kvm tools, setup: Create private directory
[linux-2.6/next.git] / fs / xfs / xfs_mount.c
blobb49b82363d203dfa8d2c19abe9e08993b34c27e9
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. 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 if (loud)
376 xfs_warn(mp, "SB sanity check 1 failed");
377 return XFS_ERROR(EFSCORRUPTED);
381 * Sanity check AG count, size fields against data size field
383 if (unlikely(
384 sbp->sb_dblocks == 0 ||
385 sbp->sb_dblocks >
386 (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
387 sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
388 sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
389 if (loud)
390 xfs_warn(mp, "SB sanity check 2 failed");
391 return XFS_ERROR(EFSCORRUPTED);
395 * Until this is fixed only page-sized or smaller data blocks work.
397 if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
398 if (loud) {
399 xfs_warn(mp,
400 "File system with blocksize %d bytes. "
401 "Only pagesize (%ld) or less will currently work.",
402 sbp->sb_blocksize, PAGE_SIZE);
404 return XFS_ERROR(ENOSYS);
408 * Currently only very few inode sizes are supported.
410 switch (sbp->sb_inodesize) {
411 case 256:
412 case 512:
413 case 1024:
414 case 2048:
415 break;
416 default:
417 if (loud)
418 xfs_warn(mp, "inode size of %d bytes not supported",
419 sbp->sb_inodesize);
420 return XFS_ERROR(ENOSYS);
423 if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
424 xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
425 if (loud)
426 xfs_warn(mp,
427 "file system too large to be mounted on this system.");
428 return XFS_ERROR(EFBIG);
431 if (unlikely(sbp->sb_inprogress)) {
432 if (loud)
433 xfs_warn(mp, "file system busy");
434 return XFS_ERROR(EFSCORRUPTED);
438 * Version 1 directory format has never worked on Linux.
440 if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
441 if (loud)
442 xfs_warn(mp,
443 "file system using version 1 directory format");
444 return XFS_ERROR(ENOSYS);
447 return 0;
451 xfs_initialize_perag(
452 xfs_mount_t *mp,
453 xfs_agnumber_t agcount,
454 xfs_agnumber_t *maxagi)
456 xfs_agnumber_t index, max_metadata;
457 xfs_agnumber_t first_initialised = 0;
458 xfs_perag_t *pag;
459 xfs_agino_t agino;
460 xfs_ino_t ino;
461 xfs_sb_t *sbp = &mp->m_sb;
462 int error = -ENOMEM;
465 * Walk the current per-ag tree so we don't try to initialise AGs
466 * that already exist (growfs case). Allocate and insert all the
467 * AGs we don't find ready for initialisation.
469 for (index = 0; index < agcount; index++) {
470 pag = xfs_perag_get(mp, index);
471 if (pag) {
472 xfs_perag_put(pag);
473 continue;
475 if (!first_initialised)
476 first_initialised = index;
478 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
479 if (!pag)
480 goto out_unwind;
481 pag->pag_agno = index;
482 pag->pag_mount = mp;
483 spin_lock_init(&pag->pag_ici_lock);
484 mutex_init(&pag->pag_ici_reclaim_lock);
485 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
486 spin_lock_init(&pag->pag_buf_lock);
487 pag->pag_buf_tree = RB_ROOT;
489 if (radix_tree_preload(GFP_NOFS))
490 goto out_unwind;
492 spin_lock(&mp->m_perag_lock);
493 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
494 BUG();
495 spin_unlock(&mp->m_perag_lock);
496 radix_tree_preload_end();
497 error = -EEXIST;
498 goto out_unwind;
500 spin_unlock(&mp->m_perag_lock);
501 radix_tree_preload_end();
505 * If we mount with the inode64 option, or no inode overflows
506 * the legacy 32-bit address space clear the inode32 option.
508 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
509 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
511 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
512 mp->m_flags |= XFS_MOUNT_32BITINODES;
513 else
514 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
516 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
518 * Calculate how much should be reserved for inodes to meet
519 * the max inode percentage.
521 if (mp->m_maxicount) {
522 __uint64_t icount;
524 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
525 do_div(icount, 100);
526 icount += sbp->sb_agblocks - 1;
527 do_div(icount, sbp->sb_agblocks);
528 max_metadata = icount;
529 } else {
530 max_metadata = agcount;
533 for (index = 0; index < agcount; index++) {
534 ino = XFS_AGINO_TO_INO(mp, index, agino);
535 if (ino > XFS_MAXINUMBER_32) {
536 index++;
537 break;
540 pag = xfs_perag_get(mp, index);
541 pag->pagi_inodeok = 1;
542 if (index < max_metadata)
543 pag->pagf_metadata = 1;
544 xfs_perag_put(pag);
546 } else {
547 for (index = 0; index < agcount; index++) {
548 pag = xfs_perag_get(mp, index);
549 pag->pagi_inodeok = 1;
550 xfs_perag_put(pag);
554 if (maxagi)
555 *maxagi = index;
556 return 0;
558 out_unwind:
559 kmem_free(pag);
560 for (; index > first_initialised; index--) {
561 pag = radix_tree_delete(&mp->m_perag_tree, index);
562 kmem_free(pag);
564 return error;
567 void
568 xfs_sb_from_disk(
569 xfs_sb_t *to,
570 xfs_dsb_t *from)
572 to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
573 to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
574 to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
575 to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
576 to->sb_rextents = be64_to_cpu(from->sb_rextents);
577 memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
578 to->sb_logstart = be64_to_cpu(from->sb_logstart);
579 to->sb_rootino = be64_to_cpu(from->sb_rootino);
580 to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
581 to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
582 to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
583 to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
584 to->sb_agcount = be32_to_cpu(from->sb_agcount);
585 to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
586 to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
587 to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
588 to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
589 to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
590 to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
591 memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
592 to->sb_blocklog = from->sb_blocklog;
593 to->sb_sectlog = from->sb_sectlog;
594 to->sb_inodelog = from->sb_inodelog;
595 to->sb_inopblog = from->sb_inopblog;
596 to->sb_agblklog = from->sb_agblklog;
597 to->sb_rextslog = from->sb_rextslog;
598 to->sb_inprogress = from->sb_inprogress;
599 to->sb_imax_pct = from->sb_imax_pct;
600 to->sb_icount = be64_to_cpu(from->sb_icount);
601 to->sb_ifree = be64_to_cpu(from->sb_ifree);
602 to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
603 to->sb_frextents = be64_to_cpu(from->sb_frextents);
604 to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
605 to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
606 to->sb_qflags = be16_to_cpu(from->sb_qflags);
607 to->sb_flags = from->sb_flags;
608 to->sb_shared_vn = from->sb_shared_vn;
609 to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
610 to->sb_unit = be32_to_cpu(from->sb_unit);
611 to->sb_width = be32_to_cpu(from->sb_width);
612 to->sb_dirblklog = from->sb_dirblklog;
613 to->sb_logsectlog = from->sb_logsectlog;
614 to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
615 to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
616 to->sb_features2 = be32_to_cpu(from->sb_features2);
617 to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
621 * Copy in core superblock to ondisk one.
623 * The fields argument is mask of superblock fields to copy.
625 void
626 xfs_sb_to_disk(
627 xfs_dsb_t *to,
628 xfs_sb_t *from,
629 __int64_t fields)
631 xfs_caddr_t to_ptr = (xfs_caddr_t)to;
632 xfs_caddr_t from_ptr = (xfs_caddr_t)from;
633 xfs_sb_field_t f;
634 int first;
635 int size;
637 ASSERT(fields);
638 if (!fields)
639 return;
641 while (fields) {
642 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
643 first = xfs_sb_info[f].offset;
644 size = xfs_sb_info[f + 1].offset - first;
646 ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
648 if (size == 1 || xfs_sb_info[f].type == 1) {
649 memcpy(to_ptr + first, from_ptr + first, size);
650 } else {
651 switch (size) {
652 case 2:
653 *(__be16 *)(to_ptr + first) =
654 cpu_to_be16(*(__u16 *)(from_ptr + first));
655 break;
656 case 4:
657 *(__be32 *)(to_ptr + first) =
658 cpu_to_be32(*(__u32 *)(from_ptr + first));
659 break;
660 case 8:
661 *(__be64 *)(to_ptr + first) =
662 cpu_to_be64(*(__u64 *)(from_ptr + first));
663 break;
664 default:
665 ASSERT(0);
669 fields &= ~(1LL << f);
674 * xfs_readsb
676 * Does the initial read of the superblock.
