Merge upstream kernel changes into 'C/H/S support' branch of libata.
[linux-2.6] / fs / xfs / xfs_iomap.c
1 /*
2  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32
33 #include "xfs.h"
34
35 #include "xfs_fs.h"
36 #include "xfs_inum.h"
37 #include "xfs_log.h"
38 #include "xfs_trans.h"
39 #include "xfs_sb.h"
40 #include "xfs_ag.h"
41 #include "xfs_dir.h"
42 #include "xfs_dir2.h"
43 #include "xfs_alloc.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_quota.h"
46 #include "xfs_mount.h"
47 #include "xfs_alloc_btree.h"
48 #include "xfs_bmap_btree.h"
49 #include "xfs_ialloc_btree.h"
50 #include "xfs_btree.h"
51 #include "xfs_ialloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode.h"
57 #include "xfs_bmap.h"
58 #include "xfs_bit.h"
59 #include "xfs_rtalloc.h"
60 #include "xfs_error.h"
61 #include "xfs_itable.h"
62 #include "xfs_rw.h"
63 #include "xfs_acl.h"
64 #include "xfs_cap.h"
65 #include "xfs_mac.h"
66 #include "xfs_attr.h"
67 #include "xfs_buf_item.h"
68 #include "xfs_trans_space.h"
69 #include "xfs_utils.h"
70 #include "xfs_iomap.h"
71
72 #if defined(XFS_RW_TRACE)
73 void
74 xfs_iomap_enter_trace(
75         int             tag,
76         xfs_iocore_t    *io,
77         xfs_off_t       offset,
78         ssize_t         count)
79 {
80         xfs_inode_t     *ip = XFS_IO_INODE(io);
81
82         if (!ip->i_rwtrace)
83                 return;
84
85         ktrace_enter(ip->i_rwtrace,
86                 (void *)((unsigned long)tag),
87                 (void *)ip,
88                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
89                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
90                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
91                 (void *)((unsigned long)(offset & 0xffffffff)),
92                 (void *)((unsigned long)count),
93                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
94                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
95                 (void *)NULL,
96                 (void *)NULL,
97                 (void *)NULL,
98                 (void *)NULL,
99                 (void *)NULL,
100                 (void *)NULL,
101                 (void *)NULL);
102 }
103
104 void
105 xfs_iomap_map_trace(
106         int             tag,
107         xfs_iocore_t    *io,
108         xfs_off_t       offset,
109         ssize_t         count,
110         xfs_iomap_t     *iomapp,
111         xfs_bmbt_irec_t *imapp,
112         int             flags)
113 {
114         xfs_inode_t     *ip = XFS_IO_INODE(io);
115
116         if (!ip->i_rwtrace)
117                 return;
118
119         ktrace_enter(ip->i_rwtrace,
120                 (void *)((unsigned long)tag),
121                 (void *)ip,
122                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
123                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
124                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
125                 (void *)((unsigned long)(offset & 0xffffffff)),
126                 (void *)((unsigned long)count),
127                 (void *)((unsigned long)flags),
128                 (void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
129                 (void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
130                 (void *)((unsigned long)(iomapp->iomap_delta)),
131                 (void *)((unsigned long)(iomapp->iomap_bsize)),
132                 (void *)((unsigned long)(iomapp->iomap_bn)),
133                 (void *)(__psint_t)(imapp->br_startoff),
134                 (void *)((unsigned long)(imapp->br_blockcount)),
135                 (void *)(__psint_t)(imapp->br_startblock));
136 }
137 #else
138 #define xfs_iomap_enter_trace(tag, io, offset, count)
139 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
140 #endif
141
142 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
143                                                 << mp->m_writeio_log)
144 #define XFS_STRAT_WRITE_IMAPS   2
145 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
146
147 STATIC int
148 xfs_imap_to_bmap(
149         xfs_iocore_t    *io,
150         xfs_off_t       offset,
151         xfs_bmbt_irec_t *imap,
152         xfs_iomap_t     *iomapp,
153         int             imaps,                  /* Number of imap entries */
154         int             iomaps,                 /* Number of iomap entries */
155         int             flags)
156 {
157         xfs_mount_t     *mp;
158         xfs_fsize_t     nisize;
159         int             pbm;
160         xfs_fsblock_t   start_block;
161
162         mp = io->io_mount;
163         nisize = XFS_SIZE(mp, io);
164         if (io->io_new_size > nisize)
165                 nisize = io->io_new_size;
166
167         for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
168                 iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
169                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
170                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
171                 iomapp->iomap_flags = flags;
172
173                 if (io->io_flags & XFS_IOCORE_RT) {
