[XFS] Implement the di_extsize allocator hint for non-realtime files as
[linux-2.6.git] / fs / xfs / xfs_iomap.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_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_dir.h"
27 #include "xfs_dir2.h"
28 #include "xfs_alloc.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_quota.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_btree.h"
42 #include "xfs_bmap.h"
43 #include "xfs_bit.h"
44 #include "xfs_rtalloc.h"
45 #include "xfs_error.h"
46 #include "xfs_itable.h"
47 #include "xfs_rw.h"
48 #include "xfs_acl.h"
49 #include "xfs_cap.h"
50 #include "xfs_mac.h"
51 #include "xfs_attr.h"
52 #include "xfs_buf_item.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_utils.h"
55 #include "xfs_iomap.h"
56
57 #if defined(XFS_RW_TRACE)
58 void
59 xfs_iomap_enter_trace(
60         int             tag,
61         xfs_iocore_t    *io,
62         xfs_off_t       offset,
63         ssize_t         count)
64 {
65         xfs_inode_t     *ip = XFS_IO_INODE(io);
66
67         if (!ip->i_rwtrace)
68                 return;
69
70         ktrace_enter(ip->i_rwtrace,
71                 (void *)((unsigned long)tag),
72                 (void *)ip,
73                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
74                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
75                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
76                 (void *)((unsigned long)(offset & 0xffffffff)),
77                 (void *)((unsigned long)count),
78                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
79                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
80                 (void *)NULL,
81                 (void *)NULL,
82                 (void *)NULL,
83                 (void *)NULL,
84                 (void *)NULL,
85                 (void *)NULL,
86                 (void *)NULL);
87 }
88
89 void
90 xfs_iomap_map_trace(
91         int             tag,
92         xfs_iocore_t    *io,
93         xfs_off_t       offset,
94         ssize_t         count,
95         xfs_iomap_t     *iomapp,
96         xfs_bmbt_irec_t *imapp,
97         int             flags)
98 {
99         xfs_inode_t     *ip = XFS_IO_INODE(io);
100
101         if (!ip->i_rwtrace)
102                 return;
103
104         ktrace_enter(ip->i_rwtrace,
105                 (void *)((unsigned long)tag),
106                 (void *)ip,
107                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
108                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
109                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
110                 (void *)((unsigned long)(offset & 0xffffffff)),
111                 (void *)((unsigned long)count),
112                 (void *)((unsigned long)flags),
113                 (void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
114                 (void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
115                 (void *)((unsigned long)(iomapp->iomap_delta)),
116                 (void *)((unsigned long)(iomapp->iomap_bsize)),
117                 (void *)((unsigned long)(iomapp->iomap_bn)),
118                 (void *)(__psint_t)(imapp->br_startoff),
119                 (void *)((unsigned long)(imapp->br_blockcount)),
120                 (void *)(__psint_t)(imapp->br_startblock));
121 }
122 #else
123 #define xfs_iomap_enter_trace(tag, io, offset, count)
124 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
125 #endif
126
127 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
128                                                 << mp->m_writeio_log)
129 #define XFS_STRAT_WRITE_IMAPS   2
130 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
131
132 STATIC int
133 xfs_imap_to_bmap(
134         xfs_iocore_t    *io,
135         xfs_off_t       offset,
136         xfs_bmbt_irec_t *imap,
137         xfs_iomap_t     *iomapp,
138         int             imaps,                  /* Number of imap entries */
139         int             iomaps,                 /* Number of iomap entries */
140         int             flags)
141 {
142         xfs_mount_t     *mp;
143         xfs_fsize_t     nisize;
144         int             pbm;
145         xfs_fsblock_t   start_block;
146
147         mp = io->io_mount;
148         nisize = XFS_SIZE(mp, io);
149         if (io->io_new_size > nisize)
150                 nisize = io->io_new_size;
151
152         for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
153                 iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
154                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
155                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
156                 iomapp->iomap_flags = flags;
157
158                 if (io->io_flags & XFS_IOCORE_RT) {
159                         iomapp->iomap_flags |= IOMAP_REALTIME;
160                         iomapp->iomap_target = mp->m_rtdev_targp;
161                 } else {
162                         iomapp->iomap_target = mp->m_ddev_targp;
163                 }
164                 start_block = imap->br_startblock;
165                 if (start_block == HOLESTARTBLOCK) {
166                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
167                         iomapp->iomap_flags |= IOMAP_HOLE;
168                 } else if (start_block == DELAYSTARTBLOCK) {
169                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
170                         iomapp->iomap_flags |= IOMAP_DELAY;
171                 } else {
172                         iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
173                         if (ISUNWRITTEN(imap))
174                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
175                 }
176
177                 if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
178                         iomapp->iomap_flags |= IOMAP_EOF;
179                 }
180
181                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
182         }
183         return pbm;     /* Return the number filled */
184 }
185
186 int
187 xfs_iomap(
188         xfs_iocore_t    *io,
189         xfs_off_t       offset,
