 * count into the total on the global superblock and
2023 * re-enable the counters. */
2025 spin_lock(&mp->m_sb_lock);
2026 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2027 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2028 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2030 mp->m_sb.sb_icount += cntp->icsb_icount;
2031 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2032 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2034 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2036 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT,
2037 XFS_ICSB_SB_LOCKED, 0);
2038 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE,
2039 XFS_ICSB_SB_LOCKED, 0);
2040 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS,
2041 XFS_ICSB_SB_LOCKED, 0);
2042 spin_unlock(&mp->m_sb_lock);
2043 xfs_icsb_unlock(mp);
2049 #endif /* CONFIG_HOTPLUG_CPU */
2052 xfs_icsb_init_counters(
2055 xfs_icsb_cnts_t *cntp;
2058 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2059 if (mp->m_sb_cnts == NULL)
2062 #ifdef CONFIG_HOTPLUG_CPU
2063 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2064 mp->m_icsb_notifier.priority = 0;
2065 register_hotcpu_notifier(&mp->m_icsb_notifier);
2066 #endif /* CONFIG_HOTPLUG_CPU */
2068 for_each_online_cpu(i) {
2069 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2070 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2073 mutex_init(&mp->m_icsb_mutex);
2076 * start with all counters disabled so that the
2077 * initial balance kicks us off correctly
2079 mp->m_icsb_counters = -1;
2084 xfs_icsb_reinit_counters(
2089 * start with all counters disabled so that the
2090 * initial balance kicks us off correctly
2092 mp->m_icsb_counters = -1;
2093 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
2094 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
2095 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
2096 xfs_icsb_unlock(mp);
2100 xfs_icsb_destroy_counters(
2103 if (mp->m_sb_cnts) {
2104 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2105 free_percpu(mp->m_sb_cnts);
2107 mutex_destroy(&mp->m_icsb_mutex);
2112 xfs_icsb_cnts_t *icsbp)
2114 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2120 xfs_icsb_unlock_cntr(
2121 xfs_icsb_cnts_t *icsbp)
2123 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2128 xfs_icsb_lock_all_counters(
2131 xfs_icsb_cnts_t *cntp;
2134 for_each_online_cpu(i) {
2135 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2136 xfs_icsb_lock_cntr(cntp);
2141 xfs_icsb_unlock_all_counters(
2144 xfs_icsb_cnts_t *cntp;
2147 for_each_online_cpu(i) {
2148 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2149 xfs_icsb_unlock_cntr(cntp);
2156 xfs_icsb_cnts_t *cnt,
2159 xfs_icsb_cnts_t *cntp;
2162 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2164 if (!(flags & XFS_ICSB_LAZY_COUNT))
2165 xfs_icsb_lock_all_counters(mp);
2167 for_each_online_cpu(i) {
2168 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2169 cnt->icsb_icount += cntp->icsb_icount;
2170 cnt->icsb_ifree += cntp->icsb_ifree;
2171 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2174 if (!(flags & XFS_ICSB_LAZY_COUNT))
2175 xfs_icsb_unlock_all_counters(mp);
2179 xfs_icsb_counter_disabled(
2181 xfs_sb_field_t field)
2183 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2184 return test_bit(field, &mp->m_icsb_counters);
2188 xfs_icsb_disable_counter(
2190 xfs_sb_field_t field)
2192 xfs_icsb_cnts_t cnt;
2194 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2197 * If we are already disabled, then there is nothing to do
2198 * here. We check before locking all the counters to avoid
2199 * the expensive lock operation when being called in the
2200 * slow path and the counter is already disabled. This is
2201 * safe because the only time we set or clear this state is under
2204 if (xfs_icsb_counter_disabled(mp, field))
2207 xfs_icsb_lock_all_counters(mp);
2208 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2209 /* drain back to superblock */
2211 xfs_icsb_count(mp, &cnt, XFS_ICSB_SB_LOCKED|XFS_ICSB_LAZY_COUNT);
2213 case XFS_SBS_ICOUNT:
2214 mp->m_sb.sb_icount = cnt.icsb_icount;
2217 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2219 case XFS_SBS_FDBLOCKS:
2220 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2227 xfs_icsb_unlock_all_counters(mp);
2233 xfs_icsb_enable_counter(
2235 xfs_sb_field_t field,
2239 xfs_icsb_cnts_t *cntp;
2242 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2244 xfs_icsb_lock_all_counters(mp);
2245 for_each_online_cpu(i) {
2246 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2248 case XFS_SBS_ICOUNT:
2249 cntp->icsb_icount = count + resid;
2252 cntp->icsb_ifree = count + resid;
2254 case XFS_SBS_FDBLOCKS:
2255 cntp->icsb_fdblocks = count + resid;
2263 clear_bit(field, &mp->m_icsb_counters);
2264 xfs_icsb_unlock_all_counters(mp);
2268 xfs_icsb_sync_counters_flags(
2272 xfs_icsb_cnts_t cnt;
2274 /* Pass 1: lock all counters */
2275 if ((flags & XFS_ICSB_SB_LOCKED) == 0)
2276 spin_lock(&mp->m_sb_lock);
2278 xfs_icsb_count(mp, &cnt, flags);
2280 /* Step 3: update mp->m_sb fields */
2281 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2282 mp->m_sb.sb_icount = cnt.icsb_icount;
2283 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2284 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2285 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2286 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2288 if ((flags & XFS_ICSB_SB_LOCKED) == 0)
2289 spin_unlock(&mp->m_sb_lock);
2293 * Accurate update of per-cpu counters to incore superblock
2296 xfs_icsb_sync_counters(
2299 xfs_icsb_sync_counters_flags(mp, 0);
2303 * Balance and enable/disable counters as necessary.
