Pull bugfix into test branch
[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_acl.h"
46 #include "xfs_cap.h"
47 #include "xfs_mac.h"
48 #include "xfs_attr.h"
49 #include "xfs_buf_item.h"
50 #include "xfs_trans_space.h"
51 #include "xfs_utils.h"
52 #include "xfs_iomap.h"
53
54 #if defined(XFS_RW_TRACE)
55 void
56 xfs_iomap_enter_trace(
57         int             tag,
58         xfs_iocore_t    *io,
59         xfs_off_t       offset,
60         ssize_t         count)
61 {
62         xfs_inode_t     *ip = XFS_IO_INODE(io);
63
64         if (!ip->i_rwtrace)
65                 return;
66
67         ktrace_enter(ip->i_rwtrace,
68                 (void *)((unsigned long)tag),
69                 (void *)ip,
70                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
71                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
72                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
73                 (void *)((unsigned long)(offset & 0xffffffff)),
74                 (void *)((unsigned long)count),
75                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
76                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
77                 (void *)((unsigned long)current_pid()),
78                 (void *)NULL,
79                 (void *)NULL,
80                 (void *)NULL,
81                 (void *)NULL,
82                 (void *)NULL,
83                 (void *)NULL);
84 }
85
86 void
87 xfs_iomap_map_trace(
88         int             tag,
89         xfs_iocore_t    *io,
90         xfs_off_t       offset,
91         ssize_t         count,
92         xfs_iomap_t     *iomapp,
93         xfs_bmbt_irec_t *imapp,
94         int             flags)
95 {
96         xfs_inode_t     *ip = XFS_IO_INODE(io);
97
98         if (!ip->i_rwtrace)
99                 return;
100
101         ktrace_enter(ip->i_rwtrace,
102                 (void *)((unsigned long)tag),
103                 (void *)ip,
104                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
105                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
106                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
107                 (void *)((unsigned long)(offset & 0xffffffff)),
108                 (void *)((unsigned long)count),
109                 (void *)((unsigned long)flags),
110                 (void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
111                 (void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
112                 (void *)((unsigned long)(iomapp->iomap_delta)),
113                 (void *)((unsigned long)(iomapp->iomap_bsize)),
114                 (void *)((unsigned long)(iomapp->iomap_bn)),
115                 (void *)(__psint_t)(imapp->br_startoff),
116                 (void *)((unsigned long)(imapp->br_blockcount)),
117                 (void *)(__psint_t)(imapp->br_startblock));
118 }
119 #else
120 #define xfs_iomap_enter_trace(tag, io, offset, count)
121 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
122 #endif
123
124 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
125                                                 << mp->m_writeio_log)
126 #define XFS_STRAT_WRITE_IMAPS   2
127 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
128
129 STATIC int
130 xfs_imap_to_bmap(
131         xfs_iocore_t    *io,
132         xfs_off_t       offset,
133         xfs_bmbt_irec_t *imap,
134         xfs_iomap_t     *iomapp,
135         int             imaps,                  /* Number of imap entries */
136         int             iomaps,                 /* Number of iomap entries */
137         int             flags)
138 {
139         xfs_mount_t     *mp;
140         xfs_fsize_t     nisize;
141         int             pbm;
142         xfs_fsblock_t   start_block;
143
144         mp = io->io_mount;
145         nisize = XFS_SIZE(mp, io);
146         if (io->io_new_size > nisize)
147                 nisize = io->io_new_size;
148
149         for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
150                 iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
151                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
152                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
153                 iomapp->iomap_flags = flags;
154
155                 if (io->io_flags & XFS_IOCORE_RT) {
156                         iomapp->iomap_flags |= IOMAP_REALTIME;
157                         iomapp->iomap_target = mp->m_rtdev_targp;
158                 } else {
159                         iomapp->iomap_target = mp->m_ddev_targp;
160                 }
161                 start_block = imap->br_startblock;
162                 if (start_block == HOLESTARTBLOCK) {
163                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
164                         iomapp->iomap_flags |= IOMAP_HOLE;
165                 } else if (start_block == DELAYSTARTBLOCK) {
166                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
167                         iomapp->iomap_flags |= IOMAP_DELAY;
168                 } else {
169                         iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
170                         if (ISUNWRITTEN(imap))
171                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
172                 }
173
174                 if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
175                         iomapp->iomap_flags |= IOMAP_EOF;
176                 }
177
178                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
179         }
180         return pbm;     /* Return the number filled */
181 }
182
183 int
184 xfs_iomap(
185         xfs_iocore_t    *io,
186         xfs_off_t       offset,
187         ssize_t         count,
188         int             flags,
189         xfs_iomap_t     *iomapp,
190         int             *niomaps)
191 {
192         xfs_mount_t     *mp = io->io_mount;
193         xfs_fileoff_t   offset_fsb, end_fsb;