190         ssize_t         count,
191         int             flags,
192         xfs_iomap_t     *iomapp,
193         int             *niomaps)
194 {
195         xfs_mount_t     *mp = io->io_mount;
196         xfs_fileoff_t   offset_fsb, end_fsb;
197         int             error = 0;
198         int             lockmode = 0;
199         xfs_bmbt_irec_t imap;
200         int             nimaps = 1;
201         int             bmapi_flags = 0;
202         int             iomap_flags = 0;
203
204         if (XFS_FORCED_SHUTDOWN(mp))
205                 return XFS_ERROR(EIO);
206
207         switch (flags &
208                 (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
209                  BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
210         case BMAPI_READ:
211                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
212                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
213                 bmapi_flags = XFS_BMAPI_ENTIRE;
214                 break;
215         case BMAPI_WRITE:
216                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
217                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
218                 if (flags & BMAPI_IGNSTATE)
219                         bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
220                 XFS_ILOCK(mp, io, lockmode);
221                 break;
222         case BMAPI_ALLOCATE:
223                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
224                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
225                 bmapi_flags = XFS_BMAPI_ENTIRE;
226                 /* Attempt non-blocking lock */
227                 if (flags & BMAPI_TRYLOCK) {
228                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
229                                 return XFS_ERROR(EAGAIN);
230                 } else {
231                         XFS_ILOCK(mp, io, lockmode);
232                 }
233                 break;
234         case BMAPI_UNWRITTEN:
235                 goto phase2;
236         case BMAPI_DEVICE:
237                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
238                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
239                         mp->m_rtdev_targp : mp->m_ddev_targp;
240                 error = 0;
241                 *niomaps = 1;
242                 goto out;
243         default:
244                 BUG();
245         }
246
247         ASSERT(offset <= mp->m_maxioffset);
248         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
249                 count = mp->m_maxioffset - offset;
250         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
251         offset_fsb = XFS_B_TO_FSBT(mp, offset);
252
253         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
254                         (xfs_filblks_t)(end_fsb - offset_fsb),
255                         bmapi_flags,  NULL, 0, &imap,
256                         &nimaps, NULL);
257
258         if (error)
259                 goto out;
260
261 phase2:
262         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
263         case BMAPI_WRITE:
264                 /* If we found an extent, return it */
265                 if (nimaps &&
266                     (imap.br_startblock != HOLESTARTBLOCK) &&
267                     (imap.br_startblock != DELAYSTARTBLOCK)) {
268                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
269                                         offset, count, iomapp, &imap, flags);
270                         break;
271                 }
272
273                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
274                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
275                                         count, flags, &imap, &nimaps, nimaps);
276                 } else {
277                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
278                                         flags, &imap, &nimaps);
279                 }
280                 if (!error) {
281                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
282                                         offset, count, iomapp, &imap, flags);
283                 }
284                 iomap_flags = IOMAP_NEW;
285                 break;
286         case BMAPI_ALLOCATE:
287                 /* If we found an extent, return it */
288                 XFS_IUNLOCK(mp, io, lockmode);
289                 lockmode = 0;
290
291                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
292                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
293                                         offset, count, iomapp, &imap, flags);
294                         break;
295                 }
296
297                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
298                                                  &imap, &nimaps);
299                 break;
300         case BMAPI_UNWRITTEN:
301                 lockmode = 0;
302                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
303                 nimaps = 0;
304                 break;
305         }
306
307         if (nimaps) {
308                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
309                                             iomapp, nimaps, *niomaps, iomap_flags);
310         } else if (niomaps) {
311                 *niomaps = 0;
312         }
313
314 out:
315         if (lockmode)
316                 XFS_IUNLOCK(mp, io, lockmode);
317         return XFS_ERROR(error);
318 }
319
320 STATIC int
321 xfs_iomap_eof_align_last_fsb(
322         xfs_mount_t     *mp,
323         xfs_iocore_t    *io,
324         xfs_fsize_t     isize,
325         xfs_extlen_t    extsize,
326         xfs_fileoff_t   *last_fsb)
327 {
328         xfs_fileoff_t   new_last_fsb = 0;
329         xfs_extlen_t    align;
330         int             eof, error;
331
332         if (io->io_flags & XFS_IOCORE_RT)
333                 ;
334         /*
335          * If mounted with the "-o swalloc" option, roundup the allocation
336          * request to a stripe width boundary if the file size is >=
337          * stripe width and we are allocating past the allocation eof.
