lustre/lvfs/fsfilt_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  lustre/lib/fsfilt_ext3.c
   5  *  Lustre filesystem abstraction routines
   6  *
   7  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   8  *   Author: Andreas Dilger <adilger@clusterfs.com>
   9  *
  10  *   This file is part of Lustre, http://www.lustre.org.
  11  *
  12  *   Lustre is free software; you can redistribute it and/or
  13  *   modify it under the terms of version 2 of the GNU General Public
  14  *   License as published by the Free Software Foundation.
  15  *
  16  *   Lustre is distributed in the hope that it will be useful,
  17  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  *   GNU General Public License for more details.
  20  *
  21  *   You should have received a copy of the GNU General Public License
  22  *   along with Lustre; if not, write to the Free Software
  23  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  24  */
  25
  26 #define DEBUG_SUBSYSTEM S_FILTER
  27
  28 #include <linux/init.h>
  29 #include <linux/module.h>
  30 #include <linux/fs.h>
  31 #include <linux/jbd.h>
  32 #include <linux/slab.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/ext3_fs.h>
  36 #include <linux/ext3_jbd.h>
  37 #include <linux/version.h>
  38 #include <linux/bitops.h>
  39 #include <linux/quota.h>
  40 #include <linux/quotaio_v1.h>
  41 #include <linux/quotaio_v2.h>
  42 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  43 #include <linux/ext3_xattr.h>
  44 #else
  45 #include <ext3/xattr.h>
  46 #endif
  47
  48 #include <libcfs/kp30.h>
  49 #include <linux/lustre_fsfilt.h>
  50 #include <linux/obd.h>
  51 #include <linux/obd_class.h>
  52 #include <linux/lustre_quota.h>
  53 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  54 #include <linux/iobuf.h>
  55 #endif
  56
  57 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  58 #include <linux/ext3_extents.h>
  59 #endif
  60
  61 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,7))
  62 # define lock_24kernel() lock_kernel()
  63 # define unlock_24kernel() unlock_kernel()
  64 #else
  65 # define lock_24kernel() do {} while (0)
  66 # define unlock_24kernel() do {} while (0)
  67 #endif
  68
  69 static kmem_cache_t *fcb_cache;
  70
  71 struct fsfilt_cb_data {
  72         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  73         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  74         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  75         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  76         void *cb_data;                  /* MDS/OST completion function data */
  77 };
  78
  79 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
  80 #define EXT3_XATTR_INDEX_TRUSTED        4
  81 #endif
  82 #define XATTR_LUSTRE_MDS_LOV_EA         "lov"
  83
  84 /*
  85  * We don't currently need any additional blocks for rmdir and
  86  * unlink transactions because we are storing the OST oa_id inside
  87  * the inode (which we will be changing anyways as part of this
  88  * transaction).
  89  */
  90 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
  91                                int logs)
  92 {
  93         /* For updates to the last received file */
  94         int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS;
  95         journal_t *journal;
  96         void *handle;
  97
  98         if (current->journal_info) {
  99                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 100                        current->journal_info);
 101                 goto journal_start;
 102         }
 103
 104         switch(op) {
 105         case FSFILT_OP_RMDIR:
 106         case FSFILT_OP_UNLINK:
 107                 /* delete one file + create/update logs for each stripe */
 108                 nblocks += EXT3_DELETE_TRANS_BLOCKS;
 109                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 110                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 111                 break;
 112         case FSFILT_OP_RENAME:
 113                 /* modify additional directory */
 114                 nblocks += EXT3_SINGLEDATA_TRANS_BLOCKS;
 115                 /* no break */
 116         case FSFILT_OP_SYMLINK:
 117                 /* additional block + block bitmap + GDT for long symlink */
 118                 nblocks += 3;
 119                 /* no break */
 120         case FSFILT_OP_CREATE: {
 121 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 122                 static int warned;
 123                 if (!warned) {
 124                         if (!test_opt(inode->i_sb, EXTENTS)) {
 125                                 warned = 1;
 126                         } else if (((EXT3_I(inode)->i_flags &
 127                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 128                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 129                                 CWARN("extent-mapped directory found - contact "
 130                                       "CFS: support@clusterfs.com\n");
 131                                 warned = 1;
 132                         }
 133                 }
 134 #endif
 135                 /* no break */
 136         }
 137         case FSFILT_OP_MKDIR:
 138         case FSFILT_OP_MKNOD:
 139                 /* modify one inode + block bitmap + GDT */
 140                 nblocks += 3;
 141                 /* no break */
 142         case FSFILT_OP_LINK:
 143                 /* modify parent directory */
 144                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 145                         EXT3_DATA_TRANS_BLOCKS;
 146                 /* create/update logs for each stripe */
 147                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 148                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 149                 break;
 150         case FSFILT_OP_SETATTR:
 151                 /* Setattr on inode */
 152                 nblocks += 1;
 153                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 154                         EXT3_DATA_TRANS_BLOCKS;
 155                 /* quota chown log for each stripe */
 156                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 157                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 158                 break;
 159         case FSFILT_OP_CANCEL_UNLINK:
 160                 /* blocks for log header bitmap update OR
 161                  * blocks for catalog header bitmap update + unlink of logs */
 162                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 163                         EXT3_DELETE_TRANS_BLOCKS * logs;
 164                 break;
 165         default: CERROR("unknown transaction start op %d\n", op);
 166                 LBUG();
 167         }
 168
 169         LASSERT(current->journal_info == desc_private);
 170         journal = EXT3_SB(inode->i_sb)->s_journal;
 171         if (nblocks > journal->j_max_transaction_buffers) {
 172                 CERROR("too many credits %d for op %ux%u using %d instead\n",
 173                        nblocks, op, logs, journal->j_max_transaction_buffers);
 174                 nblocks = journal->j_max_transaction_buffers;
 175         }
 176
 177  journal_start:
 178         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 179         lock_24kernel();
 180         handle = journal_start(EXT3_JOURNAL(inode), nblocks);
 181         unlock_24kernel();
 182
 183         if (!IS_ERR(handle))
 184                 LASSERT(current->journal_info == handle);
 185         else
 186                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 187                        op, nblocks, PTR_ERR(handle));
 188         return handle;
 189 }
 190
 191 /*
 192  * Calculate the number of buffer credits needed to write multiple pages in
 193  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 194  * doesn't have a nice API for calculating this sort of thing in advance.
