fs/xfs/xfs_iomap.c

   1 /*
   2  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it would be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  11  *
  12  * Further, this software is distributed without any warranty that it is
  13  * free of the rightful claim of any third person regarding infringement
  14  * or the like.  Any license provided herein, whether implied or
  15  * otherwise, applies only to this software file.  Patent licenses, if
  16  * any, provided herein do not apply to combinations of this program with
  17  * other software, or any other product whatsoever.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write the Free Software Foundation, Inc., 59
  21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  22  *
  23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  24  * Mountain View, CA  94043, or:
  25  *
  26  * http://www.sgi.com
  27  *
  28  * For further information regarding this notice, see:
  29  *
  30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  31  */
  32
  33 #include "xfs.h"
  34
  35 #include "xfs_fs.h"
  36 #include "xfs_inum.h"
  37 #include "xfs_log.h"
  38 #include "xfs_trans.h"
  39 #include "xfs_sb.h"
  40 #include "xfs_ag.h"
  41 #include "xfs_dir.h"
  42 #include "xfs_dir2.h"
  43 #include "xfs_alloc.h"
  44 #include "xfs_dmapi.h"
  45 #include "xfs_quota.h"
  46 #include "xfs_mount.h"
  47 #include "xfs_alloc_btree.h"
  48 #include "xfs_bmap_btree.h"
  49 #include "xfs_ialloc_btree.h"
  50 #include "xfs_btree.h"
  51 #include "xfs_ialloc.h"
  52 #include "xfs_attr_sf.h"
  53 #include "xfs_dir_sf.h"
  54 #include "xfs_dir2_sf.h"
  55 #include "xfs_dinode.h"
  56 #include "xfs_inode.h"
  57 #include "xfs_bmap.h"
  58 #include "xfs_bit.h"
  59 #include "xfs_rtalloc.h"
  60 #include "xfs_error.h"
  61 #include "xfs_itable.h"
  62 #include "xfs_rw.h"
  63 #include "xfs_acl.h"
  64 #include "xfs_cap.h"
  65 #include "xfs_mac.h"
  66 #include "xfs_attr.h"
  67 #include "xfs_buf_item.h"
  68 #include "xfs_trans_space.h"
  69 #include "xfs_utils.h"
  70 #include "xfs_iomap.h"
  71
  72 #if defined(XFS_RW_TRACE)
  73 void
  74 xfs_iomap_enter_trace(
  75         int             tag,
  76         xfs_iocore_t    *io,
  77         xfs_off_t       offset,
  78         ssize_t         count)
  79 {
  80         xfs_inode_t     *ip = XFS_IO_INODE(io);
  81
  82         if (!ip->i_rwtrace)
  83                 return;
  84
  85         ktrace_enter(ip->i_rwtrace,
  86                 (void *)((unsigned long)tag),
  87                 (void *)ip,
  88                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
  89                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
  90                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
  91                 (void *)((unsigned long)(offset & 0xffffffff)),
  92                 (void *)((unsigned long)count),
  93                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
  94                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
  95                 (void *)NULL,
  96                 (void *)NULL,
  97                 (void *)NULL,
  98                 (void *)NULL,
  99                 (void *)NULL,
 100                 (void *)NULL,
 101                 (void *)NULL);
 102 }
 103
 104 void
 105 xfs_iomap_map_trace(
 106         int             tag,
 107         xfs_iocore_t    *io,
 108         xfs_off_t       offset,
 109         ssize_t         count,
 110         xfs_iomap_t     *iomapp,
 111         xfs_bmbt_irec_t *imapp,
 112         int             flags)
 113 {
 114         xfs_inode_t     *ip = XFS_IO_INODE(io);
 115
 116         if (!ip->i_rwtrace)
 117                 return;
 118
 119         ktrace_enter(ip->i_rwtrace,
 120                 (void *)((unsigned long)tag),
 121                 (void *)ip,
 122                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
 123                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
 124                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
 125                 (void *)((unsigned long)(offset & 0xffffffff)),
 126                 (void *)((unsigned long)count),
 127                 (void *)((unsigned long)flags),
 128                 (void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
 129                 (void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
 130                 (void *)((unsigned long)(iomapp->iomap_delta)),
 131                 (void *)((unsigned long)(iomapp->iomap_bsize)),
 132                 (void *)((unsigned long)(iomapp->iomap_bn)),
 133                 (void *)(__psint_t)(imapp->br_startoff),
 134                 (void *)((unsigned long)(imapp->br_blockcount)),
 135                 (void *)(__psint_t)(imapp->br_startblock));
 136 }
 137 #else
 138 #define xfs_iomap_enter_trace(tag, io, offset, count)
 139 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
 140 #endif
 141
 142 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
 143                                                 << mp->m_writeio_log)
 144 #define XFS_STRAT_WRITE_IMAPS   2
 145 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
