git.oblomov.eu Git - 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_rtalloc.h"
  44 #include "xfs_error.h"
  45 #include "xfs_itable.h"
  46 #include "xfs_rw.h"
  47 #include "xfs_acl.h"
  48 #include "xfs_cap.h"
  49 #include "xfs_mac.h"
  50 #include "xfs_attr.h"
  51 #include "xfs_buf_item.h"
  52 #include "xfs_trans_space.h"
  53 #include "xfs_utils.h"
  54 #include "xfs_iomap.h"
  55
  56 #if defined(XFS_RW_TRACE)
  57 void
  58 xfs_iomap_enter_trace(
  59         int             tag,
  60         xfs_iocore_t    *io,
  61         xfs_off_t       offset,
  62         ssize_t         count)
  63 {
  64         xfs_inode_t     *ip = XFS_IO_INODE(io);
  65
  66         if (!ip->i_rwtrace)
  67                 return;
  68
  69         ktrace_enter(ip->i_rwtrace,
  70                 (void *)((unsigned long)tag),
  71                 (void *)ip,
  72                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
  73                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
  74                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
  75                 (void *)((unsigned long)(offset & 0xffffffff)),
  76                 (void *)((unsigned long)count),
  77                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
  78                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
  79                 (void *)NULL,
  80                 (void *)NULL,
  81                 (void *)NULL,
  82                 (void *)NULL,
  83                 (void *)NULL,
  84                 (void *)NULL,
  85                 (void *)NULL);
  86 }
  87
  88 void
  89 xfs_iomap_map_trace(
  90         int             tag,
  91         xfs_iocore_t    *io,
  92         xfs_off_t       offset,
  93         ssize_t         count,
  94         xfs_iomap_t     *iomapp,
  95         xfs_bmbt_irec_t *imapp,
  96         int             flags)
  97 {
  98         xfs_inode_t     *ip = XFS_IO_INODE(io);
  99
 100         if (!ip->i_rwtrace)
 101                 return;
 102
 103         ktrace_enter(ip->i_rwtrace,
 104                 (void *)((unsigned long)tag),
 105                 (void *)ip,
 106                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
 107                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
 108                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
 109                 (void *)((unsigned long)(offset & 0xffffffff)),
 110                 (void *)((unsigned long)count),
 111                 (void *)((unsigned long)flags),
 112                 (void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
 113                 (void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
 114                 (void *)((unsigned long)(iomapp->iomap_delta)),
 115                 (void *)((unsigned long)(iomapp->iomap_bsize)),
 116                 (void *)((unsigned long)(iomapp->iomap_bn)),
 117                 (void *)(__psint_t)(imapp->br_startoff),
 118                 (void *)((unsigned long)(imapp->br_blockcount)),
 119                 (void *)(__psint_t)(imapp->br_startblock));
 120 }
 121 #else
 122 #define xfs_iomap_enter_trace(tag, io, offset, count)
 123 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
 124 #endif
 125
 126 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
 127                                                 << mp->m_writeio_log)
 128 #define XFS_STRAT_WRITE_IMAPS   2
 129 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
 130
 131 STATIC int
 132 xfs_imap_to_bmap(
 133         xfs_iocore_t    *io,
 134         xfs_off_t       offset,
 135         xfs_bmbt_irec_t *imap,
 136         xfs_iomap_t     *iomapp,
 137         int             imaps,                  /* Number of imap entries */
 138         int             iomaps,                 /* Number of iomap entries */
 139         int             flags)
 140 {
 141         xfs_mount_t     *mp;
 142         xfs_fsize_t     nisize;
 143         int             pbm;
 144         xfs_fsblock_t   start_block;
 145
 146         mp = io->io_mount;
 147         nisize = XFS_SIZE(mp, io);
 148         if (io->io_new_size > nisize)
 149                 nisize = io->io_new_size;
 150
 151         for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
 152                 iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
 153                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
 154                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
 155                 iomapp->iomap_flags = flags;
 156
 157                 if (io->io_flags & XFS_IOCORE_RT) {
 158                         iomapp->iomap_flags |= IOMAP_REALTIME;
 159                         iomapp->iomap_target = mp->m_rtdev_targp;
 160                 } else {
 161                         iomapp->iomap_target = mp->m_ddev_targp;
 162                 }
 163                 start_block = imap->br_startblock;
 164                 if (start_block == HOLESTARTBLOCK) {
 165                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 166                         iomapp->iomap_flags |= IOMAP_HOLE;
 167                 } else if (start_block == DELAYSTARTBLOCK) {
 168                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 169                         iomapp->iomap_flags |= IOMAP_DELAY;
 170                 } else {
