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 = iattr->ia_size;
 512                 i_size_write(inode, iattr->ia_size);
 513
 514                 if (iattr->ia_valid & ATTR_UID)
 515                         inode->i_uid = iattr->ia_uid;
 516                 if (iattr->ia_valid & ATTR_GID)
 517                         inode->i_gid = iattr->ia_gid;
 518                 if (iattr->ia_valid & ATTR_ATIME)
 519                         inode->i_atime = iattr->ia_atime;
 520                 if (iattr->ia_valid & ATTR_MTIME)
 521                         inode->i_mtime = iattr->ia_mtime;
 522                 if (iattr->ia_valid & ATTR_CTIME)
 523                         inode->i_ctime = iattr->ia_ctime;
 524                 if (iattr->ia_valid & ATTR_MODE) {
 525                         inode->i_mode = iattr->ia_mode;
 526
 527                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 528                                 inode->i_mode &= ~S_ISGID;
 529                 }
 530
 531                 inode->i_sb->s_op->dirty_inode(inode);
 532
 533                 goto out;
 534         }
 535
 536         /* Don't allow setattr to change file type */
 537         if (iattr->ia_valid & ATTR_MODE)
 538                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 539                                  (iattr->ia_mode & ~S_IFMT);
 540
 541         /* We set these flags on the client, but have already checked perms
 542          * so don't confuse inode_change_ok. */
 543         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 544
 545         if (inode->i_op->setattr) {
 546                 rc = inode->i_op->setattr(dentry, iattr);
 547         } else {
 548                 rc = inode_change_ok(inode, iattr);
 549                 if (!rc)
 550                         rc = inode_setattr(inode, iattr);
 551         }
 552
 553  out:
 554         unlock_kernel();
 555         RETURN(rc);
 556 }
 557
 558 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 559                                  unsigned int cmd, unsigned long arg)
 560 {
 561         int rc = 0;
 562         ENTRY;
 563
 564         /* FIXME: Can't do this because of nested transaction deadlock */
 565         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 566                 CERROR("can't set data journal flag on file\n");
 567                 RETURN(-EPERM);
 568         }
 569
 570         if (inode->i_fop->ioctl)
 571                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 572         else
 573                 RETURN(-ENOTTY);
 574
 575         RETURN(rc);
 576 }
 577
 578 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 579                               void *lmm, int lmm_size, const char *name)
 580 {
 581         int rc;
 582
 583         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 584
 585         lock_24kernel();
 586         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 587                                    name, lmm, lmm_size, 0);
 588
 589         unlock_24kernel();
 590
 591         if (rc)
 592                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 593                        inode->i_ino, rc);
 594         return rc;
 595 }
 596
 597 /* Must be called with i_mutex held */
 598 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 599                               const char *name)
 600 {
 601         int rc;
 602
 603         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 604         lock_24kernel();
 605
 606         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 607                             name, lmm, lmm_size);
 608         unlock_24kernel();
 609
 610         /* This gives us the MD size */
 611         if (lmm == NULL)
 612                 return (rc == -ENODATA) ? 0 : rc;
 613
 614         if (rc < 0) {
 615                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 616                        EXT3_XATTR_INDEX_TRUSTED, name,
 617                        inode->i_ino, rc);
 618                 memset(lmm, 0, lmm_size);
 619                 return (rc == -ENODATA) ? 0 : rc;
 620         }
 621
 622         return rc;
 623 }
 624
 625 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 626 {
 627         submit_bio(rw, bio);
 628         return 0;
 629 }
 630
 631 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 632                                     loff_t *off)
 633 {
 634         struct inode *inode = file->f_dentry->d_inode;
 635         int rc = 0;
 636
 637         if (S_ISREG(inode->i_mode))
 638                 rc = file->f_op->read(file, buf, count, off);
 639         else {
 640                 const int blkbits = inode->i_sb->s_blocksize_bits;
 641                 const int blksize = inode->i_sb->s_blocksize;
 642
 643                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 644                        count, inode->i_ino, *off);
 645                 while (count > 0) {
 646                         struct buffer_head *bh;
 647
 648                         bh = NULL;
 649                         if (*off < i_size_read(inode)) {
 650                                 int err = 0;
 651
 652                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 653                                                 0, &err);
 654
 655                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 656
 657                                 if (bh) {
 658                                         memcpy(buf, bh->b_data, blksize);
 659                                         brelse(bh);
 660                                 } else if (err) {
 661                                         /* XXX in theory we should just fake
 662                                          * this buffer and continue like ext3,
 663                                          * especially if this is a partial read
 664                                          */
 665                                         CERROR("error read dir %lu+%llu: %d\n",
 666                                                inode->i_ino, *off, err);
 667                                         RETURN(err);
 668                                 }
 669                         }
 670                         if (!bh) {
 671                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 672
 673                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 674                                 memset(fake, 0, sizeof(*fake));
 675                                 fake->rec_len = cpu_to_le16(blksize);
 676                         }
 677                         count -= blksize;
 678                         buf += blksize;
 679                         *off += blksize;
 680                         rc += blksize;
 681                 }
 682         }
 683
 684         return rc;
 685 }
 686
 687 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 688 {
 689         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 690
 691         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 692
 693         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 694 }
 695
 696 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 697                                       void *handle, fsfilt_cb_t cb_func,
 698                                       void *cb_data)
 699 {
 700         struct fsfilt_cb_data *fcb;
 701
 702         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 703         if (fcb == NULL)
 704                 RETURN(-ENOMEM);
 705
 706         fcb->cb_func = cb_func;
 707         fcb->cb_obd = obd;
 708         fcb->cb_last_rcvd = last_rcvd;
 709         fcb->cb_data = cb_data;
 710
 711         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 712         lock_24kernel();
 713         journal_callback_set(handle, fsfilt_ext3_cb_func,
 714                              (struct journal_callback *)fcb);
 715         unlock_24kernel();
 716
 717         return 0;
 718 }
 719
 720 /*
 721  * We need to hack the return value for the free inode counts because
 722  * the current EA code requires one filesystem block per inode with EAs,
 723  * so it is possible to run out of blocks before we run out of inodes.
 724  *
 725  * This can be removed when the ext3 EA code is fixed.
