linux/fs/xfs/xfs_iomap.c
<<
>>
Prefs
   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_log.h"
  21#include "xfs_trans.h"
  22#include "xfs_sb.h"
  23#include "xfs_ag.h"
  24#include "xfs_alloc.h"
  25#include "xfs_quota.h"
  26#include "xfs_mount.h"
  27#include "xfs_bmap_btree.h"
  28#include "xfs_alloc_btree.h"
  29#include "xfs_ialloc_btree.h"
  30#include "xfs_dinode.h"
  31#include "xfs_inode.h"
  32#include "xfs_inode_item.h"
  33#include "xfs_btree.h"
  34#include "xfs_bmap.h"
  35#include "xfs_rtalloc.h"
  36#include "xfs_error.h"
  37#include "xfs_itable.h"
  38#include "xfs_attr.h"
  39#include "xfs_buf_item.h"
  40#include "xfs_trans_space.h"
  41#include "xfs_utils.h"
  42#include "xfs_iomap.h"
  43#include "xfs_trace.h"
  44#include "xfs_icache.h"
  45#include "xfs_dquot_item.h"
  46#include "xfs_dquot.h"
  47
  48
  49#define XFS_WRITEIO_ALIGN(mp,off)       (((off) >> mp->m_writeio_log) \
  50                                                << mp->m_writeio_log)
  51#define XFS_WRITE_IMAPS         XFS_BMAP_MAX_NMAP
  52
  53STATIC int
  54xfs_iomap_eof_align_last_fsb(
  55        xfs_mount_t     *mp,
  56        xfs_inode_t     *ip,
  57        xfs_extlen_t    extsize,
  58        xfs_fileoff_t   *last_fsb)
  59{
  60        xfs_fileoff_t   new_last_fsb = 0;
  61        xfs_extlen_t    align = 0;
  62        int             eof, error;
  63
  64        if (!XFS_IS_REALTIME_INODE(ip)) {
  65                /*
  66                 * Round up the allocation request to a stripe unit
  67                 * (m_dalign) boundary if the file size is >= stripe unit
  68                 * size, and we are allocating past the allocation eof.
  69                 *
  70                 * If mounted with the "-o swalloc" option the alignment is
  71                 * increased from the strip unit size to the stripe width.
  72                 */
  73                if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
  74                        align = mp->m_swidth;
  75                else if (mp->m_dalign)
  76                        align = mp->m_dalign;
  77
  78                if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
  79                        new_last_fsb = roundup_64(*last_fsb, align);
  80        }
  81
  82        /*
  83         * Always round up the allocation request to an extent boundary
  84         * (when file on a real-time subvolume or has di_extsize hint).
  85         */
  86        if (extsize) {
  87                if (new_last_fsb)
  88                        align = roundup_64(new_last_fsb, extsize);
  89                else
  90                        align = extsize;
  91                new_last_fsb = roundup_64(*last_fsb, align);
  92        }
  93
  94        if (new_last_fsb) {
  95                error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
  96                if (error)
  97                        return error;
  98                if (eof)
  99                        *last_fsb = new_last_fsb;
 100        }
 101        return 0;
 102}
 103
 104STATIC int
 105xfs_alert_fsblock_zero(
 106        xfs_inode_t     *ip,
 107        xfs_bmbt_irec_t *imap)
 108{
 109        xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
 110                        "Access to block zero in inode %llu "
 111                        "start_block: %llx start_off: %llx "
 112                        "blkcnt: %llx extent-state: %x\n",
 113                (unsigned long long)ip->i_ino,
 114                (unsigned long long)imap->br_startblock,
 115                (unsigned long long)imap->br_startoff,
 116                (unsigned long long)imap->br_blockcount,
 117                imap->br_state);
 118        return EFSCORRUPTED;
 119}
 120
 121int
 122xfs_iomap_write_direct(
 123        xfs_inode_t     *ip,
 124        xfs_off_t       offset,
 125        size_t          count,
 126        xfs_bmbt_irec_t *imap,
 127        int             nmaps)
 128{
 129        xfs_mount_t     *mp = ip->i_mount;
 130        xfs_fileoff_t   offset_fsb;
 131        xfs_fileoff_t   last_fsb;
 132        xfs_filblks_t   count_fsb, resaligned;
 133        xfs_fsblock_t   firstfsb;
 134        xfs_extlen_t    extsz, temp;
 135        int             nimaps;
 136        int             bmapi_flag;
 137        int             quota_flag;
 138        int             rt;
 139        xfs_trans_t     *tp;
 140        xfs_bmap_free_t free_list;
