Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 : * All Rights Reserved.
5 : */
6 : #include "xfs.h"
7 : #include "xfs_fs.h"
8 : #include "xfs_shared.h"
9 : #include "xfs_format.h"
10 : #include "xfs_log_format.h"
11 : #include "xfs_trans_resv.h"
12 : #include "xfs_bit.h"
13 : #include "xfs_sb.h"
14 : #include "xfs_mount.h"
15 : #include "xfs_inode.h"
16 : #include "xfs_dir2.h"
17 : #include "xfs_ialloc.h"
18 : #include "xfs_alloc.h"
19 : #include "xfs_rtalloc.h"
20 : #include "xfs_bmap.h"
21 : #include "xfs_trans.h"
22 : #include "xfs_trans_priv.h"
23 : #include "xfs_log.h"
24 : #include "xfs_log_priv.h"
25 : #include "xfs_error.h"
26 : #include "xfs_quota.h"
27 : #include "xfs_fsops.h"
28 : #include "xfs_icache.h"
29 : #include "xfs_sysfs.h"
30 : #include "xfs_rmap_btree.h"
31 : #include "xfs_refcount_btree.h"
32 : #include "xfs_reflink.h"
33 : #include "xfs_extent_busy.h"
34 : #include "xfs_health.h"
35 : #include "xfs_trace.h"
36 : #include "xfs_ag.h"
37 : #include "xfs_imeta.h"
38 : #include "xfs_rtgroup.h"
39 : #include "xfs_rtrmap_btree.h"
40 : #include "xfs_rtrefcount_btree.h"
41 : #include "scrub/stats.h"
42 :
43 : static DEFINE_MUTEX(xfs_uuid_table_mutex);
44 : static int xfs_uuid_table_size;
45 : static uuid_t *xfs_uuid_table;
46 :
47 : void
48 12 : xfs_uuid_table_free(void)
49 : {
50 12 : if (xfs_uuid_table_size == 0)
51 : return;
52 12 : kmem_free(xfs_uuid_table);
53 12 : xfs_uuid_table = NULL;
54 12 : xfs_uuid_table_size = 0;
55 : }
56 :
57 : /*
58 : * See if the UUID is unique among mounted XFS filesystems.
59 : * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
60 : */
61 : STATIC int
62 24337 : xfs_uuid_mount(
63 : struct xfs_mount *mp)
64 : {
65 24337 : uuid_t *uuid = &mp->m_sb.sb_uuid;
66 24337 : int hole, i;
67 :
68 : /* Publish UUID in struct super_block */
69 24337 : uuid_copy(&mp->m_super->s_uuid, uuid);
70 :
71 24337 : if (xfs_has_nouuid(mp))
72 : return 0;
73 :
74 24333 : if (uuid_is_null(uuid)) {
75 0 : xfs_warn(mp, "Filesystem has null UUID - can't mount");
76 0 : return -EINVAL;
77 : }
78 :
79 24333 : mutex_lock(&xfs_uuid_table_mutex);
80 106625 : for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
81 57963 : if (uuid_is_null(&xfs_uuid_table[i])) {
82 36070 : hole = i;
83 36070 : continue;
84 : }
85 21893 : if (uuid_equal(uuid, &xfs_uuid_table[i]))
86 4 : goto out_duplicate;
87 : }
88 :
89 24329 : if (hole < 0) {
90 74 : xfs_uuid_table = krealloc(xfs_uuid_table,
91 37 : (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
92 : GFP_KERNEL | __GFP_NOFAIL);
93 37 : hole = xfs_uuid_table_size++;
94 : }
95 24329 : xfs_uuid_table[hole] = *uuid;
96 24329 : mutex_unlock(&xfs_uuid_table_mutex);
97 :
98 24329 : return 0;
99 :
100 : out_duplicate:
101 4 : mutex_unlock(&xfs_uuid_table_mutex);
102 4 : xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
103 4 : return -EINVAL;
104 : }
105 :
106 : STATIC void
107 24339 : xfs_uuid_unmount(
108 : struct xfs_mount *mp)
109 : {
110 24339 : uuid_t *uuid = &mp->m_sb.sb_uuid;
111 24339 : int i;
112 :
113 24339 : if (xfs_has_nouuid(mp))
114 : return;
115 :
116 24335 : mutex_lock(&xfs_uuid_table_mutex);
117 61902 : for (i = 0; i < xfs_uuid_table_size; i++) {
118 37567 : if (uuid_is_null(&xfs_uuid_table[i]))
119 10634 : continue;
120 26933 : if (!uuid_equal(uuid, &xfs_uuid_table[i]))
121 2598 : continue;
122 24335 : memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
123 24335 : break;
124 : }
125 24335 : ASSERT(i < xfs_uuid_table_size);
126 24335 : mutex_unlock(&xfs_uuid_table_mutex);
127 : }
128 :
129 : /*
130 : * Check size of device based on the (data/realtime) block count.
131 : * Note: this check is used by the growfs code as well as mount.
132 : */
133 : int
134 49025 : xfs_sb_validate_fsb_count(
135 : xfs_sb_t *sbp,
136 : uint64_t nblocks)
137 : {
138 49025 : ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
139 49025 : ASSERT(sbp->sb_blocklog >= BBSHIFT);
140 :
141 : /* Limited by ULONG_MAX of page cache index */
142 49025 : if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
143 : return -EFBIG;
144 49025 : return 0;
145 : }
146 :
147 : /*
148 : * xfs_readsb
149 : *
150 : * Does the initial read of the superblock.
151 : */
152 : int
153 24375 : xfs_readsb(
154 : struct xfs_mount *mp,
155 : int flags)
156 : {
157 24375 : unsigned int sector_size;
158 24375 : struct xfs_buf *bp;
159 24375 : struct xfs_sb *sbp = &mp->m_sb;
160 24375 : int error;
161 24375 : int loud = !(flags & XFS_MFSI_QUIET);
162 24375 : const struct xfs_buf_ops *buf_ops;
163 :
164 24375 : ASSERT(mp->m_sb_bp == NULL);
165 24375 : ASSERT(mp->m_ddev_targp != NULL);
166 :
167 : /*
168 : * For the initial read, we must guess at the sector
169 : * size based on the block device. It's enough to
170 : * get the sb_sectsize out of the superblock and
171 : * then reread with the proper length.
172 : * We don't verify it yet, because it may not be complete.
173 : */
174 24375 : sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
175 : buf_ops = NULL;
176 :
177 : /*
178 : * Allocate a (locked) buffer to hold the superblock. This will be kept
179 : * around at all times to optimize access to the superblock. Therefore,
180 : * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
181 : * elevated.
182 : */
183 24375 : reread:
184 48748 : error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
185 48748 : BTOBB(sector_size), XBF_NO_IOACCT, &bp,
186 : buf_ops);
187 48748 : if (error) {
188 16 : if (loud)
189 16 : xfs_warn(mp, "SB validate failed with error %d.", error);
190 : /* bad CRC means corrupted metadata */
191 16 : if (error == -EFSBADCRC)
192 4 : error = -EFSCORRUPTED;
193 16 : return error;
194 : }
195 :
196 : /*
197 : * Initialize the mount structure from the superblock.
