lustre/lvfs/fsfilt_ext3.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  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_LINUX_EXPORTFS_H
  50 #include <linux/exportfs.h>
  51 #endif
  52 #include <ext4/ext4.h>
  53 #include <ext4/ext4_jbd2.h>
  54 #include <linux/version.h>
  55 #include <linux/bitops.h>
  56 #include <linux/quota.h>
  57 #ifdef HAVE_QUOTAIO_H
  58 # include <linux/quotaio_v2.h>
  59 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  60 # include <quota/quotaio_v2.h>
  61 # include <quota/quota_tree.h>
  62 # define V2_DQTREEOFF    QT_TREEOFF
  63 #elif defined(HAVE_FS_QUOTAIO_H)
  64 # include <quotaio_v2.h>
  65 # include <quota_tree.h>
  66 # define V2_DQTREEOFF    QT_TREEOFF
  67 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  68 #endif
  69
  70 #ifdef QFMT_VFS_V1
  71 #define QFMT_LUSTRE QFMT_VFS_V1
  72 #else
  73 #define QFMT_LUSTRE QFMT_VFS_V0
  74 #endif
  75
  76 #if defined(HAVE_EXT3_XATTR_H)
  77 # include <ext3/xattr.h>
  78 #elif !defined(EXT3_XATTR_INDEX_TRUSTED)
  79 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  80 /* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
  81 # define EXT3_XATTR_INDEX_TRUSTED        4
  82 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  83 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  84 #endif
  85
  86 #include <libcfs/libcfs.h>
  87 #include <lustre_fsfilt.h>
  88 #include <obd.h>
  89 #include <lustre_quota.h>
  90 #include <linux/lustre_compat25.h>
  91 #include <linux/lprocfs_status.h>
  92
  93 #include <ext4/ext4_extents.h>
  94
  95 #include "lustre_quota_fmt.h"
  96
  97 /* for kernels 2.6.18 and later */
  98 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  99
 100 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 101                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 102
 103 #define ext3_mb_discard_inode_preallocations(inode) \
 104                  ext3_discard_preallocations(inode)
 105
 106 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 107 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 108
 109 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 110 # define journal_callback ext4_journal_cb_entry
 111 # define fsfilt_journal_callback_set(handle, func, jcb) \
 112          ext4_journal_callback_add(handle, func, jcb)
 113 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
 114 # define fsfilt_journal_callback_set(handle, func, jcb) \
 115          jbd2_journal_callback_set(handle, func, jcb)
 116 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
 117 # define fsfilt_journal_callback_set(handle, func, jcb) \
 118          journal_callback_set(handle, func, jcb)
 119 #else
 120 # error missing journal commit callback
 121 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
 122
 123 static cfs_mem_cache_t *fcb_cache;
 124
 125 struct fsfilt_cb_data {
 126         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 127         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 128         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 129         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 130         void *cb_data;                  /* MDS/OST completion function data */
 131 };
 132
 133 #ifndef ext3_find_next_bit
 134 #define ext3_find_next_bit           ext2_find_next_bit
 135 #endif
 136
 137 #ifndef ext2_find_next_bit
 138 #ifdef __LITTLE_ENDIAN
 139 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 140 #else
 141 error "Need implementation of find_next_bit on big-endian systems"
 142 #endif  /* __LITTLE_ENDIAN */
 143 #endif  /* !ext2_find_next_le_bit */
 144
 145 static char *fsfilt_ext3_get_label(struct super_block *sb)
 146 {
 147         return EXT3_SB(sb)->s_es->s_volume_name;
 148 }
 149
 150 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 151 {
 152         /* see e.g. fsfilt_ext3_write_record() */
 153         journal_t *journal;
 154         handle_t *handle;
 155         int err;
 156
 157         journal = EXT3_SB(sb)->s_journal;
 158         handle = ext3_journal_start_sb(sb, 1);
 159         if (IS_ERR(handle)) {
 160                 CERROR("can't start transaction\n");
 161                 return(PTR_ERR(handle));
 162         }
 163
 164         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 165         if (err)
 166                 goto out;
 167
 168         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 169                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 170
 171         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 172
 173 out:
 174         ext3_journal_stop(handle);
 175
 176         return(err);
 177 }
 178
 179 static char *fsfilt_ext3_uuid(struct super_block *sb)
 180 {
 181         return EXT3_SB(sb)->s_es->s_uuid;
 182 }
 183
 184 #ifdef HAVE_DISK_INODE_VERSION
 185
 186 static __u64 get_i_version(struct inode *inode)
 187 {
 188         return EXT3_I(inode)->i_fs_version;
 189 }
 190
 191 static void set_i_version(struct inode *inode, __u64 new_version)
 192 {
 193         (EXT3_I(inode))->i_fs_version = new_version;
 194 }
 195
 196 /*
 197  * Get the 64-bit version for an inode.
 198  */
 199 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 200 {
 201         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 202                get_i_version(inode), inode->i_ino);
 203         return get_i_version(inode);
 204 }
 205
 206 /*
 207  * Set the 64-bit version and return the old version.
 208  */
 209 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 210 {
 211         __u64 old_version = get_i_version(inode);
 212
 213         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 214                new_version, old_version, inode->i_ino);
 215         set_i_version(inode, new_version);
 216         /* version is set after all inode operations are finished, so we should
 217          * mark it dirty here */
 218         inode->i_sb->s_op->dirty_inode(inode);
 219         return old_version;
 220 }
 221
 222 #endif
 223
 224 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
 225 #ifdef HAVE_BLOCKS_FOR_TRUNCATE
 226 # include <ext4/truncate.h>
 227 #else
 228 static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
 229 {
 230         ext4_lblk_t needed;
 231
 232         needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
 233         if (needed < 2)
 234                 needed = 2;
 235         if (needed > EXT4_MAX_TRANS_DATA)
 236                 needed = EXT4_MAX_TRANS_DATA;
 237         return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
 238 }
 239 #endif
 240
 241 /*
 242  * We don't currently need any additional blocks for rmdir and
 243  * unlink transactions because we are storing the OST oa_id inside
 244  * the inode (which we will be changing anyways as part of this
 245  * transaction).
 246  */
 247 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 248                                int logs)
 249 {
 250         /* For updates to the last received file */
 251         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 252         journal_t *journal;
 253         void *handle;
 254
 255         if (current->journal_info) {
 256                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 257                        current->journal_info);
 258                 goto journal_start;
 259         }
 260
 261         switch(op) {
 262         case FSFILT_OP_RMDIR:
 263         case FSFILT_OP_UNLINK:
 264                 /* delete one file + create/update logs for each stripe */
 265                 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
 266                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 267                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 268                 break;
 269         case FSFILT_OP_RENAME:
 270                 /* modify additional directory */
 271                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 272                 /* no break */
 273         case FSFILT_OP_SYMLINK:
 274                 /* additional block + block bitmap + GDT for long symlink */
 275                 nblocks += 3;
 276                 /* no break */
 277         case FSFILT_OP_CREATE: {
 278                 /* no break */
 279         }
 280         case FSFILT_OP_MKDIR:
 281         case FSFILT_OP_MKNOD:
 282                 /* modify one inode + block bitmap + GDT */
 283                 nblocks += 3;
 284                 /* no break */
 285         case FSFILT_OP_LINK:
 286                 /* modify parent directory */
 287                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 288                            EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
 289                 /* create/update logs for each stripe */
 290                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 291                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 292                 break;
 293         case FSFILT_OP_SETATTR:
 294                 /* Setattr on inode */
 295                 nblocks += 1;
 296                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 297                            EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
 298                 /* quota chown log for each stripe */
 299                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 300                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 301                 break;
 302         case FSFILT_OP_CANCEL_UNLINK:
 303                 LASSERT(logs == 1);
 304
 305                 /* blocks for log header bitmap update OR
 306                  * blocks for catalog header bitmap update + unlink of logs +
 307                  * blocks for delete the inode (include blocks truncating). */
 308                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 309                           EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
 310                           ext4_blocks_for_truncate(inode) + 3;
 311                 break;
 312         default: CERROR("unknown transaction start op %d\n", op);
 313                 LBUG();
 314         }
 315
 316         LASSERT(current->journal_info == desc_private);
 317         journal = EXT3_SB(inode->i_sb)->s_journal;
 318         if (nblocks > journal->j_max_transaction_buffers) {
 319                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 320                        nblocks, op, logs, journal->j_max_transaction_buffers);
 321                 nblocks = journal->j_max_transaction_buffers;
 322         }
 323
 324  journal_start:
 325         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 326         handle = ext3_journal_start(inode, nblocks);
 327
 328         if (!IS_ERR(handle))
 329                 LASSERT(current->journal_info == handle);
 330         else
 331                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 332                        op, nblocks, PTR_ERR(handle));
 333         return handle;
 334 }
 335
 336 /*
 337  * Calculate the number of buffer credits needed to write multiple pages in
 338  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 339  * doesn't have a nice API for calculating this sort of thing in advance.
