xfs: use generic percpu counters for free block counter
[deliverable/linux.git] / fs / xfs / xfs_mount.c
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_dir2.h"
31 #include "xfs_ialloc.h"
32 #include "xfs_alloc.h"
33 #include "xfs_rtalloc.h"
34 #include "xfs_bmap.h"
35 #include "xfs_trans.h"
36 #include "xfs_trans_priv.h"
37 #include "xfs_log.h"
38 #include "xfs_error.h"
39 #include "xfs_quota.h"
40 #include "xfs_fsops.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
43 #include "xfs_sysfs.h"
44
45
46 #ifdef HAVE_PERCPU_SB
47 STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
48 int);
49 STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
50 int);
51 STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
52 #else
53
54 #define xfs_icsb_balance_counter(mp, a, b) do { } while (0)
55 #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0)
56 #endif
57
58 static DEFINE_MUTEX(xfs_uuid_table_mutex);
59 static int xfs_uuid_table_size;
60 static uuid_t *xfs_uuid_table;
61
62 /*
63 * See if the UUID is unique among mounted XFS filesystems.
64 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
65 */
66 STATIC int
67 xfs_uuid_mount(
68 struct xfs_mount *mp)
69 {
70 uuid_t *uuid = &mp->m_sb.sb_uuid;
71 int hole, i;
72
73 if (mp->m_flags & XFS_MOUNT_NOUUID)
74 return 0;
75
76 if (uuid_is_nil(uuid)) {
77 xfs_warn(mp, "Filesystem has nil UUID - can't mount");
78 return -EINVAL;
79 }
80
81 mutex_lock(&xfs_uuid_table_mutex);
82 for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
83 if (uuid_is_nil(&xfs_uuid_table[i])) {
84 hole = i;
85 continue;
86 }
87 if (uuid_equal(uuid, &xfs_uuid_table[i]))
88 goto out_duplicate;
89 }
90
91 if (hole < 0) {
92 xfs_uuid_table = kmem_realloc(xfs_uuid_table,
93 (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
94 xfs_uuid_table_size * sizeof(*xfs_uuid_table),
95 KM_SLEEP);
96 hole = xfs_uuid_table_size++;
97 }
98 xfs_uuid_table[hole] = *uuid;
99 mutex_unlock(&xfs_uuid_table_mutex);
100
101 return 0;
102
103 out_duplicate:
104 mutex_unlock(&xfs_uuid_table_mutex);
105 xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
106 return -EINVAL;
107 }
108
109 STATIC void
110 xfs_uuid_unmount(
111 struct xfs_mount *mp)
112 {
113 uuid_t *uuid = &mp->m_sb.sb_uuid;
114 int i;
115
116 if (mp->m_flags & XFS_MOUNT_NOUUID)
117 return;
118
119 mutex_lock(&xfs_uuid_table_mutex);
120 for (i = 0; i < xfs_uuid_table_size; i++) {
121 if (uuid_is_nil(&xfs_uuid_table[i]))
122 continue;
123 if (!uuid_equal(uuid, &xfs_uuid_table[i]))
124 continue;
125 memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
126 break;
127 }
128 ASSERT(i < xfs_uuid_table_size);
129 mutex_unlock(&xfs_uuid_table_mutex);
130 }
131
132
133 STATIC void
134 __xfs_free_perag(
135 struct rcu_head *head)
136 {
137 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
138
139 ASSERT(atomic_read(&pag->pag_ref) == 0);
140 kmem_free(pag);
141 }
142
143 /*
144 * Free up the per-ag resources associated with the mount structure.
145 */
146 STATIC void
147 xfs_free_perag(
148 xfs_mount_t *mp)
149 {
150 xfs_agnumber_t agno;
151 struct xfs_perag *pag;
152
153 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
154 spin_lock(&mp->m_perag_lock);
155 pag = radix_tree_delete(&mp->m_perag_tree, agno);
156 spin_unlock(&mp->m_perag_lock);
157 ASSERT(pag);
158 ASSERT(atomic_read(&pag->pag_ref) == 0);
159 call_rcu(&pag->rcu_head, __xfs_free_perag);
160 }
161 }
162
163 /*
164 * Check size of device based on the (data/realtime) block count.
165 * Note: this check is used by the growfs code as well as mount.
166 */
167 int
168 xfs_sb_validate_fsb_count(
169 xfs_sb_t *sbp,
170 __uint64_t nblocks)
171 {
172 ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
173 ASSERT(sbp->sb_blocklog >= BBSHIFT);
174
175 /* Limited by ULONG_MAX of page cache index */
176 if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
177 return -EFBIG;
178 return 0;
179 }
180
181 int
182 xfs_initialize_perag(
183 xfs_mount_t *mp,
184 xfs_agnumber_t agcount,
185 xfs_agnumber_t *maxagi)
186 {
187 xfs_agnumber_t index;
188 xfs_agnumber_t first_initialised = 0;
189 xfs_perag_t *pag;
190 xfs_agino_t agino;
191 xfs_ino_t ino;
192 xfs_sb_t *sbp = &mp->m_sb;
193 int error = -ENOMEM;
194
195 /*
196 * Walk the current per-ag tree so we don't try to initialise AGs
197 * that already exist (growfs case). Allocate and insert all the
198 * AGs we don't find ready for initialisation.
199 */
200 for (index = 0; index < agcount; index++) {
201 pag = xfs_perag_get(mp, index);
202 if (pag) {
203 xfs_perag_put(pag);
204 continue;
205 }
206 if (!first_initialised)
207 first_initialised = index;
208
209 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
210 if (!pag)
211 goto out_unwind;
212 pag->pag_agno = index;
213 pag->pag_mount = mp;
214 spin_lock_init(&pag->pag_ici_lock);
215 mutex_init(&pag->pag_ici_reclaim_lock);
216 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
217 spin_lock_init(&pag->pag_buf_lock);
218 pag->pag_buf_tree = RB_ROOT;
219
220 if (radix_tree_preload(GFP_NOFS))
221 goto out_unwind;
222
223 spin_lock(&mp->m_perag_lock);
224 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
225 BUG();
226 spin_unlock(&mp->m_perag_lock);
227 radix_tree_preload_end();
228 error = -EEXIST;
229 goto out_unwind;
230 }
231 spin_unlock(&mp->m_perag_lock);
232 radix_tree_preload_end();
233 }
234
235 /*
236 * If we mount with the inode64 option, or no inode overflows
237 * the legacy 32-bit address space clear the inode32 option.
238 */
239 agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
240 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
241
242 if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
243 mp->m_flags |= XFS_MOUNT_32BITINODES;
244 else
245 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
246
247 if (mp->m_flags & XFS_MOUNT_32BITINODES)
248 index = xfs_set_inode32(mp, agcount);
249 else
250 index = xfs_set_inode64(mp, agcount);
251
252 if (maxagi)
253 *maxagi = index;
254 return 0;
255
256 out_unwind:
257 kmem_free(pag);
258 for (; index > first_initialised; index--) {
259 pag = radix_tree_delete(&mp->m_perag_tree, index);
260 kmem_free(pag);
261 }
262 return error;
263 }
264
265 /*
266 * xfs_readsb
267 *
268 * Does the initial read of the superblock.
269 */
270 int
271 xfs_readsb(
272 struct xfs_mount *mp,
273 int flags)
274 {
275 unsigned int sector_size;
276 struct xfs_buf *bp;
277 struct xfs_sb *sbp = &mp->m_sb;
278 int error;
279 int loud = !(flags & XFS_MFSI_QUIET);
280 const struct xfs_buf_ops *buf_ops;
281
282 ASSERT(mp->m_sb_bp == NULL);
283 ASSERT(mp->m_ddev_targp != NULL);
284
285 /*
286 * For the initial read, we must guess at the sector
287 * size based on the block device. It's enough to
288 * get the sb_sectsize out of the superblock and
289 * then reread with the proper length.
