xfs: add support for large btree blocks
[linux-2.6-block.git] / fs / xfs / xfs_fsops.c
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_log.h"
22 #include "xfs_trans.h"
23 #include "xfs_sb.h"
24 #include "xfs_ag.h"
25 #include "xfs_mount.h"
26 #include "xfs_bmap_btree.h"
27 #include "xfs_alloc_btree.h"
28 #include "xfs_ialloc_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
31 #include "xfs_inode_item.h"
32 #include "xfs_btree.h"
33 #include "xfs_error.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_filestream.h"
41 #include "xfs_trace.h"
42
43 /*
44  * File system operations
45  */
46
47 int
48 xfs_fs_geometry(
49         xfs_mount_t             *mp,
50         xfs_fsop_geom_t         *geo,
51         int                     new_version)
52 {
53
54         memset(geo, 0, sizeof(*geo));
55
56         geo->blocksize = mp->m_sb.sb_blocksize;
57         geo->rtextsize = mp->m_sb.sb_rextsize;
58         geo->agblocks = mp->m_sb.sb_agblocks;
59         geo->agcount = mp->m_sb.sb_agcount;
60         geo->logblocks = mp->m_sb.sb_logblocks;
61         geo->sectsize = mp->m_sb.sb_sectsize;
62         geo->inodesize = mp->m_sb.sb_inodesize;
63         geo->imaxpct = mp->m_sb.sb_imax_pct;
64         geo->datablocks = mp->m_sb.sb_dblocks;
65         geo->rtblocks = mp->m_sb.sb_rblocks;
66         geo->rtextents = mp->m_sb.sb_rextents;
67         geo->logstart = mp->m_sb.sb_logstart;
68         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
69         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
70         if (new_version >= 2) {
71                 geo->sunit = mp->m_sb.sb_unit;
72                 geo->swidth = mp->m_sb.sb_width;
73         }
74         if (new_version >= 3) {
75                 geo->version = XFS_FSOP_GEOM_VERSION;
76                 geo->flags =
77                         (xfs_sb_version_hasattr(&mp->m_sb) ?
78                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
79                         (xfs_sb_version_hasnlink(&mp->m_sb) ?
80                                 XFS_FSOP_GEOM_FLAGS_NLINK : 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_hasshared(&mp->m_sb) ?
88                                 XFS_FSOP_GEOM_FLAGS_SHARED : 0) |
89                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
90                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
91                         (xfs_sb_version_hasdirv2(&mp->m_sb) ?
92                                 XFS_FSOP_GEOM_FLAGS_DIRV2 : 0) |
93                         (xfs_sb_version_hassector(&mp->m_sb) ?
94                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
95                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
96                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
97                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
98                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
99                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
100                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
101                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
102                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0);
103                 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
104                                 mp->m_sb.sb_logsectsize : BBSIZE;
105                 geo->rtsectsize = mp->m_sb.sb_blocksize;
106                 geo->dirblocksize = mp->m_dirblksize;
107         }
108         if (new_version >= 4) {
109                 geo->flags |=
110                         (xfs_sb_version_haslogv2(&mp->m_sb) ?