174                         iomapp->iomap_flags |= IOMAP_REALTIME;
175                         iomapp->iomap_target = mp->m_rtdev_targp;
176                 } else {
177                         iomapp->iomap_target = mp->m_ddev_targp;
178                 }
179                 start_block = imap->br_startblock;
180                 if (start_block == HOLESTARTBLOCK) {
181                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
182                         iomapp->iomap_flags |= IOMAP_HOLE;
183                 } else if (start_block == DELAYSTARTBLOCK) {
184                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
185                         iomapp->iomap_flags |= IOMAP_DELAY;
186                 } else {
187                         iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
188                         if (ISUNWRITTEN(imap))
189                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
190                 }
191
192                 if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
193                         iomapp->iomap_flags |= IOMAP_EOF;
194                 }
195
196                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
197         }
198         return pbm;     /* Return the number filled */
199 }
200
201 int
202 xfs_iomap(
203         xfs_iocore_t    *io,
204         xfs_off_t       offset,
205         ssize_t         count,
206         int             flags,
207         xfs_iomap_t     *iomapp,
208         int             *niomaps)
209 {
210         xfs_mount_t     *mp = io->io_mount;
211         xfs_fileoff_t   offset_fsb, end_fsb;
212         int             error = 0;
213         int             lockmode = 0;
214         xfs_bmbt_irec_t imap;
215         int             nimaps = 1;
216         int             bmapi_flags = 0;
217         int             iomap_flags = 0;
218
219         if (XFS_FORCED_SHUTDOWN(mp))
220                 return XFS_ERROR(EIO);
221
222         switch (flags &
223                 (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
224                  BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
225         case BMAPI_READ:
226                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
227                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
228                 bmapi_flags = XFS_BMAPI_ENTIRE;
229                 if (flags & BMAPI_IGNSTATE)
230                         bmapi_flags |= XFS_BMAPI_IGSTATE;
231                 break;
232         case BMAPI_WRITE:
233                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
234                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
235                 bmapi_flags = 0;
236                 XFS_ILOCK(mp, io, lockmode);
237                 break;
238         case BMAPI_ALLOCATE:
239                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
240                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
241                 bmapi_flags = XFS_BMAPI_ENTIRE;
242                 /* Attempt non-blocking lock */
243                 if (flags & BMAPI_TRYLOCK) {
244                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
245                                 return XFS_ERROR(EAGAIN);
246                 } else {
247                         XFS_ILOCK(mp, io, lockmode);
248                 }
249                 break;
250         case BMAPI_UNWRITTEN:
251                 goto phase2;
252         case BMAPI_DEVICE:
253                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
254                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
255                         mp->m_rtdev_targp : mp->m_ddev_targp;
256                 error = 0;
257                 *niomaps = 1;
258                 goto out;
259         default:
260                 BUG();
261         }
262
263         ASSERT(offset <= mp->m_maxioffset);
264         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
265                 count = mp->m_maxioffset - offset;
266         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
267         offset_fsb = XFS_B_TO_FSBT(mp, offset);
268
269         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
270                         (xfs_filblks_t)(end_fsb - offset_fsb),
271                         bmapi_flags,  NULL, 0, &imap,
272                         &nimaps, NULL);
273
274         if (error)
275                 goto out;
276
277 phase2:
278         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
279         case BMAPI_WRITE:
280                 /* If we found an extent, return it */
281                 if (nimaps &&
282                     (imap.br_startblock != HOLESTARTBLOCK) && 
283                     (imap.br_startblock != DELAYSTARTBLOCK)) {
284                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
285                                         offset, count, iomapp, &imap, flags);
286                         break;
287                 }
288
289                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
290                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
291                                         count, flags, &imap, &nimaps, nimaps);
