Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
70a9883c | 8 | #include "xfs_shared.h" |
239880ef DC |
9 | #include "xfs_format.h" |
10 | #include "xfs_log_format.h" | |
11 | #include "xfs_trans_resv.h" | |
a844f451 | 12 | #include "xfs_bit.h" |
1da177e4 | 13 | #include "xfs_sb.h" |
1da177e4 | 14 | #include "xfs_mount.h" |
1da177e4 | 15 | #include "xfs_inode.h" |
a4fbe6ab | 16 | #include "xfs_dir2.h" |
a844f451 | 17 | #include "xfs_ialloc.h" |
1da177e4 LT |
18 | #include "xfs_alloc.h" |
19 | #include "xfs_rtalloc.h" | |
20 | #include "xfs_bmap.h" | |
a4fbe6ab DC |
21 | #include "xfs_trans.h" |
22 | #include "xfs_trans_priv.h" | |
23 | #include "xfs_log.h" | |
1da177e4 | 24 | #include "xfs_error.h" |
1da177e4 LT |
25 | #include "xfs_quota.h" |
26 | #include "xfs_fsops.h" | |
6d8b79cf | 27 | #include "xfs_icache.h" |
a31b1d3d | 28 | #include "xfs_sysfs.h" |
035e00ac | 29 | #include "xfs_rmap_btree.h" |
1946b91c | 30 | #include "xfs_refcount_btree.h" |
174edb0e | 31 | #include "xfs_reflink.h" |
ebf55872 | 32 | #include "xfs_extent_busy.h" |
39353ff6 | 33 | #include "xfs_health.h" |
13eaec4b | 34 | #include "xfs_trace.h" |
9bbafc71 | 35 | #include "xfs_ag.h" |
1da177e4 | 36 | |
27174203 CH |
37 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
38 | static int xfs_uuid_table_size; | |
39 | static uuid_t *xfs_uuid_table; | |
40 | ||
af3b6382 DW |
41 | void |
42 | xfs_uuid_table_free(void) | |
43 | { | |
44 | if (xfs_uuid_table_size == 0) | |
45 | return; | |
46 | kmem_free(xfs_uuid_table); | |
47 | xfs_uuid_table = NULL; | |
48 | xfs_uuid_table_size = 0; | |
49 | } | |
50 | ||
27174203 CH |
51 | /* |
52 | * See if the UUID is unique among mounted XFS filesystems. | |
53 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
54 | */ | |
55 | STATIC int | |
56 | xfs_uuid_mount( | |
57 | struct xfs_mount *mp) | |
58 | { | |
59 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
60 | int hole, i; | |
61 | ||
8f720d9f | 62 | /* Publish UUID in struct super_block */ |
85787090 | 63 | uuid_copy(&mp->m_super->s_uuid, uuid); |
8f720d9f | 64 | |
0560f31a | 65 | if (xfs_has_nouuid(mp)) |
27174203 CH |
66 | return 0; |
67 | ||
d905fdaa AG |
68 | if (uuid_is_null(uuid)) { |
69 | xfs_warn(mp, "Filesystem has null UUID - can't mount"); | |
2451337d | 70 | return -EINVAL; |
27174203 CH |
71 | } |
72 | ||
73 | mutex_lock(&xfs_uuid_table_mutex); | |
74 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
d905fdaa | 75 | if (uuid_is_null(&xfs_uuid_table[i])) { |
27174203 CH |
76 | hole = i; |
77 | continue; | |
78 | } | |
79 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
80 | goto out_duplicate; | |
81 | } | |
82 | ||
83 | if (hole < 0) { | |
771915c4 | 84 | xfs_uuid_table = krealloc(xfs_uuid_table, |
27174203 | 85 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), |
771915c4 | 86 | GFP_KERNEL | __GFP_NOFAIL); |
27174203 CH |
87 | hole = xfs_uuid_table_size++; |
88 | } | |
89 | xfs_uuid_table[hole] = *uuid; | |
90 | mutex_unlock(&xfs_uuid_table_mutex); | |
91 | ||
92 | return 0; | |
93 | ||
94 | out_duplicate: | |
95 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 96 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
2451337d | 97 | return -EINVAL; |
27174203 CH |
98 | } |
99 | ||
100 | STATIC void | |
101 | xfs_uuid_unmount( | |
102 | struct xfs_mount *mp) | |
103 | { | |
104 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
105 | int i; | |
106 | ||
0560f31a | 107 | if (xfs_has_nouuid(mp)) |
27174203 CH |
108 | return; |
109 | ||
110 | mutex_lock(&xfs_uuid_table_mutex); | |
111 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
d905fdaa | 112 | if (uuid_is_null(&xfs_uuid_table[i])) |
27174203 CH |
113 | continue; |
114 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
115 | continue; | |
116 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
117 | break; | |
118 | } | |
119 | ASSERT(i < xfs_uuid_table_size); | |
120 | mutex_unlock(&xfs_uuid_table_mutex); | |
121 | } | |
122 | ||
4cc929ee NS |
123 | /* |
124 | * Check size of device based on the (data/realtime) block count. | |
125 | * Note: this check is used by the growfs code as well as mount. | |
126 | */ | |
127 | int | |
128 | xfs_sb_validate_fsb_count( | |
129 | xfs_sb_t *sbp, | |
c8ce540d | 130 | uint64_t nblocks) |
4cc929ee NS |
131 | { |
132 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
133 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
134 | ||
d5cf09ba | 135 | /* Limited by ULONG_MAX of page cache index */ |
09cbfeaf | 136 | if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) |
2451337d | 137 | return -EFBIG; |
4cc929ee NS |
138 | return 0; |
139 | } | |
1da177e4 | 140 | |
1da177e4 LT |
141 | /* |
142 | * xfs_readsb | |
143 | * | |
144 | * Does the initial read of the superblock. | |
145 | */ | |
146 | int | |
ff55068c DC |
147 | xfs_readsb( |
148 | struct xfs_mount *mp, | |
149 | int flags) | |
1da177e4 LT |
150 | { |
151 | unsigned int sector_size; | |
04a1e6c5 DC |
152 | struct xfs_buf *bp; |
153 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 154 | int error; |
af34e09d | 155 | int loud = !(flags & XFS_MFSI_QUIET); |
daba5427 | 156 | const struct xfs_buf_ops *buf_ops; |
1da177e4 LT |
157 | |
158 | ASSERT(mp->m_sb_bp == NULL); | |
159 | ASSERT(mp->m_ddev_targp != NULL); | |
160 | ||
daba5427 ES |
161 | /* |
162 | * For the initial read, we must guess at the sector | |
163 | * size based on the block device. It's enough to | |
164 | * get the sb_sectsize out of the superblock and | |
165 | * then reread with the proper length. | |
166 | * We don't verify it yet, because it may not be complete. | |
167 | */ | |
168 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
169 | buf_ops = NULL; | |
170 | ||
1da177e4 | 171 | /* |
c891c30a BF |
172 | * Allocate a (locked) buffer to hold the superblock. This will be kept |
173 | * around at all times to optimize access to the superblock. Therefore, | |
174 | * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count | |
175 | * elevated. | |
1da177e4 | 176 | */ |
26af6552 | 177 | reread: |
ba372674 | 178 | error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
c891c30a BF |
179 | BTOBB(sector_size), XBF_NO_IOACCT, &bp, |
180 | buf_ops); | |
ba372674 | 181 | if (error) { |
eab4e633 | 182 | if (loud) |
e721f504 | 183 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 184 | /* bad CRC means corrupted metadata */ |
2451337d DC |
185 | if (error == -EFSBADCRC) |
186 | error = -EFSCORRUPTED; | |
ba372674 | 187 | return error; |
eab4e633 | 188 | } |
1da177e4 LT |
189 | |
190 | /* | |
191 | * Initialize the mount structure from the superblock. | |
1da177e4 | 192 | */ |
3e6e8afd | 193 | xfs_sb_from_disk(sbp, bp->b_addr); |
