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         lock_24kernel();
  95         handle = journal_start(journal, 1);
  96         unlock_24kernel();
  97         if (IS_ERR(handle)) {
  98                 CERROR("can't start transaction\n");
  99                 return(PTR_ERR(handle));
 100         }
 101
 102         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 103         if (err)
 104                 goto out;
 105
 106         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 107                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 108
 109         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 110
 111 out:
 112         lock_24kernel();
 113         journal_stop(handle);
 114         unlock_24kernel();
 115
 116         return(err);
 117 }
 118
 119 static char *fsfilt_ext3_uuid(struct super_block *sb)
 120 {
 121         return EXT3_SB(sb)->s_es->s_uuid;
 122 }
 123
 124 #ifdef HAVE_DISK_INODE_VERSION
 125 /*
 126  * Get the 64-bit version for an inode.
 127  */
 128 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 129 {
 130         return EXT3_I(inode)->i_fs_version;
 131 }
 132
 133 /*
 134  * Set the 64-bit version and return the old version.
 135  */
 136 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 137 {
 138         __u64 old_version = EXT3_I(inode)->i_fs_version;
 139
 140         (EXT3_I(inode))->i_fs_version = new_version;
 141         return old_version;
 142 }
 143
 144 #endif
 145
 146 /*
 147  * We don't currently need any additional blocks for rmdir and
 148  * unlink transactions because we are storing the OST oa_id inside
 149  * the inode (which we will be changing anyways as part of this
 150  * transaction).
 151  */
 152 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 153                                int logs)
 154 {
 155         /* For updates to the last received file */
 156         int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS;
 157         journal_t *journal;
 158         void *handle;
 159
 160         if (current->journal_info) {
 161                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 162                        current->journal_info);
 163                 goto journal_start;
 164         }
 165
 166         switch(op) {
 167         case FSFILT_OP_RMDIR:
 168         case FSFILT_OP_UNLINK:
 169                 /* delete one file + create/update logs for each stripe */
 170                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 171                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 172                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 173                 break;
 174         case FSFILT_OP_RENAME:
 175                 /* modify additional directory */
 176                 nblocks += EXT3_SINGLEDATA_TRANS_BLOCKS;
 177                 /* no break */
 178         case FSFILT_OP_SYMLINK:
 179                 /* additional block + block bitmap + GDT for long symlink */
 180                 nblocks += 3;
 181                 /* no break */
 182         case FSFILT_OP_CREATE: {
 183 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 184                 static int warned;
 185                 if (!warned) {
 186                         if (!test_opt(inode->i_sb, EXTENTS)) {
 187                                 warned = 1;
 188                         } else if (((EXT3_I(inode)->i_flags &
 189                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 190                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 191                                 CWARN("extent-mapped directory found - contact "
 192                                       "CFS: support@clusterfs.com\n");
 193                                 warned = 1;
 194                         }
 195                 }
 196 #endif
 197                 /* no break */
 198         }
 199         case FSFILT_OP_MKDIR:
 200         case FSFILT_OP_MKNOD:
 201                 /* modify one inode + block bitmap + GDT */
 202                 nblocks += 3;
 203                 /* no break */
 204         case FSFILT_OP_LINK:
 205                 /* modify parent directory */
 206                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 207                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 208                 /* create/update logs for each stripe */
 209                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 210                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 211                 break;
 212         case FSFILT_OP_SETATTR:
 213                 /* Setattr on inode */
 214                 nblocks += 1;
 215                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 216                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 217                 /* quota chown log for each stripe */
 218                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 219                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 220                 break;
 221         case FSFILT_OP_CANCEL_UNLINK:
 222                 /* blocks for log header bitmap update OR
 223                  * blocks for catalog header bitmap update + unlink of logs */
 224                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 225                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 226                 break;
 227         case FSFILT_OP_JOIN:
 228                 /* delete 2 file(file + array id) + create 1 file (array id)
 229                  * create/update logs for each stripe */
 230                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 231
 232                 /*create array log for head file*/
 233                 nblocks += 3;
 234                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 235                             EXT3_SINGLEDATA_TRANS_BLOCKS);
 236                 /*update head file array */
 237                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 238                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 239                 break;
 240         default: CERROR("unknown transaction start op %d\n", op);
 241                 LBUG();
 242         }
 243
 244         LASSERT(current->journal_info == desc_private);
 245         journal = EXT3_SB(inode->i_sb)->s_journal;
 246         if (nblocks > journal->j_max_transaction_buffers) {
 247                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 248                        nblocks, op, logs, journal->j_max_transaction_buffers);
 249                 nblocks = journal->j_max_transaction_buffers;
 250         }
 251
 252  journal_start:
 253         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 254         lock_24kernel();
 255         handle = fsfilt_ext3_journal_start(inode, nblocks);
 256         unlock_24kernel();
 257
 258         if (!IS_ERR(handle))
 259                 LASSERT(current->journal_info == handle);
 260         else
 261                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 262                        op, nblocks, PTR_ERR(handle));
 263         return handle;
 264 }
 265
 266 /*
 267  * Calculate the number of buffer credits needed to write multiple pages in
 268  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 269  * doesn't have a nice API for calculating this sort of thing in advance.
 270  *
 271  * See comment above ext3_writepage_trans_blocks for details.  We assume
 272  * no data journaling is being done, but it does allow for all of the pages
 273  * being non-contiguous.  If we are guaranteed contiguous pages we could
 274  * reduce the number of (d)indirect blocks a lot.
 275  *
 276  * With N blocks per page and P pages, for each inode we have at most:
 277  * N*P indirect
 278  * min(N*P, blocksize/4 + 1) dindirect blocks
 279  * niocount tindirect
 280  *
 281  * For the entire filesystem, we have at most:
 282  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 283  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 284  * objcount inode blocks
 285  * 1 superblock
 286  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 287  *
 288  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 289  */
 290 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 291                                       int niocount, struct niobuf_local *nb)
 292 {
 293         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 294         __u64 next_indir;
 295         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 296         int nbitmaps = 0, ngdblocks;
 297         int needed = objcount + 1; /* inodes + superblock */
 298         int i, j;
 299
 300         for (i = 0, j = 0; i < objcount; i++, fso++) {
 301                 /* two or more dindirect blocks in case we cross boundary */
 302                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 303                                     nb[j].offset) >>
 304                                    sb->s_blocksize_bits) /
 305                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 306                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 307
 308                 /* leaf, indirect, tindirect blocks for first block */
 309                 nbitmaps += blockpp + 2;
 310
 311                 j += fso->fso_bufcnt;
 312         }
 313
 314         next_indir = nb[0].offset +
 315                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 316         for (i = 1; i < niocount; i++) {
 317                 if (nb[i].offset >= next_indir) {
 318                         nbitmaps++;     /* additional indirect */
 319                         next_indir = nb[i].offset +
 320                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 321                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 322                         nbitmaps++;     /* additional indirect */
 323                 }
 324                 nbitmaps += blockpp;    /* each leaf in different group? */
 325         }
 326
 327         ngdblocks = nbitmaps;
 328         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 329                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 330         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 331                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 332
 333         needed += nbitmaps + ngdblocks;
 334
 335         /* last_rcvd update */
 336         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 337
 338 #if defined(CONFIG_QUOTA)
 339         /* We assume that there will be 1 bit set in s_dquot.flags for each
 340          * quota file that is active.  This is at least true for now.