198 : */
199 48732 : xfs_sb_from_disk(sbp, bp->b_addr);
200 :
201 : /*
202 : * If we haven't validated the superblock, do so now before we try
203 : * to check the sector size and reread the superblock appropriately.
204 : */
205 48732 : if (sbp->sb_magicnum != XFS_SB_MAGIC) {
206 0 : if (loud)
207 0 : xfs_warn(mp, "Invalid superblock magic number");
208 0 : error = -EINVAL;
209 0 : goto release_buf;
210 : }
211 :
212 : /*
213 : * We must be able to do sector-sized and sector-aligned IO.
214 : */
215 48732 : if (sector_size > sbp->sb_sectsize) {
216 0 : if (loud)
217 0 : xfs_warn(mp, "device supports %u byte sectors (not %u)",
218 : sector_size, sbp->sb_sectsize);
219 0 : error = -ENOSYS;
220 0 : goto release_buf;
221 : }
222 :
223 48732 : if (buf_ops == NULL) {
224 : /*
225 : * Re-read the superblock so the buffer is correctly sized,
226 : * and properly verified.
227 : */
228 24373 : xfs_buf_relse(bp);
229 24373 : sector_size = sbp->sb_sectsize;
230 24373 : buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
231 24373 : goto reread;
232 : }
233 :
234 24359 : mp->m_features |= xfs_sb_version_to_features(sbp);
235 24359 : xfs_reinit_percpu_counters(mp);
236 :
237 : /* no need to be quiet anymore, so reset the buf ops */
238 24359 : bp->b_ops = &xfs_sb_buf_ops;
239 :
240 24359 : mp->m_sb_bp = bp;
241 24359 : xfs_buf_unlock(bp);
242 24359 : return 0;
243 :
244 0 : release_buf:
245 0 : xfs_buf_relse(bp);
246 0 : return error;
247 : }
248 :
249 : /*
250 : * If the sunit/swidth change would move the precomputed root inode value, we
251 : * must reject the ondisk change because repair will stumble over that.
252 : * However, we allow the mount to proceed because we never rejected this
253 : * combination before. Returns true to update the sb, false otherwise.
254 : */
255 : static inline int
256 14 : xfs_check_new_dalign(
257 : struct xfs_mount *mp,
258 : int new_dalign,
259 : bool *update_sb)
260 : {
261 14 : struct xfs_sb *sbp = &mp->m_sb;
262 14 : xfs_ino_t calc_ino;
263 :
264 14 : calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
265 14 : trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
266 :
267 14 : if (sbp->sb_rootino == calc_ino) {
268 10 : *update_sb = true;
269 10 : return 0;
270 : }
271 :
272 4 : xfs_warn(mp,
273 : "Cannot change stripe alignment; would require moving root inode.");
274 :
275 : /*
276 : * XXX: Next time we add a new incompat feature, this should start
277 : * returning -EINVAL to fail the mount. Until then, spit out a warning
278 : * that we're ignoring the administrator's instructions.
279 : */
280 4 : xfs_warn(mp, "Skipping superblock stripe alignment update.");
281 4 : *update_sb = false;
282 4 : return 0;
283 : }
284 :
285 : /*
286 : * If we were provided with new sunit/swidth values as mount options, make sure
287 : * that they pass basic alignment and superblock feature checks, and convert
288 : * them into the same units (FSB) that everything else expects. This step
289 : * /must/ be done before computing the inode geometry.
290 : */
291 : STATIC int
292 24343 : xfs_validate_new_dalign(
293 : struct xfs_mount *mp)
294 : {
295 24343 : if (mp->m_dalign == 0)
296 : return 0;
297 :
298 : /*
299 : * If stripe unit and stripe width are not multiples
300 : * of the fs blocksize turn off alignment.
301 : */
302 22 : if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
303 20 : (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
304 2 : xfs_warn(mp,
305 : "alignment check failed: sunit/swidth vs. blocksize(%d)",
306 : mp->m_sb.sb_blocksize);
307 2 : return -EINVAL;
308 : }
309 :
310 : /*
311 : * Convert the stripe unit and width to FSBs.
312 : */
313 20 : mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
314 20 : if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
315 4 : xfs_warn(mp,
316 : "alignment check failed: sunit/swidth vs. agsize(%d)",
317 : mp->m_sb.sb_agblocks);
318 4 : return -EINVAL;
319 : }
320 :
321 16 : if (!mp->m_dalign) {
322 0 : xfs_warn(mp,
323 : "alignment check failed: sunit(%d) less than bsize(%d)",
324 : mp->m_dalign, mp->m_sb.sb_blocksize);
325 0 : return -EINVAL;
326 : }
327 :
328 16 : mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
329 :
330 16 : if (!xfs_has_dalign(mp)) {
331 0 : xfs_warn(mp,
332 : "cannot change alignment: superblock does not support data alignment");
333 0 : return -EINVAL;
334 : }
335 :
336 : return 0;
337 : }
338 :
339 : /* Update alignment values based on mount options and sb values. */
340 : STATIC int
341 24337 : xfs_update_alignment(
342 : struct xfs_mount *mp)
343 : {
344 24337 : struct xfs_sb *sbp = &mp->m_sb;
345 :
346 24337 : if (mp->m_dalign) {
347 16 : bool update_sb;
348 16 : int error;
349 :
350 16 : if (sbp->sb_unit == mp->m_dalign &&
351 : sbp->sb_width == mp->m_swidth)
352 6 : return 0;
353 :
354 14 : error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
355 14 : if (error || !update_sb)
356 : return error;
357 :
358 10 : sbp->sb_unit = mp->m_dalign;
359 10 : sbp->sb_width = mp->m_swidth;
360 10 : mp->m_update_sb = true;
361 24321 : } else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) {
362 92 : mp->m_dalign = sbp->sb_unit;
363 92 : mp->m_swidth = sbp->sb_width;
364 : }
365 :
366 : return 0;
367 : }
368 :
369 : /*
370 : * precalculate the low space thresholds for dynamic speculative preallocation.
371 : */
372 : void
373 24460 : xfs_set_low_space_thresholds(
374 : struct xfs_mount *mp)
375 : {
376 24460 : uint64_t dblocks = mp->m_sb.sb_dblocks;
377 24460 : uint64_t rtexts = mp->m_sb.sb_rextents;
378 24460 : int i;
379 :
380 24460 : do_div(dblocks, 100);
381 24460 : do_div(rtexts, 100);
382 :
383 146760 : for (i = 0; i < XFS_LOWSP_MAX; i++) {
384 122300 : mp->m_low_space[i] = dblocks * (i + 1);
385 122300 : mp->m_low_rtexts[i] = rtexts * (i + 1);
386 : }
387 24460 : }
388 :
389 : /*
390 : * Check that the data (and log if separate) is an ok size.
