Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_sb.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"
19#include "xfs_log.h"
20#include "xfs_log_priv.h"
21#include "xfs_ag.h"
22#include "xfs_ag_resv.h"
23#include "xfs_trace.h"
24
25/*
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.
32 *
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
35 * anything.
36 */
37static int
38xfs_resizefs_init_new_ags(
39 struct xfs_trans *tp,
40 struct aghdr_init_data *id,
41 xfs_agnumber_t oagcount,
42 xfs_agnumber_t nagcount,
43 xfs_rfsblock_t delta,
44 struct xfs_perag *last_pag,
45 bool *lastag_extended)
46{
47 struct xfs_mount *mp = tp->t_mountp;
48 xfs_rfsblock_t nb = mp->m_sb.sb_dblocks + delta;
49 int error;
50
51 *lastag_extended = false;
52
53 INIT_LIST_HEAD(&id->buffer_list);
54 for (id->agno = nagcount - 1;
55 id->agno >= oagcount;
56 id->agno--, delta -= id->agsize) {
57
58 if (id->agno == nagcount - 1)
59 id->agsize = nb - (id->agno *
60 (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
61 else
62 id->agsize = mp->m_sb.sb_agblocks;
63
64 error = xfs_ag_init_headers(mp, id);
65 if (error) {
66 xfs_buf_delwri_cancel(&id->buffer_list);
67 return error;
68 }
69 }
70
71 error = xfs_buf_delwri_submit(&id->buffer_list);
72 if (error)
73 return error;
74
75 if (delta) {
76 *lastag_extended = true;
77 error = xfs_ag_extend_space(last_pag, tp, delta);
78 }
79 return error;
80}
81
82/*
83 * growfs operations
84 */
85static int
86xfs_growfs_data_private(
87 struct xfs_mount *mp, /* mount point for filesystem */
88 struct xfs_growfs_data *in) /* growfs data input struct */
89{
90 struct xfs_buf *bp;
91 int error;
92 xfs_agnumber_t nagcount;
93 xfs_agnumber_t nagimax = 0;
94 xfs_rfsblock_t nb, nb_div, nb_mod;
95 int64_t delta;
96 bool lastag_extended = false;
97 xfs_agnumber_t oagcount;
98 struct xfs_trans *tp;
99 struct aghdr_init_data id = {};
100 struct xfs_perag *last_pag;
101
102 nb = in->newblocks;
103 error = xfs_sb_validate_fsb_count(&mp->m_sb, nb);
104 if (error)
105 return error;
106
107 if (nb > mp->m_sb.sb_dblocks) {
108 error = xfs_buf_read_uncached(mp->m_ddev_targp,
109 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
110 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
111 if (error)
112 return error;
113 xfs_buf_relse(bp);
114 }
115
116 nb_div = nb;
117 nb_mod = do_div(nb_div, mp->m_sb.sb_agblocks);
118 if (nb_mod && nb_mod >= XFS_MIN_AG_BLOCKS)
119 nb_div++;
120 else if (nb_mod)
121 nb = nb_div * mp->m_sb.sb_agblocks;
122
123 if (nb_div > XFS_MAX_AGNUMBER + 1) {
124 nb_div = XFS_MAX_AGNUMBER + 1;
125 nb = nb_div * mp->m_sb.sb_agblocks;
126 }
127 nagcount = nb_div;
128 delta = nb - mp->m_sb.sb_dblocks;
129 /*
130 * Reject filesystems with a single AG because they are not
131 * supported, and reject a shrink operation that would cause a
132 * filesystem to become unsupported.
133 */
134 if (delta < 0 && nagcount < 2)
135 return -EINVAL;
136
137 /* No work to do */
138 if (delta == 0)
139 return 0;
140
141 oagcount = mp->m_sb.sb_agcount;
142 /* allocate the new per-ag structures */
143 if (nagcount > oagcount) {
144 error = xfs_initialize_perag(mp, nagcount, nb, &nagimax);
145 if (error)
146 return error;
147 } else if (nagcount < oagcount) {
148 /* TODO: shrinking the entire AGs hasn't yet completed */
149 return -EINVAL;
150 }
151
152 if (delta > 0)
153 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
154 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
155 &tp);
156 else
157 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, -delta, 0,
158 0, &tp);
159 if (error)
160 goto out_free_unused_perag;
161
162 last_pag = xfs_perag_get(mp, oagcount - 1);
163 if (delta > 0) {
164 error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount,
165 delta, last_pag, &lastag_extended);
166 } else {
167 xfs_warn_mount(mp, XFS_OPSTATE_WARNED_SHRINK,
168 "EXPERIMENTAL online shrink feature in use. Use at your own risk!");
169
170 error = xfs_ag_shrink_space(last_pag, &tp, -delta);
171 }
172 xfs_perag_put(last_pag);
173 if (error)
174 goto out_trans_cancel;
175
176 /*
177 * Update changed superblock fields transactionally. These are not
178 * seen by the rest of the world until the transaction commit applies
179 * them atomically to the superblock.
180 */
181 if (nagcount > oagcount)
182 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
183 if (delta)
184 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta);
185 if (id.nfree)
186 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
187
188 /*
189 * Sync sb counters now to reflect the updated values. This is
190 * particularly important for shrink because the write verifier
191 * will fail if sb_fdblocks is ever larger than sb_dblocks.
192 */
193 if (xfs_has_lazysbcount(mp))
194 xfs_log_sb(tp);
195
196 xfs_trans_set_sync(tp);
197 error = xfs_trans_commit(tp);
198 if (error)
199 return error;
200
201 /* New allocation groups fully initialized, so update mount struct */
202 if (nagimax)
203 mp->m_maxagi = nagimax;
204 xfs_set_low_space_thresholds(mp);
205 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
206
207 if (delta > 0) {
208 /*
209 * If we expanded the last AG, free the per-AG reservation
210 * so we can reinitialize it with the new size.
211 */
212 if (lastag_extended) {
213 struct xfs_perag *pag;
214
215 pag = xfs_perag_get(mp, id.agno);
216 error = xfs_ag_resv_free(pag);
217 xfs_perag_put(pag);
218 if (error)
219 return error;
220 }
221 /*
222 * Reserve AG metadata blocks. ENOSPC here does not mean there
223 * was a growfs failure, just that there still isn't space for
224 * new user data after the grow has been run.
