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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/fsfilt_ext3.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/init.h>
  44 #include <linux/module.h>
  45 #include <linux/fs.h>
  46 #include <linux/slab.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/quotaops.h>
  49 #ifdef HAVE_EXT4_LDISKFS
  50 #include <ext4/ext4.h>
  51 #include <ext4/ext4_jbd2.h>
  52 #else
  53 #include <linux/jbd.h>
  54 #include <linux/ext3_fs.h>
  55 #include <linux/ext3_jbd.h>
  56 #endif
  57 #include <linux/version.h>
  58 #include <linux/bitops.h>
  59 #include <linux/quota.h>
  60 #ifdef HAVE_QUOTAIO_V1_H
  61 # include <linux/quotaio_v1.h>
  62 # include <linux/quotaio_v2.h>
  63 #else
  64 # include <quotaio_v1.h>
  65 # include <quotaio_v2.h>
  66 # include <quota_tree.h>
  67 # define V2_DQTREEOFF    QT_TREEOFF
  68 #endif
  69
  70 #if defined(HAVE_EXT3_XATTR_H)
  71 #include <ext3/xattr.h>
  72 #else
  73 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  74 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  75 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  76 #endif
  77
  78 #include <libcfs/libcfs.h>
  79 #include <lustre_fsfilt.h>
  80 #include <obd.h>
  81 #include <lustre_quota.h>
  82 #include <linux/lustre_compat25.h>
  83 #include <linux/lprocfs_status.h>
  84
  85 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  86 #ifdef HAVE_EXT4_LDISKFS
  87 #include <ext4/ext4_extents.h>
  88 #else
  89 #include <linux/ext3_extents.h>
  90 #endif
  91 #endif
  92
  93 #include "lustre_quota_fmt.h"
  94
  95 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  96 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  97 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  98 #else
  99 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 100 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 101 #endif
 102
 103 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 104 /* for kernels 2.6.18 and later */
 105 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 106 #else
 107 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 108 #endif
 109
 110 static cfs_mem_cache_t *fcb_cache;
 111
 112 struct fsfilt_cb_data {
 113         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 114         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 115         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 116         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 117         void *cb_data;                  /* MDS/OST completion function data */
 118 };
 119
 120 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 121 #define EXT3_XATTR_INDEX_TRUSTED        4
 122 #endif
 123
 124 #ifdef HAVE_EXT4_LDISKFS
 125 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 126 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 127 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 128 #else
 129 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 130 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 131 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 132 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 133 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 134 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 135 #endif
 136
 137 #ifndef ext3_find_next_bit
 138 #define ext3_find_next_bit           ext2_find_next_bit
 139 #endif
 140
 141 #ifndef ext2_find_next_bit
 142 #ifdef __LITTLE_ENDIAN
 143 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 144 #else
 145 error "Need implementation of find_next_bit on big-endian systems"
 146 #endif  /* __LITTLE_ENDIAN */
 147 #endif  /* !ext2_find_next_le_bit */
 148
 149 static char *fsfilt_ext3_get_label(struct super_block *sb)
 150 {
 151         return EXT3_SB(sb)->s_es->s_volume_name;
 152 }
 153
 154 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 155 {
 156         /* see e.g. fsfilt_ext3_write_record() */
 157         journal_t *journal;
 158         handle_t *handle;
 159         int err;
 160
 161         journal = EXT3_SB(sb)->s_journal;
 162         handle = ext3_journal_start_sb(sb, 1);
 163         if (IS_ERR(handle)) {
 164                 CERROR("can't start transaction\n");
 165                 return(PTR_ERR(handle));
 166         }
 167
 168         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 169         if (err)
 170                 goto out;
 171
 172         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 173                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 174
 175         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 176
 177 out:
 178         ext3_journal_stop(handle);
 179
 180         return(err);
 181 }
 182
 183 static char *fsfilt_ext3_uuid(struct super_block *sb)
 184 {
 185         return EXT3_SB(sb)->s_es->s_uuid;
 186 }
 187
 188 #ifdef HAVE_DISK_INODE_VERSION
 189
 190 static __u64 get_i_version(struct inode *inode)
 191 {
 192 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 193         return inode->i_version;
 194 #else
 195         return EXT3_I(inode)->i_fs_version;
 196 #endif
 197 }
 198
 199 static void set_i_version(struct inode *inode, __u64 new_version)
 200 {
 201 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 202         inode->i_version = new_version;
 203 #else
 204         (EXT3_I(inode))->i_fs_version = new_version;
 205 #endif
 206 }
 207
 208 /*
 209  * Get the 64-bit version for an inode.
 210  */
 211 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 212 {
 213         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 214                get_i_version(inode), inode->i_ino);
 215         return get_i_version(inode);
 216 }
 217
 218 /*
 219  * Set the 64-bit version and return the old version.
 220  */
 221 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 222 {
 223         __u64 old_version = get_i_version(inode);
 224
 225         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 226                new_version, old_version, inode->i_ino);
 227         set_i_version(inode, new_version);
 228         /* version is set after all inode operations are finished, so we should
 229          * mark it dirty here */
 230         inode->i_sb->s_op->dirty_inode(inode);
 231         return old_version;
 232 }
 233
 234 #endif
 235
 236 /*
 237  * We don't currently need any additional blocks for rmdir and
 238  * unlink transactions because we are storing the OST oa_id inside
 239  * the inode (which we will be changing anyways as part of this
 240  * transaction).
 241  */
 242 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 243                                int logs)
 244 {
 245         /* For updates to the last received file */
 246         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 247         journal_t *journal;
 248         void *handle;
 249
 250         if (current->journal_info) {
 251                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 252                        current->journal_info);
 253                 goto journal_start;
 254         }
 255
 256         switch(op) {
 257         case FSFILT_OP_RMDIR:
 258         case FSFILT_OP_UNLINK:
 259                 /* delete one file + create/update logs for each stripe */
 260                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 261                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 262                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 263                 break;
 264         case FSFILT_OP_RENAME:
 265                 /* modify additional directory */
 266                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 267                 /* no break */
 268         case FSFILT_OP_SYMLINK:
 269                 /* additional block + block bitmap + GDT for long symlink */
 270                 nblocks += 3;
 271                 /* no break */
 272         case FSFILT_OP_CREATE: {
 273 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 274                 static int warned;
 275                 if (!warned) {
 276                         if (!test_opt(inode->i_sb, EXTENTS)) {
 277                                 warned = 1;
 278                         } else if (((EXT3_I(inode)->i_flags &
 279                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 280                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 281                                 CWARN("extent-mapped directory found with "
 282                                       "ext3-based ldiskfs - contact "
 283                                       "http://bugzilla.lustre.org/\n");
 284                                 warned = 1;
 285                         }
 286                 }
 287 #endif
 288                 /* no break */
 289         }
 290         case FSFILT_OP_MKDIR:
 291         case FSFILT_OP_MKNOD:
 292                 /* modify one inode + block bitmap + GDT */
 293                 nblocks += 3;
 294                 /* no break */
 295         case FSFILT_OP_LINK:
 296                 /* modify parent directory */
 297                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 298                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 299                 /* create/update logs for each stripe */
 300                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 301                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 302                 break;
 303         case FSFILT_OP_SETATTR:
 304                 /* Setattr on inode */
 305                 nblocks += 1;
 306                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 307                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 308                 /* quota chown log for each stripe */
 309                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 310                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 311                 break;
 312         case FSFILT_OP_CANCEL_UNLINK:
 313                 /* blocks for log header bitmap update OR
 314                  * blocks for catalog header bitmap update + unlink of logs */
 315                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 316                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 317                 break;
 318         default: CERROR("unknown transaction start op %d\n", op);
 319                 LBUG();
 320         }
 321
 322         LASSERT(current->journal_info == desc_private);
 323         journal = EXT3_SB(inode->i_sb)->s_journal;
 324         if (nblocks > journal->j_max_transaction_buffers) {
 325                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 326                        nblocks, op, logs, journal->j_max_transaction_buffers);
 327                 nblocks = journal->j_max_transaction_buffers;
 328         }
 329
 330  journal_start:
 331         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 332         handle = ext3_journal_start(inode, nblocks);
 333
 334         if (!IS_ERR(handle))
 335                 LASSERT(current->journal_info == handle);
 336         else
 337                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 338                        op, nblocks, PTR_ERR(handle));
 339         return handle;
 340 }
 341
 342 /*
 343  * Calculate the number of buffer credits needed to write multiple pages in
 344  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 345  * doesn't have a nice API for calculating this sort of thing in advance.
 346  *
 347  * See comment above ext3_writepage_trans_blocks for details.  We assume
 348  * no data journaling is being done, but it does allow for all of the pages
 349  * being non-contiguous.  If we are guaranteed contiguous pages we could
 350  * reduce the number of (d)indirect blocks a lot.