2305 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2306 * chosen to be the same number as single on disk allocation chunk per CPU, and
2307 * free blocks is something far enough zero that we aren't going thrash when we
2308 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2309 * prevent looping endlessly when xfs_alloc_space asks for more than will
2310 * be distributed to a single CPU but each CPU has enough blocks to be
2313 * Note that we can be called when counters are already disabled.
2314 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2315 * prevent locking every per-cpu counter needlessly.
2318 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2319 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2320 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2322 xfs_icsb_balance_counter(
2324 xfs_sb_field_t field,
2328 uint64_t count, resid;
2329 int weight = num_online_cpus();
2330 uint64_t min = (uint64_t)min_per_cpu;
2332 if (!(flags & XFS_ICSB_SB_LOCKED))
2333 spin_lock(&mp->m_sb_lock);
2335 /* disable counter and sync counter */
2336 xfs_icsb_disable_counter(mp, field);
2338 /* update counters - first CPU gets residual*/
2340 case XFS_SBS_ICOUNT:
2341 count = mp->m_sb.sb_icount;
2342 resid = do_div(count, weight);
2343 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2347 count = mp->m_sb.sb_ifree;
2348 resid = do_div(count, weight);
2349 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2352 case XFS_SBS_FDBLOCKS:
2353 count = mp->m_sb.sb_fdblocks;
2354 resid = do_div(count, weight);
2355 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2360 count = resid = 0; /* quiet, gcc */
2364 xfs_icsb_enable_counter(mp, field, count, resid);
2366 if (!(flags & XFS_ICSB_SB_LOCKED))
2367 spin_unlock(&mp->m_sb_lock);
2371 xfs_icsb_modify_counters(
2373 xfs_sb_field_t field,
2377 xfs_icsb_cnts_t *icsbp;
2378 long long lcounter; /* long counter for 64 bit fields */
2384 icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2387 * if the counter is disabled, go to slow path
2389 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2391 xfs_icsb_lock_cntr(icsbp);
2392 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2393 xfs_icsb_unlock_cntr(icsbp);
2398 case XFS_SBS_ICOUNT:
2399 lcounter = icsbp->icsb_icount;
2401 if (unlikely(lcounter < 0))
2402 goto balance_counter;
2403 icsbp->icsb_icount = lcounter;
2407 lcounter = icsbp->icsb_ifree;
2409 if (unlikely(lcounter < 0))
2410 goto balance_counter;
2411 icsbp->icsb_ifree = lcounter;
2414 case XFS_SBS_FDBLOCKS:
2415 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2417 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2419 if (unlikely(lcounter < 0))
2420 goto balance_counter;
2421 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2427 xfs_icsb_unlock_cntr(icsbp);
2435 * serialise with a mutex so we don't burn lots of cpu on
2436 * the superblock lock. We still need to hold the superblock
2437 * lock, however, when we modify the global structures.
2442 * Now running atomically.
2444 * If the counter is enabled, someone has beaten us to rebalancing.
2445 * Drop the lock and try again in the fast path....
2447 if (!(xfs_icsb_counter_disabled(mp, field))) {
2448 xfs_icsb_unlock(mp);
2453 * The counter is currently disabled. Because we are
2454 * running atomically here, we know a rebalance cannot
2455 * be in progress. Hence we can go straight to operating
2456 * on the global superblock. We do not call xfs_mod_incore_sb()
2457 * here even though we need to get the m_sb_lock. Doing so
2458 * will cause us to re-enter this function and deadlock.
2459 * Hence we get the m_sb_lock ourselves and then call
2460 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2461 * directly on the global counters.
2463 spin_lock(&mp->m_sb_lock);
2464 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2465 spin_unlock(&mp->m_sb_lock);
2468 * Now that we've modified the global superblock, we
2469 * may be able to re-enable the distributed counters
2470 * (e.g. lots of space just got freed). After that
2474 xfs_icsb_balance_counter(mp, field, 0, 0);
2475 xfs_icsb_unlock(mp);
2479 xfs_icsb_unlock_cntr(icsbp);
2483 * We may have multiple threads here if multiple per-cpu
2484 * counters run dry at the same time. This will mean we can
2485 * do more balances than strictly necessary but it is not
2486 * the common slowpath case.
2491 * running atomically.
2493 * This will leave the counter in the correct state for future
2494 * accesses. After the rebalance, we simply try again and our retry
2495 * will either succeed through the fast path or slow path without
2496 * another balance operation being required.
2498 xfs_icsb_balance_counter(mp, field, 0, delta);
2499 xfs_icsb_unlock(mp);