194         int             error = 0;
195         int             lockmode = 0;
196         xfs_bmbt_irec_t imap;
197         int             nimaps = 1;
198         int             bmapi_flags = 0;
199         int             iomap_flags = 0;
200
201         if (XFS_FORCED_SHUTDOWN(mp))
202                 return XFS_ERROR(EIO);
203
204         switch (flags &
205                 (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
206                  BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
207         case BMAPI_READ:
208                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
209                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
210                 bmapi_flags = XFS_BMAPI_ENTIRE;
211                 break;
212         case BMAPI_WRITE:
213                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
214                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
215                 if (flags & BMAPI_IGNSTATE)
216                         bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
217                 XFS_ILOCK(mp, io, lockmode);
218                 break;
219         case BMAPI_ALLOCATE:
220                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
221                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
222                 bmapi_flags = XFS_BMAPI_ENTIRE;
223                 /* Attempt non-blocking lock */
224                 if (flags & BMAPI_TRYLOCK) {
225                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
226                                 return XFS_ERROR(EAGAIN);
227                 } else {
228                         XFS_ILOCK(mp, io, lockmode);
229                 }
230                 break;
231         case BMAPI_UNWRITTEN:
232                 goto phase2;
233         case BMAPI_DEVICE:
234                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
235                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
236                         mp->m_rtdev_targp : mp->m_ddev_targp;
237                 error = 0;
238                 *niomaps = 1;
239                 goto out;
240         default:
241                 BUG();
242         }
243
244         ASSERT(offset <= mp->m_maxioffset);
245         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
246                 count = mp->m_maxioffset - offset;
247         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
248         offset_fsb = XFS_B_TO_FSBT(mp, offset);
249
250         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
251                         (xfs_filblks_t)(end_fsb - offset_fsb),
252                         bmapi_flags,  NULL, 0, &imap,
253                         &nimaps, NULL, NULL);
254
255         if (error)
256                 goto out;
257
258 phase2:
259         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
260         case BMAPI_WRITE:
261                 /* If we found an extent, return it */
262                 if (nimaps &&
263                     (imap.br_startblock != HOLESTARTBLOCK) &&
264                     (imap.br_startblock != DELAYSTARTBLOCK)) {
265                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
266                                         offset, count, iomapp, &imap, flags);
267                         break;
268                 }
269
270                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
271                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
272                                         count, flags, &imap, &nimaps, nimaps);
273                 } else {
274                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
275                                         flags, &imap, &nimaps);
276                 }
277                 if (!error) {
278                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
279                                         offset, count, iomapp, &imap, flags);
280                 }
281                 iomap_flags = IOMAP_NEW;
282                 break;
283         case BMAPI_ALLOCATE:
284                 /* If we found an extent, return it */
285                 XFS_IUNLOCK(mp, io, lockmode);
286                 lockmode = 0;
287
288                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
289                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
290                                         offset, count, iomapp, &imap, flags);
291                         break;
292                 }
293
294                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
295                                                  &imap, &nimaps);
296                 break;
297         case BMAPI_UNWRITTEN:
298                 lockmode = 0;
299                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
300                 nimaps = 0;
301                 break;
302         }
303
304         if (nimaps) {
305                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
306                                             iomapp, nimaps, *niomaps, iomap_flags);
307         } else if (niomaps) {
308                 *niomaps = 0;
309         }
310
311 out:
312         if (lockmode)
313                 XFS_IUNLOCK(mp, io, lockmode);
314         return XFS_ERROR(error);
315 }
316
317 STATIC int
318 xfs_iomap_eof_align_last_fsb(
319         xfs_mount_t     *mp,
320         xfs_iocore_t    *io,
321         xfs_fsize_t     isize,
322         xfs_extlen_t    extsize,
323         xfs_fileoff_t   *last_fsb)
324 {
325         xfs_fileoff_t   new_last_fsb = 0;
326         xfs_extlen_t    align;
327         int             eof, error;
328
329         if (io->io_flags & XFS_IOCORE_RT)
330                 ;
331         /*
332          * If mounted with the "-o swalloc" option, roundup the allocation
333          * request to a stripe width boundary if the file size is >=
334          * stripe width and we are allocating past the allocation eof.