 351  *
 352  * With N blocks per page and P pages, for each inode we have at most:
 353  * N*P indirect
 354  * min(N*P, blocksize/4 + 1) dindirect blocks
 355  * niocount tindirect
 356  *
 357  * For the entire filesystem, we have at most:
 358  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 359  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 360  * objcount inode blocks
 361  * 1 superblock
 362  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 363  *
 364  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 365  */
 366 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 367                                       int niocount, struct niobuf_local *nb)
 368 {
 369         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 370         __u64 next_indir;
 371         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 372         int nbitmaps = 0, ngdblocks;
 373         int needed = objcount + 1; /* inodes + superblock */
 374         int i, j;
 375
 376         for (i = 0, j = 0; i < objcount; i++, fso++) {
 377                 /* two or more dindirect blocks in case we cross boundary */
 378                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 379                                     nb[j].offset) >>
 380                                    sb->s_blocksize_bits) /
 381                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 382                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 383
 384                 /* leaf, indirect, tindirect blocks for first block */
 385                 nbitmaps += blockpp + 2;
 386
 387                 j += fso->fso_bufcnt;
 388         }
 389
 390         next_indir = nb[0].offset +
 391                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 392         for (i = 1; i < niocount; i++) {
 393                 if (nb[i].offset >= next_indir) {
 394                         nbitmaps++;     /* additional indirect */
 395                         next_indir = nb[i].offset +
 396                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 397                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 398                         nbitmaps++;     /* additional indirect */
 399                 }
 400                 nbitmaps += blockpp;    /* each leaf in different group? */
 401         }
 402
 403         ngdblocks = nbitmaps;
 404         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 405                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 406         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 407                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 408
 409         needed += nbitmaps + ngdblocks;
 410
 411         /* last_rcvd update */
 412         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 413
 414 #if defined(CONFIG_QUOTA)
 415         /* We assume that there will be 1 bit set in s_dquot.flags for each
 416          * quota file that is active.  This is at least true for now.
 417          */
 418         needed += hweight32(ll_sb_any_quota_active(sb)) *
 419                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 420 #endif
 421
 422         return needed;
 423 }
 424
 425 /* We have to start a huge journal transaction here to hold all of the
 426  * metadata for the pages being written here.  This is necessitated by
 427  * the fact that we do lots of prepare_write operations before we do
 428  * any of the matching commit_write operations, so even if we split
 429  * up to use "smaller" transactions none of them could complete until
 430  * all of them were opened.  By having a single journal transaction,
 431  * we eliminate duplicate reservations for common blocks like the
 432  * superblock and group descriptors or bitmaps.
 433  *
 434  * We will start the transaction here, but each prepare_write will
 435  * add a refcount to the transaction, and each commit_write will
 436  * remove a refcount.  The transaction will be closed when all of
 437  * the pages have been written.
 438  */
 439 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 440                                    int niocount, struct niobuf_local *nb,
 441                                    void *desc_private, int logs)
 442 {
 443         journal_t *journal;
 444         handle_t *handle;
 445         int needed;
 446         ENTRY;
 447
 448         LASSERT(current->journal_info == desc_private);
 449         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 450         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 451
 452         /* The number of blocks we could _possibly_ dirty can very large.
 453          * We reduce our request if it is absurd (and we couldn't get that
 454          * many credits for a single handle anyways).
 455          *
 456          * At some point we have to limit the size of I/Os sent at one time,
 457          * increase the size of the journal, or we have to calculate the
 458          * actual journal requirements more carefully by checking all of
 459          * the blocks instead of being maximally pessimistic.  It remains to
 460          * be seen if this is a real problem or not.
 461          */
 462         if (needed > journal->j_max_transaction_buffers) {
 463                 CERROR("want too many journal credits (%d) using %d instead\n",
 464                        needed, journal->j_max_transaction_buffers);
 465                 needed = journal->j_max_transaction_buffers;
 466         }
 467
 468         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 469         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 470         if (IS_ERR(handle)) {
 471                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 472                        PTR_ERR(handle));
 473         } else {
 474                 LASSERT(handle->h_buffer_credits >= needed);
 475                 LASSERT(current->journal_info == handle);
 476         }
 477
 478         RETURN(handle);
 479 }
 480
 481 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 482 {
 483        handle_t *handle = h;
 484
 485        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 486        if (nblocks == 0) {
 487                handle->h_buffer_credits = 0;
 488                CWARN("setting credits of handle %p to zero by request\n", h);
 489        }
 490
 491        if (handle->h_buffer_credits > nblocks)
 492                 return 0;
 493        if (ext3_journal_extend(handle, nblocks) == 0)
 494                 return 0;
 495
 496        ext3_mark_inode_dirty(handle, inode);
 497        return ext3_journal_restart(handle, nblocks);
 498 }
 499
 500 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 501 {
 502         int rc;
 503         handle_t *handle = h;
 504
 505         LASSERT(current->journal_info == handle);
 506         if (force_sync)
 507                 handle->h_sync = 1; /* recovery likes this */
 508
 509         rc = ext3_journal_stop(handle);
 510
 511         return rc;
 512 }
 513
 514 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 515                                     void **wait_handle)
 516 {
 517         unsigned long tid;
 518         transaction_t *transaction;
 519         handle_t *handle = h;
 520         journal_t *journal;
 521         int rc;
 522
 523         LASSERT(current->journal_info == handle);
 524
 525         transaction = handle->h_transaction;
 526         journal = transaction->t_journal;
 527         tid = transaction->t_tid;
 528         /* we don't want to be blocked */
 529         handle->h_sync = 0;
 530         rc = ext3_journal_stop(handle);
 531         if (rc) {
 532                 CERROR("error while stopping transaction: %d\n", rc);
 533                 return rc;
 534         }
 535         fsfilt_log_start_commit(journal, tid);
 536
 537         *wait_handle = (void *) tid;
 538         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 539         return 0;
 540 }
 541
 542 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 543 {
 544         journal_t *journal = EXT3_JOURNAL(inode);
 545         tid_t tid = (tid_t)(long)h;
 546
 547         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 548         if (unlikely(is_journal_aborted(journal)))
 549                 return -EIO;
 550
 551         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 552
 553         if (unlikely(is_journal_aborted(journal)))
 554                 return -EIO;
 555         return 0;
 556 }
 557
 558 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 559                                struct iattr *iattr, int do_trunc)
 560 {
 561         struct inode *inode = dentry->d_inode;
 562         int rc = 0;
 563
 564         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 565          * We are already writing the inode to disk as part of this
 566          * transaction and want to avoid a lot of extra inode writeout
 567          * later on. b=9828 */
 568         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 569                 /* ATTR_SIZE would invoke truncate: clear it */
 570                 iattr->ia_valid &= ~ATTR_SIZE;
 571                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 572                 i_size_write(inode, iattr->ia_size);
 573
 574                 if (iattr->ia_valid & ATTR_UID)
 575                         inode->i_uid = iattr->ia_uid;
 576                 if (iattr->ia_valid & ATTR_GID)
 577                         inode->i_gid = iattr->ia_gid;
 578                 if (iattr->ia_valid & ATTR_ATIME)
 579                         inode->i_atime = iattr->ia_atime;
 580                 if (iattr->ia_valid & ATTR_MTIME)
 581                         inode->i_mtime = iattr->ia_mtime;
 582                 if (iattr->ia_valid & ATTR_CTIME)
 583                         inode->i_ctime = iattr->ia_ctime;
 584                 if (iattr->ia_valid & ATTR_MODE) {
 585                         inode->i_mode = iattr->ia_mode;
 586
 587                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 588                             !cfs_capable(CFS_CAP_FSETID))
 589                                 inode->i_mode &= ~S_ISGID;
 590                 }
 591
 592                 inode->i_sb->s_op->dirty_inode(inode);
 593
 594                 goto out;
 595         }
 596
 597         /* Don't allow setattr to change file type */
 598         if (iattr->ia_valid & ATTR_MODE)
 599                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 600                                  (iattr->ia_mode & ~S_IFMT);
 601
 602         /* We set these flags on the client, but have already checked perms
 603          * so don't confuse inode_change_ok. */
 604         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 605
 606         if (inode->i_op->setattr) {
 607                 rc = inode->i_op->setattr(dentry, iattr);
 608         } else {
 609                 rc = inode_change_ok(inode, iattr);
 610                 if (!rc)
 611                         rc = inode_setattr(inode, iattr);
 612         }
 613
 614  out:
 615         RETURN(rc);
 616 }
 617
 618 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 619                                  unsigned int cmd, unsigned long arg)