391 : */
392 : STATIC int
393 24333 : xfs_check_sizes(
394 : struct xfs_mount *mp)
395 : {
396 24333 : struct xfs_buf *bp;
397 24333 : xfs_daddr_t d;
398 24333 : int error;
399 :
400 24333 : d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
401 24333 : if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
402 0 : xfs_warn(mp, "filesystem size mismatch detected");
403 0 : return -EFBIG;
404 : }
405 24333 : error = xfs_buf_read_uncached(mp->m_ddev_targp,
406 24333 : d - XFS_FSS_TO_BB(mp, 1),
407 : XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
408 24333 : if (error) {
409 0 : xfs_warn(mp, "last sector read failed");
410 0 : return error;
411 : }
412 24333 : xfs_buf_relse(bp);
413 :
414 24333 : if (mp->m_logdev_targp == mp->m_ddev_targp)
415 : return 0;
416 :
417 2 : d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
418 2 : if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
419 0 : xfs_warn(mp, "log size mismatch detected");
420 0 : return -EFBIG;
421 : }
422 2 : error = xfs_buf_read_uncached(mp->m_logdev_targp,
423 2 : d - XFS_FSB_TO_BB(mp, 1),
424 : XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
425 2 : if (error) {
426 0 : xfs_warn(mp, "log device read failed");
427 0 : return error;
428 : }
429 2 : xfs_buf_relse(bp);
430 2 : return 0;
431 : }
432 :
433 : /*
434 : * Clear the quotaflags in memory and in the superblock.
435 : */
436 : int
437 34 : xfs_mount_reset_sbqflags(
438 : struct xfs_mount *mp)
439 : {
440 34 : mp->m_qflags = 0;
441 :
442 : /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
443 34 : if (mp->m_sb.sb_qflags == 0)
444 : return 0;
445 34 : spin_lock(&mp->m_sb_lock);
446 34 : mp->m_sb.sb_qflags = 0;
447 34 : spin_unlock(&mp->m_sb_lock);
448 :
449 34 : if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
450 : return 0;
451 :
452 34 : return xfs_sync_sb(mp, false);
453 : }
454 :
455 : uint64_t
456 4 : xfs_default_resblks(xfs_mount_t *mp)
457 : {
458 22102 : uint64_t resblks;
459 :
460 : /*
461 : * We default to 5% or 8192 fsbs of space reserved, whichever is
462 : * smaller. This is intended to cover concurrent allocation
463 : * transactions when we initially hit enospc. These each require a 4
464 : * block reservation. Hence by default we cover roughly 2000 concurrent
465 : * allocation reservations.
466 : */
467 22102 : resblks = mp->m_sb.sb_dblocks;
468 22102 : do_div(resblks, 20);
469 22102 : resblks = min_t(uint64_t, resblks, 8192);
470 22102 : return resblks;
471 : }
472 :
473 : /* Ensure the summary counts are correct. */
474 : STATIC int
475 24321 : xfs_check_summary_counts(
476 : struct xfs_mount *mp)
477 : {
478 24321 : int error = 0;
479 :
480 : /*
481 : * The AG0 superblock verifier rejects in-progress filesystems,
482 : * so we should never see the flag set this far into mounting.
483 : */
484 24321 : if (mp->m_sb.sb_inprogress) {
485 0 : xfs_err(mp, "sb_inprogress set after log recovery??");
486 0 : WARN_ON(1);
487 0 : return -EFSCORRUPTED;
488 : }
489 :
490 : /*
491 : * Now the log is mounted, we know if it was an unclean shutdown or
492 : * not. If it was, with the first phase of recovery has completed, we
493 : * have consistent AG blocks on disk. We have not recovered EFIs yet,
494 : * but they are recovered transactionally in the second recovery phase
495 : * later.
496 : *
497 : * If the log was clean when we mounted, we can check the summary
498 : * counters. If any of them are obviously incorrect, we can recompute
499 : * them from the AGF headers in the next step.
500 : */
501 48642 : if (xfs_is_clean(mp) &&
502 25880 : (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
503 12938 : !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
504 12938 : mp->m_sb.sb_ifree > mp->m_sb.sb_icount)) {
505 6 : xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
506 6 : xfs_fs_mark_checked(mp, XFS_SICK_FS_COUNTERS);
507 : }
508 :
509 : /*
510 : * We can safely re-initialise incore superblock counters from the
511 : * per-ag data. These may not be correct if the filesystem was not
512 : * cleanly unmounted, so we waited for recovery to finish before doing
513 : * this.
514 : *
515 : * If the filesystem was cleanly unmounted or the previous check did
516 : * not flag anything weird, then we can trust the values in the
517 : * superblock to be correct and we don't need to do anything here.
518 : * Otherwise, recalculate the summary counters.
519 : */
520 61578 : if ((xfs_has_lazysbcount(mp) && !xfs_is_clean(mp)) ||
521 : xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) {
522 11385 : error = xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
523 11385 : if (error)
524 : return error;
525 : }
526 :
527 : /*
528 : * Older kernels misused sb_frextents to reflect both incore
529 : * reservations made by running transactions and the actual count of
530 : * free rt extents in the ondisk metadata. Transactions committed
531 : * during runtime can therefore contain a superblock update that
532 : * undercounts the number of free rt extents tracked in the rt bitmap.
533 : * A clean unmount record will have the correct frextents value since
534 : * there can be no other transactions running at that point.
535 : *
536 : * If we're mounting the rt volume after recovering the log, recompute
537 : * frextents from the rtbitmap file to fix the inconsistency.
538 : */
539 24560 : if (xfs_has_realtime(mp) && !xfs_is_clean(mp)) {
540 0 : error = xfs_rtalloc_reinit_frextents(mp);
541 0 : if (error)
542 0 : return error;
543 : }
544 :
545 : return 0;
546 : }
547 :
548 : static void
549 24297 : xfs_unmount_check(
550 : struct xfs_mount *mp)
551 : {
552 48594 : if (xfs_is_shutdown(mp))
553 : return;
554 :
555 24902 : if (percpu_counter_sum(&mp->m_ifree) >
556 12451 : percpu_counter_sum(&mp->m_icount)) {
557 0 : xfs_alert(mp, "ifree/icount mismatch at unmount");
558 0 : xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
559 : }
560 : }
561 :
562 : /*
563 : * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
564 : * internal inode structures can be sitting in the CIL and AIL at this point,
565 : * so we need to unpin them, write them back and/or reclaim them before unmount
566 : * can proceed. In other words, callers are required to have inactivated all
567 : * inodes.
568 : *
569 : * An inode cluster that has been freed can have its buffer still pinned in
570 : * memory because the transaction is still sitting in a iclog. The stale inodes
571 : * on that buffer will be pinned to the buffer until the transaction hits the
572 : * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
573 : * may never see the pinned buffer, so nothing will push out the iclog and
574 : * unpin the buffer.
575 : *
576 : * Hence we need to force the log to unpin everything first. However, log
577 : * forces don't wait for the discards they issue to complete, so we have to
578 : * explicitly wait for them to complete here as well.
579 : *
580 : * Then we can tell the world we are unmounting so that error handling knows
581 : * that the filesystem is going away and we should error out anything that we
582 : * have been retrying in the background. This will prevent never-ending
583 : * retries in AIL pushing from hanging the unmount.
584 : *
585 : * Finally, we can push the AIL to clean all the remaining dirty objects, then
586 : * reclaim the remaining inodes that are still in memory at this point in time.