225 */
226 error = xfs_fs_reserve_ag_blocks(mp);
227 if (error == -ENOSPC)
228 error = 0;
229 }
230 return error;
231
232out_trans_cancel:
233 xfs_trans_cancel(tp);
234out_free_unused_perag:
235 if (nagcount > oagcount)
236 xfs_free_unused_perag_range(mp, oagcount, nagcount);
237 return error;
238}
239
240static int
241xfs_growfs_log_private(
242 struct xfs_mount *mp, /* mount point for filesystem */
243 struct xfs_growfs_log *in) /* growfs log input struct */
244{
245 xfs_extlen_t nb;
246
247 nb = in->newblocks;
248 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
249 return -EINVAL;
250 if (nb == mp->m_sb.sb_logblocks &&
251 in->isint == (mp->m_sb.sb_logstart != 0))
252 return -EINVAL;
253 /*
254 * Moving the log is hard, need new interfaces to sync
255 * the log first, hold off all activity while moving it.
256 * Can have shorter or longer log in the same space,
257 * or transform internal to external log or vice versa.
258 */
259 return -ENOSYS;
260}
261
262static int
263xfs_growfs_imaxpct(
264 struct xfs_mount *mp,
265 __u32 imaxpct)
266{
267 struct xfs_trans *tp;
268 int dpct;
269 int error;
270
271 if (imaxpct > 100)
272 return -EINVAL;
273
274 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
275 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
276 if (error)
277 return error;
278
279 dpct = imaxpct - mp->m_sb.sb_imax_pct;
280 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
281 xfs_trans_set_sync(tp);
282 return xfs_trans_commit(tp);
283}
284
285/*
286 * protected versions of growfs function acquire and release locks on the mount
287 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
288 * XFS_IOC_FSGROWFSRT
289 */
290int
291xfs_growfs_data(
292 struct xfs_mount *mp,
293 struct xfs_growfs_data *in)
294{
295 int error = 0;
296
297 if (!capable(CAP_SYS_ADMIN))
298 return -EPERM;
299 if (!mutex_trylock(&mp->m_growlock))
300 return -EWOULDBLOCK;
301
302 /* update imaxpct separately to the physical grow of the filesystem */
303 if (in->imaxpct != mp->m_sb.sb_imax_pct) {
304 error = xfs_growfs_imaxpct(mp, in->imaxpct);
305 if (error)
306 goto out_error;
307 }
308
309 if (in->newblocks != mp->m_sb.sb_dblocks) {
310 error = xfs_growfs_data_private(mp, in);
311 if (error)
312 goto out_error;
313 }
314
315 /* Post growfs calculations needed to reflect new state in operations */
316 if (mp->m_sb.sb_imax_pct) {
317 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
318 do_div(icount, 100);
319 M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount);
320 } else
321 M_IGEO(mp)->maxicount = 0;
322
323 /* Update secondary superblocks now the physical grow has completed */
324 error = xfs_update_secondary_sbs(mp);
325
326out_error:
327 /*
328 * Increment the generation unconditionally, the error could be from
329 * updating the secondary superblocks, in which case the new size
330 * is live already.
331 */
332 mp->m_generation++;
333 mutex_unlock(&mp->m_growlock);
334 return error;
335}
336
337int
338xfs_growfs_log(
339 xfs_mount_t *mp,
340 struct xfs_growfs_log *in)
341{
342 int error;
343
344 if (!capable(CAP_SYS_ADMIN))
345 return -EPERM;
346 if (!mutex_trylock(&mp->m_growlock))
347 return -EWOULDBLOCK;
348 error = xfs_growfs_log_private(mp, in);
349 mutex_unlock(&mp->m_growlock);
350 return error;
351}
352
353/*
354 * Reserve the requested number of blocks if available. Otherwise return
355 * as many as possible to satisfy the request. The actual number
356 * reserved are returned in outval.
357 */
358int
359xfs_reserve_blocks(
360 struct xfs_mount *mp,
361 uint64_t request)
362{
363 int64_t lcounter, delta;
364 int64_t fdblks_delta = 0;
365 int64_t free;
366 int error = 0;
367
368 /*
369 * With per-cpu counters, this becomes an interesting problem. we need
370 * to work out if we are freeing or allocation blocks first, then we can
371 * do the modification as necessary.
372 *
373 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
374 * hold out any changes while we work out what to do. This means that
375 * the amount of free space can change while we do this, so we need to
376 * retry if we end up trying to reserve more space than is available.
377 */
378 spin_lock(&mp->m_sb_lock);
379
380 /*
381 * If our previous reservation was larger than the current value,
382 * then move any unused blocks back to the free pool. Modify the resblks
383 * counters directly since we shouldn't have any problems unreserving
384 * space.
385 */
386 if (mp->m_resblks > request) {
387 lcounter = mp->m_resblks_avail - request;
388 if (lcounter > 0) { /* release unused blocks */
389 fdblks_delta = lcounter;
390 mp->m_resblks_avail -= lcounter;
391 }
392 mp->m_resblks = request;
393 if (fdblks_delta) {
394 spin_unlock(&mp->m_sb_lock);
395 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
396 spin_lock(&mp->m_sb_lock);
397 }
398
399 goto out;
400 }
401
402 /*
403 * If the request is larger than the current reservation, reserve the
404 * blocks before we update the reserve counters. Sample m_fdblocks and
405 * perform a partial reservation if the request exceeds free space.
406 *
407 * The code below estimates how many blocks it can request from
408 * fdblocks to stash in the reserve pool. This is a classic TOCTOU
409 * race since fdblocks updates are not always coordinated via
410 * m_sb_lock. Set the reserve size even if there's not enough free
411 * space to fill it because mod_fdblocks will refill an undersized
412 * reserve when it can.
413 */
414 free = percpu_counter_sum(&mp->m_fdblocks) -
415 xfs_fdblocks_unavailable(mp);
416 delta = request - mp->m_resblks;
417 mp->m_resblks = request;
418 if (delta > 0 && free > 0) {
419 /*
420 * We'll either succeed in getting space from the free block
421 * count or we'll get an ENOSPC. Don't set the reserved flag
422 * here - we don't want to reserve the extra reserve blocks
423 * from the reserve.