 341          */
 342         needed += hweight32(sb_any_quota_enabled(sb)) *
 343                 EXT3_SINGLEDATA_TRANS_BLOCKS;
 344 #endif
 345
 346         return needed;
 347 }
 348
 349 /* We have to start a huge journal transaction here to hold all of the
 350  * metadata for the pages being written here.  This is necessitated by
 351  * the fact that we do lots of prepare_write operations before we do
 352  * any of the matching commit_write operations, so even if we split
 353  * up to use "smaller" transactions none of them could complete until
 354  * all of them were opened.  By having a single journal transaction,
 355  * we eliminate duplicate reservations for common blocks like the
 356  * superblock and group descriptors or bitmaps.
 357  *
 358  * We will start the transaction here, but each prepare_write will
 359  * add a refcount to the transaction, and each commit_write will
 360  * remove a refcount.  The transaction will be closed when all of
 361  * the pages have been written.
 362  */
 363 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 364                                    int niocount, struct niobuf_local *nb,
 365                                    void *desc_private, int logs)
 366 {
 367         journal_t *journal;
 368         handle_t *handle;
 369         int needed;
 370         ENTRY;
 371
 372         LASSERT(current->journal_info == desc_private);
 373         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 374         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 375
 376         /* The number of blocks we could _possibly_ dirty can very large.
 377          * We reduce our request if it is absurd (and we couldn't get that
 378          * many credits for a single handle anyways).
 379          *
 380          * At some point we have to limit the size of I/Os sent at one time,
 381          * increase the size of the journal, or we have to calculate the
 382          * actual journal requirements more carefully by checking all of
 383          * the blocks instead of being maximally pessimistic.  It remains to
 384          * be seen if this is a real problem or not.
 385          */
 386         if (needed > journal->j_max_transaction_buffers) {
 387                 CERROR("want too many journal credits (%d) using %d instead\n",
 388                        needed, journal->j_max_transaction_buffers);
 389                 needed = journal->j_max_transaction_buffers;
 390         }
 391
 392         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 393         lock_24kernel();
 394         handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
 395         unlock_24kernel();
 396         if (IS_ERR(handle)) {
 397                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 398                        PTR_ERR(handle));
 399         } else {
 400                 LASSERT(handle->h_buffer_credits >= needed);
 401                 LASSERT(current->journal_info == handle);
 402         }
 403
 404         RETURN(handle);
 405 }
 406
 407 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 408 {
 409        handle_t *handle = h;
 410
 411        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 412        if (nblocks == 0) {
 413                handle->h_buffer_credits = 0;
 414                CWARN("setting credits of handle %p to zero by request\n", h);
 415        }
 416
 417        if (handle->h_buffer_credits > nblocks)
 418                 return 0;
 419        if (journal_extend(handle, nblocks) == 0)
 420                 return 0;
 421
 422        ext3_mark_inode_dirty(handle, inode);
 423        return journal_restart(handle, nblocks);
 424 }
 425
 426 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 427 {
 428         int rc;
 429         handle_t *handle = h;
 430
 431         LASSERT(current->journal_info == handle);
 432         if (force_sync)
 433                 handle->h_sync = 1; /* recovery likes this */
 434
 435         lock_24kernel();
 436         rc = fsfilt_ext3_journal_stop(handle);
 437         unlock_24kernel();
 438
 439         return rc;
 440 }
 441
 442 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 443                                     void **wait_handle)
 444 {
 445         unsigned long tid;
 446         transaction_t *transaction;
 447         handle_t *handle = h;
 448         journal_t *journal;
 449         int rc;
 450
 451         LASSERT(current->journal_info == handle);
 452
 453         lock_24kernel();
 454         transaction = handle->h_transaction;
 455         journal = transaction->t_journal;
 456         tid = transaction->t_tid;
 457         /* we don't want to be blocked */
 458         handle->h_sync = 0;
 459         rc = fsfilt_ext3_journal_stop(handle);
 460         if (rc) {
 461                 CERROR("error while stopping transaction: %d\n", rc);
 462                 unlock_24kernel();
 463                 return rc;
 464         }
 465 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 466         rtid = log_start_commit(journal, transaction);
 467         if (rtid != tid)
 468                 CERROR("strange race: %lu != %lu\n",
 469                        (unsigned long) tid, (unsigned long) rtid);
 470 #else
 471         log_start_commit(journal, tid);
 472 #endif
 473         unlock_24kernel();
 474
 475         *wait_handle = (void *) tid;
 476         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 477         return 0;
 478 }
 479
 480 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 481 {
 482         journal_t *journal = EXT3_JOURNAL(inode);
 483         tid_t tid = (tid_t)(long)h;
 484
 485         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 486         if (unlikely(is_journal_aborted(journal)))
 487                 return -EIO;
 488
 489         log_wait_commit(EXT3_JOURNAL(inode), tid);
 490
 491         if (unlikely(is_journal_aborted(journal)))
 492                 return -EIO;
 493         return 0;
 494 }
 495
 496 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 497                                struct iattr *iattr, int do_trunc)
 498 {
 499         struct inode *inode = dentry->d_inode;
 500         int rc = 0;
 501
 502         lock_kernel();
 503
 504         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 505          * We are already writing the inode to disk as part of this
 506          * transaction and want to avoid a lot of extra inode writeout
 507          * later on. b=9828 */
 508         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 509                 /* ATTR_SIZE would invoke truncate: clear it */
 510                 iattr->ia_valid &= ~ATTR_SIZE;
 511                 EXT3_I(inode)->i_disksize = inode->i_size = iattr->ia_size;
 512
 513                 if (iattr->ia_valid & ATTR_UID)
 514                         inode->i_uid = iattr->ia_uid;
 515                 if (iattr->ia_valid & ATTR_GID)
 516                         inode->i_gid = iattr->ia_gid;
 517                 if (iattr->ia_valid & ATTR_ATIME)
 518                         inode->i_atime = iattr->ia_atime;
 519                 if (iattr->ia_valid & ATTR_MTIME)
 520                         inode->i_mtime = iattr->ia_mtime;
 521                 if (iattr->ia_valid & ATTR_CTIME)
 522                         inode->i_ctime = iattr->ia_ctime;
 523                 if (iattr->ia_valid & ATTR_MODE) {
 524                         inode->i_mode = iattr->ia_mode;
 525
 526                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 527                                 inode->i_mode &= ~S_ISGID;
 528                 }
 529
 530                 inode->i_sb->s_op->dirty_inode(inode);
 531
 532                 goto out;
 533         }
 534
 535         /* Don't allow setattr to change file type */
 536         if (iattr->ia_valid & ATTR_MODE)
 537                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 538                                  (iattr->ia_mode & ~S_IFMT);
 539
 540         /* We set these flags on the client, but have already checked perms
 541          * so don't confuse inode_change_ok. */
 542         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 543
 544         if (inode->i_op->setattr) {
 545                 rc = inode->i_op->setattr(dentry, iattr);
 546         } else {
 547                 rc = inode_change_ok(inode, iattr);
 548                 if (!rc)
 549                         rc = inode_setattr(inode, iattr);
 550         }
 551
 552  out:
 553         unlock_kernel();
 554         RETURN(rc);
 555 }
 556
 557 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 558                                  unsigned int cmd, unsigned long arg)
 559 {
 560         int rc = 0;
 561         ENTRY;
 562
 563         /* FIXME: Can't do this because of nested transaction deadlock */
 564         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 565                 CERROR("can't set data journal flag on file\n");
 566                 RETURN(-EPERM);
 567         }
 568
 569         if (inode->i_fop->ioctl)
 570                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 571         else
 572                 RETURN(-ENOTTY);
 573
 574         RETURN(rc);
 575 }
 576
 577 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 578                               void *lmm, int lmm_size, const char *name)
 579 {
 580         int rc;
 581
 582         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 583