556b8883 DC |
194 | |
195 | /* | |
196 | * If we haven't validated the superblock, do so now before we try | |
197 | * to check the sector size and reread the superblock appropriately. | |
198 | */ | |
199 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
200 | if (loud) | |
201 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 202 | error = -EINVAL; |
556b8883 DC |
203 | goto release_buf; |
204 | } | |
ff55068c | 205 | |
1da177e4 LT |
206 | /* |
207 | * We must be able to do sector-sized and sector-aligned IO. | |
208 | */ | |
04a1e6c5 | 209 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
210 | if (loud) |
211 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 212 | sector_size, sbp->sb_sectsize); |
2451337d | 213 | error = -ENOSYS; |
26af6552 | 214 | goto release_buf; |
1da177e4 LT |
215 | } |
216 | ||
daba5427 | 217 | if (buf_ops == NULL) { |
556b8883 DC |
218 | /* |
219 | * Re-read the superblock so the buffer is correctly sized, | |
220 | * and properly verified. | |
221 | */ | |
1da177e4 | 222 | xfs_buf_relse(bp); |
04a1e6c5 | 223 | sector_size = sbp->sb_sectsize; |
daba5427 | 224 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 225 | goto reread; |
1da177e4 LT |
226 | } |
227 | ||
a1d86e8d | 228 | mp->m_features |= xfs_sb_version_to_features(sbp); |
5681ca40 | 229 | xfs_reinit_percpu_counters(mp); |
8d280b98 | 230 | |
04a1e6c5 DC |
231 | /* no need to be quiet anymore, so reset the buf ops */ |
232 | bp->b_ops = &xfs_sb_buf_ops; | |
233 | ||
1da177e4 | 234 | mp->m_sb_bp = bp; |
26af6552 | 235 | xfs_buf_unlock(bp); |
1da177e4 LT |
236 | return 0; |
237 | ||
26af6552 DC |
238 | release_buf: |
239 | xfs_buf_relse(bp); | |
1da177e4 LT |
240 | return error; |
241 | } | |
242 | ||
13eaec4b DW |
243 | /* |
244 | * If the sunit/swidth change would move the precomputed root inode value, we | |
245 | * must reject the ondisk change because repair will stumble over that. | |
246 | * However, we allow the mount to proceed because we never rejected this | |
247 | * combination before. Returns true to update the sb, false otherwise. | |
248 | */ | |
249 | static inline int | |
250 | xfs_check_new_dalign( | |
251 | struct xfs_mount *mp, | |
252 | int new_dalign, | |
253 | bool *update_sb) | |
254 | { | |
255 | struct xfs_sb *sbp = &mp->m_sb; | |
256 | xfs_ino_t calc_ino; | |
257 | ||
258 | calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign); | |
259 | trace_xfs_check_new_dalign(mp, new_dalign, calc_ino); | |
260 | ||
261 | if (sbp->sb_rootino == calc_ino) { | |
262 | *update_sb = true; | |
263 | return 0; | |
264 | } | |
265 | ||
266 | xfs_warn(mp, | |
267 | "Cannot change stripe alignment; would require moving root inode."); | |
268 | ||
269 | /* | |
270 | * XXX: Next time we add a new incompat feature, this should start | |
271 | * returning -EINVAL to fail the mount. Until then, spit out a warning | |
272 | * that we're ignoring the administrator's instructions. | |
273 | */ | |
274 | xfs_warn(mp, "Skipping superblock stripe alignment update."); | |
275 | *update_sb = false; | |
276 | return 0; | |
277 | } | |
278 | ||
1da177e4 | 279 | /* |
4f5b1b3a DW |
280 | * If we were provided with new sunit/swidth values as mount options, make sure |
281 | * that they pass basic alignment and superblock feature checks, and convert | |
282 | * them into the same units (FSB) that everything else expects. This step | |
283 | * /must/ be done before computing the inode geometry. | |
1da177e4 | 284 | */ |
0771fb45 | 285 | STATIC int |
4f5b1b3a DW |
286 | xfs_validate_new_dalign( |
287 | struct xfs_mount *mp) | |
1da177e4 | 288 | { |
4f5b1b3a DW |
289 | if (mp->m_dalign == 0) |
290 | return 0; | |
1da177e4 | 291 | |
4f5b1b3a DW |
292 | /* |
293 | * If stripe unit and stripe width are not multiples | |
294 | * of the fs blocksize turn off alignment. | |
295 | */ | |
296 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
297 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
298 | xfs_warn(mp, | |
299 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
300 | mp->m_sb.sb_blocksize); | |
301 | return -EINVAL; | |
302 | } else { | |
1da177e4 | 303 | /* |
4f5b1b3a | 304 | * Convert the stripe unit and width to FSBs. |
1da177e4 | 305 | */ |
4f5b1b3a DW |
306 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); |
307 | if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) { | |
39a45d84 | 308 | xfs_warn(mp, |
4f5b1b3a DW |
309 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
310 | mp->m_sb.sb_agblocks); | |
2451337d | 311 | return -EINVAL; |
4f5b1b3a DW |
312 | } else if (mp->m_dalign) { |
313 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
34d7f603 JL |
314 | } else { |
315 | xfs_warn(mp, | |
4f5b1b3a DW |
316 | "alignment check failed: sunit(%d) less than bsize(%d)", |
317 | mp->m_dalign, mp->m_sb.sb_blocksize); | |
2451337d | 318 | return -EINVAL; |
1da177e4 | 319 | } |
4f5b1b3a DW |
320 | } |
321 | ||
38c26bfd | 322 | if (!xfs_has_dalign(mp)) { |
4f5b1b3a DW |
323 | xfs_warn(mp, |
324 | "cannot change alignment: superblock does not support data alignment"); | |
325 | return -EINVAL; | |
326 | } | |
327 | ||
328 | return 0; | |
329 | } | |
330 | ||
331 | /* Update alignment values based on mount options and sb values. */ | |
332 | STATIC int | |
333 | xfs_update_alignment( | |
334 | struct xfs_mount *mp) | |
335 | { | |
336 | struct xfs_sb *sbp = &mp->m_sb; | |
337 | ||
338 | if (mp->m_dalign) { | |
13eaec4b DW |
339 | bool update_sb; |
340 | int error; | |
341 | ||
4f5b1b3a DW |
342 | if (sbp->sb_unit == mp->m_dalign && |
343 | sbp->sb_width == mp->m_swidth) | |
344 | return 0; | |
345 | ||
13eaec4b DW |
346 | error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb); |
347 | if (error || !update_sb) | |
348 | return error; | |
349 | ||
4f5b1b3a DW |
350 | sbp->sb_unit = mp->m_dalign; |
351 | sbp->sb_width = mp->m_swidth; | |
352 | mp->m_update_sb = true; | |
0560f31a | 353 | } else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) { |
4f5b1b3a DW |
354 | mp->m_dalign = sbp->sb_unit; |
355 | mp->m_swidth = sbp->sb_width; | |
1da177e4 LT |
356 | } |
357 | ||
0771fb45 ES |
358 | return 0; |
359 | } | |
1da177e4 | 360 | |
055388a3 DC |
361 | /* |
362 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
363 | */ | |
364 | void | |
365 | xfs_set_low_space_thresholds( | |
366 | struct xfs_mount *mp) | |
367 | { | |
65f03d86 DW |
368 | uint64_t dblocks = mp->m_sb.sb_dblocks; |
369 | uint64_t rtexts = mp->m_sb.sb_rextents; | |
370 | int i; | |
055388a3 | 371 | |
65f03d86 DW |
372 | do_div(dblocks, 100); |
373 | do_div(rtexts, 100); | |
055388a3 | 374 | |
65f03d86 DW |
375 | for (i = 0; i < XFS_LOWSP_MAX; i++) { |