290 * We don't verify it yet, because it may not be complete.
291 */
292 sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
293 buf_ops = NULL;
294
295 /*
296 * Allocate a (locked) buffer to hold the superblock.
297 * This will be kept around at all times to optimize
298 * access to the superblock.
299 */
300 reread:
301 error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
302 BTOBB(sector_size), 0, &bp, buf_ops);
303 if (error) {
304 if (loud)
305 xfs_warn(mp, "SB validate failed with error %d.", error);
306 /* bad CRC means corrupted metadata */
307 if (error == -EFSBADCRC)
308 error = -EFSCORRUPTED;
309 return error;
310 }
311
312 /*
313 * Initialize the mount structure from the superblock.
314 */
315 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
316
317 /*
318 * If we haven't validated the superblock, do so now before we try
319 * to check the sector size and reread the superblock appropriately.
320 */
321 if (sbp->sb_magicnum != XFS_SB_MAGIC) {
322 if (loud)
323 xfs_warn(mp, "Invalid superblock magic number");
324 error = -EINVAL;
325 goto release_buf;
326 }
327
328 /*
329 * We must be able to do sector-sized and sector-aligned IO.
330 */
331 if (sector_size > sbp->sb_sectsize) {
332 if (loud)
333 xfs_warn(mp, "device supports %u byte sectors (not %u)",
334 sector_size, sbp->sb_sectsize);
335 error = -ENOSYS;
336 goto release_buf;
337 }
338
339 if (buf_ops == NULL) {
340 /*
341 * Re-read the superblock so the buffer is correctly sized,
342 * and properly verified.
343 */
344 xfs_buf_relse(bp);
345 sector_size = sbp->sb_sectsize;
346 buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
347 goto reread;
348 }
349
350 /* Initialize per-cpu counters */
351 xfs_icsb_reinit_counters(mp);
352
353 /* no need to be quiet anymore, so reset the buf ops */
354 bp->b_ops = &xfs_sb_buf_ops;
355
356 mp->m_sb_bp = bp;
357 xfs_buf_unlock(bp);
358 return 0;
359
360 release_buf:
361 xfs_buf_relse(bp);
362 return error;
363 }
364
365 /*
366 * Update alignment values based on mount options and sb values
367 */
368 STATIC int
369 xfs_update_alignment(xfs_mount_t *mp)
370 {
371 xfs_sb_t *sbp = &(mp->m_sb);
372
373 if (mp->m_dalign) {
374 /*
375 * If stripe unit and stripe width are not multiples
376 * of the fs blocksize turn off alignment.
377 */
378 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
379 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
380 xfs_warn(mp,
381 "alignment check failed: sunit/swidth vs. blocksize(%d)",
382 sbp->sb_blocksize);
383 return -EINVAL;
384 } else {
385 /*
386 * Convert the stripe unit and width to FSBs.
387 */
388 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
389 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
390 xfs_warn(mp,
391 "alignment check failed: sunit/swidth vs. agsize(%d)",
392 sbp->sb_agblocks);
393 return -EINVAL;
394 } else if (mp->m_dalign) {
395 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
396 } else {
397 xfs_warn(mp,
398 "alignment check failed: sunit(%d) less than bsize(%d)",
399 mp->m_dalign, sbp->sb_blocksize);
400 return -EINVAL;
401 }
402 }
403
404 /*
405 * Update superblock with new values
406 * and log changes
407 */
408 if (xfs_sb_version_hasdalign(sbp)) {
409 if (sbp->sb_unit != mp->m_dalign) {
410 sbp->sb_unit = mp->m_dalign;
411 mp->m_update_sb = true;
412 }
413 if (sbp->sb_width != mp->m_swidth) {
414 sbp->sb_width = mp->m_swidth;
415 mp->m_update_sb = true;
416 }
417 } else {
418 xfs_warn(mp,
419 "cannot change alignment: superblock does not support data alignment");
420 return -EINVAL;
421 }
422 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
423 xfs_sb_version_hasdalign(&mp->m_sb)) {
424 mp->m_dalign = sbp->sb_unit;
425 mp->m_swidth = sbp->sb_width;
426 }
427
428 return 0;
429 }
430
431 /*
432 * Set the maximum inode count for this filesystem
433 */
434 STATIC void
435 xfs_set_maxicount(xfs_mount_t *mp)
436 {
437 xfs_sb_t *sbp = &(mp->m_sb);
438 __uint64_t icount;
439
440 if (sbp->sb_imax_pct) {
441 /*
442 * Make sure the maximum inode count is a multiple
443 * of the units we allocate inodes in.
444 */
445 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
446 do_div(icount, 100);
447 do_div(icount, mp->m_ialloc_blks);
448 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
449 sbp->sb_inopblog;
450 } else {
451 mp->m_maxicount = 0;
452 }
453 }
454
455 /*
456 * Set the default minimum read and write sizes unless
457 * already specified in a mount option.
458 * We use smaller I/O sizes when the file system
459 * is being used for NFS service (wsync mount option).
460 */
461 STATIC void
462 xfs_set_rw_sizes(xfs_mount_t *mp)
463 {
464 xfs_sb_t *sbp = &(mp->m_sb);
465 int readio_log, writeio_log;
466
467 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
468 if (mp->m_flags & XFS_MOUNT_WSYNC) {
469 readio_log = XFS_WSYNC_READIO_LOG;
470 writeio_log = XFS_WSYNC_WRITEIO_LOG;
471 } else {
472 readio_log = XFS_READIO_LOG_LARGE;
473 writeio_log = XFS_WRITEIO_LOG_LARGE;
474 }
475 } else {
476 readio_log = mp->m_readio_log;
477 writeio_log = mp->m_writeio_log;
478 }
479
480 if (sbp->sb_blocklog > readio_log) {
481 mp->m_readio_log = sbp->sb_blocklog;
482 } else {
483 mp->m_readio_log = readio_log;
484 }
485 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
486 if (sbp->sb_blocklog > writeio_log) {
487 mp->m_writeio_log = sbp->sb_blocklog;
488 } else {
489 mp->m_writeio_log = writeio_log;
490 }
491 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
492 }
493
494 /*
495 * precalculate the low space thresholds for dynamic speculative preallocation.
496 */
497 void
498 xfs_set_low_space_thresholds(
499 struct xfs_mount *mp)
500 {
501 int i;
502
503 for (i = 0; i < XFS_LOWSP_MAX; i++) {
504 __uint64_t space = mp->m_sb.sb_dblocks;
505
506 do_div(space, 100);
507 mp->m_low_space[i] = space * (i + 1);
508 }
509 }
510
511
512 /*
513 * Set whether we're using inode alignment.
514 */
515 STATIC void
516 xfs_set_inoalignment(xfs_mount_t *mp)
517 {
518 if (xfs_sb_version_hasalign(&mp->m_sb) &&
519 mp->m_sb.sb_inoalignmt >=
520 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
521 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
522 else
523 mp->m_inoalign_mask = 0;
524 /*
525 * If we are using stripe alignment, check whether
526 * the stripe unit is a multiple of the inode alignment
527 */
528 if (mp->m_dalign && mp->m_inoalign_mask &&
529 !(mp->m_dalign & mp->m_inoalign_mask))
530 mp->m_sinoalign = mp->m_dalign;
531 else
532 mp->m_sinoalign = 0;
533 }
534
535 /*
536 * Check that the data (and log if separate) is an ok size.
