fs/xfs/xfs_iomap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_mount.h"
  25 #include "xfs_inode.h"
  26 #include "xfs_btree.h"
  27 #include "xfs_bmap_btree.h"
  28 #include "xfs_bmap.h"
  29 #include "xfs_bmap_util.h"
  30 #include "xfs_error.h"
  31 #include "xfs_trans.h"
  32 #include "xfs_trans_space.h"
  33 #include "xfs_iomap.h"
  34 #include "xfs_trace.h"
  35 #include "xfs_icache.h"
  36 #include "xfs_quota.h"
  37 #include "xfs_dquot_item.h"
  38 #include "xfs_dquot.h"
  39
  40
  41 #define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
  42                                                 << mp->m_writeio_log)
  43 #define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
  44
  45 STATIC int
  46 xfs_iomap_eof_align_last_fsb(
  47         xfs_mount_t     *mp,
  48         xfs_inode_t     *ip,
  49         xfs_extlen_t    extsize,
  50         xfs_fileoff_t   *last_fsb)
  51 {
  52         xfs_extlen_t    align = 0;
  53         int             eof, error;
  54
  55         if (!XFS_IS_REALTIME_INODE(ip)) {
  56                 /*
  57                  * Round up the allocation request to a stripe unit
  58                  * (m_dalign) boundary if the file size is >= stripe unit
  59                  * size, and we are allocating past the allocation eof.
  60                  *
  61                  * If mounted with the "-o swalloc" option the alignment is
  62                  * increased from the strip unit size to the stripe width.
  63                  */
  64                 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
  65                         align = mp->m_swidth;
  66                 else if (mp->m_dalign)
  67                         align = mp->m_dalign;
  68
  69                 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
  70                         align = 0;
  71         }
  72
  73         /*
  74          * Always round up the allocation request to an extent boundary
  75          * (when file on a real-time subvolume or has di_extsize hint).
  76          */
  77         if (extsize) {
  78                 if (align)
  79                         align = roundup_64(align, extsize);
  80                 else
  81                         align = extsize;
  82         }
  83
  84         if (align) {
  85                 xfs_fileoff_t   new_last_fsb = roundup_64(*last_fsb, align);
  86                 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
  87                 if (error)
  88                         return error;
  89                 if (eof)
  90                         *last_fsb = new_last_fsb;
  91         }
  92         return 0;
  93 }
  94
  95 STATIC int
  96 xfs_alert_fsblock_zero(
  97         xfs_inode_t     *ip,
  98         xfs_bmbt_irec_t *imap)
  99 {
 100         xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
 101                         "Access to block zero in inode %llu "
 102                         "start_block: %llx start_off: %llx "
 103                         "blkcnt: %llx extent-state: %x",
 104                 (unsigned long long)ip->i_ino,
 105                 (unsigned long long)imap->br_startblock,
 106                 (unsigned long long)imap->br_startoff,
 107                 (unsigned long long)imap->br_blockcount,
 108                 imap->br_state);
 109         return -EFSCORRUPTED;
 110 }
 111
 112 int
 113 xfs_iomap_write_direct(
 114         xfs_inode_t     *ip,
 115         xfs_off_t       offset,
 116         size_t          count,
 117         xfs_bmbt_irec_t *imap,
 118         int             nmaps)
 119 {
 120         xfs_mount_t     *mp = ip->i_mount;
 121         xfs_fileoff_t   offset_fsb;
 122         xfs_fileoff_t   last_fsb;
 123         xfs_filblks_t   count_fsb, resaligned;
 124         xfs_fsblock_t   firstfsb;
 125         xfs_extlen_t    extsz, temp;
 126         int             nimaps;
 127         int             quota_flag;
 128         int             rt;
 129         xfs_trans_t     *tp;
 130         xfs_bmap_free_t free_list;
 131         uint            qblocks, resblks, resrtextents;
 132         int             error;
 133         int             lockmode;
 134         int             bmapi_flags = XFS_BMAPI_PREALLOC;
 135
 136         rt = XFS_IS_REALTIME_INODE(ip);
 137         extsz = xfs_get_extsz_hint(ip);
 138         lockmode = XFS_ILOCK_SHARED;    /* locked by caller */
 139
 140         ASSERT(xfs_isilocked(ip, lockmode));
 141
 142         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 143         last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 144         if ((offset + count) > XFS_ISIZE(ip)) {
 145                 /*
 146                  * Assert that the in-core extent list is present since this can
 147                  * call xfs_iread_extents() and we only have the ilock shared.
