Merge branch 'timers-for-linus-ntp' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6] / fs / xfs / xfs_iomap.c
1 /*
2  * Copyright (c) 2000-2006 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_bit.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_dir2.h"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_btree.h"
40 #include "xfs_bmap.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_itable.h"
44 #include "xfs_rw.h"
45 #include "xfs_attr.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_trans_space.h"
48 #include "xfs_utils.h"
49 #include "xfs_iomap.h"
50
51 #if defined(XFS_RW_TRACE)
52 void
53 xfs_iomap_enter_trace(
54         int             tag,
55         xfs_inode_t     *ip,
56         xfs_off_t       offset,
57         ssize_t         count)
58 {
59         if (!ip->i_rwtrace)
60                 return;
61
62         ktrace_enter(ip->i_rwtrace,
63                 (void *)((unsigned long)tag),
64                 (void *)ip,
65                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
66                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
67                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
68                 (void *)((unsigned long)(offset & 0xffffffff)),
69                 (void *)((unsigned long)count),
70                 (void *)((unsigned long)((ip->i_new_size >> 32) & 0xffffffff)),
71                 (void *)((unsigned long)(ip->i_new_size & 0xffffffff)),
72                 (void *)((unsigned long)current_pid()),
73                 (void *)NULL,
74                 (void *)NULL,
75                 (void *)NULL,
76                 (void *)NULL,
77                 (void *)NULL,
78                 (void *)NULL);
79 }
80
81 void
82 xfs_iomap_map_trace(
83         int             tag,
84         xfs_inode_t     *ip,
85         xfs_off_t       offset,
86         ssize_t         count,
87         xfs_iomap_t     *iomapp,
88         xfs_bmbt_irec_t *imapp,
89         int             flags)
90 {
91         if (!ip->i_rwtrace)
92                 return;
93
94         ktrace_enter(ip->i_rwtrace,
95                 (void *)((unsigned long)tag),
96                 (void *)ip,
97                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
98                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
99                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
100                 (void *)((unsigned long)(offset & 0xffffffff)),
101                 (void *)((unsigned long)count),
102                 (void *)((unsigned long)flags),
103                 (void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
104                 (void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
105                 (void *)((unsigned long)(iomapp->iomap_delta)),
106                 (void *)((unsigned long)(iomapp->iomap_bsize)),
107                 (void *)((unsigned long)(iomapp->iomap_bn)),
108                 (void *)(__psint_t)(imapp->br_startoff),
109                 (void *)((unsigned long)(imapp->br_blockcount)),
110                 (void *)(__psint_t)(imapp->br_startblock));
111 }
112 #else
113 #define xfs_iomap_enter_trace(tag, io, offset, count)
114 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
115 #endif
116
117 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
118                                                 << mp->m_writeio_log)
119 #define XFS_STRAT_WRITE_IMAPS   2
120 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
121
122 STATIC int
123 xfs_imap_to_bmap(
124         xfs_inode_t     *ip,
125         xfs_off_t       offset,
126         xfs_bmbt_irec_t *imap,
127         xfs_iomap_t     *iomapp,
128         int             imaps,                  /* Number of imap entries */
129         int             iomaps,                 /* Number of iomap entries */
130         int             flags)
131 {
132         xfs_mount_t     *mp = ip->i_mount;
133         int             pbm;
134         xfs_fsblock_t   start_block;
135
136
137         for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
138                 iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
139                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
140                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
141                 iomapp->iomap_flags = flags;
142
143                 if (XFS_IS_REALTIME_INODE(ip)) {
144                         iomapp->iomap_flags |= IOMAP_REALTIME;
145                         iomapp->iomap_target = mp->m_rtdev_targp;
146                 } else {
147                         iomapp->iomap_target = mp->m_ddev_targp;
148                 }
149                 start_block = imap->br_startblock;
150                 if (start_block == HOLESTARTBLOCK) {
151                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
152                         iomapp->iomap_flags |= IOMAP_HOLE;
153                 } else if (start_block == DELAYSTARTBLOCK) {
154                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
155                         iomapp->iomap_flags |= IOMAP_DELAY;
156                 } else {
157                         iomapp->iomap_bn = xfs_fsb_to_db(ip, start_block);
158                         if (ISUNWRITTEN(imap))
159                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
160                 }
161
162                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
163         }
164         return pbm;     /* Return the number filled */
165 }
166
167 int
168 xfs_iomap(
169         xfs_inode_t     *ip,
170         xfs_off_t       offset,