 195  *
 196  * See comment above ext3_writepage_trans_blocks for details.  We assume
 197  * no data journaling is being done, but it does allow for all of the pages
 198  * being non-contiguous.  If we are guaranteed contiguous pages we could
 199  * reduce the number of (d)indirect blocks a lot.
 200  *
 201  * With N blocks per page and P pages, for each inode we have at most:
 202  * N*P indirect
 203  * min(N*P, blocksize/4 + 1) dindirect blocks
 204  * niocount tindirect
 205  *
 206  * For the entire filesystem, we have at most:
 207  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 208  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 209  * objcount inode blocks
 210  * 1 superblock
 211  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 212  *
 213  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 214  */
 215 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 216                                       int niocount, struct niobuf_local *nb)
 217 {
 218         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 219         __u64 next_indir;
 220         const int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
 221         int nbitmaps = 0, ngdblocks;
 222         int needed = objcount + 1; /* inodes + superblock */
 223         int i, j;
 224
 225         for (i = 0, j = 0; i < objcount; i++, fso++) {
 226                 /* two or more dindirect blocks in case we cross boundary */
 227                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 228                                     nb[j].offset) >>
 229                                    sb->s_blocksize_bits) /
 230                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 231                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 232
 233                 /* leaf, indirect, tindirect blocks for first block */
 234                 nbitmaps += blockpp + 2;
 235
 236                 j += fso->fso_bufcnt;
 237         }
 238
 239         next_indir = nb[0].offset +
 240                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 241         for (i = 1; i < niocount; i++) {
 242                 if (nb[i].offset >= next_indir) {
 243                         nbitmaps++;     /* additional indirect */
 244                         next_indir = nb[i].offset +
 245                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 246                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 247                         nbitmaps++;     /* additional indirect */
 248                 }
 249                 nbitmaps += blockpp;    /* each leaf in different group? */
 250         }
 251
 252         ngdblocks = nbitmaps;
 253         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 254                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 255         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 256                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 257
 258         needed += nbitmaps + ngdblocks;
 259
 260         /* last_rcvd update */
 261         needed += EXT3_DATA_TRANS_BLOCKS;
 262
 263 #if defined(CONFIG_QUOTA)
 264         /* We assume that there will be 1 bit set in s_dquot.flags for each
 265          * quota file that is active.  This is at least true for now.
 266          */
 267         needed += hweight32(sb_any_quota_enabled(sb)) *
 268                 EXT3_SINGLEDATA_TRANS_BLOCKS;
 269 #endif
 270
 271         return needed;
 272 }
 273
 274 /* We have to start a huge journal transaction here to hold all of the
 275  * metadata for the pages being written here.  This is necessitated by
 276  * the fact that we do lots of prepare_write operations before we do
 277  * any of the matching commit_write operations, so even if we split
 278  * up to use "smaller" transactions none of them could complete until
 279  * all of them were opened.  By having a single journal transaction,
 280  * we eliminate duplicate reservations for common blocks like the
 281  * superblock and group descriptors or bitmaps.
 282  *
 283  * We will start the transaction here, but each prepare_write will
 284  * add a refcount to the transaction, and each commit_write will
 285  * remove a refcount.  The transaction will be closed when all of
 286  * the pages have been written.
 287  */
 288 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 289                                    int niocount, struct niobuf_local *nb,
 290                                    void *desc_private, int logs)
 291 {
 292         journal_t *journal;
 293         handle_t *handle;
 294         int needed;
 295         ENTRY;
 296
 297         LASSERT(current->journal_info == desc_private);
 298         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 299         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 300
 301         /* The number of blocks we could _possibly_ dirty can very large.
 302          * We reduce our request if it is absurd (and we couldn't get that
 303          * many credits for a single handle anyways).
 304          *
 305          * At some point we have to limit the size of I/Os sent at one time,
 306          * increase the size of the journal, or we have to calculate the
 307          * actual journal requirements more carefully by checking all of
 308          * the blocks instead of being maximally pessimistic.  It remains to
 309          * be seen if this is a real problem or not.
