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