lustre/lvfs/fsfilt_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/fsfilt_ext3.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/init.h>
  44 #include <linux/module.h>
  45 #include <linux/fs.h>
  46 #include <linux/slab.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/quotaops.h>
  49 #ifdef HAVE_EXT4_LDISKFS
  50 #include <ext4/ext4.h>
  51 #include <ext4/ext4_jbd2.h>
  52 #else
  53 #include <linux/jbd.h>
  54 #include <linux/ext3_fs.h>
  55 #include <linux/ext3_jbd.h>
  56 #endif
  57 #include <linux/version.h>
  58 #include <linux/bitops.h>
  59 #include <linux/quota.h>
  60 #ifdef HAVE_QUOTAIO_V1_H
  61 # include <linux/quotaio_v1.h>
  62 # include <linux/quotaio_v2.h>
  63 #else
  64 # include <quotaio_v1.h>
  65 # include <quotaio_v2.h>
  66 # include <quota_tree.h>
  67 # define V2_DQTREEOFF    QT_TREEOFF
  68 #endif
  69
  70 #if defined(HAVE_EXT3_XATTR_H)
  71 #include <ext3/xattr.h>
  72 #else
  73 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  74 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  75 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  76 #endif
  77
  78 #include <libcfs/libcfs.h>
  79 #include <lustre_fsfilt.h>
  80 #include <obd.h>
  81 #include <lustre_quota.h>
  82 #include <linux/lustre_compat25.h>
  83 #include <linux/lprocfs_status.h>
  84
  85 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  86 #ifdef HAVE_EXT4_LDISKFS
  87 #include <ext4/ext4_extents.h>
  88 #else
  89 #include <linux/ext3_extents.h>
  90 #endif
  91 #endif
  92
  93 #include "lustre_quota_fmt.h"
  94
  95 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  96 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  97 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  98 #else
  99 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 100 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 101 #endif
 102
 103 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 104 /* for kernels 2.6.18 and later */
 105 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 106 #else
 107 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 108 #endif
 109
 110 static cfs_mem_cache_t *fcb_cache;
 111
 112 struct fsfilt_cb_data {
 113         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 114         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 115         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 116         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 117         void *cb_data;                  /* MDS/OST completion function data */
 118 };
 119
 120 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 121 #define EXT3_XATTR_INDEX_TRUSTED        4
 122 #endif
 123
 124 #ifdef HAVE_EXT4_LDISKFS
 125 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 126 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 127 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 128 #else
 129 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 130 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 131 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 132 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 133 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 134 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 135 #endif
 136
 137 #ifndef ext3_find_next_bit
 138 #define ext3_find_next_bit           ext2_find_next_bit
 139 #endif
 140
 141 #ifndef ext2_find_next_bit
 142 #ifdef __LITTLE_ENDIAN
 143 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 144 #else
 145 error "Need implementation of find_next_bit on big-endian systems"
 146 #endif  /* __LITTLE_ENDIAN */
 147 #endif  /* !ext2_find_next_le_bit */
 148
 149 static char *fsfilt_ext3_get_label(struct super_block *sb)
 150 {
 151         return EXT3_SB(sb)->s_es->s_volume_name;
 152 }
 153
 154 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 155 {
 156         /* see e.g. fsfilt_ext3_write_record() */
 157         journal_t *journal;
 158         handle_t *handle;
 159         int err;
 160
 161         journal = EXT3_SB(sb)->s_journal;
 162         handle = ext3_journal_start_sb(sb, 1);
 163         if (IS_ERR(handle)) {
 164                 CERROR("can't start transaction\n");
 165                 return(PTR_ERR(handle));
 166         }
 167
 168         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 169         if (err)
 170                 goto out;
 171
 172         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 173                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 174
 175         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 176
 177 out:
 178         ext3_journal_stop(handle);
 179
 180         return(err);
 181 }
 182
 183 static char *fsfilt_ext3_uuid(struct super_block *sb)
 184 {
 185         return EXT3_SB(sb)->s_es->s_uuid;
 186 }
 187
 188 #ifdef HAVE_DISK_INODE_VERSION
 189
 190 static __u64 get_i_version(struct inode *inode)
 191 {
 192 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 193         return inode->i_version;
 194 #else
 195         return EXT3_I(inode)->i_fs_version;
 196 #endif
 197 }
 198
 199 static void set_i_version(struct inode *inode, __u64 new_version)
 200 {
 201 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 202         inode->i_version = new_version;
 203 #else
 204         (EXT3_I(inode))->i_fs_version = new_version;
 205 #endif
 206 }
 207
 208 /*
 209  * Get the 64-bit version for an inode.
 210  */
 211 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 212 {
 213         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 214                get_i_version(inode), inode->i_ino);
 215         return get_i_version(inode);
 216 }
 217
 218 /*
 219  * Set the 64-bit version and return the old version.
 220  */
 221 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 222 {
 223         __u64 old_version = get_i_version(inode);
 224
 225         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 226                new_version, old_version, inode->i_ino);
 227         set_i_version(inode, new_version);
 228         /* version is set after all inode operations are finished, so we should
 229          * mark it dirty here */
 230         inode->i_sb->s_op->dirty_inode(inode);
 231         return old_version;
 232 }
 233
 234 #endif
 235
 236 /*
 237  * We don't currently need any additional blocks for rmdir and
 238  * unlink transactions because we are storing the OST oa_id inside
 239  * the inode (which we will be changing anyways as part of this
 240  * transaction).
 241  */
 242 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 243                                int logs)
 244 {
 245         /* For updates to the last received file */
 246         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 247         journal_t *journal;
 248         void *handle;
 249
 250         if (current->journal_info) {
 251                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 252                        current->journal_info);
 253                 goto journal_start;
 254         }
 255
 256         switch(op) {
 257         case FSFILT_OP_RMDIR:
 258         case FSFILT_OP_UNLINK:
 259                 /* delete one file + create/update logs for each stripe */
 260                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 261                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 262                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 263                 break;
 264         case FSFILT_OP_RENAME:
 265                 /* modify additional directory */
 266                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 267                 /* no break */
 268         case FSFILT_OP_SYMLINK:
 269                 /* additional block + block bitmap + GDT for long symlink */
 270                 nblocks += 3;
 271                 /* no break */
 272         case FSFILT_OP_CREATE: {
 273 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 274                 static int warned;
 275                 if (!warned) {
 276                         if (!test_opt(inode->i_sb, EXTENTS)) {
 277                                 warned = 1;
 278                         } else if (((EXT3_I(inode)->i_flags &
 279                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 280                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 281                                 CWARN("extent-mapped directory found with "
 282                                       "ext3-based ldiskfs - contact "
 283                                       "http://bugzilla.lustre.org/\n");
 284                                 warned = 1;
 285                         }
 286                 }
 287 #endif
 288                 /* no break */
 289         }
 290         case FSFILT_OP_MKDIR:
 291         case FSFILT_OP_MKNOD:
 292                 /* modify one inode + block bitmap + GDT */
 293                 nblocks += 3;
 294                 /* no break */
 295         case FSFILT_OP_LINK:
 296                 /* modify parent directory */
 297                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 298                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 299                 /* create/update logs for each stripe */
 300                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 301                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 302                 break;
 303         case FSFILT_OP_SETATTR:
 304                 /* Setattr on inode */
 305                 nblocks += 1;
 306                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 307                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 308                 /* quota chown log for each stripe */
 309                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 310                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 311                 break;
 312         case FSFILT_OP_CANCEL_UNLINK:
 313                 /* blocks for log header bitmap update OR
 314                  * blocks for catalog header bitmap update + unlink of logs */
 315                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 316                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 317                 break;
 318         default: CERROR("unknown transaction start op %d\n", op);
 319                 LBUG();
 320         }
 321
 322         LASSERT(current->journal_info == desc_private);
 323         journal = EXT3_SB(inode->i_sb)->s_journal;
 324         if (nblocks > journal->j_max_transaction_buffers) {
 325                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 326                        nblocks, op, logs, journal->j_max_transaction_buffers);
 327                 nblocks = journal->j_max_transaction_buffers;
 328         }
 329
 330  journal_start:
 331         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 332         handle = ext3_journal_start(inode, nblocks);
 333
 334         if (!IS_ERR(handle))
 335                 LASSERT(current->journal_info == handle);
 336         else
 337                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 338                        op, nblocks, PTR_ERR(handle));
 339         return handle;
 340 }
 341
 342 /*
 343  * Calculate the number of buffer credits needed to write multiple pages in
 344  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 345  * doesn't have a nice API for calculating this sort of thing in advance.
 346  *
 347  * See comment above ext3_writepage_trans_blocks for details.  We assume
 348  * no data journaling is being done, but it does allow for all of the pages
 349  * being non-contiguous.  If we are guaranteed contiguous pages we could
 350  * reduce the number of (d)indirect blocks a lot.
