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 #include "xfs.h"
  33 #include "xfs_fs.h"
  34 #include "xfs_bit.h"
  35 #include "xfs_log.h"
  36 #include "xfs_inum.h"
  37 #include "xfs_trans.h"
  38 #include "xfs_sb.h"
  39 #include "xfs_ag.h"
  40 #include "xfs_dir.h"
  41 #include "xfs_dir2.h"
  42 #include "xfs_alloc.h"
  43 #include "xfs_dmapi.h"
  44 #include "xfs_quota.h"
  45 #include "xfs_mount.h"
  46 #include "xfs_bmap_btree.h"
  47 #include "xfs_alloc_btree.h"
  48 #include "xfs_ialloc_btree.h"
  49 #include "xfs_dir_sf.h"
  50 #include "xfs_dir2_sf.h"
  51 #include "xfs_attr_sf.h"
  52 #include "xfs_dinode.h"
  53 #include "xfs_inode.h"
  54 #include "xfs_ialloc.h"
  55 #include "xfs_btree.h"
  56 #include "xfs_bmap.h"
  57 #include "xfs_bit.h"
  58 #include "xfs_rtalloc.h"
  59 #include "xfs_error.h"
  60 #include "xfs_itable.h"
  61 #include "xfs_rw.h"
  62 #include "xfs_acl.h"
  63 #include "xfs_cap.h"
  64 #include "xfs_mac.h"
  65 #include "xfs_attr.h"
  66 #include "xfs_buf_item.h"
  67 #include "xfs_trans_space.h"
  68 #include "xfs_utils.h"
  69 #include "xfs_iomap.h"
  70
  71 #if defined(XFS_RW_TRACE)
  72 void
  73 xfs_iomap_enter_trace(
  74         int             tag,
  75         xfs_iocore_t    *io,
  76         xfs_off_t       offset,
  77         ssize_t         count)
  78 {
  79         xfs_inode_t     *ip = XFS_IO_INODE(io);
  80
  81         if (!ip->i_rwtrace)
  82                 return;
  83
  84         ktrace_enter(ip->i_rwtrace,
  85                 (void *)((unsigned long)tag),
  86                 (void *)ip,
  87                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
  88                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
  89                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
  90                 (void *)((unsigned long)(offset & 0xffffffff)),
  91                 (void *)((unsigned long)count),
  92                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
  93                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
  94                 (void *)NULL,
  95                 (void *)NULL,
  96                 (void *)NULL,
  97                 (void *)NULL,
  98                 (void *)NULL,
  99                 (void *)NULL,
 100                 (void *)NULL);
 101 }
 102
 103 void
 104 xfs_iomap_map_trace(
 105         int             tag,
 106         xfs_iocore_t    *io,
 107         xfs_off_t       offset,
 108         ssize_t         count,
 109         xfs_iomap_t     *iomapp,
 110         xfs_bmbt_irec_t *imapp,
 111         int             flags)
 112 {
 113         xfs_inode_t     *ip = XFS_IO_INODE(io);
 114
 115         if (!ip->i_rwtrace)
 116                 return;
 117
 118         ktrace_enter(ip->i_rwtrace,
 119                 (void *)((unsigned long)tag),
 120                 (void *)ip,
 121                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
 122                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
 123                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
 124                 (void *)((unsigned long)(offset & 0xffffffff)),
 125                 (void *)((unsigned long)count),
 126                 (void *)((unsigned long)flags),
 127                 (void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
 128                 (void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
 129                 (void *)((unsigned long)(iomapp->iomap_delta)),
 130                 (void *)((unsigned long)(iomapp->iomap_bsize)),
 131                 (void *)((unsigned long)(iomapp->iomap_bn)),
 132                 (void *)(__psint_t)(imapp->br_startoff),
 133                 (void *)((unsigned long)(imapp->br_blockcount)),
 134                 (void *)(__psint_t)(imapp->br_startblock));
 135 }
 136 #else
 137 #define xfs_iomap_enter_trace(tag, io, offset, count)
 138 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
 139 #endif
 140
 141 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
 142                                                 << mp->m_writeio_log)
 143 #define XFS_STRAT_WRITE_IMAPS   2
 144 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
 145
 146 STATIC int
 147 xfs_imap_to_bmap(
 148         xfs_iocore_t    *io,
 149         xfs_off_t       offset,
 150         xfs_bmbt_irec_t *imap,
 151         xfs_iomap_t     *iomapp,
 152         int             imaps,                  /* Number of imap entries */
 153         int             iomaps,                 /* Number of iomap entries */
 154         int             flags)
 155 {
 156         xfs_mount_t     *mp;
 157         xfs_fsize_t     nisize;
 158         int             pbm;
 159         xfs_fsblock_t   start_block;
 160
 161         mp = io->io_mount;
 162         nisize = XFS_SIZE(mp, io);
 163         if (io->io_new_size > nisize)
 164                 nisize = io->io_new_size;
 165