 340  *
 341  * See comment above ext3_writepage_trans_blocks for details.  We assume
 342  * no data journaling is being done, but it does allow for all of the pages
 343  * being non-contiguous.  If we are guaranteed contiguous pages we could
 344  * reduce the number of (d)indirect blocks a lot.
 345  *
 346  * With N blocks per page and P pages, for each inode we have at most:
 347  * N*P indirect
 348  * min(N*P, blocksize/4 + 1) dindirect blocks
 349  * niocount tindirect
 350  *
 351  * For the entire filesystem, we have at most:
 352  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 353  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 354  * objcount inode blocks
 355  * 1 superblock
 356  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 357  *
 358  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 359  */
 360 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 361                                       int niocount, struct niobuf_local *nb)
 362 {
 363         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 364         __u64 next_indir;
 365         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 366         int nbitmaps = 0, ngdblocks;
 367         int needed = objcount + 1; /* inodes + superblock */
 368         int i, j;
 369
 370         for (i = 0, j = 0; i < objcount; i++, fso++) {
 371                 /* two or more dindirect blocks in case we cross boundary */
 372                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 373                                     nb[j].offset) >>
 374                                    sb->s_blocksize_bits) /
 375                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 376                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 377
 378                 /* leaf, indirect, tindirect blocks for first block */
 379                 nbitmaps += blockpp + 2;
 380
 381                 j += fso->fso_bufcnt;
 382         }
 383
 384         next_indir = nb[0].offset +
 385                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 386         for (i = 1; i < niocount; i++) {
 387                 if (nb[i].offset >= next_indir) {
 388                         nbitmaps++;     /* additional indirect */
 389                         next_indir = nb[i].offset +
 390                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 391                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 392                         nbitmaps++;     /* additional indirect */
 393                 }
 394                 nbitmaps += blockpp;    /* each leaf in different group? */
 395         }
 396
 397         ngdblocks = nbitmaps;
 398         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 399                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 400         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 401                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 402
 403         needed += nbitmaps + ngdblocks;
 404
 405         /* last_rcvd update */
 406         needed += EXT3_DATA_TRANS_BLOCKS(sb);
 407
 408 #if defined(CONFIG_QUOTA)
 409         /* We assume that there will be 1 bit set in s_dquot.flags for each
 410          * quota file that is active.  This is at least true for now.
 411          */
 412         needed += hweight32(ll_sb_any_quota_active(sb)) *
 413                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 414 #endif
 415
 416         return needed;
 417 }
 418
 419 /* We have to start a huge journal transaction here to hold all of the
 420  * metadata for the pages being written here.  This is necessitated by
 421  * the fact that we do lots of prepare_write operations before we do
 422  * any of the matching commit_write operations, so even if we split
 423  * up to use "smaller" transactions none of them could complete until
 424  * all of them were opened.  By having a single journal transaction,
 425  * we eliminate duplicate reservations for common blocks like the
 426  * superblock and group descriptors or bitmaps.
 427  *
 428  * We will start the transaction here, but each prepare_write will
 429  * add a refcount to the transaction, and each commit_write will
 430  * remove a refcount.  The transaction will be closed when all of
 431  * the pages have been written.
 432  */
 433 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 434                                    int niocount, struct niobuf_local *nb,
 435                                    void *desc_private, int logs)
 436 {
 437         journal_t *journal;
 438         handle_t *handle;
 439         int needed;
 440         ENTRY;
 441
 442         LASSERT(current->journal_info == desc_private);
 443         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 444         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 445
 446         /* The number of blocks we could _possibly_ dirty can very large.
 447          * We reduce our request if it is absurd (and we couldn't get that
 448          * many credits for a single handle anyways).
 449          *
 450          * At some point we have to limit the size of I/Os sent at one time,
 451          * increase the size of the journal, or we have to calculate the
 452          * actual journal requirements more carefully by checking all of
 453          * the blocks instead of being maximally pessimistic.  It remains to
 454          * be seen if this is a real problem or not.
 455          */
 456         if (needed > journal->j_max_transaction_buffers) {
 457                 CERROR("want too many journal credits (%d) using %d instead\n",
 458                        needed, journal->j_max_transaction_buffers);
 459                 needed = journal->j_max_transaction_buffers;
 460         }
 461
 462         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 463         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 464         if (IS_ERR(handle)) {
 465                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 466                        PTR_ERR(handle));
 467         } else {
 468                 LASSERT(handle->h_buffer_credits >= needed);
 469                 LASSERT(current->journal_info == handle);
 470         }
 471
 472         RETURN(handle);
 473 }
 474
 475 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 476 {
 477        handle_t *handle = h;
 478
 479        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 480        if (nblocks == 0) {
 481                handle->h_buffer_credits = 0;
 482                CWARN("setting credits of handle %p to zero by request\n", h);
 483        }
 484
 485        if (handle->h_buffer_credits > nblocks)
 486                 return 0;
 487        if (ext3_journal_extend(handle, nblocks) == 0)
 488                 return 0;
 489
 490        ext3_mark_inode_dirty(handle, inode);
 491        return ext3_journal_restart(handle, nblocks);
 492 }
 493
 494 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 495 {
 496         int rc;
 497         handle_t *handle = h;
 498
 499         LASSERT(current->journal_info == handle);
 500         if (force_sync)
 501                 handle->h_sync = 1; /* recovery likes this */
 502
 503         rc = ext3_journal_stop(handle);
 504
 505         return rc;
 506 }
 507
 508 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 509                                     void **wait_handle)
 510 {
 511         unsigned long tid;
 512         transaction_t *transaction;
 513         handle_t *handle = h;
 514         journal_t *journal;
 515         int rc;
 516
 517         LASSERT(current->journal_info == handle);
 518
 519         transaction = handle->h_transaction;
 520         journal = transaction->t_journal;
 521         tid = transaction->t_tid;
 522         /* we don't want to be blocked */
 523         handle->h_sync = 0;
 524         rc = ext3_journal_stop(handle);
 525         if (rc) {
 526                 CERROR("error while stopping transaction: %d\n", rc);
 527                 return rc;
 528         }
 529         fsfilt_log_start_commit(journal, tid);
 530
 531         *wait_handle = (void *) tid;
 532         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 533         return 0;
 534 }
 535
 536 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 537 {
 538         journal_t *journal = EXT3_JOURNAL(inode);
 539         tid_t tid = (tid_t)(long)h;
 540
 541         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 542         if (unlikely(is_journal_aborted(journal)))
 543                 return -EIO;
 544
 545         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 546
 547         if (unlikely(is_journal_aborted(journal)))
 548                 return -EIO;
 549         return 0;
 550 }
 551
 552 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 553                                struct iattr *iattr, int do_trunc)
 554 {
 555         struct inode *inode = dentry->d_inode;
 556         int rc = 0;
 557
 558 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
 559         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
 560          * timestamps to appear to be in the far future, due old timestamp
 561          * being stored on disk as an unsigned value.  This fixes up any
 562          * bad values held by the client before storing them on disk,
 563          * and ensures any timestamp updates are correct.  LU-1042 */
 564         if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
 565                      !(iattr->ia_valid & ATTR_ATIME))) {
 566                 iattr->ia_valid |= ATTR_ATIME;
 567                 LTIME_S(iattr->ia_atime) = 0;
 568         }
 569         if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
 570                      !(iattr->ia_valid & ATTR_MTIME))) {
 571                 iattr->ia_valid |= ATTR_MTIME;
 572                 LTIME_S(iattr->ia_mtime) = 0;
 573         }
 574         if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
 575                       LTIME_S(inode->i_ctime) == 0) &&
 576                      !(iattr->ia_valid & ATTR_CTIME))) {
 577                 iattr->ia_valid |= ATTR_CTIME;
 578                 LTIME_S(iattr->ia_ctime) = 0;
 579         }
 580 #else
 581 #warning "remove old LU-221/LU-1042 workaround code"
 582 #endif
 583
 584         /* When initializating timestamps for new inodes, use the filesystem
 585          * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
 586          * that this is potentially an invalid inode.  Files with an old ctime
 587          * migrated to a newly-formatted OST with a newer s_mkfs_time will not
 588          * hit this check, since it is only for ctime == 0.  LU-1010/LU-1042 */
 589         if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
 590                 LTIME_S(iattr->ia_ctime) =
 591                         EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
 592
 593         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 594          * We are already writing the inode to disk as part of this
 595          * transaction and want to avoid a lot of extra inode writeout
 596          * later on. b=9828 */
 597         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 598                 /* ATTR_SIZE would invoke truncate: clear it */
 599                 iattr->ia_valid &= ~ATTR_SIZE;
 600                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 601                 i_size_write(inode, iattr->ia_size);
 602
 603                 if (iattr->ia_valid & ATTR_UID)
