Commit | Line | Data |
---|---|---|
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" |
0b1b213f | 48 | |
1da177e4 | 49 | |
27174203 CH |
50 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
51 | static int xfs_uuid_table_size; | |
52 | static uuid_t *xfs_uuid_table; | |
53 | ||
af3b6382 DW |
54 | void |
55 | xfs_uuid_table_free(void) | |
56 | { | |
57 | if (xfs_uuid_table_size == 0) | |
58 | return; | |
59 | kmem_free(xfs_uuid_table); | |
60 | xfs_uuid_table = NULL; | |
61 | xfs_uuid_table_size = 0; | |
62 | } | |
63 | ||
27174203 CH |
64 | /* |
65 | * See if the UUID is unique among mounted XFS filesystems. | |
66 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
67 | */ | |
68 | STATIC int | |
69 | xfs_uuid_mount( | |
70 | struct xfs_mount *mp) | |
71 | { | |
72 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
73 | int hole, i; | |
74 | ||
75 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
76 | return 0; | |
77 | ||
78 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 79 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
2451337d | 80 | return -EINVAL; |
27174203 CH |
81 | } |
82 | ||
83 | mutex_lock(&xfs_uuid_table_mutex); | |
84 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
85 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
86 | hole = i; | |
87 | continue; | |
88 | } | |
89 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
90 | goto out_duplicate; | |
91 | } | |
92 | ||
93 | if (hole < 0) { | |
94 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
95 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
27174203 CH |
96 | KM_SLEEP); |
97 | hole = xfs_uuid_table_size++; | |
98 | } | |
99 | xfs_uuid_table[hole] = *uuid; | |
100 | mutex_unlock(&xfs_uuid_table_mutex); | |
101 | ||
102 | return 0; | |
103 | ||
104 | out_duplicate: | |
105 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 106 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
2451337d | 107 | return -EINVAL; |
27174203 CH |
108 | } |
109 | ||
110 | STATIC void | |
111 | xfs_uuid_unmount( | |
112 | struct xfs_mount *mp) | |
113 | { | |
114 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
115 | int i; | |
116 | ||
117 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
118 | return; | |
119 | ||
120 | mutex_lock(&xfs_uuid_table_mutex); | |
121 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
122 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
123 | continue; | |
124 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
125 | continue; | |
126 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
127 | break; | |
128 | } | |
129 | ASSERT(i < xfs_uuid_table_size); | |
130 | mutex_unlock(&xfs_uuid_table_mutex); | |
131 | } | |
132 | ||
133 | ||
e176579e DC |
134 | STATIC void |
135 | __xfs_free_perag( | |
136 | struct rcu_head *head) | |
137 | { | |
138 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
139 | ||
140 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
141 | kmem_free(pag); | |
142 | } | |
143 | ||
1da177e4 | 144 | /* |
e176579e | 145 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 146 | */ |
c962fb79 | 147 | STATIC void |
ff4f038c | 148 | xfs_free_perag( |
745f6919 | 149 | xfs_mount_t *mp) |
1da177e4 | 150 | { |
1c1c6ebc DC |
151 | xfs_agnumber_t agno; |
152 | struct xfs_perag *pag; | |
153 | ||
154 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
155 | spin_lock(&mp->m_perag_lock); | |
156 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
157 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 158 | ASSERT(pag); |
f83282a8 | 159 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 160 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 161 | } |
1da177e4 LT |
162 | } |
163 | ||
4cc929ee NS |
164 | /* |
165 | * Check size of device based on the (data/realtime) block count. | |
166 | * Note: this check is used by the growfs code as well as mount. | |
167 | */ | |
168 | int | |
169 | xfs_sb_validate_fsb_count( | |
170 | xfs_sb_t *sbp, | |
171 | __uint64_t nblocks) | |
172 | { | |
173 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
174 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
175 | ||
d5cf09ba | 176 | /* Limited by ULONG_MAX of page cache index */ |
09cbfeaf | 177 | if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) |
2451337d | 178 | return -EFBIG; |
4cc929ee NS |
179 | return 0; |
180 | } | |
1da177e4 | 181 | |
1c1c6ebc | 182 | int |
c11e2c36 | 183 | xfs_initialize_perag( |
c11e2c36 | 184 | xfs_mount_t *mp, |
1c1c6ebc DC |
185 | xfs_agnumber_t agcount, |
186 | xfs_agnumber_t *maxagi) | |
1da177e4 | 187 | { |
2d2194f6 | 188 | xfs_agnumber_t index; |
8b26c582 | 189 | xfs_agnumber_t first_initialised = 0; |
1da177e4 | 190 | xfs_perag_t *pag; |
8b26c582 | 191 | int error = -ENOMEM; |
1da177e4 | 192 | |
1c1c6ebc DC |
193 | /* |
194 | * Walk the current per-ag tree so we don't try to initialise AGs | |
195 | * that already exist (growfs case). Allocate and insert all the | |
196 | * AGs we don't find ready for initialisation. | |
197 | */ | |
198 | for (index = 0; index < agcount; index++) { | |
199 | pag = xfs_perag_get(mp, index); | |
200 | if (pag) { | |
201 | xfs_perag_put(pag); | |
202 | continue; | |
203 | } | |
8b26c582 DC |
204 | if (!first_initialised) |
205 | first_initialised = index; | |
fb3b504a | 206 | |
1c1c6ebc DC |
207 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
208 | if (!pag) | |
8b26c582 | 209 | goto out_unwind; |
fb3b504a CH |
210 | pag->pag_agno = index; |
211 | pag->pag_mount = mp; | |
1a427ab0 | 212 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 213 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 214 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
215 | spin_lock_init(&pag->pag_buf_lock); |
216 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 217 | |
1c1c6ebc | 218 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 219 | goto out_unwind; |
fb3b504a | 220 | |
1c1c6ebc DC |
221 | spin_lock(&mp->m_perag_lock); |
222 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
223 | BUG(); | |
224 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
225 | radix_tree_preload_end(); |
226 | error = -EEXIST; | |
227 | goto out_unwind; | |
1c1c6ebc DC |
228 | } |
229 | spin_unlock(&mp->m_perag_lock); | |