171         ssize_t         count,
172         int             flags,
173         xfs_iomap_t     *iomapp,
174         int             *niomaps)
175 {
176         xfs_mount_t     *mp = ip->i_mount;
177         xfs_fileoff_t   offset_fsb, end_fsb;
178         int             error = 0;
179         int             lockmode = 0;
180         xfs_bmbt_irec_t imap;
181         int             nimaps = 1;
182         int             bmapi_flags = 0;
183         int             iomap_flags = 0;
184
185         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
186
187         if (XFS_FORCED_SHUTDOWN(mp))
188                 return XFS_ERROR(EIO);
189
190         switch (flags & (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE)) {
191         case BMAPI_READ:
192                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, ip, offset, count);
193                 lockmode = xfs_ilock_map_shared(ip);
194                 bmapi_flags = XFS_BMAPI_ENTIRE;
195                 break;
196         case BMAPI_WRITE:
197                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, ip, offset, count);
198                 lockmode = XFS_ILOCK_EXCL;
199                 if (flags & BMAPI_IGNSTATE)
200                         bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
201                 xfs_ilock(ip, lockmode);
202                 break;
203         case BMAPI_ALLOCATE:
204                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, ip, offset, count);
205                 lockmode = XFS_ILOCK_SHARED;
206                 bmapi_flags = XFS_BMAPI_ENTIRE;
207
208                 /* Attempt non-blocking lock */
209                 if (flags & BMAPI_TRYLOCK) {
210                         if (!xfs_ilock_nowait(ip, lockmode))
211                                 return XFS_ERROR(EAGAIN);
212                 } else {
213                         xfs_ilock(ip, lockmode);
214                 }
215                 break;
216         default:
217                 BUG();
218         }
219
220         ASSERT(offset <= mp->m_maxioffset);
221         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
222                 count = mp->m_maxioffset - offset;
223         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
224         offset_fsb = XFS_B_TO_FSBT(mp, offset);
225
226         error = xfs_bmapi(NULL, ip, offset_fsb,
227                         (xfs_filblks_t)(end_fsb - offset_fsb),
228                         bmapi_flags,  NULL, 0, &imap,
229                         &nimaps, NULL, NULL);
230
231         if (error)
232                 goto out;
233
234         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)) {
235         case BMAPI_WRITE:
236                 /* If we found an extent, return it */
237                 if (nimaps &&
238                     (imap.br_startblock != HOLESTARTBLOCK) &&
239                     (imap.br_startblock != DELAYSTARTBLOCK)) {
240                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, ip,
241                                         offset, count, iomapp, &imap, flags);
242                         break;
243                 }
244
245                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
246                         error = xfs_iomap_write_direct(ip, offset, count, flags,
247                                                        &imap, &nimaps, nimaps);
248                 } else {
249                         error = xfs_iomap_write_delay(ip, offset, count, flags,
250                                                       &imap, &nimaps);
251                 }
252                 if (!error) {
253                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, ip,
254                                         offset, count, iomapp, &imap, flags);
255                 }
256                 iomap_flags = IOMAP_NEW;
257                 break;
258         case BMAPI_ALLOCATE:
259                 /* If we found an extent, return it */
260                 xfs_iunlock(ip, lockmode);
261                 lockmode = 0;
262
263                 if (nimaps && !isnullstartblock(imap.br_startblock)) {
264                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, ip,
265                                         offset, count, iomapp, &imap, flags);
266                         break;
267                 }
268
269                 error = xfs_iomap_write_allocate(ip, offset, count,
270                                                  &imap, &nimaps);
271                 break;
272         }
273
274         if (nimaps) {
275                 *niomaps = xfs_imap_to_bmap(ip, offset, &imap,
276                                             iomapp, nimaps, *niomaps, iomap_flags);
277         } else if (niomaps) {
278                 *niomaps = 0;
279         }
280
281 out:
282         if (lockmode)
283                 xfs_iunlock(ip, lockmode);
284         return XFS_ERROR(error);
285 }
286
287
288 STATIC int
289 xfs_iomap_eof_align_last_fsb(
290         xfs_mount_t     *mp,
291         xfs_inode_t     *ip,
292         xfs_extlen_t    extsize,
293         xfs_fileoff_t   *last_fsb)
294 {
295         xfs_fileoff_t   new_last_fsb = 0;
296         xfs_extlen_t    align;
297         int             eof, error;
298
299         if (XFS_IS_REALTIME_INODE(ip))
300                 ;
301         /*
302          * If mounted with the "-o swalloc" option, roundup the allocation
303          * request to a stripe width boundary if the file size is >=
304          * stripe width and we are allocating past the allocation eof.