376 | mp->m_low_space[i] = dblocks * (i + 1); | |
377 | mp->m_low_rtexts[i] = rtexts * (i + 1); | |
055388a3 DC |
378 | } |
379 | } | |
380 | ||
0771fb45 | 381 | /* |
0471f62e | 382 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
383 | */ |
384 | STATIC int | |
ba372674 DC |
385 | xfs_check_sizes( |
386 | struct xfs_mount *mp) | |
0771fb45 | 387 | { |
ba372674 | 388 | struct xfs_buf *bp; |
0771fb45 | 389 | xfs_daddr_t d; |
ba372674 | 390 | int error; |
0771fb45 | 391 | |
1da177e4 LT |
392 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
393 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 394 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 395 | return -EFBIG; |
1da177e4 | 396 | } |
ba372674 | 397 | error = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 398 | d - XFS_FSS_TO_BB(mp, 1), |
ba372674 DC |
399 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
400 | if (error) { | |
0b932ccc | 401 | xfs_warn(mp, "last sector read failed"); |
ba372674 | 402 | return error; |
1da177e4 | 403 | } |
1922c949 | 404 | xfs_buf_relse(bp); |
1da177e4 | 405 | |
ba372674 DC |
406 | if (mp->m_logdev_targp == mp->m_ddev_targp) |
407 | return 0; | |
408 | ||
409 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
410 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
411 | xfs_warn(mp, "log size mismatch detected"); | |
412 | return -EFBIG; | |
413 | } | |
414 | error = xfs_buf_read_uncached(mp->m_logdev_targp, | |
1922c949 | 415 | d - XFS_FSB_TO_BB(mp, 1), |
ba372674 DC |
416 | XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); |
417 | if (error) { | |
418 | xfs_warn(mp, "log device read failed"); | |
419 | return error; | |
0771fb45 | 420 | } |
ba372674 | 421 | xfs_buf_relse(bp); |
0771fb45 ES |
422 | return 0; |
423 | } | |
424 | ||
7d095257 CH |
425 | /* |
426 | * Clear the quotaflags in memory and in the superblock. | |
427 | */ | |
428 | int | |
429 | xfs_mount_reset_sbqflags( | |
430 | struct xfs_mount *mp) | |
431 | { | |
7d095257 CH |
432 | mp->m_qflags = 0; |
433 | ||
61e63ecb | 434 | /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ |
7d095257 CH |
435 | if (mp->m_sb.sb_qflags == 0) |
436 | return 0; | |
437 | spin_lock(&mp->m_sb_lock); | |
438 | mp->m_sb.sb_qflags = 0; | |
439 | spin_unlock(&mp->m_sb_lock); | |
440 | ||
61e63ecb | 441 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
7d095257 CH |
442 | return 0; |
443 | ||
61e63ecb | 444 | return xfs_sync_sb(mp, false); |
7d095257 CH |
445 | } |
446 | ||
c8ce540d | 447 | uint64_t |
d5db0f97 ES |
448 | xfs_default_resblks(xfs_mount_t *mp) |
449 | { | |
c8ce540d | 450 | uint64_t resblks; |
d5db0f97 ES |
451 | |
452 | /* | |
8babd8a2 DC |
453 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
454 | * smaller. This is intended to cover concurrent allocation | |
455 | * transactions when we initially hit enospc. These each require a 4 | |
456 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
457 | * allocation reservations. | |
d5db0f97 ES |
458 | */ |
459 | resblks = mp->m_sb.sb_dblocks; | |
460 | do_div(resblks, 20); | |
c8ce540d | 461 | resblks = min_t(uint64_t, resblks, 8192); |
d5db0f97 ES |
462 | return resblks; |
463 | } | |
464 | ||
2e9e6481 DW |
465 | /* Ensure the summary counts are correct. */ |
466 | STATIC int | |
467 | xfs_check_summary_counts( | |
468 | struct xfs_mount *mp) | |
469 | { | |
470 | /* | |
471 | * The AG0 superblock verifier rejects in-progress filesystems, | |
472 | * so we should never see the flag set this far into mounting. | |
473 | */ | |
474 | if (mp->m_sb.sb_inprogress) { | |
475 | xfs_err(mp, "sb_inprogress set after log recovery??"); | |
476 | WARN_ON(1); | |
477 | return -EFSCORRUPTED; | |
478 | } | |
479 | ||
480 | /* | |
481 | * Now the log is mounted, we know if it was an unclean shutdown or | |
482 | * not. If it was, with the first phase of recovery has completed, we | |
483 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
484 | * but they are recovered transactionally in the second recovery phase | |
485 | * later. | |
486 | * | |
487 | * If the log was clean when we mounted, we can check the summary | |
488 | * counters. If any of them are obviously incorrect, we can recompute | |
489 | * them from the AGF headers in the next step. | |
490 | */ | |
2e973b2c | 491 | if (xfs_is_clean(mp) && |
2e9e6481 | 492 | (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks || |
00d22a1c | 493 | !xfs_verify_icount(mp, mp->m_sb.sb_icount) || |
2e9e6481 | 494 | mp->m_sb.sb_ifree > mp->m_sb.sb_icount)) |
39353ff6 | 495 | xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); |
2e9e6481 DW |
496 | |
497 | /* | |
498 | * We can safely re-initialise incore superblock counters from the | |
499 | * per-ag data. These may not be correct if the filesystem was not | |
500 | * cleanly unmounted, so we waited for recovery to finish before doing | |
501 | * this. | |
502 | * | |
503 | * If the filesystem was cleanly unmounted or the previous check did | |
504 | * not flag anything weird, then we can trust the values in the | |
505 | * superblock to be correct and we don't need to do anything here. | |
506 | * Otherwise, recalculate the summary counters. | |
507 | */ | |
2e973b2c | 508 | if ((!xfs_has_lazysbcount(mp) || xfs_is_clean(mp)) && |
39353ff6 | 509 | !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) |
2e9e6481 DW |
510 | return 0; |
511 | ||
512 | return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount); | |
513 | } | |
514 | ||
d336f7eb DW |
515 | /* |
516 | * Flush and reclaim dirty inodes in preparation for unmount. Inodes and | |
517 | * internal inode structures can be sitting in the CIL and AIL at this point, | |
518 | * so we need to unpin them, write them back and/or reclaim them before unmount | |
ab23a776 DC |
519 | * can proceed. In other words, callers are required to have inactivated all |
520 | * inodes. | |
d336f7eb DW |
521 | * |
522 | * An inode cluster that has been freed can have its buffer still pinned in | |
523 | * memory because the transaction is still sitting in a iclog. The stale inodes | |
524 | * on that buffer will be pinned to the buffer until the transaction hits the | |
525 | * disk and the callbacks run. Pushing the AIL will skip the stale inodes and | |
526 | * may never see the pinned buffer, so nothing will push out the iclog and | |
527 | * unpin the buffer. | |
528 | * | |
529 | * Hence we need to force the log to unpin everything first. However, log | |
530 | * forces don't wait for the discards they issue to complete, so we have to | |
531 | * explicitly wait for them to complete here as well. | |
532 | * | |
533 | * Then we can tell the world we are unmounting so that error handling knows | |