 351  *
 352  * With N blocks per page and P pages, for each inode we have at most:
 353  * N*P indirect
 354  * min(N*P, blocksize/4 + 1) dindirect blocks
 355  * niocount tindirect
 356  *
 357  * For the entire filesystem, we have at most:
 358  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 359  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 360  * objcount inode blocks
 361  * 1 superblock
 362  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 363  *
 364  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 365  */
 366 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 367                                       int niocount, struct niobuf_local *nb)
 368 {
 369         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 370         __u64 next_indir;
 371         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 372         int nbitmaps = 0, ngdblocks;
 373         int needed = objcount + 1; /* inodes + superblock */
 374         int i, j;
 375
 376         for (i = 0, j = 0; i < objcount; i++, fso++) {
 377                 /* two or more dindirect blocks in case we cross boundary */
 378                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 379                                     nb[j].offset) >>
 380                                    sb->s_blocksize_bits) /
 381                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 382                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 383
 384                 /* leaf, indirect, tindirect blocks for first block */
 385                 nbitmaps += blockpp + 2;
 386
 387                 j += fso->fso_bufcnt;
 388         }
 389
 390         next_indir = nb[0].offset +
 391                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 392         for (i = 1; i < niocount; i++) {
 393                 if (nb[i].offset >= next_indir) {
 394                         nbitmaps++;     /* additional indirect */
 395                         next_indir = nb[i].offset +
 396                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 397                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 398                         nbitmaps++;     /* additional indirect */
 399                 }
 400                 nbitmaps += blockpp;    /* each leaf in different group? */
 401         }
 402
 403         ngdblocks = nbitmaps;
 404         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 405                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 406         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 407                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 408
 409         needed += nbitmaps + ngdblocks;
 410
 411         /* last_rcvd update */
 412         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 413
 414 #if defined(CONFIG_QUOTA)
 415         /* We assume that there will be 1 bit set in s_dquot.flags for each
 416          * quota file that is active.  This is at least true for now.
 417          */
 418         needed += hweight32(ll_sb_any_quota_active(sb)) *
 419                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 420 #endif
 421
 422         return needed;
 423 }
 424
 425 /* We have to start a huge journal transaction here to hold all of the
 426  * metadata for the pages being written here.  This is necessitated by
 427  * the fact that we do lots of prepare_write operations before we do
 428  * any of the matching commit_write operations, so even if we split
 429  * up to use "smaller" transactions none of them could complete until
 430  * all of them were opened.  By having a single journal transaction,
 431  * we eliminate duplicate reservations for common blocks like the
 432  * superblock and group descriptors or bitmaps.
 433  *
 434  * We will start the transaction here, but each prepare_write will
 435  * add a refcount to the transaction, and each commit_write will
 436  * remove a refcount.  The transaction will be closed when all of
 437  * the pages have been written.
 438  */
 439 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 440                                    int niocount, struct niobuf_local *nb,
 441                                    void *desc_private, int logs)
 442 {
 443         journal_t *journal;
 444         handle_t *handle;
 445         int needed;
 446         ENTRY;
 447
 448         LASSERT(current->journal_info == desc_private);
 449         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 450         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 451
 452         /* The number of blocks we could _possibly_ dirty can very large.
 453          * We reduce our request if it is absurd (and we couldn't get that
 454          * many credits for a single handle anyways).
 455          *
 456          * At some point we have to limit the size of I/Os sent at one time,
 457          * increase the size of the journal, or we have to calculate the
 458          * actual journal requirements more carefully by checking all of
 459          * the blocks instead of being maximally pessimistic.  It remains to
 460          * be seen if this is a real problem or not.
 461          */
 462         if (needed > journal->j_max_transaction_buffers) {
 463                 CERROR("want too many journal credits (%d) using %d instead\n",
 464                        needed, journal->j_max_transaction_buffers);
 465                 needed = journal->j_max_transaction_buffers;
 466         }
 467
 468         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 469         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 470         if (IS_ERR(handle)) {
 471                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 472                        PTR_ERR(handle));
 473         } else {
 474                 LASSERT(handle->h_buffer_credits >= needed);
 475                 LASSERT(current->journal_info == handle);
 476         }
 477
 478         RETURN(handle);
 479 }
 480
 481 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 482 {
 483        handle_t *handle = h;
 484
 485        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 486        if (nblocks == 0) {
 487                handle->h_buffer_credits = 0;
 488                CWARN("setting credits of handle %p to zero by request\n", h);
 489        }
 490
 491        if (handle->h_buffer_credits > nblocks)
 492                 return 0;
 493        if (ext3_journal_extend(handle, nblocks) == 0)
 494                 return 0;
 495
 496        ext3_mark_inode_dirty(handle, inode);
 497        return ext3_journal_restart(handle, nblocks);
 498 }
 499
 500 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 501 {
 502         int rc;
 503         handle_t *handle = h;
 504
 505         LASSERT(current->journal_info == handle);
 506         if (force_sync)
 507                 handle->h_sync = 1; /* recovery likes this */
 508
 509         rc = ext3_journal_stop(handle);
 510
 511         return rc;
 512 }
 513
 514 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 515                                     void **wait_handle)
 516 {
 517         unsigned long tid;
 518         transaction_t *transaction;
 519         handle_t *handle = h;
 520         journal_t *journal;
 521         int rc;
 522
 523         LASSERT(current->journal_info == handle);
 524
 525         transaction = handle->h_transaction;
 526         journal = transaction->t_journal;
 527         tid = transaction->t_tid;
 528         /* we don't want to be blocked */
 529         handle->h_sync = 0;
 530         rc = ext3_journal_stop(handle);
 531         if (rc) {
 532                 CERROR("error while stopping transaction: %d\n", rc);
 533                 return rc;
 534         }
 535         fsfilt_log_start_commit(journal, tid);
 536
 537         *wait_handle = (void *) tid;
 538         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 539         return 0;
 540 }
 541
 542 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 543 {
 544         journal_t *journal = EXT3_JOURNAL(inode);
 545         tid_t tid = (tid_t)(long)h;
 546
 547         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 548         if (unlikely(is_journal_aborted(journal)))
 549                 return -EIO;
 550
 551         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 552
 553         if (unlikely(is_journal_aborted(journal)))
 554                 return -EIO;
 555         return 0;
 556 }
 557
 558 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 559                                struct iattr *iattr, int do_trunc)
 560 {
 561         struct inode *inode = dentry->d_inode;
 562         int rc = 0;
 563
 564         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 565          * We are already writing the inode to disk as part of this
 566          * transaction and want to avoid a lot of extra inode writeout
 567          * later on. b=9828 */
 568         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 569                 /* ATTR_SIZE would invoke truncate: clear it */
 570                 iattr->ia_valid &= ~ATTR_SIZE;
 571                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 572                 i_size_write(inode, iattr->ia_size);
 573
 574                 if (iattr->ia_valid & ATTR_UID)
 575                         inode->i_uid = iattr->ia_uid;
 576                 if (iattr->ia_valid & ATTR_GID)
 577                         inode->i_gid = iattr->ia_gid;
 578                 if (iattr->ia_valid & ATTR_ATIME)
 579                         inode->i_atime = iattr->ia_atime;
 580                 if (iattr->ia_valid & ATTR_MTIME)
 581                         inode->i_mtime = iattr->ia_mtime;
 582                 if (iattr->ia_valid & ATTR_CTIME)
 583                         inode->i_ctime = iattr->ia_ctime;
 584                 if (iattr->ia_valid & ATTR_MODE) {
 585                         inode->i_mode = iattr->ia_mode;
 586
 587                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 588                             !cfs_capable(CFS_CAP_FSETID))
 589                                 inode->i_mode &= ~S_ISGID;
 590                 }
 591
 592                 inode->i_sb->s_op->dirty_inode(inode);
 593
 594                 goto out;
 595         }
 596
 597         /* Don't allow setattr to change file type */
 598         if (iattr->ia_valid & ATTR_MODE)
 599                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 600                                  (iattr->ia_mode & ~S_IFMT);
 601
 602         /* We set these flags on the client, but have already checked perms
 603          * so don't confuse inode_change_ok. */
 604         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 605
 606         if (inode->i_op->setattr) {
 607                 rc = inode->i_op->setattr(dentry, iattr);
 608         } else {
 609                 rc = inode_change_ok(inode, iattr);
 610                 if (!rc)
 611                         rc = inode_setattr(inode, iattr);
 612         }
 613
 614  out:
 615         RETURN(rc);
 616 }
 617
 618 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 619                                  unsigned int cmd, unsigned long arg)
 620 {
 621         int rc = 0;
 622         ENTRY;
 623