 584         lock_24kernel();
 585         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 586                                    name, lmm, lmm_size, 0);
 587
 588         unlock_24kernel();
 589
 590         if (rc)
 591                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 592                        inode->i_ino, rc);
 593         return rc;
 594 }
 595
 596 /* Must be called with i_mutex held */
 597 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 598                               const char *name)
 599 {
 600         int rc;
 601
 602         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 603         lock_24kernel();
 604
 605         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 606                             name, lmm, lmm_size);
 607         unlock_24kernel();
 608
 609         /* This gives us the MD size */
 610         if (lmm == NULL)
 611                 return (rc == -ENODATA) ? 0 : rc;
 612
 613         if (rc < 0) {
 614                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 615                        EXT3_XATTR_INDEX_TRUSTED, name,
 616                        inode->i_ino, rc);
 617                 memset(lmm, 0, lmm_size);
 618                 return (rc == -ENODATA) ? 0 : rc;
 619         }
 620
 621         return rc;
 622 }
 623
 624 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 625 {
 626         submit_bio(rw, bio);
 627         return 0;
 628 }
 629
 630 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 631                                     loff_t *off)
 632 {
 633         struct inode *inode = file->f_dentry->d_inode;
 634         int rc = 0;
 635
 636         if (S_ISREG(inode->i_mode))
 637                 rc = file->f_op->read(file, buf, count, off);
 638         else {
 639                 const int blkbits = inode->i_sb->s_blocksize_bits;
 640                 const int blksize = inode->i_sb->s_blocksize;
 641
 642                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 643                        count, inode->i_ino, *off);
 644                 while (count > 0) {
 645                         struct buffer_head *bh;
 646
 647                         bh = NULL;
 648                         if (*off < inode->i_size) {
 649                                 int err = 0;
 650
 651                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 652                                                 0, &err);
 653
 654                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 655
 656                                 if (bh) {
 657                                         memcpy(buf, bh->b_data, blksize);
 658                                         brelse(bh);
 659                                 } else if (err) {
 660                                         /* XXX in theory we should just fake
 661                                          * this buffer and continue like ext3,
 662                                          * especially if this is a partial read
 663                                          */
 664                                         CERROR("error read dir %lu+%llu: %d\n",
 665                                                inode->i_ino, *off, err);
 666                                         RETURN(err);
 667                                 }
 668                         }
 669                         if (!bh) {
 670                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 671
 672                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 673                                 memset(fake, 0, sizeof(*fake));
 674                                 fake->rec_len = cpu_to_le16(blksize);
 675                         }
 676                         count -= blksize;
 677                         buf += blksize;
 678                         *off += blksize;
 679                         rc += blksize;
 680                 }
 681         }
 682
 683         return rc;
 684 }
 685
 686 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 687 {
 688         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 689
 690         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 691
 692         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 693 }
 694
 695 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 696                                       void *handle, fsfilt_cb_t cb_func,
 697                                       void *cb_data)
 698 {
 699         struct fsfilt_cb_data *fcb;
 700
 701         OBD_SLAB_ALLOC(fcb, fcb_cache, GFP_NOFS, sizeof *fcb);
 702         if (fcb == NULL)
 703                 RETURN(-ENOMEM);
 704
 705         fcb->cb_func = cb_func;
 706         fcb->cb_obd = obd;
 707         fcb->cb_last_rcvd = last_rcvd;
 708         fcb->cb_data = cb_data;
 709
 710         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 711         lock_24kernel();
 712         journal_callback_set(handle, fsfilt_ext3_cb_func,
 713                              (struct journal_callback *)fcb);
 714         unlock_24kernel();
 715
 716         return 0;
 717 }
 718
 719 /*
 720  * We need to hack the return value for the free inode counts because
 721  * the current EA code requires one filesystem block per inode with EAs,
 722  * so it is possible to run out of blocks before we run out of inodes.
 723  *
 724  * This can be removed when the ext3 EA code is fixed.
 725  */
 726 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 727 {
 728         struct kstatfs sfs;
 729         int rc;
 730
 731         memset(&sfs, 0, sizeof(sfs));
 732
 733         rc = ll_do_statfs(sb, &sfs);
 734
 735         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 736                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 737                 sfs.f_ffree = sfs.f_bfree;
 738         }
 739
 740         statfs_pack(osfs, &sfs);
 741         return rc;
 742 }
 743
 744 static int fsfilt_ext3_sync(struct super_block *sb)
 745 {
 746         return ext3_force_commit(sb);
 747 }
 748
 749 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 750 #warning "kernel code has old extents/mballoc patch, disabling"
 751 #undef EXT3_MULTIBLOCK_ALLOCATOR
 752 #endif
 753 #ifndef EXT3_EXTENTS_FL
 754 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 755 #endif
 756
 757 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 758 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 759 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 760 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 761 #else
 762 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 763 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 764 #endif
 765
 766 #include <linux/lustre_version.h>
 767 #if EXT3_EXT_MAGIC == 0xf301
 768 #define ee_start e_start
 769 #define ee_block e_block
 770 #define ee_len   e_num
 771 #endif
 772 #ifndef EXT3_BB_MAX_BLOCKS
 773 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 774         ext3_new_blocks(handle, inode, count, goal, err)
 775 #endif
 776
 777 struct bpointers {
 778         unsigned long *blocks;
 779         int *created;
 780         unsigned long start;
 781         int num;
 782         int init_num;
 783         int create;
 784 };
 785
 786 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 787                               unsigned long block, int *aflags)
 788 {
 789         struct ext3_inode_info *ei = EXT3_I(inode);
 790         unsigned long bg_start;
 791         unsigned long colour;
 792         int depth;
 793
 794         if (path) {
 795                 struct ext3_extent *ex;
 796                 depth = path->p_depth;
 797
 798                 /* try to predict block placement */
 799                 if ((ex = path[depth].p_ext)) {
 800 #if 0
 801                         /* This prefers to eat into a contiguous extent
 802                          * rather than find an extent that the whole
 803                          * request will fit into.  This can fragment data
 804                          * block allocation and prevents our lovely 1M I/Os
 805                          * from reaching the disk intact. */
 806                         if (ex->ee_block + ex->ee_len == block)
 807                                 *aflags |= 1;
 808 #endif
 809                         return ex->ee_start + (block - ex->ee_block);
 810                 }
 811
 812                 /* it looks index is empty
 813                  * try to find starting from index itself */
 814                 if (path[depth].p_bh)
 815                         return path[depth].p_bh->b_blocknr;
 816         }
 817
 818         /* OK. use inode's group */
 819         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 820                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 821         colour = (current->pid % 16) *
 822                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 823         return bg_start + colour + block;
 824 }
 825
 826 #define ll_unmap_underlying_metadata(sb, blocknr) \
 827         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 828
 829 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 830 static unsigned long new_blocks(handle_t *handle, struct ext3_extents_tree *tree,
 831                                 struct ext3_ext_path *path, unsigned long block,
 832                                 int *count, int *err)
 833 {