587 : */
588 : static void
589 24331 : xfs_unmount_flush_inodes(
590 : struct xfs_mount *mp)
591 : {
592 24331 : xfs_log_force(mp, XFS_LOG_SYNC);
593 24331 : xfs_extent_busy_wait_all(mp);
594 24331 : flush_workqueue(xfs_discard_wq);
595 :
596 24331 : set_bit(XFS_OPSTATE_UNMOUNTING, &mp->m_opstate);
597 :
598 24331 : xfs_ail_push_all_sync(mp->m_ail);
599 24331 : xfs_inodegc_stop(mp);
600 24331 : cancel_delayed_work_sync(&mp->m_reclaim_work);
601 24331 : xfs_reclaim_inodes(mp);
602 24331 : xfs_health_unmount(mp);
603 24331 : }
604 :
605 : static void
606 24337 : xfs_mount_setup_inode_geom(
607 : struct xfs_mount *mp)
608 : {
609 24337 : struct xfs_ino_geometry *igeo = M_IGEO(mp);
610 :
611 24337 : igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
612 24381 : ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
613 :
614 24337 : xfs_ialloc_setup_geometry(mp);
615 24337 : }
616 :
617 : STATIC int
618 24325 : xfs_mount_setup_metadir(
619 : struct xfs_mount *mp)
620 : {
621 24325 : int error;
622 :
623 : /* Load the metadata directory inode into memory. */
624 24325 : if (xfs_has_metadir(mp)) {
625 23722 : error = xfs_imeta_iget(mp, mp->m_sb.sb_metadirino,
626 : XFS_DIR3_FT_DIR, &mp->m_metadirip);
627 23722 : if (error) {
628 4 : xfs_warn(mp, "Failed metadir inode init: %d", error);
629 4 : return error;
630 : }
631 : }
632 :
633 24321 : error = xfs_imeta_mount(mp);
634 24321 : if (error) {
635 0 : xfs_warn(mp, "Failed to load metadata inode info, error %d",
636 : error);
637 0 : return error;
638 : }
639 :
640 : return 0;
641 : }
642 :
643 : /* Compute maximum possible height for per-AG btree types for this fs. */
644 : static inline void
645 : xfs_agbtree_compute_maxlevels(
646 : struct xfs_mount *mp)
647 : {
648 24337 : unsigned int levels;
649 :
650 24337 : levels = max(mp->m_alloc_maxlevels, M_IGEO(mp)->inobt_maxlevels);
651 24337 : levels = max(levels, mp->m_rmap_maxlevels);
652 24337 : mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels);
653 : }
654 :
655 : /* Compute maximum possible height for realtime btree types for this fs. */
656 : static inline void
657 : xfs_rtbtree_compute_maxlevels(
658 : struct xfs_mount *mp)
659 : {
660 24337 : unsigned int levels;
661 :
662 24337 : levels = max(mp->m_rtrmap_maxlevels, mp->m_rtrefc_maxlevels);
663 24337 : mp->m_rtbtree_maxlevels = levels;
664 : }
665 :
666 : /*
667 : * This function does the following on an initial mount of a file system:
668 : * - reads the superblock from disk and init the mount struct
669 : * - if we're a 32-bit kernel, do a size check on the superblock
670 : * so we don't mount terabyte filesystems
671 : * - init mount struct realtime fields
672 : * - allocate inode hash table for fs
673 : * - init directory manager
674 : * - perform recovery and init the log manager
675 : */
676 : int
677 24343 : xfs_mountfs(
678 : struct xfs_mount *mp)
679 : {
680 24343 : struct xfs_sb *sbp = &(mp->m_sb);
681 24343 : struct xfs_inode *rip;
682 24343 : struct xfs_ino_geometry *igeo = M_IGEO(mp);
683 24343 : uint64_t resblks;
684 24343 : uint quotamount = 0;
685 24343 : uint quotaflags = 0;
686 24343 : int error = 0;
687 :
688 24343 : xfs_sb_mount_common(mp, sbp);
689 :
690 : /*
691 : * Check for a mismatched features2 values. Older kernels read & wrote
692 : * into the wrong sb offset for sb_features2 on some platforms due to
693 : * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
694 : * which made older superblock reading/writing routines swap it as a
695 : * 64-bit value.
696 : *
697 : * For backwards compatibility, we make both slots equal.
698 : *
699 : * If we detect a mismatched field, we OR the set bits into the existing
700 : * features2 field in case it has already been modified; we don't want
701 : * to lose any features. We then update the bad location with the ORed
702 : * value so that older kernels will see any features2 flags. The
703 : * superblock writeback code ensures the new sb_features2 is copied to
704 : * sb_bad_features2 before it is logged or written to disk.
705 : */
706 24343 : if (xfs_sb_has_mismatched_features2(sbp)) {
707 4 : xfs_warn(mp, "correcting sb_features alignment problem");
708 4 : sbp->sb_features2 |= sbp->sb_bad_features2;
709 4 : mp->m_update_sb = true;
710 : }
711 :
712 :
713 : /* always use v2 inodes by default now */
714 24343 : if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
715 0 : mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
716 0 : mp->m_features |= XFS_FEAT_NLINK;
717 0 : mp->m_update_sb = true;
718 : }
719 :
720 : /*
721 : * If we were given new sunit/swidth options, do some basic validation
722 : * checks and convert the incore dalign and swidth values to the
723 : * same units (FSB) that everything else uses. This /must/ happen
724 : * before computing the inode geometry.
725 : */
726 24343 : error = xfs_validate_new_dalign(mp);
727 24343 : if (error)
728 6 : goto out;
729 :
730 24337 : xfs_alloc_compute_maxlevels(mp);
731 24337 : xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
732 24337 : xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
733 24337 : xfs_mount_setup_inode_geom(mp);
734 24337 : xfs_rmapbt_compute_maxlevels(mp);
735 24337 : xfs_rtrmapbt_compute_maxlevels(mp);
736 24337 : xfs_refcountbt_compute_maxlevels(mp);
737 24337 : xfs_rtrefcountbt_compute_maxlevels(mp);
738 :
739 24337 : xfs_agbtree_compute_maxlevels(mp);
740 24337 : xfs_rtbtree_compute_maxlevels(mp);
741 :
742 : /*
743 : * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks
744 : * is NOT aligned turn off m_dalign since allocator alignment is within
745 : * an ag, therefore ag has to be aligned at stripe boundary. Note that
746 : * we must compute the free space and rmap btree geometry before doing
747 : * this.
748 : */
749 24337 : error = xfs_update_alignment(mp);
750 24337 : if (error)
751 0 : goto out;
752 :
753 : /* enable fail_at_unmount as default */
754 24337 : mp->m_fail_unmount = true;
755 :
756 24337 : error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
757 24337 : NULL, mp->m_super->s_id);
758 24337 : if (error)
759 0 : goto out;
760 :
761 24337 : error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
762 : &mp->m_kobj, "stats");
763 24337 : if (error)
764 0 : goto out_remove_sysfs;
765 :
766 24337 : xchk_stats_register(mp->m_scrub_stats, mp->m_debugfs);
767 :
768 24337 : error = xfs_error_sysfs_init(mp);
769 24337 : if (error)
770 0 : goto out_remove_scrub_stats;
771 :
772 24337 : error = xfs_errortag_init(mp);
773 24337 : if (error)
774 0 : goto out_remove_error_sysfs;
775 :
776 24337 : error = xfs_uuid_mount(mp);
777 24337 : if (error)
778 4 : goto out_remove_errortag;
779 :
780 : /*
781 : * Update the preferred write size based on the information from the
782 : * on-disk superblock.