 620 {
 621         int rc = 0;
 622         ENTRY;
 623
 624         /* FIXME: Can't do this because of nested transaction deadlock */
 625         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 626                 CERROR("can't set data journal flag on file\n");
 627                 RETURN(-EPERM);
 628         }
 629
 630 #ifdef HAVE_EXT4_LDISKFS
 631         /* ext4_ioctl does not have a inode argument */
 632         if (inode->i_fop->unlocked_ioctl)
 633                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 634 #else
 635         if (inode->i_fop->ioctl)
 636                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 637 #endif
 638         else
 639                 RETURN(-ENOTTY);
 640
 641         RETURN(rc);
 642 }
 643
 644 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 645                               void *lmm, int lmm_size, const char *name)
 646 {
 647         int rc;
 648
 649         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 650
 651         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 652                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 653
 654
 655         if (rc && rc != -EROFS)
 656                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 657                        inode->i_ino, rc);
 658         return rc;
 659 }
 660
 661 /* Must be called with i_mutex held */
 662 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 663                               const char *name)
 664 {
 665         int rc;
 666
 667         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 668
 669         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 670                             name, lmm, lmm_size);
 671
 672         /* This gives us the MD size */
 673         if (lmm == NULL)
 674                 return (rc == -ENODATA) ? 0 : rc;
 675
 676         if (rc < 0) {
 677                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 678                        EXT3_XATTR_INDEX_TRUSTED, name,
 679                        inode->i_ino, rc);
 680                 memset(lmm, 0, lmm_size);
 681                 return (rc == -ENODATA) ? 0 : rc;
 682         }
 683
 684         return rc;
 685 }
 686
 687 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 688 {
 689         submit_bio(rw, bio);
 690         return 0;
 691 }
 692
 693 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 694                                     loff_t *off)
 695 {
 696         struct inode *inode = file->f_dentry->d_inode;
 697         int rc = 0;
 698
 699         if (S_ISREG(inode->i_mode))
 700                 rc = file->f_op->read(file, buf, count, off);
 701         else {
 702                 const int blkbits = inode->i_sb->s_blocksize_bits;
 703                 const int blksize = inode->i_sb->s_blocksize;
 704
 705                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 706                        (unsigned long)count, inode->i_ino, *off);
 707                 while (count > 0) {
 708                         struct buffer_head *bh;
 709
 710                         bh = NULL;
 711                         if (*off < i_size_read(inode)) {
 712                                 int err = 0;
 713
 714                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 715                                                 0, &err);
 716
 717                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 718
 719                                 if (bh) {
 720                                         memcpy(buf, bh->b_data, blksize);
 721                                         brelse(bh);
 722                                 } else if (err) {
 723                                         /* XXX in theory we should just fake
 724                                          * this buffer and continue like ext3,
 725                                          * especially if this is a partial read
 726                                          */
 727                                         CERROR("error read dir %lu+%llu: %d\n",
 728                                                inode->i_ino, *off, err);
 729                                         RETURN(err);
 730                                 }
 731                         }
 732                         if (!bh) {
 733                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 734
 735                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 736                                 memset(fake, 0, sizeof(*fake));
 737                                 fake->rec_len = cpu_to_le16(blksize);
 738                         }
 739                         count -= blksize;
 740                         buf += blksize;
 741                         *off += blksize;
 742                         rc += blksize;
 743                 }
 744         }
 745
 746         return rc;
 747 }
 748
 749 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 750 {
 751         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 752
 753         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 754
 755         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 756 }
 757
 758 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 759                                       void *handle, fsfilt_cb_t cb_func,
 760                                       void *cb_data)
 761 {
 762         struct fsfilt_cb_data *fcb;
 763
 764         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 765         if (fcb == NULL)
 766                 RETURN(-ENOMEM);
 767
 768         fcb->cb_func = cb_func;
 769         fcb->cb_obd = obd;
 770         fcb->cb_last_rcvd = last_rcvd;
 771         fcb->cb_data = cb_data;
 772
 773         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 774         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 775                                     (struct journal_callback *)fcb);
 776
 777         return 0;
 778 }
 779
 780 /*
 781  * We need to hack the return value for the free inode counts because
 782  * the current EA code requires one filesystem block per inode with EAs,
 783  * so it is possible to run out of blocks before we run out of inodes.
 784  *
 785  * This can be removed when the ext3 EA code is fixed.
 786  */
 787 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 788 {
 789         struct kstatfs sfs;
 790         int rc;
 791
 792         memset(&sfs, 0, sizeof(sfs));
 793         rc = ll_do_statfs(sb, &sfs);
 794         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 795                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 796                 sfs.f_ffree = sfs.f_bfree;
 797         }
 798
 799         statfs_pack(osfs, &sfs);
 800         return rc;
 801 }
 802
 803 static int fsfilt_ext3_sync(struct super_block *sb)
 804 {
 805         return ext3_force_commit(sb);
 806 }
 807
 808 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 809 #warning "kernel code has old extents/mballoc patch, disabling"
 810 #undef EXT3_MULTIBLOCK_ALLOCATOR
 811 #endif
 812 #ifndef EXT3_EXTENTS_FL
 813 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 814 #endif
 815
 816 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 817 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 818 #define fsfilt_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 819 #define fsfilt_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 820 #else
 821 #ifdef HAVE_EXT4_LDISKFS
 822 #define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem));
 823 #define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem));
 824 #else
 825 #define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 826 #define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 827 #endif
 828 #endif
 829
 830 #ifndef EXT_ASSERT
 831 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 832 #endif
 833
 834 #ifdef EXT3_EXT_HAS_NO_TREE
 835 /* for kernels 2.6.18 and later */
 836 #ifdef HAVE_EXT4_LDISKFS
 837 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 838 #else
 839 #define EXT_GENERATION(inode)           ext_generation(inode)
 840 #endif
 841 #define ext3_ext_base                   inode
 842 #define ext3_ext_base2inode(inode)      (inode)
 843 #define EXT_DEPTH(inode)                ext_depth(inode)
 844 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 845                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 846 #else
 847 #define ext3_ext_base                   ext3_extents_tree
 848 #define ext3_ext_base2inode(tree)       (tree->inode)
 849 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 850                         ext3_ext_walk_space(tree, block, num, cb);
 851 #endif
 852
 853 #include <linux/lustre_version.h>
 854
 855 struct bpointers {
 856         unsigned long *blocks;
 857         int *created;
 858         unsigned long start;
 859         int num;
 860         int init_num;
 861         int create;
 862 };
 863
 864 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 865                                unsigned long block, int *aflags)
 866 {
 867         struct ext3_inode_info *ei = EXT3_I(inode);
 868         unsigned long bg_start;
 869         unsigned long colour;
 870         int depth;
 871
 872         if (path) {
 873                 struct ext3_extent *ex;
 874                 depth = path->p_depth;
 875
 876                 /* try to predict block placement */
 877                 if ((ex = path[depth].p_ext))
 878                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 879
 880                 /* it looks index is empty
 881                  * try to find starting from index itself */
 882                 if (path[depth].p_bh)
 883                         return path[depth].p_bh->b_blocknr;
 884         }
 885
 886         /* OK. use inode's group */
 887         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 888                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 889         colour = (current->pid % 16) *
 890                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 891         return bg_start + colour + block;
 892 }
 893
 894 #define ll_unmap_underlying_metadata(sb, blocknr) \
 895         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 896
 897 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 898 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 899                                 struct ext3_ext_path *path, unsigned long block,
 900                                 unsigned long *count, int *err)
 901 {
 902         unsigned long pblock, goal;
 903         int aflags = 0;
 904         struct inode *inode = ext3_ext_base2inode(base);
 905
 906         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 907         aflags |= 2; /* block have been already reserved */
 908         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 909         return pblock;
 910
 911 }
 912 #else
 913 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 914                                 struct ext3_ext_path *path, unsigned long block,
 915                                 unsigned long *count, int *err)
 916 {