679 xfs_readsb(xfs_mount_t *mp, int flags)
681 unsigned int sector_size;
682 xfs_buf_t *bp;
683 int error;
684 int loud = !(flags & XFS_MFSI_QUIET);
686 ASSERT(mp->m_sb_bp == NULL);
687 ASSERT(mp->m_ddev_targp != NULL);
690 * Allocate a (locked) buffer to hold the superblock.
691 * This will be kept around at all times to optimize
692 * access to the superblock.
694 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
696 reread:
697 bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
698 XFS_SB_DADDR, sector_size, 0);
699 if (!bp) {
700 if (loud)
701 xfs_warn(mp, "SB buffer read failed");
702 return EIO;
706 * Initialize the mount structure from the superblock.
707 * But first do some basic consistency checking.
709 xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
710 error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
711 if (error) {
712 if (loud)
713 xfs_warn(mp, "SB validate failed");
714 goto release_buf;
718 * We must be able to do sector-sized and sector-aligned IO.
720 if (sector_size > mp->m_sb.sb_sectsize) {
721 if (loud)
722 xfs_warn(mp, "device supports %u byte sectors (not %u)",
723 sector_size, mp->m_sb.sb_sectsize);
724 error = ENOSYS;
725 goto release_buf;
729 * If device sector size is smaller than the superblock size,
730 * re-read the superblock so the buffer is correctly sized.
732 if (sector_size < mp->m_sb.sb_sectsize) {
733 xfs_buf_relse(bp);
734 sector_size = mp->m_sb.sb_sectsize;
735 goto reread;
738 /* Initialize per-cpu counters */
739 xfs_icsb_reinit_counters(mp);
741 mp->m_sb_bp = bp;
742 xfs_buf_unlock(bp);
743 return 0;
745 release_buf:
746 xfs_buf_relse(bp);
747 return error;
752 * xfs_mount_common
754 * Mount initialization code establishing various mount
755 * fields from the superblock associated with the given
756 * mount structure
758 STATIC void
759 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
761 mp->m_agfrotor = mp->m_agirotor = 0;
762 spin_lock_init(&mp->m_agirotor_lock);
763 mp->m_maxagi = mp->m_sb.sb_agcount;
764 mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
765 mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
766 mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
767 mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
768 mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
769 mp->m_blockmask = sbp->sb_blocksize - 1;
770 mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
771 mp->m_blockwmask = mp->m_blockwsize - 1;
773 mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
774 mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
775 mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
776 mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
778 mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
779 mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
780 mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
781 mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
783 mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
784 mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
785 mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
786 mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
788 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
789 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
790 sbp->sb_inopblock);
791 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
795 * xfs_initialize_perag_data
797 * Read in each per-ag structure so we can count up the number of
798 * allocated inodes, free inodes and used filesystem blocks as this
799 * information is no longer persistent in the superblock. Once we have
800 * this information, write it into the in-core superblock structure.
802 STATIC int
803 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
805 xfs_agnumber_t index;
806 xfs_perag_t *pag;
807 xfs_sb_t *sbp = &mp->m_sb;
808 uint64_t ifree = 0;
809 uint64_t ialloc = 0;
810 uint64_t bfree = 0;
811 uint64_t bfreelst = 0;
812 uint64_t btree = 0;
813 int error;
815 for (index = 0; index < agcount; index++) {
817 * read the agf, then the agi. This gets us
818 * all the information we need and populates the
819 * per-ag structures for us.
821 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
822 if (error)
823 return error;
825 error = xfs_ialloc_pagi_init(mp, NULL, index);
826 if (error)
827 return error;
828 pag = xfs_perag_get(mp, index);
829 ifree += pag->pagi_freecount;
830 ialloc += pag->pagi_count;
831 bfree += pag->pagf_freeblks;
832 bfreelst += pag->pagf_flcount;
833 btree += pag->pagf_btreeblks;
834 xfs_perag_put(pag);
837 * Overwrite incore superblock counters with just-read data
839 spin_lock(&mp->m_sb_lock);
840 sbp->sb_ifree = ifree;
841 sbp->sb_icount = ialloc;
842 sbp->sb_fdblocks = bfree + bfreelst + btree;
843 spin_unlock(&mp->m_sb_lock);
845 /* Fixup the per-cpu counters as well. */
846 xfs_icsb_reinit_counters(mp);
848 return 0;
852 * Update alignment values based on mount options and sb values
854 STATIC int
855 xfs_update_alignment(xfs_mount_t *mp)
857 xfs_sb_t *sbp = &(mp->m_sb);
859 if (mp->m_dalign) {
861 * If stripe unit and stripe width are not multiples
862 * of the fs blocksize turn off alignment.
864 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
865 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
866 if (mp->m_flags & XFS_MOUNT_RETERR) {
867 xfs_warn(mp, "alignment check 1 failed");
868 return XFS_ERROR(EINVAL);
870 mp->m_dalign = mp->m_swidth = 0;
871 } else {
873 * Convert the stripe unit and width to FSBs.
875 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
876 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
877 if (mp->m_flags & XFS_MOUNT_RETERR) {
878 return XFS_ERROR(EINVAL);
880 xfs_warn(mp,
881 "stripe alignment turned off: sunit(%d)/swidth(%d) "
882 "incompatible with agsize(%d)",
883 mp->m_dalign, mp->m_swidth,
884 sbp->sb_agblocks);
886 mp->m_dalign = 0;
887 mp->m_swidth = 0;
888 } else if (mp->m_dalign) {
889 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
890 } else {
891 if (mp->m_flags & XFS_MOUNT_RETERR) {
892 xfs_warn(mp,
893 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
894 mp->m_dalign,
895 mp->m_blockmask +1);
896 return XFS_ERROR(EINVAL);
898 mp->m_swidth = 0;
903 * Update superblock with new values
904 * and log changes
906 if (xfs_sb_version_hasdalign(sbp)) {
907 if (sbp->sb_unit != mp->m_dalign) {
908 sbp->sb_unit = mp->m_dalign;
909 mp->m_update_flags |= XFS_SB_UNIT;
911 if (sbp->sb_width != mp->m_swidth) {
912 sbp->sb_width = mp->m_swidth;
913 mp->m_update_flags |= XFS_SB_WIDTH;
916 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
917 xfs_sb_version_hasdalign(&mp->m_sb)) {
918 mp->m_dalign = sbp->sb_unit;
919 mp->m_swidth = sbp->sb_width;
922 return 0;
926 * Set the maximum inode count for this filesystem
928 STATIC void
929 xfs_set_maxicount(xfs_mount_t *mp)
931 xfs_sb_t *sbp = &(mp->m_sb);
932 __uint64_t icount;
934 if (sbp->sb_imax_pct) {
936 * Make sure the maximum inode count is a multiple
937 * of the units we allocate inodes in.
939 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
940 do_div(icount, 100);
941 do_div(icount, mp->m_ialloc_blks);
942 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
943 sbp->sb_inopblog;
944 } else {
945 mp->m_maxicount = 0;
950 * Set the default minimum read and write sizes unless
951 * already specified in a mount option.
952 * We use smaller I/O sizes when the file system
953 * is being used for NFS service (wsync mount option).