537 */
538 STATIC int
539 xfs_check_sizes(
540 struct xfs_mount *mp)
541 {
542 struct xfs_buf *bp;
543 xfs_daddr_t d;
544 int error;
545
546 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
547 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
548 xfs_warn(mp, "filesystem size mismatch detected");
549 return -EFBIG;
550 }
551 error = xfs_buf_read_uncached(mp->m_ddev_targp,
552 d - XFS_FSS_TO_BB(mp, 1),
553 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
554 if (error) {
555 xfs_warn(mp, "last sector read failed");
556 return error;
557 }
558 xfs_buf_relse(bp);
559
560 if (mp->m_logdev_targp == mp->m_ddev_targp)
561 return 0;
562
563 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
564 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
565 xfs_warn(mp, "log size mismatch detected");
566 return -EFBIG;
567 }
568 error = xfs_buf_read_uncached(mp->m_logdev_targp,
569 d - XFS_FSB_TO_BB(mp, 1),
570 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
571 if (error) {
572 xfs_warn(mp, "log device read failed");
573 return error;
574 }
575 xfs_buf_relse(bp);
576 return 0;
577 }
578
579 /*
580 * Clear the quotaflags in memory and in the superblock.
581 */
582 int
583 xfs_mount_reset_sbqflags(
584 struct xfs_mount *mp)
585 {
586 mp->m_qflags = 0;
587
588 /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
589 if (mp->m_sb.sb_qflags == 0)
590 return 0;
591 spin_lock(&mp->m_sb_lock);
592 mp->m_sb.sb_qflags = 0;
593 spin_unlock(&mp->m_sb_lock);
594
595 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
596 return 0;
597
598 return xfs_sync_sb(mp, false);
599 }
600
601 __uint64_t
602 xfs_default_resblks(xfs_mount_t *mp)
603 {
604 __uint64_t resblks;
605
606 /*
607 * We default to 5% or 8192 fsbs of space reserved, whichever is
608 * smaller. This is intended to cover concurrent allocation
609 * transactions when we initially hit enospc. These each require a 4
610 * block reservation. Hence by default we cover roughly 2000 concurrent
611 * allocation reservations.
612 */
613 resblks = mp->m_sb.sb_dblocks;
614 do_div(resblks, 20);
615 resblks = min_t(__uint64_t, resblks, 8192);
616 return resblks;
617 }
618
619 /*
620 * This function does the following on an initial mount of a file system:
621 * - reads the superblock from disk and init the mount struct
622 * - if we're a 32-bit kernel, do a size check on the superblock
623 * so we don't mount terabyte filesystems
624 * - init mount struct realtime fields
625 * - allocate inode hash table for fs
626 * - init directory manager
627 * - perform recovery and init the log manager
628 */
629 int
630 xfs_mountfs(
631 xfs_mount_t *mp)
632 {
633 xfs_sb_t *sbp = &(mp->m_sb);
634 xfs_inode_t *rip;
635 __uint64_t resblks;
636 uint quotamount = 0;
637 uint quotaflags = 0;
638 int error = 0;
639
640 xfs_sb_mount_common(mp, sbp);
641
642 /*
643 * Check for a mismatched features2 values. Older kernels read & wrote
644 * into the wrong sb offset for sb_features2 on some platforms due to
645 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
646 * which made older superblock reading/writing routines swap it as a
647 * 64-bit value.
648 *
649 * For backwards compatibility, we make both slots equal.
650 *
651 * If we detect a mismatched field, we OR the set bits into the existing
652 * features2 field in case it has already been modified; we don't want
653 * to lose any features. We then update the bad location with the ORed
654 * value so that older kernels will see any features2 flags. The
655 * superblock writeback code ensures the new sb_features2 is copied to
656 * sb_bad_features2 before it is logged or written to disk.
657 */
658 if (xfs_sb_has_mismatched_features2(sbp)) {
659 xfs_warn(mp, "correcting sb_features alignment problem");
660 sbp->sb_features2 |= sbp->sb_bad_features2;
661 mp->m_update_sb = true;
662
663 /*
664 * Re-check for ATTR2 in case it was found in bad_features2
665 * slot.
666 */
667 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
668 !(mp->m_flags & XFS_MOUNT_NOATTR2))
669 mp->m_flags |= XFS_MOUNT_ATTR2;
670 }
671
672 if (xfs_sb_version_hasattr2(&mp->m_sb) &&
673 (mp->m_flags & XFS_MOUNT_NOATTR2)) {
674 xfs_sb_version_removeattr2(&mp->m_sb);
675 mp->m_update_sb = true;
676
677 /* update sb_versionnum for the clearing of the morebits */
678 if (!sbp->sb_features2)
679 mp->m_update_sb = true;
680 }
681
682 /* always use v2 inodes by default now */
683 if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
684 mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
685 mp->m_update_sb = true;
686 }
687
688 /*
689 * Check if sb_agblocks is aligned at stripe boundary
690 * If sb_agblocks is NOT aligned turn off m_dalign since
691 * allocator alignment is within an ag, therefore ag has
692 * to be aligned at stripe boundary.
693 */
694 error = xfs_update_alignment(mp);
695 if (error)
696 goto out;
697
698 xfs_alloc_compute_maxlevels(mp);
699 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
700 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
701 xfs_ialloc_compute_maxlevels(mp);
702
703 xfs_set_maxicount(mp);
704
705 error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
706 if (error)
707 goto out;
708
709 error = xfs_uuid_mount(mp);
710 if (error)
711 goto out_remove_sysfs;
712
713 /*
714 * Set the minimum read and write sizes
715 */
716 xfs_set_rw_sizes(mp);
717
718 /* set the low space thresholds for dynamic preallocation */
719 xfs_set_low_space_thresholds(mp);
720
721 /*
722 * Set the inode cluster size.
723 * This may still be overridden by the file system
724 * block size if it is larger than the chosen cluster size.
725 *
726 * For v5 filesystems, scale the cluster size with the inode size to
727 * keep a constant ratio of inode per cluster buffer, but only if mkfs
728 * has set the inode alignment value appropriately for larger cluster
729 * sizes.
730 */
731 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
732 if (xfs_sb_version_hascrc(&mp->m_sb)) {
733 int new_size = mp->m_inode_cluster_size;
734
735 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
736 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
737 mp->m_inode_cluster_size = new_size;
738 }
739
740 /*
741 * Set inode alignment fields
742 */
743 xfs_set_inoalignment(mp);
744
745 /*
746 * Check that the data (and log if separate) is an ok size.
747 */
748 error = xfs_check_sizes(mp);
749 if (error)
750 goto out_remove_uuid;
751
752 /*
753 * Initialize realtime fields in the mount structure
754 */
755 error = xfs_rtmount_init(mp);
756 if (error) {
757 xfs_warn(mp, "RT mount failed");
758 goto out_remove_uuid;
759 }
760
761 /*
762 * Copies the low order bits of the timestamp and the randomly
763 * set "sequence" number out of a UUID.
764 */
765 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
766
767 mp->m_dmevmask = 0; /* not persistent; set after each mount */
768
769 error = xfs_da_mount(mp);
770 if (error) {
771 xfs_warn(mp, "Failed dir/attr init: %d", error);
772 goto out_remove_uuid;
773 }
774
775 /*
776 * Initialize the precomputed transaction reservations values.
777 */
778 xfs_trans_init(mp);
779
780 /*
781 * Allocate and initialize the per-ag data.