 917         struct inode *inode = ext3_ext_base2inode(base);
 918         struct ext3_allocation_request ar;
 919         unsigned long pblock;
 920         int aflags;
 921
 922         /* find neighbour allocated blocks */
 923         ar.lleft = block;
 924         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 925         if (*err)
 926                 return 0;
 927         ar.lright = block;
 928         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 929         if (*err)
 930                 return 0;
 931
 932         /* allocate new block */
 933         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 934         ar.inode = inode;
 935         ar.logical = block;
 936         ar.len = *count;
 937         ar.flags = EXT3_MB_HINT_DATA;
 938         pblock = ext3_mb_new_blocks(handle, &ar, err);
 939         *count = ar.len;
 940         return pblock;
 941 }
 942 #endif
 943
 944 #ifdef EXT3_EXT_HAS_NO_TREE
 945 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 946                                   struct ext3_ext_path *path,
 947                                   struct ext3_ext_cache *cex,
 948 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 949                                    struct ext3_extent *ex,
 950 #endif
 951                                   void *cbdata)
 952 {
 953         struct bpointers *bp = cbdata;
 954 #else
 955 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 956                                   struct ext3_ext_path *path,
 957                                   struct ext3_ext_cache *cex
 958 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 959                                   , struct ext3_extent *ex
 960 #endif
 961                                  )
 962 {
 963         struct bpointers *bp = base->private;
 964 #endif
 965         struct inode *inode = ext3_ext_base2inode(base);
 966         struct ext3_extent nex;
 967         unsigned long pblock;
 968         unsigned long tgen;
 969         int err, i;
 970         unsigned long count;
 971         handle_t *handle;
 972
 973         i = EXT_DEPTH(base);
 974         EXT_ASSERT(i == path->p_depth);
 975         EXT_ASSERT(path[i].p_hdr);
 976
 977         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 978                 err = EXT_CONTINUE;
 979                 goto map;
 980         }
 981
 982         if (bp->create == 0) {
 983                 i = 0;
 984                 if (cex->ec_block < bp->start)
 985                         i = bp->start - cex->ec_block;
 986                 if (i >= cex->ec_len)
 987                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 988                                         i, cex->ec_len);
 989                 for (; i < cex->ec_len && bp->num; i++) {
 990                         *(bp->created) = 0;
 991                         bp->created++;
 992                         *(bp->blocks) = 0;
 993                         bp->blocks++;
 994                         bp->num--;
 995                         bp->start++;
 996                 }
 997
 998                 return EXT_CONTINUE;
 999         }
1000
1001         tgen = EXT_GENERATION(base);
1002         count = ext3_ext_calc_credits_for_insert(base, path);
1003         fsfilt_up_truncate_sem(inode);
1004
1005         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1006         if (IS_ERR(handle)) {
1007                 fsfilt_down_truncate_sem(inode);
1008                 return PTR_ERR(handle);
1009         }
1010
1011         fsfilt_down_truncate_sem(inode);
1012         if (tgen != EXT_GENERATION(base)) {
1013                 /* the tree has changed. so path can be invalid at moment */
1014                 ext3_journal_stop(handle);
1015                 return EXT_REPEAT;
1016         }
1017
1018         count = cex->ec_len;
1019         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1020         if (!pblock)
1021                 goto out;
1022         EXT_ASSERT(count <= cex->ec_len);
1023
1024         /* insert new extent */
1025         nex.ee_block = cpu_to_le32(cex->ec_block);
1026         ext3_ext_store_pblock(&nex, pblock);
1027         nex.ee_len = cpu_to_le16(count);
1028         err = ext3_ext_insert_extent(handle, base, path, &nex);
1029         if (err) {
1030                 /* free data blocks we just allocated */
1031                 /* not a good idea to call discard here directly,
1032                  * but otherwise we'd need to call it every free() */
1033 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1034                 ext3_mb_discard_inode_preallocations(inode);
1035 #endif
1036                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1037                                  cpu_to_le16(nex.ee_len), 0);
1038                 goto out;
1039         }
1040
1041         /*
1042          * Putting len of the actual extent we just inserted,
1043          * we are asking ext3_ext_walk_space() to continue
1044          * scaning after that block
1045          */
1046         cex->ec_len = le16_to_cpu(nex.ee_len);
1047         cex->ec_start = ext_pblock(&nex);
1048         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1049         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1050
1051 out:
1052         ext3_journal_stop(handle);
1053 map:
1054         if (err >= 0) {
1055                 /* map blocks */
1056                 if (bp->num == 0) {
1057                         CERROR("hmm. why do we find this extent?\n");
1058                         CERROR("initial space: %lu:%u\n",
1059                                 bp->start, bp->init_num);
1060                         CERROR("current extent: %u/%u/%llu %d\n",
1061                                 cex->ec_block, cex->ec_len,
1062                                 (unsigned long long)cex->ec_start,
1063                                 cex->ec_type);
1064                 }
1065                 i = 0;
1066                 if (cex->ec_block < bp->start)
1067                         i = bp->start - cex->ec_block;
1068                 if (i >= cex->ec_len)
1069                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1070                                         i, cex->ec_len);
1071                 for (; i < cex->ec_len && bp->num; i++) {
1072                         *(bp->blocks) = cex->ec_start + i;
1073                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1074                                 *(bp->created) = 0;
1075                         } else {
1076                                 *(bp->created) = 1;
1077                                 /* unmap any possible underlying metadata from
1078                                  * the block device mapping.  bug 6998. */
1079                                 ll_unmap_underlying_metadata(inode->i_sb,
1080                                                              *(bp->blocks));
1081                         }
1082                         bp->created++;
1083                         bp->blocks++;
1084                         bp->num--;
1085                         bp->start++;
1086                 }
1087         }
1088         return err;
1089 }
1090
1091 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1092                        unsigned long num, unsigned long *blocks,
1093                        int *created, int create)
1094 {
1095 #ifdef EXT3_EXT_HAS_NO_TREE
1096         struct ext3_ext_base *base = inode;
1097 #else
1098         struct ext3_extents_tree tree;
1099         struct ext3_ext_base *base = &tree;
1100 #endif
1101         struct bpointers bp;
1102         int err;
1103
1104         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1105                block, block + num - 1, (unsigned) inode->i_ino);
1106
1107 #ifndef EXT3_EXT_HAS_NO_TREE
1108         ext3_init_tree_desc(base, inode);
1109         tree.private = &bp;
1110 #endif
1111         bp.blocks = blocks;
1112         bp.created = created;
1113         bp.start = block;
1114         bp.init_num = bp.num = num;
1115         bp.create = create;
1116
1117         fsfilt_down_truncate_sem(inode);
1118         err = fsfilt_ext3_ext_walk_space(base, block, num,
1119                                          ext3_ext_new_extent_cb, &bp);
1120         ext3_ext_invalidate_cache(base);
1121         fsfilt_up_truncate_sem(inode);
1122
1123         return err;
1124 }
1125
1126 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1127                                     int pages, unsigned long *blocks,
1128                                     int *created, int create)
1129 {
1130         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1131         int rc = 0, i = 0;
1132         struct page *fp = NULL;
1133         int clen = 0;
1134
1135         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1136                 inode->i_ino, pages, (*page)->index);
1137
1138         /* pages are sorted already. so, we just have to find
1139          * contig. space and process them properly */
1140         while (i < pages) {
1141                 if (fp == NULL) {
1142                         /* start new extent */
1143                         fp = *page++;
1144                         clen = 1;
1145                         i++;
1146                         continue;
1147                 } else if (fp->index + clen == (*page)->index) {
1148                         /* continue the extent */
1149                         page++;
1150                         clen++;
1151                         i++;
1152                         continue;
1153                 }
1154
1155                 /* process found extent */
1156                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1157                                         clen * blocks_per_page, blocks,
1158                                         created, create);
1159                 if (rc)
1160                         GOTO(cleanup, rc);
1161
1162                 /* look for next extent */
1163                 fp = NULL;
1164                 blocks += blocks_per_page * clen;
1165                 created += blocks_per_page * clen;
1166         }
1167
1168         if (fp)
1169                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1170                                         clen * blocks_per_page, blocks,
1171                                         created, create);
1172 cleanup:
1173         return rc;
1174 }
1175 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1176
1177 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1178                                unsigned long *blocks, int *created, int create);
1179 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1180                                    int pages, unsigned long *blocks,
1181                                    int *created, int create)
1182 {
1183         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1184         unsigned long *b;
1185         int rc = 0, i, *cr;
1186
1187         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1188                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1189                 if (rc) {
1190                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1191                                inode->i_ino, *b, *cr, create, rc);