 148                  * This should be safe because the lock was held around a bmapi
 149                  * call in the caller and we only need it to access the in-core
 150                  * list.
 151                  */
 152                 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
 153                                                                 XFS_IFEXTENTS);
 154                 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
 155                 if (error)
 156                         goto out_unlock;
 157         } else {
 158                 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
 159                         last_fsb = MIN(last_fsb, (xfs_fileoff_t)
 160                                         imap->br_blockcount +
 161                                         imap->br_startoff);
 162         }
 163         count_fsb = last_fsb - offset_fsb;
 164         ASSERT(count_fsb > 0);
 165
 166         resaligned = count_fsb;
 167         if (unlikely(extsz)) {
 168                 if ((temp = do_mod(offset_fsb, extsz)))
 169                         resaligned += temp;
 170                 if ((temp = do_mod(resaligned, extsz)))
 171                         resaligned += extsz - temp;
 172         }
 173
 174         if (unlikely(rt)) {
 175                 resrtextents = qblocks = resaligned;
 176                 resrtextents /= mp->m_sb.sb_rextsize;
 177                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 178                 quota_flag = XFS_QMOPT_RES_RTBLKS;
 179         } else {
 180                 resrtextents = 0;
 181                 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
 182                 quota_flag = XFS_QMOPT_RES_REGBLKS;
 183         }
 184
 185         /*
 186          * Drop the shared lock acquired by the caller, attach the dquot if
 187          * necessary and move on to transaction setup.
 188          */
 189         xfs_iunlock(ip, lockmode);
 190         error = xfs_qm_dqattach(ip, 0);
 191         if (error)
 192                 return error;
 193
 194         /*
 195          * Allocate and setup the transaction
 196          */
 197         tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
 198
 199         /*
 200          * For DAX, we do not allocate unwritten extents, but instead we zero
 201          * the block before we commit the transaction.  Ideally we'd like to do
 202          * this outside the transaction context, but if we commit and then crash
 203          * we may not have zeroed the blocks and this will be exposed on
 204          * recovery of the allocation. Hence we must zero before commit.
 205          *
 206          * Further, if we are mapping unwritten extents here, we need to zero
 207          * and convert them to written so that we don't need an unwritten extent
 208          * callback for DAX. This also means that we need to be able to dip into
 209          * the reserve block pool for bmbt block allocation if there is no space
 210          * left but we need to do unwritten extent conversion.
 211          */
 212
 213         if (IS_DAX(VFS_I(ip))) {
 214                 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
 215                 if (ISUNWRITTEN(imap)) {
 216                         tp->t_flags |= XFS_TRANS_RESERVE;
 217                         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 218                 }
 219         }
 220         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
 221                                   resblks, resrtextents);
 222         /*
 223          * Check for running out of space, note: need lock to return
 224          */
 225         if (error) {
 226                 xfs_trans_cancel(tp);
 227                 return error;
 228         }
 229
 230         lockmode = XFS_ILOCK_EXCL;
 231         xfs_ilock(ip, lockmode);
 232
 233         error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
 234         if (error)
 235                 goto out_trans_cancel;
 236
 237         xfs_trans_ijoin(tp, ip, 0);
 238
 239         /*
 240          * From this point onwards we overwrite the imap pointer that the
 241          * caller gave to us.