 834         unsigned long pblock, goal;
 835         int aflags = 0;
 836
 837         goal = ext3_ext_find_goal(tree->inode, path, block, &aflags);
 838         aflags |= 2; /* block have been already reserved */
 839         lock_24kernel();
 840         pblock = ext3_mb_new_blocks(handle, tree->inode, goal, count, aflags, err);
 841         unlock_24kernel();
 842         return pblock;
 843
 844 }
 845 #else
 846 static unsigned long new_blocks(handle_t *handle, struct ext3_extents_tree *tree,
 847                                 struct ext3_ext_path *path, unsigned long block,
 848                                 int *count, int *err)
 849 {
 850         struct ext3_allocation_request ar;
 851         unsigned long pblock;
 852         int aflags;
 853
 854         /* find neighbour allocated blocks */
 855         ar.lleft = block;
 856         *err = ext3_ext_search_left(tree, path, &ar.lleft, &ar.pleft);
 857         if (*err)
 858                 return 0;
 859         ar.lright = block;
 860         *err = ext3_ext_search_right(tree, path, &ar.lright, &ar.pright);
 861         if (*err)
 862                 return 0;
 863
 864         /* allocate new block */
 865         ar.goal = ext3_ext_find_goal(tree->inode, path, block, &aflags);
 866         ar.inode = tree->inode;
 867         ar.logical = block;
 868         ar.len = *count;
 869         ar.flags = EXT3_MB_HINT_DATA;
 870         pblock = ext3_mb_new_blocks(handle, &ar, err);
 871         *count = ar.len;
 872         return pblock;
 873
 874 }
 875 #endif
 876
 877 static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
 878                                   struct ext3_ext_path *path,
 879                                   struct ext3_ext_cache *cex)
 880 {
 881         struct inode *inode = tree->inode;
 882         struct bpointers *bp = tree->private;
 883         struct ext3_extent nex;
 884         unsigned long pblock;
 885         unsigned long tgen;
 886         int count, err, i;
 887         handle_t *handle;
 888
 889         i = EXT_DEPTH(tree);
 890         EXT_ASSERT(i == path->p_depth);
 891         EXT_ASSERT(path[i].p_hdr);
 892
 893         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 894                 err = EXT_CONTINUE;
 895                 goto map;
 896         }
 897
 898         if (bp->create == 0) {
 899                 i = 0;
 900                 if (cex->ec_block < bp->start)
 901                         i = bp->start - cex->ec_block;
 902                 if (i >= cex->ec_len)
 903                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 904                                         i, cex->ec_len);
 905                 for (; i < cex->ec_len && bp->num; i++) {
 906                         *(bp->created) = 0;
 907                         bp->created++;
 908                         *(bp->blocks) = 0;
 909                         bp->blocks++;
 910                         bp->num--;
 911                         bp->start++;
 912                 }
 913
 914                 return EXT_CONTINUE;
 915         }
 916
 917         tgen = EXT_GENERATION(tree);
 918         count = ext3_ext_calc_credits_for_insert(tree, path);
 919         ext3_up_truncate_sem(inode);
 920
 921         lock_24kernel();
 922         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 923         unlock_24kernel();
 924         if (IS_ERR(handle)) {
 925                 ext3_down_truncate_sem(inode);
 926                 return PTR_ERR(handle);
 927         }
 928
 929         ext3_down_truncate_sem(inode);
 930         if (tgen != EXT_GENERATION(tree)) {
 931                 /* the tree has changed. so path can be invalid at moment */
 932                 lock_24kernel();
 933                 fsfilt_ext3_journal_stop(handle);
 934                 unlock_24kernel();
 935                 return EXT_REPEAT;
 936         }
 937
 938         count = cex->ec_len;
 939         pblock = new_blocks(handle, tree, path, cex->ec_block, &count, &err);
 940         if (!pblock)
 941                 goto out;
 942         EXT_ASSERT(count <= cex->ec_len);
 943
 944         /* insert new extent */
 945         nex.ee_block = cex->ec_block;
 946         nex.ee_start = pblock;
 947         nex.ee_len = count;
 948         err = ext3_ext_insert_extent(handle, tree, path, &nex);
 949         if (err) {
 950                 CERROR("can't insert extent: %d\n", err);
 951                 /* XXX: export ext3_free_blocks() */
 952                 /*ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);*/
 953                 goto out;
 954         }
 955
 956         /*
 957          * Putting len of the actual extent we just inserted,
 958          * we are asking ext3_ext_walk_space() to continue
 959          * scaning after that block
 960          */
 961         cex->ec_len = nex.ee_len;
 962         cex->ec_start = nex.ee_start;
 963         BUG_ON(nex.ee_len == 0);
 964         BUG_ON(nex.ee_block != cex->ec_block);
 965
 966 out:
 967         lock_24kernel();
 968         fsfilt_ext3_journal_stop(handle);
 969         unlock_24kernel();
 970 map:
 971         if (err >= 0) {
 972                 /* map blocks */
 973                 if (bp->num == 0) {
 974                         CERROR("hmm. why do we find this extent?\n");
 975                         CERROR("initial space: %lu:%u\n",
 976                                 bp->start, bp->init_num);
 977                         CERROR("current extent: %u/%u/%u %d\n",
 978                                 cex->ec_block, cex->ec_len,
 979                                 cex->ec_start, cex->ec_type);
 980                 }
 981                 i = 0;
 982                 if (cex->ec_block < bp->start)
 983                         i = bp->start - cex->ec_block;
 984                 if (i >= cex->ec_len)
 985                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 986                                         i, cex->ec_len);
 987                 for (; i < cex->ec_len && bp->num; i++) {
 988                         *(bp->blocks) = cex->ec_start + i;
 989                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 990                                 *(bp->created) = 0;
 991                         } else {
 992                                 *(bp->created) = 1;
 993                                 /* unmap any possible underlying metadata from
 994                                  * the block device mapping.  bug 6998. */
 995                                 ll_unmap_underlying_metadata(inode->i_sb,
 996                                                              *(bp->blocks));
 997                         }
 998                         bp->created++;
 999                         bp->blocks++;
1000                         bp->num--;
1001                         bp->start++;
1002                 }
1003         }
1004         return err;
1005 }
1006
1007 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1008                        unsigned long num, unsigned long *blocks,
1009                        int *created, int create)
1010 {
1011         struct ext3_extents_tree tree;
1012         struct bpointers bp;
1013         int err;
1014
1015         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1016                block, block + num - 1, (unsigned) inode->i_ino);
1017
1018         ext3_init_tree_desc(&tree, inode);
1019         tree.private = &bp;
1020         bp.blocks = blocks;
1021         bp.created = created;
1022         bp.start = block;
1023         bp.init_num = bp.num = num;
1024         bp.create = create;
1025
1026         ext3_down_truncate_sem(inode);
1027         err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
1028         ext3_ext_invalidate_cache(&tree);
1029         ext3_up_truncate_sem(inode);
1030
1031         return err;
1032 }
1033
1034 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1035                                     int pages, unsigned long *blocks,
1036                                     int *created, int create)
1037 {
1038         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1039         int rc = 0, i = 0;
1040         struct page *fp = NULL;
1041         int clen = 0;
1042
1043         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1044                 inode->i_ino, pages, (*page)->index);
1045
1046         /* pages are sorted already. so, we just have to find
1047          * contig. space and process them properly */
1048         while (i < pages) {
1049                 if (fp == NULL) {
1050                         /* start new extent */
1051                         fp = *page++;
1052                         clen = 1;
1053                         i++;
1054                         continue;
1055                 } else if (fp->index + clen == (*page)->index) {
1056                         /* continue the extent */
1057                         page++;
1058                         clen++;
1059                         i++;
1060                         continue;
1061                 }
1062
1063                 /* process found extent */
1064                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1065                                         clen * blocks_per_page, blocks,
1066                                         created, create);
1067                 if (rc)
1068                         GOTO(cleanup, rc);
1069
1070                 /* look for next extent */
1071                 fp = NULL;
