Merge tag 'gvt-next-2017-06-08' of https://github.com/01org/gvt-linux into drm-intel...
[linux-2.6-block.git] / fs / xfs / xfs_mount.c
CommitLineData
1da177e4 1/*
7b718769
NS
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
1da177e4 4 *
7b718769
NS
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
1da177e4
LT
7 * published by the Free Software Foundation.
8 *
7b718769
NS
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.
1da177e4 13 *
7b718769
NS
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
1da177e4 17 */
1da177e4 18#include "xfs.h"
a844f451 19#include "xfs_fs.h"
70a9883c 20#include "xfs_shared.h"
239880ef
DC
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
a844f451 24#include "xfs_bit.h"
1da177e4 25#include "xfs_sb.h"
1da177e4 26#include "xfs_mount.h"
3ab78df2 27#include "xfs_defer.h"
57062787 28#include "xfs_da_format.h"
9a2cc41c 29#include "xfs_da_btree.h"
1da177e4 30#include "xfs_inode.h"
a4fbe6ab 31#include "xfs_dir2.h"
a844f451 32#include "xfs_ialloc.h"
1da177e4
LT
33#include "xfs_alloc.h"
34#include "xfs_rtalloc.h"
35#include "xfs_bmap.h"
a4fbe6ab
DC
36#include "xfs_trans.h"
37#include "xfs_trans_priv.h"
38#include "xfs_log.h"
1da177e4 39#include "xfs_error.h"
1da177e4
LT
40#include "xfs_quota.h"
41#include "xfs_fsops.h"
0b1b213f 42#include "xfs_trace.h"
6d8b79cf 43#include "xfs_icache.h"
a31b1d3d 44#include "xfs_sysfs.h"
035e00ac 45#include "xfs_rmap_btree.h"
1946b91c 46#include "xfs_refcount_btree.h"
174edb0e 47#include "xfs_reflink.h"
ebf55872 48#include "xfs_extent_busy.h"
0b1b213f 49
1da177e4 50
27174203
CH
51static DEFINE_MUTEX(xfs_uuid_table_mutex);
52static int xfs_uuid_table_size;
53static uuid_t *xfs_uuid_table;
54
af3b6382
DW
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
27174203
CH
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
8f720d9f
AG
76 /* Publish UUID in struct super_block */
77 BUILD_BUG_ON(sizeof(mp->m_super->s_uuid) != sizeof(uuid_t));
78 memcpy(&mp->m_super->s_uuid, uuid, sizeof(uuid_t));
79
27174203
CH
80 if (mp->m_flags & XFS_MOUNT_NOUUID)
81 return 0;
82
83 if (uuid_is_nil(uuid)) {
0b932ccc 84 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
2451337d 85 return -EINVAL;
27174203
CH
86 }
87
88 mutex_lock(&xfs_uuid_table_mutex);
89 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
90 if (uuid_is_nil(&xfs_uuid_table[i])) {
91 hole = i;
92 continue;
93 }
94 if (uuid_equal(uuid, &xfs_uuid_table[i]))
95 goto out_duplicate;
96 }
97
98 if (hole < 0) {
99 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
100 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
27174203
CH
101 KM_SLEEP);
102 hole = xfs_uuid_table_size++;
103 }
104 xfs_uuid_table[hole] = *uuid;
105 mutex_unlock(&xfs_uuid_table_mutex);
106
107 return 0;
108
109 out_duplicate:
110 mutex_unlock(&xfs_uuid_table_mutex);
021000e5 111 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
2451337d 112 return -EINVAL;
27174203
CH
113}
114
115STATIC void
116xfs_uuid_unmount(
117 struct xfs_mount *mp)
118{
119 uuid_t *uuid = &mp->m_sb.sb_uuid;
120 int i;
121
122 if (mp->m_flags & XFS_MOUNT_NOUUID)
123 return;
124
125 mutex_lock(&xfs_uuid_table_mutex);
126 for (i = 0; i < xfs_uuid_table_size; i++) {
127 if (uuid_is_nil(&xfs_uuid_table[i]))
128 continue;
129 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
130 continue;
131 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
132 break;
133 }
134 ASSERT(i < xfs_uuid_table_size);
135 mutex_unlock(&xfs_uuid_table_mutex);
136}
137
138
e176579e
DC
139STATIC void
140__xfs_free_perag(
141 struct rcu_head *head)
142{
143 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
144
145 ASSERT(atomic_read(&pag->pag_ref) == 0);
146 kmem_free(pag);
147}
148
1da177e4 149/*
e176579e 150 * Free up the per-ag resources associated with the mount structure.