305          */
306         else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
307                 (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_swidth)))
308                 new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
309         /*
310          * Roundup the allocation request to a stripe unit (m_dalign) boundary
311          * if the file size is >= stripe unit size, and we are allocating past
312          * the allocation eof.
313          */
314         else if (mp->m_dalign && (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_dalign)))
315                 new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
316
317         /*
318          * Always round up the allocation request to an extent boundary
319          * (when file on a real-time subvolume or has di_extsize hint).
320          */
321         if (extsize) {
322                 if (new_last_fsb)
323                         align = roundup_64(new_last_fsb, extsize);
324                 else
325                         align = extsize;
326                 new_last_fsb = roundup_64(*last_fsb, align);
327         }
328
329         if (new_last_fsb) {
330                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
331                 if (error)
332                         return error;
333                 if (eof)
334                         *last_fsb = new_last_fsb;
335         }
336         return 0;
337 }
338
339 STATIC int
340 xfs_cmn_err_fsblock_zero(
341         xfs_inode_t     *ip,
342         xfs_bmbt_irec_t *imap)
343 {
344         xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
345                         "Access to block zero in inode %llu "
346                         "start_block: %llx start_off: %llx "
347                         "blkcnt: %llx extent-state: %x\n",
348                 (unsigned long long)ip->i_ino,
349                 (unsigned long long)imap->br_startblock,
350                 (unsigned long long)imap->br_startoff,
351                 (unsigned long long)imap->br_blockcount,
352                 imap->br_state);
353         return EFSCORRUPTED;
354 }
355
356 int
357 xfs_iomap_write_direct(
358         xfs_inode_t     *ip,
359         xfs_off_t       offset,
360         size_t          count,
361         int             flags,
362         xfs_bmbt_irec_t *ret_imap,
363         int             *nmaps,
364         int             found)
365 {
366         xfs_mount_t     *mp = ip->i_mount;
367         xfs_fileoff_t   offset_fsb;
368         xfs_fileoff_t   last_fsb;
369         xfs_filblks_t   count_fsb, resaligned;
370         xfs_fsblock_t   firstfsb;
371         xfs_extlen_t    extsz, temp;
372         int             nimaps;
373         int             bmapi_flag;
374         int             quota_flag;
375         int             rt;
376         xfs_trans_t     *tp;
377         xfs_bmbt_irec_t imap;
378         xfs_bmap_free_t free_list;
379         uint            qblocks, resblks, resrtextents;
380         int             committed;
381         int             error;
382
383         /*
384          * Make sure that the dquots are there. This doesn't hold
385          * the ilock across a disk read.