 146
 147 STATIC int
 148 xfs_imap_to_bmap(
 149         xfs_iocore_t    *io,
 150         xfs_off_t       offset,
 151         xfs_bmbt_irec_t *imap,
 152         xfs_iomap_t     *iomapp,
 153         int             imaps,                  /* Number of imap entries */
 154         int             iomaps,                 /* Number of iomap entries */
 155         int             flags)
 156 {
 157         xfs_mount_t     *mp;
 158         xfs_fsize_t     nisize;
 159         int             pbm;
 160         xfs_fsblock_t   start_block;
 161
 162         mp = io->io_mount;
 163         nisize = XFS_SIZE(mp, io);
 164         if (io->io_new_size > nisize)
 165                 nisize = io->io_new_size;
 166
 167         for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
 168                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
 169                         mp->m_rtdev_targp : mp->m_ddev_targp;
 170                 iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
 171                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
 172                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
 173                 iomapp->iomap_flags = flags;
 174
 175                 start_block = imap->br_startblock;
 176                 if (start_block == HOLESTARTBLOCK) {
 177                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 178                         iomapp->iomap_flags = IOMAP_HOLE;
 179                 } else if (start_block == DELAYSTARTBLOCK) {
 180                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 181                         iomapp->iomap_flags = IOMAP_DELAY;
 182                 } else {
 183                         iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
 184                         if (ISUNWRITTEN(imap))
 185                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
 186                 }
 187
 188                 if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
 189                         iomapp->iomap_flags |= IOMAP_EOF;
 190                 }
 191
 192                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
 193         }
 194         return pbm;     /* Return the number filled */
 195 }
 196
 197 int
 198 xfs_iomap(
 199         xfs_iocore_t    *io,
 200         xfs_off_t       offset,
 201         ssize_t         count,
 202         int             flags,
 203         xfs_iomap_t     *iomapp,
 204         int             *niomaps)
 205 {
 206         xfs_mount_t     *mp = io->io_mount;
 207         xfs_fileoff_t   offset_fsb, end_fsb;
 208         int             error = 0;
 209         int             lockmode = 0;
 210         xfs_bmbt_irec_t imap;
 211         int             nimaps = 1;
 212         int             bmapi_flags = 0;
 213         int             iomap_flags = 0;
 214
 215         if (XFS_FORCED_SHUTDOWN(mp))
 216                 return XFS_ERROR(EIO);
 217
 218         switch (flags &
 219                 (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
 220                  BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
 221         case BMAPI_READ:
 222                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
 223                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 224                 bmapi_flags = XFS_BMAPI_ENTIRE;
 225                 if (flags & BMAPI_IGNSTATE)
 226                         bmapi_flags |= XFS_BMAPI_IGSTATE;
 227                 break;
 228         case BMAPI_WRITE:
 229                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
 230                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
 231                 bmapi_flags = 0;
 232                 XFS_ILOCK(mp, io, lockmode);
 233                 break;
 234         case BMAPI_ALLOCATE:
 235                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
 236                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
 237                 bmapi_flags = XFS_BMAPI_ENTIRE;
 238                 /* Attempt non-blocking lock */
 239                 if (flags & BMAPI_TRYLOCK) {
 240                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
 241                                 return XFS_ERROR(EAGAIN);
 242                 } else {
 243                         XFS_ILOCK(mp, io, lockmode);
 244                 }
 245                 break;
 246         case BMAPI_UNWRITTEN:
 247                 goto phase2;
 248         case BMAPI_DEVICE:
 249                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 250                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
 251                         mp->m_rtdev_targp : mp->m_ddev_targp;
 252                 error = 0;
 253                 *niomaps = 1;
 254                 goto out;
 255         default:
 256                 BUG();
 257         }
 258
 259         ASSERT(offset <= mp->m_maxioffset);
 260         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
 261                 count = mp->m_maxioffset - offset;
 262         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 263         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 264