 166         for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
 167                 iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
 168                 iomapp->iomap_delta = offset - iomapp->iomap_offset;
 169                 iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
 170                 iomapp->iomap_flags = flags;
 171
 172                 if (io->io_flags & XFS_IOCORE_RT) {
 173                         iomapp->iomap_flags |= IOMAP_REALTIME;
 174                         iomapp->iomap_target = mp->m_rtdev_targp;
 175                 } else {
 176                         iomapp->iomap_target = mp->m_ddev_targp;
 177                 }
 178                 start_block = imap->br_startblock;
 179                 if (start_block == HOLESTARTBLOCK) {
 180                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 181                         iomapp->iomap_flags |= IOMAP_HOLE;
 182                 } else if (start_block == DELAYSTARTBLOCK) {
 183                         iomapp->iomap_bn = IOMAP_DADDR_NULL;
 184                         iomapp->iomap_flags |= IOMAP_DELAY;
 185                 } else {
 186                         iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
 187                         if (ISUNWRITTEN(imap))
 188                                 iomapp->iomap_flags |= IOMAP_UNWRITTEN;
 189                 }
 190
 191                 if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
 192                         iomapp->iomap_flags |= IOMAP_EOF;
 193                 }
 194
 195                 offset += iomapp->iomap_bsize - iomapp->iomap_delta;
 196         }
 197         return pbm;     /* Return the number filled */
 198 }
 199
 200 int
 201 xfs_iomap(
 202         xfs_iocore_t    *io,
 203         xfs_off_t       offset,
 204         ssize_t         count,
 205         int             flags,
 206         xfs_iomap_t     *iomapp,
 207         int             *niomaps)
 208 {
 209         xfs_mount_t     *mp = io->io_mount;
 210         xfs_fileoff_t   offset_fsb, end_fsb;
 211         int             error = 0;
 212         int             lockmode = 0;
 213         xfs_bmbt_irec_t imap;
 214         int             nimaps = 1;
 215         int             bmapi_flags = 0;
 216         int             iomap_flags = 0;
 217
 218         if (XFS_FORCED_SHUTDOWN(mp))
 219                 return XFS_ERROR(EIO);
 220
 221         switch (flags &
 222                 (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
 223                  BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
 224         case BMAPI_READ:
 225                 xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
 226                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 227                 bmapi_flags = XFS_BMAPI_ENTIRE;
 228                 break;
 229         case BMAPI_WRITE:
 230                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
 231                 lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
 232                 if (flags & BMAPI_IGNSTATE)
 233                         bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
 234                 XFS_ILOCK(mp, io, lockmode);
 235                 break;
 236         case BMAPI_ALLOCATE:
 237                 xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
 238                 lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
 239                 bmapi_flags = XFS_BMAPI_ENTIRE;
 240                 /* Attempt non-blocking lock */
 241                 if (flags & BMAPI_TRYLOCK) {
 242                         if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
 243                                 return XFS_ERROR(EAGAIN);
 244                 } else {
 245                         XFS_ILOCK(mp, io, lockmode);
 246                 }
 247                 break;
 248         case BMAPI_UNWRITTEN:
 249                 goto phase2;
 250         case BMAPI_DEVICE:
 251                 lockmode = XFS_LCK_MAP_SHARED(mp, io);
 252                 iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
 253                         mp->m_rtdev_targp : mp->m_ddev_targp;
 254                 error = 0;
 255                 *niomaps = 1;
 256                 goto out;
 257         default:
 258                 BUG();
 259         }
 260
 261         ASSERT(offset <= mp->m_maxioffset);
 262         if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
 263                 count = mp->m_maxioffset - offset;
 264         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 265         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 266
 267         error = XFS_BMAPI(mp, NULL, io, offset_fsb,
 268                         (xfs_filblks_t)(end_fsb - offset_fsb),
 269                         bmapi_flags,  NULL, 0, &imap,
 270                         &nimaps, NULL);
 271
 272         if (error)
 273                 goto out;
 274
 275 phase2:
 276         switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
 277         case BMAPI_WRITE:
 278                 /* If we found an extent, return it */
 279                 if (nimaps &&
 280                     (imap.br_startblock != HOLESTARTBLOCK) &&
 281                     (imap.br_startblock != DELAYSTARTBLOCK)) {
 282                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 283                                         offset, count, iomapp, &imap, flags);
 284                         break;
 285                 }
 286
 287                 if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
 288                         error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
 289                                         count, flags, &imap, &nimaps, nimaps);
 290                 } else {
 291                         error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
 292                                         flags, &imap, &nimaps);
 293                 }
 294                 if (!error) {
 295                         xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
 296                                         offset, count, iomapp, &imap, flags);
 297                 }
 298                 iomap_flags = IOMAP_NEW;
 299                 break;
 300         case BMAPI_ALLOCATE:
 301                 /* If we found an extent, return it */
 302                 XFS_IUNLOCK(mp, io, lockmode);
 303                 lockmode = 0;
 304
 305                 if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
 306                         xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
 307                                         offset, count, iomapp, &imap, flags);
 308                         break;
 309                 }
 310
 311                 error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
 312                                                  &imap, &nimaps);
 313                 break;
 314         case BMAPI_UNWRITTEN:
 315                 lockmode = 0;
 316                 error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
 317                 nimaps = 0;
 318                 break;
 319         }
 320
 321         if (nimaps) {
 322                 *niomaps = xfs_imap_to_bmap(io, offset, &imap,
 323                                             iomapp, nimaps, *niomaps, iomap_flags);
 324         } else if (niomaps) {
 325                 *niomaps = 0;
 326         }
 327
 328 out:
 329         if (lockmode)
 330                 XFS_IUNLOCK(mp, io, lockmode);
 331         return XFS_ERROR(error);
 332 }
 333
 334 STATIC int
 335 xfs_flush_space(
 336         xfs_inode_t     *ip,
 337         int             *fsynced,
 338         int             *ioflags)
 339 {
 340         switch (*fsynced) {
 341         case 0:
 342                 if (ip->i_delayed_blks) {
 343                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 344                         xfs_flush_inode(ip);
 345                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 346                         *fsynced = 1;
 347                 } else {
 348                         *ioflags |= BMAPI_SYNC;
 349                         *fsynced = 2;
 350                 }
 351                 return 0;
 352         case 1:
 353                 *fsynced = 2;
 354                 *ioflags |= BMAPI_SYNC;
 355                 return 0;
 356         case 2:
 357                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 358                 xfs_flush_device(ip);
 359                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 360                 *fsynced = 3;
 361                 return 0;
 362         }
 363         return 1;
 364 }
 365
 366 int
 367 xfs_iomap_write_direct(
 368         xfs_inode_t     *ip,
 369         xfs_off_t       offset,
 370         size_t          count,
 371         int             flags,
 372         xfs_bmbt_irec_t *ret_imap,
 373         int             *nmaps,
 374         int             found)
 375 {
 376         xfs_mount_t     *mp = ip->i_mount;
 377         xfs_iocore_t    *io = &ip->i_iocore;
 378         xfs_fileoff_t   offset_fsb;
 379         xfs_fileoff_t   last_fsb;
 380         xfs_filblks_t   count_fsb;
 381         xfs_fsize_t     isize;
 382         xfs_fsblock_t   firstfsb;
 383         int             nimaps, maps;
 384         int             error;
 385         int             bmapi_flag;
 386         int             quota_flag;
 387         int             rt;
 388         xfs_trans_t     *tp;
 389         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS], *imapp;
 390         xfs_bmap_free_t free_list;
 391         int             aeof;
 392         xfs_filblks_t   qblocks, resblks;
 393         int             committed;
 394         int             resrtextents;
 395
 396         /*
 397          * Make sure that the dquots are there. This doesn't hold
 398          * the ilock across a disk read.