386          */
387         error = xfs_qm_dqattach_locked(ip, 0);
388         if (error)
389                 return XFS_ERROR(error);
390
391         rt = XFS_IS_REALTIME_INODE(ip);
392         extsz = xfs_get_extsz_hint(ip);
393
394         offset_fsb = XFS_B_TO_FSBT(mp, offset);
395         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
396         if ((offset + count) > ip->i_size) {
397                 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
398                 if (error)
399                         goto error_out;
400         } else {
401                 if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
402                         last_fsb = MIN(last_fsb, (xfs_fileoff_t)
403                                         ret_imap->br_blockcount +
404                                         ret_imap->br_startoff);
405         }
406         count_fsb = last_fsb - offset_fsb;
407         ASSERT(count_fsb > 0);
408
409         resaligned = count_fsb;
410         if (unlikely(extsz)) {
411                 if ((temp = do_mod(offset_fsb, extsz)))
412                         resaligned += temp;
413                 if ((temp = do_mod(resaligned, extsz)))
414                         resaligned += extsz - temp;
415         }
416
417         if (unlikely(rt)) {
418                 resrtextents = qblocks = resaligned;
419                 resrtextents /= mp->m_sb.sb_rextsize;
420                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
421                 quota_flag = XFS_QMOPT_RES_RTBLKS;
422         } else {
423                 resrtextents = 0;
424                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
425                 quota_flag = XFS_QMOPT_RES_REGBLKS;
426         }
427
428         /*
429          * Allocate and setup the transaction
430          */
431         xfs_iunlock(ip, XFS_ILOCK_EXCL);
432         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
433         error = xfs_trans_reserve(tp, resblks,
434                         XFS_WRITE_LOG_RES(mp), resrtextents,
435                         XFS_TRANS_PERM_LOG_RES,
436                         XFS_WRITE_LOG_COUNT);
437         /*
438          * Check for running out of space, note: need lock to return
439          */
440         if (error)
441                 xfs_trans_cancel(tp, 0);
442         xfs_ilock(ip, XFS_ILOCK_EXCL);
443         if (error)
444                 goto error_out;
445
446         error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
447         if (error)
448                 goto error1;
449
450         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
451         xfs_trans_ihold(tp, ip);
452
453         bmapi_flag = XFS_BMAPI_WRITE;
454         if ((flags & BMAPI_DIRECT) && (offset < ip->i_size || extsz))
455                 bmapi_flag |= XFS_BMAPI_PREALLOC;
456
457         /*
458          * Issue the xfs_bmapi() call to allocate the blocks
459          */
460         xfs_bmap_init(&free_list, &firstfsb);
461         nimaps = 1;
462         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb, bmapi_flag,
463                 &firstfsb, 0, &imap, &nimaps, &free_list, NULL);
464         if (error)
465                 goto error0;
466
467         /*
468          * Complete the transaction
469          */
470         error = xfs_bmap_finish(&tp, &free_list, &committed);
471         if (error)
472                 goto error0;
473         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
474         if (error)
475                 goto error_out;
476
477         /*
478          * Copy any maps to caller's array and return any error.
479          */
480         if (nimaps == 0) {
481                 error = ENOSPC;
482                 goto error_out;
483         }
484
485         if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) {
486                 error = xfs_cmn_err_fsblock_zero(ip, &imap);
487                 goto error_out;
488         }
489
490         *ret_imap = imap;
491         *nmaps = 1;
492         return 0;
493
494 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
495         xfs_bmap_cancel(&free_list);
496         xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
497
498 error1: /* Just cancel transaction */
499         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
500         *nmaps = 0;     /* nothing set-up here */
501
502 error_out:
503         return XFS_ERROR(error);
504 }
505
506 /*
507  * If the caller is doing a write at the end of the file, then extend the
508  * allocation out to the file system's write iosize.  We clean up any extra
509  * space left over when the file is closed in xfs_inactive().
510  */
511 STATIC int
512 xfs_iomap_eof_want_preallocate(
513         xfs_mount_t     *mp,
514         xfs_inode_t     *ip,
515         xfs_off_t       offset,
516         size_t          count,
517         int             ioflag,
518         xfs_bmbt_irec_t *imap,
519         int             nimaps,
520         int             *prealloc)
521 {
522         xfs_fileoff_t   start_fsb;
523         xfs_filblks_t   count_fsb;
524         xfs_fsblock_t   firstblock;
525         int             n, error, imaps;
526
527         *prealloc = 0;
528         if ((offset + count) <= ip->i_size)
529                 return 0;
530
531         /*
532          * If there are any real blocks past eof, then don't
533          * do any speculative allocation.