 171                         iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
 172                         if (ISUNWRITTEN(imap))
 173                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
 174                 }
 175
 176                 if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
 177                         iomapp->iomap_flags |= IOMAP_EOF;
 178                 }
 179
 180                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
 181         }
 182         return pbm;     /* Return the number filled */
 183 }
 184
 185 int
 186 xfs_iomap(
 187         xfs_iocore_t    *io,
 188         xfs_off_t       offset,
 189         ssize_t         count,
 190         int             flags,
 191         xfs_iomap_t     *iomapp,
 192         int             *niomaps)
 193 {
 194         xfs_mount_t     *mp = io->io_mount;
 195         xfs_fileoff_t   offset_fsb, end_fsb;
 196         int             error = 0;
 197         int             lockmode = 0;
 198         xfs_bmbt_irec_t imap;
 199         int             nimaps = 1;
 200         int             bmapi_flags = 0;
 201         int             iomap_flags = 0;
 202
 203         if (XFS_FORCED_SHUTDOWN(mp))
 204                 return XFS_ERROR(EIO);
 205
 206         switch (flags &
 207                 (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
 208                  BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
 209         case BMAPI_READ:
 210                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
 211                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 212                 bmapi_flags = XFS_BMAPI_ENTIRE;
 213                 break;
 214         case BMAPI_WRITE:
 215                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
 216                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
 217                 if (flags & BMAPI_IGNSTATE)
 218                         bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
 219                 XFS_ILOCK(mp, io, lockmode);
 220                 break;
 221         case BMAPI_ALLOCATE:
 222                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
 223                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
 224                 bmapi_flags = XFS_BMAPI_ENTIRE;
 225                 /* Attempt non-blocking lock */
 226                 if (flags & BMAPI_TRYLOCK) {
 227                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
 228                                 return XFS_ERROR(EAGAIN);
 229                 } else {
 230                         XFS_ILOCK(mp, io, lockmode);
 231                 }
 232                 break;
 233         case BMAPI_UNWRITTEN:
 234                 goto phase2;
 235         case BMAPI_DEVICE:
 236                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 237                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
 238                         mp->m_rtdev_targp : mp->m_ddev_targp;
 239                 error = 0;
 240                 *niomaps = 1;
 241                 goto out;
 242         default:
 243                 BUG();
 244         }
 245
 246         ASSERT(offset <= mp->m_maxioffset);
 247         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
 248                 count = mp->m_maxioffset - offset;
 249         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 250         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 251
 252         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
 253                         (xfs_filblks_t)(end_fsb - offset_fsb),
 254                         bmapi_flags,  NULL, 0, &imap,
 255                         &nimaps, NULL);
 256
 257         if (error)
 258                 goto out;
 259
 260 phase2:
 261         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
 262         case BMAPI_WRITE:
 263                 /* If we found an extent, return it */
 264                 if (nimaps &&
 265                     (imap.br_startblock != HOLESTARTBLOCK) &&
 266                     (imap.br_startblock != DELAYSTARTBLOCK)) {
 267                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 268                                         offset, count, iomapp, &imap, flags);
 269                         break;
 270                 }
 271
 272                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
 273                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
 274                                         count, flags, &imap, &nimaps, nimaps);
 275                 } else {
 276                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
 277                                         flags, &imap, &nimaps);
 278                 }
 279                 if (!error) {
 280                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
 281                                         offset, count, iomapp, &imap, flags);
 282                 }
 283                 iomap_flags = IOMAP_NEW;
 284                 break;
 285         case BMAPI_ALLOCATE:
 286                 /* If we found an extent, return it */
 287                 XFS_IUNLOCK(mp, io, lockmode);
 288                 lockmode = 0;
 289
 290                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
 291                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 292                                         offset, count, iomapp, &imap, flags);
