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 */
37 static int
xfs_resizefs_init_new_ags(struct xfs_trans * tp,struct aghdr_init_data * id,xfs_agnumber_t oagcount,xfs_agnumber_t nagcount,xfs_rfsblock_t delta,struct xfs_perag * last_pag,bool * lastag_extended)38 xfs_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 */
85 static int
xfs_growfs_data_private(struct xfs_mount * mp,struct xfs_growfs_data * in)86 xfs_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;
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 nagcount = nb_div + (nb_mod != 0);
119 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
120 nagcount--;
121 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
122 }
123 delta = nb - mp->m_sb.sb_dblocks;
124 /*
125 * Reject filesystems with a single AG because they are not
126 * supported, and reject a shrink operation that would cause a
127 * filesystem to become unsupported.
128 */
129 if (delta < 0 && nagcount < 2)
130 return -EINVAL;
131
132 oagcount = mp->m_sb.sb_agcount;
133 /* allocate the new per-ag structures */
134 if (nagcount > oagcount) {
135 error = xfs_initialize_perag(mp, nagcount, nb, &nagimax);
136 if (error)
137 return error;
138 } else if (nagcount < oagcount) {
139 /* TODO: shrinking the entire AGs hasn't yet completed */
140 return -EINVAL;
141 }
142
143 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
144 (delta > 0 ? XFS_GROWFS_SPACE_RES(mp) : -delta), 0,
145 XFS_TRANS_RESERVE, &tp);
146 if (error)
147 return error;
148
149 last_pag = xfs_perag_get(mp, oagcount - 1);
150 if (delta > 0) {
151 error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount,
152 delta, last_pag, &lastag_extended);
153 } else {
154 xfs_warn_mount(mp, XFS_OPSTATE_WARNED_SHRINK,
155 "EXPERIMENTAL online shrink feature in use. Use at your own risk!");
156
157 error = xfs_ag_shrink_space(last_pag, &tp, -delta);
158 }
159 xfs_perag_put(last_pag);
160 if (error)
161 goto out_trans_cancel;
162
163 /*
164 * Update changed superblock fields transactionally. These are not
165 * seen by the rest of the world until the transaction commit applies
166 * them atomically to the superblock.
167 */
168 if (nagcount > oagcount)
169 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
170 if (delta)
171 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta);
172 if (id.nfree)
173 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
174
175 /*
176 * Sync sb counters now to reflect the updated values. This is
177 * particularly important for shrink because the write verifier
178 * will fail if sb_fdblocks is ever larger than sb_dblocks.
179 */
180 if (xfs_has_lazysbcount(mp))
181 xfs_log_sb(tp);
182
183 xfs_trans_set_sync(tp);
184 error = xfs_trans_commit(tp);
185 if (error)
186 return error;
187
188 /* New allocation groups fully initialized, so update mount struct */
189 if (nagimax)
190 mp->m_maxagi = nagimax;
191 xfs_set_low_space_thresholds(mp);
192 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
193
194 if (delta > 0) {
195 /*
196 * If we expanded the last AG, free the per-AG reservation
197 * so we can reinitialize it with the new size.
198 */
199 if (lastag_extended) {
200 struct xfs_perag *pag;
201
202 pag = xfs_perag_get(mp, id.agno);
203 error = xfs_ag_resv_free(pag);
204 xfs_perag_put(pag);
205 if (error)
206 return error;
207 }
208 /*
209 * Reserve AG metadata blocks. ENOSPC here does not mean there
210 * was a growfs failure, just that there still isn't space for
211 * new user data after the grow has been run.
212 */
213 error = xfs_fs_reserve_ag_blocks(mp);
214 if (error == -ENOSPC)
215 error = 0;
216 }
217 return error;
218
219 out_trans_cancel:
220 xfs_trans_cancel(tp);
221 return error;
222 }
223
224 static int
xfs_growfs_log_private(struct xfs_mount * mp,struct xfs_growfs_log * in)225 xfs_growfs_log_private(
226 struct xfs_mount *mp, /* mount point for filesystem */
227 struct xfs_growfs_log *in) /* growfs log input struct */
228 {
229 xfs_extlen_t nb;
230
231 nb = in->newblocks;
232 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
233 return -EINVAL;
234 if (nb == mp->m_sb.sb_logblocks &&
235 in->isint == (mp->m_sb.sb_logstart != 0))
236 return -EINVAL;
237 /*
238 * Moving the log is hard, need new interfaces to sync
239 * the log first, hold off all activity while moving it.
