2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
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.
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.
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
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_trans.h"
28 #include "xfs_error.h"
29 #include "xfs_btree.h"
30 #include "xfs_alloc.h"
31 #include "xfs_fsops.h"
32 #include "xfs_trans_space.h"
33 #include "xfs_rtalloc.h"
34 #include "xfs_trace.h"
37 #include "xfs_ag_resv.h"
43 xfs_growfs_data_private(
44 xfs_mount_t *mp, /* mount point for filesystem */
45 xfs_growfs_data_t *in) /* growfs data input struct */
49 xfs_agnumber_t nagcount;
50 xfs_agnumber_t nagimax = 0;
51 xfs_rfsblock_t nb, nb_mod;
53 xfs_agnumber_t oagcount;
55 LIST_HEAD (buffer_list);
56 struct aghdr_init_data id = {};
59 if (nb < mp->m_sb.sb_dblocks)
61 if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
63 error = xfs_buf_read_uncached(mp->m_ddev_targp,
64 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
65 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
70 new = nb; /* use new as a temporary here */
71 nb_mod = do_div(new, mp->m_sb.sb_agblocks);
72 nagcount = new + (nb_mod != 0);
73 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
75 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
76 if (nb < mp->m_sb.sb_dblocks)
79 new = nb - mp->m_sb.sb_dblocks;
80 oagcount = mp->m_sb.sb_agcount;
82 /* allocate the new per-ag structures */
83 if (nagcount > oagcount) {
84 error = xfs_initialize_perag(mp, nagcount, &nagimax);
89 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
90 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
95 * Write new AG headers to disk. Non-transactional, but need to be
96 * written and completed prior to the growfs transaction being logged.
97 * To do this, we use a delayed write buffer list and wait for
98 * submission and IO completion of the list as a whole. This allows the
99 * IO subsystem to merge all the AG headers in a single AG into a single
100 * IO and hide most of the latency of the IO from us.
102 * This also means that if we get an error whilst building the buffer
103 * list to write, we can cancel the entire list without having written
106 INIT_LIST_HEAD(&id.buffer_list);
107 for (id.agno = nagcount - 1;
109 id.agno--, new -= id.agsize) {
111 if (id.agno == nagcount - 1)
113 (id.agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
115 id.agsize = mp->m_sb.sb_agblocks;
117 error = xfs_ag_init_headers(mp, &id);
119 xfs_buf_delwri_cancel(&id.buffer_list);
120 goto out_trans_cancel;
123 error = xfs_buf_delwri_submit(&id.buffer_list);
125 goto out_trans_cancel;
127 xfs_trans_agblocks_delta(tp, id.nfree);
129 /* If there are new blocks in the old last AG, extend it. */
131 error = xfs_ag_extend_space(mp, tp, &id, new);
133 goto out_trans_cancel;
137 * Update changed superblock fields transactionally. These are not
138 * seen by the rest of the world until the transaction commit applies
139 * them atomically to the superblock.
141 if (nagcount > oagcount)
142 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
143 if (nb > mp->m_sb.sb_dblocks)
144 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
145 nb - mp->m_sb.sb_dblocks);
147 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
148 xfs_trans_set_sync(tp);
149 error = xfs_trans_commit(tp);
153 /* New allocation groups fully initialized, so update mount struct */
155 mp->m_maxagi = nagimax;
156 xfs_set_low_space_thresholds(mp);
157 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
160 * If we expanded the last AG, free the per-AG reservation
161 * so we can reinitialize it with the new size.
164 struct xfs_perag *pag;
166 pag = xfs_perag_get(mp, id.agno);
167 error = xfs_ag_resv_free(pag);
174 * Reserve AG metadata blocks. ENOSPC here does not mean there was a
175 * growfs failure, just that there still isn't space for new user data
176 * after the grow has been run.
