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 #include <ext3/xattr.h>
  43
  44 #include <libcfs/kp30.h>
  45 #include <lustre_fsfilt.h>
  46 #include <obd.h>
  47 #include <obd_class.h>
  48 #include <lustre_quota.h>
  49 #include <linux/lustre_compat25.h>
  50 #include <linux/lprocfs_status.h>
  51
  52 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  53 #include <linux/ext3_extents.h>
  54 #endif
  55
  56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  57 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  58 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  59 #else
  60 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  61 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  62 #endif
  63
  64 #define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks)
  65 #define fsfilt_ext3_journal_stop(handle)          ext3_journal_stop(handle)
  66
  67 static cfs_mem_cache_t *fcb_cache;
  68
  69 struct fsfilt_cb_data {
  70         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  71         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  72         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  73         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  74         void *cb_data;                  /* MDS/OST completion function data */
  75 };
  76
  77 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
  78 #define EXT3_XATTR_INDEX_TRUSTED        4
  79 #endif
  80
  81 static char *fsfilt_ext3_get_label(struct super_block *sb)
  82 {
  83         return EXT3_SB(sb)->s_es->s_volume_name;
  84 }
  85
  86 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
  87 {
  88         /* see e.g. fsfilt_ext3_write_record() */
  89         journal_t *journal;
  90         handle_t *handle;
  91         int err;
  92
  93         journal = EXT3_SB(sb)->s_journal;
  94         handle = journal_start(journal, 1);
  95         if (IS_ERR(handle)) {
  96                 CERROR("can't start transaction\n");
  97                 return(PTR_ERR(handle));
  98         }
  99
 100         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 101         if (err)
 102                 goto out;
 103
 104         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 105                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 106
 107         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 108
 109 out:
 110         journal_stop(handle);
 111
 112         return(err);
 113 }
 114
 115 static char *fsfilt_ext3_uuid(struct super_block *sb)
 116 {
 117         return EXT3_SB(sb)->s_es->s_uuid;
 118 }
 119
 120 #ifdef HAVE_DISK_INODE_VERSION
 121 /*
 122  * Get the 64-bit version for an inode.
 123  */
 124 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 125 {
 126         return EXT3_I(inode)->i_fs_version;
 127 }
 128
 129 /*
 130  * Set the 64-bit version and return the old version.
 131  */
 132 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 133 {
 134         __u64 old_version = EXT3_I(inode)->i_fs_version;
 135
 136         (EXT3_I(inode))->i_fs_version = new_version;
 137         return old_version;
 138 }
 139
 140 #endif
 141
 142 /*
 143  * We don't currently need any additional blocks for rmdir and
 144  * unlink transactions because we are storing the OST oa_id inside
 145  * the inode (which we will be changing anyways as part of this
 146  * transaction).
 147  */
 148 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 149                                int logs)
 150 {
 151         /* For updates to the last received file */
 152         int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS;
 153         journal_t *journal;
 154         void *handle;
 155
 156         if (current->journal_info) {
 157                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 158                        current->journal_info);
 159                 goto journal_start;
 160         }
 161
 162         switch(op) {
 163         case FSFILT_OP_RMDIR:
 164         case FSFILT_OP_UNLINK:
 165                 /* delete one file + create/update logs for each stripe */
 166                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 167                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 168                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 169                 break;
 170         case FSFILT_OP_RENAME:
 171                 /* modify additional directory */
 172                 nblocks += EXT3_SINGLEDATA_TRANS_BLOCKS;
 173                 /* no break */
 174         case FSFILT_OP_SYMLINK:
 175                 /* additional block + block bitmap + GDT for long symlink */
 176                 nblocks += 3;
 177                 /* no break */
 178         case FSFILT_OP_CREATE: {
 179 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 180                 static int warned;
 181                 if (!warned) {
 182                         if (!test_opt(inode->i_sb, EXTENTS)) {
 183                                 warned = 1;
 184                         } else if (((EXT3_I(inode)->i_flags &
 185                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 186                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 187                                 CWARN("extent-mapped directory found - contact "
 188                                       "CFS: support@clusterfs.com\n");
 189                                 warned = 1;
 190                         }
 191                 }
 192 #endif
 193                 /* no break */
 194         }
 195         case FSFILT_OP_MKDIR:
 196         case FSFILT_OP_MKNOD:
 197                 /* modify one inode + block bitmap + GDT */
 198                 nblocks += 3;
 199                 /* no break */
 200         case FSFILT_OP_LINK:
 201                 /* modify parent directory */
 202                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 203                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 204                 /* create/update logs for each stripe */
 205                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 206                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 207                 break;
 208         case FSFILT_OP_SETATTR:
 209                 /* Setattr on inode */
 210                 nblocks += 1;
 211                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 212                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 213                 /* quota chown log for each stripe */
 214                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 215                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 216                 break;
 217         case FSFILT_OP_CANCEL_UNLINK:
 218                 /* blocks for log header bitmap update OR
 219                  * blocks for catalog header bitmap update + unlink of logs */
 220                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 221                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 222                 break;
 223         case FSFILT_OP_JOIN:
 224                 /* delete 2 file(file + array id) + create 1 file (array id)
 225                  * create/update logs for each stripe */
 226                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 227
 228                 /*create array log for head file*/
 229                 nblocks += 3;
 230                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 231                             EXT3_SINGLEDATA_TRANS_BLOCKS);
 232                 /*update head file array */
 233                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 234                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 235                 break;
 236         default: CERROR("unknown transaction start op %d\n", op);
 237                 LBUG();
 238         }
 239
 240         LASSERT(current->journal_info == desc_private);
 241         journal = EXT3_SB(inode->i_sb)->s_journal;
 242         if (nblocks > journal->j_max_transaction_buffers) {
 243                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 244                        nblocks, op, logs, journal->j_max_transaction_buffers);
 245                 nblocks = journal->j_max_transaction_buffers;
 246         }
 247
 248  journal_start:
 249         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 250         handle = fsfilt_ext3_journal_start(inode, nblocks);
 251
 252         if (!IS_ERR(handle))
 253                 LASSERT(current->journal_info == handle);
 254         else
 255                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 256                        op, nblocks, PTR_ERR(handle));
 257         return handle;
 258 }
 259
 260 /*
 261  * Calculate the number of buffer credits needed to write multiple pages in
 262  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 263  * doesn't have a nice API for calculating this sort of thing in advance.
 264  *
 265  * See comment above ext3_writepage_trans_blocks for details.  We assume
 266  * no data journaling is being done, but it does allow for all of the pages
 267  * being non-contiguous.  If we are guaranteed contiguous pages we could
 268  * reduce the number of (d)indirect blocks a lot.
 269  *
 270  * With N blocks per page and P pages, for each inode we have at most:
 271  * N*P indirect
 272  * min(N*P, blocksize/4 + 1) dindirect blocks
 273  * niocount tindirect
 274  *
 275  * For the entire filesystem, we have at most:
 276  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 277  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 278  * objcount inode blocks
 279  * 1 superblock
 280  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 281  *
 282  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 283  */
 284 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 285                                       int niocount, struct niobuf_local *nb)
 286 {
 287         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 288         __u64 next_indir;
 289         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 290         int nbitmaps = 0, ngdblocks;
 291         int needed = objcount + 1; /* inodes + superblock */
 292         int i, j;
 293
 294         for (i = 0, j = 0; i < objcount; i++, fso++) {
 295                 /* two or more dindirect blocks in case we cross boundary */
 296                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 297                                     nb[j].offset) >>
 298                                    sb->s_blocksize_bits) /
 299                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 300                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 301
 302                 /* leaf, indirect, tindirect blocks for first block */
 303                 nbitmaps += blockpp + 2;
 304
 305                 j += fso->fso_bufcnt;
 306         }
 307
 308         next_indir = nb[0].offset +
 309                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 310         for (i = 1; i < niocount; i++) {
 311                 if (nb[i].offset >= next_indir) {
 312                         nbitmaps++;     /* additional indirect */
 313                         next_indir = nb[i].offset +
 314                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 315                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 316                         nbitmaps++;     /* additional indirect */
 317                 }
 318                 nbitmaps += blockpp;    /* each leaf in different group? */
 319         }
 320
 321         ngdblocks = nbitmaps;
 322         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 323                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 324         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 325                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 326
 327         needed += nbitmaps + ngdblocks;
 328
 329         /* last_rcvd update */
 330         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 331
 332 #if defined(CONFIG_QUOTA)
 333         /* We assume that there will be 1 bit set in s_dquot.flags for each
 334          * quota file that is active.  This is at least true for now.
