SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/xfs/xfs_iomap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_bit.h"
  21 #include "xfs_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_alloc.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_quota.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_ialloc.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_bmap.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_itable.h"
  44 #include "xfs_rw.h"
  45 #include "xfs_attr.h"
  46 #include "xfs_buf_item.h"
  47 #include "xfs_trans_space.h"
  48 #include "xfs_utils.h"
  49 #include "xfs_iomap.h"
  50 #include "xfs_trace.h"
  51
  52
  53 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
  54                                                 << mp->m_writeio_log)
  55 #define XFS_STRAT_WRITE_IMAPS   2
  56 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
  57
  58 STATIC void
  59 xfs_imap_to_bmap(
  60         xfs_inode_t     *ip,
  61         xfs_off_t       offset,
  62         xfs_bmbt_irec_t *imap,
  63         xfs_iomap_t     *iomapp,
  64         int             imaps,                  /* Number of imap entries */
  65         int             flags)
  66 {
  67         xfs_mount_t     *mp = ip->i_mount;
  68         xfs_fsblock_t   start_block;
  69
  70         iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
  71         iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
  72         iomapp->iomap_flags = flags;
  73
  74         start_block = imap->br_startblock;
  75         if (start_block == HOLESTARTBLOCK) {
  76                 iomapp->iomap_bn = IOMAP_DADDR_NULL;
  77                 iomapp->iomap_flags |= IOMAP_HOLE;
  78         } else if (start_block == DELAYSTARTBLOCK) {
  79                 iomapp->iomap_bn = IOMAP_DADDR_NULL;
  80                 iomapp->iomap_flags |= IOMAP_DELAY;
  81         } else {
  82                 iomapp->iomap_bn = xfs_fsb_to_db(ip, start_block);
  83                 if (ISUNWRITTEN(imap))
  84                         iomapp->iomap_flags |= IOMAP_UNWRITTEN;
  85         }
  86 }
  87
  88 int
  89 xfs_iomap(
  90         xfs_inode_t     *ip,
  91         xfs_off_t       offset,
  92         ssize_t         count,
  93         int             flags,
  94         xfs_iomap_t     *iomapp,
  95         int             *niomaps)
  96 {
  97         xfs_mount_t     *mp = ip->i_mount;
  98         xfs_fileoff_t   offset_fsb, end_fsb;
  99         int             error = 0;
 100         int             lockmode = 0;
 101         xfs_bmbt_irec_t imap;
 102         int             nimaps = 1;
 103         int             bmapi_flags = 0;
 104         int             iomap_flags = 0;
 105
 106         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
 107         ASSERT(niomaps && *niomaps == 1);
 108
 109         if (XFS_FORCED_SHUTDOWN(mp))
 110                 return XFS_ERROR(EIO);
 111
 112         trace_xfs_iomap_enter(ip, offset, count, flags, NULL);
 113
 114         switch (flags & (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE)) {
 115         case BMAPI_READ:
 116                 lockmode = xfs_ilock_map_shared(ip);
 117                 bmapi_flags = XFS_BMAPI_ENTIRE;
 118                 break;
 119         case BMAPI_WRITE:
 120                 lockmode = XFS_ILOCK_EXCL;
 121                 if (flags & BMAPI_IGNSTATE)
 122                         bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
 123                 xfs_ilock(ip, lockmode);
 124                 break;
 125         case BMAPI_ALLOCATE:
 126                 lockmode = XFS_ILOCK_SHARED;
 127                 bmapi_flags = XFS_BMAPI_ENTIRE;
 128
 129                 /* Attempt non-blocking lock */
 130                 if (flags & BMAPI_TRYLOCK) {
 131                         if (!xfs_ilock_nowait(ip, lockmode))
 132                                 return XFS_ERROR(EAGAIN);
 133                 } else {
 134                         xfs_ilock(ip, lockmode);
 135                 }
 136                 break;
 137         default:
 138                 BUG();
 139         }
 140
 141         ASSERT(offset <= mp->m_maxioffset);