955 STATIC void
956 xfs_set_rw_sizes(xfs_mount_t *mp)
958 xfs_sb_t *sbp = &(mp->m_sb);
959 int readio_log, writeio_log;
961 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
962 if (mp->m_flags & XFS_MOUNT_WSYNC) {
963 readio_log = XFS_WSYNC_READIO_LOG;
964 writeio_log = XFS_WSYNC_WRITEIO_LOG;
965 } else {
966 readio_log = XFS_READIO_LOG_LARGE;
967 writeio_log = XFS_WRITEIO_LOG_LARGE;
969 } else {
970 readio_log = mp->m_readio_log;
971 writeio_log = mp->m_writeio_log;
974 if (sbp->sb_blocklog > readio_log) {
975 mp->m_readio_log = sbp->sb_blocklog;
976 } else {
977 mp->m_readio_log = readio_log;
979 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
980 if (sbp->sb_blocklog > writeio_log) {
981 mp->m_writeio_log = sbp->sb_blocklog;
982 } else {
983 mp->m_writeio_log = writeio_log;
985 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
989 * precalculate the low space thresholds for dynamic speculative preallocation.
991 void
992 xfs_set_low_space_thresholds(
993 struct xfs_mount *mp)
995 int i;
997 for (i = 0; i < XFS_LOWSP_MAX; i++) {
998 __uint64_t space = mp->m_sb.sb_dblocks;
1000 do_div(space, 100);
1001 mp->m_low_space[i] = space * (i + 1);
1007 * Set whether we're using inode alignment.
1009 STATIC void
1010 xfs_set_inoalignment(xfs_mount_t *mp)
1012 if (xfs_sb_version_hasalign(&mp->m_sb) &&
1013 mp->m_sb.sb_inoalignmt >=
1014 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1015 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1016 else
1017 mp->m_inoalign_mask = 0;
1019 * If we are using stripe alignment, check whether
1020 * the stripe unit is a multiple of the inode alignment
1022 if (mp->m_dalign && mp->m_inoalign_mask &&
1023 !(mp->m_dalign & mp->m_inoalign_mask))
1024 mp->m_sinoalign = mp->m_dalign;
1025 else
1026 mp->m_sinoalign = 0;
1030 * Check that the data (and log if separate) are an ok size.
1032 STATIC int
1033 xfs_check_sizes(xfs_mount_t *mp)
1035 xfs_buf_t *bp;
1036 xfs_daddr_t d;
1038 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1039 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1040 xfs_warn(mp, "filesystem size mismatch detected");
1041 return XFS_ERROR(EFBIG);
1043 bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
1044 d - XFS_FSS_TO_BB(mp, 1),
1045 BBTOB(XFS_FSS_TO_BB(mp, 1)), 0);
1046 if (!bp) {
1047 xfs_warn(mp, "last sector read failed");
1048 return EIO;
1050 xfs_buf_relse(bp);
1052 if (mp->m_logdev_targp != mp->m_ddev_targp) {
1053 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1054 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1055 xfs_warn(mp, "log size mismatch detected");
1056 return XFS_ERROR(EFBIG);
1058 bp = xfs_buf_read_uncached(mp, mp->m_logdev_targp,
1059 d - XFS_FSB_TO_BB(mp, 1),
1060 XFS_FSB_TO_B(mp, 1), 0);
1061 if (!bp) {
1062 xfs_warn(mp, "log device read failed");
1063 return EIO;
1065 xfs_buf_relse(bp);
1067 return 0;
1071 * Clear the quotaflags in memory and in the superblock.
1074 xfs_mount_reset_sbqflags(
1075 struct xfs_mount *mp)
1077 int error;
1078 struct xfs_trans *tp;
1080 mp->m_qflags = 0;
1083 * It is OK to look at sb_qflags here in mount path,
1084 * without m_sb_lock.
1086 if (mp->m_sb.sb_qflags == 0)
1087 return 0;
1088 spin_lock(&mp->m_sb_lock);
1089 mp->m_sb.sb_qflags = 0;
1090 spin_unlock(&mp->m_sb_lock);
1093 * If the fs is readonly, let the incore superblock run
1094 * with quotas off but don't flush the update out to disk
1096 if (mp->m_flags & XFS_MOUNT_RDONLY)
1097 return 0;
1099 #ifdef QUOTADEBUG
1100 xfs_notice(mp, "Writing superblock quota changes");
1101 #endif
1103 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1104 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1105 XFS_DEFAULT_LOG_COUNT);
1106 if (error) {
1107 xfs_trans_cancel(tp, 0);
1108 xfs_alert(mp, "%s: Superblock update failed!", __func__);
1109 return error;
1112 xfs_mod_sb(tp, XFS_SB_QFLAGS);
1113 return xfs_trans_commit(tp, 0);
1116 __uint64_t
1117 xfs_default_resblks(xfs_mount_t *mp)
1119 __uint64_t resblks;
1122 * We default to 5% or 8192 fsbs of space reserved, whichever is
1123 * smaller. This is intended to cover concurrent allocation
1124 * transactions when we initially hit enospc. These each require a 4
1125 * block reservation. Hence by default we cover roughly 2000 concurrent
1126 * allocation reservations.
1128 resblks = mp->m_sb.sb_dblocks;
1129 do_div(resblks, 20);
1130 resblks = min_t(__uint64_t, resblks, 8192);
1131 return resblks;
1135 * This function does the following on an initial mount of a file system:
1136 * - reads the superblock from disk and init the mount struct
1137 * - if we're a 32-bit kernel, do a size check on the superblock
1138 * so we don't mount terabyte filesystems
1139 * - init mount struct realtime fields
1140 * - allocate inode hash table for fs
1141 * - init directory manager
1142 * - perform recovery and init the log manager
1145 xfs_mountfs(
1146 xfs_mount_t *mp)
1148 xfs_sb_t *sbp = &(mp->m_sb);
1149 xfs_inode_t *rip;
1150 __uint64_t resblks;
1151 uint quotamount = 0;
1152 uint quotaflags = 0;
1153 int error = 0;
1155 xfs_mount_common(mp, sbp);
1158 * Check for a mismatched features2 values. Older kernels
1159 * read & wrote into the wrong sb offset for sb_features2
1160 * on some platforms due to xfs_sb_t not being 64bit size aligned
1161 * when sb_features2 was added, which made older superblock
1162 * reading/writing routines swap it as a 64-bit value.
1164 * For backwards compatibility, we make both slots equal.
1166 * If we detect a mismatched field, we OR the set bits into the
1167 * existing features2 field in case it has already been modified; we
1168 * don't want to lose any features. We then update the bad location
1169 * with the ORed value so that older kernels will see any features2
1170 * flags, and mark the two fields as needing updates once the
1171 * transaction subsystem is online.
1173 if (xfs_sb_has_mismatched_features2(sbp)) {
1174 xfs_warn(mp, "correcting sb_features alignment problem");
1175 sbp->sb_features2 |= sbp->sb_bad_features2;
1176 sbp->sb_bad_features2 = sbp->sb_features2;
1177 mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1180 * Re-check for ATTR2 in case it was found in bad_features2
1181 * slot.
1183 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1184 !(mp->m_flags & XFS_MOUNT_NOATTR2))
1185 mp->m_flags |= XFS_MOUNT_ATTR2;
1188 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1189 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1190 xfs_sb_version_removeattr2(&mp->m_sb);
1191 mp->m_update_flags |= XFS_SB_FEATURES2;
1193 /* update sb_versionnum for the clearing of the morebits */
1194 if (!sbp->sb_features2)
1195 mp->m_update_flags |= XFS_SB_VERSIONNUM;
1199 * Check if sb_agblocks is aligned at stripe boundary
1200 * If sb_agblocks is NOT aligned turn off m_dalign since
1201 * allocator alignment is within an ag, therefore ag has
1202 * to be aligned at stripe boundary.
1204 error = xfs_update_alignment(mp);
1205 if (error)
1206 goto out;
1208 xfs_alloc_compute_maxlevels(mp);
1209 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1210 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1211 xfs_ialloc_compute_maxlevels(mp);
1213 xfs_set_maxicount(mp);
1215 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1217 error = xfs_uuid_mount(mp);
1218 if (error)
1219 goto out;
1222 * Set the minimum read and write sizes
1224 xfs_set_rw_sizes(mp);
1226 /* set the low space thresholds for dynamic preallocation */
1227 xfs_set_low_space_thresholds(mp);
1230 * Set the inode cluster size.
