git.oblomov.eu Git - linux-2.6/blob - 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_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
 169                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
 170                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
 171                 iomapp->iomap_flags = flags;
 172
 173                 if (io->io_flags & XFS_IOCORE_RT) {
 174                         iomapp->iomap_flags |= IOMAP_REALTIME;
 175                         iomapp->iomap_target = mp->m_rtdev_targp;
 176                 } else {
 177                         iomapp->iomap_target = mp->m_ddev_targp;
 178                 }
 179                 start_block = imap->br_startblock;
 180                 if (start_block == HOLESTARTBLOCK) {
 181                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 182                         iomapp->iomap_flags |= IOMAP_HOLE;
 183                 } else if (start_block == DELAYSTARTBLOCK) {
 184                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 185                         iomapp->iomap_flags |= IOMAP_DELAY;
 186                 } else {
 187                         iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
 188                         if (ISUNWRITTEN(imap))
 189                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
 190                 }
 191
 192                 if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
 193                         iomapp->iomap_flags |= IOMAP_EOF;
 194                 }
 195
 196                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
 197         }
 198         return pbm;     /* Return the number filled */
 199 }
 200
 201 int
 202 xfs_iomap(
 203         xfs_iocore_t    *io,
 204         xfs_off_t       offset,
 205         ssize_t         count,
 206         int             flags,
 207         xfs_iomap_t     *iomapp,
 208         int             *niomaps)
 209 {
 210         xfs_mount_t     *mp = io->io_mount;
 211         xfs_fileoff_t   offset_fsb, end_fsb;
 212         int             error = 0;
 213         int             lockmode = 0;
 214         xfs_bmbt_irec_t imap;
 215         int             nimaps = 1;
 216         int             bmapi_flags = 0;
 217         int             iomap_flags = 0;
 218
 219         if (XFS_FORCED_SHUTDOWN(mp))
 220                 return XFS_ERROR(EIO);
 221
 222         switch (flags &
 223                 (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
 224                  BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
 225         case BMAPI_READ:
 226                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
 227                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 228                 bmapi_flags = XFS_BMAPI_ENTIRE;
 229                 if (flags & BMAPI_IGNSTATE)
 230                         bmapi_flags |= XFS_BMAPI_IGSTATE;
 231                 break;
 232         case BMAPI_WRITE:
 233                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
 234                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
 235                 bmapi_flags = 0;
 236                 XFS_ILOCK(mp, io, lockmode);
 237                 break;
 238         case BMAPI_ALLOCATE:
 239                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
 240                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
 241                 bmapi_flags = XFS_BMAPI_ENTIRE;
 242                 /* Attempt non-blocking lock */
 243                 if (flags & BMAPI_TRYLOCK) {
 244                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
 245                                 return XFS_ERROR(EAGAIN);
 246                 } else {
 247                         XFS_ILOCK(mp, io, lockmode);
 248                 }
 249                 break;
 250         case BMAPI_UNWRITTEN:
 251                 goto phase2;
 252         case BMAPI_DEVICE:
 253                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 254                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
 255                         mp->m_rtdev_targp : mp->m_ddev_targp;
 256                 error = 0;
 257                 *niomaps = 1;
 258                 goto out;
 259         default:
 260                 BUG();
 261         }
 262
 263         ASSERT(offset <= mp->m_maxioffset);
 264         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
 265                 count = mp->m_maxioffset - offset;
 266         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 267         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 268
 269         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
 270                         (xfs_filblks_t)(end_fsb - offset_fsb),
 271                         bmapi_flags,  NULL, 0, &imap,
 272                         &nimaps, NULL);
 273
 274         if (error)
 275                 goto out;