335          */
336         else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
337                 (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)))
338                 new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
339         /*
340          * Roundup the allocation request to a stripe unit (m_dalign) boundary
341          * if the file size is >= stripe unit size, and we are allocating past
342          * the allocation eof.
343          */
344         else if (mp->m_dalign && (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)))
345                 new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
346
347         /*
348          * Always round up the allocation request to an extent boundary
349          * (when file on a real-time subvolume or has di_extsize hint).
350          */
351         if (extsize) {
352                 if (new_last_fsb)
353                         align = roundup_64(new_last_fsb, extsize);
354                 else
355                         align = extsize;
356                 new_last_fsb = roundup_64(*last_fsb, align);
357         }
358
359         if (new_last_fsb) {
360                 error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
361                 if (error)
362                         return error;
363                 if (eof)
364                         *last_fsb = new_last_fsb;
365         }
366         return 0;
367 }
368
369 STATIC int
370 xfs_flush_space(
371         xfs_inode_t     *ip,
372         int             *fsynced,
373         int             *ioflags)
374 {
375         switch (*fsynced) {
376         case 0:
377                 if (ip->i_delayed_blks) {
378                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
379                         xfs_flush_inode(ip);
380                         xfs_ilock(ip, XFS_ILOCK_EXCL);
381                         *fsynced = 1;
382                 } else {
383                         *ioflags |= BMAPI_SYNC;
384                         *fsynced = 2;
385                 }
386                 return 0;
387         case 1:
388                 *fsynced = 2;
389                 *ioflags |= BMAPI_SYNC;
390                 return 0;
391         case 2:
392                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
393                 xfs_flush_device(ip);
394                 xfs_ilock(ip, XFS_ILOCK_EXCL);
395                 *fsynced = 3;
396                 return 0;
397         }
398         return 1;
399 }
400
401 STATIC int
402 xfs_cmn_err_fsblock_zero(
403         xfs_inode_t     *ip,
404         xfs_bmbt_irec_t *imap)
405 {
406         xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
407                         "Access to block zero in inode %llu "
408                         "start_block: %llx start_off: %llx "
409                         "blkcnt: %llx extent-state: %x\n",
410                 (unsigned long long)ip->i_ino,
411                 (unsigned long long)imap->br_startblock,
412                 (unsigned long long)imap->br_startoff,
413                 (unsigned long long)imap->br_blockcount,
414                 imap->br_state);
415         return EFSCORRUPTED;
416 }
417
418 int
419 xfs_iomap_write_direct(
420         xfs_inode_t     *ip,
421         xfs_off_t       offset,
422         size_t          count,
423         int             flags,
424         xfs_bmbt_irec_t *ret_imap,
425         int             *nmaps,
426         int             found)
427 {
428         xfs_mount_t     *mp = ip->i_mount;
429         xfs_iocore_t    *io = &ip->i_iocore;
430         xfs_fileoff_t   offset_fsb;
431         xfs_fileoff_t   last_fsb;
432         xfs_filblks_t   count_fsb, resaligned;
433         xfs_fsblock_t   firstfsb;
434         xfs_extlen_t    extsz, temp;
435         xfs_fsize_t     isize;
436         int             nimaps;
437         int             bmapi_flag;
438         int             quota_flag;
439         int             rt;
440         xfs_trans_t     *tp;
441         xfs_bmbt_irec_t imap;
442         xfs_bmap_free_t free_list;
443         uint            qblocks, resblks, resrtextents;
444         int             committed;
445         int             error;
446
447         /*
448          * Make sure that the dquots are there. This doesn't hold
449          * the ilock across a disk read.
