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/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_shared.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_sb.h"
25#include "xfs_mount.h"
26#include "xfs_defer.h"
27#include "xfs_da_format.h"
28#include "xfs_da_btree.h"
29#include "xfs_inode.h"
30#include "xfs_trans.h"
31#include "xfs_inode_item.h"
32#include "xfs_error.h"
33#include "xfs_btree.h"
34#include "xfs_alloc_btree.h"
35#include "xfs_alloc.h"
36#include "xfs_rmap_btree.h"
37#include "xfs_ialloc.h"
38#include "xfs_fsops.h"
39#include "xfs_itable.h"
40#include "xfs_trans_space.h"
41#include "xfs_rtalloc.h"
42#include "xfs_trace.h"
43#include "xfs_log.h"
44#include "xfs_filestream.h"
45#include "xfs_rmap.h"
46#include "xfs_ag_resv.h"
47
48/*
49 * File system operations
50 */
51
52static struct xfs_buf *
53xfs_growfs_get_hdr_buf(
54 struct xfs_mount *mp,
55 xfs_daddr_t blkno,
56 size_t numblks,
57 int flags,
58 const struct xfs_buf_ops *ops)
59{
60 struct xfs_buf *bp;
61
62 bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
63 if (!bp)
64 return NULL;
65
66 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
67 bp->b_bn = blkno;
68 bp->b_maps[0].bm_bn = blkno;
69 bp->b_ops = ops;
70
71 return bp;
72}
73
74static int
75xfs_growfs_data_private(
76 xfs_mount_t *mp, /* mount point for filesystem */
77 xfs_growfs_data_t *in) /* growfs data input struct */
78{
79 xfs_agf_t *agf;
80 struct xfs_agfl *agfl;
81 xfs_agi_t *agi;
82 xfs_agnumber_t agno;
83 xfs_extlen_t agsize;
84 xfs_extlen_t tmpsize;
85 xfs_alloc_rec_t *arec;
86 xfs_buf_t *bp;
87 int bucket;
88 int dpct;
89 int error, saved_error = 0;
90 xfs_agnumber_t nagcount;
91 xfs_agnumber_t nagimax = 0;
92 xfs_rfsblock_t nb, nb_mod;
93 xfs_rfsblock_t new;
94 xfs_rfsblock_t nfree;
95 xfs_agnumber_t oagcount;
96 int pct;
97 xfs_trans_t *tp;
98
99 nb = in->newblocks;
100 pct = in->imaxpct;
101 if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
102 return -EINVAL;
103 if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
104 return error;
105 dpct = pct - mp->m_sb.sb_imax_pct;
106 error = xfs_buf_read_uncached(mp->m_ddev_targp,
107 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
108 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
109 if (error)
110 return error;
111 xfs_buf_relse(bp);
112
113 new = nb; /* use new as a temporary here */
114 nb_mod = do_div(new, mp->m_sb.sb_agblocks);
115 nagcount = new + (nb_mod != 0);
116 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
117 nagcount--;
118 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
119 if (nb < mp->m_sb.sb_dblocks)
120 return -EINVAL;
121 }
122 new = nb - mp->m_sb.sb_dblocks;
123 oagcount = mp->m_sb.sb_agcount;
124
125 /* allocate the new per-ag structures */
126 if (nagcount > oagcount) {
127 error = xfs_initialize_perag(mp, nagcount, &nagimax);
128 if (error)
129 return error;
130 }
131
132 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
133 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
134 if (error)
135 return error;
136
137 /*
138 * Write new AG headers to disk. Non-transactional, but written
139 * synchronously so they are completed prior to the growfs transaction
140 * being logged.
