[XFS] Fix incorrect use of BMAPI_READ in unwritten extent handling
[linux-2.6.git] / 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                 break;
230         case BMAPI_WRITE:
231                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
232                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
233                 if (flags & BMAPI_IGNSTATE)
234                         bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
235                 XFS_ILOCK(mp, io, lockmode);
236                 break;
237         case BMAPI_ALLOCATE:
238                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
239                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
240                 bmapi_flags = XFS_BMAPI_ENTIRE;
241                 /* Attempt non-blocking lock */
242                 if (flags & BMAPI_TRYLOCK) {
243                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
244                                 return XFS_ERROR(EAGAIN);
245                 } else {
246                         XFS_ILOCK(mp, io, lockmode);
247                 }
248                 break;
249         case BMAPI_UNWRITTEN:
250                 goto phase2;
251         case BMAPI_DEVICE:
252                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
253                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
254                         mp->m_rtdev_targp : mp->m_ddev_targp;
255                 error = 0;
256                 *niomaps = 1;
257                 goto out;
258         default:
259                 BUG();
260         }
261
262         ASSERT(offset <= mp->m_maxioffset);
263         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
264                 count = mp->m_maxioffset - offset;
265         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
266         offset_fsb = XFS_B_TO_FSBT(mp, offset);
267
268         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
269                         (xfs_filblks_t)(end_fsb - offset_fsb),
270                         bmapi_flags,  NULL, 0, &imap,
271                         &nimaps, NULL);
272
273         if (error)
274                 goto out;
275
276 phase2:
277         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
278         case BMAPI_WRITE:
279                 /* If we found an extent, return it */
280                 if (nimaps &&
281                     (imap.br_startblock != HOLESTARTBLOCK) && 
282                     (imap.br_startblock != DELAYSTARTBLOCK)) {
283                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
284                                         offset, count, iomapp, &imap, flags);
285                         break;
286                 }
287
288                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
289                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
290                                         count, flags, &imap, &nimaps, nimaps);
291                 } else {
292                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
293                                         flags, &imap, &nimaps);
294                 }
295                 if (!error) {
296                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
297                                         offset, count, iomapp, &imap, flags);
298                 }
299                 iomap_flags = IOMAP_NEW;
300                 break;
301         case BMAPI_ALLOCATE:
302                 /* If we found an extent, return it */
303                 XFS_IUNLOCK(mp, io, lockmode);
304                 lockmode = 0;
305
306                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
307                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
308                                         offset, count, iomapp, &imap, flags);
309                         break;
310                 }
311
312                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
313                                                  &imap, &nimaps);
314                 break;
315         case BMAPI_UNWRITTEN:
316                 lockmode = 0;
317                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
318                 nimaps = 0;
319                 break;
320         }
321
322         if (nimaps) {
323                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
324                                             iomapp, nimaps, *niomaps, iomap_flags);
325         } else if (niomaps) {
326                 *niomaps = 0;
327         }
328
329 out:
330         if (lockmode)
331                 XFS_IUNLOCK(mp, io, lockmode);
332         return XFS_ERROR(error);
333 }
334
335 STATIC int
336 xfs_flush_space(
337         xfs_inode_t     *ip,
338         int             *fsynced,
339         int             *ioflags)
340 {
341         switch (*fsynced) {
342         case 0:
343                 if (ip->i_delayed_blks) {
344                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
345                         xfs_flush_inode(ip);
346                         xfs_ilock(ip, XFS_ILOCK_EXCL);
347                         *fsynced = 1;
348                 } else {
349                         *ioflags |= BMAPI_SYNC;
350                         *fsynced = 2;
351                 }
352                 return 0;
353         case 1:
354                 *fsynced = 2;
355                 *ioflags |= BMAPI_SYNC;
356                 return 0;
357         case 2:
358                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
359                 xfs_flush_device(ip);
360                 xfs_ilock(ip, XFS_ILOCK_EXCL);
361                 *fsynced = 3;
362                 return 0;
363         }
364         return 1;
365 }
366
367 int
368 xfs_iomap_write_direct(
369         xfs_inode_t     *ip,
370         xfs_off_t       offset,
371         size_t          count,
372         int             flags,
373         xfs_bmbt_irec_t *ret_imap,
374         int             *nmaps,
375         int             found)
376 {
377         xfs_mount_t     *mp = ip->i_mount;
378         xfs_iocore_t    *io = &ip->i_iocore;
379         xfs_fileoff_t   offset_fsb;
380         xfs_fileoff_t   last_fsb;
381         xfs_filblks_t   count_fsb;
382         xfs_fsize_t     isize;
383         xfs_fsblock_t   firstfsb;
384         int             nimaps, maps;
385         int             error;
386         int             bmapi_flag;
387         int             quota_flag;
388         int             rt;
389         xfs_trans_t     *tp;
390         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS], *imapp;
391         xfs_bmap_free_t free_list;
392         int             aeof;
393         xfs_filblks_t   qblocks, resblks;
394         int             committed;
395         int             resrtextents;
396
397         /*
398          * Make sure that the dquots are there. This doesn't hold
399          * the ilock across a disk read.