782 */
783 spin_lock_init(&mp->m_perag_lock);
784 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
785 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
786 if (error) {
787 xfs_warn(mp, "Failed per-ag init: %d", error);
788 goto out_free_dir;
789 }
790
791 if (!sbp->sb_logblocks) {
792 xfs_warn(mp, "no log defined");
793 XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
794 error = -EFSCORRUPTED;
795 goto out_free_perag;
796 }
797
798 /*
799 * log's mount-time initialization. Perform 1st part recovery if needed
800 */
801 error = xfs_log_mount(mp, mp->m_logdev_targp,
802 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
803 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
804 if (error) {
805 xfs_warn(mp, "log mount failed");
806 goto out_fail_wait;
807 }
808
809 /*
810 * Now the log is mounted, we know if it was an unclean shutdown or
811 * not. If it was, with the first phase of recovery has completed, we
812 * have consistent AG blocks on disk. We have not recovered EFIs yet,
813 * but they are recovered transactionally in the second recovery phase
814 * later.
815 *
816 * Hence we can safely re-initialise incore superblock counters from
817 * the per-ag data. These may not be correct if the filesystem was not
818 * cleanly unmounted, so we need to wait for recovery to finish before
819 * doing this.
820 *
821 * If the filesystem was cleanly unmounted, then we can trust the
822 * values in the superblock to be correct and we don't need to do
823 * anything here.
824 *
825 * If we are currently making the filesystem, the initialisation will
826 * fail as the perag data is in an undefined state.
827 */
828 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
829 !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
830 !mp->m_sb.sb_inprogress) {
831 error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
832 if (error)
833 goto out_log_dealloc;
834 }
835
836 /*
837 * Get and sanity-check the root inode.
838 * Save the pointer to it in the mount structure.
839 */
840 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
841 if (error) {
842 xfs_warn(mp, "failed to read root inode");
843 goto out_log_dealloc;
844 }
845
846 ASSERT(rip != NULL);
847
848 if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
849 xfs_warn(mp, "corrupted root inode %llu: not a directory",
850 (unsigned long long)rip->i_ino);
851 xfs_iunlock(rip, XFS_ILOCK_EXCL);
852 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
853 mp);
854 error = -EFSCORRUPTED;
855 goto out_rele_rip;
856 }
857 mp->m_rootip = rip; /* save it */
858
859 xfs_iunlock(rip, XFS_ILOCK_EXCL);
860
861 /*
862 * Initialize realtime inode pointers in the mount structure
863 */
864 error = xfs_rtmount_inodes(mp);
865 if (error) {
866 /*
867 * Free up the root inode.
868 */
869 xfs_warn(mp, "failed to read RT inodes");
870 goto out_rele_rip;
871 }
872
873 /*
874 * If this is a read-only mount defer the superblock updates until
875 * the next remount into writeable mode. Otherwise we would never
876 * perform the update e.g. for the root filesystem.
877 */
878 if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
879 error = xfs_sync_sb(mp, false);
880 if (error) {
881 xfs_warn(mp, "failed to write sb changes");
882 goto out_rtunmount;
883 }
884 }
885
886 /*
887 * Initialise the XFS quota management subsystem for this mount
888 */
889 if (XFS_IS_QUOTA_RUNNING(mp)) {
890 error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
891 if (error)
892 goto out_rtunmount;
893 } else {
894 ASSERT(!XFS_IS_QUOTA_ON(mp));
895
896 /*
897 * If a file system had quotas running earlier, but decided to
898 * mount without -o uquota/pquota/gquota options, revoke the
899 * quotachecked license.
900 */
901 if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
902 xfs_notice(mp, "resetting quota flags");
903 error = xfs_mount_reset_sbqflags(mp);
904 if (error)
905 goto out_rtunmount;
906 }
907 }
908
909 /*
910 * Finish recovering the file system. This part needed to be
911 * delayed until after the root and real-time bitmap inodes
912 * were consistently read in.
913 */
914 error = xfs_log_mount_finish(mp);
915 if (error) {
916 xfs_warn(mp, "log mount finish failed");
917 goto out_rtunmount;
918 }
919
920 /*
921 * Complete the quota initialisation, post-log-replay component.
922 */
923 if (quotamount) {
924 ASSERT(mp->m_qflags == 0);
925 mp->m_qflags = quotaflags;
926
927 xfs_qm_mount_quotas(mp);
928 }
929
930 /*
931 * Now we are mounted, reserve a small amount of unused space for
932 * privileged transactions. This is needed so that transaction
933 * space required for critical operations can dip into this pool
934 * when at ENOSPC. This is needed for operations like create with
935 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
936 * are not allowed to use this reserved space.
937 *
938 * This may drive us straight to ENOSPC on mount, but that implies
939 * we were already there on the last unmount. Warn if this occurs.
940 */
941 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
942 resblks = xfs_default_resblks(mp);
943 error = xfs_reserve_blocks(mp, &resblks, NULL);
944 if (error)
945 xfs_warn(mp,
946 "Unable to allocate reserve blocks. Continuing without reserve pool.");
947 }
948
949 return 0;
950
951 out_rtunmount:
952 xfs_rtunmount_inodes(mp);
953 out_rele_rip:
954 IRELE(rip);
955 out_log_dealloc:
956 xfs_log_unmount(mp);
957 out_fail_wait:
958 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
959 xfs_wait_buftarg(mp->m_logdev_targp);
960 xfs_wait_buftarg(mp->m_ddev_targp);
961 out_free_perag:
962 xfs_free_perag(mp);
963 out_free_dir:
964 xfs_da_unmount(mp);
965 out_remove_uuid:
966 xfs_uuid_unmount(mp);
967 out_remove_sysfs:
968 xfs_sysfs_del(&mp->m_kobj);
969 out:
970 return error;
971 }
972
973 /*
974 * This flushes out the inodes,dquots and the superblock, unmounts the
975 * log and makes sure that incore structures are freed.
976 */
977 void
978 xfs_unmountfs(
979 struct xfs_mount *mp)
980 {
981 __uint64_t resblks;
982 int error;
983
984 cancel_delayed_work_sync(&mp->m_eofblocks_work);
985
986 xfs_qm_unmount_quotas(mp);
987 xfs_rtunmount_inodes(mp);
988 IRELE(mp->m_rootip);
989
990 /*
991 * We can potentially deadlock here if we have an inode cluster
992 * that has been freed has its buffer still pinned in memory because
993 * the transaction is still sitting in a iclog. The stale inodes
994 * on that buffer will have their flush locks held until the
995 * transaction hits the disk and the callbacks run. the inode
996 * flush takes the flush lock unconditionally and with nothing to
997 * push out the iclog we will never get that unlocked. hence we
998 * need to force the log first.
999 */
1000 xfs_log_force(mp, XFS_LOG_SYNC);
1001
1002 /*
1003 * Flush all pending changes from the AIL.
1004 */
1005 xfs_ail_push_all_sync(mp->m_ail);
1006
1007 /*
1008 * And reclaim all inodes. At this point there should be no dirty
1009 * inodes and none should be pinned or locked, but use synchronous
1010 * reclaim just to be sure. We can stop background inode reclaim
1011 * here as well if it is still running.
1012 */
1013 cancel_delayed_work_sync(&mp->m_reclaim_work);
1014 xfs_reclaim_inodes(mp, SYNC_WAIT);
1015
1016 xfs_qm_unmount(mp);
1017
1018 /*
1019 * Unreserve any blocks we have so that when we unmount we don't account
1020 * the reserved free space as used. This is really only necessary for
1021 * lazy superblock counting because it trusts the incore superblock
1022 * counters to be absolutely correct on clean unmount.
1023 *
1024 * We don't bother correcting this elsewhere for lazy superblock
1025 * counting because on mount of an unclean filesystem we reconstruct the
1026 * correct counter value and this is irrelevant.
1027 *
1028 * For non-lazy counter filesystems, this doesn't matter at all because
1029 * we only every apply deltas to the superblock and hence the incore
1030 * value does not matter....
