Loading...
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_defer.h"
28#include "xfs_da_format.h"
29#include "xfs_da_btree.h"
30#include "xfs_inode.h"
31#include "xfs_dir2.h"
32#include "xfs_ialloc.h"
33#include "xfs_alloc.h"
34#include "xfs_rtalloc.h"
35#include "xfs_bmap.h"
36#include "xfs_trans.h"
37#include "xfs_trans_priv.h"
38#include "xfs_log.h"
39#include "xfs_error.h"
40#include "xfs_quota.h"
41#include "xfs_fsops.h"
42#include "xfs_trace.h"
43#include "xfs_icache.h"
44#include "xfs_sysfs.h"
45#include "xfs_rmap_btree.h"
46#include "xfs_refcount_btree.h"
47#include "xfs_reflink.h"
48#include "xfs_extent_busy.h"
49
50
51static DEFINE_MUTEX(xfs_uuid_table_mutex);
52static int xfs_uuid_table_size;
53static uuid_t *xfs_uuid_table;
54
55void
56xfs_uuid_table_free(void)
57{
58 if (xfs_uuid_table_size == 0)
59 return;
60 kmem_free(xfs_uuid_table);
61 xfs_uuid_table = NULL;
62 xfs_uuid_table_size = 0;
63}
64
65/*
66 * See if the UUID is unique among mounted XFS filesystems.
67 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
68 */
69STATIC int
70xfs_uuid_mount(
71 struct xfs_mount *mp)
72{
73 uuid_t *uuid = &mp->m_sb.sb_uuid;
74 int hole, i;
75
76 /* Publish UUID in struct super_block */
77 uuid_copy(&mp->m_super->s_uuid, uuid);
78
79 if (mp->m_flags & XFS_MOUNT_NOUUID)
80 return 0;
81
82 if (uuid_is_null(uuid)) {
83 xfs_warn(mp, "Filesystem has null UUID - can't mount");
84 return -EINVAL;
85 }
86
87 mutex_lock(&xfs_uuid_table_mutex);
88 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
89 if (uuid_is_null(&xfs_uuid_table[i])) {
90 hole = i;
91 continue;
92 }
93 if (uuid_equal(uuid, &xfs_uuid_table[i]))
94 goto out_duplicate;
95 }
96
97 if (hole < 0) {
98 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
99 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
100 KM_SLEEP);
101 hole = xfs_uuid_table_size++;
102 }
103 xfs_uuid_table[hole] = *uuid;
104 mutex_unlock(&xfs_uuid_table_mutex);
105
106 return 0;
107
108 out_duplicate:
109 mutex_unlock(&xfs_uuid_table_mutex);
110 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
111 return -EINVAL;
112}
113
114STATIC void
115xfs_uuid_unmount(
116 struct xfs_mount *mp)
117{
118 uuid_t *uuid = &mp->m_sb.sb_uuid;
119 int i;
120
121 if (mp->m_flags & XFS_MOUNT_NOUUID)
122 return;
123
124 mutex_lock(&xfs_uuid_table_mutex);
125 for (i = 0; i < xfs_uuid_table_size; i++) {
126 if (uuid_is_null(&xfs_uuid_table[i]))
127 continue;
128 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
129 continue;
130 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
131 break;
132 }
133 ASSERT(i < xfs_uuid_table_size);
134 mutex_unlock(&xfs_uuid_table_mutex);
135}
136
137
138STATIC void
139__xfs_free_perag(
140 struct rcu_head *head)
141{
142 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
143
144 ASSERT(atomic_read(&pag->pag_ref) == 0);
145 kmem_free(pag);
146}
147
148/*
149 * Free up the per-ag resources associated with the mount structure.
150 */
151STATIC void
152xfs_free_perag(
153 xfs_mount_t *mp)
154{
155 xfs_agnumber_t agno;
156 struct xfs_perag *pag;
157
158 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
159 spin_lock(&mp->m_perag_lock);
160 pag = radix_tree_delete(&mp->m_perag_tree, agno);
161 spin_unlock(&mp->m_perag_lock);
162 ASSERT(pag);
163 ASSERT(atomic_read(&pag->pag_ref) == 0);
164 xfs_buf_hash_destroy(pag);
165 mutex_destroy(&pag->pag_ici_reclaim_lock);
166 call_rcu(&pag->rcu_head, __xfs_free_perag);
167 }
168}
169
170/*
171 * Check size of device based on the (data/realtime) block count.
172 * Note: this check is used by the growfs code as well as mount.
173 */
174int
175xfs_sb_validate_fsb_count(
176 xfs_sb_t *sbp,
177 uint64_t nblocks)
178{
179 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
180 ASSERT(sbp->sb_blocklog >= BBSHIFT);
181
182 /* Limited by ULONG_MAX of page cache index */
183 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
184 return -EFBIG;
185 return 0;
186}
187
188int
189xfs_initialize_perag(
190 xfs_mount_t *mp,
191 xfs_agnumber_t agcount,
192 xfs_agnumber_t *maxagi)
193{
194 xfs_agnumber_t index;
195 xfs_agnumber_t first_initialised = NULLAGNUMBER;
196 xfs_perag_t *pag;
197 int error = -ENOMEM;
198
199 /*
200 * Walk the current per-ag tree so we don't try to initialise AGs
201 * that already exist (growfs case). Allocate and insert all the
202 * AGs we don't find ready for initialisation.
203 */
204 for (index = 0; index < agcount; index++) {
205 pag = xfs_perag_get(mp, index);
206 if (pag) {
207 xfs_perag_put(pag);
208 continue;
209 }
210
211 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
212 if (!pag)
213 goto out_unwind_new_pags;
214 pag->pag_agno = index;
215 pag->pag_mount = mp;
216 spin_lock_init(&pag->pag_ici_lock);
217 mutex_init(&pag->pag_ici_reclaim_lock);
218 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
219 if (xfs_buf_hash_init(pag))
220 goto out_free_pag;
221 init_waitqueue_head(&pag->pagb_wait);
222
223 if (radix_tree_preload(GFP_NOFS))
224 goto out_hash_destroy;
225
226 spin_lock(&mp->m_perag_lock);
227 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
228 BUG();
229 spin_unlock(&mp->m_perag_lock);
230 radix_tree_preload_end();
231 error = -EEXIST;
232 goto out_hash_destroy;
233 }
234 spin_unlock(&mp->m_perag_lock);
235 radix_tree_preload_end();
236 /* first new pag is fully initialized */
237 if (first_initialised == NULLAGNUMBER)
238 first_initialised = index;
239 }
240
241 index = xfs_set_inode_alloc(mp, agcount);
242
243 if (maxagi)
244 *maxagi = index;
245
246 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
247 return 0;
248
249out_hash_destroy:
250 xfs_buf_hash_destroy(pag);
251out_free_pag:
252 mutex_destroy(&pag->pag_ici_reclaim_lock);
253 kmem_free(pag);
254out_unwind_new_pags:
255 /* unwind any prior newly initialized pags */
256 for (index = first_initialised; index < agcount; index++) {
257 pag = radix_tree_delete(&mp->m_perag_tree, index);
258 if (!pag)
259 break;
260 xfs_buf_hash_destroy(pag);
261 mutex_destroy(&pag->pag_ici_reclaim_lock);
262 kmem_free(pag);
263 }
264 return error;
265}
266
267/*
268 * xfs_readsb
269 *
270 * Does the initial read of the superblock.
271 */
272int
273xfs_readsb(
274 struct xfs_mount *mp,
275 int flags)
276{
277 unsigned int sector_size;
278 struct xfs_buf *bp;
279 struct xfs_sb *sbp = &mp->m_sb;
280 int error;
281 int loud = !(flags & XFS_MFSI_QUIET);
282 const struct xfs_buf_ops *buf_ops;
283
284 ASSERT(mp->m_sb_bp == NULL);
285 ASSERT(mp->m_ddev_targp != NULL);
286
287 /*
288 * For the initial read, we must guess at the sector
289 * size based on the block device. It's enough to
290 * get the sb_sectsize out of the superblock and
291 * then reread with the proper length.
292 * We don't verify it yet, because it may not be complete.
293 */
294 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
295 buf_ops = NULL;
296
297 /*
298 * Allocate a (locked) buffer to hold the superblock. This will be kept
299 * around at all times to optimize access to the superblock. Therefore,
300 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
301 * elevated.
302 */
303reread:
304 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
305 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
306 buf_ops);
307 if (error) {
308 if (loud)
309 xfs_warn(mp, "SB validate failed with error %d.", error);
310 /* bad CRC means corrupted metadata */
311 if (error == -EFSBADCRC)
312 error = -EFSCORRUPTED;
313 return error;
314 }
315
316 /*
317 * Initialize the mount structure from the superblock.
318 */
319 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
320
321 /*
322 * If we haven't validated the superblock, do so now before we try
323 * to check the sector size and reread the superblock appropriately.
324 */
325 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
326 if (loud)
327 xfs_warn(mp, "Invalid superblock magic number");
328 error = -EINVAL;
329 goto release_buf;
330 }
331
332 /*
333 * We must be able to do sector-sized and sector-aligned IO.
334 */
335 if (sector_size > sbp->sb_sectsize) {
336 if (loud)
337 xfs_warn(mp, "device supports %u byte sectors (not %u)",
338 sector_size, sbp->sb_sectsize);
339 error = -ENOSYS;
340 goto release_buf;
341 }
342
343 if (buf_ops == NULL) {
344 /*
345 * Re-read the superblock so the buffer is correctly sized,
346 * and properly verified.
347 */
348 xfs_buf_relse(bp);
349 sector_size = sbp->sb_sectsize;
350 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
351 goto reread;
352 }
353
354 xfs_reinit_percpu_counters(mp);
355
356 /* no need to be quiet anymore, so reset the buf ops */
357 bp->b_ops = &xfs_sb_buf_ops;
358
359 mp->m_sb_bp = bp;
360 xfs_buf_unlock(bp);
361 return 0;
362
363release_buf:
364 xfs_buf_relse(bp);
365 return error;
366}
367
368/*
369 * Update alignment values based on mount options and sb values
370 */
371STATIC int
372xfs_update_alignment(xfs_mount_t *mp)
373{
374 xfs_sb_t *sbp = &(mp->m_sb);
375
376 if (mp->m_dalign) {
377 /*
378 * If stripe unit and stripe width are not multiples
379 * of the fs blocksize turn off alignment.
380 */
381 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
382 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
383 xfs_warn(mp,
384 "alignment check failed: sunit/swidth vs. blocksize(%d)",
385 sbp->sb_blocksize);
386 return -EINVAL;
387 } else {
388 /*
389 * Convert the stripe unit and width to FSBs.