424 *
425 * The desired reserve size can change after we drop the lock.
426 * Use mod_fdblocks to put the space into the reserve or into
427 * fdblocks as appropriate.
428 */
429 fdblks_delta = min(free, delta);
430 spin_unlock(&mp->m_sb_lock);
431 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
432 if (!error)
433 xfs_mod_fdblocks(mp, fdblks_delta, 0);
434 spin_lock(&mp->m_sb_lock);
435 }
436out:
437 spin_unlock(&mp->m_sb_lock);
438 return error;
439}
440
441int
442xfs_fs_goingdown(
443 xfs_mount_t *mp,
444 uint32_t inflags)
445{
446 switch (inflags) {
447 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
448 if (!bdev_freeze(mp->m_super->s_bdev)) {
449 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
450 bdev_thaw(mp->m_super->s_bdev);
451 }
452 break;
453 }
454 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
455 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
456 break;
457 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
458 xfs_force_shutdown(mp,
459 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
460 break;
461 default:
462 return -EINVAL;
463 }
464
465 return 0;
466}
467
468/*
469 * Force a shutdown of the filesystem instantly while keeping the filesystem
470 * consistent. We don't do an unmount here; just shutdown the shop, make sure
471 * that absolutely nothing persistent happens to this filesystem after this
472 * point.
473 *
474 * The shutdown state change is atomic, resulting in the first and only the
475 * first shutdown call processing the shutdown. This means we only shutdown the
476 * log once as it requires, and we don't spam the logs when multiple concurrent
477 * shutdowns race to set the shutdown flags.
478 */
479void
480xfs_do_force_shutdown(
481 struct xfs_mount *mp,
482 uint32_t flags,
483 char *fname,
484 int lnnum)
485{
486 int tag;
487 const char *why;
488
489
490 if (test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &mp->m_opstate)) {
491 xlog_shutdown_wait(mp->m_log);
492 return;
493 }
494 if (mp->m_sb_bp)
495 mp->m_sb_bp->b_flags |= XBF_DONE;
496
497 if (flags & SHUTDOWN_FORCE_UMOUNT)
498 xfs_alert(mp, "User initiated shutdown received.");
499
500 if (xlog_force_shutdown(mp->m_log, flags)) {
501 tag = XFS_PTAG_SHUTDOWN_LOGERROR;
502 why = "Log I/O Error";
503 } else if (flags & SHUTDOWN_CORRUPT_INCORE) {
504 tag = XFS_PTAG_SHUTDOWN_CORRUPT;
505 why = "Corruption of in-memory data";
506 } else if (flags & SHUTDOWN_CORRUPT_ONDISK) {
507 tag = XFS_PTAG_SHUTDOWN_CORRUPT;
508 why = "Corruption of on-disk metadata";
509 } else if (flags & SHUTDOWN_DEVICE_REMOVED) {
510 tag = XFS_PTAG_SHUTDOWN_IOERROR;
511 why = "Block device removal";
512 } else {
513 tag = XFS_PTAG_SHUTDOWN_IOERROR;
514 why = "Metadata I/O Error";
515 }
516
517 trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum);
518
519 xfs_alert_tag(mp, tag,
520"%s (0x%x) detected at %pS (%s:%d). Shutting down filesystem.",
521 why, flags, __return_address, fname, lnnum);
522 xfs_alert(mp,
523 "Please unmount the filesystem and rectify the problem(s)");
524 if (xfs_error_level >= XFS_ERRLEVEL_HIGH)
525 xfs_stack_trace();
526}
527
528/*
529 * Reserve free space for per-AG metadata.
530 */
531int
532xfs_fs_reserve_ag_blocks(
533 struct xfs_mount *mp)
534{
535 xfs_agnumber_t agno;
536 struct xfs_perag *pag;
537 int error = 0;
538 int err2;
539
540 mp->m_finobt_nores = false;
541 for_each_perag(mp, agno, pag) {
542 err2 = xfs_ag_resv_init(pag, NULL);
543 if (err2 && !error)
544 error = err2;
545 }
546
547 if (error && error != -ENOSPC) {
548 xfs_warn(mp,
549 "Error %d reserving per-AG metadata reserve pool.", error);
550 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
551 }
552
553 return error;
554}
555
556/*
557 * Free space reserved for per-AG metadata.
558 */
559int
560xfs_fs_unreserve_ag_blocks(
561 struct xfs_mount *mp)
562{
563 xfs_agnumber_t agno;
564 struct xfs_perag *pag;
565 int error = 0;
566 int err2;
567
568 for_each_perag(mp, agno, pag) {
569 err2 = xfs_ag_resv_free(pag);
570 if (err2 && !error)
571 error = err2;
572 }
573
574 if (error)
575 xfs_warn(mp,
576 "Error %d freeing per-AG metadata reserve pool.", error);
577
578 return error;
579}
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_sb.h"
25#include "xfs_mount.h"
26#include "xfs_defer.h"
27#include "xfs_da_format.h"
28#include "xfs_da_btree.h"
29#include "xfs_inode.h"
30#include "xfs_trans.h"
31#include "xfs_inode_item.h"
32#include "xfs_error.h"
33#include "xfs_btree.h"
34#include "xfs_alloc_btree.h"
35#include "xfs_alloc.h"
36#include "xfs_rmap_btree.h"
37#include "xfs_ialloc.h"
38#include "xfs_fsops.h"
39#include "xfs_itable.h"
40#include "xfs_trans_space.h"
41#include "xfs_rtalloc.h"
42#include "xfs_trace.h"
43#include "xfs_log.h"
44#include "xfs_filestream.h"
45#include "xfs_rmap.h"
46#include "xfs_ag_resv.h"
47
48/*
49 * File system operations
50 */
51
52int
53xfs_fs_geometry(
54 xfs_mount_t *mp,
55 xfs_fsop_geom_t *geo,
56 int new_version)
57{
58
59 memset(geo, 0, sizeof(*geo));
60
61 geo->blocksize = mp->m_sb.sb_blocksize;
62 geo->rtextsize = mp->m_sb.sb_rextsize;
63 geo->agblocks = mp->m_sb.sb_agblocks;
64 geo->agcount = mp->m_sb.sb_agcount;
65 geo->logblocks = mp->m_sb.sb_logblocks;
66 geo->sectsize = mp->m_sb.sb_sectsize;
67 geo->inodesize = mp->m_sb.sb_inodesize;
68 geo->imaxpct = mp->m_sb.sb_imax_pct;
69 geo->datablocks = mp->m_sb.sb_dblocks;
70 geo->rtblocks = mp->m_sb.sb_rblocks;
71 geo->rtextents = mp->m_sb.sb_rextents;
72 geo->logstart = mp->m_sb.sb_logstart;
73 ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
74 memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
75 if (new_version >= 2) {
76 geo->sunit = mp->m_sb.sb_unit;
77 geo->swidth = mp->m_sb.sb_width;
78 }
79 if (new_version >= 3) {
80 geo->version = XFS_FSOP_GEOM_VERSION;
81 geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
82 XFS_FSOP_GEOM_FLAGS_DIRV2 |
83 (xfs_sb_version_hasattr(&mp->m_sb) ?