1031 */
1032 resblks = 0;
1033 error = xfs_reserve_blocks(mp, &resblks, NULL);
1034 if (error)
1035 xfs_warn(mp, "Unable to free reserved block pool. "
1036 "Freespace may not be correct on next mount.");
1037
1038 error = xfs_log_sbcount(mp);
1039 if (error)
1040 xfs_warn(mp, "Unable to update superblock counters. "
1041 "Freespace may not be correct on next mount.");
1042
1043 xfs_log_unmount(mp);
1044 xfs_da_unmount(mp);
1045 xfs_uuid_unmount(mp);
1046
1047 #if defined(DEBUG)
1048 xfs_errortag_clearall(mp, 0);
1049 #endif
1050 xfs_free_perag(mp);
1051
1052 xfs_sysfs_del(&mp->m_kobj);
1053 }
1054
1055 /*
1056 * Determine whether modifications can proceed. The caller specifies the minimum
1057 * freeze level for which modifications should not be allowed. This allows
1058 * certain operations to proceed while the freeze sequence is in progress, if
1059 * necessary.
1060 */
1061 bool
1062 xfs_fs_writable(
1063 struct xfs_mount *mp,
1064 int level)
1065 {
1066 ASSERT(level > SB_UNFROZEN);
1067 if ((mp->m_super->s_writers.frozen >= level) ||
1068 XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1069 return false;
1070
1071 return true;
1072 }
1073
1074 /*
1075 * xfs_log_sbcount
1076 *
1077 * Sync the superblock counters to disk.
1078 *
1079 * Note this code can be called during the process of freezing, so we use the
1080 * transaction allocator that does not block when the transaction subsystem is
1081 * in its frozen state.
1082 */
1083 int
1084 xfs_log_sbcount(xfs_mount_t *mp)
1085 {
1086 /* allow this to proceed during the freeze sequence... */
1087 if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1088 return 0;
1089
1090 xfs_icsb_sync_counters(mp, 0);
1091
1092 /*
1093 * we don't need to do this if we are updating the superblock
1094 * counters on every modification.
1095 */
1096 if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1097 return 0;
1098
1099 return xfs_sync_sb(mp, true);
1100 }
1101
1102 int
1103 xfs_mod_icount(
1104 struct xfs_mount *mp,
1105 int64_t delta)
1106 {
1107 /* deltas are +/-64, hence the large batch size of 128. */
1108 __percpu_counter_add(&mp->m_icount, delta, 128);
1109 if (percpu_counter_compare(&mp->m_icount, 0) < 0) {
1110 ASSERT(0);
1111 percpu_counter_add(&mp->m_icount, -delta);
1112 return -EINVAL;
1113 }
1114 return 0;
1115 }
1116
1117 int
1118 xfs_mod_ifree(
1119 struct xfs_mount *mp,
1120 int64_t delta)
1121 {
1122 percpu_counter_add(&mp->m_ifree, delta);
1123 if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1124 ASSERT(0);
1125 percpu_counter_add(&mp->m_ifree, -delta);
1126 return -EINVAL;
1127 }
1128 return 0;
1129 }
1130
1131 int
1132 xfs_mod_fdblocks(
1133 struct xfs_mount *mp,
1134 int64_t delta,
1135 bool rsvd)
1136 {
1137 int64_t lcounter;
1138 long long res_used;
1139 s32 batch;
1140
1141 if (delta > 0) {
1142 /*
1143 * If the reserve pool is depleted, put blocks back into it
1144 * first. Most of the time the pool is full.
1145 */
1146 if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1147 percpu_counter_add(&mp->m_fdblocks, delta);
1148 return 0;
1149 }
1150
1151 spin_lock(&mp->m_sb_lock);
1152 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1153
1154 if (res_used > delta) {
1155 mp->m_resblks_avail += delta;
1156 } else {
1157 delta -= res_used;
1158 mp->m_resblks_avail = mp->m_resblks;
1159 percpu_counter_add(&mp->m_fdblocks, delta);
1160 }
1161 spin_unlock(&mp->m_sb_lock);
1162 return 0;
1163 }
1164
1165 /*
1166 * Taking blocks away, need to be more accurate the closer we
1167 * are to zero.
1168 *
1169 * batch size is set to a maximum of 1024 blocks - if we are
1170 * allocating of freeing extents larger than this then we aren't
1171 * going to be hammering the counter lock so a lock per update
1172 * is not a problem.
1173 *
1174 * If the counter has a value of less than 2 * max batch size,
1175 * then make everything serialise as we are real close to
1176 * ENOSPC.
1177 */
1178 #define __BATCH 1024
1179 if (percpu_counter_compare(&mp->m_fdblocks, 2 * __BATCH) < 0)
1180 batch = 1;
1181 else
1182 batch = __BATCH;
1183 #undef __BATCH
1184
1185 __percpu_counter_add(&mp->m_fdblocks, delta, batch);
1186 if (percpu_counter_compare(&mp->m_fdblocks,
1187 XFS_ALLOC_SET_ASIDE(mp)) >= 0) {
1188 /* we had space! */
1189 return 0;
1190 }
1191
1192 /*
1193 * lock up the sb for dipping into reserves before releasing the space
1194 * that took us to ENOSPC.
1195 */
1196 spin_lock(&mp->m_sb_lock);
1197 percpu_counter_add(&mp->m_fdblocks, -delta);
1198 if (!rsvd)
1199 goto fdblocks_enospc;
1200
1201 lcounter = (long long)mp->m_resblks_avail + delta;
1202 if (lcounter >= 0) {
1203 mp->m_resblks_avail = lcounter;
1204 spin_unlock(&mp->m_sb_lock);
1205 return 0;
1206 }
1207 printk_once(KERN_WARNING
1208 "Filesystem \"%s\": reserve blocks depleted! "
1209 "Consider increasing reserve pool size.",
1210 mp->m_fsname);
1211 fdblocks_enospc:
1212 spin_unlock(&mp->m_sb_lock);
1213 return -ENOSPC;
1214 }
1215
1216 /*
1217 * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply
1218 * a delta to a specified field in the in-core superblock. Simply
1219 * switch on the field indicated and apply the delta to that field.
1220 * Fields are not allowed to dip below zero, so if the delta would
1221 * do this do not apply it and return EINVAL.
1222 *
1223 * The m_sb_lock must be held when this routine is called.
1224 */
1225 STATIC int
1226 xfs_mod_incore_sb_unlocked(
1227 xfs_mount_t *mp,
1228 xfs_sb_field_t field,
1229 int64_t delta,
1230 int rsvd)
1231 {
1232 int scounter; /* short counter for 32 bit fields */
1233 long long lcounter; /* long counter for 64 bit fields */
1234
1235 /*
1236 * With the in-core superblock spin lock held, switch
1237 * on the indicated field. Apply the delta to the
1238 * proper field. If the fields value would dip below
1239 * 0, then do not apply the delta and return EINVAL.