 310          */
 311         if (needed > journal->j_max_transaction_buffers) {
 312                 CERROR("want too many journal credits (%d) using %d instead\n",
 313                        needed, journal->j_max_transaction_buffers);
 314                 needed = journal->j_max_transaction_buffers;
 315         }
 316
 317         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 318         lock_24kernel();
 319         handle = journal_start(journal, needed);
 320         unlock_24kernel();
 321         if (IS_ERR(handle)) {
 322                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 323                        PTR_ERR(handle));
 324         } else {
 325                 LASSERT(handle->h_buffer_credits >= needed);
 326                 LASSERT(current->journal_info == handle);
 327         }
 328
 329         RETURN(handle);
 330 }
 331
 332 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 333 {
 334         int rc;
 335         handle_t *handle = h;
 336
 337         LASSERT(current->journal_info == handle);
 338         if (force_sync)
 339                 handle->h_sync = 1; /* recovery likes this */
 340
 341         lock_24kernel();
 342         rc = journal_stop(handle);
 343         unlock_24kernel();
 344
 345         return rc;
 346 }
 347
 348 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 349                                     void **wait_handle)
 350 {
 351         unsigned long tid;
 352         transaction_t *transaction;
 353 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 354         unsigned long rtid;
 355 #endif
 356         handle_t *handle = h;
 357         journal_t *journal;
 358         int rc;
 359
 360         LASSERT(current->journal_info == handle);
 361
 362         lock_kernel();
 363         transaction = handle->h_transaction;
 364         journal = transaction->t_journal;
 365         tid = transaction->t_tid;
 366         /* we don't want to be blocked */
 367         handle->h_sync = 0;
 368         rc = journal_stop(handle);
 369         if (rc) {
 370                 CERROR("error while stopping transaction: %d\n", rc);
 371                 unlock_kernel();
 372                 return rc;
 373         }
 374 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 375         rtid = log_start_commit(journal, transaction);
 376         if (rtid != tid)
 377                 CERROR("strange race: %lu != %lu\n",
 378                        (unsigned long) tid, (unsigned long) rtid);
 379 #else
 380         log_start_commit(journal, transaction->t_tid);
 381 #endif
 382         unlock_kernel();
 383
 384         *wait_handle = (void *) tid;
 385         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 386         return 0;
 387 }
 388
 389 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 390 {
 391         journal_t *journal = EXT3_JOURNAL(inode);
 392         tid_t tid = (tid_t)(long)h;
 393
 394         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 395         if (unlikely(is_journal_aborted(journal)))
 396                 return -EIO;
 397
 398         log_wait_commit(EXT3_JOURNAL(inode), tid);
 399
 400         if (unlikely(is_journal_aborted(journal)))
 401                 return -EIO;
 402         return 0;
 403 }
 404
 405 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 406                                struct iattr *iattr, int do_trunc)
 407 {
 408         struct inode *inode = dentry->d_inode;
 409         int rc;
 410
 411         lock_kernel();
 412
 413         /* A _really_ horrible hack to avoid removing the data stored
 414          * in the block pointers; this is really the "small" stripe MD data.
 415          * We can avoid further hackery by virtue of the MDS file size being
 416          * zero all the time (which doesn't invoke block truncate at unlink
 417          * time), so we assert we never change the MDS file size from zero. */
 418         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 419                 /* ATTR_SIZE would invoke truncate: clear it */
 420                 iattr->ia_valid &= ~ATTR_SIZE;
 421                 EXT3_I(inode)->i_disksize = inode->i_size = iattr->ia_size;
 422
 423                 /* make sure _something_ gets set - so new inode
 424                  * goes to disk (probably won't work over XFS */
 425                 if (!(iattr->ia_valid & (ATTR_MODE | ATTR_MTIME | ATTR_CTIME))){
 426                         iattr->ia_valid |= ATTR_MODE;
 427                         iattr->ia_mode = inode->i_mode;
 428                 }
 429         }
 430
 431         /* Don't allow setattr to change file type */
 432         iattr->ia_mode = (inode->i_mode & S_IFMT)|(iattr->ia_mode & ~S_IFMT);
 433
 434         /* We set these flags on the client, but have already checked perms
 435          * so don't confuse inode_change_ok. */
 436         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 437
 438         if (inode->i_op->setattr) {
 439                 rc = inode->i_op->setattr(dentry, iattr);
 440         } else {
 441                 rc = inode_change_ok(inode, iattr);
 442                 if (!rc)
 443                         rc = inode_setattr(inode, iattr);
 444         }
 445
 446         unlock_kernel();
 447
 448         return rc;
 449 }
 450
 451 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 452                                  unsigned int cmd, unsigned long arg)
 453 {
 454         int rc = 0;
 455         ENTRY;
 456
 457         if (inode->i_fop->ioctl)
 458                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 459         else
 460                 RETURN(-ENOTTY);
 461
 462         RETURN(rc);
 463 }
 464
 465 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 466                               void *lmm, int lmm_size)
 467 {
 468         int rc;
 469
 470         LASSERT(down_trylock(&inode->i_sem) != 0);
 471
 472         if (EXT3_I(inode)->i_file_acl /* || large inode EA flag */)
 473                 CWARN("setting EA on %lu/%u again... interesting\n",
 474                        inode->i_ino, inode->i_generation);
 475
 476         lock_kernel();
 477         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 478                                    XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size, 0);
 479
 480         unlock_kernel();
 481
 482         if (rc)
 483                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 484                        inode->i_ino, rc);
 485         return rc;
 486 }
 487
 488 /* Must be called with i_sem held */
 489 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size)
 490 {
 491         int rc;
 492
 493         LASSERT(down_trylock(&inode->i_sem) != 0);
 494         lock_kernel();
 495
 496         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 497                             XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size);
 498         unlock_kernel();
 499
 500         /* This gives us the MD size */
 501         if (lmm == NULL)
 502                 return (rc == -ENODATA) ? 0 : rc;
 503
 504         if (rc < 0) {
 505                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 506                        EXT3_XATTR_INDEX_TRUSTED, XATTR_LUSTRE_MDS_LOV_EA,
 507                        inode->i_ino, rc);
 508                 memset(lmm, 0, lmm_size);
 509                 return (rc == -ENODATA) ? 0 : rc;
 510         }
 511
 512         return rc;
 513 }
 514
 515 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
 516 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 517 {
 518         submit_bio(rw, bio);
 519         return 0;
 520 }
 521 #else
 522 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct kiobuf *bio)
 523 {
 524         int rc, blk_per_page;
 525
 526         rc = brw_kiovec(rw, 1, &bio, inode->i_dev,
 527                         KIOBUF_GET_BLOCKS(bio), 1 << inode->i_blkbits);
 528         /*
 529          * brw_kiovec() returns number of bytes actually written. If error
 530          * occurred after something was written, error code is returned though
 531          * kiobuf->errno. (See bug 6854.)