 141        uint            qblocks, resblks, resrtextents;
 142        int             committed;
 143        int             error;
 144
 145        error = xfs_qm_dqattach(ip, 0);
 146        if (error)
 147                return XFS_ERROR(error);
 148
 149        rt = XFS_IS_REALTIME_INODE(ip);
 150        extsz = xfs_get_extsz_hint(ip);
 151
 152        offset_fsb = XFS_B_TO_FSBT(mp, offset);
 153        last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 154        if ((offset + count) > XFS_ISIZE(ip)) {
 155                error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
 156                if (error)
 157                        return XFS_ERROR(error);
 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         * Allocate and setup the transaction
 188         */
 189        tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
 190        error = xfs_trans_reserve(tp, resblks,
 191                        XFS_WRITE_LOG_RES(mp), resrtextents,
 192                        XFS_TRANS_PERM_LOG_RES,
 193                        XFS_WRITE_LOG_COUNT);
 194        /*
 195         * Check for running out of space, note: need lock to return
 196         */
 197        if (error) {
 198                xfs_trans_cancel(tp, 0);
 199                return XFS_ERROR(error);
 200        }
 201
 202        xfs_ilock(ip, XFS_ILOCK_EXCL);
 203
 204        error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
 205        if (error)
 206                goto out_trans_cancel;
 207
 208        xfs_trans_ijoin(tp, ip, 0);
 209
 210        bmapi_flag = 0;
 211        if (offset < XFS_ISIZE(ip) || extsz)
 212                bmapi_flag |= XFS_BMAPI_PREALLOC;
 213
 214        /*
 215         * From this point onwards we overwrite the imap pointer that the
 216         * caller gave to us.
 217         */
 218        xfs_bmap_init(&free_list, &firstfsb);
 219        nimaps = 1;
 220        error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag,
 221                                &firstfsb, 0, imap, &nimaps, &free_list);
 222        if (error)
 223                goto out_bmap_cancel;
 224
 225        /*
 226         * Complete the transaction
 227         */
 228        error = xfs_bmap_finish(&tp, &free_list, &committed);
 229        if (error)
 230                goto out_bmap_cancel;
 231        error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 232        if (error)
 233                goto out_unlock;
 234
 235        /*
 236         * Copy any maps to caller's array and return any error.
 237         */
 238        if (nimaps == 0) {
 239                error = XFS_ERROR(ENOSPC);
 240                goto out_unlock;
 241        }
 242
 243        if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
 244                error = xfs_alert_fsblock_zero(ip, imap);
 245
 246out_unlock:
 247        xfs_iunlock(ip, XFS_ILOCK_EXCL);
 248        return error;
 249
 250out_bmap_cancel:
 251        xfs_bmap_cancel(&free_list);
 252        xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
 253out_trans_cancel:
 254        xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 255        goto out_unlock;
 256}
 257
 258/*
 259 * If the caller is doing a write at the end of the file, then extend the
 260 * allocation out to the file system's write iosize.  We clean up any extra
 261 * space left over when the file is closed in xfs_inactive().
 262 *
 263 * If we find we already have delalloc preallocation beyond EOF, don't do more
 264 * preallocation as it it not needed.
 265 */
 266STATIC int
 267xfs_iomap_eof_want_preallocate(
 268        xfs_mount_t     *mp,
 269        xfs_inode_t     *ip,
 270        xfs_off_t       offset,
 271        size_t          count,
 272        xfs_bmbt_irec_t *imap,
 273        int             nimaps,
 274        int             *prealloc)
 275{
 276        xfs_fileoff_t   start_fsb;
 277        xfs_filblks_t   count_fsb;
 278        xfs_fsblock_t   firstblock;
 279        int             n, error, imaps;
 280        int             found_delalloc = 0;
 281
 282        *prealloc = 0;
 283        if (offset + count <= XFS_ISIZE(ip))
 284                return 0;
 285
 286        /*
 287         * If the file is smaller than the minimum prealloc and we are using
 288         * dynamic preallocation, don't do any preallocation at all as it is
 289         * likely this is the only write to the file that is going to be done.