 604                         inode->i_uid = iattr->ia_uid;
 605                 if (iattr->ia_valid & ATTR_GID)
 606                         inode->i_gid = iattr->ia_gid;
 607                 if (iattr->ia_valid & ATTR_ATIME)
 608                         inode->i_atime = iattr->ia_atime;
 609                 if (iattr->ia_valid & ATTR_MTIME)
 610                         inode->i_mtime = iattr->ia_mtime;
 611                 if (iattr->ia_valid & ATTR_CTIME)
 612                         inode->i_ctime = iattr->ia_ctime;
 613                 if (iattr->ia_valid & ATTR_MODE) {
 614                         inode->i_mode = iattr->ia_mode;
 615
 616                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 617                             !cfs_capable(CFS_CAP_FSETID))
 618                                 inode->i_mode &= ~S_ISGID;
 619                 }
 620
 621                 inode->i_sb->s_op->dirty_inode(inode);
 622
 623                 goto out;
 624         }
 625
 626         /* Don't allow setattr to change file type */
 627         if (iattr->ia_valid & ATTR_MODE)
 628                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 629                                  (iattr->ia_mode & ~S_IFMT);
 630
 631         /* We set these flags on the client, but have already checked perms
 632          * so don't confuse inode_change_ok. */
 633         iattr->ia_valid &= ~TIMES_SET_FLAGS;
 634
 635         if (inode->i_op->setattr) {
 636                 rc = inode->i_op->setattr(dentry, iattr);
 637         } else {
 638 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
 639                 rc = inode_change_ok(inode, iattr);
 640                 if (!rc)
 641 #endif
 642                         rc = simple_setattr(dentry, iattr);
 643         }
 644
 645  out:
 646         RETURN(rc);
 647 }
 648
 649 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 650                                  unsigned int cmd, unsigned long arg)
 651 {
 652         int rc = 0;
 653         ENTRY;
 654
 655         /* FIXME: Can't do this because of nested transaction deadlock */
 656         if (cmd == EXT3_IOC_SETFLAGS) {
 657                 /* We can't enable data journaling on OST objects, because
 658                 * this forces the transaction to be closed in order to
 659                 * flush the journal, but the caller will already have a
 660                 * compound transaction open to update the last_rcvd file,
 661                 * and this thread would deadlock trying to set the flag. */
 662                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 663                         CERROR("can't set data journal flag on file\n");
 664                         RETURN(-EPERM);
 665                 }
 666                 /* Because the MDS does not see the EXTENTS_FL set on the
 667                  * OST objects, mask this flag into all set flags.  It is
 668                  * not legal to clear this flag in any case, so we are not
 669                  * changing the functionality by doing this.  b=22911 */
 670                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 671         }
 672
 673         /* ext4_ioctl does not have a inode argument */
 674         if (inode->i_fop->unlocked_ioctl)
 675                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 676         else
 677                 RETURN(-ENOTTY);
 678
 679         RETURN(rc);
 680 }
 681
 682 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 683                               void *lmm, int lmm_size, const char *name)
 684 {
 685         int rc;
 686
 687         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 688
 689         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 690                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 691
 692         if (rc && rc != -EROFS)
 693                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 694                        inode->i_ino, rc);
 695         return rc;
 696 }
 697
 698 /* Must be called with i_mutex held */
 699 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 700                               const char *name)
 701 {
 702         int rc;
 703
 704         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 705
 706         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 707                             name, lmm, lmm_size);
 708
 709         /* This gives us the MD size */
 710         if (lmm == NULL)
 711                 return (rc == -ENODATA) ? 0 : rc;
 712
 713         if (rc < 0) {
 714                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 715                        EXT3_XATTR_INDEX_TRUSTED, name,
 716                        inode->i_ino, rc);
 717                 memset(lmm, 0, lmm_size);
 718                 return (rc == -ENODATA) ? 0 : rc;
 719         }
 720
 721         return rc;
 722 }
 723
 724 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 725 {
 726         submit_bio(rw, bio);
 727         return 0;
 728 }
 729
 730 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 731                                     loff_t *off)
 732 {
 733         struct inode *inode = file->f_dentry->d_inode;
 734         int rc = 0;
 735
 736         if (S_ISREG(inode->i_mode))
 737                 rc = file->f_op->read(file, buf, count, off);
 738         else {
 739                 const int blkbits = inode->i_sb->s_blocksize_bits;
 740                 const int blksize = inode->i_sb->s_blocksize;
 741
 742                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 743                        (unsigned long)count, inode->i_ino, *off);
 744                 while (count > 0) {
 745                         struct buffer_head *bh;
 746
 747                         bh = NULL;
 748                         if (*off < i_size_read(inode)) {
 749                                 int err = 0;
 750
 751                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 752                                                 0, &err);
 753
 754                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 755
 756                                 if (bh) {
 757                                         memcpy(buf, bh->b_data, blksize);
 758                                         brelse(bh);
 759                                 } else if (err) {
 760                                         /* XXX in theory we should just fake
 761                                          * this buffer and continue like ext3,
 762                                          * especially if this is a partial read
 763                                          */
 764                                         CERROR("error read dir %lu+%llu: %d\n",
 765                                                inode->i_ino, *off, err);
 766                                         RETURN(err);
 767                                 }
 768                         }
 769                         if (!bh) {
 770                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 771
 772                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 773                                 memset(fake, 0, sizeof(*fake));
 774                                 fake->rec_len = cpu_to_le16(blksize);
 775                         }
 776                         count -= blksize;
 777                         buf += blksize;
 778                         *off += blksize;
 779                         rc += blksize;
 780                 }
 781         }
 782
 783         return rc;
 784 }
 785
 786 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 787 static void fsfilt_ext3_cb_func(struct super_block *sb,
 788                                 struct journal_callback *jcb, int error)
 789 #else
 790 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 791 #endif
 792 {
 793         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 794
 795         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 796
 797         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 798 }
 799
 800 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 801                                       void *handle, fsfilt_cb_t cb_func,
 802                                       void *cb_data)
 803 {
 804         struct fsfilt_cb_data *fcb;
 805
 806         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 807         if (fcb == NULL)
 808                 RETURN(-ENOMEM);
 809
 810         fcb->cb_func = cb_func;
 811         fcb->cb_obd = obd;
 812         fcb->cb_last_rcvd = last_rcvd;
 813         fcb->cb_data = cb_data;
 814
 815         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 816         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 817
 818         return 0;
 819 }
 820
 821 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 822 {
 823         struct kstatfs sfs;
 824         int rc;
 825
 826         memset(&sfs, 0, sizeof(sfs));
 827         rc = sb->s_op->statfs(sb->s_root, &sfs);
 828         statfs_pack(osfs, &sfs);
 829         return rc;
 830 }
 831
 832 static int fsfilt_ext3_sync(struct super_block *sb)
 833 {
 834         return ext3_force_commit(sb);
 835 }
 836
 837 #ifndef EXT3_EXTENTS_FL
 838 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 839 #endif
 840
 841 #ifndef EXT_ASSERT
 842 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 843 #endif
 844
 845 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 846 #define ext3_ext_base                   inode
 847 #define ext3_ext_base2inode(inode)      (inode)
 848 #define EXT_DEPTH(inode)                ext_depth(inode)
 849 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 850                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 851
 852 struct bpointers {
 853         unsigned long *blocks;
 854         int *created;
 855         unsigned long start;
 856         int num;
 857         int init_num;
 858         int create;
 859 };
 860
 861 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 862                                unsigned long block, int *aflags)
 863 {
 864         struct ext3_inode_info *ei = EXT3_I(inode);
 865         unsigned long bg_start;
 866         unsigned long colour;
 867         int depth;
 868
 869         if (path) {
 870                 struct ext3_extent *ex;
 871                 depth = path->p_depth;
 872
 873                 /* try to predict block placement */
 874                 if ((ex = path[depth].p_ext))
 875                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 876
 877                 /* it looks index is empty
 878                  * try to find starting from index itself */
 879                 if (path[depth].p_bh)
 880                         return path[depth].p_bh->b_blocknr;
 881         }
 882
 883         /* OK. use inode's group */
 884         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 885                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 886         colour = (current->pid % 16) *
 887                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 888         return bg_start + colour + block;
 889 }
 890
 891 #define ll_unmap_underlying_metadata(sb, blocknr) \
 892         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 893
 894 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 895 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 896                                 struct ext3_ext_path *path, unsigned long block,