111                                 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
112                 geo->logsunit = mp->m_sb.sb_logsunit;
113         }
114         return 0;
115 }
116
117 static struct xfs_buf *
118 xfs_growfs_get_hdr_buf(
119         struct xfs_mount        *mp,
120         xfs_daddr_t             blkno,
121         size_t                  numblks,
122         int                     flags,
123         const struct xfs_buf_ops *ops)
124 {
125         struct xfs_buf          *bp;
126
127         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
128         if (!bp)
129                 return NULL;
130
131         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
132         bp->b_bn = blkno;
133         bp->b_maps[0].bm_bn = blkno;
134         bp->b_ops = ops;
135
136         return bp;
137 }
138
139 static int
140 xfs_growfs_data_private(
141         xfs_mount_t             *mp,            /* mount point for filesystem */
142         xfs_growfs_data_t       *in)            /* growfs data input struct */
143 {
144         xfs_agf_t               *agf;
145         struct xfs_agfl         *agfl;
146         xfs_agi_t               *agi;
147         xfs_agnumber_t          agno;
148         xfs_extlen_t            agsize;
149         xfs_extlen_t            tmpsize;
150         xfs_alloc_rec_t         *arec;
151         xfs_buf_t               *bp;
152         int                     bucket;
153         int                     dpct;
154         int                     error;
155         xfs_agnumber_t          nagcount;
156         xfs_agnumber_t          nagimax = 0;
157         xfs_rfsblock_t          nb, nb_mod;
158         xfs_rfsblock_t          new;
159         xfs_rfsblock_t          nfree;
160         xfs_agnumber_t          oagcount;
161         int                     pct;
162         xfs_trans_t             *tp;
163
164         nb = in->newblocks;
165         pct = in->imaxpct;
166         if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
167                 return XFS_ERROR(EINVAL);
168         if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
169                 return error;
170         dpct = pct - mp->m_sb.sb_imax_pct;
171         bp = xfs_buf_read_uncached(mp->m_ddev_targp,
172                                 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
173                                 XFS_FSS_TO_BB(mp, 1), 0, NULL);
174         if (!bp)
175                 return EIO;
176         if (bp->b_error) {
177                 int     error = bp->b_error;
178                 xfs_buf_relse(bp);
179                 return error;
180         }
181         xfs_buf_relse(bp);
182
183         new = nb;       /* use new as a temporary here */
184         nb_mod = do_div(new, mp->m_sb.sb_agblocks);
185         nagcount = new + (nb_mod != 0);
186         if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
187                 nagcount--;
188                 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
189                 if (nb < mp->m_sb.sb_dblocks)
190                         return XFS_ERROR(EINVAL);
191         }
192         new = nb - mp->m_sb.sb_dblocks;
193         oagcount = mp->m_sb.sb_agcount;
194
195         /* allocate the new per-ag structures */
196         if (nagcount > oagcount) {
197                 error = xfs_initialize_perag(mp, nagcount, &nagimax);
198                 if (error)
199                         return error;
200         }
201
202         tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
203         tp->t_flags |= XFS_TRANS_RESERVE;
204         if ((error = xfs_trans_reserve(tp, XFS_GROWFS_SPACE_RES(mp),
205                         XFS_GROWDATA_LOG_RES(mp), 0, 0, 0))) {
206                 xfs_trans_cancel(tp, 0);
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                 /*
218                  * AG freespace header block
219                  */
220                 bp = xfs_growfs_get_hdr_buf(mp,
221                                 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
222                                 XFS_FSS_TO_BB(mp, 1), 0,
223                                 &xfs_agf_buf_ops);
224                 if (!bp) {
225                         error = ENOMEM;
226                         goto error0;
227                 }
228
229                 agf = XFS_BUF_TO_AGF(bp);
230                 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
231                 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
232                 agf->agf_seqno = cpu_to_be32(agno);
233                 if (agno == nagcount - 1)
234                         agsize =
235                                 nb -
236                                 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
237                 else
238                         agsize = mp->m_sb.sb_agblocks;
239                 agf->agf_length = cpu_to_be32(agsize);
240                 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
241                 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
242                 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
243                 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
244                 agf->agf_flfirst = 0;
245                 agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
246                 agf->agf_flcount = 0;
247                 tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
248                 agf->agf_freeblks = cpu_to_be32(tmpsize);
249                 agf->agf_longest = cpu_to_be32(tmpsize);
250                 error = xfs_bwrite(bp);
251                 xfs_buf_relse(bp);
252                 if (error)
253                         goto error0;
254
255                 /*
256                  * AG freelist header block
257                  */
258                 bp = xfs_growfs_get_hdr_buf(mp,
259                                 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
260                                 XFS_FSS_TO_BB(mp, 1), 0,
261                                 &xfs_agfl_buf_ops);
262                 if (!bp) {
263                         error = ENOMEM;
264                         goto error0;
265                 }
266
267                 agfl = XFS_BUF_TO_AGFL(bp);
