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_defer.h"
15 #include "xfs_trans.h"
16 #include "xfs_error.h"
17 #include "xfs_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_fsops.h"
20 #include "xfs_trans_space.h"
21 #include "xfs_rtalloc.h"
22 #include "xfs_trace.h"
25 #include "xfs_ag_resv.h"
31 xfs_growfs_data_private(
32 xfs_mount_t *mp, /* mount point for filesystem */
33 xfs_growfs_data_t *in) /* growfs data input struct */
37 xfs_agnumber_t nagcount;
38 xfs_agnumber_t nagimax = 0;
39 xfs_rfsblock_t nb, nb_mod;
41 xfs_agnumber_t oagcount;
43 struct aghdr_init_data id = {};
46 if (nb < mp->m_sb.sb_dblocks)
48 if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
50 error = xfs_buf_read_uncached(mp->m_ddev_targp,
51 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
52 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
57 new = nb; /* use new as a temporary here */
58 nb_mod = do_div(new, mp->m_sb.sb_agblocks);
59 nagcount = new + (nb_mod != 0);
60 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
62 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
63 if (nb < mp->m_sb.sb_dblocks)
66 new = nb - mp->m_sb.sb_dblocks;
67 oagcount = mp->m_sb.sb_agcount;
69 /* allocate the new per-ag structures */
70 if (nagcount > oagcount) {
71 error = xfs_initialize_perag(mp, nagcount, &nagimax);
76 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
77 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
82 * Write new AG headers to disk. Non-transactional, but need to be
83 * written and completed prior to the growfs transaction being logged.
84 * To do this, we use a delayed write buffer list and wait for
85 * submission and IO completion of the list as a whole. This allows the
86 * IO subsystem to merge all the AG headers in a single AG into a single
87 * IO and hide most of the latency of the IO from us.
89 * This also means that if we get an error whilst building the buffer
90 * list to write, we can cancel the entire list without having written
93 INIT_LIST_HEAD(&id.buffer_list);
94 for (id.agno = nagcount - 1;
96 id.agno--, new -= id.agsize) {
98 if (id.agno == nagcount - 1)
100 (id.agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
102 id.agsize = mp->m_sb.sb_agblocks;
104 error = xfs_ag_init_headers(mp, &id);
106 xfs_buf_delwri_cancel(&id.buffer_list);
107 goto out_trans_cancel;
110 error = xfs_buf_delwri_submit(&id.buffer_list);
112 goto out_trans_cancel;
114 xfs_trans_agblocks_delta(tp, id.nfree);
116 /* If there are new blocks in the old last AG, extend it. */
118 error = xfs_ag_extend_space(mp, tp, &id, new);
120 goto out_trans_cancel;
124 * Update changed superblock fields transactionally. These are not
125 * seen by the rest of the world until the transaction commit applies
126 * them atomically to the superblock.
128 if (nagcount > oagcount)
129 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
130 if (nb > mp->m_sb.sb_dblocks)
131 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
132 nb - mp->m_sb.sb_dblocks);
134 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
135 xfs_trans_set_sync(tp);
136 error = xfs_trans_commit(tp);
140 /* New allocation groups fully initialized, so update mount struct */
142 mp->m_maxagi = nagimax;
143 xfs_set_low_space_thresholds(mp);
144 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
147 * If we expanded the last AG, free the per-AG reservation
148 * so we can reinitialize it with the new size.
151 struct xfs_perag *pag;
153 pag = xfs_perag_get(mp, id.agno);
154 error = xfs_ag_resv_free(pag);
161 * Reserve AG metadata blocks. ENOSPC here does not mean there was a
162 * growfs failure, just that there still isn't space for new user data
163 * after the grow has been run.