 726  */
 727 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 728 {
 729         struct kstatfs sfs;
 730         int rc;
 731
 732         memset(&sfs, 0, sizeof(sfs));
 733
 734         rc = ll_do_statfs(sb, &sfs);
 735
 736         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 737                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 738                 sfs.f_ffree = sfs.f_bfree;
 739         }
 740
 741         statfs_pack(osfs, &sfs);
 742         return rc;
 743 }
 744
 745 static int fsfilt_ext3_sync(struct super_block *sb)
 746 {
 747         return ext3_force_commit(sb);
 748 }
 749
 750 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 751 #warning "kernel code has old extents/mballoc patch, disabling"
 752 #undef EXT3_MULTIBLOCK_ALLOCATOR
 753 #endif
 754 #ifndef EXT3_EXTENTS_FL
 755 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 756 #endif
 757
 758 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 759 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 760 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 761 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 762 #else
 763 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 764 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 765 #endif
 766
 767 #include <linux/lustre_version.h>
 768 #if EXT3_EXT_MAGIC == 0xf301
 769 #define ee_start e_start
 770 #define ee_block e_block
 771 #define ee_len   e_num
 772 #endif
 773 #ifndef EXT3_BB_MAX_BLOCKS
 774 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 775         ext3_new_blocks(handle, inode, count, goal, err)
 776 #endif
 777
 778 struct bpointers {
 779         unsigned long *blocks;
 780         int *created;
 781         unsigned long start;
 782         int num;
 783         int init_num;
 784         int create;
 785 };
 786
 787 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 788                               unsigned long block, int *aflags)
 789 {
 790         struct ext3_inode_info *ei = EXT3_I(inode);
 791         unsigned long bg_start;
 792         unsigned long colour;
 793         int depth;
 794
 795         if (path) {
 796                 struct ext3_extent *ex;
 797                 depth = path->p_depth;
 798
 799                 /* try to predict block placement */
 800                 if ((ex = path[depth].p_ext)) {
 801 #if 0
 802                         /* This prefers to eat into a contiguous extent
 803                          * rather than find an extent that the whole
 804                          * request will fit into.  This can fragment data
 805                          * block allocation and prevents our lovely 1M I/Os
 806                          * from reaching the disk intact. */
 807                         if (ex->ee_block + ex->ee_len == block)
 808                                 *aflags |= 1;
 809 #endif
 810                         return ex->ee_start + (block - ex->ee_block);
 811                 }
 812
 813                 /* it looks index is empty
 814                  * try to find starting from index itself */
 815                 if (path[depth].p_bh)
 816                         return path[depth].p_bh->b_blocknr;
 817         }
 818
 819         /* OK. use inode's group */
 820         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 821                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 822         colour = (current->pid % 16) *
 823                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 824         return bg_start + colour + block;
 825 }
 826
 827 #define ll_unmap_underlying_metadata(sb, blocknr) \
 828         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 829
 830 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 831 static unsigned long new_blocks(handle_t *handle, struct ext3_extents_tree *tree,
 832                                 struct ext3_ext_path *path, unsigned long block,
 833                                 int *count, int *err)
 834 {
 835         unsigned long pblock, goal;
 836         int aflags = 0;
 837
 838         goal = ext3_ext_find_goal(tree->inode, path, block, &aflags);
 839         aflags |= 2; /* block have been already reserved */
 840         lock_24kernel();
 841         pblock = ext3_mb_new_blocks(handle, tree->inode, goal, count, aflags, err);
 842         unlock_24kernel();
 843         return pblock;
 844
 845 }
 846 #else
 847 static unsigned long new_blocks(handle_t *handle, struct ext3_extents_tree *tree,
 848                                 struct ext3_ext_path *path, unsigned long block,
 849                                 int *count, int *err)
 850 {
 851         struct ext3_allocation_request ar;
 852         unsigned long pblock;
 853         int aflags;
 854
 855         /* find neighbour allocated blocks */
 856         ar.lleft = block;
 857         *err = ext3_ext_search_left(tree, path, &ar.lleft, &ar.pleft);
 858         if (*err)
 859                 return 0;
 860         ar.lright = block;
 861         *err = ext3_ext_search_right(tree, path, &ar.lright, &ar.pright);
 862         if (*err)
 863                 return 0;
 864
 865         /* allocate new block */
 866         ar.goal = ext3_ext_find_goal(tree->inode, path, block, &aflags);
 867         ar.inode = tree->inode;
 868         ar.logical = block;
 869         ar.len = *count;
 870         ar.flags = EXT3_MB_HINT_DATA;
 871         pblock = ext3_mb_new_blocks(handle, &ar, err);
 872         *count = ar.len;
 873         return pblock;
 874
 875 }
 876 #endif
 877
 878 static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
 879                                   struct ext3_ext_path *path,
 880                                   struct ext3_ext_cache *cex)
 881 {
 882         struct inode *inode = tree->inode;
 883         struct bpointers *bp = tree->private;
 884         struct ext3_extent nex;
 885         unsigned long pblock;
 886         unsigned long tgen;
 887         int count, err, i;
 888         handle_t *handle;
 889
 890         i = EXT_DEPTH(tree);
 891         EXT_ASSERT(i == path->p_depth);
 892         EXT_ASSERT(path[i].p_hdr);
 893
 894         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 895                 err = EXT_CONTINUE;
 896                 goto map;
 897         }
 898
 899         if (bp->create == 0) {
 900                 i = 0;
 901                 if (cex->ec_block < bp->start)
 902                         i = bp->start - cex->ec_block;
 903                 if (i >= cex->ec_len)
 904                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 905                                         i, cex->ec_len);
 906                 for (; i < cex->ec_len && bp->num; i++) {
 907                         *(bp->created) = 0;
 908                         bp->created++;
 909                         *(bp->blocks) = 0;
 910                         bp->blocks++;
 911                         bp->num--;
 912                         bp->start++;
 913                 }
 914
 915                 return EXT_CONTINUE;
 916         }
 917
 918         tgen = EXT_GENERATION(tree);
 919         count = ext3_ext_calc_credits_for_insert(tree, path);
 920         ext3_up_truncate_sem(inode);
 921
 922         lock_24kernel();
 923         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 924         unlock_24kernel();
 925         if (IS_ERR(handle)) {
 926                 ext3_down_truncate_sem(inode);