 399          */
 400         error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
 401         if (error)
 402                 return XFS_ERROR(error);
 403
 404         maps = min(XFS_WRITE_IMAPS, *nmaps);
 405         nimaps = maps;
 406
 407         isize = ip->i_d.di_size;
 408         aeof = (offset + count) > isize;
 409
 410         if (io->io_new_size > isize)
 411                 isize = io->io_new_size;
 412
 413         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 414         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 415         count_fsb = last_fsb - offset_fsb;
 416         if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
 417                 xfs_fileoff_t   map_last_fsb;
 418
 419                 map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
 420                 if (map_last_fsb < last_fsb) {
 421                         last_fsb = map_last_fsb;
 422                         count_fsb = last_fsb - offset_fsb;
 423                 }
 424                 ASSERT(count_fsb > 0);
 425         }
 426
 427         /*
 428          * Determine if reserving space on the data or realtime partition.
 429          */
 430         if ((rt = XFS_IS_REALTIME_INODE(ip))) {
 431                 xfs_extlen_t    extsz;
 432
 433                 if (!(extsz = ip->i_d.di_extsize))
 434                         extsz = mp->m_sb.sb_rextsize;
 435                 resrtextents = qblocks = (count_fsb + extsz - 1);
 436                 do_div(resrtextents, mp->m_sb.sb_rextsize);
 437                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 438                 quota_flag = XFS_QMOPT_RES_RTBLKS;
 439         } else {
 440                 resrtextents = 0;
 441                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, count_fsb);
 442                 quota_flag = XFS_QMOPT_RES_REGBLKS;
 443         }
 444
 445         /*
 446          * Allocate and setup the transaction
 447          */
 448         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 449         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
 450         error = xfs_trans_reserve(tp, resblks,
 451                         XFS_WRITE_LOG_RES(mp), resrtextents,
 452                         XFS_TRANS_PERM_LOG_RES,
 453                         XFS_WRITE_LOG_COUNT);
 454
 455         /*
 456          * Check for running out of space, note: need lock to return
 457          */
 458         if (error)
 459                 xfs_trans_cancel(tp, 0);
 460         xfs_ilock(ip, XFS_ILOCK_EXCL);
 461         if (error)
 462                 goto error_out;
 463
 464         if (XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag)) {
 465                 error = (EDQUOT);
 466                 goto error1;
 467         }
 468
 469         bmapi_flag = XFS_BMAPI_WRITE;
 470         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 471         xfs_trans_ihold(tp, ip);
 472
 473         if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
 474                 bmapi_flag |= XFS_BMAPI_PREALLOC;
 475
 476         /*
 477          * Issue the bmapi() call to allocate the blocks
 478          */
 479         XFS_BMAP_INIT(&free_list, &firstfsb);
 480         nimaps = 1;
 481         imapp = &imap[0];
 482         error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 483                 bmapi_flag, &firstfsb, 0, imapp, &nimaps, &free_list);
 484         if (error)
 485                 goto error0;
 486
 487         /*
 488          * Complete the transaction
 489          */
 490         error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
 491         if (error)
 492                 goto error0;
 493         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 494         if (error)
 495                 goto error_out;
 496
 497         /*
 498          * Copy any maps to caller's array and return any error.
