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 <lustre_fsfilt.h>
  50 #include <obd.h>
  51 #include <obd_class.h>
  52 #include <lustre_quota.h>
  53 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  54 #include <linux/iobuf.h>
  55 #endif
  56 #include <linux/lustre_compat25.h>
  57 #include <linux/lprocfs_status.h>
  58
  59 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  60 #include <linux/ext3_extents.h>
  61 #endif
  62
  63 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  64 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  65 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  66 #else
  67 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  68 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  69 #endif
  70
  71 static kmem_cache_t *fcb_cache;
  72
  73 struct fsfilt_cb_data {
  74         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  75         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  76         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  77         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  78         void *cb_data;                  /* MDS/OST completion function data */
  79 };
  80
  81 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
  82 #define EXT3_XATTR_INDEX_TRUSTED        4
  83 #endif
  84
  85 static char *fsfilt_ext3_get_label(struct super_block *sb)
  86 {
  87         return EXT3_SB(sb)->s_es->s_volume_name;
  88 }
  89
  90 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
  91 {
  92         /* see e.g. fsfilt_ext3_write_record() */
  93         journal_t *journal;
  94         handle_t *handle;
  95         int err;
  96
  97         journal = EXT3_SB(sb)->s_journal;
  98         lock_24kernel();
  99         handle = journal_start(journal, 1);
 100         unlock_24kernel();
 101         if (IS_ERR(handle)) {
 102                 CERROR("can't start transaction\n");
 103                 return(PTR_ERR(handle));
 104         }
 105
 106         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 107         if (err)
 108                 goto out;
 109
 110         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 111                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 112
 113         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 114
 115 out:
 116         lock_24kernel();
 117         journal_stop(handle);
 118         unlock_24kernel();
 119
 120         return(err);
 121 }
 122
 123 static char *fsfilt_ext3_uuid(struct super_block *sb)
 124 {
 125         return EXT3_SB(sb)->s_es->s_uuid;
 126 }
 127
 128 /*
 129  * We don't currently need any additional blocks for rmdir and
 130  * unlink transactions because we are storing the OST oa_id inside
 131  * the inode (which we will be changing anyways as part of this
 132  * transaction).
 133  */
 134 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 135                                int logs)
 136 {
 137         /* For updates to the last received file */
 138         int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS;
 139         journal_t *journal;
 140         void *handle;
 141
 142         if (current->journal_info) {
 143                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 144                        current->journal_info);
 145                 goto journal_start;
 146         }
 147
 148         switch(op) {
 149         case FSFILT_OP_RMDIR:
 150         case FSFILT_OP_UNLINK:
 151                 /* delete one file + create/update logs for each stripe */
 152                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 153                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 154                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 155                 break;
 156         case FSFILT_OP_RENAME:
 157                 /* modify additional directory */
 158                 nblocks += EXT3_SINGLEDATA_TRANS_BLOCKS;
 159                 /* no break */
 160         case FSFILT_OP_SYMLINK:
 161                 /* additional block + block bitmap + GDT for long symlink */
 162                 nblocks += 3;
 163                 /* no break */
 164         case FSFILT_OP_CREATE: {
 165 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 166                 static int warned;
 167                 if (!warned) {
 168                         if (!test_opt(inode->i_sb, EXTENTS)) {
 169                                 warned = 1;
 170                         } else if (((EXT3_I(inode)->i_flags &
 171                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 172                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 173                                 CWARN("extent-mapped directory found - contact "
 174                                       "CFS: support@clusterfs.com\n");
 175                                 warned = 1;
 176                         }
 177                 }
 178 #endif
 179                 /* no break */
 180         }
 181         case FSFILT_OP_MKDIR:
 182         case FSFILT_OP_MKNOD:
 183                 /* modify one inode + block bitmap + GDT */
 184                 nblocks += 3;
 185                 /* no break */
 186         case FSFILT_OP_LINK:
 187                 /* modify parent directory */
 188                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 189                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 190                 /* create/update logs for each stripe */
 191                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 192                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 193                 break;
 194         case FSFILT_OP_SETATTR:
 195                 /* Setattr on inode */
 196                 nblocks += 1;
 197                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 198                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 199                 /* quota chown log for each stripe */
 200                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 201                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 202                 break;
 203         case FSFILT_OP_CANCEL_UNLINK:
 204                 /* blocks for log header bitmap update OR
 205                  * blocks for catalog header bitmap update + unlink of logs */
 206                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 207                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 208                 break;
 209         case FSFILT_OP_JOIN:
 210                 /* delete 2 file(file + array id) + create 1 file (array id)
 211                  * create/update logs for each stripe */
 212                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 213
 214                 /*create array log for head file*/
 215                 nblocks += 3;
 216                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 217                             EXT3_SINGLEDATA_TRANS_BLOCKS);
 218                 /*update head file array */
 219                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 220                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 221                 break;
 222         default: CERROR("unknown transaction start op %d\n", op);
 223                 LBUG();
 224         }
 225
 226         LASSERT(current->journal_info == desc_private);
 227         journal = EXT3_SB(inode->i_sb)->s_journal;
 228         if (nblocks > journal->j_max_transaction_buffers) {
 229                 CERROR("too many credits %d for op %ux%u using %d instead\n",
 230                        nblocks, op, logs, journal->j_max_transaction_buffers);
 231                 nblocks = journal->j_max_transaction_buffers;
 232         }
 233
 234  journal_start:
 235         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 236         lock_24kernel();
 237         handle = journal_start(EXT3_JOURNAL(inode), nblocks);
 238         unlock_24kernel();
 239
 240         if (!IS_ERR(handle))
 241                 LASSERT(current->journal_info == handle);
 242         else
 243                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 244                        op, nblocks, PTR_ERR(handle));
 245         return handle;
 246 }
 247
 248 /*
 249  * Calculate the number of buffer credits needed to write multiple pages in
 250  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 251  * doesn't have a nice API for calculating this sort of thing in advance.
 252  *
 253  * See comment above ext3_writepage_trans_blocks for details.  We assume
 254  * no data journaling is being done, but it does allow for all of the pages
 255  * being non-contiguous.  If we are guaranteed contiguous pages we could
 256  * reduce the number of (d)indirect blocks a lot.
 257  *
 258  * With N blocks per page and P pages, for each inode we have at most:
 259  * N*P indirect
 260  * min(N*P, blocksize/4 + 1) dindirect blocks
 261  * niocount tindirect
 262  *
 263  * For the entire filesystem, we have at most:
 264  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 265  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 266  * objcount inode blocks
 267  * 1 superblock
 268  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 269  *
 270  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 271  */
 272 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 273                                       int niocount, struct niobuf_local *nb)
 274 {
 275         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 276         __u64 next_indir;
 277         const int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
 278         int nbitmaps = 0, ngdblocks;
 279         int needed = objcount + 1; /* inodes + superblock */
 280         int i, j;
 281
 282         for (i = 0, j = 0; i < objcount; i++, fso++) {
 283                 /* two or more dindirect blocks in case we cross boundary */
 284                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 285                                     nb[j].offset) >>
 286                                    sb->s_blocksize_bits) /
 287                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 288                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 289
 290                 /* leaf, indirect, tindirect blocks for first block */
 291                 nbitmaps += blockpp + 2;
 292
 293                 j += fso->fso_bufcnt;
 294         }
 295
 296         next_indir = nb[0].offset +
 297                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 298         for (i = 1; i < niocount; i++) {
 299                 if (nb[i].offset >= next_indir) {
 300                         nbitmaps++;     /* additional indirect */
 301                         next_indir = nb[i].offset +
 302                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 303                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 304                         nbitmaps++;     /* additional indirect */
 305                 }
 306                 nbitmaps += blockpp;    /* each leaf in different group? */
 307         }
 308
 309         ngdblocks = nbitmaps;
 310         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 311                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 312         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 313                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 314
 315         needed += nbitmaps + ngdblocks;
 316
 317         /* last_rcvd update */
 318         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 319
 320 #if defined(CONFIG_QUOTA)
 321         /* We assume that there will be 1 bit set in s_dquot.flags for each
 322          * quota file that is active.  This is at least true for now.