84 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
85 (xfs_sb_version_hasquota(&mp->m_sb) ?
86 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
87 (xfs_sb_version_hasalign(&mp->m_sb) ?
88 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
89 (xfs_sb_version_hasdalign(&mp->m_sb) ?
90 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
91 (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
92 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
93 (xfs_sb_version_hassector(&mp->m_sb) ?
94 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
95 (xfs_sb_version_hasasciici(&mp->m_sb) ?
96 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
97 (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
98 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
99 (xfs_sb_version_hasattr2(&mp->m_sb) ?
100 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
101 (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
102 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
103 (xfs_sb_version_hascrc(&mp->m_sb) ?
104 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
105 (xfs_sb_version_hasftype(&mp->m_sb) ?
106 XFS_FSOP_GEOM_FLAGS_FTYPE : 0) |
107 (xfs_sb_version_hasfinobt(&mp->m_sb) ?
108 XFS_FSOP_GEOM_FLAGS_FINOBT : 0) |
109 (xfs_sb_version_hassparseinodes(&mp->m_sb) ?
110 XFS_FSOP_GEOM_FLAGS_SPINODES : 0) |
111 (xfs_sb_version_hasrmapbt(&mp->m_sb) ?
112 XFS_FSOP_GEOM_FLAGS_RMAPBT : 0) |
113 (xfs_sb_version_hasreflink(&mp->m_sb) ?
114 XFS_FSOP_GEOM_FLAGS_REFLINK : 0);
115 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
116 mp->m_sb.sb_logsectsize : BBSIZE;
117 geo->rtsectsize = mp->m_sb.sb_blocksize;
118 geo->dirblocksize = mp->m_dir_geo->blksize;
119 }
120 if (new_version >= 4) {
121 geo->flags |=
122 (xfs_sb_version_haslogv2(&mp->m_sb) ?
123 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
124 geo->logsunit = mp->m_sb.sb_logsunit;
125 }
126 return 0;
127}
128
129static struct xfs_buf *
130xfs_growfs_get_hdr_buf(
131 struct xfs_mount *mp,
132 xfs_daddr_t blkno,
133 size_t numblks,
134 int flags,
135 const struct xfs_buf_ops *ops)
136{
137 struct xfs_buf *bp;
138
139 bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
140 if (!bp)
141 return NULL;
142
143 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
144 bp->b_bn = blkno;
145 bp->b_maps[0].bm_bn = blkno;
146 bp->b_ops = ops;
147
148 return bp;
149}
150
151static int
152xfs_growfs_data_private(
153 xfs_mount_t *mp, /* mount point for filesystem */
154 xfs_growfs_data_t *in) /* growfs data input struct */
155{
156 xfs_agf_t *agf;
157 struct xfs_agfl *agfl;
158 xfs_agi_t *agi;
159 xfs_agnumber_t agno;
160 xfs_extlen_t agsize;
161 xfs_extlen_t tmpsize;
162 xfs_alloc_rec_t *arec;
163 xfs_buf_t *bp;
164 int bucket;
165 int dpct;
166 int error, saved_error = 0;
167 xfs_agnumber_t nagcount;
168 xfs_agnumber_t nagimax = 0;
169 xfs_rfsblock_t nb, nb_mod;
170 xfs_rfsblock_t new;
171 xfs_rfsblock_t nfree;
172 xfs_agnumber_t oagcount;
173 int pct;
174 xfs_trans_t *tp;
175
176 nb = in->newblocks;
177 pct = in->imaxpct;
178 if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
179 return -EINVAL;
180 if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
181 return error;
182 dpct = pct - mp->m_sb.sb_imax_pct;
183 error = xfs_buf_read_uncached(mp->m_ddev_targp,
184 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
185 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
186 if (error)
187 return error;
188 xfs_buf_relse(bp);
189
190 new = nb; /* use new as a temporary here */
191 nb_mod = do_div(new, mp->m_sb.sb_agblocks);
192 nagcount = new + (nb_mod != 0);
193 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
194 nagcount--;
195 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
196 if (nb < mp->m_sb.sb_dblocks)
197 return -EINVAL;
198 }
199 new = nb - mp->m_sb.sb_dblocks;
200 oagcount = mp->m_sb.sb_agcount;
201
202 /* allocate the new per-ag structures */
203 if (nagcount > oagcount) {
204 error = xfs_initialize_perag(mp, nagcount, &nagimax);
205 if (error)
206 return error;
207 }
208
209 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
210 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
211 if (error)
212 return error;
213
214 /*
215 * Write new AG headers to disk. Non-transactional, but written
216 * synchronously so they are completed prior to the growfs transaction
217 * being logged.