1192                         break;
1193                 }
1194
1195                 b += blocks_per_page;
1196                 cr += blocks_per_page;
1197         }
1198         return rc;
1199 }
1200
1201 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1202                                 int pages, unsigned long *blocks,
1203                                 int *created, int create,
1204                                 cfs_semaphore_t *optional_sem)
1205 {
1206         int rc;
1207 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1208         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1209                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1210                                                      blocks, created, create);
1211                 return rc;
1212         }
1213 #endif
1214         if (optional_sem != NULL)
1215                 cfs_down(optional_sem);
1216         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1217                                             created, create);
1218         if (optional_sem != NULL)
1219                 cfs_up(optional_sem);
1220
1221         return rc;
1222 }
1223
1224 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1225 {
1226         unsigned long block;
1227         struct buffer_head *bh;
1228         int err, blocksize, csize, boffs, osize = size;
1229
1230         /* prevent reading after eof */
1231         cfs_lock_kernel();
1232         if (i_size_read(inode) < *offs + size) {
1233                 size = i_size_read(inode) - *offs;
1234                 cfs_unlock_kernel();
1235                 if (size < 0) {
1236                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1237                                i_size_read(inode), *offs);
1238                         return -EBADR;
1239                 } else if (size == 0) {
1240                         return 0;
1241                 }
1242         } else {
1243                 cfs_unlock_kernel();
1244         }
1245
1246         blocksize = 1 << inode->i_blkbits;
1247
1248         while (size > 0) {
1249                 block = *offs >> inode->i_blkbits;
1250                 boffs = *offs & (blocksize - 1);
1251                 csize = min(blocksize - boffs, size);
1252                 bh = ext3_bread(NULL, inode, block, 0, &err);
1253                 if (!bh) {
1254                         CERROR("can't read block: %d\n", err);
1255                         return err;
1256                 }
1257
1258                 memcpy(buf, bh->b_data + boffs, csize);
1259                 brelse(bh);
1260
1261                 *offs += csize;
1262                 buf += csize;
1263                 size -= csize;
1264         }
1265         return osize;
1266 }
1267 EXPORT_SYMBOL(fsfilt_ext3_read);
1268
1269 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1270                                    int size, loff_t *offs)
1271 {
1272         int rc;
1273         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1274         if (rc > 0)
1275                 rc = 0;
1276         return rc;
1277 }
1278
1279 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1280                                 loff_t *offs, handle_t *handle)
1281 {
1282         struct buffer_head *bh = NULL;
1283         loff_t old_size = i_size_read(inode), offset = *offs;
1284         loff_t new_size = i_size_read(inode);
1285         unsigned long block;
1286         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1287
1288         while (bufsize > 0) {
1289                 if (bh != NULL)
1290                         brelse(bh);
1291
1292                 block = offset >> inode->i_blkbits;
1293                 boffs = offset & (blocksize - 1);
1294                 size = min(blocksize - boffs, bufsize);
1295                 bh = ext3_bread(handle, inode, block, 1, &err);
1296                 if (!bh) {
1297                         CERROR("can't read/create block: %d\n", err);
1298                         break;
1299                 }
1300
1301                 err = ext3_journal_get_write_access(handle, bh);
1302                 if (err) {
1303                         CERROR("journal_get_write_access() returned error %d\n",
1304                                err);
1305                         break;
1306                 }
1307                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1308                 memcpy(bh->b_data + boffs, buf, size);
1309                 err = ext3_journal_dirty_metadata(handle, bh);
1310                 if (err) {
1311                         CERROR("journal_dirty_metadata() returned error %d\n",
1312                                err);
1313                         break;
1314                 }
1315                 if (offset + size > new_size)
1316                         new_size = offset + size;
1317                 offset += size;
1318                 bufsize -= size;
1319                 buf += size;
1320         }
1321         if (bh)
1322                 brelse(bh);
1323
1324         /* correct in-core and on-disk sizes */
1325         if (new_size > i_size_read(inode)) {
1326                 cfs_lock_kernel();
1327                 if (new_size > i_size_read(inode))
1328                         i_size_write(inode, new_size);
1329                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1330                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1331                 if (i_size_read(inode) > old_size)
1332                         mark_inode_dirty(inode);
1333                 cfs_unlock_kernel();
1334         }
1335
1336         if (err == 0)
1337                 *offs = offset;
1338         return err;
1339 }
1340 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1341
1342 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1343                                     loff_t *offs, int force_sync)
1344 {
1345         struct inode *inode = file->f_dentry->d_inode;
1346         handle_t *handle;
1347         int err, block_count = 0, blocksize;
1348
1349         /* Determine how many transaction credits are needed */
1350         blocksize = 1 << inode->i_blkbits;
1351         block_count = (*offs & (blocksize - 1)) + bufsize;
1352         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1353
1354         handle = ext3_journal_start(inode,
1355                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1356         if (IS_ERR(handle)) {
1357                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1358                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1359                 return PTR_ERR(handle);
1360         }
1361
1362         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1363
1364         if (!err && force_sync)
1365                 handle->h_sync = 1; /* recovery likes this */
1366
1367         ext3_journal_stop(handle);
1368
1369         return err;
1370 }
1371
1372 static int fsfilt_ext3_setup(struct super_block *sb)
1373 {
1374 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1375      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1376         struct ext3_sb_info *sbi = EXT3_SB(sb);
1377 #if 0
1378         sbi->dx_lock = fsfilt_ext3_dx_lock;
1379         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1380 #endif
1381 #endif
1382 #ifdef S_PDIROPS
1383         CWARN("Enabling PDIROPS\n");
1384         set_opt(sbi->s_mount_opt, PDIROPS);
1385         sb->s_flags |= S_PDIROPS;
1386 #endif
1387         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1388                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1389 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1390         /* enable journaled quota support */
1391         /* kfreed in ext3_put_super() */
1392         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1393         if (!sbi->s_qf_names[USRQUOTA])
1394                 return -ENOMEM;
1395         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1396         if (!sbi->s_qf_names[GRPQUOTA]) {
1397                 kfree(sbi->s_qf_names[USRQUOTA]);
1398                 sbi->s_qf_names[USRQUOTA] = NULL;
1399                 return -ENOMEM;
1400         }
1401         sbi->s_jquota_fmt = QFMT_VFS_V0;
1402 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1403         set_opt(sbi->s_mount_opt, QUOTA);
1404 #endif
1405 #endif
1406         return 0;
1407 }
1408
1409 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1410    objects. Logs is number of logfiles to update */
1411 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1412 {
1413         if ( !fso ) {
1414                 switch(op) {
1415                 case FSFILT_OP_CREATE:
1416                                  /* directory leaf, index & indirect & EA*/
1417                         return 4 + 3 * logs;
1418                 case FSFILT_OP_UNLINK:
1419                         return 3 * logs;
1420                 }
1421         } else {
1422                 int i;
1423                 int needed = 0;
1424                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1425                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1426                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1427                 for (i = 0; i < op; i++, fso++) {
1428                         int nblocks = fso->fso_bufcnt * blockpp;
1429                         int ndindirect = min(nblocks, addrpp + 1);
1430                         int nindir = nblocks + ndindirect + 1;
1431
1432                         needed += nindir;
1433                 }
1434                 return needed + 3 * logs;
1435         }
1436
1437         return 0;
1438 }
1439
1440 #ifdef HAVE_QUOTA_SUPPORT
1441 #define DQINFO_COPY(out, in)                    \
1442 do {                                            \
1443         Q_COPY(out, in, dqi_bgrace);            \
1444         Q_COPY(out, in, dqi_igrace);            \
1445         Q_COPY(out, in, dqi_flags);             \
1446         Q_COPY(out, in, dqi_valid);             \
1447 } while (0)
1448
1449 #define DQBLK_COPY(out, in)                     \
1450 do {                                            \
1451         Q_COPY(out, in, dqb_bhardlimit);        \
1452         Q_COPY(out, in, dqb_bsoftlimit);        \
1453         Q_COPY(out, in, dqb_curspace);          \
1454         Q_COPY(out, in, dqb_ihardlimit);        \
1455         Q_COPY(out, in, dqb_isoftlimit);        \
1456         Q_COPY(out, in, dqb_curinodes);         \
1457         Q_COPY(out, in, dqb_btime);             \
1458         Q_COPY(out, in, dqb_itime);             \
1459         Q_COPY(out, in, dqb_valid);             \
1460 } while (0)
1461
1462 static int fsfilt_ext3_quotactl(struct super_block *sb,
1463                                 struct obd_quotactl *oqc)
1464 {
1465         int i, rc = 0, error = 0;
1466         struct quotactl_ops *qcop;
1467         struct if_dqinfo *info;
1468         struct if_dqblk *dqblk;
1469         ENTRY;
1470
1471         if (!sb->s_qcop)
1472                 RETURN(-ENOSYS);
1473
1474         OBD_ALLOC_PTR(info);
1475         if (!info)
1476                 RETURN(-ENOMEM);
1477         OBD_ALLOC_PTR(dqblk);
