bridge: fix a possible use after free
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / xfs / xfs_mount.c
blob0081657ad985ee27417ba4b73c142cb3d43e78b7
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_ialloc.h"
36 #include "xfs_alloc.h"
37 #include "xfs_rtalloc.h"
38 #include "xfs_bmap.h"
39 #include "xfs_error.h"
40 #include "xfs_rw.h"
41 #include "xfs_quota.h"
42 #include "xfs_fsops.h"
43 #include "xfs_utils.h"
44 #include "xfs_trace.h"
47 STATIC void xfs_unmountfs_wait(xfs_mount_t *);
50 #ifdef HAVE_PERCPU_SB
51 STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
52 int);
53 STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
54 int);
55 STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
56 #else
58 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
59 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
60 #endif
62 static const struct {
63 short offset;
64 short type; /* 0 = integer
65 * 1 = binary / string (no translation)
67 } xfs_sb_info[] = {
68 { offsetof(xfs_sb_t, sb_magicnum), 0 },
69 { offsetof(xfs_sb_t, sb_blocksize), 0 },
70 { offsetof(xfs_sb_t, sb_dblocks), 0 },
71 { offsetof(xfs_sb_t, sb_rblocks), 0 },
72 { offsetof(xfs_sb_t, sb_rextents), 0 },
73 { offsetof(xfs_sb_t, sb_uuid), 1 },
74 { offsetof(xfs_sb_t, sb_logstart), 0 },
75 { offsetof(xfs_sb_t, sb_rootino), 0 },
76 { offsetof(xfs_sb_t, sb_rbmino), 0 },
77 { offsetof(xfs_sb_t, sb_rsumino), 0 },
78 { offsetof(xfs_sb_t, sb_rextsize), 0 },
79 { offsetof(xfs_sb_t, sb_agblocks), 0 },
80 { offsetof(xfs_sb_t, sb_agcount), 0 },
81 { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
82 { offsetof(xfs_sb_t, sb_logblocks), 0 },
83 { offsetof(xfs_sb_t, sb_versionnum), 0 },
84 { offsetof(xfs_sb_t, sb_sectsize), 0 },
85 { offsetof(xfs_sb_t, sb_inodesize), 0 },
86 { offsetof(xfs_sb_t, sb_inopblock), 0 },
87 { offsetof(xfs_sb_t, sb_fname[0]), 1 },
88 { offsetof(xfs_sb_t, sb_blocklog), 0 },
89 { offsetof(xfs_sb_t, sb_sectlog), 0 },
90 { offsetof(xfs_sb_t, sb_inodelog), 0 },
91 { offsetof(xfs_sb_t, sb_inopblog), 0 },
92 { offsetof(xfs_sb_t, sb_agblklog), 0 },
93 { offsetof(xfs_sb_t, sb_rextslog), 0 },
94 { offsetof(xfs_sb_t, sb_inprogress), 0 },
95 { offsetof(xfs_sb_t, sb_imax_pct), 0 },
96 { offsetof(xfs_sb_t, sb_icount), 0 },
97 { offsetof(xfs_sb_t, sb_ifree), 0 },
98 { offsetof(xfs_sb_t, sb_fdblocks), 0 },
99 { offsetof(xfs_sb_t, sb_frextents), 0 },
100 { offsetof(xfs_sb_t, sb_uquotino), 0 },
101 { offsetof(xfs_sb_t, sb_gquotino), 0 },
102 { offsetof(xfs_sb_t, sb_qflags), 0 },
103 { offsetof(xfs_sb_t, sb_flags), 0 },
104 { offsetof(xfs_sb_t, sb_shared_vn), 0 },
105 { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
106 { offsetof(xfs_sb_t, sb_unit), 0 },
107 { offsetof(xfs_sb_t, sb_width), 0 },
108 { offsetof(xfs_sb_t, sb_dirblklog), 0 },
109 { offsetof(xfs_sb_t, sb_logsectlog), 0 },
110 { offsetof(xfs_sb_t, sb_logsectsize),0 },
111 { offsetof(xfs_sb_t, sb_logsunit), 0 },
112 { offsetof(xfs_sb_t, sb_features2), 0 },
113 { offsetof(xfs_sb_t, sb_bad_features2), 0 },
114 { sizeof(xfs_sb_t), 0 }
117 static DEFINE_MUTEX(xfs_uuid_table_mutex);
118 static int xfs_uuid_table_size;
119 static uuid_t *xfs_uuid_table;
122 * See if the UUID is unique among mounted XFS filesystems.
123 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
125 STATIC int
126 xfs_uuid_mount(
127 struct xfs_mount *mp)
129 uuid_t *uuid = &mp->m_sb.sb_uuid;
130 int hole, i;
132 if (mp->m_flags & XFS_MOUNT_NOUUID)
133 return 0;
135 if (uuid_is_nil(uuid)) {
136 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
137 return XFS_ERROR(EINVAL);
140 mutex_lock(&xfs_uuid_table_mutex);
141 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
142 if (uuid_is_nil(&xfs_uuid_table[i])) {
143 hole = i;
144 continue;
146 if (uuid_equal(uuid, &xfs_uuid_table[i]))
147 goto out_duplicate;
150 if (hole < 0) {
151 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
152 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
153 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
154 KM_SLEEP);
155 hole = xfs_uuid_table_size++;
157 xfs_uuid_table[hole] = *uuid;
158 mutex_unlock(&xfs_uuid_table_mutex);
160 return 0;
162 out_duplicate:
163 mutex_unlock(&xfs_uuid_table_mutex);
164 xfs_warn(mp, "Filesystem has duplicate UUID - can't mount");
165 return XFS_ERROR(EINVAL);
168 STATIC void
169 xfs_uuid_unmount(
170 struct xfs_mount *mp)
172 uuid_t *uuid = &mp->m_sb.sb_uuid;
173 int i;
175 if (mp->m_flags & XFS_MOUNT_NOUUID)
176 return;
178 mutex_lock(&xfs_uuid_table_mutex);
179 for (i = 0; i < xfs_uuid_table_size; i++) {
180 if (uuid_is_nil(&xfs_uuid_table[i]))
181 continue;
182 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
183 continue;
184 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
185 break;
187 ASSERT(i < xfs_uuid_table_size);
188 mutex_unlock(&xfs_uuid_table_mutex);
193 * Reference counting access wrappers to the perag structures.
194 * Because we never free per-ag structures, the only thing we
195 * have to protect against changes is the tree structure itself.
197 struct xfs_perag *
198 xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
200 struct xfs_perag *pag;
201 int ref = 0;
203 rcu_read_lock();
204 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
205 if (pag) {
206 ASSERT(atomic_read(&pag->pag_ref) >= 0);
207 ref = atomic_inc_return(&pag->pag_ref);
209 rcu_read_unlock();
210 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
211 return pag;
215 * search from @first to find the next perag with the given tag set.
217 struct xfs_perag *
218 xfs_perag_get_tag(
219 struct xfs_mount *mp,
220 xfs_agnumber_t first,
221 int tag)
223 struct xfs_perag *pag;
224 int found;
225 int ref;
227 rcu_read_lock();
228 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
229 (void **)&pag, first, 1, tag);
230 if (found <= 0) {
231 rcu_read_unlock();
232 return NULL;
234 ref = atomic_inc_return(&pag->pag_ref);
235 rcu_read_unlock();
236 trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
237 return pag;
240 void
241 xfs_perag_put(struct xfs_perag *pag)
243 int ref;
245 ASSERT(atomic_read(&pag->pag_ref) > 0);
246 ref = atomic_dec_return(&pag->pag_ref);
247 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
250 STATIC void
251 __xfs_free_perag(
252 struct rcu_head *head)
254 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
256 ASSERT(atomic_read(&pag->pag_ref) == 0);
257 kmem_free(pag);
261 * Free up the per-ag resources associated with the mount structure.
263 STATIC void
264 xfs_free_perag(
265 xfs_mount_t *mp)
267 xfs_agnumber_t agno;
268 struct xfs_perag *pag;
270 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
271 spin_lock(&mp->m_perag_lock);
272 pag = radix_tree_delete(&mp->m_perag_tree, agno);
273 spin_unlock(&mp->m_perag_lock);
274 ASSERT(pag);
275 ASSERT(atomic_read(&pag->pag_ref) == 0);
276 call_rcu(&pag->rcu_head, __xfs_free_perag);
281 * Check size of device based on the (data/realtime) block count.
282 * Note: this check is used by the growfs code as well as mount.
285 xfs_sb_validate_fsb_count(
286 xfs_sb_t *sbp,
287 __uint64_t nblocks)
289 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
290 ASSERT(sbp->sb_blocklog >= BBSHIFT);
292 #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
293 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
294 return EFBIG;
295 #else /* Limited by UINT_MAX of sectors */
296 if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
297 return EFBIG;
298 #endif
299 return 0;
303 * Check the validity of the SB found.
305 STATIC int
306 xfs_mount_validate_sb(
307 xfs_mount_t *mp,
308 xfs_sb_t *sbp,
309 int flags)
311 int loud = !(flags & XFS_MFSI_QUIET);
314 * If the log device and data device have the
315 * same device number, the log is internal.
