fs/xfs/xfs_iomap.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_bit.h"
  21 #include "xfs_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_alloc.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_quota.h"
  31 #include "xfs_mount.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_ialloc_btree.h"
  35 #include "xfs_dir_sf.h"
  36 #include "xfs_dir2_sf.h"
  37 #include "xfs_attr_sf.h"
  38 #include "xfs_dinode.h"
  39 #include "xfs_inode.h"
  40 #include "xfs_ialloc.h"
  41 #include "xfs_btree.h"
  42 #include "xfs_bmap.h"
  43 #include "xfs_rtalloc.h"
  44 #include "xfs_error.h"
  45 #include "xfs_itable.h"
  46 #include "xfs_rw.h"
  47 #include "xfs_acl.h"
  48 #include "xfs_cap.h"
  49 #include "xfs_mac.h"
  50 #include "xfs_attr.h"
  51 #include "xfs_buf_item.h"
  52 #include "xfs_trans_space.h"
  53 #include "xfs_utils.h"
  54 #include "xfs_iomap.h"
  55
  56 #if defined(XFS_RW_TRACE)
  57 void
  58 xfs_iomap_enter_trace(
  59         int             tag,
  60         xfs_iocore_t    *io,
  61         xfs_off_t       offset,
  62         ssize_t         count)
  63 {
  64         xfs_inode_t     *ip = XFS_IO_INODE(io);
  65
  66         if (!ip->i_rwtrace)
  67                 return;
  68
  69         ktrace_enter(ip->i_rwtrace,
  70                 (void *)((unsigned long)tag),
  71                 (void *)ip,
  72                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
  73                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
  74                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
  75                 (void *)((unsigned long)(offset & 0xffffffff)),
  76                 (void *)((unsigned long)count),
  77                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
  78                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
  79                 (void *)NULL,
  80                 (void *)NULL,
  81                 (void *)NULL,
  82                 (void *)NULL,
  83                 (void *)NULL,
  84                 (void *)NULL,
  85                 (void *)NULL);
  86 }
  87
  88 void
  89 xfs_iomap_map_trace(
  90         int             tag,
  91         xfs_iocore_t    *io,
  92         xfs_off_t       offset,
  93         ssize_t         count,
  94         xfs_iomap_t     *iomapp,
  95         xfs_bmbt_irec_t *imapp,
  96         int             flags)
  97 {
  98         xfs_inode_t     *ip = XFS_IO_INODE(io);
  99
 100         if (!ip->i_rwtrace)
 101                 return;
 102
 103         ktrace_enter(ip->i_rwtrace,
 104                 (void *)((unsigned long)tag),
 105                 (void *)ip,
 106                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
 107                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
 108                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
 109                 (void *)((unsigned long)(offset & 0xffffffff)),
 110                 (void *)((unsigned long)count),
 111                 (void *)((unsigned long)flags),
 112                 (void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
 113                 (void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
 114                 (void *)((unsigned long)(iomapp->iomap_delta)),
 115                 (void *)((unsigned long)(iomapp->iomap_bsize)),
 116                 (void *)((unsigned long)(iomapp->iomap_bn)),
 117                 (void *)(__psint_t)(imapp->br_startoff),
 118                 (void *)((unsigned long)(imapp->br_blockcount)),
 119                 (void *)(__psint_t)(imapp->br_startblock));
 120 }
 121 #else
 122 #define xfs_iomap_enter_trace(tag, io, offset, count)
 123 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
 124 #endif
 125
 126 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
 127                                                 << mp->m_writeio_log)
 128 #define XFS_STRAT_WRITE_IMAPS   2
 129 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
 130
 131 STATIC int
 132 xfs_imap_to_bmap(
 133         xfs_iocore_t    *io,
 134         xfs_off_t       offset,
 135         xfs_bmbt_irec_t *imap,
 136         xfs_iomap_t     *iomapp,
 137         int             imaps,                  /* Number of imap entries */
 138         int             iomaps,                 /* Number of iomap entries */
 139         int             flags)
 140 {
 141         xfs_mount_t     *mp;
 142         xfs_fsize_t     nisize;
 143         int             pbm;
 144         xfs_fsblock_t   start_block;
 145
 146         mp = io->io_mount;
 147         nisize = XFS_SIZE(mp, io);
 148         if (io->io_new_size > nisize)
 149                 nisize = io->io_new_size;
 150
 151         for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
 152                 iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