1478         if (!dqblk) {
1479                 OBD_FREE_PTR(info);
1480                 RETURN(-ENOMEM);
1481         }
1482
1483         DQINFO_COPY(info, &oqc->qc_dqinfo);
1484         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1485
1486         qcop = sb->s_qcop;
1487         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1488                 for (i = 0; i < MAXQUOTAS; i++) {
1489                         if (!Q_TYPESET(oqc, i))
1490                                 continue;
1491
1492                         if (oqc->qc_cmd == Q_QUOTAON) {
1493                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1494
1495                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1496
1497                                 rc = ll_quota_on(sb, i, QFMT_VFS_V0,
1498                                                  name[i], 0);
1499                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1500                                 rc = ll_quota_off(sb, i, 0);
1501                         }
1502
1503                         if (rc == -EBUSY)
1504                                 error = rc;
1505                         else if (rc)
1506                                 GOTO(out, rc);
1507                 }
1508                 GOTO(out, rc ?: error);
1509         }
1510
1511         switch (oqc->qc_cmd) {
1512         case Q_GETOINFO:
1513         case Q_GETINFO:
1514                 if (!qcop->get_info)
1515                         GOTO(out, rc = -ENOSYS);
1516                 rc = qcop->get_info(sb, oqc->qc_type, info);
1517                 break;
1518         case Q_SETQUOTA:
1519         case Q_INITQUOTA:
1520                 if (!qcop->set_dqblk)
1521                         GOTO(out, rc = -ENOSYS);
1522                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1523                 break;
1524         case Q_GETOQUOTA:
1525         case Q_GETQUOTA:
1526                 if (!qcop->get_dqblk)
1527                         GOTO(out, rc = -ENOSYS);
1528                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1529                 if (!rc)
1530                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1531                 break;
1532         case Q_SYNC:
1533                 if (!sb->s_qcop->quota_sync)
1534                         GOTO(out, rc = -ENOSYS);
1535                 qcop->quota_sync(sb, oqc->qc_type);
1536                 break;
1537         case Q_FINVALIDATE:
1538                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1539                 for (i = 0; i < MAXQUOTAS; i++) {
1540                         struct file *fp;
1541                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1542
1543                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1544
1545                         if (!Q_TYPESET(oqc, i))
1546                                 continue;
1547
1548                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1549                         if (IS_ERR(fp)) {
1550                                 rc = PTR_ERR(fp);
1551                                 CERROR("error invalidating operational quota file"
1552                                        " %s (rc:%d)\n", name[i], rc);
1553                         } else {
1554                                 filp_close(fp, 0);
1555                         }
1556
1557                 }
1558                 break;
1559         default:
1560                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1561                 LBUG();
1562         }
1563 out:
1564         DQINFO_COPY(&oqc->qc_dqinfo, info);
1565         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1566
1567         OBD_FREE_PTR(info);
1568         OBD_FREE_PTR(dqblk);
1569
1570         if (rc)
1571                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1572                                 "failed: %d\n",
1573                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1574         RETURN(rc);
1575 }
1576
1577 struct chk_dqblk{
1578         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1579         cfs_list_t              dqb_list;        /** in list also */
1580         qid_t                   dqb_id;          /** uid/gid */
1581         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1582         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1583         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1584         qsize_t                 dqb_curspace;    /** current space */
1585         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1586         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1587         qsize_t                 dqb_curinodes;   /** current inodes */
1588         __u64                   dqb_btime;       /** block grace time */
1589         __u64                   dqb_itime;       /** inode grace time */
1590         __u32                   dqb_valid;       /** flag for above fields */
1591 };
1592
1593 static inline unsigned int chkquot_hash(qid_t id, int type)
1594                                         __attribute__((__const__));
1595
1596 static inline unsigned int chkquot_hash(qid_t id, int type)
1597 {
1598         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1599 }
1600
1601 static inline struct chk_dqblk *
1602 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1603 {
1604         cfs_hlist_node_t *node;
1605         struct chk_dqblk *cdqb;
1606
1607         cfs_hlist_for_each(node, head) {
1608                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1609                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1610                         return cdqb;
1611         }
1612
1613         return NULL;
1614 }
1615
1616 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1617 {
1618         struct chk_dqblk *cdqb;
1619
1620         OBD_ALLOC_PTR(cdqb);
1621         if (cdqb) {
1622                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1623                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1624                 cdqb->dqb_id = id;
1625                 cdqb->dqb_type = type;
1626         }
1627
1628         return cdqb;
1629 }
1630
1631 static struct chk_dqblk *
1632 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1633       cfs_list_t *list, qid_t id, int type, int first_check)
1634 {
1635         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1636         struct if_dqblk dqb;
1637         struct chk_dqblk *cdqb;
1638         int rc;
1639
1640         cdqb = find_chkquot(head, id, type);
1641         if (cdqb)
1642                 return cdqb;
1643
1644         cdqb = alloc_chkquot(id, type);
1645         if (!cdqb)
1646                 return NULL;
1647
1648         if (!first_check) {
1649                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1650                 if (rc) {
1651                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1652                                id, type, rc);
1653                 } else {
1654                         DQBLK_COPY(cdqb, &dqb);
1655                         cdqb->dqb_curspace = 0;
1656                         cdqb->dqb_curinodes = 0;
1657                 }
1658         }
1659
1660         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1661         cfs_list_add_tail(&cdqb->dqb_list, list);
1662
1663         return cdqb;
1664 }
1665
1666 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1667 {
1668         struct obd_quotactl *oqctl;
1669         int rc;
1670
1671         OBD_ALLOC_PTR(oqctl);
1672         if (!oqctl)
1673                 RETURN(-ENOMEM);
1674
1675         oqctl->qc_cmd = cmd;
1676         oqctl->qc_id = qfmt;
1677         oqctl->qc_type = type;
1678         rc = fsfilt_ext3_quotactl(sb, oqctl);
1679
1680         OBD_FREE_PTR(oqctl);
1681         return rc;
1682 }
1683
1684 static inline int read_old_dqinfo(struct super_block *sb, int type,
1685                                   struct if_dqinfo *dqinfo)
1686 {
1687         struct obd_quotactl *oqctl;
1688         int rc;
1689         ENTRY;
1690
1691         OBD_ALLOC_PTR(oqctl);
1692         if (!oqctl)
1693                 RETURN(-ENOMEM);
1694
1695         oqctl->qc_cmd = Q_GETINFO;
1696         oqctl->qc_type = type;
1697         rc = fsfilt_ext3_quotactl(sb, oqctl);
1698         if (!rc)
1699                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1700
1701         OBD_FREE_PTR(oqctl);
1702         RETURN(rc);
1703 }
1704
1705 #ifndef HAVE_EXT4_LDISKFS
1706 static inline struct ext3_group_desc *
1707 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1708 {
1709         unsigned long desc_block, desc;
1710         struct ext3_group_desc *gdp;
1711
1712         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1713         desc = group % EXT3_DESC_PER_BLOCK(sb);
1714         gdp = (struct ext3_group_desc *)
1715               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1716
1717         return gdp + desc;
1718 }
1719
1720 static inline struct buffer_head *
1721 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1722 {
1723         struct ext3_group_desc *desc;
1724         struct buffer_head *bh;
1725
1726         desc = get_group_desc(sb, group, NULL);
1727         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1728         return bh;
1729 }
1730
1731 static __u32 ext3_itable_unused_count(struct super_block *sb,
1732                                struct ext3_group_desc *bg) {
1733        return le16_to_cpu(bg->bg_itable_unused);
1734 }
1735 #else
1736 #define get_group_desc ext3_get_group_desc
1737 #endif
1738
1739 struct qchk_ctxt {
1740         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1741         cfs_list_t              qckt_list;                 /* quotacheck list */
1742         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1743         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1744 };
1745
1746 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1747                            struct obd_quotactl *oqc)
1748 {
1749         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1750         loff_t size = 0;
1751         qid_t qid[MAXQUOTAS];
1752         int cnt, i, rc = 0;
1753
1754         if (!inode)
1755                 return 0;
1756
1757         qid[USRQUOTA] = inode->i_uid;
1758         qid[GRPQUOTA] = inode->i_gid;
1759
1760         if (S_ISDIR(inode->i_mode) ||
1761             S_ISREG(inode->i_mode) ||
1762             S_ISLNK(inode->i_mode))
1763                 size = inode_get_bytes(inode);
1764
1765         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1766                 if (!Q_TYPESET(oqc, cnt))
1767                         continue;
1768
1769                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1770                                 &qctxt->qckt_list, qid[cnt], cnt,
1771                                 qctxt->qckt_first_check[cnt]);
1772                 if (!cdqb[cnt]) {
1773                         rc = -ENOMEM;
1774                         break;
1775                 }
1776
1777                 cdqb[cnt]->dqb_curspace += size;
1778                 cdqb[cnt]->dqb_curinodes++;
1779         }
1780
1781         if (rc) {