 927                 return PTR_ERR(handle);
 928         }
 929
 930         ext3_down_truncate_sem(inode);
 931         if (tgen != EXT_GENERATION(tree)) {
 932                 /* the tree has changed. so path can be invalid at moment */
 933                 lock_24kernel();
 934                 fsfilt_ext3_journal_stop(handle);
 935                 unlock_24kernel();
 936                 return EXT_REPEAT;
 937         }
 938
 939         count = cex->ec_len;
 940         pblock = new_blocks(handle, tree, path, cex->ec_block, &count, &err);
 941         if (!pblock)
 942                 goto out;
 943         EXT_ASSERT(count <= cex->ec_len);
 944
 945         /* insert new extent */
 946         nex.ee_block = cex->ec_block;
 947         nex.ee_start = pblock;
 948         nex.ee_len = count;
 949         err = ext3_ext_insert_extent(handle, tree, path, &nex);
 950         if (err) {
 951                 CERROR("can't insert extent: %d\n", err);
 952                 /* XXX: export ext3_free_blocks() */
 953                 /*ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);*/
 954                 goto out;
 955         }
 956
 957         /*
 958          * Putting len of the actual extent we just inserted,
 959          * we are asking ext3_ext_walk_space() to continue
 960          * scaning after that block
 961          */
 962         cex->ec_len = nex.ee_len;
 963         cex->ec_start = nex.ee_start;
 964         BUG_ON(nex.ee_len == 0);
 965         BUG_ON(nex.ee_block != cex->ec_block);
 966
 967 out:
 968         lock_24kernel();
 969         fsfilt_ext3_journal_stop(handle);
 970         unlock_24kernel();
 971 map:
 972         if (err >= 0) {
 973                 /* map blocks */
 974                 if (bp->num == 0) {
 975                         CERROR("hmm. why do we find this extent?\n");
 976                         CERROR("initial space: %lu:%u\n",
 977                                 bp->start, bp->init_num);
 978                         CERROR("current extent: %u/%u/%u %d\n",
 979                                 cex->ec_block, cex->ec_len,
 980                                 cex->ec_start, cex->ec_type);
 981                 }
 982                 i = 0;
 983                 if (cex->ec_block < bp->start)
 984                         i = bp->start - cex->ec_block;
 985                 if (i >= cex->ec_len)
 986                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 987                                         i, cex->ec_len);
 988                 for (; i < cex->ec_len && bp->num; i++) {
 989                         *(bp->blocks) = cex->ec_start + i;
 990                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 991                                 *(bp->created) = 0;
 992                         } else {
 993                                 *(bp->created) = 1;
 994                                 /* unmap any possible underlying metadata from
 995                                  * the block device mapping.  bug 6998. */
 996                                 ll_unmap_underlying_metadata(inode->i_sb,
 997                                                              *(bp->blocks));
 998                         }
 999                         bp->created++;
1000                         bp->blocks++;
1001                         bp->num--;
1002                         bp->start++;
1003                 }
1004         }
1005         return err;
1006 }
1007
1008 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1009                        unsigned long num, unsigned long *blocks,
1010                        int *created, int create)
1011 {
1012         struct ext3_extents_tree tree;
1013         struct bpointers bp;
1014         int err;
1015
1016         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1017                block, block + num - 1, (unsigned) inode->i_ino);
1018
1019         ext3_init_tree_desc(&tree, inode);
1020         tree.private = &bp;
1021         bp.blocks = blocks;
1022         bp.created = created;
1023         bp.start = block;
1024         bp.init_num = bp.num = num;
1025         bp.create = create;
1026
1027         ext3_down_truncate_sem(inode);
1028         err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
1029         ext3_ext_invalidate_cache(&tree);
1030         ext3_up_truncate_sem(inode);
1031
1032         return err;
1033 }
1034
1035 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1036                                     int pages, unsigned long *blocks,
1037                                     int *created, int create)
1038 {
1039         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1040         int rc = 0, i = 0;
1041         struct page *fp = NULL;
1042         int clen = 0;
1043
1044         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1045                 inode->i_ino, pages, (*page)->index);
1046
1047         /* pages are sorted already. so, we just have to find
1048          * contig. space and process them properly */
1049         while (i < pages) {
1050                 if (fp == NULL) {
1051                         /* start new extent */
1052                         fp = *page++;
1053                         clen = 1;
1054                         i++;
1055                         continue;
1056                 } else if (fp->index + clen == (*page)->index) {
1057                         /* continue the extent */
1058                         page++;
1059                         clen++;
1060                         i++;
1061                         continue;
1062                 }
1063
1064                 /* process found extent */
1065                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1066                                         clen * blocks_per_page, blocks,
1067                                         created, create);
1068                 if (rc)
1069                         GOTO(cleanup, rc);
1070
1071                 /* look for next extent */
1072                 fp = NULL;
1073                 blocks += blocks_per_page * clen;
1074                 created += blocks_per_page * clen;
1075         }
1076
1077         if (fp)
1078                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1079                                         clen * blocks_per_page, blocks,
1080                                         created, create);
1081 cleanup:
1082         return rc;
1083 }
1084 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1085
1086 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1087                                unsigned long *blocks, int *created, int create);
1088 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1089                                    int pages, unsigned long *blocks,
1090                                    int *created, int create)
1091 {
1092         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1093         unsigned long *b;
1094         int rc = 0, i, *cr;
1095
1096         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1097                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1098                 if (rc) {
1099                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1100                                inode->i_ino, *b, *cr, create, rc);
1101                         break;
1102                 }
1103
1104                 b += blocks_per_page;
1105                 cr += blocks_per_page;
1106         }
1107         return rc;
1108 }