218 */
219 nfree = 0;
220 for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
221 __be32 *agfl_bno;
222
223 /*
224 * AG freespace header block
225 */
226 bp = xfs_growfs_get_hdr_buf(mp,
227 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
228 XFS_FSS_TO_BB(mp, 1), 0,
229 &xfs_agf_buf_ops);
230 if (!bp) {
231 error = -ENOMEM;
232 goto error0;
233 }
234
235 agf = XFS_BUF_TO_AGF(bp);
236 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
237 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
238 agf->agf_seqno = cpu_to_be32(agno);
239 if (agno == nagcount - 1)
240 agsize =
241 nb -
242 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
243 else
244 agsize = mp->m_sb.sb_agblocks;
245 agf->agf_length = cpu_to_be32(agsize);
246 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
247 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
248 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
249 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
250 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
251 agf->agf_roots[XFS_BTNUM_RMAPi] =
252 cpu_to_be32(XFS_RMAP_BLOCK(mp));
253 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
254 agf->agf_rmap_blocks = cpu_to_be32(1);
255 }
256
257 agf->agf_flfirst = cpu_to_be32(1);
258 agf->agf_fllast = 0;
259 agf->agf_flcount = 0;
260 tmpsize = agsize - mp->m_ag_prealloc_blocks;
261 agf->agf_freeblks = cpu_to_be32(tmpsize);
262 agf->agf_longest = cpu_to_be32(tmpsize);
263 if (xfs_sb_version_hascrc(&mp->m_sb))
264 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
265 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
266 agf->agf_refcount_root = cpu_to_be32(
267 xfs_refc_block(mp));
268 agf->agf_refcount_level = cpu_to_be32(1);
269 agf->agf_refcount_blocks = cpu_to_be32(1);
270 }
271
272 error = xfs_bwrite(bp);
273 xfs_buf_relse(bp);
274 if (error)
275 goto error0;
276
277 /*
278 * AG freelist header block
279 */
280 bp = xfs_growfs_get_hdr_buf(mp,
281 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
282 XFS_FSS_TO_BB(mp, 1), 0,
283 &xfs_agfl_buf_ops);
284 if (!bp) {
285 error = -ENOMEM;
286 goto error0;
287 }
288
289 agfl = XFS_BUF_TO_AGFL(bp);
290 if (xfs_sb_version_hascrc(&mp->m_sb)) {
291 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
292 agfl->agfl_seqno = cpu_to_be32(agno);
293 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
294 }
295
296 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
297 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
298 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
299
300 error = xfs_bwrite(bp);
301 xfs_buf_relse(bp);
302 if (error)
303 goto error0;
304
305 /*
306 * AG inode header block
307 */
308 bp = xfs_growfs_get_hdr_buf(mp,
309 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
310 XFS_FSS_TO_BB(mp, 1), 0,
311 &xfs_agi_buf_ops);
312 if (!bp) {
313 error = -ENOMEM;
314 goto error0;
315 }
316
317 agi = XFS_BUF_TO_AGI(bp);
318 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
319 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
320 agi->agi_seqno = cpu_to_be32(agno);
321 agi->agi_length = cpu_to_be32(agsize);
322 agi->agi_count = 0;
323 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
324 agi->agi_level = cpu_to_be32(1);
325 agi->agi_freecount = 0;
326 agi->agi_newino = cpu_to_be32(NULLAGINO);
327 agi->agi_dirino = cpu_to_be32(NULLAGINO);
328 if (xfs_sb_version_hascrc(&mp->m_sb))
329 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
330 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
331 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
332 agi->agi_free_level = cpu_to_be32(1);
333 }
334 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
335 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
336
337 error = xfs_bwrite(bp);
338 xfs_buf_relse(bp);
339 if (error)
340 goto error0;
341
342 /*
343 * BNO btree root block
344 */
345 bp = xfs_growfs_get_hdr_buf(mp,
346 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
347 BTOBB(mp->m_sb.sb_blocksize), 0,
348 &xfs_allocbt_buf_ops);
349
350 if (!bp) {
351 error = -ENOMEM;
352 goto error0;
353 }
354
355 if (xfs_sb_version_hascrc(&mp->m_sb))
356 xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
357 agno, XFS_BTREE_CRC_BLOCKS);
358 else
359 xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
360 agno, 0);
361
362 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
363 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
364 arec->ar_blockcount = cpu_to_be32(
365 agsize - be32_to_cpu(arec->ar_startblock));
366
367 error = xfs_bwrite(bp);
368 xfs_buf_relse(bp);
369 if (error)
370 goto error0;
371
372 /*
373 * CNT btree root block
374 */
375 bp = xfs_growfs_get_hdr_buf(mp,
376 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
377 BTOBB(mp->m_sb.sb_blocksize), 0,
378 &xfs_allocbt_buf_ops);
379 if (!bp) {
380 error = -ENOMEM;
381 goto error0;
382 }
383
384 if (xfs_sb_version_hascrc(&mp->m_sb))
385 xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
386 agno, XFS_BTREE_CRC_BLOCKS);
387 else
388 xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
389 agno, 0);
390
391 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
392 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
393 arec->ar_blockcount = cpu_to_be32(
394 agsize - be32_to_cpu(arec->ar_startblock));
395 nfree += be32_to_cpu(arec->ar_blockcount);
396
397 error = xfs_bwrite(bp);
398 xfs_buf_relse(bp);
399 if (error)
400 goto error0;
401
402 /* RMAP btree root block */
403 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
404 struct xfs_rmap_rec *rrec;
405 struct xfs_btree_block *block;
406
407 bp = xfs_growfs_get_hdr_buf(mp,
408 XFS_AGB_TO_DADDR(mp, agno, XFS_RMAP_BLOCK(mp)),
409 BTOBB(mp->m_sb.sb_blocksize), 0,
410 &xfs_rmapbt_buf_ops);
411 if (!bp) {
412 error = -ENOMEM;
413 goto error0;
414 }
415
416 xfs_btree_init_block(mp, bp, XFS_RMAP_CRC_MAGIC, 0, 0,
417 agno, XFS_BTREE_CRC_BLOCKS);
418 block = XFS_BUF_TO_BLOCK(bp);
419
420
421 /*
422 * mark the AG header regions as static metadata The BNO
423 * btree block is the first block after the headers, so
424 * it's location defines the size of region the static
425 * metadata consumes.