 335          */
 336         needed += hweight32(sb_any_quota_enabled(sb)) *
 337                 EXT3_SINGLEDATA_TRANS_BLOCKS;
 338 #endif
 339
 340         return needed;
 341 }
 342
 343 /* We have to start a huge journal transaction here to hold all of the
 344  * metadata for the pages being written here.  This is necessitated by
 345  * the fact that we do lots of prepare_write operations before we do
 346  * any of the matching commit_write operations, so even if we split
 347  * up to use "smaller" transactions none of them could complete until
 348  * all of them were opened.  By having a single journal transaction,
 349  * we eliminate duplicate reservations for common blocks like the
 350  * superblock and group descriptors or bitmaps.
 351  *
 352  * We will start the transaction here, but each prepare_write will
 353  * add a refcount to the transaction, and each commit_write will
 354  * remove a refcount.  The transaction will be closed when all of
 355  * the pages have been written.
 356  */
 357 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 358                                    int niocount, struct niobuf_local *nb,
 359                                    void *desc_private, int logs)
 360 {
 361         journal_t *journal;
 362         handle_t *handle;
 363         int needed;
 364         ENTRY;
 365
 366         LASSERT(current->journal_info == desc_private);
 367         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 368         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 369
 370         /* The number of blocks we could _possibly_ dirty can very large.
 371          * We reduce our request if it is absurd (and we couldn't get that
 372          * many credits for a single handle anyways).
 373          *
 374          * At some point we have to limit the size of I/Os sent at one time,
 375          * increase the size of the journal, or we have to calculate the
 376          * actual journal requirements more carefully by checking all of
 377          * the blocks instead of being maximally pessimistic.  It remains to
 378          * be seen if this is a real problem or not.
 379          */
 380         if (needed > journal->j_max_transaction_buffers) {
 381                 CERROR("want too many journal credits (%d) using %d instead\n",
 382                        needed, journal->j_max_transaction_buffers);
 383                 needed = journal->j_max_transaction_buffers;
 384         }
 385
 386         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 387         handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
 388         if (IS_ERR(handle)) {
 389                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 390                        PTR_ERR(handle));
 391         } else {
 392                 LASSERT(handle->h_buffer_credits >= needed);
 393                 LASSERT(current->journal_info == handle);
 394         }
 395
 396         RETURN(handle);
 397 }
 398
 399 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 400 {
 401        handle_t *handle = h;
 402
 403        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 404        if (nblocks == 0) {
 405                handle->h_buffer_credits = 0;
 406                CWARN("setting credits of handle %p to zero by request\n", h);
 407        }
 408
 409        if (handle->h_buffer_credits > nblocks)
 410                 return 0;
 411        if (journal_extend(handle, nblocks) == 0)
 412                 return 0;
 413
 414        ext3_mark_inode_dirty(handle, inode);
 415        return journal_restart(handle, nblocks);
 416 }
 417
 418 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 419 {
 420         int rc;
 421         handle_t *handle = h;
 422
 423         LASSERT(current->journal_info == handle);
 424         if (force_sync)
 425                 handle->h_sync = 1; /* recovery likes this */
 426
 427         rc = fsfilt_ext3_journal_stop(handle);
 428
 429         return rc;
 430 }
 431
 432 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 433                                     void **wait_handle)
 434 {
 435         unsigned long tid;
 436         transaction_t *transaction;
 437         handle_t *handle = h;
 438         journal_t *journal;
 439         int rc;
 440
 441         LASSERT(current->journal_info == handle);
 442
 443         transaction = handle->h_transaction;
 444         journal = transaction->t_journal;
 445         tid = transaction->t_tid;
 446         /* we don't want to be blocked */
 447         handle->h_sync = 0;
 448         rc = fsfilt_ext3_journal_stop(handle);
 449         if (rc) {
 450                 CERROR("error while stopping transaction: %d\n", rc);
 451                 return rc;
 452         }
 453         log_start_commit(journal, tid);
 454
 455         *wait_handle = (void *) tid;
 456         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 457         return 0;
 458 }
 459
 460 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 461 {
 462         journal_t *journal = EXT3_JOURNAL(inode);
 463         tid_t tid = (tid_t)(long)h;
 464
 465         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 466         if (unlikely(is_journal_aborted(journal)))
 467                 return -EIO;
 468
 469         log_wait_commit(EXT3_JOURNAL(inode), tid);
 470
 471         if (unlikely(is_journal_aborted(journal)))
 472                 return -EIO;
 473         return 0;
 474 }
 475
 476 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 477                                struct iattr *iattr, int do_trunc)
 478 {
 479         struct inode *inode = dentry->d_inode;
 480         int rc = 0;
 481
 482         lock_24kernel();
 483
 484         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 485          * We are already writing the inode to disk as part of this
 486          * transaction and want to avoid a lot of extra inode writeout
 487          * later on. b=9828 */
 488         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 489                 /* ATTR_SIZE would invoke truncate: clear it */
 490                 iattr->ia_valid &= ~ATTR_SIZE;
 491                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 492                 i_size_write(inode, iattr->ia_size);
 493
 494                 if (iattr->ia_valid & ATTR_UID)
 495                         inode->i_uid = iattr->ia_uid;
 496                 if (iattr->ia_valid & ATTR_GID)
 497                         inode->i_gid = iattr->ia_gid;
 498                 if (iattr->ia_valid & ATTR_ATIME)
 499                         inode->i_atime = iattr->ia_atime;
 500                 if (iattr->ia_valid & ATTR_MTIME)
 501                         inode->i_mtime = iattr->ia_mtime;
 502                 if (iattr->ia_valid & ATTR_CTIME)
 503                         inode->i_ctime = iattr->ia_ctime;
 504                 if (iattr->ia_valid & ATTR_MODE) {
 505                         inode->i_mode = iattr->ia_mode;
 506
 507                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 508                                 inode->i_mode &= ~S_ISGID;
 509                 }
 510
 511                 inode->i_sb->s_op->dirty_inode(inode);
 512
 513                 goto out;
 514         }
 515
 516         /* Don't allow setattr to change file type */
 517         if (iattr->ia_valid & ATTR_MODE)
 518                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 519                                  (iattr->ia_mode & ~S_IFMT);
 520
 521         /* We set these flags on the client, but have already checked perms
 522          * so don't confuse inode_change_ok. */
 523         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 524
 525         if (inode->i_op->setattr) {
 526                 rc = inode->i_op->setattr(dentry, iattr);
 527         } else {
 528                 rc = inode_change_ok(inode, iattr);
 529                 if (!rc)
 530                         rc = inode_setattr(inode, iattr);
 531         }
 532
 533  out:
 534         unlock_24kernel();
 535         RETURN(rc);
 536 }
 537
 538 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 539                                  unsigned int cmd, unsigned long arg)
 540 {
 541         int rc = 0;
 542         ENTRY;
 543
 544         /* FIXME: Can't do this because of nested transaction deadlock */
 545         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 546                 CERROR("can't set data journal flag on file\n");
 547                 RETURN(-EPERM);
 548         }
 549
 550         if (inode->i_fop->ioctl)
 551                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 552         else
 553                 RETURN(-ENOTTY);
 554
 555         RETURN(rc);
 556 }
 557
 558 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 559                               void *lmm, int lmm_size, const char *name)
 560 {
 561         int rc;
 562
 563         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 564
 565         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 566                                    name, lmm, lmm_size, 0);
 567
 568
 569         if (rc)
 570                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 571                        inode->i_ino, rc);
 572         return rc;
 573 }
 574
 575 /* Must be called with i_mutex held */