390 */
391 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
392 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
393 xfs_warn(mp,
394 "alignment check failed: sunit/swidth vs. agsize(%d)",
395 sbp->sb_agblocks);
396 return -EINVAL;
397 } else if (mp->m_dalign) {
398 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
399 } else {
400 xfs_warn(mp,
401 "alignment check failed: sunit(%d) less than bsize(%d)",
402 mp->m_dalign, sbp->sb_blocksize);
403 return -EINVAL;
404 }
405 }
406
407 /*
408 * Update superblock with new values
409 * and log changes
410 */
411 if (xfs_sb_version_hasdalign(sbp)) {
412 if (sbp->sb_unit != mp->m_dalign) {
413 sbp->sb_unit = mp->m_dalign;
414 mp->m_update_sb = true;
415 }
416 if (sbp->sb_width != mp->m_swidth) {
417 sbp->sb_width = mp->m_swidth;
418 mp->m_update_sb = true;
419 }
420 } else {
421 xfs_warn(mp,
422 "cannot change alignment: superblock does not support data alignment");
423 return -EINVAL;
424 }
425 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
426 xfs_sb_version_hasdalign(&mp->m_sb)) {
427 mp->m_dalign = sbp->sb_unit;
428 mp->m_swidth = sbp->sb_width;
429 }
430
431 return 0;
432}
433
434/*
435 * Set the maximum inode count for this filesystem
436 */
437STATIC void
438xfs_set_maxicount(xfs_mount_t *mp)
439{
440 xfs_sb_t *sbp = &(mp->m_sb);
441 uint64_t icount;
442
443 if (sbp->sb_imax_pct) {
444 /*
445 * Make sure the maximum inode count is a multiple
446 * of the units we allocate inodes in.
447 */
448 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
449 do_div(icount, 100);
450 do_div(icount, mp->m_ialloc_blks);
451 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
452 sbp->sb_inopblog;
453 } else {
454 mp->m_maxicount = 0;
455 }
456}
457
458/*
459 * Set the default minimum read and write sizes unless
460 * already specified in a mount option.
461 * We use smaller I/O sizes when the file system
462 * is being used for NFS service (wsync mount option).
463 */
464STATIC void
465xfs_set_rw_sizes(xfs_mount_t *mp)
466{
467 xfs_sb_t *sbp = &(mp->m_sb);
468 int readio_log, writeio_log;
469
470 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
471 if (mp->m_flags & XFS_MOUNT_WSYNC) {
472 readio_log = XFS_WSYNC_READIO_LOG;
473 writeio_log = XFS_WSYNC_WRITEIO_LOG;
474 } else {
475 readio_log = XFS_READIO_LOG_LARGE;
476 writeio_log = XFS_WRITEIO_LOG_LARGE;
477 }
478 } else {
479 readio_log = mp->m_readio_log;
480 writeio_log = mp->m_writeio_log;
481 }
482
483 if (sbp->sb_blocklog > readio_log) {
484 mp->m_readio_log = sbp->sb_blocklog;
485 } else {
486 mp->m_readio_log = readio_log;
487 }
488 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
489 if (sbp->sb_blocklog > writeio_log) {
490 mp->m_writeio_log = sbp->sb_blocklog;
491 } else {
492 mp->m_writeio_log = writeio_log;
493 }
494 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
495}
496
497/*
498 * precalculate the low space thresholds for dynamic speculative preallocation.
499 */
500void
501xfs_set_low_space_thresholds(
502 struct xfs_mount *mp)
503{
504 int i;
505
506 for (i = 0; i < XFS_LOWSP_MAX; i++) {
507 uint64_t space = mp->m_sb.sb_dblocks;
508
509 do_div(space, 100);
510 mp->m_low_space[i] = space * (i + 1);
511 }
512}
513
514
515/*
516 * Set whether we're using inode alignment.
517 */
518STATIC void
519xfs_set_inoalignment(xfs_mount_t *mp)
520{
521 if (xfs_sb_version_hasalign(&mp->m_sb) &&
522 mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
523 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
524 else
525 mp->m_inoalign_mask = 0;
526 /*
527 * If we are using stripe alignment, check whether
528 * the stripe unit is a multiple of the inode alignment
529 */
530 if (mp->m_dalign && mp->m_inoalign_mask &&
531 !(mp->m_dalign & mp->m_inoalign_mask))
532 mp->m_sinoalign = mp->m_dalign;
533 else
534 mp->m_sinoalign = 0;
535}
536
537/*
538 * Check that the data (and log if separate) is an ok size.
539 */
540STATIC int
541xfs_check_sizes(
542 struct xfs_mount *mp)
543{
544 struct xfs_buf *bp;
545 xfs_daddr_t d;
546 int error;
547
548 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
549 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
550 xfs_warn(mp, "filesystem size mismatch detected");
551 return -EFBIG;
552 }
553 error = xfs_buf_read_uncached(mp->m_ddev_targp,
554 d - XFS_FSS_TO_BB(mp, 1),
555 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
556 if (error) {
557 xfs_warn(mp, "last sector read failed");
558 return error;
559 }
560 xfs_buf_relse(bp);
561
562 if (mp->m_logdev_targp == mp->m_ddev_targp)
563 return 0;
564
565 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
566 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
567 xfs_warn(mp, "log size mismatch detected");
568 return -EFBIG;
569 }
570 error = xfs_buf_read_uncached(mp->m_logdev_targp,
571 d - XFS_FSB_TO_BB(mp, 1),
572 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
573 if (error) {
574 xfs_warn(mp, "log device read failed");
575 return error;
576 }
577 xfs_buf_relse(bp);
578 return 0;
579}
580
581/*
582 * Clear the quotaflags in memory and in the superblock.
583 */
584int
585xfs_mount_reset_sbqflags(
586 struct xfs_mount *mp)
587{
588 mp->m_qflags = 0;
589
590 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
591 if (mp->m_sb.sb_qflags == 0)
592 return 0;
593 spin_lock(&mp->m_sb_lock);
594 mp->m_sb.sb_qflags = 0;
595 spin_unlock(&mp->m_sb_lock);
596
597 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
598 return 0;
599
600 return xfs_sync_sb(mp, false);
601}
602
603uint64_t
604xfs_default_resblks(xfs_mount_t *mp)
605{
606 uint64_t resblks;
607
608 /*
609 * We default to 5% or 8192 fsbs of space reserved, whichever is
610 * smaller. This is intended to cover concurrent allocation
611 * transactions when we initially hit enospc. These each require a 4
612 * block reservation. Hence by default we cover roughly 2000 concurrent
613 * allocation reservations.
614 */
615 resblks = mp->m_sb.sb_dblocks;
616 do_div(resblks, 20);
617 resblks = min_t(uint64_t, resblks, 8192);
618 return resblks;
619}
620
621/*
622 * This function does the following on an initial mount of a file system:
623 * - reads the superblock from disk and init the mount struct
624 * - if we're a 32-bit kernel, do a size check on the superblock
625 * so we don't mount terabyte filesystems
626 * - init mount struct realtime fields
627 * - allocate inode hash table for fs
628 * - init directory manager
629 * - perform recovery and init the log manager
630 */
631int
632xfs_mountfs(
633 struct xfs_mount *mp)
634{
635 struct xfs_sb *sbp = &(mp->m_sb);
636 struct xfs_inode *rip;
637 uint64_t resblks;
638 uint quotamount = 0;
639 uint quotaflags = 0;
640 int error = 0;
641
642 xfs_sb_mount_common(mp, sbp);
643
644 /*
645 * Check for a mismatched features2 values. Older kernels read & wrote
646 * into the wrong sb offset for sb_features2 on some platforms due to
647 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
648 * which made older superblock reading/writing routines swap it as a
649 * 64-bit value.
650 *
651 * For backwards compatibility, we make both slots equal.
652 *
653 * If we detect a mismatched field, we OR the set bits into the existing
654 * features2 field in case it has already been modified; we don't want
655 * to lose any features. We then update the bad location with the ORed
656 * value so that older kernels will see any features2 flags. The
657 * superblock writeback code ensures the new sb_features2 is copied to
658 * sb_bad_features2 before it is logged or written to disk.
659 */
660 if (xfs_sb_has_mismatched_features2(sbp)) {
661 xfs_warn(mp, "correcting sb_features alignment problem");
662 sbp->sb_features2 |= sbp->sb_bad_features2;
663 mp->m_update_sb = true;
664
665 /*
666 * Re-check for ATTR2 in case it was found in bad_features2
667 * slot.
668 */
669 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
670 !(mp->m_flags & XFS_MOUNT_NOATTR2))
671 mp->m_flags |= XFS_MOUNT_ATTR2;
672 }
673
674 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
675 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
676 xfs_sb_version_removeattr2(&mp->m_sb);
677 mp->m_update_sb = true;
678
679 /* update sb_versionnum for the clearing of the morebits */
680 if (!sbp->sb_features2)
681 mp->m_update_sb = true;
682 }
683
684 /* always use v2 inodes by default now */
685 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
686 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
687 mp->m_update_sb = true;
688 }
689
690 /*
691 * Check if sb_agblocks is aligned at stripe boundary
692 * If sb_agblocks is NOT aligned turn off m_dalign since
693 * allocator alignment is within an ag, therefore ag has
694 * to be aligned at stripe boundary.
695 */
696 error = xfs_update_alignment(mp);
697 if (error)
698 goto out;
699
700 xfs_alloc_compute_maxlevels(mp);
701 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
702 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
703 xfs_ialloc_compute_maxlevels(mp);
704 xfs_rmapbt_compute_maxlevels(mp);
705 xfs_refcountbt_compute_maxlevels(mp);
706
707 xfs_set_maxicount(mp);
708
709 /* enable fail_at_unmount as default */
710 mp->m_fail_unmount = true;
711
712 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
713 if (error)
714 goto out;
715
716 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
717 &mp->m_kobj, "stats");
718 if (error)
719 goto out_remove_sysfs;
720
721 error = xfs_error_sysfs_init(mp);
722 if (error)
723 goto out_del_stats;
724
725 error = xfs_errortag_init(mp);
726 if (error)
727 goto out_remove_error_sysfs;
728
729 error = xfs_uuid_mount(mp);
730 if (error)
731 goto out_remove_errortag;
732
733 /*
734 * Set the minimum read and write sizes
735 */
736 xfs_set_rw_sizes(mp);
737
738 /* set the low space thresholds for dynamic preallocation */
739 xfs_set_low_space_thresholds(mp);
740
741 /*
742 * Set the inode cluster size.
743 * This may still be overridden by the file system
744 * block size if it is larger than the chosen cluster size.
745 *
746 * For v5 filesystems, scale the cluster size with the inode size to
747 * keep a constant ratio of inode per cluster buffer, but only if mkfs
748 * has set the inode alignment value appropriately for larger cluster
749 * sizes.
750 */
751 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
752 if (xfs_sb_version_hascrc(&mp->m_sb)) {
753 int new_size = mp->m_inode_cluster_size;
754
755 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
756 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
757 mp->m_inode_cluster_size = new_size;
758 }
759
760 /*
761 * If enabled, sparse inode chunk alignment is expected to match the
762 * cluster size. Full inode chunk alignment must match the chunk size,
763 * but that is checked on sb read verification...
764 */
765 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
766 mp->m_sb.sb_spino_align !=
767 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
768 xfs_warn(mp,
769 "Sparse inode block alignment (%u) must match cluster size (%llu).",
770 mp->m_sb.sb_spino_align,
771 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
772 error = -EINVAL;
773 goto out_remove_uuid;
774 }
775
776 /*
777 * Set inode alignment fields
778 */
779 xfs_set_inoalignment(mp);
780
781 /*
782 * Check that the data (and log if separate) is an ok size.
783 */
784 error = xfs_check_sizes(mp);
785 if (error)
786 goto out_remove_uuid;
787
788 /*
789 * Initialize realtime fields in the mount structure
790 */
791 error = xfs_rtmount_init(mp);
792 if (error) {
793 xfs_warn(mp, "RT mount failed");
794 goto out_remove_uuid;
795 }
796
797 /*
798 * Copies the low order bits of the timestamp and the randomly
799 * set "sequence" number out of a UUID.