1240 */
1241 switch (field) {
1242 case XFS_SBS_ICOUNT:
1243 case XFS_SBS_IFREE:
1244 case XFS_SBS_FDBLOCKS:
1245 ASSERT(0);
1246 return -EINVAL;
1247 case XFS_SBS_FREXTENTS:
1248 lcounter = (long long)mp->m_sb.sb_frextents;
1249 lcounter += delta;
1250 if (lcounter < 0) {
1251 return -ENOSPC;
1252 }
1253 mp->m_sb.sb_frextents = lcounter;
1254 return 0;
1255 case XFS_SBS_DBLOCKS:
1256 lcounter = (long long)mp->m_sb.sb_dblocks;
1257 lcounter += delta;
1258 if (lcounter < 0) {
1259 ASSERT(0);
1260 return -EINVAL;
1261 }
1262 mp->m_sb.sb_dblocks = lcounter;
1263 return 0;
1264 case XFS_SBS_AGCOUNT:
1265 scounter = mp->m_sb.sb_agcount;
1266 scounter += delta;
1267 if (scounter < 0) {
1268 ASSERT(0);
1269 return -EINVAL;
1270 }
1271 mp->m_sb.sb_agcount = scounter;
1272 return 0;
1273 case XFS_SBS_IMAX_PCT:
1274 scounter = mp->m_sb.sb_imax_pct;
1275 scounter += delta;
1276 if (scounter < 0) {
1277 ASSERT(0);
1278 return -EINVAL;
1279 }
1280 mp->m_sb.sb_imax_pct = scounter;
1281 return 0;
1282 case XFS_SBS_REXTSIZE:
1283 scounter = mp->m_sb.sb_rextsize;
1284 scounter += delta;
1285 if (scounter < 0) {
1286 ASSERT(0);
1287 return -EINVAL;
1288 }
1289 mp->m_sb.sb_rextsize = scounter;
1290 return 0;
1291 case XFS_SBS_RBMBLOCKS:
1292 scounter = mp->m_sb.sb_rbmblocks;
1293 scounter += delta;
1294 if (scounter < 0) {
1295 ASSERT(0);
1296 return -EINVAL;
1297 }
1298 mp->m_sb.sb_rbmblocks = scounter;
1299 return 0;
1300 case XFS_SBS_RBLOCKS:
1301 lcounter = (long long)mp->m_sb.sb_rblocks;
1302 lcounter += delta;
1303 if (lcounter < 0) {
1304 ASSERT(0);
1305 return -EINVAL;
1306 }
1307 mp->m_sb.sb_rblocks = lcounter;
1308 return 0;
1309 case XFS_SBS_REXTENTS:
1310 lcounter = (long long)mp->m_sb.sb_rextents;
1311 lcounter += delta;
1312 if (lcounter < 0) {
1313 ASSERT(0);
1314 return -EINVAL;
1315 }
1316 mp->m_sb.sb_rextents = lcounter;
1317 return 0;
1318 case XFS_SBS_REXTSLOG:
1319 scounter = mp->m_sb.sb_rextslog;
1320 scounter += delta;
1321 if (scounter < 0) {
1322 ASSERT(0);
1323 return -EINVAL;
1324 }
1325 mp->m_sb.sb_rextslog = scounter;
1326 return 0;
1327 default:
1328 ASSERT(0);
1329 return -EINVAL;
1330 }
1331 }
1332
1333 /*
1334 * xfs_mod_incore_sb() is used to change a field in the in-core
1335 * superblock structure by the specified delta. This modification
1336 * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked()
1337 * routine to do the work.
1338 */
1339 int
1340 xfs_mod_incore_sb(
1341 struct xfs_mount *mp,
1342 xfs_sb_field_t field,
1343 int64_t delta,
1344 int rsvd)
1345 {
1346 int status;
1347
1348 #ifdef HAVE_PERCPU_SB
1349 ASSERT(field < XFS_SBS_IFREE || field > XFS_SBS_FDBLOCKS);
1350 #endif
1351
1352 spin_lock(&mp->m_sb_lock);
1353 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1354 spin_unlock(&mp->m_sb_lock);
1355
1356 return status;
1357 }
1358
1359 /*
1360 * Change more than one field in the in-core superblock structure at a time.
1361 *
1362 * The fields and changes to those fields are specified in the array of
1363 * xfs_mod_sb structures passed in. Either all of the specified deltas
1364 * will be applied or none of them will. If any modified field dips below 0,
1365 * then all modifications will be backed out and EINVAL will be returned.
1366 *
1367 * Note that this function may not be used for the superblock values that
1368 * are tracked with the in-memory per-cpu counters - a direct call to
1369 * xfs_mod_incore_sb is required for these.
1370 */
1371 int
1372 xfs_mod_incore_sb_batch(
1373 struct xfs_mount *mp,
1374 xfs_mod_sb_t *msb,
1375 uint nmsb,
1376 int rsvd)
1377 {
1378 xfs_mod_sb_t *msbp;
1379 int error = 0;
1380
1381 /*
1382 * Loop through the array of mod structures and apply each individually.
1383 * If any fail, then back out all those which have already been applied.
1384 * Do all of this within the scope of the m_sb_lock so that all of the
1385 * changes will be atomic.
1386 */
1387 spin_lock(&mp->m_sb_lock);
1388 for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1389 ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1390 msbp->msb_field > XFS_SBS_FDBLOCKS);
1391
1392 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1393 msbp->msb_delta, rsvd);
1394 if (error)
1395 goto unwind;
1396 }
1397 spin_unlock(&mp->m_sb_lock);
1398 return 0;
1399
1400 unwind:
1401 while (--msbp >= msb) {
1402 error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1403 -msbp->msb_delta, rsvd);
1404 ASSERT(error == 0);
1405 }
1406 spin_unlock(&mp->m_sb_lock);
1407 return error;
1408 }
1409
1410 /*
1411 * xfs_getsb() is called to obtain the buffer for the superblock.
1412 * The buffer is returned locked and read in from disk.
1413 * The buffer should be released with a call to xfs_brelse().
1414 *
1415 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1416 * the superblock buffer if it can be locked without sleeping.
1417 * If it can't then we'll return NULL.
1418 */
1419 struct xfs_buf *
1420 xfs_getsb(
1421 struct xfs_mount *mp,
1422 int flags)
1423 {
1424 struct xfs_buf *bp = mp->m_sb_bp;
1425
1426 if (!xfs_buf_trylock(bp)) {
1427 if (flags & XBF_TRYLOCK)
1428 return NULL;
1429 xfs_buf_lock(bp);
1430 }
1431
1432 xfs_buf_hold(bp);
1433 ASSERT(XFS_BUF_ISDONE(bp));
1434 return bp;
1435 }
1436
1437 /*
1438 * Used to free the superblock along various error paths.
1439 */
1440 void
1441 xfs_freesb(
1442 struct xfs_mount *mp)
1443 {
1444 struct xfs_buf *bp = mp->m_sb_bp;
1445
1446 xfs_buf_lock(bp);
1447 mp->m_sb_bp = NULL;
1448 xfs_buf_relse(bp);
1449 }
1450
1451 /*
1452 * If the underlying (data/log/rt) device is readonly, there are some
1453 * operations that cannot proceed.
1454 */
1455 int
1456 xfs_dev_is_read_only(
1457 struct xfs_mount *mp,
1458 char *message)
1459 {
1460 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1461 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1462 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1463 xfs_notice(mp, "%s required on read-only device.", message);
1464 xfs_notice(mp, "write access unavailable, cannot proceed.");
1465 return -EROFS;
1466 }
1467 return 0;
1468 }
1469
1470 #ifdef HAVE_PERCPU_SB
1471 /*
1472 * Per-cpu incore superblock counters
1473 *
1474 * Simple concept, difficult implementation
1475 *
1476 * Basically, replace the incore superblock counters with a distributed per cpu
1477 * counter for contended fields (e.g. free block count).
1478 *
1479 * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1480 * hence needs to be accurately read when we are running low on space. Hence
1481 * there is a method to enable and disable the per-cpu counters based on how
1482 * much "stuff" is available in them.
1483 *
1484 * Basically, a counter is enabled if there is enough free resource to justify
1485 * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1486 * ENOSPC), then we disable the counters to synchronise all callers and
1487 * re-distribute the available resources.
1488 *
1489 * If, once we redistributed the available resources, we still get a failure,
1490 * we disable the per-cpu counter and go through the slow path.