1da177e4 151 */
c962fb79 152STATIC void
ff4f038c 153xfs_free_perag(
745f6919 154 xfs_mount_t *mp)
1da177e4 155{
1c1c6ebc
DC
156 xfs_agnumber_t agno;
157 struct xfs_perag *pag;
158
159 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
160 spin_lock(&mp->m_perag_lock);
161 pag = radix_tree_delete(&mp->m_perag_tree, agno);
162 spin_unlock(&mp->m_perag_lock);
e176579e 163 ASSERT(pag);
f83282a8 164 ASSERT(atomic_read(&pag->pag_ref) == 0);
6031e73a 165 xfs_buf_hash_destroy(pag);
e176579e 166 call_rcu(&pag->rcu_head, __xfs_free_perag);
1da177e4 167 }
1da177e4
LT
168}
169
4cc929ee
NS
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
d5cf09ba 182 /* Limited by ULONG_MAX of page cache index */
09cbfeaf 183 if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
2451337d 184 return -EFBIG;
4cc929ee
NS
185 return 0;
186}
1da177e4 187
1c1c6ebc 188int
c11e2c36 189xfs_initialize_perag(
c11e2c36 190 xfs_mount_t *mp,
1c1c6ebc
DC
191 xfs_agnumber_t agcount,
192 xfs_agnumber_t *maxagi)
1da177e4 193{
2d2194f6 194 xfs_agnumber_t index;
b20fe473 195 xfs_agnumber_t first_initialised = NULLAGNUMBER;
1da177e4 196 xfs_perag_t *pag;
8b26c582 197 int error = -ENOMEM;
1da177e4 198
1c1c6ebc
DC
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 }
fb3b504a 210
1c1c6ebc
DC
211 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
212 if (!pag)
b20fe473 213 goto out_unwind_new_pags;
fb3b504a
CH
214 pag->pag_agno = index;
215 pag->pag_mount = mp;
1a427ab0 216 spin_lock_init(&pag->pag_ici_lock);
69b491c2 217 mutex_init(&pag->pag_ici_reclaim_lock);
fb3b504a 218 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
6031e73a 219 if (xfs_buf_hash_init(pag))
b20fe473 220 goto out_free_pag;
ebf55872 221 init_waitqueue_head(&pag->pagb_wait);
fb3b504a 222
1c1c6ebc 223 if (radix_tree_preload(GFP_NOFS))
b20fe473 224 goto out_hash_destroy;
fb3b504a 225
1c1c6ebc
DC
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);
8b26c582
DC
230 radix_tree_preload_end();
231 error = -EEXIST;
b20fe473 232 goto out_hash_destroy;
1c1c6ebc
DC
233 }
234 spin_unlock(&mp->m_perag_lock);
235 radix_tree_preload_end();
b20fe473
BD
236 /* first new pag is fully initialized */
237 if (first_initialised == NULLAGNUMBER)
238 first_initialised = index;
1c1c6ebc
DC
239 }
240
12c3f05c 241 index = xfs_set_inode_alloc(mp, agcount);
fb3b504a 242
1c1c6ebc
DC
243 if (maxagi)
244 *maxagi = index;
8018026e
DW
245
246 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
1c1c6ebc 247 return 0;
8b26c582 248
b20fe473 249out_hash_destroy:
6031e73a 250 xfs_buf_hash_destroy(pag);
b20fe473 251out_free_pag:
8b26c582 252 kmem_free(pag);
b20fe473
BD
253out_unwind_new_pags:
254 /* unwind any prior newly initialized pags */
255 for (index = first_initialised; index < agcount; index++) {
8b26c582 256 pag = radix_tree_delete(&mp->m_perag_tree, index);
b20fe473
BD
257 if (!pag)
258 break;
6031e73a 259 xfs_buf_hash_destroy(pag);
8b26c582
DC
260 kmem_free(pag);
261 }
262 return error;
1da177e4
LT
263}
264
1da177e4
LT
265/*
266 * xfs_readsb
267 *
268 * Does the initial read of the superblock.
269 */
270int
ff55068c
DC
271xfs_readsb(
272 struct xfs_mount *mp,
273 int flags)
1da177e4
LT
274{
275 unsigned int sector_size;
04a1e6c5
DC
276 struct xfs_buf *bp;
277 struct xfs_sb *sbp = &mp->m_sb;
1da177e4 278 int error;
af34e09d 279 int loud = !(flags & XFS_MFSI_QUIET);
daba5427 280 const struct xfs_buf_ops *buf_ops;
1da177e4
LT
281
282 ASSERT(mp->m_sb_bp == NULL);
283 ASSERT(mp->m_ddev_targp != NULL);
284
daba5427
ES
285 /*
286 * For the initial read, we must guess at the sector
287 * size based on the block device. It's enough to
288 * get the sb_sectsize out of the superblock and
289 * then reread with the proper length.
290 * We don't verify it yet, because it may not be complete.
291 */
292 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
293 buf_ops = NULL;
294
1da177e4 295 /*
c891c30a
BF
296 * Allocate a (locked) buffer to hold the superblock. This will be kept
297 * around at all times to optimize access to the superblock. Therefore,
298 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
299 * elevated.
1da177e4 300 */
26af6552 301reread:
ba372674 302 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
c891c30a
BF
303 BTOBB(sector_size), XBF_NO_IOACCT, &bp,
304 buf_ops);
ba372674 305 if (error) {
eab4e633 306 if (loud)
e721f504 307 xfs_warn(mp, "SB validate failed with error %d.", error);
ac75a1f7 308 /* bad CRC means corrupted metadata */
2451337d
DC
309 if (error == -EFSBADCRC)
310 error = -EFSCORRUPTED;
ba372674 311 return error;
eab4e633 312 }
1da177e4
LT
313
314 /*
315 * Initialize the mount structure from the superblock.
1da177e4 316 */
556b8883 317 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
556b8883
DC
318
319 /*
320 * If we haven't validated the superblock, do so now before we try
321 * to check the sector size and reread the superblock appropriately.
322 */
323 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
324 if (loud)
325 xfs_warn(mp, "Invalid superblock magic number");
2451337d 326 error = -EINVAL;
556b8883
DC
327 goto release_buf;
328 }
ff55068c 329
1da177e4
LT
330 /*
331 * We must be able to do sector-sized and sector-aligned IO.
332 */
04a1e6c5 333 if (sector_size > sbp->sb_sectsize) {
af34e09d
DC
334 if (loud)
335 xfs_warn(mp, "device supports %u byte sectors (not %u)",
04a1e6c5 336 sector_size, sbp->sb_sectsize);
2451337d 337 error = -ENOSYS;
26af6552 338 goto release_buf;
1da177e4
LT
339 }
340
daba5427 341 if (buf_ops == NULL) {
556b8883
DC
342 /*
343 * Re-read the superblock so the buffer is correctly sized,
344 * and properly verified.
345 */
1da177e4 346 xfs_buf_relse(bp);
04a1e6c5 347 sector_size = sbp->sb_sectsize;
daba5427 348 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
26af6552 349 goto reread;
1da177e4
LT
350 }
351
5681ca40 352 xfs_reinit_percpu_counters(mp);
8d280b98 353
04a1e6c5
DC
354 /* no need to be quiet anymore, so reset the buf ops */
355 bp->b_ops = &xfs_sb_buf_ops;
356
1da177e4 357 mp->m_sb_bp = bp;
26af6552 358 xfs_buf_unlock(bp);
1da177e4
LT
359 return 0;
360
26af6552
DC
361release_buf:
362 xfs_buf_relse(bp);
1da177e4
LT
363 return error;
364}
365
1da177e4 366/*
0771fb45 367 * Update alignment values based on mount options and sb values
1da177e4 368 */
0771fb45 369STATIC int
7884bc86 370xfs_update_alignment(xfs_mount_t *mp)
1da177e4 371{
1da177e4 372 xfs_sb_t *sbp = &(mp->m_sb);
1da177e4 373
4249023a 374 if (mp->m_dalign) {
1da177e4
LT
375 /*
376 * If stripe unit and stripe width are not multiples
377 * of the fs blocksize turn off alignment.
378 */
379 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
380 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
39a45d84
JL
381 xfs_warn(mp,
382 "alignment check failed: sunit/swidth vs. blocksize(%d)",
383 sbp->sb_blocksize);
2451337d 384 return -EINVAL;
1da177e4
LT
385 } else {
386 /*
387 * Convert the stripe unit and width to FSBs.