1109
1110 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1111                                 int pages, unsigned long *blocks,
1112                                 int *created, int create,
1113                                 struct semaphore *optional_sem)
1114 {
1115         int rc;
1116 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1117         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1118                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1119                                                      blocks, created, create);
1120                 return rc;
1121         }
1122 #endif
1123         if (optional_sem != NULL)
1124                 down(optional_sem);
1125         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1126                                             created, create);
1127         if (optional_sem != NULL)
1128                 up(optional_sem);
1129
1130         return rc;
1131 }
1132
1133 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1134 {
1135         unsigned long block;
1136         struct buffer_head *bh;
1137         int err, blocksize, csize, boffs, osize = size;
1138
1139         /* prevent reading after eof */
1140         lock_kernel();
1141         if (i_size_read(inode) < *offs + size) {
1142                 size = i_size_read(inode) - *offs;
1143                 unlock_kernel();
1144                 if (size < 0) {
1145                         CERROR("size %llu is too short for read %u@%llu\n",
1146                                i_size_read(inode), size, *offs);
1147                         return -EIO;
1148                 } else if (size == 0) {
1149                         return 0;
1150                 }
1151         } else {
1152                 unlock_kernel();
1153         }
1154
1155         blocksize = 1 << inode->i_blkbits;
1156
1157         while (size > 0) {
1158                 block = *offs >> inode->i_blkbits;
1159                 boffs = *offs & (blocksize - 1);
1160                 csize = min(blocksize - boffs, size);
1161                 bh = ext3_bread(NULL, inode, block, 0, &err);
1162                 if (!bh) {
1163                         CERROR("can't read block: %d\n", err);
1164                         return err;
1165                 }
1166
1167                 memcpy(buf, bh->b_data + boffs, csize);
1168                 brelse(bh);
1169
1170                 *offs += csize;
1171                 buf += csize;
1172                 size -= csize;
1173         }
1174         return osize;
1175 }
1176 EXPORT_SYMBOL(fsfilt_ext3_read);
1177
1178 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1179                                    int size, loff_t *offs)
1180 {
1181         int rc;
1182         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1183         if (rc > 0)
1184                 rc = 0;
1185         return rc;
1186 }
1187
1188 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1189                                 loff_t *offs, handle_t *handle)
1190 {
1191         struct buffer_head *bh = NULL;
1192         loff_t old_size = i_size_read(inode), offset = *offs;
1193         loff_t new_size = i_size_read(inode);
1194         unsigned long block;
1195         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1196
1197         while (bufsize > 0) {
1198                 if (bh != NULL)
1199                         brelse(bh);
1200
1201                 block = offset >> inode->i_blkbits;
1202                 boffs = offset & (blocksize - 1);
1203                 size = min(blocksize - boffs, bufsize);
1204                 bh = ext3_bread(handle, inode, block, 1, &err);
1205                 if (!bh) {
1206                         CERROR("can't read/create block: %d\n", err);
1207                         break;
1208                 }
1209
1210                 err = ext3_journal_get_write_access(handle, bh);
1211                 if (err) {
1212                         CERROR("journal_get_write_access() returned error %d\n",
1213                                err);
1214                         break;
1215                 }
1216                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1217                 memcpy(bh->b_data + boffs, buf, size);
1218                 err = ext3_journal_dirty_metadata(handle, bh);
1219                 if (err) {
1220                         CERROR("journal_dirty_metadata() returned error %d\n",
1221                                err);
1222                         break;
1223                 }
1224                 if (offset + size > new_size)
1225                         new_size = offset + size;
1226                 offset += size;
1227                 bufsize -= size;
1228                 buf += size;
1229         }
1230         if (bh)
1231                 brelse(bh);
1232
1233         /* correct in-core and on-disk sizes */
1234         if (new_size > i_size_read(inode)) {
1235                 lock_kernel();
1236                 if (new_size > i_size_read(inode))
1237                         i_size_write(inode, new_size);
1238                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1239                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1240                 if (i_size_read(inode) > old_size)
1241                         mark_inode_dirty(inode);
1242                 unlock_kernel();
1243         }
1244
1245         if (err == 0)
1246                 *offs = offset;
1247         return err;
1248 }
1249 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1250
1251 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1252                                     loff_t *offs, int force_sync)
1253 {
1254         struct inode *inode = file->f_dentry->d_inode;
1255         handle_t *handle;
1256         int err, block_count = 0, blocksize;
1257
1258         /* Determine how many transaction credits are needed */
1259         blocksize = 1 << inode->i_blkbits;
1260         block_count = (*offs & (blocksize - 1)) + bufsize;
1261         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1262
1263         lock_24kernel();
1264         handle = fsfilt_ext3_journal_start(inode,
1265                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1266         unlock_24kernel();
1267         if (IS_ERR(handle)) {
1268                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1269                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1270                 return PTR_ERR(handle);
1271         }
1272
1273         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1274
1275         if (!err && force_sync)
1276                 handle->h_sync = 1; /* recovery likes this */
1277
1278         lock_24kernel();
1279         fsfilt_ext3_journal_stop(handle);
1280         unlock_24kernel();
1281
1282         return err;
1283 }
1284
1285 static int fsfilt_ext3_setup(struct super_block *sb)
1286 {
1287 #if 0
1288         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1289         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1290 #endif
1291 #ifdef S_PDIROPS
1292         CWARN("Enabling PDIROPS\n");
1293         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1294         sb->s_flags |= S_PDIROPS;
1295 #endif
1296         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1297                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1298 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13)) && HAVE_QUOTA_SUPPORT