 153                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
 154                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
 155                 iomapp->iomap_flags = flags;
 156
 157                 if (io->io_flags & XFS_IOCORE_RT) {
 158                         iomapp->iomap_flags |= IOMAP_REALTIME;
 159                         iomapp->iomap_target = mp->m_rtdev_targp;
 160                 } else {
 161                         iomapp->iomap_target = mp->m_ddev_targp;
 162                 }
 163                 start_block = imap->br_startblock;
 164                 if (start_block == HOLESTARTBLOCK) {
 165                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 166                         iomapp->iomap_flags |= IOMAP_HOLE;
 167                 } else if (start_block == DELAYSTARTBLOCK) {
 168                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 169                         iomapp->iomap_flags |= IOMAP_DELAY;
 170                 } else {
 171                         iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
 172                         if (ISUNWRITTEN(imap))
 173                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
 174                 }
 175
 176                 if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
 177                         iomapp->iomap_flags |= IOMAP_EOF;
 178                 }
 179
 180                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
 181         }
 182         return pbm;     /* Return the number filled */
 183 }
 184
 185 int
 186 xfs_iomap(
 187         xfs_iocore_t    *io,
 188         xfs_off_t       offset,
 189         ssize_t         count,
 190         int             flags,
 191         xfs_iomap_t     *iomapp,
 192         int             *niomaps)
 193 {
 194         xfs_mount_t     *mp = io->io_mount;
 195         xfs_fileoff_t   offset_fsb, end_fsb;
 196         int             error = 0;
 197         int             lockmode = 0;
 198         xfs_bmbt_irec_t imap;
 199         int             nimaps = 1;
 200         int             bmapi_flags = 0;
 201         int             iomap_flags = 0;
 202
 203         if (XFS_FORCED_SHUTDOWN(mp))
 204                 return XFS_ERROR(EIO);
 205
 206         switch (flags &
 207                 (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
 208                  BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
 209         case BMAPI_READ:
 210                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
 211                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 212                 bmapi_flags = XFS_BMAPI_ENTIRE;
 213                 break;
 214         case BMAPI_WRITE:
 215                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
 216                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
 217                 if (flags & BMAPI_IGNSTATE)
 218                         bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
 219                 XFS_ILOCK(mp, io, lockmode);
 220                 break;
 221         case BMAPI_ALLOCATE:
 222                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
 223                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
 224                 bmapi_flags = XFS_BMAPI_ENTIRE;
 225                 /* Attempt non-blocking lock */
 226                 if (flags & BMAPI_TRYLOCK) {
 227                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
 228                                 return XFS_ERROR(EAGAIN);
 229                 } else {
 230                         XFS_ILOCK(mp, io, lockmode);
 231                 }
 232                 break;
 233         case BMAPI_UNWRITTEN:
 234                 goto phase2;
 235         case BMAPI_DEVICE:
 236                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 237                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
 238                         mp->m_rtdev_targp : mp->m_ddev_targp;
 239                 error = 0;
 240                 *niomaps = 1;
 241                 goto out;
 242         default:
 243                 BUG();
 244         }
 245
 246         ASSERT(offset <= mp->m_maxioffset);
 247         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
 248                 count = mp->m_maxioffset - offset;
 249         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 250         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 251
 252         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
 253                         (xfs_filblks_t)(end_fsb - offset_fsb),
 254                         bmapi_flags,  NULL, 0, &imap,
 255                         &nimaps, NULL);
 256
 257         if (error)
 258                 goto out;
 259