426 *
427 * Note: unlike mkfs, we never have to account for log
428 * space when growing the data regions
429 */
430 rrec = XFS_RMAP_REC_ADDR(block, 1);
431 rrec->rm_startblock = 0;
432 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
433 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
434 rrec->rm_offset = 0;
435 be16_add_cpu(&block->bb_numrecs, 1);
436
437 /* account freespace btree root blocks */
438 rrec = XFS_RMAP_REC_ADDR(block, 2);
439 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
440 rrec->rm_blockcount = cpu_to_be32(2);
441 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
442 rrec->rm_offset = 0;
443 be16_add_cpu(&block->bb_numrecs, 1);
444
445 /* account inode btree root blocks */
446 rrec = XFS_RMAP_REC_ADDR(block, 3);
447 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
448 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
449 XFS_IBT_BLOCK(mp));
450 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
451 rrec->rm_offset = 0;
452 be16_add_cpu(&block->bb_numrecs, 1);
453
454 /* account for rmap btree root */
455 rrec = XFS_RMAP_REC_ADDR(block, 4);
456 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
457 rrec->rm_blockcount = cpu_to_be32(1);
458 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
459 rrec->rm_offset = 0;
460 be16_add_cpu(&block->bb_numrecs, 1);
461
462 /* account for refc btree root */
463 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
464 rrec = XFS_RMAP_REC_ADDR(block, 5);
465 rrec->rm_startblock = cpu_to_be32(
466 xfs_refc_block(mp));
467 rrec->rm_blockcount = cpu_to_be32(1);
468 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
469 rrec->rm_offset = 0;
470 be16_add_cpu(&block->bb_numrecs, 1);
471 }
472
473 error = xfs_bwrite(bp);
474 xfs_buf_relse(bp);
475 if (error)
476 goto error0;
477 }
478
479 /*
480 * INO btree root block
481 */
482 bp = xfs_growfs_get_hdr_buf(mp,
483 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
484 BTOBB(mp->m_sb.sb_blocksize), 0,
485 &xfs_inobt_buf_ops);
486 if (!bp) {
487 error = -ENOMEM;
488 goto error0;
489 }
490
491 if (xfs_sb_version_hascrc(&mp->m_sb))
492 xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
493 agno, XFS_BTREE_CRC_BLOCKS);
494 else
495 xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
496 agno, 0);
497
498 error = xfs_bwrite(bp);
499 xfs_buf_relse(bp);
500 if (error)
501 goto error0;
502
503 /*
504 * FINO btree root block
505 */
506 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
507 bp = xfs_growfs_get_hdr_buf(mp,
508 XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
509 BTOBB(mp->m_sb.sb_blocksize), 0,
510 &xfs_inobt_buf_ops);
511 if (!bp) {
512 error = -ENOMEM;
513 goto error0;
514 }
515
516 if (xfs_sb_version_hascrc(&mp->m_sb))
517 xfs_btree_init_block(mp, bp, XFS_FIBT_CRC_MAGIC,
518 0, 0, agno,
519 XFS_BTREE_CRC_BLOCKS);
520 else
521 xfs_btree_init_block(mp, bp, XFS_FIBT_MAGIC, 0,
522 0, agno, 0);
523
524 error = xfs_bwrite(bp);
525 xfs_buf_relse(bp);
526 if (error)
527 goto error0;
528 }
529
530 /*
531 * refcount btree root block
532 */
533 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
534 bp = xfs_growfs_get_hdr_buf(mp,
535 XFS_AGB_TO_DADDR(mp, agno, xfs_refc_block(mp)),
536 BTOBB(mp->m_sb.sb_blocksize), 0,
537 &xfs_refcountbt_buf_ops);
538 if (!bp) {
539 error = -ENOMEM;
540 goto error0;
541 }
542
543 xfs_btree_init_block(mp, bp, XFS_REFC_CRC_MAGIC,
544 0, 0, agno,
545 XFS_BTREE_CRC_BLOCKS);
546
547 error = xfs_bwrite(bp);
548 xfs_buf_relse(bp);
549 if (error)
550 goto error0;
551 }
552 }
553 xfs_trans_agblocks_delta(tp, nfree);
554 /*
555 * There are new blocks in the old last a.g.
556 */
557 if (new) {
558 struct xfs_owner_info oinfo;
559
560 /*
561 * Change the agi length.
562 */
563 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
564 if (error) {
565 goto error0;
566 }
567 ASSERT(bp);
568 agi = XFS_BUF_TO_AGI(bp);
569 be32_add_cpu(&agi->agi_length, new);
570 ASSERT(nagcount == oagcount ||
571 be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
572 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
573 /*
574 * Change agf length.
575 */
576 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
577 if (error) {
578 goto error0;
579 }
580 ASSERT(bp);
581 agf = XFS_BUF_TO_AGF(bp);
582 be32_add_cpu(&agf->agf_length, new);
583 ASSERT(be32_to_cpu(agf->agf_length) ==
584 be32_to_cpu(agi->agi_length));
585
586 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
587
588 /*
589 * Free the new space.
590 *
591 * XFS_RMAP_OWN_NULL is used here to tell the rmap btree that
592 * this doesn't actually exist in the rmap btree.
593 */
594 xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_NULL);
595 error = xfs_free_extent(tp,
596 XFS_AGB_TO_FSB(mp, agno,
597 be32_to_cpu(agf->agf_length) - new),
598 new, &oinfo, XFS_AG_RESV_NONE);
599 if (error)
600 goto error0;
601 }
602
603 /*
604 * Update changed superblock fields transactionally. These are not
605 * seen by the rest of the world until the transaction commit applies
606 * them atomically to the superblock.
607 */
608 if (nagcount > oagcount)
609 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
610 if (nb > mp->m_sb.sb_dblocks)
611 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
612 nb - mp->m_sb.sb_dblocks);
613 if (nfree)
614 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
615 if (dpct)
616 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
617 xfs_trans_set_sync(tp);
618 error = xfs_trans_commit(tp);
619 if (error)
620 return error;
621
622 /* New allocation groups fully initialized, so update mount struct */
623 if (nagimax)
624 mp->m_maxagi = nagimax;
625 if (mp->m_sb.sb_imax_pct) {
626 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
627 do_div(icount, 100);
628 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
629 } else
630 mp->m_maxicount = 0;
631 xfs_set_low_space_thresholds(mp);
632 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
633
634 /*
635 * If we expanded the last AG, free the per-AG reservation
636 * so we can reinitialize it with the new size.