 142         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
 143                 count = mp->m_maxioffset - offset;
 144         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 145         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 146
 147         error = xfs_bmapi(NULL, ip, offset_fsb,
 148                         (xfs_filblks_t)(end_fsb - offset_fsb),
 149                         bmapi_flags,  NULL, 0, &imap,
 150                         &nimaps, NULL, NULL);
 151
 152         if (error)
 153                 goto out;
 154
 155         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)) {
 156         case BMAPI_WRITE:
 157                 /* If we found an extent, return it */
 158                 if (nimaps &&
 159                     (imap.br_startblock != HOLESTARTBLOCK) &&
 160                     (imap.br_startblock != DELAYSTARTBLOCK)) {
 161                         trace_xfs_iomap_found(ip, offset, count, flags, &imap);
 162                         break;
 163                 }
 164
 165                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
 166                         error = xfs_iomap_write_direct(ip, offset, count, flags,
 167                                                        &imap, &nimaps, nimaps);
 168                 } else {
 169                         error = xfs_iomap_write_delay(ip, offset, count, flags,
 170                                                       &imap, &nimaps);
 171                 }
 172                 if (!error) {
 173                         trace_xfs_iomap_alloc(ip, offset, count, flags, &imap);
 174                 }
 175                 iomap_flags = IOMAP_NEW;
 176                 break;
 177         case BMAPI_ALLOCATE:
 178                 /* If we found an extent, return it */
 179                 xfs_iunlock(ip, lockmode);
 180                 lockmode = 0;
 181
 182                 if (nimaps && !isnullstartblock(imap.br_startblock)) {
 183                         trace_xfs_iomap_found(ip, offset, count, flags, &imap);
 184                         break;
 185                 }
 186
 187                 error = xfs_iomap_write_allocate(ip, offset, count,
 188                                                  &imap, &nimaps);
 189                 break;
 190         }
 191
 192         ASSERT(nimaps <= 1);
 193
 194         if (nimaps)
 195                 xfs_imap_to_bmap(ip, offset, &imap, iomapp, nimaps, iomap_flags);
 196         *niomaps = nimaps;
 197
 198 out:
 199         if (lockmode)
 200                 xfs_iunlock(ip, lockmode);
 201         return XFS_ERROR(error);
 202 }
 203
 204
 205 STATIC int
 206 xfs_iomap_eof_align_last_fsb(
 207         xfs_mount_t     *mp,
 208         xfs_inode_t     *ip,
 209         xfs_extlen_t    extsize,
 210         xfs_fileoff_t   *last_fsb)
 211 {
 212         xfs_fileoff_t   new_last_fsb = 0;
 213         xfs_extlen_t    align;
 214         int             eof, error;
 215
 216         if (XFS_IS_REALTIME_INODE(ip))
 217                 ;
 218         /*
 219          * If mounted with the "-o swalloc" option, roundup the allocation
 220          * request to a stripe width boundary if the file size is >=
 221          * stripe width and we are allocating past the allocation eof.
 222          */
 223         else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
 224                 (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_swidth)))
 225                 new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
 226         /*
 227          * Roundup the allocation request to a stripe unit (m_dalign) boundary
 228          * if the file size is >= stripe unit size, and we are allocating past
 229          * the allocation eof.
 230          */
 231         else if (mp->m_dalign && (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_dalign)))
 232                 new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
 233
 234         /*
 235          * Always round up the allocation request to an extent boundary
 236          * (when file on a real-time subvolume or has di_extsize hint).