1491 *
1492 * The slow path is the current xfs_mod_incore_sb() function. This means that
1493 * when we disable a per-cpu counter, we need to drain its resources back to
1494 * the global superblock. We do this after disabling the counter to prevent
1495 * more threads from queueing up on the counter.
1496 *
1497 * Essentially, this means that we still need a lock in the fast path to enable
1498 * synchronisation between the global counters and the per-cpu counters. This
1499 * is not a problem because the lock will be local to a CPU almost all the time
1500 * and have little contention except when we get to ENOSPC conditions.
1501 *
1502 * Basically, this lock becomes a barrier that enables us to lock out the fast
1503 * path while we do things like enabling and disabling counters and
1504 * synchronising the counters.
1505 *
1506 * Locking rules:
1507 *
1508 * 1. m_sb_lock before picking up per-cpu locks
1509 * 2. per-cpu locks always picked up via for_each_online_cpu() order
1510 * 3. accurate counter sync requires m_sb_lock + per cpu locks
1511 * 4. modifying per-cpu counters requires holding per-cpu lock
1512 * 5. modifying global counters requires holding m_sb_lock
1513 * 6. enabling or disabling a counter requires holding the m_sb_lock
1514 * and _none_ of the per-cpu locks.
1515 *
1516 * Disabled counters are only ever re-enabled by a balance operation
1517 * that results in more free resources per CPU than a given threshold.
1518 * To ensure counters don't remain disabled, they are rebalanced when
1519 * the global resource goes above a higher threshold (i.e. some hysteresis
1520 * is present to prevent thrashing).
1521 */
1522
1523 #ifdef CONFIG_HOTPLUG_CPU
1524 /*
1525 * hot-plug CPU notifier support.
1526 *
1527 * We need a notifier per filesystem as we need to be able to identify
1528 * the filesystem to balance the counters out. This is achieved by
1529 * having a notifier block embedded in the xfs_mount_t and doing pointer
1530 * magic to get the mount pointer from the notifier block address.
1531 */
1532 STATIC int
1533 xfs_icsb_cpu_notify(
1534 struct notifier_block *nfb,
1535 unsigned long action,
1536 void *hcpu)
1537 {
1538 xfs_icsb_cnts_t *cntp;
1539 xfs_mount_t *mp;
1540
1541 mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1542 cntp = (xfs_icsb_cnts_t *)
1543 per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1544 switch (action) {
1545 case CPU_UP_PREPARE:
1546 case CPU_UP_PREPARE_FROZEN:
1547 /* Easy Case - initialize the area and locks, and
1548 * then rebalance when online does everything else for us. */
1549 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1550 break;
1551 case CPU_ONLINE:
1552 case CPU_ONLINE_FROZEN:
1553 xfs_icsb_lock(mp);
1554 xfs_icsb_unlock(mp);
1555 break;
1556 case CPU_DEAD:
1557 case CPU_DEAD_FROZEN:
1558 /* Disable all the counters, then fold the dead cpu's
1559 * count into the total on the global superblock and
1560 * re-enable the counters. */
1561 xfs_icsb_lock(mp);
1562 spin_lock(&mp->m_sb_lock);
1563
1564 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1565
1566 spin_unlock(&mp->m_sb_lock);
1567 xfs_icsb_unlock(mp);
1568 break;
1569 }
1570
1571 return NOTIFY_OK;
1572 }
1573 #endif /* CONFIG_HOTPLUG_CPU */
1574
1575 int
1576 xfs_icsb_init_counters(
1577 xfs_mount_t *mp)
1578 {
1579 xfs_icsb_cnts_t *cntp;
1580 int error;
1581 int i;
1582
1583 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1584 if (error)
1585 return error;
1586
1587 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1588 if (error)
1589 goto free_icount;
1590
1591 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1592 if (error)
1593 goto free_ifree;
1594
1595 mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1596 if (!mp->m_sb_cnts) {
1597 error = -ENOMEM;
1598 goto free_fdblocks;
1599 }
1600
1601 for_each_online_cpu(i) {
1602 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1603 memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1604 }
1605
1606 mutex_init(&mp->m_icsb_mutex);
1607
1608 /*
1609 * start with all counters disabled so that the
1610 * initial balance kicks us off correctly
1611 */
1612 mp->m_icsb_counters = -1;
1613
1614 #ifdef CONFIG_HOTPLUG_CPU
1615 mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1616 mp->m_icsb_notifier.priority = 0;
1617 register_hotcpu_notifier(&mp->m_icsb_notifier);
1618 #endif /* CONFIG_HOTPLUG_CPU */
1619
1620 return 0;
1621
1622 free_fdblocks:
1623 percpu_counter_destroy(&mp->m_fdblocks);
1624 free_ifree:
1625 percpu_counter_destroy(&mp->m_ifree);
1626 free_icount:
1627 percpu_counter_destroy(&mp->m_icount);
1628 return error;
1629 }
1630
1631 void
1632 xfs_icsb_reinit_counters(
1633 xfs_mount_t *mp)
1634 {
1635 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1636 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1637 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1638
1639 xfs_icsb_lock(mp);
1640 /*
1641 * start with all counters disabled so that the
1642 * initial balance kicks us off correctly
1643 */
1644 mp->m_icsb_counters = -1;
1645 xfs_icsb_unlock(mp);
1646 }
1647
1648 void
1649 xfs_icsb_destroy_counters(
1650 xfs_mount_t *mp)
1651 {
1652 if (mp->m_sb_cnts) {
1653 unregister_hotcpu_notifier(&mp->m_icsb_notifier);
1654 free_percpu(mp->m_sb_cnts);
1655 }
1656
1657 percpu_counter_destroy(&mp->m_icount);
1658 percpu_counter_destroy(&mp->m_ifree);
1659 percpu_counter_destroy(&mp->m_fdblocks);
1660
1661 mutex_destroy(&mp->m_icsb_mutex);
1662 }
1663
1664 STATIC void
1665 xfs_icsb_lock_cntr(
1666 xfs_icsb_cnts_t *icsbp)
1667 {
1668 while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1669 ndelay(1000);
1670 }
1671 }
1672
1673 STATIC void
1674 xfs_icsb_unlock_cntr(
1675 xfs_icsb_cnts_t *icsbp)
1676 {
1677 clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1678 }
1679
1680
1681 STATIC void
1682 xfs_icsb_lock_all_counters(
1683 xfs_mount_t *mp)
1684 {
1685 xfs_icsb_cnts_t *cntp;
1686 int i;
1687
1688 for_each_online_cpu(i) {
1689 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1690 xfs_icsb_lock_cntr(cntp);
1691 }
1692 }
1693
1694 STATIC void
1695 xfs_icsb_unlock_all_counters(
1696 xfs_mount_t *mp)
1697 {
1698 xfs_icsb_cnts_t *cntp;
1699 int i;
1700
1701 for_each_online_cpu(i) {
1702 cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1703 xfs_icsb_unlock_cntr(cntp);
1704 }
1705 }
1706
1707 STATIC void
1708 xfs_icsb_count(
1709 xfs_mount_t *mp,
1710 xfs_icsb_cnts_t *cnt,
1711 int flags)
1712 {
1713 memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1714
1715 if (!(flags & XFS_ICSB_LAZY_COUNT))
1716 xfs_icsb_lock_all_counters(mp);
1717
1718
1719 if (!(flags & XFS_ICSB_LAZY_COUNT))
1720 xfs_icsb_unlock_all_counters(mp);
1721 }
1722
1723 STATIC int
1724 xfs_icsb_counter_disabled(
1725 xfs_mount_t *mp,
1726 xfs_sb_field_t field)
1727 {
1728 return test_bit(field, &mp->m_icsb_counters);
1729 }
1730
1731 STATIC void
1732 xfs_icsb_disable_counter(
1733 xfs_mount_t *mp,
1734 xfs_sb_field_t field)
1735 {
1736 xfs_icsb_cnts_t cnt;
1737
1738 /*
1739 * If we are already disabled, then there is nothing to do
1740 * here. We check before locking all the counters to avoid
1741 * the expensive lock operation when being called in the
1742 * slow path and the counter is already disabled. This is
1743 * safe because the only time we set or clear this state is under
1744 * the m_icsb_mutex.