 242          */
 243         xfs_bmap_init(&free_list, &firstfsb);
 244         nimaps = 1;
 245         error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
 246                                 bmapi_flags, &firstfsb, resblks, imap,
 247                                 &nimaps, &free_list);
 248         if (error)
 249                 goto out_bmap_cancel;
 250
 251         /*
 252          * Complete the transaction
 253          */
 254         error = xfs_bmap_finish(&tp, &free_list, NULL);
 255         if (error)
 256                 goto out_bmap_cancel;
 257
 258         error = xfs_trans_commit(tp);
 259         if (error)
 260                 goto out_unlock;
 261
 262         /*
 263          * Copy any maps to caller's array and return any error.
 264          */
 265         if (nimaps == 0) {
 266                 error = -ENOSPC;
 267                 goto out_unlock;
 268         }
 269
 270         if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
 271                 error = xfs_alert_fsblock_zero(ip, imap);
 272
 273 out_unlock:
 274         xfs_iunlock(ip, lockmode);
 275         return error;
 276
 277 out_bmap_cancel:
 278         xfs_bmap_cancel(&free_list);
 279         xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
 280 out_trans_cancel:
 281         xfs_trans_cancel(tp);
 282         goto out_unlock;
 283 }
 284
 285 /*
 286  * If the caller is doing a write at the end of the file, then extend the
 287  * allocation out to the file system's write iosize.  We clean up any extra
 288  * space left over when the file is closed in xfs_inactive().
 289  *
 290  * If we find we already have delalloc preallocation beyond EOF, don't do more
 291  * preallocation as it it not needed.
 292  */
 293 STATIC int
 294 xfs_iomap_eof_want_preallocate(
 295         xfs_mount_t     *mp,
 296         xfs_inode_t     *ip,
 297         xfs_off_t       offset,
 298         size_t          count,
 299         xfs_bmbt_irec_t *imap,
 300         int             nimaps,
 301         int             *prealloc)
 302 {
 303         xfs_fileoff_t   start_fsb;
 304         xfs_filblks_t   count_fsb;
 305         int             n, error, imaps;
 306         int             found_delalloc = 0;
 307
 308         *prealloc = 0;
 309         if (offset + count <= XFS_ISIZE(ip))
 310                 return 0;
 311
 312         /*
 313          * If the file is smaller than the minimum prealloc and we are using
 314          * dynamic preallocation, don't do any preallocation at all as it is
 315          * likely this is the only write to the file that is going to be done.
 316          */
 317         if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
 318             XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))
 319                 return 0;
 320
 321         /*
 322          * If there are any real blocks past eof, then don't
 323          * do any speculative allocation.
 324          */
 325         start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
 326         count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 327         while (count_fsb > 0) {
 328                 imaps = nimaps;
 329                 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
 330                                        0);
 331                 if (error)
 332                         return error;
 333                 for (n = 0; n < imaps; n++) {
 334                         if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
 335                             (imap[n].br_startblock != DELAYSTARTBLOCK))
 336                                 return 0;
 337                         start_fsb += imap[n].br_blockcount;
 338                         count_fsb -= imap[n].br_blockcount;
 339
 340                         if (imap[n].br_startblock == DELAYSTARTBLOCK)
 341                                 found_delalloc = 1;
 342                 }
 343         }
 344         if (!found_delalloc)
 345                 *prealloc = 1;
 346         return 0;
 347 }
 348
 349 /*
 350  * Determine the initial size of the preallocation. We are beyond the current
 351  * EOF here, but we need to take into account whether this is a sparse write or
 352  * an extending write when determining the preallocation size.  Hence we need to
 353  * look up the extent that ends at the current write offset and use the result
 354  * to determine the preallocation size.
 355  *
 356  * If the extent is a hole, then preallocation is essentially disabled.