400          */
401         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
402         if (error)
403                 return XFS_ERROR(error);
404
405         maps = min(XFS_WRITE_IMAPS, *nmaps);
406         nimaps = maps;
407
408         isize = ip->i_d.di_size;
409         aeof = (offset + count) > isize;
410
411         if (io->io_new_size > isize)
412                 isize = io->io_new_size;
413
414         offset_fsb = XFS_B_TO_FSBT(mp, offset);
415         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
416         count_fsb = last_fsb - offset_fsb;
417         if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
418                 xfs_fileoff_t   map_last_fsb;
419
420                 map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
421                 if (map_last_fsb < last_fsb) {
422                         last_fsb = map_last_fsb;
423                         count_fsb = last_fsb - offset_fsb;
424                 }
425                 ASSERT(count_fsb > 0);
426         }
427
428         /*
429          * Determine if reserving space on the data or realtime partition.
430          */
431         if ((rt = XFS_IS_REALTIME_INODE(ip))) {
432                 xfs_extlen_t    extsz;
433
434                 if (!(extsz = ip->i_d.di_extsize))
435                         extsz = mp->m_sb.sb_rextsize;
436                 resrtextents = qblocks = (count_fsb + extsz - 1);
437                 do_div(resrtextents, mp->m_sb.sb_rextsize);
438                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
439                 quota_flag = XFS_QMOPT_RES_RTBLKS;
440         } else {
441                 resrtextents = 0;
442                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, count_fsb);
443                 quota_flag = XFS_QMOPT_RES_REGBLKS;
444         }
445
446         /*
447          * Allocate and setup the transaction
448          */
449         xfs_iunlock(ip, XFS_ILOCK_EXCL);
450         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
451         error = xfs_trans_reserve(tp, resblks,
452                         XFS_WRITE_LOG_RES(mp), resrtextents,
453                         XFS_TRANS_PERM_LOG_RES,
454                         XFS_WRITE_LOG_COUNT);
455
456         /*
457          * Check for running out of space, note: need lock to return
458          */
459         if (error)
460                 xfs_trans_cancel(tp, 0);
461         xfs_ilock(ip, XFS_ILOCK_EXCL);
462         if (error)
463                 goto error_out;
464
465         if (XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag)) {
466                 error = (EDQUOT);
467                 goto error1;
468         }
469
470         bmapi_flag = XFS_BMAPI_WRITE;
471         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
472         xfs_trans_ihold(tp, ip);
473
474         if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
475                 bmapi_flag |= XFS_BMAPI_PREALLOC;
476
477         /*
478          * Issue the bmapi() call to allocate the blocks
479          */
480         XFS_BMAP_INIT(&free_list, &firstfsb);
481         nimaps = 1;
482         imapp = &imap[0];
483         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
484                 bmapi_flag, &firstfsb, 0, imapp, &nimaps, &free_list);
485         if (error)
486                 goto error0;
487
488         /*
489          * Complete the transaction
490          */
491         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
492         if (error)
493                 goto error0;
494         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
495         if (error)
496                 goto error_out;
497
498         /*
499          * Copy any maps to caller's array and return any error.