141 */
142 nfree = 0;
143 for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
144 __be32 *agfl_bno;
145
146 /*
147 * AG freespace header block
148 */
149 bp = xfs_growfs_get_hdr_buf(mp,
150 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
151 XFS_FSS_TO_BB(mp, 1), 0,
152 &xfs_agf_buf_ops);
153 if (!bp) {
154 error = -ENOMEM;
155 goto error0;
156 }
157
158 agf = XFS_BUF_TO_AGF(bp);
159 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
160 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
161 agf->agf_seqno = cpu_to_be32(agno);
162 if (agno == nagcount - 1)
163 agsize =
164 nb -
165 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
166 else
167 agsize = mp->m_sb.sb_agblocks;
168 agf->agf_length = cpu_to_be32(agsize);
169 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
170 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
171 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
172 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
173 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
174 agf->agf_roots[XFS_BTNUM_RMAPi] =
175 cpu_to_be32(XFS_RMAP_BLOCK(mp));
176 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
177 agf->agf_rmap_blocks = cpu_to_be32(1);
178 }
179
180 agf->agf_flfirst = cpu_to_be32(1);
181 agf->agf_fllast = 0;
182 agf->agf_flcount = 0;
183 tmpsize = agsize - mp->m_ag_prealloc_blocks;
184 agf->agf_freeblks = cpu_to_be32(tmpsize);
185 agf->agf_longest = cpu_to_be32(tmpsize);
186 if (xfs_sb_version_hascrc(&mp->m_sb))
187 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
188 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
189 agf->agf_refcount_root = cpu_to_be32(
190 xfs_refc_block(mp));
191 agf->agf_refcount_level = cpu_to_be32(1);
192 agf->agf_refcount_blocks = cpu_to_be32(1);
193 }
194
195 error = xfs_bwrite(bp);
196 xfs_buf_relse(bp);
197 if (error)
198 goto error0;
199
200 /*
201 * AG freelist header block
202 */
203 bp = xfs_growfs_get_hdr_buf(mp,
204 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
205 XFS_FSS_TO_BB(mp, 1), 0,
206 &xfs_agfl_buf_ops);
207 if (!bp) {
208 error = -ENOMEM;
209 goto error0;
210 }
211
212 agfl = XFS_BUF_TO_AGFL(bp);
213 if (xfs_sb_version_hascrc(&mp->m_sb)) {
214 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
215 agfl->agfl_seqno = cpu_to_be32(agno);
216 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
217 }
218
219 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
220 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
221 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
222
223 error = xfs_bwrite(bp);
224 xfs_buf_relse(bp);
225 if (error)
226 goto error0;
227
228 /*
229 * AG inode header block
230 */
231 bp = xfs_growfs_get_hdr_buf(mp,
232 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
233 XFS_FSS_TO_BB(mp, 1), 0,
234 &xfs_agi_buf_ops);
235 if (!bp) {
236 error = -ENOMEM;
237 goto error0;
238 }
239
240 agi = XFS_BUF_TO_AGI(bp);
241 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
242 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
243 agi->agi_seqno = cpu_to_be32(agno);
244 agi->agi_length = cpu_to_be32(agsize);
245 agi->agi_count = 0;
246 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
247 agi->agi_level = cpu_to_be32(1);
248 agi->agi_freecount = 0;
249 agi->agi_newino = cpu_to_be32(NULLAGINO);
250 agi->agi_dirino = cpu_to_be32(NULLAGINO);
251 if (xfs_sb_version_hascrc(&mp->m_sb))
252 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
253 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
254 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
255 agi->agi_free_level = cpu_to_be32(1);
256 }
257 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
258 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
259
260 error = xfs_bwrite(bp);
261 xfs_buf_relse(bp);
262 if (error)
263 goto error0;
264
265 /*
266 * BNO btree root block
267 */
268 bp = xfs_growfs_get_hdr_buf(mp,