 323          */
 324         needed += hweight32(sb_any_quota_enabled(sb)) *
 325                 EXT3_SINGLEDATA_TRANS_BLOCKS;
 326 #endif
 327
 328         return needed;
 329 }
 330
 331 /* We have to start a huge journal transaction here to hold all of the
 332  * metadata for the pages being written here.  This is necessitated by
 333  * the fact that we do lots of prepare_write operations before we do
 334  * any of the matching commit_write operations, so even if we split
 335  * up to use "smaller" transactions none of them could complete until
 336  * all of them were opened.  By having a single journal transaction,
 337  * we eliminate duplicate reservations for common blocks like the
 338  * superblock and group descriptors or bitmaps.
 339  *
 340  * We will start the transaction here, but each prepare_write will
 341  * add a refcount to the transaction, and each commit_write will
 342  * remove a refcount.  The transaction will be closed when all of
 343  * the pages have been written.
 344  */
 345 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 346                                    int niocount, struct niobuf_local *nb,
 347                                    void *desc_private, int logs)
 348 {
 349         journal_t *journal;
 350         handle_t *handle;
 351         int needed;
 352         ENTRY;
 353
 354         LASSERT(current->journal_info == desc_private);
 355         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 356         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 357
 358         /* The number of blocks we could _possibly_ dirty can very large.
 359          * We reduce our request if it is absurd (and we couldn't get that
 360          * many credits for a single handle anyways).
 361          *
 362          * At some point we have to limit the size of I/Os sent at one time,
 363          * increase the size of the journal, or we have to calculate the
 364          * actual journal requirements more carefully by checking all of
 365          * the blocks instead of being maximally pessimistic.  It remains to
 366          * be seen if this is a real problem or not.
 367          */
 368         if (needed > journal->j_max_transaction_buffers) {
 369                 CERROR("want too many journal credits (%d) using %d instead\n",
 370                        needed, journal->j_max_transaction_buffers);
 371                 needed = journal->j_max_transaction_buffers;
 372         }
 373
 374         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 375         lock_24kernel();
 376         handle = journal_start(journal, needed);
 377         unlock_24kernel();
 378         if (IS_ERR(handle)) {
 379                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 380                        PTR_ERR(handle));
 381         } else {
 382                 LASSERT(handle->h_buffer_credits >= needed);
 383                 LASSERT(current->journal_info == handle);
 384         }
 385
 386         RETURN(handle);
 387 }
 388
 389 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 390 {
 391         int rc;
 392         handle_t *handle = h;
 393
 394         LASSERT(current->journal_info == handle);
 395         if (force_sync)
 396                 handle->h_sync = 1; /* recovery likes this */
 397
 398         lock_24kernel();
 399         rc = journal_stop(handle);
 400         unlock_24kernel();
 401
 402         return rc;
 403 }
 404
 405 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 406                                     void **wait_handle)
 407 {
 408         unsigned long tid;
 409         transaction_t *transaction;
 410 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 411         unsigned long rtid;
 412 #endif
 413         handle_t *handle = h;
 414         journal_t *journal;
 415         int rc;
 416
 417         LASSERT(current->journal_info == handle);
 418
 419         lock_24kernel();
 420         transaction = handle->h_transaction;
 421         journal = transaction->t_journal;
 422         tid = transaction->t_tid;
 423         /* we don't want to be blocked */
 424         handle->h_sync = 0;
 425         rc = journal_stop(handle);
 426         if (rc) {
 427                 CERROR("error while stopping transaction: %d\n", rc);
 428                 unlock_24kernel();
 429                 return rc;
 430         }
 431 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 432         rtid = log_start_commit(journal, transaction);
 433         if (rtid != tid)
 434                 CERROR("strange race: %lu != %lu\n",
 435                        (unsigned long) tid, (unsigned long) rtid);
 436 #else
 437         log_start_commit(journal, tid);
 438 #endif
 439         unlock_24kernel();
 440
 441         *wait_handle = (void *) tid;
 442         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 443         return 0;
 444 }
 445
 446 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 447 {
 448         journal_t *journal = EXT3_JOURNAL(inode);
 449         tid_t tid = (tid_t)(long)h;
 450
 451         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 452         if (unlikely(is_journal_aborted(journal)))
 453                 return -EIO;
 454
 455         log_wait_commit(EXT3_JOURNAL(inode), tid);
 456
 457         if (unlikely(is_journal_aborted(journal)))
 458                 return -EIO;
 459         return 0;
 460 }
 461
 462 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 463                                struct iattr *iattr, int do_trunc)
 464 {
 465         struct inode *inode = dentry->d_inode;
 466         int rc = 0;
 467
 468         lock_kernel();
 469
 470         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 471          * We are already writing the inode to disk as part of this
 472          * transaction and want to avoid a lot of extra inode writeout
 473          * later on. b=9828 */
 474         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 475                 /* ATTR_SIZE would invoke truncate: clear it */
 476                 iattr->ia_valid &= ~ATTR_SIZE;
 477                 EXT3_I(inode)->i_disksize = inode->i_size = iattr->ia_size;
 478
 479                 if (iattr->ia_valid & ATTR_UID)
 480                         inode->i_uid = iattr->ia_uid;
 481                 if (iattr->ia_valid & ATTR_GID)
 482                         inode->i_gid = iattr->ia_gid;
 483                 if (iattr->ia_valid & ATTR_ATIME)
 484                         inode->i_atime = iattr->ia_atime;
 485                 if (iattr->ia_valid & ATTR_MTIME)
 486                         inode->i_mtime = iattr->ia_mtime;
 487                 if (iattr->ia_valid & ATTR_CTIME)
 488                         inode->i_ctime = iattr->ia_ctime;
 489                 if (iattr->ia_valid & ATTR_MODE) {
 490                         inode->i_mode = iattr->ia_mode;
 491
 492                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 493                                 inode->i_mode &= ~S_ISGID;
 494                 }
 495
 496                 inode->i_sb->s_op->dirty_inode(inode);
 497
 498                 goto out;
 499         }
 500
 501         /* Don't allow setattr to change file type */
 502         if (iattr->ia_valid & ATTR_MODE)
 503                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 504                                  (iattr->ia_mode & ~S_IFMT);
 505
 506         /* We set these flags on the client, but have already checked perms
 507          * so don't confuse inode_change_ok. */
 508         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 509
 510         if (inode->i_op->setattr) {
 511                 rc = inode->i_op->setattr(dentry, iattr);
 512         } else {
 513                 rc = inode_change_ok(inode, iattr);
 514                 if (!rc)
 515                         rc = inode_setattr(inode, iattr);
 516         }
 517
 518  out:
 519         unlock_kernel();
 520         RETURN(rc);
 521 }
 522
 523 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 524                                  unsigned int cmd, unsigned long arg)
 525 {
 526         int rc = 0;
 527         ENTRY;
 528
 529         /* FIXME: Can't do this because of nested transaction deadlock */
 530         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 531                 CERROR("can't set data journal flag on file\n");
 532                 RETURN(-EPERM);
 533         }
 534
 535         if (inode->i_fop->ioctl)
 536                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 537         else
 538                 RETURN(-ENOTTY);
 539
 540         RETURN(rc);
 541 }
 542
 543 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 544                               void *lmm, int lmm_size, const char *name)
 545 {
 546         int rc;
 547
 548         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 549
 550         lock_24kernel();
 551         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 552                                    name, lmm, lmm_size, 0);
 553
 554         unlock_24kernel();
 555
 556         if (rc)
 557                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 558                        inode->i_ino, rc);
 559         return rc;
 560 }
 561
 562 /* Must be called with i_mutex held */
 563 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 564                               const char *name)
 565 {
 566         int rc;
 567
 568         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 569         lock_24kernel();
 570
 571         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 572                             name, lmm, lmm_size);
 573         unlock_24kernel();
 574
 575         /* This gives us the MD size */
 576         if (lmm == NULL)
 577                 return (rc == -ENODATA) ? 0 : rc;
 578
 579         if (rc < 0) {
 580                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 581                        EXT3_XATTR_INDEX_TRUSTED, name,
 582                        inode->i_ino, rc);
 583                 memset(lmm, 0, lmm_size);
 584                 return (rc == -ENODATA) ? 0 : rc;
 585         }
 586
 587         return rc;
 588 }
 589
 590 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
 591 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 592 {
 593         submit_bio(rw, bio);
 594         return 0;
 595 }
 596 #else
 597 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct kiobuf *bio)
 598 {
 599         int rc, blk_per_page;
 600
 601         rc = brw_kiovec(rw, 1, &bio, inode->i_dev,
 602                         KIOBUF_GET_BLOCKS(bio), 1 << inode->i_blkbits);
 603         /*
 604          * brw_kiovec() returns number of bytes actually written. If error
 605          * occurred after something was written, error code is returned though
 606          * kiobuf->errno. (See bug 6854.)