800 */
801 mp->m_fixedfsid[0] =
802 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
803 get_unaligned_be16(&sbp->sb_uuid.b[4]);
804 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
805
806 error = xfs_da_mount(mp);
807 if (error) {
808 xfs_warn(mp, "Failed dir/attr init: %d", error);
809 goto out_remove_uuid;
810 }
811
812 /*
813 * Initialize the precomputed transaction reservations values.
814 */
815 xfs_trans_init(mp);
816
817 /*
818 * Allocate and initialize the per-ag data.
819 */
820 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
821 if (error) {
822 xfs_warn(mp, "Failed per-ag init: %d", error);
823 goto out_free_dir;
824 }
825
826 if (!sbp->sb_logblocks) {
827 xfs_warn(mp, "no log defined");
828 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
829 error = -EFSCORRUPTED;
830 goto out_free_perag;
831 }
832
833 /*
834 * Log's mount-time initialization. The first part of recovery can place
835 * some items on the AIL, to be handled when recovery is finished or
836 * cancelled.
837 */
838 error = xfs_log_mount(mp, mp->m_logdev_targp,
839 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
840 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
841 if (error) {
842 xfs_warn(mp, "log mount failed");
843 goto out_fail_wait;
844 }
845
846 /*
847 * Now the log is mounted, we know if it was an unclean shutdown or
848 * not. If it was, with the first phase of recovery has completed, we
849 * have consistent AG blocks on disk. We have not recovered EFIs yet,
850 * but they are recovered transactionally in the second recovery phase
851 * later.
852 *
853 * Hence we can safely re-initialise incore superblock counters from
854 * the per-ag data. These may not be correct if the filesystem was not
855 * cleanly unmounted, so we need to wait for recovery to finish before
856 * doing this.
857 *
858 * If the filesystem was cleanly unmounted, then we can trust the
859 * values in the superblock to be correct and we don't need to do
860 * anything here.
861 *
862 * If we are currently making the filesystem, the initialisation will
863 * fail as the perag data is in an undefined state.
864 */
865 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
866 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
867 !mp->m_sb.sb_inprogress) {
868 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
869 if (error)
870 goto out_log_dealloc;
871 }
872
873 /*
874 * Get and sanity-check the root inode.
875 * Save the pointer to it in the mount structure.
876 */
877 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
878 if (error) {
879 xfs_warn(mp, "failed to read root inode");
880 goto out_log_dealloc;
881 }
882
883 ASSERT(rip != NULL);
884
885 if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
886 xfs_warn(mp, "corrupted root inode %llu: not a directory",
887 (unsigned long long)rip->i_ino);
888 xfs_iunlock(rip, XFS_ILOCK_EXCL);
889 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
890 mp);
891 error = -EFSCORRUPTED;
892 goto out_rele_rip;
893 }
894 mp->m_rootip = rip; /* save it */
895
896 xfs_iunlock(rip, XFS_ILOCK_EXCL);
897
898 /*
899 * Initialize realtime inode pointers in the mount structure
900 */
901 error = xfs_rtmount_inodes(mp);
902 if (error) {
903 /*
904 * Free up the root inode.
905 */
906 xfs_warn(mp, "failed to read RT inodes");
907 goto out_rele_rip;
908 }
909
910 /*
911 * If this is a read-only mount defer the superblock updates until
912 * the next remount into writeable mode. Otherwise we would never
913 * perform the update e.g. for the root filesystem.
914 */
915 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
916 error = xfs_sync_sb(mp, false);
917 if (error) {
918 xfs_warn(mp, "failed to write sb changes");
919 goto out_rtunmount;
920 }
921 }
922
923 /*
924 * Initialise the XFS quota management subsystem for this mount
925 */
926 if (XFS_IS_QUOTA_RUNNING(mp)) {
927 error = xfs_qm_newmount(mp, "amount, "aflags);
928 if (error)
929 goto out_rtunmount;
930 } else {
931 ASSERT(!XFS_IS_QUOTA_ON(mp));
932
933 /*
934 * If a file system had quotas running earlier, but decided to
935 * mount without -o uquota/pquota/gquota options, revoke the
936 * quotachecked license.
937 */
938 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
939 xfs_notice(mp, "resetting quota flags");
940 error = xfs_mount_reset_sbqflags(mp);
941 if (error)
942 goto out_rtunmount;
943 }
944 }
945
946 /*
947 * Finish recovering the file system. This part needed to be delayed
948 * until after the root and real-time bitmap inodes were consistently
949 * read in.
950 */
951 error = xfs_log_mount_finish(mp);
952 if (error) {
953 xfs_warn(mp, "log mount finish failed");
954 goto out_rtunmount;
955 }
956
957 /*
958 * Now the log is fully replayed, we can transition to full read-only
959 * mode for read-only mounts. This will sync all the metadata and clean
960 * the log so that the recovery we just performed does not have to be
961 * replayed again on the next mount.
962 *
963 * We use the same quiesce mechanism as the rw->ro remount, as they are
964 * semantically identical operations.
965 */
966 if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
967 XFS_MOUNT_RDONLY) {
968 xfs_quiesce_attr(mp);
969 }
970
971 /*
972 * Complete the quota initialisation, post-log-replay component.
973 */
974 if (quotamount) {
975 ASSERT(mp->m_qflags == 0);
976 mp->m_qflags = quotaflags;
977
978 xfs_qm_mount_quotas(mp);
979 }
980
981 /*
982 * Now we are mounted, reserve a small amount of unused space for
983 * privileged transactions. This is needed so that transaction
984 * space required for critical operations can dip into this pool
985 * when at ENOSPC. This is needed for operations like create with
986 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
987 * are not allowed to use this reserved space.
988 *
989 * This may drive us straight to ENOSPC on mount, but that implies
990 * we were already there on the last unmount. Warn if this occurs.
991 */
992 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
993 resblks = xfs_default_resblks(mp);
994 error = xfs_reserve_blocks(mp, &resblks, NULL);
995 if (error)
996 xfs_warn(mp,
997 "Unable to allocate reserve blocks. Continuing without reserve pool.");
998
999 /* Recover any CoW blocks that never got remapped. */
1000 error = xfs_reflink_recover_cow(mp);
1001 if (error) {
1002 xfs_err(mp,
1003 "Error %d recovering leftover CoW allocations.", error);
1004 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1005 goto out_quota;
1006 }
1007
1008 /* Reserve AG blocks for future btree expansion. */
1009 error = xfs_fs_reserve_ag_blocks(mp);
1010 if (error && error != -ENOSPC)
1011 goto out_agresv;
1012 }
1013
1014 return 0;
1015
1016 out_agresv:
1017 xfs_fs_unreserve_ag_blocks(mp);
1018 out_quota:
1019 xfs_qm_unmount_quotas(mp);
1020 out_rtunmount:
1021 xfs_rtunmount_inodes(mp);
1022 out_rele_rip:
1023 IRELE(rip);
1024 /* Clean out dquots that might be in memory after quotacheck. */
1025 xfs_qm_unmount(mp);
1026 /*
1027 * Cancel all delayed reclaim work and reclaim the inodes directly.
1028 * We have to do this /after/ rtunmount and qm_unmount because those
1029 * two will have scheduled delayed reclaim for the rt/quota inodes.
1030 *
1031 * This is slightly different from the unmountfs call sequence
1032 * because we could be tearing down a partially set up mount. In
1033 * particular, if log_mount_finish fails we bail out without calling
1034 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1035 * quota inodes.
1036 */
1037 cancel_delayed_work_sync(&mp->m_reclaim_work);
1038 xfs_reclaim_inodes(mp, SYNC_WAIT);
1039 out_log_dealloc:
1040 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1041 xfs_log_mount_cancel(mp);
1042 out_fail_wait:
1043 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1044 xfs_wait_buftarg(mp->m_logdev_targp);
1045 xfs_wait_buftarg(mp->m_ddev_targp);
1046 out_free_perag:
1047 xfs_free_perag(mp);
1048 out_free_dir:
1049 xfs_da_unmount(mp);
1050 out_remove_uuid:
1051 xfs_uuid_unmount(mp);
1052 out_remove_errortag:
1053 xfs_errortag_del(mp);
1054 out_remove_error_sysfs:
1055 xfs_error_sysfs_del(mp);
1056 out_del_stats:
1057 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1058 out_remove_sysfs:
1059 xfs_sysfs_del(&mp->m_kobj);
1060 out:
1061 return error;
1062}
1063
1064/*
1065 * This flushes out the inodes,dquots and the superblock, unmounts the
1066 * log and makes sure that incore structures are freed.
1067 */
1068void
1069xfs_unmountfs(
1070 struct xfs_mount *mp)
1071{
1072 uint64_t resblks;
1073 int error;
1074
1075 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1076 cancel_delayed_work_sync(&mp->m_cowblocks_work);
1077
1078 xfs_fs_unreserve_ag_blocks(mp);
1079 xfs_qm_unmount_quotas(mp);
1080 xfs_rtunmount_inodes(mp);
1081 IRELE(mp->m_rootip);
1082
1083 /*
1084 * We can potentially deadlock here if we have an inode cluster
1085 * that has been freed has its buffer still pinned in memory because
1086 * the transaction is still sitting in a iclog. The stale inodes
1087 * on that buffer will have their flush locks held until the
1088 * transaction hits the disk and the callbacks run. the inode
1089 * flush takes the flush lock unconditionally and with nothing to
1090 * push out the iclog we will never get that unlocked. hence we
1091 * need to force the log first.
1092 */
1093 xfs_log_force(mp, XFS_LOG_SYNC);
1094
1095 /*
1096 * Wait for all busy extents to be freed, including completion of
1097 * any discard operation.
1098 */
1099 xfs_extent_busy_wait_all(mp);
1100 flush_workqueue(xfs_discard_wq);
1101
1102 /*
1103 * We now need to tell the world we are unmounting. This will allow
1104 * us to detect that the filesystem is going away and we should error
1105 * out anything that we have been retrying in the background. This will
1106 * prevent neverending retries in AIL pushing from hanging the unmount.
1107 */
1108 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1109
1110 /*
1111 * Flush all pending changes from the AIL.
1112 */
1113 xfs_ail_push_all_sync(mp->m_ail);
1114
1115 /*
1116 * And reclaim all inodes. At this point there should be no dirty
1117 * inodes and none should be pinned or locked, but use synchronous
1118 * reclaim just to be sure. We can stop background inode reclaim
1119 * here as well if it is still running.
1120 */
1121 cancel_delayed_work_sync(&mp->m_reclaim_work);
1122 xfs_reclaim_inodes(mp, SYNC_WAIT);
1123
1124 xfs_qm_unmount(mp);
1125
1126 /*
1127 * Unreserve any blocks we have so that when we unmount we don't account
1128 * the reserved free space as used. This is really only necessary for
1129 * lazy superblock counting because it trusts the incore superblock
1130 * counters to be absolutely correct on clean unmount.
1131 *
1132 * We don't bother correcting this elsewhere for lazy superblock
1133 * counting because on mount of an unclean filesystem we reconstruct the
1134 * correct counter value and this is irrelevant.
1135 *
1136 * For non-lazy counter filesystems, this doesn't matter at all because
1137 * we only every apply deltas to the superblock and hence the incore
1138 * value does not matter....