269 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
270 BTOBB(mp->m_sb.sb_blocksize), 0,
271 &xfs_allocbt_buf_ops);
272
273 if (!bp) {
274 error = -ENOMEM;
275 goto error0;
276 }
277
278 xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, agno, 0);
279
280 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
281 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
282 arec->ar_blockcount = cpu_to_be32(
283 agsize - be32_to_cpu(arec->ar_startblock));
284
285 error = xfs_bwrite(bp);
286 xfs_buf_relse(bp);
287 if (error)
288 goto error0;
289
290 /*
291 * CNT btree root block
292 */
293 bp = xfs_growfs_get_hdr_buf(mp,
294 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
295 BTOBB(mp->m_sb.sb_blocksize), 0,
296 &xfs_allocbt_buf_ops);
297 if (!bp) {
298 error = -ENOMEM;
299 goto error0;
300 }
301
302 xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, agno, 0);
303
304 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
305 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
306 arec->ar_blockcount = cpu_to_be32(
307 agsize - be32_to_cpu(arec->ar_startblock));
308 nfree += be32_to_cpu(arec->ar_blockcount);
309
310 error = xfs_bwrite(bp);
311 xfs_buf_relse(bp);
312 if (error)
313 goto error0;
314
315 /* RMAP btree root block */
316 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
317 struct xfs_rmap_rec *rrec;
318 struct xfs_btree_block *block;
319
320 bp = xfs_growfs_get_hdr_buf(mp,
321 XFS_AGB_TO_DADDR(mp, agno, XFS_RMAP_BLOCK(mp)),
322 BTOBB(mp->m_sb.sb_blocksize), 0,
323 &xfs_rmapbt_buf_ops);
324 if (!bp) {
325 error = -ENOMEM;
326 goto error0;
327 }
328
329 xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 0,
330 agno, 0);
331 block = XFS_BUF_TO_BLOCK(bp);
332
333
334 /*
335 * mark the AG header regions as static metadata The BNO
336 * btree block is the first block after the headers, so
337 * it's location defines the size of region the static
338 * metadata consumes.
339 *
340 * Note: unlike mkfs, we never have to account for log
341 * space when growing the data regions
342 */
343 rrec = XFS_RMAP_REC_ADDR(block, 1);
344 rrec->rm_startblock = 0;
345 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
346 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
347 rrec->rm_offset = 0;
348 be16_add_cpu(&block->bb_numrecs, 1);
349
350 /* account freespace btree root blocks */
351 rrec = XFS_RMAP_REC_ADDR(block, 2);
352 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
353 rrec->rm_blockcount = cpu_to_be32(2);
354 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
355 rrec->rm_offset = 0;
356 be16_add_cpu(&block->bb_numrecs, 1);
357
358 /* account inode btree root blocks */
359 rrec = XFS_RMAP_REC_ADDR(block, 3);
360 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
361 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
362 XFS_IBT_BLOCK(mp));
363 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
364 rrec->rm_offset = 0;
365 be16_add_cpu(&block->bb_numrecs, 1);
366
367 /* account for rmap btree root */
368 rrec = XFS_RMAP_REC_ADDR(block, 4);
369 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
370 rrec->rm_blockcount = cpu_to_be32(1);
371 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
372 rrec->rm_offset = 0;
373 be16_add_cpu(&block->bb_numrecs, 1);
374
375 /* account for refc btree root */
376 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
377 rrec = XFS_RMAP_REC_ADDR(block, 5);
378 rrec->rm_startblock = cpu_to_be32(
379 xfs_refc_block(mp));
380 rrec->rm_blockcount = cpu_to_be32(1);
381 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
382 rrec->rm_offset = 0;
383 be16_add_cpu(&block->bb_numrecs, 1);
384 }
385
386 error = xfs_bwrite(bp);
387 xfs_buf_relse(bp);
388 if (error)
389 goto error0;
390 }
391
392 /*
393 * INO btree root block
394 */
395 bp = xfs_growfs_get_hdr_buf(mp,
396 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