230 | radix_tree_preload_end(); | |
231 | } | |
232 | ||
12c3f05c | 233 | index = xfs_set_inode_alloc(mp, agcount); |
fb3b504a | 234 | |
1c1c6ebc DC |
235 | if (maxagi) |
236 | *maxagi = index; | |
8018026e DW |
237 | |
238 | mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp); | |
1c1c6ebc | 239 | return 0; |
8b26c582 DC |
240 | |
241 | out_unwind: | |
242 | kmem_free(pag); | |
243 | for (; index > first_initialised; index--) { | |
244 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
245 | kmem_free(pag); | |
246 | } | |
247 | return error; | |
1da177e4 LT |
248 | } |
249 | ||
1da177e4 LT |
250 | /* |
251 | * xfs_readsb | |
252 | * | |
253 | * Does the initial read of the superblock. | |
254 | */ | |
255 | int | |
ff55068c DC |
256 | xfs_readsb( |
257 | struct xfs_mount *mp, | |
258 | int flags) | |
1da177e4 LT |
259 | { |
260 | unsigned int sector_size; | |
04a1e6c5 DC |
261 | struct xfs_buf *bp; |
262 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 263 | int error; |
af34e09d | 264 | int loud = !(flags & XFS_MFSI_QUIET); |
daba5427 | 265 | const struct xfs_buf_ops *buf_ops; |
1da177e4 LT |
266 | |
267 | ASSERT(mp->m_sb_bp == NULL); | |
268 | ASSERT(mp->m_ddev_targp != NULL); | |
269 | ||
daba5427 ES |
270 | /* |
271 | * For the initial read, we must guess at the sector | |
272 | * size based on the block device. It's enough to | |
273 | * get the sb_sectsize out of the superblock and | |
274 | * then reread with the proper length. | |
275 | * We don't verify it yet, because it may not be complete. | |
276 | */ | |
277 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
278 | buf_ops = NULL; | |
279 | ||
1da177e4 | 280 | /* |
c891c30a BF |
281 | * Allocate a (locked) buffer to hold the superblock. This will be kept |
282 | * around at all times to optimize access to the superblock. Therefore, | |
283 | * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count | |
284 | * elevated. | |
1da177e4 | 285 | */ |
26af6552 | 286 | reread: |
ba372674 | 287 | error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
c891c30a BF |
288 | BTOBB(sector_size), XBF_NO_IOACCT, &bp, |
289 | buf_ops); | |
ba372674 | 290 | if (error) { |
eab4e633 | 291 | if (loud) |
e721f504 | 292 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 293 | /* bad CRC means corrupted metadata */ |
2451337d DC |
294 | if (error == -EFSBADCRC) |
295 | error = -EFSCORRUPTED; | |
ba372674 | 296 | return error; |
eab4e633 | 297 | } |
1da177e4 LT |
298 | |
299 | /* | |
300 | * Initialize the mount structure from the superblock. | |
1da177e4 | 301 | */ |
556b8883 | 302 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
556b8883 DC |
303 | |
304 | /* | |
305 | * If we haven't validated the superblock, do so now before we try | |
306 | * to check the sector size and reread the superblock appropriately. | |
307 | */ | |
308 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
309 | if (loud) | |
310 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 311 | error = -EINVAL; |
556b8883 DC |
312 | goto release_buf; |
313 | } | |
ff55068c | 314 | |
1da177e4 LT |
315 | /* |
316 | * We must be able to do sector-sized and sector-aligned IO. | |
317 | */ | |
04a1e6c5 | 318 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
319 | if (loud) |
320 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 321 | sector_size, sbp->sb_sectsize); |
2451337d | 322 | error = -ENOSYS; |
26af6552 | 323 | goto release_buf; |
1da177e4 LT |
324 | } |
325 | ||
daba5427 | 326 | if (buf_ops == NULL) { |
556b8883 DC |
327 | /* |
328 | * Re-read the superblock so the buffer is correctly sized, | |
329 | * and properly verified. | |
330 | */ | |
1da177e4 | 331 | xfs_buf_relse(bp); |
04a1e6c5 | 332 | sector_size = sbp->sb_sectsize; |
daba5427 | 333 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 334 | goto reread; |
1da177e4 LT |
335 | } |
336 | ||
5681ca40 | 337 | xfs_reinit_percpu_counters(mp); |
8d280b98 | 338 | |
04a1e6c5 DC |
339 | /* no need to be quiet anymore, so reset the buf ops */ |
340 | bp->b_ops = &xfs_sb_buf_ops; | |
341 | ||
1da177e4 | 342 | mp->m_sb_bp = bp; |
26af6552 | 343 | xfs_buf_unlock(bp); |
1da177e4 LT |
344 | return 0; |
345 | ||
26af6552 DC |
346 | release_buf: |
347 | xfs_buf_relse(bp); | |
1da177e4 LT |
348 | return error; |
349 | } | |
350 | ||
1da177e4 | 351 | /* |
0771fb45 | 352 | * Update alignment values based on mount options and sb values |
1da177e4 | 353 | */ |
0771fb45 | 354 | STATIC int |
7884bc86 | 355 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 356 | { |
1da177e4 | 357 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 358 | |
4249023a | 359 | if (mp->m_dalign) { |
1da177e4 LT |
360 | /* |
361 | * If stripe unit and stripe width are not multiples | |
362 | * of the fs blocksize turn off alignment. | |
363 | */ | |
364 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
365 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
366 | xfs_warn(mp, |
367 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
368 | sbp->sb_blocksize); | |
2451337d | 369 | return -EINVAL; |
1da177e4 LT |
370 | } else { |
371 | /* | |
372 | * Convert the stripe unit and width to FSBs. | |
373 | */ | |
374 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
375 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 376 | xfs_warn(mp, |
39a45d84 JL |
377 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
378 | sbp->sb_agblocks); | |
2451337d | 379 | return -EINVAL; |
1da177e4 LT |
380 | } else if (mp->m_dalign) { |
381 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
382 | } else { | |
39a45d84 JL |
383 | xfs_warn(mp, |
384 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
385 | mp->m_dalign, sbp->sb_blocksize); | |
2451337d | 386 | return -EINVAL; |
1da177e4 LT |
387 | } |
388 | } | |
389 | ||
390 | /* | |
391 | * Update superblock with new values | |
392 | * and log changes | |
393 | */ | |
62118709 | 394 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
395 | if (sbp->sb_unit != mp->m_dalign) { |
396 | sbp->sb_unit = mp->m_dalign; | |
61e63ecb | 397 | mp->m_update_sb = true; |
1da177e4 LT |
398 | } |
399 | if (sbp->sb_width != mp->m_swidth) { | |
400 | sbp->sb_width = mp->m_swidth; | |
61e63ecb | 401 | mp->m_update_sb = true; |
1da177e4 | 402 | } |