783 : */
784 24333 : mp->m_allocsize_log =
785 24333 : max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
786 24333 : mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
787 :
788 : /* set the low space thresholds for dynamic preallocation */
789 24333 : xfs_set_low_space_thresholds(mp);
790 :
791 : /*
792 : * If enabled, sparse inode chunk alignment is expected to match the
793 : * cluster size. Full inode chunk alignment must match the chunk size,
794 : * but that is checked on sb read verification...
795 : */
796 24333 : if (xfs_has_sparseinodes(mp) &&
797 24291 : mp->m_sb.sb_spino_align !=
798 24291 : XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
799 0 : xfs_warn(mp,
800 : "Sparse inode block alignment (%u) must match cluster size (%llu).",
801 : mp->m_sb.sb_spino_align,
802 : XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
803 0 : error = -EINVAL;
804 0 : goto out_remove_uuid;
805 : }
806 :
807 : /*
808 : * Check that the data (and log if separate) is an ok size.
809 : */
810 24333 : error = xfs_check_sizes(mp);
811 24333 : if (error)
812 0 : goto out_remove_uuid;
813 :
814 : /*
815 : * Initialize realtime fields in the mount structure
816 : */
817 24333 : error = xfs_rtmount_init(mp);
818 24333 : if (error) {
819 0 : xfs_warn(mp, "RT mount failed");
820 0 : goto out_remove_uuid;
821 : }
822 :
823 : /*
824 : * Copies the low order bits of the timestamp and the randomly
825 : * set "sequence" number out of a UUID.
826 : */
827 24333 : mp->m_fixedfsid[0] =
828 24333 : (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
829 24333 : get_unaligned_be16(&sbp->sb_uuid.b[4]);
830 24333 : mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
831 :
832 24333 : error = xfs_da_mount(mp);
833 24333 : if (error) {
834 0 : xfs_warn(mp, "Failed dir/attr init: %d", error);
835 0 : goto out_remove_uuid;
836 : }
837 :
838 : /*
839 : * Initialize the precomputed transaction reservations values.
840 : */
841 24333 : xfs_trans_init(mp);
842 :
843 : /*
844 : * Allocate and initialize the per-ag data.
845 : */
846 24333 : error = xfs_initialize_perag(mp, sbp->sb_agcount, mp->m_sb.sb_dblocks,
847 : &mp->m_maxagi);
848 24333 : if (error) {
849 0 : xfs_warn(mp, "Failed per-ag init: %d", error);
850 0 : goto out_free_dir;
851 : }
852 :
853 24333 : error = xfs_initialize_rtgroups(mp, sbp->sb_rgcount);
854 24333 : if (error) {
855 0 : xfs_warn(mp, "Failed rtgroup init: %d", error);
856 0 : goto out_free_perag;
857 : }
858 :
859 24333 : if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
860 0 : xfs_warn(mp, "no log defined");
861 0 : error = -EFSCORRUPTED;
862 0 : goto out_free_rtgroup;
863 : }
864 :
865 24333 : error = xfs_inodegc_register_shrinker(mp);
866 24333 : if (error)
867 0 : goto out_fail_wait;
868 :
869 : /*
870 : * Log's mount-time initialization. The first part of recovery can place
871 : * some items on the AIL, to be handled when recovery is finished or
872 : * cancelled.
873 : */
874 24333 : error = xfs_log_mount(mp, mp->m_logdev_targp,
875 24333 : XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
876 24333 : XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
877 24333 : if (error) {
878 8 : xfs_warn(mp, "log mount failed");
879 8 : goto out_inodegc_shrinker;
880 : }
881 :
882 : /* Enable background inode inactivation workers. */
883 24325 : xfs_inodegc_start(mp);
884 24325 : xfs_blockgc_start(mp);
885 :
886 : /*
887 : * Now that we've recovered any pending superblock feature bit
888 : * additions, we can finish setting up the attr2 behaviour for the
889 : * mount. The noattr2 option overrides the superblock flag, so only
890 : * check the superblock feature flag if the mount option is not set.
891 : */
892 24325 : if (xfs_has_noattr2(mp)) {
893 2 : mp->m_features &= ~XFS_FEAT_ATTR2;
894 24323 : } else if (!xfs_has_attr2(mp) &&
895 6 : (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)) {
896 4 : mp->m_features |= XFS_FEAT_ATTR2;
897 : }
898 :
899 24325 : error = xfs_mount_setup_metadir(mp);
900 24325 : if (error)
901 4 : goto out_free_metadir;
902 :
903 : /*
904 : * Get and sanity-check the root inode.
905 : * Save the pointer to it in the mount structure.
906 : */
907 24321 : error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
908 : XFS_ILOCK_EXCL, &rip);
909 24321 : if (error) {
910 0 : xfs_warn(mp,
911 : "Failed to read root inode 0x%llx, error %d",
912 : sbp->sb_rootino, -error);
913 0 : goto out_free_metadir;
914 : }
915 :
916 24321 : ASSERT(rip != NULL);
917 :
918 24321 : if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
919 0 : xfs_warn(mp, "corrupted root inode %llu: not a directory",
920 : (unsigned long long)rip->i_ino);
921 0 : xfs_iunlock(rip, XFS_ILOCK_EXCL);
922 0 : error = -EFSCORRUPTED;
923 0 : goto out_rele_rip;
924 : }
925 24321 : mp->m_rootip = rip; /* save it */
926 :
927 24321 : xfs_iunlock(rip, XFS_ILOCK_EXCL);
928 :
929 : /*
930 : * Initialize realtime inode pointers in the mount structure
931 : */
932 24321 : error = xfs_rtmount_inodes(mp);
933 24321 : if (error) {
934 : /*
935 : * Free up the root inode.
936 : */
937 0 : xfs_warn(mp, "failed to read RT inodes");
938 0 : goto out_rele_rip;
939 : }
940 :
941 : /* Make sure the summary counts are ok. */
942 24321 : error = xfs_check_summary_counts(mp);
943 24321 : if (error)
944 2 : goto out_rtunmount;
945 :
946 : /*
947 : * If this is a read-only mount defer the superblock updates until
948 : * the next remount into writeable mode. Otherwise we would never
949 : * perform the update e.g. for the root filesystem.
950 : */
951 24333 : if (mp->m_update_sb && !xfs_is_readonly(mp)) {
952 12 : error = xfs_sync_sb(mp, false);
953 12 : if (error) {
954 0 : xfs_warn(mp, "failed to write sb changes");
955 0 : goto out_rtunmount;
956 : }
957 : }
958 :
959 : /*
960 : * Initialise the XFS quota management subsystem for this mount
961 : */
962 24319 : if (XFS_IS_QUOTA_ON(mp)) {
963 22322 : error = xfs_qm_newmount(mp, "amount, "aflags);
964 22322 : if (error)
965 4 : goto out_rtunmount;
966 : } else {
967 : /*
968 : * If a file system had quotas running earlier, but decided to
969 : * mount without -o uquota/pquota/gquota options, revoke the
970 : * quotachecked license.