1072                 blocks += blocks_per_page * clen;
1073                 created += blocks_per_page * clen;
1074         }
1075
1076         if (fp)
1077                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1078                                         clen * blocks_per_page, blocks,
1079                                         created, create);
1080 cleanup:
1081         return rc;
1082 }
1083 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1084
1085 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1086                                unsigned long *blocks, int *created, int create);
1087 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1088                                    int pages, unsigned long *blocks,
1089                                    int *created, int create)
1090 {
1091         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1092         unsigned long *b;
1093         int rc = 0, i, *cr;
1094
1095         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1096                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1097                 if (rc) {
1098                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1099                                inode->i_ino, *b, *cr, create, rc);
1100                         break;
1101                 }
1102
1103                 b += blocks_per_page;
1104                 cr += blocks_per_page;
1105         }
1106         return rc;
1107 }
1108
1109 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1110                                 int pages, unsigned long *blocks,
1111                                 int *created, int create,
1112                                 struct semaphore *optional_sem)
1113 {
1114         int rc;
1115 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1116         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1117                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1118                                                      blocks, created, create);
1119                 return rc;
1120         }
1121 #endif
1122         if (optional_sem != NULL)
1123                 down(optional_sem);
1124         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1125                                             created, create);
1126         if (optional_sem != NULL)
1127                 up(optional_sem);
1128
1129         return rc;
1130 }
1131
1132 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1133 {
1134         unsigned long block;
1135         struct buffer_head *bh;
1136         int err, blocksize, csize, boffs, osize = size;
1137
1138         /* prevent reading after eof */
1139         lock_kernel();
1140         if (inode->i_size < *offs + size) {
1141                 size = inode->i_size - *offs;
1142                 unlock_kernel();
1143                 if (size < 0) {
1144                         CERROR("size %llu is too short for read %u@%llu\n",
1145                                inode->i_size, size, *offs);
1146                         return -EIO;
1147                 } else if (size == 0) {
1148                         return 0;
1149                 }
1150         } else {
1151                 unlock_kernel();
1152         }
1153
1154         blocksize = 1 << inode->i_blkbits;
1155
1156         while (size > 0) {
1157                 block = *offs >> inode->i_blkbits;
1158                 boffs = *offs & (blocksize - 1);
1159                 csize = min(blocksize - boffs, size);
1160                 bh = ext3_bread(NULL, inode, block, 0, &err);
1161                 if (!bh) {
1162                         CERROR("can't read block: %d\n", err);
1163                         return err;
1164                 }
1165
1166                 memcpy(buf, bh->b_data + boffs, csize);
1167                 brelse(bh);
1168
1169                 *offs += csize;
1170                 buf += csize;
1171                 size -= csize;
1172         }
1173         return osize;
1174 }
1175 EXPORT_SYMBOL(fsfilt_ext3_read);
1176
1177 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1178                                    int size, loff_t *offs)
1179 {
1180         int rc;
1181         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1182         if (rc > 0)
1183                 rc = 0;
1184         return rc;
1185 }
1186
1187 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1188                                 loff_t *offs, handle_t *handle)
1189 {
1190         struct buffer_head *bh = NULL;
1191         loff_t old_size = inode->i_size, offset = *offs;
1192         loff_t new_size = inode->i_size;
1193         unsigned long block;
1194         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
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                         break;
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                         break;
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                         break;
1222                 }
1223                 if (offset + size > new_size)
1224                         new_size = offset + size;
1225                 offset += size;
1226                 bufsize -= size;
1227                 buf += size;
1228         }
1229         if (bh)
1230                 brelse(bh);
1231
1232         /* correct in-core and on-disk sizes */
1233         if (new_size > inode->i_size) {
1234                 lock_kernel();
1235                 if (new_size > inode->i_size)
1236                         inode->i_size = new_size;
1237                 if (inode->i_size > EXT3_I(inode)->i_disksize)
1238                         EXT3_I(inode)->i_disksize = inode->i_size;
1239                 if (inode->i_size > old_size)
1240                         mark_inode_dirty(inode);
1241                 unlock_kernel();
1242         }
1243
1244         if (err == 0)
1245                 *offs = offset;
1246         return err;
1247 }
1248 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1249
1250 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1251                                     loff_t *offs, int force_sync)
1252 {
1253         struct inode *inode = file->f_dentry->d_inode;
1254         handle_t *handle;
1255         int err, block_count = 0, blocksize;
1256
1257         /* Determine how many transaction credits are needed */
1258         blocksize = 1 << inode->i_blkbits;
1259         block_count = (*offs & (blocksize - 1)) + bufsize;
1260         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1261
1262         lock_24kernel();
1263         handle = fsfilt_ext3_journal_start(inode,
1264                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1265         unlock_24kernel();
1266         if (IS_ERR(handle)) {
1267                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1268                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1269                 return PTR_ERR(handle);
1270         }
1271
1272         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1273
1274         if (!err && force_sync)
1275                 handle->h_sync = 1; /* recovery likes this */
1276
1277         lock_24kernel();
1278         fsfilt_ext3_journal_stop(handle);
1279         unlock_24kernel();
1280
1281         return err;
1282 }
1283
1284 static int fsfilt_ext3_setup(struct super_block *sb)
1285 {
1286 #if 0
1287         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1288         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1289 #endif
1290 #ifdef S_PDIROPS
1291         CWARN("Enabling PDIROPS\n");
1292         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1293         sb->s_flags |= S_PDIROPS;
1294 #endif
1295         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1296                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1297 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13)) && HAVE_QUOTA_SUPPORT
1298         set_opt(EXT3_SB(sb)->s_mount_opt, QUOTA);
1299 #endif
1300         return 0;
1301 }
1302
1303 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1304    objects. Logs is number of logfiles to update */
1305 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1306 {
1307         if ( !fso ) {
1308                 switch(op) {
1309                 case FSFILT_OP_CREATE:
1310                                  /* directory leaf, index & indirect & EA*/
1311                         return 4 + 3 * logs;
1312                 case FSFILT_OP_UNLINK:
1313                         return 3 * logs;
1314                 }
1315         } else {
1316                 int i;
1317                 int needed = 0;
1318                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1319                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1320                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1321                 for (i = 0; i < op; i++, fso++) {
1322                         int nblocks = fso->fso_bufcnt * blockpp;
1323                         int ndindirect = min(nblocks, addrpp + 1);
1324                         int nindir = nblocks + ndindirect + 1;
1325
1326                         needed += nindir;
1327                 }
1328                 return needed + 3 * logs;
1329         }
1330
1331         return 0;
1332 }
1333
1334 static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
1335
1336 #define DQINFO_COPY(out, in)                    \