 290         */
 291        if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
 292            XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))
 293                return 0;
 294
 295        /*
 296         * If there are any real blocks past eof, then don't
 297         * do any speculative allocation.
 298         */
 299        start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
 300        count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 301        while (count_fsb > 0) {
 302                imaps = nimaps;
 303                firstblock = NULLFSBLOCK;
 304                error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
 305                                       0);
 306                if (error)
 307                        return error;
 308                for (n = 0; n < imaps; n++) {
 309                        if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
 310                            (imap[n].br_startblock != DELAYSTARTBLOCK))
 311                                return 0;
 312                        start_fsb += imap[n].br_blockcount;
 313                        count_fsb -= imap[n].br_blockcount;
 314
 315                        if (imap[n].br_startblock == DELAYSTARTBLOCK)
 316                                found_delalloc = 1;
 317                }
 318        }
 319        if (!found_delalloc)
 320                *prealloc = 1;
 321        return 0;
 322}
 323
 324/*
 325 * Determine the initial size of the preallocation. We are beyond the current
 326 * EOF here, but we need to take into account whether this is a sparse write or
 327 * an extending write when determining the preallocation size.  Hence we need to
 328 * look up the extent that ends at the current write offset and use the result
 329 * to determine the preallocation size.
 330 *
 331 * If the extent is a hole, then preallocation is essentially disabled.
 332 * Otherwise we take the size of the preceeding data extent as the basis for the
 333 * preallocation size. If the size of the extent is greater than half the
 334 * maximum extent length, then use the current offset as the basis. This ensures
 335 * that for large files the preallocation size always extends to MAXEXTLEN
 336 * rather than falling short due to things like stripe unit/width alignment of
 337 * real extents.
 338 */
 339STATIC xfs_fsblock_t
 340xfs_iomap_eof_prealloc_initial_size(
 341        struct xfs_mount        *mp,
 342        struct xfs_inode        *ip,
 343        xfs_off_t               offset,
 344        xfs_bmbt_irec_t         *imap,
 345        int                     nimaps)
 346{
 347        xfs_fileoff_t   start_fsb;
 348        int             imaps = 1;
 349        int             error;
 350
 351        ASSERT(nimaps >= imaps);
 352
 353        /* if we are using a specific prealloc size, return now */
 354        if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
 355                return 0;
 356
 357        /* If the file is small, then use the minimum prealloc */
 358        if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign))
 359                return 0;
 360
 361        /*
 362         * As we write multiple pages, the offset will always align to the
 363         * start of a page and hence point to a hole at EOF. i.e. if the size is
 364         * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
 365         * will return FSB 1. Hence if there are blocks in the file, we want to
 366         * point to the block prior to the EOF block and not the hole that maps
 367         * directly at @offset.
 368         */
 369        start_fsb = XFS_B_TO_FSB(mp, offset);
 370        if (start_fsb)
 371                start_fsb--;
 372        error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
 373        if (error)
 374                return 0;
 375
 376        ASSERT(imaps == 1);
 377        if (imap[0].br_startblock == HOLESTARTBLOCK)
 378                return 0;
 379        if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
 380                return imap[0].br_blockcount << 1;
 381        return XFS_B_TO_FSB(mp, offset);
 382}
 383
 384STATIC bool
 385xfs_quota_need_throttle(
 386        struct xfs_inode *ip,
 387        int type,
 388        xfs_fsblock_t alloc_blocks)
 389{
 390        struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
 391
 392        if (!dq || !xfs_this_quota_on(ip->i_mount, type))
 393                return false;
 394
 395        /* no hi watermark, no throttle */
 396        if (!dq->q_prealloc_hi_wmark)
 397                return false;
 398
 399        /* under the lo watermark, no throttle */
 400        if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
 401                return false;
 402
 403        return true;
 404}
 405
 406STATIC void
 407xfs_quota_calc_throttle(
 408        struct xfs_inode *ip,
 409        int type,
 410        xfs_fsblock_t *qblocks,
 411        int *qshift)
 412{
 413        int64_t freesp;
 414        int shift = 0;
 415        struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
 416
 417        /* over hi wmark, squash the prealloc completely */
 418        if (dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
 419                *qblocks = 0;
 420                return;
 421        }
 422
 423        freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
 424        if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
 425                shift = 2;
 426                if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
 427                        shift += 2;
 428                if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
 429                        shift += 2;
 430        }
 431
 432        /* only overwrite the throttle values if we are more aggressive */
 433        if ((freesp >> shift) < (*qblocks >> *qshift)) {
 434                *qblocks = freesp;
 435                *qshift = shift;
 436        }
 437}
 438
 439/*
 440 * If we don't have a user specified preallocation size, dynamically increase
 441 * the preallocation size as the size of the file grows. Cap the maximum size
 442 * at a single extent or less if the filesystem is near full. The closer the
 443 * filesystem is to full, the smaller the maximum prealocation.