 624         /* FIXME: Can't do this because of nested transaction deadlock */
 625         if (cmd == EXT3_IOC_SETFLAGS) {
 626                 /* We can't enable data journaling on OST objects, because
 627                 * this forces the transaction to be closed in order to
 628                 * flush the journal, but the caller will already have a
 629                 * compound transaction open to update the last_rcvd file,
 630                 * and this thread would deadlock trying to set the flag. */
 631                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 632                         CERROR("can't set data journal flag on file\n");
 633                         RETURN(-EPERM);
 634                 }
 635                 /* Because the MDS does not see the EXTENTS_FL set on the
 636                  * OST objects, mask this flag into all set flags.  It is
 637                  * not legal to clear this flag in any case, so we are not
 638                  * changing the functionality by doing this.  b=22911 */
 639                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 640         }
 641
 642 #ifdef HAVE_EXT4_LDISKFS
 643         /* ext4_ioctl does not have a inode argument */
 644         if (inode->i_fop->unlocked_ioctl)
 645                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 646 #else
 647         if (inode->i_fop->ioctl)
 648                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 649 #endif
 650         else
 651                 RETURN(-ENOTTY);
 652
 653         RETURN(rc);
 654 }
 655
 656 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 657                               void *lmm, int lmm_size, const char *name)
 658 {
 659         int rc;
 660
 661         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 662
 663         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 664                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 665
 666
 667         if (rc && rc != -EROFS)
 668                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 669                        inode->i_ino, rc);
 670         return rc;
 671 }
 672
 673 /* Must be called with i_mutex held */
 674 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 675                               const char *name)
 676 {
 677         int rc;
 678
 679         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 680
 681         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 682                             name, lmm, lmm_size);
 683
 684         /* This gives us the MD size */
 685         if (lmm == NULL)
 686                 return (rc == -ENODATA) ? 0 : rc;
 687
 688         if (rc < 0) {
 689                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 690                        EXT3_XATTR_INDEX_TRUSTED, name,
 691                        inode->i_ino, rc);
 692                 memset(lmm, 0, lmm_size);
 693                 return (rc == -ENODATA) ? 0 : rc;
 694         }
 695
 696         return rc;
 697 }
 698
 699 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 700 {
 701         submit_bio(rw, bio);
 702         return 0;
 703 }
 704
 705 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 706                                     loff_t *off)
 707 {
 708         struct inode *inode = file->f_dentry->d_inode;
 709         int rc = 0;
 710
 711         if (S_ISREG(inode->i_mode))
 712                 rc = file->f_op->read(file, buf, count, off);
 713         else {
 714                 const int blkbits = inode->i_sb->s_blocksize_bits;
 715                 const int blksize = inode->i_sb->s_blocksize;
 716
 717                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 718                        (unsigned long)count, inode->i_ino, *off);
 719                 while (count > 0) {
 720                         struct buffer_head *bh;
 721
 722                         bh = NULL;
 723                         if (*off < i_size_read(inode)) {
 724                                 int err = 0;
 725
 726                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 727                                                 0, &err);
 728
 729                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 730
 731                                 if (bh) {
 732                                         memcpy(buf, bh->b_data, blksize);
 733                                         brelse(bh);
 734                                 } else if (err) {
 735                                         /* XXX in theory we should just fake
 736                                          * this buffer and continue like ext3,
 737                                          * especially if this is a partial read
 738                                          */
 739                                         CERROR("error read dir %lu+%llu: %d\n",
 740                                                inode->i_ino, *off, err);
 741                                         RETURN(err);
 742                                 }
 743                         }
 744                         if (!bh) {
 745                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 746
 747                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 748                                 memset(fake, 0, sizeof(*fake));
 749                                 fake->rec_len = cpu_to_le16(blksize);
 750                         }
 751                         count -= blksize;
 752                         buf += blksize;
 753                         *off += blksize;
 754                         rc += blksize;
 755                 }
 756         }
 757
 758         return rc;
 759 }
 760
 761 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 762 {
 763         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 764
 765         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 766
 767         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 768 }
 769
 770 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 771                                       void *handle, fsfilt_cb_t cb_func,
 772                                       void *cb_data)
 773 {
 774         struct fsfilt_cb_data *fcb;
 775
 776         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 777         if (fcb == NULL)
 778                 RETURN(-ENOMEM);
 779
 780         fcb->cb_func = cb_func;
 781         fcb->cb_obd = obd;
 782         fcb->cb_last_rcvd = last_rcvd;
 783         fcb->cb_data = cb_data;
 784
 785         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 786         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 787                                     (struct journal_callback *)fcb);
 788
 789         return 0;
 790 }
 791
 792 /*
 793  * We need to hack the return value for the free inode counts because
 794  * the current EA code requires one filesystem block per inode with EAs,
 795  * so it is possible to run out of blocks before we run out of inodes.
 796  *
 797  * This can be removed when the ext3 EA code is fixed.
 798  */
 799 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 800 {
 801         struct kstatfs sfs;
 802         int rc;
 803
 804         memset(&sfs, 0, sizeof(sfs));
 805         rc = ll_do_statfs(sb, &sfs);
 806         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 807                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 808                 sfs.f_ffree = sfs.f_bfree;
 809         }
 810
 811         statfs_pack(osfs, &sfs);
 812         return rc;
 813 }
 814
 815 static int fsfilt_ext3_sync(struct super_block *sb)
 816 {
 817         return ext3_force_commit(sb);
 818 }
 819
 820 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 821 #warning "kernel code has old extents/mballoc patch, disabling"
 822 #undef EXT3_MULTIBLOCK_ALLOCATOR
 823 #endif
 824 #ifndef EXT3_EXTENTS_FL
 825 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 826 #endif
 827
 828 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 829 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 830 #define fsfilt_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 831 #define fsfilt_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 832 #else
 833 #ifdef HAVE_EXT4_LDISKFS
 834 #define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem));
 835 #define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem));
 836 #else
 837 #define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 838 #define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 839 #endif
 840 #endif
 841
 842 #ifndef EXT_ASSERT
 843 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 844 #endif
 845
 846 #ifdef EXT3_EXT_HAS_NO_TREE
 847 /* for kernels 2.6.18 and later */
 848 #ifdef HAVE_EXT4_LDISKFS
 849 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 850 #else
 851 #define EXT_GENERATION(inode)           ext_generation(inode)
 852 #endif
 853 #define ext3_ext_base                   inode
 854 #define ext3_ext_base2inode(inode)      (inode)
 855 #define EXT_DEPTH(inode)                ext_depth(inode)
 856 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 857                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 858 #else
 859 #define ext3_ext_base                   ext3_extents_tree
 860 #define ext3_ext_base2inode(tree)       (tree->inode)
 861 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 862                         ext3_ext_walk_space(tree, block, num, cb);
 863 #endif
 864
 865 #include <linux/lustre_version.h>
 866
 867 struct bpointers {
 868         unsigned long *blocks;
 869         int *created;
 870         unsigned long start;
 871         int num;
 872         int init_num;
 873         int create;
 874 };
 875
 876 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 877                                unsigned long block, int *aflags)
 878 {
 879         struct ext3_inode_info *ei = EXT3_I(inode);
 880         unsigned long bg_start;
 881         unsigned long colour;
 882         int depth;
 883
 884         if (path) {
 885                 struct ext3_extent *ex;
 886                 depth = path->p_depth;
 887
 888                 /* try to predict block placement */
 889                 if ((ex = path[depth].p_ext))
 890                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 891
 892                 /* it looks index is empty
 893                  * try to find starting from index itself */
 894                 if (path[depth].p_bh)
 895                         return path[depth].p_bh->b_blocknr;
 896         }
 897
 898         /* OK. use inode's group */
 899         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 900                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 901         colour = (current->pid % 16) *
 902                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 903         return bg_start + colour + block;
 904 }
 905
 906 #define ll_unmap_underlying_metadata(sb, blocknr) \
 907         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 908
 909 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 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         unsigned long pblock, goal;
 915         int aflags = 0;
 916         struct inode *inode = ext3_ext_base2inode(base);