 532          */
 533
 534         blk_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
 535
 536         if (rc != (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page) {
 537                 CERROR("short write?  expected %d, wrote %d (%d)\n",
 538                        (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
 539                        rc, bio->errno);
 540         }
 541         if (bio->errno != 0) {
 542                 CERROR("IO error. Wrote %d of %d (%d)\n",
 543                        rc,
 544                        (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
 545                        bio->errno);
 546                 rc = bio->errno;
 547         }
 548
 549         return rc;
 550 }
 551 #endif
 552
 553 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 554                                     loff_t *off)
 555 {
 556         struct inode *inode = file->f_dentry->d_inode;
 557         int rc = 0;
 558
 559         if (S_ISREG(inode->i_mode))
 560                 rc = file->f_op->read(file, buf, count, off);
 561         else {
 562                 const int blkbits = inode->i_sb->s_blocksize_bits;
 563                 const int blksize = inode->i_sb->s_blocksize;
 564
 565                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 566                        count, inode->i_ino, *off);
 567                 while (count > 0) {
 568                         struct buffer_head *bh;
 569
 570                         bh = NULL;
 571                         if (*off < inode->i_size) {
 572                                 int err = 0;
 573
 574                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 575                                                 0, &err);
 576
 577                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 578
 579                                 if (bh) {
 580                                         memcpy(buf, bh->b_data, blksize);
 581                                         brelse(bh);
 582                                 } else if (err) {
 583                                         /* XXX in theory we should just fake
 584                                          * this buffer and continue like ext3,
 585                                          * especially if this is a partial read
 586                                          */
 587                                         CERROR("error read dir %lu+%llu: %d\n",
 588                                                inode->i_ino, *off, err);
 589                                         RETURN(err);
 590                                 }
 591                         }
 592                         if (!bh) {
 593                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 594
 595                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 596                                 memset(fake, 0, sizeof(*fake));
 597                                 fake->rec_len = cpu_to_le32(blksize);
 598                         }
 599                         count -= blksize;
 600                         buf += blksize;
 601                         *off += blksize;
 602                         rc += blksize;
 603                 }
 604         }
 605
 606         return rc;
 607 }
 608
 609 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 610 {
 611         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 612
 613         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 614
 615         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 616 }
 617
 618 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 619                                       void *handle, fsfilt_cb_t cb_func,
 620                                       void *cb_data)
 621 {
 622         struct fsfilt_cb_data *fcb;
 623
 624         OBD_SLAB_ALLOC(fcb, fcb_cache, GFP_NOFS, sizeof *fcb);
 625         if (fcb == NULL)
 626                 RETURN(-ENOMEM);
 627
 628         fcb->cb_func = cb_func;
 629         fcb->cb_obd = obd;
 630         fcb->cb_last_rcvd = last_rcvd;
 631         fcb->cb_data = cb_data;
 632
 633         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 634         lock_24kernel();
 635         journal_callback_set(handle, fsfilt_ext3_cb_func,
 636                              (struct journal_callback *)fcb);
 637         unlock_24kernel();
 638
 639         return 0;
 640 }
 641
 642 /*
 643  * We need to hack the return value for the free inode counts because
 644  * the current EA code requires one filesystem block per inode with EAs,
 645  * so it is possible to run out of blocks before we run out of inodes.
 646  *
 647  * This can be removed when the ext3 EA code is fixed.
