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