Loading...
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_types.h"
21#include "xfs_bit.h"
22#include "xfs_inum.h"
23#include "xfs_log.h"
24#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
27#include "xfs_mount.h"
28#include "xfs_bmap_btree.h"
29#include "xfs_alloc_btree.h"
30#include "xfs_ialloc_btree.h"
31#include "xfs_dinode.h"
32#include "xfs_inode.h"
33#include "xfs_inode_item.h"
34#include "xfs_btree.h"
35#include "xfs_error.h"
36#include "xfs_alloc.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_rw.h"
43#include "xfs_filestream.h"
44#include "xfs_trace.h"
45
46/*
47 * File system operations
48 */
49
50int
51xfs_fs_geometry(
52 xfs_mount_t *mp,
53 xfs_fsop_geom_t *geo,
54 int new_version)
55{
56
57 memset(geo, 0, sizeof(*geo));
58
59 geo->blocksize = mp->m_sb.sb_blocksize;
60 geo->rtextsize = mp->m_sb.sb_rextsize;
61 geo->agblocks = mp->m_sb.sb_agblocks;
62 geo->agcount = mp->m_sb.sb_agcount;
63 geo->logblocks = mp->m_sb.sb_logblocks;
64 geo->sectsize = mp->m_sb.sb_sectsize;
65 geo->inodesize = mp->m_sb.sb_inodesize;
66 geo->imaxpct = mp->m_sb.sb_imax_pct;
67 geo->datablocks = mp->m_sb.sb_dblocks;
68 geo->rtblocks = mp->m_sb.sb_rblocks;
69 geo->rtextents = mp->m_sb.sb_rextents;
70 geo->logstart = mp->m_sb.sb_logstart;
71 ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
72 memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
73 if (new_version >= 2) {
74 geo->sunit = mp->m_sb.sb_unit;
75 geo->swidth = mp->m_sb.sb_width;
76 }
77 if (new_version >= 3) {
78 geo->version = XFS_FSOP_GEOM_VERSION;
79 geo->flags =
80 (xfs_sb_version_hasattr(&mp->m_sb) ?
81 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
82 (xfs_sb_version_hasnlink(&mp->m_sb) ?
83 XFS_FSOP_GEOM_FLAGS_NLINK : 0) |
84 (xfs_sb_version_hasquota(&mp->m_sb) ?
85 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
86 (xfs_sb_version_hasalign(&mp->m_sb) ?
87 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
88 (xfs_sb_version_hasdalign(&mp->m_sb) ?
89 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
90 (xfs_sb_version_hasshared(&mp->m_sb) ?
91 XFS_FSOP_GEOM_FLAGS_SHARED : 0) |
92 (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
93 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
94 (xfs_sb_version_hasdirv2(&mp->m_sb) ?
95 XFS_FSOP_GEOM_FLAGS_DIRV2 : 0) |
96 (xfs_sb_version_hassector(&mp->m_sb) ?
97 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
98 (xfs_sb_version_hasasciici(&mp->m_sb) ?
99 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
100 (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
101 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
102 (xfs_sb_version_hasattr2(&mp->m_sb) ?
103 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0);
104 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
105 mp->m_sb.sb_logsectsize : BBSIZE;
106 geo->rtsectsize = mp->m_sb.sb_blocksize;
107 geo->dirblocksize = mp->m_dirblksize;
108 }
109 if (new_version >= 4) {
110 geo->flags |=
111 (xfs_sb_version_haslogv2(&mp->m_sb) ?