 357  * Otherwise we take the size of the preceeding data extent as the basis for the
 358  * preallocation size. If the size of the extent is greater than half the
 359  * maximum extent length, then use the current offset as the basis. This ensures
 360  * that for large files the preallocation size always extends to MAXEXTLEN
 361  * rather than falling short due to things like stripe unit/width alignment of
 362  * real extents.
 363  */
 364 STATIC xfs_fsblock_t
 365 xfs_iomap_eof_prealloc_initial_size(
 366         struct xfs_mount        *mp,
 367         struct xfs_inode        *ip,
 368         xfs_off_t               offset,
 369         xfs_bmbt_irec_t         *imap,
 370         int                     nimaps)
 371 {
 372         xfs_fileoff_t   start_fsb;
 373         int             imaps = 1;
 374         int             error;
 375
 376         ASSERT(nimaps >= imaps);
 377
 378         /* if we are using a specific prealloc size, return now */
 379         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
 380                 return 0;
 381
 382         /* If the file is small, then use the minimum prealloc */
 383         if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign))
 384                 return 0;
 385
 386         /*
 387          * As we write multiple pages, the offset will always align to the
 388          * start of a page and hence point to a hole at EOF. i.e. if the size is
 389          * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
 390          * will return FSB 1. Hence if there are blocks in the file, we want to
 391          * point to the block prior to the EOF block and not the hole that maps
 392          * directly at @offset.
 393          */
 394         start_fsb = XFS_B_TO_FSB(mp, offset);
 395         if (start_fsb)
 396                 start_fsb--;
 397         error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
 398         if (error)
 399                 return 0;
 400
 401         ASSERT(imaps == 1);
 402         if (imap[0].br_startblock == HOLESTARTBLOCK)
 403                 return 0;
 404         if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
 405                 return imap[0].br_blockcount << 1;
 406         return XFS_B_TO_FSB(mp, offset);
 407 }
 408
 409 STATIC bool
 410 xfs_quota_need_throttle(
 411         struct xfs_inode *ip,
 412         int type,
 413         xfs_fsblock_t alloc_blocks)
 414 {
 415         struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
 416
 417         if (!dq || !xfs_this_quota_on(ip->i_mount, type))
 418                 return false;
 419
 420         /* no hi watermark, no throttle */
 421         if (!dq->q_prealloc_hi_wmark)
 422                 return false;
 423
 424         /* under the lo watermark, no throttle */
 425         if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
 426                 return false;
 427
 428         return true;
 429 }
 430
 431 STATIC void
 432 xfs_quota_calc_throttle(
 433         struct xfs_inode *ip,
 434         int type,
 435         xfs_fsblock_t *qblocks,
 436         int *qshift,
 437         int64_t *qfreesp)
 438 {
 439         int64_t freesp;
 440         int shift = 0;
 441         struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
 442
 443         /* no dq, or over hi wmark, squash the prealloc completely */
 444         if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
 445                 *qblocks = 0;
 446                 *qfreesp = 0;
 447                 return;
 448         }
 449
 450         freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
 451         if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
 452                 shift = 2;
 453                 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
 454                         shift += 2;
 455                 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
 456                         shift += 2;
 457         }
 458
 459         if (freesp < *qfreesp)
 460                 *qfreesp = freesp;
 461
 462         /* only overwrite the throttle values if we are more aggressive */
 463         if ((freesp >> shift) < (*qblocks >> *qshift)) {
 464                 *qblocks = freesp;
 465                 *qshift = shift;
 466         }
 467 }
 468
 469 /*
 470  * If we don't have a user specified preallocation size, dynamically increase
 471  * the preallocation size as the size of the file grows. Cap the maximum size
 472  * at a single extent or less if the filesystem is near full. The closer the
 473  * filesystem is to full, the smaller the maximum prealocation.