388 */
389 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
390 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
53487786 391 xfs_warn(mp,
39a45d84
JL
392 "alignment check failed: sunit/swidth vs. agsize(%d)",
393 sbp->sb_agblocks);
2451337d 394 return -EINVAL;
1da177e4
LT
395 } else if (mp->m_dalign) {
396 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
397 } else {
39a45d84
JL
398 xfs_warn(mp,
399 "alignment check failed: sunit(%d) less than bsize(%d)",
400 mp->m_dalign, sbp->sb_blocksize);
2451337d 401 return -EINVAL;
1da177e4
LT
402 }
403 }
404
405 /*
406 * Update superblock with new values
407 * and log changes
408 */
62118709 409 if (xfs_sb_version_hasdalign(sbp)) {
1da177e4
LT
410 if (sbp->sb_unit != mp->m_dalign) {
411 sbp->sb_unit = mp->m_dalign;
61e63ecb 412 mp->m_update_sb = true;
1da177e4
LT
413 }
414 if (sbp->sb_width != mp->m_swidth) {
415 sbp->sb_width = mp->m_swidth;
61e63ecb 416 mp->m_update_sb = true;
1da177e4 417 }
34d7f603
JL
418 } else {
419 xfs_warn(mp,
420 "cannot change alignment: superblock does not support data alignment");
2451337d 421 return -EINVAL;
1da177e4
LT
422 }
423 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
62118709 424 xfs_sb_version_hasdalign(&mp->m_sb)) {
1da177e4
LT
425 mp->m_dalign = sbp->sb_unit;
426 mp->m_swidth = sbp->sb_width;
427 }
428
0771fb45
ES
429 return 0;
430}
1da177e4 431
0771fb45
ES
432/*
433 * Set the maximum inode count for this filesystem
434 */
435STATIC void
436xfs_set_maxicount(xfs_mount_t *mp)
437{
438 xfs_sb_t *sbp = &(mp->m_sb);
439 __uint64_t icount;
1da177e4 440
0771fb45
ES
441 if (sbp->sb_imax_pct) {
442 /*
443 * Make sure the maximum inode count is a multiple
444 * of the units we allocate inodes in.
1da177e4 445 */
1da177e4
LT
446 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
447 do_div(icount, 100);
448 do_div(icount, mp->m_ialloc_blks);
449 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
450 sbp->sb_inopblog;
0771fb45 451 } else {
1da177e4 452 mp->m_maxicount = 0;
1da177e4 453 }
0771fb45
ES
454}
455
456/*
457 * Set the default minimum read and write sizes unless
458 * already specified in a mount option.
459 * We use smaller I/O sizes when the file system
460 * is being used for NFS service (wsync mount option).
461 */
462STATIC void
463xfs_set_rw_sizes(xfs_mount_t *mp)
464{
465 xfs_sb_t *sbp = &(mp->m_sb);
466 int readio_log, writeio_log;
1da177e4 467
1da177e4
LT
468 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
469 if (mp->m_flags & XFS_MOUNT_WSYNC) {
470 readio_log = XFS_WSYNC_READIO_LOG;
471 writeio_log = XFS_WSYNC_WRITEIO_LOG;
472 } else {
473 readio_log = XFS_READIO_LOG_LARGE;
474 writeio_log = XFS_WRITEIO_LOG_LARGE;
475 }
476 } else {
477 readio_log = mp->m_readio_log;
478 writeio_log = mp->m_writeio_log;
479 }
480
1da177e4
LT
481 if (sbp->sb_blocklog > readio_log) {
482 mp->m_readio_log = sbp->sb_blocklog;
483 } else {
484 mp->m_readio_log = readio_log;
485 }
486 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
487 if (sbp->sb_blocklog > writeio_log) {
488 mp->m_writeio_log = sbp->sb_blocklog;
489 } else {
490 mp->m_writeio_log = writeio_log;
491 }
492 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
0771fb45 493}
1da177e4 494
055388a3
DC
495/*
496 * precalculate the low space thresholds for dynamic speculative preallocation.
497 */
498void
499xfs_set_low_space_thresholds(
500 struct xfs_mount *mp)
501{
502 int i;
503
504 for (i = 0; i < XFS_LOWSP_MAX; i++) {
505 __uint64_t space = mp->m_sb.sb_dblocks;
506
507 do_div(space, 100);
508 mp->m_low_space[i] = space * (i + 1);
509 }
510}
511
512
0771fb45
ES
513/*
514 * Set whether we're using inode alignment.
515 */
516STATIC void
517xfs_set_inoalignment(xfs_mount_t *mp)
518{
62118709 519 if (xfs_sb_version_hasalign(&mp->m_sb) &&
d5825712 520 mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
1da177e4
LT
521 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
522 else
523 mp->m_inoalign_mask = 0;
524 /*
525 * If we are using stripe alignment, check whether
526 * the stripe unit is a multiple of the inode alignment
527 */
528 if (mp->m_dalign && mp->m_inoalign_mask &&
529 !(mp->m_dalign & mp->m_inoalign_mask))
530 mp->m_sinoalign = mp->m_dalign;
531 else
532 mp->m_sinoalign = 0;
0771fb45
ES
533}
534
535/*
0471f62e 536 * Check that the data (and log if separate) is an ok size.
0771fb45
ES
537 */
538STATIC int
ba372674
DC
539xfs_check_sizes(
540 struct xfs_mount *mp)
0771fb45 541{
ba372674 542 struct xfs_buf *bp;
0771fb45 543 xfs_daddr_t d;
ba372674 544 int error;
0771fb45 545
1da177e4
LT
546 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
547 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
0b932ccc 548 xfs_warn(mp, "filesystem size mismatch detected");
2451337d 549 return -EFBIG;
1da177e4 550 }
ba372674 551 error = xfs_buf_read_uncached(mp->m_ddev_targp,
1922c949 552 d - XFS_FSS_TO_BB(mp, 1),
ba372674
DC
553 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
554 if (error) {
0b932ccc 555 xfs_warn(mp, "last sector read failed");
ba372674 556 return error;
1da177e4 557 }
1922c949 558 xfs_buf_relse(bp);
1da177e4 559
ba372674
DC
560 if (mp->m_logdev_targp == mp->m_ddev_targp)
561 return 0;
562
563 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
564 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
565 xfs_warn(mp, "log size mismatch detected");
566 return -EFBIG;
567 }
568 error = xfs_buf_read_uncached(mp->m_logdev_targp,
1922c949 569 d - XFS_FSB_TO_BB(mp, 1),
ba372674
DC
570 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
571 if (error) {
572 xfs_warn(mp, "log device read failed");
573 return error;
0771fb45 574 }
ba372674 575 xfs_buf_relse(bp);
0771fb45
ES
576 return 0;
577}
578
7d095257
CH
579/*
580 * Clear the quotaflags in memory and in the superblock.