338          */
339         else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
340                 (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)))
341                 new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
342         /*
343          * Roundup the allocation request to a stripe unit (m_dalign) boundary
344          * if the file size is >= stripe unit size, and we are allocating past
345          * the allocation eof.
346          */
347         else if (mp->m_dalign && (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)))
348                 new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
349
350         /*
351          * Always round up the allocation request to an extent boundary
352          * (when file on a real-time subvolume or has di_extsize hint).
353          */
354         if (extsize) {
355                 if (new_last_fsb)
356                         align = roundup_64(new_last_fsb, extsize);
357                 else
358                         align = extsize;
359                 new_last_fsb = roundup_64(*last_fsb, align);
360         }
361
362         if (new_last_fsb) {
363                 error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
364                 if (error)
365                         return error;
366                 if (eof)
367                         *last_fsb = new_last_fsb;
368         }
369         return 0;
370 }
371
372 STATIC int
373 xfs_flush_space(
374         xfs_inode_t     *ip,
375         int             *fsynced,
376         int             *ioflags)
377 {
378         switch (*fsynced) {
379         case 0:
380                 if (ip->i_delayed_blks) {
381                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
382                         xfs_flush_inode(ip);
383                         xfs_ilock(ip, XFS_ILOCK_EXCL);
384                         *fsynced = 1;
385                 } else {
386                         *ioflags |= BMAPI_SYNC;
387                         *fsynced = 2;
388                 }
389                 return 0;
390         case 1:
391                 *fsynced = 2;
392                 *ioflags |= BMAPI_SYNC;
393                 return 0;
394         case 2:
395                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
396                 xfs_flush_device(ip);
397                 xfs_ilock(ip, XFS_ILOCK_EXCL);
398                 *fsynced = 3;
399                 return 0;
400         }
401         return 1;
402 }
403
404 int
405 xfs_iomap_write_direct(
406         xfs_inode_t     *ip,
407         xfs_off_t       offset,
408         size_t          count,
409         int             flags,
410         xfs_bmbt_irec_t *ret_imap,
411         int             *nmaps,
412         int             found)
413 {
414         xfs_mount_t     *mp = ip->i_mount;
415         xfs_iocore_t    *io = &ip->i_iocore;
416         xfs_fileoff_t   offset_fsb;
417         xfs_fileoff_t   last_fsb;
418         xfs_filblks_t   count_fsb, resaligned;
419         xfs_fsblock_t   firstfsb;
420         xfs_extlen_t    extsz, temp;
421         xfs_fsize_t     isize;
422         int             nimaps;
423         int             bmapi_flag;
424         int             quota_flag;
425         int             rt;
426         xfs_trans_t     *tp;
427         xfs_bmbt_irec_t imap;
428         xfs_bmap_free_t free_list;
429         uint            qblocks, resblks, resrtextents;
430         int             committed;
431         int             error;
432
433         /*
434          * Make sure that the dquots are there. This doesn't hold
435          * the ilock across a disk read.