500          */
501         if (nimaps == 0) {
502                 error = (ENOSPC);
503                 goto error_out;
504         }
505
506         *ret_imap = imap[0];
507         *nmaps = 1;
508         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
509                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
510                         "start_block : %llx start_off : %llx blkcnt : %llx "
511                         "extent-state : %x \n",
512                         (ip->i_mount)->m_fsname,
513                         (long long)ip->i_ino,
514                         ret_imap->br_startblock, ret_imap->br_startoff,
515                         ret_imap->br_blockcount,ret_imap->br_state);
516         }
517         return 0;
518
519 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
520         xfs_bmap_cancel(&free_list);
521         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
522
523 error1: /* Just cancel transaction */
524         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
525         *nmaps = 0;     /* nothing set-up here */
526
527 error_out:
528         return XFS_ERROR(error);
529 }
530
531 int
532 xfs_iomap_write_delay(
533         xfs_inode_t     *ip,
534         xfs_off_t       offset,
535         size_t          count,
536         int             ioflag,
537         xfs_bmbt_irec_t *ret_imap,
538         int             *nmaps)
539 {
540         xfs_mount_t     *mp = ip->i_mount;
541         xfs_iocore_t    *io = &ip->i_iocore;
542         xfs_fileoff_t   offset_fsb;
543         xfs_fileoff_t   last_fsb;
544         xfs_fsize_t     isize;
545         xfs_fsblock_t   firstblock;
546         int             nimaps;
547         int             error;
548         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
549         int             aeof;
550         int             fsynced = 0;
551
552         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
553
554         /*
555          * Make sure that the dquots are there. This doesn't hold
556          * the ilock across a disk read.
557          */
558
559         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
560         if (error)
561                 return XFS_ERROR(error);
562
563 retry:
564         isize = ip->i_d.di_size;
565         if (io->io_new_size > isize) {
566                 isize = io->io_new_size;
567         }
568
569         aeof = 0;
570         offset_fsb = XFS_B_TO_FSBT(mp, offset);
571         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
572         /*
573          * If the caller is doing a write at the end of the file,
574          * then extend the allocation (and the buffer used for the write)
575          * out to the file system's write iosize.  We clean up any extra
576          * space left over when the file is closed in xfs_inactive().
577          *
578          * For sync writes, we are flushing delayed allocate space to
579          * try to make additional space available for allocation near
580          * the filesystem full boundary - preallocation hurts in that
581          * situation, of course.
582          */
583         if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
584                 xfs_off_t       aligned_offset;
585                 xfs_filblks_t   count_fsb;
586                 unsigned int    iosize;
587                 xfs_fileoff_t   ioalign;
588                 int             n;
589                 xfs_fileoff_t   start_fsb;
590
591                 /*
592                  * If there are any real blocks past eof, then don't
593                  * do any speculative allocation.
594                  */
595                 start_fsb = XFS_B_TO_FSBT(mp,
596                                         ((xfs_ufsize_t)(offset + count - 1)));
597                 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
598                 while (count_fsb > 0) {
599                         nimaps = XFS_WRITE_IMAPS;
600                         error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
601                                         0, &firstblock, 0, imap, &nimaps, NULL);
602                         if (error) {
603                                 return error;
604                         }
605                         for (n = 0; n < nimaps; n++) {
606                                 if ( !(io->io_flags & XFS_IOCORE_RT)  && 
607                                         !imap[n].br_startblock) {
608                                         cmn_err(CE_PANIC,"Access to block "
609                                                 "zero:  fs <%s> inode: %lld "
610                                                 "start_block : %llx start_off "
611                                                 ": %llx blkcnt : %llx "
612                                                 "extent-state : %x \n",
613                                                 (ip->i_mount)->m_fsname,
614                                                 (long long)ip->i_ino,
615                                                 imap[n].br_startblock,
616                                                 imap[n].br_startoff,
617                                                 imap[n].br_blockcount,
618                                                 imap[n].br_state);
619                                 }
620                                 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
621                                     (imap[n].br_startblock != DELAYSTARTBLOCK)) {
622                                         goto write_map;
623                                 }
624                                 start_fsb += imap[n].br_blockcount;
625                                 count_fsb -= imap[n].br_blockcount;
626                         }
627                 }
628                 iosize = mp->m_writeio_blocks;
629                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
630                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
631                 last_fsb = ioalign + iosize;
632                 aeof = 1;
633         }
634 write_map:
635         nimaps = XFS_WRITE_IMAPS;
636         firstblock = NULLFSBLOCK;
637
638         /*
639          * If mounted with the "-o swalloc" option, roundup the allocation
640          * request to a stripe width boundary if the file size is >=
641          * stripe width and we are allocating past the allocation eof.