 474  */
 475 STATIC xfs_fsblock_t
 476 xfs_iomap_prealloc_size(
 477         struct xfs_mount        *mp,
 478         struct xfs_inode        *ip,
 479         xfs_off_t               offset,
 480         struct xfs_bmbt_irec    *imap,
 481         int                     nimaps)
 482 {
 483         xfs_fsblock_t           alloc_blocks = 0;
 484         int                     shift = 0;
 485         int64_t                 freesp;
 486         xfs_fsblock_t           qblocks;
 487         int                     qshift = 0;
 488
 489         alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
 490                                                            imap, nimaps);
 491         if (!alloc_blocks)
 492                 goto check_writeio;
 493         qblocks = alloc_blocks;
 494
 495         /*
 496          * MAXEXTLEN is not a power of two value but we round the prealloc down
 497          * to the nearest power of two value after throttling. To prevent the
 498          * round down from unconditionally reducing the maximum supported prealloc
 499          * size, we round up first, apply appropriate throttling, round down and
 500          * cap the value to MAXEXTLEN.
 501          */
 502         alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
 503                                        alloc_blocks);
 504
 505         freesp = percpu_counter_read_positive(&mp->m_fdblocks);
 506         if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
 507                 shift = 2;
 508                 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
 509                         shift++;
 510                 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
 511                         shift++;
 512                 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
 513                         shift++;
 514                 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
 515                         shift++;
 516         }
 517
 518         /*
 519          * Check each quota to cap the prealloc size, provide a shift value to
 520          * throttle with and adjust amount of available space.
 521          */
 522         if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
 523                 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
 524                                         &freesp);
 525         if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
 526                 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
 527                                         &freesp);
 528         if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
 529                 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
 530                                         &freesp);
 531
 532         /*
 533          * The final prealloc size is set to the minimum of free space available
 534          * in each of the quotas and the overall filesystem.
 535          *
 536          * The shift throttle value is set to the maximum value as determined by
 537          * the global low free space values and per-quota low free space values.
 538          */
 539         alloc_blocks = MIN(alloc_blocks, qblocks);
 540         shift = MAX(shift, qshift);
 541
 542         if (shift)
 543                 alloc_blocks >>= shift;
 544         /*
 545          * rounddown_pow_of_two() returns an undefined result if we pass in
 546          * alloc_blocks = 0.
 547          */
 548         if (alloc_blocks)
 549                 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
 550         if (alloc_blocks > MAXEXTLEN)
 551                 alloc_blocks = MAXEXTLEN;
 552
 553         /*
 554          * If we are still trying to allocate more space than is
 555          * available, squash the prealloc hard. This can happen if we
 556          * have a large file on a small filesystem and the above
 557          * lowspace thresholds are smaller than MAXEXTLEN.
 558          */
 559         while (alloc_blocks && alloc_blocks >= freesp)
 560                 alloc_blocks >>= 4;
 561
 562 check_writeio:
 563         if (alloc_blocks < mp->m_writeio_blocks)
 564                 alloc_blocks = mp->m_writeio_blocks;
 565
 566         trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
 567                                       mp->m_writeio_blocks);
 568
 569         return alloc_blocks;
 570 }
 571
 572 int
 573 xfs_iomap_write_delay(
 574         xfs_inode_t     *ip,
 575         xfs_off_t       offset,
 576         size_t          count,
 577         xfs_bmbt_irec_t *ret_imap)
 578 {
 579         xfs_mount_t     *mp = ip->i_mount;
 580         xfs_fileoff_t   offset_fsb;
 581         xfs_fileoff_t   last_fsb;
 582         xfs_off_t       aligned_offset;
 583         xfs_fileoff_t   ioalign;
 584         xfs_extlen_t    extsz;
 585         int             nimaps;
 586         xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
 587         int             prealloc;
 588         int             error;
 589
 590         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 591
 592         /*
 593          * Make sure that the dquots are there. This doesn't hold
 594          * the ilock across a disk read.