397 BTOBB(mp->m_sb.sb_blocksize), 0,
398 &xfs_inobt_buf_ops);
399 if (!bp) {
400 error = -ENOMEM;
401 goto error0;
402 }
403
404 xfs_btree_init_block(mp, bp, XFS_BTNUM_INO , 0, 0, agno, 0);
405
406 error = xfs_bwrite(bp);
407 xfs_buf_relse(bp);
408 if (error)
409 goto error0;
410
411 /*
412 * FINO btree root block
413 */
414 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
415 bp = xfs_growfs_get_hdr_buf(mp,
416 XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
417 BTOBB(mp->m_sb.sb_blocksize), 0,
418 &xfs_inobt_buf_ops);
419 if (!bp) {
420 error = -ENOMEM;
421 goto error0;
422 }
423
424 xfs_btree_init_block(mp, bp, XFS_BTNUM_FINO,
425 0, 0, agno, 0);
426
427 error = xfs_bwrite(bp);
428 xfs_buf_relse(bp);
429 if (error)
430 goto error0;
431 }
432
433 /*
434 * refcount btree root block
435 */
436 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
437 bp = xfs_growfs_get_hdr_buf(mp,
438 XFS_AGB_TO_DADDR(mp, agno, xfs_refc_block(mp)),
439 BTOBB(mp->m_sb.sb_blocksize), 0,
440 &xfs_refcountbt_buf_ops);
441 if (!bp) {
442 error = -ENOMEM;
443 goto error0;
444 }
445
446 xfs_btree_init_block(mp, bp, XFS_BTNUM_REFC,
447 0, 0, agno, 0);
448
449 error = xfs_bwrite(bp);
450 xfs_buf_relse(bp);
451 if (error)
452 goto error0;
453 }
454 }
455 xfs_trans_agblocks_delta(tp, nfree);
456 /*
457 * There are new blocks in the old last a.g.
458 */
459 if (new) {
460 struct xfs_owner_info oinfo;
461
462 /*
463 * Change the agi length.
464 */
465 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
466 if (error) {
467 goto error0;
468 }
469 ASSERT(bp);
470 agi = XFS_BUF_TO_AGI(bp);
471 be32_add_cpu(&agi->agi_length, new);
472 ASSERT(nagcount == oagcount ||
473 be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
474 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
475 /*
476 * Change agf length.
477 */
478 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
479 if (error) {
480 goto error0;
481 }
482 ASSERT(bp);
483 agf = XFS_BUF_TO_AGF(bp);
484 be32_add_cpu(&agf->agf_length, new);
485 ASSERT(be32_to_cpu(agf->agf_length) ==
486 be32_to_cpu(agi->agi_length));
487
488 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
489
490 /*
491 * Free the new space.
492 *
493 * XFS_RMAP_OWN_NULL is used here to tell the rmap btree that
494 * this doesn't actually exist in the rmap btree.
495 */
496 xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_NULL);
497 error = xfs_rmap_free(tp, bp, agno,
498 be32_to_cpu(agf->agf_length) - new,
499 new, &oinfo);
500 if (error)
501 goto error0;
502 error = xfs_free_extent(tp,
503 XFS_AGB_TO_FSB(mp, agno,
504 be32_to_cpu(agf->agf_length) - new),
505 new, &oinfo, XFS_AG_RESV_NONE);
506 if (error)
507 goto error0;
508 }
509
510 /*
511 * Update changed superblock fields transactionally. These are not
512 * seen by the rest of the world until the transaction commit applies
513 * them atomically to the superblock.
514 */
515 if (nagcount > oagcount)
516 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
517 if (nb > mp->m_sb.sb_dblocks)
518 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
519 nb - mp->m_sb.sb_dblocks);
520 if (nfree)
521 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
522 if (dpct)
523 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
524 xfs_trans_set_sync(tp);
525 error = xfs_trans_commit(tp);
526 if (error)
527 return error;
528
529 /* New allocation groups fully initialized, so update mount struct */
530 if (nagimax)
531 mp->m_maxagi = nagimax;
532 if (mp->m_sb.sb_imax_pct) {
533 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
534 do_div(icount, 100);
535 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
536 } else
537 mp->m_maxicount = 0;
538 xfs_set_low_space_thresholds(mp);
539 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
540
541 /*
542 * If we expanded the last AG, free the per-AG reservation
543 * so we can reinitialize it with the new size.