 607          */
 608
 609         blk_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
 610
 611         if (rc != (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page) {
 612                 CERROR("short write?  expected %d, wrote %d (%d)\n",
 613                        (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
 614                        rc, bio->errno);
 615         }
 616         if (bio->errno != 0) {
 617                 CERROR("IO error. Wrote %d of %d (%d)\n",
 618                        rc,
 619                        (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
 620                        bio->errno);
 621                 rc = bio->errno;
 622         }
 623
 624         return rc;
 625 }
 626 #endif
 627
 628 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 629                                     loff_t *off)
 630 {
 631         struct inode *inode = file->f_dentry->d_inode;
 632         int rc = 0;
 633
 634         if (S_ISREG(inode->i_mode))
 635                 rc = file->f_op->read(file, buf, count, off);
 636         else {
 637                 const int blkbits = inode->i_sb->s_blocksize_bits;
 638                 const int blksize = inode->i_sb->s_blocksize;
 639
 640                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 641                        count, inode->i_ino, *off);
 642                 while (count > 0) {
 643                         struct buffer_head *bh;
 644
 645                         bh = NULL;
 646                         if (*off < inode->i_size) {
 647                                 int err = 0;
 648
 649                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 650                                                 0, &err);
 651
 652                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 653
 654                                 if (bh) {
 655                                         memcpy(buf, bh->b_data, blksize);
 656                                         brelse(bh);
 657                                 } else if (err) {
 658                                         /* XXX in theory we should just fake
 659                                          * this buffer and continue like ext3,
 660                                          * especially if this is a partial read
 661                                          */
 662                                         CERROR("error read dir %lu+%llu: %d\n",
 663                                                inode->i_ino, *off, err);
 664                                         RETURN(err);
 665                                 }
 666                         }
 667                         if (!bh) {
 668                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 669
 670                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 671                                 memset(fake, 0, sizeof(*fake));
 672                                 fake->rec_len = cpu_to_le16(blksize);
 673                         }
 674                         count -= blksize;
 675                         buf += blksize;
 676                         *off += blksize;
 677                         rc += blksize;
 678                 }
 679         }
 680
 681         return rc;
 682 }
 683
 684 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 685 {
 686         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 687
 688         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 689
 690         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 691 }
 692
 693 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 694                                       void *handle, fsfilt_cb_t cb_func,
 695                                       void *cb_data)
 696 {
 697         struct fsfilt_cb_data *fcb;
 698
 699         OBD_SLAB_ALLOC(fcb, fcb_cache, GFP_NOFS, sizeof *fcb);
 700         if (fcb == NULL)
 701                 RETURN(-ENOMEM);
 702
 703         fcb->cb_func = cb_func;
 704         fcb->cb_obd = obd;
 705         fcb->cb_last_rcvd = last_rcvd;
 706         fcb->cb_data = cb_data;
 707
 708         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 709         lock_24kernel();
 710         journal_callback_set(handle, fsfilt_ext3_cb_func,
 711                              (struct journal_callback *)fcb);
 712         unlock_24kernel();
 713
 714         return 0;
 715 }
 716
 717 /*
 718  * We need to hack the return value for the free inode counts because
 719  * the current EA code requires one filesystem block per inode with EAs,
 720  * so it is possible to run out of blocks before we run out of inodes.
 721  *
 722  * This can be removed when the ext3 EA code is fixed.
 723  */
 724 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 725 {
 726         struct kstatfs sfs;
 727         int rc;
 728
 729         memset(&sfs, 0, sizeof(sfs));
 730
 731         rc = sb->s_op->statfs(sb, &sfs);
 732
 733         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 734                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 735                 sfs.f_ffree = sfs.f_bfree;
 736         }
 737
 738         statfs_pack(osfs, &sfs);
 739         return rc;
 740 }
 741
 742 static int fsfilt_ext3_sync(struct super_block *sb)
 743 {
 744         return ext3_force_commit(sb);
 745 }
 746
 747 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 748 #warning "kernel code has old extents/mballoc patch, disabling"
 749 #undef EXT3_MULTIBLOCK_ALLOCATOR
 750 #endif
 751 #ifndef EXT3_EXTENTS_FL
 752 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 753 #endif
 754
 755 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 756 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 757 #define ext3_up_truncate_sem(inode)  up_write(&EXT3_I(inode)->truncate_sem);
 758 #define ext3_down_truncate_sem(inode)  down_write(&EXT3_I(inode)->truncate_sem);
 759 #elif (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 760 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 761 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 762 #else
 763 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 764 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 765 #endif
 766
 767 #include <linux/lustre_version.h>
 768 #if EXT3_EXT_MAGIC == 0xf301
 769 #define ee_start e_start
 770 #define ee_block e_block
 771 #define ee_len   e_num
 772 #endif
 773 #ifndef EXT3_BB_MAX_BLOCKS
 774 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 775         ext3_new_blocks(handle, inode, count, goal, err)
 776 #endif
 777
 778 struct bpointers {
 779         unsigned long *blocks;
 780         int *created;
 781         unsigned long start;
 782         int num;
 783         int init_num;
 784         int create;
 785 };
 786
 787 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 788                               unsigned long block, int *aflags)
 789 {
 790         struct ext3_inode_info *ei = EXT3_I(inode);
 791         unsigned long bg_start;
 792         unsigned long colour;
 793         int depth;
 794
 795         if (path) {
 796                 struct ext3_extent *ex;
 797                 depth = path->p_depth;
 798
 799                 /* try to predict block placement */
 800                 if ((ex = path[depth].p_ext)) {
 801 #if 0
 802                         /* This prefers to eat into a contiguous extent
 803                          * rather than find an extent that the whole
 804                          * request will fit into.  This can fragment data
 805                          * block allocation and prevents our lovely 1M I/Os
 806                          * from reaching the disk intact. */
 807                         if (ex->ee_block + ex->ee_len == block)
 808                                 *aflags |= 1;
 809 #endif
 810                         return ex->ee_start + (block - ex->ee_block);
 811                 }
 812
 813                 /* it looks index is empty
 814                  * try to find starting from index itself */
 815                 if (path[depth].p_bh)
 816                         return path[depth].p_bh->b_blocknr;
 817         }
 818
 819         /* OK. use inode's group */
 820         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 821                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 822         colour = (current->pid % 16) *
 823                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 824         return bg_start + colour + block;
 825 }
 826
 827 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 828 #include <linux/locks.h>
 829 static void ll_unmap_underlying_metadata(struct super_block *sb,
 830                                          unsigned long blocknr)
 831 {
 832         struct buffer_head *old_bh;
 833
 834         old_bh = get_hash_table(sb->s_dev, blocknr, sb->s_blocksize);
 835         if (old_bh) {
 836                 mark_buffer_clean(old_bh);
 837                 wait_on_buffer(old_bh);
 838                 clear_bit(BH_Req, &old_bh->b_state);
 839                 __brelse(old_bh);
 840         }
 841 }
 842 #else
 843 #define ll_unmap_underlying_metadata(sb, blocknr) \
 844         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 845 #endif
 846