 293                         break;
 294                 }
 295
 296                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
 297                                                  &imap, &nimaps);
 298                 break;
 299         case BMAPI_UNWRITTEN:
 300                 lockmode = 0;
 301                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
 302                 nimaps = 0;
 303                 break;
 304         }
 305
 306         if (nimaps) {
 307                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
 308                                             iomapp, nimaps, *niomaps, iomap_flags);
 309         } else if (niomaps) {
 310                 *niomaps = 0;
 311         }
 312
 313 out:
 314         if (lockmode)
 315                 XFS_IUNLOCK(mp, io, lockmode);
 316         return XFS_ERROR(error);
 317 }
 318
 319 STATIC int
 320 xfs_iomap_eof_align_last_fsb(
 321         xfs_mount_t     *mp,
 322         xfs_iocore_t    *io,
 323         xfs_fsize_t     isize,
 324         xfs_extlen_t    extsize,
 325         xfs_fileoff_t   *last_fsb)
 326 {
 327         xfs_fileoff_t   new_last_fsb = 0;
 328         xfs_extlen_t    align;
 329         int             eof, error;
 330
 331         if (io->io_flags & XFS_IOCORE_RT)
 332                 ;
 333         /*
 334          * If mounted with the "-o swalloc" option, roundup the allocation
 335          * request to a stripe width boundary if the file size is >=
 336          * stripe width and we are allocating past the allocation eof.
 337          */
 338         else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
 339                 (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)))
 340                 new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
 341         /*
 342          * Roundup the allocation request to a stripe unit (m_dalign) boundary
 343          * if the file size is >= stripe unit size, and we are allocating past
 344          * the allocation eof.
 345          */
 346         else if (mp->m_dalign && (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)))
 347                 new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
 348
 349         /*
 350          * Always round up the allocation request to an extent boundary
 351          * (when file on a real-time subvolume or has di_extsize hint).
 352          */
 353         if (extsize) {
 354                 if (new_last_fsb)
 355                         align = roundup_64(new_last_fsb, extsize);
 356                 else
 357                         align = extsize;
 358                 new_last_fsb = roundup_64(*last_fsb, align);
 359         }
 360
 361         if (new_last_fsb) {
 362                 error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
 363                 if (error)
 364                         return error;
 365                 if (eof)
 366                         *last_fsb = new_last_fsb;
 367         }
 368         return 0;
 369 }
 370
 371 STATIC int
 372 xfs_flush_space(
 373         xfs_inode_t     *ip,
 374         int             *fsynced,
 375         int             *ioflags)
 376 {
 377         switch (*fsynced) {
 378         case 0:
 379                 if (ip->i_delayed_blks) {
 380                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 381                         xfs_flush_inode(ip);
 382                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 383                         *fsynced = 1;
 384                 } else {
 385                         *ioflags |= BMAPI_SYNC;
 386                         *fsynced = 2;
 387                 }
 388                 return 0;
 389         case 1:
 390                 *fsynced = 2;
 391                 *ioflags |= BMAPI_SYNC;
 392                 return 0;
 393         case 2:
 394                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 395                 xfs_flush_device(ip);
 396                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 397                 *fsynced = 3;
 398                 return 0;
 399         }
 400         return 1;
 401 }
 402
 403 int
 404 xfs_iomap_write_direct(
 405         xfs_inode_t     *ip,
 406         xfs_off_t       offset,
 407         size_t          count,
 408         int             flags,
 409         xfs_bmbt_irec_t *ret_imap,
 410         int             *nmaps,
 411         int             found)
 412 {
 413         xfs_mount_t     *mp = ip->i_mount;
 414         xfs_iocore_t    *io = &ip->i_iocore;
 415         xfs_fileoff_t   offset_fsb;
 416         xfs_fileoff_t   last_fsb;
 417         xfs_filblks_t   count_fsb, resaligned;
 418         xfs_fsblock_t   firstfsb;
 419         xfs_extlen_t    extsz, temp;
 420         xfs_fsize_t     isize;
 421         int             nimaps;
 422         int             bmapi_flag;
 423         int             quota_flag;
 424         int             rt;
 425         xfs_trans_t     *tp;
 426         xfs_bmbt_irec_t imap;
 427         xfs_bmap_free_t free_list;
 428         uint            qblocks, resblks, resrtextents;
 429         int             committed;
 430         int             error;
 431
 432         /*
 433          * Make sure that the dquots are there. This doesn't hold
 434          * the ilock across a disk read.