 237          */
 238         if (extsize) {
 239                 if (new_last_fsb)
 240                         align = roundup_64(new_last_fsb, extsize);
 241                 else
 242                         align = extsize;
 243                 new_last_fsb = roundup_64(*last_fsb, align);
 244         }
 245
 246         if (new_last_fsb) {
 247                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 248                 if (error)
 249                         return error;
 250                 if (eof)
 251                         *last_fsb = new_last_fsb;
 252         }
 253         return 0;
 254 }
 255
 256 STATIC int
 257 xfs_cmn_err_fsblock_zero(
 258         xfs_inode_t     *ip,
 259         xfs_bmbt_irec_t *imap)
 260 {
 261         xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
 262                         "Access to block zero in inode %llu "
 263                         "start_block: %llx start_off: %llx "
 264                         "blkcnt: %llx extent-state: %x\n",
 265                 (unsigned long long)ip->i_ino,
 266                 (unsigned long long)imap->br_startblock,
 267                 (unsigned long long)imap->br_startoff,
 268                 (unsigned long long)imap->br_blockcount,
 269                 imap->br_state);
 270         return EFSCORRUPTED;
 271 }
 272
 273 int
 274 xfs_iomap_write_direct(
 275         xfs_inode_t     *ip,
 276         xfs_off_t       offset,
 277         size_t          count,
 278         int             flags,
 279         xfs_bmbt_irec_t *ret_imap,
 280         int             *nmaps,
 281         int             found)
 282 {
 283         xfs_mount_t     *mp = ip->i_mount;
 284         xfs_fileoff_t   offset_fsb;
 285         xfs_fileoff_t   last_fsb;
 286         xfs_filblks_t   count_fsb, resaligned;
 287         xfs_fsblock_t   firstfsb;
 288         xfs_extlen_t    extsz, temp;
 289         int             nimaps;
 290         int             bmapi_flag;
 291         int             quota_flag;
 292         int             rt;
 293         xfs_trans_t     *tp;
 294         xfs_bmbt_irec_t imap;
 295         xfs_bmap_free_t free_list;
 296         uint            qblocks, resblks, resrtextents;
 297         int             committed;
 298         int             error;
 299
 300         /*
 301          * Make sure that the dquots are there. This doesn't hold
 302          * the ilock across a disk read.
 303          */
 304         error = xfs_qm_dqattach_locked(ip, 0);
 305         if (error)
 306                 return XFS_ERROR(error);
 307
 308         rt = XFS_IS_REALTIME_INODE(ip);
 309         extsz = xfs_get_extsz_hint(ip);
 310
 311         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 312         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 313         if ((offset + count) > ip->i_size) {
 314                 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
 315                 if (error)
 316                         goto error_out;
 317         } else {
 318                 if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
 319                         last_fsb = MIN(last_fsb, (xfs_fileoff_t)
 320                                         ret_imap->br_blockcount +
 321                                         ret_imap->br_startoff);
 322         }
 323         count_fsb = last_fsb - offset_fsb;
 324         ASSERT(count_fsb > 0);
 325
 326         resaligned = count_fsb;
 327         if (unlikely(extsz)) {
 328                 if ((temp = do_mod(offset_fsb, extsz)))
 329                         resaligned += temp;
 330                 if ((temp = do_mod(resaligned, extsz)))
 331                         resaligned += extsz - temp;
 332         }
 333
 334         if (unlikely(rt)) {
 335                 resrtextents = qblocks = resaligned;
 336                 resrtextents /= mp->m_sb.sb_rextsize;
 337                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 338                 quota_flag = XFS_QMOPT_RES_RTBLKS;
 339         } else {
 340                 resrtextents = 0;
 341                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
 342                 quota_flag = XFS_QMOPT_RES_REGBLKS;
 343         }
 344
 345         /*
 346          * Allocate and setup the transaction
 347          */
 348         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 349         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
 350         error = xfs_trans_reserve(tp, resblks,
 351                         XFS_WRITE_LOG_RES(mp), resrtextents,
 352                         XFS_TRANS_PERM_LOG_RES,
 353                         XFS_WRITE_LOG_COUNT);
 354         /*
 355          * Check for running out of space, note: need lock to return
 356          */
 357         if (error)
 358                 xfs_trans_cancel(tp, 0);
 359         xfs_ilock(ip, XFS_ILOCK_EXCL);
 360         if (error)
 361                 goto error_out;
 362
 363         error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
 364         if (error)
 365                 goto error1;
 366
 367         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 368         xfs_trans_ihold(tp, ip);
 369
 370         bmapi_flag = XFS_BMAPI_WRITE;
 371         if ((flags & BMAPI_DIRECT) && (offset < ip->i_size || extsz))
 372                 bmapi_flag |= XFS_BMAPI_PREALLOC;
 373
 374         /*
 375          * Issue the xfs_bmapi() call to allocate the blocks
 376          */
 377         xfs_bmap_init(&free_list, &firstfsb);
 378         nimaps = 1;
 379         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb, bmapi_flag,
 380                 &firstfsb, 0, &imap, &nimaps, &free_list, NULL);
 381         if (error)
 382                 goto error0;
 383
 384         /*
 385          * Complete the transaction
 386          */
 387         error = xfs_bmap_finish(&tp, &free_list, &committed);
 388         if (error)
 389                 goto error0;
 390         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 391         if (error)
 392                 goto error_out;
 393
 394         /*
 395          * Copy any maps to caller's array and return any error.