 576 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 577                               const char *name)
 578 {
 579         int rc;
 580
 581         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 582
 583         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 584                             name, lmm, lmm_size);
 585
 586         /* This gives us the MD size */
 587         if (lmm == NULL)
 588                 return (rc == -ENODATA) ? 0 : rc;
 589
 590         if (rc < 0) {
 591                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 592                        EXT3_XATTR_INDEX_TRUSTED, name,
 593                        inode->i_ino, rc);
 594                 memset(lmm, 0, lmm_size);
 595                 return (rc == -ENODATA) ? 0 : rc;
 596         }
 597
 598         return rc;
 599 }
 600
 601 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 602 {
 603         submit_bio(rw, bio);
 604         return 0;
 605 }
 606
 607 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 608                                     loff_t *off)
 609 {
 610         struct inode *inode = file->f_dentry->d_inode;
 611         int rc = 0;
 612
 613         if (S_ISREG(inode->i_mode))
 614                 rc = file->f_op->read(file, buf, count, off);
 615         else {
 616                 const int blkbits = inode->i_sb->s_blocksize_bits;
 617                 const int blksize = inode->i_sb->s_blocksize;
 618
 619                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 620                        count, inode->i_ino, *off);
 621                 while (count > 0) {
 622                         struct buffer_head *bh;
 623
 624                         bh = NULL;
 625                         if (*off < i_size_read(inode)) {
 626                                 int err = 0;
 627
 628                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 629                                                 0, &err);
 630
 631                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 632
 633                                 if (bh) {
 634                                         memcpy(buf, bh->b_data, blksize);
 635                                         brelse(bh);
 636                                 } else if (err) {
 637                                         /* XXX in theory we should just fake
 638                                          * this buffer and continue like ext3,
 639                                          * especially if this is a partial read
 640                                          */
 641                                         CERROR("error read dir %lu+%llu: %d\n",
 642                                                inode->i_ino, *off, err);
 643                                         RETURN(err);
 644                                 }
 645                         }
 646                         if (!bh) {
 647                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 648
 649                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 650                                 memset(fake, 0, sizeof(*fake));
 651                                 fake->rec_len = cpu_to_le16(blksize);
 652                         }
 653                         count -= blksize;
 654                         buf += blksize;
 655                         *off += blksize;
 656                         rc += blksize;
 657                 }
 658         }
 659
 660         return rc;
 661 }
 662
 663 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 664 {
 665         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 666
 667         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 668
 669         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 670 }
 671
 672 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 673                                       void *handle, fsfilt_cb_t cb_func,
 674                                       void *cb_data)
 675 {
 676         struct fsfilt_cb_data *fcb;
 677
 678         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 679         if (fcb == NULL)
 680                 RETURN(-ENOMEM);
 681
 682         fcb->cb_func = cb_func;
 683         fcb->cb_obd = obd;
 684         fcb->cb_last_rcvd = last_rcvd;
 685         fcb->cb_data = cb_data;
 686
 687         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 688         journal_callback_set(handle, fsfilt_ext3_cb_func,
 689                              (struct journal_callback *)fcb);
 690
 691         return 0;
 692 }
 693
 694 /*
 695  * We need to hack the return value for the free inode counts because
 696  * the current EA code requires one filesystem block per inode with EAs,
 697  * so it is possible to run out of blocks before we run out of inodes.
 698  *
 699  * This can be removed when the ext3 EA code is fixed.
 700  */
 701 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 702 {
 703         struct kstatfs sfs;
 704         int rc;
 705
 706         memset(&sfs, 0, sizeof(sfs));
 707
 708         rc = ll_do_statfs(sb, &sfs);
 709
 710         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 711                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 712                 sfs.f_ffree = sfs.f_bfree;
 713         }
 714
 715         statfs_pack(osfs, &sfs);
 716         return rc;
 717 }
 718
 719 static int fsfilt_ext3_sync(struct super_block *sb)
 720 {
 721         return ext3_force_commit(sb);
 722 }
 723
 724 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 725 #warning "kernel code has old extents/mballoc patch, disabling"
 726 #undef EXT3_MULTIBLOCK_ALLOCATOR
 727 #endif
 728 #ifndef EXT3_EXTENTS_FL
 729 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 730 #endif
 731
 732 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 733 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 734 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 735 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 736 #else
 737 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 738 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 739 #endif
 740
 741 #include <linux/lustre_version.h>
 742 #if EXT3_EXT_MAGIC == 0xf301
 743 #define ee_start e_start
 744 #define ee_block e_block
 745 #define ee_len   e_num
 746 #endif
 747 #ifndef EXT3_BB_MAX_BLOCKS
 748 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 749         ext3_new_blocks(handle, inode, count, goal, err)
 750 #endif
 751
 752 struct bpointers {
 753         unsigned long *blocks;
 754         int *created;
 755         unsigned long start;
 756         int num;
 757         int init_num;
 758         int create;
 759 };
 760
 761 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 762                               unsigned long block, int *aflags)
 763 {
 764         struct ext3_inode_info *ei = EXT3_I(inode);
 765         unsigned long bg_start;
 766         unsigned long colour;
 767         int depth;
 768
 769         if (path) {
 770                 struct ext3_extent *ex;
 771                 depth = path->p_depth;
 772
 773                 /* try to predict block placement */
 774                 if ((ex = path[depth].p_ext)) {
 775 #if 0
 776                         /* This prefers to eat into a contiguous extent
 777                          * rather than find an extent that the whole
 778                          * request will fit into.  This can fragment data
 779                          * block allocation and prevents our lovely 1M I/Os
 780                          * from reaching the disk intact. */
 781                         if (ex->ee_block + ex->ee_len == block)
 782                                 *aflags |= 1;
 783 #endif
 784                         return ex->ee_start + (block - ex->ee_block);
 785                 }
 786
 787                 /* it looks index is empty
 788                  * try to find starting from index itself */
 789                 if (path[depth].p_bh)
 790                         return path[depth].p_bh->b_blocknr;
 791         }
 792
 793         /* OK. use inode's group */
 794         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 795                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 796         colour = (current->pid % 16) *
 797                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 798         return bg_start + colour + block;
 799 }
 800
 801 #define ll_unmap_underlying_metadata(sb, blocknr) \
 802         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 803
 804 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 805 static unsigned long new_blocks(handle_t *handle, struct ext3_extents_tree *tree,
 806                                 struct ext3_ext_path *path, unsigned long block,
 807                                 int *count, int *err)
 808 {
 809         unsigned long pblock, goal;
 810         int aflags = 0;
 811
 812         goal = ext3_ext_find_goal(tree->inode, path, block, &aflags);
 813         aflags |= 2; /* block have been already reserved */
 814         pblock = ext3_mb_new_blocks(handle, tree->inode, goal, count, aflags, err);
 815         return pblock;
 816
 817 }
 818 #else
 819 static unsigned long new_blocks(handle_t *handle, struct ext3_extents_tree *tree,