 276
 277 phase2:
 278         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
 279         case BMAPI_WRITE:
 280                 /* If we found an extent, return it */
 281                 if (nimaps &&
 282                     (imap.br_startblock != HOLESTARTBLOCK) &&
 283                     (imap.br_startblock != DELAYSTARTBLOCK)) {
 284                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 285                                         offset, count, iomapp, &imap, flags);
 286                         break;
 287                 }
 288
 289                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
 290                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
 291                                         count, flags, &imap, &nimaps, nimaps);
 292                 } else {
 293                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
 294                                         flags, &imap, &nimaps);
 295                 }
 296                 if (!error) {
 297                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
 298                                         offset, count, iomapp, &imap, flags);
 299                 }
 300                 iomap_flags = IOMAP_NEW;
 301                 break;
 302         case BMAPI_ALLOCATE:
 303                 /* If we found an extent, return it */
 304                 XFS_IUNLOCK(mp, io, lockmode);
 305                 lockmode = 0;
 306
 307                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
 308                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 309                                         offset, count, iomapp, &imap, flags);
 310                         break;
 311                 }
 312
 313                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
 314                                                  &imap, &nimaps);
 315                 break;
 316         case BMAPI_UNWRITTEN:
 317                 lockmode = 0;
 318                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
 319                 nimaps = 0;
 320                 break;
 321         }
 322
 323         if (nimaps) {
 324                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
 325                                             iomapp, nimaps, *niomaps, iomap_flags);
 326         } else if (niomaps) {
 327                 *niomaps = 0;
 328         }
 329
 330 out:
 331         if (lockmode)
 332                 XFS_IUNLOCK(mp, io, lockmode);
 333         return XFS_ERROR(error);
 334 }
 335
 336 STATIC int
 337 xfs_flush_space(
 338         xfs_inode_t     *ip,
 339         int             *fsynced,
 340         int             *ioflags)
 341 {
 342         switch (*fsynced) {
 343         case 0:
 344                 if (ip->i_delayed_blks) {
 345                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 346                         xfs_flush_inode(ip);
 347                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 348                         *fsynced = 1;
 349                 } else {
 350                         *ioflags |= BMAPI_SYNC;
 351                         *fsynced = 2;
 352                 }
 353                 return 0;
 354         case 1:
 355                 *fsynced = 2;
 356                 *ioflags |= BMAPI_SYNC;
 357                 return 0;
 358         case 2:
 359                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 360                 xfs_flush_device(ip);
 361                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 362                 *fsynced = 3;
 363                 return 0;
 364         }
 365         return 1;
 366 }
 367
 368 int
 369 xfs_iomap_write_direct(
 370         xfs_inode_t     *ip,
 371         xfs_off_t       offset,
 372         size_t          count,
 373         int             flags,
 374         xfs_bmbt_irec_t *ret_imap,
 375         int             *nmaps,
 376         int             found)
 377 {
 378         xfs_mount_t     *mp = ip->i_mount;
 379         xfs_iocore_t    *io = &ip->i_iocore;
 380         xfs_fileoff_t   offset_fsb;
 381         xfs_fileoff_t   last_fsb;
 382         xfs_filblks_t   count_fsb;
 383         xfs_fsize_t     isize;
 384         xfs_fsblock_t   firstfsb;
 385         int             nimaps, maps;
 386         int             error;
 387         int             bmapi_flag;
 388         int             quota_flag;
 389         int             rt;
 390         xfs_trans_t     *tp;
 391         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS], *imapp;
 392         xfs_bmap_free_t free_list;
 393         int             aeof;
 394         xfs_filblks_t   qblocks, resblks;
 395         int             committed;
 396         int             resrtextents;
 397
 398         /*
 399          * Make sure that the dquots are there. This doesn't hold
 400          * the ilock across a disk read.