 444 */
 445STATIC xfs_fsblock_t
 446xfs_iomap_prealloc_size(
 447        struct xfs_mount        *mp,
 448        struct xfs_inode        *ip,
 449        xfs_off_t               offset,
 450        struct xfs_bmbt_irec    *imap,
 451        int                     nimaps)
 452{
 453        xfs_fsblock_t           alloc_blocks = 0;
 454        int                     shift = 0;
 455        int64_t                 freesp;
 456        xfs_fsblock_t           qblocks;
 457        int                     qshift = 0;
 458
 459        alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
 460                                                           imap, nimaps);
 461        if (!alloc_blocks)
 462                goto check_writeio;
 463        qblocks = alloc_blocks;
 464
 465        /*
 466         * MAXEXTLEN is not a power of two value but we round the prealloc down
 467         * to the nearest power of two value after throttling. To prevent the
 468         * round down from unconditionally reducing the maximum supported prealloc
 469         * size, we round up first, apply appropriate throttling, round down and
 470         * cap the value to MAXEXTLEN.
 471         */
 472        alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
 473                                       alloc_blocks);
 474
 475        xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 476        freesp = mp->m_sb.sb_fdblocks;
 477        if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
 478                shift = 2;
 479                if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
 480                        shift++;
 481                if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
 482                        shift++;
 483                if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
 484                        shift++;
 485                if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
 486                        shift++;
 487        }
 488
 489        /*
 490         * Check each quota to cap the prealloc size and provide a shift
 491         * value to throttle with.
 492         */
 493        if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
 494                xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift);
 495        if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
 496                xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift);
 497        if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
 498                xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift);
 499
 500        /*
 501         * The final prealloc size is set to the minimum of free space available
 502         * in each of the quotas and the overall filesystem.
 503         *
 504         * The shift throttle value is set to the maximum value as determined by
 505         * the global low free space values and per-quota low free space values.
 506         */
 507        alloc_blocks = MIN(alloc_blocks, qblocks);
 508        shift = MAX(shift, qshift);
 509
 510        if (shift)
 511                alloc_blocks >>= shift;
 512        /*
 513         * rounddown_pow_of_two() returns an undefined result if we pass in
 514         * alloc_blocks = 0.
 515         */
 516        if (alloc_blocks)
 517                alloc_blocks = rounddown_pow_of_two(alloc_blocks);
 518        if (alloc_blocks > MAXEXTLEN)
 519                alloc_blocks = MAXEXTLEN;
 520
 521        /*
 522         * If we are still trying to allocate more space than is
 523         * available, squash the prealloc hard. This can happen if we
 524         * have a large file on a small filesystem and the above
 525         * lowspace thresholds are smaller than MAXEXTLEN.
 526         */
 527        while (alloc_blocks && alloc_blocks >= freesp)
 528                alloc_blocks >>= 4;
 529
 530check_writeio:
 531        if (alloc_blocks < mp->m_writeio_blocks)
 532                alloc_blocks = mp->m_writeio_blocks;
 533
 534        trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
 535                                      mp->m_writeio_blocks);
 536
 537        return alloc_blocks;
 538}
 539
 540int
 541xfs_iomap_write_delay(
 542        xfs_inode_t     *ip,
 543        xfs_off_t       offset,
 544        size_t          count,
 545        xfs_bmbt_irec_t *ret_imap)
 546{
 547        xfs_mount_t     *mp = ip->i_mount;
 548        xfs_fileoff_t   offset_fsb;
 549        xfs_fileoff_t   last_fsb;
 550        xfs_off_t       aligned_offset;
 551        xfs_fileoff_t   ioalign;
 552        xfs_extlen_t    extsz;
 553        int             nimaps;
 554        xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
 555        int             prealloc;
 556        int             error;
 557
 558        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 559
 560        /*
 561         * Make sure that the dquots are there. This doesn't hold
 562         * the ilock across a disk read.