 648  */
 649 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 650 {
 651         struct kstatfs sfs;
 652         int rc;
 653
 654         memset(&sfs, 0, sizeof(sfs));
 655
 656         rc = sb->s_op->statfs(sb, &sfs);
 657
 658         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 659                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 660                 sfs.f_ffree = sfs.f_bfree;
 661         }
 662
 663         statfs_pack(osfs, &sfs);
 664         return rc;
 665 }
 666
 667 static int fsfilt_ext3_sync(struct super_block *sb)
 668 {
 669         return ext3_force_commit(sb);
 670 }
 671
 672 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 673 #warning "kernel code has old extents/mballoc patch, disabling"
 674 #undef EXT3_MULTIBLOCK_ALLOCATOR
 675 #endif
 676 #ifndef EXT3_EXTENTS_FL
 677 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 678 #endif
 679
 680 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 681 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 682 #define ext3_up_truncate_sem(inode)  up_write(&EXT3_I(inode)->truncate_sem);
 683 #define ext3_down_truncate_sem(inode)  down_write(&EXT3_I(inode)->truncate_sem);
 684 #else
 685 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 686 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 687 #endif
 688
 689 #include <linux/lustre_version.h>
 690 #if EXT3_EXT_MAGIC == 0xf301
 691 #define ee_start e_start
 692 #define ee_block e_block
 693 #define ee_len   e_num
 694 #endif
 695 #ifndef EXT3_BB_MAX_BLOCKS
 696 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 697         ext3_new_blocks(handle, inode, count, goal, err)
 698 #endif
 699
 700 struct bpointers {
 701         unsigned long *blocks;
 702         int *created;
 703         unsigned long start;
 704         int num;
 705         int init_num;
 706         int create;
 707 };
 708
 709 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 710                               unsigned long block, int *aflags)
 711 {
 712         struct ext3_inode_info *ei = EXT3_I(inode);
 713         unsigned long bg_start;
 714         unsigned long colour;
 715         int depth;
 716
 717         if (path) {
 718                 struct ext3_extent *ex;
 719                 depth = path->p_depth;
 720
 721                 /* try to predict block placement */
 722                 if ((ex = path[depth].p_ext)) {
 723 #if 0
 724                         /* This prefers to eat into a contiguous extent
 725                          * rather than find an extent that the whole
 726                          * request will fit into.  This can fragment data
 727                          * block allocation and prevents our lovely 1M I/Os
 728                          * from reaching the disk intact. */
 729                         if (ex->ee_block + ex->ee_len == block)
 730                                 *aflags |= 1;
 731 #endif
 732                         return ex->ee_start + (block - ex->ee_block);
 733                 }
 734
 735                 /* it looks index is empty
 736                  * try to find starting from index itself */
 737                 if (path[depth].p_bh)
 738                         return path[depth].p_bh->b_blocknr;
 739         }
 740
 741         /* OK. use inode's group */
 742         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 743                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 744         colour = (current->pid % 16) *
 745                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 746         return bg_start + colour + block;
 747 }
 748
 749 static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
 750                                   struct ext3_ext_path *path,
 751                                   struct ext3_ext_cache *cex)
 752 {
 753         struct inode *inode = tree->inode;
 754         struct bpointers *bp = tree->private;
 755         struct ext3_extent nex;
 756         int count, err, goal;
 757         unsigned long pblock;
 758         unsigned long tgen;
 759         loff_t new_i_size;
 760         handle_t *handle;
 761         int i, aflags = 0;
 762
 763         i = EXT_DEPTH(tree);
 764         EXT_ASSERT(i == path->p_depth);
 765         EXT_ASSERT(path[i].p_hdr);
 766
 767         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 768                 err = EXT_CONTINUE;
 769                 goto map;
 770         }
 771
 772         if (bp->create == 0) {
 773                 i = 0;
 774                 if (cex->ec_block < bp->start)
 775                         i = bp->start - cex->ec_block;
 776                 if (i >= cex->ec_len)
 777                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 778                                         i, cex->ec_len);
 779                 for (; i < cex->ec_len && bp->num; i++) {
 780                         *(bp->created) = 0;
 781                         bp->created++;
 782                         *(bp->blocks) = 0;
 783                         bp->blocks++;
 784                         bp->num--;
 785                         bp->start++;
 786                 }
 787
 788                 return EXT_CONTINUE;
 789         }
 790
 791         tgen = EXT_GENERATION(tree);
 792         count = ext3_ext_calc_credits_for_insert(tree, path);
 793         ext3_up_truncate_sem(inode);
 794
 795         lock_24kernel();
 796         handle = journal_start(EXT3_JOURNAL(inode), count+EXT3_ALLOC_NEEDED+1);
 797         unlock_24kernel();
 798         if (IS_ERR(handle)) {
 799                 ext3_down_truncate_sem(inode);
 800                 return PTR_ERR(handle);
 801         }
 802
 803         ext3_down_truncate_sem(inode);
 804         if (tgen != EXT_GENERATION(tree)) {
 805                 /* the tree has changed. so path can be invalid at moment */
 806                 lock_24kernel();
 807                 journal_stop(handle);
 808                 unlock_24kernel();
 809                 return EXT_REPEAT;
 810         }
 811
 812         count = cex->ec_len;
 813         goal = ext3_ext_find_goal(inode, path, cex->ec_block, &aflags);
 814         aflags |= 2; /* block have been already reserved */
 815         pblock = ext3_mb_new_blocks(handle, inode, goal, &count, aflags, &err);
 816         if (!pblock)
 817                 goto out;