1231 * This may still be overridden by the file system
1232 * block size if it is larger than the chosen cluster size.
1234 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1237 * Set inode alignment fields
1239 xfs_set_inoalignment(mp);
1242 * Check that the data (and log if separate) are an ok size.
1244 error = xfs_check_sizes(mp);
1245 if (error)
1246 goto out_remove_uuid;
1249 * Initialize realtime fields in the mount structure
1251 error = xfs_rtmount_init(mp);
1252 if (error) {
1253 xfs_warn(mp, "RT mount failed");
1254 goto out_remove_uuid;
1258 * Copies the low order bits of the timestamp and the randomly
1259 * set "sequence" number out of a UUID.
1261 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1263 mp->m_dmevmask = 0; /* not persistent; set after each mount */
1265 xfs_dir_mount(mp);
1268 * Initialize the attribute manager's entries.
1270 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1273 * Initialize the precomputed transaction reservations values.
1275 xfs_trans_init(mp);
1278 * Allocate and initialize the per-ag data.
1280 spin_lock_init(&mp->m_perag_lock);
1281 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1282 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1283 if (error) {
1284 xfs_warn(mp, "Failed per-ag init: %d", error);
1285 goto out_remove_uuid;
1288 if (!sbp->sb_logblocks) {
1289 xfs_warn(mp, "no log defined");
1290 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1291 error = XFS_ERROR(EFSCORRUPTED);
1292 goto out_free_perag;
1296 * log's mount-time initialization. Perform 1st part recovery if needed
1298 error = xfs_log_mount(mp, mp->m_logdev_targp,
1299 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1300 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1301 if (error) {
1302 xfs_warn(mp, "log mount failed");
1303 goto out_free_perag;
1307 * Now the log is mounted, we know if it was an unclean shutdown or
1308 * not. If it was, with the first phase of recovery has completed, we
1309 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1310 * but they are recovered transactionally in the second recovery phase
1311 * later.
1313 * Hence we can safely re-initialise incore superblock counters from
1314 * the per-ag data. These may not be correct if the filesystem was not
1315 * cleanly unmounted, so we need to wait for recovery to finish before
1316 * doing this.
1318 * If the filesystem was cleanly unmounted, then we can trust the
1319 * values in the superblock to be correct and we don't need to do
1320 * anything here.
1322 * If we are currently making the filesystem, the initialisation will
1323 * fail as the perag data is in an undefined state.
1325 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1326 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1327 !mp->m_sb.sb_inprogress) {
1328 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1329 if (error)
1330 goto out_free_perag;
1334 * Get and sanity-check the root inode.
1335 * Save the pointer to it in the mount structure.
1337 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1338 if (error) {
1339 xfs_warn(mp, "failed to read root inode");
1340 goto out_log_dealloc;
1343 ASSERT(rip != NULL);
1345 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1346 xfs_warn(mp, "corrupted root inode %llu: not a directory",
1347 (unsigned long long)rip->i_ino);
1348 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1349 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1350 mp);
1351 error = XFS_ERROR(EFSCORRUPTED);
1352 goto out_rele_rip;
1354 mp->m_rootip = rip; /* save it */
1356 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1359 * Initialize realtime inode pointers in the mount structure
1361 error = xfs_rtmount_inodes(mp);
1362 if (error) {
1364 * Free up the root inode.
1366 xfs_warn(mp, "failed to read RT inodes");
1367 goto out_rele_rip;
1371 * If this is a read-only mount defer the superblock updates until
1372 * the next remount into writeable mode. Otherwise we would never
1373 * perform the update e.g. for the root filesystem.
1375 if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1376 error = xfs_mount_log_sb(mp, mp->m_update_flags);
1377 if (error) {
1378 xfs_warn(mp, "failed to write sb changes");
1379 goto out_rtunmount;
1384 * Initialise the XFS quota management subsystem for this mount
1386 if (XFS_IS_QUOTA_RUNNING(mp)) {
1387 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1388 if (error)
1389 goto out_rtunmount;
1390 } else {
1391 ASSERT(!XFS_IS_QUOTA_ON(mp));
1394 * If a file system had quotas running earlier, but decided to
1395 * mount without -o uquota/pquota/gquota options, revoke the
1396 * quotachecked license.
1398 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1399 xfs_notice(mp, "resetting quota flags");
1400 error = xfs_mount_reset_sbqflags(mp);
1401 if (error)
1402 return error;
1407 * Finish recovering the file system. This part needed to be
1408 * delayed until after the root and real-time bitmap inodes
1409 * were consistently read in.
1411 error = xfs_log_mount_finish(mp);
1412 if (error) {
1413 xfs_warn(mp, "log mount finish failed");
1414 goto out_rtunmount;
1418 * Complete the quota initialisation, post-log-replay component.
1420 if (quotamount) {
1421 ASSERT(mp->m_qflags == 0);
1422 mp->m_qflags = quotaflags;
1424 xfs_qm_mount_quotas(mp);
1428 * Now we are mounted, reserve a small amount of unused space for
1429 * privileged transactions. This is needed so that transaction
1430 * space required for critical operations can dip into this pool
1431 * when at ENOSPC. This is needed for operations like create with
1432 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1433 * are not allowed to use this reserved space.
1435 * This may drive us straight to ENOSPC on mount, but that implies
1436 * we were already there on the last unmount. Warn if this occurs.
1438 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1439 resblks = xfs_default_resblks(mp);
1440 error = xfs_reserve_blocks(mp, &resblks, NULL);
1441 if (error)
1442 xfs_warn(mp,
1443 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1446 return 0;
1448 out_rtunmount:
1449 xfs_rtunmount_inodes(mp);
1450 out_rele_rip:
1451 IRELE(rip);
1452 out_log_dealloc:
1453 xfs_log_unmount(mp);
1454 out_free_perag:
1455 xfs_free_perag(mp);
1456 out_remove_uuid:
1457 xfs_uuid_unmount(mp);
1458 out:
1459 return error;
1463 * This flushes out the inodes,dquots and the superblock, unmounts the
1464 * log and makes sure that incore structures are freed.
1466 void
1467 xfs_unmountfs(
1468 struct xfs_mount *mp)
1470 __uint64_t resblks;
1471 int error;
1473 xfs_qm_unmount_quotas(mp);
1474 xfs_rtunmount_inodes(mp);
1475 IRELE(mp->m_rootip);
1478 * We can potentially deadlock here if we have an inode cluster
1479 * that has been freed has its buffer still pinned in memory because
1480 * the transaction is still sitting in a iclog. The stale inodes
1481 * on that buffer will have their flush locks held until the
1482 * transaction hits the disk and the callbacks run. the inode
1483 * flush takes the flush lock unconditionally and with nothing to
1484 * push out the iclog we will never get that unlocked. hence we
1485 * need to force the log first.
1487 xfs_log_force(mp, XFS_LOG_SYNC);
1490 * Do a delwri reclaim pass first so that as many dirty inodes are
1491 * queued up for IO as possible. Then flush the buffers before making
1492 * a synchronous path to catch all the remaining inodes are reclaimed.
1493 * This makes the reclaim process as quick as possible by avoiding
1494 * synchronous writeout and blocking on inodes already in the delwri
1495 * state as much as possible.
1497 xfs_reclaim_inodes(mp, 0);
1498 XFS_bflush(mp->m_ddev_targp);
1499 xfs_reclaim_inodes(mp, SYNC_WAIT);
1501 xfs_qm_unmount(mp);
1504 * Flush out the log synchronously so that we know for sure
1505 * that nothing is pinned. This is important because bflush()
1506 * will skip pinned buffers.