316 * Consequently, the sb_logstart should be non-zero. If
317 * we have a zero sb_logstart in this case, we may be trying to mount
318 * a volume filesystem in a non-volume manner.
320 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
321 if (loud)
322 xfs_warn(mp, "bad magic number");
323 return XFS_ERROR(EWRONGFS);
326 if (!xfs_sb_good_version(sbp)) {
327 if (loud)
328 xfs_warn(mp, "bad version");
329 return XFS_ERROR(EWRONGFS);
332 if (unlikely(
333 sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
334 if (loud)
335 xfs_warn(mp,
336 "filesystem is marked as having an external log; "
337 "specify logdev on the mount command line.");
338 return XFS_ERROR(EINVAL);
341 if (unlikely(
342 sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
343 if (loud)
344 xfs_warn(mp,
345 "filesystem is marked as having an internal log; "
346 "do not specify logdev on the mount command line.");
347 return XFS_ERROR(EINVAL);
351 * More sanity checking. Most of these were stolen directly from
352 * xfs_repair.
354 if (unlikely(
355 sbp->sb_agcount <= 0 ||
356 sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
357 sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
358 sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
359 sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
360 sbp->sb_sectsize != (1 << sbp->sb_sectlog) ||
361 sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
362 sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
363 sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
364 sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
365 sbp->sb_blocksize != (1 << sbp->sb_blocklog) ||
366 sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
367 sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
368 sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
369 sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
370 sbp->sb_inodesize != (1 << sbp->sb_inodelog) ||
371 (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
372 (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
373 (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
374 (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) ||
375 sbp->sb_dblocks == 0 ||
376 sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) ||
377 sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
378 if (loud)
379 XFS_CORRUPTION_ERROR("SB sanity check failed",
380 XFS_ERRLEVEL_LOW, mp, sbp);
381 return XFS_ERROR(EFSCORRUPTED);
385 * Until this is fixed only page-sized or smaller data blocks work.
387 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
388 if (loud) {
389 xfs_warn(mp,
390 "File system with blocksize %d bytes. "
391 "Only pagesize (%ld) or less will currently work.",
392 sbp->sb_blocksize, PAGE_SIZE);
394 return XFS_ERROR(ENOSYS);
398 * Currently only very few inode sizes are supported.
400 switch (sbp->sb_inodesize) {
401 case 256:
402 case 512:
403 case 1024:
404 case 2048:
405 break;
406 default:
407 if (loud)
408 xfs_warn(mp, "inode size of %d bytes not supported",
409 sbp->sb_inodesize);
410 return XFS_ERROR(ENOSYS);
413 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
414 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
415 if (loud)
416 xfs_warn(mp,
417 "file system too large to be mounted on this system.");
418 return XFS_ERROR(EFBIG);
421 if (unlikely(sbp->sb_inprogress)) {
422 if (loud)
423 xfs_warn(mp, "file system busy");
424 return XFS_ERROR(EFSCORRUPTED);
428 * Version 1 directory format has never worked on Linux.
430 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
431 if (loud)
432 xfs_warn(mp,
433 "file system using version 1 directory format");
434 return XFS_ERROR(ENOSYS);
437 return 0;
441 xfs_initialize_perag(
442 xfs_mount_t *mp,
443 xfs_agnumber_t agcount,
444 xfs_agnumber_t *maxagi)
446 xfs_agnumber_t index, max_metadata;
447 xfs_agnumber_t first_initialised = 0;
448 xfs_perag_t *pag;
449 xfs_agino_t agino;
450 xfs_ino_t ino;
451 xfs_sb_t *sbp = &mp->m_sb;
452 int error = -ENOMEM;
455 * Walk the current per-ag tree so we don't try to initialise AGs
456 * that already exist (growfs case). Allocate and insert all the
457 * AGs we don't find ready for initialisation.
459 for (index = 0; index < agcount; index++) {
460 pag = xfs_perag_get(mp, index);
461 if (pag) {
462 xfs_perag_put(pag);
463 continue;
465 if (!first_initialised)
466 first_initialised = index;
468 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
469 if (!pag)
470 goto out_unwind;
471 pag->pag_agno = index;
472 pag->pag_mount = mp;
473 spin_lock_init(&pag->pag_ici_lock);
474 mutex_init(&pag->pag_ici_reclaim_lock);
475 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
476 spin_lock_init(&pag->pag_buf_lock);
477 pag->pag_buf_tree = RB_ROOT;
479 if (radix_tree_preload(GFP_NOFS))
480 goto out_unwind;
482 spin_lock(&mp->m_perag_lock);
483 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
484 BUG();
485 spin_unlock(&mp->m_perag_lock);
486 radix_tree_preload_end();
487 error = -EEXIST;
488 goto out_unwind;
490 spin_unlock(&mp->m_perag_lock);
491 radix_tree_preload_end();
495 * If we mount with the inode64 option, or no inode overflows
496 * the legacy 32-bit address space clear the inode32 option.
498 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
499 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
501 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
502 mp->m_flags |= XFS_MOUNT_32BITINODES;
503 else
504 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
506 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
508 * Calculate how much should be reserved for inodes to meet
509 * the max inode percentage.
511 if (mp->m_maxicount) {
512 __uint64_t icount;
514 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
515 do_div(icount, 100);
516 icount += sbp->sb_agblocks - 1;
517 do_div(icount, sbp->sb_agblocks);
518 max_metadata = icount;
519 } else {
520 max_metadata = agcount;
523 for (index = 0; index < agcount; index++) {
524 ino = XFS_AGINO_TO_INO(mp, index, agino);
525 if (ino > XFS_MAXINUMBER_32) {
526 index++;
527 break;
530 pag = xfs_perag_get(mp, index);
531 pag->pagi_inodeok = 1;
532 if (index < max_metadata)
533 pag->pagf_metadata = 1;
534 xfs_perag_put(pag);
536 } else {
537 for (index = 0; index < agcount; index++) {
538 pag = xfs_perag_get(mp, index);
539 pag->pagi_inodeok = 1;
540 xfs_perag_put(pag);
544 if (maxagi)
545 *maxagi = index;
546 return 0;
548 out_unwind:
549 kmem_free(pag);
550 for (; index > first_initialised; index--) {
551 pag = radix_tree_delete(&mp->m_perag_tree, index);
552 kmem_free(pag);
554 return error;
557 void
558 xfs_sb_from_disk(
559 xfs_sb_t *to,
560 xfs_dsb_t *from)
562 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
563 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
564 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
565 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
566 to->sb_rextents = be64_to_cpu(from->sb_rextents);
567 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
568 to->sb_logstart = be64_to_cpu(from->sb_logstart);
569 to->sb_rootino = be64_to_cpu(from->sb_rootino);
570 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
571 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
572 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
573 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
574 to->sb_agcount = be32_to_cpu(from->sb_agcount);
575 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
576 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
577 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
578 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
579 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
580 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
581 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
582 to->sb_blocklog = from->sb_blocklog;
583 to->sb_sectlog = from->sb_sectlog;
584 to->sb_inodelog = from->sb_inodelog;
585 to->sb_inopblog = from->sb_inopblog;
586 to->sb_agblklog = from->sb_agblklog;
587 to->sb_rextslog = from->sb_rextslog;
588 to->sb_inprogress = from->sb_inprogress;
589 to->sb_imax_pct = from->sb_imax_pct;
590 to->sb_icount = be64_to_cpu(from->sb_icount);
591 to->sb_ifree = be64_to_cpu(from->sb_ifree);
592 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
593 to->sb_frextents = be64_to_cpu(from->sb_frextents);
594 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
595 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
596 to->sb_qflags = be16_to_cpu(from->sb_qflags);
597 to->sb_flags = from->sb_flags;
598 to->sb_shared_vn = from->sb_shared_vn;
599 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
600 to->sb_unit = be32_to_cpu(from->sb_unit);
601 to->sb_width = be32_to_cpu(from->sb_width);
602 to->sb_dirblklog = from->sb_dirblklog;
603 to->sb_logsectlog = from->sb_logsectlog;
604 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
605 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
606 to->sb_features2 = be32_to_cpu(from->sb_features2);
607 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
611 * Copy in core superblock to ondisk one.
613 * The fields argument is mask of superblock fields to copy.
615 void
616 xfs_sb_to_disk(
617 xfs_dsb_t *to,
618 xfs_sb_t *from,
619 __int64_t fields)
621 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
622 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
623 xfs_sb_field_t f;
624 int first;
625 int size;
627 ASSERT(fields);
628 if (!fields)
629 return;
631 while (fields) {
632 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
633 first = xfs_sb_info[f].offset;
634 size = xfs_sb_info[f + 1].offset - first;
636 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
638 if (size == 1 || xfs_sb_info[f].type == 1) {
639 memcpy(to_ptr + first, from_ptr + first, size);
640 } else {
641 switch (size) {
642 case 2:
643 *(__be16 *)(to_ptr + first) =
644 cpu_to_be16(*(__u16 *)(from_ptr + first));
645 break;
646 case 4:
647 *(__be32 *)(to_ptr + first) =
648 cpu_to_be32(*(__u32 *)(from_ptr + first));
649 break;
650 case 8:
651 *(__be64 *)(to_ptr + first) =
652 cpu_to_be64(*(__u64 *)(from_ptr + first));
653 break;
654 default:
655 ASSERT(0);
659 fields &= ~(1LL << f);
664 * xfs_readsb
666 * Does the initial read of the superblock.