 260 phase2:
 261         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
 262         case BMAPI_WRITE:
 263                 /* If we found an extent, return it */
 264                 if (nimaps &&
 265                     (imap.br_startblock != HOLESTARTBLOCK) &&
 266                     (imap.br_startblock != DELAYSTARTBLOCK)) {
 267                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 268                                         offset, count, iomapp, &imap, flags);
 269                         break;
 270                 }
 271
 272                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
 273                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
 274                                         count, flags, &imap, &nimaps, nimaps);
 275                 } else {
 276                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
 277                                         flags, &imap, &nimaps);
 278                 }
 279                 if (!error) {
 280                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
 281                                         offset, count, iomapp, &imap, flags);
 282                 }
 283                 iomap_flags = IOMAP_NEW;
 284                 break;
 285         case BMAPI_ALLOCATE:
 286                 /* If we found an extent, return it */
 287                 XFS_IUNLOCK(mp, io, lockmode);
 288                 lockmode = 0;
 289
 290                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
 291                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 292                                         offset, count, iomapp, &imap, flags);
 293                         break;
 294                 }
 295
 296                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
 297                                                  &imap, &nimaps);
 298                 break;
 299         case BMAPI_UNWRITTEN:
 300                 lockmode = 0;
 301                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
 302                 nimaps = 0;
 303                 break;
 304         }
 305
 306         if (nimaps) {
 307                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
 308                                             iomapp, nimaps, *niomaps, iomap_flags);
 309         } else if (niomaps) {
 310                 *niomaps = 0;
 311         }
 312
 313 out:
 314         if (lockmode)
 315                 XFS_IUNLOCK(mp, io, lockmode);
 316         return XFS_ERROR(error);
 317 }
 318
 319 STATIC int
 320 xfs_flush_space(
 321         xfs_inode_t     *ip,
 322         int             *fsynced,
 323         int             *ioflags)
 324 {
 325         switch (*fsynced) {
 326         case 0:
 327                 if (ip->i_delayed_blks) {
 328                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 329                         xfs_flush_inode(ip);
 330                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 331                         *fsynced = 1;
 332                 } else {
 333                         *ioflags |= BMAPI_SYNC;
 334                         *fsynced = 2;
 335                 }
 336                 return 0;
 337         case 1:
 338                 *fsynced = 2;
 339                 *ioflags |= BMAPI_SYNC;
 340                 return 0;
 341         case 2:
 342                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 343                 xfs_flush_device(ip);
 344                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 345                 *fsynced = 3;
 346                 return 0;
 347         }
 348         return 1;
 349 }
 350
 351 int
 352 xfs_iomap_write_direct(
 353         xfs_inode_t     *ip,
 354         xfs_off_t       offset,
 355         size_t          count,
 356         int             flags,
 357         xfs_bmbt_irec_t *ret_imap,
 358         int             *nmaps,
 359         int             found)
 360 {
 361         xfs_mount_t     *mp = ip->i_mount;
 362         xfs_iocore_t    *io = &ip->i_iocore;
 363         xfs_fileoff_t   offset_fsb;
 364         xfs_fileoff_t   last_fsb;
 365         xfs_filblks_t   count_fsb;
 366         xfs_fsblock_t   firstfsb;
 367         int             nimaps;
 368         int             error;
 369         int             bmapi_flag;
 370         int             quota_flag;
 371         int             rt;
 372         xfs_trans_t     *tp;
 373         xfs_bmbt_irec_t imap;
 374         xfs_bmap_free_t free_list;
 375         xfs_filblks_t   qblocks, resblks;
 376         int             committed;
 377         int             resrtextents;
 378
 379         /*
 380          * Make sure that the dquots are there. This doesn't hold
 381          * the ilock across a disk read.