34d7f603 JL |
403 | } else { |
404 | xfs_warn(mp, | |
405 | "cannot change alignment: superblock does not support data alignment"); | |
2451337d | 406 | return -EINVAL; |
1da177e4 LT |
407 | } |
408 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 409 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
410 | mp->m_dalign = sbp->sb_unit; |
411 | mp->m_swidth = sbp->sb_width; | |
412 | } | |
413 | ||
0771fb45 ES |
414 | return 0; |
415 | } | |
1da177e4 | 416 | |
0771fb45 ES |
417 | /* |
418 | * Set the maximum inode count for this filesystem | |
419 | */ | |
420 | STATIC void | |
421 | xfs_set_maxicount(xfs_mount_t *mp) | |
422 | { | |
423 | xfs_sb_t *sbp = &(mp->m_sb); | |
424 | __uint64_t icount; | |
1da177e4 | 425 | |
0771fb45 ES |
426 | if (sbp->sb_imax_pct) { |
427 | /* | |
428 | * Make sure the maximum inode count is a multiple | |
429 | * of the units we allocate inodes in. | |
1da177e4 | 430 | */ |
1da177e4 LT |
431 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
432 | do_div(icount, 100); | |
433 | do_div(icount, mp->m_ialloc_blks); | |
434 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
435 | sbp->sb_inopblog; | |
0771fb45 | 436 | } else { |
1da177e4 | 437 | mp->m_maxicount = 0; |
1da177e4 | 438 | } |
0771fb45 ES |
439 | } |
440 | ||
441 | /* | |
442 | * Set the default minimum read and write sizes unless | |
443 | * already specified in a mount option. | |
444 | * We use smaller I/O sizes when the file system | |
445 | * is being used for NFS service (wsync mount option). | |
446 | */ | |
447 | STATIC void | |
448 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
449 | { | |
450 | xfs_sb_t *sbp = &(mp->m_sb); | |
451 | int readio_log, writeio_log; | |
1da177e4 | 452 | |
1da177e4 LT |
453 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
454 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
455 | readio_log = XFS_WSYNC_READIO_LOG; | |
456 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
457 | } else { | |
458 | readio_log = XFS_READIO_LOG_LARGE; | |
459 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
460 | } | |
461 | } else { | |
462 | readio_log = mp->m_readio_log; | |
463 | writeio_log = mp->m_writeio_log; | |
464 | } | |
465 | ||
1da177e4 LT |
466 | if (sbp->sb_blocklog > readio_log) { |
467 | mp->m_readio_log = sbp->sb_blocklog; | |
468 | } else { | |
469 | mp->m_readio_log = readio_log; | |
470 | } | |
471 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
472 | if (sbp->sb_blocklog > writeio_log) { | |
473 | mp->m_writeio_log = sbp->sb_blocklog; | |
474 | } else { | |
475 | mp->m_writeio_log = writeio_log; | |
476 | } | |
477 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 478 | } |
1da177e4 | 479 | |
055388a3 DC |
480 | /* |
481 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
482 | */ | |
483 | void | |
484 | xfs_set_low_space_thresholds( | |
485 | struct xfs_mount *mp) | |
486 | { | |
487 | int i; | |
488 | ||
489 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
490 | __uint64_t space = mp->m_sb.sb_dblocks; | |
491 | ||
492 | do_div(space, 100); | |
493 | mp->m_low_space[i] = space * (i + 1); | |
494 | } | |
495 | } | |
496 | ||
497 | ||
0771fb45 ES |
498 | /* |
499 | * Set whether we're using inode alignment. | |
500 | */ | |
501 | STATIC void | |
502 | xfs_set_inoalignment(xfs_mount_t *mp) | |
503 | { | |
62118709 | 504 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
505 | mp->m_sb.sb_inoalignmt >= |
506 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
507 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
508 | else | |
509 | mp->m_inoalign_mask = 0; | |
510 | /* | |
511 | * If we are using stripe alignment, check whether | |
512 | * the stripe unit is a multiple of the inode alignment | |
513 | */ | |
514 | if (mp->m_dalign && mp->m_inoalign_mask && | |
515 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
516 | mp->m_sinoalign = mp->m_dalign; | |
517 | else | |
518 | mp->m_sinoalign = 0; | |
0771fb45 ES |
519 | } |
520 | ||
521 | /* | |
0471f62e | 522 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
523 | */ |
524 | STATIC int | |
ba372674 DC |
525 | xfs_check_sizes( |
526 | struct xfs_mount *mp) | |
0771fb45 | 527 | { |
ba372674 | 528 | struct xfs_buf *bp; |
0771fb45 | 529 | xfs_daddr_t d; |
ba372674 | 530 | int error; |
0771fb45 | 531 | |
1da177e4 LT |
532 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
533 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 534 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 535 | return -EFBIG; |
1da177e4 | 536 | } |
ba372674 | 537 | error = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 538 | d - XFS_FSS_TO_BB(mp, 1), |
ba372674 DC |
539 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
540 | if (error) { | |
0b932ccc | 541 | xfs_warn(mp, "last sector read failed"); |
ba372674 | 542 | return error; |
1da177e4 | 543 | } |
1922c949 | 544 | xfs_buf_relse(bp); |
1da177e4 | 545 | |
ba372674 DC |
546 | if (mp->m_logdev_targp == mp->m_ddev_targp) |
547 | return 0; | |
548 | ||
549 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
550 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
551 | xfs_warn(mp, "log size mismatch detected"); | |
552 | return -EFBIG; | |
553 | } | |
554 | error = xfs_buf_read_uncached(mp->m_logdev_targp, | |
1922c949 | 555 | d - XFS_FSB_TO_BB(mp, 1), |
ba372674 DC |
556 | XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); |
557 | if (error) { | |
558 | xfs_warn(mp, "log device read failed"); | |
559 | return error; | |
0771fb45 | 560 | } |
ba372674 | 561 | xfs_buf_relse(bp); |
0771fb45 ES |
562 | return 0; |
563 | } | |
564 | ||
7d095257 CH |
565 | /* |
566 | * Clear the quotaflags in memory and in the superblock. | |
567 | */ | |
568 | int | |
569 | xfs_mount_reset_sbqflags( | |
570 | struct xfs_mount *mp) | |
571 | { | |
7d095257 CH |
572 | mp->m_qflags = 0; |
573 | ||
61e63ecb | 574 | /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ |
7d095257 CH |
575 | if (mp->m_sb.sb_qflags == 0) |
576 | return 0; | |
577 | spin_lock(&mp->m_sb_lock); | |
578 | mp->m_sb.sb_qflags = 0; | |
579 | spin_unlock(&mp->m_sb_lock); | |
580 | ||
61e63ecb | 581 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
7d095257 CH |
582 | return 0; |
583 | ||