292                 } else {
293                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
294                                         flags, &imap, &nimaps);
295                 }
296                 if (!error) {
297                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
298                                         offset, count, iomapp, &imap, flags);
299                 }
300                 iomap_flags = IOMAP_NEW;
301                 break;
302         case BMAPI_ALLOCATE:
303                 /* If we found an extent, return it */
304                 XFS_IUNLOCK(mp, io, lockmode);
305                 lockmode = 0;
306
307                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
308                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
309                                         offset, count, iomapp, &imap, flags);
310                         break;
311                 }
312
313                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
314                                                  &imap, &nimaps);
315                 break;
316         case BMAPI_UNWRITTEN:
317                 lockmode = 0;
318                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
319                 nimaps = 0;
320                 break;
321         }
322
323         if (nimaps) {
324                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
325                                             iomapp, nimaps, *niomaps, iomap_flags);
326         } else if (niomaps) {
327                 *niomaps = 0;
328         }
329
330 out:
331         if (lockmode)
332                 XFS_IUNLOCK(mp, io, lockmode);
333         return XFS_ERROR(error);
334 }
335
336 STATIC int
337 xfs_flush_space(
338         xfs_inode_t     *ip,
339         int             *fsynced,
340         int             *ioflags)
341 {
342         switch (*fsynced) {
343         case 0:
344                 if (ip->i_delayed_blks) {
345                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
346                         xfs_flush_inode(ip);
347                         xfs_ilock(ip, XFS_ILOCK_EXCL);
348                         *fsynced = 1;
349                 } else {
350                         *ioflags |= BMAPI_SYNC;
351                         *fsynced = 2;
352                 }
353                 return 0;
354         case 1:
355                 *fsynced = 2;
356                 *ioflags |= BMAPI_SYNC;
357                 return 0;
358         case 2:
359                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
360                 xfs_flush_device(ip);
361                 xfs_ilock(ip, XFS_ILOCK_EXCL);
362                 *fsynced = 3;
363                 return 0;
364         }
365         return 1;
366 }
367
368 int
369 xfs_iomap_write_direct(
370         xfs_inode_t     *ip,
371         xfs_off_t       offset,
372         size_t          count,
373         int             flags,
374         xfs_bmbt_irec_t *ret_imap,
375         int             *nmaps,
376         int             found)
377 {
378         xfs_mount_t     *mp = ip->i_mount;
379         xfs_iocore_t    *io = &ip->i_iocore;
380         xfs_fileoff_t   offset_fsb;
381         xfs_fileoff_t   last_fsb;
382         xfs_filblks_t   count_fsb;
383         xfs_fsize_t     isize;
384         xfs_fsblock_t   firstfsb;
385         int             nimaps, maps;
386         int             error;
387         int             bmapi_flag;
388         int             quota_flag;
389         int             rt;
390         xfs_trans_t     *tp;
391         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS], *imapp;
392         xfs_bmap_free_t free_list;
393         int             aeof;
394         xfs_filblks_t   datablocks, qblocks, resblks;
395         int             committed;
396         int             numrtextents;
397
398         /*
399          * Make sure that the dquots are there. This doesn't hold
400          * the ilock across a disk read.
401          */
402         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
403         if (error)
404                 return XFS_ERROR(error);
405
406         maps = min(XFS_WRITE_IMAPS, *nmaps);
407         nimaps = maps;
408
409         isize = ip->i_d.di_size;
410         aeof = (offset + count) > isize;
411
412         if (io->io_new_size > isize)
413                 isize = io->io_new_size;
414
415         offset_fsb = XFS_B_TO_FSBT(mp, offset);
416         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
417         count_fsb = last_fsb - offset_fsb;
418         if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
419                 xfs_fileoff_t   map_last_fsb;
420
421                 map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
422                 if (map_last_fsb < last_fsb) {
423                         last_fsb = map_last_fsb;
424                         count_fsb = last_fsb - offset_fsb;
425                 }
426                 ASSERT(count_fsb > 0);
427         }
428
429         /*
430          * Determine if reserving space on the data or realtime partition.