 435          */
 436         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
 437         if (error)
 438                 return XFS_ERROR(error);
 439
 440         rt = XFS_IS_REALTIME_INODE(ip);
 441         if (unlikely(rt)) {
 442                 if (!(extsz = ip->i_d.di_extsize))
 443                         extsz = mp->m_sb.sb_rextsize;
 444         } else {
 445                 extsz = ip->i_d.di_extsize;
 446         }
 447
 448         isize = ip->i_d.di_size;
 449         if (io->io_new_size > isize)
 450                 isize = io->io_new_size;
 451
 452         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 453         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 454         if ((offset + count) > isize) {
 455                 error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
 456                                                         &last_fsb);
 457                 if (error)
 458                         goto error_out;
 459         } else {
 460                 if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
 461                         last_fsb = MIN(last_fsb, (xfs_fileoff_t)
 462                                         ret_imap->br_blockcount +
 463                                         ret_imap->br_startoff);
 464         }
 465         count_fsb = last_fsb - offset_fsb;
 466         ASSERT(count_fsb > 0);
 467
 468         resaligned = count_fsb;
 469         if (unlikely(extsz)) {
 470                 if ((temp = do_mod(offset_fsb, extsz)))
 471                         resaligned += temp;
 472                 if ((temp = do_mod(resaligned, extsz)))
 473                         resaligned += extsz - temp;
 474         }
 475
 476         if (unlikely(rt)) {
 477                 resrtextents = qblocks = resaligned;
 478                 resrtextents /= mp->m_sb.sb_rextsize;
 479                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 480                 quota_flag = XFS_QMOPT_RES_RTBLKS;
 481         } else {
 482                 resrtextents = 0;
 483                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
 484                 quota_flag = XFS_QMOPT_RES_REGBLKS;
 485         }
 486
 487         /*
 488          * Allocate and setup the transaction
 489          */
 490         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 491         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
 492         error = xfs_trans_reserve(tp, resblks,
 493                         XFS_WRITE_LOG_RES(mp), resrtextents,
 494                         XFS_TRANS_PERM_LOG_RES,
 495                         XFS_WRITE_LOG_COUNT);
 496         /*
 497          * Check for running out of space, note: need lock to return
 498          */
 499         if (error)
 500                 xfs_trans_cancel(tp, 0);
 501         xfs_ilock(ip, XFS_ILOCK_EXCL);
 502         if (error)
 503                 goto error_out;
 504
 505         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
 506                                               qblocks, 0, quota_flag);
 507         if (error)
 508                 goto error1;
 509
 510         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 511         xfs_trans_ihold(tp, ip);
 512
 513         bmapi_flag = XFS_BMAPI_WRITE;
 514         if ((flags & BMAPI_DIRECT) && (offset < ip->i_d.di_size || extsz))
 515                 bmapi_flag |= XFS_BMAPI_PREALLOC;
 516
 517         /*
 518          * Issue the xfs_bmapi() call to allocate the blocks
 519          */
 520         XFS_BMAP_INIT(&free_list, &firstfsb);
 521         nimaps = 1;
 522         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 523                 bmapi_flag, &firstfsb, 0, &imap, &nimaps, &free_list);
 524         if (error)
 525                 goto error0;
 526
 527         /*
 528          * Complete the transaction
 529          */
 530         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
 531         if (error)
 532                 goto error0;
 533         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 534         if (error)
 535                 goto error_out;
 536
 537         /*
 538          * Copy any maps to caller's array and return any error.