61e63ecb | 584 | return xfs_sync_sb(mp, false); |
7d095257 CH |
585 | } |
586 | ||
d5db0f97 ES |
587 | __uint64_t |
588 | xfs_default_resblks(xfs_mount_t *mp) | |
589 | { | |
590 | __uint64_t resblks; | |
591 | ||
592 | /* | |
8babd8a2 DC |
593 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
594 | * smaller. This is intended to cover concurrent allocation | |
595 | * transactions when we initially hit enospc. These each require a 4 | |
596 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
597 | * allocation reservations. | |
d5db0f97 ES |
598 | */ |
599 | resblks = mp->m_sb.sb_dblocks; | |
600 | do_div(resblks, 20); | |
8babd8a2 | 601 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
602 | return resblks; |
603 | } | |
604 | ||
0771fb45 | 605 | /* |
0771fb45 ES |
606 | * This function does the following on an initial mount of a file system: |
607 | * - reads the superblock from disk and init the mount struct | |
608 | * - if we're a 32-bit kernel, do a size check on the superblock | |
609 | * so we don't mount terabyte filesystems | |
610 | * - init mount struct realtime fields | |
611 | * - allocate inode hash table for fs | |
612 | * - init directory manager | |
613 | * - perform recovery and init the log manager | |
614 | */ | |
615 | int | |
616 | xfs_mountfs( | |
f0b2efad | 617 | struct xfs_mount *mp) |
0771fb45 | 618 | { |
f0b2efad BF |
619 | struct xfs_sb *sbp = &(mp->m_sb); |
620 | struct xfs_inode *rip; | |
621 | __uint64_t resblks; | |
622 | uint quotamount = 0; | |
623 | uint quotaflags = 0; | |
624 | int error = 0; | |
0771fb45 | 625 | |
ff55068c | 626 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 627 | |
ee1c0908 | 628 | /* |
074e427b DC |
629 | * Check for a mismatched features2 values. Older kernels read & wrote |
630 | * into the wrong sb offset for sb_features2 on some platforms due to | |
631 | * xfs_sb_t not being 64bit size aligned when sb_features2 was added, | |
632 | * which made older superblock reading/writing routines swap it as a | |
633 | * 64-bit value. | |
ee1c0908 | 634 | * |
e6957ea4 ES |
635 | * For backwards compatibility, we make both slots equal. |
636 | * | |
074e427b DC |
637 | * If we detect a mismatched field, we OR the set bits into the existing |
638 | * features2 field in case it has already been modified; we don't want | |
639 | * to lose any features. We then update the bad location with the ORed | |
640 | * value so that older kernels will see any features2 flags. The | |
641 | * superblock writeback code ensures the new sb_features2 is copied to | |
642 | * sb_bad_features2 before it is logged or written to disk. | |
ee1c0908 | 643 | */ |
e6957ea4 | 644 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 645 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 646 | sbp->sb_features2 |= sbp->sb_bad_features2; |
61e63ecb | 647 | mp->m_update_sb = true; |
e6957ea4 ES |
648 | |
649 | /* | |
650 | * Re-check for ATTR2 in case it was found in bad_features2 | |
651 | * slot. | |
652 | */ | |
7c12f296 TS |
653 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
654 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 655 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
656 | } |
657 | ||
658 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
659 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
660 | xfs_sb_version_removeattr2(&mp->m_sb); | |
61e63ecb | 661 | mp->m_update_sb = true; |
e6957ea4 | 662 | |
7c12f296 TS |
663 | /* update sb_versionnum for the clearing of the morebits */ |
664 | if (!sbp->sb_features2) | |
61e63ecb | 665 | mp->m_update_sb = true; |
ee1c0908 DC |
666 | } |
667 | ||
263997a6 DC |
668 | /* always use v2 inodes by default now */ |
669 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
670 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
61e63ecb | 671 | mp->m_update_sb = true; |
263997a6 DC |
672 | } |
673 | ||
0771fb45 ES |
674 | /* |
675 | * Check if sb_agblocks is aligned at stripe boundary | |
676 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
677 | * allocator alignment is within an ag, therefore ag has | |
678 | * to be aligned at stripe boundary. | |
679 | */ | |
7884bc86 | 680 | error = xfs_update_alignment(mp); |
0771fb45 | 681 | if (error) |
f9057e3d | 682 | goto out; |
0771fb45 ES |
683 | |
684 | xfs_alloc_compute_maxlevels(mp); | |
685 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
686 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
687 | xfs_ialloc_compute_maxlevels(mp); | |
035e00ac | 688 | xfs_rmapbt_compute_maxlevels(mp); |
1946b91c | 689 | xfs_refcountbt_compute_maxlevels(mp); |
0771fb45 ES |
690 | |
691 | xfs_set_maxicount(mp); | |
692 | ||
e6b3bb78 CM |
693 | /* enable fail_at_unmount as default */ |
694 | mp->m_fail_unmount = 1; | |
695 | ||
a31b1d3d | 696 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); |
27174203 CH |
697 | if (error) |
698 | goto out; | |
1da177e4 | 699 | |
225e4635 BD |
700 | error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, |
701 | &mp->m_kobj, "stats"); | |
a31b1d3d BF |
702 | if (error) |
703 | goto out_remove_sysfs; | |
704 | ||
192852be | 705 | error = xfs_error_sysfs_init(mp); |
225e4635 BD |
706 | if (error) |
707 | goto out_del_stats; | |
708 | ||
192852be CM |
709 | |
710 | error = xfs_uuid_mount(mp); | |
711 | if (error) | |
712 | goto out_remove_error_sysfs; | |
713 | ||
0771fb45 ES |
714 | /* |
715 | * Set the minimum read and write sizes | |
716 | */ | |
717 | xfs_set_rw_sizes(mp); | |
718 | ||
055388a3 DC |
719 | /* set the low space thresholds for dynamic preallocation */ |
720 | xfs_set_low_space_thresholds(mp); | |
721 | ||
0771fb45 ES |
722 | /* |
723 | * Set the inode cluster size. | |
724 | * This may still be overridden by the file system | |
725 | * block size if it is larger than the chosen cluster size. | |
8f80587b DC |
726 | * |
727 | * For v5 filesystems, scale the cluster size with the inode size to | |
728 | * keep a constant ratio of inode per cluster buffer, but only if mkfs | |
729 | * has set the inode alignment value appropriately for larger cluster | |
730 | * sizes. | |
0771fb45 ES |
731 | */ |
732 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
8f80587b DC |
733 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
734 | int new_size = mp->m_inode_cluster_size; | |