 401          */
 402         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
 403         if (error)
 404                 return XFS_ERROR(error);
 405
 406         maps = min(XFS_WRITE_IMAPS, *nmaps);
 407         nimaps = maps;
 408
 409         isize = ip->i_d.di_size;
 410         aeof = (offset + count) > isize;
 411
 412         if (io->io_new_size > isize)
 413                 isize = io->io_new_size;
 414
 415         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 416         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 417         count_fsb = last_fsb - offset_fsb;
 418         if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
 419                 xfs_fileoff_t   map_last_fsb;
 420
 421                 map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
 422                 if (map_last_fsb < last_fsb) {
 423                         last_fsb = map_last_fsb;
 424                         count_fsb = last_fsb - offset_fsb;
 425                 }
 426                 ASSERT(count_fsb > 0);
 427         }
 428
 429         /*
 430          * Determine if reserving space on the data or realtime partition.
 431          */
 432         if ((rt = XFS_IS_REALTIME_INODE(ip))) {
 433                 xfs_extlen_t    extsz;
 434
 435                 if (!(extsz = ip->i_d.di_extsize))
 436                         extsz = mp->m_sb.sb_rextsize;
 437                 resrtextents = qblocks = (count_fsb + extsz - 1);
 438                 do_div(resrtextents, mp->m_sb.sb_rextsize);
 439                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 440                 quota_flag = XFS_QMOPT_RES_RTBLKS;
 441         } else {
 442                 resrtextents = 0;
 443                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, count_fsb);
 444                 quota_flag = XFS_QMOPT_RES_REGBLKS;
 445         }
 446
 447         /*
 448          * Allocate and setup the transaction
 449          */
 450         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 451         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
 452         error = xfs_trans_reserve(tp, resblks,
 453                         XFS_WRITE_LOG_RES(mp), resrtextents,
 454                         XFS_TRANS_PERM_LOG_RES,
 455                         XFS_WRITE_LOG_COUNT);
 456
 457         /*
 458          * Check for running out of space, note: need lock to return
 459          */
 460         if (error)
 461                 xfs_trans_cancel(tp, 0);
 462         xfs_ilock(ip, XFS_ILOCK_EXCL);
 463         if (error)
 464                 goto error_out;
 465
 466         if (XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag)) {
 467                 error = (EDQUOT);
 468                 goto error1;
 469         }
 470
 471         bmapi_flag = XFS_BMAPI_WRITE;
 472         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 473         xfs_trans_ihold(tp, ip);
 474
 475         if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
 476                 bmapi_flag |= XFS_BMAPI_PREALLOC;
 477
 478         /*
 479          * Issue the bmapi() call to allocate the blocks
 480          */
 481         XFS_BMAP_INIT(&free_list, &firstfsb);
 482         nimaps = 1;
 483         imapp = &imap[0];
 484         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 485                 bmapi_flag, &firstfsb, 0, imapp, &nimaps, &free_list);
 486         if (error)
 487                 goto error0;
 488
 489         /*
 490          * Complete the transaction
 491          */
 492         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
 493         if (error)
 494                 goto error0;
 495         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 496         if (error)
 497                 goto error_out;
 498
 499         /*
 500          * Copy any maps to caller's array and return any error.
 501          */
 502         if (nimaps == 0) {
 503                 error = (ENOSPC);
 504                 goto error_out;
 505         }
 506
 507         *ret_imap = imap[0];
 508         *nmaps = 1;
 509         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 510                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 511                         "start_block : %llx start_off : %llx blkcnt : %llx "
 512                         "extent-state : %x \n",
 513                         (ip->i_mount)->m_fsname,
 514                         (long long)ip->i_ino,
 515                         ret_imap->br_startblock, ret_imap->br_startoff,
 516                         ret_imap->br_blockcount,ret_imap->br_state);
 517         }
 518         return 0;
 519
 520 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
 521         xfs_bmap_cancel(&free_list);
 522         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
 523
 524 error1: /* Just cancel transaction */
 525         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 526         *nmaps = 0;     /* nothing set-up here */
 527
 528 error_out:
 529         return XFS_ERROR(error);
 530 }
 531
 532 int
 533 xfs_iomap_write_delay(
 534         xfs_inode_t     *ip,
 535         xfs_off_t       offset,
 536         size_t          count,
 537         int             ioflag,
 538         xfs_bmbt_irec_t *ret_imap,
 539         int             *nmaps)
 540 {
 541         xfs_mount_t     *mp = ip->i_mount;
 542         xfs_iocore_t    *io = &ip->i_iocore;
 543         xfs_fileoff_t   offset_fsb;
 544         xfs_fileoff_t   last_fsb;
 545         xfs_fsize_t     isize;
 546         xfs_fsblock_t   firstblock;
 547         int             nimaps;
 548         int             error;
 549         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
 550         int             aeof;
 551         int             fsynced = 0;
 552
 553         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
 554
 555         /*
 556          * Make sure that the dquots are there. This doesn't hold
 557          * the ilock across a disk read.