431          */
432         if ((rt = XFS_IS_REALTIME_INODE(ip))) {
433                 xfs_extlen_t    extsz;
434
435                 if (!(extsz = ip->i_d.di_extsize))
436                         extsz = mp->m_sb.sb_rextsize;
437                 numrtextents = qblocks = (count_fsb + extsz - 1);
438                 do_div(numrtextents, mp->m_sb.sb_rextsize);
439                 quota_flag = XFS_QMOPT_RES_RTBLKS;
440                 datablocks = 0;
441         } else {
442                 datablocks = qblocks = count_fsb;
443                 quota_flag = XFS_QMOPT_RES_REGBLKS;
444                 numrtextents = 0;
445         }
446
447         /*
448          * Allocate and setup the transaction
449          */
450         xfs_iunlock(ip, XFS_ILOCK_EXCL);
451         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
452         resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
453         error = xfs_trans_reserve(tp, resblks,
454                         XFS_WRITE_LOG_RES(mp), numrtextents,
455                         XFS_TRANS_PERM_LOG_RES,
456                         XFS_WRITE_LOG_COUNT);
457
458         /*
459          * Check for running out of space, note: need lock to return
460          */
461         if (error)
462                 xfs_trans_cancel(tp, 0);
463         xfs_ilock(ip, XFS_ILOCK_EXCL);
464         if (error)
465                 goto error_out;
466
467         if (XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag)) {
468                 error = (EDQUOT);
469                 goto error1;
470         }
471
472         bmapi_flag = XFS_BMAPI_WRITE;
473         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
474         xfs_trans_ihold(tp, ip);
475
476         if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
477                 bmapi_flag |= XFS_BMAPI_PREALLOC;
478
479         /*
480          * Issue the bmapi() call to allocate the blocks
481          */
482         XFS_BMAP_INIT(&free_list, &firstfsb);
483         nimaps = 1;
484         imapp = &imap[0];
485         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
486                 bmapi_flag, &firstfsb, 0, imapp, &nimaps, &free_list);
487         if (error)
488                 goto error0;
489
490         /*
491          * Complete the transaction
492          */
493         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
494         if (error)
495                 goto error0;
496         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
497         if (error)
498                 goto error_out;
499
500         /*
501          * Copy any maps to caller's array and return any error.
502          */
503         if (nimaps == 0) {
504                 error = (ENOSPC);
505                 goto error_out;
506         }
507
508         *ret_imap = imap[0];
509         *nmaps = 1;
510         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
511                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
512                         "start_block : %llx start_off : %llx blkcnt : %llx "
513                         "extent-state : %x \n",
514                         (ip->i_mount)->m_fsname,
515                         (long long)ip->i_ino,
516                         ret_imap->br_startblock, ret_imap->br_startoff,
517                         ret_imap->br_blockcount,ret_imap->br_state);
518         }
519         return 0;
520
521 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
522         xfs_bmap_cancel(&free_list);
523         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
524
525 error1: /* Just cancel transaction */
526         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
527         *nmaps = 0;     /* nothing set-up here */
528
529 error_out:
530         return XFS_ERROR(error);
531 }
532
533 int
534 xfs_iomap_write_delay(
535         xfs_inode_t     *ip,
536         xfs_off_t       offset,
537         size_t          count,
538         int             ioflag,
539         xfs_bmbt_irec_t *ret_imap,
540         int             *nmaps)
541 {
542         xfs_mount_t     *mp = ip->i_mount;
543         xfs_iocore_t    *io = &ip->i_iocore;
544         xfs_fileoff_t   offset_fsb;
545         xfs_fileoff_t   last_fsb;
546         xfs_fsize_t     isize;
547         xfs_fsblock_t   firstblock;
548         int             nimaps;
549         int             error;
550         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
551         int             aeof;
552         int             fsynced = 0;
553
554         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
555
556         /*
557          * Make sure that the dquots are there. This doesn't hold
558          * the ilock across a disk read.
559          */
560
561         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
562         if (error)
563                 return XFS_ERROR(error);
564
565 retry:
566         isize = ip->i_d.di_size;
567         if (io->io_new_size > isize) {
568                 isize = io->io_new_size;
569         }
570
571         aeof = 0;
572         offset_fsb = XFS_B_TO_FSBT(mp, offset);
573         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
574         /*
575          * If the caller is doing a write at the end of the file,
576          * then extend the allocation (and the buffer used for the write)
577          * out to the file system's write iosize.  We clean up any extra
578          * space left over when the file is closed in xfs_inactive().
579          *
580          * For sync writes, we are flushing delayed allocate space to
581          * try to make additional space available for allocation near
582          * the filesystem full boundary - preallocation hurts in that
583          * situation, of course.
584          */
585         if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
586                 xfs_off_t       aligned_offset;
587                 xfs_filblks_t   count_fsb;
588                 unsigned int    iosize;
589                 xfs_fileoff_t   ioalign;
590                 int             n;
591                 xfs_fileoff_t   start_fsb;
592
593                 /*
594                  * If there are any real blocks past eof, then don't
595                  * do any speculative allocation.