178 error = xfs_fs_reserve_ag_blocks(mp);
179 if (error == -ENOSPC)
184 xfs_trans_cancel(tp);
189 xfs_growfs_log_private(
190 xfs_mount_t *mp, /* mount point for filesystem */
191 xfs_growfs_log_t *in) /* growfs log input struct */
196 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
198 if (nb == mp->m_sb.sb_logblocks &&
199 in->isint == (mp->m_sb.sb_logstart != 0))
202 * Moving the log is hard, need new interfaces to sync
203 * the log first, hold off all activity while moving it.
204 * Can have shorter or longer log in the same space,
205 * or transform internal to external log or vice versa.
212 struct xfs_mount *mp,
215 struct xfs_trans *tp;
222 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
223 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
227 dpct = imaxpct - mp->m_sb.sb_imax_pct;
228 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
229 xfs_trans_set_sync(tp);
230 return xfs_trans_commit(tp);
234 * protected versions of growfs function acquire and release locks on the mount
235 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
240 struct xfs_mount *mp,
241 struct xfs_growfs_data *in)
245 if (!capable(CAP_SYS_ADMIN))
247 if (!mutex_trylock(&mp->m_growlock))
250 /* update imaxpct separately to the physical grow of the filesystem */
251 if (in->imaxpct != mp->m_sb.sb_imax_pct) {
252 error = xfs_growfs_imaxpct(mp, in->imaxpct);
257 if (in->newblocks != mp->m_sb.sb_dblocks) {
258 error = xfs_growfs_data_private(mp, in);
263 /* Post growfs calculations needed to reflect new state in operations */
264 if (mp->m_sb.sb_imax_pct) {
265 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
267 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
271 /* Update secondary superblocks now the physical grow has completed */
272 error = xfs_update_secondary_sbs(mp);
276 * Increment the generation unconditionally, the error could be from
277 * updating the secondary superblocks, in which case the new size
281 mutex_unlock(&mp->m_growlock);
288 xfs_growfs_log_t *in)
292 if (!capable(CAP_SYS_ADMIN))
294 if (!mutex_trylock(&mp->m_growlock))
296 error = xfs_growfs_log_private(mp, in);
297 mutex_unlock(&mp->m_growlock);
302 * exported through ioctl XFS_IOC_FSCOUNTS
308 xfs_fsop_counts_t *cnt)
310 cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
311 cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
312 cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
313 mp->m_alloc_set_aside;
315 spin_lock(&mp->m_sb_lock);
316 cnt->freertx = mp->m_sb.sb_frextents;
317 spin_unlock(&mp->m_sb_lock);
322 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
324 * xfs_reserve_blocks is called to set m_resblks
325 * in the in-core mount table. The number of unused reserved blocks
326 * is kept in m_resblks_avail.
328 * Reserve the requested number of blocks if available. Otherwise return
329 * as many as possible to satisfy the request. The actual number
330 * reserved are returned in outval
332 * A null inval pointer indicates that only the current reserved blocks
333 * available should be returned no settings are changed.
340 xfs_fsop_resblks_t *outval)
342 int64_t lcounter, delta;
343 int64_t fdblks_delta = 0;
348 /* If inval is null, report current values and return */
349 if (inval == (uint64_t *)NULL) {
352 outval->resblks = mp->m_resblks;
353 outval->resblks_avail = mp->m_resblks_avail;
360 * With per-cpu counters, this becomes an interesting problem. we need
361 * to work out if we are freeing or allocation blocks first, then we can
362 * do the modification as necessary.
364 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
365 * hold out any changes while we work out what to do. This means that
366 * the amount of free space can change while we do this, so we need to
367 * retry if we end up trying to reserve more space than is available.