 847 static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
 848                                   struct ext3_ext_path *path,
 849                                   struct ext3_ext_cache *cex)
 850 {
 851         struct inode *inode = tree->inode;
 852         struct bpointers *bp = tree->private;
 853         struct ext3_extent nex;
 854         int count, err, goal;
 855         unsigned long pblock;
 856         unsigned long tgen;
 857         loff_t new_i_size;
 858         handle_t *handle;
 859         int i, aflags = 0;
 860
 861         i = EXT_DEPTH(tree);
 862         EXT_ASSERT(i == path->p_depth);
 863         EXT_ASSERT(path[i].p_hdr);
 864
 865         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 866                 err = EXT_CONTINUE;
 867                 goto map;
 868         }
 869
 870         if (bp->create == 0) {
 871                 i = 0;
 872                 if (cex->ec_block < bp->start)
 873                         i = bp->start - cex->ec_block;
 874                 if (i >= cex->ec_len)
 875                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 876                                         i, cex->ec_len);
 877                 for (; i < cex->ec_len && bp->num; i++) {
 878                         *(bp->created) = 0;
 879                         bp->created++;
 880                         *(bp->blocks) = 0;
 881                         bp->blocks++;
 882                         bp->num--;
 883                         bp->start++;
 884                 }
 885
 886                 return EXT_CONTINUE;
 887         }
 888
 889         tgen = EXT_GENERATION(tree);
 890         count = ext3_ext_calc_credits_for_insert(tree, path);
 891         ext3_up_truncate_sem(inode);
 892
 893         lock_24kernel();
 894         handle = journal_start(EXT3_JOURNAL(inode), count+EXT3_ALLOC_NEEDED+1);
 895         unlock_24kernel();
 896         if (IS_ERR(handle)) {
 897                 ext3_down_truncate_sem(inode);
 898                 return PTR_ERR(handle);
 899         }
 900
 901         ext3_down_truncate_sem(inode);
 902         if (tgen != EXT_GENERATION(tree)) {
 903                 /* the tree has changed. so path can be invalid at moment */
 904                 lock_24kernel();
 905                 journal_stop(handle);
 906                 unlock_24kernel();
 907                 return EXT_REPEAT;
 908         }
 909
 910         count = cex->ec_len;
 911         goal = ext3_ext_find_goal(inode, path, cex->ec_block, &aflags);
 912         aflags |= 2; /* block have been already reserved */
 913         lock_24kernel();
 914         pblock = ext3_mb_new_blocks(handle, inode, goal, &count, aflags, &err);
 915         unlock_24kernel();
 916         if (!pblock)
 917                 goto out;
 918         EXT_ASSERT(count <= cex->ec_len);
 919
 920         /* insert new extent */
 921         nex.ee_block = cex->ec_block;
 922         nex.ee_start = pblock;
 923         nex.ee_len = count;
 924         err = ext3_ext_insert_extent(handle, tree, path, &nex);
 925         if (err)
 926                 goto out;
 927
 928         /*
 929          * Putting len of the actual extent we just inserted,
 930          * we are asking ext3_ext_walk_space() to continue
 931          * scaning after that block
 932          */
 933         cex->ec_len = nex.ee_len;
 934         cex->ec_start = nex.ee_start;
 935         BUG_ON(nex.ee_len == 0);
 936         BUG_ON(nex.ee_block != cex->ec_block);
 937
 938         /* correct on-disk inode size */
 939         if (nex.ee_len > 0) {
 940                 new_i_size = (loff_t) nex.ee_block + nex.ee_len;
 941                 new_i_size = new_i_size << inode->i_blkbits;
 942                 if (new_i_size > EXT3_I(inode)->i_disksize) {
 943                         EXT3_I(inode)->i_disksize = new_i_size;
 944                         err = ext3_mark_inode_dirty(handle, inode);
 945                 }
 946         }
 947
 948 out:
 949         lock_24kernel();
 950         journal_stop(handle);
 951         unlock_24kernel();
 952 map:
 953         if (err >= 0) {
 954                 /* map blocks */
 955                 if (bp->num == 0) {
 956                         CERROR("hmm. why do we find this extent?\n");
 957                         CERROR("initial space: %lu:%u\n",
 958                                 bp->start, bp->init_num);
 959                         CERROR("current extent: %u/%u/%u %d\n",
 960                                 cex->ec_block, cex->ec_len,
 961                                 cex->ec_start, cex->ec_type);
 962                 }
 963                 i = 0;
 964                 if (cex->ec_block < bp->start)
 965                         i = bp->start - cex->ec_block;
 966                 if (i >= cex->ec_len)
 967                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 968                                         i, cex->ec_len);
 969                 for (; i < cex->ec_len && bp->num; i++) {
 970                         *(bp->blocks) = cex->ec_start + i;
 971                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 972                                 *(bp->created) = 0;
 973                         } else {
 974                                 *(bp->created) = 1;
 975                                 /* unmap any possible underlying metadata from
 976                                  * the block device mapping.  bug 6998. */
 977                                 ll_unmap_underlying_metadata(inode->i_sb,
 978                                                              *(bp->blocks));
 979                         }
 980                         bp->created++;
 981                         bp->blocks++;
 982                         bp->num--;
 983                         bp->start++;
 984                 }
 985         }
 986         return err;
 987 }
 988
 989 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
 990                        unsigned long num, unsigned long *blocks,
 991                        int *created, int create)
 992 {
 993         struct ext3_extents_tree tree;
 994         struct bpointers bp;
 995         int err;
 996
 997         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
 998                 block, block + num, (unsigned) inode->i_ino);
 999
1000         ext3_init_tree_desc(&tree, inode);
1001         tree.private = &bp;
1002         bp.blocks = blocks;
1003         bp.created = created;
1004         bp.start = block;
1005         bp.init_num = bp.num = num;
1006         bp.create = create;
1007
1008         ext3_down_truncate_sem(inode);
1009         err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
1010         ext3_ext_invalidate_cache(&tree);
1011         ext3_up_truncate_sem(inode);
1012
1013         return err;
1014 }
1015
1016 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1017                                     int pages, unsigned long *blocks,
1018                                     int *created, int create)
1019 {
1020         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
1021         int rc = 0, i = 0;
1022         struct page *fp = NULL;
1023         int clen = 0;
1024
1025         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1026                 inode->i_ino, pages, (*page)->index);
1027
1028         /* pages are sorted already. so, we just have to find
1029          * contig. space and process them properly */
1030         while (i < pages) {
1031                 if (fp == NULL) {
1032                         /* start new extent */
1033                         fp = *page++;
1034                         clen = 1;
1035                         i++;
1036                         continue;
1037                 } else if (fp->index + clen == (*page)->index) {
1038                         /* continue the extent */
1039                         page++;
1040                         clen++;
1041                         i++;
1042                         continue;
1043                 }
1044
1045                 /* process found extent */
1046                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1047                                         clen * blocks_per_page, blocks,
1048                                         created, create);
1049                 if (rc)
1050                         GOTO(cleanup, rc);
1051
1052                 /* look for next extent */
1053                 fp = NULL;
1054                 blocks += blocks_per_page * clen;
1055                 created += blocks_per_page * clen;
1056         }
1057
1058         if (fp)
1059                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1060                                         clen * blocks_per_page, blocks,
1061                                         created, create);
1062 cleanup:
1063         return rc;
1064 }
1065 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1066
1067 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1068                                unsigned long *blocks, int *created, int create);
1069 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1070                                    int pages, unsigned long *blocks,
1071                                    int *created, int create)
1072 {
1073         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