450          */
451         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
452         if (error)
453                 return XFS_ERROR(error);
454
455         rt = XFS_IS_REALTIME_INODE(ip);
456         if (unlikely(rt)) {
457                 if (!(extsz = ip->i_d.di_extsize))
458                         extsz = mp->m_sb.sb_rextsize;
459         } else {
460                 extsz = ip->i_d.di_extsize;
461         }
462
463         isize = ip->i_d.di_size;
464         if (io->io_new_size > isize)
465                 isize = io->io_new_size;
466
467         offset_fsb = XFS_B_TO_FSBT(mp, offset);
468         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
469         if ((offset + count) > isize) {
470                 error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
471                                                         &last_fsb);
472                 if (error)
473                         goto error_out;
474         } else {
475                 if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
476                         last_fsb = MIN(last_fsb, (xfs_fileoff_t)
477                                         ret_imap->br_blockcount +
478                                         ret_imap->br_startoff);
479         }
480         count_fsb = last_fsb - offset_fsb;
481         ASSERT(count_fsb > 0);
482
483         resaligned = count_fsb;
484         if (unlikely(extsz)) {
485                 if ((temp = do_mod(offset_fsb, extsz)))
486                         resaligned += temp;
487                 if ((temp = do_mod(resaligned, extsz)))
488                         resaligned += extsz - temp;
489         }
490
491         if (unlikely(rt)) {
492                 resrtextents = qblocks = resaligned;
493                 resrtextents /= mp->m_sb.sb_rextsize;
494                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
495                 quota_flag = XFS_QMOPT_RES_RTBLKS;
496         } else {
497                 resrtextents = 0;
498                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
499                 quota_flag = XFS_QMOPT_RES_REGBLKS;
500         }
501
502         /*
503          * Allocate and setup the transaction
504          */
505         xfs_iunlock(ip, XFS_ILOCK_EXCL);
506         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
507         error = xfs_trans_reserve(tp, resblks,
508                         XFS_WRITE_LOG_RES(mp), resrtextents,
509                         XFS_TRANS_PERM_LOG_RES,
510                         XFS_WRITE_LOG_COUNT);
511         /*
512          * Check for running out of space, note: need lock to return
513          */
514         if (error)
515                 xfs_trans_cancel(tp, 0);
516         xfs_ilock(ip, XFS_ILOCK_EXCL);
517         if (error)
518                 goto error_out;
519
520         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
521                                               qblocks, 0, quota_flag);
522         if (error)
523                 goto error1;
524
525         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
526         xfs_trans_ihold(tp, ip);
527
528         bmapi_flag = XFS_BMAPI_WRITE;
529         if ((flags & BMAPI_DIRECT) && (offset < ip->i_d.di_size || extsz))
530                 bmapi_flag |= XFS_BMAPI_PREALLOC;
531
532         /*
533          * Issue the xfs_bmapi() call to allocate the blocks
534          */
535         XFS_BMAP_INIT(&free_list, &firstfsb);
536         nimaps = 1;
537         error = XFS_BMAPI(mp, tp, io, offset_fsb, count_fsb, bmapi_flag,
538                 &firstfsb, 0, &imap, &nimaps, &free_list, NULL);
539         if (error)
540                 goto error0;
541
542         /*
543          * Complete the transaction
544          */
545         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
546         if (error)
547                 goto error0;
548         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
549         if (error)
550                 goto error_out;
551
552         /*
553          * Copy any maps to caller's array and return any error.
554          */
555         if (nimaps == 0) {
556                 error = ENOSPC;
557                 goto error_out;
558         }
559
560         if (unlikely(!imap.br_startblock && !(io->io_flags & XFS_IOCORE_RT))) {
561                 error = xfs_cmn_err_fsblock_zero(ip, &imap);
562                 goto error_out;
563         }
564
565         *ret_imap = imap;
566         *nmaps = 1;
567         return 0;
568
569 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
570         xfs_bmap_cancel(&free_list);
571         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
572
573 error1: /* Just cancel transaction */
574         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
575         *nmaps = 0;     /* nothing set-up here */
576
577 error_out:
578         return XFS_ERROR(error);
579 }
580
581 /*
582  * If the caller is doing a write at the end of the file,
583  * then extend the allocation out to the file system's write
584  * iosize.  We clean up any extra space left over when the
585  * file is closed in xfs_inactive().
586  *
587  * For sync writes, we are flushing delayed allocate space to
588  * try to make additional space available for allocation near
589  * the filesystem full boundary - preallocation hurts in that
590  * situation, of course.