 563         */
 564        error = xfs_qm_dqattach_locked(ip, 0);
 565        if (error)
 566                return XFS_ERROR(error);
 567
 568        extsz = xfs_get_extsz_hint(ip);
 569        offset_fsb = XFS_B_TO_FSBT(mp, offset);
 570
 571        error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
 572                                imap, XFS_WRITE_IMAPS, &prealloc);
 573        if (error)
 574                return error;
 575
 576retry:
 577        if (prealloc) {
 578                xfs_fsblock_t   alloc_blocks;
 579
 580                alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
 581                                                       XFS_WRITE_IMAPS);
 582
 583                aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
 584                ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
 585                last_fsb = ioalign + alloc_blocks;
 586        } else {
 587                last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 588        }
 589
 590        if (prealloc || extsz) {
 591                error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
 592                if (error)
 593                        return error;
 594        }
 595
 596        /*
 597         * Make sure preallocation does not create extents beyond the range we
 598         * actually support in this filesystem.
 599         */
 600        if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
 601                last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 602
 603        ASSERT(last_fsb > offset_fsb);
 604
 605        nimaps = XFS_WRITE_IMAPS;
 606        error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
 607                                imap, &nimaps, XFS_BMAPI_ENTIRE);
 608        switch (error) {
 609        case 0:
 610        case ENOSPC:
 611        case EDQUOT:
 612                break;
 613        default:
 614                return XFS_ERROR(error);
 615        }
 616
 617        /*
 618         * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
 619         * without EOF preallocation.
 620         */
 621        if (nimaps == 0) {
 622                trace_xfs_delalloc_enospc(ip, offset, count);
 623                if (prealloc) {
 624                        prealloc = 0;
 625                        error = 0;
 626                        goto retry;
 627                }
 628                return XFS_ERROR(error ? error : ENOSPC);
 629        }
 630
 631        if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
 632                return xfs_alert_fsblock_zero(ip, &imap[0]);
 633
 634        /*
 635         * Tag the inode as speculatively preallocated so we can reclaim this
 636         * space on demand, if necessary.
 637         */
 638        if (prealloc)
 639                xfs_inode_set_eofblocks_tag(ip);
 640
 641        *ret_imap = imap[0];
 642        return 0;
 643}
 644
 645/*
 646 * Pass in a delayed allocate extent, convert it to real extents;
 647 * return to the caller the extent we create which maps on top of
 648 * the originating callers request.
 649 *
 650 * Called without a lock on the inode.
 651 *
 652 * We no longer bother to look at the incoming map - all we have to
 653 * guarantee is that whatever we allocate fills the required range.
 654 */
 655int
 656xfs_iomap_write_allocate(
 657        xfs_inode_t     *ip,
 658        xfs_off_t       offset,
 659        size_t          count,
 660        xfs_bmbt_irec_t *imap)
 661{
 662        xfs_mount_t     *mp = ip->i_mount;
 663        xfs_fileoff_t   offset_fsb, last_block;
 664        xfs_fileoff_t   end_fsb, map_start_fsb;
 665        xfs_fsblock_t   first_block;
 666        xfs_bmap_free_t free_list;
 667        xfs_filblks_t   count_fsb;
 668        xfs_trans_t     *tp;
 669        int             nimaps, committed;
 670        int             error = 0;
 671        int             nres;
 672
 673        /*
 674         * Make sure that the dquots are there.
 675         */
 676        error = xfs_qm_dqattach(ip, 0);
 677        if (error)
 678                return XFS_ERROR(error);
 679
 680        offset_fsb = XFS_B_TO_FSBT(mp, offset);
 681        count_fsb = imap->br_blockcount;
 682        map_start_fsb = imap->br_startoff;
 683
 684        XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 685
 686        while (count_fsb != 0) {
 687                /*
 688                 * Set up a transaction with which to allocate the
 689                 * backing store for the file.  Do allocations in a
 690                 * loop until we get some space in the range we are
 691                 * interested in.  The other space that might be allocated
 692                 * is in the delayed allocation extent on which we sit
 693                 * but before our buffer starts.