112 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
113 geo->logsunit = mp->m_sb.sb_logsunit;
114 }
115 return 0;
116}
117
118static int
119xfs_growfs_data_private(
120 xfs_mount_t *mp, /* mount point for filesystem */
121 xfs_growfs_data_t *in) /* growfs data input struct */
122{
123 xfs_agf_t *agf;
124 xfs_agi_t *agi;
125 xfs_agnumber_t agno;
126 xfs_extlen_t agsize;
127 xfs_extlen_t tmpsize;
128 xfs_alloc_rec_t *arec;
129 struct xfs_btree_block *block;
130 xfs_buf_t *bp;
131 int bucket;
132 int dpct;
133 int error;
134 xfs_agnumber_t nagcount;
135 xfs_agnumber_t nagimax = 0;
136 xfs_rfsblock_t nb, nb_mod;
137 xfs_rfsblock_t new;
138 xfs_rfsblock_t nfree;
139 xfs_agnumber_t oagcount;
140 int pct;
141 xfs_trans_t *tp;
142
143 nb = in->newblocks;
144 pct = in->imaxpct;
145 if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
146 return XFS_ERROR(EINVAL);
147 if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
148 return error;
149 dpct = pct - mp->m_sb.sb_imax_pct;
150 bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
151 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
152 BBTOB(XFS_FSS_TO_BB(mp, 1)), 0);
153 if (!bp)
154 return EIO;
155 xfs_buf_relse(bp);
156
157 new = nb; /* use new as a temporary here */
158 nb_mod = do_div(new, mp->m_sb.sb_agblocks);
159 nagcount = new + (nb_mod != 0);
160 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
161 nagcount--;
162 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
163 if (nb < mp->m_sb.sb_dblocks)
164 return XFS_ERROR(EINVAL);
165 }
166 new = nb - mp->m_sb.sb_dblocks;
167 oagcount = mp->m_sb.sb_agcount;
168
169 /* allocate the new per-ag structures */
170 if (nagcount > oagcount) {
171 error = xfs_initialize_perag(mp, nagcount, &nagimax);
172 if (error)
173 return error;
174 }
175
176 tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
177 tp->t_flags |= XFS_TRANS_RESERVE;
178 if ((error = xfs_trans_reserve(tp, XFS_GROWFS_SPACE_RES(mp),
179 XFS_GROWDATA_LOG_RES(mp), 0, 0, 0))) {
180 xfs_trans_cancel(tp, 0);
181 return error;
182 }
183
184 /*
185 * Write new AG headers to disk. Non-transactional, but written
186 * synchronously so they are completed prior to the growfs transaction
187 * being logged.
188 */
189 nfree = 0;
190 for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
191 /*
192 * AG freelist header block
193 */
194 bp = xfs_buf_get(mp->m_ddev_targp,
195 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
196 XFS_FSS_TO_BB(mp, 1), XBF_LOCK | XBF_MAPPED);
197 agf = XFS_BUF_TO_AGF(bp);
198 memset(agf, 0, mp->m_sb.sb_sectsize);
199 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
200 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
201 agf->agf_seqno = cpu_to_be32(agno);
202 if (agno == nagcount - 1)
203 agsize =
204 nb -
205 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
206 else
207 agsize = mp->m_sb.sb_agblocks;
208 agf->agf_length = cpu_to_be32(agsize);
209 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
210 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
211 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
212 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
213 agf->agf_flfirst = 0;
214 agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
215 agf->agf_flcount = 0;
216 tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
217 agf->agf_freeblks = cpu_to_be32(tmpsize);
218 agf->agf_longest = cpu_to_be32(tmpsize);
219 error = xfs_bwrite(mp, bp);
220 if (error) {
221 goto error0;
222 }
223 /*
224 * AG inode header block
225 */
226 bp = xfs_buf_get(mp->m_ddev_targp,
227 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
228 XFS_FSS_TO_BB(mp, 1), XBF_LOCK | XBF_MAPPED);
229 agi = XFS_BUF_TO_AGI(bp);
230 memset(agi, 0, mp->m_sb.sb_sectsize);
231 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
232 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
233 agi->agi_seqno = cpu_to_be32(agno);
234 agi->agi_length = cpu_to_be32(agsize);
235 agi->agi_count = 0;
236 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
237 agi->agi_level = cpu_to_be32(1);
238 agi->agi_freecount = 0;
239 agi->agi_newino = cpu_to_be32(NULLAGINO);
240 agi->agi_dirino = cpu_to_be32(NULLAGINO);
241 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
242 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
243 error = xfs_bwrite(mp, bp);
244 if (error) {
245 goto error0;
246 }
247 /*
248 * BNO btree root block
249 */
250 bp = xfs_buf_get(mp->m_ddev_targp,
251 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
252 BTOBB(mp->m_sb.sb_blocksize),
253 XBF_LOCK | XBF_MAPPED);
254 block = XFS_BUF_TO_BLOCK(bp);