240 * Can have shorter or longer log in the same space,
241 * or transform internal to external log or vice versa.
242 */
243 return -ENOSYS;
244 }
245
246 static int
xfs_growfs_imaxpct(struct xfs_mount * mp,__u32 imaxpct)247 xfs_growfs_imaxpct(
248 struct xfs_mount *mp,
249 __u32 imaxpct)
250 {
251 struct xfs_trans *tp;
252 int dpct;
253 int error;
254
255 if (imaxpct > 100)
256 return -EINVAL;
257
258 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
259 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
260 if (error)
261 return error;
262
263 dpct = imaxpct - mp->m_sb.sb_imax_pct;
264 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
265 xfs_trans_set_sync(tp);
266 return xfs_trans_commit(tp);
267 }
268
269 /*
270 * protected versions of growfs function acquire and release locks on the mount
271 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
272 * XFS_IOC_FSGROWFSRT
273 */
274 int
xfs_growfs_data(struct xfs_mount * mp,struct xfs_growfs_data * in)275 xfs_growfs_data(
276 struct xfs_mount *mp,
277 struct xfs_growfs_data *in)
278 {
279 int error = 0;
280
281 if (!capable(CAP_SYS_ADMIN))
282 return -EPERM;
283 if (!mutex_trylock(&mp->m_growlock))
284 return -EWOULDBLOCK;
285
286 /* update imaxpct separately to the physical grow of the filesystem */
287 if (in->imaxpct != mp->m_sb.sb_imax_pct) {
288 error = xfs_growfs_imaxpct(mp, in->imaxpct);
289 if (error)
290 goto out_error;
291 }
292
293 if (in->newblocks != mp->m_sb.sb_dblocks) {
294 error = xfs_growfs_data_private(mp, in);
295 if (error)
296 goto out_error;
297 }
298
299 /* Post growfs calculations needed to reflect new state in operations */
300 if (mp->m_sb.sb_imax_pct) {
301 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
302 do_div(icount, 100);
303 M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount);
304 } else
305 M_IGEO(mp)->maxicount = 0;
306
307 /* Update secondary superblocks now the physical grow has completed */
308 error = xfs_update_secondary_sbs(mp);
309
310 out_error:
311 /*
312 * Increment the generation unconditionally, the error could be from
313 * updating the secondary superblocks, in which case the new size
314 * is live already.
315 */
316 mp->m_generation++;
317 mutex_unlock(&mp->m_growlock);
318 return error;
319 }
320
321 int
xfs_growfs_log(xfs_mount_t * mp,struct xfs_growfs_log * in)322 xfs_growfs_log(
323 xfs_mount_t *mp,
324 struct xfs_growfs_log *in)
325 {
326 int error;
327
328 if (!capable(CAP_SYS_ADMIN))
329 return -EPERM;
330 if (!mutex_trylock(&mp->m_growlock))
331 return -EWOULDBLOCK;
332 error = xfs_growfs_log_private(mp, in);
333 mutex_unlock(&mp->m_growlock);
334 return error;
335 }
336
337 /*
338 * exported through ioctl XFS_IOC_FSCOUNTS
339 */
340
341 void
xfs_fs_counts(xfs_mount_t * mp,xfs_fsop_counts_t * cnt)342 xfs_fs_counts(
343 xfs_mount_t *mp,
344 xfs_fsop_counts_t *cnt)
345 {
346 cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
347 cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
348 cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
349 xfs_fdblocks_unavailable(mp);
350 cnt->freertx = percpu_counter_read_positive(&mp->m_frextents);
351 }
352
353 /*
354 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
355 *
356 * xfs_reserve_blocks is called to set m_resblks
357 * in the in-core mount table. The number of unused reserved blocks
358 * is kept in m_resblks_avail.