 558          */
 559
 560         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
 561         if (error)
 562                 return XFS_ERROR(error);
 563
 564 retry:
 565         isize = ip->i_d.di_size;
 566         if (io->io_new_size > isize) {
 567                 isize = io->io_new_size;
 568         }
 569
 570         aeof = 0;
 571         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 572         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 573         /*
 574          * If the caller is doing a write at the end of the file,
 575          * then extend the allocation (and the buffer used for the write)
 576          * out to the file system's write iosize.  We clean up any extra
 577          * space left over when the file is closed in xfs_inactive().
 578          *
 579          * For sync writes, we are flushing delayed allocate space to
 580          * try to make additional space available for allocation near
 581          * the filesystem full boundary - preallocation hurts in that
 582          * situation, of course.
 583          */
 584         if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
 585                 xfs_off_t       aligned_offset;
 586                 xfs_filblks_t   count_fsb;
 587                 unsigned int    iosize;
 588                 xfs_fileoff_t   ioalign;
 589                 int             n;
 590                 xfs_fileoff_t   start_fsb;
 591
 592                 /*
 593                  * If there are any real blocks past eof, then don't
 594                  * do any speculative allocation.
 595                  */
 596                 start_fsb = XFS_B_TO_FSBT(mp,
 597                                         ((xfs_ufsize_t)(offset + count - 1)));
 598                 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
 599                 while (count_fsb > 0) {
 600                         nimaps = XFS_WRITE_IMAPS;
 601                         error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
 602                                         0, &firstblock, 0, imap, &nimaps, NULL);
 603                         if (error) {
 604                                 return error;
 605                         }
 606                         for (n = 0; n < nimaps; n++) {
 607                                 if ( !(io->io_flags & XFS_IOCORE_RT)  &&
 608                                         !imap[n].br_startblock) {
 609                                         cmn_err(CE_PANIC,"Access to block "
 610                                                 "zero:  fs <%s> inode: %lld "
 611                                                 "start_block : %llx start_off "
 612                                                 ": %llx blkcnt : %llx "
 613                                                 "extent-state : %x \n",
 614                                                 (ip->i_mount)->m_fsname,
 615                                                 (long long)ip->i_ino,
 616                                                 imap[n].br_startblock,
 617                                                 imap[n].br_startoff,
 618                                                 imap[n].br_blockcount,
 619                                                 imap[n].br_state);
 620                                 }
 621                                 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
 622                                     (imap[n].br_startblock != DELAYSTARTBLOCK)) {
 623                                         goto write_map;
 624                                 }
 625                                 start_fsb += imap[n].br_blockcount;
 626                                 count_fsb -= imap[n].br_blockcount;
 627                         }
 628                 }
 629                 iosize = mp->m_writeio_blocks;
 630                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 631                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
 632                 last_fsb = ioalign + iosize;
 633                 aeof = 1;
 634         }
 635 write_map:
 636         nimaps = XFS_WRITE_IMAPS;
 637         firstblock = NULLFSBLOCK;
 638
 639         /*
 640          * If mounted with the "-o swalloc" option, roundup the allocation
 641          * request to a stripe width boundary if the file size is >=
 642          * stripe width and we are allocating past the allocation eof.