268                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
269                         agfl->agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
270
271                 error = xfs_bwrite(bp);
272                 xfs_buf_relse(bp);
273                 if (error)
274                         goto error0;
275
276                 /*
277                  * AG inode header block
278                  */
279                 bp = xfs_growfs_get_hdr_buf(mp,
280                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
281                                 XFS_FSS_TO_BB(mp, 1), 0,
282                                 &xfs_agi_buf_ops);
283                 if (!bp) {
284                         error = ENOMEM;
285                         goto error0;
286                 }
287
288                 agi = XFS_BUF_TO_AGI(bp);
289                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
290                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
291                 agi->agi_seqno = cpu_to_be32(agno);
292                 agi->agi_length = cpu_to_be32(agsize);
293                 agi->agi_count = 0;
294                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
295                 agi->agi_level = cpu_to_be32(1);
296                 agi->agi_freecount = 0;
297                 agi->agi_newino = cpu_to_be32(NULLAGINO);
298                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
299                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
300                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
301                 error = xfs_bwrite(bp);
302                 xfs_buf_relse(bp);
303                 if (error)
304                         goto error0;
305
306                 /*
307                  * BNO btree root block
308                  */
309                 bp = xfs_growfs_get_hdr_buf(mp,
310                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
311                                 BTOBB(mp->m_sb.sb_blocksize), 0,
312                                 &xfs_allocbt_buf_ops);
313
314                 if (!bp) {
315                         error = ENOMEM;
316                         goto error0;
317                 }
318
319                 if (xfs_sb_version_hascrc(&mp->m_sb))
320                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
321                                                 agno, XFS_BTREE_CRC_BLOCKS);
322                 else
323                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
324                                                 agno, 0);
325
326                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
327                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
328                 arec->ar_blockcount = cpu_to_be32(
329                         agsize - be32_to_cpu(arec->ar_startblock));
330
331                 error = xfs_bwrite(bp);
332                 xfs_buf_relse(bp);
333                 if (error)
334                         goto error0;
335
336                 /*
337                  * CNT btree root block
338                  */
339                 bp = xfs_growfs_get_hdr_buf(mp,
340                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
341                                 BTOBB(mp->m_sb.sb_blocksize), 0,
342                                 &xfs_allocbt_buf_ops);
343                 if (!bp) {
344                         error = ENOMEM;
345                         goto error0;
346                 }
347
348                 if (xfs_sb_version_hascrc(&mp->m_sb))
349                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
350                                                 agno, XFS_BTREE_CRC_BLOCKS);
351                 else
352                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
353                                                 agno, 0);
354
355                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
356                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
357                 arec->ar_blockcount = cpu_to_be32(
358                         agsize - be32_to_cpu(arec->ar_startblock));
359                 nfree += be32_to_cpu(arec->ar_blockcount);
360
361                 error = xfs_bwrite(bp);
362                 xfs_buf_relse(bp);
363                 if (error)
364                         goto error0;
365
366                 /*
367                  * INO btree root block
368                  */
369                 bp = xfs_growfs_get_hdr_buf(mp,
370                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
371                                 BTOBB(mp->m_sb.sb_blocksize), 0,
372                                 &xfs_inobt_buf_ops);
373                 if (!bp) {
374                         error = ENOMEM;
375                         goto error0;
376                 }
377
378                 if (xfs_sb_version_hascrc(&mp->m_sb))
379                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
380                                                 agno, XFS_BTREE_CRC_BLOCKS);
381                 else
382                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
383                                                 agno, 0);
384
385                 error = xfs_bwrite(bp);
386                 xfs_buf_relse(bp);
387                 if (error)
388                         goto error0;
389         }
390         xfs_trans_agblocks_delta(tp, nfree);
391         /*
392          * There are new blocks in the old last a.g.
393          */
394         if (new) {
395                 /*
396                  * Change the agi length.
397                  */
398                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
399                 if (error) {
400                         goto error0;
401                 }
402                 ASSERT(bp);
403                 agi = XFS_BUF_TO_AGI(bp);
404                 be32_add_cpu(&agi->agi_length, new);
405                 ASSERT(nagcount == oagcount ||
406                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
407                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
408                 /*
409                  * Change agf length.