1782                 for (i = 0; i < cnt; i++) {
1783                         if (!Q_TYPESET(oqc, i))
1784                                 continue;
1785                         LASSERT(cdqb[i]);
1786                         cdqb[i]->dqb_curspace -= size;
1787                         cdqb[i]->dqb_curinodes--;
1788                 }
1789         }
1790
1791         return rc;
1792 }
1793
1794 /* write dqinfo struct in a new quota file */
1795 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1796 {
1797         struct v2_disk_dqinfo dqinfo;
1798         __u32 blocks = V2_DQTREEOFF + 1;
1799         loff_t offset = V2_DQINFOOFF;
1800
1801         if (info) {
1802                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1803                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1804                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1805                                                ~DQF_INFO_DIRTY);
1806         } else {
1807                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1808                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1809                 dqinfo.dqi_flags = 0;
1810         }
1811
1812         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1813         dqinfo.dqi_free_blk = 0;
1814         dqinfo.dqi_free_entry = 0;
1815
1816         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1817 }
1818
1819 static int v3_write_dqheader(struct file *f, int type)
1820 {
1821         static const __u32 quota_magics[] = V2_INITQMAGICS;
1822         static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1823         struct v2_disk_dqheader dqhead;
1824         loff_t offset = 0;
1825
1826         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1827         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1828
1829         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1830         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1831
1832         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1833 }
1834
1835 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1836                                   struct obd_quotactl *oqc)
1837 {
1838         int i, rc = 0;
1839         ENTRY;
1840
1841         for (i = 0; i < MAXQUOTAS; i++) {
1842                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1843                                          NULL : &qctxt->qckt_dqinfo[i];
1844                 struct file *file;
1845                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1846
1847                 if (!Q_TYPESET(oqc, i))
1848                         continue;
1849
1850                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1851
1852                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1853                 if (IS_ERR(file)) {
1854                         rc = PTR_ERR(file);
1855                         CERROR("can't create %s file: rc = %d\n",
1856                                name[i], rc);
1857                         GOTO(out, rc);
1858                 }
1859
1860                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1861                         CERROR("file %s is not regular", name[i]);
1862                         filp_close(file, 0);
1863                         GOTO(out, rc = -EINVAL);
1864                 }
1865
1866                 DQUOT_DROP(file->f_dentry->d_inode);
1867
1868                 rc = v3_write_dqheader(file, i);
1869                 if (rc) {
1870                         filp_close(file, 0);
1871                         GOTO(out, rc);
1872                 }
1873
1874                 rc = v3_write_dqinfo(file, i, info);
1875                 filp_close(file, 0);
1876                 if (rc)
1877                         GOTO(out, rc);
1878         }
1879
1880 out:
1881         RETURN(rc);
1882 }
1883
1884
1885 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1886                           struct chk_dqblk *cdqb)
1887 {
1888         struct obd_quotactl *oqc;
1889         long now;
1890         int rc;
1891         ENTRY;
1892
1893         OBD_ALLOC_PTR(oqc);
1894         if (!oqc)
1895                 RETURN(-ENOMEM);
1896
1897         now = cfs_time_current_sec();
1898
1899         if (cdqb->dqb_bsoftlimit &&
1900             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1901             !cdqb->dqb_btime)
1902                 cdqb->dqb_btime =
1903                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1904
1905         if (cdqb->dqb_isoftlimit &&
1906             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1907             !cdqb->dqb_itime)
1908                 cdqb->dqb_itime =
1909                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1910
1911         cdqb->dqb_valid = QIF_ALL;
1912
1913         oqc->qc_cmd = Q_SETQUOTA;
1914         oqc->qc_type = cdqb->dqb_type;
1915         oqc->qc_id = cdqb->dqb_id;
1916         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1917
1918         rc = fsfilt_ext3_quotactl(sb, oqc);
1919         OBD_FREE_PTR(oqc);
1920         RETURN(rc);
1921 }
1922
1923 static int prune_chkquots(struct super_block *sb,
1924                           struct qchk_ctxt *qctxt, int error)
1925 {
1926         struct chk_dqblk *cdqb, *tmp;
1927         int rc;
1928
1929         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1930                 if (!error) {
1931                         rc = commit_chkquot(sb, qctxt, cdqb);
1932                         if (rc)
1933                                 error = rc;
1934                 }
1935                 cfs_hlist_del_init(&cdqb->dqb_hash);
1936                 cfs_list_del(&cdqb->dqb_list);
1937                 OBD_FREE_PTR(cdqb);
1938         }
1939
1940         return error;
1941 }
1942
1943 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1944 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1945 #endif
1946
1947 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1948                                   struct obd_quotactl *oqc)
1949 {
1950         struct ext3_sb_info *sbi = EXT3_SB(sb);
1951         int i, group, uninit_feat = 0;
1952         struct qchk_ctxt *qctxt;
1953         struct buffer_head *bitmap_bh = NULL;
1954         unsigned long ino, inode_inuse;
1955         struct inode *inode;
1956         int rc = 0;
1957         ENTRY;
1958
1959         /* turn on quota and read dqinfo if existed */
1960         OBD_ALLOC_PTR(qctxt);
1961         if (!qctxt) {
1962                 oqc->qc_stat = -ENOMEM;
1963                 RETURN(-ENOMEM);
1964         }
1965
1966         for (i = 0; i < NR_DQHASH; i++)
1967                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1968         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
1969
1970         for (i = 0; i < MAXQUOTAS; i++) {
1971                 if (!Q_TYPESET(oqc, i))
1972                         continue;
1973
1974                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1975                 if (!rc || rc == -EBUSY) {
1976                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1977                         if (rc)
1978                                 GOTO(out, rc);
1979                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1980                         qctxt->qckt_first_check[i] = 1;
1981                 } else if (rc) {
1982                         GOTO(out, rc);
1983                 }
1984         }
1985         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
1986                 /* This filesystem supports the uninit group feature */
1987                 uninit_feat = 1;
1988
1989         /* number of inodes that have been allocated */
1990         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
1991                       percpu_counter_sum(&sbi->s_freeinodes_counter);
1992
1993         /* check quota and update in hash */
1994         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
1995              group++) {
1996                 unsigned long used_count = sbi->s_inodes_per_group;
1997
1998                 if (uninit_feat) {
1999                         struct ext3_group_desc *desc;
2000                         desc = get_group_desc(sb, group, NULL);
2001                         if (!desc)
2002                                 GOTO(out, -EIO);
2003
2004                         /* we don't really need to take the group lock here,
2005                          * but it may be useful if one day we support online
2006                          * quotacheck */
2007                         spin_lock(sb_bgl_lock(sbi, group));
2008                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2009                                 /* no inode in use in this group, just skip it */
2010                                 spin_unlock(sb_bgl_lock(sbi, group));
2011                                 continue;
2012                         }
2013                         used_count -= ext3_itable_unused_count(sb, desc);
2014                         spin_unlock(sb_bgl_lock(sbi, group));
2015                 }
2016
2017                 ino = group * sbi->s_inodes_per_group + 1;
2018                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2019                 if (!bitmap_bh) {
2020                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2021                                sb->s_id, group);
2022                         GOTO(out, -EIO);
2023                 }
2024
2025                 i = 0;
2026                 while (i < used_count &&
2027                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2028                                                used_count, i)) < used_count) {
2029                         inode_inuse--;
2030                         i++;
2031                         ino = i + group * sbi->s_inodes_per_group;
2032                         if (ino < sbi->s_first_ino)
2033                                 continue;
2034 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2035                         inode = ext3_iget(sb, ino);
2036 #else
2037                         inode = iget(sb, ino);
2038 #endif
2039                         if (!inode || IS_ERR(inode))
2040                                 continue;
2041
2042                         rc = add_inode_quota(inode, qctxt, oqc);
2043                         iput(inode);
2044                         if (rc) {
2045                                 brelse(bitmap_bh);
2046                                 GOTO(out, rc);
2047                         }
2048                 }
2049                 brelse(bitmap_bh);
2050         }
2051
2052         /* read old quota limits from old quota file. (only for the user
2053          * has limits but hasn't file) */
2054 #ifdef HAVE_QUOTA_SUPPORT
2055         for (i = 0; i < MAXQUOTAS; i++) {
2056                 cfs_list_t id_list;
2057                 struct dquot_id *dqid, *tmp;
2058
2059                 if (!Q_TYPESET(oqc, i))
2060                         continue;
2061
2062                 if (qctxt->qckt_first_check[i])
2063                         continue;
2064
2065
2066                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2067                 CFS_INIT_LIST_HEAD(&id_list);