 818         EXT_ASSERT(count <= cex->ec_len);
 819
 820         /* insert new extent */
 821         nex.ee_block = cex->ec_block;
 822         nex.ee_start = pblock;
 823         nex.ee_len = count;
 824         err = ext3_ext_insert_extent(handle, tree, path, &nex);
 825         if (err)
 826                 goto out;
 827
 828         /*
 829          * Putting len of the actual extent we just inserted,
 830          * we are asking ext3_ext_walk_space() to continue
 831          * scaning after that block
 832          */
 833         cex->ec_len = nex.ee_len;
 834         cex->ec_start = nex.ee_start;
 835         BUG_ON(nex.ee_len == 0);
 836         BUG_ON(nex.ee_block != cex->ec_block);
 837
 838         /* correct on-disk inode size */
 839         if (nex.ee_len > 0) {
 840                 new_i_size = (loff_t) nex.ee_block + nex.ee_len;
 841                 new_i_size = new_i_size << inode->i_blkbits;
 842                 if (new_i_size > EXT3_I(inode)->i_disksize) {
 843                         EXT3_I(inode)->i_disksize = new_i_size;
 844                         err = ext3_mark_inode_dirty(handle, inode);
 845                 }
 846         }
 847
 848 out:
 849         lock_24kernel();
 850         journal_stop(handle);
 851         unlock_24kernel();
 852 map:
 853         if (err >= 0) {
 854                 /* map blocks */
 855                 if (bp->num == 0) {
 856                         CERROR("hmm. why do we find this extent?\n");
 857                         CERROR("initial space: %lu:%u\n",
 858                                 bp->start, bp->init_num);
 859                         CERROR("current extent: %u/%u/%u %d\n",
 860                                 cex->ec_block, cex->ec_len,
 861                                 cex->ec_start, cex->ec_type);
 862                 }
 863                 i = 0;
 864                 if (cex->ec_block < bp->start)
 865                         i = bp->start - cex->ec_block;
 866                 if (i >= cex->ec_len)
 867                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 868                                         i, cex->ec_len);
 869                 for (; i < cex->ec_len && bp->num; i++) {
 870                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT)
 871                                 *(bp->created) = 0;
 872                         else
 873                                 *(bp->created) = 1;
 874                         bp->created++;
 875                         *(bp->blocks) = cex->ec_start + i;
 876                         bp->blocks++;
 877                         bp->num--;
 878                         bp->start++;
 879                 }
 880         }
 881         return err;
 882 }
 883
 884 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
 885                        unsigned long num, unsigned long *blocks,
 886                        int *created, int create)
 887 {
 888         struct ext3_extents_tree tree;
 889         struct bpointers bp;
 890         int err;
 891
 892         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
 893                 block, block + num, (unsigned) inode->i_ino);
 894
 895         ext3_init_tree_desc(&tree, inode);
 896         tree.private = &bp;
 897         bp.blocks = blocks;
 898         bp.created = created;
 899         bp.start = block;
 900         bp.init_num = bp.num = num;
 901         bp.create = create;
 902
 903         ext3_down_truncate_sem(inode);
 904         err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
 905         ext3_ext_invalidate_cache(&tree);
 906         ext3_up_truncate_sem(inode);
 907
 908         return err;
 909 }
 910
 911 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
 912                                     int pages, unsigned long *blocks,
 913                                     int *created, int create)
 914 {
 915         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 916         int rc = 0, i = 0;
 917         struct page *fp = NULL;
 918         int clen = 0;
 919
 920         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
 921                 inode->i_ino, pages, (*page)->index);
 922
 923         /* pages are sorted already. so, we just have to find
 924          * contig. space and process them properly */
 925         while (i < pages) {
 926                 if (fp == NULL) {
 927                         /* start new extent */
 928                         fp = *page++;
 929                         clen = 1;
 930                         i++;
 931                         continue;
 932                 } else if (fp->index + clen == (*page)->index) {
 933                         /* continue the extent */
 934                         page++;
 935                         clen++;
 936                         i++;
 937                         continue;
 938                 }
 939
 940                 /* process found extent */
 941                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 942                                         clen * blocks_per_page, blocks,
 943                                         created, create);
 944                 if (rc)
 945                         GOTO(cleanup, rc);
 946
 947                 /* look for next extent */
 948                 fp = NULL;
 949                 blocks += blocks_per_page * clen;
 950                 created += blocks_per_page * clen;
 951         }
 952
 953         if (fp)
 954                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 955                                         clen * blocks_per_page, blocks,
 956                                         created, create);
 957 cleanup:
 958         return rc;
 959 }
 960 #endif
 961
 962 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
 963                                unsigned long *blocks, int *created, int create);
 964 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
 965                                    int pages, unsigned long *blocks,
 966                                    int *created, int create)
 967 {
 968         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 969         unsigned long *b;
 970         int rc = 0, i, *cr;
 971
 972         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
 973                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