165 error = xfs_fs_reserve_ag_blocks(mp);
166 if (error == -ENOSPC)
171 xfs_trans_cancel(tp);
176 xfs_growfs_log_private(
177 xfs_mount_t *mp, /* mount point for filesystem */
178 xfs_growfs_log_t *in) /* growfs log input struct */
183 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
185 if (nb == mp->m_sb.sb_logblocks &&
186 in->isint == (mp->m_sb.sb_logstart != 0))
189 * Moving the log is hard, need new interfaces to sync
190 * the log first, hold off all activity while moving it.
191 * Can have shorter or longer log in the same space,
192 * or transform internal to external log or vice versa.
199 struct xfs_mount *mp,
202 struct xfs_trans *tp;
209 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
210 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
214 dpct = imaxpct - mp->m_sb.sb_imax_pct;
215 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
216 xfs_trans_set_sync(tp);
217 return xfs_trans_commit(tp);
221 * protected versions of growfs function acquire and release locks on the mount
222 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
227 struct xfs_mount *mp,
228 struct xfs_growfs_data *in)
232 if (!capable(CAP_SYS_ADMIN))
234 if (!mutex_trylock(&mp->m_growlock))
237 /* update imaxpct separately to the physical grow of the filesystem */
238 if (in->imaxpct != mp->m_sb.sb_imax_pct) {
239 error = xfs_growfs_imaxpct(mp, in->imaxpct);
244 if (in->newblocks != mp->m_sb.sb_dblocks) {
245 error = xfs_growfs_data_private(mp, in);
250 /* Post growfs calculations needed to reflect new state in operations */
251 if (mp->m_sb.sb_imax_pct) {
252 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
254 mp->m_maxicount = XFS_FSB_TO_INO(mp, icount);
258 /* Update secondary superblocks now the physical grow has completed */
259 error = xfs_update_secondary_sbs(mp);
263 * Increment the generation unconditionally, the error could be from
264 * updating the secondary superblocks, in which case the new size
268 mutex_unlock(&mp->m_growlock);
275 xfs_growfs_log_t *in)
279 if (!capable(CAP_SYS_ADMIN))
281 if (!mutex_trylock(&mp->m_growlock))
283 error = xfs_growfs_log_private(mp, in);
284 mutex_unlock(&mp->m_growlock);
289 * exported through ioctl XFS_IOC_FSCOUNTS
295 xfs_fsop_counts_t *cnt)
297 cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
298 cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
299 cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
300 mp->m_alloc_set_aside;
302 spin_lock(&mp->m_sb_lock);
303 cnt->freertx = mp->m_sb.sb_frextents;
304 spin_unlock(&mp->m_sb_lock);
308 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
310 * xfs_reserve_blocks is called to set m_resblks
311 * in the in-core mount table. The number of unused reserved blocks
312 * is kept in m_resblks_avail.
314 * Reserve the requested number of blocks if available. Otherwise return
315 * as many as possible to satisfy the request. The actual number
316 * reserved are returned in outval
318 * A null inval pointer indicates that only the current reserved blocks
319 * available should be returned no settings are changed.
326 xfs_fsop_resblks_t *outval)
328 int64_t lcounter, delta;
329 int64_t fdblks_delta = 0;
334 /* If inval is null, report current values and return */
335 if (inval == (uint64_t *)NULL) {
338 outval->resblks = mp->m_resblks;
339 outval->resblks_avail = mp->m_resblks_avail;
346 * With per-cpu counters, this becomes an interesting problem. we need
347 * to work out if we are freeing or allocation blocks first, then we can
348 * do the modification as necessary.
350 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
351 * hold out any changes while we work out what to do. This means that
352 * the amount of free space can change while we do this, so we need to
353 * retry if we end up trying to reserve more space than is available.
355 spin_lock(&mp->m_sb_lock);
358 * If our previous reservation was larger than the current value,
359 * then move any unused blocks back to the free pool. Modify the resblks
360 * counters directly since we shouldn't have any problems unreserving
363 if (mp->m_resblks > request) {
364 lcounter = mp->m_resblks_avail - request;
365 if (lcounter > 0) { /* release unused blocks */
366 fdblks_delta = lcounter;
367 mp->m_resblks_avail -= lcounter;
369 mp->m_resblks = request;
371 spin_unlock(&mp->m_sb_lock);
372 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
373 spin_lock(&mp->m_sb_lock);
380 * If the request is larger than the current reservation, reserve the
381 * blocks before we update the reserve counters. Sample m_fdblocks and
382 * perform a partial reservation if the request exceeds free space.