735 | ||
736 | new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; | |
737 | if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) | |
738 | mp->m_inode_cluster_size = new_size; | |
8f80587b | 739 | } |
0771fb45 | 740 | |
e5376fc1 BF |
741 | /* |
742 | * If enabled, sparse inode chunk alignment is expected to match the | |
743 | * cluster size. Full inode chunk alignment must match the chunk size, | |
744 | * but that is checked on sb read verification... | |
745 | */ | |
746 | if (xfs_sb_version_hassparseinodes(&mp->m_sb) && | |
747 | mp->m_sb.sb_spino_align != | |
748 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) { | |
749 | xfs_warn(mp, | |
750 | "Sparse inode block alignment (%u) must match cluster size (%llu).", | |
751 | mp->m_sb.sb_spino_align, | |
752 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)); | |
753 | error = -EINVAL; | |
754 | goto out_remove_uuid; | |
755 | } | |
756 | ||
0771fb45 ES |
757 | /* |
758 | * Set inode alignment fields | |
759 | */ | |
760 | xfs_set_inoalignment(mp); | |
761 | ||
762 | /* | |
c2bfbc9b | 763 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 764 | */ |
4249023a | 765 | error = xfs_check_sizes(mp); |
0771fb45 | 766 | if (error) |
f9057e3d | 767 | goto out_remove_uuid; |
0771fb45 | 768 | |
1da177e4 LT |
769 | /* |
770 | * Initialize realtime fields in the mount structure | |
771 | */ | |
0771fb45 ES |
772 | error = xfs_rtmount_init(mp); |
773 | if (error) { | |
0b932ccc | 774 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 775 | goto out_remove_uuid; |
1da177e4 LT |
776 | } |
777 | ||
1da177e4 LT |
778 | /* |
779 | * Copies the low order bits of the timestamp and the randomly | |
780 | * set "sequence" number out of a UUID. | |
781 | */ | |
782 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
783 | ||
1da177e4 LT |
784 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
785 | ||
0650b554 DC |
786 | error = xfs_da_mount(mp); |
787 | if (error) { | |
788 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
789 | goto out_remove_uuid; | |
790 | } | |
1da177e4 LT |
791 | |
792 | /* | |
793 | * Initialize the precomputed transaction reservations values. | |
794 | */ | |
795 | xfs_trans_init(mp); | |
796 | ||
1da177e4 LT |
797 | /* |
798 | * Allocate and initialize the per-ag data. | |
799 | */ | |
1c1c6ebc | 800 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 801 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
802 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
803 | if (error) { | |
0b932ccc | 804 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 805 | goto out_free_dir; |
1c1c6ebc | 806 | } |
1da177e4 | 807 | |
f9057e3d | 808 | if (!sbp->sb_logblocks) { |
0b932ccc | 809 | xfs_warn(mp, "no log defined"); |
f9057e3d | 810 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
2451337d | 811 | error = -EFSCORRUPTED; |
f9057e3d CH |
812 | goto out_free_perag; |
813 | } | |
814 | ||
1da177e4 | 815 | /* |
f0b2efad BF |
816 | * Log's mount-time initialization. The first part of recovery can place |
817 | * some items on the AIL, to be handled when recovery is finished or | |
818 | * cancelled. | |
1da177e4 | 819 | */ |
f9057e3d CH |
820 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
821 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
822 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
823 | if (error) { | |
0b932ccc | 824 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 825 | goto out_fail_wait; |
1da177e4 LT |
826 | } |
827 | ||
92821e2b DC |
828 | /* |
829 | * Now the log is mounted, we know if it was an unclean shutdown or | |
830 | * not. If it was, with the first phase of recovery has completed, we | |
831 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
832 | * but they are recovered transactionally in the second recovery phase | |
833 | * later. | |
834 | * | |
835 | * Hence we can safely re-initialise incore superblock counters from | |
836 | * the per-ag data. These may not be correct if the filesystem was not | |
837 | * cleanly unmounted, so we need to wait for recovery to finish before | |
838 | * doing this. | |
839 | * | |
840 | * If the filesystem was cleanly unmounted, then we can trust the | |
841 | * values in the superblock to be correct and we don't need to do | |
842 | * anything here. | |
843 | * | |
844 | * If we are currently making the filesystem, the initialisation will | |
845 | * fail as the perag data is in an undefined state. | |
846 | */ | |
92821e2b DC |
847 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
848 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
849 | !mp->m_sb.sb_inprogress) { | |
850 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 851 | if (error) |
6eee8972 | 852 | goto out_log_dealloc; |
92821e2b | 853 | } |
f9057e3d | 854 | |
1da177e4 LT |
855 | /* |
856 | * Get and sanity-check the root inode. | |
857 | * Save the pointer to it in the mount structure. | |
858 | */ | |
7b6259e7 | 859 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 860 | if (error) { |
0b932ccc | 861 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 862 | goto out_log_dealloc; |
1da177e4 LT |
863 | } |
864 | ||
865 | ASSERT(rip != NULL); | |
1da177e4 | 866 | |
c19b3b05 | 867 | if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) { |
0b932ccc | 868 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 869 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
870 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
871 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
872 | mp); | |
2451337d | 873 | error = -EFSCORRUPTED; |
f9057e3d | 874 | goto out_rele_rip; |
1da177e4 LT |
875 | } |
876 | mp->m_rootip = rip; /* save it */ | |
877 | ||
878 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
879 | ||
880 | /* | |
881 | * Initialize realtime inode pointers in the mount structure | |
882 | */ | |
0771fb45 ES |
883 | error = xfs_rtmount_inodes(mp); |
884 | if (error) { | |
1da177e4 LT |
885 | /* |
886 | * Free up the root inode. | |
887 | */ | |
0b932ccc | 888 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 889 | goto out_rele_rip; |
1da177e4 LT |
890 | } |
891 | ||
892 | /* | |
7884bc86 CH |
893 | * If this is a read-only mount defer the superblock updates until |