369 spin_lock(&mp->m_sb_lock);
372 * If our previous reservation was larger than the current value,
373 * then move any unused blocks back to the free pool. Modify the resblks
374 * counters directly since we shouldn't have any problems unreserving
377 if (mp->m_resblks > request) {
378 lcounter = mp->m_resblks_avail - request;
379 if (lcounter > 0) { /* release unused blocks */
380 fdblks_delta = lcounter;
381 mp->m_resblks_avail -= lcounter;
383 mp->m_resblks = request;
385 spin_unlock(&mp->m_sb_lock);
386 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
387 spin_lock(&mp->m_sb_lock);
394 * If the request is larger than the current reservation, reserve the
395 * blocks before we update the reserve counters. Sample m_fdblocks and
396 * perform a partial reservation if the request exceeds free space.
400 free = percpu_counter_sum(&mp->m_fdblocks) -
401 mp->m_alloc_set_aside;
405 delta = request - mp->m_resblks;
406 lcounter = free - delta;
408 /* We can't satisfy the request, just get what we can */
411 fdblks_delta = delta;
414 * We'll either succeed in getting space from the free block
415 * count or we'll get an ENOSPC. If we get a ENOSPC, it means
416 * things changed while we were calculating fdblks_delta and so
417 * we should try again to see if there is anything left to
420 * Don't set the reserved flag here - we don't want to reserve
421 * the extra reserve blocks from the reserve.....
423 spin_unlock(&mp->m_sb_lock);
424 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
425 spin_lock(&mp->m_sb_lock);
426 } while (error == -ENOSPC);
429 * Update the reserve counters if blocks have been successfully
432 if (!error && fdblks_delta) {
433 mp->m_resblks += fdblks_delta;
434 mp->m_resblks_avail += fdblks_delta;
439 outval->resblks = mp->m_resblks;
440 outval->resblks_avail = mp->m_resblks_avail;
443 spin_unlock(&mp->m_sb_lock);
453 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
454 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
456 if (sb && !IS_ERR(sb)) {
457 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
458 thaw_bdev(sb->s_bdev, sb);
463 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
464 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
466 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
467 xfs_force_shutdown(mp,
468 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
478 * Force a shutdown of the filesystem instantly while keeping the filesystem
479 * consistent. We don't do an unmount here; just shutdown the shop, make sure
480 * that absolutely nothing persistent happens to this filesystem after this
484 xfs_do_force_shutdown(
492 logerror = flags & SHUTDOWN_LOG_IO_ERROR;
494 if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
496 "%s(0x%x) called from line %d of file %s. Return address = "PTR_FMT,
497 __func__, flags, lnnum, fname, __return_address);
500 * No need to duplicate efforts.
502 if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
506 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
507 * queue up anybody new on the log reservations, and wakes up
508 * everybody who's sleeping on log reservations to tell them
511 if (xfs_log_force_umount(mp, logerror))
514 if (flags & SHUTDOWN_CORRUPT_INCORE) {
515 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
516 "Corruption of in-memory data detected. Shutting down filesystem");
517 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
519 } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
521 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
522 "Log I/O Error Detected. Shutting down filesystem");
523 } else if (flags & SHUTDOWN_DEVICE_REQ) {
524 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
525 "All device paths lost. Shutting down filesystem");
526 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
527 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
528 "I/O Error Detected. Shutting down filesystem");
531 if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
533 "Please umount the filesystem and rectify the problem(s)");
538 * Reserve free space for per-AG metadata.
541 xfs_fs_reserve_ag_blocks(
542 struct xfs_mount *mp)
545 struct xfs_perag *pag;
549 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
550 pag = xfs_perag_get(mp, agno);
551 err2 = xfs_ag_resv_init(pag);
557 if (error && error != -ENOSPC) {
559 "Error %d reserving per-AG metadata reserve pool.", error);
560 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
567 * Free space reserved for per-AG metadata.
570 xfs_fs_unreserve_ag_blocks(
571 struct xfs_mount *mp)
574 struct xfs_perag *pag;
578 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
579 pag = xfs_perag_get(mp, agno);
580 err2 = xfs_ag_resv_free(pag);
588 "Error %d freeing per-AG metadata reserve pool.", error);