581 */
582int
583xfs_mount_reset_sbqflags(
584 struct xfs_mount *mp)
585{
7d095257
CH
586 mp->m_qflags = 0;
587
61e63ecb 588 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
7d095257
CH
589 if (mp->m_sb.sb_qflags == 0)
590 return 0;
591 spin_lock(&mp->m_sb_lock);
592 mp->m_sb.sb_qflags = 0;
593 spin_unlock(&mp->m_sb_lock);
594
61e63ecb 595 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
7d095257
CH
596 return 0;
597
61e63ecb 598 return xfs_sync_sb(mp, false);
7d095257
CH
599}
600
d5db0f97
ES
601__uint64_t
602xfs_default_resblks(xfs_mount_t *mp)
603{
604 __uint64_t resblks;
605
606 /*
8babd8a2
DC
607 * We default to 5% or 8192 fsbs of space reserved, whichever is
608 * smaller. This is intended to cover concurrent allocation
609 * transactions when we initially hit enospc. These each require a 4
610 * block reservation. Hence by default we cover roughly 2000 concurrent
611 * allocation reservations.
d5db0f97
ES
612 */
613 resblks = mp->m_sb.sb_dblocks;
614 do_div(resblks, 20);
8babd8a2 615 resblks = min_t(__uint64_t, resblks, 8192);
d5db0f97
ES
616 return resblks;
617}
618
0771fb45 619/*
0771fb45
ES
620 * This function does the following on an initial mount of a file system:
621 * - reads the superblock from disk and init the mount struct
622 * - if we're a 32-bit kernel, do a size check on the superblock
623 * so we don't mount terabyte filesystems
624 * - init mount struct realtime fields
625 * - allocate inode hash table for fs
626 * - init directory manager
627 * - perform recovery and init the log manager
628 */
629int
630xfs_mountfs(
f0b2efad 631 struct xfs_mount *mp)
0771fb45 632{
f0b2efad
BF
633 struct xfs_sb *sbp = &(mp->m_sb);
634 struct xfs_inode *rip;
635 __uint64_t resblks;
636 uint quotamount = 0;
637 uint quotaflags = 0;
638 int error = 0;
0771fb45 639
ff55068c 640 xfs_sb_mount_common(mp, sbp);
0771fb45 641
ee1c0908 642 /*
074e427b
DC
643 * Check for a mismatched features2 values. Older kernels read & wrote
644 * into the wrong sb offset for sb_features2 on some platforms due to
645 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
646 * which made older superblock reading/writing routines swap it as a
647 * 64-bit value.
ee1c0908 648 *
e6957ea4
ES
649 * For backwards compatibility, we make both slots equal.
650 *
074e427b
DC
651 * If we detect a mismatched field, we OR the set bits into the existing
652 * features2 field in case it has already been modified; we don't want
653 * to lose any features. We then update the bad location with the ORed
654 * value so that older kernels will see any features2 flags. The
655 * superblock writeback code ensures the new sb_features2 is copied to
656 * sb_bad_features2 before it is logged or written to disk.
ee1c0908 657 */
e6957ea4 658 if (xfs_sb_has_mismatched_features2(sbp)) {
0b932ccc 659 xfs_warn(mp, "correcting sb_features alignment problem");
ee1c0908 660 sbp->sb_features2 |= sbp->sb_bad_features2;
61e63ecb 661 mp->m_update_sb = true;
e6957ea4
ES
662
663 /*
664 * Re-check for ATTR2 in case it was found in bad_features2
665 * slot.
666 */
7c12f296
TS
667 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
668 !(mp->m_flags & XFS_MOUNT_NOATTR2))
e6957ea4 669 mp->m_flags |= XFS_MOUNT_ATTR2;
7c12f296
TS
670 }
671
672 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
673 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
674 xfs_sb_version_removeattr2(&mp->m_sb);
61e63ecb 675 mp->m_update_sb = true;
e6957ea4 676
7c12f296
TS
677 /* update sb_versionnum for the clearing of the morebits */
678 if (!sbp->sb_features2)
61e63ecb 679 mp->m_update_sb = true;
ee1c0908
DC
680 }
681
263997a6
DC
682 /* always use v2 inodes by default now */
683 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
684 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
61e63ecb 685 mp->m_update_sb = true;
263997a6
DC
686 }
687
0771fb45
ES
688 /*
689 * Check if sb_agblocks is aligned at stripe boundary
690 * If sb_agblocks is NOT aligned turn off m_dalign since
691 * allocator alignment is within an ag, therefore ag has
692 * to be aligned at stripe boundary.
693 */
7884bc86 694 error = xfs_update_alignment(mp);
0771fb45 695 if (error)
f9057e3d 696 goto out;
0771fb45
ES
697
698 xfs_alloc_compute_maxlevels(mp);
699 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
700 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
701 xfs_ialloc_compute_maxlevels(mp);
035e00ac 702 xfs_rmapbt_compute_maxlevels(mp);
1946b91c 703 xfs_refcountbt_compute_maxlevels(mp);
0771fb45
ES
704
705 xfs_set_maxicount(mp);
706
e6b3bb78
CM
707 /* enable fail_at_unmount as default */
708 mp->m_fail_unmount = 1;
709
a31b1d3d 710 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
27174203
CH
711 if (error)
712 goto out;
1da177e4 713
225e4635
BD
714 error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
715 &mp->m_kobj, "stats");
a31b1d3d
BF
716 if (error)
717 goto out_remove_sysfs;
718
192852be 719 error = xfs_error_sysfs_init(mp);
225e4635
BD
720 if (error)
721 goto out_del_stats;
722
192852be
CM
723
724 error = xfs_uuid_mount(mp);
725 if (error)
726 goto out_remove_error_sysfs;
727
0771fb45
ES
728 /*
729 * Set the minimum read and write sizes
730 */
731 xfs_set_rw_sizes(mp);
732
055388a3
DC
733 /* set the low space thresholds for dynamic preallocation */
734 xfs_set_low_space_thresholds(mp);
735
0771fb45
ES
736 /*
737 * Set the inode cluster size.