1508 xfs_log_force(mp, XFS_LOG_SYNC);
1510 xfs_binval(mp->m_ddev_targp);
1511 if (mp->m_rtdev_targp) {
1512 xfs_binval(mp->m_rtdev_targp);
1516 * Unreserve any blocks we have so that when we unmount we don't account
1517 * the reserved free space as used. This is really only necessary for
1518 * lazy superblock counting because it trusts the incore superblock
1519 * counters to be absolutely correct on clean unmount.
1521 * We don't bother correcting this elsewhere for lazy superblock
1522 * counting because on mount of an unclean filesystem we reconstruct the
1523 * correct counter value and this is irrelevant.
1525 * For non-lazy counter filesystems, this doesn't matter at all because
1526 * we only every apply deltas to the superblock and hence the incore
1527 * value does not matter....
1529 resblks = 0;
1530 error = xfs_reserve_blocks(mp, &resblks, NULL);
1531 if (error)
1532 xfs_warn(mp, "Unable to free reserved block pool. "
1533 "Freespace may not be correct on next mount.");
1535 error = xfs_log_sbcount(mp, 1);
1536 if (error)
1537 xfs_warn(mp, "Unable to update superblock counters. "
1538 "Freespace may not be correct on next mount.");
1539 xfs_unmountfs_writesb(mp);
1540 xfs_unmountfs_wait(mp); /* wait for async bufs */
1541 xfs_log_unmount_write(mp);
1542 xfs_log_unmount(mp);
1543 xfs_uuid_unmount(mp);
1545 #if defined(DEBUG)
1546 xfs_errortag_clearall(mp, 0);
1547 #endif
1548 xfs_free_perag(mp);
1551 STATIC void
1552 xfs_unmountfs_wait(xfs_mount_t *mp)
1554 if (mp->m_logdev_targp != mp->m_ddev_targp)
1555 xfs_wait_buftarg(mp->m_logdev_targp);
1556 if (mp->m_rtdev_targp)
1557 xfs_wait_buftarg(mp->m_rtdev_targp);
1558 xfs_wait_buftarg(mp->m_ddev_targp);
1562 xfs_fs_writable(xfs_mount_t *mp)
1564 return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1565 (mp->m_flags & XFS_MOUNT_RDONLY));
1569 * xfs_log_sbcount
1571 * Called either periodically to keep the on disk superblock values
1572 * roughly up to date or from unmount to make sure the values are
1573 * correct on a clean unmount.
1575 * Note this code can be called during the process of freezing, so
1576 * we may need to use the transaction allocator which does not not
1577 * block when the transaction subsystem is in its frozen state.
1580 xfs_log_sbcount(
1581 xfs_mount_t *mp,
1582 uint sync)
1584 xfs_trans_t *tp;
1585 int error;
1587 if (!xfs_fs_writable(mp))
1588 return 0;
1590 xfs_icsb_sync_counters(mp, 0);
1593 * we don't need to do this if we are updating the superblock
1594 * counters on every modification.
1596 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1597 return 0;
1599 tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1600 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1601 XFS_DEFAULT_LOG_COUNT);
1602 if (error) {
1603 xfs_trans_cancel(tp, 0);
1604 return error;
1607 xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1608 if (sync)
1609 xfs_trans_set_sync(tp);
1610 error = xfs_trans_commit(tp, 0);
1611 return error;
1615 xfs_unmountfs_writesb(xfs_mount_t *mp)
1617 xfs_buf_t *sbp;
1618 int error = 0;
1621 * skip superblock write if fs is read-only, or
1622 * if we are doing a forced umount.
1624 if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1625 XFS_FORCED_SHUTDOWN(mp))) {
1627 sbp = xfs_getsb(mp, 0);
1629 XFS_BUF_UNDONE(sbp);
1630 XFS_BUF_UNREAD(sbp);
1631 XFS_BUF_UNDELAYWRITE(sbp);
1632 XFS_BUF_WRITE(sbp);
1633 XFS_BUF_UNASYNC(sbp);
1634 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1635 xfsbdstrat(mp, sbp);
1636 error = xfs_buf_iowait(sbp);
1637 if (error)
1638 xfs_ioerror_alert("xfs_unmountfs_writesb",
1639 mp, sbp, XFS_BUF_ADDR(sbp));
1640 xfs_buf_relse(sbp);
1642 return error;
1646 * xfs_mod_sb() can be used to copy arbitrary changes to the
1647 * in-core superblock into the superblock buffer to be logged.
1648 * It does not provide the higher level of locking that is
1649 * needed to protect the in-core superblock from concurrent
1650 * access.
1652 void
1653 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1655 xfs_buf_t *bp;
1656 int first;
1657 int last;
1658 xfs_mount_t *mp;
1659 xfs_sb_field_t f;
1661 ASSERT(fields);
1662 if (!fields)
1663 return;
1664 mp = tp->t_mountp;
1665 bp = xfs_trans_getsb(tp, mp, 0);
1666 first = sizeof(xfs_sb_t);
1667 last = 0;
1669 /* translate/copy */
1671 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1673 /* find modified range */
1674 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1675 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1676 last = xfs_sb_info[f + 1].offset - 1;
1678 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1679 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1680 first = xfs_sb_info[f].offset;
1682 xfs_trans_log_buf(tp, bp, first, last);
1687 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1688 * a delta to a specified field in the in-core superblock. Simply
1689 * switch on the field indicated and apply the delta to that field.
1690 * Fields are not allowed to dip below zero, so if the delta would
1691 * do this do not apply it and return EINVAL.
1693 * The m_sb_lock must be held when this routine is called.
1695 STATIC int
1696 xfs_mod_incore_sb_unlocked(
1697 xfs_mount_t *mp,
1698 xfs_sb_field_t field,
1699 int64_t delta,
1700 int rsvd)
1702 int scounter; /* short counter for 32 bit fields */
1703 long long lcounter; /* long counter for 64 bit fields */
1704 long long res_used, rem;
1707 * With the in-core superblock spin lock held, switch
1708 * on the indicated field. Apply the delta to the
1709 * proper field. If the fields value would dip below
1710 * 0, then do not apply the delta and return EINVAL.
1712 switch (field) {
1713 case XFS_SBS_ICOUNT:
1714 lcounter = (long long)mp->m_sb.sb_icount;
1715 lcounter += delta;
1716 if (lcounter < 0) {
1717 ASSERT(0);
1718 return XFS_ERROR(EINVAL);
1720 mp->m_sb.sb_icount = lcounter;
1721 return 0;
1722 case XFS_SBS_IFREE:
1723 lcounter = (long long)mp->m_sb.sb_ifree;
1724 lcounter += delta;
1725 if (lcounter < 0) {
1726 ASSERT(0);
1727 return XFS_ERROR(EINVAL);
1729 mp->m_sb.sb_ifree = lcounter;
1730 return 0;
1731 case XFS_SBS_FDBLOCKS:
1732 lcounter = (long long)
1733 mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1734 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1736 if (delta > 0) { /* Putting blocks back */
1737 if (res_used > delta) {
1738 mp->m_resblks_avail += delta;
1739 } else {
1740 rem = delta - res_used;
1741 mp->m_resblks_avail = mp->m_resblks;
1742 lcounter += rem;
1744 } else { /* Taking blocks away */
1745 lcounter += delta;
1746 if (lcounter >= 0) {
1747 mp->m_sb.sb_fdblocks = lcounter +
1748 XFS_ALLOC_SET_ASIDE(mp);
1749 return 0;
1753 * We are out of blocks, use any available reserved
1754 * blocks if were allowed to.