534 | * that the filesystem is going away and we should error out anything that we | |
535 | * have been retrying in the background. This will prevent never-ending | |
536 | * retries in AIL pushing from hanging the unmount. | |
537 | * | |
538 | * Finally, we can push the AIL to clean all the remaining dirty objects, then | |
539 | * reclaim the remaining inodes that are still in memory at this point in time. | |
540 | */ | |
541 | static void | |
542 | xfs_unmount_flush_inodes( | |
543 | struct xfs_mount *mp) | |
544 | { | |
545 | xfs_log_force(mp, XFS_LOG_SYNC); | |
546 | xfs_extent_busy_wait_all(mp); | |
547 | flush_workqueue(xfs_discard_wq); | |
548 | ||
2e973b2c | 549 | set_bit(XFS_OPSTATE_UNMOUNTING, &mp->m_opstate); |
d336f7eb DW |
550 | |
551 | xfs_ail_push_all_sync(mp->m_ail); | |
ab23a776 | 552 | xfs_inodegc_stop(mp); |
d336f7eb DW |
553 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
554 | xfs_reclaim_inodes(mp); | |
555 | xfs_health_unmount(mp); | |
556 | } | |
557 | ||
b2941046 DC |
558 | static void |
559 | xfs_mount_setup_inode_geom( | |
560 | struct xfs_mount *mp) | |
561 | { | |
562 | struct xfs_ino_geometry *igeo = M_IGEO(mp); | |
563 | ||
564 | igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp); | |
565 | ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp)); | |
566 | ||
567 | xfs_ialloc_setup_geometry(mp); | |
568 | } | |
569 | ||
0771fb45 | 570 | /* |
0771fb45 ES |
571 | * This function does the following on an initial mount of a file system: |
572 | * - reads the superblock from disk and init the mount struct | |
573 | * - if we're a 32-bit kernel, do a size check on the superblock | |
574 | * so we don't mount terabyte filesystems | |
575 | * - init mount struct realtime fields | |
576 | * - allocate inode hash table for fs | |
577 | * - init directory manager | |
578 | * - perform recovery and init the log manager | |
579 | */ | |
580 | int | |
581 | xfs_mountfs( | |
f0b2efad | 582 | struct xfs_mount *mp) |
0771fb45 | 583 | { |
f0b2efad BF |
584 | struct xfs_sb *sbp = &(mp->m_sb); |
585 | struct xfs_inode *rip; | |
ef325959 | 586 | struct xfs_ino_geometry *igeo = M_IGEO(mp); |
c8ce540d | 587 | uint64_t resblks; |
f0b2efad BF |
588 | uint quotamount = 0; |
589 | uint quotaflags = 0; | |
590 | int error = 0; | |
0771fb45 | 591 | |
ff55068c | 592 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 593 | |
ee1c0908 | 594 | /* |
074e427b DC |
595 | * Check for a mismatched features2 values. Older kernels read & wrote |
596 | * into the wrong sb offset for sb_features2 on some platforms due to | |
597 | * xfs_sb_t not being 64bit size aligned when sb_features2 was added, | |
598 | * which made older superblock reading/writing routines swap it as a | |
599 | * 64-bit value. | |
ee1c0908 | 600 | * |
e6957ea4 ES |
601 | * For backwards compatibility, we make both slots equal. |
602 | * | |
074e427b DC |
603 | * If we detect a mismatched field, we OR the set bits into the existing |
604 | * features2 field in case it has already been modified; we don't want | |
605 | * to lose any features. We then update the bad location with the ORed | |
606 | * value so that older kernels will see any features2 flags. The | |
607 | * superblock writeback code ensures the new sb_features2 is copied to | |
608 | * sb_bad_features2 before it is logged or written to disk. | |
ee1c0908 | 609 | */ |
e6957ea4 | 610 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 611 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 612 | sbp->sb_features2 |= sbp->sb_bad_features2; |
61e63ecb | 613 | mp->m_update_sb = true; |
7c12f296 TS |
614 | } |
615 | ||
ee1c0908 | 616 | |
263997a6 DC |
617 | /* always use v2 inodes by default now */ |
618 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
619 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
38c26bfd | 620 | mp->m_features |= XFS_FEAT_NLINK; |
61e63ecb | 621 | mp->m_update_sb = true; |
263997a6 DC |
622 | } |
623 | ||
0771fb45 | 624 | /* |
4f5b1b3a DW |
625 | * If we were given new sunit/swidth options, do some basic validation |
626 | * checks and convert the incore dalign and swidth values to the | |
627 | * same units (FSB) that everything else uses. This /must/ happen | |
628 | * before computing the inode geometry. | |
0771fb45 | 629 | */ |
4f5b1b3a | 630 | error = xfs_validate_new_dalign(mp); |
0771fb45 | 631 | if (error) |
f9057e3d | 632 | goto out; |
0771fb45 ES |
633 | |
634 | xfs_alloc_compute_maxlevels(mp); | |
635 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
636 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
b2941046 | 637 | xfs_mount_setup_inode_geom(mp); |
035e00ac | 638 | xfs_rmapbt_compute_maxlevels(mp); |
1946b91c | 639 | xfs_refcountbt_compute_maxlevels(mp); |
0771fb45 | 640 | |
4f5b1b3a DW |
641 | /* |
642 | * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks | |
643 | * is NOT aligned turn off m_dalign since allocator alignment is within | |
644 | * an ag, therefore ag has to be aligned at stripe boundary. Note that | |
645 | * we must compute the free space and rmap btree geometry before doing | |
646 | * this. | |
647 | */ | |
648 | error = xfs_update_alignment(mp); | |
649 | if (error) | |
650 | goto out; | |
651 | ||
e6b3bb78 | 652 | /* enable fail_at_unmount as default */ |
749f24f3 | 653 | mp->m_fail_unmount = true; |
e6b3bb78 | 654 | |
e1d3d218 IK |
655 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, |
656 | NULL, mp->m_super->s_id); | |
27174203 CH |
657 | if (error) |
658 | goto out; | |
1da177e4 | 659 | |
225e4635 BD |
660 | error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, |
661 | &mp->m_kobj, "stats"); | |
a31b1d3d BF |
662 | if (error) |
663 | goto out_remove_sysfs; | |
664 | ||
192852be | 665 | error = xfs_error_sysfs_init(mp); |
225e4635 BD |
666 | if (error) |
667 | goto out_del_stats; | |
668 | ||
31965ef3 DW |
669 | error = xfs_errortag_init(mp); |
670 | if (error) | |
671 | goto out_remove_error_sysfs; | |
192852be CM |
672 | |
673 | error = xfs_uuid_mount(mp); | |
674 | if (error) | |
31965ef3 | 675 | goto out_remove_errortag; |
192852be | 676 | |
0771fb45 | 677 | /* |
2fcddee8 CH |
678 | * Update the preferred write size based on the information from the |
679 | * on-disk superblock. | |
0771fb45 | 680 | */ |
2fcddee8 CH |
681 | mp->m_allocsize_log = |
682 | max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log); | |
683 | mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog); | |
0771fb45 | 684 | |
055388a3 DC |
685 | /* set the low space thresholds for dynamic preallocation */ |
686 | xfs_set_low_space_thresholds(mp); | |
687 | ||
e5376fc1 BF |
688 | /* |
689 | * If enabled, sparse inode chunk alignment is expected to match the | |