359 *
360 * Reserve the requested number of blocks if available. Otherwise return
361 * as many as possible to satisfy the request. The actual number
362 * reserved are returned in outval
363 *
364 * A null inval pointer indicates that only the current reserved blocks
365 * available should be returned no settings are changed.
366 */
367
368 int
xfs_reserve_blocks(xfs_mount_t * mp,uint64_t * inval,xfs_fsop_resblks_t * outval)369 xfs_reserve_blocks(
370 xfs_mount_t *mp,
371 uint64_t *inval,
372 xfs_fsop_resblks_t *outval)
373 {
374 int64_t lcounter, delta;
375 int64_t fdblks_delta = 0;
376 uint64_t request;
377 int64_t free;
378 int error = 0;
379
380 /* If inval is null, report current values and return */
381 if (inval == (uint64_t *)NULL) {
382 if (!outval)
383 return -EINVAL;
384 outval->resblks = mp->m_resblks;
385 outval->resblks_avail = mp->m_resblks_avail;
386 return 0;
387 }
388
389 request = *inval;
390
391 /*
392 * With per-cpu counters, this becomes an interesting problem. we need
393 * to work out if we are freeing or allocation blocks first, then we can
394 * do the modification as necessary.
395 *
396 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
397 * hold out any changes while we work out what to do. This means that
398 * the amount of free space can change while we do this, so we need to
399 * retry if we end up trying to reserve more space than is available.
400 */
401 spin_lock(&mp->m_sb_lock);
402
403 /*
404 * If our previous reservation was larger than the current value,
405 * then move any unused blocks back to the free pool. Modify the resblks
406 * counters directly since we shouldn't have any problems unreserving
407 * space.
408 */
409 if (mp->m_resblks > request) {
410 lcounter = mp->m_resblks_avail - request;
411 if (lcounter > 0) { /* release unused blocks */
412 fdblks_delta = lcounter;
413 mp->m_resblks_avail -= lcounter;
414 }
415 mp->m_resblks = request;
416 if (fdblks_delta) {
417 spin_unlock(&mp->m_sb_lock);
418 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
419 spin_lock(&mp->m_sb_lock);
420 }
421
422 goto out;
423 }
424
425 /*
426 * If the request is larger than the current reservation, reserve the
427 * blocks before we update the reserve counters. Sample m_fdblocks and
428 * perform a partial reservation if the request exceeds free space.
429 *
430 * The code below estimates how many blocks it can request from
431 * fdblocks to stash in the reserve pool. This is a classic TOCTOU
432 * race since fdblocks updates are not always coordinated via
433 * m_sb_lock. Set the reserve size even if there's not enough free
434 * space to fill it because mod_fdblocks will refill an undersized
435 * reserve when it can.
436 */
437 free = percpu_counter_sum(&mp->m_fdblocks) -
438 xfs_fdblocks_unavailable(mp);
439 delta = request - mp->m_resblks;
440 mp->m_resblks = request;
441 if (delta > 0 && free > 0) {
442 /*
443 * We'll either succeed in getting space from the free block
444 * count or we'll get an ENOSPC. Don't set the reserved flag
445 * here - we don't want to reserve the extra reserve blocks
446 * from the reserve.
447 *
448 * The desired reserve size can change after we drop the lock.
449 * Use mod_fdblocks to put the space into the reserve or into
450 * fdblocks as appropriate.
451 */
452 fdblks_delta = min(free, delta);
453 spin_unlock(&mp->m_sb_lock);
454 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
455 if (!error)
456 xfs_mod_fdblocks(mp, fdblks_delta, 0);
457 spin_lock(&mp->m_sb_lock);
458 }
459 out:
460 if (outval) {
461 outval->resblks = mp->m_resblks;
462 outval->resblks_avail = mp->m_resblks_avail;
463 }
464
465 spin_unlock(&mp->m_sb_lock);
466 return error;
467 }
468
469 int
xfs_fs_goingdown(xfs_mount_t * mp,uint32_t inflags)470 xfs_fs_goingdown(
471 xfs_mount_t *mp,
472 uint32_t inflags)
473 {
474 switch (inflags) {
475 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
476 if (!freeze_bdev(mp->m_super->s_bdev)) {
477 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
478 thaw_bdev(mp->m_super->s_bdev);
479 }
480 break;
481 }
482 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
483 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
484 break;
485 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
486 xfs_force_shutdown(mp,
487 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
488 break;
489 default:
490 return -EINVAL;
491 }
492
493 return 0;
494 }
495
496 /*
497 * Force a shutdown of the filesystem instantly while keeping the filesystem
498 * consistent. We don't do an unmount here; just shutdown the shop, make sure
499 * that absolutely nothing persistent happens to this filesystem after this
500 * point.