1756 if (!rsvd)
1757 return XFS_ERROR(ENOSPC);
1759 lcounter = (long long)mp->m_resblks_avail + delta;
1760 if (lcounter >= 0) {
1761 mp->m_resblks_avail = lcounter;
1762 return 0;
1764 printk_once(KERN_WARNING
1765 "Filesystem \"%s\": reserve blocks depleted! "
1766 "Consider increasing reserve pool size.",
1767 mp->m_fsname);
1768 return XFS_ERROR(ENOSPC);
1771 mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1772 return 0;
1773 case XFS_SBS_FREXTENTS:
1774 lcounter = (long long)mp->m_sb.sb_frextents;
1775 lcounter += delta;
1776 if (lcounter < 0) {
1777 return XFS_ERROR(ENOSPC);
1779 mp->m_sb.sb_frextents = lcounter;
1780 return 0;
1781 case XFS_SBS_DBLOCKS:
1782 lcounter = (long long)mp->m_sb.sb_dblocks;
1783 lcounter += delta;
1784 if (lcounter < 0) {
1785 ASSERT(0);
1786 return XFS_ERROR(EINVAL);
1788 mp->m_sb.sb_dblocks = lcounter;
1789 return 0;
1790 case XFS_SBS_AGCOUNT:
1791 scounter = mp->m_sb.sb_agcount;
1792 scounter += delta;
1793 if (scounter < 0) {
1794 ASSERT(0);
1795 return XFS_ERROR(EINVAL);
1797 mp->m_sb.sb_agcount = scounter;
1798 return 0;
1799 case XFS_SBS_IMAX_PCT:
1800 scounter = mp->m_sb.sb_imax_pct;
1801 scounter += delta;
1802 if (scounter < 0) {
1803 ASSERT(0);
1804 return XFS_ERROR(EINVAL);
1806 mp->m_sb.sb_imax_pct = scounter;
1807 return 0;
1808 case XFS_SBS_REXTSIZE:
1809 scounter = mp->m_sb.sb_rextsize;
1810 scounter += delta;
1811 if (scounter < 0) {
1812 ASSERT(0);
1813 return XFS_ERROR(EINVAL);
1815 mp->m_sb.sb_rextsize = scounter;
1816 return 0;
1817 case XFS_SBS_RBMBLOCKS:
1818 scounter = mp->m_sb.sb_rbmblocks;
1819 scounter += delta;
1820 if (scounter < 0) {
1821 ASSERT(0);
1822 return XFS_ERROR(EINVAL);
1824 mp->m_sb.sb_rbmblocks = scounter;
1825 return 0;
1826 case XFS_SBS_RBLOCKS:
1827 lcounter = (long long)mp->m_sb.sb_rblocks;
1828 lcounter += delta;
1829 if (lcounter < 0) {
1830 ASSERT(0);
1831 return XFS_ERROR(EINVAL);
1833 mp->m_sb.sb_rblocks = lcounter;
1834 return 0;
1835 case XFS_SBS_REXTENTS:
1836 lcounter = (long long)mp->m_sb.sb_rextents;
1837 lcounter += delta;
1838 if (lcounter < 0) {
1839 ASSERT(0);
1840 return XFS_ERROR(EINVAL);
1842 mp->m_sb.sb_rextents = lcounter;
1843 return 0;
1844 case XFS_SBS_REXTSLOG:
1845 scounter = mp->m_sb.sb_rextslog;
1846 scounter += delta;
1847 if (scounter < 0) {
1848 ASSERT(0);
1849 return XFS_ERROR(EINVAL);
1851 mp->m_sb.sb_rextslog = scounter;
1852 return 0;
1853 default:
1854 ASSERT(0);
1855 return XFS_ERROR(EINVAL);
1860 * xfs_mod_incore_sb() is used to change a field in the in-core
1861 * superblock structure by the specified delta. This modification
1862 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1863 * routine to do the work.
1866 xfs_mod_incore_sb(
1867 struct xfs_mount *mp,
1868 xfs_sb_field_t field,
1869 int64_t delta,
1870 int rsvd)
1872 int status;
1874 #ifdef HAVE_PERCPU_SB
1875 ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1876 #endif
1877 spin_lock(&mp->m_sb_lock);
1878 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1879 spin_unlock(&mp->m_sb_lock);
1881 return status;
1885 * Change more than one field in the in-core superblock structure at a time.
1887 * The fields and changes to those fields are specified in the array of
1888 * xfs_mod_sb structures passed in. Either all of the specified deltas
1889 * will be applied or none of them will. If any modified field dips below 0,
1890 * then all modifications will be backed out and EINVAL will be returned.
1892 * Note that this function may not be used for the superblock values that
1893 * are tracked with the in-memory per-cpu counters - a direct call to
1894 * xfs_icsb_modify_counters is required for these.
1897 xfs_mod_incore_sb_batch(
1898 struct xfs_mount *mp,
1899 xfs_mod_sb_t *msb,
1900 uint nmsb,
1901 int rsvd)
1903 xfs_mod_sb_t *msbp;
1904 int error = 0;
1907 * Loop through the array of mod structures and apply each individually.
1908 * If any fail, then back out all those which have already been applied.
1909 * Do all of this within the scope of the m_sb_lock so that all of the
1910 * changes will be atomic.
1912 spin_lock(&mp->m_sb_lock);
1913 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1914 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1915 msbp->msb_field > XFS_SBS_FDBLOCKS);
1917 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1918 msbp->msb_delta, rsvd);
1919 if (error)
1920 goto unwind;
1922 spin_unlock(&mp->m_sb_lock);
1923 return 0;
1925 unwind:
1926 while (--msbp >= msb) {
1927 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1928 -msbp->msb_delta, rsvd);
1929 ASSERT(error == 0);
1931 spin_unlock(&mp->m_sb_lock);
1932 return error;
1936 * xfs_getsb() is called to obtain the buffer for the superblock.
1937 * The buffer is returned locked and read in from disk.
1938 * The buffer should be released with a call to xfs_brelse().
1940 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1941 * the superblock buffer if it can be locked without sleeping.
1942 * If it can't then we'll return NULL.
1944 xfs_buf_t *
1945 xfs_getsb(
1946 xfs_mount_t *mp,
1947 int flags)
1949 xfs_buf_t *bp;
1951 ASSERT(mp->m_sb_bp != NULL);
1952 bp = mp->m_sb_bp;
1953 if (flags & XBF_TRYLOCK) {
1954 if (!XFS_BUF_CPSEMA(bp)) {
1955 return NULL;
1957 } else {
1958 XFS_BUF_PSEMA(bp, PRIBIO);
1960 XFS_BUF_HOLD(bp);
1961 ASSERT(XFS_BUF_ISDONE(bp));
1962 return bp;
1966 * Used to free the superblock along various error paths.
1968 void
1969 xfs_freesb(
1970 struct xfs_mount *mp)
1972 struct xfs_buf *bp = mp->m_sb_bp;
1974 xfs_buf_lock(bp);
1975 mp->m_sb_bp = NULL;
1976 xfs_buf_relse(bp);
1980 * Used to log changes to the superblock unit and width fields which could
1981 * be altered by the mount options, as well as any potential sb_features2
1982 * fixup. Only the first superblock is updated.
1985 xfs_mount_log_sb(
1986 xfs_mount_t *mp,
1987 __int64_t fields)
1989 xfs_trans_t *tp;
1990 int error;
1992 ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1993 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1994 XFS_SB_VERSIONNUM));
1996 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1997 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1998 XFS_DEFAULT_LOG_COUNT);
1999 if (error) {
2000 xfs_trans_cancel(tp, 0);
2001 return error;
2003 xfs_mod_sb(tp, fields);
2004 error = xfs_trans_commit(tp, 0);
2005 return error;
2009 * If the underlying (data/log/rt) device is readonly, there are some
2010 * operations that cannot proceed.
2013 xfs_dev_is_read_only(
2014 struct xfs_mount *mp,
2015 char *message)
2017 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
2018 xfs_readonly_buftarg(mp->m_logdev_targp) ||
2019 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
2020 xfs_notice(mp, "%s required on read-only device.", message);
2021 xfs_notice(mp, "write access unavailable, cannot proceed.");
2022 return EROFS;
2024 return 0;
2027 #ifdef HAVE_PERCPU_SB
2029 * Per-cpu incore superblock counters
2031 * Simple concept, difficult implementation
2033 * Basically, replace the incore superblock counters with a distributed per cpu
2034 * counter for contended fields (e.g. free block count).
2036 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2037 * hence needs to be accurately read when we are running low on space. Hence
2038 * there is a method to enable and disable the per-cpu counters based on how
2039 * much "stuff" is available in them.