 917
 918         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 919         aflags |= 2; /* block have been already reserved */
 920         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 921         return pblock;
 922
 923 }
 924 #else
 925 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 926                                 struct ext3_ext_path *path, unsigned long block,
 927                                 unsigned long *count, int *err)
 928 {
 929         struct inode *inode = ext3_ext_base2inode(base);
 930         struct ext3_allocation_request ar;
 931         unsigned long pblock;
 932         int aflags;
 933
 934         /* find neighbour allocated blocks */
 935         ar.lleft = block;
 936         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 937         if (*err)
 938                 return 0;
 939         ar.lright = block;
 940         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 941         if (*err)
 942                 return 0;
 943
 944         /* allocate new block */
 945         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 946         ar.inode = inode;
 947         ar.logical = block;
 948         ar.len = *count;
 949         ar.flags = EXT3_MB_HINT_DATA;
 950         pblock = ext3_mb_new_blocks(handle, &ar, err);
 951         *count = ar.len;
 952         return pblock;
 953 }
 954 #endif
 955
 956 #ifdef EXT3_EXT_HAS_NO_TREE
 957 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 958                                   struct ext3_ext_path *path,
 959                                   struct ext3_ext_cache *cex,
 960 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 961                                    struct ext3_extent *ex,
 962 #endif
 963                                   void *cbdata)
 964 {
 965         struct bpointers *bp = cbdata;
 966 #else
 967 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 968                                   struct ext3_ext_path *path,
 969                                   struct ext3_ext_cache *cex
 970 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 971                                   , struct ext3_extent *ex
 972 #endif
 973                                  )
 974 {
 975         struct bpointers *bp = base->private;
 976 #endif
 977         struct inode *inode = ext3_ext_base2inode(base);
 978         struct ext3_extent nex;
 979         unsigned long pblock;
 980         unsigned long tgen;
 981         int err, i;
 982         unsigned long count;
 983         handle_t *handle;
 984
 985         i = EXT_DEPTH(base);
 986         EXT_ASSERT(i == path->p_depth);
 987         EXT_ASSERT(path[i].p_hdr);
 988
 989         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 990                 err = EXT_CONTINUE;
 991                 goto map;
 992         }
 993
 994         if (bp->create == 0) {
 995                 i = 0;
 996                 if (cex->ec_block < bp->start)
 997                         i = bp->start - cex->ec_block;
 998                 if (i >= cex->ec_len)
 999                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1000                                         i, cex->ec_len);
1001                 for (; i < cex->ec_len && bp->num; i++) {
1002                         *(bp->created) = 0;
1003                         bp->created++;
1004                         *(bp->blocks) = 0;
1005                         bp->blocks++;
1006                         bp->num--;
1007                         bp->start++;
1008                 }
1009
1010                 return EXT_CONTINUE;
1011         }
1012
1013         tgen = EXT_GENERATION(base);
1014         count = ext3_ext_calc_credits_for_insert(base, path);
1015         fsfilt_up_truncate_sem(inode);
1016
1017         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1018         if (IS_ERR(handle)) {
1019                 fsfilt_down_truncate_sem(inode);
1020                 return PTR_ERR(handle);
1021         }
1022
1023         fsfilt_down_truncate_sem(inode);
1024         if (tgen != EXT_GENERATION(base)) {
1025                 /* the tree has changed. so path can be invalid at moment */
1026                 ext3_journal_stop(handle);
1027                 return EXT_REPEAT;
1028         }
1029
1030         count = cex->ec_len;
1031         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1032         if (!pblock)
1033                 goto out;
1034         EXT_ASSERT(count <= cex->ec_len);
1035
1036         /* insert new extent */
1037         nex.ee_block = cpu_to_le32(cex->ec_block);
1038         ext3_ext_store_pblock(&nex, pblock);
1039         nex.ee_len = cpu_to_le16(count);
1040         err = ext3_ext_insert_extent(handle, base, path, &nex);
1041         if (err) {
1042                 /* free data blocks we just allocated */
1043                 /* not a good idea to call discard here directly,
1044                  * but otherwise we'd need to call it every free() */
1045 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1046                 ext3_mb_discard_inode_preallocations(inode);
1047 #endif
1048                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1049                                  cpu_to_le16(nex.ee_len), 0);
1050                 goto out;
1051         }
1052
1053         /*
1054          * Putting len of the actual extent we just inserted,
1055          * we are asking ext3_ext_walk_space() to continue
1056          * scaning after that block
1057          */
1058         cex->ec_len = le16_to_cpu(nex.ee_len);
1059         cex->ec_start = ext_pblock(&nex);
1060         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1061         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1062
1063 out:
1064         ext3_journal_stop(handle);
1065 map:
1066         if (err >= 0) {
1067                 /* map blocks */
1068                 if (bp->num == 0) {
1069                         CERROR("hmm. why do we find this extent?\n");
1070                         CERROR("initial space: %lu:%u\n",
1071                                 bp->start, bp->init_num);
1072                         CERROR("current extent: %u/%u/%llu %d\n",
1073                                 cex->ec_block, cex->ec_len,
1074                                 (unsigned long long)cex->ec_start,
1075                                 cex->ec_type);
1076                 }
1077                 i = 0;
1078                 if (cex->ec_block < bp->start)
1079                         i = bp->start - cex->ec_block;
1080                 if (i >= cex->ec_len)
1081                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1082                                         i, cex->ec_len);
1083                 for (; i < cex->ec_len && bp->num; i++) {
1084                         *(bp->blocks) = cex->ec_start + i;
1085                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1086                                 *(bp->created) = 0;
1087                         } else {
1088                                 *(bp->created) = 1;
1089                                 /* unmap any possible underlying metadata from
1090                                  * the block device mapping.  bug 6998. */
1091                                 ll_unmap_underlying_metadata(inode->i_sb,
1092                                                              *(bp->blocks));
1093                         }
1094                         bp->created++;
1095                         bp->blocks++;
1096                         bp->num--;
1097                         bp->start++;
1098                 }
1099         }
1100         return err;
1101 }
1102
1103 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1104                        unsigned long num, unsigned long *blocks,
1105                        int *created, int create)
1106 {
1107 #ifdef EXT3_EXT_HAS_NO_TREE
1108         struct ext3_ext_base *base = inode;
1109 #else
1110         struct ext3_extents_tree tree;
1111         struct ext3_ext_base *base = &tree;
1112 #endif
1113         struct bpointers bp;
1114         int err;
1115
1116         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1117                block, block + num - 1, (unsigned) inode->i_ino);
1118
1119 #ifndef EXT3_EXT_HAS_NO_TREE
1120         ext3_init_tree_desc(base, inode);
1121         tree.private = &bp;
1122 #endif
1123         bp.blocks = blocks;
1124         bp.created = created;
1125         bp.start = block;
1126         bp.init_num = bp.num = num;
1127         bp.create = create;
1128
1129         fsfilt_down_truncate_sem(inode);
1130         err = fsfilt_ext3_ext_walk_space(base, block, num,
1131                                          ext3_ext_new_extent_cb, &bp);
1132         ext3_ext_invalidate_cache(base);
1133         fsfilt_up_truncate_sem(inode);
1134
1135         return err;
1136 }
1137
1138 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1139                                     int pages, unsigned long *blocks,
1140                                     int *created, int create)
1141 {
1142         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1143         int rc = 0, i = 0;
1144         struct page *fp = NULL;
1145         int clen = 0;
1146
1147         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1148                 inode->i_ino, pages, (*page)->index);
1149
1150         /* pages are sorted already. so, we just have to find
1151          * contig. space and process them properly */
1152         while (i < pages) {
1153                 if (fp == NULL) {
1154                         /* start new extent */
1155                         fp = *page++;
1156                         clen = 1;
1157                         i++;
1158                         continue;
1159                 } else if (fp->index + clen == (*page)->index) {
1160                         /* continue the extent */
1161                         page++;
1162                         clen++;
1163                         i++;
1164                         continue;
1165                 }
1166
1167                 /* process found extent */
1168                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1169                                         clen * blocks_per_page, blocks,
1170                                         created, create);
1171                 if (rc)
1172                         GOTO(cleanup, rc);
1173
1174                 /* look for next extent */
1175                 fp = NULL;
1176                 blocks += blocks_per_page * clen;
1177                 created += blocks_per_page * clen;
1178         }
1179
1180         if (fp)
1181                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1182                                         clen * blocks_per_page, blocks,
1183                                         created, create);
1184 cleanup:
1185         return rc;
1186 }
1187 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1188
1189 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1190                                unsigned long *blocks, int *created, int create);
1191 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1192                                    int pages, unsigned long *blocks,
1193                                    int *created, int create)