591  */
592 STATIC int
593 xfs_iomap_eof_want_preallocate(
594         xfs_mount_t     *mp,
595         xfs_iocore_t    *io,
596         xfs_fsize_t     isize,
597         xfs_off_t       offset,
598         size_t          count,
599         int             ioflag,
600         xfs_bmbt_irec_t *imap,
601         int             nimaps,
602         int             *prealloc)
603 {
604         xfs_fileoff_t   start_fsb;
605         xfs_filblks_t   count_fsb;
606         xfs_fsblock_t   firstblock;
607         int             n, error, imaps;
608
609         *prealloc = 0;
610         if ((ioflag & BMAPI_SYNC) || (offset + count) <= isize)
611                 return 0;
612
613         /*
614          * If there are any real blocks past eof, then don't
615          * do any speculative allocation.
616          */
617         start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
618         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
619         while (count_fsb > 0) {
620                 imaps = nimaps;
621                 firstblock = NULLFSBLOCK;
622                 error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb, 0,
623                                   &firstblock, 0, imap, &imaps, NULL, NULL);
624                 if (error)
625                         return error;
626                 for (n = 0; n < imaps; n++) {
627                         if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
628                             (imap[n].br_startblock != DELAYSTARTBLOCK))
629                                 return 0;
630                         start_fsb += imap[n].br_blockcount;
631                         count_fsb -= imap[n].br_blockcount;
632                 }
633         }
634         *prealloc = 1;
635         return 0;
636 }
637
638 int
639 xfs_iomap_write_delay(
640         xfs_inode_t     *ip,
641         xfs_off_t       offset,
642         size_t          count,
643         int             ioflag,
644         xfs_bmbt_irec_t *ret_imap,
645         int             *nmaps)
646 {
647         xfs_mount_t     *mp = ip->i_mount;
648         xfs_iocore_t    *io = &ip->i_iocore;
649         xfs_fileoff_t   offset_fsb;
650         xfs_fileoff_t   last_fsb;
651         xfs_off_t       aligned_offset;
652         xfs_fileoff_t   ioalign;
653         xfs_fsblock_t   firstblock;
654         xfs_extlen_t    extsz;
655         xfs_fsize_t     isize;
656         int             nimaps;
657         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
658         int             prealloc, fsynced = 0;
659         int             error;
660
661         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
662
663         /*
664          * Make sure that the dquots are there. This doesn't hold
665          * the ilock across a disk read.
666          */
667         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
668         if (error)
669                 return XFS_ERROR(error);
670
671         if (XFS_IS_REALTIME_INODE(ip)) {
672                 if (!(extsz = ip->i_d.di_extsize))
673                         extsz = mp->m_sb.sb_rextsize;
674         } else {
675                 extsz = ip->i_d.di_extsize;
676         }
677
678         offset_fsb = XFS_B_TO_FSBT(mp, offset);
679
680 retry:
681         isize = ip->i_d.di_size;
682         if (io->io_new_size > isize)
683                 isize = io->io_new_size;
684
685         error = xfs_iomap_eof_want_preallocate(mp, io, isize, offset, count,
686                                 ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
687         if (error)
688                 return error;
689
690         if (prealloc) {
691                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
692                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
693                 last_fsb = ioalign + mp->m_writeio_blocks;
694         } else {
695                 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
696         }
697
698         if (prealloc || extsz) {
699                 error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
700                                                         &last_fsb);
701                 if (error)
702                         return error;
703         }
704
705         nimaps = XFS_WRITE_IMAPS;
706         firstblock = NULLFSBLOCK;
707         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
708                           (xfs_filblks_t)(last_fsb - offset_fsb),
709                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
710                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
711                           &nimaps, NULL, NULL);
712         if (error && (error != ENOSPC))
713                 return XFS_ERROR(error);
714
715         /*
716          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
717          * then we must have run out of space - flush delalloc, and retry..
718          */
719         if (nimaps == 0) {
720                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
721                                         io, offset, count);
722                 if (xfs_flush_space(ip, &fsynced, &ioflag))
723                         return XFS_ERROR(ENOSPC);
724
725                 error = 0;
726                 goto retry;
727         }
728
729         if (unlikely(!imap[0].br_startblock && !(io->io_flags & XFS_IOCORE_RT)))
730                 return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
731
732         *ret_imap = imap[0];
733         *nmaps = 1;
734
735         return 0;
736 }
737
738 /*
739  * Pass in a delayed allocate extent, convert it to real extents;
740  * return to the caller the extent we create which maps on top of
741  * the originating callers request.
