1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_trans.h"
15 #include "xfs_error.h"
16 #include "xfs_alloc.h"
17 #include "xfs_fsops.h"
18 #include "xfs_trans_space.h"
20 #include "xfs_log_priv.h"
22 #include "xfs_ag_resv.h"
23 #include "xfs_trace.h"
26 * Write new AG headers to disk. Non-transactional, but need to be
27 * written and completed prior to the growfs transaction being logged.
28 * To do this, we use a delayed write buffer list and wait for
29 * submission and IO completion of the list as a whole. This allows the
30 * IO subsystem to merge all the AG headers in a single AG into a single
31 * IO and hide most of the latency of the IO from us.
33 * This also means that if we get an error whilst building the buffer
34 * list to write, we can cancel the entire list without having written
38 xfs_resizefs_init_new_ags(
40 struct aghdr_init_data *id,
41 xfs_agnumber_t oagcount,
42 xfs_agnumber_t nagcount,
44 bool *lastag_extended)
46 struct xfs_mount *mp = tp->t_mountp;
47 xfs_rfsblock_t nb = mp->m_sb.sb_dblocks + delta;
50 *lastag_extended = false;
52 INIT_LIST_HEAD(&id->buffer_list);
53 for (id->agno = nagcount - 1;
55 id->agno--, delta -= id->agsize) {
57 if (id->agno == nagcount - 1)
58 id->agsize = nb - (id->agno *
59 (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
61 id->agsize = mp->m_sb.sb_agblocks;
63 error = xfs_ag_init_headers(mp, id);
65 xfs_buf_delwri_cancel(&id->buffer_list);
70 error = xfs_buf_delwri_submit(&id->buffer_list);
75 *lastag_extended = true;
76 error = xfs_ag_extend_space(mp, tp, id, delta);
85 xfs_growfs_data_private(
86 struct xfs_mount *mp, /* mount point for filesystem */
87 struct xfs_growfs_data *in) /* growfs data input struct */
91 xfs_agnumber_t nagcount;
92 xfs_agnumber_t nagimax = 0;
93 xfs_rfsblock_t nb, nb_div, nb_mod;
96 xfs_agnumber_t oagcount;
98 struct aghdr_init_data id = {};
101 error = xfs_sb_validate_fsb_count(&mp->m_sb, nb);
105 if (nb > mp->m_sb.sb_dblocks) {
106 error = xfs_buf_read_uncached(mp->m_ddev_targp,
107 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
108 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
115 nb_mod = do_div(nb_div, mp->m_sb.sb_agblocks);
116 nagcount = nb_div + (nb_mod != 0);
117 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
119 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
121 delta = nb - mp->m_sb.sb_dblocks;
123 * Reject filesystems with a single AG because they are not
124 * supported, and reject a shrink operation that would cause a
125 * filesystem to become unsupported.
127 if (delta < 0 && nagcount < 2)
130 oagcount = mp->m_sb.sb_agcount;
132 /* allocate the new per-ag structures */
133 if (nagcount > oagcount) {
134 error = xfs_initialize_perag(mp, nagcount, &nagimax);
137 } else if (nagcount < oagcount) {
138 /* TODO: shrinking the entire AGs hasn't yet completed */
142 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
143 (delta > 0 ? XFS_GROWFS_SPACE_RES(mp) : -delta), 0,
144 XFS_TRANS_RESERVE, &tp);
149 error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount,
150 delta, &lastag_extended);
152 static struct ratelimit_state shrink_warning = \
153 RATELIMIT_STATE_INIT("shrink_warning", 86400 * HZ, 1);
154 ratelimit_set_flags(&shrink_warning, RATELIMIT_MSG_ON_RELEASE);
156 if (__ratelimit(&shrink_warning))
158 "EXPERIMENTAL online shrink feature in use. Use at your own risk!");
160 error = xfs_ag_shrink_space(mp, &tp, nagcount - 1, -delta);
163 goto out_trans_cancel;
166 * Update changed superblock fields transactionally. These are not
167 * seen by the rest of the world until the transaction commit applies
168 * them atomically to the superblock.
170 if (nagcount > oagcount)
171 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
173 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta);
175 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
178 * Sync sb counters now to reflect the updated values. This is
179 * particularly important for shrink because the write verifier
180 * will fail if sb_fdblocks is ever larger than sb_dblocks.
182 if (xfs_has_lazysbcount(mp))
185 xfs_trans_set_sync(tp);
186 error = xfs_trans_commit(tp);
190 /* New allocation groups fully initialized, so update mount struct */
192 mp->m_maxagi = nagimax;
193 xfs_set_low_space_thresholds(mp);
194 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
198 * If we expanded the last AG, free the per-AG reservation
199 * so we can reinitialize it with the new size.