501 *
502 * The shutdown state change is atomic, resulting in the first and only the
503 * first shutdown call processing the shutdown. This means we only shutdown the
504 * log once as it requires, and we don't spam the logs when multiple concurrent
505 * shutdowns race to set the shutdown flags.
506 */
507 void
xfs_do_force_shutdown(struct xfs_mount * mp,uint32_t flags,char * fname,int lnnum)508 xfs_do_force_shutdown(
509 struct xfs_mount *mp,
510 uint32_t flags,
511 char *fname,
512 int lnnum)
513 {
514 int tag;
515 const char *why;
516
517
518 if (test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &mp->m_opstate)) {
519 xlog_shutdown_wait(mp->m_log);
520 return;
521 }
522 if (mp->m_sb_bp)
523 mp->m_sb_bp->b_flags |= XBF_DONE;
524
525 if (flags & SHUTDOWN_FORCE_UMOUNT)
526 xfs_alert(mp, "User initiated shutdown received.");
527
528 if (xlog_force_shutdown(mp->m_log, flags)) {
529 tag = XFS_PTAG_SHUTDOWN_LOGERROR;
530 why = "Log I/O Error";
531 } else if (flags & SHUTDOWN_CORRUPT_INCORE) {
532 tag = XFS_PTAG_SHUTDOWN_CORRUPT;
533 why = "Corruption of in-memory data";
534 } else if (flags & SHUTDOWN_CORRUPT_ONDISK) {
535 tag = XFS_PTAG_SHUTDOWN_CORRUPT;
536 why = "Corruption of on-disk metadata";
537 } else {
538 tag = XFS_PTAG_SHUTDOWN_IOERROR;
539 why = "Metadata I/O Error";
540 }
541
542 trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum);
543
544 xfs_alert_tag(mp, tag,
545 "%s (0x%x) detected at %pS (%s:%d). Shutting down filesystem.",
546 why, flags, __return_address, fname, lnnum);
547 xfs_alert(mp,
548 "Please unmount the filesystem and rectify the problem(s)");
549 if (xfs_error_level >= XFS_ERRLEVEL_HIGH)
550 xfs_stack_trace();
551 }
552
553 /*
554 * Reserve free space for per-AG metadata.
555 */
556 int
xfs_fs_reserve_ag_blocks(struct xfs_mount * mp)557 xfs_fs_reserve_ag_blocks(
558 struct xfs_mount *mp)
559 {
560 xfs_agnumber_t agno;
561 struct xfs_perag *pag;
562 int error = 0;
563 int err2;
564
565 mp->m_finobt_nores = false;
566 for_each_perag(mp, agno, pag) {
567 err2 = xfs_ag_resv_init(pag, NULL);
568 if (err2 && !error)
569 error = err2;
570 }
571
572 if (error && error != -ENOSPC) {
573 xfs_warn(mp,
574 "Error %d reserving per-AG metadata reserve pool.", error);
575 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
576 }
577
578 return error;
579 }
580
581 /*
582 * Free space reserved for per-AG metadata.
583 */
584 int
xfs_fs_unreserve_ag_blocks(struct xfs_mount * mp)585 xfs_fs_unreserve_ag_blocks(
586 struct xfs_mount *mp)
587 {
588 xfs_agnumber_t agno;
589 struct xfs_perag *pag;
590 int error = 0;
591 int err2;
592
593 for_each_perag(mp, agno, pag) {
594 err2 = xfs_ag_resv_free(pag);
595 if (err2 && !error)
596 error = err2;
597 }
598
599 if (error)
600 xfs_warn(mp,
601 "Error %d freeing per-AG metadata reserve pool.", error);
602
603 return error;
604 }
605