 643          */
 644         if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_swidth
 645             && (mp->m_flags & XFS_MOUNT_SWALLOC)
 646             && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
 647                 int eof;
 648                 xfs_fileoff_t new_last_fsb;
 649
 650                 new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
 651                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 652                 if (error) {
 653                         return error;
 654                 }
 655                 if (eof) {
 656                         last_fsb = new_last_fsb;
 657                 }
 658         /*
 659          * Roundup the allocation request to a stripe unit (m_dalign) boundary
 660          * if the file size is >= stripe unit size, and we are allocating past
 661          * the allocation eof.
 662          */
 663         } else if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_dalign &&
 664                    (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)) && aeof) {
 665                 int eof;
 666                 xfs_fileoff_t new_last_fsb;
 667                 new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
 668                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 669                 if (error) {
 670                         return error;
 671                 }
 672                 if (eof) {
 673                         last_fsb = new_last_fsb;
 674                 }
 675         /*
 676          * Round up the allocation request to a real-time extent boundary
 677          * if the file is on the real-time subvolume.
 678          */
 679         } else if (io->io_flags & XFS_IOCORE_RT && aeof) {
 680                 int eof;
 681                 xfs_fileoff_t new_last_fsb;
 682
 683                 new_last_fsb = roundup_64(last_fsb, mp->m_sb.sb_rextsize);
 684                 error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
 685                 if (error) {
 686                         return error;
 687                 }
 688                 if (eof)
 689                         last_fsb = new_last_fsb;
 690         }
 691         error = xfs_bmapi(NULL, ip, offset_fsb,
 692                           (xfs_filblks_t)(last_fsb - offset_fsb),
 693                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
 694                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
 695                           &nimaps, NULL);
 696         /*
 697          * This can be EDQUOT, if nimaps == 0
 698          */
 699         if (error && (error != ENOSPC)) {
 700                 return XFS_ERROR(error);
 701         }
 702         /*
 703          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
 704          * then we must have run out of space.
 705          */
 706         if (nimaps == 0) {
 707                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
 708                                         io, offset, count);
 709                 if (xfs_flush_space(ip, &fsynced, &ioflag))
 710                         return XFS_ERROR(ENOSPC);
 711
 712                 error = 0;
 713                 goto retry;
 714         }
 715
 716         *ret_imap = imap[0];
 717         *nmaps = 1;
 718         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 719                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 720                         "start_block : %llx start_off : %llx blkcnt : %llx "
 721                         "extent-state : %x \n",
 722                         (ip->i_mount)->m_fsname,
 723                         (long long)ip->i_ino,
 724                         ret_imap->br_startblock, ret_imap->br_startoff,
 725                         ret_imap->br_blockcount,ret_imap->br_state);
 726         }
 727         return 0;
 728 }
 729
 730 /*
 731  * Pass in a delayed allocate extent, convert it to real extents;
 732  * return to the caller the extent we create which maps on top of
 733  * the originating callers request.
 734  *
 735  * Called without a lock on the inode.
 736  */
 737 int
 738 xfs_iomap_write_allocate(
 739         xfs_inode_t     *ip,
 740         xfs_off_t       offset,
 741         size_t          count,
 742         xfs_bmbt_irec_t *map,
 743         int             *retmap)
 744 {
 745         xfs_mount_t     *mp = ip->i_mount;
 746         xfs_iocore_t    *io = &ip->i_iocore;
 747         xfs_fileoff_t   offset_fsb, last_block;
 748         xfs_fileoff_t   end_fsb, map_start_fsb;
 749         xfs_fsblock_t   first_block;
 750         xfs_bmap_free_t free_list;
 751         xfs_filblks_t   count_fsb;
 752         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
 753         xfs_trans_t     *tp;
 754         int             i, nimaps, committed;
 755         int             error = 0;
 756         int             nres;
 757
 758         *retmap = 0;
 759
 760         /*
 761          * Make sure that the dquots are there.