690 | * cluster size. Full inode chunk alignment must match the chunk size, | |
691 | * but that is checked on sb read verification... | |
692 | */ | |
38c26bfd | 693 | if (xfs_has_sparseinodes(mp) && |
e5376fc1 | 694 | mp->m_sb.sb_spino_align != |
490d451f | 695 | XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) { |
e5376fc1 BF |
696 | xfs_warn(mp, |
697 | "Sparse inode block alignment (%u) must match cluster size (%llu).", | |
698 | mp->m_sb.sb_spino_align, | |
490d451f | 699 | XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)); |
e5376fc1 BF |
700 | error = -EINVAL; |
701 | goto out_remove_uuid; | |
702 | } | |
703 | ||
0771fb45 | 704 | /* |
c2bfbc9b | 705 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 706 | */ |
4249023a | 707 | error = xfs_check_sizes(mp); |
0771fb45 | 708 | if (error) |
f9057e3d | 709 | goto out_remove_uuid; |
0771fb45 | 710 | |
1da177e4 LT |
711 | /* |
712 | * Initialize realtime fields in the mount structure | |
713 | */ | |
0771fb45 ES |
714 | error = xfs_rtmount_init(mp); |
715 | if (error) { | |
0b932ccc | 716 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 717 | goto out_remove_uuid; |
1da177e4 LT |
718 | } |
719 | ||
1da177e4 LT |
720 | /* |
721 | * Copies the low order bits of the timestamp and the randomly | |
722 | * set "sequence" number out of a UUID. | |
723 | */ | |
cb0ba6cc CH |
724 | mp->m_fixedfsid[0] = |
725 | (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) | | |
726 | get_unaligned_be16(&sbp->sb_uuid.b[4]); | |
727 | mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]); | |
1da177e4 | 728 | |
0650b554 DC |
729 | error = xfs_da_mount(mp); |
730 | if (error) { | |
731 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
732 | goto out_remove_uuid; | |
733 | } | |
1da177e4 LT |
734 | |
735 | /* | |
736 | * Initialize the precomputed transaction reservations values. | |
737 | */ | |
738 | xfs_trans_init(mp); | |
739 | ||
1da177e4 LT |
740 | /* |
741 | * Allocate and initialize the per-ag data. | |
742 | */ | |
1c1c6ebc DC |
743 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
744 | if (error) { | |
0b932ccc | 745 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 746 | goto out_free_dir; |
1c1c6ebc | 747 | } |
1da177e4 | 748 | |
a71895c5 | 749 | if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) { |
0b932ccc | 750 | xfs_warn(mp, "no log defined"); |
2451337d | 751 | error = -EFSCORRUPTED; |
f9057e3d CH |
752 | goto out_free_perag; |
753 | } | |
754 | ||
40b1de00 DW |
755 | error = xfs_inodegc_register_shrinker(mp); |
756 | if (error) | |
757 | goto out_fail_wait; | |
758 | ||
1da177e4 | 759 | /* |
f0b2efad BF |
760 | * Log's mount-time initialization. The first part of recovery can place |
761 | * some items on the AIL, to be handled when recovery is finished or | |
762 | * cancelled. | |
1da177e4 | 763 | */ |
f9057e3d CH |
764 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
765 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
766 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
767 | if (error) { | |
0b932ccc | 768 | xfs_warn(mp, "log mount failed"); |
40b1de00 | 769 | goto out_inodegc_shrinker; |
1da177e4 LT |
770 | } |
771 | ||
2e9e6481 DW |
772 | /* Make sure the summary counts are ok. */ |
773 | error = xfs_check_summary_counts(mp); | |
774 | if (error) | |
775 | goto out_log_dealloc; | |
f9057e3d | 776 | |
ab23a776 DC |
777 | /* Enable background inode inactivation workers. */ |
778 | xfs_inodegc_start(mp); | |
6f649091 | 779 | xfs_blockgc_start(mp); |
ab23a776 | 780 | |
e23b55d5 DC |
781 | /* |
782 | * Now that we've recovered any pending superblock feature bit | |
783 | * additions, we can finish setting up the attr2 behaviour for the | |
0560f31a DC |
784 | * mount. The noattr2 option overrides the superblock flag, so only |
785 | * check the superblock feature flag if the mount option is not set. | |
e23b55d5 | 786 | */ |
0560f31a DC |
787 | if (xfs_has_noattr2(mp)) { |
788 | mp->m_features &= ~XFS_FEAT_ATTR2; | |
789 | } else if (!xfs_has_attr2(mp) && | |
790 | (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)) { | |
791 | mp->m_features |= XFS_FEAT_ATTR2; | |
792 | } | |
e23b55d5 | 793 | |
1da177e4 LT |
794 | /* |
795 | * Get and sanity-check the root inode. | |
796 | * Save the pointer to it in the mount structure. | |
797 | */ | |
541b5acc DC |
798 | error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED, |
799 | XFS_ILOCK_EXCL, &rip); | |
1da177e4 | 800 | if (error) { |
541b5acc DC |
801 | xfs_warn(mp, |
802 | "Failed to read root inode 0x%llx, error %d", | |
803 | sbp->sb_rootino, -error); | |
f9057e3d | 804 | goto out_log_dealloc; |
1da177e4 LT |
805 | } |
806 | ||
807 | ASSERT(rip != NULL); | |
1da177e4 | 808 | |
a71895c5 | 809 | if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) { |
0b932ccc | 810 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 811 | (unsigned long long)rip->i_ino); |
1da177e4 | 812 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
2451337d | 813 | error = -EFSCORRUPTED; |
f9057e3d | 814 | goto out_rele_rip; |
1da177e4 LT |
815 | } |
816 | mp->m_rootip = rip; /* save it */ | |
817 | ||
818 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
819 | ||
820 | /* | |
821 | * Initialize realtime inode pointers in the mount structure | |
822 | */ | |
0771fb45 ES |
823 | error = xfs_rtmount_inodes(mp); |
824 | if (error) { | |
1da177e4 LT |
825 | /* |
826 | * Free up the root inode. | |
827 | */ | |
0b932ccc | 828 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 829 | goto out_rele_rip; |
1da177e4 LT |
830 | } |
831 | ||
832 | /* | |
7884bc86 CH |
833 | * If this is a read-only mount defer the superblock updates until |
834 | * the next remount into writeable mode. Otherwise we would never | |
835 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 836 | */ |
2e973b2c | 837 | if (mp->m_update_sb && !xfs_is_readonly(mp)) { |
61e63ecb | 838 | error = xfs_sync_sb(mp, false); |
e5720eec | 839 | if (error) { |
0b932ccc | 840 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 841 | goto out_rtunmount; |
e5720eec DC |
842 | } |
843 | } | |
1da177e4 LT |
844 | |
845 | /* | |
846 | * Initialise the XFS quota management subsystem for this mount | |
847 | */ | |
149e53af | 848 | if (XFS_IS_QUOTA_ON(mp)) { |
7d095257 CH |
849 | error = xfs_qm_newmount(mp, "amount, "aflags); |
850 | if (error) | |
851 | goto out_rtunmount; | |
852 | } else { | |
7d095257 CH |
853 | /* |
854 | * If a file system had quotas running earlier, but decided to | |
855 | * mount without -o uquota/pquota/gquota options, revoke the | |
856 | * quotachecked license. | |
857 | */ | |