 974                 if (rc) {
 975                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
 976                                inode->i_ino, *b, *cr, create, rc);
 977                         break;
 978                 }
 979
 980                 b += blocks_per_page;
 981                 cr += blocks_per_page;
 982         }
 983         return rc;
 984 }
 985
 986 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
 987                                 int pages, unsigned long *blocks,
 988                                 int *created, int create,
 989                                 struct semaphore *optional_sem)
 990 {
 991         int rc;
 992 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 993         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
 994                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
 995                                                      blocks, created, create);
 996                 return rc;
 997         }
 998 #endif
 999         if (optional_sem != NULL)
1000                 down(optional_sem);
1001         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1002                                             created, create);
1003         if (optional_sem != NULL)
1004                 up(optional_sem);
1005
1006         return rc;
1007 }
1008
1009 extern int ext3_prep_san_write(struct inode *inode, long *blocks,
1010                                int nblocks, loff_t newsize);
1011 static int fsfilt_ext3_prep_san_write(struct inode *inode, long *blocks,
1012                                       int nblocks, loff_t newsize)
1013 {
1014         return ext3_prep_san_write(inode, blocks, nblocks, newsize);
1015 }
1016
1017 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1018                                    int size, loff_t *offs)
1019 {
1020         struct inode *inode = file->f_dentry->d_inode;
1021         unsigned long block;
1022         struct buffer_head *bh;
1023         int err, blocksize, csize, boffs;
1024
1025         /* prevent reading after eof */
1026         lock_kernel();
1027         if (inode->i_size < *offs + size) {
1028                 size = inode->i_size - *offs;
1029                 unlock_kernel();
1030                 if (size < 0) {
1031                         CERROR("size %llu is too short for read %u@%llu\n",
1032                                inode->i_size, size, *offs);
1033                         return -EIO;
1034                 } else if (size == 0) {
1035                         return 0;
1036                 }
1037         } else {
1038                 unlock_kernel();
1039         }
1040
1041         blocksize = 1 << inode->i_blkbits;
1042
1043         while (size > 0) {
1044                 block = *offs >> inode->i_blkbits;
1045                 boffs = *offs & (blocksize - 1);
1046                 csize = min(blocksize - boffs, size);
1047                 bh = ext3_bread(NULL, inode, block, 0, &err);
1048                 if (!bh) {
1049                         CERROR("can't read block: %d\n", err);
1050                         return err;
1051                 }
1052
1053                 memcpy(buf, bh->b_data + boffs, csize);
1054                 brelse(bh);
1055
1056                 *offs += csize;
1057                 buf += csize;
1058                 size -= csize;
1059         }
1060         return 0;
1061 }
1062
1063 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1064                                     loff_t *offs, int force_sync)
1065 {
1066         struct buffer_head *bh = NULL;
1067         unsigned long block;
1068         struct inode *inode = file->f_dentry->d_inode;
1069         loff_t old_size = inode->i_size, offset = *offs;
1070         loff_t new_size = inode->i_size;
1071         journal_t *journal;
1072         handle_t *handle;
1073         int err, block_count = 0, blocksize, size, boffs;
1074
1075         /* Determine how many transaction credits are needed */
1076         blocksize = 1 << inode->i_blkbits;
1077         block_count = (*offs & (blocksize - 1)) + bufsize;
1078         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1079
1080         journal = EXT3_SB(inode->i_sb)->s_journal;
1081         lock_24kernel();
1082         handle = journal_start(journal,
1083                                block_count * EXT3_DATA_TRANS_BLOCKS + 2);
1084         unlock_24kernel();
1085         if (IS_ERR(handle)) {
1086                 CERROR("can't start transaction\n");
1087                 return PTR_ERR(handle);
1088         }
1089
1090         while (bufsize > 0) {
1091                 if (bh != NULL)
1092                         brelse(bh);
1093
1094                 block = offset >> inode->i_blkbits;
1095                 boffs = offset & (blocksize - 1);
1096                 size = min(blocksize - boffs, bufsize);
1097                 bh = ext3_bread(handle, inode, block, 1, &err);
1098                 if (!bh) {
1099                         CERROR("can't read/create block: %d\n", err);
1100                         goto out;
1101                 }
1102
1103                 err = ext3_journal_get_write_access(handle, bh);
1104                 if (err) {
1105                         CERROR("journal_get_write_access() returned error %d\n",
1106                                err);
1107                         goto out;
1108                 }
1109                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1110                 memcpy(bh->b_data + boffs, buf, size);
1111                 err = ext3_journal_dirty_metadata(handle, bh);
1112                 if (err) {
1113                         CERROR("journal_dirty_metadata() returned error %d\n",
1114                                err);
1115                         goto out;
1116                 }
1117                 if (offset + size > new_size)
1118                         new_size = offset + size;
1119                 offset += size;
1120                 bufsize -= size;
1121                 buf += size;
1122         }
1123
1124         if (force_sync)
1125                 handle->h_sync = 1; /* recovery likes this */
1126 out:
1127         if (bh)
1128                 brelse(bh);
1129
1130         /* correct in-core and on-disk sizes */
1131         if (new_size > inode->i_size) {
1132                 lock_kernel();
1133                 if (new_size > inode->i_size)