436          */
437         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
438         if (error)
439                 return XFS_ERROR(error);
440
441         rt = XFS_IS_REALTIME_INODE(ip);
442         if (unlikely(rt)) {
443                 if (!(extsz = ip->i_d.di_extsize))
444                         extsz = mp->m_sb.sb_rextsize;
445         } else {
446                 extsz = ip->i_d.di_extsize;
447         }
448
449         isize = ip->i_d.di_size;
450         if (io->io_new_size > isize)
451                 isize = io->io_new_size;
452
453         offset_fsb = XFS_B_TO_FSBT(mp, offset);
454         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
455         if ((offset + count) > isize) {
456                 error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
457                                                         &last_fsb);
458                 if (error)
459                         goto error_out;
460         } else {
461                 if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
462                         last_fsb = MIN(last_fsb, (xfs_fileoff_t)
463                                         ret_imap->br_blockcount +
464                                         ret_imap->br_startoff);
465         }
466         count_fsb = last_fsb - offset_fsb;
467         ASSERT(count_fsb > 0);
468
469         resaligned = count_fsb;
470         if (unlikely(extsz)) {
471                 if ((temp = do_mod(offset_fsb, extsz)))
472                         resaligned += temp;
473                 if ((temp = do_mod(resaligned, extsz)))
474                         resaligned += extsz - temp;
475         }
476
477         if (unlikely(rt)) {
478                 resrtextents = qblocks = resaligned;
479                 resrtextents /= mp->m_sb.sb_rextsize;
480                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
481                 quota_flag = XFS_QMOPT_RES_RTBLKS;
482         } else {
483                 resrtextents = 0;
484                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
485                 quota_flag = XFS_QMOPT_RES_REGBLKS;
486         }
487
488         /*
489          * Allocate and setup the transaction
490          */
491         xfs_iunlock(ip, XFS_ILOCK_EXCL);
492         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
493         error = xfs_trans_reserve(tp, resblks,
494                         XFS_WRITE_LOG_RES(mp), resrtextents,
495                         XFS_TRANS_PERM_LOG_RES,
496                         XFS_WRITE_LOG_COUNT);
497         /*
498          * Check for running out of space, note: need lock to return
499          */
500         if (error)
501                 xfs_trans_cancel(tp, 0);
502         xfs_ilock(ip, XFS_ILOCK_EXCL);
503         if (error)
504                 goto error_out;
505
506         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
507                                               qblocks, 0, quota_flag);
508         if (error)
509                 goto error1;
510
511         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
512         xfs_trans_ihold(tp, ip);
513
514         bmapi_flag = XFS_BMAPI_WRITE;
515         if ((flags & BMAPI_DIRECT) && (offset < ip->i_d.di_size || extsz))
516                 bmapi_flag |= XFS_BMAPI_PREALLOC;
517
518         /*
519          * Issue the xfs_bmapi() call to allocate the blocks
520          */
521         XFS_BMAP_INIT(&free_list, &firstfsb);
522         nimaps = 1;
523         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
524                 bmapi_flag, &firstfsb, 0, &imap, &nimaps, &free_list);
525         if (error)
526                 goto error0;
527
528         /*
529          * Complete the transaction
530          */
531         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
532         if (error)
533                 goto error0;
534         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
535         if (error)
536                 goto error_out;
537
538         /*
539          * Copy any maps to caller's array and return any error.
540          */
541         if (nimaps == 0) {
542                 error = (ENOSPC);
543                 goto error_out;
544         }
545
546         *ret_imap = imap;
547         *nmaps = 1;
548         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
549                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
550                         "start_block : %llx start_off : %llx blkcnt : %llx "
551                         "extent-state : %x \n",
552                         (ip->i_mount)->m_fsname,
553                         (long long)ip->i_ino,
554                         ret_imap->br_startblock, ret_imap->br_startoff,
555                         ret_imap->br_blockcount,ret_imap->br_state);
556         }
557         return 0;
558
559 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
560         xfs_bmap_cancel(&free_list);
561         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
562
563 error1: /* Just cancel transaction */
564         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
565         *nmaps = 0;     /* nothing set-up here */
566
567 error_out:
568         return XFS_ERROR(error);
569 }
570
571 /*
572  * If the caller is doing a write at the end of the file,
573  * then extend the allocation out to the file system's write
574  * iosize.  We clean up any extra space left over when the
575  * file is closed in xfs_inactive().
576  *
577  * For sync writes, we are flushing delayed allocate space to
578  * try to make additional space available for allocation near
579  * the filesystem full boundary - preallocation hurts in that
580  * situation, of course.
