perf record: Add --sample-cpu option
[deliverable/linux.git] / fs / xfs / xfs_mount.c
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_dir2.h"
31 #include "xfs_ialloc.h"
32 #include "xfs_alloc.h"
33 #include "xfs_rtalloc.h"
34 #include "xfs_bmap.h"
35 #include "xfs_trans.h"
36 #include "xfs_trans_priv.h"
37 #include "xfs_log.h"
38 #include "xfs_error.h"
39 #include "xfs_quota.h"
40 #include "xfs_fsops.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
43 #include "xfs_sysfs.h"
44
45
46 static DEFINE_MUTEX(xfs_uuid_table_mutex);
47 static int xfs_uuid_table_size;
48 static uuid_t *xfs_uuid_table;
49
50 void
51 xfs_uuid_table_free(void)
52 {
53 if (xfs_uuid_table_size == 0)
54 return;
55 kmem_free(xfs_uuid_table);
56 xfs_uuid_table = NULL;
57 xfs_uuid_table_size = 0;
58 }
59
60 /*
61 * See if the UUID is unique among mounted XFS filesystems.
62 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
63 */
64 STATIC int
65 xfs_uuid_mount(
66 struct xfs_mount *mp)
67 {
68 uuid_t *uuid = &mp->m_sb.sb_uuid;
69 int hole, i;
70
71 if (mp->m_flags & XFS_MOUNT_NOUUID)
72 return 0;
73
74 if (uuid_is_nil(uuid)) {
75 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
76 return -EINVAL;
77 }
78
79 mutex_lock(&xfs_uuid_table_mutex);
80 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
81 if (uuid_is_nil(&xfs_uuid_table[i])) {
82 hole = i;
83 continue;
84 }
85 if (uuid_equal(uuid, &xfs_uuid_table[i]))
86 goto out_duplicate;
87 }
88
89 if (hole < 0) {
90 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
91 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
92 KM_SLEEP);
93 hole = xfs_uuid_table_size++;
94 }
95 xfs_uuid_table[hole] = *uuid;
96 mutex_unlock(&xfs_uuid_table_mutex);
97
98 return 0;
99
100 out_duplicate:
101 mutex_unlock(&xfs_uuid_table_mutex);
102 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
103 return -EINVAL;
104 }
105
106 STATIC void
107 xfs_uuid_unmount(
108 struct xfs_mount *mp)
109 {
110 uuid_t *uuid = &mp->m_sb.sb_uuid;
111 int i;
112
113 if (mp->m_flags & XFS_MOUNT_NOUUID)
114 return;
115
116 mutex_lock(&xfs_uuid_table_mutex);
117 for (i = 0; i < xfs_uuid_table_size; i++) {
118 if (uuid_is_nil(&xfs_uuid_table[i]))
119 continue;
120 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
121 continue;
122 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
123 break;
124 }
125 ASSERT(i < xfs_uuid_table_size);
126 mutex_unlock(&xfs_uuid_table_mutex);
127 }
128
129
130 STATIC void
131 __xfs_free_perag(
132 struct rcu_head *head)
133 {
134 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
135
136 ASSERT(atomic_read(&pag->pag_ref) == 0);
137 kmem_free(pag);
138 }
139
140 /*
141 * Free up the per-ag resources associated with the mount structure.
142 */
143 STATIC void
144 xfs_free_perag(
145 xfs_mount_t *mp)
146 {
147 xfs_agnumber_t agno;
148 struct xfs_perag *pag;
149
150 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
151 spin_lock(&mp->m_perag_lock);
152 pag = radix_tree_delete(&mp->m_perag_tree, agno);
153 spin_unlock(&mp->m_perag_lock);
154 ASSERT(pag);
155 ASSERT(atomic_read(&pag->pag_ref) == 0);
156 call_rcu(&pag->rcu_head, __xfs_free_perag);
157 }
158 }
159
160 /*
161 * Check size of device based on the (data/realtime) block count.
162 * Note: this check is used by the growfs code as well as mount.
163 */
164 int
165 xfs_sb_validate_fsb_count(
166 xfs_sb_t *sbp,
167 __uint64_t nblocks)
168 {
169 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
170 ASSERT(sbp->sb_blocklog >= BBSHIFT);
171
172 /* Limited by ULONG_MAX of page cache index */
173 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
174 return -EFBIG;
175 return 0;
176 }
177
178 int
179 xfs_initialize_perag(
180 xfs_mount_t *mp,
181 xfs_agnumber_t agcount,
182 xfs_agnumber_t *maxagi)
183 {
184 xfs_agnumber_t index;
185 xfs_agnumber_t first_initialised = 0;
186 xfs_perag_t *pag;
187 int error = -ENOMEM;
188
189 /*
190 * Walk the current per-ag tree so we don't try to initialise AGs
191 * that already exist (growfs case). Allocate and insert all the
192 * AGs we don't find ready for initialisation.