544 */
545 if (new) {
546 struct xfs_perag *pag;
547
548 pag = xfs_perag_get(mp, agno);
549 error = xfs_ag_resv_free(pag);
550 xfs_perag_put(pag);
551 if (error)
552 goto out;
553 }
554
555 /* Reserve AG metadata blocks. */
556 error = xfs_fs_reserve_ag_blocks(mp);
557 if (error && error != -ENOSPC)
558 goto out;
559
560 /* update secondary superblocks. */
561 for (agno = 1; agno < nagcount; agno++) {
562 error = 0;
563 /*
564 * new secondary superblocks need to be zeroed, not read from
565 * disk as the contents of the new area we are growing into is
566 * completely unknown.
567 */
568 if (agno < oagcount) {
569 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
570 XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
571 XFS_FSS_TO_BB(mp, 1), 0, &bp,
572 &xfs_sb_buf_ops);
573 } else {
574 bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
575 XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
576 XFS_FSS_TO_BB(mp, 1), 0);
577 if (bp) {
578 bp->b_ops = &xfs_sb_buf_ops;
579 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
580 } else
581 error = -ENOMEM;
582 }
583
584 /*
585 * If we get an error reading or writing alternate superblocks,
586 * continue. xfs_repair chooses the "best" superblock based
587 * on most matches; if we break early, we'll leave more
588 * superblocks un-updated than updated, and xfs_repair may
589 * pick them over the properly-updated primary.
590 */
591 if (error) {
592 xfs_warn(mp,
593 "error %d reading secondary superblock for ag %d",
594 error, agno);
595 saved_error = error;
596 continue;
597 }
598 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
599
600 error = xfs_bwrite(bp);
601 xfs_buf_relse(bp);
602 if (error) {
603 xfs_warn(mp,
604 "write error %d updating secondary superblock for ag %d",
605 error, agno);
606 saved_error = error;
607 continue;
608 }
609 }
610
611 out:
612 return saved_error ? saved_error : error;
613
614 error0:
615 xfs_trans_cancel(tp);
616 return error;
617}
618
619static int
620xfs_growfs_log_private(
621 xfs_mount_t *mp, /* mount point for filesystem */
622 xfs_growfs_log_t *in) /* growfs log input struct */
623{
624 xfs_extlen_t nb;
625
626 nb = in->newblocks;
627 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
628 return -EINVAL;
629 if (nb == mp->m_sb.sb_logblocks &&
630 in->isint == (mp->m_sb.sb_logstart != 0))
631 return -EINVAL;
632 /*
633 * Moving the log is hard, need new interfaces to sync
634 * the log first, hold off all activity while moving it.
635 * Can have shorter or longer log in the same space,
636 * or transform internal to external log or vice versa.
637 */
638 return -ENOSYS;
639}
640
641/*
642 * protected versions of growfs function acquire and release locks on the mount
643 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
644 * XFS_IOC_FSGROWFSRT
645 */
646
647
648int
649xfs_growfs_data(
650 xfs_mount_t *mp,
651 xfs_growfs_data_t *in)
652{
653 int error;
654
655 if (!capable(CAP_SYS_ADMIN))
656 return -EPERM;
657 if (!mutex_trylock(&mp->m_growlock))
658 return -EWOULDBLOCK;
659 error = xfs_growfs_data_private(mp, in);
660 /*
661 * Increment the generation unconditionally, the error could be from
662 * updating the secondary superblocks, in which case the new size
663 * is live already.
664 */
665 mp->m_generation++;
666 mutex_unlock(&mp->m_growlock);
667 return error;
668}
669
670int
671xfs_growfs_log(
672 xfs_mount_t *mp,
673 xfs_growfs_log_t *in)
674{
675 int error;
676
677 if (!capable(CAP_SYS_ADMIN))
678 return -EPERM;
679 if (!mutex_trylock(&mp->m_growlock))
680 return -EWOULDBLOCK;
681 error = xfs_growfs_log_private(mp, in);
682 mutex_unlock(&mp->m_growlock);
683 return error;
684}
685
686/*
687 * exported through ioctl XFS_IOC_FSCOUNTS
688 */
689
690int
691xfs_fs_counts(
692 xfs_mount_t *mp,
693 xfs_fsop_counts_t *cnt)
694{
695 cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
696 cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
697 cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
698 mp->m_alloc_set_aside;
699
700 spin_lock(&mp->m_sb_lock);
701 cnt->freertx = mp->m_sb.sb_frextents;
702 spin_unlock(&mp->m_sb_lock);
703 return 0;
704}
705
706/*
707 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
708 *
709 * xfs_reserve_blocks is called to set m_resblks
710 * in the in-core mount table. The number of unused reserved blocks
711 * is kept in m_resblks_avail.