1074         unsigned long *b;
1075         int rc = 0, i, *cr;
1076
1077         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1078                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1079                 if (rc) {
1080                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1081                                inode->i_ino, *b, *cr, create, rc);
1082                         break;
1083                 }
1084
1085                 b += blocks_per_page;
1086                 cr += blocks_per_page;
1087         }
1088         return rc;
1089 }
1090
1091 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1092                                 int pages, unsigned long *blocks,
1093                                 int *created, int create,
1094                                 struct semaphore *optional_sem)
1095 {
1096         int rc;
1097 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1098         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1099                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1100                                                      blocks, created, create);
1101                 return rc;
1102         }
1103 #endif
1104         if (optional_sem != NULL)
1105                 down(optional_sem);
1106         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1107                                             created, create);
1108         if (optional_sem != NULL)
1109                 up(optional_sem);
1110
1111         return rc;
1112 }
1113
1114 extern int ext3_prep_san_write(struct inode *inode, long *blocks,
1115                                int nblocks, loff_t newsize);
1116 static int fsfilt_ext3_prep_san_write(struct inode *inode, long *blocks,
1117                                       int nblocks, loff_t newsize)
1118 {
1119         return ext3_prep_san_write(inode, blocks, nblocks, newsize);
1120 }
1121
1122 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1123                                    int size, loff_t *offs)
1124 {
1125         struct inode *inode = file->f_dentry->d_inode;
1126         unsigned long block;
1127         struct buffer_head *bh;
1128         int err, blocksize, csize, boffs;
1129
1130         /* prevent reading after eof */
1131         lock_kernel();
1132         if (inode->i_size < *offs + size) {
1133                 size = inode->i_size - *offs;
1134                 unlock_kernel();
1135                 if (size < 0) {
1136                         CERROR("size %llu is too short for read %u@%llu\n",
1137                                inode->i_size, size, *offs);
1138                         return -EIO;
1139                 } else if (size == 0) {
1140                         return 0;
1141                 }
1142         } else {
1143                 unlock_kernel();
1144         }
1145
1146         blocksize = 1 << inode->i_blkbits;
1147
1148         while (size > 0) {
1149                 block = *offs >> inode->i_blkbits;
1150                 boffs = *offs & (blocksize - 1);
1151                 csize = min(blocksize - boffs, size);
1152                 bh = ext3_bread(NULL, inode, block, 0, &err);
1153                 if (!bh) {
1154                         CERROR("can't read block: %d\n", err);
1155                         return err;
1156                 }
1157
1158                 memcpy(buf, bh->b_data + boffs, csize);
1159                 brelse(bh);
1160
1161                 *offs += csize;
1162                 buf += csize;
1163                 size -= csize;
1164         }
1165         return 0;
1166 }
1167
1168 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1169                                     loff_t *offs, int force_sync)
1170 {
1171         struct buffer_head *bh = NULL;
1172         unsigned long block;
1173         struct inode *inode = file->f_dentry->d_inode;
1174         loff_t old_size = inode->i_size, offset = *offs;
1175         loff_t new_size = inode->i_size;
1176         journal_t *journal;
1177         handle_t *handle;
1178         int err, block_count = 0, blocksize, size, boffs;
1179
1180         /* Determine how many transaction credits are needed */
1181         blocksize = 1 << inode->i_blkbits;
1182         block_count = (*offs & (blocksize - 1)) + bufsize;
1183         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1184
1185         journal = EXT3_SB(inode->i_sb)->s_journal;
1186         lock_24kernel();
1187         handle = journal_start(journal,
1188                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1189         unlock_24kernel();
1190         if (IS_ERR(handle)) {
1191                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1192                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1193                 return PTR_ERR(handle);
1194         }
1195
1196         while (bufsize > 0) {
1197                 if (bh != NULL)
1198                         brelse(bh);
1199
1200                 block = offset >> inode->i_blkbits;
1201                 boffs = offset & (blocksize - 1);
1202                 size = min(blocksize - boffs, bufsize);
1203                 bh = ext3_bread(handle, inode, block, 1, &err);
1204                 if (!bh) {
1205                         CERROR("can't read/create block: %d\n", err);
1206                         goto out;
1207                 }
1208
1209                 err = ext3_journal_get_write_access(handle, bh);
1210                 if (err) {
1211                         CERROR("journal_get_write_access() returned error %d\n",
1212                                err);
1213                         goto out;
1214                 }
1215                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1216                 memcpy(bh->b_data + boffs, buf, size);
1217                 err = ext3_journal_dirty_metadata(handle, bh);
1218                 if (err) {
1219                         CERROR("journal_dirty_metadata() returned error %d\n",
1220                                err);
1221                         goto out;
1222                 }
1223                 if (offset + size > new_size)
1224                         new_size = offset + size;
1225                 offset += size;
1226                 bufsize -= size;
1227                 buf += size;
1228         }
1229
1230         if (force_sync)
1231                 handle->h_sync = 1; /* recovery likes this */
1232 out:
1233         if (bh)
1234                 brelse(bh);
1235
1236         /* correct in-core and on-disk sizes */
1237         if (new_size > inode->i_size) {
1238                 lock_kernel();
1239                 if (new_size > inode->i_size)
1240                         inode->i_size = new_size;
1241                 if (inode->i_size > EXT3_I(inode)->i_disksize)
1242                         EXT3_I(inode)->i_disksize = inode->i_size;
1243                 if (inode->i_size > old_size)
1244                         mark_inode_dirty(inode);
1245                 unlock_kernel();
1246         }
1247
1248         lock_24kernel();
1249         journal_stop(handle);
1250         unlock_24kernel();
1251
1252         if (err == 0)
1253                 *offs = offset;
1254         return err;
1255 }
1256
1257 static int fsfilt_ext3_setup(struct super_block *sb)
1258 {
1259 #if 0
1260         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1261         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1262 #endif
1263 #ifdef S_PDIROPS
1264         CWARN("Enabling PDIROPS\n");
1265         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1266         sb->s_flags |= S_PDIROPS;
1267 #endif
1268         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1269                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1270         return 0;
1271 }
1272
1273 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1274    objects. Logs is number of logfiles to update */
1275 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1276 {
1277         if ( !fso ) {
1278                 switch(op) {
1279                 case FSFILT_OP_CREATE:
1280                                  /* directory leaf, index & indirect & EA*/
1281                         return 4 + 3 * logs;
1282                 case FSFILT_OP_UNLINK:
1283                         return 3 * logs;
1284                 }
1285         } else {
1286                 int i;
1287                 int needed = 0;
1288                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1289                 int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
1290                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1291                 for (i = 0; i < op; i++, fso++) {
1292                         int nblocks = fso->fso_bufcnt * blockpp;
1293                         int ndindirect = min(nblocks, addrpp + 1);
1294                         int nindir = nblocks + ndindirect + 1;
1295
1296                         needed += nindir;
1297                 }
1298                 return needed + 3 * logs;
1299         }
1300
1301         return 0;
1302 }
1303
1304 static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
1305
1306 #define DQINFO_COPY(out, in)                    \
1307 do {                                            \
1308         Q_COPY(out, in, dqi_bgrace);            \