669 xfs_readsb(xfs_mount_t *mp, int flags)
671 unsigned int sector_size;
672 xfs_buf_t *bp;
673 int error;
674 int loud = !(flags & XFS_MFSI_QUIET);
676 ASSERT(mp->m_sb_bp == NULL);
677 ASSERT(mp->m_ddev_targp != NULL);
680 * Allocate a (locked) buffer to hold the superblock.
681 * This will be kept around at all times to optimize
682 * access to the superblock.
684 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
686 reread:
687 bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
688 XFS_SB_DADDR, sector_size, 0);
689 if (!bp) {
690 if (loud)
691 xfs_warn(mp, "SB buffer read failed");
692 return EIO;
696 * Initialize the mount structure from the superblock.
697 * But first do some basic consistency checking.
699 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
700 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
701 if (error) {
702 if (loud)
703 xfs_warn(mp, "SB validate failed");
704 goto release_buf;
708 * We must be able to do sector-sized and sector-aligned IO.
710 if (sector_size > mp->m_sb.sb_sectsize) {
711 if (loud)
712 xfs_warn(mp, "device supports %u byte sectors (not %u)",
713 sector_size, mp->m_sb.sb_sectsize);
714 error = ENOSYS;
715 goto release_buf;
719 * If device sector size is smaller than the superblock size,
720 * re-read the superblock so the buffer is correctly sized.
722 if (sector_size < mp->m_sb.sb_sectsize) {
723 xfs_buf_relse(bp);
724 sector_size = mp->m_sb.sb_sectsize;
725 goto reread;
728 /* Initialize per-cpu counters */
729 xfs_icsb_reinit_counters(mp);
731 mp->m_sb_bp = bp;
732 xfs_buf_unlock(bp);
733 return 0;
735 release_buf:
736 xfs_buf_relse(bp);
737 return error;
742 * xfs_mount_common
744 * Mount initialization code establishing various mount
745 * fields from the superblock associated with the given
746 * mount structure
748 STATIC void
749 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
751 mp->m_agfrotor = mp->m_agirotor = 0;
752 spin_lock_init(&mp->m_agirotor_lock);
753 mp->m_maxagi = mp->m_sb.sb_agcount;
754 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
755 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
756 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
757 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
758 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
759 mp->m_blockmask = sbp->sb_blocksize - 1;
760 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
761 mp->m_blockwmask = mp->m_blockwsize - 1;
763 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
764 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
765 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
766 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
768 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
769 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
770 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
771 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
773 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
774 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
775 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
776 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
778 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
779 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
780 sbp->sb_inopblock);
781 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
785 * xfs_initialize_perag_data
787 * Read in each per-ag structure so we can count up the number of
788 * allocated inodes, free inodes and used filesystem blocks as this
789 * information is no longer persistent in the superblock. Once we have
790 * this information, write it into the in-core superblock structure.
792 STATIC int
793 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
795 xfs_agnumber_t index;
796 xfs_perag_t *pag;
797 xfs_sb_t *sbp = &mp->m_sb;
798 uint64_t ifree = 0;
799 uint64_t ialloc = 0;
800 uint64_t bfree = 0;
801 uint64_t bfreelst = 0;
802 uint64_t btree = 0;
803 int error;
805 for (index = 0; index < agcount; index++) {
807 * read the agf, then the agi. This gets us
808 * all the information we need and populates the
809 * per-ag structures for us.
811 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
812 if (error)
813 return error;
815 error = xfs_ialloc_pagi_init(mp, NULL, index);
816 if (error)
817 return error;
818 pag = xfs_perag_get(mp, index);
819 ifree += pag->pagi_freecount;
820 ialloc += pag->pagi_count;
821 bfree += pag->pagf_freeblks;
822 bfreelst += pag->pagf_flcount;
823 btree += pag->pagf_btreeblks;
824 xfs_perag_put(pag);
827 * Overwrite incore superblock counters with just-read data
829 spin_lock(&mp->m_sb_lock);
830 sbp->sb_ifree = ifree;
831 sbp->sb_icount = ialloc;
832 sbp->sb_fdblocks = bfree + bfreelst + btree;
833 spin_unlock(&mp->m_sb_lock);
835 /* Fixup the per-cpu counters as well. */
836 xfs_icsb_reinit_counters(mp);
838 return 0;
842 * Update alignment values based on mount options and sb values
844 STATIC int
845 xfs_update_alignment(xfs_mount_t *mp)
847 xfs_sb_t *sbp = &(mp->m_sb);
849 if (mp->m_dalign) {
851 * If stripe unit and stripe width are not multiples
852 * of the fs blocksize turn off alignment.
854 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
855 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
856 if (mp->m_flags & XFS_MOUNT_RETERR) {
857 xfs_warn(mp, "alignment check failed: "
858 "(sunit/swidth vs. blocksize)");
859 return XFS_ERROR(EINVAL);
861 mp->m_dalign = mp->m_swidth = 0;
862 } else {
864 * Convert the stripe unit and width to FSBs.
866 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
867 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
868 if (mp->m_flags & XFS_MOUNT_RETERR) {
869 xfs_warn(mp, "alignment check failed: "
870 "(sunit/swidth vs. ag size)");
871 return XFS_ERROR(EINVAL);
873 xfs_warn(mp,
874 "stripe alignment turned off: sunit(%d)/swidth(%d) "
875 "incompatible with agsize(%d)",
876 mp->m_dalign, mp->m_swidth,
877 sbp->sb_agblocks);
879 mp->m_dalign = 0;
880 mp->m_swidth = 0;
881 } else if (mp->m_dalign) {
882 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
883 } else {
884 if (mp->m_flags & XFS_MOUNT_RETERR) {
885 xfs_warn(mp, "alignment check failed: "
886 "sunit(%d) less than bsize(%d)",
887 mp->m_dalign,
888 mp->m_blockmask +1);
889 return XFS_ERROR(EINVAL);
891 mp->m_swidth = 0;
896 * Update superblock with new values
897 * and log changes
899 if (xfs_sb_version_hasdalign(sbp)) {
900 if (sbp->sb_unit != mp->m_dalign) {
901 sbp->sb_unit = mp->m_dalign;
902 mp->m_update_flags |= XFS_SB_UNIT;
904 if (sbp->sb_width != mp->m_swidth) {
905 sbp->sb_width = mp->m_swidth;
906 mp->m_update_flags |= XFS_SB_WIDTH;
909 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
910 xfs_sb_version_hasdalign(&mp->m_sb)) {
911 mp->m_dalign = sbp->sb_unit;
912 mp->m_swidth = sbp->sb_width;
915 return 0;
919 * Set the maximum inode count for this filesystem
921 STATIC void
922 xfs_set_maxicount(xfs_mount_t *mp)
924 xfs_sb_t *sbp = &(mp->m_sb);
925 __uint64_t icount;
927 if (sbp->sb_imax_pct) {
929 * Make sure the maximum inode count is a multiple
930 * of the units we allocate inodes in.
932 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
933 do_div(icount, 100);
934 do_div(icount, mp->m_ialloc_blks);
935 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
936 sbp->sb_inopblog;
937 } else {
938 mp->m_maxicount = 0;
943 * Set the default minimum read and write sizes unless
944 * already specified in a mount option.
945 * We use smaller I/O sizes when the file system
946 * is being used for NFS service (wsync mount option).
948 STATIC void
949 xfs_set_rw_sizes(xfs_mount_t *mp)
951 xfs_sb_t *sbp = &(mp->m_sb);
952 int readio_log, writeio_log;
954 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
955 if (mp->m_flags & XFS_MOUNT_WSYNC) {
956 readio_log = XFS_WSYNC_READIO_LOG;
957 writeio_log = XFS_WSYNC_WRITEIO_LOG;
958 } else {
959 readio_log = XFS_READIO_LOG_LARGE;
960 writeio_log = XFS_WRITEIO_LOG_LARGE;
962 } else {
963 readio_log = mp->m_readio_log;
964 writeio_log = mp->m_writeio_log;
967 if (sbp->sb_blocklog > readio_log) {
968 mp->m_readio_log = sbp->sb_blocklog;
969 } else {
970 mp->m_readio_log = readio_log;
972 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
973 if (sbp->sb_blocklog > writeio_log) {
974 mp->m_writeio_log = sbp->sb_blocklog;
975 } else {
976 mp->m_writeio_log = writeio_log;
978 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
982 * precalculate the low space thresholds for dynamic speculative preallocation.
984 void
985 xfs_set_low_space_thresholds(
986 struct xfs_mount *mp)
988 int i;
990 for (i = 0; i < XFS_LOWSP_MAX; i++) {
991 __uint64_t space = mp->m_sb.sb_dblocks;
993 do_div(space, 100);
994 mp->m_low_space[i] = space * (i + 1);
1000 * Set whether we're using inode alignment.
