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