 382          */
 383         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
 384         if (error)
 385                 return XFS_ERROR(error);
 386
 387         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 388         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 389         count_fsb = last_fsb - offset_fsb;
 390         if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
 391                 xfs_fileoff_t   map_last_fsb;
 392
 393                 map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
 394                 if (map_last_fsb < last_fsb) {
 395                         last_fsb = map_last_fsb;
 396                         count_fsb = last_fsb - offset_fsb;
 397                 }
 398                 ASSERT(count_fsb > 0);
 399         }
 400
 401         /*
 402          * Determine if reserving space on the data or realtime partition.
 403          */
 404         if ((rt = XFS_IS_REALTIME_INODE(ip))) {
 405                 xfs_extlen_t    extsz;
 406
 407                 if (!(extsz = ip->i_d.di_extsize))
 408                         extsz = mp->m_sb.sb_rextsize;
 409                 resrtextents = qblocks = (count_fsb + extsz - 1);
 410                 do_div(resrtextents, mp->m_sb.sb_rextsize);
 411                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 412                 quota_flag = XFS_QMOPT_RES_RTBLKS;
 413         } else {
 414                 resrtextents = 0;
 415                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, count_fsb);
 416                 quota_flag = XFS_QMOPT_RES_REGBLKS;
 417         }
 418
 419         /*
 420          * Allocate and setup the transaction
 421          */
 422         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 423         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
 424         error = xfs_trans_reserve(tp, resblks,
 425                         XFS_WRITE_LOG_RES(mp), resrtextents,
 426                         XFS_TRANS_PERM_LOG_RES,
 427                         XFS_WRITE_LOG_COUNT);
 428
 429         /*
 430          * Check for running out of space, note: need lock to return
 431          */
 432         if (error)
 433                 xfs_trans_cancel(tp, 0);
 434         xfs_ilock(ip, XFS_ILOCK_EXCL);
 435         if (error)
 436                 goto error_out;
 437
 438         if (XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag)) {
 439                 error = (EDQUOT);
 440                 goto error1;
 441         }
 442
 443         bmapi_flag = XFS_BMAPI_WRITE;
 444         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 445         xfs_trans_ihold(tp, ip);
 446
 447         if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
 448                 bmapi_flag |= XFS_BMAPI_PREALLOC;
 449
 450         /*
 451          * Issue the bmapi() call to allocate the blocks
 452          */
 453         XFS_BMAP_INIT(&free_list, &firstfsb);
 454         nimaps = 1;
 455         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 456                 bmapi_flag, &firstfsb, 0, &imap, &nimaps, &free_list);
 457         if (error)
 458                 goto error0;
 459
 460         /*
 461          * Complete the transaction
 462          */
 463         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
 464         if (error)
 465                 goto error0;
 466         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 467         if (error)
 468                 goto error_out;
 469
 470         /*
 471          * Copy any maps to caller's array and return any error.
 472          */
 473         if (nimaps == 0) {
 474                 error = (ENOSPC);
 475                 goto error_out;
 476         }
 477
 478         *ret_imap = imap;
 479         *nmaps = 1;
 480         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 481                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 482                         "start_block : %llx start_off : %llx blkcnt : %llx "
 483                         "extent-state : %x \n",
 484                         (ip->i_mount)->m_fsname,
 485                         (long long)ip->i_ino,
 486                         ret_imap->br_startblock, ret_imap->br_startoff,
 487                         ret_imap->br_blockcount,ret_imap->br_state);
 488         }
 489         return 0;
 490
 491 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
 492         xfs_bmap_cancel(&free_list);
 493         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
 494
 495 error1: /* Just cancel transaction */
 496         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 497         *nmaps = 0;     /* nothing set-up here */
 498
 499 error_out:
 500         return XFS_ERROR(error);
 501 }
 502
 503 int
 504 xfs_iomap_write_delay(
 505         xfs_inode_t     *ip,
 506         xfs_off_t       offset,
 507         size_t          count,
 508         int             ioflag,
 509         xfs_bmbt_irec_t *ret_imap,
 510         int             *nmaps)
 511 {
 512         xfs_mount_t     *mp = ip->i_mount;
 513         xfs_iocore_t    *io = &ip->i_iocore;
 514         xfs_fileoff_t   offset_fsb;
 515         xfs_fileoff_t   last_fsb;
 516         xfs_fsize_t     isize;
 517         xfs_fsblock_t   firstblock;
 518         int             nimaps;
 519         int             error;
 520         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
 521         int             aeof;
 522         int             fsynced = 0;
 523
 524         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
 525
 526         /*
 527          * Make sure that the dquots are there. This doesn't hold
 528          * the ilock across a disk read.