1299         set_opt(EXT3_SB(sb)->s_mount_opt, QUOTA);
1300 #endif
1301         return 0;
1302 }
1303
1304 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1305    objects. Logs is number of logfiles to update */
1306 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1307 {
1308         if ( !fso ) {
1309                 switch(op) {
1310                 case FSFILT_OP_CREATE:
1311                                  /* directory leaf, index & indirect & EA*/
1312                         return 4 + 3 * logs;
1313                 case FSFILT_OP_UNLINK:
1314                         return 3 * logs;
1315                 }
1316         } else {
1317                 int i;
1318                 int needed = 0;
1319                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1320                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1321                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1322                 for (i = 0; i < op; i++, fso++) {
1323                         int nblocks = fso->fso_bufcnt * blockpp;
1324                         int ndindirect = min(nblocks, addrpp + 1);
1325                         int nindir = nblocks + ndindirect + 1;
1326
1327                         needed += nindir;
1328                 }
1329                 return needed + 3 * logs;
1330         }
1331
1332         return 0;
1333 }
1334
1335 static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
1336
1337 #define DQINFO_COPY(out, in)                    \
1338 do {                                            \
1339         Q_COPY(out, in, dqi_bgrace);            \
1340         Q_COPY(out, in, dqi_igrace);            \
1341         Q_COPY(out, in, dqi_flags);             \
1342         Q_COPY(out, in, dqi_valid);             \
1343 } while (0)
1344
1345 #define DQBLK_COPY(out, in)                     \
1346 do {                                            \
1347         Q_COPY(out, in, dqb_bhardlimit);        \
1348         Q_COPY(out, in, dqb_bsoftlimit);        \
1349         Q_COPY(out, in, dqb_curspace);          \
1350         Q_COPY(out, in, dqb_ihardlimit);        \
1351         Q_COPY(out, in, dqb_isoftlimit);        \
1352         Q_COPY(out, in, dqb_curinodes);         \
1353         Q_COPY(out, in, dqb_btime);             \
1354         Q_COPY(out, in, dqb_itime);             \
1355         Q_COPY(out, in, dqb_valid);             \
1356 } while (0)
1357
1358
1359
1360 static int fsfilt_ext3_quotactl(struct super_block *sb,
1361                                 struct obd_quotactl *oqc)
1362 {
1363         int i, rc = 0, error = 0;
1364         struct quotactl_ops *qcop;
1365         struct if_dqinfo *info;
1366         struct if_dqblk *dqblk;
1367         ENTRY;
1368
1369         if (!sb->s_qcop)
1370                 RETURN(-ENOSYS);
1371
1372         OBD_ALLOC_PTR(info);
1373         if (!info)
1374                 RETURN(-ENOMEM);
1375         OBD_ALLOC_PTR(dqblk);
1376         if (!dqblk) {
1377                 OBD_FREE_PTR(info);
1378                 RETURN(-ENOMEM);
1379         }
1380
1381         DQINFO_COPY(info, &oqc->qc_dqinfo);
1382         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1383
1384         qcop = sb->s_qcop;
1385         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1386                 for (i = 0; i < MAXQUOTAS; i++) {
1387                         if (!Q_TYPESET(oqc, i))
1388                                 continue;
1389
1390                         if (oqc->qc_cmd == Q_QUOTAON) {
1391                                 if (!qcop->quota_on)
1392                                         GOTO(out, rc = -ENOSYS);
1393                                 rc = qcop->quota_on(sb, i, oqc->qc_id,
1394                                                     (char *)op_quotafile[i]);
1395                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1396                                 if (!qcop->quota_off)
1397                                         GOTO(out, rc = -ENOSYS);
1398                                 rc = qcop->quota_off(sb, i);
1399                         }
1400
1401                         if (rc == -EBUSY)
1402                                 error = rc;
1403                         else if (rc)
1404                                 GOTO(out, rc);
1405                 }
1406                 GOTO(out, rc ?: error);
1407         }
1408
1409         switch (oqc->qc_cmd) {
1410         case Q_GETOINFO:
1411         case Q_GETINFO:
1412                 if (!qcop->get_info)
1413                         GOTO(out, rc = -ENOSYS);
1414                 rc = qcop->get_info(sb, oqc->qc_type, info);
1415                 break;
1416         case Q_SETQUOTA:
1417         case Q_INITQUOTA:
1418                 if (!qcop->set_dqblk)
1419                         GOTO(out, rc = -ENOSYS);
1420                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1421                 break;
1422         case Q_GETOQUOTA:
1423         case Q_GETQUOTA:
1424                 if (!qcop->get_dqblk)
1425                         GOTO(out, rc = -ENOSYS);
1426                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1427                 break;
1428         case Q_SYNC:
1429                 if (!sb->s_qcop->quota_sync)
1430                         GOTO(out, rc = -ENOSYS);
1431                 qcop->quota_sync(sb, oqc->qc_type);
1432                 break;
1433         default:
1434                 CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
1435                 LBUG();
1436         }
1437 out:
1438         DQINFO_COPY(&oqc->qc_dqinfo, info);
1439         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1440
1441         OBD_FREE_PTR(info);
1442         OBD_FREE_PTR(dqblk);
1443
1444         if (rc)
1445                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1446                                 "failed: %d\n",
1447                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1448         RETURN(rc);
1449 }
1450
1451 struct chk_dqblk{
1452         struct hlist_node       dqb_hash;        /* quotacheck hash */
1453         struct list_head        dqb_list;        /* in list also */
1454         qid_t                   dqb_id;          /* uid/gid */
1455         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1456         __u32                   dqb_bhardlimit;  /* block hard limit */
1457         __u32                   dqb_bsoftlimit;  /* block soft limit */
1458         qsize_t                 dqb_curspace;    /* current space */
1459         __u32                   dqb_ihardlimit;  /* inode hard limit */
1460         __u32                   dqb_isoftlimit;  /* inode soft limit */
1461         __u32                   dqb_curinodes;   /* current inodes */
1462         __u64                   dqb_btime;       /* block grace time */
1463         __u64                   dqb_itime;       /* inode grace time */
1464         __u32                   dqb_valid;       /* flag for above fields */
1465 };
1466
1467 static inline unsigned int chkquot_hash(qid_t id, int type)
1468                                         __attribute__((__const__));
1469
1470 static inline unsigned int chkquot_hash(qid_t id, int type)
1471 {
1472         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1473 }
1474
1475 static inline struct chk_dqblk *
1476 find_chkquot(struct hlist_head *head, qid_t id, int type)
1477 {
1478         struct hlist_node *node;
1479         struct chk_dqblk *cdqb;