 265         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
 266                         (xfs_filblks_t)(end_fsb - offset_fsb),
 267                         bmapi_flags,  NULL, 0, &imap,
 268                         &nimaps, NULL);
 269
 270         if (error)
 271                 goto out;
 272
 273 phase2:
 274         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
 275         case BMAPI_WRITE:
 276                 /* If we found an extent, return it */
 277                 if (nimaps && (imap.br_startblock != HOLESTARTBLOCK)) {
 278                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 279                                         offset, count, iomapp, &imap, flags);
 280                         break;
 281                 }
 282
 283                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
 284                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
 285                                         count, flags, &imap, &nimaps, nimaps);
 286                 } else {
 287                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
 288                                         flags, &imap, &nimaps);
 289                 }
 290                 if (!error) {
 291                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
 292                                         offset, count, iomapp, &imap, flags);
 293                 }
 294                 iomap_flags = IOMAP_NEW;
 295                 break;
 296         case BMAPI_ALLOCATE:
 297                 /* If we found an extent, return it */
 298                 XFS_IUNLOCK(mp, io, lockmode);
 299                 lockmode = 0;
 300
 301                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
 302                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 303                                         offset, count, iomapp, &imap, flags);
 304                         break;
 305                 }
 306
 307                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, &imap, &nimaps);
 308                 break;
 309         case BMAPI_UNWRITTEN:
 310                 lockmode = 0;
 311                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
 312                 nimaps = 0;
 313                 break;
 314         }
 315
 316         if (nimaps) {
 317                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
 318                                             iomapp, nimaps, *niomaps, iomap_flags);
 319         } else if (niomaps) {
 320                 *niomaps = 0;
 321         }
 322
 323 out:
 324         if (lockmode)
 325                 XFS_IUNLOCK(mp, io, lockmode);
 326         return XFS_ERROR(error);
 327 }
 328
 329 STATIC int
 330 xfs_flush_space(
 331         xfs_inode_t     *ip,
 332         int             *fsynced,
 333         int             *ioflags)
 334 {
 335         switch (*fsynced) {
 336         case 0:
 337                 if (ip->i_delayed_blks) {
 338                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 339                         xfs_flush_inode(ip);
 340                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 341                         *fsynced = 1;
 342                 } else {
 343                         *ioflags |= BMAPI_SYNC;
 344                         *fsynced = 2;
 345                 }
 346                 return 0;
 347         case 1:
 348                 *fsynced = 2;
 349                 *ioflags |= BMAPI_SYNC;
 350                 return 0;
 351         case 2:
 352                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 353                 xfs_flush_device(ip);
 354                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 355                 *fsynced = 3;
 356                 return 0;
 357         }
 358         return 1;
 359 }
 360
 361 int
 362 xfs_iomap_write_direct(
 363         xfs_inode_t     *ip,
 364         loff_t          offset,
 365         size_t          count,
 366         int             flags,
 367         xfs_bmbt_irec_t *ret_imap,
 368         int             *nmaps,
 369         int             found)
 370 {
 371         xfs_mount_t     *mp = ip->i_mount;
 372         xfs_iocore_t    *io = &ip->i_iocore;
 373         xfs_fileoff_t   offset_fsb;
 374         xfs_fileoff_t   last_fsb;
 375         xfs_filblks_t   count_fsb;
 376         xfs_fsize_t     isize;
 377         xfs_fsblock_t   firstfsb;
 378         int             nimaps, maps;
 379         int             error;
 380         int             bmapi_flag;
 381         int             rt;
 382         xfs_trans_t     *tp;
 383         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS], *imapp;
 384         xfs_bmap_free_t free_list;
 385         int             aeof;
 386         xfs_filblks_t   datablocks;
 387         int             committed;
 388         int             numrtextents;
 389         uint            resblks;
 390
 391         /*
 392          * Make sure that the dquots are there. This doesn't hold
 393          * the ilock across a disk read.