193 */
194 for (index = 0; index < agcount; index++) {
195 pag = xfs_perag_get(mp, index);
196 if (pag) {
197 xfs_perag_put(pag);
198 continue;
199 }
200 if (!first_initialised)
201 first_initialised = index;
202
203 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
204 if (!pag)
205 goto out_unwind;
206 pag->pag_agno = index;
207 pag->pag_mount = mp;
208 spin_lock_init(&pag->pag_ici_lock);
209 mutex_init(&pag->pag_ici_reclaim_lock);
210 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
211 spin_lock_init(&pag->pag_buf_lock);
212 pag->pag_buf_tree = RB_ROOT;
213
214 if (radix_tree_preload(GFP_NOFS))
215 goto out_unwind;
216
217 spin_lock(&mp->m_perag_lock);
218 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
219 BUG();
220 spin_unlock(&mp->m_perag_lock);
221 radix_tree_preload_end();
222 error = -EEXIST;
223 goto out_unwind;
224 }
225 spin_unlock(&mp->m_perag_lock);
226 radix_tree_preload_end();
227 }
228
229 index = xfs_set_inode_alloc(mp, agcount);
230
231 if (maxagi)
232 *maxagi = index;
233 return 0;
234
235 out_unwind:
236 kmem_free(pag);
237 for (; index > first_initialised; index--) {
238 pag = radix_tree_delete(&mp->m_perag_tree, index);
239 kmem_free(pag);
240 }
241 return error;
242 }
243
244 /*
245 * xfs_readsb
246 *
247 * Does the initial read of the superblock.
248 */
249 int
250 xfs_readsb(
251 struct xfs_mount *mp,
252 int flags)
253 {
254 unsigned int sector_size;
255 struct xfs_buf *bp;
256 struct xfs_sb *sbp = &mp->m_sb;
257 int error;
258 int loud = !(flags & XFS_MFSI_QUIET);
259 const struct xfs_buf_ops *buf_ops;
260
261 ASSERT(mp->m_sb_bp == NULL);
262 ASSERT(mp->m_ddev_targp != NULL);
263
264 /*
265 * For the initial read, we must guess at the sector
266 * size based on the block device. It's enough to
267 * get the sb_sectsize out of the superblock and
268 * then reread with the proper length.
269 * We don't verify it yet, because it may not be complete.
270 */
271 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
272 buf_ops = NULL;
273
274 /*
275 * Allocate a (locked) buffer to hold the superblock.
276 * This will be kept around at all times to optimize
277 * access to the superblock.
278 */
279 reread:
280 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
281 BTOBB(sector_size), 0, &bp, buf_ops);
282 if (error) {
283 if (loud)
284 xfs_warn(mp, "SB validate failed with error %d.", error);
285 /* bad CRC means corrupted metadata */
286 if (error == -EFSBADCRC)
287 error = -EFSCORRUPTED;
288 return error;
289 }
290
291 /*
292 * Initialize the mount structure from the superblock.
293 */
294 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
295
296 /*
297 * If we haven't validated the superblock, do so now before we try
298 * to check the sector size and reread the superblock appropriately.
299 */
300 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
301 if (loud)
302 xfs_warn(mp, "Invalid superblock magic number");
303 error = -EINVAL;
304 goto release_buf;
305 }
306
307 /*
308 * We must be able to do sector-sized and sector-aligned IO.
309 */
310 if (sector_size > sbp->sb_sectsize) {
311 if (loud)
312 xfs_warn(mp, "device supports %u byte sectors (not %u)",
313 sector_size, sbp->sb_sectsize);
314 error = -ENOSYS;
315 goto release_buf;
316 }
317
318 if (buf_ops == NULL) {
319 /*
320 * Re-read the superblock so the buffer is correctly sized,
321 * and properly verified.
322 */
323 xfs_buf_relse(bp);
324 sector_size = sbp->sb_sectsize;
325 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
326 goto reread;
327 }
328
329 xfs_reinit_percpu_counters(mp);
330
331 /* no need to be quiet anymore, so reset the buf ops */
332 bp->b_ops = &xfs_sb_buf_ops;
333
334 mp->m_sb_bp = bp;
335 xfs_buf_unlock(bp);
336 return 0;
337
338 release_buf:
339 xfs_buf_relse(bp);
340 return error;
341 }
342
343 /*
344 * Update alignment values based on mount options and sb values
345 */
346 STATIC int
347 xfs_update_alignment(xfs_mount_t *mp)
348 {
349 xfs_sb_t *sbp = &(mp->m_sb);
350
351 if (mp->m_dalign) {
352 /*
353 * If stripe unit and stripe width are not multiples
354 * of the fs blocksize turn off alignment.
355 */
356 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
357 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
358 xfs_warn(mp,
359 "alignment check failed: sunit/swidth vs. blocksize(%d)",
360 sbp->sb_blocksize);
361 return -EINVAL;
362 } else {
363 /*
364 * Convert the stripe unit and width to FSBs.