255 memset(block, 0, mp->m_sb.sb_blocksize);
256 block->bb_magic = cpu_to_be32(XFS_ABTB_MAGIC);
257 block->bb_level = 0;
258 block->bb_numrecs = cpu_to_be16(1);
259 block->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
260 block->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
261 arec = XFS_ALLOC_REC_ADDR(mp, block, 1);
262 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
263 arec->ar_blockcount = cpu_to_be32(
264 agsize - be32_to_cpu(arec->ar_startblock));
265 error = xfs_bwrite(mp, bp);
266 if (error) {
267 goto error0;
268 }
269 /*
270 * CNT btree root block
271 */
272 bp = xfs_buf_get(mp->m_ddev_targp,
273 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
274 BTOBB(mp->m_sb.sb_blocksize),
275 XBF_LOCK | XBF_MAPPED);
276 block = XFS_BUF_TO_BLOCK(bp);
277 memset(block, 0, mp->m_sb.sb_blocksize);
278 block->bb_magic = cpu_to_be32(XFS_ABTC_MAGIC);
279 block->bb_level = 0;
280 block->bb_numrecs = cpu_to_be16(1);
281 block->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
282 block->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
283 arec = XFS_ALLOC_REC_ADDR(mp, block, 1);
284 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
285 arec->ar_blockcount = cpu_to_be32(
286 agsize - be32_to_cpu(arec->ar_startblock));
287 nfree += be32_to_cpu(arec->ar_blockcount);
288 error = xfs_bwrite(mp, bp);
289 if (error) {
290 goto error0;
291 }
292 /*
293 * INO btree root block
294 */
295 bp = xfs_buf_get(mp->m_ddev_targp,
296 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
297 BTOBB(mp->m_sb.sb_blocksize),
298 XBF_LOCK | XBF_MAPPED);
299 block = XFS_BUF_TO_BLOCK(bp);
300 memset(block, 0, mp->m_sb.sb_blocksize);
301 block->bb_magic = cpu_to_be32(XFS_IBT_MAGIC);
302 block->bb_level = 0;
303 block->bb_numrecs = 0;
304 block->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
305 block->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
306 error = xfs_bwrite(mp, bp);
307 if (error) {
308 goto error0;
309 }
310 }
311 xfs_trans_agblocks_delta(tp, nfree);
312 /*
313 * There are new blocks in the old last a.g.
314 */
315 if (new) {
316 /*
317 * Change the agi length.
318 */
319 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
320 if (error) {
321 goto error0;
322 }
323 ASSERT(bp);
324 agi = XFS_BUF_TO_AGI(bp);
325 be32_add_cpu(&agi->agi_length, new);
326 ASSERT(nagcount == oagcount ||
327 be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
328 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
329 /*
330 * Change agf length.
331 */
332 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
333 if (error) {
334 goto error0;
335 }
336 ASSERT(bp);
337 agf = XFS_BUF_TO_AGF(bp);
338 be32_add_cpu(&agf->agf_length, new);
339 ASSERT(be32_to_cpu(agf->agf_length) ==
340 be32_to_cpu(agi->agi_length));
341
342 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
343 /*
344 * Free the new space.
345 */
346 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
347 be32_to_cpu(agf->agf_length) - new), new);
348 if (error) {
349 goto error0;
350 }
351 }
352
353 /*
354 * Update changed superblock fields transactionally. These are not
355 * seen by the rest of the world until the transaction commit applies
356 * them atomically to the superblock.
357 */
358 if (nagcount > oagcount)
359 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
360 if (nb > mp->m_sb.sb_dblocks)
361 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
362 nb - mp->m_sb.sb_dblocks);
363 if (nfree)
364 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
365 if (dpct)
366 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
367 error = xfs_trans_commit(tp, 0);
368 if (error)
369 return error;
370
371 /* New allocation groups fully initialized, so update mount struct */
372 if (nagimax)
373 mp->m_maxagi = nagimax;
374 if (mp->m_sb.sb_imax_pct) {
375 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
376 do_div(icount, 100);
377 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
378 } else
379 mp->m_maxicount = 0;
380 xfs_set_low_space_thresholds(mp);
381
382 /* update secondary superblocks. */
383 for (agno = 1; agno < nagcount; agno++) {
384 error = xfs_read_buf(mp, mp->m_ddev_targp,
385 XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
386 XFS_FSS_TO_BB(mp, 1), 0, &bp);
387 if (error) {
388 xfs_warn(mp,
389 "error %d reading secondary superblock for ag %d",
390 error, agno);
391 break;
392 }
393 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
394 /*
395 * If we get an error writing out the alternate superblocks,
396 * just issue a warning and continue. The real work is
397 * already done and committed.