 394          */
 395         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
 396         if (error)
 397                 return XFS_ERROR(error);
 398
 399         maps = min(XFS_WRITE_IMAPS, *nmaps);
 400         nimaps = maps;
 401
 402         isize = ip->i_d.di_size;
 403         aeof = (offset + count) > isize;
 404
 405         if (io->io_new_size > isize)
 406                 isize = io->io_new_size;
 407
 408         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 409         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 410         count_fsb = last_fsb - offset_fsb;
 411         if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
 412                 xfs_fileoff_t   map_last_fsb;
 413
 414                 map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
 415
 416                 if (map_last_fsb < last_fsb) {
 417                         last_fsb = map_last_fsb;
 418                         count_fsb = last_fsb - offset_fsb;
 419                 }
 420                 ASSERT(count_fsb > 0);
 421         }
 422
 423         /*
 424          * determine if reserving space on
 425          * the data or realtime partition.
 426          */
 427         if ((rt = XFS_IS_REALTIME_INODE(ip))) {
 428                 int     sbrtextsize, iprtextsize;
 429
 430                 sbrtextsize = mp->m_sb.sb_rextsize;
 431                 iprtextsize =
 432                         ip->i_d.di_extsize ? ip->i_d.di_extsize : sbrtextsize;
 433                 numrtextents = (count_fsb + iprtextsize - 1);
 434                 do_div(numrtextents, sbrtextsize);
 435                 datablocks = 0;
 436         } else {
 437                 datablocks = count_fsb;
 438                 numrtextents = 0;
 439         }
 440
 441         /*
 442          * allocate and setup the transaction
 443          */
 444         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 445         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
 446
 447         resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
 448
 449         error = xfs_trans_reserve(tp, resblks,
 450                         XFS_WRITE_LOG_RES(mp), numrtextents,
 451                         XFS_TRANS_PERM_LOG_RES,
 452                         XFS_WRITE_LOG_COUNT);
 453
 454         /*
 455          * check for running out of space
 456          */
 457         if (error)
 458                 /*
 459                  * Free the transaction structure.
 460                  */
 461                 xfs_trans_cancel(tp, 0);
 462
 463         xfs_ilock(ip, XFS_ILOCK_EXCL);
 464
 465         if (error)
 466                 goto error_out; /* Don't return in above if .. trans ..,
 467                                         need lock to return */
 468
 469         if (XFS_TRANS_RESERVE_BLKQUOTA(mp, tp, ip, resblks)) {
 470                 error = (EDQUOT);
 471                 goto error1;
 472         }
 473         nimaps = 1;
 474
 475         bmapi_flag = XFS_BMAPI_WRITE;
 476         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 477         xfs_trans_ihold(tp, ip);
 478
 479         if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
 480                 bmapi_flag |= XFS_BMAPI_PREALLOC;
 481
 482         /*
 483          * issue the bmapi() call to allocate the blocks
 484          */
 485         XFS_BMAP_INIT(&free_list, &firstfsb);
 486         imapp = &imap[0];
 487         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 488                 bmapi_flag, &firstfsb, 0, imapp, &nimaps, &free_list);
 489         if (error) {
 490                 goto error0;
 491         }
 492
 493         /*
 494          * complete the transaction
 495          */
 496
 497         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
 498         if (error) {
 499                 goto error0;
 500         }
 501
 502         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 503         if (error) {
 504                 goto error_out;
 505         }
 506
 507         /* copy any maps to caller's array and return any error. */
 508         if (nimaps == 0) {
 509                 error = (ENOSPC);
 510                 goto error_out;
 511         }
 512
 513         *ret_imap = imap[0];
 514         *nmaps = 1;
 515         return 0;
 516
 517  error0:        /* Cancel bmap, unlock inode, and cancel trans */
 518         xfs_bmap_cancel(&free_list);
 519
 520  error1:        /* Just cancel transaction */
 521         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 522         *nmaps = 0;     /* nothing set-up here */
 523
 524 error_out:
 525         return XFS_ERROR(error);
 526 }
 527
 528 int
 529 xfs_iomap_write_delay(
 530         xfs_inode_t     *ip,
 531         loff_t          offset,
 532         size_t          count,
 533         int             ioflag,
 534         xfs_bmbt_irec_t *ret_imap,
 535         int             *nmaps)
 536 {
 537         xfs_mount_t     *mp = ip->i_mount;
 538         xfs_iocore_t    *io = &ip->i_iocore;
 539         xfs_fileoff_t   offset_fsb;
 540         xfs_fileoff_t   last_fsb;
 541         xfs_fsize_t     isize;
 542         xfs_fsblock_t   firstblock;
 543         int             nimaps;
 544         int             error;
 545         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
 546         int             aeof;
 547         int             fsynced = 0;
 548
 549         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
 550
 551         /*
 552          * Make sure that the dquots are there. This doesn't hold
 553          * the ilock across a disk read.