1309         Q_COPY(out, in, dqi_igrace);            \
1310         Q_COPY(out, in, dqi_flags);             \
1311         Q_COPY(out, in, dqi_valid);             \
1312 } while (0)
1313
1314 #define DQBLK_COPY(out, in)                     \
1315 do {                                            \
1316         Q_COPY(out, in, dqb_bhardlimit);        \
1317         Q_COPY(out, in, dqb_bsoftlimit);        \
1318         Q_COPY(out, in, dqb_curspace);          \
1319         Q_COPY(out, in, dqb_ihardlimit);        \
1320         Q_COPY(out, in, dqb_isoftlimit);        \
1321         Q_COPY(out, in, dqb_curinodes);         \
1322         Q_COPY(out, in, dqb_btime);             \
1323         Q_COPY(out, in, dqb_itime);             \
1324         Q_COPY(out, in, dqb_valid);             \
1325 } while (0)
1326
1327
1328
1329 static int fsfilt_ext3_quotactl(struct super_block *sb,
1330                                 struct obd_quotactl *oqc)
1331 {
1332         int i, rc = 0, error = 0;
1333         struct quotactl_ops *qcop;
1334         struct if_dqinfo *info;
1335         struct if_dqblk *dqblk;
1336         ENTRY;
1337
1338         if (!sb->s_qcop)
1339                 RETURN(-ENOSYS);
1340
1341         OBD_ALLOC_PTR(info);
1342         if (!info)
1343                 RETURN(-ENOMEM);
1344         OBD_ALLOC_PTR(dqblk);
1345         if (!dqblk) {
1346                 OBD_FREE_PTR(info);
1347                 RETURN(-ENOMEM);
1348         }
1349
1350         DQINFO_COPY(info, &oqc->qc_dqinfo);
1351         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1352
1353         qcop = sb->s_qcop;
1354         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1355                 for (i = 0; i < MAXQUOTAS; i++) {
1356                         if (!Q_TYPESET(oqc, i))
1357                                 continue;
1358
1359                         if (oqc->qc_cmd == Q_QUOTAON) {
1360                                 if (!qcop->quota_on)
1361                                         GOTO(out, rc = -ENOSYS);
1362                                 rc = qcop->quota_on(sb, i, oqc->qc_id,
1363                                                     (char *)op_quotafile[i]);
1364                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1365                                 if (!qcop->quota_off)
1366                                         GOTO(out, rc = -ENOSYS);
1367                                 rc = qcop->quota_off(sb, i);
1368                         }
1369
1370                         if (rc == -EBUSY)
1371                                 error = rc;
1372                         else if (rc)
1373                                 GOTO(out, rc);
1374                 }
1375                 GOTO(out, rc ?: error);
1376         }
1377
1378         switch (oqc->qc_cmd) {
1379         case Q_GETOINFO:
1380         case Q_GETINFO:
1381                 if (!qcop->get_info)
1382                         GOTO(out, rc = -ENOSYS);
1383                 rc = qcop->get_info(sb, oqc->qc_type, info);
1384                 break;
1385         case Q_SETQUOTA:
1386         case Q_INITQUOTA:
1387                 if (!qcop->set_dqblk)
1388                         GOTO(out, rc = -ENOSYS);
1389                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1390                 break;
1391         case Q_GETOQUOTA:
1392         case Q_GETQUOTA:
1393                 if (!qcop->get_dqblk)
1394                         GOTO(out, rc = -ENOSYS);
1395                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1396                 break;
1397         case Q_SYNC:
1398                 if (!sb->s_qcop->quota_sync)
1399                         GOTO(out, rc = -ENOSYS);
1400                 qcop->quota_sync(sb, oqc->qc_type);
1401                 break;
1402         default:
1403                 CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
1404                 LBUG();
1405         }
1406 out:
1407         DQINFO_COPY(&oqc->qc_dqinfo, info);
1408         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1409
1410         OBD_FREE_PTR(info);
1411         OBD_FREE_PTR(dqblk);
1412
1413         if (rc)
1414                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1415                                 "failed: %d\n",
1416                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1417         RETURN(rc);
1418 }
1419
1420 struct chk_dqblk{
1421         struct hlist_node       dqb_hash;        /* quotacheck hash */
1422         struct list_head        dqb_list;        /* in list also */
1423         qid_t                   dqb_id;          /* uid/gid */
1424         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1425         __u32                   dqb_bhardlimit;  /* block hard limit */
1426         __u32                   dqb_bsoftlimit;  /* block soft limit */
1427         qsize_t                 dqb_curspace;    /* current space */
1428         __u32                   dqb_ihardlimit;  /* inode hard limit */
1429         __u32                   dqb_isoftlimit;  /* inode soft limit */
1430         __u32                   dqb_curinodes;   /* current inodes */
1431         __u64                   dqb_btime;       /* block grace time */
1432         __u64                   dqb_itime;       /* inode grace time */
1433         __u32                   dqb_valid;       /* flag for above fields */
1434 };
1435
1436 static inline unsigned int chkquot_hash(qid_t id, int type)
1437                                         __attribute__((__const__));
1438
1439 static inline unsigned int chkquot_hash(qid_t id, int type)
1440 {
1441         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1442 }
1443
1444 static inline struct chk_dqblk *
1445 find_chkquot(struct hlist_head *head, qid_t id, int type)
1446 {
1447         struct hlist_node *node;
1448         struct chk_dqblk *cdqb;
1449
1450         hlist_for_each(node, head) {
1451                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1452                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1453                         return cdqb;
1454         }
1455
1456         return NULL;
1457 }
1458
1459 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1460 {
1461         struct chk_dqblk *cdqb;
1462
1463         OBD_ALLOC_PTR(cdqb);
1464         if (cdqb) {
1465                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1466                 INIT_LIST_HEAD(&cdqb->dqb_list);
1467                 cdqb->dqb_id = id;
1468                 cdqb->dqb_type = type;
1469         }
1470
1471         return cdqb;
1472 }
1473
1474 static struct chk_dqblk *
1475 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1476       qid_t id, int type, int first_check)
1477 {
1478         struct hlist_head *head = hash + chkquot_hash(id, type);
1479         struct if_dqblk dqb;
1480         struct chk_dqblk *cdqb;
1481         int rc;
1482
1483         cdqb = find_chkquot(head, id, type);
1484         if (cdqb)
1485                 return cdqb;
1486
1487         cdqb = alloc_chkquot(id, type);
1488         if (!cdqb)
1489                 return NULL;
1490
1491         if (!first_check) {
1492                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1493                 if (rc) {
1494                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1495                                id, type, rc);
1496                 } else {
1497                         DQBLK_COPY(cdqb, &dqb);
1498                         cdqb->dqb_curspace = 0;
1499                         cdqb->dqb_curinodes = 0;
1500                 }
1501         }
1502
1503         hlist_add_head(&cdqb->dqb_hash, head);
1504         list_add_tail(&cdqb->dqb_list, list);
1505
1506         return cdqb;
1507 }
1508
1509 static inline int quota_onoff(struct super_block *sb, int cmd, int type)
1510 {
1511         struct obd_quotactl *oqctl;
1512         int rc;
1513
1514         OBD_ALLOC_PTR(oqctl);
1515         if (!oqctl)
1516                 RETURN(-ENOMEM);
1517
1518         oqctl->qc_cmd = cmd;
1519         oqctl->qc_id = QFMT_LDISKFS;
1520         oqctl->qc_type = type;
1521         rc = fsfilt_ext3_quotactl(sb, oqctl);
1522
1523         OBD_FREE_PTR(oqctl);
1524         return rc;
1525 }
1526
1527 static inline int read_old_dqinfo(struct super_block *sb, int type,
1528                                   struct if_dqinfo *dqinfo)
1529 {
1530         struct obd_quotactl *oqctl;
1531         int rc;
1532         ENTRY;
1533
1534         OBD_ALLOC_PTR(oqctl);
1535         if (!oqctl)
1536                 RETURN(-ENOMEM);
1537
1538         oqctl->qc_cmd = Q_GETINFO;
1539         oqctl->qc_type = type;
1540         rc = fsfilt_ext3_quotactl(sb, oqctl);
1541         if (!rc)
1542                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1543
1544         OBD_FREE_PTR(oqctl);
1545         RETURN(rc);
1546 }
1547
1548 static inline struct ext3_group_desc *
1549 get_group_desc(struct super_block *sb, int group)
1550 {
1551         unsigned long desc_block, desc;
1552         struct ext3_group_desc *gdp;
1553
1554         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1555         desc = group % EXT3_DESC_PER_BLOCK(sb);
1556         gdp = (struct ext3_group_desc *)