894 | * the next remount into writeable mode. Otherwise we would never | |
895 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 896 | */ |
61e63ecb DC |
897 | if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
898 | error = xfs_sync_sb(mp, false); | |
e5720eec | 899 | if (error) { |
0b932ccc | 900 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 901 | goto out_rtunmount; |
e5720eec DC |
902 | } |
903 | } | |
1da177e4 LT |
904 | |
905 | /* | |
906 | * Initialise the XFS quota management subsystem for this mount | |
907 | */ | |
7d095257 CH |
908 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
909 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
910 | if (error) | |
911 | goto out_rtunmount; | |
912 | } else { | |
913 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
914 | ||
915 | /* | |
916 | * If a file system had quotas running earlier, but decided to | |
917 | * mount without -o uquota/pquota/gquota options, revoke the | |
918 | * quotachecked license. | |
919 | */ | |
920 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 921 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
922 | error = xfs_mount_reset_sbqflags(mp); |
923 | if (error) | |
a70a4fa5 | 924 | goto out_rtunmount; |
7d095257 CH |
925 | } |
926 | } | |
1da177e4 | 927 | |
17c12bcd DW |
928 | /* |
929 | * During the second phase of log recovery, we need iget and | |
930 | * iput to behave like they do for an active filesystem. | |
931 | * xfs_fs_drop_inode needs to be able to prevent the deletion | |
932 | * of inodes before we're done replaying log items on those | |
933 | * inodes. | |
934 | */ | |
935 | mp->m_super->s_flags |= MS_ACTIVE; | |
936 | ||
1da177e4 | 937 | /* |
f0b2efad BF |
938 | * Finish recovering the file system. This part needed to be delayed |
939 | * until after the root and real-time bitmap inodes were consistently | |
940 | * read in. | |
1da177e4 | 941 | */ |
4249023a | 942 | error = xfs_log_mount_finish(mp); |
1da177e4 | 943 | if (error) { |
0b932ccc | 944 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 945 | goto out_rtunmount; |
1da177e4 LT |
946 | } |
947 | ||
ddeb14f4 DC |
948 | /* |
949 | * Now the log is fully replayed, we can transition to full read-only | |
950 | * mode for read-only mounts. This will sync all the metadata and clean | |
951 | * the log so that the recovery we just performed does not have to be | |
952 | * replayed again on the next mount. | |
953 | * | |
954 | * We use the same quiesce mechanism as the rw->ro remount, as they are | |
955 | * semantically identical operations. | |
956 | */ | |
957 | if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) == | |
958 | XFS_MOUNT_RDONLY) { | |
959 | xfs_quiesce_attr(mp); | |
960 | } | |
961 | ||
1da177e4 LT |
962 | /* |
963 | * Complete the quota initialisation, post-log-replay component. | |
964 | */ | |
7d095257 CH |
965 | if (quotamount) { |
966 | ASSERT(mp->m_qflags == 0); | |
967 | mp->m_qflags = quotaflags; | |
968 | ||
969 | xfs_qm_mount_quotas(mp); | |
970 | } | |
971 | ||
84e1e99f DC |
972 | /* |
973 | * Now we are mounted, reserve a small amount of unused space for | |
974 | * privileged transactions. This is needed so that transaction | |
975 | * space required for critical operations can dip into this pool | |
976 | * when at ENOSPC. This is needed for operations like create with | |
977 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
978 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
979 | * |
980 | * This may drive us straight to ENOSPC on mount, but that implies | |
981 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 982 | */ |
d5db0f97 ES |
983 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
984 | resblks = xfs_default_resblks(mp); | |
985 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
986 | if (error) | |
0b932ccc DC |
987 | xfs_warn(mp, |
988 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
174edb0e DW |
989 | |
990 | /* Recover any CoW blocks that never got remapped. */ | |
991 | error = xfs_reflink_recover_cow(mp); | |
992 | if (error) { | |
993 | xfs_err(mp, | |
994 | "Error %d recovering leftover CoW allocations.", error); | |
995 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
996 | goto out_quota; | |
997 | } | |
84d69619 DW |
998 | |
999 | /* Reserve AG blocks for future btree expansion. */ | |
1000 | error = xfs_fs_reserve_ag_blocks(mp); | |
1001 | if (error && error != -ENOSPC) | |
1002 | goto out_agresv; | |
d5db0f97 | 1003 | } |
84e1e99f | 1004 | |
1da177e4 LT |
1005 | return 0; |
1006 | ||
84d69619 DW |
1007 | out_agresv: |
1008 | xfs_fs_unreserve_ag_blocks(mp); | |
174edb0e DW |
1009 | out_quota: |
1010 | xfs_qm_unmount_quotas(mp); | |
b93b6e43 | 1011 | out_rtunmount: |
d0992452 | 1012 | mp->m_super->s_flags &= ~MS_ACTIVE; |
b93b6e43 | 1013 | xfs_rtunmount_inodes(mp); |
f9057e3d | 1014 | out_rele_rip: |
43355099 | 1015 | IRELE(rip); |
0ae120f8 BF |
1016 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
1017 | xfs_reclaim_inodes(mp, SYNC_WAIT); | |
f9057e3d | 1018 | out_log_dealloc: |
e6b3bb78 | 1019 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; |
f0b2efad | 1020 | xfs_log_mount_cancel(mp); |
d4f3512b DC |
1021 | out_fail_wait: |
1022 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
1023 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1024 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 1025 | out_free_perag: |
ff4f038c | 1026 | xfs_free_perag(mp); |
0650b554 DC |
1027 | out_free_dir: |
1028 | xfs_da_unmount(mp); | |
f9057e3d | 1029 | out_remove_uuid: |
27174203 | 1030 | xfs_uuid_unmount(mp); |
192852be CM |
1031 | out_remove_error_sysfs: |
1032 | xfs_error_sysfs_del(mp); | |
225e4635 BD |
1033 | out_del_stats: |
1034 | xfs_sysfs_del(&mp->m_stats.xs_kobj); | |
a31b1d3d BF |
1035 | out_remove_sysfs: |
1036 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 1037 | out: |
1da177e4 LT |
1038 | return error; |
1039 | } | |
1040 | ||
1041 | /* | |
1da177e4 LT |
1042 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1043 | * log and makes sure that incore structures are freed. | |
1044 | */ | |
41b5c2e7 CH |
1045 | void |
1046 | xfs_unmountfs( | |
1047 | struct xfs_mount *mp) | |
1da177e4 | 1048 | { |
41b5c2e7 CH |
1049 | __uint64_t resblks; |