 554          */
 555
 556         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
 557         if (error)
 558                 return XFS_ERROR(error);
 559
 560 retry:
 561         isize = ip->i_d.di_size;
 562         if (io->io_new_size > isize) {
 563                 isize = io->io_new_size;
 564         }
 565
 566         aeof = 0;
 567         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 568         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 569         /*
 570          * If the caller is doing a write at the end of the file,
 571          * then extend the allocation (and the buffer used for the write)
 572          * out to the file system's write iosize.  We clean up any extra
 573          * space left over when the file is closed in xfs_inactive().
 574          *
 575          * We don't bother with this for sync writes, because we need
 576          * to minimize the amount we write for good performance.
 577          */
 578         if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
 579                 xfs_off_t       aligned_offset;
 580                 xfs_filblks_t   count_fsb;
 581                 unsigned int    iosize;
 582                 xfs_fileoff_t   ioalign;
 583                 int             n;
 584                 xfs_fileoff_t   start_fsb;
 585
 586                 /*
 587                  * If there are any real blocks past eof, then don't
 588                  * do any speculative allocation.
 589                  */
 590                 start_fsb = XFS_B_TO_FSBT(mp,
 591                                         ((xfs_ufsize_t)(offset + count - 1)));
 592                 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
 593                 while (count_fsb > 0) {
 594                         nimaps = XFS_WRITE_IMAPS;
 595                         error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
 596                                         0, &firstblock, 0, imap, &nimaps, NULL);
 597                         if (error) {
 598                                 return error;
 599                         }
 600                         for (n = 0; n < nimaps; n++) {
 601                                 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
 602                                     (imap[n].br_startblock != DELAYSTARTBLOCK)) {
 603                                         goto write_map;
 604                                 }
 605                                 start_fsb += imap[n].br_blockcount;
 606                                 count_fsb -= imap[n].br_blockcount;
 607                         }
 608                 }
 609                 iosize = mp->m_writeio_blocks;
 610                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 611                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
 612                 last_fsb = ioalign + iosize;
 613                 aeof = 1;
 614         }
 615 write_map:
 616         nimaps = XFS_WRITE_IMAPS;
 617         firstblock = NULLFSBLOCK;
 618
 619         /*
 620          * If mounted with the "-o swalloc" option, roundup the allocation
 621          * request to a stripe width boundary if the file size is >=
 622          * stripe width and we are allocating past the allocation eof.
 623          */
 624         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)
 625             && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
 626                 int eof;
 627                 xfs_fileoff_t new_last_fsb;
 628
 629                 new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
 630                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 631                 if (error) {
 632                         return error;
 633                 }
 634                 if (eof) {
 635                         last_fsb = new_last_fsb;
 636                 }
 637         /*
 638          * Roundup the allocation request to a stripe unit (m_dalign) boundary
 639          * if the file size is >= stripe unit size, and we are allocating past
 640          * the allocation eof.
 641          */
 642         } else if (mp->m_dalign && (isize >= XFS_FSB_TO_B(mp, mp->m_dalign))
 643                    && aeof) {
 644                 int eof;
 645                 xfs_fileoff_t new_last_fsb;
 646                 new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
 647                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 648                 if (error) {
 649                         return error;
 650                 }
 651                 if (eof) {
 652                         last_fsb = new_last_fsb;
 653                 }
 654         }
 655
 656         error = xfs_bmapi(NULL, ip, offset_fsb,
 657                           (xfs_filblks_t)(last_fsb - offset_fsb),
 658                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
 659                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
 660                           &nimaps, NULL);
 661         /*
 662          * This can be EDQUOT, if nimaps == 0
 663          */
 664         if (error && (error != ENOSPC)) {
 665                 return XFS_ERROR(error);
 666         }
 667         /*
 668          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
 669          * then we must have run out of space.