 539          */
 540         if (nimaps == 0) {
 541                 error = (ENOSPC);
 542                 goto error_out;
 543         }
 544
 545         *ret_imap = imap;
 546         *nmaps = 1;
 547         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 548                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 549                         "start_block : %llx start_off : %llx blkcnt : %llx "
 550                         "extent-state : %x \n",
 551                         (ip->i_mount)->m_fsname,
 552                         (long long)ip->i_ino,
 553                         (unsigned long long)ret_imap->br_startblock,
 554                         (unsigned long long)ret_imap->br_startoff,
 555                         (unsigned long long)ret_imap->br_blockcount,
 556                         ret_imap->br_state);
 557         }
 558         return 0;
 559
 560 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
 561         xfs_bmap_cancel(&free_list);
 562         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
 563
 564 error1: /* Just cancel transaction */
 565         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 566         *nmaps = 0;     /* nothing set-up here */
 567
 568 error_out:
 569         return XFS_ERROR(error);
 570 }
 571
 572 /*
 573  * If the caller is doing a write at the end of the file,
 574  * then extend the allocation out to the file system's write
 575  * iosize.  We clean up any extra space left over when the
 576  * file is closed in xfs_inactive().
 577  *
 578  * For sync writes, we are flushing delayed allocate space to
 579  * try to make additional space available for allocation near
 580  * the filesystem full boundary - preallocation hurts in that
 581  * situation, of course.
 582  */
 583 STATIC int
 584 xfs_iomap_eof_want_preallocate(
 585         xfs_mount_t     *mp,
 586         xfs_iocore_t    *io,
 587         xfs_fsize_t     isize,
 588         xfs_off_t       offset,
 589         size_t          count,
 590         int             ioflag,
 591         xfs_bmbt_irec_t *imap,
 592         int             nimaps,
 593         int             *prealloc)
 594 {
 595         xfs_fileoff_t   start_fsb;
 596         xfs_filblks_t   count_fsb;
 597         xfs_fsblock_t   firstblock;
 598         int             n, error, imaps;
 599
 600         *prealloc = 0;
 601         if ((ioflag & BMAPI_SYNC) || (offset + count) <= isize)
 602                 return 0;
 603
 604         /*
 605          * If there are any real blocks past eof, then don't
 606          * do any speculative allocation.
 607          */
 608         start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
 609         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
 610         while (count_fsb > 0) {
 611                 imaps = nimaps;
 612                 firstblock = NULLFSBLOCK;
 613                 error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
 614                                   0, &firstblock, 0, imap, &imaps, NULL);
 615                 if (error)
 616                         return error;
 617                 for (n = 0; n < imaps; n++) {
 618                         if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
 619                             (imap[n].br_startblock != DELAYSTARTBLOCK))
 620                                 return 0;
 621                         start_fsb += imap[n].br_blockcount;
 622                         count_fsb -= imap[n].br_blockcount;
 623                 }
 624         }
 625         *prealloc = 1;
 626         return 0;
 627 }
 628
 629 int
 630 xfs_iomap_write_delay(
 631         xfs_inode_t     *ip,
 632         xfs_off_t       offset,
 633         size_t          count,
 634         int             ioflag,
 635         xfs_bmbt_irec_t *ret_imap,
 636         int             *nmaps)
 637 {
 638         xfs_mount_t     *mp = ip->i_mount;
 639         xfs_iocore_t    *io = &ip->i_iocore;
 640         xfs_fileoff_t   offset_fsb;
 641         xfs_fileoff_t   last_fsb;
 642         xfs_off_t       aligned_offset;
 643         xfs_fileoff_t   ioalign;
 644         xfs_fsblock_t   firstblock;
 645         xfs_extlen_t    extsz;
 646         xfs_fsize_t     isize;
 647         int             nimaps;
 648         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
 649         int             prealloc, fsynced = 0;
 650         int             error;
 651
 652         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
 653
 654         /*
 655          * Make sure that the dquots are there. This doesn't hold
 656          * the ilock across a disk read.