1139 */
1140 resblks = 0;
1141 error = xfs_reserve_blocks(mp, &resblks, NULL);
1142 if (error)
1143 xfs_warn(mp, "Unable to free reserved block pool. "
1144 "Freespace may not be correct on next mount.");
1145
1146 error = xfs_log_sbcount(mp);
1147 if (error)
1148 xfs_warn(mp, "Unable to update superblock counters. "
1149 "Freespace may not be correct on next mount.");
1150
1151
1152 xfs_log_unmount(mp);
1153 xfs_da_unmount(mp);
1154 xfs_uuid_unmount(mp);
1155
1156#if defined(DEBUG)
1157 xfs_errortag_clearall(mp);
1158#endif
1159 xfs_free_perag(mp);
1160
1161 xfs_errortag_del(mp);
1162 xfs_error_sysfs_del(mp);
1163 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1164 xfs_sysfs_del(&mp->m_kobj);
1165}
1166
1167/*
1168 * Determine whether modifications can proceed. The caller specifies the minimum
1169 * freeze level for which modifications should not be allowed. This allows
1170 * certain operations to proceed while the freeze sequence is in progress, if
1171 * necessary.
1172 */
1173bool
1174xfs_fs_writable(
1175 struct xfs_mount *mp,
1176 int level)
1177{
1178 ASSERT(level > SB_UNFROZEN);
1179 if ((mp->m_super->s_writers.frozen >= level) ||
1180 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1181 return false;
1182
1183 return true;
1184}
1185
1186/*
1187 * xfs_log_sbcount
1188 *
1189 * Sync the superblock counters to disk.
1190 *
1191 * Note this code can be called during the process of freezing, so we use the
1192 * transaction allocator that does not block when the transaction subsystem is
1193 * in its frozen state.
1194 */
1195int
1196xfs_log_sbcount(xfs_mount_t *mp)
1197{
1198 /* allow this to proceed during the freeze sequence... */
1199 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1200 return 0;
1201
1202 /*
1203 * we don't need to do this if we are updating the superblock
1204 * counters on every modification.
1205 */
1206 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1207 return 0;
1208
1209 return xfs_sync_sb(mp, true);
1210}
1211
1212/*
1213 * Deltas for the inode count are +/-64, hence we use a large batch size
1214 * of 128 so we don't need to take the counter lock on every update.
1215 */
1216#define XFS_ICOUNT_BATCH 128
1217int
1218xfs_mod_icount(
1219 struct xfs_mount *mp,
1220 int64_t delta)
1221{
1222 percpu_counter_add_batch(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1223 if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1224 ASSERT(0);
1225 percpu_counter_add(&mp->m_icount, -delta);
1226 return -EINVAL;
1227 }
1228 return 0;
1229}
1230
1231int
1232xfs_mod_ifree(
1233 struct xfs_mount *mp,
1234 int64_t delta)
1235{
1236 percpu_counter_add(&mp->m_ifree, delta);
1237 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1238 ASSERT(0);
1239 percpu_counter_add(&mp->m_ifree, -delta);
1240 return -EINVAL;
1241 }
1242 return 0;
1243}
1244
1245/*
1246 * Deltas for the block count can vary from 1 to very large, but lock contention
1247 * only occurs on frequent small block count updates such as in the delayed
1248 * allocation path for buffered writes (page a time updates). Hence we set
1249 * a large batch count (1024) to minimise global counter updates except when
1250 * we get near to ENOSPC and we have to be very accurate with our updates.
1251 */
1252#define XFS_FDBLOCKS_BATCH 1024
1253int
1254xfs_mod_fdblocks(
1255 struct xfs_mount *mp,
1256 int64_t delta,
1257 bool rsvd)
1258{
1259 int64_t lcounter;
1260 long long res_used;
1261 s32 batch;
1262
1263 if (delta > 0) {
1264 /*
1265 * If the reserve pool is depleted, put blocks back into it
1266 * first. Most of the time the pool is full.
1267 */
1268 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1269 percpu_counter_add(&mp->m_fdblocks, delta);
1270 return 0;
1271 }
1272
1273 spin_lock(&mp->m_sb_lock);
1274 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1275
1276 if (res_used > delta) {
1277 mp->m_resblks_avail += delta;
1278 } else {
1279 delta -= res_used;
1280 mp->m_resblks_avail = mp->m_resblks;
1281 percpu_counter_add(&mp->m_fdblocks, delta);
1282 }
1283 spin_unlock(&mp->m_sb_lock);
1284 return 0;
1285 }
1286
1287 /*
1288 * Taking blocks away, need to be more accurate the closer we
1289 * are to zero.
1290 *
1291 * If the counter has a value of less than 2 * max batch size,
1292 * then make everything serialise as we are real close to
1293 * ENOSPC.
1294 */
1295 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1296 XFS_FDBLOCKS_BATCH) < 0)
1297 batch = 1;
1298 else
1299 batch = XFS_FDBLOCKS_BATCH;
1300
1301 percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1302 if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1303 XFS_FDBLOCKS_BATCH) >= 0) {
1304 /* we had space! */
1305 return 0;
1306 }
1307
1308 /*
1309 * lock up the sb for dipping into reserves before releasing the space
1310 * that took us to ENOSPC.
1311 */
1312 spin_lock(&mp->m_sb_lock);
1313 percpu_counter_add(&mp->m_fdblocks, -delta);
1314 if (!rsvd)
1315 goto fdblocks_enospc;
1316
1317 lcounter = (long long)mp->m_resblks_avail + delta;
1318 if (lcounter >= 0) {
1319 mp->m_resblks_avail = lcounter;
1320 spin_unlock(&mp->m_sb_lock);
1321 return 0;
1322 }
1323 printk_once(KERN_WARNING
1324 "Filesystem \"%s\": reserve blocks depleted! "
1325 "Consider increasing reserve pool size.",
1326 mp->m_fsname);
1327fdblocks_enospc:
1328 spin_unlock(&mp->m_sb_lock);
1329 return -ENOSPC;
1330}
1331
1332int
1333xfs_mod_frextents(
1334 struct xfs_mount *mp,
1335 int64_t delta)
1336{
1337 int64_t lcounter;
1338 int ret = 0;
1339
1340 spin_lock(&mp->m_sb_lock);
1341 lcounter = mp->m_sb.sb_frextents + delta;
1342 if (lcounter < 0)
1343 ret = -ENOSPC;
1344 else
1345 mp->m_sb.sb_frextents = lcounter;
1346 spin_unlock(&mp->m_sb_lock);
1347 return ret;
1348}
1349
1350/*
1351 * xfs_getsb() is called to obtain the buffer for the superblock.
1352 * The buffer is returned locked and read in from disk.
1353 * The buffer should be released with a call to xfs_brelse().
1354 *
1355 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1356 * the superblock buffer if it can be locked without sleeping.
1357 * If it can't then we'll return NULL.
1358 */
1359struct xfs_buf *
1360xfs_getsb(
1361 struct xfs_mount *mp,
1362 int flags)
1363{
1364 struct xfs_buf *bp = mp->m_sb_bp;
1365
1366 if (!xfs_buf_trylock(bp)) {
1367 if (flags & XBF_TRYLOCK)
1368 return NULL;
1369 xfs_buf_lock(bp);
1370 }
1371
1372 xfs_buf_hold(bp);
1373 ASSERT(bp->b_flags & XBF_DONE);
1374 return bp;
1375}
1376
1377/*
1378 * Used to free the superblock along various error paths.
1379 */
1380void
1381xfs_freesb(
1382 struct xfs_mount *mp)
1383{
1384 struct xfs_buf *bp = mp->m_sb_bp;
1385
1386 xfs_buf_lock(bp);
1387 mp->m_sb_bp = NULL;
1388 xfs_buf_relse(bp);
1389}
1390
1391/*
1392 * If the underlying (data/log/rt) device is readonly, there are some
1393 * operations that cannot proceed.
1394 */
1395int
1396xfs_dev_is_read_only(
1397 struct xfs_mount *mp,
1398 char *message)
1399{
1400 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1401 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1402 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1403 xfs_notice(mp, "%s required on read-only device.", message);
1404 xfs_notice(mp, "write access unavailable, cannot proceed.");
1405 return -EROFS;
1406 }
1407 return 0;
1408}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_bit.h"
13#include "xfs_sb.h"
14#include "xfs_mount.h"
15#include "xfs_inode.h"
16#include "xfs_dir2.h"
17#include "xfs_ialloc.h"
18#include "xfs_alloc.h"
19#include "xfs_rtalloc.h"
20#include "xfs_bmap.h"
21#include "xfs_trans.h"
22#include "xfs_trans_priv.h"
23#include "xfs_log.h"
24#include "xfs_log_priv.h"
25#include "xfs_error.h"
26#include "xfs_quota.h"
27#include "xfs_fsops.h"
28#include "xfs_icache.h"
29#include "xfs_sysfs.h"
30#include "xfs_rmap_btree.h"
31#include "xfs_refcount_btree.h"
32#include "xfs_reflink.h"
33#include "xfs_extent_busy.h"
34#include "xfs_health.h"
35#include "xfs_trace.h"
36#include "xfs_ag.h"
37#include "xfs_rtbitmap.h"
38#include "xfs_metafile.h"
39#include "xfs_rtgroup.h"
40#include "scrub/stats.h"
41
42static DEFINE_MUTEX(xfs_uuid_table_mutex);
43static int xfs_uuid_table_size;
44static uuid_t *xfs_uuid_table;
45
46void
47xfs_uuid_table_free(void)
48{
49 if (xfs_uuid_table_size == 0)
50 return;
51 kfree(xfs_uuid_table);
52 xfs_uuid_table = NULL;
53 xfs_uuid_table_size = 0;
54}
55
56/*
57 * See if the UUID is unique among mounted XFS filesystems.
58 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
59 */
60STATIC int
61xfs_uuid_mount(
62 struct xfs_mount *mp)
63{
64 uuid_t *uuid = &mp->m_sb.sb_uuid;
65 int hole, i;
66
67 /* Publish UUID in struct super_block */
68 super_set_uuid(mp->m_super, uuid->b, sizeof(*uuid));
69
70 if (xfs_has_nouuid(mp))
71 return 0;
72
73 if (uuid_is_null(uuid)) {
74 xfs_warn(mp, "Filesystem has null UUID - can't mount");
75 return -EINVAL;
76 }
77
78 mutex_lock(&xfs_uuid_table_mutex);
79 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
80 if (uuid_is_null(&xfs_uuid_table[i])) {
81 hole = i;
82 continue;
83 }
84 if (uuid_equal(uuid, &xfs_uuid_table[i]))
85 goto out_duplicate;
86 }
87
88 if (hole < 0) {
89 xfs_uuid_table = krealloc(xfs_uuid_table,
90 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
91 GFP_KERNEL | __GFP_NOFAIL);
92 hole = xfs_uuid_table_size++;
93 }
94 xfs_uuid_table[hole] = *uuid;
95 mutex_unlock(&xfs_uuid_table_mutex);
96
97 return 0;
98
99 out_duplicate:
100 mutex_unlock(&xfs_uuid_table_mutex);
101 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
102 return -EINVAL;
103}
104
105STATIC void
106xfs_uuid_unmount(
107 struct xfs_mount *mp)
108{
109 uuid_t *uuid = &mp->m_sb.sb_uuid;
110 int i;
111
112 if (xfs_has_nouuid(mp))
113 return;
114
115 mutex_lock(&xfs_uuid_table_mutex);
116 for (i = 0; i < xfs_uuid_table_size; i++) {
117 if (uuid_is_null(&xfs_uuid_table[i]))
118 continue;
119 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
120 continue;
121 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
122 break;
123 }
124 ASSERT(i < xfs_uuid_table_size);
125 mutex_unlock(&xfs_uuid_table_mutex);
126}
127
128/*
129 * Check size of device based on the (data/realtime) block count.