1337 do {                                            \
1338         Q_COPY(out, in, dqi_bgrace);            \
1339         Q_COPY(out, in, dqi_igrace);            \
1340         Q_COPY(out, in, dqi_flags);             \
1341         Q_COPY(out, in, dqi_valid);             \
1342 } while (0)
1343
1344 #define DQBLK_COPY(out, in)                     \
1345 do {                                            \
1346         Q_COPY(out, in, dqb_bhardlimit);        \
1347         Q_COPY(out, in, dqb_bsoftlimit);        \
1348         Q_COPY(out, in, dqb_curspace);          \
1349         Q_COPY(out, in, dqb_ihardlimit);        \
1350         Q_COPY(out, in, dqb_isoftlimit);        \
1351         Q_COPY(out, in, dqb_curinodes);         \
1352         Q_COPY(out, in, dqb_btime);             \
1353         Q_COPY(out, in, dqb_itime);             \
1354         Q_COPY(out, in, dqb_valid);             \
1355 } while (0)
1356
1357
1358
1359 static int fsfilt_ext3_quotactl(struct super_block *sb,
1360                                 struct obd_quotactl *oqc)
1361 {
1362         int i, rc = 0, error = 0;
1363         struct quotactl_ops *qcop;
1364         struct if_dqinfo *info;
1365         struct if_dqblk *dqblk;
1366         ENTRY;
1367
1368         if (!sb->s_qcop)
1369                 RETURN(-ENOSYS);
1370
1371         OBD_ALLOC_PTR(info);
1372         if (!info)
1373                 RETURN(-ENOMEM);
1374         OBD_ALLOC_PTR(dqblk);
1375         if (!dqblk) {
1376                 OBD_FREE_PTR(info);
1377                 RETURN(-ENOMEM);
1378         }
1379
1380         DQINFO_COPY(info, &oqc->qc_dqinfo);
1381         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1382
1383         qcop = sb->s_qcop;
1384         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1385                 for (i = 0; i < MAXQUOTAS; i++) {
1386                         if (!Q_TYPESET(oqc, i))
1387                                 continue;
1388
1389                         if (oqc->qc_cmd == Q_QUOTAON) {
1390                                 if (!qcop->quota_on)
1391                                         GOTO(out, rc = -ENOSYS);
1392                                 rc = qcop->quota_on(sb, i, oqc->qc_id,
1393                                                     (char *)op_quotafile[i]);
1394                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1395                                 if (!qcop->quota_off)
1396                                         GOTO(out, rc = -ENOSYS);
1397                                 rc = qcop->quota_off(sb, i);
1398                         }
1399
1400                         if (rc == -EBUSY)
1401                                 error = rc;
1402                         else if (rc)
1403                                 GOTO(out, rc);
1404                 }
1405                 GOTO(out, rc ?: error);
1406         }
1407
1408         switch (oqc->qc_cmd) {
1409         case Q_GETOINFO:
1410         case Q_GETINFO:
1411                 if (!qcop->get_info)
1412                         GOTO(out, rc = -ENOSYS);
1413                 rc = qcop->get_info(sb, oqc->qc_type, info);
1414                 break;
1415         case Q_SETQUOTA:
1416         case Q_INITQUOTA:
1417                 if (!qcop->set_dqblk)
1418                         GOTO(out, rc = -ENOSYS);
1419                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1420                 break;
1421         case Q_GETOQUOTA:
1422         case Q_GETQUOTA:
1423                 if (!qcop->get_dqblk)
1424                         GOTO(out, rc = -ENOSYS);
1425                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1426                 break;
1427         case Q_SYNC:
1428                 if (!sb->s_qcop->quota_sync)
1429                         GOTO(out, rc = -ENOSYS);
1430                 qcop->quota_sync(sb, oqc->qc_type);
1431                 break;
1432         default:
1433                 CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
1434                 LBUG();
1435         }
1436 out:
1437         DQINFO_COPY(&oqc->qc_dqinfo, info);
1438         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1439
1440         OBD_FREE_PTR(info);
1441         OBD_FREE_PTR(dqblk);
1442
1443         if (rc)
1444                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1445                                 "failed: %d\n",
1446                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1447         RETURN(rc);
1448 }
1449
1450 struct chk_dqblk{
1451         struct hlist_node       dqb_hash;        /* quotacheck hash */
1452         struct list_head        dqb_list;        /* in list also */
1453         qid_t                   dqb_id;          /* uid/gid */
1454         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1455         __u32                   dqb_bhardlimit;  /* block hard limit */
1456         __u32                   dqb_bsoftlimit;  /* block soft limit */
1457         qsize_t                 dqb_curspace;    /* current space */
1458         __u32                   dqb_ihardlimit;  /* inode hard limit */
1459         __u32                   dqb_isoftlimit;  /* inode soft limit */
1460         __u32                   dqb_curinodes;   /* current inodes */
1461         __u64                   dqb_btime;       /* block grace time */
1462         __u64                   dqb_itime;       /* inode grace time */
1463         __u32                   dqb_valid;       /* flag for above fields */
1464 };
1465
1466 static inline unsigned int chkquot_hash(qid_t id, int type)
1467                                         __attribute__((__const__));
1468
1469 static inline unsigned int chkquot_hash(qid_t id, int type)
1470 {
1471         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1472 }
1473
1474 static inline struct chk_dqblk *
1475 find_chkquot(struct hlist_head *head, qid_t id, int type)
1476 {
1477         struct hlist_node *node;
1478         struct chk_dqblk *cdqb;
1479
1480         hlist_for_each(node, head) {
1481                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1482                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1483                         return cdqb;
1484         }
1485
1486         return NULL;
1487 }
1488
1489 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1490 {
1491         struct chk_dqblk *cdqb;
1492
1493         OBD_ALLOC_PTR(cdqb);
1494         if (cdqb) {
1495                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1496                 INIT_LIST_HEAD(&cdqb->dqb_list);
1497                 cdqb->dqb_id = id;
1498                 cdqb->dqb_type = type;
1499         }
1500
1501         return cdqb;
1502 }
1503
1504 static struct chk_dqblk *
1505 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1506       qid_t id, int type, int first_check)
1507 {
1508         struct hlist_head *head = hash + chkquot_hash(id, type);
1509         struct if_dqblk dqb;
1510         struct chk_dqblk *cdqb;
1511         int rc;
1512
1513         cdqb = find_chkquot(head, id, type);
1514         if (cdqb)
1515                 return cdqb;
1516
1517         cdqb = alloc_chkquot(id, type);
1518         if (!cdqb)
1519                 return NULL;
1520
1521         if (!first_check) {
1522                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1523                 if (rc) {
1524                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1525                                id, type, rc);
1526                 } else {
1527                         DQBLK_COPY(cdqb, &dqb);
1528                         cdqb->dqb_curspace = 0;
1529                         cdqb->dqb_curinodes = 0;
1530                 }
1531         }
1532
1533         hlist_add_head(&cdqb->dqb_hash, head);
1534         list_add_tail(&cdqb->dqb_list, list);
1535
1536         return cdqb;
1537 }
1538
1539 static inline int quota_onoff(struct super_block *sb, int cmd, int type)
1540 {
1541         struct obd_quotactl *oqctl;
1542         int rc;
1543
1544         OBD_ALLOC_PTR(oqctl);
1545         if (!oqctl)
1546                 RETURN(-ENOMEM);
1547
1548         oqctl->qc_cmd = cmd;
1549         oqctl->qc_id = QFMT_LDISKFS;
1550         oqctl->qc_type = type;
1551         rc = fsfilt_ext3_quotactl(sb, oqctl);
1552
1553         OBD_FREE_PTR(oqctl);
1554         return rc;
1555 }
1556
1557 static inline int read_old_dqinfo(struct super_block *sb, int type,
1558                                   struct if_dqinfo *dqinfo)
1559 {
1560         struct obd_quotactl *oqctl;
1561         int rc;
1562         ENTRY;
1563
1564         OBD_ALLOC_PTR(oqctl);
1565         if (!oqctl)
1566                 RETURN(-ENOMEM);
1567
1568         oqctl->qc_cmd = Q_GETINFO;
1569         oqctl->qc_type = type;
1570         rc = fsfilt_ext3_quotactl(sb, oqctl);
1571         if (!rc)
1572                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1573
1574         OBD_FREE_PTR(oqctl);
1575         RETURN(rc);
1576 }
1577
1578 static inline struct ext3_group_desc *
1579 get_group_desc(struct super_block *sb, int group)
1580 {
1581         unsigned long desc_block, desc;
1582         struct ext3_group_desc *gdp;
1583
1584         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1585         desc = group % EXT3_DESC_PER_BLOCK(sb);
1586         gdp = (struct ext3_group_desc *)
1587               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1588
1589         return gdp + desc;
1590 }
1591