2041 * Basically, a counter is enabled if there is enough free resource to justify
2042 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2043 * ENOSPC), then we disable the counters to synchronise all callers and
2044 * re-distribute the available resources.
2046 * If, once we redistributed the available resources, we still get a failure,
2047 * we disable the per-cpu counter and go through the slow path.
2049 * The slow path is the current xfs_mod_incore_sb() function. This means that
2050 * when we disable a per-cpu counter, we need to drain its resources back to
2051 * the global superblock. We do this after disabling the counter to prevent
2052 * more threads from queueing up on the counter.
2054 * Essentially, this means that we still need a lock in the fast path to enable
2055 * synchronisation between the global counters and the per-cpu counters. This
2056 * is not a problem because the lock will be local to a CPU almost all the time
2057 * and have little contention except when we get to ENOSPC conditions.
2059 * Basically, this lock becomes a barrier that enables us to lock out the fast
2060 * path while we do things like enabling and disabling counters and
2061 * synchronising the counters.
2063 * Locking rules:
2065 * 1. m_sb_lock before picking up per-cpu locks
2066 * 2. per-cpu locks always picked up via for_each_online_cpu() order
2067 * 3. accurate counter sync requires m_sb_lock + per cpu locks
2068 * 4. modifying per-cpu counters requires holding per-cpu lock
2069 * 5. modifying global counters requires holding m_sb_lock
2070 * 6. enabling or disabling a counter requires holding the m_sb_lock
2071 * and _none_ of the per-cpu locks.
2073 * Disabled counters are only ever re-enabled by a balance operation
2074 * that results in more free resources per CPU than a given threshold.
2075 * To ensure counters don't remain disabled, they are rebalanced when
2076 * the global resource goes above a higher threshold (i.e. some hysteresis
2077 * is present to prevent thrashing).
2080 #ifdef CONFIG_HOTPLUG_CPU
2082 * hot-plug CPU notifier support.
2084 * We need a notifier per filesystem as we need to be able to identify
2085 * the filesystem to balance the counters out. This is achieved by
2086 * having a notifier block embedded in the xfs_mount_t and doing pointer
2087 * magic to get the mount pointer from the notifier block address.
2089 STATIC int
2090 xfs_icsb_cpu_notify(
2091 struct notifier_block *nfb,
2092 unsigned long action,
2093 void *hcpu)
2095 xfs_icsb_cnts_t *cntp;
2096 xfs_mount_t *mp;
2098 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2099 cntp = (xfs_icsb_cnts_t *)
2100 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2101 switch (action) {
2102 case CPU_UP_PREPARE:
2103 case CPU_UP_PREPARE_FROZEN:
2104 /* Easy Case - initialize the area and locks, and
2105 * then rebalance when online does everything else for us. */
2106 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2107 break;
2108 case CPU_ONLINE:
2109 case CPU_ONLINE_FROZEN:
2110 xfs_icsb_lock(mp);
2111 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2112 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2113 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2114 xfs_icsb_unlock(mp);
2115 break;
2116 case CPU_DEAD:
2117 case CPU_DEAD_FROZEN:
2118 /* Disable all the counters, then fold the dead cpu's
2119 * count into the total on the global superblock and
2120 * re-enable the counters. */
2121 xfs_icsb_lock(mp);
2122 spin_lock(&mp->m_sb_lock);
2123 xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2124 xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2125 xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2127 mp->m_sb.sb_icount += cntp->icsb_icount;
2128 mp->m_sb.sb_ifree += cntp->icsb_ifree;
2129 mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2131 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2133 xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2134 xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2135 xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2136 spin_unlock(&mp->m_sb_lock);
2137 xfs_icsb_unlock(mp);
2138 break;
2141 return NOTIFY_OK;
2143 #endif /* CONFIG_HOTPLUG_CPU */
2146 xfs_icsb_init_counters(
2147 xfs_mount_t *mp)
2149 xfs_icsb_cnts_t *cntp;
2150 int i;
2152 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2153 if (mp->m_sb_cnts == NULL)
2154 return -ENOMEM;
2156 #ifdef CONFIG_HOTPLUG_CPU
2157 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2158 mp->m_icsb_notifier.priority = 0;
2159 register_hotcpu_notifier(&mp->m_icsb_notifier);
2160 #endif /* CONFIG_HOTPLUG_CPU */
2162 for_each_online_cpu(i) {
2163 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2164 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2167 mutex_init(&mp->m_icsb_mutex);
2170 * start with all counters disabled so that the
2171 * initial balance kicks us off correctly
2173 mp->m_icsb_counters = -1;
2174 return 0;
2177 void
2178 xfs_icsb_reinit_counters(
2179 xfs_mount_t *mp)
2181 xfs_icsb_lock(mp);
2183 * start with all counters disabled so that the
2184 * initial balance kicks us off correctly
2186 mp->m_icsb_counters = -1;
2187 xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2188 xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2189 xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2190 xfs_icsb_unlock(mp);
2193 void
2194 xfs_icsb_destroy_counters(
2195 xfs_mount_t *mp)
2197 if (mp->m_sb_cnts) {
2198 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2199 free_percpu(mp->m_sb_cnts);
2201 mutex_destroy(&mp->m_icsb_mutex);
2204 STATIC void
2205 xfs_icsb_lock_cntr(
2206 xfs_icsb_cnts_t *icsbp)
2208 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2209 ndelay(1000);
2213 STATIC void
2214 xfs_icsb_unlock_cntr(
2215 xfs_icsb_cnts_t *icsbp)
2217 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2221 STATIC void
2222 xfs_icsb_lock_all_counters(
2223 xfs_mount_t *mp)
2225 xfs_icsb_cnts_t *cntp;
2226 int i;
2228 for_each_online_cpu(i) {
2229 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2230 xfs_icsb_lock_cntr(cntp);
2234 STATIC void
2235 xfs_icsb_unlock_all_counters(
2236 xfs_mount_t *mp)
2238 xfs_icsb_cnts_t *cntp;
2239 int i;
2241 for_each_online_cpu(i) {
2242 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2243 xfs_icsb_unlock_cntr(cntp);
2247 STATIC void
2248 xfs_icsb_count(
2249 xfs_mount_t *mp,
2250 xfs_icsb_cnts_t *cnt,
2251 int flags)
2253 xfs_icsb_cnts_t *cntp;
2254 int i;
2256 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2258 if (!(flags & XFS_ICSB_LAZY_COUNT))
2259 xfs_icsb_lock_all_counters(mp);
2261 for_each_online_cpu(i) {
2262 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2263 cnt->icsb_icount += cntp->icsb_icount;
2264 cnt->icsb_ifree += cntp->icsb_ifree;
2265 cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2268 if (!(flags & XFS_ICSB_LAZY_COUNT))
2269 xfs_icsb_unlock_all_counters(mp);
2272 STATIC int
2273 xfs_icsb_counter_disabled(
2274 xfs_mount_t *mp,
2275 xfs_sb_field_t field)
2277 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2278 return test_bit(field, &mp->m_icsb_counters);
2281 STATIC void
2282 xfs_icsb_disable_counter(
2283 xfs_mount_t *mp,
2284 xfs_sb_field_t field)
2286 xfs_icsb_cnts_t cnt;
2288 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2291 * If we are already disabled, then there is nothing to do
2292 * here. We check before locking all the counters to avoid
2293 * the expensive lock operation when being called in the
2294 * slow path and the counter is already disabled. This is
2295 * safe because the only time we set or clear this state is under
2296 * the m_icsb_mutex.