 396          */
 397         if (nimaps == 0) {
 398                 error = ENOSPC;
 399                 goto error_out;
 400         }
 401
 402         if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) {
 403                 error = xfs_cmn_err_fsblock_zero(ip, &imap);
 404                 goto error_out;
 405         }
 406
 407         *ret_imap = imap;
 408         *nmaps = 1;
 409         return 0;
 410
 411 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
 412         xfs_bmap_cancel(&free_list);
 413         xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
 414
 415 error1: /* Just cancel transaction */
 416         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 417         *nmaps = 0;     /* nothing set-up here */
 418
 419 error_out:
 420         return XFS_ERROR(error);
 421 }
 422
 423 /*
 424  * If the caller is doing a write at the end of the file, then extend the
 425  * allocation out to the file system's write iosize.  We clean up any extra
 426  * space left over when the file is closed in xfs_inactive().
 427  */
 428 STATIC int
 429 xfs_iomap_eof_want_preallocate(
 430         xfs_mount_t     *mp,
 431         xfs_inode_t     *ip,
 432         xfs_off_t       offset,
 433         size_t          count,
 434         int             ioflag,
 435         xfs_bmbt_irec_t *imap,
 436         int             nimaps,
 437         int             *prealloc)
 438 {
 439         xfs_fileoff_t   start_fsb;
 440         xfs_filblks_t   count_fsb;
 441         xfs_fsblock_t   firstblock;
 442         int             n, error, imaps;
 443
 444         *prealloc = 0;
 445         if ((offset + count) <= ip->i_size)
 446                 return 0;
 447
 448         /*
 449          * If there are any real blocks past eof, then don't
 450          * do any speculative allocation.
 451          */
 452         start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
 453         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
 454         while (count_fsb > 0) {
 455                 imaps = nimaps;
 456                 firstblock = NULLFSBLOCK;
 457                 error = xfs_bmapi(NULL, ip, start_fsb, count_fsb, 0,
 458                                   &firstblock, 0, imap, &imaps, NULL, NULL);
 459                 if (error)
 460                         return error;
 461                 for (n = 0; n < imaps; n++) {
 462                         if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
 463                             (imap[n].br_startblock != DELAYSTARTBLOCK))
 464                                 return 0;
 465                         start_fsb += imap[n].br_blockcount;
 466                         count_fsb -= imap[n].br_blockcount;
 467                 }
 468         }
 469         *prealloc = 1;
 470         return 0;
 471 }
 472
 473 int
 474 xfs_iomap_write_delay(
 475         xfs_inode_t     *ip,
 476         xfs_off_t       offset,
 477         size_t          count,
 478         int             ioflag,
 479         xfs_bmbt_irec_t *ret_imap,
 480         int             *nmaps)
 481 {
 482         xfs_mount_t     *mp = ip->i_mount;
 483         xfs_fileoff_t   offset_fsb;
 484         xfs_fileoff_t   last_fsb;
 485         xfs_off_t       aligned_offset;
 486         xfs_fileoff_t   ioalign;
 487         xfs_fsblock_t   firstblock;
 488         xfs_extlen_t    extsz;
 489         int             nimaps;
 490         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
 491         int             prealloc, flushed = 0;
 492         int             error;
 493
 494         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 495
 496         /*
 497          * Make sure that the dquots are there. This doesn't hold
 498          * the ilock across a disk read.