712 *
713 * Reserve the requested number of blocks if available. Otherwise return
714 * as many as possible to satisfy the request. The actual number
715 * reserved are returned in outval
716 *
717 * A null inval pointer indicates that only the current reserved blocks
718 * available should be returned no settings are changed.
719 */
720
721int
722xfs_reserve_blocks(
723 xfs_mount_t *mp,
724 uint64_t *inval,
725 xfs_fsop_resblks_t *outval)
726{
727 int64_t lcounter, delta;
728 int64_t fdblks_delta = 0;
729 uint64_t request;
730 int64_t free;
731 int error = 0;
732
733 /* If inval is null, report current values and return */
734 if (inval == (uint64_t *)NULL) {
735 if (!outval)
736 return -EINVAL;
737 outval->resblks = mp->m_resblks;
738 outval->resblks_avail = mp->m_resblks_avail;
739 return 0;
740 }
741
742 request = *inval;
743
744 /*
745 * With per-cpu counters, this becomes an interesting problem. we need
746 * to work out if we are freeing or allocation blocks first, then we can
747 * do the modification as necessary.
748 *
749 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
750 * hold out any changes while we work out what to do. This means that
751 * the amount of free space can change while we do this, so we need to
752 * retry if we end up trying to reserve more space than is available.
753 */
754 spin_lock(&mp->m_sb_lock);
755
756 /*
757 * If our previous reservation was larger than the current value,
758 * then move any unused blocks back to the free pool. Modify the resblks
759 * counters directly since we shouldn't have any problems unreserving
760 * space.
761 */
762 if (mp->m_resblks > request) {
763 lcounter = mp->m_resblks_avail - request;
764 if (lcounter > 0) { /* release unused blocks */
765 fdblks_delta = lcounter;
766 mp->m_resblks_avail -= lcounter;
767 }
768 mp->m_resblks = request;
769 if (fdblks_delta) {
770 spin_unlock(&mp->m_sb_lock);
771 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
772 spin_lock(&mp->m_sb_lock);
773 }
774
775 goto out;
776 }
777
778 /*
779 * If the request is larger than the current reservation, reserve the
780 * blocks before we update the reserve counters. Sample m_fdblocks and
781 * perform a partial reservation if the request exceeds free space.
782 */
783 error = -ENOSPC;
784 do {
785 free = percpu_counter_sum(&mp->m_fdblocks) -
786 mp->m_alloc_set_aside;
787 if (!free)
788 break;
789
790 delta = request - mp->m_resblks;
791 lcounter = free - delta;
792 if (lcounter < 0)
793 /* We can't satisfy the request, just get what we can */
794 fdblks_delta = free;
795 else
796 fdblks_delta = delta;
797
798 /*
799 * We'll either succeed in getting space from the free block
800 * count or we'll get an ENOSPC. If we get a ENOSPC, it means
801 * things changed while we were calculating fdblks_delta and so
802 * we should try again to see if there is anything left to
803 * reserve.
804 *
805 * Don't set the reserved flag here - we don't want to reserve
806 * the extra reserve blocks from the reserve.....
807 */
808 spin_unlock(&mp->m_sb_lock);
809 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
810 spin_lock(&mp->m_sb_lock);
811 } while (error == -ENOSPC);
812
813 /*
814 * Update the reserve counters if blocks have been successfully
815 * allocated.