 897                                 unsigned long *count, int *err)
 898 {
 899         unsigned long pblock, goal;
 900         int aflags = 0;
 901         struct inode *inode = ext3_ext_base2inode(base);
 902
 903         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 904         aflags |= 2; /* block have been already reserved */
 905         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 906         return pblock;
 907
 908 }
 909 #else
 910 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 911                                 struct ext3_ext_path *path, unsigned long block,
 912                                 unsigned long *count, int *err)
 913 {
 914         struct inode *inode = ext3_ext_base2inode(base);
 915         struct ext3_allocation_request ar;
 916         unsigned long pblock;
 917         int aflags;
 918
 919         /* find neighbour allocated blocks */
 920         ar.lleft = block;
 921         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 922         if (*err)
 923                 return 0;
 924         ar.lright = block;
 925         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 926         if (*err)
 927                 return 0;
 928
 929         /* allocate new block */
 930         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 931         ar.inode = inode;
 932         ar.logical = block;
 933         ar.len = *count;
 934         ar.flags = EXT3_MB_HINT_DATA;
 935         pblock = ext3_mb_new_blocks(handle, &ar, err);
 936         *count = ar.len;
 937         return pblock;
 938 }
 939 #endif
 940
 941 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 942                                   struct ext3_ext_path *path,
 943                                   struct ext3_ext_cache *cex,
 944 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 945                                    struct ext3_extent *ex,
 946 #endif
 947                                   void *cbdata)
 948 {
 949         struct bpointers *bp = cbdata;
 950         struct inode *inode = ext3_ext_base2inode(base);
 951         struct ext3_extent nex;
 952         unsigned long pblock;
 953         unsigned long tgen;
 954         int err, i;
 955         unsigned long count;
 956         handle_t *handle;
 957
 958         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 959                 err = EXT_CONTINUE;
 960                 goto map;
 961         }
 962
 963         if (bp->create == 0) {
 964                 i = 0;
 965                 if (cex->ec_block < bp->start)
 966                         i = bp->start - cex->ec_block;
 967                 if (i >= cex->ec_len)
 968                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 969                                         i, cex->ec_len);
 970                 for (; i < cex->ec_len && bp->num; i++) {
 971                         *(bp->created) = 0;
 972                         bp->created++;
 973                         *(bp->blocks) = 0;
 974                         bp->blocks++;
 975                         bp->num--;
 976                         bp->start++;
 977                 }
 978
 979                 return EXT_CONTINUE;
 980         }
 981
 982         tgen = EXT_GENERATION(base);
 983         count = ext3_ext_calc_credits_for_insert(base, path);
 984
 985         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 986         if (IS_ERR(handle)) {
 987                 return PTR_ERR(handle);
 988         }
 989
 990         if (tgen != EXT_GENERATION(base)) {
 991                 /* the tree has changed. so path can be invalid at moment */
 992                 ext3_journal_stop(handle);
 993                 return EXT_REPEAT;
 994         }
 995
 996         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
 997          * protected by i_data_sem as whole. so we patch it to store
 998          * generation to path and now verify the tree hasn't changed */
 999         down_write((&EXT4_I(inode)->i_data_sem));
1000
1001         /* validate extent, make sure the extent tree does not changed */
1002         if (EXT_GENERATION(base) != path[0].p_generation) {
1003                 /* cex is invalid, try again */
1004                 up_write(&EXT4_I(inode)->i_data_sem);
1005                 ext3_journal_stop(handle);
1006                 return EXT_REPEAT;
1007         }
1008
1009         count = cex->ec_len;
1010         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1011         if (!pblock)
1012                 goto out;
1013         EXT_ASSERT(count <= cex->ec_len);
1014
1015         /* insert new extent */
1016         nex.ee_block = cpu_to_le32(cex->ec_block);
1017         ext3_ext_store_pblock(&nex, pblock);
1018         nex.ee_len = cpu_to_le16(count);
1019         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1020         if (err) {
1021                 /* free data blocks we just allocated */
1022                 /* not a good idea to call discard here directly,
1023                  * but otherwise we'd need to call it every free() */
1024 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1025                 ext3_mb_discard_inode_preallocations(inode);
1026 #endif
1027                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1028                                  cpu_to_le16(nex.ee_len), 0);
1029                 goto out;
1030         }
1031
1032         /*
1033          * Putting len of the actual extent we just inserted,
1034          * we are asking ext3_ext_walk_space() to continue
1035          * scaning after that block
1036          */
1037         cex->ec_len = le16_to_cpu(nex.ee_len);
1038         cex->ec_start = ext_pblock(&nex);
1039         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1040         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1041
1042 out:
1043         up_write((&EXT4_I(inode)->i_data_sem));
1044         ext3_journal_stop(handle);
1045 map:
1046         if (err >= 0) {
1047                 /* map blocks */
1048                 if (bp->num == 0) {
1049                         CERROR("hmm. why do we find this extent?\n");
1050                         CERROR("initial space: %lu:%u\n",
1051                                 bp->start, bp->init_num);
1052                         CERROR("current extent: %u/%u/%llu %d\n",
1053                                 cex->ec_block, cex->ec_len,
1054                                 (unsigned long long)cex->ec_start,
1055                                 cex->ec_type);
1056                 }
1057                 i = 0;
1058                 if (cex->ec_block < bp->start)
1059                         i = bp->start - cex->ec_block;
1060                 if (i >= cex->ec_len)
1061                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1062                                         i, cex->ec_len);
1063                 for (; i < cex->ec_len && bp->num; i++) {
1064                         *(bp->blocks) = cex->ec_start + i;
1065                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1066                                 *(bp->created) = 0;
1067                         } else {
1068                                 *(bp->created) = 1;
1069                                 /* unmap any possible underlying metadata from
1070                                  * the block device mapping.  bug 6998. */
1071                                 ll_unmap_underlying_metadata(inode->i_sb,
1072                                                              *(bp->blocks));
1073                         }
1074                         bp->created++;
1075                         bp->blocks++;
1076                         bp->num--;
1077                         bp->start++;
1078                 }
1079         }
1080         return err;
1081 }
1082
1083 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1084                        unsigned long num, unsigned long *blocks,
1085                        int *created, int create)
1086 {
1087         struct ext3_ext_base *base = inode;
1088         struct bpointers bp;
1089         int err;
1090
1091         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1092                block, block + num - 1, (unsigned) inode->i_ino);
1093
1094         bp.blocks = blocks;
1095         bp.created = created;
1096         bp.start = block;
1097         bp.init_num = bp.num = num;
1098         bp.create = create;
1099
1100         err = fsfilt_ext3_ext_walk_space(base, block, num,
1101                                          ext3_ext_new_extent_cb, &bp);
1102         ext3_ext_invalidate_cache(base);
1103
1104         return err;
1105 }
1106
1107 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1108                                     int pages, unsigned long *blocks,
1109                                     int *created, int create)
1110 {
1111         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1112         int rc = 0, i = 0;
1113         struct page *fp = NULL;
1114         int clen = 0;
1115
1116         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1117                 inode->i_ino, pages, (*page)->index);
1118
1119         /* pages are sorted already. so, we just have to find
1120          * contig. space and process them properly */
1121         while (i < pages) {
1122                 if (fp == NULL) {
1123                         /* start new extent */
1124                         fp = *page++;
1125                         clen = 1;
1126                         i++;
1127                         continue;
1128                 } else if (fp->index + clen == (*page)->index) {
1129                         /* continue the extent */
1130                         page++;
1131                         clen++;
1132                         i++;
1133                         continue;
1134                 }
1135
1136                 /* process found extent */
1137                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1138                                         clen * blocks_per_page, blocks,
1139                                         created, create);
1140                 if (rc)
1141                         GOTO(cleanup, rc);
1142
1143                 /* look for next extent */
1144                 fp = NULL;
1145                 blocks += blocks_per_page * clen;
1146                 created += blocks_per_page * clen;
1147         }
1148
1149         if (fp)
1150                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1151                                         clen * blocks_per_page, blocks,
1152                                         created, create);
1153 cleanup:
1154         return rc;
1155 }
1156
1157 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1158                                unsigned long *blocks, int *created, int create);
1159 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1160                                    int pages, unsigned long *blocks,
1161                                    int *created, int create)
1162 {
1163         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1164         unsigned long *b;
1165         int rc = 0, i, *cr;
1166
1167         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1168                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1169                 if (rc) {
1170                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1171                                inode->i_ino, *b, *cr, create, rc);
1172                         break;
1173                 }
1174