581  */
582 STATIC int
583 xfs_iomap_eof_want_preallocate(
584         xfs_mount_t     *mp,
585         xfs_iocore_t    *io,
586         xfs_fsize_t     isize,
587         xfs_off_t       offset,
588         size_t          count,
589         int             ioflag,
590         xfs_bmbt_irec_t *imap,
591         int             nimaps,
592         int             *prealloc)
593 {
594         xfs_fileoff_t   start_fsb;
595         xfs_filblks_t   count_fsb;
596         xfs_fsblock_t   firstblock;
597         int             n, error, imaps;
598
599         *prealloc = 0;
600         if ((ioflag & BMAPI_SYNC) || (offset + count) <= isize)
601                 return 0;
602
603         /*
604          * If there are any real blocks past eof, then don't
605          * do any speculative allocation.
606          */
607         start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
608         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
609         while (count_fsb > 0) {
610                 imaps = nimaps;
611                 error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
612                                   0, &firstblock, 0, imap, &imaps, NULL);
613                 if (error)
614                         return error;
615                 for (n = 0; n < imaps; n++) {
616                         if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
617                             (imap[n].br_startblock != DELAYSTARTBLOCK))
618                                 return 0;
619                         start_fsb += imap[n].br_blockcount;
620                         count_fsb -= imap[n].br_blockcount;
621                 }
622         }
623         *prealloc = 1;
624         return 0;
625 }
626
627 int
628 xfs_iomap_write_delay(
629         xfs_inode_t     *ip,
630         xfs_off_t       offset,
631         size_t          count,
632         int             ioflag,
633         xfs_bmbt_irec_t *ret_imap,
634         int             *nmaps)
635 {
636         xfs_mount_t     *mp = ip->i_mount;
637         xfs_iocore_t    *io = &ip->i_iocore;
638         xfs_fileoff_t   offset_fsb;
639         xfs_fileoff_t   last_fsb;
640         xfs_off_t       aligned_offset;
641         xfs_fileoff_t   ioalign;
642         xfs_fsblock_t   firstblock;
643         xfs_extlen_t    extsz;
644         xfs_fsize_t     isize;
645         int             nimaps;
646         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
647         int             prealloc, fsynced = 0;
648         int             error;
649
650         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
651
652         /*
653          * Make sure that the dquots are there. This doesn't hold
654          * the ilock across a disk read.
655          */
656         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
657         if (error)
658                 return XFS_ERROR(error);
659
660         if (XFS_IS_REALTIME_INODE(ip)) {
661                 if (!(extsz = ip->i_d.di_extsize))
662                         extsz = mp->m_sb.sb_rextsize;
663         } else {
664                 extsz = ip->i_d.di_extsize;
665         }
666
667         offset_fsb = XFS_B_TO_FSBT(mp, offset);
668
669 retry:
670         isize = ip->i_d.di_size;
671         if (io->io_new_size > isize)
672                 isize = io->io_new_size;
673
674         error = xfs_iomap_eof_want_preallocate(mp, io, isize, offset, count,
675                                 ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
676         if (error)
677                 return error;
678
679         if (prealloc) {
680                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
681                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
682                 last_fsb = ioalign + mp->m_writeio_blocks;
683         } else {
684                 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
685         }
686
687         if (prealloc || extsz) {
688                 error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
689                                                         &last_fsb);
690                 if (error)
691                         return error;
692         }
693
694         nimaps = XFS_WRITE_IMAPS;
695         firstblock = NULLFSBLOCK;
696         error = xfs_bmapi(NULL, ip, offset_fsb,
697                           (xfs_filblks_t)(last_fsb - offset_fsb),
698                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
699                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
700                           &nimaps, NULL);
701         if (error && (error != ENOSPC))
702                 return XFS_ERROR(error);
703
704         /*
705          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
706          * then we must have run out of space - flush delalloc, and retry..
707          */
708         if (nimaps == 0) {
709                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
710                                         io, offset, count);
711                 if (xfs_flush_space(ip, &fsynced, &ioflag))
712                         return XFS_ERROR(ENOSPC);
713
714                 error = 0;
715                 goto retry;
716         }
717
718         if (!(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
719                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
720                         "start_block : %llx start_off : %llx blkcnt : %llx "
721                         "extent-state : %x \n",
722                         (ip->i_mount)->m_fsname,
723                         (long long)ip->i_ino,
724                         ret_imap->br_startblock, ret_imap->br_startoff,
725                         ret_imap->br_blockcount,ret_imap->br_state);
726         }
727
728         *ret_imap = imap[0];
729         *nmaps = 1;
730
731         return 0;
732 }
733
734 /*
735  * Pass in a delayed allocate extent, convert it to real extents;
736  * return to the caller the extent we create which maps on top of
737  * the originating callers request.