 657          */
 658         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
 659         if (error)
 660                 return XFS_ERROR(error);
 661
 662         if (XFS_IS_REALTIME_INODE(ip)) {
 663                 if (!(extsz = ip->i_d.di_extsize))
 664                         extsz = mp->m_sb.sb_rextsize;
 665         } else {
 666                 extsz = ip->i_d.di_extsize;
 667         }
 668
 669         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 670
 671 retry:
 672         isize = ip->i_d.di_size;
 673         if (io->io_new_size > isize)
 674                 isize = io->io_new_size;
 675
 676         error = xfs_iomap_eof_want_preallocate(mp, io, isize, offset, count,
 677                                 ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
 678         if (error)
 679                 return error;
 680
 681         if (prealloc) {
 682                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 683                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
 684                 last_fsb = ioalign + mp->m_writeio_blocks;
 685         } else {
 686                 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 687         }
 688
 689         if (prealloc || extsz) {
 690                 error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
 691                                                         &last_fsb);
 692                 if (error)
 693                         return error;
 694         }
 695
 696         nimaps = XFS_WRITE_IMAPS;
 697         firstblock = NULLFSBLOCK;
 698         error = xfs_bmapi(NULL, ip, offset_fsb,
 699                           (xfs_filblks_t)(last_fsb - offset_fsb),
 700                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
 701                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
 702                           &nimaps, NULL);
 703         if (error && (error != ENOSPC))
 704                 return XFS_ERROR(error);
 705
 706         /*
 707          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
 708          * then we must have run out of space - flush delalloc, and retry..
 709          */
 710         if (nimaps == 0) {
 711                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
 712                                         io, offset, count);
 713                 if (xfs_flush_space(ip, &fsynced, &ioflag))
 714                         return XFS_ERROR(ENOSPC);
 715
 716                 error = 0;
 717                 goto retry;
 718         }
 719
 720         if (!(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 721                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 722                         "start_block : %llx start_off : %llx blkcnt : %llx "
 723                         "extent-state : %x \n",
 724                         (ip->i_mount)->m_fsname,
 725                         (long long)ip->i_ino,
 726                         (unsigned long long)ret_imap->br_startblock,
 727                         (unsigned long long)ret_imap->br_startoff,
 728                         (unsigned long long)ret_imap->br_blockcount,
 729                         ret_imap->br_state);
 730         }
 731
 732         *ret_imap = imap[0];
 733         *nmaps = 1;
 734
 735         return 0;
 736 }
 737
 738 /*
 739  * Pass in a delayed allocate extent, convert it to real extents;
 740  * return to the caller the extent we create which maps on top of
 741  * the originating callers request.
 742  *
 743  * Called without a lock on the inode.
 744  */
 745 int
 746 xfs_iomap_write_allocate(
 747         xfs_inode_t     *ip,
 748         xfs_off_t       offset,
 749         size_t          count,
 750         xfs_bmbt_irec_t *map,
 751         int             *retmap)
 752 {
 753         xfs_mount_t     *mp = ip->i_mount;
 754         xfs_iocore_t    *io = &ip->i_iocore;
 755         xfs_fileoff_t   offset_fsb, last_block;
 756         xfs_fileoff_t   end_fsb, map_start_fsb;
 757         xfs_fsblock_t   first_block;
 758         xfs_bmap_free_t free_list;
 759         xfs_filblks_t   count_fsb;
 760         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
 761         xfs_trans_t     *tp;
 762         int             i, nimaps, committed;
 763         int             error = 0;
 764         int             nres;
 765
 766         *retmap = 0;
 767
 768         /*
 769          * Make sure that the dquots are there.