 529          */
 530
 531         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
 532         if (error)
 533                 return XFS_ERROR(error);
 534
 535 retry:
 536         isize = ip->i_d.di_size;
 537         if (io->io_new_size > isize) {
 538                 isize = io->io_new_size;
 539         }
 540
 541         aeof = 0;
 542         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 543         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 544         /*
 545          * If the caller is doing a write at the end of the file,
 546          * then extend the allocation (and the buffer used for the write)
 547          * out to the file system's write iosize.  We clean up any extra
 548          * space left over when the file is closed in xfs_inactive().
 549          *
 550          * For sync writes, we are flushing delayed allocate space to
 551          * try to make additional space available for allocation near
 552          * the filesystem full boundary - preallocation hurts in that
 553          * situation, of course.
 554          */
 555         if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
 556                 xfs_off_t       aligned_offset;
 557                 xfs_filblks_t   count_fsb;
 558                 unsigned int    iosize;
 559                 xfs_fileoff_t   ioalign;
 560                 int             n;
 561                 xfs_fileoff_t   start_fsb;
 562
 563                 /*
 564                  * If there are any real blocks past eof, then don't
 565                  * do any speculative allocation.
 566                  */
 567                 start_fsb = XFS_B_TO_FSBT(mp,
 568                                         ((xfs_ufsize_t)(offset + count - 1)));
 569                 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
 570                 while (count_fsb > 0) {
 571                         nimaps = XFS_WRITE_IMAPS;
 572                         error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
 573                                         0, &firstblock, 0, imap, &nimaps, NULL);
 574                         if (error) {
 575                                 return error;
 576                         }
 577                         for (n = 0; n < nimaps; n++) {
 578                                 if ( !(io->io_flags & XFS_IOCORE_RT)  &&
 579                                         !imap[n].br_startblock) {
 580                                         cmn_err(CE_PANIC,"Access to block "
 581                                                 "zero:  fs <%s> inode: %lld "
 582                                                 "start_block : %llx start_off "
 583                                                 ": %llx blkcnt : %llx "
 584                                                 "extent-state : %x \n",
 585                                                 (ip->i_mount)->m_fsname,
 586                                                 (long long)ip->i_ino,
 587                                                 imap[n].br_startblock,
 588                                                 imap[n].br_startoff,
 589                                                 imap[n].br_blockcount,
 590                                                 imap[n].br_state);
 591                                 }
 592                                 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
 593                                     (imap[n].br_startblock != DELAYSTARTBLOCK)) {
 594                                         goto write_map;
 595                                 }
 596                                 start_fsb += imap[n].br_blockcount;
 597                                 count_fsb -= imap[n].br_blockcount;
 598                         }
 599                 }
 600                 iosize = mp->m_writeio_blocks;
 601                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 602                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
 603                 last_fsb = ioalign + iosize;
 604                 aeof = 1;
 605         }
 606 write_map:
 607         nimaps = XFS_WRITE_IMAPS;
 608         firstblock = NULLFSBLOCK;
 609
 610         /*
 611          * If mounted with the "-o swalloc" option, roundup the allocation
 612          * request to a stripe width boundary if the file size is >=
 613          * stripe width and we are allocating past the allocation eof.