596                  */
597                 start_fsb = XFS_B_TO_FSBT(mp,
598                                         ((xfs_ufsize_t)(offset + count - 1)));
599                 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
600                 while (count_fsb > 0) {
601                         nimaps = XFS_WRITE_IMAPS;
602                         error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
603                                         0, &firstblock, 0, imap, &nimaps, NULL);
604                         if (error) {
605                                 return error;
606                         }
607                         for (n = 0; n < nimaps; n++) {
608                                 if ( !(io->io_flags & XFS_IOCORE_RT)  && 
609                                         !imap[n].br_startblock) {
610                                         cmn_err(CE_PANIC,"Access to block "
611                                                 "zero:  fs <%s> inode: %lld "
612                                                 "start_block : %llx start_off "
613                                                 ": %llx blkcnt : %llx "
614                                                 "extent-state : %x \n",
615                                                 (ip->i_mount)->m_fsname,
616                                                 (long long)ip->i_ino,
617                                                 imap[n].br_startblock,
618                                                 imap[n].br_startoff,
619                                                 imap[n].br_blockcount,
620                                                 imap[n].br_state);
621                                 }
622                                 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
623                                     (imap[n].br_startblock != DELAYSTARTBLOCK)) {
624                                         goto write_map;
625                                 }
626                                 start_fsb += imap[n].br_blockcount;
627                                 count_fsb -= imap[n].br_blockcount;
628                         }
629                 }
630                 iosize = mp->m_writeio_blocks;
631                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
632                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
633                 last_fsb = ioalign + iosize;
634                 aeof = 1;
635         }
636 write_map:
637         nimaps = XFS_WRITE_IMAPS;
638         firstblock = NULLFSBLOCK;
639
640         /*
641          * If mounted with the "-o swalloc" option, roundup the allocation
642          * request to a stripe width boundary if the file size is >=
643          * stripe width and we are allocating past the allocation eof.
644          */
645         if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_swidth 
646             && (mp->m_flags & XFS_MOUNT_SWALLOC)
647             && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
648                 int eof;
649                 xfs_fileoff_t new_last_fsb;
650
651                 new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
652                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
653                 if (error) {
654                         return error;
655                 }
656                 if (eof) {
657                         last_fsb = new_last_fsb;
658                 }
659         /*
660          * Roundup the allocation request to a stripe unit (m_dalign) boundary
661          * if the file size is >= stripe unit size, and we are allocating past
662          * the allocation eof.
663          */
664         } else if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_dalign &&
665                    (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)) && aeof) {
666                 int eof;
667                 xfs_fileoff_t new_last_fsb;
668                 new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
669                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
670                 if (error) {
671                         return error;
672                 }
673                 if (eof) {
674                         last_fsb = new_last_fsb;
675                 }
676         /*
677          * Round up the allocation request to a real-time extent boundary
678          * if the file is on the real-time subvolume.
679          */
680         } else if (io->io_flags & XFS_IOCORE_RT && aeof) {
681                 int eof;
682                 xfs_fileoff_t new_last_fsb;
683
684                 new_last_fsb = roundup_64(last_fsb, mp->m_sb.sb_rextsize);
685                 error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
686                 if (error) {
687                         return error;
688                 }
689                 if (eof)
690                         last_fsb = new_last_fsb;
691         }
692         error = xfs_bmapi(NULL, ip, offset_fsb,
693                           (xfs_filblks_t)(last_fsb - offset_fsb),
694                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
695                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
696                           &nimaps, NULL);
697         /*
698          * This can be EDQUOT, if nimaps == 0
699          */
700         if (error && (error != ENOSPC)) {
701                 return XFS_ERROR(error);
702         }
703         /*
704          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
705          * then we must have run out of space.
706          */
707         if (nimaps == 0) {
708                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
709                                         io, offset, count);
710                 if (xfs_flush_space(ip, &fsynced, &ioflag))
711                         return XFS_ERROR(ENOSPC);
712
713                 error = 0;
714                 goto retry;
715         }
716
717         *ret_imap = imap[0];
718         *nmaps = 1;
719         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
720                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
721                         "start_block : %llx start_off : %llx blkcnt : %llx "
722                         "extent-state : %x \n",
723                         (ip->i_mount)->m_fsname,
724                         (long long)ip->i_ino,
725                         ret_imap->br_startblock, ret_imap->br_startoff,
726                         ret_imap->br_blockcount,ret_imap->br_state);
727         }
728         return 0;
729 }
730
731 /*
732  * Pass in a delayed allocate extent, convert it to real extents;
733  * return to the caller the extent we create which maps on top of
734  * the originating callers request.