858 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 859 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
860 | error = xfs_mount_reset_sbqflags(mp); |
861 | if (error) | |
a70a4fa5 | 862 | goto out_rtunmount; |
7d095257 CH |
863 | } |
864 | } | |
1da177e4 LT |
865 | |
866 | /* | |
f0b2efad BF |
867 | * Finish recovering the file system. This part needed to be delayed |
868 | * until after the root and real-time bitmap inodes were consistently | |
81ed9475 DW |
869 | * read in. Temporarily create per-AG space reservations for metadata |
870 | * btree shape changes because space freeing transactions (for inode | |
871 | * inactivation) require the per-AG reservation in lieu of reserving | |
872 | * blocks. | |
1da177e4 | 873 | */ |
81ed9475 DW |
874 | error = xfs_fs_reserve_ag_blocks(mp); |
875 | if (error && error == -ENOSPC) | |
876 | xfs_warn(mp, | |
877 | "ENOSPC reserving per-AG metadata pool, log recovery may fail."); | |
4249023a | 878 | error = xfs_log_mount_finish(mp); |
81ed9475 | 879 | xfs_fs_unreserve_ag_blocks(mp); |
1da177e4 | 880 | if (error) { |
0b932ccc | 881 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 882 | goto out_rtunmount; |
1da177e4 LT |
883 | } |
884 | ||
ddeb14f4 DC |
885 | /* |
886 | * Now the log is fully replayed, we can transition to full read-only | |
887 | * mode for read-only mounts. This will sync all the metadata and clean | |
888 | * the log so that the recovery we just performed does not have to be | |
889 | * replayed again on the next mount. | |
890 | * | |
891 | * We use the same quiesce mechanism as the rw->ro remount, as they are | |
892 | * semantically identical operations. | |
893 | */ | |
2e973b2c | 894 | if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp)) |
ea2064da | 895 | xfs_log_clean(mp); |
ddeb14f4 | 896 | |
1da177e4 LT |
897 | /* |
898 | * Complete the quota initialisation, post-log-replay component. | |
899 | */ | |
7d095257 CH |
900 | if (quotamount) { |
901 | ASSERT(mp->m_qflags == 0); | |
902 | mp->m_qflags = quotaflags; | |
903 | ||
904 | xfs_qm_mount_quotas(mp); | |
905 | } | |
906 | ||
84e1e99f DC |
907 | /* |
908 | * Now we are mounted, reserve a small amount of unused space for | |
909 | * privileged transactions. This is needed so that transaction | |
910 | * space required for critical operations can dip into this pool | |
911 | * when at ENOSPC. This is needed for operations like create with | |
912 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
913 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
914 | * |
915 | * This may drive us straight to ENOSPC on mount, but that implies | |
916 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 917 | */ |
2e973b2c | 918 | if (!xfs_is_readonly(mp)) { |
d5db0f97 ES |
919 | resblks = xfs_default_resblks(mp); |
920 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
921 | if (error) | |
0b932ccc DC |
922 | xfs_warn(mp, |
923 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
174edb0e DW |
924 | |
925 | /* Recover any CoW blocks that never got remapped. */ | |
926 | error = xfs_reflink_recover_cow(mp); | |
927 | if (error) { | |
928 | xfs_err(mp, | |
929 | "Error %d recovering leftover CoW allocations.", error); | |
930 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
931 | goto out_quota; | |
932 | } | |
84d69619 DW |
933 | |
934 | /* Reserve AG blocks for future btree expansion. */ | |
935 | error = xfs_fs_reserve_ag_blocks(mp); | |
936 | if (error && error != -ENOSPC) | |
937 | goto out_agresv; | |
d5db0f97 | 938 | } |
84e1e99f | 939 | |
1da177e4 LT |
940 | return 0; |
941 | ||
84d69619 DW |
942 | out_agresv: |
943 | xfs_fs_unreserve_ag_blocks(mp); | |
174edb0e DW |
944 | out_quota: |
945 | xfs_qm_unmount_quotas(mp); | |
b93b6e43 CH |
946 | out_rtunmount: |
947 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 948 | out_rele_rip: |
44a8736b | 949 | xfs_irele(rip); |
77aff8c7 DW |
950 | /* Clean out dquots that might be in memory after quotacheck. */ |
951 | xfs_qm_unmount(mp); | |
ab23a776 DC |
952 | |
953 | /* | |
954 | * Inactivate all inodes that might still be in memory after a log | |
955 | * intent recovery failure so that reclaim can free them. Metadata | |
956 | * inodes and the root directory shouldn't need inactivation, but the | |
957 | * mount failed for some reason, so pull down all the state and flee. | |
958 | */ | |
959 | xfs_inodegc_flush(mp); | |
960 | ||
2d1d1da3 | 961 | /* |
d336f7eb | 962 | * Flush all inode reclamation work and flush the log. |
2d1d1da3 DW |
963 | * We have to do this /after/ rtunmount and qm_unmount because those |
964 | * two will have scheduled delayed reclaim for the rt/quota inodes. | |
965 | * | |
966 | * This is slightly different from the unmountfs call sequence | |
967 | * because we could be tearing down a partially set up mount. In | |
968 | * particular, if log_mount_finish fails we bail out without calling | |
969 | * qm_unmount_quotas and therefore rely on qm_unmount to release the | |
970 | * quota inodes. | |
971 | */ | |
d336f7eb | 972 | xfs_unmount_flush_inodes(mp); |
f9057e3d | 973 | out_log_dealloc: |
f0b2efad | 974 | xfs_log_mount_cancel(mp); |
40b1de00 DW |
975 | out_inodegc_shrinker: |
976 | unregister_shrinker(&mp->m_inodegc_shrinker); | |
d4f3512b DC |
977 | out_fail_wait: |
978 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
10fb9ac1 BF |
979 | xfs_buftarg_drain(mp->m_logdev_targp); |
980 | xfs_buftarg_drain(mp->m_ddev_targp); | |
f9057e3d | 981 | out_free_perag: |
ff4f038c | 982 | xfs_free_perag(mp); |
0650b554 DC |
983 | out_free_dir: |
984 | xfs_da_unmount(mp); | |
f9057e3d | 985 | out_remove_uuid: |
27174203 | 986 | xfs_uuid_unmount(mp); |
31965ef3 DW |
987 | out_remove_errortag: |
988 | xfs_errortag_del(mp); | |
192852be CM |
989 | out_remove_error_sysfs: |
990 | xfs_error_sysfs_del(mp); | |
225e4635 BD |
991 | out_del_stats: |
992 | xfs_sysfs_del(&mp->m_stats.xs_kobj); | |
a31b1d3d BF |
993 | out_remove_sysfs: |
994 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 995 | out: |
1da177e4 LT |
996 | return error; |
997 | } | |
998 | ||
999 | /* | |
1da177e4 LT |
1000 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1001 | * log and makes sure that incore structures are freed. | |
1002 | */ | |
41b5c2e7 CH |
1003 | void |
1004 | xfs_unmountfs( | |
1005 | struct xfs_mount *mp) | |
1da177e4 | 1006 | { |
c8ce540d | 1007 | uint64_t resblks; |
41b5c2e7 | 1008 | int error; |
1da177e4 | 1009 | |
ab23a776 DC |
1010 | /* |
1011 | * Perform all on-disk metadata updates required to inactivate inodes | |
1012 | * that the VFS evicted earlier in the unmount process. Freeing inodes | |
1013 | * and discarding CoW fork preallocations can cause shape changes to | |
