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