971 : */
972 1997 : if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
973 34 : xfs_notice(mp, "resetting quota flags");
974 34 : error = xfs_mount_reset_sbqflags(mp);
975 34 : if (error)
976 0 : goto out_rtunmount;
977 : }
978 : }
979 :
980 : /*
981 : * Finish recovering the file system. This part needed to be delayed
982 : * until after the root and real-time bitmap inodes were consistently
983 : * read in. Temporarily create per-AG space reservations for metadata
984 : * btree shape changes because space freeing transactions (for inode
985 : * inactivation) require the per-AG reservation in lieu of reserving
986 : * blocks.
987 : */
988 24315 : error = xfs_fs_reserve_ag_blocks(mp);
989 24315 : if (error && error == -ENOSPC)
990 0 : xfs_warn(mp,
991 : "ENOSPC reserving per-AG metadata pool, log recovery may fail.");
992 24315 : error = xfs_log_mount_finish(mp);
993 24315 : xfs_fs_unreserve_ag_blocks(mp);
994 24315 : if (error) {
995 2 : xfs_warn(mp, "log mount finish failed");
996 2 : goto out_rtunmount;
997 : }
998 :
999 : /*
1000 : * Now the log is fully replayed, we can transition to full read-only
1001 : * mode for read-only mounts. This will sync all the metadata and clean
1002 : * the log so that the recovery we just performed does not have to be
1003 : * replayed again on the next mount.
1004 : *
1005 : * We use the same quiesce mechanism as the rw->ro remount, as they are
1006 : * semantically identical operations.
1007 : */
1008 48626 : if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp))
1009 2207 : xfs_log_clean(mp);
1010 :
1011 : /*
1012 : * Complete the quota initialisation, post-log-replay component.
1013 : */
1014 24313 : if (quotamount) {
1015 3081 : ASSERT(mp->m_qflags == 0);
1016 3081 : mp->m_qflags = quotaflags;
1017 :
1018 3081 : error = xfs_qm_mount_quotas(mp);
1019 3081 : if (error)
1020 0 : goto out_rtunmount;
1021 : }
1022 :
1023 : /*
1024 : * Now we are mounted, reserve a small amount of unused space for
1025 : * privileged transactions. This is needed so that transaction
1026 : * space required for critical operations can dip into this pool
1027 : * when at ENOSPC. This is needed for operations like create with
1028 : * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1029 : * are not allowed to use this reserved space.
1030 : *
1031 : * This may drive us straight to ENOSPC on mount, but that implies
1032 : * we were already there on the last unmount. Warn if this occurs.
1033 : */
1034 48626 : if (!xfs_is_readonly(mp)) {
1035 22098 : resblks = xfs_default_resblks(mp);
1036 22098 : error = xfs_reserve_blocks(mp, &resblks, NULL);
1037 22098 : if (error)
1038 0 : xfs_warn(mp,
1039 : "Unable to allocate reserve blocks. Continuing without reserve pool.");
1040 :
1041 : /* Reserve AG blocks for future btree expansion. */
1042 22098 : error = xfs_fs_reserve_ag_blocks(mp);
1043 22098 : if (error && error != -ENOSPC)
1044 22 : goto out_agresv;
1045 : }
1046 :
1047 : return 0;
1048 :
1049 : out_agresv:
1050 22 : xfs_fs_unreserve_ag_blocks(mp);
1051 22 : xfs_qm_unmount_quotas(mp);
1052 30 : out_rtunmount:
1053 30 : xfs_rtunmount_inodes(mp);
1054 30 : out_rele_rip:
1055 30 : xfs_irele(rip);
1056 : /* Clean out dquots that might be in memory after quotacheck. */
1057 30 : xfs_qm_unmount(mp);
1058 34 : out_free_metadir:
1059 34 : if (mp->m_metadirip)
1060 30 : xfs_imeta_irele(mp->m_metadirip);
1061 :
1062 : /*
1063 : * Inactivate all inodes that might still be in memory after a log
1064 : * intent recovery failure so that reclaim can free them. Metadata
1065 : * inodes and the root directory shouldn't need inactivation, but the
1066 : * mount failed for some reason, so pull down all the state and flee.
1067 : */
1068 34 : xfs_inodegc_flush(mp);
1069 :
1070 : /*
1071 : * Flush all inode reclamation work and flush the log.
1072 : * We have to do this /after/ rtunmount and qm_unmount because those
1073 : * two will have scheduled delayed reclaim for the rt/quota inodes.
1074 : *
1075 : * This is slightly different from the unmountfs call sequence
1076 : * because we could be tearing down a partially set up mount. In
1077 : * particular, if log_mount_finish fails we bail out without calling
1078 : * qm_unmount_quotas and therefore rely on qm_unmount to release the
1079 : * quota inodes.
1080 : */
1081 34 : xfs_unmount_flush_inodes(mp);
1082 34 : xfs_log_mount_cancel(mp);
1083 42 : out_inodegc_shrinker:
1084 42 : unregister_shrinker(&mp->m_inodegc_shrinker);
1085 42 : out_fail_wait:
1086 42 : if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1087 0 : xfs_buftarg_drain(mp->m_logdev_targp);
1088 42 : xfs_buftarg_drain(mp->m_ddev_targp);
1089 42 : out_free_rtgroup:
1090 42 : xfs_free_rtgroups(mp);
1091 42 : out_free_perag:
1092 42 : xfs_free_perag(mp);
1093 42 : out_free_dir:
1094 42 : xfs_da_unmount(mp);
1095 42 : out_remove_uuid:
1096 42 : xfs_uuid_unmount(mp);
1097 46 : out_remove_errortag:
1098 46 : xfs_errortag_del(mp);
1099 46 : out_remove_error_sysfs:
1100 46 : xfs_error_sysfs_del(mp);
1101 46 : out_remove_scrub_stats:
1102 46 : xchk_stats_unregister(mp->m_scrub_stats);
1103 46 : xfs_sysfs_del(&mp->m_stats.xs_kobj);
1104 46 : out_remove_sysfs:
1105 46 : xfs_sysfs_del(&mp->m_kobj);
1106 : out:
1107 : return error;
1108 : }
1109 :
1110 : /*
1111 : * This flushes out the inodes,dquots and the superblock, unmounts the
1112 : * log and makes sure that incore structures are freed.
1113 : */
1114 : void
1115 24297 : xfs_unmountfs(
1116 : struct xfs_mount *mp)
1117 : {
1118 24297 : uint64_t resblks;
1119 24297 : int error;
1120 :
1121 : /*
1122 : * Perform all on-disk metadata updates required to inactivate inodes
1123 : * that the VFS evicted earlier in the unmount process. Freeing inodes
1124 : * and discarding CoW fork preallocations can cause shape changes to
1125 : * the free inode and refcount btrees, respectively, so we must finish
1126 : * this before we discard the metadata space reservations. Metadata
1127 : * inodes and the root directory do not require inactivation.