738 * This may still be overridden by the file system
739 * block size if it is larger than the chosen cluster size.
8f80587b
DC
740 *
741 * For v5 filesystems, scale the cluster size with the inode size to
742 * keep a constant ratio of inode per cluster buffer, but only if mkfs
743 * has set the inode alignment value appropriately for larger cluster
744 * sizes.
0771fb45
ES
745 */
746 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
8f80587b
DC
747 if (xfs_sb_version_hascrc(&mp->m_sb)) {
748 int new_size = mp->m_inode_cluster_size;
749
750 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
751 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
752 mp->m_inode_cluster_size = new_size;
8f80587b 753 }
0771fb45 754
e5376fc1
BF
755 /*
756 * If enabled, sparse inode chunk alignment is expected to match the
757 * cluster size. Full inode chunk alignment must match the chunk size,
758 * but that is checked on sb read verification...
759 */
760 if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
761 mp->m_sb.sb_spino_align !=
762 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
763 xfs_warn(mp,
764 "Sparse inode block alignment (%u) must match cluster size (%llu).",
765 mp->m_sb.sb_spino_align,
766 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
767 error = -EINVAL;
768 goto out_remove_uuid;
769 }
770
0771fb45
ES
771 /*
772 * Set inode alignment fields
773 */
774 xfs_set_inoalignment(mp);
775
776 /*
c2bfbc9b 777 * Check that the data (and log if separate) is an ok size.
0771fb45 778 */
4249023a 779 error = xfs_check_sizes(mp);
0771fb45 780 if (error)
f9057e3d 781 goto out_remove_uuid;
0771fb45 782
1da177e4
LT
783 /*
784 * Initialize realtime fields in the mount structure
785 */
0771fb45
ES
786 error = xfs_rtmount_init(mp);
787 if (error) {
0b932ccc 788 xfs_warn(mp, "RT mount failed");
f9057e3d 789 goto out_remove_uuid;
1da177e4
LT
790 }
791
1da177e4
LT
792 /*
793 * Copies the low order bits of the timestamp and the randomly
794 * set "sequence" number out of a UUID.
795 */
796 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
797
1da177e4
LT
798 mp->m_dmevmask = 0; /* not persistent; set after each mount */
799
0650b554
DC
800 error = xfs_da_mount(mp);
801 if (error) {
802 xfs_warn(mp, "Failed dir/attr init: %d", error);
803 goto out_remove_uuid;
804 }
1da177e4
LT
805
806 /*
807 * Initialize the precomputed transaction reservations values.
808 */
809 xfs_trans_init(mp);
810
1da177e4
LT
811 /*
812 * Allocate and initialize the per-ag data.
813 */
1c1c6ebc 814 spin_lock_init(&mp->m_perag_lock);
9b98b6f3 815 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1c1c6ebc
DC
816 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
817 if (error) {
0b932ccc 818 xfs_warn(mp, "Failed per-ag init: %d", error);
0650b554 819 goto out_free_dir;
1c1c6ebc 820 }
1da177e4 821
f9057e3d 822 if (!sbp->sb_logblocks) {
0b932ccc 823 xfs_warn(mp, "no log defined");
f9057e3d 824 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
2451337d 825 error = -EFSCORRUPTED;
f9057e3d
CH
826 goto out_free_perag;
827 }
828
1da177e4 829 /*
f0b2efad
BF
830 * Log's mount-time initialization. The first part of recovery can place
831 * some items on the AIL, to be handled when recovery is finished or
832 * cancelled.
1da177e4 833 */
f9057e3d
CH
834 error = xfs_log_mount(mp, mp->m_logdev_targp,
835 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
836 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
837 if (error) {
0b932ccc 838 xfs_warn(mp, "log mount failed");
d4f3512b 839 goto out_fail_wait;
1da177e4
LT
840 }
841
92821e2b
DC
842 /*
843 * Now the log is mounted, we know if it was an unclean shutdown or
844 * not. If it was, with the first phase of recovery has completed, we
845 * have consistent AG blocks on disk. We have not recovered EFIs yet,
846 * but they are recovered transactionally in the second recovery phase
847 * later.
848 *
849 * Hence we can safely re-initialise incore superblock counters from
850 * the per-ag data. These may not be correct if the filesystem was not
851 * cleanly unmounted, so we need to wait for recovery to finish before
852 * doing this.
853 *
854 * If the filesystem was cleanly unmounted, then we can trust the
855 * values in the superblock to be correct and we don't need to do
856 * anything here.
857 *
858 * If we are currently making the filesystem, the initialisation will
859 * fail as the perag data is in an undefined state.
860 */
92821e2b
DC
861 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
862 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
863 !mp->m_sb.sb_inprogress) {
864 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
f9057e3d 865 if (error)
6eee8972 866 goto out_log_dealloc;
92821e2b 867 }
f9057e3d 868
1da177e4
LT
869 /*
870 * Get and sanity-check the root inode.
871 * Save the pointer to it in the mount structure.
872 */
7b6259e7 873 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1da177e4 874 if (error) {
0b932ccc 875 xfs_warn(mp, "failed to read root inode");
f9057e3d 876 goto out_log_dealloc;
1da177e4
LT
877 }
878
879 ASSERT(rip != NULL);
1da177e4 880
c19b3b05 881 if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
0b932ccc 882 xfs_warn(mp, "corrupted root inode %llu: not a directory",
b6574520 883 (unsigned long long)rip->i_ino);
1da177e4
LT
884 xfs_iunlock(rip, XFS_ILOCK_EXCL);
885 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
886 mp);
2451337d 887 error = -EFSCORRUPTED;
f9057e3d 888 goto out_rele_rip;
1da177e4
LT
889 }
890 mp->m_rootip = rip; /* save it */
891
892 xfs_iunlock(rip, XFS_ILOCK_EXCL);
893
894 /*
895 * Initialize realtime inode pointers in the mount structure
896 */
0771fb45
ES
897 error = xfs_rtmount_inodes(mp);
898 if (error) {
1da177e4
LT
899 /*
900 * Free up the root inode.
901 */
0b932ccc 902 xfs_warn(mp, "failed to read RT inodes");
f9057e3d 903 goto out_rele_rip;
1da177e4
LT
904 }
905
906 /*
7884bc86
CH
907 * If this is a read-only mount defer the superblock updates until
908 * the next remount into writeable mode. Otherwise we would never
909 * perform the update e.g. for the root filesystem.