130 * Note: this check is used by the growfs code as well as mount.
131 */
132int
133xfs_sb_validate_fsb_count(
134 xfs_sb_t *sbp,
135 uint64_t nblocks)
136{
137 uint64_t max_bytes;
138
139 ASSERT(sbp->sb_blocklog >= BBSHIFT);
140
141 if (check_shl_overflow(nblocks, sbp->sb_blocklog, &max_bytes))
142 return -EFBIG;
143
144 /* Limited by ULONG_MAX of page cache index */
145 if (max_bytes >> PAGE_SHIFT > ULONG_MAX)
146 return -EFBIG;
147 return 0;
148}
149
150/*
151 * xfs_readsb
152 *
153 * Does the initial read of the superblock.
154 */
155int
156xfs_readsb(
157 struct xfs_mount *mp,
158 int flags)
159{
160 unsigned int sector_size;
161 struct xfs_buf *bp;
162 struct xfs_sb *sbp = &mp->m_sb;
163 int error;
164 int loud = !(flags & XFS_MFSI_QUIET);
165 const struct xfs_buf_ops *buf_ops;
166
167 ASSERT(mp->m_sb_bp == NULL);
168 ASSERT(mp->m_ddev_targp != NULL);
169
170 /*
171 * For the initial read, we must guess at the sector
172 * size based on the block device. It's enough to
173 * get the sb_sectsize out of the superblock and
174 * then reread with the proper length.
175 * We don't verify it yet, because it may not be complete.
176 */
177 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
178 buf_ops = NULL;
179
180 /*
181 * Allocate a (locked) buffer to hold the superblock. This will be kept
182 * around at all times to optimize access to the superblock. Therefore,
183 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
184 * elevated.
185 */
186reread:
187 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
188 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
189 buf_ops);
190 if (error) {
191 if (loud)
192 xfs_warn(mp, "SB validate failed with error %d.", error);
193 /* bad CRC means corrupted metadata */
194 if (error == -EFSBADCRC)
195 error = -EFSCORRUPTED;
196 return error;
197 }
198
199 /*
200 * Initialize the mount structure from the superblock.
201 */
202 xfs_sb_from_disk(sbp, bp->b_addr);
203
204 /*
205 * If we haven't validated the superblock, do so now before we try
206 * to check the sector size and reread the superblock appropriately.
207 */
208 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
209 if (loud)
210 xfs_warn(mp, "Invalid superblock magic number");
211 error = -EINVAL;
212 goto release_buf;
213 }
214
215 /*
216 * We must be able to do sector-sized and sector-aligned IO.
217 */
218 if (sector_size > sbp->sb_sectsize) {
219 if (loud)
220 xfs_warn(mp, "device supports %u byte sectors (not %u)",
221 sector_size, sbp->sb_sectsize);
222 error = -ENOSYS;
223 goto release_buf;
224 }
225
226 if (buf_ops == NULL) {
227 /*
228 * Re-read the superblock so the buffer is correctly sized,
229 * and properly verified.
230 */
231 xfs_buf_relse(bp);
232 sector_size = sbp->sb_sectsize;
233 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
234 goto reread;
235 }
236
237 mp->m_features |= xfs_sb_version_to_features(sbp);
238 xfs_reinit_percpu_counters(mp);
239
240 /*
241 * If logged xattrs are enabled after log recovery finishes, then set
242 * the opstate so that log recovery will work properly.
243 */
244 if (xfs_sb_version_haslogxattrs(&mp->m_sb))
245 xfs_set_using_logged_xattrs(mp);
246
247 /* no need to be quiet anymore, so reset the buf ops */
248 bp->b_ops = &xfs_sb_buf_ops;
249
250 mp->m_sb_bp = bp;
251 xfs_buf_unlock(bp);
252 return 0;
253
254release_buf:
255 xfs_buf_relse(bp);
256 return error;
257}
258
259/*
260 * If the sunit/swidth change would move the precomputed root inode value, we
261 * must reject the ondisk change because repair will stumble over that.
262 * However, we allow the mount to proceed because we never rejected this
263 * combination before. Returns true to update the sb, false otherwise.
264 */
265static inline int
266xfs_check_new_dalign(
267 struct xfs_mount *mp,
268 int new_dalign,
269 bool *update_sb)
270{
271 struct xfs_sb *sbp = &mp->m_sb;
272 xfs_ino_t calc_ino;
273
274 calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
275 trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
276
277 if (sbp->sb_rootino == calc_ino) {
278 *update_sb = true;
279 return 0;
280 }
281
282 xfs_warn(mp,
283"Cannot change stripe alignment; would require moving root inode.");
284
285 /*
286 * XXX: Next time we add a new incompat feature, this should start
287 * returning -EINVAL to fail the mount. Until then, spit out a warning
288 * that we're ignoring the administrator's instructions.
289 */
290 xfs_warn(mp, "Skipping superblock stripe alignment update.");
291 *update_sb = false;
292 return 0;
293}
294
295/*
296 * If we were provided with new sunit/swidth values as mount options, make sure
297 * that they pass basic alignment and superblock feature checks, and convert
298 * them into the same units (FSB) that everything else expects. This step
299 * /must/ be done before computing the inode geometry.
300 */
301STATIC int
302xfs_validate_new_dalign(
303 struct xfs_mount *mp)
304{
305 if (mp->m_dalign == 0)
306 return 0;
307
308 /*
309 * If stripe unit and stripe width are not multiples
310 * of the fs blocksize turn off alignment.
311 */
312 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
313 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
314 xfs_warn(mp,
315 "alignment check failed: sunit/swidth vs. blocksize(%d)",
316 mp->m_sb.sb_blocksize);
317 return -EINVAL;
318 }
319
320 /*
321 * Convert the stripe unit and width to FSBs.
322 */
323 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
324 if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
325 xfs_warn(mp,
326 "alignment check failed: sunit/swidth vs. agsize(%d)",
327 mp->m_sb.sb_agblocks);
328 return -EINVAL;
329 }
330
331 if (!mp->m_dalign) {
332 xfs_warn(mp,
333 "alignment check failed: sunit(%d) less than bsize(%d)",
334 mp->m_dalign, mp->m_sb.sb_blocksize);
335 return -EINVAL;
336 }
337
338 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
339
340 if (!xfs_has_dalign(mp)) {
341 xfs_warn(mp,
342"cannot change alignment: superblock does not support data alignment");
343 return -EINVAL;
344 }
345
346 return 0;
347}
348
349/* Update alignment values based on mount options and sb values. */
350STATIC int
351xfs_update_alignment(
352 struct xfs_mount *mp)
353{
354 struct xfs_sb *sbp = &mp->m_sb;
355
356 if (mp->m_dalign) {
357 bool update_sb;
358 int error;
359
360 if (sbp->sb_unit == mp->m_dalign &&
361 sbp->sb_width == mp->m_swidth)
362 return 0;
363
364 error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
365 if (error || !update_sb)
366 return error;
367
368 sbp->sb_unit = mp->m_dalign;
369 sbp->sb_width = mp->m_swidth;
370 mp->m_update_sb = true;
371 } else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) {
372 mp->m_dalign = sbp->sb_unit;
373 mp->m_swidth = sbp->sb_width;
374 }
375
376 return 0;
377}
378
379/*
380 * precalculate the low space thresholds for dynamic speculative preallocation.
381 */
382void
383xfs_set_low_space_thresholds(
384 struct xfs_mount *mp)
385{
386 uint64_t dblocks = mp->m_sb.sb_dblocks;
387 uint64_t rtexts = mp->m_sb.sb_rextents;
388 int i;
389
390 do_div(dblocks, 100);
391 do_div(rtexts, 100);
392
393 for (i = 0; i < XFS_LOWSP_MAX; i++) {
394 mp->m_low_space[i] = dblocks * (i + 1);
395 mp->m_low_rtexts[i] = rtexts * (i + 1);
396 }
397}
398
399/*
400 * Check that the data (and log if separate) is an ok size.
401 */
402STATIC int
403xfs_check_sizes(
404 struct xfs_mount *mp)
405{
406 struct xfs_buf *bp;
407 xfs_daddr_t d;
408 int error;
409
410 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
411 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
412 xfs_warn(mp, "filesystem size mismatch detected");
413 return -EFBIG;
414 }
415 error = xfs_buf_read_uncached(mp->m_ddev_targp,
416 d - XFS_FSS_TO_BB(mp, 1),
417 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
418 if (error) {
419 xfs_warn(mp, "last sector read failed");
420 return error;
421 }
422 xfs_buf_relse(bp);
423
424 if (mp->m_logdev_targp == mp->m_ddev_targp)
425 return 0;
426
427 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
428 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
429 xfs_warn(mp, "log size mismatch detected");
430 return -EFBIG;
431 }
432 error = xfs_buf_read_uncached(mp->m_logdev_targp,
433 d - XFS_FSB_TO_BB(mp, 1),
434 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
435 if (error) {
436 xfs_warn(mp, "log device read failed");
437 return error;
438 }
439 xfs_buf_relse(bp);
440 return 0;
441}
442
443/*
444 * Clear the quotaflags in memory and in the superblock.
445 */
446int
447xfs_mount_reset_sbqflags(
448 struct xfs_mount *mp)
449{
450 mp->m_qflags = 0;
451
452 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
453 if (mp->m_sb.sb_qflags == 0)
454 return 0;
455 spin_lock(&mp->m_sb_lock);
456 mp->m_sb.sb_qflags = 0;
457 spin_unlock(&mp->m_sb_lock);
458
459 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
460 return 0;
461
462 return xfs_sync_sb(mp, false);
463}
464
465uint64_t
466xfs_default_resblks(xfs_mount_t *mp)
467{
468 uint64_t resblks;
469
470 /*
471 * We default to 5% or 8192 fsbs of space reserved, whichever is
472 * smaller. This is intended to cover concurrent allocation
473 * transactions when we initially hit enospc. These each require a 4
474 * block reservation. Hence by default we cover roughly 2000 concurrent
475 * allocation reservations.
476 */
477 resblks = mp->m_sb.sb_dblocks;
478 do_div(resblks, 20);
479 resblks = min_t(uint64_t, resblks, 8192);
480 return resblks;
481}
482
483/* Ensure the summary counts are correct. */
484STATIC int
485xfs_check_summary_counts(
486 struct xfs_mount *mp)
487{
488 int error = 0;
489
490 /*
491 * The AG0 superblock verifier rejects in-progress filesystems,
492 * so we should never see the flag set this far into mounting.
493 */
494 if (mp->m_sb.sb_inprogress) {
495 xfs_err(mp, "sb_inprogress set after log recovery??");
496 WARN_ON(1);
497 return -EFSCORRUPTED;
498 }
499
500 /*
501 * Now the log is mounted, we know if it was an unclean shutdown or
502 * not. If it was, with the first phase of recovery has completed, we
503 * have consistent AG blocks on disk. We have not recovered EFIs yet,
504 * but they are recovered transactionally in the second recovery phase
505 * later.
506 *
507 * If the log was clean when we mounted, we can check the summary
508 * counters. If any of them are obviously incorrect, we can recompute
509 * them from the AGF headers in the next step.