 820                                 struct ext3_ext_path *path, unsigned long block,
 821                                 int *count, int *err)
 822 {
 823         struct ext3_allocation_request ar;
 824         unsigned long pblock;
 825         int aflags;
 826
 827         /* find neighbour allocated blocks */
 828         ar.lleft = block;
 829         *err = ext3_ext_search_left(tree, path, &ar.lleft, &ar.pleft);
 830         if (*err)
 831                 return 0;
 832         ar.lright = block;
 833         *err = ext3_ext_search_right(tree, path, &ar.lright, &ar.pright);
 834         if (*err)
 835                 return 0;
 836
 837         /* allocate new block */
 838         ar.goal = ext3_ext_find_goal(tree->inode, path, block, &aflags);
 839         ar.inode = tree->inode;
 840         ar.logical = block;
 841         ar.len = *count;
 842         ar.flags = EXT3_MB_HINT_DATA;
 843         pblock = ext3_mb_new_blocks(handle, &ar, err);
 844         *count = ar.len;
 845         return pblock;
 846
 847 }
 848 #endif
 849
 850 static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
 851                                   struct ext3_ext_path *path,
 852                                   struct ext3_ext_cache *cex)
 853 {
 854         struct inode *inode = tree->inode;
 855         struct bpointers *bp = tree->private;
 856         struct ext3_extent nex;
 857         unsigned long pblock;
 858         unsigned long tgen;
 859         int count, err, i;
 860         handle_t *handle;
 861
 862         i = EXT_DEPTH(tree);
 863         EXT_ASSERT(i == path->p_depth);
 864         EXT_ASSERT(path[i].p_hdr);
 865
 866         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 867                 err = EXT_CONTINUE;
 868                 goto map;
 869         }
 870
 871         if (bp->create == 0) {
 872                 i = 0;
 873                 if (cex->ec_block < bp->start)
 874                         i = bp->start - cex->ec_block;
 875                 if (i >= cex->ec_len)
 876                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 877                                         i, cex->ec_len);
 878                 for (; i < cex->ec_len && bp->num; i++) {
 879                         *(bp->created) = 0;
 880                         bp->created++;
 881                         *(bp->blocks) = 0;
 882                         bp->blocks++;
 883                         bp->num--;
 884                         bp->start++;
 885                 }
 886
 887                 return EXT_CONTINUE;
 888         }
 889
 890         tgen = EXT_GENERATION(tree);
 891         count = ext3_ext_calc_credits_for_insert(tree, path);
 892         ext3_up_truncate_sem(inode);
 893
 894         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 895         if (IS_ERR(handle)) {
 896                 ext3_down_truncate_sem(inode);
 897                 return PTR_ERR(handle);
 898         }
 899
 900         ext3_down_truncate_sem(inode);
 901         if (tgen != EXT_GENERATION(tree)) {
 902                 /* the tree has changed. so path can be invalid at moment */
 903                 fsfilt_ext3_journal_stop(handle);
 904                 return EXT_REPEAT;
 905         }
 906
 907         count = cex->ec_len;
 908         pblock = new_blocks(handle, tree, path, cex->ec_block, &count, &err);
 909         if (!pblock)
 910                 goto out;
 911         EXT_ASSERT(count <= cex->ec_len);
 912
 913         /* insert new extent */
 914         nex.ee_block = cex->ec_block;
 915         nex.ee_start = pblock;
 916         nex.ee_len = count;
 917         err = ext3_ext_insert_extent(handle, tree, path, &nex);
 918         if (err) {
 919                 CERROR("can't insert extent: %d\n", err);
 920                 /* XXX: export ext3_free_blocks() */
 921                 /*ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);*/
 922                 goto out;
 923         }
 924
 925         /*
 926          * Putting len of the actual extent we just inserted,
 927          * we are asking ext3_ext_walk_space() to continue
 928          * scaning after that block
 929          */
 930         cex->ec_len = nex.ee_len;
 931         cex->ec_start = nex.ee_start;
 932         BUG_ON(nex.ee_len == 0);
 933         BUG_ON(nex.ee_block != cex->ec_block);
 934
 935 out:
 936         fsfilt_ext3_journal_stop(handle);
 937 map:
 938         if (err >= 0) {
 939                 /* map blocks */
 940                 if (bp->num == 0) {
 941                         CERROR("hmm. why do we find this extent?\n");
 942                         CERROR("initial space: %lu:%u\n",
 943                                 bp->start, bp->init_num);
 944                         CERROR("current extent: %u/%u/%u %d\n",
 945                                 cex->ec_block, cex->ec_len,
 946                                 cex->ec_start, cex->ec_type);
 947                 }
 948                 i = 0;
 949                 if (cex->ec_block < bp->start)
 950                         i = bp->start - cex->ec_block;
 951                 if (i >= cex->ec_len)
 952                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 953                                         i, cex->ec_len);
 954                 for (; i < cex->ec_len && bp->num; i++) {
 955                         *(bp->blocks) = cex->ec_start + i;
 956                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 957                                 *(bp->created) = 0;
 958                         } else {
 959                                 *(bp->created) = 1;
 960                                 /* unmap any possible underlying metadata from
 961                                  * the block device mapping.  bug 6998. */
 962                                 ll_unmap_underlying_metadata(inode->i_sb,
 963                                                              *(bp->blocks));
 964                         }
 965                         bp->created++;
 966                         bp->blocks++;
 967                         bp->num--;
 968                         bp->start++;
 969                 }
 970         }
 971         return err;
 972 }
 973
 974 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
 975                        unsigned long num, unsigned long *blocks,
 976                        int *created, int create)
 977 {
 978         struct ext3_extents_tree tree;
 979         struct bpointers bp;
 980         int err;
 981
 982         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
 983                block, block + num - 1, (unsigned) inode->i_ino);
 984
 985         ext3_init_tree_desc(&tree, inode);
 986         tree.private = &bp;
 987         bp.blocks = blocks;
 988         bp.created = created;
 989         bp.start = block;
 990         bp.init_num = bp.num = num;
 991         bp.create = create;
 992
 993         ext3_down_truncate_sem(inode);
 994         err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
 995         ext3_ext_invalidate_cache(&tree);
 996         ext3_up_truncate_sem(inode);
 997
 998         return err;
 999 }
1000
1001 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1002                                     int pages, unsigned long *blocks,
1003                                     int *created, int create)
1004 {
1005         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1006         int rc = 0, i = 0;
1007         struct page *fp = NULL;
1008         int clen = 0;
1009
1010         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1011                 inode->i_ino, pages, (*page)->index);
1012
1013         /* pages are sorted already. so, we just have to find
1014          * contig. space and process them properly */
1015         while (i < pages) {
1016                 if (fp == NULL) {
1017                         /* start new extent */
1018                         fp = *page++;
1019                         clen = 1;
1020                         i++;
1021                         continue;
1022                 } else if (fp->index + clen == (*page)->index) {
1023                         /* continue the extent */
1024                         page++;
1025                         clen++;
1026                         i++;
1027                         continue;
1028                 }
1029
1030                 /* process found extent */
1031                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1032                                         clen * blocks_per_page, blocks,
1033                                         created, create);
1034                 if (rc)
1035                         GOTO(cleanup, rc);
1036
1037                 /* look for next extent */
1038                 fp = NULL;
1039                 blocks += blocks_per_page * clen;
1040                 created += blocks_per_page * clen;
1041         }
1042
1043         if (fp)
1044                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1045                                         clen * blocks_per_page, blocks,
1046                                         created, create);
1047 cleanup:
1048         return rc;
1049 }
1050 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1051
1052 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1053                                unsigned long *blocks, int *created, int create);