1002 STATIC void
1003 xfs_set_inoalignment(xfs_mount_t *mp)
1005 if (xfs_sb_version_hasalign(&mp->m_sb) &&
1006 mp->m_sb.sb_inoalignmt >=
1007 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1008 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1009 else
1010 mp->m_inoalign_mask = 0;
1012 * If we are using stripe alignment, check whether
1013 * the stripe unit is a multiple of the inode alignment
1015 if (mp->m_dalign && mp->m_inoalign_mask &&
1016 !(mp->m_dalign & mp->m_inoalign_mask))
1017 mp->m_sinoalign = mp->m_dalign;
1018 else
1019 mp->m_sinoalign = 0;
1023 * Check that the data (and log if separate) are an ok size.
1025 STATIC int
1026 xfs_check_sizes(xfs_mount_t *mp)
1028 xfs_buf_t *bp;
1029 xfs_daddr_t d;
1031 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1032 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1033 xfs_warn(mp, "filesystem size mismatch detected");
1034 return XFS_ERROR(EFBIG);
1036 bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
1037 d - XFS_FSS_TO_BB(mp, 1),
1038 BBTOB(XFS_FSS_TO_BB(mp, 1)), 0);
1039 if (!bp) {
1040 xfs_warn(mp, "last sector read failed");
1041 return EIO;
1043 xfs_buf_relse(bp);
1045 if (mp->m_logdev_targp != mp->m_ddev_targp) {
1046 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1047 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1048 xfs_warn(mp, "log size mismatch detected");
1049 return XFS_ERROR(EFBIG);
1051 bp = xfs_buf_read_uncached(mp, mp->m_logdev_targp,
1052 d - XFS_FSB_TO_BB(mp, 1),
1053 XFS_FSB_TO_B(mp, 1), 0);
1054 if (!bp) {
1055 xfs_warn(mp, "log device read failed");
1056 return EIO;
1058 xfs_buf_relse(bp);
1060 return 0;
1064 * Clear the quotaflags in memory and in the superblock.
1067 xfs_mount_reset_sbqflags(
1068 struct xfs_mount *mp)
1070 int error;
1071 struct xfs_trans *tp;
1073 mp->m_qflags = 0;
1076 * It is OK to look at sb_qflags here in mount path,
1077 * without m_sb_lock.
1079 if (mp->m_sb.sb_qflags == 0)
1080 return 0;
1081 spin_lock(&mp->m_sb_lock);
1082 mp->m_sb.sb_qflags = 0;
1083 spin_unlock(&mp->m_sb_lock);
1086 * If the fs is readonly, let the incore superblock run
1087 * with quotas off but don't flush the update out to disk
1089 if (mp->m_flags & XFS_MOUNT_RDONLY)
1090 return 0;
1092 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1093 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1094 XFS_DEFAULT_LOG_COUNT);
1095 if (error) {
1096 xfs_trans_cancel(tp, 0);
1097 xfs_alert(mp, "%s: Superblock update failed!", __func__);
1098 return error;
1101 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1102 return xfs_trans_commit(tp, 0);
1105 __uint64_t
1106 xfs_default_resblks(xfs_mount_t *mp)
1108 __uint64_t resblks;
1111 * We default to 5% or 8192 fsbs of space reserved, whichever is
1112 * smaller. This is intended to cover concurrent allocation
1113 * transactions when we initially hit enospc. These each require a 4
1114 * block reservation. Hence by default we cover roughly 2000 concurrent
1115 * allocation reservations.
1117 resblks = mp->m_sb.sb_dblocks;
1118 do_div(resblks, 20);
1119 resblks = min_t(__uint64_t, resblks, 8192);
1120 return resblks;
1124 * This function does the following on an initial mount of a file system:
1125 * - reads the superblock from disk and init the mount struct
1126 * - if we're a 32-bit kernel, do a size check on the superblock
1127 * so we don't mount terabyte filesystems
1128 * - init mount struct realtime fields
1129 * - allocate inode hash table for fs
1130 * - init directory manager
1131 * - perform recovery and init the log manager
1134 xfs_mountfs(
1135 xfs_mount_t *mp)
1137 xfs_sb_t *sbp = &(mp->m_sb);
1138 xfs_inode_t *rip;
1139 __uint64_t resblks;
1140 uint quotamount = 0;
1141 uint quotaflags = 0;
1142 int error = 0;
1144 xfs_mount_common(mp, sbp);
1147 * Check for a mismatched features2 values. Older kernels
1148 * read & wrote into the wrong sb offset for sb_features2
1149 * on some platforms due to xfs_sb_t not being 64bit size aligned
1150 * when sb_features2 was added, which made older superblock
1151 * reading/writing routines swap it as a 64-bit value.
1153 * For backwards compatibility, we make both slots equal.
1155 * If we detect a mismatched field, we OR the set bits into the
1156 * existing features2 field in case it has already been modified; we
1157 * don't want to lose any features. We then update the bad location
1158 * with the ORed value so that older kernels will see any features2
1159 * flags, and mark the two fields as needing updates once the
1160 * transaction subsystem is online.
1162 if (xfs_sb_has_mismatched_features2(sbp)) {
1163 xfs_warn(mp, "correcting sb_features alignment problem");
1164 sbp->sb_features2 |= sbp->sb_bad_features2;
1165 sbp->sb_bad_features2 = sbp->sb_features2;
1166 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1169 * Re-check for ATTR2 in case it was found in bad_features2
1170 * slot.
1172 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1173 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1174 mp->m_flags |= XFS_MOUNT_ATTR2;
1177 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1178 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1179 xfs_sb_version_removeattr2(&mp->m_sb);
1180 mp->m_update_flags |= XFS_SB_FEATURES2;
1182 /* update sb_versionnum for the clearing of the morebits */
1183 if (!sbp->sb_features2)
1184 mp->m_update_flags |= XFS_SB_VERSIONNUM;
1188 * Check if sb_agblocks is aligned at stripe boundary
1189 * If sb_agblocks is NOT aligned turn off m_dalign since
1190 * allocator alignment is within an ag, therefore ag has
1191 * to be aligned at stripe boundary.
1193 error = xfs_update_alignment(mp);
1194 if (error)
1195 goto out;
1197 xfs_alloc_compute_maxlevels(mp);
1198 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1199 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1200 xfs_ialloc_compute_maxlevels(mp);
1202 xfs_set_maxicount(mp);
1204 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1206 error = xfs_uuid_mount(mp);
1207 if (error)
1208 goto out;
1211 * Set the minimum read and write sizes
1213 xfs_set_rw_sizes(mp);
1215 /* set the low space thresholds for dynamic preallocation */
1216 xfs_set_low_space_thresholds(mp);
1219 * Set the inode cluster size.
1220 * This may still be overridden by the file system
1221 * block size if it is larger than the chosen cluster size.
1223 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1226 * Set inode alignment fields
1228 xfs_set_inoalignment(mp);
1231 * Check that the data (and log if separate) are an ok size.
1233 error = xfs_check_sizes(mp);
1234 if (error)
1235 goto out_remove_uuid;
1238 * Initialize realtime fields in the mount structure
1240 error = xfs_rtmount_init(mp);
1241 if (error) {
1242 xfs_warn(mp, "RT mount failed");
1243 goto out_remove_uuid;
1247 * Copies the low order bits of the timestamp and the randomly
1248 * set "sequence" number out of a UUID.
1250 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1252 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1254 xfs_dir_mount(mp);
1257 * Initialize the attribute manager's entries.
1259 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1262 * Initialize the precomputed transaction reservations values.
1264 xfs_trans_init(mp);
1267 * Allocate and initialize the per-ag data.
1269 spin_lock_init(&mp->m_perag_lock);
1270 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1271 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1272 if (error) {
1273 xfs_warn(mp, "Failed per-ag init: %d", error);
1274 goto out_remove_uuid;
1277 if (!sbp->sb_logblocks) {
1278 xfs_warn(mp, "no log defined");
1279 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1280 error = XFS_ERROR(EFSCORRUPTED);
1281 goto out_free_perag;
1285 * log's mount-time initialization. Perform 1st part recovery if needed
1287 error = xfs_log_mount(mp, mp->m_logdev_targp,
1288 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1289 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1290 if (error) {
1291 xfs_warn(mp, "log mount failed");
1292 goto out_free_perag;
1296 * Now the log is mounted, we know if it was an unclean shutdown or
1297 * not. If it was, with the first phase of recovery has completed, we
1298 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1299 * but they are recovered transactionally in the second recovery phase
1300 * later.
1302 * Hence we can safely re-initialise incore superblock counters from
1303 * the per-ag data. These may not be correct if the filesystem was not
1304 * cleanly unmounted, so we need to wait for recovery to finish before
1305 * doing this.
1307 * If the filesystem was cleanly unmounted, then we can trust the
1308 * values in the superblock to be correct and we don't need to do
1309 * anything here.
1311 * If we are currently making the filesystem, the initialisation will
1312 * fail as the perag data is in an undefined state.