534          */
535         start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
536         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
537         while (count_fsb > 0) {
538                 imaps = nimaps;
539                 firstblock = NULLFSBLOCK;
540                 error = xfs_bmapi(NULL, ip, start_fsb, count_fsb, 0,
541                                   &firstblock, 0, imap, &imaps, NULL, NULL);
542                 if (error)
543                         return error;
544                 for (n = 0; n < imaps; n++) {
545                         if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
546                             (imap[n].br_startblock != DELAYSTARTBLOCK))
547                                 return 0;
548                         start_fsb += imap[n].br_blockcount;
549                         count_fsb -= imap[n].br_blockcount;
550                 }
551         }
552         *prealloc = 1;
553         return 0;
554 }
555
556 int
557 xfs_iomap_write_delay(
558         xfs_inode_t     *ip,
559         xfs_off_t       offset,
560         size_t          count,
561         int             ioflag,
562         xfs_bmbt_irec_t *ret_imap,
563         int             *nmaps)
564 {
565         xfs_mount_t     *mp = ip->i_mount;
566         xfs_fileoff_t   offset_fsb;
567         xfs_fileoff_t   last_fsb;
568         xfs_off_t       aligned_offset;
569         xfs_fileoff_t   ioalign;
570         xfs_fsblock_t   firstblock;
571         xfs_extlen_t    extsz;
572         int             nimaps;
573         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
574         int             prealloc, flushed = 0;
575         int             error;
576
577         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
578
579         /*
580          * Make sure that the dquots are there. This doesn't hold
581          * the ilock across a disk read.
582          */
583         error = xfs_qm_dqattach_locked(ip, 0);
584         if (error)
585                 return XFS_ERROR(error);
586
587         extsz = xfs_get_extsz_hint(ip);
588         offset_fsb = XFS_B_TO_FSBT(mp, offset);
589
590         error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
591                                 ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
592         if (error)
593                 return error;
594
595 retry:
596         if (prealloc) {
597                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
598                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
599                 last_fsb = ioalign + mp->m_writeio_blocks;
600         } else {
601                 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
602         }
603
604         if (prealloc || extsz) {
605                 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
606                 if (error)
607                         return error;
608         }
609
610         nimaps = XFS_WRITE_IMAPS;
611         firstblock = NULLFSBLOCK;
612         error = xfs_bmapi(NULL, ip, offset_fsb,
613                           (xfs_filblks_t)(last_fsb - offset_fsb),
614                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
615                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
616                           &nimaps, NULL, NULL);
617         if (error && (error != ENOSPC))
618                 return XFS_ERROR(error);
619
620         /*
621          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
622          * then we must have run out of space - flush all other inodes with
623          * delalloc blocks and retry without EOF preallocation.
624          */
625         if (nimaps == 0) {
626                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
627                                         ip, offset, count);
628                 if (flushed)
629                         return XFS_ERROR(ENOSPC);
630
631                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
632                 xfs_flush_inodes(ip);
633                 xfs_ilock(ip, XFS_ILOCK_EXCL);
634
635                 flushed = 1;
636                 error = 0;
637                 prealloc = 0;
638                 goto retry;
639         }
640
641         if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
642                 return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
643
644         *ret_imap = imap[0];
645         *nmaps = 1;
646
647         return 0;
648 }
649
650 /*
651  * Pass in a delayed allocate extent, convert it to real extents;
652  * return to the caller the extent we create which maps on top of
653  * the originating callers request.
654  *
655  * Called without a lock on the inode.
656  *
657  * We no longer bother to look at the incoming map - all we have to
658  * guarantee is that whatever we allocate fills the required range.
659  */
660 int
661 xfs_iomap_write_allocate(
662         xfs_inode_t     *ip,
663         xfs_off_t       offset,
664         size_t          count,
665         xfs_bmbt_irec_t *map,
666         int             *retmap)
667 {
668         xfs_mount_t     *mp = ip->i_mount;
669         xfs_fileoff_t   offset_fsb, last_block;
670         xfs_fileoff_t   end_fsb, map_start_fsb;
671         xfs_fsblock_t   first_block;
672         xfs_bmap_free_t free_list;
673         xfs_filblks_t   count_fsb;
674         xfs_bmbt_irec_t imap;
675         xfs_trans_t     *tp;
676         int             nimaps, committed;
677         int             error = 0;
678         int             nres;
679
680         *retmap = 0;
681
682         /*
683          * Make sure that the dquots are there.