738  *
739  * Called without a lock on the inode.
740  */
741 int
742 xfs_iomap_write_allocate(
743         xfs_inode_t     *ip,
744         xfs_off_t       offset,
745         size_t          count,
746         xfs_bmbt_irec_t *map,
747         int             *retmap)
748 {
749         xfs_mount_t     *mp = ip->i_mount;
750         xfs_iocore_t    *io = &ip->i_iocore;
751         xfs_fileoff_t   offset_fsb, last_block;
752         xfs_fileoff_t   end_fsb, map_start_fsb;
753         xfs_fsblock_t   first_block;
754         xfs_bmap_free_t free_list;
755         xfs_filblks_t   count_fsb;
756         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
757         xfs_trans_t     *tp;
758         int             i, nimaps, committed;
759         int             error = 0;
760         int             nres;
761
762         *retmap = 0;
763
764         /*
765          * Make sure that the dquots are there.
766          */
767         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
768                 return XFS_ERROR(error);
769
770         offset_fsb = XFS_B_TO_FSBT(mp, offset);
771         count_fsb = map->br_blockcount;
772         map_start_fsb = map->br_startoff;
773
774         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
775
776         while (count_fsb != 0) {
777                 /*
778                  * Set up a transaction with which to allocate the
779                  * backing store for the file.  Do allocations in a
780                  * loop until we get some space in the range we are
781                  * interested in.  The other space that might be allocated
782                  * is in the delayed allocation extent on which we sit
783                  * but before our buffer starts.
784                  */
785
786                 nimaps = 0;
787                 while (nimaps == 0) {
788                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
789                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
790                         error = xfs_trans_reserve(tp, nres,
791                                         XFS_WRITE_LOG_RES(mp),
792                                         0, XFS_TRANS_PERM_LOG_RES,
793                                         XFS_WRITE_LOG_COUNT);
794                         if (error == ENOSPC) {
795                                 error = xfs_trans_reserve(tp, 0,
796                                                 XFS_WRITE_LOG_RES(mp),
797                                                 0,
798                                                 XFS_TRANS_PERM_LOG_RES,
799                                                 XFS_WRITE_LOG_COUNT);
800                         }
801                         if (error) {
802                                 xfs_trans_cancel(tp, 0);
803                                 return XFS_ERROR(error);
804                         }
805                         xfs_ilock(ip, XFS_ILOCK_EXCL);
806                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
807                         xfs_trans_ihold(tp, ip);
808
809                         XFS_BMAP_INIT(&free_list, &first_block);
810
811                         nimaps = XFS_STRAT_WRITE_IMAPS;
812                         /*
813                          * Ensure we don't go beyond eof - it is possible
814                          * the extents changed since we did the read call,
815                          * we dropped the ilock in the interim.
816                          */
817
818                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
819                         xfs_bmap_last_offset(NULL, ip, &last_block,
820                                 XFS_DATA_FORK);
821                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
822                         if ((map_start_fsb + count_fsb) > last_block) {
823                                 count_fsb = last_block - map_start_fsb;
824                                 if (count_fsb == 0) {
825                                         error = EAGAIN;
826                                         goto trans_cancel;
827                                 }
828                         }
829
830                         /* Go get the actual blocks */
831                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
832                                         XFS_BMAPI_WRITE, &first_block, 1,
833                                         imap, &nimaps, &free_list);
834                         if (error)
835                                 goto trans_cancel;
836
837                         error = xfs_bmap_finish(&tp, &free_list,
838                                         first_block, &committed);
839                         if (error)
840                                 goto trans_cancel;
841
842                         error = xfs_trans_commit(tp,
843                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
844                         if (error)
845                                 goto error0;
846
847                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
848                 }
849
850                 /*
851                  * See if we were able to allocate an extent that
852                  * covers at least part of the callers request
853                  */
854
855                 for (i = 0; i < nimaps; i++) {
856                         if ( !(io->io_flags & XFS_IOCORE_RT)  && 
857                                 !imap[i].br_startblock) {
858                                 cmn_err(CE_PANIC,"Access to block zero:  "
859                                         "fs <%s> inode: %lld "
860                                         "start_block : %llx start_off : %llx " 
861                                         "blkcnt : %llx extent-state : %x \n",
862                                         (ip->i_mount)->m_fsname,
863                                         (long long)ip->i_ino,
864                                         imap[i].br_startblock,
865                                         imap[i].br_startoff,
866                                         imap[i].br_blockcount,imap[i].br_state);
867                         }
868                         if ((offset_fsb >= imap[i].br_startoff) &&
869                             (offset_fsb < (imap[i].br_startoff +
870                                            imap[i].br_blockcount))) {
871                                 *map = imap[i];
872                                 *retmap = 1;
873                                 XFS_STATS_INC(xs_xstrat_quick);
874                                 return 0;
875                         }
876                         count_fsb -= imap[i].br_blockcount;
877                 }
878
879                 /* So far we have not mapped the requested part of the
880                  * file, just surrounding data, try again.