1314 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1315 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1316 !mp->m_sb.sb_inprogress) {
1317 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1318 if (error)
1319 goto out_free_perag;
1323 * Get and sanity-check the root inode.
1324 * Save the pointer to it in the mount structure.
1326 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1327 if (error) {
1328 xfs_warn(mp, "failed to read root inode");
1329 goto out_log_dealloc;
1332 ASSERT(rip != NULL);
1334 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
1335 xfs_warn(mp, "corrupted root inode %llu: not a directory",
1336 (unsigned long long)rip->i_ino);
1337 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1338 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1339 mp);
1340 error = XFS_ERROR(EFSCORRUPTED);
1341 goto out_rele_rip;
1343 mp->m_rootip = rip; /* save it */
1345 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1348 * Initialize realtime inode pointers in the mount structure
1350 error = xfs_rtmount_inodes(mp);
1351 if (error) {
1353 * Free up the root inode.
1355 xfs_warn(mp, "failed to read RT inodes");
1356 goto out_rele_rip;
1360 * If this is a read-only mount defer the superblock updates until
1361 * the next remount into writeable mode. Otherwise we would never
1362 * perform the update e.g. for the root filesystem.
1364 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1365 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1366 if (error) {
1367 xfs_warn(mp, "failed to write sb changes");
1368 goto out_rtunmount;
1373 * Initialise the XFS quota management subsystem for this mount
1375 if (XFS_IS_QUOTA_RUNNING(mp)) {
1376 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1377 if (error)
1378 goto out_rtunmount;
1379 } else {
1380 ASSERT(!XFS_IS_QUOTA_ON(mp));
1383 * If a file system had quotas running earlier, but decided to
1384 * mount without -o uquota/pquota/gquota options, revoke the
1385 * quotachecked license.
1387 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1388 xfs_notice(mp, "resetting quota flags");
1389 error = xfs_mount_reset_sbqflags(mp);
1390 if (error)
1391 return error;
1396 * Finish recovering the file system. This part needed to be
1397 * delayed until after the root and real-time bitmap inodes
1398 * were consistently read in.
1400 error = xfs_log_mount_finish(mp);
1401 if (error) {
1402 xfs_warn(mp, "log mount finish failed");
1403 goto out_rtunmount;
1407 * Complete the quota initialisation, post-log-replay component.
1409 if (quotamount) {
1410 ASSERT(mp->m_qflags == 0);
1411 mp->m_qflags = quotaflags;
1413 xfs_qm_mount_quotas(mp);
1417 * Now we are mounted, reserve a small amount of unused space for
1418 * privileged transactions. This is needed so that transaction
1419 * space required for critical operations can dip into this pool
1420 * when at ENOSPC. This is needed for operations like create with
1421 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1422 * are not allowed to use this reserved space.
1424 * This may drive us straight to ENOSPC on mount, but that implies
1425 * we were already there on the last unmount. Warn if this occurs.
1427 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1428 resblks = xfs_default_resblks(mp);
1429 error = xfs_reserve_blocks(mp, &resblks, NULL);
1430 if (error)
1431 xfs_warn(mp,
1432 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1435 return 0;
1437 out_rtunmount:
1438 xfs_rtunmount_inodes(mp);
1439 out_rele_rip:
1440 IRELE(rip);
1441 out_log_dealloc:
1442 xfs_log_unmount(mp);
1443 out_free_perag:
1444 xfs_free_perag(mp);
1445 out_remove_uuid:
1446 xfs_uuid_unmount(mp);
1447 out:
1448 return error;
1452 * This flushes out the inodes,dquots and the superblock, unmounts the
1453 * log and makes sure that incore structures are freed.
1455 void
1456 xfs_unmountfs(
1457 struct xfs_mount *mp)
1459 __uint64_t resblks;
1460 int error;
1462 xfs_qm_unmount_quotas(mp);
1463 xfs_rtunmount_inodes(mp);
1464 IRELE(mp->m_rootip);
1467 * We can potentially deadlock here if we have an inode cluster
1468 * that has been freed has its buffer still pinned in memory because
1469 * the transaction is still sitting in a iclog. The stale inodes
1470 * on that buffer will have their flush locks held until the
1471 * transaction hits the disk and the callbacks run. the inode
1472 * flush takes the flush lock unconditionally and with nothing to
1473 * push out the iclog we will never get that unlocked. hence we
1474 * need to force the log first.
1476 xfs_log_force(mp, XFS_LOG_SYNC);
1479 * Do a delwri reclaim pass first so that as many dirty inodes are
1480 * queued up for IO as possible. Then flush the buffers before making
1481 * a synchronous path to catch all the remaining inodes are reclaimed.
1482 * This makes the reclaim process as quick as possible by avoiding
1483 * synchronous writeout and blocking on inodes already in the delwri
1484 * state as much as possible.
1486 xfs_reclaim_inodes(mp, 0);
1487 XFS_bflush(mp->m_ddev_targp);
1488 xfs_reclaim_inodes(mp, SYNC_WAIT);
1490 xfs_qm_unmount(mp);
1493 * Flush out the log synchronously so that we know for sure
1494 * that nothing is pinned. This is important because bflush()
1495 * will skip pinned buffers.
1497 xfs_log_force(mp, XFS_LOG_SYNC);
1499 xfs_binval(mp->m_ddev_targp);
1500 if (mp->m_rtdev_targp) {
1501 xfs_binval(mp->m_rtdev_targp);
1505 * Unreserve any blocks we have so that when we unmount we don't account
1506 * the reserved free space as used. This is really only necessary for
1507 * lazy superblock counting because it trusts the incore superblock
1508 * counters to be absolutely correct on clean unmount.
1510 * We don't bother correcting this elsewhere for lazy superblock
1511 * counting because on mount of an unclean filesystem we reconstruct the
1512 * correct counter value and this is irrelevant.
1514 * For non-lazy counter filesystems, this doesn't matter at all because
1515 * we only every apply deltas to the superblock and hence the incore
1516 * value does not matter....
1518 resblks = 0;
1519 error = xfs_reserve_blocks(mp, &resblks, NULL);
1520 if (error)
1521 xfs_warn(mp, "Unable to free reserved block pool. "
1522 "Freespace may not be correct on next mount.");
1524 error = xfs_log_sbcount(mp);
1525 if (error)
1526 xfs_warn(mp, "Unable to update superblock counters. "
1527 "Freespace may not be correct on next mount.");
1528 xfs_unmountfs_writesb(mp);
1529 xfs_unmountfs_wait(mp); /* wait for async bufs */
1530 xfs_log_unmount_write(mp);
1531 xfs_log_unmount(mp);
1532 xfs_uuid_unmount(mp);
1534 #if defined(DEBUG)
1535 xfs_errortag_clearall(mp, 0);
1536 #endif
1537 xfs_free_perag(mp);
1540 STATIC void
1541 xfs_unmountfs_wait(xfs_mount_t *mp)
1543 if (mp->m_logdev_targp != mp->m_ddev_targp)
1544 xfs_wait_buftarg(mp->m_logdev_targp);
1545 if (mp->m_rtdev_targp)
1546 xfs_wait_buftarg(mp->m_rtdev_targp);
1547 xfs_wait_buftarg(mp->m_ddev_targp);
1551 xfs_fs_writable(xfs_mount_t *mp)
1553 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1554 (mp->m_flags & XFS_MOUNT_RDONLY));
1558 * xfs_log_sbcount
1560 * Sync the superblock counters to disk.
1562 * Note this code can be called during the process of freezing, so
1563 * we may need to use the transaction allocator which does not
1564 * block when the transaction subsystem is in its frozen state.
1567 xfs_log_sbcount(xfs_mount_t *mp)
1569 xfs_trans_t *tp;
1570 int error;
1572 if (!xfs_fs_writable(mp))
1573 return 0;
1575 xfs_icsb_sync_counters(mp, 0);
1578 * we don't need to do this if we are updating the superblock
1579 * counters on every modification.
1581 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1582 return 0;
1584 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1585 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1586 XFS_DEFAULT_LOG_COUNT);
1587 if (error) {
1588 xfs_trans_cancel(tp, 0);
1589 return error;
1592 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1593 xfs_trans_set_sync(tp);
1594 error = xfs_trans_commit(tp, 0);
1595 return error;
1599 xfs_unmountfs_writesb(xfs_mount_t *mp)
1601 xfs_buf_t *sbp;
1602 int error = 0;
1605 * skip superblock write if fs is read-only, or
1606 * if we are doing a forced umount.
1608 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1609 XFS_FORCED_SHUTDOWN(mp))) {
1611 sbp = xfs_getsb(mp, 0);
1613 XFS_BUF_UNDONE(sbp);
1614 XFS_BUF_UNREAD(sbp);
1615 XFS_BUF_UNDELAYWRITE(sbp);
1616 XFS_BUF_WRITE(sbp);
1617 XFS_BUF_UNASYNC(sbp);
1618 ASSERT(sbp->b_target == mp->m_ddev_targp);
1619 xfsbdstrat(mp, sbp);
1620 error = xfs_buf_iowait(sbp);
1621 if (error)
1622 xfs_ioerror_alert("xfs_unmountfs_writesb",
1623 mp, sbp, XFS_BUF_ADDR(sbp));
1624 xfs_buf_relse(sbp);
1626 return error;
1630 * xfs_mod_sb() can be used to copy arbitrary changes to the
1631 * in-core superblock into the superblock buffer to be logged.