365 */
366 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
367 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
368 xfs_warn(mp,
369 "alignment check failed: sunit/swidth vs. agsize(%d)",
370 sbp->sb_agblocks);
371 return -EINVAL;
372 } else if (mp->m_dalign) {
373 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
374 } else {
375 xfs_warn(mp,
376 "alignment check failed: sunit(%d) less than bsize(%d)",
377 mp->m_dalign, sbp->sb_blocksize);
378 return -EINVAL;
379 }
380 }
381
382 /*
383 * Update superblock with new values
384 * and log changes
385 */
386 if (xfs_sb_version_hasdalign(sbp)) {
387 if (sbp->sb_unit != mp->m_dalign) {
388 sbp->sb_unit = mp->m_dalign;
389 mp->m_update_sb = true;
390 }
391 if (sbp->sb_width != mp->m_swidth) {
392 sbp->sb_width = mp->m_swidth;
393 mp->m_update_sb = true;
394 }
395 } else {
396 xfs_warn(mp,
397 "cannot change alignment: superblock does not support data alignment");
398 return -EINVAL;
399 }
400 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
401 xfs_sb_version_hasdalign(&mp->m_sb)) {
402 mp->m_dalign = sbp->sb_unit;
403 mp->m_swidth = sbp->sb_width;
404 }
405
406 return 0;
407 }
408
409 /*
410 * Set the maximum inode count for this filesystem
411 */
412 STATIC void
413 xfs_set_maxicount(xfs_mount_t *mp)
414 {
415 xfs_sb_t *sbp = &(mp->m_sb);
416 __uint64_t icount;
417
418 if (sbp->sb_imax_pct) {
419 /*
420 * Make sure the maximum inode count is a multiple
421 * of the units we allocate inodes in.
422 */
423 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
424 do_div(icount, 100);
425 do_div(icount, mp->m_ialloc_blks);
426 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
427 sbp->sb_inopblog;
428 } else {
429 mp->m_maxicount = 0;
430 }
431 }
432
433 /*
434 * Set the default minimum read and write sizes unless
435 * already specified in a mount option.
436 * We use smaller I/O sizes when the file system
437 * is being used for NFS service (wsync mount option).
438 */
439 STATIC void
440 xfs_set_rw_sizes(xfs_mount_t *mp)
441 {
442 xfs_sb_t *sbp = &(mp->m_sb);
443 int readio_log, writeio_log;
444
445 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
446 if (mp->m_flags & XFS_MOUNT_WSYNC) {
447 readio_log = XFS_WSYNC_READIO_LOG;
448 writeio_log = XFS_WSYNC_WRITEIO_LOG;
449 } else {
450 readio_log = XFS_READIO_LOG_LARGE;
451 writeio_log = XFS_WRITEIO_LOG_LARGE;
452 }
453 } else {
454 readio_log = mp->m_readio_log;
455 writeio_log = mp->m_writeio_log;
456 }
457
458 if (sbp->sb_blocklog > readio_log) {
459 mp->m_readio_log = sbp->sb_blocklog;
460 } else {
461 mp->m_readio_log = readio_log;
462 }
463 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
464 if (sbp->sb_blocklog > writeio_log) {
465 mp->m_writeio_log = sbp->sb_blocklog;
466 } else {
467 mp->m_writeio_log = writeio_log;
468 }
469 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
470 }
471
472 /*
473 * precalculate the low space thresholds for dynamic speculative preallocation.
474 */
475 void
476 xfs_set_low_space_thresholds(
477 struct xfs_mount *mp)
478 {
479 int i;
480
481 for (i = 0; i < XFS_LOWSP_MAX; i++) {
482 __uint64_t space = mp->m_sb.sb_dblocks;
483
484 do_div(space, 100);
485 mp->m_low_space[i] = space * (i + 1);
486 }
487 }
488
489
490 /*
491 * Set whether we're using inode alignment.
492 */
493 STATIC void
494 xfs_set_inoalignment(xfs_mount_t *mp)
495 {
496 if (xfs_sb_version_hasalign(&mp->m_sb) &&
497 mp->m_sb.sb_inoalignmt >=
498 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
499 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
500 else
501 mp->m_inoalign_mask = 0;
502 /*
503 * If we are using stripe alignment, check whether
504 * the stripe unit is a multiple of the inode alignment
505 */
506 if (mp->m_dalign && mp->m_inoalign_mask &&
507 !(mp->m_dalign & mp->m_inoalign_mask))
508 mp->m_sinoalign = mp->m_dalign;
509 else
510 mp->m_sinoalign = 0;
511 }
512
513 /*
514 * Check that the data (and log if separate) is an ok size.