1480
1481         hlist_for_each(node, head) {
1482                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1483                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1484                         return cdqb;
1485         }
1486
1487         return NULL;
1488 }
1489
1490 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1491 {
1492         struct chk_dqblk *cdqb;
1493
1494         OBD_ALLOC_PTR(cdqb);
1495         if (cdqb) {
1496                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1497                 INIT_LIST_HEAD(&cdqb->dqb_list);
1498                 cdqb->dqb_id = id;
1499                 cdqb->dqb_type = type;
1500         }
1501
1502         return cdqb;
1503 }
1504
1505 static struct chk_dqblk *
1506 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1507       qid_t id, int type, int first_check)
1508 {
1509         struct hlist_head *head = hash + chkquot_hash(id, type);
1510         struct if_dqblk dqb;
1511         struct chk_dqblk *cdqb;
1512         int rc;
1513
1514         cdqb = find_chkquot(head, id, type);
1515         if (cdqb)
1516                 return cdqb;
1517
1518         cdqb = alloc_chkquot(id, type);
1519         if (!cdqb)
1520                 return NULL;
1521
1522         if (!first_check) {
1523                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1524                 if (rc) {
1525                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1526                                id, type, rc);
1527                 } else {
1528                         DQBLK_COPY(cdqb, &dqb);
1529                         cdqb->dqb_curspace = 0;
1530                         cdqb->dqb_curinodes = 0;
1531                 }
1532         }
1533
1534         hlist_add_head(&cdqb->dqb_hash, head);
1535         list_add_tail(&cdqb->dqb_list, list);
1536
1537         return cdqb;
1538 }
1539
1540 static inline int quota_onoff(struct super_block *sb, int cmd, int type)
1541 {
1542         struct obd_quotactl *oqctl;
1543         int rc;
1544
1545         OBD_ALLOC_PTR(oqctl);
1546         if (!oqctl)
1547                 RETURN(-ENOMEM);
1548
1549         oqctl->qc_cmd = cmd;
1550         oqctl->qc_id = QFMT_LDISKFS;
1551         oqctl->qc_type = type;
1552         rc = fsfilt_ext3_quotactl(sb, oqctl);
1553
1554         OBD_FREE_PTR(oqctl);
1555         return rc;
1556 }
1557
1558 static inline int read_old_dqinfo(struct super_block *sb, int type,
1559                                   struct if_dqinfo *dqinfo)
1560 {
1561         struct obd_quotactl *oqctl;
1562         int rc;
1563         ENTRY;
1564
1565         OBD_ALLOC_PTR(oqctl);
1566         if (!oqctl)
1567                 RETURN(-ENOMEM);
1568
1569         oqctl->qc_cmd = Q_GETINFO;
1570         oqctl->qc_type = type;
1571         rc = fsfilt_ext3_quotactl(sb, oqctl);
1572         if (!rc)
1573                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1574
1575         OBD_FREE_PTR(oqctl);
1576         RETURN(rc);
1577 }
1578
1579 static inline struct ext3_group_desc *
1580 get_group_desc(struct super_block *sb, int group)
1581 {
1582         unsigned long desc_block, desc;
1583         struct ext3_group_desc *gdp;
1584
1585         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1586         desc = group % EXT3_DESC_PER_BLOCK(sb);
1587         gdp = (struct ext3_group_desc *)
1588               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1589
1590         return gdp + desc;
1591 }
1592
1593 static inline struct buffer_head *
1594 read_inode_bitmap(struct super_block *sb, unsigned long group)
1595 {
1596         struct ext3_group_desc *desc;
1597         struct buffer_head *bh;
1598
1599         desc = get_group_desc(sb, group);
1600         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1601
1602         return bh;
1603 }
1604
1605 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1606                                      struct buffer_head *bitmap_bh,
1607                                      int index, unsigned long ino)
1608 {
1609         struct inode *inode = NULL;
1610
1611         if (ext3_test_bit(index, bitmap_bh->b_data))
1612                 inode = iget(sb, ino);
1613
1614         return inode;
1615 }
1616
1617 struct qchk_ctxt {
1618         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1619         struct list_head        qckt_list;                   /* quotacheck list */
1620         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1621         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1622 };
1623
1624 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1625                            struct obd_quotactl *oqc)
1626 {
1627         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1628         loff_t size = 0;
1629         qid_t qid[MAXQUOTAS];
1630         int cnt, i, rc = 0;
1631
1632         if (!inode)
1633                 return 0;
1634
1635         qid[USRQUOTA] = inode->i_uid;
1636         qid[GRPQUOTA] = inode->i_gid;
1637
1638         if (S_ISDIR(inode->i_mode) ||
1639             S_ISREG(inode->i_mode) ||
1640             S_ISLNK(inode->i_mode))
1641                 size = inode_get_bytes(inode);
1642
1643         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1644                 if (!Q_TYPESET(oqc, cnt))
1645                         continue;
1646
1647                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1648                                 &qctxt->qckt_list, qid[cnt], cnt,
1649                                 qctxt->qckt_first_check[cnt]);
1650                 if (!cdqb[cnt]) {
1651                         rc = -ENOMEM;
1652                         break;
1653                 }
1654
1655                 cdqb[cnt]->dqb_curspace += size;
1656                 cdqb[cnt]->dqb_curinodes++;
1657         }
1658
1659         if (rc) {
1660                 for (i = 0; i < cnt; i++) {
1661                         if (!Q_TYPESET(oqc, i))
1662                                 continue;
1663                         LASSERT(cdqb[i]);
1664                         cdqb[i]->dqb_curspace -= size;
1665                         cdqb[i]->dqb_curinodes--;
1666                 }
1667         }
1668
1669         return rc;
1670 }
1671
1672 static int v2_write_dqheader(struct file *f, int type)
1673 {
1674         static const __u32 quota_magics[] = V2_INITQMAGICS;
1675         static const __u32 quota_versions[] = V2_INITQVERSIONS;
1676         struct v2_disk_dqheader dqhead;
1677         loff_t offset = 0;
1678
1679         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1680         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1681
1682         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1683         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1684
1685         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1686 }
1687
1688 /* write dqinfo struct in a new quota file */
1689 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1690 {
1691         struct v2_disk_dqinfo dqinfo;
1692         __u32 blocks = V2_DQTREEOFF + 1;
1693         loff_t offset = V2_DQINFOOFF;
1694
1695         if (info) {
1696                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1697                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1698                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1699                                                ~DQF_INFO_DIRTY);
1700         } else {