510 */
511 if (xfs_is_clean(mp) &&
512 (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
513 !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
514 mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
515 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
516
517 /*
518 * We can safely re-initialise incore superblock counters from the
519 * per-ag data. These may not be correct if the filesystem was not
520 * cleanly unmounted, so we waited for recovery to finish before doing
521 * this.
522 *
523 * If the filesystem was cleanly unmounted or the previous check did
524 * not flag anything weird, then we can trust the values in the
525 * superblock to be correct and we don't need to do anything here.
526 * Otherwise, recalculate the summary counters.
527 */
528 if ((xfs_has_lazysbcount(mp) && !xfs_is_clean(mp)) ||
529 xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) {
530 error = xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
531 if (error)
532 return error;
533 }
534
535 /*
536 * Older kernels misused sb_frextents to reflect both incore
537 * reservations made by running transactions and the actual count of
538 * free rt extents in the ondisk metadata. Transactions committed
539 * during runtime can therefore contain a superblock update that
540 * undercounts the number of free rt extents tracked in the rt bitmap.
541 * A clean unmount record will have the correct frextents value since
542 * there can be no other transactions running at that point.
543 *
544 * If we're mounting the rt volume after recovering the log, recompute
545 * frextents from the rtbitmap file to fix the inconsistency.
546 */
547 if (xfs_has_realtime(mp) && !xfs_is_clean(mp)) {
548 error = xfs_rtalloc_reinit_frextents(mp);
549 if (error)
550 return error;
551 }
552
553 return 0;
554}
555
556static void
557xfs_unmount_check(
558 struct xfs_mount *mp)
559{
560 if (xfs_is_shutdown(mp))
561 return;
562
563 if (percpu_counter_sum(&mp->m_ifree) >
564 percpu_counter_sum(&mp->m_icount)) {
565 xfs_alert(mp, "ifree/icount mismatch at unmount");
566 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
567 }
568}
569
570/*
571 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
572 * internal inode structures can be sitting in the CIL and AIL at this point,
573 * so we need to unpin them, write them back and/or reclaim them before unmount
574 * can proceed. In other words, callers are required to have inactivated all
575 * inodes.
576 *
577 * An inode cluster that has been freed can have its buffer still pinned in
578 * memory because the transaction is still sitting in a iclog. The stale inodes
579 * on that buffer will be pinned to the buffer until the transaction hits the
580 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
581 * may never see the pinned buffer, so nothing will push out the iclog and
582 * unpin the buffer.
583 *
584 * Hence we need to force the log to unpin everything first. However, log
585 * forces don't wait for the discards they issue to complete, so we have to
586 * explicitly wait for them to complete here as well.
587 *
588 * Then we can tell the world we are unmounting so that error handling knows
589 * that the filesystem is going away and we should error out anything that we
590 * have been retrying in the background. This will prevent never-ending
591 * retries in AIL pushing from hanging the unmount.
592 *
593 * Finally, we can push the AIL to clean all the remaining dirty objects, then
594 * reclaim the remaining inodes that are still in memory at this point in time.
595 */
596static void
597xfs_unmount_flush_inodes(
598 struct xfs_mount *mp)
599{
600 xfs_log_force(mp, XFS_LOG_SYNC);
601 xfs_extent_busy_wait_all(mp);
602 flush_workqueue(xfs_discard_wq);
603
604 xfs_set_unmounting(mp);
605
606 xfs_ail_push_all_sync(mp->m_ail);
607 xfs_inodegc_stop(mp);
608 cancel_delayed_work_sync(&mp->m_reclaim_work);
609 xfs_reclaim_inodes(mp);
610 xfs_health_unmount(mp);
611}
612
613static void
614xfs_mount_setup_inode_geom(
615 struct xfs_mount *mp)
616{
617 struct xfs_ino_geometry *igeo = M_IGEO(mp);
618
619 igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
620 ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
621
622 xfs_ialloc_setup_geometry(mp);
623}
624
625/* Mount the metadata directory tree root. */
626STATIC int
627xfs_mount_setup_metadir(
628 struct xfs_mount *mp)
629{
630 int error;
631
632 /* Load the metadata directory root inode into memory. */
633 error = xfs_metafile_iget(mp, mp->m_sb.sb_metadirino, XFS_METAFILE_DIR,
634 &mp->m_metadirip);
635 if (error)
636 xfs_warn(mp, "Failed to load metadir root directory, error %d",
637 error);
638 return error;
639}
640
641/* Compute maximum possible height for per-AG btree types for this fs. */
642static inline void
643xfs_agbtree_compute_maxlevels(
644 struct xfs_mount *mp)
645{
646 unsigned int levels;
647
648 levels = max(mp->m_alloc_maxlevels, M_IGEO(mp)->inobt_maxlevels);
649 levels = max(levels, mp->m_rmap_maxlevels);
650 mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels);
651}
652
653/*
654 * This function does the following on an initial mount of a file system:
655 * - reads the superblock from disk and init the mount struct
656 * - if we're a 32-bit kernel, do a size check on the superblock
657 * so we don't mount terabyte filesystems
658 * - init mount struct realtime fields
659 * - allocate inode hash table for fs
660 * - init directory manager
661 * - perform recovery and init the log manager
662 */
663int
664xfs_mountfs(
665 struct xfs_mount *mp)
666{
667 struct xfs_sb *sbp = &(mp->m_sb);
668 struct xfs_inode *rip;
669 struct xfs_ino_geometry *igeo = M_IGEO(mp);
670 uint quotamount = 0;
671 uint quotaflags = 0;
672 int error = 0;
673
674 xfs_sb_mount_common(mp, sbp);
675
676 /*
677 * Check for a mismatched features2 values. Older kernels read & wrote
678 * into the wrong sb offset for sb_features2 on some platforms due to
679 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
680 * which made older superblock reading/writing routines swap it as a
681 * 64-bit value.
682 *
683 * For backwards compatibility, we make both slots equal.
684 *
685 * If we detect a mismatched field, we OR the set bits into the existing
686 * features2 field in case it has already been modified; we don't want
687 * to lose any features. We then update the bad location with the ORed
688 * value so that older kernels will see any features2 flags. The
689 * superblock writeback code ensures the new sb_features2 is copied to
690 * sb_bad_features2 before it is logged or written to disk.
691 */
692 if (xfs_sb_has_mismatched_features2(sbp)) {
693 xfs_warn(mp, "correcting sb_features alignment problem");
694 sbp->sb_features2 |= sbp->sb_bad_features2;
695 mp->m_update_sb = true;
696 }
697
698
699 /* always use v2 inodes by default now */
700 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
701 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
702 mp->m_features |= XFS_FEAT_NLINK;
703 mp->m_update_sb = true;
704 }
705
706 /*
707 * If we were given new sunit/swidth options, do some basic validation
708 * checks and convert the incore dalign and swidth values to the
709 * same units (FSB) that everything else uses. This /must/ happen
710 * before computing the inode geometry.
711 */
712 error = xfs_validate_new_dalign(mp);
713 if (error)
714 goto out;
715
716 xfs_alloc_compute_maxlevels(mp);
717 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
718 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
719 xfs_mount_setup_inode_geom(mp);
720 xfs_rmapbt_compute_maxlevels(mp);
721 xfs_refcountbt_compute_maxlevels(mp);
722
723 xfs_agbtree_compute_maxlevels(mp);
724
725 /*
726 * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
727 * is NOT aligned turn off m_dalign since allocator alignment is within
728 * an ag, therefore ag has to be aligned at stripe boundary. Note that
729 * we must compute the free space and rmap btree geometry before doing
730 * this.
731 */
732 error = xfs_update_alignment(mp);
733 if (error)
734 goto out;
735
736 /* enable fail_at_unmount as default */
737 mp->m_fail_unmount = true;
738
739 super_set_sysfs_name_id(mp->m_super);
740
741 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
742 NULL, mp->m_super->s_id);
743 if (error)
744 goto out;
745
746 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
747 &mp->m_kobj, "stats");
748 if (error)
749 goto out_remove_sysfs;
750
751 xchk_stats_register(mp->m_scrub_stats, mp->m_debugfs);
752
753 error = xfs_error_sysfs_init(mp);
754 if (error)
755 goto out_remove_scrub_stats;
756
757 error = xfs_errortag_init(mp);
758 if (error)
759 goto out_remove_error_sysfs;
760
761 error = xfs_uuid_mount(mp);
762 if (error)
763 goto out_remove_errortag;
764
765 /*
766 * Update the preferred write size based on the information from the
767 * on-disk superblock.
768 */
769 mp->m_allocsize_log =
770 max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
771 mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
772
773 /* set the low space thresholds for dynamic preallocation */
774 xfs_set_low_space_thresholds(mp);
775
776 /*
777 * If enabled, sparse inode chunk alignment is expected to match the
778 * cluster size. Full inode chunk alignment must match the chunk size,
779 * but that is checked on sb read verification...
780 */
781 if (xfs_has_sparseinodes(mp) &&
782 mp->m_sb.sb_spino_align !=
783 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
784 xfs_warn(mp,
785 "Sparse inode block alignment (%u) must match cluster size (%llu).",
786 mp->m_sb.sb_spino_align,
787 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
788 error = -EINVAL;
789 goto out_remove_uuid;
790 }
791
792 /*
793 * Check that the data (and log if separate) is an ok size.
794 */
795 error = xfs_check_sizes(mp);
796 if (error)
797 goto out_remove_uuid;
798
799 /*
800 * Initialize realtime fields in the mount structure
801 */
802 error = xfs_rtmount_init(mp);
803 if (error) {
804 xfs_warn(mp, "RT mount failed");
805 goto out_remove_uuid;
806 }
807
808 /*
809 * Copies the low order bits of the timestamp and the randomly
810 * set "sequence" number out of a UUID.
811 */
812 mp->m_fixedfsid[0] =
813 (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
814 get_unaligned_be16(&sbp->sb_uuid.b[4]);
815 mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
816
817 error = xfs_da_mount(mp);
818 if (error) {
819 xfs_warn(mp, "Failed dir/attr init: %d", error);
820 goto out_remove_uuid;
821 }
822
823 /*
824 * Initialize the precomputed transaction reservations values.
825 */
826 xfs_trans_init(mp);
827
828 /*
829 * Allocate and initialize the per-ag data.
830 */
831 error = xfs_initialize_perag(mp, 0, sbp->sb_agcount,
832 mp->m_sb.sb_dblocks, &mp->m_maxagi);
833 if (error) {
834 xfs_warn(mp, "Failed per-ag init: %d", error);
835 goto out_free_dir;
836 }
837
838 error = xfs_initialize_rtgroups(mp, 0, sbp->sb_rgcount,
839 mp->m_sb.sb_rextents);
840 if (error) {
841 xfs_warn(mp, "Failed rtgroup init: %d", error);
842 goto out_free_perag;
843 }
844
845 if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
846 xfs_warn(mp, "no log defined");
847 error = -EFSCORRUPTED;
848 goto out_free_rtgroup;
849 }
850
851 error = xfs_inodegc_register_shrinker(mp);
852 if (error)
853 goto out_fail_wait;
854
855 /*
856 * If we're resuming quota status, pick up the preliminary qflags from
857 * the ondisk superblock so that we know if we should recover dquots.
858 */
859 if (xfs_is_resuming_quotaon(mp))
860 xfs_qm_resume_quotaon(mp);
861
862 /*
863 * Log's mount-time initialization. The first part of recovery can place
864 * some items on the AIL, to be handled when recovery is finished or
865 * cancelled.