515 */
516 STATIC int
517 xfs_check_sizes(
518 struct xfs_mount *mp)
519 {
520 struct xfs_buf *bp;
521 xfs_daddr_t d;
522 int error;
523
524 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
525 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
526 xfs_warn(mp, "filesystem size mismatch detected");
527 return -EFBIG;
528 }
529 error = xfs_buf_read_uncached(mp->m_ddev_targp,
530 d - XFS_FSS_TO_BB(mp, 1),
531 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
532 if (error) {
533 xfs_warn(mp, "last sector read failed");
534 return error;
535 }
536 xfs_buf_relse(bp);
537
538 if (mp->m_logdev_targp == mp->m_ddev_targp)
539 return 0;
540
541 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
542 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
543 xfs_warn(mp, "log size mismatch detected");
544 return -EFBIG;
545 }
546 error = xfs_buf_read_uncached(mp->m_logdev_targp,
547 d - XFS_FSB_TO_BB(mp, 1),
548 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
549 if (error) {
550 xfs_warn(mp, "log device read failed");
551 return error;
552 }
553 xfs_buf_relse(bp);
554 return 0;
555 }
556
557 /*
558 * Clear the quotaflags in memory and in the superblock.
559 */
560 int
561 xfs_mount_reset_sbqflags(
562 struct xfs_mount *mp)
563 {
564 mp->m_qflags = 0;
565
566 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
567 if (mp->m_sb.sb_qflags == 0)
568 return 0;
569 spin_lock(&mp->m_sb_lock);
570 mp->m_sb.sb_qflags = 0;
571 spin_unlock(&mp->m_sb_lock);
572
573 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
574 return 0;
575
576 return xfs_sync_sb(mp, false);
577 }
578
579 __uint64_t
580 xfs_default_resblks(xfs_mount_t *mp)
581 {
582 __uint64_t resblks;
583
584 /*
585 * We default to 5% or 8192 fsbs of space reserved, whichever is
586 * smaller. This is intended to cover concurrent allocation
587 * transactions when we initially hit enospc. These each require a 4
588 * block reservation. Hence by default we cover roughly 2000 concurrent
589 * allocation reservations.
590 */
591 resblks = mp->m_sb.sb_dblocks;
592 do_div(resblks, 20);
593 resblks = min_t(__uint64_t, resblks, 8192);
594 return resblks;
595 }
596
597 /*
598 * This function does the following on an initial mount of a file system:
599 * - reads the superblock from disk and init the mount struct
600 * - if we're a 32-bit kernel, do a size check on the superblock
601 * so we don't mount terabyte filesystems
602 * - init mount struct realtime fields
603 * - allocate inode hash table for fs
604 * - init directory manager
605 * - perform recovery and init the log manager
606 */
607 int
608 xfs_mountfs(
609 struct xfs_mount *mp)
610 {
611 struct xfs_sb *sbp = &(mp->m_sb);
612 struct xfs_inode *rip;
613 __uint64_t resblks;
614 uint quotamount = 0;
615 uint quotaflags = 0;
616 int error = 0;
617
618 xfs_sb_mount_common(mp, sbp);
619
620 /*
621 * Check for a mismatched features2 values. Older kernels read & wrote
622 * into the wrong sb offset for sb_features2 on some platforms due to
623 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
624 * which made older superblock reading/writing routines swap it as a
625 * 64-bit value.
626 *
627 * For backwards compatibility, we make both slots equal.
628 *
629 * If we detect a mismatched field, we OR the set bits into the existing
630 * features2 field in case it has already been modified; we don't want
631 * to lose any features. We then update the bad location with the ORed
632 * value so that older kernels will see any features2 flags. The
633 * superblock writeback code ensures the new sb_features2 is copied to
634 * sb_bad_features2 before it is logged or written to disk.
635 */
636 if (xfs_sb_has_mismatched_features2(sbp)) {
637 xfs_warn(mp, "correcting sb_features alignment problem");
638 sbp->sb_features2 |= sbp->sb_bad_features2;
639 mp->m_update_sb = true;
640
641 /*
642 * Re-check for ATTR2 in case it was found in bad_features2
643 * slot.
644 */
645 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
646 !(mp->m_flags & XFS_MOUNT_NOATTR2))
647 mp->m_flags |= XFS_MOUNT_ATTR2;
648 }
649
650 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
651 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
652 xfs_sb_version_removeattr2(&mp->m_sb);
653 mp->m_update_sb = true;
654
655 /* update sb_versionnum for the clearing of the morebits */
656 if (!sbp->sb_features2)
657 mp->m_update_sb = true;
658 }
659
660 /* always use v2 inodes by default now */
661 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
662 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
663 mp->m_update_sb = true;
664 }
665
666 /*
667 * Check if sb_agblocks is aligned at stripe boundary
668 * If sb_agblocks is NOT aligned turn off m_dalign since
669 * allocator alignment is within an ag, therefore ag has
670 * to be aligned at stripe boundary.
671 */
672 error = xfs_update_alignment(mp);
673 if (error)
674 goto out;
675
676 xfs_alloc_compute_maxlevels(mp);
677 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
678 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
679 xfs_ialloc_compute_maxlevels(mp);
680
681 xfs_set_maxicount(mp);
682
683 /* enable fail_at_unmount as default */
684 mp->m_fail_unmount = 1;
685
686 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
687 if (error)
688 goto out;
689
690 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
691 &mp->m_kobj, "stats");
692 if (error)
693 goto out_remove_sysfs;
694
695 error = xfs_error_sysfs_init(mp);
696 if (error)
697 goto out_del_stats;
698
699
700 error = xfs_uuid_mount(mp);
701 if (error)
702 goto out_remove_error_sysfs;
703
704 /*
705 * Set the minimum read and write sizes
706 */
707 xfs_set_rw_sizes(mp);
708
709 /* set the low space thresholds for dynamic preallocation */
710 xfs_set_low_space_thresholds(mp);
711
712 /*
713 * Set the inode cluster size.