 770          */
 771         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
 772                 return XFS_ERROR(error);
 773
 774         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 775         count_fsb = map->br_blockcount;
 776         map_start_fsb = map->br_startoff;
 777
 778         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 779
 780         while (count_fsb != 0) {
 781                 /*
 782                  * Set up a transaction with which to allocate the
 783                  * backing store for the file.  Do allocations in a
 784                  * loop until we get some space in the range we are
 785                  * interested in.  The other space that might be allocated
 786                  * is in the delayed allocation extent on which we sit
 787                  * but before our buffer starts.
 788                  */
 789
 790                 nimaps = 0;
 791                 while (nimaps == 0) {
 792                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 793                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 794                         error = xfs_trans_reserve(tp, nres,
 795                                         XFS_WRITE_LOG_RES(mp),
 796                                         0, XFS_TRANS_PERM_LOG_RES,
 797                                         XFS_WRITE_LOG_COUNT);
 798                         if (error == ENOSPC) {
 799                                 error = xfs_trans_reserve(tp, 0,
 800                                                 XFS_WRITE_LOG_RES(mp),
 801                                                 0,
 802                                                 XFS_TRANS_PERM_LOG_RES,
 803                                                 XFS_WRITE_LOG_COUNT);
 804                         }
 805                         if (error) {
 806                                 xfs_trans_cancel(tp, 0);
 807                                 return XFS_ERROR(error);
 808                         }
 809                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 810                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 811                         xfs_trans_ihold(tp, ip);
 812
 813                         XFS_BMAP_INIT(&free_list, &first_block);
 814
 815                         nimaps = XFS_STRAT_WRITE_IMAPS;
 816                         /*
 817                          * Ensure we don't go beyond eof - it is possible
 818                          * the extents changed since we did the read call,
 819                          * we dropped the ilock in the interim.
 820                          */
 821
 822                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
 823                         xfs_bmap_last_offset(NULL, ip, &last_block,
 824                                 XFS_DATA_FORK);
 825                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 826                         if ((map_start_fsb + count_fsb) > last_block) {
 827                                 count_fsb = last_block - map_start_fsb;
 828                                 if (count_fsb == 0) {
 829                                         error = EAGAIN;
 830                                         goto trans_cancel;
 831                                 }
 832                         }
 833
 834                         /* Go get the actual blocks */
 835                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
 836                                         XFS_BMAPI_WRITE, &first_block, 1,
 837                                         imap, &nimaps, &free_list);
 838                         if (error)
 839                                 goto trans_cancel;
 840
 841                         error = xfs_bmap_finish(&tp, &free_list,
 842                                         first_block, &committed);
 843                         if (error)
 844                                 goto trans_cancel;
 845
 846                         error = xfs_trans_commit(tp,
 847                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
 848                         if (error)
 849                                 goto error0;
 850
 851                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 852                 }
 853
 854                 /*
 855                  * See if we were able to allocate an extent that
 856                  * covers at least part of the callers request
 857                  */
 858
 859                 for (i = 0; i < nimaps; i++) {
 860                         if (!(io->io_flags & XFS_IOCORE_RT)  &&
 861                             !imap[i].br_startblock) {
 862                                 cmn_err(CE_PANIC,"Access to block zero:  "
 863                                         "fs <%s> inode: %lld "
 864                                         "start_block : %llx start_off : %llx "
 865                                         "blkcnt : %llx extent-state : %x \n",
 866                                         (ip->i_mount)->m_fsname,
 867                                         (long long)ip->i_ino,
 868                                         (unsigned long long)
 869                                                 imap[i].br_startblock,
 870                                         (unsigned long long)
 871                                                 imap[i].br_startoff,
 872                                         (unsigned long long)
 873                                                 imap[i].br_blockcount,
 874                                         imap[i].br_state);
 875                         }
 876                         if ((offset_fsb >= imap[i].br_startoff) &&
 877                             (offset_fsb < (imap[i].br_startoff +
 878                                            imap[i].br_blockcount))) {
 879                                 *map = imap[i];
 880                                 *retmap = 1;
 881                                 XFS_STATS_INC(xs_xstrat_quick);
 882                                 return 0;
 883                         }
 884                         count_fsb -= imap[i].br_blockcount;
 885                 }
 886
 887                 /* So far we have not mapped the requested part of the
 888                  * file, just surrounding data, try again.