642          */
643         if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_swidth 
644             && (mp->m_flags & XFS_MOUNT_SWALLOC)
645             && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
646                 int eof;
647                 xfs_fileoff_t new_last_fsb;
648
649                 new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
650                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
651                 if (error) {
652                         return error;
653                 }
654                 if (eof) {
655                         last_fsb = new_last_fsb;
656                 }
657         /*
658          * Roundup the allocation request to a stripe unit (m_dalign) boundary
659          * if the file size is >= stripe unit size, and we are allocating past
660          * the allocation eof.
661          */
662         } else if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_dalign &&
663                    (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)) && aeof) {
664                 int eof;
665                 xfs_fileoff_t new_last_fsb;
666                 new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
667                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
668                 if (error) {
669                         return error;
670                 }
671                 if (eof) {
672                         last_fsb = new_last_fsb;
673                 }
674         /*
675          * Round up the allocation request to a real-time extent boundary
676          * if the file is on the real-time subvolume.
677          */
678         } else if (io->io_flags & XFS_IOCORE_RT && aeof) {
679                 int eof;
680                 xfs_fileoff_t new_last_fsb;
681
682                 new_last_fsb = roundup_64(last_fsb, mp->m_sb.sb_rextsize);
683                 error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
684                 if (error) {
685                         return error;
686                 }
687                 if (eof)
688                         last_fsb = new_last_fsb;
689         }
690         error = xfs_bmapi(NULL, ip, offset_fsb,
691                           (xfs_filblks_t)(last_fsb - offset_fsb),
692                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
693                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
694                           &nimaps, NULL);
695         /*
696          * This can be EDQUOT, if nimaps == 0
697          */
698         if (error && (error != ENOSPC)) {
699                 return XFS_ERROR(error);
700         }
701         /*
702          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
703          * then we must have run out of space.
704          */
705         if (nimaps == 0) {
706                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
707                                         io, offset, count);
708                 if (xfs_flush_space(ip, &fsynced, &ioflag))
709                         return XFS_ERROR(ENOSPC);
710
711                 error = 0;
712                 goto retry;
713         }
714
715         *ret_imap = imap[0];
716         *nmaps = 1;
717         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
718                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
719                         "start_block : %llx start_off : %llx blkcnt : %llx "
720                         "extent-state : %x \n",
721                         (ip->i_mount)->m_fsname,
722                         (long long)ip->i_ino,
723                         ret_imap->br_startblock, ret_imap->br_startoff,
724                         ret_imap->br_blockcount,ret_imap->br_state);
725         }
726         return 0;
727 }
728
729 /*
730  * Pass in a delayed allocate extent, convert it to real extents;
731  * return to the caller the extent we create which maps on top of
732  * the originating callers request.
733  *
734  * Called without a lock on the inode.
735  */
736 int
737 xfs_iomap_write_allocate(
738         xfs_inode_t     *ip,
739         xfs_off_t       offset,
740         size_t          count,
741         xfs_bmbt_irec_t *map,
742         int             *retmap)
743 {
744         xfs_mount_t     *mp = ip->i_mount;
745         xfs_iocore_t    *io = &ip->i_iocore;
746         xfs_fileoff_t   offset_fsb, last_block;
747         xfs_fileoff_t   end_fsb, map_start_fsb;
748         xfs_fsblock_t   first_block;
749         xfs_bmap_free_t free_list;
750         xfs_filblks_t   count_fsb;
751         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
752         xfs_trans_t     *tp;
753         int             i, nimaps, committed;
754         int             error = 0;
755         int             nres;
756
757         *retmap = 0;
758
759         /*
760          * Make sure that the dquots are there.