742  *
743  * Called without a lock on the inode.
744  */
745 int
746 xfs_iomap_write_allocate(
747         xfs_inode_t     *ip,
748         xfs_off_t       offset,
749         size_t          count,
750         xfs_bmbt_irec_t *map,
751         int             *retmap)
752 {
753         xfs_mount_t     *mp = ip->i_mount;
754         xfs_iocore_t    *io = &ip->i_iocore;
755         xfs_fileoff_t   offset_fsb, last_block;
756         xfs_fileoff_t   end_fsb, map_start_fsb;
757         xfs_fsblock_t   first_block;
758         xfs_bmap_free_t free_list;
759         xfs_filblks_t   count_fsb;
760         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
761         xfs_trans_t     *tp;
762         int             i, nimaps, committed;
763         int             error = 0;
764         int             nres;
765
766         *retmap = 0;
767
768         /*
769          * Make sure that the dquots are there.
770          */
771         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
772                 return XFS_ERROR(error);
773
774         offset_fsb = XFS_B_TO_FSBT(mp, offset);
775         count_fsb = map->br_blockcount;
776         map_start_fsb = map->br_startoff;
777
778         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
779
780         while (count_fsb != 0) {
781                 /*
782                  * Set up a transaction with which to allocate the
783                  * backing store for the file.  Do allocations in a
784                  * loop until we get some space in the range we are
785                  * interested in.  The other space that might be allocated
786                  * is in the delayed allocation extent on which we sit
787                  * but before our buffer starts.
788                  */
789
790                 nimaps = 0;
791                 while (nimaps == 0) {
792                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
793                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
794                         error = xfs_trans_reserve(tp, nres,
795                                         XFS_WRITE_LOG_RES(mp),
796                                         0, XFS_TRANS_PERM_LOG_RES,
797                                         XFS_WRITE_LOG_COUNT);
798                         if (error == ENOSPC) {
799                                 error = xfs_trans_reserve(tp, 0,
800                                                 XFS_WRITE_LOG_RES(mp),
801                                                 0,
802                                                 XFS_TRANS_PERM_LOG_RES,
803                                                 XFS_WRITE_LOG_COUNT);
804                         }
805                         if (error) {
806                                 xfs_trans_cancel(tp, 0);
807                                 return XFS_ERROR(error);
808                         }
809                         xfs_ilock(ip, XFS_ILOCK_EXCL);
810                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
811                         xfs_trans_ihold(tp, ip);
812
813                         XFS_BMAP_INIT(&free_list, &first_block);
814
815                         nimaps = XFS_STRAT_WRITE_IMAPS;
816                         /*
817                          * Ensure we don't go beyond eof - it is possible
818                          * the extents changed since we did the read call,
819                          * we dropped the ilock in the interim.
820                          */
821
822                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
823                         xfs_bmap_last_offset(NULL, ip, &last_block,
824                                 XFS_DATA_FORK);
825                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
826                         if ((map_start_fsb + count_fsb) > last_block) {
827                                 count_fsb = last_block - map_start_fsb;
828                                 if (count_fsb == 0) {
829                                         error = EAGAIN;
830                                         goto trans_cancel;
831                                 }
832                         }
833
834                         /* Go get the actual blocks */
835                         error = XFS_BMAPI(mp, tp, io, map_start_fsb, count_fsb,
836                                         XFS_BMAPI_WRITE, &first_block, 1,
837                                         imap, &nimaps, &free_list, NULL);
838                         if (error)
839                                 goto trans_cancel;
840
841                         error = xfs_bmap_finish(&tp, &free_list,
842                                         first_block, &committed);
843                         if (error)
844                                 goto trans_cancel;
845
846                         error = xfs_trans_commit(tp,
847                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
848                         if (error)
849                                 goto error0;
850
851                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
852                 }
853
854                 /*
855                  * See if we were able to allocate an extent that
856                  * covers at least part of the callers request
857                  */
858                 for (i = 0; i < nimaps; i++) {
859                         if (unlikely(!imap[i].br_startblock &&
860                                      !(io->io_flags & XFS_IOCORE_RT)))
861                                 return xfs_cmn_err_fsblock_zero(ip, &imap[i]);
862                         if ((offset_fsb >= imap[i].br_startoff) &&
863                             (offset_fsb < (imap[i].br_startoff +
864                                            imap[i].br_blockcount))) {
865                                 *map = imap[i];
866                                 *retmap = 1;
867                                 XFS_STATS_INC(xs_xstrat_quick);
868                                 return 0;
869                         }
870                         count_fsb -= imap[i].br_blockcount;
871                 }
872
873                 /* So far we have not mapped the requested part of the
874                  * file, just surrounding data, try again.