 889                  */
 890                 nimaps--;
 891                 map_start_fsb = imap[nimaps].br_startoff +
 892                                 imap[nimaps].br_blockcount;
 893         }
 894
 895 trans_cancel:
 896         xfs_bmap_cancel(&free_list);
 897         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 898 error0:
 899         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 900         return XFS_ERROR(error);
 901 }
 902
 903 int
 904 xfs_iomap_write_unwritten(
 905         xfs_inode_t     *ip,
 906         xfs_off_t       offset,
 907         size_t          count)
 908 {
 909         xfs_mount_t     *mp = ip->i_mount;
 910         xfs_iocore_t    *io = &ip->i_iocore;
 911         xfs_fileoff_t   offset_fsb;
 912         xfs_filblks_t   count_fsb;
 913         xfs_filblks_t   numblks_fsb;
 914         xfs_fsblock_t   firstfsb;
 915         int             nimaps;
 916         xfs_trans_t     *tp;
 917         xfs_bmbt_irec_t imap;
 918         xfs_bmap_free_t free_list;
 919         uint            resblks;
 920         int             committed;
 921         int             error;
 922
 923         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
 924                                 &ip->i_iocore, offset, count);
 925
 926         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 927         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 928         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 929
 930         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 931
 932         do {
 933                 /*
 934                  * set up a transaction to convert the range of extents
 935                  * from unwritten to real. Do allocations in a loop until
 936                  * we have covered the range passed in.
 937                  */
 938
 939                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 940                 error = xfs_trans_reserve(tp, resblks,
 941                                 XFS_WRITE_LOG_RES(mp), 0,
 942                                 XFS_TRANS_PERM_LOG_RES,
 943                                 XFS_WRITE_LOG_COUNT);
 944                 if (error) {
 945                         xfs_trans_cancel(tp, 0);
 946                         goto error0;
 947                 }
 948
 949                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 950                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 951                 xfs_trans_ihold(tp, ip);
 952
 953                 /*
 954                  * Modify the unwritten extent state of the buffer.
 955                  */
 956                 XFS_BMAP_INIT(&free_list, &firstfsb);
 957                 nimaps = 1;
 958                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 959                                   XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
 960                                   1, &imap, &nimaps, &free_list);
 961                 if (error)
 962                         goto error_on_bmapi_transaction;
 963
 964                 error = xfs_bmap_finish(&(tp), &(free_list),
 965                                 firstfsb, &committed);
 966                 if (error)
 967                         goto error_on_bmapi_transaction;
 968
 969                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 970                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 971                 if (error)
 972                         goto error0;
 973
 974                 if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
 975                         cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
 976                                 "inode: %lld start_block : %llx start_off : "
 977                                 "%llx blkcnt : %llx extent-state : %x \n",
 978                                 (ip->i_mount)->m_fsname,
 979                                 (long long)ip->i_ino,
 980                                 (unsigned long long)imap.br_startblock,
 981                                 (unsigned long long)imap.br_startoff,
 982                                 (unsigned long long)imap.br_blockcount,
 983                                 imap.br_state);
 984                 }
 985
 986                 if ((numblks_fsb = imap.br_blockcount) == 0) {
 987                         /*
 988                          * The numblks_fsb value should always get
 989                          * smaller, otherwise the loop is stuck.
 990                          */
 991                         ASSERT(imap.br_blockcount);
 992                         break;
 993                 }
 994                 offset_fsb += numblks_fsb;
 995                 count_fsb -= numblks_fsb;
 996         } while (count_fsb > 0);
 997
 998         return 0;
 999
1000 error_on_bmapi_transaction:
1001         xfs_bmap_cancel(&free_list);
1002         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
1003         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1004 error0:
1005         return XFS_ERROR(error);
1006 }