761          */
762         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
763                 return XFS_ERROR(error);
764
765         offset_fsb = XFS_B_TO_FSBT(mp, offset);
766         count_fsb = map->br_blockcount;
767         map_start_fsb = map->br_startoff;
768
769         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
770
771         while (count_fsb != 0) {
772                 /*
773                  * Set up a transaction with which to allocate the
774                  * backing store for the file.  Do allocations in a
775                  * loop until we get some space in the range we are
776                  * interested in.  The other space that might be allocated
777                  * is in the delayed allocation extent on which we sit
778                  * but before our buffer starts.
779                  */
780
781                 nimaps = 0;
782                 while (nimaps == 0) {
783                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
784                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
785                         error = xfs_trans_reserve(tp, nres,
786                                         XFS_WRITE_LOG_RES(mp),
787                                         0, XFS_TRANS_PERM_LOG_RES,
788                                         XFS_WRITE_LOG_COUNT);
789                         if (error == ENOSPC) {
790                                 error = xfs_trans_reserve(tp, 0,
791                                                 XFS_WRITE_LOG_RES(mp),
792                                                 0,
793                                                 XFS_TRANS_PERM_LOG_RES,
794                                                 XFS_WRITE_LOG_COUNT);
795                         }
796                         if (error) {
797                                 xfs_trans_cancel(tp, 0);
798                                 return XFS_ERROR(error);
799                         }
800                         xfs_ilock(ip, XFS_ILOCK_EXCL);
801                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
802                         xfs_trans_ihold(tp, ip);
803
804                         XFS_BMAP_INIT(&free_list, &first_block);
805
806                         nimaps = XFS_STRAT_WRITE_IMAPS;
807                         /*
808                          * Ensure we don't go beyond eof - it is possible
809                          * the extents changed since we did the read call,
810                          * we dropped the ilock in the interim.
811                          */
812
813                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
814                         xfs_bmap_last_offset(NULL, ip, &last_block,
815                                 XFS_DATA_FORK);
816                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
817                         if ((map_start_fsb + count_fsb) > last_block) {
818                                 count_fsb = last_block - map_start_fsb;
819                                 if (count_fsb == 0) {
820                                         error = EAGAIN;
821                                         goto trans_cancel;
822                                 }
823                         }
824
825                         /* Go get the actual blocks */
826                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
827                                         XFS_BMAPI_WRITE, &first_block, 1,
828                                         imap, &nimaps, &free_list);
829                         if (error)
830                                 goto trans_cancel;
831
832                         error = xfs_bmap_finish(&tp, &free_list,
833                                         first_block, &committed);
834                         if (error)
835                                 goto trans_cancel;
836
837                         error = xfs_trans_commit(tp,
838                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
839                         if (error)
840                                 goto error0;
841
842                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
843                 }
844
845                 /*
846                  * See if we were able to allocate an extent that
847                  * covers at least part of the callers request
848                  */
849
850                 for (i = 0; i < nimaps; i++) {
851                         if ( !(io->io_flags & XFS_IOCORE_RT)  && 
852                                 !imap[i].br_startblock) {
853                                 cmn_err(CE_PANIC,"Access to block zero:  "
854                                         "fs <%s> inode: %lld "
855                                         "start_block : %llx start_off : %llx " 
856                                         "blkcnt : %llx extent-state : %x \n",
857                                         (ip->i_mount)->m_fsname,
858                                         (long long)ip->i_ino,
859                                         imap[i].br_startblock,
860                                         imap[i].br_startoff,
861                                         imap[i].br_blockcount,imap[i].br_state);
862                         }
863                         if ((offset_fsb >= imap[i].br_startoff) &&
864                             (offset_fsb < (imap[i].br_startoff +
865                                            imap[i].br_blockcount))) {
866                                 *map = imap[i];
867                                 *retmap = 1;
868                                 XFS_STATS_INC(xs_xstrat_quick);
869                                 return 0;
870                         }
871                         count_fsb -= imap[i].br_blockcount;
872                 }
873
874                 /* So far we have not mapped the requested part of the
875                  * file, just surrounding data, try again.