 614          */
 615         if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_swidth
 616             && (mp->m_flags & XFS_MOUNT_SWALLOC)
 617             && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
 618                 int eof;
 619                 xfs_fileoff_t new_last_fsb;
 620
 621                 new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
 622                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 623                 if (error) {
 624                         return error;
 625                 }
 626                 if (eof) {
 627                         last_fsb = new_last_fsb;
 628                 }
 629         /*
 630          * Roundup the allocation request to a stripe unit (m_dalign) boundary
 631          * if the file size is >= stripe unit size, and we are allocating past
 632          * the allocation eof.
 633          */
 634         } else if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_dalign &&
 635                    (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)) && aeof) {
 636                 int eof;
 637                 xfs_fileoff_t new_last_fsb;
 638                 new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
 639                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 640                 if (error) {
 641                         return error;
 642                 }
 643                 if (eof) {
 644                         last_fsb = new_last_fsb;
 645                 }
 646         /*
 647          * Round up the allocation request to a real-time extent boundary
 648          * if the file is on the real-time subvolume.
 649          */
 650         } else if (io->io_flags & XFS_IOCORE_RT && aeof) {
 651                 int eof;
 652                 xfs_fileoff_t new_last_fsb;
 653
 654                 new_last_fsb = roundup_64(last_fsb, mp->m_sb.sb_rextsize);
 655                 error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
 656                 if (error) {
 657                         return error;
 658                 }
 659                 if (eof)
 660                         last_fsb = new_last_fsb;
 661         }
 662         error = xfs_bmapi(NULL, ip, offset_fsb,
 663                           (xfs_filblks_t)(last_fsb - offset_fsb),
 664                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
 665                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
 666                           &nimaps, NULL);
 667         /*
 668          * This can be EDQUOT, if nimaps == 0
 669          */
 670         if (error && (error != ENOSPC)) {
 671                 return XFS_ERROR(error);
 672         }
 673         /*
 674          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
 675          * then we must have run out of space.
 676          */
 677         if (nimaps == 0) {
 678                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
 679                                         io, offset, count);
 680                 if (xfs_flush_space(ip, &fsynced, &ioflag))
 681                         return XFS_ERROR(ENOSPC);
 682
 683                 error = 0;
 684                 goto retry;
 685         }
 686
 687         *ret_imap = imap[0];
 688         *nmaps = 1;
 689         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 690                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 691                         "start_block : %llx start_off : %llx blkcnt : %llx "
 692                         "extent-state : %x \n",
 693                         (ip->i_mount)->m_fsname,
 694                         (long long)ip->i_ino,
 695                         ret_imap->br_startblock, ret_imap->br_startoff,
 696                         ret_imap->br_blockcount,ret_imap->br_state);
 697         }
 698         return 0;
 699 }
 700
 701 /*
 702  * Pass in a delayed allocate extent, convert it to real extents;
 703  * return to the caller the extent we create which maps on top of
 704  * the originating callers request.
 705  *
 706  * Called without a lock on the inode.
 707  */
 708 int
 709 xfs_iomap_write_allocate(
 710         xfs_inode_t     *ip,
 711         xfs_off_t       offset,
 712         size_t          count,
 713         xfs_bmbt_irec_t *map,
 714         int             *retmap)
 715 {
 716         xfs_mount_t     *mp = ip->i_mount;
 717         xfs_iocore_t    *io = &ip->i_iocore;
 718         xfs_fileoff_t   offset_fsb, last_block;
 719         xfs_fileoff_t   end_fsb, map_start_fsb;
 720         xfs_fsblock_t   first_block;
 721         xfs_bmap_free_t free_list;
 722         xfs_filblks_t   count_fsb;
 723         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
 724         xfs_trans_t     *tp;
 725         int             i, nimaps, committed;
 726         int             error = 0;
 727         int             nres;
 728
 729         *retmap = 0;
 730
 731         /*
 732          * Make sure that the dquots are there.