1054 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1055                                    int pages, unsigned long *blocks,
1056                                    int *created, int create)
1057 {
1058         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1059         unsigned long *b;
1060         int rc = 0, i, *cr;
1061
1062         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1063                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1064                 if (rc) {
1065                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1066                                inode->i_ino, *b, *cr, create, rc);
1067                         break;
1068                 }
1069
1070                 b += blocks_per_page;
1071                 cr += blocks_per_page;
1072         }
1073         return rc;
1074 }
1075
1076 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1077                                 int pages, unsigned long *blocks,
1078                                 int *created, int create,
1079                                 struct semaphore *optional_sem)
1080 {
1081         int rc;
1082 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1083         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1084                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1085                                                      blocks, created, create);
1086                 return rc;
1087         }
1088 #endif
1089         if (optional_sem != NULL)
1090                 down(optional_sem);
1091         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1092                                             created, create);
1093         if (optional_sem != NULL)
1094                 up(optional_sem);
1095
1096         return rc;
1097 }
1098
1099 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1100 {
1101         unsigned long block;
1102         struct buffer_head *bh;
1103         int err, blocksize, csize, boffs, osize = size;
1104
1105         /* prevent reading after eof */
1106         lock_kernel();
1107         if (i_size_read(inode) < *offs + size) {
1108                 size = i_size_read(inode) - *offs;
1109                 unlock_kernel();
1110                 if (size < 0) {
1111                         CERROR("size %llu is too short for read %u@%llu\n",
1112                                i_size_read(inode), size, *offs);
1113                         return -EIO;
1114                 } else if (size == 0) {
1115                         return 0;
1116                 }
1117         } else {
1118                 unlock_kernel();
1119         }
1120
1121         blocksize = 1 << inode->i_blkbits;
1122
1123         while (size > 0) {
1124                 block = *offs >> inode->i_blkbits;
1125                 boffs = *offs & (blocksize - 1);
1126                 csize = min(blocksize - boffs, size);
1127                 bh = ext3_bread(NULL, inode, block, 0, &err);
1128                 if (!bh) {
1129                         CERROR("can't read block: %d\n", err);
1130                         return err;
1131                 }
1132
1133                 memcpy(buf, bh->b_data + boffs, csize);
1134                 brelse(bh);
1135
1136                 *offs += csize;
1137                 buf += csize;
1138                 size -= csize;
1139         }
1140         return osize;
1141 }
1142 EXPORT_SYMBOL(fsfilt_ext3_read);
1143
1144 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1145                                    int size, loff_t *offs)
1146 {
1147         int rc;
1148         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1149         if (rc > 0)
1150                 rc = 0;
1151         return rc;
1152 }
1153
1154 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1155                                 loff_t *offs, handle_t *handle)
1156 {
1157         struct buffer_head *bh = NULL;
1158         loff_t old_size = i_size_read(inode), offset = *offs;
1159         loff_t new_size = i_size_read(inode);
1160         unsigned long block;
1161         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1162
1163         while (bufsize > 0) {
1164                 if (bh != NULL)
1165                         brelse(bh);
1166
1167                 block = offset >> inode->i_blkbits;
1168                 boffs = offset & (blocksize - 1);
1169                 size = min(blocksize - boffs, bufsize);
1170                 bh = ext3_bread(handle, inode, block, 1, &err);
1171                 if (!bh) {
1172                         CERROR("can't read/create block: %d\n", err);
1173                         break;
1174                 }
1175
1176                 err = ext3_journal_get_write_access(handle, bh);
1177                 if (err) {
1178                         CERROR("journal_get_write_access() returned error %d\n",
1179                                err);
1180                         break;
1181                 }
1182                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1183                 memcpy(bh->b_data + boffs, buf, size);
1184                 err = ext3_journal_dirty_metadata(handle, bh);
1185                 if (err) {
1186                         CERROR("journal_dirty_metadata() returned error %d\n",
1187                                err);
1188                         break;
1189                 }
1190                 if (offset + size > new_size)
1191                         new_size = offset + size;
1192                 offset += size;
1193                 bufsize -= size;
1194                 buf += size;
1195         }
1196         if (bh)
1197                 brelse(bh);
1198
1199         /* correct in-core and on-disk sizes */
1200         if (new_size > i_size_read(inode)) {
1201                 lock_kernel();
1202                 if (new_size > i_size_read(inode))
1203                         i_size_write(inode, new_size);
1204                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1205                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1206                 if (i_size_read(inode) > old_size)
1207                         mark_inode_dirty(inode);
1208                 unlock_kernel();
1209         }
1210
1211         if (err == 0)
1212                 *offs = offset;
1213         return err;
1214 }
1215 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1216
1217 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1218                                     loff_t *offs, int force_sync)
1219 {
1220         struct inode *inode = file->f_dentry->d_inode;
1221         handle_t *handle;
1222         int err, block_count = 0, blocksize;
1223
1224         /* Determine how many transaction credits are needed */
1225         blocksize = 1 << inode->i_blkbits;
1226         block_count = (*offs & (blocksize - 1)) + bufsize;
1227         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1228
1229         handle = fsfilt_ext3_journal_start(inode,
1230                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1231         if (IS_ERR(handle)) {
1232                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1233                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1234                 return PTR_ERR(handle);
1235         }
1236
1237         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1238
1239         if (!err && force_sync)
1240                 handle->h_sync = 1; /* recovery likes this */
1241
1242         fsfilt_ext3_journal_stop(handle);
1243
1244         return err;
1245 }
1246
1247 static int fsfilt_ext3_setup(struct super_block *sb)
1248 {
1249 #if 0
1250         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1251         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1252 #endif
1253 #ifdef S_PDIROPS
1254         CWARN("Enabling PDIROPS\n");
1255         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1256         sb->s_flags |= S_PDIROPS;
1257 #endif
1258         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1259                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1260 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13)) && HAVE_QUOTA_SUPPORT
1261         set_opt(EXT3_SB(sb)->s_mount_opt, QUOTA);
1262 #endif
1263         return 0;
1264 }
1265
1266 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1267    objects. Logs is number of logfiles to update */
1268 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1269 {
1270         if ( !fso ) {
1271                 switch(op) {
1272                 case FSFILT_OP_CREATE:
1273                                  /* directory leaf, index & indirect & EA*/
1274                         return 4 + 3 * logs;
1275                 case FSFILT_OP_UNLINK:
1276                         return 3 * logs;
1277                 }
1278         } else {
1279                 int i;
1280                 int needed = 0;
1281                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1282                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1283                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1284                 for (i = 0; i < op; i++, fso++) {
1285                         int nblocks = fso->fso_bufcnt * blockpp;
1286                         int ndindirect = min(nblocks, addrpp + 1);