 694                 */
 695
 696                nimaps = 0;
 697                while (nimaps == 0) {
 698                        tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
 699                        tp->t_flags |= XFS_TRANS_RESERVE;
 700                        nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 701                        error = xfs_trans_reserve(tp, nres,
 702                                        XFS_WRITE_LOG_RES(mp),
 703                                        0, XFS_TRANS_PERM_LOG_RES,
 704                                        XFS_WRITE_LOG_COUNT);
 705                        if (error) {
 706                                xfs_trans_cancel(tp, 0);
 707                                return XFS_ERROR(error);
 708                        }
 709                        xfs_ilock(ip, XFS_ILOCK_EXCL);
 710                        xfs_trans_ijoin(tp, ip, 0);
 711
 712                        xfs_bmap_init(&free_list, &first_block);
 713
 714                        /*
 715                         * it is possible that the extents have changed since
 716                         * we did the read call as we dropped the ilock for a
 717                         * while. We have to be careful about truncates or hole
 718                         * punchs here - we are not allowed to allocate
 719                         * non-delalloc blocks here.
 720                         *
 721                         * The only protection against truncation is the pages
 722                         * for the range we are being asked to convert are
 723                         * locked and hence a truncate will block on them
 724                         * first.
 725                         *
 726                         * As a result, if we go beyond the range we really
 727                         * need and hit an delalloc extent boundary followed by
 728                         * a hole while we have excess blocks in the map, we
 729                         * will fill the hole incorrectly and overrun the
 730                         * transaction reservation.
 731                         *
 732                         * Using a single map prevents this as we are forced to
 733                         * check each map we look for overlap with the desired
 734                         * range and abort as soon as we find it. Also, given
 735                         * that we only return a single map, having one beyond
 736                         * what we can return is probably a bit silly.
 737                         *
 738                         * We also need to check that we don't go beyond EOF;
 739                         * this is a truncate optimisation as a truncate sets
 740                         * the new file size before block on the pages we
 741                         * currently have locked under writeback. Because they
 742                         * are about to be tossed, we don't need to write them
 743                         * back....
 744                         */
 745                        nimaps = 1;
 746                        end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
 747                        error = xfs_bmap_last_offset(NULL, ip, &last_block,
 748                                                        XFS_DATA_FORK);
 749                        if (error)
 750                                goto trans_cancel;
 751
 752                        last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 753                        if ((map_start_fsb + count_fsb) > last_block) {
 754                                count_fsb = last_block - map_start_fsb;
 755                                if (count_fsb == 0) {
 756                                        error = EAGAIN;
 757                                        goto trans_cancel;
 758                                }
 759                        }
 760
 761                        /*
 762                         * From this point onwards we overwrite the imap
 763                         * pointer that the caller gave to us.
 764                         */
 765                        error = xfs_bmapi_write(tp, ip, map_start_fsb,
 766                                                count_fsb,
 767                                                XFS_BMAPI_STACK_SWITCH,
 768                                                &first_block, 1,
 769                                                imap, &nimaps, &free_list);
 770                        if (error)
 771                                goto trans_cancel;
 772
 773                        error = xfs_bmap_finish(&tp, &free_list, &committed);
 774                        if (error)
 775                                goto trans_cancel;
 776
 777                        error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 778                        if (error)
 779                                goto error0;
 780
 781                        xfs_iunlock(ip, XFS_ILOCK_EXCL);
 782                }
 783
 784                /*
 785                 * See if we were able to allocate an extent that
 786                 * covers at least part of the callers request
 787                 */
 788                if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
 789                        return xfs_alert_fsblock_zero(ip, imap);
 790
 791                if ((offset_fsb >= imap->br_startoff) &&
 792                    (offset_fsb < (imap->br_startoff +
 793                                   imap->br_blockcount))) {
 794                        XFS_STATS_INC(xs_xstrat_quick);
 795                        return 0;
 796                }
 797
 798                /*
 799                 * So far we have not mapped the requested part of the
 800                 * file, just surrounding data, try again.