714 * This may still be overridden by the file system
715 * block size if it is larger than the chosen cluster size.
716 *
717 * For v5 filesystems, scale the cluster size with the inode size to
718 * keep a constant ratio of inode per cluster buffer, but only if mkfs
719 * has set the inode alignment value appropriately for larger cluster
720 * sizes.
721 */
722 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
723 if (xfs_sb_version_hascrc(&mp->m_sb)) {
724 int new_size = mp->m_inode_cluster_size;
725
726 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
727 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
728 mp->m_inode_cluster_size = new_size;
729 }
730
731 /*
732 * If enabled, sparse inode chunk alignment is expected to match the
733 * cluster size. Full inode chunk alignment must match the chunk size,
734 * but that is checked on sb read verification...
735 */
736 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
737 mp->m_sb.sb_spino_align !=
738 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
739 xfs_warn(mp,
740 "Sparse inode block alignment (%u) must match cluster size (%llu).",
741 mp->m_sb.sb_spino_align,
742 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
743 error = -EINVAL;
744 goto out_remove_uuid;
745 }
746
747 /*
748 * Set inode alignment fields
749 */
750 xfs_set_inoalignment(mp);
751
752 /*
753 * Check that the data (and log if separate) is an ok size.
754 */
755 error = xfs_check_sizes(mp);
756 if (error)
757 goto out_remove_uuid;
758
759 /*
760 * Initialize realtime fields in the mount structure
761 */
762 error = xfs_rtmount_init(mp);
763 if (error) {
764 xfs_warn(mp, "RT mount failed");
765 goto out_remove_uuid;
766 }
767
768 /*
769 * Copies the low order bits of the timestamp and the randomly
770 * set "sequence" number out of a UUID.
771 */
772 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
773
774 mp->m_dmevmask = 0; /* not persistent; set after each mount */
775
776 error = xfs_da_mount(mp);
777 if (error) {
778 xfs_warn(mp, "Failed dir/attr init: %d", error);
779 goto out_remove_uuid;
780 }
781
782 /*
783 * Initialize the precomputed transaction reservations values.
784 */
785 xfs_trans_init(mp);
786
787 /*
788 * Allocate and initialize the per-ag data.
789 */
790 spin_lock_init(&mp->m_perag_lock);
791 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
792 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
793 if (error) {
794 xfs_warn(mp, "Failed per-ag init: %d", error);
795 goto out_free_dir;
796 }
797
798 if (!sbp->sb_logblocks) {
799 xfs_warn(mp, "no log defined");
800 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
801 error = -EFSCORRUPTED;
802 goto out_free_perag;
803 }
804
805 /*
806 * Log's mount-time initialization. The first part of recovery can place
807 * some items on the AIL, to be handled when recovery is finished or
808 * cancelled.
809 */
810 error = xfs_log_mount(mp, mp->m_logdev_targp,
811 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
812 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
813 if (error) {
814 xfs_warn(mp, "log mount failed");
815 goto out_fail_wait;
816 }
817
818 /*
819 * Now the log is mounted, we know if it was an unclean shutdown or
820 * not. If it was, with the first phase of recovery has completed, we
821 * have consistent AG blocks on disk. We have not recovered EFIs yet,
822 * but they are recovered transactionally in the second recovery phase
823 * later.
824 *
825 * Hence we can safely re-initialise incore superblock counters from
826 * the per-ag data. These may not be correct if the filesystem was not
827 * cleanly unmounted, so we need to wait for recovery to finish before
828 * doing this.
829 *
830 * If the filesystem was cleanly unmounted, then we can trust the
831 * values in the superblock to be correct and we don't need to do
832 * anything here.
833 *
834 * If we are currently making the filesystem, the initialisation will
835 * fail as the perag data is in an undefined state.
836 */
837 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
838 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
839 !mp->m_sb.sb_inprogress) {
840 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
841 if (error)
842 goto out_log_dealloc;
843 }
844
845 /*
846 * Get and sanity-check the root inode.
847 * Save the pointer to it in the mount structure.
848 */
849 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
850 if (error) {
851 xfs_warn(mp, "failed to read root inode");
852 goto out_log_dealloc;
853 }
854
855 ASSERT(rip != NULL);
856
857 if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
858 xfs_warn(mp, "corrupted root inode %llu: not a directory",
859 (unsigned long long)rip->i_ino);
860 xfs_iunlock(rip, XFS_ILOCK_EXCL);
861 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
862 mp);
863 error = -EFSCORRUPTED;
864 goto out_rele_rip;
865 }
866 mp->m_rootip = rip; /* save it */
867
868 xfs_iunlock(rip, XFS_ILOCK_EXCL);
869
870 /*
871 * Initialize realtime inode pointers in the mount structure
872 */
873 error = xfs_rtmount_inodes(mp);
874 if (error) {
875 /*
876 * Free up the root inode.