 499          */
 500         error = xfs_qm_dqattach_locked(ip, 0);
 501         if (error)
 502                 return XFS_ERROR(error);
 503
 504         extsz = xfs_get_extsz_hint(ip);
 505         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 506
 507         error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
 508                                 ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
 509         if (error)
 510                 return error;
 511
 512 retry:
 513         if (prealloc) {
 514                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 515                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
 516                 last_fsb = ioalign + mp->m_writeio_blocks;
 517         } else {
 518                 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 519         }
 520
 521         if (prealloc || extsz) {
 522                 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
 523                 if (error)
 524                         return error;
 525         }
 526
 527         nimaps = XFS_WRITE_IMAPS;
 528         firstblock = NULLFSBLOCK;
 529         error = xfs_bmapi(NULL, ip, offset_fsb,
 530                           (xfs_filblks_t)(last_fsb - offset_fsb),
 531                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
 532                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
 533                           &nimaps, NULL, NULL);
 534         if (error && (error != ENOSPC))
 535                 return XFS_ERROR(error);
 536
 537         /*
 538          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
 539          * then we must have run out of space - flush all other inodes with
 540          * delalloc blocks and retry without EOF preallocation.
 541          */
 542         if (nimaps == 0) {
 543                 trace_xfs_delalloc_enospc(ip, offset, count);
 544                 if (flushed)
 545                         return XFS_ERROR(ENOSPC);
 546
 547                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 548                 xfs_flush_inodes(ip);
 549                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 550
 551                 flushed = 1;
 552                 error = 0;
 553                 prealloc = 0;
 554                 goto retry;
 555         }
 556
 557         if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
 558                 return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
 559
 560         *ret_imap = imap[0];
 561         *nmaps = 1;
 562
 563         return 0;
 564 }
 565
 566 /*
 567  * Pass in a delayed allocate extent, convert it to real extents;
 568  * return to the caller the extent we create which maps on top of
 569  * the originating callers request.
 570  *
 571  * Called without a lock on the inode.
 572  *
 573  * We no longer bother to look at the incoming map - all we have to
 574  * guarantee is that whatever we allocate fills the required range.
 575  */
 576 int
 577 xfs_iomap_write_allocate(
 578         xfs_inode_t     *ip,
 579         xfs_off_t       offset,
 580         size_t          count,
 581         xfs_bmbt_irec_t *map,
 582         int             *retmap)
 583 {
 584         xfs_mount_t     *mp = ip->i_mount;
 585         xfs_fileoff_t   offset_fsb, last_block;
 586         xfs_fileoff_t   end_fsb, map_start_fsb;
 587         xfs_fsblock_t   first_block;
 588         xfs_bmap_free_t free_list;
 589         xfs_filblks_t   count_fsb;
 590         xfs_bmbt_irec_t imap;
 591         xfs_trans_t     *tp;
 592         int             nimaps, committed;
 593         int             error = 0;
 594         int             nres;
 595
 596         *retmap = 0;
 597
 598         /*
 599          * Make sure that the dquots are there.
 600          */
 601         error = xfs_qm_dqattach(ip, 0);
 602         if (error)
 603                 return XFS_ERROR(error);
 604
 605         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 606         count_fsb = map->br_blockcount;
 607         map_start_fsb = map->br_startoff;
 608
 609         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 610
 611         while (count_fsb != 0) {
 612                 /*
 613                  * Set up a transaction with which to allocate the
 614                  * backing store for the file.  Do allocations in a
 615                  * loop until we get some space in the range we are
 616                  * interested in.  The other space that might be allocated
 617                  * is in the delayed allocation extent on which we sit
 618                  * but before our buffer starts.