1175                 b += blocks_per_page;
1176                 cr += blocks_per_page;
1177         }
1178         return rc;
1179 }
1180
1181 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1182                                 int pages, unsigned long *blocks,
1183                                 int *created, int create,
1184                                 cfs_mutex_t *optional_mutex)
1185 {
1186         int rc;
1187
1188         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1189                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1190                                                      blocks, created, create);
1191                 return rc;
1192         }
1193         if (optional_mutex != NULL)
1194                 cfs_mutex_lock(optional_mutex);
1195         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1196                                             created, create);
1197         if (optional_mutex != NULL)
1198                 cfs_mutex_unlock(optional_mutex);
1199
1200         return rc;
1201 }
1202
1203 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1204 {
1205         unsigned long block;
1206         struct buffer_head *bh;
1207         int err, blocksize, csize, boffs, osize = size;
1208
1209         /* prevent reading after eof */
1210         spin_lock(&inode->i_lock);
1211         if (i_size_read(inode) < *offs + size) {
1212                 size = i_size_read(inode) - *offs;
1213                 spin_unlock(&inode->i_lock);
1214                 if (size < 0) {
1215                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1216                                i_size_read(inode), *offs);
1217                         return -EBADR;
1218                 } else if (size == 0) {
1219                         return 0;
1220                 }
1221         } else {
1222                 spin_unlock(&inode->i_lock);
1223         }
1224
1225         blocksize = 1 << inode->i_blkbits;
1226
1227         while (size > 0) {
1228                 block = *offs >> inode->i_blkbits;
1229                 boffs = *offs & (blocksize - 1);
1230                 csize = min(blocksize - boffs, size);
1231                 bh = ext3_bread(NULL, inode, block, 0, &err);
1232                 if (!bh) {
1233                         CERROR("can't read block: %d\n", err);
1234                         return err;
1235                 }
1236
1237                 memcpy(buf, bh->b_data + boffs, csize);
1238                 brelse(bh);
1239
1240                 *offs += csize;
1241                 buf += csize;
1242                 size -= csize;
1243         }
1244         return osize;
1245 }
1246 EXPORT_SYMBOL(fsfilt_ext3_read);
1247
1248 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1249                                    int size, loff_t *offs)
1250 {
1251         int rc;
1252         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1253         if (rc > 0)
1254                 rc = 0;
1255         return rc;
1256 }
1257
1258 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1259                                 loff_t *offs, handle_t *handle)
1260 {
1261         struct buffer_head *bh = NULL;
1262         loff_t old_size = i_size_read(inode), offset = *offs;
1263         loff_t new_size = i_size_read(inode);
1264         unsigned long block;
1265         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1266
1267         while (bufsize > 0) {
1268                 if (bh != NULL)
1269                         brelse(bh);
1270
1271                 block = offset >> inode->i_blkbits;
1272                 boffs = offset & (blocksize - 1);
1273                 size = min(blocksize - boffs, bufsize);
1274                 bh = ext3_bread(handle, inode, block, 1, &err);
1275                 if (!bh) {
1276                         CERROR("can't read/create block: %d\n", err);
1277                         break;
1278                 }
1279
1280                 err = ext3_journal_get_write_access(handle, bh);
1281                 if (err) {
1282                         CERROR("journal_get_write_access() returned error %d\n",
1283                                err);
1284                         break;
1285                 }
1286                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1287                 memcpy(bh->b_data + boffs, buf, size);
1288                 err = ext3_journal_dirty_metadata(handle, bh);
1289                 if (err) {
1290                         CERROR("journal_dirty_metadata() returned error %d\n",
1291                                err);
1292                         break;
1293                 }
1294                 if (offset + size > new_size)
1295                         new_size = offset + size;
1296                 offset += size;
1297                 bufsize -= size;
1298                 buf += size;
1299         }
1300         if (bh)
1301                 brelse(bh);
1302
1303         /* correct in-core and on-disk sizes */
1304         if (new_size > i_size_read(inode)) {
1305                 spin_lock(&inode->i_lock);
1306                 if (new_size > i_size_read(inode))
1307                         i_size_write(inode, new_size);
1308                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1309                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1310                 if (i_size_read(inode) > old_size) {
1311                         spin_unlock(&inode->i_lock);
1312                         mark_inode_dirty(inode);
1313                 } else {
1314                         spin_unlock(&inode->i_lock);
1315                 }
1316         }
1317
1318         if (err == 0)
1319                 *offs = offset;
1320         return err;
1321 }
1322 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1323
1324 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1325                                     loff_t *offs, int force_sync)
1326 {
1327         struct inode *inode = file->f_dentry->d_inode;
1328         handle_t *handle;
1329         int err, block_count = 0, blocksize;
1330
1331         /* Determine how many transaction credits are needed */
1332         blocksize = 1 << inode->i_blkbits;
1333         block_count = (*offs & (blocksize - 1)) + bufsize;
1334         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1335
1336         handle = ext3_journal_start(inode,
1337                         block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1338         if (IS_ERR(handle)) {
1339                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1340                        block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
1341                        bufsize);
1342                 return PTR_ERR(handle);
1343         }
1344
1345         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1346
1347         if (!err && force_sync)
1348                 handle->h_sync = 1; /* recovery likes this */
1349
1350         ext3_journal_stop(handle);
1351
1352         return err;
1353 }
1354
1355 static int fsfilt_ext3_setup(struct super_block *sb)
1356 {
1357 #if defined(HAVE_QUOTA_SUPPORT) || defined(S_PDIROPS)
1358         struct ext3_sb_info *sbi = EXT3_SB(sb);
1359 #if 0
1360         sbi->dx_lock = fsfilt_ext3_dx_lock;
1361         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1362 #endif
1363 #endif
1364         if (!EXT3_HAS_COMPAT_FEATURE(sb,
1365                                 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1366                 CERROR("ext3 mounted without journal\n");
1367                 return -EINVAL;
1368         }
1369
1370 #ifdef S_PDIROPS
1371         CWARN("Enabling PDIROPS\n");
1372         set_opt(sbi->s_mount_opt, PDIROPS);
1373         sb->s_flags |= S_PDIROPS;
1374 #endif
1375         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1376                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1377 #ifdef HAVE_QUOTA_SUPPORT
1378         /* enable journaled quota support */
1379         /* kfreed in ext3_put_super() */
1380         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1381         if (!sbi->s_qf_names[USRQUOTA])
1382                 return -ENOMEM;
1383         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1384         if (!sbi->s_qf_names[GRPQUOTA]) {
1385                 kfree(sbi->s_qf_names[USRQUOTA]);
1386                 sbi->s_qf_names[USRQUOTA] = NULL;
1387                 return -ENOMEM;
1388         }
1389         sbi->s_jquota_fmt = QFMT_LUSTRE;
1390         set_opt(sbi->s_mount_opt, QUOTA);
1391 #endif
1392         return 0;
1393 }
1394
1395 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1396    objects. Logs is number of logfiles to update */
1397 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1398 {
1399         if ( !fso ) {
1400                 switch(op) {
1401                 case FSFILT_OP_CREATE:
1402                                  /* directory leaf, index & indirect & EA*/
1403                         return 4 + 3 * logs;
1404                 case FSFILT_OP_UNLINK:
1405                         return 3 * logs;
1406                 }
1407         } else {
1408                 int i;
1409                 int needed = 0;
1410                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1411                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1412                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1413                 for (i = 0; i < op; i++, fso++) {
1414                         int nblocks = fso->fso_bufcnt * blockpp;
1415                         int ndindirect = min(nblocks, addrpp + 1);
1416                         int nindir = nblocks + ndindirect + 1;
1417
1418                         needed += nindir;
1419                 }
1420                 return needed + 3 * logs;
1421         }
1422
1423         return 0;
1424 }
1425
1426 #ifdef HAVE_QUOTA_SUPPORT
1427 #define DQINFO_COPY(out, in)                    \
1428 do {                                            \
1429         Q_COPY(out, in, dqi_bgrace);            \
1430         Q_COPY(out, in, dqi_igrace);            \
1431         Q_COPY(out, in, dqi_flags);             \
1432         Q_COPY(out, in, dqi_valid);             \
1433 } while (0)
1434
1435 #define DQBLK_COPY(out, in)                     \
1436 do {                                            \
1437         Q_COPY(out, in, dqb_bhardlimit);        \
1438         Q_COPY(out, in, dqb_bsoftlimit);        \
1439         Q_COPY(out, in, dqb_curspace);          \
1440         Q_COPY(out, in, dqb_ihardlimit);        \
1441         Q_COPY(out, in, dqb_isoftlimit);        \
1442         Q_COPY(out, in, dqb_curinodes);         \
1443         Q_COPY(out, in, dqb_btime);             \
1444         Q_COPY(out, in, dqb_itime);             \
1445         Q_COPY(out, in, dqb_valid);             \
1446 } while (0)
1447
1448 static int fsfilt_ext3_quotactl(struct super_block *sb,
1449                                 struct obd_quotactl *oqc)
1450 {
1451         int i, rc = 0, error = 0;
1452         const struct quotactl_ops *qcop;
1453         struct if_dqinfo *info;
1454         struct if_dqblk *dqblk;
1455         ENTRY;
1456
1457         if (!sb->s_qcop)
1458                 RETURN(-ENOSYS);
1459
1460         OBD_ALLOC_PTR(info);
1461         if (!info)
1462                 RETURN(-ENOMEM);
1463         OBD_ALLOC_PTR(dqblk);
1464         if (!dqblk) {
1465                 OBD_FREE_PTR(info);
1466                 RETURN(-ENOMEM);
1467         }
1468
1469         DQINFO_COPY(info, &oqc->qc_dqinfo);