637 */
638 if (new) {
639 struct xfs_perag *pag;
640
641 pag = xfs_perag_get(mp, agno);
642 error = xfs_ag_resv_free(pag);
643 xfs_perag_put(pag);
644 if (error)
645 goto out;
646 }
647
648 /* Reserve AG metadata blocks. */
649 error = xfs_fs_reserve_ag_blocks(mp);
650 if (error && error != -ENOSPC)
651 goto out;
652
653 /* update secondary superblocks. */
654 for (agno = 1; agno < nagcount; agno++) {
655 error = 0;
656 /*
657 * new secondary superblocks need to be zeroed, not read from
658 * disk as the contents of the new area we are growing into is
659 * completely unknown.
660 */
661 if (agno < oagcount) {
662 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
663 XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
664 XFS_FSS_TO_BB(mp, 1), 0, &bp,
665 &xfs_sb_buf_ops);
666 } else {
667 bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
668 XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
669 XFS_FSS_TO_BB(mp, 1), 0);
670 if (bp) {
671 bp->b_ops = &xfs_sb_buf_ops;
672 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
673 } else
674 error = -ENOMEM;
675 }
676
677 /*
678 * If we get an error reading or writing alternate superblocks,
679 * continue. xfs_repair chooses the "best" superblock based
680 * on most matches; if we break early, we'll leave more
681 * superblocks un-updated than updated, and xfs_repair may
682 * pick them over the properly-updated primary.
683 */
684 if (error) {
685 xfs_warn(mp,
686 "error %d reading secondary superblock for ag %d",
687 error, agno);
688 saved_error = error;
689 continue;
690 }
691 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
692
693 error = xfs_bwrite(bp);
694 xfs_buf_relse(bp);
695 if (error) {
696 xfs_warn(mp,
697 "write error %d updating secondary superblock for ag %d",
698 error, agno);
699 saved_error = error;
700 continue;
701 }
702 }
703
704 out:
705 return saved_error ? saved_error : error;
706
707 error0:
708 xfs_trans_cancel(tp);
709 return error;
710}
711
712static int
713xfs_growfs_log_private(
714 xfs_mount_t *mp, /* mount point for filesystem */
715 xfs_growfs_log_t *in) /* growfs log input struct */
716{
717 xfs_extlen_t nb;
718
719 nb = in->newblocks;
720 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
721 return -EINVAL;
722 if (nb == mp->m_sb.sb_logblocks &&
723 in->isint == (mp->m_sb.sb_logstart != 0))
724 return -EINVAL;
725 /*
726 * Moving the log is hard, need new interfaces to sync
727 * the log first, hold off all activity while moving it.
728 * Can have shorter or longer log in the same space,
729 * or transform internal to external log or vice versa.
730 */
731 return -ENOSYS;
732}
733
734/*
735 * protected versions of growfs function acquire and release locks on the mount
736 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
737 * XFS_IOC_FSGROWFSRT
738 */
739
740
741int
742xfs_growfs_data(
743 xfs_mount_t *mp,
744 xfs_growfs_data_t *in)
745{
746 int error;
747
748 if (!capable(CAP_SYS_ADMIN))
749 return -EPERM;
750 if (!mutex_trylock(&mp->m_growlock))
751 return -EWOULDBLOCK;
752 error = xfs_growfs_data_private(mp, in);
753 /*
754 * Increment the generation unconditionally, the error could be from
755 * updating the secondary superblocks, in which case the new size
756 * is live already.
757 */
758 mp->m_generation++;
759 mutex_unlock(&mp->m_growlock);
760 return error;
761}
762
763int
764xfs_growfs_log(
765 xfs_mount_t *mp,
766 xfs_growfs_log_t *in)
767{
768 int error;
769
770 if (!capable(CAP_SYS_ADMIN))
771 return -EPERM;
772 if (!mutex_trylock(&mp->m_growlock))
773 return -EWOULDBLOCK;
774 error = xfs_growfs_log_private(mp, in);
775 mutex_unlock(&mp->m_growlock);
776 return error;
777}
778
779/*
780 * exported through ioctl XFS_IOC_FSCOUNTS
781 */
782
783int
784xfs_fs_counts(
785 xfs_mount_t *mp,
786 xfs_fsop_counts_t *cnt)
787{
788 cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
789 cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
790 cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
791 mp->m_alloc_set_aside;
792
793 spin_lock(&mp->m_sb_lock);
794 cnt->freertx = mp->m_sb.sb_frextents;
795 spin_unlock(&mp->m_sb_lock);
796 return 0;
797}
798
799/*
800 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
801 *
802 * xfs_reserve_blocks is called to set m_resblks
803 * in the in-core mount table. The number of unused reserved blocks
804 * is kept in m_resblks_avail.
805 *
806 * Reserve the requested number of blocks if available. Otherwise return
807 * as many as possible to satisfy the request. The actual number
808 * reserved are returned in outval
809 *
810 * A null inval pointer indicates that only the current reserved blocks
811 * available should be returned no settings are changed.
812 */
813
814int
815xfs_reserve_blocks(
816 xfs_mount_t *mp,
817 __uint64_t *inval,
818 xfs_fsop_resblks_t *outval)
819{
820 __int64_t lcounter, delta;
821 __int64_t fdblks_delta = 0;
822 __uint64_t request;
823 __int64_t free;
824 int error = 0;
825
826 /* If inval is null, report current values and return */
827 if (inval == (__uint64_t *)NULL) {
828 if (!outval)
829 return -EINVAL;
830 outval->resblks = mp->m_resblks;
831 outval->resblks_avail = mp->m_resblks_avail;
832 return 0;
833 }
834
835 request = *inval;
836
837 /*
838 * With per-cpu counters, this becomes an interesting problem. we need
839 * to work out if we are freeing or allocation blocks first, then we can
840 * do the modification as necessary.
841 *
842 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
843 * hold out any changes while we work out what to do. This means that
844 * the amount of free space can change while we do this, so we need to
845 * retry if we end up trying to reserve more space than is available.