1194 {
1195         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1196         unsigned long *b;
1197         int rc = 0, i, *cr;
1198
1199         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1200                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1201                 if (rc) {
1202                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1203                                inode->i_ino, *b, *cr, create, rc);
1204                         break;
1205                 }
1206
1207                 b += blocks_per_page;
1208                 cr += blocks_per_page;
1209         }
1210         return rc;
1211 }
1212
1213 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1214                                 int pages, unsigned long *blocks,
1215                                 int *created, int create,
1216                                 cfs_semaphore_t *optional_sem)
1217 {
1218         int rc;
1219 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1220         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1221                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1222                                                      blocks, created, create);
1223                 return rc;
1224         }
1225 #endif
1226         if (optional_sem != NULL)
1227                 cfs_down(optional_sem);
1228         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1229                                             created, create);
1230         if (optional_sem != NULL)
1231                 cfs_up(optional_sem);
1232
1233         return rc;
1234 }
1235
1236 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1237 {
1238         unsigned long block;
1239         struct buffer_head *bh;
1240         int err, blocksize, csize, boffs, osize = size;
1241
1242         /* prevent reading after eof */
1243         cfs_lock_kernel();
1244         if (i_size_read(inode) < *offs + size) {
1245                 size = i_size_read(inode) - *offs;
1246                 cfs_unlock_kernel();
1247                 if (size < 0) {
1248                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1249                                i_size_read(inode), *offs);
1250                         return -EBADR;
1251                 } else if (size == 0) {
1252                         return 0;
1253                 }
1254         } else {
1255                 cfs_unlock_kernel();
1256         }
1257
1258         blocksize = 1 << inode->i_blkbits;
1259
1260         while (size > 0) {
1261                 block = *offs >> inode->i_blkbits;
1262                 boffs = *offs & (blocksize - 1);
1263                 csize = min(blocksize - boffs, size);
1264                 bh = ext3_bread(NULL, inode, block, 0, &err);
1265                 if (!bh) {
1266                         CERROR("can't read block: %d\n", err);
1267                         return err;
1268                 }
1269
1270                 memcpy(buf, bh->b_data + boffs, csize);
1271                 brelse(bh);
1272
1273                 *offs += csize;
1274                 buf += csize;
1275                 size -= csize;
1276         }
1277         return osize;
1278 }
1279 EXPORT_SYMBOL(fsfilt_ext3_read);
1280
1281 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1282                                    int size, loff_t *offs)
1283 {
1284         int rc;
1285         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1286         if (rc > 0)
1287                 rc = 0;
1288         return rc;
1289 }
1290
1291 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1292                                 loff_t *offs, handle_t *handle)
1293 {
1294         struct buffer_head *bh = NULL;
1295         loff_t old_size = i_size_read(inode), offset = *offs;
1296         loff_t new_size = i_size_read(inode);
1297         unsigned long block;
1298         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1299
1300         while (bufsize > 0) {
1301                 if (bh != NULL)
1302                         brelse(bh);
1303
1304                 block = offset >> inode->i_blkbits;
1305                 boffs = offset & (blocksize - 1);
1306                 size = min(blocksize - boffs, bufsize);
1307                 bh = ext3_bread(handle, inode, block, 1, &err);
1308                 if (!bh) {
1309                         CERROR("can't read/create block: %d\n", err);
1310                         break;
1311                 }
1312
1313                 err = ext3_journal_get_write_access(handle, bh);
1314                 if (err) {
1315                         CERROR("journal_get_write_access() returned error %d\n",
1316                                err);
1317                         break;
1318                 }
1319                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1320                 memcpy(bh->b_data + boffs, buf, size);
1321                 err = ext3_journal_dirty_metadata(handle, bh);
1322                 if (err) {
1323                         CERROR("journal_dirty_metadata() returned error %d\n",
1324                                err);
1325                         break;
1326                 }
1327                 if (offset + size > new_size)
1328                         new_size = offset + size;
1329                 offset += size;
1330                 bufsize -= size;
1331                 buf += size;
1332         }
1333         if (bh)
1334                 brelse(bh);
1335
1336         /* correct in-core and on-disk sizes */
1337         if (new_size > i_size_read(inode)) {
1338                 cfs_lock_kernel();
1339                 if (new_size > i_size_read(inode))
1340                         i_size_write(inode, new_size);
1341                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1342                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1343                 if (i_size_read(inode) > old_size)
1344                         mark_inode_dirty(inode);
1345                 cfs_unlock_kernel();
1346         }
1347
1348         if (err == 0)
1349                 *offs = offset;
1350         return err;
1351 }
1352 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1353
1354 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1355                                     loff_t *offs, int force_sync)
1356 {
1357         struct inode *inode = file->f_dentry->d_inode;
1358         handle_t *handle;
1359         int err, block_count = 0, blocksize;
1360
1361         /* Determine how many transaction credits are needed */
1362         blocksize = 1 << inode->i_blkbits;
1363         block_count = (*offs & (blocksize - 1)) + bufsize;
1364         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1365
1366         handle = ext3_journal_start(inode,
1367                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1368         if (IS_ERR(handle)) {
1369                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1370                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1371                 return PTR_ERR(handle);
1372         }
1373
1374         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1375
1376         if (!err && force_sync)
1377                 handle->h_sync = 1; /* recovery likes this */
1378
1379         ext3_journal_stop(handle);
1380
1381         return err;
1382 }
1383
1384 static int fsfilt_ext3_setup(struct super_block *sb)
1385 {
1386 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1387      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1388         struct ext3_sb_info *sbi = EXT3_SB(sb);
1389 #if 0
1390         sbi->dx_lock = fsfilt_ext3_dx_lock;
1391         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1392 #endif
1393 #endif
1394 #ifdef S_PDIROPS
1395         CWARN("Enabling PDIROPS\n");
1396         set_opt(sbi->s_mount_opt, PDIROPS);
1397         sb->s_flags |= S_PDIROPS;
1398 #endif
1399         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1400                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1401 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1402         /* enable journaled quota support */
1403         /* kfreed in ext3_put_super() */
1404         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1405         if (!sbi->s_qf_names[USRQUOTA])
1406                 return -ENOMEM;
1407         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1408         if (!sbi->s_qf_names[GRPQUOTA]) {
1409                 kfree(sbi->s_qf_names[USRQUOTA]);
1410                 sbi->s_qf_names[USRQUOTA] = NULL;
1411                 return -ENOMEM;
1412         }
1413         sbi->s_jquota_fmt = QFMT_VFS_V0;
1414 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1415         set_opt(sbi->s_mount_opt, QUOTA);
1416 #endif
1417 #endif
1418         return 0;
1419 }
1420
1421 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1422    objects. Logs is number of logfiles to update */
1423 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1424 {
1425         if ( !fso ) {
1426                 switch(op) {
1427                 case FSFILT_OP_CREATE:
1428                                  /* directory leaf, index & indirect & EA*/
1429                         return 4 + 3 * logs;
1430                 case FSFILT_OP_UNLINK:
1431                         return 3 * logs;
1432                 }
1433         } else {
1434                 int i;
1435                 int needed = 0;
1436                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1437                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1438                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1439                 for (i = 0; i < op; i++, fso++) {
1440                         int nblocks = fso->fso_bufcnt * blockpp;
1441                         int ndindirect = min(nblocks, addrpp + 1);
1442                         int nindir = nblocks + ndindirect + 1;
1443
1444                         needed += nindir;
1445                 }
1446                 return needed + 3 * logs;
1447         }
1448
1449         return 0;
1450 }
1451
1452 #ifdef HAVE_QUOTA_SUPPORT
1453 #define DQINFO_COPY(out, in)                    \
1454 do {                                            \
1455         Q_COPY(out, in, dqi_bgrace);            \
1456         Q_COPY(out, in, dqi_igrace);            \
1457         Q_COPY(out, in, dqi_flags);             \
1458         Q_COPY(out, in, dqi_valid);             \
1459 } while (0)
1460
1461 #define DQBLK_COPY(out, in)                     \
1462 do {                                            \
1463         Q_COPY(out, in, dqb_bhardlimit);        \
1464         Q_COPY(out, in, dqb_bsoftlimit);        \
1465         Q_COPY(out, in, dqb_curspace);          \
1466         Q_COPY(out, in, dqb_ihardlimit);        \
1467         Q_COPY(out, in, dqb_isoftlimit);        \
1468         Q_COPY(out, in, dqb_curinodes);         \
1469         Q_COPY(out, in, dqb_btime);             \
1470         Q_COPY(out, in, dqb_itime);             \
1471         Q_COPY(out, in, dqb_valid);             \
1472 } while (0)
1473
1474 static int fsfilt_ext3_quotactl(struct super_block *sb,
1475                                 struct obd_quotactl *oqc)
1476 {
1477         int i, rc = 0, error = 0;
1478         struct quotactl_ops *qcop;