876                  */
877                 nimaps--;
878                 map_start_fsb = imap[nimaps].br_startoff +
879                                 imap[nimaps].br_blockcount;
880         }
881
882 trans_cancel:
883         xfs_bmap_cancel(&free_list);
884         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
885 error0:
886         xfs_iunlock(ip, XFS_ILOCK_EXCL);
887         return XFS_ERROR(error);
888 }
889
890 int
891 xfs_iomap_write_unwritten(
892         xfs_inode_t     *ip,
893         xfs_off_t       offset,
894         size_t          count)
895 {
896         xfs_mount_t     *mp = ip->i_mount;
897         xfs_iocore_t    *io = &ip->i_iocore;
898         xfs_trans_t     *tp;
899         xfs_fileoff_t   offset_fsb;
900         xfs_filblks_t   count_fsb;
901         xfs_filblks_t   numblks_fsb;
902         xfs_bmbt_irec_t imap;
903         int             committed;
904         int             error;
905         int             nres;
906         int             nimaps;
907         xfs_fsblock_t   firstfsb;
908         xfs_bmap_free_t free_list;
909
910         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
911                                 &ip->i_iocore, offset, count);
912
913         offset_fsb = XFS_B_TO_FSBT(mp, offset);
914         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
915         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
916
917         do {
918                 nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
919
920                 /*
921                  * set up a transaction to convert the range of extents
922                  * from unwritten to real. Do allocations in a loop until
923                  * we have covered the range passed in.
924                  */
925
926                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
927                 error = xfs_trans_reserve(tp, nres,
928                                 XFS_WRITE_LOG_RES(mp), 0,
929                                 XFS_TRANS_PERM_LOG_RES,
930                                 XFS_WRITE_LOG_COUNT);
931                 if (error) {
932                         xfs_trans_cancel(tp, 0);
933                         goto error0;
934                 }
935
936                 xfs_ilock(ip, XFS_ILOCK_EXCL);
937                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
938                 xfs_trans_ihold(tp, ip);
939
940                 /*
941                  * Modify the unwritten extent state of the buffer.
942                  */
943                 XFS_BMAP_INIT(&free_list, &firstfsb);
944                 nimaps = 1;
945                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
946                                   XFS_BMAPI_WRITE, &firstfsb,
947                                   1, &imap, &nimaps, &free_list);
948                 if (error)
949                         goto error_on_bmapi_transaction;
950
951                 error = xfs_bmap_finish(&(tp), &(free_list),
952                                 firstfsb, &committed);
953                 if (error)
954                         goto error_on_bmapi_transaction;
955
956                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
957                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
958                 if (error)
959                         goto error0;
960                 
961                 if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
962                         cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
963                                 "inode: %lld start_block : %llx start_off : "
964                                 "%llx blkcnt : %llx extent-state : %x \n",
965                                 (ip->i_mount)->m_fsname,
966                                 (long long)ip->i_ino,
967                                 imap.br_startblock,imap.br_startoff,
968                                 imap.br_blockcount,imap.br_state);
969                 }
970
971                 if ((numblks_fsb = imap.br_blockcount) == 0) {
972                         /*
973                          * The numblks_fsb value should always get
974                          * smaller, otherwise the loop is stuck.
975                          */
976                         ASSERT(imap.br_blockcount);
977                         break;
978                 }
979                 offset_fsb += numblks_fsb;
980                 count_fsb -= numblks_fsb;
981         } while (count_fsb > 0);
982
983         return 0;
984
985 error_on_bmapi_transaction:
986         xfs_bmap_cancel(&free_list);
987         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
988         xfs_iunlock(ip, XFS_ILOCK_EXCL);
989 error0:
990         return XFS_ERROR(error);
991 }