 670          */
 671         if (nimaps == 0) {
 672                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
 673                                         io, offset, count);
 674                 if (xfs_flush_space(ip, &fsynced, &ioflag))
 675                         return XFS_ERROR(ENOSPC);
 676
 677                 error = 0;
 678                 goto retry;
 679         }
 680
 681         *ret_imap = imap[0];
 682         *nmaps = 1;
 683         return 0;
 684 }
 685
 686 /*
 687  * Pass in a delayed allocate extent, convert it to real extents;
 688  * return to the caller the extent we create which maps on top of
 689  * the originating callers request.
 690  *
 691  * Called without a lock on the inode.
 692  */
 693 int
 694 xfs_iomap_write_allocate(
 695         xfs_inode_t     *ip,
 696         xfs_bmbt_irec_t *map,
 697         int             *retmap)
 698 {
 699         xfs_mount_t     *mp = ip->i_mount;
 700         xfs_fileoff_t   offset_fsb, last_block;
 701         xfs_fileoff_t   end_fsb, map_start_fsb;
 702         xfs_fsblock_t   first_block;
 703         xfs_bmap_free_t free_list;
 704         xfs_filblks_t   count_fsb;
 705         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
 706         xfs_trans_t     *tp;
 707         int             i, nimaps, committed;
 708         int             error = 0;
 709         int             nres;
 710
 711         *retmap = 0;
 712
 713         /*
 714          * Make sure that the dquots are there.
 715          */
 716         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
 717                 return XFS_ERROR(error);
 718
 719         offset_fsb = map->br_startoff;
 720         count_fsb = map->br_blockcount;
 721         map_start_fsb = offset_fsb;
 722
 723         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 724
 725         while (count_fsb != 0) {
 726                 /*
 727                  * Set up a transaction with which to allocate the
 728                  * backing store for the file.  Do allocations in a
 729                  * loop until we get some space in the range we are
 730                  * interested in.  The other space that might be allocated
 731                  * is in the delayed allocation extent on which we sit
 732                  * but before our buffer starts.
 733                  */
 734
 735                 nimaps = 0;
 736                 while (nimaps == 0) {
 737                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 738                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 739                         error = xfs_trans_reserve(tp, nres,
 740                                         XFS_WRITE_LOG_RES(mp),
 741                                         0, XFS_TRANS_PERM_LOG_RES,
 742                                         XFS_WRITE_LOG_COUNT);
 743                         if (error == ENOSPC) {
 744                                 error = xfs_trans_reserve(tp, 0,
 745                                                 XFS_WRITE_LOG_RES(mp),
 746                                                 0,
 747                                                 XFS_TRANS_PERM_LOG_RES,
 748                                                 XFS_WRITE_LOG_COUNT);
 749                         }
 750                         if (error) {
 751                                 xfs_trans_cancel(tp, 0);
 752                                 return XFS_ERROR(error);
 753                         }
 754                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 755                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 756                         xfs_trans_ihold(tp, ip);
 757
 758                         XFS_BMAP_INIT(&free_list, &first_block);
 759
 760                         nimaps = XFS_STRAT_WRITE_IMAPS;
 761                         /*
 762                          * Ensure we don't go beyond eof - it is possible
 763                          * the extents changed since we did the read call,
 764                          * we dropped the ilock in the interim.