1557               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1558
1559         return gdp + desc;
1560 }
1561
1562 static inline struct buffer_head *
1563 read_inode_bitmap(struct super_block *sb, unsigned long group)
1564 {
1565         struct ext3_group_desc *desc;
1566         struct buffer_head *bh;
1567
1568         desc = get_group_desc(sb, group);
1569         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1570
1571         return bh;
1572 }
1573
1574 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1575                                      struct buffer_head *bitmap_bh,
1576                                      int index, unsigned long ino)
1577 {
1578         struct inode *inode = NULL;
1579
1580         if (ext3_test_bit(index, bitmap_bh->b_data))
1581                 inode = iget(sb, ino);
1582
1583         return inode;
1584 }
1585
1586 struct qchk_ctxt {
1587         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1588         struct list_head        qckt_list;                   /* quotacheck list */
1589         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1590         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1591 };
1592
1593 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1594                            struct obd_quotactl *oqc)
1595 {
1596         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1597         loff_t size = 0;
1598         qid_t qid[MAXQUOTAS];
1599         int cnt, i, rc = 0;
1600
1601         if (!inode)
1602                 return 0;
1603
1604         qid[USRQUOTA] = inode->i_uid;
1605         qid[GRPQUOTA] = inode->i_gid;
1606
1607         if (S_ISDIR(inode->i_mode) ||
1608             S_ISREG(inode->i_mode) ||
1609             S_ISLNK(inode->i_mode))
1610                 size = inode_get_bytes(inode);
1611
1612         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1613                 if (!Q_TYPESET(oqc, cnt))
1614                         continue;
1615
1616                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1617                                 &qctxt->qckt_list, qid[cnt], cnt,
1618                                 qctxt->qckt_first_check[cnt]);
1619                 if (!cdqb[cnt]) {
1620                         rc = -ENOMEM;
1621                         break;
1622                 }
1623
1624                 cdqb[cnt]->dqb_curspace += size;
1625                 cdqb[cnt]->dqb_curinodes++;
1626         }
1627
1628         if (rc) {
1629                 for (i = 0; i < cnt; i++) {
1630                         if (!Q_TYPESET(oqc, i))
1631                                 continue;
1632                         LASSERT(cdqb[i]);
1633                         cdqb[i]->dqb_curspace -= size;
1634                         cdqb[i]->dqb_curinodes--;
1635                 }
1636         }
1637
1638         return rc;
1639 }
1640
1641 static int v2_write_dqheader(struct file *f, int type)
1642 {
1643         static const __u32 quota_magics[] = V2_INITQMAGICS;
1644         static const __u32 quota_versions[] = V2_INITQVERSIONS;
1645         struct v2_disk_dqheader dqhead;
1646         loff_t offset = 0;
1647
1648         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1649         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1650
1651         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1652         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1653
1654         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1655 }
1656
1657 /* write dqinfo struct in a new quota file */
1658 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1659 {
1660         struct v2_disk_dqinfo dqinfo;
1661         __u32 blocks = V2_DQTREEOFF + 1;
1662         loff_t offset = V2_DQINFOOFF;
1663
1664         if (info) {
1665                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1666                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1667                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1668                                                ~DQF_INFO_DIRTY);
1669         } else {
1670                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1671                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1672                 dqinfo.dqi_flags = 0;
1673         }
1674
1675         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1676         dqinfo.dqi_free_blk = 0;
1677         dqinfo.dqi_free_entry = 0;
1678
1679         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1680 }
1681
1682 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1683                                   struct obd_quotactl *oqc)
1684 {
1685         int i, rc = 0;
1686         ENTRY;
1687
1688         for (i = 0; i < MAXQUOTAS; i++) {
1689                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1690                                          NULL : &qctxt->qckt_dqinfo[i];
1691                 struct file *file;
1692
1693                 if (!Q_TYPESET(oqc, i))
1694                         continue;
1695
1696                 file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
1697                                  0644);
1698                 if (IS_ERR(file)) {
1699                         rc = PTR_ERR(file);
1700                         CERROR("can't create %s file: rc = %d\n",
1701                                op_quotafile[i], rc);
1702                         GOTO(out, rc);
1703                 }
1704
1705                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1706                         CERROR("file %s is not regular", op_quotafile[i]);
1707                         filp_close(file, 0);
1708                         GOTO(out, rc = -EINVAL);
1709                 }
1710
1711                 rc = v2_write_dqheader(file, i);
1712                 if (rc) {
1713                         filp_close(file, 0);
1714                         GOTO(out, rc);
1715                 }
1716
1717                 rc = v2_write_dqinfo(file, i, info);
1718                 filp_close(file, 0);
1719                 if (rc)
1720                         GOTO(out, rc);
1721         }
1722
1723 out:
1724         RETURN(rc);
1725 }
1726
1727
1728 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1729                           struct chk_dqblk *cdqb)
1730 {
1731         struct obd_quotactl *oqc;
1732         struct timeval now;
1733         int rc;
1734         ENTRY;
1735
1736         OBD_ALLOC_PTR(oqc);
1737         if (!oqc)
1738                 RETURN(-ENOMEM);
1739
1740         do_gettimeofday(&now);
1741
1742         if (cdqb->dqb_bsoftlimit &&
1743             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1744             !cdqb->dqb_btime)
1745                 cdqb->dqb_btime = now.tv_sec +
1746                                qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1747
1748         if (cdqb->dqb_isoftlimit &&
1749             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1750             !cdqb->dqb_itime)
1751                 cdqb->dqb_itime = now.tv_sec +
1752                                qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1753
1754         cdqb->dqb_valid = QIF_ALL;
1755
1756         oqc->qc_cmd = Q_SETQUOTA;
1757         oqc->qc_type = cdqb->dqb_type;
1758         oqc->qc_id = cdqb->dqb_id;
1759         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1760
1761         rc = fsfilt_ext3_quotactl(sb, oqc);
1762         OBD_FREE_PTR(oqc);
1763         RETURN(rc);
1764 }
1765
1766 static int prune_chkquots(struct super_block *sb,
1767                           struct qchk_ctxt *qctxt, int error)
1768 {
1769         struct chk_dqblk *cdqb, *tmp;
1770         int rc;
1771
1772         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1773                 if (!error) {
1774                         rc = commit_chkquot(sb, qctxt, cdqb);
1775                         if (rc)
1776                                 error = rc;
1777                 }
1778                 hlist_del_init(&cdqb->dqb_hash);
1779                 list_del(&cdqb->dqb_list);
1780                 OBD_FREE_PTR(cdqb);
1781         }
1782
1783         return error;
1784 }
1785
1786 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1787                                   struct obd_quotactl *oqc)
1788 {
1789         struct ext3_sb_info *sbi = EXT3_SB(sb);
1790         int i, group;
1791         struct qchk_ctxt *qctxt;
1792         struct buffer_head *bitmap_bh = NULL;
1793         unsigned long ino;
1794         struct inode *inode;
1795         int rc = 0;
1796         ENTRY;
1797
1798         /* turn on quota and read dqinfo if existed */
1799         OBD_ALLOC_PTR(qctxt);
1800         if (!qctxt) {
1801                 oqc->qc_stat = -ENOMEM;
1802                 RETURN(-ENOMEM);
1803         }
1804
1805         for (i = 0; i < NR_DQHASH; i++)
1806                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1807         INIT_LIST_HEAD(&qctxt->qckt_list);
1808
1809         for (i = 0; i < MAXQUOTAS; i++) {
1810                 if (!Q_TYPESET(oqc, i))
1811                         continue;
1812
1813                 rc = quota_onoff(sb, Q_QUOTAON, i);
1814                 if (!rc || rc == -EBUSY) {
1815                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1816                         if (rc)