410                  */
411                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
412                 if (error) {
413                         goto error0;
414                 }
415                 ASSERT(bp);
416                 agf = XFS_BUF_TO_AGF(bp);
417                 be32_add_cpu(&agf->agf_length, new);
418                 ASSERT(be32_to_cpu(agf->agf_length) ==
419                        be32_to_cpu(agi->agi_length));
420
421                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
422                 /*
423                  * Free the new space.
424                  */
425                 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
426                         be32_to_cpu(agf->agf_length) - new), new);
427                 if (error) {
428                         goto error0;
429                 }
430         }
431
432         /*
433          * Update changed superblock fields transactionally. These are not
434          * seen by the rest of the world until the transaction commit applies
435          * them atomically to the superblock.
436          */
437         if (nagcount > oagcount)
438                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
439         if (nb > mp->m_sb.sb_dblocks)
440                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
441                                  nb - mp->m_sb.sb_dblocks);
442         if (nfree)
443                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
444         if (dpct)
445                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
446         error = xfs_trans_commit(tp, 0);
447         if (error)
448                 return error;
449
450         /* New allocation groups fully initialized, so update mount struct */
451         if (nagimax)
452                 mp->m_maxagi = nagimax;
453         if (mp->m_sb.sb_imax_pct) {
454                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
455                 do_div(icount, 100);
456                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
457         } else
458                 mp->m_maxicount = 0;
459         xfs_set_low_space_thresholds(mp);
460
461         /* update secondary superblocks. */
462         for (agno = 1; agno < nagcount; agno++) {
463                 error = 0;
464                 /*
465                  * new secondary superblocks need to be zeroed, not read from
466                  * disk as the contents of the new area we are growing into is
467                  * completely unknown.
468                  */
469                 if (agno < oagcount) {
470                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
471                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
472                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
473                                   &xfs_sb_buf_ops);
474                 } else {
475                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
476                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
477                                   XFS_FSS_TO_BB(mp, 1), 0);
478                         if (bp) {
479                                 bp->b_ops = &xfs_sb_buf_ops;
480                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
481                         } else
482                                 error = ENOMEM;
483                 }
484
485                 if (error) {
486                         xfs_warn(mp,
487                 "error %d reading secondary superblock for ag %d",
488                                 error, agno);
489                         break;
490                 }
491                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
492
493                 /*
494                  * If we get an error writing out the alternate superblocks,
495                  * just issue a warning and continue.  The real work is
496                  * already done and committed.
497                  */
498                 error = xfs_bwrite(bp);
499                 xfs_buf_relse(bp);
500                 if (error) {
501                         xfs_warn(mp,
502                 "write error %d updating secondary superblock for ag %d",
503                                 error, agno);
504                         break; /* no point in continuing */
505                 }
506         }
507         return error;
508
509  error0:
510         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
511         return error;
512 }
513
514 static int
515 xfs_growfs_log_private(
516         xfs_mount_t             *mp,    /* mount point for filesystem */
517         xfs_growfs_log_t        *in)    /* growfs log input struct */
518 {
519         xfs_extlen_t            nb;
520
521         nb = in->newblocks;
522         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
523                 return XFS_ERROR(EINVAL);
524         if (nb == mp->m_sb.sb_logblocks &&
525             in->isint == (mp->m_sb.sb_logstart != 0))
526                 return XFS_ERROR(EINVAL);
527         /*
528          * Moving the log is hard, need new interfaces to sync
529          * the log first, hold off all activity while moving it.
530          * Can have shorter or longer log in the same space,
531          * or transform internal to external log or vice versa.