 499          */
 500         if (nimaps == 0) {
 501                 error = (ENOSPC);
 502                 goto error_out;
 503         }
 504
 505         *ret_imap = imap[0];
 506         *nmaps = 1;
 507         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 508                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 509                         "start_block : %llx start_off : %llx blkcnt : %llx "
 510                         "extent-state : %x \n",
 511                         (ip->i_mount)->m_fsname,
 512                         (long long)ip->i_ino,
 513                         ret_imap->br_startblock, ret_imap->br_startoff,
 514                         ret_imap->br_blockcount,ret_imap->br_state);
 515         }
 516         return 0;
 517
 518 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
 519         xfs_bmap_cancel(&free_list);
 520         XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
 521
 522 error1: /* Just cancel transaction */
 523         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 524         *nmaps = 0;     /* nothing set-up here */
 525
 526 error_out:
 527         return XFS_ERROR(error);
 528 }
 529
 530 int
 531 xfs_iomap_write_delay(
 532         xfs_inode_t     *ip,
 533         xfs_off_t       offset,
 534         size_t          count,
 535         int             ioflag,
 536         xfs_bmbt_irec_t *ret_imap,
 537         int             *nmaps)
 538 {
 539         xfs_mount_t     *mp = ip->i_mount;
 540         xfs_iocore_t    *io = &ip->i_iocore;
 541         xfs_fileoff_t   offset_fsb;
 542         xfs_fileoff_t   last_fsb;
 543         xfs_fsize_t     isize;
 544         xfs_fsblock_t   firstblock;
 545         int             nimaps;
 546         int             error;
 547         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
 548         int             aeof;
 549         int             fsynced = 0;
 550
 551         ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
 552
 553         /*
 554          * Make sure that the dquots are there. This doesn't hold
 555          * the ilock across a disk read.
 556          */
 557
 558         error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
 559         if (error)
 560                 return XFS_ERROR(error);
 561
 562 retry:
 563         isize = ip->i_d.di_size;
 564         if (io->io_new_size > isize) {
 565                 isize = io->io_new_size;
 566         }
 567
 568         aeof = 0;
 569         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 570         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 571         /*
 572          * If the caller is doing a write at the end of the file,
 573          * then extend the allocation (and the buffer used for the write)
 574          * out to the file system's write iosize.  We clean up any extra
 575          * space left over when the file is closed in xfs_inactive().
 576          *
 577          * For sync writes, we are flushing delayed allocate space to
 578          * try to make additional space available for allocation near
 579          * the filesystem full boundary - preallocation hurts in that
 580          * situation, of course.
 581          */
 582         if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
 583                 xfs_off_t       aligned_offset;
 584                 xfs_filblks_t   count_fsb;
 585                 unsigned int    iosize;
 586                 xfs_fileoff_t   ioalign;
 587                 int             n;
 588                 xfs_fileoff_t   start_fsb;
 589
 590                 /*
 591                  * If there are any real blocks past eof, then don't
 592                  * do any speculative allocation.
 593                  */
 594                 start_fsb = XFS_B_TO_FSBT(mp,
 595                                         ((xfs_ufsize_t)(offset + count - 1)));
 596                 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
 597                 while (count_fsb > 0) {
 598                         nimaps = XFS_WRITE_IMAPS;
 599                         error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
 600                                         0, &firstblock, 0, imap, &nimaps, NULL);
 601                         if (error) {
 602                                 return error;
 603                         }
 604                         for (n = 0; n < nimaps; n++) {
 605                                 if ( !(io->io_flags & XFS_IOCORE_RT)  &&
 606                                         !imap[n].br_startblock) {
 607                                         cmn_err(CE_PANIC,"Access to block "
 608                                                 "zero:  fs <%s> inode: %lld "
 609                                                 "start_block : %llx start_off "
 610                                                 ": %llx blkcnt : %llx "
 611                                                 "extent-state : %x \n",
 612                                                 (ip->i_mount)->m_fsname,
 613                                                 (long long)ip->i_ino,
 614                                                 imap[n].br_startblock,
 615                                                 imap[n].br_startoff,
 616                                                 imap[n].br_blockcount,
 617                                                 imap[n].br_state);
 618                                 }
 619                                 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
 620                                     (imap[n].br_startblock != DELAYSTARTBLOCK)) {
 621                                         goto write_map;
 622                                 }
 623                                 start_fsb += imap[n].br_blockcount;
 624                                 count_fsb -= imap[n].br_blockcount;
 625                         }
 626                 }
 627                 iosize = mp->m_writeio_blocks;
 628                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 629                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
 630                 last_fsb = ioalign + iosize;
 631                 aeof = 1;
 632         }
 633 write_map:
 634         nimaps = XFS_WRITE_IMAPS;
 635         firstblock = NULLFSBLOCK;
 636
 637         /*
 638          * If mounted with the "-o swalloc" option, roundup the allocation
 639          * request to a stripe width boundary if the file size is >=
 640          * stripe width and we are allocating past the allocation eof.