2298 if (xfs_icsb_counter_disabled(mp, field))
2299 return;
2301 xfs_icsb_lock_all_counters(mp);
2302 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2303 /* drain back to superblock */
2305 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2306 switch(field) {
2307 case XFS_SBS_ICOUNT:
2308 mp->m_sb.sb_icount = cnt.icsb_icount;
2309 break;
2310 case XFS_SBS_IFREE:
2311 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2312 break;
2313 case XFS_SBS_FDBLOCKS:
2314 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2315 break;
2316 default:
2317 BUG();
2321 xfs_icsb_unlock_all_counters(mp);
2324 STATIC void
2325 xfs_icsb_enable_counter(
2326 xfs_mount_t *mp,
2327 xfs_sb_field_t field,
2328 uint64_t count,
2329 uint64_t resid)
2331 xfs_icsb_cnts_t *cntp;
2332 int i;
2334 ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2336 xfs_icsb_lock_all_counters(mp);
2337 for_each_online_cpu(i) {
2338 cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2339 switch (field) {
2340 case XFS_SBS_ICOUNT:
2341 cntp->icsb_icount = count + resid;
2342 break;
2343 case XFS_SBS_IFREE:
2344 cntp->icsb_ifree = count + resid;
2345 break;
2346 case XFS_SBS_FDBLOCKS:
2347 cntp->icsb_fdblocks = count + resid;
2348 break;
2349 default:
2350 BUG();
2351 break;
2353 resid = 0;
2355 clear_bit(field, &mp->m_icsb_counters);
2356 xfs_icsb_unlock_all_counters(mp);
2359 void
2360 xfs_icsb_sync_counters_locked(
2361 xfs_mount_t *mp,
2362 int flags)
2364 xfs_icsb_cnts_t cnt;
2366 xfs_icsb_count(mp, &cnt, flags);
2368 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2369 mp->m_sb.sb_icount = cnt.icsb_icount;
2370 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2371 mp->m_sb.sb_ifree = cnt.icsb_ifree;
2372 if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2373 mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2377 * Accurate update of per-cpu counters to incore superblock
2379 void
2380 xfs_icsb_sync_counters(
2381 xfs_mount_t *mp,
2382 int flags)
2384 spin_lock(&mp->m_sb_lock);
2385 xfs_icsb_sync_counters_locked(mp, flags);
2386 spin_unlock(&mp->m_sb_lock);
2390 * Balance and enable/disable counters as necessary.
2392 * Thresholds for re-enabling counters are somewhat magic. inode counts are
2393 * chosen to be the same number as single on disk allocation chunk per CPU, and
2394 * free blocks is something far enough zero that we aren't going thrash when we
2395 * get near ENOSPC. We also need to supply a minimum we require per cpu to
2396 * prevent looping endlessly when xfs_alloc_space asks for more than will
2397 * be distributed to a single CPU but each CPU has enough blocks to be
2398 * reenabled.
2400 * Note that we can be called when counters are already disabled.
2401 * xfs_icsb_disable_counter() optimises the counter locking in this case to
2402 * prevent locking every per-cpu counter needlessly.
2405 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
2406 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2407 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2408 STATIC void
2409 xfs_icsb_balance_counter_locked(
2410 xfs_mount_t *mp,
2411 xfs_sb_field_t field,
2412 int min_per_cpu)
2414 uint64_t count, resid;
2415 int weight = num_online_cpus();
2416 uint64_t min = (uint64_t)min_per_cpu;
2418 /* disable counter and sync counter */
2419 xfs_icsb_disable_counter(mp, field);
2421 /* update counters - first CPU gets residual*/
2422 switch (field) {
2423 case XFS_SBS_ICOUNT:
2424 count = mp->m_sb.sb_icount;
2425 resid = do_div(count, weight);
2426 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2427 return;
2428 break;
2429 case XFS_SBS_IFREE:
2430 count = mp->m_sb.sb_ifree;
2431 resid = do_div(count, weight);
2432 if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2433 return;
2434 break;
2435 case XFS_SBS_FDBLOCKS:
2436 count = mp->m_sb.sb_fdblocks;
2437 resid = do_div(count, weight);
2438 if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2439 return;
2440 break;
2441 default:
2442 BUG();
2443 count = resid = 0; /* quiet, gcc */
2444 break;
2447 xfs_icsb_enable_counter(mp, field, count, resid);
2450 STATIC void
2451 xfs_icsb_balance_counter(
2452 xfs_mount_t *mp,
2453 xfs_sb_field_t fields,
2454 int min_per_cpu)
2456 spin_lock(&mp->m_sb_lock);
2457 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2458 spin_unlock(&mp->m_sb_lock);
2462 xfs_icsb_modify_counters(
2463 xfs_mount_t *mp,
2464 xfs_sb_field_t field,
2465 int64_t delta,
2466 int rsvd)
2468 xfs_icsb_cnts_t *icsbp;
2469 long long lcounter; /* long counter for 64 bit fields */
2470 int ret = 0;
2472 might_sleep();
2473 again:
2474 preempt_disable();
2475 icsbp = this_cpu_ptr(mp->m_sb_cnts);
2478 * if the counter is disabled, go to slow path
2480 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2481 goto slow_path;
2482 xfs_icsb_lock_cntr(icsbp);
2483 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2484 xfs_icsb_unlock_cntr(icsbp);
2485 goto slow_path;
2488 switch (field) {
2489 case XFS_SBS_ICOUNT:
2490 lcounter = icsbp->icsb_icount;
2491 lcounter += delta;
2492 if (unlikely(lcounter < 0))
2493 goto balance_counter;
2494 icsbp->icsb_icount = lcounter;
2495 break;
2497 case XFS_SBS_IFREE:
2498 lcounter = icsbp->icsb_ifree;
2499 lcounter += delta;
2500 if (unlikely(lcounter < 0))
2501 goto balance_counter;
2502 icsbp->icsb_ifree = lcounter;
2503 break;
2505 case XFS_SBS_FDBLOCKS:
2506 BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2508 lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2509 lcounter += delta;
2510 if (unlikely(lcounter < 0))
2511 goto balance_counter;
2512 icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2513 break;
2514 default:
2515 BUG();
2516 break;
2518 xfs_icsb_unlock_cntr(icsbp);
2519 preempt_enable();
2520 return 0;
2522 slow_path:
2523 preempt_enable();
2526 * serialise with a mutex so we don't burn lots of cpu on
2527 * the superblock lock. We still need to hold the superblock
2528 * lock, however, when we modify the global structures.
2530 xfs_icsb_lock(mp);
2533 * Now running atomically.
2535 * If the counter is enabled, someone has beaten us to rebalancing.
2536 * Drop the lock and try again in the fast path....
2538 if (!(xfs_icsb_counter_disabled(mp, field))) {
2539 xfs_icsb_unlock(mp);
2540 goto again;
2544 * The counter is currently disabled. Because we are
2545 * running atomically here, we know a rebalance cannot
2546 * be in progress. Hence we can go straight to operating
2547 * on the global superblock. We do not call xfs_mod_incore_sb()
2548 * here even though we need to get the m_sb_lock. Doing so
2549 * will cause us to re-enter this function and deadlock.
2550 * Hence we get the m_sb_lock ourselves and then call
2551 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2552 * directly on the global counters.
2554 spin_lock(&mp->m_sb_lock);
2555 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2556 spin_unlock(&mp->m_sb_lock);
2559 * Now that we've modified the global superblock, we
2560 * may be able to re-enable the distributed counters
2561 * (e.g. lots of space just got freed). After that
2562 * we are done.
2564 if (ret != ENOSPC)
2565 xfs_icsb_balance_counter(mp, field, 0);
2566 xfs_icsb_unlock(mp);
2567 return ret;
2569 balance_counter:
2570 xfs_icsb_unlock_cntr(icsbp);
2571 preempt_enable();
2574 * We may have multiple threads here if multiple per-cpu
2575 * counters run dry at the same time. This will mean we can
2576 * do more balances than strictly necessary but it is not
2577 * the common slowpath case.
2579 xfs_icsb_lock(mp);
2582 * running atomically.
2584 * This will leave the counter in the correct state for future
2585 * accesses. After the rebalance, we simply try again and our retry
2586 * will either succeed through the fast path or slow path without
2587 * another balance operation being required.
2589 xfs_icsb_balance_counter(mp, field, delta);
2590 xfs_icsb_unlock(mp);
2591 goto again;
2594 #endif