 801                 */
 802                count_fsb -= imap->br_blockcount;
 803                map_start_fsb = imap->br_startoff + imap->br_blockcount;
 804        }
 805
 806trans_cancel:
 807        xfs_bmap_cancel(&free_list);
 808        xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
 809error0:
 810        xfs_iunlock(ip, XFS_ILOCK_EXCL);
 811        return XFS_ERROR(error);
 812}
 813
 814int
 815xfs_iomap_write_unwritten(
 816        xfs_inode_t     *ip,
 817        xfs_off_t       offset,
 818        size_t          count)
 819{
 820        xfs_mount_t     *mp = ip->i_mount;
 821        xfs_fileoff_t   offset_fsb;
 822        xfs_filblks_t   count_fsb;
 823        xfs_filblks_t   numblks_fsb;
 824        xfs_fsblock_t   firstfsb;
 825        int             nimaps;
 826        xfs_trans_t     *tp;
 827        xfs_bmbt_irec_t imap;
 828        xfs_bmap_free_t free_list;
 829        xfs_fsize_t     i_size;
 830        uint            resblks;
 831        int             committed;
 832        int             error;
 833
 834        trace_xfs_unwritten_convert(ip, offset, count);
 835
 836        offset_fsb = XFS_B_TO_FSBT(mp, offset);
 837        count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 838        count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 839
 840        /*
 841         * Reserve enough blocks in this transaction for two complete extent
 842         * btree splits.  We may be converting the middle part of an unwritten
 843         * extent and in this case we will insert two new extents in the btree
 844         * each of which could cause a full split.
 845         *
 846         * This reservation amount will be used in the first call to
 847         * xfs_bmbt_split() to select an AG with enough space to satisfy the
 848         * rest of the operation.
 849         */
 850        resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 851
 852        do {
 853                /*
 854                 * set up a transaction to convert the range of extents
 855                 * from unwritten to real. Do allocations in a loop until
 856                 * we have covered the range passed in.
 857                 *
 858                 * Note that we open code the transaction allocation here
 859                 * to pass KM_NOFS--we can't risk to recursing back into
 860                 * the filesystem here as we might be asked to write out
 861                 * the same inode that we complete here and might deadlock
 862                 * on the iolock.
 863                 */
 864                sb_start_intwrite(mp->m_super);
 865                tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
 866                tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT;
 867                error = xfs_trans_reserve(tp, resblks,
 868                                XFS_WRITE_LOG_RES(mp), 0,
 869                                XFS_TRANS_PERM_LOG_RES,
 870                                XFS_WRITE_LOG_COUNT);
 871                if (error) {
 872                        xfs_trans_cancel(tp, 0);
 873                        return XFS_ERROR(error);
 874                }
 875
 876                xfs_ilock(ip, XFS_ILOCK_EXCL);
 877                xfs_trans_ijoin(tp, ip, 0);
 878
 879                /*
 880                 * Modify the unwritten extent state of the buffer.
 881                 */
 882                xfs_bmap_init(&free_list, &firstfsb);
 883                nimaps = 1;
 884                error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
 885                                  XFS_BMAPI_CONVERT, &firstfsb,
 886                                  1, &imap, &nimaps, &free_list);
 887                if (error)
 888                        goto error_on_bmapi_transaction;
 889
 890                /*
 891                 * Log the updated inode size as we go.  We have to be careful
 892                 * to only log it up to the actual write offset if it is
 893                 * halfway into a block.
 894                 */
 895                i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
 896                if (i_size > offset + count)
 897                        i_size = offset + count;
 898
 899                i_size = xfs_new_eof(ip, i_size);
 900                if (i_size) {
 901                        ip->i_d.di_size = i_size;
 902                        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 903                }
 904
 905                error = xfs_bmap_finish(&tp, &free_list, &committed);
 906                if (error)
 907                        goto error_on_bmapi_transaction;
 908
 909                error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 910                xfs_iunlock(ip, XFS_ILOCK_EXCL);
 911                if (error)
 912                        return XFS_ERROR(error);
 913
 914                if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
 915                        return xfs_alert_fsblock_zero(ip, &imap);
 916
 917                if ((numblks_fsb = imap.br_blockcount) == 0) {
 918                        /*
 919                         * The numblks_fsb value should always get
 920                         * smaller, otherwise the loop is stuck.
 921                         */
 922                        ASSERT(imap.br_blockcount);
 923                        break;
 924                }
 925                offset_fsb += numblks_fsb;
 926                count_fsb -= numblks_fsb;
 927        } while (count_fsb > 0);
 928
 929        return 0;
 930
 931error_on_bmapi_transaction:
 932        xfs_bmap_cancel(&free_list);
 933        xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
 934        xfs_iunlock(ip, XFS_ILOCK_EXCL);
 935        return XFS_ERROR(error);
 936}
 937
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.