532          */
533         return XFS_ERROR(ENOSYS);
534 }
535
536 /*
537  * protected versions of growfs function acquire and release locks on the mount
538  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
539  * XFS_IOC_FSGROWFSRT
540  */
541
542
543 int
544 xfs_growfs_data(
545         xfs_mount_t             *mp,
546         xfs_growfs_data_t       *in)
547 {
548         int error;
549
550         if (!capable(CAP_SYS_ADMIN))
551                 return XFS_ERROR(EPERM);
552         if (!mutex_trylock(&mp->m_growlock))
553                 return XFS_ERROR(EWOULDBLOCK);
554         error = xfs_growfs_data_private(mp, in);
555         mutex_unlock(&mp->m_growlock);
556         return error;
557 }
558
559 int
560 xfs_growfs_log(
561         xfs_mount_t             *mp,
562         xfs_growfs_log_t        *in)
563 {
564         int error;
565
566         if (!capable(CAP_SYS_ADMIN))
567                 return XFS_ERROR(EPERM);
568         if (!mutex_trylock(&mp->m_growlock))
569                 return XFS_ERROR(EWOULDBLOCK);
570         error = xfs_growfs_log_private(mp, in);
571         mutex_unlock(&mp->m_growlock);
572         return error;
573 }
574
575 /*
576  * exported through ioctl XFS_IOC_FSCOUNTS
577  */
578
579 int
580 xfs_fs_counts(
581         xfs_mount_t             *mp,
582         xfs_fsop_counts_t       *cnt)
583 {
584         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
585         spin_lock(&mp->m_sb_lock);
586         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
587         cnt->freertx = mp->m_sb.sb_frextents;
588         cnt->freeino = mp->m_sb.sb_ifree;
589         cnt->allocino = mp->m_sb.sb_icount;
590         spin_unlock(&mp->m_sb_lock);
591         return 0;
592 }
593
594 /*
595  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
596  *
597  * xfs_reserve_blocks is called to set m_resblks
598  * in the in-core mount table. The number of unused reserved blocks
599  * is kept in m_resblks_avail.
600  *
601  * Reserve the requested number of blocks if available. Otherwise return
602  * as many as possible to satisfy the request. The actual number
603  * reserved are returned in outval
604  *
605  * A null inval pointer indicates that only the current reserved blocks
606  * available  should  be returned no settings are changed.
607  */
608
609 int
610 xfs_reserve_blocks(
611         xfs_mount_t             *mp,
612         __uint64_t              *inval,
613         xfs_fsop_resblks_t      *outval)
614 {
615         __int64_t               lcounter, delta, fdblks_delta;
616         __uint64_t              request;
617
618         /* If inval is null, report current values and return */
619         if (inval == (__uint64_t *)NULL) {
620                 if (!outval)
621                         return EINVAL;
622                 outval->resblks = mp->m_resblks;
623                 outval->resblks_avail = mp->m_resblks_avail;
624                 return 0;
625         }
626
627         request = *inval;
628
629         /*
630          * With per-cpu counters, this becomes an interesting
631          * problem. we needto work out if we are freeing or allocation
632          * blocks first, then we can do the modification as necessary.
633          *
634          * We do this under the m_sb_lock so that if we are near
635          * ENOSPC, we will hold out any changes while we work out
636          * what to do. This means that the amount of free space can
637          * change while we do this, so we need to retry if we end up
638          * trying to reserve more space than is available.
639          *
640          * We also use the xfs_mod_incore_sb() interface so that we
641          * don't have to care about whether per cpu counter are
642          * enabled, disabled or even compiled in....
643          */
644 retry:
645         spin_lock(&mp->m_sb_lock);
646         xfs_icsb_sync_counters_locked(mp, 0);
647
648         /*
649          * If our previous reservation was larger than the current value,
650          * then move any unused blocks back to the free pool.
651          */
652         fdblks_delta = 0;
653         if (mp->m_resblks > request) {
654                 lcounter = mp->m_resblks_avail - request;
655                 if (lcounter  > 0) {            /* release unused blocks */
656                         fdblks_delta = lcounter;
657                         mp->m_resblks_avail -= lcounter;
658                 }
659                 mp->m_resblks = request;
660         } else {
661                 __int64_t       free;
662
663                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
664                 if (!free)
665                         goto out; /* ENOSPC and fdblks_delta = 0 */
666
667                 delta = request - mp->m_resblks;
668                 lcounter = free - delta;
669                 if (lcounter < 0) {
670                         /* We can't satisfy the request, just get what we can */
671                         mp->m_resblks += free;
672                         mp->m_resblks_avail += free;
673                         fdblks_delta = -free;
674                 } else {
675                         fdblks_delta = -delta;
676                         mp->m_resblks = request;
677                         mp->m_resblks_avail += delta;
678                 }
679         }
680 out:
681         if (outval) {
682                 outval->resblks = mp->m_resblks;
683                 outval->resblks_avail = mp->m_resblks_avail;
684         }
685         spin_unlock(&mp->m_sb_lock);
686
687         if (fdblks_delta) {
688                 /*
689                  * If we are putting blocks back here, m_resblks_avail is
690                  * already at its max so this will put it in the free pool.