875                  */
876                 nimaps--;
877                 map_start_fsb = imap[nimaps].br_startoff +
878                                 imap[nimaps].br_blockcount;
879         }
880
881 trans_cancel:
882         xfs_bmap_cancel(&free_list);
883         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
884 error0:
885         xfs_iunlock(ip, XFS_ILOCK_EXCL);
886         return XFS_ERROR(error);
887 }
888
889 int
890 xfs_iomap_write_unwritten(
891         xfs_inode_t     *ip,
892         xfs_off_t       offset,
893         size_t          count)
894 {
895         xfs_mount_t     *mp = ip->i_mount;
896         xfs_iocore_t    *io = &ip->i_iocore;
897         xfs_fileoff_t   offset_fsb;
898         xfs_filblks_t   count_fsb;
899         xfs_filblks_t   numblks_fsb;
900         xfs_fsblock_t   firstfsb;
901         int             nimaps;
902         xfs_trans_t     *tp;
903         xfs_bmbt_irec_t imap;
904         xfs_bmap_free_t free_list;
905         uint            resblks;
906         int             committed;
907         int             error;
908
909         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
910                                 &ip->i_iocore, offset, count);
911
912         offset_fsb = XFS_B_TO_FSBT(mp, offset);
913         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
914         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
915
916         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
917
918         do {
919                 /*
920                  * set up a transaction to convert the range of extents
921                  * from unwritten to real. Do allocations in a loop until
922                  * we have covered the range passed in.
923                  */
924
925                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
926                 error = xfs_trans_reserve(tp, resblks,
927                                 XFS_WRITE_LOG_RES(mp), 0,
928                                 XFS_TRANS_PERM_LOG_RES,
929                                 XFS_WRITE_LOG_COUNT);
930                 if (error) {
931                         xfs_trans_cancel(tp, 0);
932                         return XFS_ERROR(error);
933                 }
934
935                 xfs_ilock(ip, XFS_ILOCK_EXCL);
936                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
937                 xfs_trans_ihold(tp, ip);
938
939                 /*
940                  * Modify the unwritten extent state of the buffer.
941                  */
942                 XFS_BMAP_INIT(&free_list, &firstfsb);
943                 nimaps = 1;
944                 error = XFS_BMAPI(mp, tp, io, offset_fsb, count_fsb,
945                                   XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
946                                   1, &imap, &nimaps, &free_list, NULL);
947                 if (error)
948                         goto error_on_bmapi_transaction;
949
950                 error = xfs_bmap_finish(&(tp), &(free_list),
951                                 firstfsb, &committed);
952                 if (error)
953                         goto error_on_bmapi_transaction;
954
955                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
956                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
957                 if (error)
958                         return XFS_ERROR(error);
959
960                 if (unlikely(!imap.br_startblock &&
961                              !(io->io_flags & XFS_IOCORE_RT)))
962                         return xfs_cmn_err_fsblock_zero(ip, &imap);
963
964                 if ((numblks_fsb = imap.br_blockcount) == 0) {
965                         /*
966                          * The numblks_fsb value should always get
967                          * smaller, otherwise the loop is stuck.
968                          */
969                         ASSERT(imap.br_blockcount);
970                         break;
971                 }
972                 offset_fsb += numblks_fsb;
973                 count_fsb -= numblks_fsb;
974         } while (count_fsb > 0);
975
976         return 0;
977
978 error_on_bmapi_transaction:
979         xfs_bmap_cancel(&free_list);
980         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
981         xfs_iunlock(ip, XFS_ILOCK_EXCL);
982         return XFS_ERROR(error);
983 }