1479         struct if_dqinfo *info;
1480         struct if_dqblk *dqblk;
1481         ENTRY;
1482
1483         if (!sb->s_qcop)
1484                 RETURN(-ENOSYS);
1485
1486         OBD_ALLOC_PTR(info);
1487         if (!info)
1488                 RETURN(-ENOMEM);
1489         OBD_ALLOC_PTR(dqblk);
1490         if (!dqblk) {
1491                 OBD_FREE_PTR(info);
1492                 RETURN(-ENOMEM);
1493         }
1494
1495         DQINFO_COPY(info, &oqc->qc_dqinfo);
1496         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1497
1498         qcop = sb->s_qcop;
1499         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1500                 for (i = 0; i < MAXQUOTAS; i++) {
1501                         if (!Q_TYPESET(oqc, i))
1502                                 continue;
1503
1504                         if (oqc->qc_cmd == Q_QUOTAON) {
1505                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1506
1507                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1508
1509                                 rc = ll_quota_on(sb, i, QFMT_VFS_V0,
1510                                                  name[i], 0);
1511                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1512                                 rc = ll_quota_off(sb, i, 0);
1513                         }
1514
1515                         if (rc == -EBUSY)
1516                                 error = rc;
1517                         else if (rc)
1518                                 GOTO(out, rc);
1519                 }
1520                 GOTO(out, rc ?: error);
1521         }
1522
1523         switch (oqc->qc_cmd) {
1524         case Q_GETOINFO:
1525         case Q_GETINFO:
1526                 if (!qcop->get_info)
1527                         GOTO(out, rc = -ENOSYS);
1528                 rc = qcop->get_info(sb, oqc->qc_type, info);
1529                 break;
1530         case Q_SETQUOTA:
1531         case Q_INITQUOTA:
1532                 if (!qcop->set_dqblk)
1533                         GOTO(out, rc = -ENOSYS);
1534                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1535                 break;
1536         case Q_GETOQUOTA:
1537         case Q_GETQUOTA:
1538                 if (!qcop->get_dqblk)
1539                         GOTO(out, rc = -ENOSYS);
1540                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1541                 if (!rc)
1542                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1543                 break;
1544         case Q_SYNC:
1545                 if (!sb->s_qcop->quota_sync)
1546                         GOTO(out, rc = -ENOSYS);
1547                 qcop->quota_sync(sb, oqc->qc_type);
1548                 break;
1549         case Q_FINVALIDATE:
1550                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1551                 for (i = 0; i < MAXQUOTAS; i++) {
1552                         struct file *fp;
1553                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1554
1555                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1556
1557                         if (!Q_TYPESET(oqc, i))
1558                                 continue;
1559
1560                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1561                         if (IS_ERR(fp)) {
1562                                 rc = PTR_ERR(fp);
1563                                 CERROR("error invalidating operational quota file"
1564                                        " %s (rc:%d)\n", name[i], rc);
1565                         } else {
1566                                 filp_close(fp, 0);
1567                         }
1568
1569                 }
1570                 break;
1571         default:
1572                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1573                 LBUG();
1574         }
1575 out:
1576         DQINFO_COPY(&oqc->qc_dqinfo, info);
1577         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1578
1579         OBD_FREE_PTR(info);
1580         OBD_FREE_PTR(dqblk);
1581
1582         if (rc)
1583                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1584                                 "failed: %d\n",
1585                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1586         RETURN(rc);
1587 }
1588
1589 struct chk_dqblk{
1590         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1591         cfs_list_t              dqb_list;        /** in list also */
1592         qid_t                   dqb_id;          /** uid/gid */
1593         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1594         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1595         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1596         qsize_t                 dqb_curspace;    /** current space */
1597         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1598         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1599         qsize_t                 dqb_curinodes;   /** current inodes */
1600         __u64                   dqb_btime;       /** block grace time */
1601         __u64                   dqb_itime;       /** inode grace time */
1602         __u32                   dqb_valid;       /** flag for above fields */
1603 };
1604
1605 static inline unsigned int chkquot_hash(qid_t id, int type)
1606                                         __attribute__((__const__));
1607
1608 static inline unsigned int chkquot_hash(qid_t id, int type)
1609 {
1610         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1611 }
1612
1613 static inline struct chk_dqblk *
1614 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1615 {
1616         cfs_hlist_node_t *node;
1617         struct chk_dqblk *cdqb;
1618
1619         cfs_hlist_for_each(node, head) {
1620                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1621                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1622                         return cdqb;
1623         }
1624
1625         return NULL;
1626 }
1627
1628 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1629 {
1630         struct chk_dqblk *cdqb;
1631
1632         OBD_ALLOC_PTR(cdqb);
1633         if (cdqb) {
1634                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1635                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1636                 cdqb->dqb_id = id;
1637                 cdqb->dqb_type = type;
1638         }
1639
1640         return cdqb;
1641 }
1642
1643 static struct chk_dqblk *
1644 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1645       cfs_list_t *list, qid_t id, int type, int first_check)
1646 {
1647         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1648         struct if_dqblk dqb;
1649         struct chk_dqblk *cdqb;
1650         int rc;
1651
1652         cdqb = find_chkquot(head, id, type);
1653         if (cdqb)
1654                 return cdqb;
1655
1656         cdqb = alloc_chkquot(id, type);
1657         if (!cdqb)
1658                 return NULL;
1659
1660         if (!first_check) {
1661                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1662                 if (rc) {
1663                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1664                                id, type, rc);
1665                 } else {
1666                         DQBLK_COPY(cdqb, &dqb);
1667                         cdqb->dqb_curspace = 0;
1668                         cdqb->dqb_curinodes = 0;
1669                 }
1670         }
1671
1672         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1673         cfs_list_add_tail(&cdqb->dqb_list, list);
1674
1675         return cdqb;
1676 }
1677
1678 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1679 {
1680         struct obd_quotactl *oqctl;
1681         int rc;
1682
1683         OBD_ALLOC_PTR(oqctl);
1684         if (!oqctl)
1685                 RETURN(-ENOMEM);
1686
1687         oqctl->qc_cmd = cmd;
1688         oqctl->qc_id = qfmt;
1689         oqctl->qc_type = type;
1690         rc = fsfilt_ext3_quotactl(sb, oqctl);
1691
1692         OBD_FREE_PTR(oqctl);
1693         return rc;
1694 }
1695
1696 static inline int read_old_dqinfo(struct super_block *sb, int type,
1697                                   struct if_dqinfo *dqinfo)
1698 {
1699         struct obd_quotactl *oqctl;
1700         int rc;
1701         ENTRY;
1702
1703         OBD_ALLOC_PTR(oqctl);
1704         if (!oqctl)
1705                 RETURN(-ENOMEM);
1706
1707         oqctl->qc_cmd = Q_GETINFO;
1708         oqctl->qc_type = type;
1709         rc = fsfilt_ext3_quotactl(sb, oqctl);
1710         if (!rc)
1711                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1712
1713         OBD_FREE_PTR(oqctl);
1714         RETURN(rc);
1715 }
1716
1717 #ifndef HAVE_EXT4_LDISKFS
1718 static inline struct ext3_group_desc *
1719 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1720 {
1721         unsigned long desc_block, desc;
1722         struct ext3_group_desc *gdp;
1723
1724         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1725         desc = group % EXT3_DESC_PER_BLOCK(sb);
1726         gdp = (struct ext3_group_desc *)
1727               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1728
1729         return gdp + desc;
1730 }
1731
1732 static inline struct buffer_head *
1733 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1734 {
1735         struct ext3_group_desc *desc;
1736         struct buffer_head *bh;
1737
1738         desc = get_group_desc(sb, group, NULL);
1739         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1740         return bh;
1741 }
1742
1743 static __u32 ext3_itable_unused_count(struct super_block *sb,
1744                                struct ext3_group_desc *bg) {
1745        return le16_to_cpu(bg->bg_itable_unused);
1746 }
1747 #else
1748 #define get_group_desc ext3_get_group_desc
1749 #endif
1750
1751 struct qchk_ctxt {
1752         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1753         cfs_list_t              qckt_list;                 /* quotacheck list */
1754         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1755         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1756 };
1757
1758 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1759                            struct obd_quotactl *oqc)
1760 {
1761         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1762         loff_t size = 0;
1763         qid_t qid[MAXQUOTAS];
1764         int cnt, i, rc = 0;
1765
1766         if (!inode)
1767                 return 0;
1768
1769         qid[USRQUOTA] = inode->i_uid;
1770         qid[GRPQUOTA] = inode->i_gid;
1771
1772         if (S_ISDIR(inode->i_mode) ||
1773             S_ISREG(inode->i_mode) ||
1774             S_ISLNK(inode->i_mode))
1775                 size = inode_get_bytes(inode);
1776
1777         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1778                 if (!Q_TYPESET(oqc, cnt))
1779                         continue;
1780
1781                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1782                                 &qctxt->qckt_list, qid[cnt], cnt,
1783                                 qctxt->qckt_first_check[cnt]);
1784                 if (!cdqb[cnt]) {
1785                         rc = -ENOMEM;
