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