 733          */
 734         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
 735                 return XFS_ERROR(error);
 736
 737         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 738         count_fsb = map->br_blockcount;
 739         map_start_fsb = map->br_startoff;
 740
 741         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 742
 743         while (count_fsb != 0) {
 744                 /*
 745                  * Set up a transaction with which to allocate the
 746                  * backing store for the file.  Do allocations in a
 747                  * loop until we get some space in the range we are
 748                  * interested in.  The other space that might be allocated
 749                  * is in the delayed allocation extent on which we sit
 750                  * but before our buffer starts.
 751                  */
 752
 753                 nimaps = 0;
 754                 while (nimaps == 0) {
 755                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 756                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 757                         error = xfs_trans_reserve(tp, nres,
 758                                         XFS_WRITE_LOG_RES(mp),
 759                                         0, XFS_TRANS_PERM_LOG_RES,
 760                                         XFS_WRITE_LOG_COUNT);
 761                         if (error == ENOSPC) {
 762                                 error = xfs_trans_reserve(tp, 0,
 763                                                 XFS_WRITE_LOG_RES(mp),
 764                                                 0,
 765                                                 XFS_TRANS_PERM_LOG_RES,
 766                                                 XFS_WRITE_LOG_COUNT);
 767                         }
 768                         if (error) {
 769                                 xfs_trans_cancel(tp, 0);
 770                                 return XFS_ERROR(error);
 771                         }
 772                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 773                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 774                         xfs_trans_ihold(tp, ip);
 775
 776                         XFS_BMAP_INIT(&free_list, &first_block);
 777
 778                         nimaps = XFS_STRAT_WRITE_IMAPS;
 779                         /*
 780                          * Ensure we don't go beyond eof - it is possible
 781                          * the extents changed since we did the read call,
 782                          * we dropped the ilock in the interim.
 783                          */
 784
 785                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
 786                         xfs_bmap_last_offset(NULL, ip, &last_block,
 787                                 XFS_DATA_FORK);
 788                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 789                         if ((map_start_fsb + count_fsb) > last_block) {
 790                                 count_fsb = last_block - map_start_fsb;
 791                                 if (count_fsb == 0) {
 792                                         error = EAGAIN;
 793                                         goto trans_cancel;
 794                                 }
 795                         }
 796
 797                         /* Go get the actual blocks */
 798                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
 799                                         XFS_BMAPI_WRITE, &first_block, 1,
 800                                         imap, &nimaps, &free_list);
 801                         if (error)
 802                                 goto trans_cancel;
 803
 804                         error = xfs_bmap_finish(&tp, &free_list,
 805                                         first_block, &committed);
 806                         if (error)
 807                                 goto trans_cancel;
 808
 809                         error = xfs_trans_commit(tp,
 810                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
 811                         if (error)
 812                                 goto error0;
 813
 814                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 815                 }
 816
 817                 /*
 818                  * See if we were able to allocate an extent that
 819                  * covers at least part of the callers request
 820                  */
 821
 822                 for (i = 0; i < nimaps; i++) {
 823                         if ( !(io->io_flags & XFS_IOCORE_RT)  &&
 824                                 !imap[i].br_startblock) {
 825                                 cmn_err(CE_PANIC,"Access to block zero:  "
 826                                         "fs <%s> inode: %lld "
 827                                         "start_block : %llx start_off : %llx "
 828                                         "blkcnt : %llx extent-state : %x \n",
 829                                         (ip->i_mount)->m_fsname,
 830                                         (long long)ip->i_ino,
 831                                         imap[i].br_startblock,
 832                                         imap[i].br_startoff,
 833                                         imap[i].br_blockcount,imap[i].br_state);
 834                         }
 835                         if ((offset_fsb >= imap[i].br_startoff) &&
 836                             (offset_fsb < (imap[i].br_startoff +
 837                                            imap[i].br_blockcount))) {
 838                                 *map = imap[i];
 839                                 *retmap = 1;
 840                                 XFS_STATS_INC(xs_xstrat_quick);
 841                                 return 0;
 842                         }
 843                         count_fsb -= imap[i].br_blockcount;
 844                 }
 845
 846                 /* So far we have not mapped the requested part of the
 847                  * file, just surrounding data, try again.