1287                         int nindir = nblocks + ndindirect + 1;
1288
1289                         needed += nindir;
1290                 }
1291                 return needed + 3 * logs;
1292         }
1293
1294         return 0;
1295 }
1296
1297 static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
1298
1299 #define DQINFO_COPY(out, in)                    \
1300 do {                                            \
1301         Q_COPY(out, in, dqi_bgrace);            \
1302         Q_COPY(out, in, dqi_igrace);            \
1303         Q_COPY(out, in, dqi_flags);             \
1304         Q_COPY(out, in, dqi_valid);             \
1305 } while (0)
1306
1307 #define DQBLK_COPY(out, in)                     \
1308 do {                                            \
1309         Q_COPY(out, in, dqb_bhardlimit);        \
1310         Q_COPY(out, in, dqb_bsoftlimit);        \
1311         Q_COPY(out, in, dqb_curspace);          \
1312         Q_COPY(out, in, dqb_ihardlimit);        \
1313         Q_COPY(out, in, dqb_isoftlimit);        \
1314         Q_COPY(out, in, dqb_curinodes);         \
1315         Q_COPY(out, in, dqb_btime);             \
1316         Q_COPY(out, in, dqb_itime);             \
1317         Q_COPY(out, in, dqb_valid);             \
1318 } while (0)
1319
1320
1321
1322 static int fsfilt_ext3_quotactl(struct super_block *sb,
1323                                 struct obd_quotactl *oqc)
1324 {
1325         int i, rc = 0, error = 0;
1326         struct quotactl_ops *qcop;
1327         struct if_dqinfo *info;
1328         struct if_dqblk *dqblk;
1329         ENTRY;
1330
1331         if (!sb->s_qcop)
1332                 RETURN(-ENOSYS);
1333
1334         OBD_ALLOC_PTR(info);
1335         if (!info)
1336                 RETURN(-ENOMEM);
1337         OBD_ALLOC_PTR(dqblk);
1338         if (!dqblk) {
1339                 OBD_FREE_PTR(info);
1340                 RETURN(-ENOMEM);
1341         }
1342
1343         DQINFO_COPY(info, &oqc->qc_dqinfo);
1344         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1345
1346         qcop = sb->s_qcop;
1347         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1348                 for (i = 0; i < MAXQUOTAS; i++) {
1349                         if (!Q_TYPESET(oqc, i))
1350                                 continue;
1351
1352                         if (oqc->qc_cmd == Q_QUOTAON) {
1353                                 if (!qcop->quota_on)
1354                                         GOTO(out, rc = -ENOSYS);
1355                                 rc = qcop->quota_on(sb, i, oqc->qc_id,
1356                                                     (char *)op_quotafile[i]);
1357                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1358                                 if (!qcop->quota_off)
1359                                         GOTO(out, rc = -ENOSYS);
1360                                 rc = qcop->quota_off(sb, i);
1361                         }
1362
1363                         if (rc == -EBUSY)
1364                                 error = rc;
1365                         else if (rc)
1366                                 GOTO(out, rc);
1367                 }
1368                 GOTO(out, rc ?: error);
1369         }
1370
1371         switch (oqc->qc_cmd) {
1372         case Q_GETOINFO:
1373         case Q_GETINFO:
1374                 if (!qcop->get_info)
1375                         GOTO(out, rc = -ENOSYS);
1376                 rc = qcop->get_info(sb, oqc->qc_type, info);
1377                 break;
1378         case Q_SETQUOTA:
1379         case Q_INITQUOTA:
1380                 if (!qcop->set_dqblk)
1381                         GOTO(out, rc = -ENOSYS);
1382                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1383                 break;
1384         case Q_GETOQUOTA:
1385         case Q_GETQUOTA:
1386                 if (!qcop->get_dqblk)
1387                         GOTO(out, rc = -ENOSYS);
1388                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1389                 break;
1390         case Q_SYNC:
1391                 if (!sb->s_qcop->quota_sync)
1392                         GOTO(out, rc = -ENOSYS);
1393                 qcop->quota_sync(sb, oqc->qc_type);
1394                 break;
1395         default:
1396                 CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
1397                 LBUG();
1398         }
1399 out:
1400         DQINFO_COPY(&oqc->qc_dqinfo, info);
1401         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1402
1403         OBD_FREE_PTR(info);
1404         OBD_FREE_PTR(dqblk);
1405
1406         if (rc)
1407                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1408                                 "failed: %d\n",
1409                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1410         RETURN(rc);
1411 }
1412
1413 struct chk_dqblk{
1414         struct hlist_node       dqb_hash;        /* quotacheck hash */
1415         struct list_head        dqb_list;        /* in list also */
1416         qid_t                   dqb_id;          /* uid/gid */
1417         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1418         __u32                   dqb_bhardlimit;  /* block hard limit */
1419         __u32                   dqb_bsoftlimit;  /* block soft limit */
1420         qsize_t                 dqb_curspace;    /* current space */
1421         __u32                   dqb_ihardlimit;  /* inode hard limit */
1422         __u32                   dqb_isoftlimit;  /* inode soft limit */
1423         __u32                   dqb_curinodes;   /* current inodes */
1424         __u64                   dqb_btime;       /* block grace time */
1425         __u64                   dqb_itime;       /* inode grace time */
1426         __u32                   dqb_valid;       /* flag for above fields */
1427 };
1428
1429 static inline unsigned int chkquot_hash(qid_t id, int type)
1430                                         __attribute__((__const__));
1431
1432 static inline unsigned int chkquot_hash(qid_t id, int type)
1433 {
1434         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1435 }
1436
1437 static inline struct chk_dqblk *
1438 find_chkquot(struct hlist_head *head, qid_t id, int type)
1439 {
1440         struct hlist_node *node;
1441         struct chk_dqblk *cdqb;
1442
1443         hlist_for_each(node, head) {
1444                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1445                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1446                         return cdqb;
1447         }
1448
1449         return NULL;
1450 }
1451
1452 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1453 {
1454         struct chk_dqblk *cdqb;
1455
1456         OBD_ALLOC_PTR(cdqb);
1457         if (cdqb) {
1458                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1459                 INIT_LIST_HEAD(&cdqb->dqb_list);
1460                 cdqb->dqb_id = id;
1461                 cdqb->dqb_type = type;
1462         }
1463
1464         return cdqb;
1465 }
1466
1467 static struct chk_dqblk *
1468 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1469       qid_t id, int type, int first_check)
1470 {
1471         struct hlist_head *head = hash + chkquot_hash(id, type);
1472         struct if_dqblk dqb;
1473         struct chk_dqblk *cdqb;
1474         int rc;
1475
1476         cdqb = find_chkquot(head, id, type);
1477         if (cdqb)
1478                 return cdqb;
1479
1480         cdqb = alloc_chkquot(id, type);
1481         if (!cdqb)
1482                 return NULL;
1483
1484         if (!first_check) {
1485                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1486                 if (rc) {
1487                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1488                                id, type, rc);
1489                 } else {
1490                         DQBLK_COPY(cdqb, &dqb);
1491                         cdqb->dqb_curspace = 0;
1492                         cdqb->dqb_curinodes = 0;
1493                 }
1494         }
1495
1496         hlist_add_head(&cdqb->dqb_hash, head);
1497         list_add_tail(&cdqb->dqb_list, list);
1498
1499         return cdqb;
1500 }
1501
1502 static inline int quota_onoff(struct super_block *sb, int cmd, int type)
1503 {
1504         struct obd_quotactl *oqctl;
1505         int rc;
1506
1507         OBD_ALLOC_PTR(oqctl);
1508         if (!oqctl)
1509                 RETURN(-ENOMEM);
1510
1511         oqctl->qc_cmd = cmd;
1512         oqctl->qc_id = QFMT_LDISKFS;
1513         oqctl->qc_type = type;
1514         rc = fsfilt_ext3_quotactl(sb, oqctl);
1515
1516         OBD_FREE_PTR(oqctl);
1517         return rc;
1518 }
1519
1520 static inline int read_old_dqinfo(struct super_block *sb, int type,
1521                                   struct if_dqinfo *dqinfo)
1522 {
1523         struct obd_quotactl *oqctl;
1524         int rc;
1525         ENTRY;
1526
1527         OBD_ALLOC_PTR(oqctl);
1528         if (!oqctl)
1529                 RETURN(-ENOMEM);
1530
1531         oqctl->qc_cmd = Q_GETINFO;
1532         oqctl->qc_type = type;
1533         rc = fsfilt_ext3_quotactl(sb, oqctl);
1534         if (!rc)
1535                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1536
1537         OBD_FREE_PTR(oqctl);
1538         RETURN(rc);