1128 : */
1129 24297 : xfs_inodegc_flush(mp);
1130 :
1131 24297 : xfs_blockgc_stop(mp);
1132 24297 : xfs_fs_unreserve_ag_blocks(mp);
1133 24297 : xfs_qm_unmount_quotas(mp);
1134 24297 : xfs_rtunmount_inodes(mp);
1135 24297 : xfs_irele(mp->m_rootip);
1136 24297 : if (mp->m_metadirip)
1137 23694 : xfs_imeta_irele(mp->m_metadirip);
1138 :
1139 24297 : xfs_unmount_flush_inodes(mp);
1140 :
1141 24297 : xfs_qm_unmount(mp);
1142 :
1143 : /*
1144 : * Unreserve any blocks we have so that when we unmount we don't account
1145 : * the reserved free space as used. This is really only necessary for
1146 : * lazy superblock counting because it trusts the incore superblock
1147 : * counters to be absolutely correct on clean unmount.
1148 : *
1149 : * We don't bother correcting this elsewhere for lazy superblock
1150 : * counting because on mount of an unclean filesystem we reconstruct the
1151 : * correct counter value and this is irrelevant.
1152 : *
1153 : * For non-lazy counter filesystems, this doesn't matter at all because
1154 : * we only every apply deltas to the superblock and hence the incore
1155 : * value does not matter....
1156 : */
1157 24297 : resblks = 0;
1158 24297 : error = xfs_reserve_blocks(mp, &resblks, NULL);
1159 24297 : if (error)
1160 0 : xfs_warn(mp, "Unable to free reserved block pool. "
1161 : "Freespace may not be correct on next mount.");
1162 24297 : xfs_unmount_check(mp);
1163 :
1164 24297 : xfs_log_unmount(mp);
1165 24297 : xfs_da_unmount(mp);
1166 24297 : xfs_uuid_unmount(mp);
1167 :
1168 : #if defined(DEBUG)
1169 24297 : xfs_errortag_clearall(mp);
1170 : #endif
1171 24297 : unregister_shrinker(&mp->m_inodegc_shrinker);
1172 24297 : xfs_free_rtgroups(mp);
1173 24297 : xfs_free_perag(mp);
1174 :
1175 24297 : xfs_errortag_del(mp);
1176 24297 : xfs_error_sysfs_del(mp);
1177 24297 : xchk_stats_unregister(mp->m_scrub_stats);
1178 24297 : xfs_sysfs_del(&mp->m_stats.xs_kobj);
1179 24297 : xfs_sysfs_del(&mp->m_kobj);
1180 24297 : }
1181 :
1182 : /*
1183 : * Determine whether modifications can proceed. The caller specifies the minimum
1184 : * freeze level for which modifications should not be allowed. This allows
1185 : * certain operations to proceed while the freeze sequence is in progress, if
1186 : * necessary.
1187 : */
1188 : bool
1189 12085 : xfs_fs_writable(
1190 : struct xfs_mount *mp,
1191 : int level)
1192 : {
1193 12085 : ASSERT(level > SB_UNFROZEN);
1194 24145 : if ((mp->m_super->s_writers.frozen >= level) ||
1195 12060 : xfs_is_shutdown(mp) || xfs_is_readonly(mp))
1196 25 : return false;
1197 :
1198 : return true;
1199 : }
1200 :
1201 : /* Adjust m_fdblocks or m_frextents. */
1202 : int
1203 2265303188 : xfs_mod_freecounter(
1204 : struct xfs_mount *mp,
1205 : struct percpu_counter *counter,
1206 : int64_t delta,
1207 : bool rsvd)
1208 : {
1209 2265303188 : int64_t lcounter;
1210 2265303188 : long long res_used;
1211 2265303188 : uint64_t set_aside = 0;
1212 2265303188 : s32 batch;
1213 2265303188 : bool has_resv_pool;
1214 :
1215 2265303188 : ASSERT(counter == &mp->m_fdblocks || counter == &mp->m_frextents);
1216 2265303188 : has_resv_pool = (counter == &mp->m_fdblocks);
1217 2265303188 : if (rsvd)
1218 86683357 : ASSERT(has_resv_pool);
1219 :
1220 2265303188 : if (delta > 0) {
1221 : /*
1222 : * If the reserve pool is depleted, put blocks back into it
1223 : * first. Most of the time the pool is full.
1224 : */
1225 1136171537 : if (likely(!has_resv_pool ||
1226 : mp->m_resblks == mp->m_resblks_avail)) {
1227 1135925784 : percpu_counter_add(counter, delta);
1228 1135925784 : return 0;
1229 : }
1230 :
1231 245753 : spin_lock(&mp->m_sb_lock);
1232 245763 : res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1233 :
1234 245763 : if (res_used > delta) {
1235 53605 : mp->m_resblks_avail += delta;
1236 : } else {
1237 192158 : delta -= res_used;
1238 192158 : mp->m_resblks_avail = mp->m_resblks;
1239 192158 : percpu_counter_add(counter, delta);
1240 : }
1241 245763 : spin_unlock(&mp->m_sb_lock);
1242 245763 : return 0;
1243 : }
1244 :
1245 : /*
1246 : * Taking blocks away, need to be more accurate the closer we
1247 : * are to zero.
1248 : *
1249 : * If the counter has a value of less than 2 * max batch size,
1250 : * then make everything serialise as we are real close to
1251 : * ENOSPC.
1252 : */
1253 1129131651 : if (__percpu_counter_compare(counter, 2 * XFS_FDBLOCKS_BATCH,
1254 : XFS_FDBLOCKS_BATCH) < 0)
1255 : batch = 1;
1256 : else
1257 1112432923 : batch = XFS_FDBLOCKS_BATCH;
1258 :
1259 : /*
1260 : * Set aside allocbt blocks because these blocks are tracked as free
1261 : * space but not available for allocation. Technically this means that a
1262 : * single reservation cannot consume all remaining free space, but the
1263 : * ratio of allocbt blocks to usable free blocks should be rather small.
1264 : * The tradeoff without this is that filesystems that maintain high
1265 : * perag block reservations can over reserve physical block availability
1266 : * and fail physical allocation, which leads to much more serious
1267 : * problems (i.e. transaction abort, pagecache discards, etc.) than
1268 : * slightly premature -ENOSPC.
1269 : */
1270 1129186990 : if (has_resv_pool)
1271 1003683108 : set_aside = xfs_fdblocks_unavailable(mp);
1272 1129186990 : percpu_counter_add_batch(counter, delta, batch);
1273 1129212091 : if (__percpu_counter_compare(counter, set_aside,
1274 : XFS_FDBLOCKS_BATCH) >= 0) {
1275 : /* we had space! */
1276 : return 0;
1277 : }
1278 :
1279 : /*
1280 : * lock up the sb for dipping into reserves before releasing the space
1281 : * that took us to ENOSPC.