881                  */
882                 nimaps--;
883                 map_start_fsb = imap[nimaps].br_startoff +
884                                 imap[nimaps].br_blockcount;
885         }
886
887 trans_cancel:
888         xfs_bmap_cancel(&free_list);
889         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
890 error0:
891         xfs_iunlock(ip, XFS_ILOCK_EXCL);
892         return XFS_ERROR(error);
893 }
894
895 int
896 xfs_iomap_write_unwritten(
897         xfs_inode_t     *ip,
898         xfs_off_t       offset,
899         size_t          count)
900 {
901         xfs_mount_t     *mp = ip->i_mount;
902         xfs_iocore_t    *io = &ip->i_iocore;
903         xfs_fileoff_t   offset_fsb;
904         xfs_filblks_t   count_fsb;
905         xfs_filblks_t   numblks_fsb;
906         xfs_fsblock_t   firstfsb;
907         int             nimaps;
908         xfs_trans_t     *tp;
909         xfs_bmbt_irec_t imap;
910         xfs_bmap_free_t free_list;
911         uint            resblks;
912         int             committed;
913         int             error;
914
915         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
916                                 &ip->i_iocore, offset, count);
917
918         offset_fsb = XFS_B_TO_FSBT(mp, offset);
919         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
920         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
921
922         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
923
924         do {
925                 /*
926                  * set up a transaction to convert the range of extents
927                  * from unwritten to real. Do allocations in a loop until
928                  * we have covered the range passed in.
929                  */
930
931                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
932                 error = xfs_trans_reserve(tp, resblks,
933                                 XFS_WRITE_LOG_RES(mp), 0,
934                                 XFS_TRANS_PERM_LOG_RES,
935                                 XFS_WRITE_LOG_COUNT);
936                 if (error) {
937                         xfs_trans_cancel(tp, 0);
938                         goto error0;
939                 }
940
941                 xfs_ilock(ip, XFS_ILOCK_EXCL);
942                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
943                 xfs_trans_ihold(tp, ip);
944
945                 /*
946                  * Modify the unwritten extent state of the buffer.
947                  */
948                 XFS_BMAP_INIT(&free_list, &firstfsb);
949                 nimaps = 1;
950                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
951                                   XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
952                                   1, &imap, &nimaps, &free_list);
953                 if (error)
954                         goto error_on_bmapi_transaction;
955
956                 error = xfs_bmap_finish(&(tp), &(free_list),
957                                 firstfsb, &committed);
958                 if (error)
959                         goto error_on_bmapi_transaction;
960
961                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
962                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
963                 if (error)
964                         goto error0;
965
966                 if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
967                         cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
968                                 "inode: %lld start_block : %llx start_off : "
969                                 "%llx blkcnt : %llx extent-state : %x \n",
970                                 (ip->i_mount)->m_fsname,
971                                 (long long)ip->i_ino,
972                                 imap.br_startblock,imap.br_startoff,
973                                 imap.br_blockcount,imap.br_state);
974                 }
975
976                 if ((numblks_fsb = imap.br_blockcount) == 0) {
977                         /*
978                          * The numblks_fsb value should always get
979                          * smaller, otherwise the loop is stuck.
980                          */
981                         ASSERT(imap.br_blockcount);
982                         break;
983                 }
984                 offset_fsb += numblks_fsb;
985                 count_fsb -= numblks_fsb;
986         } while (count_fsb > 0);
987
988         return 0;
989
990 error_on_bmapi_transaction:
991         xfs_bmap_cancel(&free_list);
992         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
993         xfs_iunlock(ip, XFS_ILOCK_EXCL);
994 error0:
995         return XFS_ERROR(error);
996 }