1539 }
1540
1541 static inline struct ext3_group_desc *
1542 get_group_desc(struct super_block *sb, int group)
1543 {
1544         unsigned long desc_block, desc;
1545         struct ext3_group_desc *gdp;
1546
1547         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1548         desc = group % EXT3_DESC_PER_BLOCK(sb);
1549         gdp = (struct ext3_group_desc *)
1550               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1551
1552         return gdp + desc;
1553 }
1554
1555 static inline struct buffer_head *
1556 read_inode_bitmap(struct super_block *sb, unsigned long group)
1557 {
1558         struct ext3_group_desc *desc;
1559         struct buffer_head *bh;
1560
1561         desc = get_group_desc(sb, group);
1562         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1563
1564         return bh;
1565 }
1566
1567 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1568                                      struct buffer_head *bitmap_bh,
1569                                      int index, unsigned long ino)
1570 {
1571         struct inode *inode = NULL;
1572
1573         if (ext3_test_bit(index, bitmap_bh->b_data))
1574                 inode = iget(sb, ino);
1575
1576         return inode;
1577 }
1578
1579 struct qchk_ctxt {
1580         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1581         struct list_head        qckt_list;                   /* quotacheck list */
1582         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1583         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1584 };
1585
1586 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1587                            struct obd_quotactl *oqc)
1588 {
1589         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1590         loff_t size = 0;
1591         qid_t qid[MAXQUOTAS];
1592         int cnt, i, rc = 0;
1593
1594         if (!inode)
1595                 return 0;
1596
1597         qid[USRQUOTA] = inode->i_uid;
1598         qid[GRPQUOTA] = inode->i_gid;
1599
1600         if (S_ISDIR(inode->i_mode) ||
1601             S_ISREG(inode->i_mode) ||
1602             S_ISLNK(inode->i_mode))
1603                 size = inode_get_bytes(inode);
1604
1605         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1606                 if (!Q_TYPESET(oqc, cnt))
1607                         continue;
1608
1609                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1610                                 &qctxt->qckt_list, qid[cnt], cnt,
1611                                 qctxt->qckt_first_check[cnt]);
1612                 if (!cdqb[cnt]) {
1613                         rc = -ENOMEM;
1614                         break;
1615                 }
1616
1617                 cdqb[cnt]->dqb_curspace += size;
1618                 cdqb[cnt]->dqb_curinodes++;
1619         }
1620
1621         if (rc) {
1622                 for (i = 0; i < cnt; i++) {
1623                         if (!Q_TYPESET(oqc, i))
1624                                 continue;
1625                         LASSERT(cdqb[i]);
1626                         cdqb[i]->dqb_curspace -= size;
1627                         cdqb[i]->dqb_curinodes--;
1628                 }
1629         }
1630
1631         return rc;
1632 }
1633
1634 static int v2_write_dqheader(struct file *f, int type)
1635 {
1636         static const __u32 quota_magics[] = V2_INITQMAGICS;
1637         static const __u32 quota_versions[] = V2_INITQVERSIONS;
1638         struct v2_disk_dqheader dqhead;
1639         loff_t offset = 0;
1640
1641         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1642         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1643
1644         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1645         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1646
1647         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1648 }
1649
1650 /* write dqinfo struct in a new quota file */
1651 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1652 {
1653         struct v2_disk_dqinfo dqinfo;
1654         __u32 blocks = V2_DQTREEOFF + 1;
1655         loff_t offset = V2_DQINFOOFF;
1656
1657         if (info) {
1658                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1659                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1660                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1661                                                ~DQF_INFO_DIRTY);
1662         } else {
1663                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1664                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1665                 dqinfo.dqi_flags = 0;
1666         }
1667
1668         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1669         dqinfo.dqi_free_blk = 0;
1670         dqinfo.dqi_free_entry = 0;
1671
1672         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1673 }
1674
1675 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1676                                   struct obd_quotactl *oqc)
1677 {
1678         int i, rc = 0;
1679         ENTRY;
1680
1681         for (i = 0; i < MAXQUOTAS; i++) {
1682                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1683                                          NULL : &qctxt->qckt_dqinfo[i];
1684                 struct file *file;
1685
1686                 if (!Q_TYPESET(oqc, i))
1687                         continue;
1688
1689                 file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
1690                                  0644);
1691                 if (IS_ERR(file)) {
1692                         rc = PTR_ERR(file);
1693                         CERROR("can't create %s file: rc = %d\n",
1694                                op_quotafile[i], rc);
1695                         GOTO(out, rc);
1696                 }
1697
1698                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1699                         CERROR("file %s is not regular", op_quotafile[i]);
1700                         filp_close(file, 0);
1701                         GOTO(out, rc = -EINVAL);
1702                 }
1703
1704                 rc = v2_write_dqheader(file, i);
1705                 if (rc) {
1706                         filp_close(file, 0);
1707                         GOTO(out, rc);
1708                 }
1709
1710                 rc = v2_write_dqinfo(file, i, info);
1711                 filp_close(file, 0);
1712                 if (rc)
1713                         GOTO(out, rc);
1714         }
1715
1716 out:
1717         RETURN(rc);
1718 }
1719
1720
1721 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1722                           struct chk_dqblk *cdqb)
1723 {
1724         struct obd_quotactl *oqc;
1725         long now;
1726         int rc;
1727         ENTRY;
1728
1729         OBD_ALLOC_PTR(oqc);
1730         if (!oqc)
1731                 RETURN(-ENOMEM);
1732
1733         now = CURRENT_SECONDS;
1734
1735         if (cdqb->dqb_bsoftlimit &&
1736             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1737             !cdqb->dqb_btime)
1738                 cdqb->dqb_btime =
1739                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1740
1741         if (cdqb->dqb_isoftlimit &&
1742             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1743             !cdqb->dqb_itime)
1744                 cdqb->dqb_itime =
1745                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1746
1747         cdqb->dqb_valid = QIF_ALL;
1748
1749         oqc->qc_cmd = Q_SETQUOTA;
1750         oqc->qc_type = cdqb->dqb_type;
1751         oqc->qc_id = cdqb->dqb_id;
1752         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1753
1754         rc = fsfilt_ext3_quotactl(sb, oqc);
1755         OBD_FREE_PTR(oqc);
1756         RETURN(rc);
1757 }
1758
1759 static int prune_chkquots(struct super_block *sb,
1760                           struct qchk_ctxt *qctxt, int error)
1761 {
1762         struct chk_dqblk *cdqb, *tmp;
1763         int rc;
1764
1765         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1766                 if (!error) {
1767                         rc = commit_chkquot(sb, qctxt, cdqb);
1768                         if (rc)
1769                                 error = rc;
1770                 }
1771                 hlist_del_init(&cdqb->dqb_hash);
1772                 list_del(&cdqb->dqb_list);
1773                 OBD_FREE_PTR(cdqb);
1774         }
1775
1776         return error;
1777 }
1778
1779 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1780                                   struct obd_quotactl *oqc)
1781 {
1782         struct ext3_sb_info *sbi = EXT3_SB(sb);
1783         int i, group;
1784         struct qchk_ctxt *qctxt;
1785         struct buffer_head *bitmap_bh = NULL;
1786         unsigned long ino;
1787         struct inode *inode;
1788         int rc = 0;
1789         ENTRY;
1790
1791         /* turn on quota and read dqinfo if existed */
1792         OBD_ALLOC_PTR(qctxt);
1793         if (!qctxt) {
1794                 oqc->qc_stat = -ENOMEM;
1795                 RETURN(-ENOMEM);
1796         }
1797
1798         for (i = 0; i < NR_DQHASH; i++)
1799                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1800         INIT_LIST_HEAD(&qctxt->qckt_list);
1801
1802         for (i = 0; i < MAXQUOTAS; i++) {