1282 : */
1283 9763640 : spin_lock(&mp->m_sb_lock);
1284 9766248 : percpu_counter_add(counter, -delta);
1285 9766248 : if (!has_resv_pool || !rsvd)
1286 9600754 : goto fdblocks_enospc;
1287 :
1288 165494 : lcounter = (long long)mp->m_resblks_avail + delta;
1289 165494 : if (lcounter >= 0) {
1290 161389 : mp->m_resblks_avail = lcounter;
1291 161389 : spin_unlock(&mp->m_sb_lock);
1292 161389 : return 0;
1293 : }
1294 4105 : xfs_warn_once(mp,
1295 : "Reserve blocks depleted! Consider increasing reserve pool size.");
1296 :
1297 9604859 : fdblocks_enospc:
1298 9604859 : spin_unlock(&mp->m_sb_lock);
1299 9604859 : return -ENOSPC;
1300 : }
1301 :
1302 : /*
1303 : * Used to free the superblock along various error paths.
1304 : */
1305 : void
1306 24365 : xfs_freesb(
1307 : struct xfs_mount *mp)
1308 : {
1309 24365 : struct xfs_buf *bp = mp->m_sb_bp;
1310 :
1311 24365 : xfs_buf_lock(bp);
1312 24365 : mp->m_sb_bp = NULL;
1313 24365 : xfs_buf_relse(bp);
1314 24365 : }
1315 :
1316 : /*
1317 : * If the underlying (data/log/rt) device is readonly, there are some
1318 : * operations that cannot proceed.
1319 : */
1320 : int
1321 14428 : xfs_dev_is_read_only(
1322 : struct xfs_mount *mp,
1323 : char *message)
1324 : {
1325 28850 : if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1326 14422 : xfs_readonly_buftarg(mp->m_logdev_targp) ||
1327 14422 : (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1328 6 : xfs_notice(mp, "%s required on read-only device.", message);
1329 6 : xfs_notice(mp, "write access unavailable, cannot proceed.");
1330 6 : return -EROFS;
1331 : }
1332 : return 0;
1333 : }
1334 :
1335 : /* Force the summary counters to be recalculated at next mount. */
1336 : void
1337 434579 : xfs_force_summary_recalc(
1338 : struct xfs_mount *mp)
1339 : {
1340 434579 : if (!xfs_has_lazysbcount(mp))
1341 : return;
1342 :
1343 434579 : xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1344 434579 : xfs_fs_mark_checked(mp, XFS_SICK_FS_COUNTERS);
1345 : }
1346 :
1347 : /*
1348 : * Enable a log incompat feature flag in the primary superblock. The caller
1349 : * cannot have any other transactions in progress.
1350 : */
1351 : int
1352 122311 : xfs_add_incompat_log_feature(
1353 : struct xfs_mount *mp,
1354 : uint32_t feature)
1355 : {
1356 122311 : struct xfs_dsb *dsb;
1357 122311 : int error;
1358 :
1359 244622 : ASSERT(hweight32(feature) == 1);
1360 122311 : ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
1361 :
1362 : /*
1363 : * Force the log to disk and kick the background AIL thread to reduce
1364 : * the chances that the bwrite will stall waiting for the AIL to unpin
1365 : * the primary superblock buffer. This isn't a data integrity
1366 : * operation, so we don't need a synchronous push.
1367 : */
1368 122311 : error = xfs_log_force(mp, XFS_LOG_SYNC);
1369 122308 : if (error)
1370 : return error;
1371 122297 : xfs_ail_push_all(mp->m_ail);
1372 :
1373 : /*
1374 : * Lock the primary superblock buffer to serialize all callers that
1375 : * are trying to set feature bits.
1376 : */
1377 122320 : xfs_buf_lock(mp->m_sb_bp);
1378 122328 : xfs_buf_hold(mp->m_sb_bp);
1379 :
1380 244656 : if (xfs_is_shutdown(mp)) {
1381 0 : error = -EIO;
1382 0 : goto rele;
1383 : }
1384 :
1385 122328 : if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature))
1386 52610 : goto rele;
1387 :
1388 : /*
1389 : * Write the primary superblock to disk immediately, because we need
1390 : * the log_incompat bit to be set in the primary super now to protect
1391 : * the log items that we're going to commit later.
1392 : */
1393 69718 : dsb = mp->m_sb_bp->b_addr;
1394 69718 : xfs_sb_to_disk(dsb, &mp->m_sb);
1395 69718 : dsb->sb_features_log_incompat |= cpu_to_be32(feature);
1396 69718 : error = xfs_bwrite(mp->m_sb_bp);
1397 69718 : if (error)
1398 92 : goto shutdown;
1399 :
1400 : /*
1401 : * Add the feature bits to the incore superblock before we unlock the
1402 : * buffer.
1403 : */
1404 69626 : xfs_sb_add_incompat_log_features(&mp->m_sb, feature);
1405 69626 : xfs_buf_relse(mp->m_sb_bp);
1406 :
1407 : /* Log the superblock to disk. */
1408 69626 : return xfs_sync_sb(mp, false);
1409 : shutdown:
1410 92 : xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1411 52702 : rele:
1412 52702 : xfs_buf_relse(mp->m_sb_bp);
1413 52702 : return error;
1414 : }
1415 :
1416 : /*
1417 : * Clear all the log incompat flags from the superblock.
1418 : *
1419 : * The caller cannot be in a transaction, must ensure that the log does not
1420 : * contain any log items protected by any log incompat bit, and must ensure
1421 : * that there are no other threads that depend on the state of the log incompat
1422 : * feature flags in the primary super.
1423 : *
1424 : * Returns true if the superblock is dirty.
1425 : */
1426 : bool
1427 86701 : xfs_clear_incompat_log_features(
1428 : struct xfs_mount *mp,
1429 : uint32_t features)
1430 : {
1431 86701 : bool ret = false;
1432 :
1433 86701 : if (!xfs_has_crc(mp) ||
1434 70350 : !xfs_sb_has_incompat_log_feature(&mp->m_sb, features) ||
1435 : xfs_is_shutdown(mp))
1436 : return false;
1437 :
1438 : /*
1439 : * Update the incore superblock. We synchronize on the primary super
1440 : * buffer lock to be consistent with the add function, though at least
1441 : * in theory this shouldn't be necessary.
1442 : */
1443 60239 : xfs_buf_lock(mp->m_sb_bp);
1444 60239 : xfs_buf_hold(mp->m_sb_bp);
1445 :
1446 60239 : if (xfs_sb_has_incompat_log_feature(&mp->m_sb, features)) {
1447 60239 : xfs_sb_remove_incompat_log_features(&mp->m_sb, features);
1448 60239 : ret = true;
1449 : }
1450 :
1451 60239 : xfs_buf_relse(mp->m_sb_bp);
1452 60239 : return ret;
1453 : }
1454 :
1455 : /*
1456 : * Update the in-core delayed block counter.
1457 : *
1458 : * We prefer to update the counter without having to take a spinlock for every
1459 : * counter update (i.e. batching). Each change to delayed allocation
1460 : * reservations can change can easily exceed the default percpu counter
1461 : * batching, so we use a larger batch factor here.
1462 : *
1463 : * Note that we don't currently have any callers requiring fast summation
1464 : * (e.g. percpu_counter_read) so we can use a big batch value here.
1465 : */
1466 : #define XFS_DELALLOC_BATCH (4096)
1467 : void
1468 57077947 : xfs_mod_delalloc(
1469 : struct xfs_mount *mp,
1470 : int64_t delta)
1471 : {
1472 57077947 : percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1473 : XFS_DELALLOC_BATCH);
1474 57078078 : }
|