735  *
736  * Called without a lock on the inode.
737  */
738 int
739 xfs_iomap_write_allocate(
740         xfs_inode_t     *ip,
741         xfs_off_t       offset,
742         size_t          count,
743         xfs_bmbt_irec_t *map,
744         int             *retmap)
745 {
746         xfs_mount_t     *mp = ip->i_mount;
747         xfs_iocore_t    *io = &ip->i_iocore;
748         xfs_fileoff_t   offset_fsb, last_block;
749         xfs_fileoff_t   end_fsb, map_start_fsb;
750         xfs_fsblock_t   first_block;
751         xfs_bmap_free_t free_list;
752         xfs_filblks_t   count_fsb;
753         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
754         xfs_trans_t     *tp;
755         int             i, nimaps, committed;
756         int             error = 0;
757         int             nres;
758
759         *retmap = 0;
760
761         /*
762          * Make sure that the dquots are there.
763          */
764         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
765                 return XFS_ERROR(error);
766
767         offset_fsb = XFS_B_TO_FSBT(mp, offset);
768         count_fsb = map->br_blockcount;
769         map_start_fsb = map->br_startoff;
770
771         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
772
773         while (count_fsb != 0) {
774                 /*
775                  * Set up a transaction with which to allocate the
776                  * backing store for the file.  Do allocations in a
777                  * loop until we get some space in the range we are
778                  * interested in.  The other space that might be allocated
779                  * is in the delayed allocation extent on which we sit
780                  * but before our buffer starts.
781                  */
782
783                 nimaps = 0;
784                 while (nimaps == 0) {
785                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
786                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
787                         error = xfs_trans_reserve(tp, nres,
788                                         XFS_WRITE_LOG_RES(mp),
789                                         0, XFS_TRANS_PERM_LOG_RES,
790                                         XFS_WRITE_LOG_COUNT);
791                         if (error == ENOSPC) {
792                                 error = xfs_trans_reserve(tp, 0,
793                                                 XFS_WRITE_LOG_RES(mp),
794                                                 0,
795                                                 XFS_TRANS_PERM_LOG_RES,
796                                                 XFS_WRITE_LOG_COUNT);
797                         }
798                         if (error) {
799                                 xfs_trans_cancel(tp, 0);
800                                 return XFS_ERROR(error);
801                         }
802                         xfs_ilock(ip, XFS_ILOCK_EXCL);
803                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
804                         xfs_trans_ihold(tp, ip);
805
806                         XFS_BMAP_INIT(&free_list, &first_block);
807
808                         nimaps = XFS_STRAT_WRITE_IMAPS;
809                         /*
810                          * Ensure we don't go beyond eof - it is possible
811                          * the extents changed since we did the read call,
812                          * we dropped the ilock in the interim.
813                          */
814
815                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
816                         xfs_bmap_last_offset(NULL, ip, &last_block,
817                                 XFS_DATA_FORK);
818                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
819                         if ((map_start_fsb + count_fsb) > last_block) {
820                                 count_fsb = last_block - map_start_fsb;
821                                 if (count_fsb == 0) {
822                                         error = EAGAIN;
823                                         goto trans_cancel;
824                                 }
825                         }
826
827                         /* Go get the actual blocks */
828                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
829                                         XFS_BMAPI_WRITE, &first_block, 1,
830                                         imap, &nimaps, &free_list);
831                         if (error)
832                                 goto trans_cancel;
833
834                         error = xfs_bmap_finish(&tp, &free_list,
835                                         first_block, &committed);
836                         if (error)
837                                 goto trans_cancel;
838
839                         error = xfs_trans_commit(tp,
840                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
841                         if (error)
842                                 goto error0;
843
844                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
845                 }
846
847                 /*
848                  * See if we were able to allocate an extent that
849                  * covers at least part of the callers request
850                  */
851
852                 for (i = 0; i < nimaps; i++) {
853                         if ( !(io->io_flags & XFS_IOCORE_RT)  && 
854                                 !imap[i].br_startblock) {
855                                 cmn_err(CE_PANIC,"Access to block zero:  "
856                                         "fs <%s> inode: %lld "
857                                         "start_block : %llx start_off : %llx " 
858                                         "blkcnt : %llx extent-state : %x \n",
859                                         (ip->i_mount)->m_fsname,
860                                         (long long)ip->i_ino,
861                                         imap[i].br_startblock,
862                                         imap[i].br_startoff,
863                                         imap[i].br_blockcount,imap[i].br_state);
864                         }
865                         if ((offset_fsb >= imap[i].br_startoff) &&
866                             (offset_fsb < (imap[i].br_startoff +
867                                            imap[i].br_blockcount))) {
868                                 *map = imap[i];
869                                 *retmap = 1;
870                                 XFS_STATS_INC(xs_xstrat_quick);
871                                 return 0;
872                         }
873                         count_fsb -= imap[i].br_blockcount;
874                 }
875
876                 /* So far we have not mapped the requested part of the
877                  * file, just surrounding data, try again.