1014 | * the free inode and refcount btrees, respectively, so we must finish | |
1015 | * this before we discard the metadata space reservations. Metadata | |
1016 | * inodes and the root directory do not require inactivation. | |
1017 | */ | |
1018 | xfs_inodegc_flush(mp); | |
1019 | ||
c9a6526f | 1020 | xfs_blockgc_stop(mp); |
84d69619 | 1021 | xfs_fs_unreserve_ag_blocks(mp); |
7d095257 | 1022 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1023 | xfs_rtunmount_inodes(mp); |
44a8736b | 1024 | xfs_irele(mp->m_rootip); |
77508ec8 | 1025 | |
d336f7eb | 1026 | xfs_unmount_flush_inodes(mp); |
1da177e4 | 1027 | |
7d095257 | 1028 | xfs_qm_unmount(mp); |
a357a121 | 1029 | |
84e1e99f DC |
1030 | /* |
1031 | * Unreserve any blocks we have so that when we unmount we don't account | |
1032 | * the reserved free space as used. This is really only necessary for | |
1033 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1034 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1035 | * |
1036 | * We don't bother correcting this elsewhere for lazy superblock | |
1037 | * counting because on mount of an unclean filesystem we reconstruct the | |
1038 | * correct counter value and this is irrelevant. | |
1039 | * | |
1040 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1041 | * we only every apply deltas to the superblock and hence the incore | |
1042 | * value does not matter.... | |
1043 | */ | |
1044 | resblks = 0; | |
714082bc DC |
1045 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1046 | if (error) | |
0b932ccc | 1047 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1048 | "Freespace may not be correct on next mount."); |
1049 | ||
21b699c8 | 1050 | xfs_log_unmount(mp); |
0650b554 | 1051 | xfs_da_unmount(mp); |
27174203 | 1052 | xfs_uuid_unmount(mp); |
1da177e4 | 1053 | |
1550d0b0 | 1054 | #if defined(DEBUG) |
31965ef3 | 1055 | xfs_errortag_clearall(mp); |
1da177e4 | 1056 | #endif |
40b1de00 | 1057 | unregister_shrinker(&mp->m_inodegc_shrinker); |
ff4f038c | 1058 | xfs_free_perag(mp); |
a31b1d3d | 1059 | |
31965ef3 | 1060 | xfs_errortag_del(mp); |
192852be | 1061 | xfs_error_sysfs_del(mp); |
225e4635 | 1062 | xfs_sysfs_del(&mp->m_stats.xs_kobj); |
a31b1d3d | 1063 | xfs_sysfs_del(&mp->m_kobj); |
1da177e4 LT |
1064 | } |
1065 | ||
91ee575f BF |
1066 | /* |
1067 | * Determine whether modifications can proceed. The caller specifies the minimum | |
1068 | * freeze level for which modifications should not be allowed. This allows | |
1069 | * certain operations to proceed while the freeze sequence is in progress, if | |
1070 | * necessary. | |
1071 | */ | |
1072 | bool | |
1073 | xfs_fs_writable( | |
1074 | struct xfs_mount *mp, | |
1075 | int level) | |
92821e2b | 1076 | { |
91ee575f BF |
1077 | ASSERT(level > SB_UNFROZEN); |
1078 | if ((mp->m_super->s_writers.frozen >= level) || | |
2e973b2c | 1079 | XFS_FORCED_SHUTDOWN(mp) || xfs_is_readonly(mp)) |
91ee575f BF |
1080 | return false; |
1081 | ||
1082 | return true; | |
92821e2b DC |
1083 | } |
1084 | ||
0d485ada DC |
1085 | int |
1086 | xfs_mod_fdblocks( | |
1087 | struct xfs_mount *mp, | |
1088 | int64_t delta, | |
1089 | bool rsvd) | |
1090 | { | |
1091 | int64_t lcounter; | |
1092 | long long res_used; | |
1093 | s32 batch; | |
fd43cf60 | 1094 | uint64_t set_aside; |
0d485ada DC |
1095 | |
1096 | if (delta > 0) { | |
1097 | /* | |
1098 | * If the reserve pool is depleted, put blocks back into it | |
1099 | * first. Most of the time the pool is full. | |
1100 | */ | |
1101 | if (likely(mp->m_resblks == mp->m_resblks_avail)) { | |
1102 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1103 | return 0; | |
1104 | } | |
1105 | ||
1106 | spin_lock(&mp->m_sb_lock); | |
1107 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); | |
1108 | ||
1109 | if (res_used > delta) { | |
1110 | mp->m_resblks_avail += delta; | |
1111 | } else { | |
1112 | delta -= res_used; | |
1113 | mp->m_resblks_avail = mp->m_resblks; | |
1114 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1115 | } | |
1116 | spin_unlock(&mp->m_sb_lock); | |
1117 | return 0; | |
1118 | } | |
1119 | ||
1120 | /* | |
1121 | * Taking blocks away, need to be more accurate the closer we | |
1122 | * are to zero. | |
1123 | * | |
0d485ada DC |
1124 | * If the counter has a value of less than 2 * max batch size, |
1125 | * then make everything serialise as we are real close to | |
1126 | * ENOSPC. | |
1127 | */ | |
8c1903d3 DC |
1128 | if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, |
1129 | XFS_FDBLOCKS_BATCH) < 0) | |
0d485ada DC |
1130 | batch = 1; |
1131 | else | |
8c1903d3 | 1132 | batch = XFS_FDBLOCKS_BATCH; |
0d485ada | 1133 | |
fd43cf60 BF |
1134 | /* |
1135 | * Set aside allocbt blocks because these blocks are tracked as free | |
1136 | * space but not available for allocation. Technically this means that a | |
1137 | * single reservation cannot consume all remaining free space, but the | |
1138 | * ratio of allocbt blocks to usable free blocks should be rather small. | |
1139 | * The tradeoff without this is that filesystems that maintain high | |
1140 | * perag block reservations can over reserve physical block availability | |
1141 | * and fail physical allocation, which leads to much more serious | |
1142 | * problems (i.e. transaction abort, pagecache discards, etc.) than | |
1143 | * slightly premature -ENOSPC. | |
1144 | */ | |
1145 | set_aside = mp->m_alloc_set_aside + atomic64_read(&mp->m_allocbt_blks); | |
104b4e51 | 1146 | percpu_counter_add_batch(&mp->m_fdblocks, delta, batch); |
fd43cf60 | 1147 | if (__percpu_counter_compare(&mp->m_fdblocks, set_aside, |
8c1903d3 | 1148 | XFS_FDBLOCKS_BATCH) >= 0) { |
0d485ada DC |
1149 | /* we had space! */ |
1150 | return 0; | |
1151 | } | |
1152 | ||
1153 | /* | |
1154 | * lock up the sb for dipping into reserves before releasing the space | |
1155 | * that took us to ENOSPC. | |
1156 | */ | |
1157 | spin_lock(&mp->m_sb_lock); | |
1158 | percpu_counter_add(&mp->m_fdblocks, -delta); | |
1159 | if (!rsvd) | |
1160 | goto fdblocks_enospc; | |
1161 | ||
1162 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1163 | if (lcounter >= 0) { | |
1164 | mp->m_resblks_avail = lcounter; | |
1165 | spin_unlock(&mp->m_sb_lock); | |
1166 | return 0; | |
1167 | } | |
ec43f6da ES |
1168 | xfs_warn_once(mp, |
1169 | "Reserve blocks depleted! Consider increasing reserve pool size."); | |
1170 | ||
0d485ada DC |
1171 | fdblocks_enospc: |
1172 | spin_unlock(&mp->m_sb_lock); | |
1173 | return -ENOSPC; | |
1174 | } | |
1175 | ||
bab98bbe DC |
1176 | int |
1177 | xfs_mod_frextents( | |
1178 | struct xfs_mount *mp, | |
1179 | int64_t delta) | |
1180 | { | |
1181 | int64_t lcounter; | |
1182 | int ret = 0; | |