1745 */
1746 if (xfs_icsb_counter_disabled(mp, field))
1747 return;
1748
1749 xfs_icsb_lock_all_counters(mp);
1750 if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1751 /* drain back to superblock */
1752
1753 xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
1754 switch(field) {
1755 default:
1756 BUG();
1757 }
1758 }
1759
1760 xfs_icsb_unlock_all_counters(mp);
1761 }
1762
1763 STATIC void
1764 xfs_icsb_enable_counter(
1765 xfs_mount_t *mp,
1766 xfs_sb_field_t field,
1767 uint64_t count,
1768 uint64_t resid)
1769 {
1770 int i;
1771
1772 xfs_icsb_lock_all_counters(mp);
1773 for_each_online_cpu(i) {
1774 switch (field) {
1775 default:
1776 BUG();
1777 break;
1778 }
1779 resid = 0;
1780 }
1781 clear_bit(field, &mp->m_icsb_counters);
1782 xfs_icsb_unlock_all_counters(mp);
1783 }
1784
1785 void
1786 xfs_icsb_sync_counters_locked(
1787 xfs_mount_t *mp,
1788 int flags)
1789 {
1790 xfs_icsb_cnts_t cnt;
1791
1792 xfs_icsb_count(mp, &cnt, flags);
1793 }
1794
1795 /*
1796 * Accurate update of per-cpu counters to incore superblock
1797 */
1798 void
1799 xfs_icsb_sync_counters(
1800 xfs_mount_t *mp,
1801 int flags)
1802 {
1803 spin_lock(&mp->m_sb_lock);
1804 xfs_icsb_sync_counters_locked(mp, flags);
1805 spin_unlock(&mp->m_sb_lock);
1806 }
1807
1808 /*
1809 * Balance and enable/disable counters as necessary.
1810 *
1811 * Thresholds for re-enabling counters are somewhat magic. inode counts are
1812 * chosen to be the same number as single on disk allocation chunk per CPU, and
1813 * free blocks is something far enough zero that we aren't going thrash when we
1814 * get near ENOSPC. We also need to supply a minimum we require per cpu to
1815 * prevent looping endlessly when xfs_alloc_space asks for more than will
1816 * be distributed to a single CPU but each CPU has enough blocks to be
1817 * reenabled.
1818 *
1819 * Note that we can be called when counters are already disabled.
1820 * xfs_icsb_disable_counter() optimises the counter locking in this case to
1821 * prevent locking every per-cpu counter needlessly.
1822 */
1823
1824 #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
1825 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
1826 (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
1827 STATIC void
1828 xfs_icsb_balance_counter_locked(
1829 xfs_mount_t *mp,
1830 xfs_sb_field_t field,
1831 int min_per_cpu)
1832 {
1833 uint64_t count, resid;
1834
1835 /* disable counter and sync counter */
1836 xfs_icsb_disable_counter(mp, field);
1837
1838 /* update counters - first CPU gets residual*/
1839 switch (field) {
1840 default:
1841 BUG();
1842 count = resid = 0; /* quiet, gcc */
1843 break;
1844 }
1845
1846 xfs_icsb_enable_counter(mp, field, count, resid);
1847 }
1848
1849 STATIC void
1850 xfs_icsb_balance_counter(
1851 xfs_mount_t *mp,
1852 xfs_sb_field_t fields,
1853 int min_per_cpu)
1854 {
1855 spin_lock(&mp->m_sb_lock);
1856 xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
1857 spin_unlock(&mp->m_sb_lock);
1858 }
1859
1860 int
1861 xfs_icsb_modify_counters(
1862 xfs_mount_t *mp,
1863 xfs_sb_field_t field,
1864 int64_t delta,
1865 int rsvd)
1866 {
1867 xfs_icsb_cnts_t *icsbp;
1868 int ret = 0;
1869
1870 might_sleep();
1871 again:
1872 preempt_disable();
1873 icsbp = this_cpu_ptr(mp->m_sb_cnts);
1874
1875 /*
1876 * if the counter is disabled, go to slow path
1877 */
1878 if (unlikely(xfs_icsb_counter_disabled(mp, field)))
1879 goto slow_path;
1880 xfs_icsb_lock_cntr(icsbp);
1881 if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
1882 xfs_icsb_unlock_cntr(icsbp);
1883 goto slow_path;
1884 }
1885
1886 switch (field) {
1887 default:
1888 BUG();
1889 goto balance_counter; /* be still, gcc */
1890 }
1891 xfs_icsb_unlock_cntr(icsbp);
1892 preempt_enable();
1893 return 0;
1894
1895 slow_path:
1896 preempt_enable();
1897
1898 /*
1899 * serialise with a mutex so we don't burn lots of cpu on
1900 * the superblock lock. We still need to hold the superblock
1901 * lock, however, when we modify the global structures.
1902 */
1903 xfs_icsb_lock(mp);
1904
1905 /*
1906 * Now running atomically.
1907 *
1908 * If the counter is enabled, someone has beaten us to rebalancing.
1909 * Drop the lock and try again in the fast path....
1910 */
1911 if (!(xfs_icsb_counter_disabled(mp, field))) {
1912 xfs_icsb_unlock(mp);
1913 goto again;
1914 }
1915
1916 /*
1917 * The counter is currently disabled. Because we are
1918 * running atomically here, we know a rebalance cannot
1919 * be in progress. Hence we can go straight to operating
1920 * on the global superblock. We do not call xfs_mod_incore_sb()
1921 * here even though we need to get the m_sb_lock. Doing so
1922 * will cause us to re-enter this function and deadlock.
1923 * Hence we get the m_sb_lock ourselves and then call
1924 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
1925 * directly on the global counters.
1926 */
1927 spin_lock(&mp->m_sb_lock);
1928 ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1929 spin_unlock(&mp->m_sb_lock);
1930
1931 /*
1932 * Now that we've modified the global superblock, we
1933 * may be able to re-enable the distributed counters
1934 * (e.g. lots of space just got freed). After that
1935 * we are done.
1936 */
1937 if (ret != -ENOSPC)
1938 xfs_icsb_balance_counter(mp, field, 0);
1939 xfs_icsb_unlock(mp);
1940 return ret;
1941
1942 balance_counter:
1943 xfs_icsb_unlock_cntr(icsbp);
1944 preempt_enable();
1945
1946 /*
1947 * We may have multiple threads here if multiple per-cpu
1948 * counters run dry at the same time. This will mean we can
1949 * do more balances than strictly necessary but it is not
1950 * the common slowpath case.
1951 */
1952 xfs_icsb_lock(mp);
1953
1954 /*
1955 * running atomically.
1956 *
1957 * This will leave the counter in the correct state for future
1958 * accesses. After the rebalance, we simply try again and our retry
1959 * will either succeed through the fast path or slow path without
1960 * another balance operation being required.
1961 */
1962 xfs_icsb_balance_counter(mp, field, delta);
1963 xfs_icsb_unlock(mp);
1964 goto again;
1965 }
1966
1967 #endif
This page took 0.082882 seconds and 5 git commands to generate.