1470         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1471
1472         qcop = sb->s_qcop;
1473         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1474                 for (i = 0; i < MAXQUOTAS; i++) {
1475                         if (!Q_TYPESET(oqc, i))
1476                                 continue;
1477
1478                         if (oqc->qc_cmd == Q_QUOTAON) {
1479                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1480
1481                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1482
1483                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1484                                                  name[i], 0);
1485                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1486                                 rc = ll_quota_off(sb, i, 0);
1487                         }
1488
1489                         if (rc == -EBUSY)
1490                                 error = rc;
1491                         else if (rc)
1492                                 GOTO(out, rc);
1493                 }
1494                 GOTO(out, rc ?: error);
1495         }
1496
1497         switch (oqc->qc_cmd) {
1498         case Q_GETOINFO:
1499         case Q_GETINFO:
1500                 if (!qcop->get_info)
1501                         GOTO(out, rc = -ENOSYS);
1502                 rc = qcop->get_info(sb, oqc->qc_type, info);
1503                 break;
1504         case Q_SETQUOTA:
1505         case Q_INITQUOTA:
1506                 if (!qcop->set_dqblk)
1507                         GOTO(out, rc = -ENOSYS);
1508                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1509                 break;
1510         case Q_GETOQUOTA:
1511         case Q_GETQUOTA:
1512                 if (!qcop->get_dqblk)
1513                         GOTO(out, rc = -ENOSYS);
1514                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1515                 if (!rc)
1516                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1517                 break;
1518         case Q_SYNC:
1519                 if (!sb->s_qcop->quota_sync)
1520                         GOTO(out, rc = -ENOSYS);
1521                 qcop->quota_sync(sb, oqc->qc_type);
1522                 break;
1523         case Q_FINVALIDATE:
1524                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1525                 for (i = 0; i < MAXQUOTAS; i++) {
1526                         struct file *fp;
1527                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1528
1529                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1530
1531                         if (!Q_TYPESET(oqc, i))
1532                                 continue;
1533
1534                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1535                         if (IS_ERR(fp)) {
1536                                 rc = PTR_ERR(fp);
1537                                 CERROR("error invalidating operational quota file"
1538                                        " %s (rc:%d)\n", name[i], rc);
1539                         } else {
1540                                 filp_close(fp, 0);
1541                         }
1542
1543                 }
1544                 break;
1545         default:
1546                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1547                 LBUG();
1548         }
1549 out:
1550         DQINFO_COPY(&oqc->qc_dqinfo, info);
1551         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1552
1553         OBD_FREE_PTR(info);
1554         OBD_FREE_PTR(dqblk);
1555
1556         if (rc)
1557                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1558                                 "failed: %d\n",
1559                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1560         RETURN(rc);
1561 }
1562
1563 struct chk_dqblk{
1564         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1565         cfs_list_t              dqb_list;        /** in list also */
1566         qid_t                   dqb_id;          /** uid/gid */
1567         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1568         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1569         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1570         qsize_t                 dqb_curspace;    /** current space */
1571         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1572         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1573         qsize_t                 dqb_curinodes;   /** current inodes */
1574         obd_time                dqb_btime;       /** block grace time */
1575         obd_time                dqb_itime;       /** inode grace time */
1576         __u32                   dqb_valid;       /** flag for above fields */
1577 };
1578
1579 static inline unsigned int chkquot_hash(qid_t id, int type)
1580                                         __attribute__((__const__));
1581
1582 static inline unsigned int chkquot_hash(qid_t id, int type)
1583 {
1584         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1585 }
1586
1587 static inline struct chk_dqblk *
1588 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1589 {
1590         cfs_hlist_node_t *node;
1591         struct chk_dqblk *cdqb;
1592
1593         cfs_hlist_for_each(node, head) {
1594                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1595                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1596                         return cdqb;
1597         }
1598
1599         return NULL;
1600 }
1601
1602 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1603 {
1604         struct chk_dqblk *cdqb;
1605
1606         OBD_ALLOC_PTR(cdqb);
1607         if (cdqb) {
1608                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1609                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1610                 cdqb->dqb_id = id;
1611                 cdqb->dqb_type = type;
1612         }
1613
1614         return cdqb;
1615 }
1616
1617 static struct chk_dqblk *
1618 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1619       cfs_list_t *list, qid_t id, int type, int first_check)
1620 {
1621         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1622         struct if_dqblk dqb;
1623         struct chk_dqblk *cdqb;
1624         int rc;
1625
1626         cdqb = find_chkquot(head, id, type);
1627         if (cdqb)
1628                 return cdqb;
1629
1630         cdqb = alloc_chkquot(id, type);
1631         if (!cdqb)
1632                 return NULL;
1633
1634         if (!first_check) {
1635                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1636                 if (rc) {
1637                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1638                                id, type, rc);
1639                 } else {
1640                         DQBLK_COPY(cdqb, &dqb);
1641                         cdqb->dqb_curspace = 0;
1642                         cdqb->dqb_curinodes = 0;
1643                 }
1644         }
1645
1646         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1647         cfs_list_add_tail(&cdqb->dqb_list, list);
1648
1649         return cdqb;
1650 }
1651
1652 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1653 {
1654         struct obd_quotactl *oqctl;
1655         int rc;
1656
1657         OBD_ALLOC_PTR(oqctl);
1658         if (!oqctl)
1659                 RETURN(-ENOMEM);
1660
1661         oqctl->qc_cmd = cmd;
1662         oqctl->qc_id = qfmt;
1663         oqctl->qc_type = type;
1664         rc = fsfilt_ext3_quotactl(sb, oqctl);
1665
1666         OBD_FREE_PTR(oqctl);
1667         return rc;
1668 }
1669
1670 static inline int read_old_dqinfo(struct super_block *sb, int type,
1671                                   struct if_dqinfo *dqinfo)
1672 {
1673         struct obd_quotactl *oqctl;
1674         int rc;
1675         ENTRY;
1676
1677         OBD_ALLOC_PTR(oqctl);
1678         if (!oqctl)
1679                 RETURN(-ENOMEM);
1680
1681         oqctl->qc_cmd = Q_GETINFO;
1682         oqctl->qc_type = type;
1683         rc = fsfilt_ext3_quotactl(sb, oqctl);
1684         if (!rc)
1685                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1686
1687         OBD_FREE_PTR(oqctl);
1688         RETURN(rc);
1689 }
1690
1691 struct qchk_ctxt {
1692         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1693         cfs_list_t              qckt_list;                 /* quotacheck list */
1694         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1695         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1696 };
1697
1698 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1699                            struct obd_quotactl *oqc)
1700 {
1701         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1702         loff_t size = 0;
1703         qid_t qid[MAXQUOTAS];
1704         int cnt, i, rc = 0;
1705
1706         if (!inode)
1707                 return 0;
1708
1709         qid[USRQUOTA] = inode->i_uid;
1710         qid[GRPQUOTA] = inode->i_gid;
1711
1712         if (S_ISDIR(inode->i_mode) ||
1713             S_ISREG(inode->i_mode) ||
1714             S_ISLNK(inode->i_mode))
1715                 size = inode_get_bytes(inode);
1716
1717         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1718                 if (!Q_TYPESET(oqc, cnt))
1719                         continue;
1720
1721                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1722                                 &qctxt->qckt_list, qid[cnt], cnt,
1723                                 qctxt->qckt_first_check[cnt]);
1724                 if (!cdqb[cnt]) {
1725                         rc = -ENOMEM;
1726                         break;
1727                 }
1728
1729                 cdqb[cnt]->dqb_curspace += size;
1730                 cdqb[cnt]->dqb_curinodes++;
1731         }
1732
1733         if (rc) {
1734                 for (i = 0; i < cnt; i++) {
1735                         if (!Q_TYPESET(oqc, i))
1736                                 continue;
1737                         LASSERT(cdqb[i]);
1738                         cdqb[i]->dqb_curspace -= size;
1739                         cdqb[i]->dqb_curinodes--;
1740                 }
1741         }
1742
1743         return rc;
1744 }
1745
1746 /* write dqinfo struct in a new quota file */
1747 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1748 {
1749         struct v2_disk_dqinfo dqinfo;
1750         __u32 blocks = V2_DQTREEOFF + 1;
1751         loff_t offset = V2_DQINFOOFF;
1752
1753         if (info) {
1754                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1755                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1756                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1757                                                ~DQF_INFO_DIRTY);
1758         } else {
1759                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1760                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1761                 dqinfo.dqi_flags = 0;
1762         }
1763
1764         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1765         dqinfo.dqi_free_blk = 0;
1766         dqinfo.dqi_free_entry = 0;
1767
1768         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1769 }
1770
1771 static int v3_write_dqheader(struct file *f, int type)
1772 {
1773         static const __u32 quota_magics[] = V2_INITQMAGICS;
1774         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1775         struct v2_disk_dqheader dqhead;