386 free = percpu_counter_sum(&mp->m_fdblocks) -
387 mp->m_alloc_set_aside;
391 delta = request - mp->m_resblks;
392 lcounter = free - delta;
394 /* We can't satisfy the request, just get what we can */
397 fdblks_delta = delta;
400 * We'll either succeed in getting space from the free block
401 * count or we'll get an ENOSPC. If we get a ENOSPC, it means
402 * things changed while we were calculating fdblks_delta and so
403 * we should try again to see if there is anything left to
406 * Don't set the reserved flag here - we don't want to reserve
407 * the extra reserve blocks from the reserve.....
409 spin_unlock(&mp->m_sb_lock);
410 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
411 spin_lock(&mp->m_sb_lock);
412 } while (error == -ENOSPC);
415 * Update the reserve counters if blocks have been successfully
418 if (!error && fdblks_delta) {
419 mp->m_resblks += fdblks_delta;
420 mp->m_resblks_avail += fdblks_delta;
425 outval->resblks = mp->m_resblks;
426 outval->resblks_avail = mp->m_resblks_avail;
429 spin_unlock(&mp->m_sb_lock);
439 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
440 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
442 if (sb && !IS_ERR(sb)) {
443 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
444 thaw_bdev(sb->s_bdev, sb);
449 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
450 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
452 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
453 xfs_force_shutdown(mp,
454 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
464 * Force a shutdown of the filesystem instantly while keeping the filesystem
465 * consistent. We don't do an unmount here; just shutdown the shop, make sure
466 * that absolutely nothing persistent happens to this filesystem after this
470 xfs_do_force_shutdown(
471 struct xfs_mount *mp,
476 bool logerror = flags & SHUTDOWN_LOG_IO_ERROR;
479 * No need to duplicate efforts.
481 if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
485 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
486 * queue up anybody new on the log reservations, and wakes up
487 * everybody who's sleeping on log reservations to tell them
490 if (xfs_log_force_umount(mp, logerror))
493 if (flags & SHUTDOWN_FORCE_UMOUNT) {
495 "User initiated shutdown received. Shutting down filesystem");
500 "%s(0x%x) called from line %d of file %s. Return address = "PTR_FMT,
501 __func__, flags, lnnum, fname, __return_address);
503 if (flags & SHUTDOWN_CORRUPT_INCORE) {
504 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
505 "Corruption of in-memory data detected. Shutting down filesystem");
506 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
508 } else if (logerror) {
509 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
510 "Log I/O Error Detected. Shutting down filesystem");
511 } else if (flags & SHUTDOWN_DEVICE_REQ) {
512 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
513 "All device paths lost. Shutting down filesystem");
514 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
515 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
516 "I/O Error Detected. Shutting down filesystem");
520 "Please unmount the filesystem and rectify the problem(s)");
524 * Reserve free space for per-AG metadata.
527 xfs_fs_reserve_ag_blocks(
528 struct xfs_mount *mp)
531 struct xfs_perag *pag;
535 mp->m_finobt_nores = false;
536 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
537 pag = xfs_perag_get(mp, agno);
538 err2 = xfs_ag_resv_init(pag, NULL);
544 if (error && error != -ENOSPC) {
546 "Error %d reserving per-AG metadata reserve pool.", error);
547 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
554 * Free space reserved for per-AG metadata.
557 xfs_fs_unreserve_ag_blocks(
558 struct xfs_mount *mp)
561 struct xfs_perag *pag;
565 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
566 pag = xfs_perag_get(mp, agno);
567 err2 = xfs_ag_resv_free(pag);
575 "Error %d freeing per-AG metadata reserve pool.", error);