816 */
817 if (!error && fdblks_delta) {
818 mp->m_resblks += fdblks_delta;
819 mp->m_resblks_avail += fdblks_delta;
820 }
821
822out:
823 if (outval) {
824 outval->resblks = mp->m_resblks;
825 outval->resblks_avail = mp->m_resblks_avail;
826 }
827
828 spin_unlock(&mp->m_sb_lock);
829 return error;
830}
831
832int
833xfs_fs_goingdown(
834 xfs_mount_t *mp,
835 uint32_t inflags)
836{
837 switch (inflags) {
838 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
839 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
840
841 if (sb && !IS_ERR(sb)) {
842 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
843 thaw_bdev(sb->s_bdev, sb);
844 }
845
846 break;
847 }
848 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
849 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
850 break;
851 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
852 xfs_force_shutdown(mp,
853 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
854 break;
855 default:
856 return -EINVAL;
857 }
858
859 return 0;
860}
861
862/*
863 * Force a shutdown of the filesystem instantly while keeping the filesystem
864 * consistent. We don't do an unmount here; just shutdown the shop, make sure
865 * that absolutely nothing persistent happens to this filesystem after this
866 * point.
867 */
868void
869xfs_do_force_shutdown(
870 xfs_mount_t *mp,
871 int flags,
872 char *fname,
873 int lnnum)
874{
875 int logerror;
876
877 logerror = flags & SHUTDOWN_LOG_IO_ERROR;
878
879 if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
880 xfs_notice(mp,
881 "%s(0x%x) called from line %d of file %s. Return address = "PTR_FMT,
882 __func__, flags, lnnum, fname, __return_address);
883 }
884 /*
885 * No need to duplicate efforts.
886 */
887 if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
888 return;
889
890 /*
891 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
892 * queue up anybody new on the log reservations, and wakes up
893 * everybody who's sleeping on log reservations to tell them
894 * the bad news.
895 */
896 if (xfs_log_force_umount(mp, logerror))
897 return;
898
899 if (flags & SHUTDOWN_CORRUPT_INCORE) {
900 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
901 "Corruption of in-memory data detected. Shutting down filesystem");
902 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
903 xfs_stack_trace();
904 } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
905 if (logerror) {
906 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
907 "Log I/O Error Detected. Shutting down filesystem");
908 } else if (flags & SHUTDOWN_DEVICE_REQ) {
909 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
910 "All device paths lost. Shutting down filesystem");
911 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
912 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
913 "I/O Error Detected. Shutting down filesystem");
914 }
915 }
916 if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
917 xfs_alert(mp,
918 "Please umount the filesystem and rectify the problem(s)");
919 }
920}
921
922/*
923 * Reserve free space for per-AG metadata.
924 */
925int
926xfs_fs_reserve_ag_blocks(
927 struct xfs_mount *mp)
928{
929 xfs_agnumber_t agno;
930 struct xfs_perag *pag;
931 int error = 0;
932 int err2;
933
934 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
935 pag = xfs_perag_get(mp, agno);
936 err2 = xfs_ag_resv_init(pag);
937 xfs_perag_put(pag);
938 if (err2 && !error)
939 error = err2;
940 }
941
942 if (error && error != -ENOSPC) {
943 xfs_warn(mp,
944 "Error %d reserving per-AG metadata reserve pool.", error);
945 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
946 }
947
948 return error;
949}
950
951/*
952 * Free space reserved for per-AG metadata.
953 */
954int
955xfs_fs_unreserve_ag_blocks(
956 struct xfs_mount *mp)
957{
958 xfs_agnumber_t agno;
959 struct xfs_perag *pag;
960 int error = 0;
961 int err2;
962
963 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
964 pag = xfs_perag_get(mp, agno);
965 err2 = xfs_ag_resv_free(pag);
966 xfs_perag_put(pag);
967 if (err2 && !error)
968 error = err2;
969 }
970
971 if (error)
972 xfs_warn(mp,
973 "Error %d freeing per-AG metadata reserve pool.", error);
974
975 return error;
976}