1776         loff_t offset = 0;
1777
1778         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1779         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1780
1781         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1782         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1783
1784         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1785 }
1786
1787 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1788                                   struct obd_quotactl *oqc)
1789 {
1790         int i, rc = 0;
1791         ENTRY;
1792
1793         for (i = 0; i < MAXQUOTAS; i++) {
1794                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1795                                          NULL : &qctxt->qckt_dqinfo[i];
1796                 struct file *file;
1797                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1798
1799                 if (!Q_TYPESET(oqc, i))
1800                         continue;
1801
1802                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1803
1804                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1805                 if (IS_ERR(file)) {
1806                         rc = PTR_ERR(file);
1807                         CERROR("can't create %s file: rc = %d\n",
1808                                name[i], rc);
1809                         GOTO(out, rc);
1810                 }
1811
1812                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1813                         CERROR("file %s is not regular", name[i]);
1814                         filp_close(file, 0);
1815                         GOTO(out, rc = -EINVAL);
1816                 }
1817
1818                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1819
1820                 rc = v3_write_dqheader(file, i);
1821                 if (rc) {
1822                         filp_close(file, 0);
1823                         GOTO(out, rc);
1824                 }
1825
1826                 rc = v3_write_dqinfo(file, i, info);
1827                 filp_close(file, 0);
1828                 if (rc)
1829                         GOTO(out, rc);
1830         }
1831
1832 out:
1833         RETURN(rc);
1834 }
1835
1836
1837 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1838                           struct chk_dqblk *cdqb)
1839 {
1840         struct obd_quotactl *oqc;
1841         long now;
1842         int rc;
1843         ENTRY;
1844
1845         OBD_ALLOC_PTR(oqc);
1846         if (!oqc)
1847                 RETURN(-ENOMEM);
1848
1849         now = cfs_time_current_sec();
1850
1851         if (cdqb->dqb_bsoftlimit &&
1852             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1853             !cdqb->dqb_btime)
1854                 cdqb->dqb_btime =
1855                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1856
1857         if (cdqb->dqb_isoftlimit &&
1858             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1859             !cdqb->dqb_itime)
1860                 cdqb->dqb_itime =
1861                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1862
1863         cdqb->dqb_valid = QIF_ALL;
1864
1865         oqc->qc_cmd = Q_SETQUOTA;
1866         oqc->qc_type = cdqb->dqb_type;
1867         oqc->qc_id = cdqb->dqb_id;
1868         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1869
1870         rc = fsfilt_ext3_quotactl(sb, oqc);
1871         OBD_FREE_PTR(oqc);
1872         RETURN(rc);
1873 }
1874
1875 static int prune_chkquots(struct super_block *sb,
1876                           struct qchk_ctxt *qctxt, int error)
1877 {
1878         struct chk_dqblk *cdqb, *tmp;
1879         int rc;
1880
1881         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1882                 if (!error) {
1883                         rc = commit_chkquot(sb, qctxt, cdqb);
1884                         if (rc)
1885                                 error = rc;
1886                 }
1887                 cfs_hlist_del_init(&cdqb->dqb_hash);
1888                 cfs_list_del(&cdqb->dqb_list);
1889                 OBD_FREE_PTR(cdqb);
1890         }
1891
1892         return error;
1893 }
1894
1895 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1896 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1897 #endif
1898
1899 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1900                                   struct obd_quotactl *oqc)
1901 {
1902         struct ext3_sb_info *sbi = EXT3_SB(sb);
1903         int i, group, uninit_feat = 0;
1904         struct qchk_ctxt *qctxt;
1905         struct buffer_head *bitmap_bh = NULL;
1906         unsigned long ino, inode_inuse;
1907         struct inode *inode;
1908         int rc = 0;
1909         ENTRY;
1910
1911         /* turn on quota and read dqinfo if existed */
1912         OBD_ALLOC_PTR(qctxt);
1913         if (!qctxt) {
1914                 oqc->qc_stat = -ENOMEM;
1915                 RETURN(-ENOMEM);
1916         }
1917
1918         for (i = 0; i < NR_DQHASH; i++)
1919                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1920         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
1921
1922         for (i = 0; i < MAXQUOTAS; i++) {
1923                 if (!Q_TYPESET(oqc, i))
1924                         continue;
1925
1926                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1927                 if (!rc || rc == -EBUSY) {
1928                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1929                         if (rc)
1930                                 GOTO(out, rc);
1931                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1932                         qctxt->qckt_first_check[i] = 1;
1933                 } else if (rc) {
1934                         GOTO(out, rc);
1935                 }
1936         }
1937         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
1938                 /* This filesystem supports the uninit group feature */
1939                 uninit_feat = 1;
1940
1941         /* number of inodes that have been allocated */
1942         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
1943                       percpu_counter_sum(&sbi->s_freeinodes_counter);
1944
1945         /* check quota and update in hash */
1946         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
1947              group++) {
1948                 unsigned long used_count = sbi->s_inodes_per_group;
1949
1950                 if (uninit_feat) {
1951                         struct ext3_group_desc *desc;
1952                         desc = ext3_get_group_desc(sb, group, NULL);
1953                         if (!desc)
1954                                 GOTO(out, -EIO);
1955
1956                         /* we don't really need to take the group lock here,
1957                          * but it may be useful if one day we support online
1958                          * quotacheck */
1959                         ext4_lock_group(sb, group);
1960                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
1961                                 /* no inode in use in this group, just skip it */
1962                                 ext3_unlock_group(sb, group);
1963                                 continue;
1964                         }
1965
1966                         used_count -= ext3_itable_unused_count(sb, desc);
1967                         ext3_unlock_group(sb, group);
1968                 }
1969
1970                 ino = group * sbi->s_inodes_per_group + 1;
1971                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
1972                 if (!bitmap_bh) {
1973                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
1974                                sb->s_id, group);
1975                         GOTO(out, -EIO);
1976                 }
1977
1978                 i = 0;
1979                 while (i < used_count &&
1980                        (i = ext3_find_next_bit(bitmap_bh->b_data,
1981                                                used_count, i)) < used_count) {
1982                         inode_inuse--;
1983                         i++;
1984                         ino = i + group * sbi->s_inodes_per_group;
1985                         if (ino < sbi->s_first_ino)
1986                                 continue;
1987                         inode = ext3_iget(sb, ino);
1988                         if (!inode || IS_ERR(inode))
1989                                 continue;
1990
1991                         rc = add_inode_quota(inode, qctxt, oqc);
1992                         iput(inode);
1993                         if (rc) {
1994                                 brelse(bitmap_bh);
1995                                 GOTO(out, rc);
1996                         }
1997                 }
1998                 brelse(bitmap_bh);
1999         }
2000
2001         /* read old quota limits from old quota file. (only for the user
2002          * has limits but hasn't file) */
2003 #ifdef HAVE_QUOTA_SUPPORT
2004         for (i = 0; i < MAXQUOTAS; i++) {
2005                 cfs_list_t id_list;
2006                 struct dquot_id *dqid, *tmp;
2007
2008                 if (!Q_TYPESET(oqc, i))
2009                         continue;
2010
2011                 if (qctxt->qckt_first_check[i])
2012                         continue;
2013
2014
2015                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2016                 CFS_INIT_LIST_HEAD(&id_list);
2017                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2018                 if (rc)
2019                         CERROR("read old limits failed. (rc:%d)\n", rc);
2020
2021                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2022                         cfs_list_del_init(&dqid->di_link);
2023
2024                         if (!rc)
2025                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2026                                       dqid->di_id, i,
2027                                       qctxt->qckt_first_check[i]);
2028                         OBD_FREE_PTR(dqid);
2029                 }
2030         }
2031 #endif
2032         /* turn off quota cause we are to dump chk_dqblk to files */
2033         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2034
2035         rc = create_new_quota_files(qctxt, oqc);
2036         if (rc)
2037                 GOTO(out, rc);
2038
2039         /* we use vfs functions to set dqblk, so turn quota on */
2040         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2041 out:
2042         /* dump and free chk_dqblk */
2043         rc = prune_chkquots(sb, qctxt, rc);
2044         OBD_FREE_PTR(qctxt);
2045
2046         /* turn off quota, `lfs quotacheck` will turn on when all
2047          * nodes quotacheck finish. */
2048         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2049
2050         oqc->qc_stat = rc;
2051         if (rc)
2052                 CERROR("quotacheck failed: rc = %d\n", rc);
2053
2054         RETURN(rc);
2055 }
2056
2057 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2058                                  int cmd)
2059 {
2060         int rc = 0;
2061         ENTRY;
2062
2063         if (lqi->qi_files[type] == NULL) {
2064                 CERROR("operate qinfo before it's enabled!\n");
2065                 RETURN(-ESRCH);
2066         }
2067
2068         switch (cmd) {
2069         case QFILE_CHK:
2070                 rc = lustre_check_quota_file(lqi, type);
2071                 break;
2072         case QFILE_RD_INFO:
2073                 rc = lustre_read_quota_info(lqi, type);
2074                 break;
2075         case QFILE_WR_INFO:
2076                 rc = lustre_write_quota_info(lqi, type);
2077                 break;
2078         case QFILE_INIT_INFO:
2079                 rc = lustre_init_quota_info(lqi, type);
2080                 break;
2081         case QFILE_CONVERT:
2082                 rc = -ENOTSUPP;
2083                 CERROR("quota CONVERT command is not supported\n");
2084                 break;
2085         default:
2086                 rc = -ENOTSUPP;
2087                 CERROR("Unsupported admin quota file cmd %d\n"
2088                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2089                        cmd);
2090                 break;
2091         }
2092         RETURN(rc);
2093 }
2094
2095 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2096                             cfs_list_t *list)
2097 {
2098         return lustre_get_qids(file, inode, type, list);
2099 }
2100
2101 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2102 {
2103         int rc = 0;
2104         ENTRY;
2105
2106         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2107                 CERROR("operate dquot before it's enabled!\n");
2108                 RETURN(-ESRCH);
2109         }
2110
2111         switch (cmd) {
2112         case QFILE_RD_DQUOT:
2113                 rc = lustre_read_dquot(dquot);
2114                 break;
2115         case QFILE_WR_DQUOT:
2116                 if (dquot->dq_dqb.dqb_ihardlimit ||
2117                     dquot->dq_dqb.dqb_isoftlimit ||
2118                     dquot->dq_dqb.dqb_bhardlimit ||
2119                     dquot->dq_dqb.dqb_bsoftlimit)
2120                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2121                 else
2122                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2123
2124                 rc = lustre_commit_dquot(dquot);
2125                 if (rc >= 0)
2126                         rc = 0;
2127                 break;
2128         default:
2129                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2130                 LBUG();
2131                 break;
2132         }
2133         RETURN(rc);
2134 }
2135
2136 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2137                                 struct inode *inode, int frags)
2138 {
2139         struct ext3_ext_base *base = inode;
2140         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2141          * metablocks at maxium b=16542 */
2142         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2143         return 0;
2144 }
2145
2146 #endif
2147
2148 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2149 {
2150         return (EXT3_SB(sb)->journal_bdev);
2151 }
2152 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2153
2154 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2155                           char *buf, int count, loff_t pos)
2156 {
2157         loff_t p = pos;
2158         int rc;
2159
2160         if (!f && !inode) {
2161                 CERROR("lustre_read_quota failed for no quota file!\n");
2162                 libcfs_debug_dumpstack(NULL);
2163                 return -EINVAL;
2164         }
2165
2166         /* Support for both adm and op quota files must be provided */
2167         if (f) {
2168                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2169                 rc = rc < 0 ? rc : p - pos;
2170         } else {
2171                 struct super_block *sb = inode->i_sb;
2172                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2173         }
2174         return rc;
2175 }
2176
2177 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2178 {
2179         loff_t p = pos;
2180         int rc;
2181
2182         /* Only adm quota files are supported, op updates are handled by vfs */
2183         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2184         rc = rc < 0 ? rc : p - pos;
2185
2186         return rc;
2187 }
2188
2189 void *lustre_quota_journal_start(struct inode *inode, int delete)
2190 {
2191         handle_t *handle;
2192         unsigned block_count;
2193
2194         if (delete) {
2195                 /* each indirect block (+4) may become free, attaching to the
2196                  * header list of free blocks (+1); the data block (+1) may
2197                  * become a free block (+0) or a block with free dqentries (+0) */
2198                 block_count = (4 + 1) + 1;
2199                 handle = ext3_journal_start(inode,
2200                              block_count*EXT3_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2201         } else {
2202                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2203                  * freeblk reusage with a header update (+1); dqentry is either reused
2204                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2205                  * a new block allocation (+1) with a header update (+1)              */
2206                 block_count = (4 + 1) + 3;
2207                 handle = ext3_journal_start(inode,
2208                              block_count*EXT3_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2209
2210         }
2211
2212         return handle;
2213 }
2214
2215 void lustre_quota_journal_stop(void *handle)
2216 {
2217         ext3_journal_stop((handle_t *)handle);
2218 }
2219
2220 static int ll_decode_fh_accept(void *context, struct dentry *de)
2221 {
2222         return 1;
2223 }
2224
2225 #ifdef HAVE_EXPORTFS_DECODE_FH
2226 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2227          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2228                             acceptable, context)
2229 #else
2230 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2231          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2232                                      type, acceptable, context)
2233 # define FILEID_INO32_GEN 1
2234 extern struct export_operations export_op_default;
2235 #endif
2236
2237 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2238                                       struct fsfilt_fid *fid, int ignore_gen)
2239 {
2240         struct inode  *inode;
2241         struct dentry *result;
2242
2243         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2244                                        ll_decode_fh_accept, NULL);
2245         if (IS_ERR(result)) {
2246                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2247                        fid->ino, fid->gen, PTR_ERR(result));
2248                 return result;
2249         }
2250
2251         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2252                fid->ino, fid->gen);
2253         inode = result->d_inode;
2254         if (inode == NULL)
2255                 goto err_out;
2256
2257         if (inode->i_nlink == 0 &&
2258             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2259                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2260                               " ctime -- this may indicate disk "
2261                               "corruption (inode: %lu, link: %lu, "
2262                               "count: %d)\n", inode->i_ino,
2263                               (unsigned long)inode->i_nlink,
2264                               atomic_read(&inode->i_count));
2265                 goto err_out;
2266         }
2267         if (fid->gen && inode->i_generation != fid->gen) {
2268                 /* we didn't find the right inode.. */
2269                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2270                        "count: %d, generation %u/%u\n",
2271                        inode->i_ino, (unsigned long)inode->i_nlink,
2272                        atomic_read(&inode->i_count), inode->i_generation,
2273                        fid->gen);
2274                 goto err_out;
2275         }
2276
2277         return result;
2278 err_out:
2279         l_dput(result);
2280         return ERR_PTR(-ENOENT);
2281 }
2282
2283 static struct fsfilt_operations fsfilt_ext3_ops = {
2284         .fs_type                = "ext3",
2285         .fs_owner               = THIS_MODULE,
2286         .fs_getlabel            = fsfilt_ext3_get_label,
2287         .fs_setlabel            = fsfilt_ext3_set_label,
2288         .fs_uuid                = fsfilt_ext3_uuid,
2289         .fs_start               = fsfilt_ext3_start,
2290         .fs_brw_start           = fsfilt_ext3_brw_start,
2291         .fs_extend              = fsfilt_ext3_extend,
2292         .fs_commit              = fsfilt_ext3_commit,
2293         .fs_commit_async        = fsfilt_ext3_commit_async,
2294         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2295         .fs_setattr             = fsfilt_ext3_setattr,
2296         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2297         .fs_set_md              = fsfilt_ext3_set_md,
2298         .fs_get_md              = fsfilt_ext3_get_md,
2299         .fs_readpage            = fsfilt_ext3_readpage,
2300         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2301         .fs_statfs              = fsfilt_ext3_statfs,
2302         .fs_sync                = fsfilt_ext3_sync,
2303         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2304         .fs_write_record        = fsfilt_ext3_write_record,
2305         .fs_read_record         = fsfilt_ext3_read_record,
2306         .fs_setup               = fsfilt_ext3_setup,
2307         .fs_send_bio            = fsfilt_ext3_send_bio,
2308         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2309 #ifdef HAVE_DISK_INODE_VERSION
2310         .fs_get_version         = fsfilt_ext3_get_version,
2311         .fs_set_version         = fsfilt_ext3_set_version,
2312 #endif
2313 #ifdef HAVE_QUOTA_SUPPORT
2314         .fs_quotactl            = fsfilt_ext3_quotactl,
2315         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2316         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2317         .fs_qids                = fsfilt_ext3_qids,
2318         .fs_dquot               = fsfilt_ext3_dquot,
2319         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2320 #endif
2321         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2322         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2323 };
2324
2325 static int __init fsfilt_ext3_init(void)
2326 {
2327         int rc;
2328
2329         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2330                                          sizeof(struct fsfilt_cb_data), 0, 0);
2331         if (!fcb_cache) {
2332                 CERROR("error allocating fsfilt journal callback cache\n");
2333                 GOTO(out, rc = -ENOMEM);
2334         }
2335
2336         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2337
2338         if (rc) {
2339                 int err = cfs_mem_cache_destroy(fcb_cache);
2340                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2341         }
2342 out:
2343         return rc;
2344 }
2345
2346 static void __exit fsfilt_ext3_exit(void)
2347 {
2348         int rc;
2349
2350         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2351         rc = cfs_mem_cache_destroy(fcb_cache);
2352         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2353 }
2354
2355 module_init(fsfilt_ext3_init);
2356 module_exit(fsfilt_ext3_exit);
2357
2358 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2359 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2360 MODULE_LICENSE("GPL");