1da177e4 910 */
61e63ecb
DC
911 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
912 error = xfs_sync_sb(mp, false);
e5720eec 913 if (error) {
0b932ccc 914 xfs_warn(mp, "failed to write sb changes");
b93b6e43 915 goto out_rtunmount;
e5720eec
DC
916 }
917 }
1da177e4
LT
918
919 /*
920 * Initialise the XFS quota management subsystem for this mount
921 */
7d095257
CH
922 if (XFS_IS_QUOTA_RUNNING(mp)) {
923 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
924 if (error)
925 goto out_rtunmount;
926 } else {
927 ASSERT(!XFS_IS_QUOTA_ON(mp));
928
929 /*
930 * If a file system had quotas running earlier, but decided to
931 * mount without -o uquota/pquota/gquota options, revoke the
932 * quotachecked license.
933 */
934 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
0b932ccc 935 xfs_notice(mp, "resetting quota flags");
7d095257
CH
936 error = xfs_mount_reset_sbqflags(mp);
937 if (error)
a70a4fa5 938 goto out_rtunmount;
7d095257
CH
939 }
940 }
1da177e4 941
17c12bcd
DW
942 /*
943 * During the second phase of log recovery, we need iget and
944 * iput to behave like they do for an active filesystem.
945 * xfs_fs_drop_inode needs to be able to prevent the deletion
946 * of inodes before we're done replaying log items on those
947 * inodes.
948 */
949 mp->m_super->s_flags |= MS_ACTIVE;
950
1da177e4 951 /*
f0b2efad
BF
952 * Finish recovering the file system. This part needed to be delayed
953 * until after the root and real-time bitmap inodes were consistently
954 * read in.
1da177e4 955 */
4249023a 956 error = xfs_log_mount_finish(mp);
1da177e4 957 if (error) {
0b932ccc 958 xfs_warn(mp, "log mount finish failed");
b93b6e43 959 goto out_rtunmount;
1da177e4
LT
960 }
961
ddeb14f4
DC
962 /*
963 * Now the log is fully replayed, we can transition to full read-only
964 * mode for read-only mounts. This will sync all the metadata and clean
965 * the log so that the recovery we just performed does not have to be
966 * replayed again on the next mount.
967 *
968 * We use the same quiesce mechanism as the rw->ro remount, as they are
969 * semantically identical operations.
970 */
971 if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
972 XFS_MOUNT_RDONLY) {
973 xfs_quiesce_attr(mp);
974 }
975
1da177e4
LT
976 /*
977 * Complete the quota initialisation, post-log-replay component.
978 */
7d095257
CH
979 if (quotamount) {
980 ASSERT(mp->m_qflags == 0);
981 mp->m_qflags = quotaflags;
982
983 xfs_qm_mount_quotas(mp);
984 }
985
84e1e99f
DC
986 /*
987 * Now we are mounted, reserve a small amount of unused space for
988 * privileged transactions. This is needed so that transaction
989 * space required for critical operations can dip into this pool
990 * when at ENOSPC. This is needed for operations like create with
991 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
992 * are not allowed to use this reserved space.
8babd8a2
DC
993 *
994 * This may drive us straight to ENOSPC on mount, but that implies
995 * we were already there on the last unmount. Warn if this occurs.
84e1e99f 996 */
d5db0f97
ES
997 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
998 resblks = xfs_default_resblks(mp);
999 error = xfs_reserve_blocks(mp, &resblks, NULL);
1000 if (error)
0b932ccc
DC
1001 xfs_warn(mp,
1002 "Unable to allocate reserve blocks. Continuing without reserve pool.");
174edb0e
DW
1003
1004 /* Recover any CoW blocks that never got remapped. */
1005 error = xfs_reflink_recover_cow(mp);
1006 if (error) {
1007 xfs_err(mp,
1008 "Error %d recovering leftover CoW allocations.", error);
1009 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1010 goto out_quota;
1011 }
84d69619
DW
1012
1013 /* Reserve AG blocks for future btree expansion. */
1014 error = xfs_fs_reserve_ag_blocks(mp);
1015 if (error && error != -ENOSPC)
1016 goto out_agresv;
d5db0f97 1017 }
84e1e99f 1018
1da177e4
LT
1019 return 0;
1020
84d69619
DW
1021 out_agresv:
1022 xfs_fs_unreserve_ag_blocks(mp);
174edb0e
DW
1023 out_quota:
1024 xfs_qm_unmount_quotas(mp);
b93b6e43 1025 out_rtunmount:
d0992452 1026 mp->m_super->s_flags &= ~MS_ACTIVE;
b93b6e43 1027 xfs_rtunmount_inodes(mp);
f9057e3d 1028 out_rele_rip:
43355099 1029 IRELE(rip);
0ae120f8
BF
1030 cancel_delayed_work_sync(&mp->m_reclaim_work);
1031 xfs_reclaim_inodes(mp, SYNC_WAIT);
f9057e3d 1032 out_log_dealloc:
e6b3bb78 1033 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
f0b2efad 1034 xfs_log_mount_cancel(mp);
d4f3512b
DC
1035 out_fail_wait:
1036 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1037 xfs_wait_buftarg(mp->m_logdev_targp);
1038 xfs_wait_buftarg(mp->m_ddev_targp);
f9057e3d 1039 out_free_perag:
ff4f038c 1040 xfs_free_perag(mp);
0650b554
DC
1041 out_free_dir:
1042 xfs_da_unmount(mp);
f9057e3d 1043 out_remove_uuid:
27174203 1044 xfs_uuid_unmount(mp);
192852be
CM
1045 out_remove_error_sysfs:
1046 xfs_error_sysfs_del(mp);
225e4635
BD
1047 out_del_stats:
1048 xfs_sysfs_del(&mp->m_stats.xs_kobj);
a31b1d3d
BF
1049 out_remove_sysfs:
1050 xfs_sysfs_del(&mp->m_kobj);
f9057e3d 1051 out:
1da177e4
LT
1052 return error;
1053}
1054
1055/*
1da177e4
LT
1056 * This flushes out the inodes,dquots and the superblock, unmounts the
1057 * log and makes sure that incore structures are freed.