 641          */
 642         if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_swidth
 643             && (mp->m_flags & XFS_MOUNT_SWALLOC)
 644             && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
 645                 int eof;
 646                 xfs_fileoff_t new_last_fsb;
 647
 648                 new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
 649                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 650                 if (error) {
 651                         return error;
 652                 }
 653                 if (eof) {
 654                         last_fsb = new_last_fsb;
 655                 }
 656         /*
 657          * Roundup the allocation request to a stripe unit (m_dalign) boundary
 658          * if the file size is >= stripe unit size, and we are allocating past
 659          * the allocation eof.
 660          */
 661         } else if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_dalign &&
 662                    (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)) && aeof) {
 663                 int eof;
 664                 xfs_fileoff_t new_last_fsb;
 665                 new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
 666                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 667                 if (error) {
 668                         return error;
 669                 }
 670                 if (eof) {
 671                         last_fsb = new_last_fsb;
 672                 }
 673         /*
 674          * Round up the allocation request to a real-time extent boundary
 675          * if the file is on the real-time subvolume.
 676          */
 677         } else if (io->io_flags & XFS_IOCORE_RT && aeof) {
 678                 int eof;
 679                 xfs_fileoff_t new_last_fsb;
 680
 681                 new_last_fsb = roundup_64(last_fsb, mp->m_sb.sb_rextsize);
 682                 error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
 683                 if (error) {
 684                         return error;
 685                 }
 686                 if (eof)
 687                         last_fsb = new_last_fsb;
 688         }
 689         error = xfs_bmapi(NULL, ip, offset_fsb,
 690                           (xfs_filblks_t)(last_fsb - offset_fsb),
 691                           XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
 692                           XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
 693                           &nimaps, NULL);
 694         /*
 695          * This can be EDQUOT, if nimaps == 0
 696          */
 697         if (error && (error != ENOSPC)) {
 698                 return XFS_ERROR(error);
 699         }
 700         /*
 701          * If bmapi returned us nothing, and if we didn't get back EDQUOT,
 702          * then we must have run out of space.
 703          */
 704         if (nimaps == 0) {
 705                 xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
 706                                         io, offset, count);
 707                 if (xfs_flush_space(ip, &fsynced, &ioflag))
 708                         return XFS_ERROR(ENOSPC);
 709
 710                 error = 0;
 711                 goto retry;
 712         }
 713
 714         *ret_imap = imap[0];
 715         *nmaps = 1;
 716         if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
 717                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
 718                         "start_block : %llx start_off : %llx blkcnt : %llx "
 719                         "extent-state : %x \n",
 720                         (ip->i_mount)->m_fsname,
 721                         (long long)ip->i_ino,
 722                         ret_imap->br_startblock, ret_imap->br_startoff,
 723                         ret_imap->br_blockcount,ret_imap->br_state);
 724         }
 725         return 0;
 726 }
 727
 728 /*
 729  * Pass in a delayed allocate extent, convert it to real extents;
 730  * return to the caller the extent we create which maps on top of
 731  * the originating callers request.
 732  *
 733  * Called without a lock on the inode.
 734  */
 735 int
 736 xfs_iomap_write_allocate(
 737         xfs_inode_t     *ip,
 738         xfs_off_t       offset,
 739         size_t          count,
 740         xfs_bmbt_irec_t *map,
 741         int             *retmap)
 742 {
 743         xfs_mount_t     *mp = ip->i_mount;
 744         xfs_iocore_t    *io = &ip->i_iocore;
 745         xfs_fileoff_t   offset_fsb, last_block;
 746         xfs_fileoff_t   end_fsb, map_start_fsb;
 747         xfs_fsblock_t   first_block;
 748         xfs_bmap_free_t free_list;
 749         xfs_filblks_t   count_fsb;
 750         xfs_bmbt_irec_t imap[XFS_STRAT_WRITE_IMAPS];
 751         xfs_trans_t     *tp;
 752         int             i, nimaps, committed;
 753         int             error = 0;
 754         int             nres;
 755
 756         *retmap = 0;
 757
 758         /*
 759          * Make sure that the dquots are there.