684          */
685         error = xfs_qm_dqattach(ip, 0);
686         if (error)
687                 return XFS_ERROR(error);
688
689         offset_fsb = XFS_B_TO_FSBT(mp, offset);
690         count_fsb = map->br_blockcount;
691         map_start_fsb = map->br_startoff;
692
693         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
694
695         while (count_fsb != 0) {
696                 /*
697                  * Set up a transaction with which to allocate the
698                  * backing store for the file.  Do allocations in a
699                  * loop until we get some space in the range we are
700                  * interested in.  The other space that might be allocated
701                  * is in the delayed allocation extent on which we sit
702                  * but before our buffer starts.
703                  */
704
705                 nimaps = 0;
706                 while (nimaps == 0) {
707                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
708                         tp->t_flags |= XFS_TRANS_RESERVE;
709                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
710                         error = xfs_trans_reserve(tp, nres,
711                                         XFS_WRITE_LOG_RES(mp),
712                                         0, XFS_TRANS_PERM_LOG_RES,
713                                         XFS_WRITE_LOG_COUNT);
714                         if (error) {
715                                 xfs_trans_cancel(tp, 0);
716                                 return XFS_ERROR(error);
717                         }
718                         xfs_ilock(ip, XFS_ILOCK_EXCL);
719                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
720                         xfs_trans_ihold(tp, ip);
721
722                         xfs_bmap_init(&free_list, &first_block);
723
724                         /*
725                          * it is possible that the extents have changed since
726                          * we did the read call as we dropped the ilock for a
727                          * while. We have to be careful about truncates or hole
728                          * punchs here - we are not allowed to allocate
729                          * non-delalloc blocks here.
730                          *
731                          * The only protection against truncation is the pages
732                          * for the range we are being asked to convert are
733                          * locked and hence a truncate will block on them
734                          * first.
735                          *
736                          * As a result, if we go beyond the range we really
737                          * need and hit an delalloc extent boundary followed by
738                          * a hole while we have excess blocks in the map, we
739                          * will fill the hole incorrectly and overrun the
740                          * transaction reservation.
741                          *
742                          * Using a single map prevents this as we are forced to
743                          * check each map we look for overlap with the desired
744                          * range and abort as soon as we find it. Also, given
745                          * that we only return a single map, having one beyond
746                          * what we can return is probably a bit silly.
747                          *
748                          * We also need to check that we don't go beyond EOF;
749                          * this is a truncate optimisation as a truncate sets
750                          * the new file size before block on the pages we
751                          * currently have locked under writeback. Because they
752                          * are about to be tossed, we don't need to write them
753                          * back....
754                          */
755                         nimaps = 1;
756                         end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
757                         error = xfs_bmap_last_offset(NULL, ip, &last_block,
758                                                         XFS_DATA_FORK);
759                         if (error)
760                                 goto trans_cancel;
761
762                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
763                         if ((map_start_fsb + count_fsb) > last_block) {
764                                 count_fsb = last_block - map_start_fsb;
765                                 if (count_fsb == 0) {
766                                         error = EAGAIN;
767                                         goto trans_cancel;
768                                 }
769                         }
770
771                         /* Go get the actual blocks */
772                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
773                                         XFS_BMAPI_WRITE, &first_block, 1,
774                                         &imap, &nimaps, &free_list, NULL);
775                         if (error)
776                                 goto trans_cancel;
777
778                         error = xfs_bmap_finish(&tp, &free_list, &committed);
779                         if (error)
780                                 goto trans_cancel;
781
782                         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
783                         if (error)
784                                 goto error0;
785
786                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
787                 }
788
789                 /*
790                  * See if we were able to allocate an extent that
791                  * covers at least part of the callers request
792                  */
793                 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
794                         return xfs_cmn_err_fsblock_zero(ip, &imap);
795
796                 if ((offset_fsb >= imap.br_startoff) &&
797                     (offset_fsb < (imap.br_startoff +
798                                    imap.br_blockcount))) {
799                         *map = imap;
800                         *retmap = 1;
801                         XFS_STATS_INC(xs_xstrat_quick);
802                         return 0;
803                 }
804
805                 /*
806                  * So far we have not mapped the requested part of the
807                  * file, just surrounding data, try again.
