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