1786                         break;
1787                 }
1788
1789                 cdqb[cnt]->dqb_curspace += size;
1790                 cdqb[cnt]->dqb_curinodes++;
1791         }
1792
1793         if (rc) {
1794                 for (i = 0; i < cnt; i++) {
1795                         if (!Q_TYPESET(oqc, i))
1796                                 continue;
1797                         LASSERT(cdqb[i]);
1798                         cdqb[i]->dqb_curspace -= size;
1799                         cdqb[i]->dqb_curinodes--;
1800                 }
1801         }
1802
1803         return rc;
1804 }
1805
1806 /* write dqinfo struct in a new quota file */
1807 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1808 {
1809         struct v2_disk_dqinfo dqinfo;
1810         __u32 blocks = V2_DQTREEOFF + 1;
1811         loff_t offset = V2_DQINFOOFF;
1812
1813         if (info) {
1814                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1815                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1816                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1817                                                ~DQF_INFO_DIRTY);
1818         } else {
1819                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1820                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1821                 dqinfo.dqi_flags = 0;
1822         }
1823
1824         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1825         dqinfo.dqi_free_blk = 0;
1826         dqinfo.dqi_free_entry = 0;
1827
1828         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1829 }
1830
1831 static int v3_write_dqheader(struct file *f, int type)
1832 {
1833         static const __u32 quota_magics[] = V2_INITQMAGICS;
1834         static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1835         struct v2_disk_dqheader dqhead;
1836         loff_t offset = 0;
1837
1838         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1839         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1840
1841         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1842         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1843
1844         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1845 }
1846
1847 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1848                                   struct obd_quotactl *oqc)
1849 {
1850         int i, rc = 0;
1851         ENTRY;
1852
1853         for (i = 0; i < MAXQUOTAS; i++) {
1854                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1855                                          NULL : &qctxt->qckt_dqinfo[i];
1856                 struct file *file;
1857                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1858
1859                 if (!Q_TYPESET(oqc, i))
1860                         continue;
1861
1862                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1863
1864                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1865                 if (IS_ERR(file)) {
1866                         rc = PTR_ERR(file);
1867                         CERROR("can't create %s file: rc = %d\n",
1868                                name[i], rc);
1869                         GOTO(out, rc);
1870                 }
1871
1872                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1873                         CERROR("file %s is not regular", name[i]);
1874                         filp_close(file, 0);
1875                         GOTO(out, rc = -EINVAL);
1876                 }
1877
1878                 DQUOT_DROP(file->f_dentry->d_inode);
1879
1880                 rc = v3_write_dqheader(file, i);
1881                 if (rc) {
1882                         filp_close(file, 0);
1883                         GOTO(out, rc);
1884                 }
1885
1886                 rc = v3_write_dqinfo(file, i, info);
1887                 filp_close(file, 0);
1888                 if (rc)
1889                         GOTO(out, rc);
1890         }
1891
1892 out:
1893         RETURN(rc);
1894 }
1895
1896
1897 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1898                           struct chk_dqblk *cdqb)
1899 {
1900         struct obd_quotactl *oqc;
1901         long now;
1902         int rc;
1903         ENTRY;
1904
1905         OBD_ALLOC_PTR(oqc);
1906         if (!oqc)
1907                 RETURN(-ENOMEM);
1908
1909         now = cfs_time_current_sec();
1910
1911         if (cdqb->dqb_bsoftlimit &&
1912             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1913             !cdqb->dqb_btime)
1914                 cdqb->dqb_btime =
1915                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1916
1917         if (cdqb->dqb_isoftlimit &&
1918             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1919             !cdqb->dqb_itime)
1920                 cdqb->dqb_itime =
1921                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1922
1923         cdqb->dqb_valid = QIF_ALL;
1924
1925         oqc->qc_cmd = Q_SETQUOTA;
1926         oqc->qc_type = cdqb->dqb_type;
1927         oqc->qc_id = cdqb->dqb_id;
1928         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1929
1930         rc = fsfilt_ext3_quotactl(sb, oqc);
1931         OBD_FREE_PTR(oqc);
1932         RETURN(rc);
1933 }
1934
1935 static int prune_chkquots(struct super_block *sb,
1936                           struct qchk_ctxt *qctxt, int error)
1937 {
1938         struct chk_dqblk *cdqb, *tmp;
1939         int rc;
1940
1941         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1942                 if (!error) {
1943                         rc = commit_chkquot(sb, qctxt, cdqb);
1944                         if (rc)
1945                                 error = rc;
1946                 }
1947                 cfs_hlist_del_init(&cdqb->dqb_hash);
1948                 cfs_list_del(&cdqb->dqb_list);
1949                 OBD_FREE_PTR(cdqb);
1950         }
1951
1952         return error;
1953 }
1954
1955 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1956 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1957 #endif
1958
1959 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1960                                   struct obd_quotactl *oqc)
1961 {
1962         struct ext3_sb_info *sbi = EXT3_SB(sb);
1963         int i, group, uninit_feat = 0;
1964         struct qchk_ctxt *qctxt;
1965         struct buffer_head *bitmap_bh = NULL;
1966         unsigned long ino, inode_inuse;
1967         struct inode *inode;
1968         int rc = 0;
1969         ENTRY;
1970
1971         /* turn on quota and read dqinfo if existed */
1972         OBD_ALLOC_PTR(qctxt);
1973         if (!qctxt) {
1974                 oqc->qc_stat = -ENOMEM;
1975                 RETURN(-ENOMEM);
1976         }
1977
1978         for (i = 0; i < NR_DQHASH; i++)
1979                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1980         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
1981
1982         for (i = 0; i < MAXQUOTAS; i++) {
1983                 if (!Q_TYPESET(oqc, i))
1984                         continue;
1985
1986                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1987                 if (!rc || rc == -EBUSY) {
1988                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1989                         if (rc)
1990                                 GOTO(out, rc);
1991                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1992                         qctxt->qckt_first_check[i] = 1;
1993                 } else if (rc) {
1994                         GOTO(out, rc);
1995                 }
1996         }
1997         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
1998                 /* This filesystem supports the uninit group feature */
1999                 uninit_feat = 1;
2000
2001         /* number of inodes that have been allocated */
2002         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2003                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2004
2005         /* check quota and update in hash */
2006         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2007              group++) {
2008                 unsigned long used_count = sbi->s_inodes_per_group;
2009
2010                 if (uninit_feat) {
2011                         struct ext3_group_desc *desc;
2012                         desc = get_group_desc(sb, group, NULL);
2013                         if (!desc)
2014                                 GOTO(out, -EIO);
2015
2016                         /* we don't really need to take the group lock here,
2017                          * but it may be useful if one day we support online
2018                          * quotacheck */
2019                         spin_lock(sb_bgl_lock(sbi, group));
2020                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2021                                 /* no inode in use in this group, just skip it */
2022                                 spin_unlock(sb_bgl_lock(sbi, group));
2023                                 continue;
2024                         }
2025                         used_count -= ext3_itable_unused_count(sb, desc);
2026                         spin_unlock(sb_bgl_lock(sbi, group));
2027                 }
2028
2029                 ino = group * sbi->s_inodes_per_group + 1;
2030                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2031                 if (!bitmap_bh) {
2032                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2033                                sb->s_id, group);
2034                         GOTO(out, -EIO);
2035                 }
2036
2037                 i = 0;
2038                 while (i < used_count &&
2039                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2040                                                used_count, i)) < used_count) {
2041                         inode_inuse--;
2042                         i++;
2043                         ino = i + group * sbi->s_inodes_per_group;
2044                         if (ino < sbi->s_first_ino)
2045                                 continue;
2046 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2047                         inode = ext3_iget(sb, ino);
2048 #else
2049                         inode = iget(sb, ino);
2050 #endif
2051                         if (!inode || IS_ERR(inode))
2052                                 continue;
2053
2054                         rc = add_inode_quota(inode, qctxt, oqc);
2055                         iput(inode);
2056                         if (rc) {
2057                                 brelse(bitmap_bh);
2058                                 GOTO(out, rc);
2059                         }
2060                 }
2061                 brelse(bitmap_bh);
2062         }
2063
2064         /* read old quota limits from old quota file. (only for the user
2065          * has limits but hasn't file) */
2066 #ifdef HAVE_QUOTA_SUPPORT
2067         for (i = 0; i < MAXQUOTAS; i++) {
2068                 cfs_list_t id_list;
2069                 struct dquot_id *dqid, *tmp;
2070
2071                 if (!Q_TYPESET(oqc, i))
2072                         continue;
2073
2074                 if (qctxt->qckt_first_check[i])
2075                         continue;
2076