 762          */
 763         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
 764                 return XFS_ERROR(error);
 765
 766         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 767         count_fsb = map->br_blockcount;
 768         map_start_fsb = map->br_startoff;
 769
 770         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 771
 772         while (count_fsb != 0) {
 773                 /*
 774                  * Set up a transaction with which to allocate the
 775                  * backing store for the file.  Do allocations in a
 776                  * loop until we get some space in the range we are
 777                  * interested in.  The other space that might be allocated
 778                  * is in the delayed allocation extent on which we sit
 779                  * but before our buffer starts.
 780                  */
 781
 782                 nimaps = 0;
 783                 while (nimaps == 0) {
 784                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 785                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 786                         error = xfs_trans_reserve(tp, nres,
 787                                         XFS_WRITE_LOG_RES(mp),
 788                                         0, XFS_TRANS_PERM_LOG_RES,
 789                                         XFS_WRITE_LOG_COUNT);
 790                         if (error == ENOSPC) {
 791                                 error = xfs_trans_reserve(tp, 0,
 792                                                 XFS_WRITE_LOG_RES(mp),
 793                                                 0,
 794                                                 XFS_TRANS_PERM_LOG_RES,
 795                                                 XFS_WRITE_LOG_COUNT);
 796                         }
 797                         if (error) {
 798                                 xfs_trans_cancel(tp, 0);
 799                                 return XFS_ERROR(error);
 800                         }
 801                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 802                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 803                         xfs_trans_ihold(tp, ip);
 804
 805                         XFS_BMAP_INIT(&free_list, &first_block);
 806
 807                         nimaps = XFS_STRAT_WRITE_IMAPS;
 808                         /*
 809                          * Ensure we don't go beyond eof - it is possible
 810                          * the extents changed since we did the read call,
 811                          * we dropped the ilock in the interim.
 812                          */
 813
 814                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
 815                         xfs_bmap_last_offset(NULL, ip, &last_block,
 816                                 XFS_DATA_FORK);
 817                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 818                         if ((map_start_fsb + count_fsb) > last_block) {
 819                                 count_fsb = last_block - map_start_fsb;
 820                                 if (count_fsb == 0) {
 821                                         error = EAGAIN;
 822                                         goto trans_cancel;
 823                                 }
 824                         }
 825
 826                         /* Go get the actual blocks */
 827                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
 828                                         XFS_BMAPI_WRITE, &first_block, 1,
 829                                         imap, &nimaps, &free_list);
 830                         if (error)
 831                                 goto trans_cancel;
 832
 833                         error = xfs_bmap_finish(&tp, &free_list,
 834                                         first_block, &committed);
 835                         if (error)
 836                                 goto trans_cancel;
 837
 838                         error = xfs_trans_commit(tp,
 839                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
 840                         if (error)
 841                                 goto error0;
 842
 843                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 844                 }
 845
 846                 /*
 847                  * See if we were able to allocate an extent that
 848                  * covers at least part of the callers request
 849                  */
 850
 851                 for (i = 0; i < nimaps; i++) {
 852                         if ( !(io->io_flags & XFS_IOCORE_RT)  &&
 853                                 !imap[i].br_startblock) {
 854                                 cmn_err(CE_PANIC,"Access to block zero:  "
 855                                         "fs <%s> inode: %lld "
 856                                         "start_block : %llx start_off : %llx "
 857                                         "blkcnt : %llx extent-state : %x \n",
 858                                         (ip->i_mount)->m_fsname,
 859                                         (long long)ip->i_ino,
 860                                         imap[i].br_startblock,
 861                                         imap[i].br_startoff,
 862                                         imap[i].br_blockcount,imap[i].br_state);
 863                         }
 864                         if ((offset_fsb >= imap[i].br_startoff) &&
 865                             (offset_fsb < (imap[i].br_startoff +
 866                                            imap[i].br_blockcount))) {
 867                                 *map = imap[i];
 868                                 *retmap = 1;
 869                                 XFS_STATS_INC(xs_xstrat_quick);
 870                                 return 0;
 871                         }
 872                         count_fsb -= imap[i].br_blockcount;
 873                 }
 874
 875                 /* So far we have not mapped the requested part of the
 876                  * file, just surrounding data, try again.