 619                  */
 620
 621                 nimaps = 0;
 622                 while (nimaps == 0) {
 623                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 624                         tp->t_flags |= XFS_TRANS_RESERVE;
 625                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 626                         error = xfs_trans_reserve(tp, nres,
 627                                         XFS_WRITE_LOG_RES(mp),
 628                                         0, XFS_TRANS_PERM_LOG_RES,
 629                                         XFS_WRITE_LOG_COUNT);
 630                         if (error) {
 631                                 xfs_trans_cancel(tp, 0);
 632                                 return XFS_ERROR(error);
 633                         }
 634                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 635                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 636                         xfs_trans_ihold(tp, ip);
 637
 638                         xfs_bmap_init(&free_list, &first_block);
 639
 640                         /*
 641                          * it is possible that the extents have changed since
 642                          * we did the read call as we dropped the ilock for a
 643                          * while. We have to be careful about truncates or hole
 644                          * punchs here - we are not allowed to allocate
 645                          * non-delalloc blocks here.
 646                          *
 647                          * The only protection against truncation is the pages
 648                          * for the range we are being asked to convert are
 649                          * locked and hence a truncate will block on them
 650                          * first.
 651                          *
 652                          * As a result, if we go beyond the range we really
 653                          * need and hit an delalloc extent boundary followed by
 654                          * a hole while we have excess blocks in the map, we
 655                          * will fill the hole incorrectly and overrun the
 656                          * transaction reservation.
 657                          *
 658                          * Using a single map prevents this as we are forced to
 659                          * check each map we look for overlap with the desired
 660                          * range and abort as soon as we find it. Also, given
 661                          * that we only return a single map, having one beyond
 662                          * what we can return is probably a bit silly.
 663                          *
 664                          * We also need to check that we don't go beyond EOF;
 665                          * this is a truncate optimisation as a truncate sets
 666                          * the new file size before block on the pages we
 667                          * currently have locked under writeback. Because they
 668                          * are about to be tossed, we don't need to write them
 669                          * back....
 670                          */
 671                         nimaps = 1;
 672                         end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
 673                         error = xfs_bmap_last_offset(NULL, ip, &last_block,
 674                                                         XFS_DATA_FORK);
 675                         if (error)
 676                                 goto trans_cancel;
 677
 678                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 679                         if ((map_start_fsb + count_fsb) > last_block) {
 680                                 count_fsb = last_block - map_start_fsb;
 681                                 if (count_fsb == 0) {
 682                                         error = EAGAIN;
 683                                         goto trans_cancel;
 684                                 }
 685                         }
 686
 687                         /* Go get the actual blocks */
 688                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
 689                                         XFS_BMAPI_WRITE, &first_block, 1,
 690                                         &imap, &nimaps, &free_list, NULL);
 691                         if (error)
 692                                 goto trans_cancel;
 693
 694                         error = xfs_bmap_finish(&tp, &free_list, &committed);
 695                         if (error)
 696                                 goto trans_cancel;
 697
 698                         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 699                         if (error)
 700                                 goto error0;
 701
 702                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 703                 }
 704
 705                 /*
 706                  * See if we were able to allocate an extent that
 707                  * covers at least part of the callers request
 708                  */
 709                 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
 710                         return xfs_cmn_err_fsblock_zero(ip, &imap);
 711
 712                 if ((offset_fsb >= imap.br_startoff) &&
 713                     (offset_fsb < (imap.br_startoff +
 714                                    imap.br_blockcount))) {
 715                         *map = imap;
 716                         *retmap = 1;
 717                         XFS_STATS_INC(xs_xstrat_quick);
 718                         return 0;
 719                 }
 720
 721                 /*
 722                  * So far we have not mapped the requested part of the
 723                  * file, just surrounding data, try again.