2077
2078                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2079                 CFS_INIT_LIST_HEAD(&id_list);
2080 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2081                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2082 #else
2083                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2084 #endif
2085                 if (rc)
2086                         CERROR("read old limits failed. (rc:%d)\n", rc);
2087
2088                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2089                         cfs_list_del_init(&dqid->di_link);
2090
2091                         if (!rc)
2092                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2093                                       dqid->di_id, i,
2094                                       qctxt->qckt_first_check[i]);
2095                         OBD_FREE_PTR(dqid);
2096                 }
2097         }
2098 #endif
2099         /* turn off quota cause we are to dump chk_dqblk to files */
2100         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2101
2102         rc = create_new_quota_files(qctxt, oqc);
2103         if (rc)
2104                 GOTO(out, rc);
2105
2106         /* we use vfs functions to set dqblk, so turn quota on */
2107         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2108 out:
2109         /* dump and free chk_dqblk */
2110         rc = prune_chkquots(sb, qctxt, rc);
2111         OBD_FREE_PTR(qctxt);
2112
2113         /* turn off quota, `lfs quotacheck` will turn on when all
2114          * nodes quotacheck finish. */
2115         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2116
2117         oqc->qc_stat = rc;
2118         if (rc)
2119                 CERROR("quotacheck failed: rc = %d\n", rc);
2120
2121         RETURN(rc);
2122 }
2123
2124 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2125                                  int cmd)
2126 {
2127         int rc = 0;
2128         ENTRY;
2129
2130         if (lqi->qi_files[type] == NULL) {
2131                 CERROR("operate qinfo before it's enabled!\n");
2132                 RETURN(-ESRCH);
2133         }
2134
2135         switch (cmd) {
2136         case QFILE_CHK:
2137                 rc = lustre_check_quota_file(lqi, type);
2138                 break;
2139         case QFILE_RD_INFO:
2140                 rc = lustre_read_quota_info(lqi, type);
2141                 break;
2142         case QFILE_WR_INFO:
2143                 rc = lustre_write_quota_info(lqi, type);
2144                 break;
2145         case QFILE_INIT_INFO:
2146                 rc = lustre_init_quota_info(lqi, type);
2147                 break;
2148         case QFILE_CONVERT:
2149                 rc = -ENOTSUPP;
2150                 CERROR("quota CONVERT command is not supported\n");
2151                 break;
2152         default:
2153                 rc = -ENOTSUPP;
2154                 CERROR("Unsupported admin quota file cmd %d\n"
2155                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2156                        cmd);
2157                 break;
2158         }
2159         RETURN(rc);
2160 }
2161
2162 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2163                             cfs_list_t *list)
2164 {
2165         return lustre_get_qids(file, inode, type, list);
2166 }
2167
2168 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2169 {
2170         int rc = 0;
2171         ENTRY;
2172
2173         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2174                 CERROR("operate dquot before it's enabled!\n");
2175                 RETURN(-ESRCH);
2176         }
2177
2178         switch (cmd) {
2179         case QFILE_RD_DQUOT:
2180                 rc = lustre_read_dquot(dquot);
2181                 break;
2182         case QFILE_WR_DQUOT:
2183                 if (dquot->dq_dqb.dqb_ihardlimit ||
2184                     dquot->dq_dqb.dqb_isoftlimit ||
2185                     dquot->dq_dqb.dqb_bhardlimit ||
2186                     dquot->dq_dqb.dqb_bsoftlimit)
2187                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2188                 else
2189                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2190
2191                 rc = lustre_commit_dquot(dquot);
2192                 if (rc >= 0)
2193                         rc = 0;
2194                 break;
2195         default:
2196                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2197                 LBUG();
2198                 break;
2199         }
2200         RETURN(rc);
2201 }
2202
2203 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2204                                 struct inode *inode, int frags)
2205 {
2206 #ifdef EXT3_EXT_HAS_NO_TREE
2207         struct ext3_ext_base *base = inode;
2208 #else
2209         struct ext3_extents_tree tree;
2210         struct ext3_ext_base *base = &tree;
2211
2212         ext3_init_tree_desc(base, inode);
2213 #endif
2214         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2215          * metablocks at maxium b=16542 */
2216         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2217         return 0;
2218 }
2219
2220 #endif
2221
2222 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2223 {
2224         return (EXT3_SB(sb)->journal_bdev);
2225 }
2226 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2227
2228 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2229                           char *buf, int count, loff_t pos)
2230 {
2231         loff_t p = pos;
2232         int rc;
2233
2234         if (!f && !inode) {
2235                 CERROR("lustre_read_quota failed for no quota file!\n");
2236                 libcfs_debug_dumpstack(NULL);
2237                 return -EINVAL;
2238         }
2239
2240         /* Support for both adm and op quota files must be provided */
2241         if (f) {
2242                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2243                 rc = rc < 0 ? rc : p - pos;
2244         } else {
2245                 struct super_block *sb = inode->i_sb;
2246                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2247         }
2248         return rc;
2249 }
2250
2251 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2252 {
2253         loff_t p = pos;
2254         int rc;
2255
2256         /* Only adm quota files are supported, op updates are handled by vfs */
2257         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2258         rc = rc < 0 ? rc : p - pos;
2259
2260         return rc;
2261 }
2262
2263 void *lustre_quota_journal_start(struct inode *inode, int delete)
2264 {
2265         handle_t *handle;
2266         unsigned block_count;
2267
2268         if (delete) {
2269                 /* each indirect block (+4) may become free, attaching to the
2270                  * header list of free blocks (+1); the data block (+1) may
2271                  * become a free block (+0) or a block with free dqentries (+0) */
2272                 block_count = (4 + 1) + 1;
2273                 handle = ext3_journal_start(inode,
2274                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2275         } else {
2276                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2277                  * freeblk reusage with a header update (+1); dqentry is either reused
2278                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2279                  * a new block allocation (+1) with a header update (+1)              */
2280                 block_count = (4 + 1) + 3;
2281                 handle = ext3_journal_start(inode,
2282                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2283
2284         }
2285
2286         return handle;
2287 }
2288
2289 void lustre_quota_journal_stop(void *handle)
2290 {
2291         ext3_journal_stop((handle_t *)handle);
2292 }
2293
2294 static struct fsfilt_operations fsfilt_ext3_ops = {
2295         .fs_type                = "ext3",
2296         .fs_owner               = THIS_MODULE,
2297         .fs_getlabel            = fsfilt_ext3_get_label,
2298         .fs_setlabel            = fsfilt_ext3_set_label,
2299         .fs_uuid                = fsfilt_ext3_uuid,
2300         .fs_start               = fsfilt_ext3_start,
2301         .fs_brw_start           = fsfilt_ext3_brw_start,
2302         .fs_extend              = fsfilt_ext3_extend,
2303         .fs_commit              = fsfilt_ext3_commit,
2304         .fs_commit_async        = fsfilt_ext3_commit_async,
2305         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2306         .fs_setattr             = fsfilt_ext3_setattr,
2307         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2308         .fs_set_md              = fsfilt_ext3_set_md,
2309         .fs_get_md              = fsfilt_ext3_get_md,
2310         .fs_readpage            = fsfilt_ext3_readpage,
2311         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2312         .fs_statfs              = fsfilt_ext3_statfs,
2313         .fs_sync                = fsfilt_ext3_sync,
2314         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2315         .fs_write_record        = fsfilt_ext3_write_record,
2316         .fs_read_record         = fsfilt_ext3_read_record,
2317         .fs_setup               = fsfilt_ext3_setup,
2318         .fs_send_bio            = fsfilt_ext3_send_bio,
2319         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2320 #ifdef HAVE_DISK_INODE_VERSION
2321         .fs_get_version         = fsfilt_ext3_get_version,
2322         .fs_set_version         = fsfilt_ext3_set_version,
2323 #endif
2324 #ifdef HAVE_QUOTA_SUPPORT
2325         .fs_quotactl            = fsfilt_ext3_quotactl,
2326         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2327         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2328         .fs_qids                = fsfilt_ext3_qids,
2329         .fs_dquot               = fsfilt_ext3_dquot,
2330         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2331 #endif
2332         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2333 };
2334
2335 static int __init fsfilt_ext3_init(void)
2336 {
2337         int rc;
2338
2339         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2340                                          sizeof(struct fsfilt_cb_data), 0, 0);
2341         if (!fcb_cache) {
2342                 CERROR("error allocating fsfilt journal callback cache\n");
2343                 GOTO(out, rc = -ENOMEM);
2344         }
2345
2346         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2347
2348         if (rc) {
2349                 int err = cfs_mem_cache_destroy(fcb_cache);
2350                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2351         }
2352 out:
2353         return rc;
2354 }
2355
2356 static void __exit fsfilt_ext3_exit(void)
2357 {
2358         int rc;
2359
2360         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2361         rc = cfs_mem_cache_destroy(fcb_cache);
2362         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2363 }
2364
2365 module_init(fsfilt_ext3_init);
2366 module_exit(fsfilt_ext3_exit);
2367
2368 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2369 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2370 MODULE_LICENSE("GPL");