846 */
847 spin_lock(&mp->m_sb_lock);
848
849 /*
850 * If our previous reservation was larger than the current value,
851 * then move any unused blocks back to the free pool. Modify the resblks
852 * counters directly since we shouldn't have any problems unreserving
853 * space.
854 */
855 if (mp->m_resblks > request) {
856 lcounter = mp->m_resblks_avail - request;
857 if (lcounter > 0) { /* release unused blocks */
858 fdblks_delta = lcounter;
859 mp->m_resblks_avail -= lcounter;
860 }
861 mp->m_resblks = request;
862 if (fdblks_delta) {
863 spin_unlock(&mp->m_sb_lock);
864 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
865 spin_lock(&mp->m_sb_lock);
866 }
867
868 goto out;
869 }
870
871 /*
872 * If the request is larger than the current reservation, reserve the
873 * blocks before we update the reserve counters. Sample m_fdblocks and
874 * perform a partial reservation if the request exceeds free space.
875 */
876 error = -ENOSPC;
877 do {
878 free = percpu_counter_sum(&mp->m_fdblocks) -
879 mp->m_alloc_set_aside;
880 if (!free)
881 break;
882
883 delta = request - mp->m_resblks;
884 lcounter = free - delta;
885 if (lcounter < 0)
886 /* We can't satisfy the request, just get what we can */
887 fdblks_delta = free;
888 else
889 fdblks_delta = delta;
890
891 /*
892 * We'll either succeed in getting space from the free block
893 * count or we'll get an ENOSPC. If we get a ENOSPC, it means
894 * things changed while we were calculating fdblks_delta and so
895 * we should try again to see if there is anything left to
896 * reserve.
897 *
898 * Don't set the reserved flag here - we don't want to reserve
899 * the extra reserve blocks from the reserve.....
900 */
901 spin_unlock(&mp->m_sb_lock);
902 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
903 spin_lock(&mp->m_sb_lock);
904 } while (error == -ENOSPC);
905
906 /*
907 * Update the reserve counters if blocks have been successfully
908 * allocated.
909 */
910 if (!error && fdblks_delta) {
911 mp->m_resblks += fdblks_delta;
912 mp->m_resblks_avail += fdblks_delta;
913 }
914
915out:
916 if (outval) {
917 outval->resblks = mp->m_resblks;
918 outval->resblks_avail = mp->m_resblks_avail;
919 }
920
921 spin_unlock(&mp->m_sb_lock);
922 return error;
923}
924
925int
926xfs_fs_goingdown(
927 xfs_mount_t *mp,
928 __uint32_t inflags)
929{
930 switch (inflags) {
931 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
932 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
933
934 if (sb && !IS_ERR(sb)) {
935 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
936 thaw_bdev(sb->s_bdev, sb);
937 }
938
939 break;
940 }
941 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
942 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
943 break;
944 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
945 xfs_force_shutdown(mp,
946 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
947 break;
948 default:
949 return -EINVAL;
950 }
951
952 return 0;
953}
954
955/*
956 * Force a shutdown of the filesystem instantly while keeping the filesystem
957 * consistent. We don't do an unmount here; just shutdown the shop, make sure
958 * that absolutely nothing persistent happens to this filesystem after this
959 * point.
960 */
961void
962xfs_do_force_shutdown(
963 xfs_mount_t *mp,
964 int flags,
965 char *fname,
966 int lnnum)
967{
968 int logerror;
969
970 logerror = flags & SHUTDOWN_LOG_IO_ERROR;
971
972 if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
973 xfs_notice(mp,
974 "%s(0x%x) called from line %d of file %s. Return address = 0x%p",
975 __func__, flags, lnnum, fname, __return_address);
976 }
977 /*
978 * No need to duplicate efforts.
979 */
980 if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
981 return;
982
983 /*
984 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
985 * queue up anybody new on the log reservations, and wakes up
986 * everybody who's sleeping on log reservations to tell them
987 * the bad news.
988 */
989 if (xfs_log_force_umount(mp, logerror))
990 return;
991
992 if (flags & SHUTDOWN_CORRUPT_INCORE) {
993 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
994 "Corruption of in-memory data detected. Shutting down filesystem");
995 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
996 xfs_stack_trace();
997 } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
998 if (logerror) {
999 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
1000 "Log I/O Error Detected. Shutting down filesystem");
1001 } else if (flags & SHUTDOWN_DEVICE_REQ) {
1002 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
1003 "All device paths lost. Shutting down filesystem");
1004 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
1005 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
1006 "I/O Error Detected. Shutting down filesystem");
1007 }
1008 }
1009 if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
1010 xfs_alert(mp,
1011 "Please umount the filesystem and rectify the problem(s)");
1012 }
1013}
1014
1015/*
1016 * Reserve free space for per-AG metadata.
1017 */
1018int
1019xfs_fs_reserve_ag_blocks(
1020 struct xfs_mount *mp)
1021{
1022 xfs_agnumber_t agno;
1023 struct xfs_perag *pag;
1024 int error = 0;
1025 int err2;
1026
1027 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
1028 pag = xfs_perag_get(mp, agno);
1029 err2 = xfs_ag_resv_init(pag);
1030 xfs_perag_put(pag);
1031 if (err2 && !error)
1032 error = err2;
1033 }
1034
1035 if (error && error != -ENOSPC) {
1036 xfs_warn(mp,
1037 "Error %d reserving per-AG metadata reserve pool.", error);
1038 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1039 }
1040
1041 return error;
1042}
1043
1044/*
1045 * Free space reserved for per-AG metadata.
1046 */
1047int
1048xfs_fs_unreserve_ag_blocks(
1049 struct xfs_mount *mp)
1050{
1051 xfs_agnumber_t agno;
1052 struct xfs_perag *pag;
1053 int error = 0;
1054 int err2;
1055
1056 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
1057 pag = xfs_perag_get(mp, agno);
1058 err2 = xfs_ag_resv_free(pag);
1059 xfs_perag_put(pag);
1060 if (err2 && !error)
1061 error = err2;
1062 }
1063
1064 if (error)
1065 xfs_warn(mp,
1066 "Error %d freeing per-AG metadata reserve pool.", error);
1067
1068 return error;
1069}