1058 */
41b5c2e7
CH
1059void
1060xfs_unmountfs(
1061 struct xfs_mount *mp)
1da177e4 1062{
41b5c2e7
CH
1063 __uint64_t resblks;
1064 int error;
1da177e4 1065
579b62fa 1066 cancel_delayed_work_sync(&mp->m_eofblocks_work);
83104d44 1067 cancel_delayed_work_sync(&mp->m_cowblocks_work);
579b62fa 1068
84d69619 1069 xfs_fs_unreserve_ag_blocks(mp);
7d095257 1070 xfs_qm_unmount_quotas(mp);
b93b6e43 1071 xfs_rtunmount_inodes(mp);
77508ec8
CH
1072 IRELE(mp->m_rootip);
1073
641c56fb
DC
1074 /*
1075 * We can potentially deadlock here if we have an inode cluster
9da096fd 1076 * that has been freed has its buffer still pinned in memory because
641c56fb
DC
1077 * the transaction is still sitting in a iclog. The stale inodes
1078 * on that buffer will have their flush locks held until the
1079 * transaction hits the disk and the callbacks run. the inode
1080 * flush takes the flush lock unconditionally and with nothing to
1081 * push out the iclog we will never get that unlocked. hence we
1082 * need to force the log first.
1083 */
a14a348b 1084 xfs_log_force(mp, XFS_LOG_SYNC);
c854363e 1085
ebf55872
CH
1086 /*
1087 * Wait for all busy extents to be freed, including completion of
1088 * any discard operation.
1089 */
1090 xfs_extent_busy_wait_all(mp);
4560e78f 1091 flush_workqueue(xfs_discard_wq);
ebf55872 1092
e6b3bb78
CM
1093 /*
1094 * We now need to tell the world we are unmounting. This will allow
1095 * us to detect that the filesystem is going away and we should error
1096 * out anything that we have been retrying in the background. This will
1097 * prevent neverending retries in AIL pushing from hanging the unmount.
1098 */
1099 mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1100
c854363e 1101 /*
211e4d43
CH
1102 * Flush all pending changes from the AIL.
1103 */
1104 xfs_ail_push_all_sync(mp->m_ail);
1105
1106 /*
1107 * And reclaim all inodes. At this point there should be no dirty
7e18530b
DC
1108 * inodes and none should be pinned or locked, but use synchronous
1109 * reclaim just to be sure. We can stop background inode reclaim
1110 * here as well if it is still running.
c854363e 1111 */
7e18530b 1112 cancel_delayed_work_sync(&mp->m_reclaim_work);
c854363e 1113 xfs_reclaim_inodes(mp, SYNC_WAIT);
1da177e4 1114
7d095257 1115 xfs_qm_unmount(mp);
a357a121 1116
84e1e99f
DC
1117 /*
1118 * Unreserve any blocks we have so that when we unmount we don't account
1119 * the reserved free space as used. This is really only necessary for
1120 * lazy superblock counting because it trusts the incore superblock
9da096fd 1121 * counters to be absolutely correct on clean unmount.
84e1e99f
DC
1122 *
1123 * We don't bother correcting this elsewhere for lazy superblock
1124 * counting because on mount of an unclean filesystem we reconstruct the
1125 * correct counter value and this is irrelevant.
1126 *
1127 * For non-lazy counter filesystems, this doesn't matter at all because
1128 * we only every apply deltas to the superblock and hence the incore
1129 * value does not matter....
1130 */
1131 resblks = 0;
714082bc
DC
1132 error = xfs_reserve_blocks(mp, &resblks, NULL);
1133 if (error)
0b932ccc 1134 xfs_warn(mp, "Unable to free reserved block pool. "
714082bc
DC
1135 "Freespace may not be correct on next mount.");
1136
adab0f67 1137 error = xfs_log_sbcount(mp);
e5720eec 1138 if (error)
0b932ccc 1139 xfs_warn(mp, "Unable to update superblock counters. "
e5720eec 1140 "Freespace may not be correct on next mount.");
87c7bec7 1141
225e4635 1142
21b699c8 1143 xfs_log_unmount(mp);
0650b554 1144 xfs_da_unmount(mp);
27174203 1145 xfs_uuid_unmount(mp);
1da177e4 1146
1550d0b0 1147#if defined(DEBUG)
0ce4cfd4 1148 xfs_errortag_clearall(mp, 0);
1da177e4 1149#endif
ff4f038c 1150 xfs_free_perag(mp);
a31b1d3d 1151
192852be 1152 xfs_error_sysfs_del(mp);
225e4635 1153 xfs_sysfs_del(&mp->m_stats.xs_kobj);
a31b1d3d 1154 xfs_sysfs_del(&mp->m_kobj);
1da177e4
LT
1155}
1156
91ee575f
BF
1157/*
1158 * Determine whether modifications can proceed. The caller specifies the minimum
1159 * freeze level for which modifications should not be allowed. This allows
1160 * certain operations to proceed while the freeze sequence is in progress, if
1161 * necessary.
1162 */
1163bool
1164xfs_fs_writable(
1165 struct xfs_mount *mp,
1166 int level)
92821e2b 1167{
91ee575f
BF
1168 ASSERT(level > SB_UNFROZEN);
1169 if ((mp->m_super->s_writers.frozen >= level) ||
1170 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1171 return false;
1172
1173 return true;
92821e2b
DC
1174}
1175
1176/*
b2ce3974
AE
1177 * xfs_log_sbcount
1178 *
adab0f67 1179 * Sync the superblock counters to disk.
b2ce3974 1180 *
91ee575f
BF
1181 * Note this code can be called during the process of freezing, so we use the
1182 * transaction allocator that does not block when the transaction subsystem is
1183 * in its frozen state.
92821e2b
DC
1184 */
1185int
adab0f67 1186xfs_log_sbcount(xfs_mount_t *mp)
92821e2b 1187{
91ee575f
BF
1188 /* allow this to proceed during the freeze sequence... */
1189 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
92821e2b
DC
1190 return 0;
1191
92821e2b
DC
1192 /*
1193 * we don't need to do this if we are updating the superblock
1194 * counters on every modification.
1195 */
1196 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1197 return 0;
1198
61e63ecb 1199 return xfs_sync_sb(mp, true);
92821e2b
DC
1200}
1201
8c1903d3
DC
1202/*
1203 * Deltas for the inode count are +/-64, hence we use a large batch size
1204 * of 128 so we don't need to take the counter lock on every update.