 595          */
 596         error = xfs_qm_dqattach_locked(ip, 0);
 597         if (error)
 598                 return error;
 599
 600         extsz = xfs_get_extsz_hint(ip);
 601         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 602
 603         error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
 604                                 imap, XFS_WRITE_IMAPS, &prealloc);
 605         if (error)
 606                 return error;
 607
 608 retry:
 609         if (prealloc) {
 610                 xfs_fsblock_t   alloc_blocks;
 611
 612                 alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
 613                                                        XFS_WRITE_IMAPS);
 614
 615                 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 616                 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
 617                 last_fsb = ioalign + alloc_blocks;
 618         } else {
 619                 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 620         }
 621
 622         if (prealloc || extsz) {
 623                 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
 624                 if (error)
 625                         return error;
 626         }
 627
 628         /*
 629          * Make sure preallocation does not create extents beyond the range we
 630          * actually support in this filesystem.
 631          */
 632         if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
 633                 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 634
 635         ASSERT(last_fsb > offset_fsb);
 636
 637         nimaps = XFS_WRITE_IMAPS;
 638         error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
 639                                 imap, &nimaps, XFS_BMAPI_ENTIRE);
 640         switch (error) {
 641         case 0:
 642         case -ENOSPC:
 643         case -EDQUOT:
 644                 break;
 645         default:
 646                 return error;
 647         }
 648
 649         /*
 650          * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
 651          * without EOF preallocation.
 652          */
 653         if (nimaps == 0) {
 654                 trace_xfs_delalloc_enospc(ip, offset, count);
 655                 if (prealloc) {
 656                         prealloc = 0;
 657                         error = 0;
 658                         goto retry;
 659                 }
 660                 return error ? error : -ENOSPC;
 661         }
 662
 663         if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
 664                 return xfs_alert_fsblock_zero(ip, &imap[0]);
 665
 666         /*
 667          * Tag the inode as speculatively preallocated so we can reclaim this
 668          * space on demand, if necessary.
 669          */
 670         if (prealloc)
 671                 xfs_inode_set_eofblocks_tag(ip);
 672
 673         *ret_imap = imap[0];
 674         return 0;
 675 }
 676
 677 /*
 678  * Pass in a delayed allocate extent, convert it to real extents;
 679  * return to the caller the extent we create which maps on top of
 680  * the originating callers request.
 681  *
 682  * Called without a lock on the inode.
 683  *
 684  * We no longer bother to look at the incoming map - all we have to
 685  * guarantee is that whatever we allocate fills the required range.
 686  */
 687 int
 688 xfs_iomap_write_allocate(
 689         xfs_inode_t     *ip,
 690         xfs_off_t       offset,
 691         xfs_bmbt_irec_t *imap)
 692 {
 693         xfs_mount_t     *mp = ip->i_mount;
 694         xfs_fileoff_t   offset_fsb, last_block;
 695         xfs_fileoff_t   end_fsb, map_start_fsb;
 696         xfs_fsblock_t   first_block;
 697         xfs_bmap_free_t free_list;
 698         xfs_filblks_t   count_fsb;
 699         xfs_trans_t     *tp;
 700         int             nimaps;
 701         int             error = 0;
 702         int             nres;
 703
 704         /*
 705          * Make sure that the dquots are there.
 706          */
 707         error = xfs_qm_dqattach(ip, 0);
 708         if (error)
 709                 return error;
 710
 711         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 712         count_fsb = imap->br_blockcount;
 713         map_start_fsb = imap->br_startoff;
 714
 715         XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 716
 717         while (count_fsb != 0) {
 718                 /*
 719                  * Set up a transaction with which to allocate the
 720                  * backing store for the file.  Do allocations in a
 721                  * loop until we get some space in the range we are
 722                  * interested in.  The other space that might be allocated
 723                  * is in the delayed allocation extent on which we sit
 724                  * but before our buffer starts.