691                  *
692                  * If we need space, we'll either succeed in getting it
693                  * from the free block count or we'll get an enospc. If
694                  * we get a ENOSPC, it means things changed while we were
695                  * calculating fdblks_delta and so we should try again to
696                  * see if there is anything left to reserve.
697                  *
698                  * Don't set the reserved flag here - we don't want to reserve
699                  * the extra reserve blocks from the reserve.....
700                  */
701                 int error;
702                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
703                                                  fdblks_delta, 0);
704                 if (error == ENOSPC)
705                         goto retry;
706         }
707         return 0;
708 }
709
710 /*
711  * Dump a transaction into the log that contains no real change. This is needed
712  * to be able to make the log dirty or stamp the current tail LSN into the log
713  * during the covering operation.
714  *
715  * We cannot use an inode here for this - that will push dirty state back up
716  * into the VFS and then periodic inode flushing will prevent log covering from
717  * making progress. Hence we log a field in the superblock instead and use a
718  * synchronous transaction to ensure the superblock is immediately unpinned
719  * and can be written back.
720  */
721 int
722 xfs_fs_log_dummy(
723         xfs_mount_t     *mp)
724 {
725         xfs_trans_t     *tp;
726         int             error;
727
728         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
729         error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0,
730                                   XFS_DEFAULT_LOG_COUNT);
731         if (error) {
732                 xfs_trans_cancel(tp, 0);
733                 return error;
734         }
735
736         /* log the UUID because it is an unchanging field */
737         xfs_mod_sb(tp, XFS_SB_UUID);
738         xfs_trans_set_sync(tp);
739         return xfs_trans_commit(tp, 0);
740 }
741
742 int
743 xfs_fs_goingdown(
744         xfs_mount_t     *mp,
745         __uint32_t      inflags)
746 {
747         switch (inflags) {
748         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
749                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
750
751                 if (sb && !IS_ERR(sb)) {
752                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
753                         thaw_bdev(sb->s_bdev, sb);
754                 }
755
756                 break;
757         }
758         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
759                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
760                 break;
761         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
762                 xfs_force_shutdown(mp,
763                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
764                 break;
765         default:
766                 return XFS_ERROR(EINVAL);
767         }
768
769         return 0;
770 }
771
772 /*
773  * Force a shutdown of the filesystem instantly while keeping the filesystem
774  * consistent. We don't do an unmount here; just shutdown the shop, make sure
775  * that absolutely nothing persistent happens to this filesystem after this
776  * point.
777  */
778 void
779 xfs_do_force_shutdown(
780         xfs_mount_t     *mp,
781         int             flags,
782         char            *fname,
783         int             lnnum)
784 {
785         int             logerror;
786
787         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
788
789         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
790                 xfs_notice(mp,
791         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
792                         __func__, flags, lnnum, fname, __return_address);
793         }
794         /*
795          * No need to duplicate efforts.
796          */
797         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
798                 return;
799
800         /*
801          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
802          * queue up anybody new on the log reservations, and wakes up
803          * everybody who's sleeping on log reservations to tell them
804          * the bad news.
805          */
806         if (xfs_log_force_umount(mp, logerror))
807                 return;
808
809         if (flags & SHUTDOWN_CORRUPT_INCORE) {
810                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
811     "Corruption of in-memory data detected.  Shutting down filesystem");
812                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
813                         xfs_stack_trace();
814         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
815                 if (logerror) {
816                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
817                 "Log I/O Error Detected.  Shutting down filesystem");
818                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
819                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
820                 "All device paths lost.  Shutting down filesystem");
821                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
822                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
823                 "I/O Error Detected. Shutting down filesystem");
824                 }
825         }
826         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
827                 xfs_alert(mp,
828         "Please umount the filesystem and rectify the problem(s)");
829         }
830 }