1632 * It does not provide the higher level of locking that is
1633 * needed to protect the in-core superblock from concurrent
1634 * access.
1636 void
1637 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1639 xfs_buf_t *bp;
1640 int first;
1641 int last;
1642 xfs_mount_t *mp;
1643 xfs_sb_field_t f;
1645 ASSERT(fields);
1646 if (!fields)
1647 return;
1648 mp = tp->t_mountp;
1649 bp = xfs_trans_getsb(tp, mp, 0);
1650 first = sizeof(xfs_sb_t);
1651 last = 0;
1653 /* translate/copy */
1655 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1657 /* find modified range */
1658 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1659 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1660 last = xfs_sb_info[f + 1].offset - 1;
1662 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1663 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1664 first = xfs_sb_info[f].offset;
1666 xfs_trans_log_buf(tp, bp, first, last);
1671 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1672 * a delta to a specified field in the in-core superblock. Simply
1673 * switch on the field indicated and apply the delta to that field.
1674 * Fields are not allowed to dip below zero, so if the delta would
1675 * do this do not apply it and return EINVAL.
1677 * The m_sb_lock must be held when this routine is called.
1679 STATIC int
1680 xfs_mod_incore_sb_unlocked(
1681 xfs_mount_t *mp,
1682 xfs_sb_field_t field,
1683 int64_t delta,
1684 int rsvd)
1686 int scounter; /* short counter for 32 bit fields */
1687 long long lcounter; /* long counter for 64 bit fields */
1688 long long res_used, rem;
1691 * With the in-core superblock spin lock held, switch
1692 * on the indicated field. Apply the delta to the
1693 * proper field. If the fields value would dip below
1694 * 0, then do not apply the delta and return EINVAL.
1696 switch (field) {
1697 case XFS_SBS_ICOUNT:
1698 lcounter = (long long)mp->m_sb.sb_icount;
1699 lcounter += delta;
1700 if (lcounter < 0) {
1701 ASSERT(0);
1702 return XFS_ERROR(EINVAL);
1704 mp->m_sb.sb_icount = lcounter;
1705 return 0;
1706 case XFS_SBS_IFREE:
1707 lcounter = (long long)mp->m_sb.sb_ifree;
1708 lcounter += delta;
1709 if (lcounter < 0) {
1710 ASSERT(0);
1711 return XFS_ERROR(EINVAL);
1713 mp->m_sb.sb_ifree = lcounter;
1714 return 0;
1715 case XFS_SBS_FDBLOCKS:
1716 lcounter = (long long)
1717 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1718 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1720 if (delta > 0) { /* Putting blocks back */
1721 if (res_used > delta) {
1722 mp->m_resblks_avail += delta;
1723 } else {
1724 rem = delta - res_used;
1725 mp->m_resblks_avail = mp->m_resblks;
1726 lcounter += rem;
1728 } else { /* Taking blocks away */
1729 lcounter += delta;
1730 if (lcounter >= 0) {
1731 mp->m_sb.sb_fdblocks = lcounter +
1732 XFS_ALLOC_SET_ASIDE(mp);
1733 return 0;
1737 * We are out of blocks, use any available reserved
1738 * blocks if were allowed to.
1740 if (!rsvd)
1741 return XFS_ERROR(ENOSPC);
1743 lcounter = (long long)mp->m_resblks_avail + delta;
1744 if (lcounter >= 0) {
1745 mp->m_resblks_avail = lcounter;
1746 return 0;
1748 printk_once(KERN_WARNING
1749 "Filesystem \"%s\": reserve blocks depleted! "
1750 "Consider increasing reserve pool size.",
1751 mp->m_fsname);
1752 return XFS_ERROR(ENOSPC);
1755 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1756 return 0;
1757 case XFS_SBS_FREXTENTS:
1758 lcounter = (long long)mp->m_sb.sb_frextents;
1759 lcounter += delta;
1760 if (lcounter < 0) {
1761 return XFS_ERROR(ENOSPC);
1763 mp->m_sb.sb_frextents = lcounter;
1764 return 0;
1765 case XFS_SBS_DBLOCKS:
1766 lcounter = (long long)mp->m_sb.sb_dblocks;
1767 lcounter += delta;
1768 if (lcounter < 0) {
1769 ASSERT(0);
1770 return XFS_ERROR(EINVAL);
1772 mp->m_sb.sb_dblocks = lcounter;
1773 return 0;
1774 case XFS_SBS_AGCOUNT:
1775 scounter = mp->m_sb.sb_agcount;
1776 scounter += delta;
1777 if (scounter < 0) {
1778 ASSERT(0);
1779 return XFS_ERROR(EINVAL);
1781 mp->m_sb.sb_agcount = scounter;
1782 return 0;
1783 case XFS_SBS_IMAX_PCT:
1784 scounter = mp->m_sb.sb_imax_pct;
1785 scounter += delta;
1786 if (scounter < 0) {
1787 ASSERT(0);
1788 return XFS_ERROR(EINVAL);
1790 mp->m_sb.sb_imax_pct = scounter;
1791 return 0;
1792 case XFS_SBS_REXTSIZE:
1793 scounter = mp->m_sb.sb_rextsize;
1794 scounter += delta;
1795 if (scounter < 0) {
1796 ASSERT(0);
1797 return XFS_ERROR(EINVAL);
1799 mp->m_sb.sb_rextsize = scounter;
1800 return 0;
1801 case XFS_SBS_RBMBLOCKS:
1802 scounter = mp->m_sb.sb_rbmblocks;
1803 scounter += delta;
1804 if (scounter < 0) {
1805 ASSERT(0);
1806 return XFS_ERROR(EINVAL);
1808 mp->m_sb.sb_rbmblocks = scounter;
1809 return 0;
1810 case XFS_SBS_RBLOCKS:
1811 lcounter = (long long)mp->m_sb.sb_rblocks;
1812 lcounter += delta;
1813 if (lcounter < 0) {
1814 ASSERT(0);
1815 return XFS_ERROR(EINVAL);
1817 mp->m_sb.sb_rblocks = lcounter;
1818 return 0;
1819 case XFS_SBS_REXTENTS:
1820 lcounter = (long long)mp->m_sb.sb_rextents;
1821 lcounter += delta;
1822 if (lcounter < 0) {
1823 ASSERT(0);
1824 return XFS_ERROR(EINVAL);
1826 mp->m_sb.sb_rextents = lcounter;
1827 return 0;
1828 case XFS_SBS_REXTSLOG:
1829 scounter = mp->m_sb.sb_rextslog;
1830 scounter += delta;
1831 if (scounter < 0) {
1832 ASSERT(0);
1833 return XFS_ERROR(EINVAL);
1835 mp->m_sb.sb_rextslog = scounter;
1836 return 0;
1837 default:
1838 ASSERT(0);
1839 return XFS_ERROR(EINVAL);
1844 * xfs_mod_incore_sb() is used to change a field in the in-core
1845 * superblock structure by the specified delta. This modification
1846 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1847 * routine to do the work.
1850 xfs_mod_incore_sb(
1851 struct xfs_mount *mp,
1852 xfs_sb_field_t field,
1853 int64_t delta,
1854 int rsvd)
1856 int status;
1858 #ifdef HAVE_PERCPU_SB
1859 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1860 #endif
1861 spin_lock(&mp->m_sb_lock);
1862 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1863 spin_unlock(&mp->m_sb_lock);
1865 return status;
1869 * Change more than one field in the in-core superblock structure at a time.
1871 * The fields and changes to those fields are specified in the array of
1872 * xfs_mod_sb structures passed in. Either all of the specified deltas
1873 * will be applied or none of them will. If any modified field dips below 0,
1874 * then all modifications will be backed out and EINVAL will be returned.
1876 * Note that this function may not be used for the superblock values that
1877 * are tracked with the in-memory per-cpu counters - a direct call to
1878 * xfs_icsb_modify_counters is required for these.
1881 xfs_mod_incore_sb_batch(
1882 struct xfs_mount *mp,
1883 xfs_mod_sb_t *msb,
1884 uint nmsb,
1885 int rsvd)
1887 xfs_mod_sb_t *msbp;
1888 int error = 0;
1891 * Loop through the array of mod structures and apply each individually.
1892 * If any fail, then back out all those which have already been applied.
1893 * Do all of this within the scope of the m_sb_lock so that all of the
1894 * changes will be atomic.
1896 spin_lock(&mp->m_sb_lock);
1897 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1898 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1899 msbp->msb_field > XFS_SBS_FDBLOCKS);
1901 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1902 msbp->msb_delta, rsvd);
1903 if (error)
1904 goto unwind;
1906 spin_unlock(&mp->m_sb_lock);
1907 return 0;
1909 unwind:
1910 while (--msbp >= msb) {
1911 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1912 -msbp->msb_delta, rsvd);
1913 ASSERT(error == 0);
1915 spin_unlock(&mp->m_sb_lock);
1916 return error;
1920 * xfs_getsb() is called to obtain the buffer for the superblock.