1050 | int error; | |
1da177e4 | 1051 | |
579b62fa | 1052 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
83104d44 | 1053 | cancel_delayed_work_sync(&mp->m_cowblocks_work); |
579b62fa | 1054 | |
84d69619 | 1055 | xfs_fs_unreserve_ag_blocks(mp); |
7d095257 | 1056 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1057 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1058 | IRELE(mp->m_rootip); |
1059 | ||
641c56fb DC |
1060 | /* |
1061 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1062 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1063 | * the transaction is still sitting in a iclog. The stale inodes |
1064 | * on that buffer will have their flush locks held until the | |
1065 | * transaction hits the disk and the callbacks run. the inode | |
1066 | * flush takes the flush lock unconditionally and with nothing to | |
1067 | * push out the iclog we will never get that unlocked. hence we | |
1068 | * need to force the log first. | |
1069 | */ | |
a14a348b | 1070 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e | 1071 | |
e6b3bb78 CM |
1072 | /* |
1073 | * We now need to tell the world we are unmounting. This will allow | |
1074 | * us to detect that the filesystem is going away and we should error | |
1075 | * out anything that we have been retrying in the background. This will | |
1076 | * prevent neverending retries in AIL pushing from hanging the unmount. | |
1077 | */ | |
1078 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; | |
1079 | ||
c854363e | 1080 | /* |
211e4d43 CH |
1081 | * Flush all pending changes from the AIL. |
1082 | */ | |
1083 | xfs_ail_push_all_sync(mp->m_ail); | |
1084 | ||
1085 | /* | |
1086 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1087 | * inodes and none should be pinned or locked, but use synchronous |
1088 | * reclaim just to be sure. We can stop background inode reclaim | |
1089 | * here as well if it is still running. | |
c854363e | 1090 | */ |
7e18530b | 1091 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1092 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1093 | |
7d095257 | 1094 | xfs_qm_unmount(mp); |
a357a121 | 1095 | |
84e1e99f DC |
1096 | /* |
1097 | * Unreserve any blocks we have so that when we unmount we don't account | |
1098 | * the reserved free space as used. This is really only necessary for | |
1099 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1100 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1101 | * |
1102 | * We don't bother correcting this elsewhere for lazy superblock | |
1103 | * counting because on mount of an unclean filesystem we reconstruct the | |
1104 | * correct counter value and this is irrelevant. | |
1105 | * | |
1106 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1107 | * we only every apply deltas to the superblock and hence the incore | |
1108 | * value does not matter.... | |
1109 | */ | |
1110 | resblks = 0; | |
714082bc DC |
1111 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1112 | if (error) | |
0b932ccc | 1113 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1114 | "Freespace may not be correct on next mount."); |
1115 | ||
adab0f67 | 1116 | error = xfs_log_sbcount(mp); |
e5720eec | 1117 | if (error) |
0b932ccc | 1118 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1119 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1120 | |
225e4635 | 1121 | |
21b699c8 | 1122 | xfs_log_unmount(mp); |
0650b554 | 1123 | xfs_da_unmount(mp); |
27174203 | 1124 | xfs_uuid_unmount(mp); |
1da177e4 | 1125 | |
1550d0b0 | 1126 | #if defined(DEBUG) |
0ce4cfd4 | 1127 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1128 | #endif |
ff4f038c | 1129 | xfs_free_perag(mp); |
a31b1d3d | 1130 | |
192852be | 1131 | xfs_error_sysfs_del(mp); |
225e4635 | 1132 | xfs_sysfs_del(&mp->m_stats.xs_kobj); |
a31b1d3d | 1133 | xfs_sysfs_del(&mp->m_kobj); |
1da177e4 LT |
1134 | } |
1135 | ||
91ee575f BF |
1136 | /* |
1137 | * Determine whether modifications can proceed. The caller specifies the minimum | |
1138 | * freeze level for which modifications should not be allowed. This allows | |
1139 | * certain operations to proceed while the freeze sequence is in progress, if | |
1140 | * necessary. | |
1141 | */ | |
1142 | bool | |
1143 | xfs_fs_writable( | |
1144 | struct xfs_mount *mp, | |
1145 | int level) | |
92821e2b | 1146 | { |
91ee575f BF |
1147 | ASSERT(level > SB_UNFROZEN); |
1148 | if ((mp->m_super->s_writers.frozen >= level) || | |
1149 | XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) | |
1150 | return false; | |
1151 | ||
1152 | return true; | |
92821e2b DC |
1153 | } |
1154 | ||
1155 | /* | |
b2ce3974 AE |
1156 | * xfs_log_sbcount |
1157 | * | |
adab0f67 | 1158 | * Sync the superblock counters to disk. |
b2ce3974 | 1159 | * |
91ee575f BF |
1160 | * Note this code can be called during the process of freezing, so we use the |
1161 | * transaction allocator that does not block when the transaction subsystem is | |
1162 | * in its frozen state. | |
92821e2b DC |
1163 | */ |
1164 | int | |
adab0f67 | 1165 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b | 1166 | { |
91ee575f BF |
1167 | /* allow this to proceed during the freeze sequence... */ |
1168 | if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) | |
92821e2b DC |
1169 | return 0; |
1170 | ||
92821e2b DC |
1171 | /* |
1172 | * we don't need to do this if we are updating the superblock | |
1173 | * counters on every modification. | |
1174 | */ | |
1175 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1176 | return 0; | |
1177 | ||
61e63ecb | 1178 | return xfs_sync_sb(mp, true); |
92821e2b DC |
1179 | } |
1180 | ||
8c1903d3 DC |
1181 | /* |
1182 | * Deltas for the inode count are +/-64, hence we use a large batch size | |
1183 | * of 128 so we don't need to take the counter lock on every update. | |
1184 | */ | |
1185 | #define XFS_ICOUNT_BATCH 128 | |
501ab323 DC |
1186 | int |
1187 | xfs_mod_icount( | |
1188 | struct xfs_mount *mp, | |
1189 | int64_t delta) | |
1190 | { | |
8c1903d3 DC |
1191 | __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH); |
1192 | if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) { | |
501ab323 DC |
1193 | ASSERT(0); |
1194 | percpu_counter_add(&mp->m_icount, -delta); | |
1195 | return -EINVAL; | |