866 */
867 error = xfs_log_mount(mp, mp->m_logdev_targp,
868 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
869 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
870 if (error) {
871 xfs_warn(mp, "log mount failed");
872 goto out_inodegc_shrinker;
873 }
874
875 /*
876 * If we're resuming quota status and recovered the log, re-sample the
877 * qflags from the ondisk superblock now that we've recovered it, just
878 * in case someone shut down enforcement just before a crash.
879 */
880 if (xfs_clear_resuming_quotaon(mp) && xlog_recovery_needed(mp->m_log))
881 xfs_qm_resume_quotaon(mp);
882
883 /*
884 * If logged xattrs are still enabled after log recovery finishes, then
885 * they'll be available until unmount. Otherwise, turn them off.
886 */
887 if (xfs_sb_version_haslogxattrs(&mp->m_sb))
888 xfs_set_using_logged_xattrs(mp);
889 else
890 xfs_clear_using_logged_xattrs(mp);
891
892 /* Enable background inode inactivation workers. */
893 xfs_inodegc_start(mp);
894 xfs_blockgc_start(mp);
895
896 /*
897 * Now that we've recovered any pending superblock feature bit
898 * additions, we can finish setting up the attr2 behaviour for the
899 * mount. The noattr2 option overrides the superblock flag, so only
900 * check the superblock feature flag if the mount option is not set.
901 */
902 if (xfs_has_noattr2(mp)) {
903 mp->m_features &= ~XFS_FEAT_ATTR2;
904 } else if (!xfs_has_attr2(mp) &&
905 (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)) {
906 mp->m_features |= XFS_FEAT_ATTR2;
907 }
908
909 if (xfs_has_metadir(mp)) {
910 error = xfs_mount_setup_metadir(mp);
911 if (error)
912 goto out_free_metadir;
913 }
914
915 /*
916 * Get and sanity-check the root inode.
917 * Save the pointer to it in the mount structure.
918 */
919 error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
920 XFS_ILOCK_EXCL, &rip);
921 if (error) {
922 xfs_warn(mp,
923 "Failed to read root inode 0x%llx, error %d",
924 sbp->sb_rootino, -error);
925 goto out_free_metadir;
926 }
927
928 ASSERT(rip != NULL);
929
930 if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
931 xfs_warn(mp, "corrupted root inode %llu: not a directory",
932 (unsigned long long)rip->i_ino);
933 xfs_iunlock(rip, XFS_ILOCK_EXCL);
934 error = -EFSCORRUPTED;
935 goto out_rele_rip;
936 }
937 mp->m_rootip = rip; /* save it */
938
939 xfs_iunlock(rip, XFS_ILOCK_EXCL);
940
941 /*
942 * Initialize realtime inode pointers in the mount structure
943 */
944 error = xfs_rtmount_inodes(mp);
945 if (error) {
946 /*
947 * Free up the root inode.
948 */
949 xfs_warn(mp, "failed to read RT inodes");
950 goto out_rele_rip;
951 }
952
953 /* Make sure the summary counts are ok. */
954 error = xfs_check_summary_counts(mp);
955 if (error)
956 goto out_rtunmount;
957
958 /*
959 * If this is a read-only mount defer the superblock updates until
960 * the next remount into writeable mode. Otherwise we would never
961 * perform the update e.g. for the root filesystem.
962 */
963 if (mp->m_update_sb && !xfs_is_readonly(mp)) {
964 error = xfs_sync_sb(mp, false);
965 if (error) {
966 xfs_warn(mp, "failed to write sb changes");
967 goto out_rtunmount;
968 }
969 }
970
971 /*
972 * Initialise the XFS quota management subsystem for this mount
973 */
974 if (XFS_IS_QUOTA_ON(mp)) {
975 error = xfs_qm_newmount(mp, "amount, "aflags);
976 if (error)
977 goto out_rtunmount;
978 } else {
979 /*
980 * If a file system had quotas running earlier, but decided to
981 * mount without -o uquota/pquota/gquota options, revoke the
982 * quotachecked license.
983 */
984 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
985 xfs_notice(mp, "resetting quota flags");
986 error = xfs_mount_reset_sbqflags(mp);
987 if (error)
988 goto out_rtunmount;
989 }
990 }
991
992 /*
993 * Finish recovering the file system. This part needed to be delayed
994 * until after the root and real-time bitmap inodes were consistently
995 * read in. Temporarily create per-AG space reservations for metadata
996 * btree shape changes because space freeing transactions (for inode
997 * inactivation) require the per-AG reservation in lieu of reserving
998 * blocks.
999 */
1000 error = xfs_fs_reserve_ag_blocks(mp);
1001 if (error && error == -ENOSPC)
1002 xfs_warn(mp,
1003 "ENOSPC reserving per-AG metadata pool, log recovery may fail.");
1004 error = xfs_log_mount_finish(mp);
1005 xfs_fs_unreserve_ag_blocks(mp);
1006 if (error) {
1007 xfs_warn(mp, "log mount finish failed");
1008 goto out_rtunmount;
1009 }
1010
1011 /*
1012 * Now the log is fully replayed, we can transition to full read-only
1013 * mode for read-only mounts. This will sync all the metadata and clean
1014 * the log so that the recovery we just performed does not have to be
1015 * replayed again on the next mount.
1016 *
1017 * We use the same quiesce mechanism as the rw->ro remount, as they are
1018 * semantically identical operations.
1019 */
1020 if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp))
1021 xfs_log_clean(mp);
1022
1023 /*
1024 * Complete the quota initialisation, post-log-replay component.
1025 */
1026 if (quotamount) {
1027 ASSERT(mp->m_qflags == 0);
1028 mp->m_qflags = quotaflags;
1029
1030 xfs_qm_mount_quotas(mp);
1031 }
1032
1033 /*
1034 * Now we are mounted, reserve a small amount of unused space for
1035 * privileged transactions. This is needed so that transaction
1036 * space required for critical operations can dip into this pool
1037 * when at ENOSPC. This is needed for operations like create with
1038 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1039 * are not allowed to use this reserved space.
1040 *
1041 * This may drive us straight to ENOSPC on mount, but that implies
1042 * we were already there on the last unmount. Warn if this occurs.
1043 */
1044 if (!xfs_is_readonly(mp)) {
1045 error = xfs_reserve_blocks(mp, xfs_default_resblks(mp));
1046 if (error)
1047 xfs_warn(mp,
1048 "Unable to allocate reserve blocks. Continuing without reserve pool.");
1049
1050 /* Reserve AG blocks for future btree expansion. */
1051 error = xfs_fs_reserve_ag_blocks(mp);
1052 if (error && error != -ENOSPC)
1053 goto out_agresv;
1054 }
1055
1056 return 0;
1057
1058 out_agresv:
1059 xfs_fs_unreserve_ag_blocks(mp);
1060 xfs_qm_unmount_quotas(mp);
1061 out_rtunmount:
1062 xfs_rtunmount_inodes(mp);
1063 out_rele_rip:
1064 xfs_irele(rip);
1065 /* Clean out dquots that might be in memory after quotacheck. */
1066 xfs_qm_unmount(mp);
1067 out_free_metadir:
1068 if (mp->m_metadirip)
1069 xfs_irele(mp->m_metadirip);
1070
1071 /*
1072 * Inactivate all inodes that might still be in memory after a log
1073 * intent recovery failure so that reclaim can free them. Metadata
1074 * inodes and the root directory shouldn't need inactivation, but the
1075 * mount failed for some reason, so pull down all the state and flee.
1076 */
1077 xfs_inodegc_flush(mp);
1078
1079 /*
1080 * Flush all inode reclamation work and flush the log.
1081 * We have to do this /after/ rtunmount and qm_unmount because those
1082 * two will have scheduled delayed reclaim for the rt/quota inodes.
1083 *
1084 * This is slightly different from the unmountfs call sequence
1085 * because we could be tearing down a partially set up mount. In
1086 * particular, if log_mount_finish fails we bail out without calling
1087 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1088 * quota inodes.
1089 */
1090 xfs_unmount_flush_inodes(mp);
1091 xfs_log_mount_cancel(mp);
1092 out_inodegc_shrinker:
1093 shrinker_free(mp->m_inodegc_shrinker);
1094 out_fail_wait:
1095 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1096 xfs_buftarg_drain(mp->m_logdev_targp);
1097 xfs_buftarg_drain(mp->m_ddev_targp);
1098 out_free_rtgroup:
1099 xfs_free_rtgroups(mp, 0, mp->m_sb.sb_rgcount);
1100 out_free_perag:
1101 xfs_free_perag_range(mp, 0, mp->m_sb.sb_agcount);
1102 out_free_dir:
1103 xfs_da_unmount(mp);
1104 out_remove_uuid:
1105 xfs_uuid_unmount(mp);
1106 out_remove_errortag:
1107 xfs_errortag_del(mp);
1108 out_remove_error_sysfs:
1109 xfs_error_sysfs_del(mp);
1110 out_remove_scrub_stats:
1111 xchk_stats_unregister(mp->m_scrub_stats);
1112 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1113 out_remove_sysfs:
1114 xfs_sysfs_del(&mp->m_kobj);
1115 out:
1116 return error;
1117}
1118
1119/*
1120 * This flushes out the inodes,dquots and the superblock, unmounts the
1121 * log and makes sure that incore structures are freed.
1122 */
1123void
1124xfs_unmountfs(
1125 struct xfs_mount *mp)
1126{
1127 int error;
1128
1129 /*
1130 * Perform all on-disk metadata updates required to inactivate inodes
1131 * that the VFS evicted earlier in the unmount process. Freeing inodes
1132 * and discarding CoW fork preallocations can cause shape changes to
1133 * the free inode and refcount btrees, respectively, so we must finish
1134 * this before we discard the metadata space reservations. Metadata
1135 * inodes and the root directory do not require inactivation.
1136 */
1137 xfs_inodegc_flush(mp);
1138
1139 xfs_blockgc_stop(mp);
1140 xfs_fs_unreserve_ag_blocks(mp);
1141 xfs_qm_unmount_quotas(mp);
1142 xfs_rtunmount_inodes(mp);
1143 xfs_irele(mp->m_rootip);
1144 if (mp->m_metadirip)
1145 xfs_irele(mp->m_metadirip);
1146
1147 xfs_unmount_flush_inodes(mp);
1148
1149 xfs_qm_unmount(mp);
1150
1151 /*
1152 * Unreserve any blocks we have so that when we unmount we don't account
1153 * the reserved free space as used. This is really only necessary for
1154 * lazy superblock counting because it trusts the incore superblock
1155 * counters to be absolutely correct on clean unmount.
1156 *
1157 * We don't bother correcting this elsewhere for lazy superblock
1158 * counting because on mount of an unclean filesystem we reconstruct the
1159 * correct counter value and this is irrelevant.
1160 *
1161 * For non-lazy counter filesystems, this doesn't matter at all because
1162 * we only every apply deltas to the superblock and hence the incore
1163 * value does not matter....
1164 */
1165 error = xfs_reserve_blocks(mp, 0);
1166 if (error)
1167 xfs_warn(mp, "Unable to free reserved block pool. "
1168 "Freespace may not be correct on next mount.");
1169 xfs_unmount_check(mp);
1170
1171 /*
1172 * Indicate that it's ok to clear log incompat bits before cleaning
1173 * the log and writing the unmount record.