 760          */
 761         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
 762                 return XFS_ERROR(error);
 763
 764         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 765         count_fsb = map->br_blockcount;
 766         map_start_fsb = map->br_startoff;
 767
 768         XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 769
 770         while (count_fsb != 0) {
 771                 /*
 772                  * Set up a transaction with which to allocate the
 773                  * backing store for the file.  Do allocations in a
 774                  * loop until we get some space in the range we are
 775                  * interested in.  The other space that might be allocated
 776                  * is in the delayed allocation extent on which we sit
 777                  * but before our buffer starts.
 778                  */
 779
 780                 nimaps = 0;
 781                 while (nimaps == 0) {
 782                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 783                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 784                         error = xfs_trans_reserve(tp, nres,
 785                                         XFS_WRITE_LOG_RES(mp),
 786                                         0, XFS_TRANS_PERM_LOG_RES,
 787                                         XFS_WRITE_LOG_COUNT);
 788                         if (error == ENOSPC) {
 789                                 error = xfs_trans_reserve(tp, 0,
 790                                                 XFS_WRITE_LOG_RES(mp),
 791                                                 0,
 792                                                 XFS_TRANS_PERM_LOG_RES,
 793                                                 XFS_WRITE_LOG_COUNT);
 794                         }
 795                         if (error) {
 796                                 xfs_trans_cancel(tp, 0);
 797                                 return XFS_ERROR(error);
 798                         }
 799                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 800                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 801                         xfs_trans_ihold(tp, ip);
 802
 803                         XFS_BMAP_INIT(&free_list, &first_block);
 804
 805                         nimaps = XFS_STRAT_WRITE_IMAPS;
 806                         /*
 807                          * Ensure we don't go beyond eof - it is possible
 808                          * the extents changed since we did the read call,
 809                          * we dropped the ilock in the interim.
 810                          */
 811
 812                         end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
 813                         xfs_bmap_last_offset(NULL, ip, &last_block,
 814                                 XFS_DATA_FORK);
 815                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 816                         if ((map_start_fsb + count_fsb) > last_block) {
 817                                 count_fsb = last_block - map_start_fsb;
 818                                 if (count_fsb == 0) {
 819                                         error = EAGAIN;
 820                                         goto trans_cancel;
 821                                 }
 822                         }
 823
 824                         /* Go get the actual blocks */
 825                         error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
 826                                         XFS_BMAPI_WRITE, &first_block, 1,
 827                                         imap, &nimaps, &free_list);
 828                         if (error)
 829                                 goto trans_cancel;
 830
 831                         error = xfs_bmap_finish(&tp, &free_list,
 832                                         first_block, &committed);
 833                         if (error)
 834                                 goto trans_cancel;
 835
 836                         error = xfs_trans_commit(tp,
 837                                         XFS_TRANS_RELEASE_LOG_RES, NULL);
 838                         if (error)
 839                                 goto error0;
 840
 841                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 842                 }
 843
 844                 /*
 845                  * See if we were able to allocate an extent that
 846                  * covers at least part of the callers request
 847                  */
 848
 849                 for (i = 0; i < nimaps; i++) {
 850                         if ( !(io->io_flags & XFS_IOCORE_RT)  &&
 851                                 !imap[i].br_startblock) {
 852                                 cmn_err(CE_PANIC,"Access to block zero:  "
 853                                         "fs <%s> inode: %lld "
 854                                         "start_block : %llx start_off : %llx "
 855                                         "blkcnt : %llx extent-state : %x \n",
 856                                         (ip->i_mount)->m_fsname,
 857                                         (long long)ip->i_ino,
 858                                         imap[i].br_startblock,
 859                                         imap[i].br_startoff,
 860                                         imap[i].br_blockcount,imap[i].br_state);
 861                         }
 862                         if ((offset_fsb >= imap[i].br_startoff) &&
 863                             (offset_fsb < (imap[i].br_startoff +
 864                                            imap[i].br_blockcount))) {
 865                                 *map = imap[i];
 866                                 *retmap = 1;
 867                                 XFS_STATS_INC(xs_xstrat_quick);
 868                                 return 0;
 869                         }
 870                         count_fsb -= imap[i].br_blockcount;
 871                 }
 872
 873                 /* So far we have not mapped the requested part of the
 874                  * file, just surrounding data, try again.