1196 | } | |
1197 | return 0; | |
1198 | } | |
1199 | ||
e88b64ea DC |
1200 | int |
1201 | xfs_mod_ifree( | |
1202 | struct xfs_mount *mp, | |
1203 | int64_t delta) | |
1204 | { | |
1205 | percpu_counter_add(&mp->m_ifree, delta); | |
1206 | if (percpu_counter_compare(&mp->m_ifree, 0) < 0) { | |
1207 | ASSERT(0); | |
1208 | percpu_counter_add(&mp->m_ifree, -delta); | |
1209 | return -EINVAL; | |
1210 | } | |
1211 | return 0; | |
1212 | } | |
0d485ada | 1213 | |
8c1903d3 DC |
1214 | /* |
1215 | * Deltas for the block count can vary from 1 to very large, but lock contention | |
1216 | * only occurs on frequent small block count updates such as in the delayed | |
1217 | * allocation path for buffered writes (page a time updates). Hence we set | |
1218 | * a large batch count (1024) to minimise global counter updates except when | |
1219 | * we get near to ENOSPC and we have to be very accurate with our updates. | |
1220 | */ | |
1221 | #define XFS_FDBLOCKS_BATCH 1024 | |
0d485ada DC |
1222 | int |
1223 | xfs_mod_fdblocks( | |
1224 | struct xfs_mount *mp, | |
1225 | int64_t delta, | |
1226 | bool rsvd) | |
1227 | { | |
1228 | int64_t lcounter; | |
1229 | long long res_used; | |
1230 | s32 batch; | |
1231 | ||
1232 | if (delta > 0) { | |
1233 | /* | |
1234 | * If the reserve pool is depleted, put blocks back into it | |
1235 | * first. Most of the time the pool is full. | |
1236 | */ | |
1237 | if (likely(mp->m_resblks == mp->m_resblks_avail)) { | |
1238 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1239 | return 0; | |
1240 | } | |
1241 | ||
1242 | spin_lock(&mp->m_sb_lock); | |
1243 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); | |
1244 | ||
1245 | if (res_used > delta) { | |
1246 | mp->m_resblks_avail += delta; | |
1247 | } else { | |
1248 | delta -= res_used; | |
1249 | mp->m_resblks_avail = mp->m_resblks; | |
1250 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1251 | } | |
1252 | spin_unlock(&mp->m_sb_lock); | |
1253 | return 0; | |
1254 | } | |
1255 | ||
1256 | /* | |
1257 | * Taking blocks away, need to be more accurate the closer we | |
1258 | * are to zero. | |
1259 | * | |
0d485ada DC |
1260 | * If the counter has a value of less than 2 * max batch size, |
1261 | * then make everything serialise as we are real close to | |
1262 | * ENOSPC. | |
1263 | */ | |
8c1903d3 DC |
1264 | if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, |
1265 | XFS_FDBLOCKS_BATCH) < 0) | |
0d485ada DC |
1266 | batch = 1; |
1267 | else | |
8c1903d3 | 1268 | batch = XFS_FDBLOCKS_BATCH; |
0d485ada DC |
1269 | |
1270 | __percpu_counter_add(&mp->m_fdblocks, delta, batch); | |
52548852 | 1271 | if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside, |
8c1903d3 | 1272 | XFS_FDBLOCKS_BATCH) >= 0) { |
0d485ada DC |
1273 | /* we had space! */ |
1274 | return 0; | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * lock up the sb for dipping into reserves before releasing the space | |
1279 | * that took us to ENOSPC. | |
1280 | */ | |
1281 | spin_lock(&mp->m_sb_lock); | |
1282 | percpu_counter_add(&mp->m_fdblocks, -delta); | |
1283 | if (!rsvd) | |
1284 | goto fdblocks_enospc; | |
1285 | ||
1286 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1287 | if (lcounter >= 0) { | |
1288 | mp->m_resblks_avail = lcounter; | |
1289 | spin_unlock(&mp->m_sb_lock); | |
1290 | return 0; | |
1291 | } | |
1292 | printk_once(KERN_WARNING | |
1293 | "Filesystem \"%s\": reserve blocks depleted! " | |
1294 | "Consider increasing reserve pool size.", | |
1295 | mp->m_fsname); | |
1296 | fdblocks_enospc: | |
1297 | spin_unlock(&mp->m_sb_lock); | |
1298 | return -ENOSPC; | |
1299 | } | |
1300 | ||
bab98bbe DC |
1301 | int |
1302 | xfs_mod_frextents( | |
1303 | struct xfs_mount *mp, | |
1304 | int64_t delta) | |
1305 | { | |
1306 | int64_t lcounter; | |
1307 | int ret = 0; | |
1308 | ||
1309 | spin_lock(&mp->m_sb_lock); | |
1310 | lcounter = mp->m_sb.sb_frextents + delta; | |
1311 | if (lcounter < 0) | |
1312 | ret = -ENOSPC; | |
1313 | else | |
1314 | mp->m_sb.sb_frextents = lcounter; | |
1315 | spin_unlock(&mp->m_sb_lock); | |
1316 | return ret; | |
1317 | } | |
1318 | ||
1da177e4 LT |
1319 | /* |
1320 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1321 | * The buffer is returned locked and read in from disk. | |
1322 | * The buffer should be released with a call to xfs_brelse(). | |
1323 | * | |
1324 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1325 | * the superblock buffer if it can be locked without sleeping. | |
1326 | * If it can't then we'll return NULL. | |
1327 | */ | |
0c842ad4 | 1328 | struct xfs_buf * |
1da177e4 | 1329 | xfs_getsb( |
0c842ad4 CH |
1330 | struct xfs_mount *mp, |
1331 | int flags) | |
1da177e4 | 1332 | { |
0c842ad4 | 1333 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1334 | |
0c842ad4 CH |
1335 | if (!xfs_buf_trylock(bp)) { |
1336 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1337 | return NULL; |
0c842ad4 | 1338 | xfs_buf_lock(bp); |
1da177e4 | 1339 | } |
0c842ad4 | 1340 | |
72790aa1 | 1341 | xfs_buf_hold(bp); |
b0388bf1 | 1342 | ASSERT(bp->b_flags & XBF_DONE); |
014c2544 | 1343 | return bp; |
1da177e4 LT |
1344 | } |
1345 | ||
1346 | /* | |
1347 | * Used to free the superblock along various error paths. | |
1348 | */ | |
1349 | void | |
1350 | xfs_freesb( | |
26af6552 | 1351 | struct xfs_mount *mp) |
1da177e4 | 1352 | { |
26af6552 | 1353 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1354 | |
26af6552 | 1355 | xfs_buf_lock(bp); |
1da177e4 | 1356 | mp->m_sb_bp = NULL; |
26af6552 | 1357 | xfs_buf_relse(bp); |
1da177e4 LT |
1358 | } |
1359 | ||
dda35b8f CH |
1360 | /* |
1361 | * If the underlying (data/log/rt) device is readonly, there are some | |
1362 | * operations that cannot proceed. | |
1363 | */ | |
1364 | int | |
1365 | xfs_dev_is_read_only( | |
1366 | struct xfs_mount *mp, | |
1367 | char *message) | |
1368 | { | |
1369 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1370 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1371 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1372 | xfs_notice(mp, "%s required on read-only device.", message); |
1373 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1374 | return -EROFS; |
dda35b8f CH |
1375 | } |
1376 | return 0; | |
1377 | } |