 877                  */
 878                 nimaps--;
 879                 map_start_fsb = imap[nimaps].br_startoff +
 880                                 imap[nimaps].br_blockcount;
 881         }
 882
 883 trans_cancel:
 884         xfs_bmap_cancel(&free_list);
 885         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 886 error0:
 887         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 888         return XFS_ERROR(error);
 889 }
 890
 891 int
 892 xfs_iomap_write_unwritten(
 893         xfs_inode_t     *ip,
 894         xfs_off_t       offset,
 895         size_t          count)
 896 {
 897         xfs_mount_t     *mp = ip->i_mount;
 898         xfs_iocore_t    *io = &ip->i_iocore;
 899         xfs_trans_t     *tp;
 900         xfs_fileoff_t   offset_fsb;
 901         xfs_filblks_t   count_fsb;
 902         xfs_filblks_t   numblks_fsb;
 903         xfs_bmbt_irec_t imap;
 904         int             committed;
 905         int             error;
 906         int             nres;
 907         int             nimaps;
 908         xfs_fsblock_t   firstfsb;
 909         xfs_bmap_free_t free_list;
 910
 911         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
 912                                 &ip->i_iocore, offset, count);
 913
 914         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 915         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 916         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 917
 918         do {
 919                 nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 920
 921                 /*
 922                  * set up a transaction to convert the range of extents
 923                  * from unwritten to real. Do allocations in a loop until
 924                  * we have covered the range passed in.
 925                  */
 926
 927                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 928                 error = xfs_trans_reserve(tp, nres,
 929                                 XFS_WRITE_LOG_RES(mp), 0,
 930                                 XFS_TRANS_PERM_LOG_RES,
 931                                 XFS_WRITE_LOG_COUNT);
 932                 if (error) {
 933                         xfs_trans_cancel(tp, 0);
 934                         goto error0;
 935                 }
 936
 937                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 938                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 939                 xfs_trans_ihold(tp, ip);
 940
 941                 /*
 942                  * Modify the unwritten extent state of the buffer.
 943                  */
 944                 XFS_BMAP_INIT(&free_list, &firstfsb);
 945                 nimaps = 1;
 946                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 947                                   XFS_BMAPI_WRITE, &firstfsb,
 948                                   1, &imap, &nimaps, &free_list);
 949                 if (error)
 950                         goto error_on_bmapi_transaction;
 951
 952                 error = xfs_bmap_finish(&(tp), &(free_list),
 953                                 firstfsb, &committed);
 954                 if (error)
 955                         goto error_on_bmapi_transaction;
 956
 957                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 958                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 959                 if (error)
 960                         goto error0;
 961
 962                 if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
 963                         cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
 964                                 "inode: %lld start_block : %llx start_off : "
 965                                 "%llx blkcnt : %llx extent-state : %x \n",
 966                                 (ip->i_mount)->m_fsname,
 967                                 (long long)ip->i_ino,
 968                                 imap.br_startblock,imap.br_startoff,
 969                                 imap.br_blockcount,imap.br_state);
 970                 }
 971
 972                 if ((numblks_fsb = imap.br_blockcount) == 0) {
 973                         /*
 974                          * The numblks_fsb value should always get
 975                          * smaller, otherwise the loop is stuck.
 976                          */
 977                         ASSERT(imap.br_blockcount);
 978                         break;
 979                 }
 980                 offset_fsb += numblks_fsb;
 981                 count_fsb -= numblks_fsb;
 982         } while (count_fsb > 0);
 983
 984         return 0;
 985
 986 error_on_bmapi_transaction:
 987         xfs_bmap_cancel(&free_list);
 988         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
 989         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 990 error0:
 991         return XFS_ERROR(error);
 992 }