2068 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2069                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2070 #else
2071                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2072 #endif
2073                 if (rc)
2074                         CERROR("read old limits failed. (rc:%d)\n", rc);
2075
2076                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2077                         cfs_list_del_init(&dqid->di_link);
2078
2079                         if (!rc)
2080                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2081                                       dqid->di_id, i,
2082                                       qctxt->qckt_first_check[i]);
2083                         OBD_FREE_PTR(dqid);
2084                 }
2085         }
2086 #endif
2087         /* turn off quota cause we are to dump chk_dqblk to files */
2088         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2089
2090         rc = create_new_quota_files(qctxt, oqc);
2091         if (rc)
2092                 GOTO(out, rc);
2093
2094         /* we use vfs functions to set dqblk, so turn quota on */
2095         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2096 out:
2097         /* dump and free chk_dqblk */
2098         rc = prune_chkquots(sb, qctxt, rc);
2099         OBD_FREE_PTR(qctxt);
2100
2101         /* turn off quota, `lfs quotacheck` will turn on when all
2102          * nodes quotacheck finish. */
2103         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2104
2105         oqc->qc_stat = rc;
2106         if (rc)
2107                 CERROR("quotacheck failed: rc = %d\n", rc);
2108
2109         RETURN(rc);
2110 }
2111
2112 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2113                                  int cmd)
2114 {
2115         int rc = 0;
2116         ENTRY;
2117
2118         if (lqi->qi_files[type] == NULL) {
2119                 CERROR("operate qinfo before it's enabled!\n");
2120                 RETURN(-ESRCH);
2121         }
2122
2123         switch (cmd) {
2124         case QFILE_CHK:
2125                 rc = lustre_check_quota_file(lqi, type);
2126                 break;
2127         case QFILE_RD_INFO:
2128                 rc = lustre_read_quota_info(lqi, type);
2129                 break;
2130         case QFILE_WR_INFO:
2131                 rc = lustre_write_quota_info(lqi, type);
2132                 break;
2133         case QFILE_INIT_INFO:
2134                 rc = lustre_init_quota_info(lqi, type);
2135                 break;
2136         case QFILE_CONVERT:
2137                 rc = -ENOTSUPP;
2138                 CERROR("quota CONVERT command is not supported\n");
2139                 break;
2140         default:
2141                 rc = -ENOTSUPP;
2142                 CERROR("Unsupported admin quota file cmd %d\n"
2143                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2144                        cmd);
2145                 break;
2146         }
2147         RETURN(rc);
2148 }
2149
2150 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2151                             cfs_list_t *list)
2152 {
2153         return lustre_get_qids(file, inode, type, list);
2154 }
2155
2156 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2157 {
2158         int rc = 0;
2159         ENTRY;
2160
2161         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2162                 CERROR("operate dquot before it's enabled!\n");
2163                 RETURN(-ESRCH);
2164         }
2165
2166         switch (cmd) {
2167         case QFILE_RD_DQUOT:
2168                 rc = lustre_read_dquot(dquot);
2169                 break;
2170         case QFILE_WR_DQUOT:
2171                 if (dquot->dq_dqb.dqb_ihardlimit ||
2172                     dquot->dq_dqb.dqb_isoftlimit ||
2173                     dquot->dq_dqb.dqb_bhardlimit ||
2174                     dquot->dq_dqb.dqb_bsoftlimit)
2175                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2176                 else
2177                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2178
2179                 rc = lustre_commit_dquot(dquot);
2180                 if (rc >= 0)
2181                         rc = 0;
2182                 break;
2183         default:
2184                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2185                 LBUG();
2186                 break;
2187         }
2188         RETURN(rc);
2189 }
2190
2191 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2192                                 struct inode *inode, int frags)
2193 {
2194 #ifdef EXT3_EXT_HAS_NO_TREE
2195         struct ext3_ext_base *base = inode;
2196 #else
2197         struct ext3_extents_tree tree;
2198         struct ext3_ext_base *base = &tree;
2199
2200         ext3_init_tree_desc(base, inode);
2201 #endif
2202         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2203          * metablocks at maxium b=16542 */
2204         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2205         return 0;
2206 }
2207
2208 #endif
2209
2210 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2211 {
2212         return (EXT3_SB(sb)->journal_bdev);
2213 }
2214 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2215
2216 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2217                           char *buf, int count, loff_t pos)
2218 {
2219         loff_t p = pos;
2220         int rc;
2221
2222         if (!f && !inode) {
2223                 CERROR("lustre_read_quota failed for no quota file!\n");
2224                 libcfs_debug_dumpstack(NULL);
2225                 return -EINVAL;
2226         }
2227
2228         /* Support for both adm and op quota files must be provided */
2229         if (f) {
2230                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2231                 rc = rc < 0 ? rc : p - pos;
2232         } else {
2233                 struct super_block *sb = inode->i_sb;
2234                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2235         }
2236         return rc;
2237 }
2238
2239 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2240 {
2241         loff_t p = pos;
2242         int rc;
2243
2244         /* Only adm quota files are supported, op updates are handled by vfs */
2245         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2246         rc = rc < 0 ? rc : p - pos;
2247
2248         return rc;
2249 }
2250
2251 void *lustre_quota_journal_start(struct inode *inode, int delete)
2252 {
2253         handle_t *handle;
2254         unsigned block_count;
2255
2256         if (delete) {
2257                 /* each indirect block (+4) may become free, attaching to the
2258                  * header list of free blocks (+1); the data block (+1) may
2259                  * become a free block (+0) or a block with free dqentries (+0) */
2260                 block_count = (4 + 1) + 1;
2261                 handle = ext3_journal_start(inode,
2262                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2263         } else {
2264                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2265                  * freeblk reusage with a header update (+1); dqentry is either reused
2266                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2267                  * a new block allocation (+1) with a header update (+1)              */
2268                 block_count = (4 + 1) + 3;
2269                 handle = ext3_journal_start(inode,
2270                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2271
2272         }
2273
2274         return handle;
2275 }
2276
2277 void lustre_quota_journal_stop(void *handle)
2278 {
2279         ext3_journal_stop((handle_t *)handle);
2280 }
2281
2282 static struct fsfilt_operations fsfilt_ext3_ops = {
2283         .fs_type                = "ext3",
2284         .fs_owner               = THIS_MODULE,
2285         .fs_getlabel            = fsfilt_ext3_get_label,
2286         .fs_setlabel            = fsfilt_ext3_set_label,
2287         .fs_uuid                = fsfilt_ext3_uuid,
2288         .fs_start               = fsfilt_ext3_start,
2289         .fs_brw_start           = fsfilt_ext3_brw_start,
2290         .fs_extend              = fsfilt_ext3_extend,
2291         .fs_commit              = fsfilt_ext3_commit,
2292         .fs_commit_async        = fsfilt_ext3_commit_async,
2293         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2294         .fs_setattr             = fsfilt_ext3_setattr,
2295         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2296         .fs_set_md              = fsfilt_ext3_set_md,
2297         .fs_get_md              = fsfilt_ext3_get_md,
2298         .fs_readpage            = fsfilt_ext3_readpage,
2299         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2300         .fs_statfs              = fsfilt_ext3_statfs,
2301         .fs_sync                = fsfilt_ext3_sync,
2302         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2303         .fs_write_record        = fsfilt_ext3_write_record,
2304         .fs_read_record         = fsfilt_ext3_read_record,
2305         .fs_setup               = fsfilt_ext3_setup,
2306         .fs_send_bio            = fsfilt_ext3_send_bio,
2307         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2308 #ifdef HAVE_DISK_INODE_VERSION
2309         .fs_get_version         = fsfilt_ext3_get_version,
2310         .fs_set_version         = fsfilt_ext3_set_version,
2311 #endif
2312 #ifdef HAVE_QUOTA_SUPPORT
2313         .fs_quotactl            = fsfilt_ext3_quotactl,
2314         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2315         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2316         .fs_qids                = fsfilt_ext3_qids,
2317         .fs_dquot               = fsfilt_ext3_dquot,
2318         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2319 #endif
2320         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2321 };
2322
2323 static int __init fsfilt_ext3_init(void)
2324 {
2325         int rc;
2326
2327         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2328                                          sizeof(struct fsfilt_cb_data), 0, 0);
2329         if (!fcb_cache) {
2330                 CERROR("error allocating fsfilt journal callback cache\n");
2331                 GOTO(out, rc = -ENOMEM);
2332         }
2333
2334         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2335
2336         if (rc) {
2337                 int err = cfs_mem_cache_destroy(fcb_cache);
2338                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2339         }
2340 out:
2341         return rc;
2342 }
2343
2344 static void __exit fsfilt_ext3_exit(void)
2345 {
2346         int rc;
2347
2348         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2349         rc = cfs_mem_cache_destroy(fcb_cache);
2350         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2351 }
2352
2353 module_init(fsfilt_ext3_init);
2354 module_exit(fsfilt_ext3_exit);
2355
2356 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2357 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2358 MODULE_LICENSE("GPL");