 725                  */
 726
 727                 nimaps = 0;
 728                 while (nimaps == 0) {
 729                         tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 730                         tp->t_flags |= XFS_TRANS_RESERVE;
 731                         nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 732                         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
 733                                                   nres, 0);
 734                         if (error) {
 735                                 xfs_trans_cancel(tp);
 736                                 return error;
 737                         }
 738                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 739                         xfs_trans_ijoin(tp, ip, 0);
 740
 741                         xfs_bmap_init(&free_list, &first_block);
 742
 743                         /*
 744                          * it is possible that the extents have changed since
 745                          * we did the read call as we dropped the ilock for a
 746                          * while. We have to be careful about truncates or hole
 747                          * punchs here - we are not allowed to allocate
 748                          * non-delalloc blocks here.
 749                          *
 750                          * The only protection against truncation is the pages
 751                          * for the range we are being asked to convert are
 752                          * locked and hence a truncate will block on them
 753                          * first.
 754                          *
 755                          * As a result, if we go beyond the range we really
 756                          * need and hit an delalloc extent boundary followed by
 757                          * a hole while we have excess blocks in the map, we
 758                          * will fill the hole incorrectly and overrun the
 759                          * transaction reservation.
 760                          *
 761                          * Using a single map prevents this as we are forced to
 762                          * check each map we look for overlap with the desired
 763                          * range and abort as soon as we find it. Also, given
 764                          * that we only return a single map, having one beyond
 765                          * what we can return is probably a bit silly.
 766                          *
 767                          * We also need to check that we don't go beyond EOF;
 768                          * this is a truncate optimisation as a truncate sets
 769                          * the new file size before block on the pages we
 770                          * currently have locked under writeback. Because they
 771                          * are about to be tossed, we don't need to write them
 772                          * back....
 773                          */
 774                         nimaps = 1;
 775                         end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
 776                         error = xfs_bmap_last_offset(ip, &last_block,
 777                                                         XFS_DATA_FORK);
 778                         if (error)
 779                                 goto trans_cancel;
 780
 781                         last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 782                         if ((map_start_fsb + count_fsb) > last_block) {
 783                                 count_fsb = last_block - map_start_fsb;
 784                                 if (count_fsb == 0) {
 785                                         error = -EAGAIN;
 786                                         goto trans_cancel;
 787                                 }
 788                         }
 789
 790                         /*
 791                          * From this point onwards we overwrite the imap
 792                          * pointer that the caller gave to us.
 793                          */
 794                         error = xfs_bmapi_write(tp, ip, map_start_fsb,
 795                                                 count_fsb, 0, &first_block,
 796                                                 nres, imap, &nimaps,
 797                                                 &free_list);
 798                         if (error)
 799                                 goto trans_cancel;
 800
 801                         error = xfs_bmap_finish(&tp, &free_list, NULL);
 802                         if (error)
 803                                 goto trans_cancel;
 804
 805                         error = xfs_trans_commit(tp);
 806                         if (error)
 807                                 goto error0;
 808
 809                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 810                 }
 811
 812                 /*
 813                  * See if we were able to allocate an extent that
 814                  * covers at least part of the callers request
 815                  */
 816                 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
 817                         return xfs_alert_fsblock_zero(ip, imap);
 818
 819                 if ((offset_fsb >= imap->br_startoff) &&
 820                     (offset_fsb < (imap->br_startoff +
 821                                    imap->br_blockcount))) {
 822                         XFS_STATS_INC(mp, xs_xstrat_quick);
 823                         return 0;
 824                 }
 825
 826                 /*
 827                  * So far we have not mapped the requested part of the
 828                  * file, just surrounding data, try again.