398 */
399 if (!(error = xfs_bwrite(mp, bp))) {
400 continue;
401 } else {
402 xfs_warn(mp,
403 "write error %d updating secondary superblock for ag %d",
404 error, agno);
405 break; /* no point in continuing */
406 }
407 }
408 return 0;
409
410 error0:
411 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
412 return error;
413}
414
415static int
416xfs_growfs_log_private(
417 xfs_mount_t *mp, /* mount point for filesystem */
418 xfs_growfs_log_t *in) /* growfs log input struct */
419{
420 xfs_extlen_t nb;
421
422 nb = in->newblocks;
423 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
424 return XFS_ERROR(EINVAL);
425 if (nb == mp->m_sb.sb_logblocks &&
426 in->isint == (mp->m_sb.sb_logstart != 0))
427 return XFS_ERROR(EINVAL);
428 /*
429 * Moving the log is hard, need new interfaces to sync
430 * the log first, hold off all activity while moving it.
431 * Can have shorter or longer log in the same space,
432 * or transform internal to external log or vice versa.
433 */
434 return XFS_ERROR(ENOSYS);
435}
436
437/*
438 * protected versions of growfs function acquire and release locks on the mount
439 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
440 * XFS_IOC_FSGROWFSRT
441 */
442
443
444int
445xfs_growfs_data(
446 xfs_mount_t *mp,
447 xfs_growfs_data_t *in)
448{
449 int error;
450
451 if (!capable(CAP_SYS_ADMIN))
452 return XFS_ERROR(EPERM);
453 if (!mutex_trylock(&mp->m_growlock))
454 return XFS_ERROR(EWOULDBLOCK);
455 error = xfs_growfs_data_private(mp, in);
456 mutex_unlock(&mp->m_growlock);
457 return error;
458}
459
460int
461xfs_growfs_log(
462 xfs_mount_t *mp,
463 xfs_growfs_log_t *in)
464{
465 int error;
466
467 if (!capable(CAP_SYS_ADMIN))
468 return XFS_ERROR(EPERM);
469 if (!mutex_trylock(&mp->m_growlock))
470 return XFS_ERROR(EWOULDBLOCK);
471 error = xfs_growfs_log_private(mp, in);
472 mutex_unlock(&mp->m_growlock);
473 return error;
474}
475
476/*
477 * exported through ioctl XFS_IOC_FSCOUNTS
478 */
479
480int
481xfs_fs_counts(
482 xfs_mount_t *mp,
483 xfs_fsop_counts_t *cnt)
484{
485 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
486 spin_lock(&mp->m_sb_lock);
487 cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
488 cnt->freertx = mp->m_sb.sb_frextents;
489 cnt->freeino = mp->m_sb.sb_ifree;
490 cnt->allocino = mp->m_sb.sb_icount;
491 spin_unlock(&mp->m_sb_lock);
492 return 0;
493}
494
495/*
496 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
497 *
498 * xfs_reserve_blocks is called to set m_resblks
499 * in the in-core mount table. The number of unused reserved blocks
500 * is kept in m_resblks_avail.
501 *
502 * Reserve the requested number of blocks if available. Otherwise return
503 * as many as possible to satisfy the request. The actual number
504 * reserved are returned in outval
505 *
506 * A null inval pointer indicates that only the current reserved blocks
507 * available should be returned no settings are changed.
508 */
509
510int
511xfs_reserve_blocks(
512 xfs_mount_t *mp,
513 __uint64_t *inval,
514 xfs_fsop_resblks_t *outval)
515{
516 __int64_t lcounter, delta, fdblks_delta;
517 __uint64_t request;
518
519 /* If inval is null, report current values and return */
520 if (inval == (__uint64_t *)NULL) {
521 if (!outval)
522 return EINVAL;
523 outval->resblks = mp->m_resblks;
524 outval->resblks_avail = mp->m_resblks_avail;
525 return 0;
526 }
527
528 request = *inval;
529
530 /*
531 * With per-cpu counters, this becomes an interesting
532 * problem. we needto work out if we are freeing or allocation
533 * blocks first, then we can do the modification as necessary.