878                  */
879                 nimaps--;
880                 map_start_fsb = imap[nimaps].br_startoff +
881                                 imap[nimaps].br_blockcount;
882         }
883
884 trans_cancel:
885         xfs_bmap_cancel(&free_list);
886         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
887 error0:
888         xfs_iunlock(ip, XFS_ILOCK_EXCL);
889         return XFS_ERROR(error);
890 }
891
892 int
893 xfs_iomap_write_unwritten(
894         xfs_inode_t     *ip,
895         xfs_off_t       offset,
896         size_t          count)
897 {
898         xfs_mount_t     *mp = ip->i_mount;
899         xfs_iocore_t    *io = &ip->i_iocore;
900         xfs_trans_t     *tp;
901         xfs_fileoff_t   offset_fsb;
902         xfs_filblks_t   count_fsb;
903         xfs_filblks_t   numblks_fsb;
904         xfs_bmbt_irec_t imap;
905         int             committed;
906         int             error;
907         int             nres;
908         int             nimaps;
909         xfs_fsblock_t   firstfsb;
910         xfs_bmap_free_t free_list;
911
912         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
913                                 &ip->i_iocore, offset, count);
914
915         offset_fsb = XFS_B_TO_FSBT(mp, offset);
916         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
917         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
918
919         do {
920                 nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
921
922                 /*
923                  * set up a transaction to convert the range of extents
924                  * from unwritten to real. Do allocations in a loop until
925                  * we have covered the range passed in.
926                  */
927
928                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
929                 error = xfs_trans_reserve(tp, nres,
930                                 XFS_WRITE_LOG_RES(mp), 0,
931                                 XFS_TRANS_PERM_LOG_RES,
932                                 XFS_WRITE_LOG_COUNT);
933                 if (error) {
934                         xfs_trans_cancel(tp, 0);
935                         goto error0;
936                 }
937
938                 xfs_ilock(ip, XFS_ILOCK_EXCL);
939                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
940                 xfs_trans_ihold(tp, ip);
941
942                 /*
943                  * Modify the unwritten extent state of the buffer.
944                  */
945                 XFS_BMAP_INIT(&free_list, &firstfsb);
946                 nimaps = 1;
947                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
948                                   XFS_BMAPI_WRITE, &firstfsb,
949                                   1, &imap, &nimaps, &free_list);
950                 if (error)
951                         goto error_on_bmapi_transaction;
952
953                 error = xfs_bmap_finish(&(tp), &(free_list),
954                                 firstfsb, &committed);
955                 if (error)
956                         goto error_on_bmapi_transaction;
957
958                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
959                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
960                 if (error)
961                         goto error0;
962                 
963                 if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
964                         cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
965                                 "inode: %lld start_block : %llx start_off : "
966                                 "%llx blkcnt : %llx extent-state : %x \n",
967                                 (ip->i_mount)->m_fsname,
968                                 (long long)ip->i_ino,
969                                 imap.br_startblock,imap.br_startoff,
970                                 imap.br_blockcount,imap.br_state);
971                 }
972
973                 if ((numblks_fsb = imap.br_blockcount) == 0) {
974                         /*
975                          * The numblks_fsb value should always get
976                          * smaller, otherwise the loop is stuck.
977                          */
978                         ASSERT(imap.br_blockcount);
979                         break;
980                 }
981                 offset_fsb += numblks_fsb;
982                 count_fsb -= numblks_fsb;
983         } while (count_fsb > 0);
984
985         return 0;
986
987 error_on_bmapi_transaction:
988         xfs_bmap_cancel(&free_list);
989         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
990         xfs_iunlock(ip, XFS_ILOCK_EXCL);
991 error0:
992         return XFS_ERROR(error);
993 }