201 if (lastag_extended) {
202 struct xfs_perag *pag;
204 pag = xfs_perag_get(mp, id.agno);
205 error = xfs_ag_resv_free(pag);
211 * Reserve AG metadata blocks. ENOSPC here does not mean there
212 * was a growfs failure, just that there still isn't space for
213 * new user data after the grow has been run.
215 error = xfs_fs_reserve_ag_blocks(mp);
216 if (error == -ENOSPC)
222 xfs_trans_cancel(tp);
227 xfs_growfs_log_private(
228 struct xfs_mount *mp, /* mount point for filesystem */
229 struct xfs_growfs_log *in) /* growfs log input struct */
234 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
236 if (nb == mp->m_sb.sb_logblocks &&
237 in->isint == (mp->m_sb.sb_logstart != 0))
240 * Moving the log is hard, need new interfaces to sync
241 * the log first, hold off all activity while moving it.
242 * Can have shorter or longer log in the same space,
243 * or transform internal to external log or vice versa.
250 struct xfs_mount *mp,
253 struct xfs_trans *tp;
260 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
261 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
265 dpct = imaxpct - mp->m_sb.sb_imax_pct;
266 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
267 xfs_trans_set_sync(tp);
268 return xfs_trans_commit(tp);
272 * protected versions of growfs function acquire and release locks on the mount
273 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
278 struct xfs_mount *mp,
279 struct xfs_growfs_data *in)
283 if (!capable(CAP_SYS_ADMIN))
285 if (!mutex_trylock(&mp->m_growlock))
288 /* update imaxpct separately to the physical grow of the filesystem */
289 if (in->imaxpct != mp->m_sb.sb_imax_pct) {
290 error = xfs_growfs_imaxpct(mp, in->imaxpct);
295 if (in->newblocks != mp->m_sb.sb_dblocks) {
296 error = xfs_growfs_data_private(mp, in);
301 /* Post growfs calculations needed to reflect new state in operations */
302 if (mp->m_sb.sb_imax_pct) {
303 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
305 M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount);
307 M_IGEO(mp)->maxicount = 0;
309 /* Update secondary superblocks now the physical grow has completed */
310 error = xfs_update_secondary_sbs(mp);
314 * Increment the generation unconditionally, the error could be from
315 * updating the secondary superblocks, in which case the new size
319 mutex_unlock(&mp->m_growlock);
326 struct xfs_growfs_log *in)
330 if (!capable(CAP_SYS_ADMIN))
332 if (!mutex_trylock(&mp->m_growlock))
334 error = xfs_growfs_log_private(mp, in);
335 mutex_unlock(&mp->m_growlock);
340 * exported through ioctl XFS_IOC_FSCOUNTS
346 xfs_fsop_counts_t *cnt)
348 cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
349 cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
350 cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
351 xfs_fdblocks_unavailable(mp);
353 spin_lock(&mp->m_sb_lock);
354 cnt->freertx = mp->m_sb.sb_frextents;
355 spin_unlock(&mp->m_sb_lock);
359 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
361 * xfs_reserve_blocks is called to set m_resblks
362 * in the in-core mount table. The number of unused reserved blocks
363 * is kept in m_resblks_avail.
365 * Reserve the requested number of blocks if available. Otherwise return
366 * as many as possible to satisfy the request. The actual number
367 * reserved are returned in outval
369 * A null inval pointer indicates that only the current reserved blocks
370 * available should be returned no settings are changed.
377 xfs_fsop_resblks_t *outval)
379 int64_t lcounter, delta;
380 int64_t fdblks_delta = 0;
385 /* If inval is null, report current values and return */
386 if (inval == (uint64_t *)NULL) {
389 outval->resblks = mp->m_resblks;
390 outval->resblks_avail = mp->m_resblks_avail;
397 * With per-cpu counters, this becomes an interesting problem. we need
398 * to work out if we are freeing or allocation blocks first, then we can
399 * do the modification as necessary.
401 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
402 * hold out any changes while we work out what to do. This means that
403 * the amount of free space can change while we do this, so we need to
404 * retry if we end up trying to reserve more space than is available.