1817                                 GOTO(out, rc);
1818                 } else if (rc == -ENOENT) {
1819                         qctxt->qckt_first_check[i] = 1;
1820                 } else if (rc) {
1821                         GOTO(out, rc);
1822                 }
1823         }
1824
1825         /* check quota and update in hash */
1826         for (group = 0; group < sbi->s_groups_count; group++) {
1827                 ino = group * sbi->s_inodes_per_group + 1;
1828                 bitmap_bh = read_inode_bitmap(sb, group);
1829                 if (!bitmap_bh) {
1830                         CERROR("read_inode_bitmap group %d failed", group);
1831                         GOTO(out, -EIO);
1832                 }
1833
1834                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1835                         if (ino < sbi->s_first_ino)
1836                                 continue;
1837
1838                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1839                         rc = add_inode_quota(inode, qctxt, oqc);
1840                         iput(inode);
1841                         if (rc) {
1842                                 brelse(bitmap_bh);
1843                                 GOTO(out, rc);
1844                         }
1845                 }
1846
1847                 brelse(bitmap_bh);
1848         }
1849
1850         /* read old quota limits from old quota file. (only for the user
1851          * has limits but hasn't file) */
1852 #ifdef HAVE_QUOTA_SUPPORT
1853         for (i = 0; i < MAXQUOTAS; i++) {
1854                 struct list_head id_list;
1855                 struct dquot_id *dqid, *tmp;
1856
1857                 if (!Q_TYPESET(oqc, i))
1858                         continue;
1859
1860                 if (qctxt->qckt_first_check[i])
1861                         continue;
1862
1863
1864                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1865                 INIT_LIST_HEAD(&id_list);
1866 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,12)
1867                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1868 #else
1869                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1870 #endif
1871                 if (rc)
1872                         CERROR("read old limits failed. (rc:%d)\n", rc);
1873
1874                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1875                         list_del_init(&dqid->di_link);
1876
1877                         if (!rc)
1878                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1879                                       dqid->di_id, i,
1880                                       qctxt->qckt_first_check[i]);
1881                         kfree(dqid);
1882                 }
1883         }
1884 #endif
1885         /* turn off quota cause we are to dump chk_dqblk to files */
1886         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1887
1888         rc = create_new_quota_files(qctxt, oqc);
1889         if (rc)
1890                 GOTO(out, rc);
1891
1892         /* we use vfs functions to set dqblk, so turn quota on */
1893         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
1894 out:
1895         /* dump and free chk_dqblk */
1896         rc = prune_chkquots(sb, qctxt, rc);
1897         OBD_FREE_PTR(qctxt);
1898
1899         /* turn off quota, `lfs quotacheck` will turn on when all
1900          * nodes quotacheck finish. */
1901         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1902
1903         oqc->qc_stat = rc;
1904         if (rc)
1905                 CERROR("quotacheck failed: rc = %d\n", rc);
1906
1907         RETURN(rc);
1908 }
1909
1910 #ifdef HAVE_QUOTA_SUPPORT
1911 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
1912                                  int cmd)
1913 {
1914         int rc = 0;
1915         ENTRY;
1916
1917         if (lqi->qi_files[type] == NULL) {
1918                 CERROR("operate qinfo before it's enabled!\n");
1919                 RETURN(-EIO);
1920         }
1921
1922         switch (cmd) {
1923         case QFILE_CHK:
1924                 rc = lustre_check_quota_file(lqi, type);
1925                 break;
1926         case QFILE_RD_INFO:
1927                 rc = lustre_read_quota_info(lqi, type);
1928                 break;
1929         case QFILE_WR_INFO:
1930                 rc = lustre_write_quota_info(lqi, type);
1931                 break;
1932         case QFILE_INIT_INFO:
1933                 rc = lustre_init_quota_info(lqi, type);
1934                 break;
1935         default:
1936                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1937                 LBUG();
1938                 break;
1939         }
1940         RETURN(rc);
1941 }
1942
1943 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
1944                             struct list_head *list)
1945 {
1946         return lustre_get_qids(file, inode, type, list);
1947 }
1948
1949 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
1950 {
1951         int rc = 0;
1952         ENTRY;
1953
1954         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
1955                 CERROR("operate dquot before it's enabled!\n");
1956                 RETURN(-EIO);
1957         }
1958
1959         switch (cmd) {
1960         case QFILE_RD_DQUOT:
1961                 rc = lustre_read_dquot(dquot);
1962                 break;
1963         case QFILE_WR_DQUOT:
1964                 if (dquot->dq_dqb.dqb_ihardlimit ||
1965                     dquot->dq_dqb.dqb_isoftlimit ||
1966                     dquot->dq_dqb.dqb_bhardlimit ||
1967                     dquot->dq_dqb.dqb_bsoftlimit)
1968                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1969                 else
1970                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
1971
1972                 rc = lustre_commit_dquot(dquot);
1973                 if (rc >= 0)
1974                         rc = 0;
1975                 break;
1976         default:
1977                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1978                 LBUG();
1979                 break;
1980         }
1981         RETURN(rc);
1982 }
1983 #endif
1984
1985 static struct fsfilt_operations fsfilt_ext3_ops = {
1986         .fs_type                = "ext3",
1987         .fs_owner               = THIS_MODULE,
1988         .fs_getlabel            = fsfilt_ext3_get_label,
1989         .fs_setlabel            = fsfilt_ext3_set_label,
1990         .fs_uuid                = fsfilt_ext3_uuid,
1991         .fs_start               = fsfilt_ext3_start,
1992         .fs_brw_start           = fsfilt_ext3_brw_start,
1993         .fs_commit              = fsfilt_ext3_commit,
1994         .fs_commit_async        = fsfilt_ext3_commit_async,
1995         .fs_commit_wait         = fsfilt_ext3_commit_wait,
1996         .fs_setattr             = fsfilt_ext3_setattr,
1997         .fs_iocontrol           = fsfilt_ext3_iocontrol,
1998         .fs_set_md              = fsfilt_ext3_set_md,
1999         .fs_get_md              = fsfilt_ext3_get_md,
2000         .fs_readpage            = fsfilt_ext3_readpage,
2001         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2002         .fs_statfs              = fsfilt_ext3_statfs,
2003         .fs_sync                = fsfilt_ext3_sync,
2004         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2005         .fs_prep_san_write      = fsfilt_ext3_prep_san_write,
2006         .fs_write_record        = fsfilt_ext3_write_record,
2007         .fs_read_record         = fsfilt_ext3_read_record,
2008         .fs_setup               = fsfilt_ext3_setup,
2009         .fs_send_bio            = fsfilt_ext3_send_bio,
2010         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2011         .fs_quotactl            = fsfilt_ext3_quotactl,
2012         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2013 #ifdef HAVE_QUOTA_SUPPORT
2014         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2015         .fs_qids                = fsfilt_ext3_qids,
2016         .fs_dquot               = fsfilt_ext3_dquot,
2017 #endif
2018 };
2019
2020 static int __init fsfilt_ext3_init(void)
2021 {
2022         int rc;
2023
2024         fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
2025                                       sizeof(struct fsfilt_cb_data), 0,
2026                                       0, NULL, NULL);
2027         if (!fcb_cache) {
2028                 CERROR("error allocating fsfilt journal callback cache\n");
2029                 GOTO(out, rc = -ENOMEM);
2030         }
2031
2032         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2033
2034         if (rc)
2035                 kmem_cache_destroy(fcb_cache);
2036 out:
2037         return rc;
2038 }
2039
2040 static void __exit fsfilt_ext3_exit(void)
2041 {
2042         int rc;
2043
2044         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2045         rc = kmem_cache_destroy(fcb_cache);
2046         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2047 }
2048
2049 module_init(fsfilt_ext3_init);
2050 module_exit(fsfilt_ext3_exit);
2051
2052 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2053 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2054 MODULE_LICENSE("GPL");