877 */
878 xfs_warn(mp, "failed to read RT inodes");
879 goto out_rele_rip;
880 }
881
882 /*
883 * If this is a read-only mount defer the superblock updates until
884 * the next remount into writeable mode. Otherwise we would never
885 * perform the update e.g. for the root filesystem.
886 */
887 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
888 error = xfs_sync_sb(mp, false);
889 if (error) {
890 xfs_warn(mp, "failed to write sb changes");
891 goto out_rtunmount;
892 }
893 }
894
895 /*
896 * Initialise the XFS quota management subsystem for this mount
897 */
898 if (XFS_IS_QUOTA_RUNNING(mp)) {
899 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
900 if (error)
901 goto out_rtunmount;
902 } else {
903 ASSERT(!XFS_IS_QUOTA_ON(mp));
904
905 /*
906 * If a file system had quotas running earlier, but decided to
907 * mount without -o uquota/pquota/gquota options, revoke the
908 * quotachecked license.
909 */
910 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
911 xfs_notice(mp, "resetting quota flags");
912 error = xfs_mount_reset_sbqflags(mp);
913 if (error)
914 goto out_rtunmount;
915 }
916 }
917
918 /*
919 * Finish recovering the file system. This part needed to be delayed
920 * until after the root and real-time bitmap inodes were consistently
921 * read in.
922 */
923 error = xfs_log_mount_finish(mp);
924 if (error) {
925 xfs_warn(mp, "log mount finish failed");
926 goto out_rtunmount;
927 }
928
929 /*
930 * Complete the quota initialisation, post-log-replay component.
931 */
932 if (quotamount) {
933 ASSERT(mp->m_qflags == 0);
934 mp->m_qflags = quotaflags;
935
936 xfs_qm_mount_quotas(mp);
937 }
938
939 /*
940 * Now we are mounted, reserve a small amount of unused space for
941 * privileged transactions. This is needed so that transaction
942 * space required for critical operations can dip into this pool
943 * when at ENOSPC. This is needed for operations like create with
944 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
945 * are not allowed to use this reserved space.
946 *
947 * This may drive us straight to ENOSPC on mount, but that implies
948 * we were already there on the last unmount. Warn if this occurs.
949 */
950 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
951 resblks = xfs_default_resblks(mp);
952 error = xfs_reserve_blocks(mp, &resblks, NULL);
953 if (error)
954 xfs_warn(mp,
955 "Unable to allocate reserve blocks. Continuing without reserve pool.");
956 }
957
958 return 0;
959
960 out_rtunmount:
961 xfs_rtunmount_inodes(mp);
962 out_rele_rip:
963 IRELE(rip);
964 cancel_delayed_work_sync(&mp->m_reclaim_work);
965 xfs_reclaim_inodes(mp, SYNC_WAIT);
966 out_log_dealloc:
967 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
968 xfs_log_mount_cancel(mp);
969 out_fail_wait:
970 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
971 xfs_wait_buftarg(mp->m_logdev_targp);
972 xfs_wait_buftarg(mp->m_ddev_targp);
973 out_free_perag:
974 xfs_free_perag(mp);
975 out_free_dir:
976 xfs_da_unmount(mp);
977 out_remove_uuid:
978 xfs_uuid_unmount(mp);
979 out_remove_error_sysfs:
980 xfs_error_sysfs_del(mp);
981 out_del_stats:
982 xfs_sysfs_del(&mp->m_stats.xs_kobj);
983 out_remove_sysfs:
984 xfs_sysfs_del(&mp->m_kobj);
985 out:
986 return error;
987 }
988
989 /*
990 * This flushes out the inodes,dquots and the superblock, unmounts the
991 * log and makes sure that incore structures are freed.
992 */
993 void
994 xfs_unmountfs(
995 struct xfs_mount *mp)
996 {
997 __uint64_t resblks;
998 int error;
999
1000 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1001
1002 xfs_qm_unmount_quotas(mp);
1003 xfs_rtunmount_inodes(mp);
1004 IRELE(mp->m_rootip);
1005
1006 /*
1007 * We can potentially deadlock here if we have an inode cluster
1008 * that has been freed has its buffer still pinned in memory because
1009 * the transaction is still sitting in a iclog. The stale inodes
1010 * on that buffer will have their flush locks held until the
1011 * transaction hits the disk and the callbacks run. the inode
1012 * flush takes the flush lock unconditionally and with nothing to
1013 * push out the iclog we will never get that unlocked. hence we
1014 * need to force the log first.
1015 */
1016 xfs_log_force(mp, XFS_LOG_SYNC);
1017
1018 /*
1019 * We now need to tell the world we are unmounting. This will allow
1020 * us to detect that the filesystem is going away and we should error
1021 * out anything that we have been retrying in the background. This will
1022 * prevent neverending retries in AIL pushing from hanging the unmount.