 875                  */
 876                 nimaps--;
 877                 map_start_fsb = imap[nimaps].br_startoff +
 878                                 imap[nimaps].br_blockcount;
 879         }
 880
 881 trans_cancel:
 882         xfs_bmap_cancel(&free_list);
 883         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 884 error0:
 885         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 886         return XFS_ERROR(error);
 887 }
 888
 889 int
 890 xfs_iomap_write_unwritten(
 891         xfs_inode_t     *ip,
 892         xfs_off_t       offset,
 893         size_t          count)
 894 {
 895         xfs_mount_t     *mp = ip->i_mount;
 896         xfs_iocore_t    *io = &ip->i_iocore;
 897         xfs_trans_t     *tp;
 898         xfs_fileoff_t   offset_fsb;
 899         xfs_filblks_t   count_fsb;
 900         xfs_filblks_t   numblks_fsb;
 901         xfs_bmbt_irec_t imap;
 902         int             committed;
 903         int             error;
 904         int             nres;
 905         int             nimaps;
 906         xfs_fsblock_t   firstfsb;
 907         xfs_bmap_free_t free_list;
 908
 909         xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
 910                                 &ip->i_iocore, offset, count);
 911
 912         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 913         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 914         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 915
 916         do {
 917                 nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 918
 919                 /*
 920                  * set up a transaction to convert the range of extents
 921                  * from unwritten to real. Do allocations in a loop until
 922                  * we have covered the range passed in.
 923                  */
 924
 925                 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 926                 error = xfs_trans_reserve(tp, nres,
 927                                 XFS_WRITE_LOG_RES(mp), 0,
 928                                 XFS_TRANS_PERM_LOG_RES,
 929                                 XFS_WRITE_LOG_COUNT);
 930                 if (error) {
 931                         xfs_trans_cancel(tp, 0);
 932                         goto error0;
 933                 }
 934
 935                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 936                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 937                 xfs_trans_ihold(tp, ip);
 938
 939                 /*
 940                  * Modify the unwritten extent state of the buffer.
 941                  */
 942                 XFS_BMAP_INIT(&free_list, &firstfsb);
 943                 nimaps = 1;
 944                 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
 945                                   XFS_BMAPI_WRITE, &firstfsb,
 946                                   1, &imap, &nimaps, &free_list);
 947                 if (error)
 948                         goto error_on_bmapi_transaction;
 949
 950                 error = xfs_bmap_finish(&(tp), &(free_list),
 951                                 firstfsb, &committed);
 952                 if (error)
 953                         goto error_on_bmapi_transaction;
 954
 955                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
 956                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 957                 if (error)
 958                         goto error0;
 959
 960                 if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
 961                         cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
 962                                 "inode: %lld start_block : %llx start_off : "
 963                                 "%llx blkcnt : %llx extent-state : %x \n",
 964                                 (ip->i_mount)->m_fsname,
 965                                 (long long)ip->i_ino,
 966                                 imap.br_startblock,imap.br_startoff,
 967                                 imap.br_blockcount,imap.br_state);
 968                 }
 969
 970                 if ((numblks_fsb = imap.br_blockcount) == 0) {
 971                         /*
 972                          * The numblks_fsb value should always get
 973                          * smaller, otherwise the loop is stuck.
 974                          */
 975                         ASSERT(imap.br_blockcount);
 976                         break;
 977                 }
 978                 offset_fsb += numblks_fsb;
 979                 count_fsb -= numblks_fsb;
 980         } while (count_fsb > 0);
 981
 982         return 0;
 983
 984 error_on_bmapi_transaction:
 985         xfs_bmap_cancel(&free_list);
 986         xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
 987         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 988 error0:
 989         return XFS_ERROR(error);
 990 }