 765                          */
 766
 767                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
 768                         xfs_bmap_last_offset(NULL, ip, &last_block,
 769                                 XFS_DATA_FORK);
 770                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 771                         if ((map_start_fsb + count_fsb) > last_block) {
 772                                 count_fsb = last_block - map_start_fsb;
 773                                 if (count_fsb == 0) {
 774                                         error = EAGAIN;
 775                                         goto trans_cancel;
 776                                 }
 777                         }
 778
 779                         /* Go get the actual blocks */
 780                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
 781                                         XFS_BMAPI_WRITE, &first_block, 1,
 782                                         imap, &nimaps, &free_list);
 783                         if (error)
 784                                 goto trans_cancel;
 785
 786                         error = xfs_bmap_finish(&tp, &free_list,
 787                                         first_block, &committed);
 788                         if (error)
 789                                 goto trans_cancel;
 790
 791                         error = xfs_trans_commit(tp,
 792                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
 793                         if (error)
 794                                 goto error0;
 795
 796                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 797                 }
 798
 799                 /*
 800                  * See if we were able to allocate an extent that
 801                  * covers at least part of the callers request
 802                  */
 803
 804                 for (i = 0; i < nimaps; i++) {
 805                         if ((map->br_startoff >= imap[i].br_startoff) &&
 806                             (map->br_startoff < (imap[i].br_startoff +
 807                                                  imap[i].br_blockcount))) {
 808                                 *map = imap[i];
 809                                 *retmap = 1;
 810                                 XFS_STATS_INC(xs_xstrat_quick);
 811                                 return 0;
 812                         }
 813                         count_fsb -= imap[i].br_blockcount;
 814                 }
 815
 816                 /* So far we have not mapped the requested part of the
 817                  * file, just surrounding data, try again.
 818                  */
 819                 nimaps--;
 820                 offset_fsb = imap[nimaps].br_startoff +
 821                              imap[nimaps].br_blockcount;
 822                 map_start_fsb = offset_fsb;
 823         }
 824
 825 trans_cancel:
 826         xfs_bmap_cancel(&free_list);
 827         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 828 error0:
 829         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 830         return XFS_ERROR(error);
 831 }
 832
 833 int
 834 xfs_iomap_write_unwritten(
 835         xfs_inode_t     *ip,
 836         loff_t          offset,
 837         size_t          count)
 838 {
 839         xfs_mount_t     *mp = ip->i_mount;
 840         xfs_trans_t     *tp;
 841         xfs_fileoff_t   offset_fsb;
 842         xfs_filblks_t   count_fsb;
 843         xfs_filblks_t   numblks_fsb;
 844         xfs_bmbt_irec_t imap;
 845         int             committed;
 846         int             error;
 847         int             nres;
 848         int             nimaps;
 849         xfs_fsblock_t   firstfsb;
 850         xfs_bmap_free_t free_list;
 851
 852         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
 853                                 &ip->i_iocore, offset, count);
 854
 855         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 856         count_fsb = XFS_B_TO_FSB(mp, count);
 857
 858         do {
 859                 nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 860
 861                 /*
 862                  * set up a transaction to convert the range of extents
 863                  * from unwritten to real. Do allocations in a loop until
 864                  * we have covered the range passed in.
 865                  */
 866
 867                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 868                 error = xfs_trans_reserve(tp, nres,
 869                                 XFS_WRITE_LOG_RES(mp), 0,
 870                                 XFS_TRANS_PERM_LOG_RES,
 871                                 XFS_WRITE_LOG_COUNT);
 872                 if (error) {
 873                         xfs_trans_cancel(tp, 0);
 874                         goto error0;
 875                 }
 876
 877                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 878                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 879                 xfs_trans_ihold(tp, ip);
 880
 881                 /*
 882                  * Modify the unwritten extent state of the buffer.
 883                  */
 884                 XFS_BMAP_INIT(&free_list, &firstfsb);
 885                 nimaps = 1;
 886                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 887                                   XFS_BMAPI_WRITE, &firstfsb,
 888                                   1, &imap, &nimaps, &free_list);
 889                 if (error)
 890                         goto error_on_bmapi_transaction;
 891
 892                 error = xfs_bmap_finish(&(tp), &(free_list),
 893                                 firstfsb, &committed);
 894                 if (error)
 895                         goto error_on_bmapi_transaction;
 896
 897                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 898                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 899                 if (error)
 900                         goto error0;
 901
 902                 if ((numblks_fsb = imap.br_blockcount) == 0) {
 903                         /*
 904                          * The numblks_fsb value should always get
 905                          * smaller, otherwise the loop is stuck.
 906                          */
 907                         ASSERT(imap.br_blockcount);
 908                         break;
 909                 }
 910                 offset_fsb += numblks_fsb;
 911                 count_fsb -= numblks_fsb;
 912         } while (count_fsb > 0);
 913
 914         return 0;
 915
 916 error_on_bmapi_transaction:
 917         xfs_bmap_cancel(&free_list);
 918         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
 919         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 920 error0:
 921         return XFS_ERROR(error);
 922 }