1183 | ||
1184 | spin_lock(&mp->m_sb_lock); | |
1185 | lcounter = mp->m_sb.sb_frextents + delta; | |
1186 | if (lcounter < 0) | |
1187 | ret = -ENOSPC; | |
1188 | else | |
1189 | mp->m_sb.sb_frextents = lcounter; | |
1190 | spin_unlock(&mp->m_sb_lock); | |
1191 | return ret; | |
1192 | } | |
1193 | ||
1da177e4 LT |
1194 | /* |
1195 | * Used to free the superblock along various error paths. | |
1196 | */ | |
1197 | void | |
1198 | xfs_freesb( | |
26af6552 | 1199 | struct xfs_mount *mp) |
1da177e4 | 1200 | { |
26af6552 | 1201 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1202 | |
26af6552 | 1203 | xfs_buf_lock(bp); |
1da177e4 | 1204 | mp->m_sb_bp = NULL; |
26af6552 | 1205 | xfs_buf_relse(bp); |
1da177e4 LT |
1206 | } |
1207 | ||
dda35b8f CH |
1208 | /* |
1209 | * If the underlying (data/log/rt) device is readonly, there are some | |
1210 | * operations that cannot proceed. | |
1211 | */ | |
1212 | int | |
1213 | xfs_dev_is_read_only( | |
1214 | struct xfs_mount *mp, | |
1215 | char *message) | |
1216 | { | |
1217 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1218 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1219 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1220 | xfs_notice(mp, "%s required on read-only device.", message); |
1221 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1222 | return -EROFS; |
dda35b8f CH |
1223 | } |
1224 | return 0; | |
1225 | } | |
f467cad9 DW |
1226 | |
1227 | /* Force the summary counters to be recalculated at next mount. */ | |
1228 | void | |
1229 | xfs_force_summary_recalc( | |
1230 | struct xfs_mount *mp) | |
1231 | { | |
38c26bfd | 1232 | if (!xfs_has_lazysbcount(mp)) |
f467cad9 DW |
1233 | return; |
1234 | ||
39353ff6 | 1235 | xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); |
f467cad9 | 1236 | } |
9fe82b8c | 1237 | |
908ce71e DW |
1238 | /* |
1239 | * Enable a log incompat feature flag in the primary superblock. The caller | |
1240 | * cannot have any other transactions in progress. | |
1241 | */ | |
1242 | int | |
1243 | xfs_add_incompat_log_feature( | |
1244 | struct xfs_mount *mp, | |
1245 | uint32_t feature) | |
1246 | { | |
1247 | struct xfs_dsb *dsb; | |
1248 | int error; | |
1249 | ||
1250 | ASSERT(hweight32(feature) == 1); | |
1251 | ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); | |
1252 | ||
1253 | /* | |
1254 | * Force the log to disk and kick the background AIL thread to reduce | |
1255 | * the chances that the bwrite will stall waiting for the AIL to unpin | |
1256 | * the primary superblock buffer. This isn't a data integrity | |
1257 | * operation, so we don't need a synchronous push. | |
1258 | */ | |
1259 | error = xfs_log_force(mp, XFS_LOG_SYNC); | |
1260 | if (error) | |
1261 | return error; | |
1262 | xfs_ail_push_all(mp->m_ail); | |
1263 | ||
1264 | /* | |
1265 | * Lock the primary superblock buffer to serialize all callers that | |
1266 | * are trying to set feature bits. | |
1267 | */ | |
1268 | xfs_buf_lock(mp->m_sb_bp); | |
1269 | xfs_buf_hold(mp->m_sb_bp); | |
1270 | ||
1271 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1272 | error = -EIO; | |
1273 | goto rele; | |
1274 | } | |
1275 | ||
1276 | if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature)) | |
1277 | goto rele; | |
1278 | ||
1279 | /* | |
1280 | * Write the primary superblock to disk immediately, because we need | |
1281 | * the log_incompat bit to be set in the primary super now to protect | |
1282 | * the log items that we're going to commit later. | |
1283 | */ | |
1284 | dsb = mp->m_sb_bp->b_addr; | |
1285 | xfs_sb_to_disk(dsb, &mp->m_sb); | |
1286 | dsb->sb_features_log_incompat |= cpu_to_be32(feature); | |
1287 | error = xfs_bwrite(mp->m_sb_bp); | |
1288 | if (error) | |
1289 | goto shutdown; | |
1290 | ||
1291 | /* | |
1292 | * Add the feature bits to the incore superblock before we unlock the | |
1293 | * buffer. | |
1294 | */ | |
1295 | xfs_sb_add_incompat_log_features(&mp->m_sb, feature); | |
1296 | xfs_buf_relse(mp->m_sb_bp); | |
1297 | ||
1298 | /* Log the superblock to disk. */ | |
1299 | return xfs_sync_sb(mp, false); | |
1300 | shutdown: | |
1301 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1302 | rele: | |
1303 | xfs_buf_relse(mp->m_sb_bp); | |
1304 | return error; | |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * Clear all the log incompat flags from the superblock. | |
1309 | * | |
1310 | * The caller cannot be in a transaction, must ensure that the log does not | |
1311 | * contain any log items protected by any log incompat bit, and must ensure | |
1312 | * that there are no other threads that depend on the state of the log incompat | |
1313 | * feature flags in the primary super. | |
1314 | * | |
1315 | * Returns true if the superblock is dirty. | |
1316 | */ | |
1317 | bool | |
1318 | xfs_clear_incompat_log_features( | |
1319 | struct xfs_mount *mp) | |
1320 | { | |
1321 | bool ret = false; | |
1322 | ||
1323 | if (!xfs_sb_version_hascrc(&mp->m_sb) || | |
1324 | !xfs_sb_has_incompat_log_feature(&mp->m_sb, | |
1325 | XFS_SB_FEAT_INCOMPAT_LOG_ALL) || | |
1326 | XFS_FORCED_SHUTDOWN(mp)) | |
1327 | return false; | |
1328 | ||
1329 | /* | |
1330 | * Update the incore superblock. We synchronize on the primary super | |
1331 | * buffer lock to be consistent with the add function, though at least | |
1332 | * in theory this shouldn't be necessary. | |
1333 | */ | |
1334 | xfs_buf_lock(mp->m_sb_bp); | |
1335 | xfs_buf_hold(mp->m_sb_bp); | |
1336 | ||
1337 | if (xfs_sb_has_incompat_log_feature(&mp->m_sb, | |
1338 | XFS_SB_FEAT_INCOMPAT_LOG_ALL)) { | |
1339 | xfs_info(mp, "Clearing log incompat feature flags."); | |
1340 | xfs_sb_remove_incompat_log_features(&mp->m_sb); | |
1341 | ret = true; | |
1342 | } | |
1343 | ||
1344 | xfs_buf_relse(mp->m_sb_bp); | |
1345 | return ret; | |
1346 | } | |
1347 | ||
9fe82b8c DW |
1348 | /* |
1349 | * Update the in-core delayed block counter. | |
1350 | * | |
1351 | * We prefer to update the counter without having to take a spinlock for every | |
1352 | * counter update (i.e. batching). Each change to delayed allocation | |
1353 | * reservations can change can easily exceed the default percpu counter | |
1354 | * batching, so we use a larger batch factor here. | |
1355 | * | |
1356 | * Note that we don't currently have any callers requiring fast summation | |
1357 | * (e.g. percpu_counter_read) so we can use a big batch value here. | |
1358 | */ | |
1359 | #define XFS_DELALLOC_BATCH (4096) | |
1360 | void | |
1361 | xfs_mod_delalloc( | |
1362 | struct xfs_mount *mp, | |
1363 | int64_t delta) | |
1364 | { | |
1365 | percpu_counter_add_batch(&mp->m_delalloc_blks, delta, | |
1366 | XFS_DELALLOC_BATCH); | |
1367 | } |