1174 */
1175 xfs_set_done_with_log_incompat(mp);
1176 xfs_log_unmount(mp);
1177 xfs_da_unmount(mp);
1178 xfs_uuid_unmount(mp);
1179
1180#if defined(DEBUG)
1181 xfs_errortag_clearall(mp);
1182#endif
1183 shrinker_free(mp->m_inodegc_shrinker);
1184 xfs_free_rtgroups(mp, 0, mp->m_sb.sb_rgcount);
1185 xfs_free_perag_range(mp, 0, mp->m_sb.sb_agcount);
1186 xfs_errortag_del(mp);
1187 xfs_error_sysfs_del(mp);
1188 xchk_stats_unregister(mp->m_scrub_stats);
1189 xfs_sysfs_del(&mp->m_stats.xs_kobj);
1190 xfs_sysfs_del(&mp->m_kobj);
1191}
1192
1193/*
1194 * Determine whether modifications can proceed. The caller specifies the minimum
1195 * freeze level for which modifications should not be allowed. This allows
1196 * certain operations to proceed while the freeze sequence is in progress, if
1197 * necessary.
1198 */
1199bool
1200xfs_fs_writable(
1201 struct xfs_mount *mp,
1202 int level)
1203{
1204 ASSERT(level > SB_UNFROZEN);
1205 if ((mp->m_super->s_writers.frozen >= level) ||
1206 xfs_is_shutdown(mp) || xfs_is_readonly(mp))
1207 return false;
1208
1209 return true;
1210}
1211
1212void
1213xfs_add_freecounter(
1214 struct xfs_mount *mp,
1215 struct percpu_counter *counter,
1216 uint64_t delta)
1217{
1218 bool has_resv_pool = (counter == &mp->m_fdblocks);
1219 uint64_t res_used;
1220
1221 /*
1222 * If the reserve pool is depleted, put blocks back into it first.
1223 * Most of the time the pool is full.
1224 */
1225 if (!has_resv_pool || mp->m_resblks == mp->m_resblks_avail) {
1226 percpu_counter_add(counter, delta);
1227 return;
1228 }
1229
1230 spin_lock(&mp->m_sb_lock);
1231 res_used = mp->m_resblks - mp->m_resblks_avail;
1232 if (res_used > delta) {
1233 mp->m_resblks_avail += delta;
1234 } else {
1235 delta -= res_used;
1236 mp->m_resblks_avail = mp->m_resblks;
1237 percpu_counter_add(counter, delta);
1238 }
1239 spin_unlock(&mp->m_sb_lock);
1240}
1241
1242int
1243xfs_dec_freecounter(
1244 struct xfs_mount *mp,
1245 struct percpu_counter *counter,
1246 uint64_t delta,
1247 bool rsvd)
1248{
1249 int64_t lcounter;
1250 uint64_t set_aside = 0;
1251 s32 batch;
1252 bool has_resv_pool;
1253
1254 ASSERT(counter == &mp->m_fdblocks || counter == &mp->m_frextents);
1255 has_resv_pool = (counter == &mp->m_fdblocks);
1256 if (rsvd)
1257 ASSERT(has_resv_pool);
1258
1259 /*
1260 * Taking blocks away, need to be more accurate the closer we
1261 * are to zero.
1262 *
1263 * If the counter has a value of less than 2 * max batch size,
1264 * then make everything serialise as we are real close to
1265 * ENOSPC.
1266 */
1267 if (__percpu_counter_compare(counter, 2 * XFS_FDBLOCKS_BATCH,
1268 XFS_FDBLOCKS_BATCH) < 0)
1269 batch = 1;
1270 else
1271 batch = XFS_FDBLOCKS_BATCH;
1272
1273 /*
1274 * Set aside allocbt blocks because these blocks are tracked as free
1275 * space but not available for allocation. Technically this means that a
1276 * single reservation cannot consume all remaining free space, but the
1277 * ratio of allocbt blocks to usable free blocks should be rather small.
1278 * The tradeoff without this is that filesystems that maintain high
1279 * perag block reservations can over reserve physical block availability
1280 * and fail physical allocation, which leads to much more serious
1281 * problems (i.e. transaction abort, pagecache discards, etc.) than
1282 * slightly premature -ENOSPC.
1283 */
1284 if (has_resv_pool)
1285 set_aside = xfs_fdblocks_unavailable(mp);
1286 percpu_counter_add_batch(counter, -((int64_t)delta), batch);
1287 if (__percpu_counter_compare(counter, set_aside,
1288 XFS_FDBLOCKS_BATCH) >= 0) {
1289 /* we had space! */
1290 return 0;
1291 }
1292
1293 /*
1294 * lock up the sb for dipping into reserves before releasing the space
1295 * that took us to ENOSPC.
1296 */
1297 spin_lock(&mp->m_sb_lock);
1298 percpu_counter_add(counter, delta);
1299 if (!has_resv_pool || !rsvd)
1300 goto fdblocks_enospc;
1301
1302 lcounter = (long long)mp->m_resblks_avail - delta;
1303 if (lcounter >= 0) {
1304 mp->m_resblks_avail = lcounter;
1305 spin_unlock(&mp->m_sb_lock);
1306 return 0;
1307 }
1308 xfs_warn_once(mp,
1309"Reserve blocks depleted! Consider increasing reserve pool size.");
1310
1311fdblocks_enospc:
1312 spin_unlock(&mp->m_sb_lock);
1313 return -ENOSPC;
1314}
1315
1316/*
1317 * Used to free the superblock along various error paths.
1318 */
1319void
1320xfs_freesb(
1321 struct xfs_mount *mp)
1322{
1323 struct xfs_buf *bp = mp->m_sb_bp;
1324
1325 xfs_buf_lock(bp);
1326 mp->m_sb_bp = NULL;
1327 xfs_buf_relse(bp);
1328}
1329
1330/*
1331 * If the underlying (data/log/rt) device is readonly, there are some
1332 * operations that cannot proceed.
1333 */
1334int
1335xfs_dev_is_read_only(
1336 struct xfs_mount *mp,
1337 char *message)
1338{
1339 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1340 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1341 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1342 xfs_notice(mp, "%s required on read-only device.", message);
1343 xfs_notice(mp, "write access unavailable, cannot proceed.");
1344 return -EROFS;
1345 }
1346 return 0;
1347}
1348
1349/* Force the summary counters to be recalculated at next mount. */
1350void
1351xfs_force_summary_recalc(
1352 struct xfs_mount *mp)
1353{
1354 if (!xfs_has_lazysbcount(mp))
1355 return;
1356
1357 xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1358}
1359
1360/*
1361 * Enable a log incompat feature flag in the primary superblock. The caller
1362 * cannot have any other transactions in progress.
1363 */
1364int
1365xfs_add_incompat_log_feature(
1366 struct xfs_mount *mp,
1367 uint32_t feature)
1368{
1369 struct xfs_dsb *dsb;
1370 int error;
1371
1372 ASSERT(hweight32(feature) == 1);
1373 ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
1374
1375 /*
1376 * Force the log to disk and kick the background AIL thread to reduce
1377 * the chances that the bwrite will stall waiting for the AIL to unpin
1378 * the primary superblock buffer. This isn't a data integrity
1379 * operation, so we don't need a synchronous push.
1380 */
1381 error = xfs_log_force(mp, XFS_LOG_SYNC);
1382 if (error)
1383 return error;
1384 xfs_ail_push_all(mp->m_ail);
1385
1386 /*
1387 * Lock the primary superblock buffer to serialize all callers that
1388 * are trying to set feature bits.
1389 */
1390 xfs_buf_lock(mp->m_sb_bp);
1391 xfs_buf_hold(mp->m_sb_bp);
1392
1393 if (xfs_is_shutdown(mp)) {
1394 error = -EIO;
1395 goto rele;
1396 }
1397
1398 if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature))
1399 goto rele;
1400
1401 /*
1402 * Write the primary superblock to disk immediately, because we need
1403 * the log_incompat bit to be set in the primary super now to protect
1404 * the log items that we're going to commit later.
1405 */
1406 dsb = mp->m_sb_bp->b_addr;
1407 xfs_sb_to_disk(dsb, &mp->m_sb);
1408 dsb->sb_features_log_incompat |= cpu_to_be32(feature);
1409 error = xfs_bwrite(mp->m_sb_bp);
1410 if (error)
1411 goto shutdown;
1412
1413 /*
1414 * Add the feature bits to the incore superblock before we unlock the
1415 * buffer.
1416 */
1417 xfs_sb_add_incompat_log_features(&mp->m_sb, feature);
1418 xfs_buf_relse(mp->m_sb_bp);
1419
1420 /* Log the superblock to disk. */
1421 return xfs_sync_sb(mp, false);
1422shutdown:
1423 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1424rele:
1425 xfs_buf_relse(mp->m_sb_bp);
1426 return error;
1427}
1428
1429/*
1430 * Clear all the log incompat flags from the superblock.
1431 *
1432 * The caller cannot be in a transaction, must ensure that the log does not
1433 * contain any log items protected by any log incompat bit, and must ensure
1434 * that there are no other threads that depend on the state of the log incompat
1435 * feature flags in the primary super.
1436 *
1437 * Returns true if the superblock is dirty.
1438 */
1439bool
1440xfs_clear_incompat_log_features(
1441 struct xfs_mount *mp)
1442{
1443 bool ret = false;
1444
1445 if (!xfs_has_crc(mp) ||
1446 !xfs_sb_has_incompat_log_feature(&mp->m_sb,
1447 XFS_SB_FEAT_INCOMPAT_LOG_ALL) ||
1448 xfs_is_shutdown(mp) ||
1449 !xfs_is_done_with_log_incompat(mp))
1450 return false;
1451
1452 /*
1453 * Update the incore superblock. We synchronize on the primary super
1454 * buffer lock to be consistent with the add function, though at least
1455 * in theory this shouldn't be necessary.
1456 */
1457 xfs_buf_lock(mp->m_sb_bp);
1458 xfs_buf_hold(mp->m_sb_bp);
1459
1460 if (xfs_sb_has_incompat_log_feature(&mp->m_sb,
1461 XFS_SB_FEAT_INCOMPAT_LOG_ALL)) {
1462 xfs_sb_remove_incompat_log_features(&mp->m_sb);
1463 ret = true;
1464 }
1465
1466 xfs_buf_relse(mp->m_sb_bp);
1467 return ret;
1468}
1469
1470/*
1471 * Update the in-core delayed block counter.
1472 *
1473 * We prefer to update the counter without having to take a spinlock for every
1474 * counter update (i.e. batching). Each change to delayed allocation
1475 * reservations can change can easily exceed the default percpu counter
1476 * batching, so we use a larger batch factor here.
1477 *
1478 * Note that we don't currently have any callers requiring fast summation
1479 * (e.g. percpu_counter_read) so we can use a big batch value here.
1480 */
1481#define XFS_DELALLOC_BATCH (4096)
1482void
1483xfs_mod_delalloc(
1484 struct xfs_inode *ip,
1485 int64_t data_delta,
1486 int64_t ind_delta)
1487{
1488 struct xfs_mount *mp = ip->i_mount;
1489
1490 if (XFS_IS_REALTIME_INODE(ip)) {
1491 percpu_counter_add_batch(&mp->m_delalloc_rtextents,
1492 xfs_blen_to_rtbxlen(mp, data_delta),
1493 XFS_DELALLOC_BATCH);
1494 if (!ind_delta)
1495 return;
1496 data_delta = 0;
1497 }
1498 percpu_counter_add_batch(&mp->m_delalloc_blks, data_delta + ind_delta,
1499 XFS_DELALLOC_BATCH);
1500}