406 spin_lock(&mp->m_sb_lock);
409 * If our previous reservation was larger than the current value,
410 * then move any unused blocks back to the free pool. Modify the resblks
411 * counters directly since we shouldn't have any problems unreserving
414 if (mp->m_resblks > request) {
415 lcounter = mp->m_resblks_avail - request;
416 if (lcounter > 0) { /* release unused blocks */
417 fdblks_delta = lcounter;
418 mp->m_resblks_avail -= lcounter;
420 mp->m_resblks = request;
422 spin_unlock(&mp->m_sb_lock);
423 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
424 spin_lock(&mp->m_sb_lock);
431 * If the request is larger than the current reservation, reserve the
432 * blocks before we update the reserve counters. Sample m_fdblocks and
433 * perform a partial reservation if the request exceeds free space.
435 * The code below estimates how many blocks it can request from
436 * fdblocks to stash in the reserve pool. This is a classic TOCTOU
437 * race since fdblocks updates are not always coordinated via
438 * m_sb_lock. Set the reserve size even if there's not enough free
439 * space to fill it because mod_fdblocks will refill an undersized
440 * reserve when it can.
442 free = percpu_counter_sum(&mp->m_fdblocks) -
443 xfs_fdblocks_unavailable(mp);
444 delta = request - mp->m_resblks;
445 mp->m_resblks = request;
446 if (delta > 0 && free > 0) {
448 * We'll either succeed in getting space from the free block
449 * count or we'll get an ENOSPC. Don't set the reserved flag
450 * here - we don't want to reserve the extra reserve blocks
453 * The desired reserve size can change after we drop the lock.
454 * Use mod_fdblocks to put the space into the reserve or into
455 * fdblocks as appropriate.
457 fdblks_delta = min(free, delta);
458 spin_unlock(&mp->m_sb_lock);
459 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
461 xfs_mod_fdblocks(mp, fdblks_delta, 0);
462 spin_lock(&mp->m_sb_lock);
466 outval->resblks = mp->m_resblks;
467 outval->resblks_avail = mp->m_resblks_avail;
470 spin_unlock(&mp->m_sb_lock);
480 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
481 if (!freeze_bdev(mp->m_super->s_bdev)) {
482 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
483 thaw_bdev(mp->m_super->s_bdev);
487 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
488 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
490 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
491 xfs_force_shutdown(mp,
492 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
502 * Force a shutdown of the filesystem instantly while keeping the filesystem
503 * consistent. We don't do an unmount here; just shutdown the shop, make sure
504 * that absolutely nothing persistent happens to this filesystem after this
507 * The shutdown state change is atomic, resulting in the first and only the
508 * first shutdown call processing the shutdown. This means we only shutdown the
509 * log once as it requires, and we don't spam the logs when multiple concurrent
510 * shutdowns race to set the shutdown flags.
513 xfs_do_force_shutdown(
514 struct xfs_mount *mp,
523 if (test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &mp->m_opstate)) {
524 xlog_shutdown_wait(mp->m_log);
528 mp->m_sb_bp->b_flags |= XBF_DONE;
530 if (flags & SHUTDOWN_FORCE_UMOUNT)
531 xfs_alert(mp, "User initiated shutdown received.");
533 if (xlog_force_shutdown(mp->m_log, flags)) {
534 tag = XFS_PTAG_SHUTDOWN_LOGERROR;
535 why = "Log I/O Error";
536 } else if (flags & SHUTDOWN_CORRUPT_INCORE) {
537 tag = XFS_PTAG_SHUTDOWN_CORRUPT;
538 why = "Corruption of in-memory data";
540 tag = XFS_PTAG_SHUTDOWN_IOERROR;
541 why = "Metadata I/O Error";
544 trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum);
546 xfs_alert_tag(mp, tag,
547 "%s (0x%x) detected at %pS (%s:%d). Shutting down filesystem.",
548 why, flags, __return_address, fname, lnnum);
550 "Please unmount the filesystem and rectify the problem(s)");
551 if (xfs_error_level >= XFS_ERRLEVEL_HIGH)
556 * Reserve free space for per-AG metadata.
559 xfs_fs_reserve_ag_blocks(
560 struct xfs_mount *mp)
563 struct xfs_perag *pag;
567 mp->m_finobt_nores = false;
568 for_each_perag(mp, agno, pag) {
569 err2 = xfs_ag_resv_init(pag, NULL);
574 if (error && error != -ENOSPC) {
576 "Error %d reserving per-AG metadata reserve pool.", error);
577 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
584 * Free space reserved for per-AG metadata.
587 xfs_fs_unreserve_ag_blocks(
588 struct xfs_mount *mp)
591 struct xfs_perag *pag;
595 for_each_perag(mp, agno, pag) {
596 err2 = xfs_ag_resv_free(pag);
603 "Error %d freeing per-AG metadata reserve pool.", error);