1701                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1702                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1703                 dqinfo.dqi_flags = 0;
1704         }
1705
1706         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1707         dqinfo.dqi_free_blk = 0;
1708         dqinfo.dqi_free_entry = 0;
1709
1710         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1711 }
1712
1713 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1714                                   struct obd_quotactl *oqc)
1715 {
1716         int i, rc = 0;
1717         ENTRY;
1718
1719         for (i = 0; i < MAXQUOTAS; i++) {
1720                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1721                                          NULL : &qctxt->qckt_dqinfo[i];
1722                 struct file *file;
1723
1724                 if (!Q_TYPESET(oqc, i))
1725                         continue;
1726
1727                 file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
1728                                  0644);
1729                 if (IS_ERR(file)) {
1730                         rc = PTR_ERR(file);
1731                         CERROR("can't create %s file: rc = %d\n",
1732                                op_quotafile[i], rc);
1733                         GOTO(out, rc);
1734                 }
1735
1736                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1737                         CERROR("file %s is not regular", op_quotafile[i]);
1738                         filp_close(file, 0);
1739                         GOTO(out, rc = -EINVAL);
1740                 }
1741
1742                 rc = v2_write_dqheader(file, i);
1743                 if (rc) {
1744                         filp_close(file, 0);
1745                         GOTO(out, rc);
1746                 }
1747
1748                 rc = v2_write_dqinfo(file, i, info);
1749                 filp_close(file, 0);
1750                 if (rc)
1751                         GOTO(out, rc);
1752         }
1753
1754 out:
1755         RETURN(rc);
1756 }
1757
1758
1759 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1760                           struct chk_dqblk *cdqb)
1761 {
1762         struct obd_quotactl *oqc;
1763         long now;
1764         int rc;
1765         ENTRY;
1766
1767         OBD_ALLOC_PTR(oqc);
1768         if (!oqc)
1769                 RETURN(-ENOMEM);
1770
1771         now = CURRENT_SECONDS;
1772
1773         if (cdqb->dqb_bsoftlimit &&
1774             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1775             !cdqb->dqb_btime)
1776                 cdqb->dqb_btime =
1777                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1778
1779         if (cdqb->dqb_isoftlimit &&
1780             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1781             !cdqb->dqb_itime)
1782                 cdqb->dqb_itime =
1783                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1784
1785         cdqb->dqb_valid = QIF_ALL;
1786
1787         oqc->qc_cmd = Q_SETQUOTA;
1788         oqc->qc_type = cdqb->dqb_type;
1789         oqc->qc_id = cdqb->dqb_id;
1790         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1791
1792         rc = fsfilt_ext3_quotactl(sb, oqc);
1793         OBD_FREE_PTR(oqc);
1794         RETURN(rc);
1795 }
1796
1797 static int prune_chkquots(struct super_block *sb,
1798                           struct qchk_ctxt *qctxt, int error)
1799 {
1800         struct chk_dqblk *cdqb, *tmp;
1801         int rc;
1802
1803         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1804                 if (!error) {
1805                         rc = commit_chkquot(sb, qctxt, cdqb);
1806                         if (rc)
1807                                 error = rc;
1808                 }
1809                 hlist_del_init(&cdqb->dqb_hash);
1810                 list_del(&cdqb->dqb_list);
1811                 OBD_FREE_PTR(cdqb);
1812         }
1813
1814         return error;
1815 }
1816
1817 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1818                                   struct obd_quotactl *oqc)
1819 {
1820         struct ext3_sb_info *sbi = EXT3_SB(sb);
1821         int i, group;
1822         struct qchk_ctxt *qctxt;
1823         struct buffer_head *bitmap_bh = NULL;
1824         unsigned long ino;
1825         struct inode *inode;
1826         int rc = 0;
1827         ENTRY;
1828
1829         /* turn on quota and read dqinfo if existed */
1830         OBD_ALLOC_PTR(qctxt);
1831         if (!qctxt) {
1832                 oqc->qc_stat = -ENOMEM;
1833                 RETURN(-ENOMEM);
1834         }
1835
1836         for (i = 0; i < NR_DQHASH; i++)
1837                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1838         INIT_LIST_HEAD(&qctxt->qckt_list);
1839
1840         for (i = 0; i < MAXQUOTAS; i++) {
1841                 if (!Q_TYPESET(oqc, i))
1842                         continue;
1843
1844                 rc = quota_onoff(sb, Q_QUOTAON, i);
1845                 if (!rc || rc == -EBUSY) {
1846                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1847                         if (rc)
1848                                 GOTO(out, rc);
1849                 } else if (rc == -ENOENT) {
1850                         qctxt->qckt_first_check[i] = 1;
1851                 } else if (rc) {
1852                         GOTO(out, rc);
1853                 }
1854         }
1855
1856         /* check quota and update in hash */
1857         for (group = 0; group < sbi->s_groups_count; group++) {
1858                 ino = group * sbi->s_inodes_per_group + 1;
1859                 bitmap_bh = read_inode_bitmap(sb, group);
1860                 if (!bitmap_bh) {
1861                         CERROR("read_inode_bitmap group %d failed", group);
1862                         GOTO(out, rc = -EIO);
1863                 }
1864
1865                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1866                         if (ino < sbi->s_first_ino)
1867                                 continue;
1868
1869                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1870                         rc = add_inode_quota(inode, qctxt, oqc);
1871                         iput(inode);
1872                         if (rc) {
1873                                 brelse(bitmap_bh);
1874                                 GOTO(out, rc);
1875                         }
1876                 }
1877
1878                 brelse(bitmap_bh);
1879         }
1880
1881         /* read old quota limits from old quota file. (only for the user
1882          * has limits but hasn't file) */
1883 #ifdef HAVE_QUOTA_SUPPORT
1884         for (i = 0; i < MAXQUOTAS; i++) {
1885                 struct list_head id_list;
1886                 struct dquot_id *dqid, *tmp;
1887
1888                 if (!Q_TYPESET(oqc, i))
1889                         continue;
1890
1891                 if (qctxt->qckt_first_check[i])
1892                         continue;
1893
1894
1895                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1896                 INIT_LIST_HEAD(&id_list);
1897 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1898                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1899 #else
1900                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1901 #endif
1902                 if (rc)
1903                         CERROR("read old limits failed. (rc:%d)\n", rc);