 829                  */
 830                 count_fsb -= imap->br_blockcount;
 831                 map_start_fsb = imap->br_startoff + imap->br_blockcount;
 832         }
 833
 834 trans_cancel:
 835         xfs_bmap_cancel(&free_list);
 836         xfs_trans_cancel(tp);
 837 error0:
 838         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 839         return error;
 840 }
 841
 842 int
 843 xfs_iomap_write_unwritten(
 844         xfs_inode_t     *ip,
 845         xfs_off_t       offset,
 846         xfs_off_t       count)
 847 {
 848         xfs_mount_t     *mp = ip->i_mount;
 849         xfs_fileoff_t   offset_fsb;
 850         xfs_filblks_t   count_fsb;
 851         xfs_filblks_t   numblks_fsb;
 852         xfs_fsblock_t   firstfsb;
 853         int             nimaps;
 854         xfs_trans_t     *tp;
 855         xfs_bmbt_irec_t imap;
 856         xfs_bmap_free_t free_list;
 857         xfs_fsize_t     i_size;
 858         uint            resblks;
 859         int             error;
 860
 861         trace_xfs_unwritten_convert(ip, offset, count);
 862
 863         offset_fsb = XFS_B_TO_FSBT(mp, offset);
 864         count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 865         count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 866
 867         /*
 868          * Reserve enough blocks in this transaction for two complete extent
 869          * btree splits.  We may be converting the middle part of an unwritten
 870          * extent and in this case we will insert two new extents in the btree
 871          * each of which could cause a full split.
 872          *
 873          * This reservation amount will be used in the first call to
 874          * xfs_bmbt_split() to select an AG with enough space to satisfy the
 875          * rest of the operation.
 876          */
 877         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 878
 879         do {
 880                 /*
 881                  * set up a transaction to convert the range of extents
 882                  * from unwritten to real. Do allocations in a loop until
 883                  * we have covered the range passed in.
 884                  *
 885                  * Note that we open code the transaction allocation here
 886                  * to pass KM_NOFS--we can't risk to recursing back into
 887                  * the filesystem here as we might be asked to write out
 888                  * the same inode that we complete here and might deadlock
 889                  * on the iolock.
 890                  */
 891                 sb_start_intwrite(mp->m_super);
 892                 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
 893                 tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT;
 894                 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
 895                                           resblks, 0);
 896                 if (error) {
 897                         xfs_trans_cancel(tp);
 898                         return error;
 899                 }
 900
 901                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 902                 xfs_trans_ijoin(tp, ip, 0);
 903
 904                 /*
 905                  * Modify the unwritten extent state of the buffer.
 906                  */
 907                 xfs_bmap_init(&free_list, &firstfsb);
 908                 nimaps = 1;
 909                 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
 910                                         XFS_BMAPI_CONVERT, &firstfsb, resblks,
 911                                         &imap, &nimaps, &free_list);
 912                 if (error)
 913                         goto error_on_bmapi_transaction;
 914
 915                 /*
 916                  * Log the updated inode size as we go.  We have to be careful
 917                  * to only log it up to the actual write offset if it is
 918                  * halfway into a block.
 919                  */
 920                 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
 921                 if (i_size > offset + count)
 922                         i_size = offset + count;
 923
 924                 i_size = xfs_new_eof(ip, i_size);
 925                 if (i_size) {
 926                         ip->i_d.di_size = i_size;
 927                         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 928                 }
 929
 930                 error = xfs_bmap_finish(&tp, &free_list, NULL);
 931                 if (error)
 932                         goto error_on_bmapi_transaction;
 933
 934                 error = xfs_trans_commit(tp);
 935                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 936                 if (error)
 937                         return error;
 938
 939                 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
 940                         return xfs_alert_fsblock_zero(ip, &imap);
 941
 942                 if ((numblks_fsb = imap.br_blockcount) == 0) {
 943                         /*
 944                          * The numblks_fsb value should always get
 945                          * smaller, otherwise the loop is stuck.
 946                          */
 947                         ASSERT(imap.br_blockcount);
 948                         break;
 949                 }
 950                 offset_fsb += numblks_fsb;
 951                 count_fsb -= numblks_fsb;
 952         } while (count_fsb > 0);
 953
 954         return 0;
 955
 956 error_on_bmapi_transaction:
 957         xfs_bmap_cancel(&free_list);
 958         xfs_trans_cancel(tp);
 959         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 960         return error;
 961 }