 848                  */
 849                 nimaps--;
 850                 map_start_fsb = imap[nimaps].br_startoff +
 851                                 imap[nimaps].br_blockcount;
 852         }
 853
 854 trans_cancel:
 855         xfs_bmap_cancel(&free_list);
 856         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 857 error0:
 858         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 859         return XFS_ERROR(error);
 860 }
 861
 862 int
 863 xfs_iomap_write_unwritten(
 864         xfs_inode_t     *ip,
 865         xfs_off_t       offset,
 866         size_t          count)
 867 {
 868         xfs_mount_t     *mp = ip->i_mount;
 869         xfs_iocore_t    *io = &ip->i_iocore;
 870         xfs_trans_t     *tp;
 871         xfs_fileoff_t   offset_fsb;
 872         xfs_filblks_t   count_fsb;
 873         xfs_filblks_t   numblks_fsb;
 874         xfs_bmbt_irec_t imap;
 875         int             committed;
 876         int             error;
 877         int             nres;
 878         int             nimaps;
 879         xfs_fsblock_t   firstfsb;
 880         xfs_bmap_free_t free_list;
 881
 882         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
 883                                 &ip->i_iocore, offset, count);
 884
 885         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 886         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 887         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 888
 889         do {
 890                 nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 891
 892                 /*
 893                  * set up a transaction to convert the range of extents
 894                  * from unwritten to real. Do allocations in a loop until
 895                  * we have covered the range passed in.
 896                  */
 897
 898                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 899                 error = xfs_trans_reserve(tp, nres,
 900                                 XFS_WRITE_LOG_RES(mp), 0,
 901                                 XFS_TRANS_PERM_LOG_RES,
 902                                 XFS_WRITE_LOG_COUNT);
 903                 if (error) {
 904                         xfs_trans_cancel(tp, 0);
 905                         goto error0;
 906                 }
 907
 908                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 909                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 910                 xfs_trans_ihold(tp, ip);
 911
 912                 /*
 913                  * Modify the unwritten extent state of the buffer.
 914                  */
 915                 XFS_BMAP_INIT(&free_list, &firstfsb);
 916                 nimaps = 1;
 917                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 918                                   XFS_BMAPI_WRITE, &firstfsb,
 919                                   1, &imap, &nimaps, &free_list);
 920                 if (error)
 921                         goto error_on_bmapi_transaction;
 922
 923                 error = xfs_bmap_finish(&(tp), &(free_list),
 924                                 firstfsb, &committed);
 925                 if (error)
 926                         goto error_on_bmapi_transaction;
 927
 928                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 929                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 930                 if (error)
 931                         goto error0;
 932
 933                 if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
 934                         cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
 935                                 "inode: %lld start_block : %llx start_off : "
 936                                 "%llx blkcnt : %llx extent-state : %x \n",
 937                                 (ip->i_mount)->m_fsname,
 938                                 (long long)ip->i_ino,
 939                                 imap.br_startblock,imap.br_startoff,
 940                                 imap.br_blockcount,imap.br_state);
 941                 }
 942
 943                 if ((numblks_fsb = imap.br_blockcount) == 0) {
 944                         /*
 945                          * The numblks_fsb value should always get
 946                          * smaller, otherwise the loop is stuck.
 947                          */
 948                         ASSERT(imap.br_blockcount);
 949                         break;
 950                 }
 951                 offset_fsb += numblks_fsb;
 952                 count_fsb -= numblks_fsb;
 953         } while (count_fsb > 0);
 954
 955         return 0;
 956
 957 error_on_bmapi_transaction:
 958         xfs_bmap_cancel(&free_list);
 959         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
 960         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 961 error0:
 962         return XFS_ERROR(error);
 963 }