1904
1905                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1906                         list_del_init(&dqid->di_link);
1907
1908                         if (!rc)
1909                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1910                                       dqid->di_id, i,
1911                                       qctxt->qckt_first_check[i]);
1912                         kfree(dqid);
1913                 }
1914         }
1915 #endif
1916         /* turn off quota cause we are to dump chk_dqblk to files */
1917         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1918
1919         rc = create_new_quota_files(qctxt, oqc);
1920         if (rc)
1921                 GOTO(out, rc);
1922
1923         /* we use vfs functions to set dqblk, so turn quota on */
1924         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
1925 out:
1926         /* dump and free chk_dqblk */
1927         rc = prune_chkquots(sb, qctxt, rc);
1928         OBD_FREE_PTR(qctxt);
1929
1930         /* turn off quota, `lfs quotacheck` will turn on when all
1931          * nodes quotacheck finish. */
1932         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1933
1934         oqc->qc_stat = rc;
1935         if (rc)
1936                 CERROR("quotacheck failed: rc = %d\n", rc);
1937
1938         RETURN(rc);
1939 }
1940
1941 #ifdef HAVE_QUOTA_SUPPORT
1942 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
1943                                  int cmd)
1944 {
1945         int rc = 0;
1946         ENTRY;
1947
1948         if (lqi->qi_files[type] == NULL) {
1949                 CERROR("operate qinfo before it's enabled!\n");
1950                 RETURN(-EIO);
1951         }
1952
1953         switch (cmd) {
1954         case QFILE_CHK:
1955                 rc = lustre_check_quota_file(lqi, type);
1956                 break;
1957         case QFILE_RD_INFO:
1958                 rc = lustre_read_quota_info(lqi, type);
1959                 break;
1960         case QFILE_WR_INFO:
1961                 rc = lustre_write_quota_info(lqi, type);
1962                 break;
1963         case QFILE_INIT_INFO:
1964                 rc = lustre_init_quota_info(lqi, type);
1965                 break;
1966         default:
1967                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1968                 LBUG();
1969                 break;
1970         }
1971         RETURN(rc);
1972 }
1973
1974 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
1975                             struct list_head *list)
1976 {
1977         return lustre_get_qids(file, inode, type, list);
1978 }
1979
1980 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
1981 {
1982         int rc = 0;
1983         ENTRY;
1984
1985         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
1986                 CERROR("operate dquot before it's enabled!\n");
1987                 RETURN(-EIO);
1988         }
1989
1990         switch (cmd) {
1991         case QFILE_RD_DQUOT:
1992                 rc = lustre_read_dquot(dquot);
1993                 break;
1994         case QFILE_WR_DQUOT:
1995                 if (dquot->dq_dqb.dqb_ihardlimit ||
1996                     dquot->dq_dqb.dqb_isoftlimit ||
1997                     dquot->dq_dqb.dqb_bhardlimit ||
1998                     dquot->dq_dqb.dqb_bsoftlimit)
1999                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2000                 else
2001                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2002
2003                 rc = lustre_commit_dquot(dquot);
2004                 if (rc >= 0)
2005                         rc = 0;
2006                 break;
2007         default:
2008                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2009                 LBUG();
2010                 break;
2011         }
2012         RETURN(rc);
2013 }
2014 #endif
2015
2016 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2017 {
2018         return (EXT3_SB(sb)->journal_bdev);
2019 }
2020 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2021
2022 static struct fsfilt_operations fsfilt_ext3_ops = {
2023         .fs_type                = "ext3",
2024         .fs_owner               = THIS_MODULE,
2025         .fs_getlabel            = fsfilt_ext3_get_label,
2026         .fs_setlabel            = fsfilt_ext3_set_label,
2027         .fs_uuid                = fsfilt_ext3_uuid,
2028         .fs_start               = fsfilt_ext3_start,
2029         .fs_brw_start           = fsfilt_ext3_brw_start,
2030         .fs_extend              = fsfilt_ext3_extend,
2031         .fs_commit              = fsfilt_ext3_commit,
2032         .fs_commit_async        = fsfilt_ext3_commit_async,
2033         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2034         .fs_setattr             = fsfilt_ext3_setattr,
2035         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2036         .fs_set_md              = fsfilt_ext3_set_md,
2037         .fs_get_md              = fsfilt_ext3_get_md,
2038         .fs_readpage            = fsfilt_ext3_readpage,
2039         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2040         .fs_statfs              = fsfilt_ext3_statfs,
2041         .fs_sync                = fsfilt_ext3_sync,
2042         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2043         .fs_write_record        = fsfilt_ext3_write_record,
2044         .fs_read_record         = fsfilt_ext3_read_record,
2045         .fs_setup               = fsfilt_ext3_setup,
2046         .fs_send_bio            = fsfilt_ext3_send_bio,
2047         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2048         .fs_quotactl            = fsfilt_ext3_quotactl,
2049         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2050 #ifdef HAVE_DISK_INODE_VERSION
2051         .fs_get_version         = fsfilt_ext3_get_version,
2052         .fs_set_version         = fsfilt_ext3_set_version,
2053 #endif
2054 #ifdef HAVE_QUOTA_SUPPORT
2055         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2056         .fs_qids                = fsfilt_ext3_qids,
2057         .fs_dquot               = fsfilt_ext3_dquot,
2058 #endif
2059         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2060 };
2061
2062 static int __init fsfilt_ext3_init(void)
2063 {
2064         int rc;
2065
2066         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2067                                       sizeof(struct fsfilt_cb_data), 0, 0);
2068         if (!fcb_cache) {
2069                 CERROR("error allocating fsfilt journal callback cache\n");
2070                 GOTO(out, rc = -ENOMEM);
2071         }
2072
2073         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2074
2075         if (rc) {
2076                 int err = cfs_mem_cache_destroy(fcb_cache);
2077                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2078         }
2079 out:
2080         return rc;
2081 }
2082
2083 static void __exit fsfilt_ext3_exit(void)
2084 {
2085         int rc;
2086
2087         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2088         rc = cfs_mem_cache_destroy(fcb_cache);
2089         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2090 }
2091
2092 module_init(fsfilt_ext3_init);
2093 module_exit(fsfilt_ext3_exit);
2094
2095 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2096 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2097 MODULE_LICENSE("GPL");