1592 static inline struct buffer_head *
1593 read_inode_bitmap(struct super_block *sb, unsigned long group)
1594 {
1595         struct ext3_group_desc *desc;
1596         struct buffer_head *bh;
1597
1598         desc = get_group_desc(sb, group);
1599         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1600
1601         return bh;
1602 }
1603
1604 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1605                                      struct buffer_head *bitmap_bh,
1606                                      int index, unsigned long ino)
1607 {
1608         struct inode *inode = NULL;
1609
1610         if (ext3_test_bit(index, bitmap_bh->b_data))
1611                 inode = iget(sb, ino);
1612
1613         return inode;
1614 }
1615
1616 struct qchk_ctxt {
1617         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1618         struct list_head        qckt_list;                   /* quotacheck list */
1619         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1620         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1621 };
1622
1623 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1624                            struct obd_quotactl *oqc)
1625 {
1626         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1627         loff_t size = 0;
1628         qid_t qid[MAXQUOTAS];
1629         int cnt, i, rc = 0;
1630
1631         if (!inode)
1632                 return 0;
1633
1634         qid[USRQUOTA] = inode->i_uid;
1635         qid[GRPQUOTA] = inode->i_gid;
1636
1637         if (S_ISDIR(inode->i_mode) ||
1638             S_ISREG(inode->i_mode) ||
1639             S_ISLNK(inode->i_mode))
1640                 size = inode_get_bytes(inode);
1641
1642         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1643                 if (!Q_TYPESET(oqc, cnt))
1644                         continue;
1645
1646                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1647                                 &qctxt->qckt_list, qid[cnt], cnt,
1648                                 qctxt->qckt_first_check[cnt]);
1649                 if (!cdqb[cnt]) {
1650                         rc = -ENOMEM;
1651                         break;
1652                 }
1653
1654                 cdqb[cnt]->dqb_curspace += size;
1655                 cdqb[cnt]->dqb_curinodes++;
1656         }
1657
1658         if (rc) {
1659                 for (i = 0; i < cnt; i++) {
1660                         if (!Q_TYPESET(oqc, i))
1661                                 continue;
1662                         LASSERT(cdqb[i]);
1663                         cdqb[i]->dqb_curspace -= size;
1664                         cdqb[i]->dqb_curinodes--;
1665                 }
1666         }
1667
1668         return rc;
1669 }
1670
1671 static int v2_write_dqheader(struct file *f, int type)
1672 {
1673         static const __u32 quota_magics[] = V2_INITQMAGICS;
1674         static const __u32 quota_versions[] = V2_INITQVERSIONS;
1675         struct v2_disk_dqheader dqhead;
1676         loff_t offset = 0;
1677
1678         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1679         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1680
1681         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1682         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1683
1684         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1685 }
1686
1687 /* write dqinfo struct in a new quota file */
1688 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1689 {
1690         struct v2_disk_dqinfo dqinfo;
1691         __u32 blocks = V2_DQTREEOFF + 1;
1692         loff_t offset = V2_DQINFOOFF;
1693
1694         if (info) {
1695                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1696                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1697                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1698                                                ~DQF_INFO_DIRTY);
1699         } else {
1700                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1701                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1702                 dqinfo.dqi_flags = 0;
1703         }
1704
1705         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1706         dqinfo.dqi_free_blk = 0;
1707         dqinfo.dqi_free_entry = 0;
1708
1709         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1710 }
1711
1712 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1713                                   struct obd_quotactl *oqc)
1714 {
1715         int i, rc = 0;
1716         ENTRY;
1717
1718         for (i = 0; i < MAXQUOTAS; i++) {
1719                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1720                                          NULL : &qctxt->qckt_dqinfo[i];
1721                 struct file *file;
1722
1723                 if (!Q_TYPESET(oqc, i))
1724                         continue;
1725
1726                 file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
1727                                  0644);
1728                 if (IS_ERR(file)) {
1729                         rc = PTR_ERR(file);
1730                         CERROR("can't create %s file: rc = %d\n",
1731                                op_quotafile[i], rc);
1732                         GOTO(out, rc);
1733                 }
1734
1735                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1736                         CERROR("file %s is not regular", op_quotafile[i]);
1737                         filp_close(file, 0);
1738                         GOTO(out, rc = -EINVAL);
1739                 }
1740
1741                 rc = v2_write_dqheader(file, i);
1742                 if (rc) {
1743                         filp_close(file, 0);
1744                         GOTO(out, rc);
1745                 }
1746
1747                 rc = v2_write_dqinfo(file, i, info);
1748                 filp_close(file, 0);
1749                 if (rc)
1750                         GOTO(out, rc);
1751         }
1752
1753 out:
1754         RETURN(rc);
1755 }
1756
1757
1758 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1759                           struct chk_dqblk *cdqb)
1760 {
1761         struct obd_quotactl *oqc;
1762         long now;
1763         int rc;
1764         ENTRY;
1765
1766         OBD_ALLOC_PTR(oqc);
1767         if (!oqc)
1768                 RETURN(-ENOMEM);
1769
1770         now = CURRENT_SECONDS;
1771
1772         if (cdqb->dqb_bsoftlimit &&
1773             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1774             !cdqb->dqb_btime)
1775                 cdqb->dqb_btime =
1776                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1777
1778         if (cdqb->dqb_isoftlimit &&
1779             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1780             !cdqb->dqb_itime)
1781                 cdqb->dqb_itime =
1782                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1783
1784         cdqb->dqb_valid = QIF_ALL;
1785
1786         oqc->qc_cmd = Q_SETQUOTA;
1787         oqc->qc_type = cdqb->dqb_type;
1788         oqc->qc_id = cdqb->dqb_id;
1789         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1790
1791         rc = fsfilt_ext3_quotactl(sb, oqc);
1792         OBD_FREE_PTR(oqc);
1793         RETURN(rc);
1794 }
1795
1796 static int prune_chkquots(struct super_block *sb,
1797                           struct qchk_ctxt *qctxt, int error)
1798 {
1799         struct chk_dqblk *cdqb, *tmp;
1800         int rc;
1801
1802         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1803                 if (!error) {
1804                         rc = commit_chkquot(sb, qctxt, cdqb);
1805                         if (rc)
1806                                 error = rc;
1807                 }
1808                 hlist_del_init(&cdqb->dqb_hash);
1809                 list_del(&cdqb->dqb_list);
1810                 OBD_FREE_PTR(cdqb);
1811         }
1812
1813         return error;
1814 }
1815
1816 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1817                                   struct obd_quotactl *oqc)
1818 {
1819         struct ext3_sb_info *sbi = EXT3_SB(sb);
1820         int i, group;
1821         struct qchk_ctxt *qctxt;
1822         struct buffer_head *bitmap_bh = NULL;
1823         unsigned long ino;
1824         struct inode *inode;
1825         int rc = 0;
1826         ENTRY;
1827
1828         /* turn on quota and read dqinfo if existed */
1829         OBD_ALLOC_PTR(qctxt);
1830         if (!qctxt) {
1831                 oqc->qc_stat = -ENOMEM;
1832                 RETURN(-ENOMEM);
1833         }
1834
1835         for (i = 0; i < NR_DQHASH; i++)
1836                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1837         INIT_LIST_HEAD(&qctxt->qckt_list);
1838
1839         for (i = 0; i < MAXQUOTAS; i++) {
1840                 if (!Q_TYPESET(oqc, i))
1841                         continue;
1842
1843                 rc = quota_onoff(sb, Q_QUOTAON, i);
1844                 if (!rc || rc == -EBUSY) {
1845                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1846                         if (rc)
1847                                 GOTO(out, rc);
1848                 } else if (rc == -ENOENT) {
1849                         qctxt->qckt_first_check[i] = 1;
1850                 } else if (rc) {
1851                         GOTO(out, rc);
1852                 }
1853         }
1854