 724                  */
 725                 count_fsb -= imap.br_blockcount;
 726                 map_start_fsb = imap.br_startoff + imap.br_blockcount;
 727         }
 728
 729 trans_cancel:
 730         xfs_bmap_cancel(&free_list);
 731         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 732 error0:
 733         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 734         return XFS_ERROR(error);
 735 }
 736
 737 int
 738 xfs_iomap_write_unwritten(
 739         xfs_inode_t     *ip,
 740         xfs_off_t       offset,
 741         size_t          count)
 742 {
 743         xfs_mount_t     *mp = ip->i_mount;
 744         xfs_fileoff_t   offset_fsb;
 745         xfs_filblks_t   count_fsb;
 746         xfs_filblks_t   numblks_fsb;
 747         xfs_fsblock_t   firstfsb;
 748         int             nimaps;
 749         xfs_trans_t     *tp;
 750         xfs_bmbt_irec_t imap;
 751         xfs_bmap_free_t free_list;
 752         uint            resblks;
 753         int             committed;
 754         int             error;
 755
 756         trace_xfs_unwritten_convert(ip, offset, count);
 757
 758         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 759         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 760         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 761
 762         /*
 763          * Reserve enough blocks in this transaction for two complete extent
 764          * btree splits.  We may be converting the middle part of an unwritten
 765          * extent and in this case we will insert two new extents in the btree
 766          * each of which could cause a full split.
 767          *
 768          * This reservation amount will be used in the first call to
 769          * xfs_bmbt_split() to select an AG with enough space to satisfy the
 770          * rest of the operation.
 771          */
 772         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 773
 774         do {
 775                 /*
 776                  * set up a transaction to convert the range of extents
 777                  * from unwritten to real. Do allocations in a loop until
 778                  * we have covered the range passed in.
 779                  *
 780                  * Note that we open code the transaction allocation here
 781                  * to pass KM_NOFS--we can't risk to recursing back into
 782                  * the filesystem here as we might be asked to write out
 783                  * the same inode that we complete here and might deadlock
 784                  * on the iolock.
 785                  */
 786                 xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
 787                 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
 788                 tp->t_flags |= XFS_TRANS_RESERVE;
 789                 error = xfs_trans_reserve(tp, resblks,
 790                                 XFS_WRITE_LOG_RES(mp), 0,
 791                                 XFS_TRANS_PERM_LOG_RES,
 792                                 XFS_WRITE_LOG_COUNT);
 793                 if (error) {
 794                         xfs_trans_cancel(tp, 0);
 795                         return XFS_ERROR(error);
 796                 }
 797
 798                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 799                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 800                 xfs_trans_ihold(tp, ip);
 801
 802                 /*
 803                  * Modify the unwritten extent state of the buffer.
 804                  */
 805                 xfs_bmap_init(&free_list, &firstfsb);
 806                 nimaps = 1;
 807                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 808                                   XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
 809                                   1, &imap, &nimaps, &free_list, NULL);
 810                 if (error)
 811                         goto error_on_bmapi_transaction;
 812
 813                 error = xfs_bmap_finish(&(tp), &(free_list), &committed);
 814                 if (error)
 815                         goto error_on_bmapi_transaction;
 816
 817                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 818                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 819                 if (error)
 820                         return XFS_ERROR(error);
 821
 822                 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
 823                         return xfs_cmn_err_fsblock_zero(ip, &imap);
 824
 825                 if ((numblks_fsb = imap.br_blockcount) == 0) {
 826                         /*
 827                          * The numblks_fsb value should always get
 828                          * smaller, otherwise the loop is stuck.
 829                          */
 830                         ASSERT(imap.br_blockcount);
 831                         break;
 832                 }
 833                 offset_fsb += numblks_fsb;
 834                 count_fsb -= numblks_fsb;
 835         } while (count_fsb > 0);
 836
 837         return 0;
 838
 839 error_on_bmapi_transaction:
 840         xfs_bmap_cancel(&free_list);
 841         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
 842         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 843         return XFS_ERROR(error);
 844 }