1023 */
1024 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1025
1026 /*
1027 * Flush all pending changes from the AIL.
1028 */
1029 xfs_ail_push_all_sync(mp->m_ail);
1030
1031 /*
1032 * And reclaim all inodes. At this point there should be no dirty
1033 * inodes and none should be pinned or locked, but use synchronous
1034 * reclaim just to be sure. We can stop background inode reclaim
1035 * here as well if it is still running.
1036 */
1037 cancel_delayed_work_sync(&mp->m_reclaim_work);
1038 xfs_reclaim_inodes(mp, SYNC_WAIT);
1039
1040 xfs_qm_unmount(mp);
1041
1042 /*
1043 * Unreserve any blocks we have so that when we unmount we don't account
1044 * the reserved free space as used. This is really only necessary for
1045 * lazy superblock counting because it trusts the incore superblock
1046 * counters to be absolutely correct on clean unmount.
1047 *
1048 * We don't bother correcting this elsewhere for lazy superblock
1049 * counting because on mount of an unclean filesystem we reconstruct the
1050 * correct counter value and this is irrelevant.
1051 *
1052 * For non-lazy counter filesystems, this doesn't matter at all because
1053 * we only every apply deltas to the superblock and hence the incore
1054 * value does not matter....
1055 */
1056 resblks = 0;
1057 error = xfs_reserve_blocks(mp, &resblks, NULL);
1058 if (error)
1059 xfs_warn(mp, "Unable to free reserved block pool. "
1060 "Freespace may not be correct on next mount.");
1061
1062 error = xfs_log_sbcount(mp);
1063 if (error)
1064 xfs_warn(mp, "Unable to update superblock counters. "
1065 "Freespace may not be correct on next mount.");
1066
1067
1068 xfs_log_unmount(mp);
1069 xfs_da_unmount(mp);
1070 xfs_uuid_unmount(mp);
1071
1072 #if defined(DEBUG)
1073 xfs_errortag_clearall(mp, 0);
1074 #endif
1075 xfs_free_perag(mp);
1076
1077 xfs_error_sysfs_del(mp);
1078 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1079 xfs_sysfs_del(&mp->m_kobj);
1080 }
1081
1082 /*
1083 * Determine whether modifications can proceed. The caller specifies the minimum
1084 * freeze level for which modifications should not be allowed. This allows
1085 * certain operations to proceed while the freeze sequence is in progress, if
1086 * necessary.
1087 */
1088 bool
1089 xfs_fs_writable(
1090 struct xfs_mount *mp,
1091 int level)
1092 {
1093 ASSERT(level > SB_UNFROZEN);
1094 if ((mp->m_super->s_writers.frozen >= level) ||
1095 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1096 return false;
1097
1098 return true;
1099 }
1100
1101 /*
1102 * xfs_log_sbcount
1103 *
1104 * Sync the superblock counters to disk.
1105 *
1106 * Note this code can be called during the process of freezing, so we use the
1107 * transaction allocator that does not block when the transaction subsystem is
1108 * in its frozen state.
1109 */
1110 int
1111 xfs_log_sbcount(xfs_mount_t *mp)
1112 {
1113 /* allow this to proceed during the freeze sequence... */
1114 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1115 return 0;
1116
1117 /*
1118 * we don't need to do this if we are updating the superblock
1119 * counters on every modification.
1120 */
1121 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1122 return 0;
1123
1124 return xfs_sync_sb(mp, true);
1125 }
1126
1127 /*
1128 * Deltas for the inode count are +/-64, hence we use a large batch size
1129 * of 128 so we don't need to take the counter lock on every update.
1130 */
1131 #define XFS_ICOUNT_BATCH 128
1132 int
1133 xfs_mod_icount(
1134 struct xfs_mount *mp,
1135 int64_t delta)
1136 {
1137 __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1138 if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1139 ASSERT(0);
1140 percpu_counter_add(&mp->m_icount, -delta);
1141 return -EINVAL;
1142 }
1143 return 0;
1144 }
1145
1146 int
1147 xfs_mod_ifree(
1148 struct xfs_mount *mp,
1149 int64_t delta)
1150 {
1151 percpu_counter_add(&mp->m_ifree, delta);
1152 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1153 ASSERT(0);
1154 percpu_counter_add(&mp->m_ifree, -delta);
1155 return -EINVAL;
1156 }
1157 return 0;
1158 }
1159
1160 /*
1161 * Deltas for the block count can vary from 1 to very large, but lock contention
1162 * only occurs on frequent small block count updates such as in the delayed
1163 * allocation path for buffered writes (page a time updates). Hence we set
1164 * a large batch count (1024) to minimise global counter updates except when
1165 * we get near to ENOSPC and we have to be very accurate with our updates.