1205 */
1206#define XFS_ICOUNT_BATCH 128
501ab323
DC
1207int
1208xfs_mod_icount(
1209 struct xfs_mount *mp,
1210 int64_t delta)
1211{
8c1903d3
DC
1212 __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1213 if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
501ab323
DC
1214 ASSERT(0);
1215 percpu_counter_add(&mp->m_icount, -delta);
1216 return -EINVAL;
1217 }
1218 return 0;
1219}
1220
e88b64ea
DC
1221int
1222xfs_mod_ifree(
1223 struct xfs_mount *mp,
1224 int64_t delta)
1225{
1226 percpu_counter_add(&mp->m_ifree, delta);
1227 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1228 ASSERT(0);
1229 percpu_counter_add(&mp->m_ifree, -delta);
1230 return -EINVAL;
1231 }
1232 return 0;
1233}
0d485ada 1234
8c1903d3
DC
1235/*
1236 * Deltas for the block count can vary from 1 to very large, but lock contention
1237 * only occurs on frequent small block count updates such as in the delayed
1238 * allocation path for buffered writes (page a time updates). Hence we set
1239 * a large batch count (1024) to minimise global counter updates except when
1240 * we get near to ENOSPC and we have to be very accurate with our updates.
1241 */
1242#define XFS_FDBLOCKS_BATCH 1024
0d485ada
DC
1243int
1244xfs_mod_fdblocks(
1245 struct xfs_mount *mp,
1246 int64_t delta,
1247 bool rsvd)
1248{
1249 int64_t lcounter;
1250 long long res_used;
1251 s32 batch;
1252
1253 if (delta > 0) {
1254 /*
1255 * If the reserve pool is depleted, put blocks back into it
1256 * first. Most of the time the pool is full.
1257 */
1258 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1259 percpu_counter_add(&mp->m_fdblocks, delta);
1260 return 0;
1261 }
1262
1263 spin_lock(&mp->m_sb_lock);
1264 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1265
1266 if (res_used > delta) {
1267 mp->m_resblks_avail += delta;
1268 } else {
1269 delta -= res_used;
1270 mp->m_resblks_avail = mp->m_resblks;
1271 percpu_counter_add(&mp->m_fdblocks, delta);
1272 }
1273 spin_unlock(&mp->m_sb_lock);
1274 return 0;
1275 }
1276
1277 /*
1278 * Taking blocks away, need to be more accurate the closer we
1279 * are to zero.
1280 *
0d485ada
DC
1281 * If the counter has a value of less than 2 * max batch size,
1282 * then make everything serialise as we are real close to
1283 * ENOSPC.
1284 */
8c1903d3
DC
1285 if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1286 XFS_FDBLOCKS_BATCH) < 0)
0d485ada
DC
1287 batch = 1;
1288 else
8c1903d3 1289 batch = XFS_FDBLOCKS_BATCH;
0d485ada
DC
1290
1291 __percpu_counter_add(&mp->m_fdblocks, delta, batch);
52548852 1292 if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
8c1903d3 1293 XFS_FDBLOCKS_BATCH) >= 0) {
0d485ada
DC
1294 /* we had space! */
1295 return 0;
1296 }
1297
1298 /*
1299 * lock up the sb for dipping into reserves before releasing the space
1300 * that took us to ENOSPC.
1301 */
1302 spin_lock(&mp->m_sb_lock);
1303 percpu_counter_add(&mp->m_fdblocks, -delta);
1304 if (!rsvd)
1305 goto fdblocks_enospc;
1306
1307 lcounter = (long long)mp->m_resblks_avail + delta;
1308 if (lcounter >= 0) {
1309 mp->m_resblks_avail = lcounter;
1310 spin_unlock(&mp->m_sb_lock);
1311 return 0;
1312 }
1313 printk_once(KERN_WARNING
1314 "Filesystem \"%s\": reserve blocks depleted! "
1315 "Consider increasing reserve pool size.",
1316 mp->m_fsname);
1317fdblocks_enospc:
1318 spin_unlock(&mp->m_sb_lock);
1319 return -ENOSPC;
1320}
1321
bab98bbe
DC
1322int
1323xfs_mod_frextents(
1324 struct xfs_mount *mp,
1325 int64_t delta)
1326{
1327 int64_t lcounter;
1328 int ret = 0;
1329
1330 spin_lock(&mp->m_sb_lock);
1331 lcounter = mp->m_sb.sb_frextents + delta;
1332 if (lcounter < 0)
1333 ret = -ENOSPC;
1334 else
1335 mp->m_sb.sb_frextents = lcounter;
1336 spin_unlock(&mp->m_sb_lock);
1337 return ret;
1338}
1339
1da177e4
LT
1340/*
1341 * xfs_getsb() is called to obtain the buffer for the superblock.
1342 * The buffer is returned locked and read in from disk.
1343 * The buffer should be released with a call to xfs_brelse().
1344 *
1345 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1346 * the superblock buffer if it can be locked without sleeping.
1347 * If it can't then we'll return NULL.
1348 */
0c842ad4 1349struct xfs_buf *
1da177e4 1350xfs_getsb(
0c842ad4
CH
1351 struct xfs_mount *mp,
1352 int flags)
1da177e4 1353{
0c842ad4 1354 struct xfs_buf *bp = mp->m_sb_bp;
1da177e4 1355
0c842ad4
CH
1356 if (!xfs_buf_trylock(bp)) {
1357 if (flags & XBF_TRYLOCK)
1da177e4 1358 return NULL;
0c842ad4 1359 xfs_buf_lock(bp);
1da177e4 1360 }
0c842ad4 1361
72790aa1 1362 xfs_buf_hold(bp);
b0388bf1 1363 ASSERT(bp->b_flags & XBF_DONE);
014c2544 1364 return bp;
1da177e4
LT
1365}
1366
1367/*
1368 * Used to free the superblock along various error paths.
1369 */
1370void
1371xfs_freesb(
26af6552 1372 struct xfs_mount *mp)
1da177e4 1373{
26af6552 1374 struct xfs_buf *bp = mp->m_sb_bp;
1da177e4 1375
26af6552 1376 xfs_buf_lock(bp);
1da177e4 1377 mp->m_sb_bp = NULL;
26af6552 1378 xfs_buf_relse(bp);
1da177e4
LT
1379}
1380
dda35b8f
CH
1381/*
1382 * If the underlying (data/log/rt) device is readonly, there are some
1383 * operations that cannot proceed.
1384 */
1385int
1386xfs_dev_is_read_only(
1387 struct xfs_mount *mp,
1388 char *message)
1389{
1390 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1391 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1392 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
0b932ccc
DC
1393 xfs_notice(mp, "%s required on read-only device.", message);
1394 xfs_notice(mp, "write access unavailable, cannot proceed.");
2451337d 1395 return -EROFS;
dda35b8f
CH
1396 }
1397 return 0;
1398}