1921 * The buffer is returned locked and read in from disk.
1922 * The buffer should be released with a call to xfs_brelse().
1924 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1925 * the superblock buffer if it can be locked without sleeping.
1926 * If it can't then we'll return NULL.
1928 struct xfs_buf *
1929 xfs_getsb(
1930 struct xfs_mount *mp,
1931 int flags)
1933 struct xfs_buf *bp = mp->m_sb_bp;
1935 if (!xfs_buf_trylock(bp)) {
1936 if (flags & XBF_TRYLOCK)
1937 return NULL;
1938 xfs_buf_lock(bp);
1941 xfs_buf_hold(bp);
1942 ASSERT(XFS_BUF_ISDONE(bp));
1943 return bp;
1947 * Used to free the superblock along various error paths.
1949 void
1950 xfs_freesb(
1951 struct xfs_mount *mp)
1953 struct xfs_buf *bp = mp->m_sb_bp;
1955 xfs_buf_lock(bp);
1956 mp->m_sb_bp = NULL;
1957 xfs_buf_relse(bp);
1961 * Used to log changes to the superblock unit and width fields which could
1962 * be altered by the mount options, as well as any potential sb_features2
1963 * fixup. Only the first superblock is updated.
1966 xfs_mount_log_sb(
1967 xfs_mount_t *mp,
1968 __int64_t fields)
1970 xfs_trans_t *tp;
1971 int error;
1973 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1974 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1975 XFS_SB_VERSIONNUM));
1977 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1978 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1979 XFS_DEFAULT_LOG_COUNT);
1980 if (error) {
1981 xfs_trans_cancel(tp, 0);
1982 return error;
1984 xfs_mod_sb(tp, fields);
1985 error = xfs_trans_commit(tp, 0);
1986 return error;
1990 * If the underlying (data/log/rt) device is readonly, there are some
1991 * operations that cannot proceed.
1994 xfs_dev_is_read_only(
1995 struct xfs_mount *mp,
1996 char *message)
1998 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1999 xfs_readonly_buftarg(mp->m_logdev_targp) ||
2000 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
2001 xfs_notice(mp, "%s required on read-only device.", message);
2002 xfs_notice(mp, "write access unavailable, cannot proceed.");
2003 return EROFS;
2005 return 0;
2008 #ifdef HAVE_PERCPU_SB
2010 * Per-cpu incore superblock counters
2012 * Simple concept, difficult implementation
2014 * Basically, replace the incore superblock counters with a distributed per cpu
2015 * counter for contended fields (e.g. free block count).
2017 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2018 * hence needs to be accurately read when we are running low on space. Hence
2019 * there is a method to enable and disable the per-cpu counters based on how
2020 * much "stuff" is available in them.
2022 * Basically, a counter is enabled if there is enough free resource to justify
2023 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2024 * ENOSPC), then we disable the counters to synchronise all callers and
2025 * re-distribute the available resources.
2027 * If, once we redistributed the available resources, we still get a failure,
2028 * we disable the per-cpu counter and go through the slow path.
2030 * The slow path is the current xfs_mod_incore_sb() function. This means that
2031 * when we disable a per-cpu counter, we need to drain its resources back to
2032 * the global superblock. We do this after disabling the counter to prevent
2033 * more threads from queueing up on the counter.
2035 * Essentially, this means that we still need a lock in the fast path to enable
2036 * synchronisation between the global counters and the per-cpu counters. This
2037 * is not a problem because the lock will be local to a CPU almost all the time
2038 * and have little contention except when we get to ENOSPC conditions.
2040 * Basically, this lock becomes a barrier that enables us to lock out the fast
2041 * path while we do things like enabling and disabling counters and
2042 * synchronising the counters.
2044 * Locking rules:
2046 * 1. m_sb_lock before picking up per-cpu locks
2047 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2048 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2049 * 4. modifying per-cpu counters requires holding per-cpu lock
2050 * 5. modifying global counters requires holding m_sb_lock
2051 * 6. enabling or disabling a counter requires holding the m_sb_lock
2052 * and _none_ of the per-cpu locks.
2054 * Disabled counters are only ever re-enabled by a balance operation
2055 * that results in more free resources per CPU than a given threshold.
2056 * To ensure counters don't remain disabled, they are rebalanced when
2057 * the global resource goes above a higher threshold (i.e. some hysteresis
2058 * is present to prevent thrashing).
2061 #ifdef CONFIG_HOTPLUG_CPU
2063 * hot-plug CPU notifier support.
2065 * We need a notifier per filesystem as we need to be able to identify
2066 * the filesystem to balance the counters out. This is achieved by
2067 * having a notifier block embedded in the xfs_mount_t and doing pointer
2068 * magic to get the mount pointer from the notifier block address.
2070 STATIC int
2071 xfs_icsb_cpu_notify(
2072 struct notifier_block *nfb,
2073 unsigned long action,
2074 void *hcpu)
2076 xfs_icsb_cnts_t *cntp;
2077 xfs_mount_t *mp;
2079 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2080 cntp = (xfs_icsb_cnts_t *)
2081 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2082 switch (action) {
2083 case CPU_UP_PREPARE:
2084 case CPU_UP_PREPARE_FROZEN:
2085 /* Easy Case - initialize the area and locks, and
2086 * then rebalance when online does everything else for us. */
2087 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2088 break;
2089 case CPU_ONLINE:
2090 case CPU_ONLINE_FROZEN:
2091 xfs_icsb_lock(mp);
2092 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2093 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2094 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2095 xfs_icsb_unlock(mp);
2096 break;
2097 case CPU_DEAD:
2098 case CPU_DEAD_FROZEN:
2099 /* Disable all the counters, then fold the dead cpu's
2100 * count into the total on the global superblock and
2101 * re-enable the counters. */
2102 xfs_icsb_lock(mp);
2103 spin_lock(&mp->m_sb_lock);
2104 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2105 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2106 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2108 mp->m_sb.sb_icount += cntp->icsb_icount;
2109 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2110 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2112 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2114 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2115 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2116 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2117 spin_unlock(&mp->m_sb_lock);
2118 xfs_icsb_unlock(mp);
2119 break;
2122 return NOTIFY_OK;
2124 #endif /* CONFIG_HOTPLUG_CPU */
2127 xfs_icsb_init_counters(
2128 xfs_mount_t *mp)
2130 xfs_icsb_cnts_t *cntp;
2131 int i;
2133 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2134 if (mp->m_sb_cnts == NULL)
2135 return -ENOMEM;
2137 #ifdef CONFIG_HOTPLUG_CPU
2138 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2139 mp->m_icsb_notifier.priority = 0;
2140 register_hotcpu_notifier(&mp->m_icsb_notifier);
2141 #endif /* CONFIG_HOTPLUG_CPU */
2143 for_each_online_cpu(i) {
2144 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2145 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2148 mutex_init(&mp->m_icsb_mutex);
2151 * start with all counters disabled so that the
2152 * initial balance kicks us off correctly
2154 mp->m_icsb_counters = -1;
2155 return 0;
2158 void
2159 xfs_icsb_reinit_counters(
2160 xfs_mount_t *mp)
2162 xfs_icsb_lock(mp);
2164 * start with all counters disabled so that the
2165 * initial balance kicks us off correctly
2167 mp->m_icsb_counters = -1;
2168 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2169 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2170 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2171 xfs_icsb_unlock(mp);
2174 void
2175 xfs_icsb_destroy_counters(
2176 xfs_mount_t *mp)
2178 if (mp->m_sb_cnts) {
2179 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2180 free_percpu(mp->m_sb_cnts);
2182 mutex_destroy(&mp->m_icsb_mutex);
2185 STATIC void
2186 xfs_icsb_lock_cntr(
2187 xfs_icsb_cnts_t *icsbp)
2189 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2190 ndelay(1000);
2194 STATIC void
2195 xfs_icsb_unlock_cntr(
2196 xfs_icsb_cnts_t *icsbp)
2198 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2202 STATIC void
2203 xfs_icsb_lock_all_counters(
2204 xfs_mount_t *mp)
2206 xfs_icsb_cnts_t *cntp;
2207 int i;
2209 for_each_online_cpu(i) {
2210 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2211 xfs_icsb_lock_cntr(cntp);
2215 STATIC void
2216 xfs_icsb_unlock_all_counters(
2217 xfs_mount_t *mp)
2219 xfs_icsb_cnts_t *cntp;
2220 int i;
2222 for_each_online_cpu(i) {
2223 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2224 xfs_icsb_unlock_cntr(cntp);
2228 STATIC void
2229 xfs_icsb_count(
2230 xfs_mount_t *mp,
2231 xfs_icsb_cnts_t *cnt,
2232 int flags)
2234 xfs_icsb_cnts_t *cntp;
2235 int i;
2237 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2239 if (!(flags & XFS_ICSB_LAZY_COUNT))
2240 xfs_icsb_lock_all_counters(mp);
2242 for_each_online_cpu(i) {
2243 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2244 cnt->icsb_icount += cntp->icsb_icount;
2245 cnt->icsb_ifree += cntp->icsb_ifree;
2246 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2249 if (!(flags & XFS_ICSB_LAZY_COUNT))
2250 xfs_icsb_unlock_all_counters(mp);
2253 STATIC int
2254 xfs_icsb_counter_disabled(
2255 xfs_mount_t *mp,
2256 xfs_sb_field_t field)
2258 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2259 return test_bit(field, &mp->m_icsb_counters);
2262 STATIC void
2263 xfs_icsb_disable_counter(
2264 xfs_mount_t *mp,
2265 xfs_sb_field_t field)
2267 xfs_icsb_cnts_t cnt;
2269 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2272 * If we are already disabled, then there is nothing to do
2273 * here. We check before locking all the counters to avoid
2274 * the expensive lock operation when being called in the
2275 * slow path and the counter is already disabled. This is
2276 * safe because the only time we set or clear this state is under
2277 * the m_icsb_mutex.