1855         /* check quota and update in hash */
1856         for (group = 0; group < sbi->s_groups_count; group++) {
1857                 ino = group * sbi->s_inodes_per_group + 1;
1858                 bitmap_bh = read_inode_bitmap(sb, group);
1859                 if (!bitmap_bh) {
1860                         CERROR("read_inode_bitmap group %d failed", group);
1861                         GOTO(out, rc = -EIO);
1862                 }
1863
1864                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1865                         if (ino < sbi->s_first_ino)
1866                                 continue;
1867
1868                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1869                         rc = add_inode_quota(inode, qctxt, oqc);
1870                         iput(inode);
1871                         if (rc) {
1872                                 brelse(bitmap_bh);
1873                                 GOTO(out, rc);
1874                         }
1875                 }
1876
1877                 brelse(bitmap_bh);
1878         }
1879
1880         /* read old quota limits from old quota file. (only for the user
1881          * has limits but hasn't file) */
1882 #ifdef HAVE_QUOTA_SUPPORT
1883         for (i = 0; i < MAXQUOTAS; i++) {
1884                 struct list_head id_list;
1885                 struct dquot_id *dqid, *tmp;
1886
1887                 if (!Q_TYPESET(oqc, i))
1888                         continue;
1889
1890                 if (qctxt->qckt_first_check[i])
1891                         continue;
1892
1893
1894                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1895                 INIT_LIST_HEAD(&id_list);
1896 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1897                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1898 #else
1899                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1900 #endif
1901                 if (rc)
1902                         CERROR("read old limits failed. (rc:%d)\n", rc);
1903
1904                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1905                         list_del_init(&dqid->di_link);
1906
1907                         if (!rc)
1908                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1909                                       dqid->di_id, i,
1910                                       qctxt->qckt_first_check[i]);
1911                         kfree(dqid);
1912                 }
1913         }
1914 #endif
1915         /* turn off quota cause we are to dump chk_dqblk to files */
1916         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1917
1918         rc = create_new_quota_files(qctxt, oqc);
1919         if (rc)
1920                 GOTO(out, rc);
1921
1922         /* we use vfs functions to set dqblk, so turn quota on */
1923         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
1924 out:
1925         /* dump and free chk_dqblk */
1926         rc = prune_chkquots(sb, qctxt, rc);
1927         OBD_FREE_PTR(qctxt);
1928
1929         /* turn off quota, `lfs quotacheck` will turn on when all
1930          * nodes quotacheck finish. */
1931         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1932
1933         oqc->qc_stat = rc;
1934         if (rc)
1935                 CERROR("quotacheck failed: rc = %d\n", rc);
1936
1937         RETURN(rc);
1938 }
1939
1940 #ifdef HAVE_QUOTA_SUPPORT
1941 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
1942                                  int cmd)
1943 {
1944         int rc = 0;
1945         ENTRY;
1946
1947         if (lqi->qi_files[type] == NULL) {
1948                 CERROR("operate qinfo before it's enabled!\n");
1949                 RETURN(-EIO);
1950         }
1951
1952         switch (cmd) {
1953         case QFILE_CHK:
1954                 rc = lustre_check_quota_file(lqi, type);
1955                 break;
1956         case QFILE_RD_INFO:
1957                 rc = lustre_read_quota_info(lqi, type);
1958                 break;
1959         case QFILE_WR_INFO:
1960                 rc = lustre_write_quota_info(lqi, type);
1961                 break;
1962         case QFILE_INIT_INFO:
1963                 rc = lustre_init_quota_info(lqi, type);
1964                 break;
1965         default:
1966                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1967                 LBUG();
1968                 break;
1969         }
1970         RETURN(rc);
1971 }
1972
1973 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
1974                             struct list_head *list)
1975 {
1976         return lustre_get_qids(file, inode, type, list);
1977 }
1978
1979 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
1980 {
1981         int rc = 0;
1982         ENTRY;
1983
1984         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
1985                 CERROR("operate dquot before it's enabled!\n");
1986                 RETURN(-EIO);
1987         }
1988
1989         switch (cmd) {
1990         case QFILE_RD_DQUOT:
1991                 rc = lustre_read_dquot(dquot);
1992                 break;
1993         case QFILE_WR_DQUOT:
1994                 if (dquot->dq_dqb.dqb_ihardlimit ||
1995                     dquot->dq_dqb.dqb_isoftlimit ||
1996                     dquot->dq_dqb.dqb_bhardlimit ||
1997                     dquot->dq_dqb.dqb_bsoftlimit)
1998                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1999                 else
2000                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2001
2002                 rc = lustre_commit_dquot(dquot);
2003                 if (rc >= 0)
2004                         rc = 0;
2005                 break;
2006         default:
2007                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2008                 LBUG();
2009                 break;
2010         }
2011         RETURN(rc);
2012 }
2013 #endif
2014
2015 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2016 {
2017         return (EXT3_SB(sb)->journal_bdev);
2018 }
2019 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2020
2021 static struct fsfilt_operations fsfilt_ext3_ops = {
2022         .fs_type                = "ext3",
2023         .fs_owner               = THIS_MODULE,
2024         .fs_getlabel            = fsfilt_ext3_get_label,
2025         .fs_setlabel            = fsfilt_ext3_set_label,
2026         .fs_uuid                = fsfilt_ext3_uuid,
2027         .fs_start               = fsfilt_ext3_start,
2028         .fs_brw_start           = fsfilt_ext3_brw_start,
2029         .fs_extend              = fsfilt_ext3_extend,
2030         .fs_commit              = fsfilt_ext3_commit,
2031         .fs_commit_async        = fsfilt_ext3_commit_async,
2032         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2033         .fs_setattr             = fsfilt_ext3_setattr,
2034         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2035         .fs_set_md              = fsfilt_ext3_set_md,
2036         .fs_get_md              = fsfilt_ext3_get_md,
2037         .fs_readpage            = fsfilt_ext3_readpage,
2038         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2039         .fs_statfs              = fsfilt_ext3_statfs,
2040         .fs_sync                = fsfilt_ext3_sync,
2041         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2042         .fs_write_record        = fsfilt_ext3_write_record,
2043         .fs_read_record         = fsfilt_ext3_read_record,
2044         .fs_setup               = fsfilt_ext3_setup,
2045         .fs_send_bio            = fsfilt_ext3_send_bio,
2046         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2047         .fs_quotactl            = fsfilt_ext3_quotactl,
2048         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2049 #ifdef HAVE_DISK_INODE_VERSION
2050         .fs_get_version         = fsfilt_ext3_get_version,
2051         .fs_set_version         = fsfilt_ext3_set_version,
2052 #endif
2053 #ifdef HAVE_QUOTA_SUPPORT
2054         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2055         .fs_qids                = fsfilt_ext3_qids,
2056         .fs_dquot               = fsfilt_ext3_dquot,
2057 #endif
2058         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2059 };
2060
2061 static int __init fsfilt_ext3_init(void)
2062 {
2063         int rc;
2064
2065         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2066                                       sizeof(struct fsfilt_cb_data), 0, 0);
2067         if (!fcb_cache) {
2068                 CERROR("error allocating fsfilt journal callback cache\n");
2069                 GOTO(out, rc = -ENOMEM);
2070         }
2071
2072         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2073
2074         if (rc) {
2075                 int err = cfs_mem_cache_destroy(fcb_cache);
2076                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2077         }
2078 out:
2079         return rc;
2080 }
2081
2082 static void __exit fsfilt_ext3_exit(void)
2083 {
2084         int rc;
2085
2086         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2087         rc = cfs_mem_cache_destroy(fcb_cache);
2088         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2089 }
2090
2091 module_init(fsfilt_ext3_init);
2092 module_exit(fsfilt_ext3_exit);
2093
2094 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2095 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2096 MODULE_LICENSE("GPL");