534 *
535 * We do this under the m_sb_lock so that if we are near
536 * ENOSPC, we will hold out any changes while we work out
537 * what to do. This means that the amount of free space can
538 * change while we do this, so we need to retry if we end up
539 * trying to reserve more space than is available.
540 *
541 * We also use the xfs_mod_incore_sb() interface so that we
542 * don't have to care about whether per cpu counter are
543 * enabled, disabled or even compiled in....
544 */
545retry:
546 spin_lock(&mp->m_sb_lock);
547 xfs_icsb_sync_counters_locked(mp, 0);
548
549 /*
550 * If our previous reservation was larger than the current value,
551 * then move any unused blocks back to the free pool.
552 */
553 fdblks_delta = 0;
554 if (mp->m_resblks > request) {
555 lcounter = mp->m_resblks_avail - request;
556 if (lcounter > 0) { /* release unused blocks */
557 fdblks_delta = lcounter;
558 mp->m_resblks_avail -= lcounter;
559 }
560 mp->m_resblks = request;
561 } else {
562 __int64_t free;
563
564 free = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
565 if (!free)
566 goto out; /* ENOSPC and fdblks_delta = 0 */
567
568 delta = request - mp->m_resblks;
569 lcounter = free - delta;
570 if (lcounter < 0) {
571 /* We can't satisfy the request, just get what we can */
572 mp->m_resblks += free;
573 mp->m_resblks_avail += free;
574 fdblks_delta = -free;
575 } else {
576 fdblks_delta = -delta;
577 mp->m_resblks = request;
578 mp->m_resblks_avail += delta;
579 }
580 }
581out:
582 if (outval) {
583 outval->resblks = mp->m_resblks;
584 outval->resblks_avail = mp->m_resblks_avail;
585 }
586 spin_unlock(&mp->m_sb_lock);
587
588 if (fdblks_delta) {
589 /*
590 * If we are putting blocks back here, m_resblks_avail is
591 * already at its max so this will put it in the free pool.
592 *
593 * If we need space, we'll either succeed in getting it
594 * from the free block count or we'll get an enospc. If
595 * we get a ENOSPC, it means things changed while we were
596 * calculating fdblks_delta and so we should try again to
597 * see if there is anything left to reserve.
598 *
599 * Don't set the reserved flag here - we don't want to reserve
600 * the extra reserve blocks from the reserve.....
601 */
602 int error;
603 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
604 fdblks_delta, 0);
605 if (error == ENOSPC)
606 goto retry;
607 }
608 return 0;
609}
610
611/*
612 * Dump a transaction into the log that contains no real change. This is needed
613 * to be able to make the log dirty or stamp the current tail LSN into the log
614 * during the covering operation.
615 *
616 * We cannot use an inode here for this - that will push dirty state back up
617 * into the VFS and then periodic inode flushing will prevent log covering from
618 * making progress. Hence we log a field in the superblock instead and use a
619 * synchronous transaction to ensure the superblock is immediately unpinned
620 * and can be written back.
621 */
622int
623xfs_fs_log_dummy(
624 xfs_mount_t *mp)
625{
626 xfs_trans_t *tp;
627 int error;
628
629 tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
630 error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
631 XFS_DEFAULT_LOG_COUNT);
632 if (error) {
633 xfs_trans_cancel(tp, 0);
634 return error;
635 }
636
637 /* log the UUID because it is an unchanging field */
638 xfs_mod_sb(tp, XFS_SB_UUID);
639 xfs_trans_set_sync(tp);
640 return xfs_trans_commit(tp, 0);
641}
642
643int
644xfs_fs_goingdown(
645 xfs_mount_t *mp,
646 __uint32_t inflags)
647{
648 switch (inflags) {
649 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
650 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
651
652 if (sb && !IS_ERR(sb)) {
653 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
654 thaw_bdev(sb->s_bdev, sb);
655 }
656
657 break;
658 }
659 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
660 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
661 break;
662 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
663 xfs_force_shutdown(mp,
664 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
665 break;
666 default:
667 return XFS_ERROR(EINVAL);
668 }
669
670 return 0;
671}
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}