2279 if (xfs_icsb_counter_disabled(mp, field))
2280 return;
2282 xfs_icsb_lock_all_counters(mp);
2283 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2284 /* drain back to superblock */
2286 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2287 switch(field) {
2288 case XFS_SBS_ICOUNT:
2289 mp->m_sb.sb_icount = cnt.icsb_icount;
2290 break;
2291 case XFS_SBS_IFREE:
2292 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2293 break;
2294 case XFS_SBS_FDBLOCKS:
2295 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2296 break;
2297 default:
2298 BUG();
2302 xfs_icsb_unlock_all_counters(mp);
2305 STATIC void
2306 xfs_icsb_enable_counter(
2307 xfs_mount_t *mp,
2308 xfs_sb_field_t field,
2309 uint64_t count,
2310 uint64_t resid)
2312 xfs_icsb_cnts_t *cntp;
2313 int i;
2315 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2317 xfs_icsb_lock_all_counters(mp);
2318 for_each_online_cpu(i) {
2319 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2320 switch (field) {
2321 case XFS_SBS_ICOUNT:
2322 cntp->icsb_icount = count + resid;
2323 break;
2324 case XFS_SBS_IFREE:
2325 cntp->icsb_ifree = count + resid;
2326 break;
2327 case XFS_SBS_FDBLOCKS:
2328 cntp->icsb_fdblocks = count + resid;
2329 break;
2330 default:
2331 BUG();
2332 break;
2334 resid = 0;
2336 clear_bit(field, &mp->m_icsb_counters);
2337 xfs_icsb_unlock_all_counters(mp);
2340 void
2341 xfs_icsb_sync_counters_locked(
2342 xfs_mount_t *mp,
2343 int flags)
2345 xfs_icsb_cnts_t cnt;
2347 xfs_icsb_count(mp, &cnt, flags);
2349 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2350 mp->m_sb.sb_icount = cnt.icsb_icount;
2351 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2352 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2353 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2354 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2358 * Accurate update of per-cpu counters to incore superblock
2360 void
2361 xfs_icsb_sync_counters(
2362 xfs_mount_t *mp,
2363 int flags)
2365 spin_lock(&mp->m_sb_lock);
2366 xfs_icsb_sync_counters_locked(mp, flags);
2367 spin_unlock(&mp->m_sb_lock);
2371 * Balance and enable/disable counters as necessary.
2373 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2374 * chosen to be the same number as single on disk allocation chunk per CPU, and
2375 * free blocks is something far enough zero that we aren't going thrash when we
2376 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2377 * prevent looping endlessly when xfs_alloc_space asks for more than will
2378 * be distributed to a single CPU but each CPU has enough blocks to be
2379 * reenabled.
2381 * Note that we can be called when counters are already disabled.
2382 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2383 * prevent locking every per-cpu counter needlessly.
2386 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2387 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2388 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2389 STATIC void
2390 xfs_icsb_balance_counter_locked(
2391 xfs_mount_t *mp,
2392 xfs_sb_field_t field,
2393 int min_per_cpu)
2395 uint64_t count, resid;
2396 int weight = num_online_cpus();
2397 uint64_t min = (uint64_t)min_per_cpu;
2399 /* disable counter and sync counter */
2400 xfs_icsb_disable_counter(mp, field);
2402 /* update counters - first CPU gets residual*/
2403 switch (field) {
2404 case XFS_SBS_ICOUNT:
2405 count = mp->m_sb.sb_icount;
2406 resid = do_div(count, weight);
2407 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2408 return;
2409 break;
2410 case XFS_SBS_IFREE:
2411 count = mp->m_sb.sb_ifree;
2412 resid = do_div(count, weight);
2413 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2414 return;
2415 break;
2416 case XFS_SBS_FDBLOCKS:
2417 count = mp->m_sb.sb_fdblocks;
2418 resid = do_div(count, weight);
2419 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2420 return;
2421 break;
2422 default:
2423 BUG();
2424 count = resid = 0; /* quiet, gcc */
2425 break;
2428 xfs_icsb_enable_counter(mp, field, count, resid);
2431 STATIC void
2432 xfs_icsb_balance_counter(
2433 xfs_mount_t *mp,
2434 xfs_sb_field_t fields,
2435 int min_per_cpu)
2437 spin_lock(&mp->m_sb_lock);
2438 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2439 spin_unlock(&mp->m_sb_lock);
2443 xfs_icsb_modify_counters(
2444 xfs_mount_t *mp,
2445 xfs_sb_field_t field,
2446 int64_t delta,
2447 int rsvd)
2449 xfs_icsb_cnts_t *icsbp;
2450 long long lcounter; /* long counter for 64 bit fields */
2451 int ret = 0;
2453 might_sleep();
2454 again:
2455 preempt_disable();
2456 icsbp = this_cpu_ptr(mp->m_sb_cnts);
2459 * if the counter is disabled, go to slow path
2461 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2462 goto slow_path;
2463 xfs_icsb_lock_cntr(icsbp);
2464 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2465 xfs_icsb_unlock_cntr(icsbp);
2466 goto slow_path;
2469 switch (field) {
2470 case XFS_SBS_ICOUNT:
2471 lcounter = icsbp->icsb_icount;
2472 lcounter += delta;
2473 if (unlikely(lcounter < 0))
2474 goto balance_counter;
2475 icsbp->icsb_icount = lcounter;
2476 break;
2478 case XFS_SBS_IFREE:
2479 lcounter = icsbp->icsb_ifree;
2480 lcounter += delta;
2481 if (unlikely(lcounter < 0))
2482 goto balance_counter;
2483 icsbp->icsb_ifree = lcounter;
2484 break;
2486 case XFS_SBS_FDBLOCKS:
2487 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2489 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2490 lcounter += delta;
2491 if (unlikely(lcounter < 0))
2492 goto balance_counter;
2493 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2494 break;
2495 default:
2496 BUG();
2497 break;
2499 xfs_icsb_unlock_cntr(icsbp);
2500 preempt_enable();
2501 return 0;
2503 slow_path:
2504 preempt_enable();
2507 * serialise with a mutex so we don't burn lots of cpu on
2508 * the superblock lock. We still need to hold the superblock
2509 * lock, however, when we modify the global structures.
2511 xfs_icsb_lock(mp);
2514 * Now running atomically.
2516 * If the counter is enabled, someone has beaten us to rebalancing.
2517 * Drop the lock and try again in the fast path....
2519 if (!(xfs_icsb_counter_disabled(mp, field))) {
2520 xfs_icsb_unlock(mp);
2521 goto again;
2525 * The counter is currently disabled. Because we are
2526 * running atomically here, we know a rebalance cannot
2527 * be in progress. Hence we can go straight to operating
2528 * on the global superblock. We do not call xfs_mod_incore_sb()
2529 * here even though we need to get the m_sb_lock. Doing so
2530 * will cause us to re-enter this function and deadlock.
2531 * Hence we get the m_sb_lock ourselves and then call
2532 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2533 * directly on the global counters.
2535 spin_lock(&mp->m_sb_lock);
2536 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2537 spin_unlock(&mp->m_sb_lock);
2540 * Now that we've modified the global superblock, we
2541 * may be able to re-enable the distributed counters
2542 * (e.g. lots of space just got freed). After that
2543 * we are done.
2545 if (ret != ENOSPC)
2546 xfs_icsb_balance_counter(mp, field, 0);
2547 xfs_icsb_unlock(mp);
2548 return ret;
2550 balance_counter:
2551 xfs_icsb_unlock_cntr(icsbp);
2552 preempt_enable();
2555 * We may have multiple threads here if multiple per-cpu
2556 * counters run dry at the same time. This will mean we can
2557 * do more balances than strictly necessary but it is not
2558 * the common slowpath case.
2560 xfs_icsb_lock(mp);
2563 * running atomically.
2565 * This will leave the counter in the correct state for future
2566 * accesses. After the rebalance, we simply try again and our retry
2567 * will either succeed through the fast path or slow path without
2568 * another balance operation being required.
2570 xfs_icsb_balance_counter(mp, field, delta);
2571 xfs_icsb_unlock(mp);
2572 goto again;
2575 #endif