1166 */
1167 #define XFS_FDBLOCKS_BATCH 1024
1168 int
1169 xfs_mod_fdblocks(
1170 struct xfs_mount *mp,
1171 int64_t delta,
1172 bool rsvd)
1173 {
1174 int64_t lcounter;
1175 long long res_used;
1176 s32 batch;
1177
1178 if (delta > 0) {
1179 /*
1180 * If the reserve pool is depleted, put blocks back into it
1181 * first. Most of the time the pool is full.
1182 */
1183 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1184 percpu_counter_add(&mp->m_fdblocks, delta);
1185 return 0;
1186 }
1187
1188 spin_lock(&mp->m_sb_lock);
1189 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1190
1191 if (res_used > delta) {
1192 mp->m_resblks_avail += delta;
1193 } else {
1194 delta -= res_used;
1195 mp->m_resblks_avail = mp->m_resblks;
1196 percpu_counter_add(&mp->m_fdblocks, delta);
1197 }
1198 spin_unlock(&mp->m_sb_lock);
1199 return 0;
1200 }
1201
1202 /*
1203 * Taking blocks away, need to be more accurate the closer we
1204 * are to zero.
1205 *
1206 * If the counter has a value of less than 2 * max batch size,
1207 * then make everything serialise as we are real close to
1208 * ENOSPC.
1209 */
1210 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1211 XFS_FDBLOCKS_BATCH) < 0)
1212 batch = 1;
1213 else
1214 batch = XFS_FDBLOCKS_BATCH;
1215
1216 __percpu_counter_add(&mp->m_fdblocks, delta, batch);
1217 if (__percpu_counter_compare(&mp->m_fdblocks, XFS_ALLOC_SET_ASIDE(mp),
1218 XFS_FDBLOCKS_BATCH) >= 0) {
1219 /* we had space! */
1220 return 0;
1221 }
1222
1223 /*
1224 * lock up the sb for dipping into reserves before releasing the space
1225 * that took us to ENOSPC.
1226 */
1227 spin_lock(&mp->m_sb_lock);
1228 percpu_counter_add(&mp->m_fdblocks, -delta);
1229 if (!rsvd)
1230 goto fdblocks_enospc;
1231
1232 lcounter = (long long)mp->m_resblks_avail + delta;
1233 if (lcounter >= 0) {
1234 mp->m_resblks_avail = lcounter;
1235 spin_unlock(&mp->m_sb_lock);
1236 return 0;
1237 }
1238 printk_once(KERN_WARNING
1239 "Filesystem \"%s\": reserve blocks depleted! "
1240 "Consider increasing reserve pool size.",
1241 mp->m_fsname);
1242 fdblocks_enospc:
1243 spin_unlock(&mp->m_sb_lock);
1244 return -ENOSPC;
1245 }
1246
1247 int
1248 xfs_mod_frextents(
1249 struct xfs_mount *mp,
1250 int64_t delta)
1251 {
1252 int64_t lcounter;
1253 int ret = 0;
1254
1255 spin_lock(&mp->m_sb_lock);
1256 lcounter = mp->m_sb.sb_frextents + delta;
1257 if (lcounter < 0)
1258 ret = -ENOSPC;
1259 else
1260 mp->m_sb.sb_frextents = lcounter;
1261 spin_unlock(&mp->m_sb_lock);
1262 return ret;
1263 }
1264
1265 /*
1266 * xfs_getsb() is called to obtain the buffer for the superblock.
1267 * The buffer is returned locked and read in from disk.
1268 * The buffer should be released with a call to xfs_brelse().
1269 *
1270 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1271 * the superblock buffer if it can be locked without sleeping.
1272 * If it can't then we'll return NULL.
1273 */
1274 struct xfs_buf *
1275 xfs_getsb(
1276 struct xfs_mount *mp,
1277 int flags)
1278 {
1279 struct xfs_buf *bp = mp->m_sb_bp;
1280
1281 if (!xfs_buf_trylock(bp)) {
1282 if (flags & XBF_TRYLOCK)
1283 return NULL;
1284 xfs_buf_lock(bp);
1285 }
1286
1287 xfs_buf_hold(bp);
1288 ASSERT(bp->b_flags & XBF_DONE);
1289 return bp;
1290 }
1291
1292 /*
1293 * Used to free the superblock along various error paths.
1294 */
1295 void
1296 xfs_freesb(
1297 struct xfs_mount *mp)
1298 {
1299 struct xfs_buf *bp = mp->m_sb_bp;
1300
1301 xfs_buf_lock(bp);
1302 mp->m_sb_bp = NULL;
1303 xfs_buf_relse(bp);
1304 }
1305
1306 /*
1307 * If the underlying (data/log/rt) device is readonly, there are some
1308 * operations that cannot proceed.
1309 */
1310 int
1311 xfs_dev_is_read_only(
1312 struct xfs_mount *mp,
1313 char *message)
1314 {
1315 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1316 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1317 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1318 xfs_notice(mp, "%s required on read-only device.", message);
1319 xfs_notice(mp, "write access unavailable, cannot proceed.");
1320 return -EROFS;
1321 }
1322 return 0;
1323 }
This page took 0.05835 seconds and 5 git commands to generate.