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