1134                         inode->i_size = new_size;
1135                 if (inode->i_size > EXT3_I(inode)->i_disksize)
1136                         EXT3_I(inode)->i_disksize = inode->i_size;
1137                 if (inode->i_size > old_size)
1138                         mark_inode_dirty(inode);
1139                 unlock_kernel();
1140         }
1141
1142         lock_24kernel();
1143         journal_stop(handle);
1144         unlock_24kernel();
1145
1146         if (err == 0)
1147                 *offs = offset;
1148         return err;
1149 }
1150
1151 static int fsfilt_ext3_setup(struct super_block *sb)
1152 {
1153 #if 0
1154         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1155         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1156 #endif
1157 #ifdef S_PDIROPS
1158         CWARN("Enabling PDIROPS\n");
1159         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1160         sb->s_flags |= S_PDIROPS;
1161 #endif
1162         return 0;
1163 }
1164
1165 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1166    objects. Logs is number of logfiles to update */
1167 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1168 {
1169         if ( !fso ) {
1170                 switch(op) {
1171                 case FSFILT_OP_CREATE:
1172                                  /* directory leaf, index & indirect & EA*/
1173                         return 4 + 3 * logs;
1174                 case FSFILT_OP_UNLINK:
1175                         return 3 * logs;
1176                 }
1177         } else {
1178                 int i;
1179                 int needed = 0;
1180                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1181                 int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
1182                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1183                 for (i = 0; i < op; i++, fso++) {
1184                         int nblocks = fso->fso_bufcnt * blockpp;
1185                         int ndindirect = min(nblocks, addrpp + 1);
1186                         int nindir = nblocks + ndindirect + 1;
1187
1188                         needed += nindir;
1189                 }
1190                 return needed + 3 * logs;
1191         }
1192
1193         return 0;
1194 }
1195
1196 static inline struct ext3_group_desc *
1197 get_group_desc(struct super_block *sb, int group)
1198 {
1199         unsigned long desc_block, desc;
1200         struct ext3_group_desc *gdp;
1201
1202         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1203         desc = group % EXT3_DESC_PER_BLOCK(sb);
1204         gdp = (struct ext3_group_desc *)
1205               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1206
1207         return gdp + desc;
1208 }
1209
1210 static inline struct buffer_head *
1211 read_inode_bitmap(struct super_block *sb, unsigned long group)
1212 {
1213         struct ext3_group_desc *desc;
1214         struct buffer_head *bh;
1215
1216         desc = get_group_desc(sb, group);
1217         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1218
1219         return bh;
1220 }
1221
1222 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1223                                      struct buffer_head *bitmap_bh,
1224                                      int index, unsigned long ino)
1225 {
1226         struct inode *inode = NULL;
1227
1228         if (ext3_test_bit(index, bitmap_bh->b_data))
1229                 inode = iget(sb, ino);
1230
1231         return inode;
1232 }
1233
1234 #ifdef HAVE_QUOTA_SUPPORT
1235 # include "fsfilt_ext3_quota.h"
1236 #endif
1237
1238 static struct fsfilt_operations fsfilt_ext3_ops = {
1239         .fs_type                = "ext3",
1240         .fs_owner               = THIS_MODULE,
1241         .fs_start               = fsfilt_ext3_start,
1242         .fs_brw_start           = fsfilt_ext3_brw_start,
1243         .fs_commit              = fsfilt_ext3_commit,
1244         .fs_commit_async        = fsfilt_ext3_commit_async,
1245         .fs_commit_wait         = fsfilt_ext3_commit_wait,
1246         .fs_setattr             = fsfilt_ext3_setattr,
1247         .fs_iocontrol           = fsfilt_ext3_iocontrol,
1248         .fs_set_md              = fsfilt_ext3_set_md,
1249         .fs_get_md              = fsfilt_ext3_get_md,
1250         .fs_readpage            = fsfilt_ext3_readpage,
1251         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
1252         .fs_statfs              = fsfilt_ext3_statfs,
1253         .fs_sync                = fsfilt_ext3_sync,
1254         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
1255         .fs_prep_san_write      = fsfilt_ext3_prep_san_write,
1256         .fs_write_record        = fsfilt_ext3_write_record,
1257         .fs_read_record         = fsfilt_ext3_read_record,
1258         .fs_setup               = fsfilt_ext3_setup,
1259         .fs_send_bio            = fsfilt_ext3_send_bio,
1260         .fs_get_op_len          = fsfilt_ext3_get_op_len,
1261 #ifdef HAVE_QUOTA_SUPPORT
1262         .fs_quotactl            = fsfilt_ext3_quotactl,
1263         .fs_quotacheck          = fsfilt_ext3_quotacheck,
1264         .fs_quotainfo           = fsfilt_ext3_quotainfo,
1265         .fs_dquot               = fsfilt_ext3_dquot,
1266 #endif
1267 };
1268
1269 static int __init fsfilt_ext3_init(void)
1270 {
1271         int rc;
1272
1273         fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
1274                                       sizeof(struct fsfilt_cb_data), 0,
1275                                       0, NULL, NULL);
1276         if (!fcb_cache) {
1277                 CERROR("error allocating fsfilt journal callback cache\n");
1278                 GOTO(out, rc = -ENOMEM);
1279         }
1280
1281         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
1282
1283         if (rc)
1284                 kmem_cache_destroy(fcb_cache);
1285 out:
1286         return rc;
1287 }
1288
1289 static void __exit fsfilt_ext3_exit(void)
1290 {
1291         fsfilt_unregister_ops(&fsfilt_ext3_ops);
1292         LASSERT(kmem_cache_destroy(fcb_cache) == 0);
1293 }
1294
1295 module_init(fsfilt_ext3_init);
1296 module_exit(fsfilt_ext3_exit);
1297
1298 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1299 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
1300 MODULE_LICENSE("GPL");