808                  */
809                 count_fsb -= imap.br_blockcount;
810                 map_start_fsb = imap.br_startoff + imap.br_blockcount;
811         }
812
813 trans_cancel:
814         xfs_bmap_cancel(&free_list);
815         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
816 error0:
817         xfs_iunlock(ip, XFS_ILOCK_EXCL);
818         return XFS_ERROR(error);
819 }
820
821 int
822 xfs_iomap_write_unwritten(
823         xfs_inode_t     *ip,
824         xfs_off_t       offset,
825         size_t          count)
826 {
827         xfs_mount_t     *mp = ip->i_mount;
828         xfs_fileoff_t   offset_fsb;
829         xfs_filblks_t   count_fsb;
830         xfs_filblks_t   numblks_fsb;
831         xfs_fsblock_t   firstfsb;
832         int             nimaps;
833         xfs_trans_t     *tp;
834         xfs_bmbt_irec_t imap;
835         xfs_bmap_free_t free_list;
836         uint            resblks;
837         int             committed;
838         int             error;
839
840         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN, ip, offset, count);
841
842         offset_fsb = XFS_B_TO_FSBT(mp, offset);
843         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
844         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
845
846         /*
847          * Reserve enough blocks in this transaction for two complete extent
848          * btree splits.  We may be converting the middle part of an unwritten
849          * extent and in this case we will insert two new extents in the btree
850          * each of which could cause a full split.
851          *
852          * This reservation amount will be used in the first call to
853          * xfs_bmbt_split() to select an AG with enough space to satisfy the
854          * rest of the operation.
855          */
856         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
857
858         do {
859                 /*
860                  * set up a transaction to convert the range of extents
861                  * from unwritten to real. Do allocations in a loop until
862                  * we have covered the range passed in.
863                  */
864                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
865                 tp->t_flags |= XFS_TRANS_RESERVE;
866                 error = xfs_trans_reserve(tp, resblks,
867                                 XFS_WRITE_LOG_RES(mp), 0,
868                                 XFS_TRANS_PERM_LOG_RES,
869                                 XFS_WRITE_LOG_COUNT);
870                 if (error) {
871                         xfs_trans_cancel(tp, 0);
872                         return XFS_ERROR(error);
873                 }
874
875                 xfs_ilock(ip, XFS_ILOCK_EXCL);
876                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
877                 xfs_trans_ihold(tp, ip);
878
879                 /*
880                  * Modify the unwritten extent state of the buffer.
881                  */
882                 xfs_bmap_init(&free_list, &firstfsb);
883                 nimaps = 1;
884                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
885                                   XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
886                                   1, &imap, &nimaps, &free_list, NULL);
887                 if (error)
888                         goto error_on_bmapi_transaction;
889
890                 error = xfs_bmap_finish(&(tp), &(free_list), &committed);
891                 if (error)
892                         goto error_on_bmapi_transaction;
893
894                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
895                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
896                 if (error)
897                         return XFS_ERROR(error);
898
899                 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
900                         return xfs_cmn_err_fsblock_zero(ip, &imap);
901
902                 if ((numblks_fsb = imap.br_blockcount) == 0) {
903                         /*
904                          * The numblks_fsb value should always get
905                          * smaller, otherwise the loop is stuck.
906                          */
907                         ASSERT(imap.br_blockcount);
908                         break;
909                 }
910                 offset_fsb += numblks_fsb;
911                 count_fsb -= numblks_fsb;
912         } while (count_fsb > 0);
913
914         return 0;
915
916 error_on_bmapi_transaction:
917         xfs_bmap_cancel(&free_list);
918         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
919         xfs_iunlock(ip, XFS_ILOCK_EXCL);
920         return XFS_ERROR(error);
921 }