1803                 if (!Q_TYPESET(oqc, i))
1804                         continue;
1805
1806                 rc = quota_onoff(sb, Q_QUOTAON, i);
1807                 if (!rc || rc == -EBUSY) {
1808                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1809                         if (rc)
1810                                 GOTO(out, rc);
1811                 } else if (rc == -ENOENT) {
1812                         qctxt->qckt_first_check[i] = 1;
1813                 } else if (rc) {
1814                         GOTO(out, rc);
1815                 }
1816         }
1817
1818         /* check quota and update in hash */
1819         for (group = 0; group < sbi->s_groups_count; group++) {
1820                 ino = group * sbi->s_inodes_per_group + 1;
1821                 bitmap_bh = read_inode_bitmap(sb, group);
1822                 if (!bitmap_bh) {
1823                         CERROR("read_inode_bitmap group %d failed", group);
1824                         GOTO(out, rc = -EIO);
1825                 }
1826
1827                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1828                         if (ino < sbi->s_first_ino)
1829                                 continue;
1830
1831                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1832                         rc = add_inode_quota(inode, qctxt, oqc);
1833                         iput(inode);
1834                         if (rc) {
1835                                 brelse(bitmap_bh);
1836                                 GOTO(out, rc);
1837                         }
1838                 }
1839
1840                 brelse(bitmap_bh);
1841         }
1842
1843         /* read old quota limits from old quota file. (only for the user
1844          * has limits but hasn't file) */
1845 #ifdef HAVE_QUOTA_SUPPORT
1846         for (i = 0; i < MAXQUOTAS; i++) {
1847                 struct list_head id_list;
1848                 struct dquot_id *dqid, *tmp;
1849
1850                 if (!Q_TYPESET(oqc, i))
1851                         continue;
1852
1853                 if (qctxt->qckt_first_check[i])
1854                         continue;
1855
1856
1857                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1858                 INIT_LIST_HEAD(&id_list);
1859 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1860                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1861 #else
1862                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1863 #endif
1864                 if (rc)
1865                         CERROR("read old limits failed. (rc:%d)\n", rc);
1866
1867                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1868                         list_del_init(&dqid->di_link);
1869
1870                         if (!rc)
1871                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1872                                       dqid->di_id, i,
1873                                       qctxt->qckt_first_check[i]);
1874                         kfree(dqid);
1875                 }
1876         }
1877 #endif
1878         /* turn off quota cause we are to dump chk_dqblk to files */
1879         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1880
1881         rc = create_new_quota_files(qctxt, oqc);
1882         if (rc)
1883                 GOTO(out, rc);
1884
1885         /* we use vfs functions to set dqblk, so turn quota on */
1886         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
1887 out:
1888         /* dump and free chk_dqblk */
1889         rc = prune_chkquots(sb, qctxt, rc);
1890         OBD_FREE_PTR(qctxt);
1891
1892         /* turn off quota, `lfs quotacheck` will turn on when all
1893          * nodes quotacheck finish. */
1894         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1895
1896         oqc->qc_stat = rc;
1897         if (rc)
1898                 CERROR("quotacheck failed: rc = %d\n", rc);
1899
1900         RETURN(rc);
1901 }
1902
1903 #ifdef HAVE_QUOTA_SUPPORT
1904 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
1905                                  int cmd)
1906 {
1907         int rc = 0;
1908         ENTRY;
1909
1910         if (lqi->qi_files[type] == NULL) {
1911                 CERROR("operate qinfo before it's enabled!\n");
1912                 RETURN(-EIO);
1913         }
1914
1915         switch (cmd) {
1916         case QFILE_CHK:
1917                 rc = lustre_check_quota_file(lqi, type);
1918                 break;
1919         case QFILE_RD_INFO:
1920                 rc = lustre_read_quota_info(lqi, type);
1921                 break;
1922         case QFILE_WR_INFO:
1923                 rc = lustre_write_quota_info(lqi, type);
1924                 break;
1925         case QFILE_INIT_INFO:
1926                 rc = lustre_init_quota_info(lqi, type);
1927                 break;
1928         default:
1929                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1930                 LBUG();
1931                 break;
1932         }
1933         RETURN(rc);
1934 }
1935
1936 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
1937                             struct list_head *list)
1938 {
1939         return lustre_get_qids(file, inode, type, list);
1940 }
1941
1942 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
1943 {
1944         int rc = 0;
1945         ENTRY;
1946
1947         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
1948                 CERROR("operate dquot before it's enabled!\n");
1949                 RETURN(-EIO);
1950         }
1951
1952         switch (cmd) {
1953         case QFILE_RD_DQUOT:
1954                 rc = lustre_read_dquot(dquot);
1955                 break;
1956         case QFILE_WR_DQUOT:
1957                 if (dquot->dq_dqb.dqb_ihardlimit ||
1958                     dquot->dq_dqb.dqb_isoftlimit ||
1959                     dquot->dq_dqb.dqb_bhardlimit ||
1960                     dquot->dq_dqb.dqb_bsoftlimit)
1961                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1962                 else
1963                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
1964
1965                 rc = lustre_commit_dquot(dquot);
1966                 if (rc >= 0)
1967                         rc = 0;
1968                 break;
1969         default:
1970                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1971                 LBUG();
1972                 break;
1973         }
1974         RETURN(rc);
1975 }
1976 #endif
1977
1978 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
1979 {
1980         return (EXT3_SB(sb)->journal_bdev);
1981 }
1982 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
1983
1984 static struct fsfilt_operations fsfilt_ext3_ops = {
1985         .fs_type                = "ext3",
1986         .fs_owner               = THIS_MODULE,
1987         .fs_getlabel            = fsfilt_ext3_get_label,
1988         .fs_setlabel            = fsfilt_ext3_set_label,
1989         .fs_uuid                = fsfilt_ext3_uuid,
1990         .fs_start               = fsfilt_ext3_start,
1991         .fs_brw_start           = fsfilt_ext3_brw_start,
1992         .fs_extend              = fsfilt_ext3_extend,
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_write_record        = fsfilt_ext3_write_record,
2006         .fs_read_record         = fsfilt_ext3_read_record,
2007         .fs_setup               = fsfilt_ext3_setup,
2008         .fs_send_bio            = fsfilt_ext3_send_bio,
2009         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2010         .fs_quotactl            = fsfilt_ext3_quotactl,
2011         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2012 #ifdef HAVE_DISK_INODE_VERSION
2013         .fs_get_version         = fsfilt_ext3_get_version,
2014         .fs_set_version         = fsfilt_ext3_set_version,
2015 #endif
2016 #ifdef HAVE_QUOTA_SUPPORT
2017         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2018         .fs_qids                = fsfilt_ext3_qids,
2019         .fs_dquot               = fsfilt_ext3_dquot,
2020 #endif
2021         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2022 };
2023
2024 static int __init fsfilt_ext3_init(void)
2025 {
2026         int rc;
2027
2028         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2029                                       sizeof(struct fsfilt_cb_data), 0, 0);
2030         if (!fcb_cache) {
2031                 CERROR("error allocating fsfilt journal callback cache\n");
2032                 GOTO(out, rc = -ENOMEM);
2033         }
2034
2035         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2036
2037         if (rc) {
2038                 int err = cfs_mem_cache_destroy(fcb_cache);
2039                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2040         }
2041 out:
2042         return rc;
2043 }
2044
2045 static void __exit fsfilt_ext3_exit(void)
2046 {
2047         int rc;
2048
2049         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2050         rc = cfs_mem_cache_destroy(fcb_cache);
2051         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2052 }
2053
2054 module_init(fsfilt_ext3_init);
2055 module_exit(fsfilt_ext3_exit);
2056
2057 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2058 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2059 MODULE_LICENSE("GPL");