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  2008 Sun Microsystems, Inc. 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 #include <linux/parser.h>
  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/kp30.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 HAVE_EXT4_LDISKFS
  86 #include <ext4/ext4_extents.h>
  87 #else
  88 #include <linux/ext3_extents.h>
  89 #endif
  90
  91 #include "lustre_quota_fmt.h"
  92
  93 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  94 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  95 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  96 #else
  97 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  98 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  99 #endif
 100
 101 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 102 /* for kernels 2.6.18 and later */
 103 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 104 #else
 105 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 106 #endif
 107
 108 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 109 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 110                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 111 #else
 112 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 113                ext3_ext_insert_extent(handle, inode, path, newext)
 114 #endif
 115
 116
 117 static cfs_mem_cache_t *fcb_cache;
 118
 119 struct fsfilt_cb_data {
 120         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 121         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 122         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 123         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 124         void *cb_data;                  /* MDS/OST completion function data */
 125 };
 126
 127 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 128 #define EXT3_XATTR_INDEX_TRUSTED        4
 129 #endif
 130 #ifndef XATTR_NO_CTIME
 131 #define XATTR_NO_CTIME 0x80
 132 #endif
 133
 134 #ifdef HAVE_EXT4_LDISKFS
 135 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 136 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 137 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 138 #else
 139 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 140 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 141 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 142 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 143 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 144 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 145 #endif
 146
 147 #ifndef ext3_find_next_bit
 148 #define ext3_find_next_bit           ext2_find_next_bit
 149 #endif
 150
 151 #ifndef ext2_find_next_bit
 152 #ifdef __LITTLE_ENDIAN
 153 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 154 #else
 155 error "Need implementation of find_next_bit on big-endian systems"
 156 #endif  /* __LITTLE_ENDIAN */
 157 #endif  /* !ext2_find_next_le_bit */
 158
 159 static char *fsfilt_ext3_get_label(struct super_block *sb)
 160 {
 161         return EXT3_SB(sb)->s_es->s_volume_name;
 162 }
 163
 164 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 165 {
 166         /* see e.g. fsfilt_ext3_write_record() */
 167         journal_t *journal;
 168         handle_t *handle;
 169         int err;
 170
 171         journal = EXT3_SB(sb)->s_journal;
 172         handle = ext3_journal_start_sb(sb, 1);
 173         if (IS_ERR(handle)) {
 174                 CERROR("can't start transaction\n");
 175                 return(PTR_ERR(handle));
 176         }
 177
 178         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 179         if (err)
 180                 goto out;
 181
 182         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 183                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 184
 185         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 186
 187 out:
 188         ext3_journal_stop(handle);
 189
 190         return(err);
 191 }
 192
 193 static char *fsfilt_ext3_uuid(struct super_block *sb)
 194 {
 195         return EXT3_SB(sb)->s_es->s_uuid;
 196 }
 197
 198 #ifdef HAVE_DISK_INODE_VERSION
 199
 200 static __u64 get_i_version(struct inode *inode)
 201 {
 202 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && !defined(USE_I_FS_VERSION)
 203         return inode->i_version;
 204 #else
 205         return EXT3_I(inode)->i_fs_version;
 206 #endif
 207 }
 208
 209 static void set_i_version(struct inode *inode, __u64 new_version)
 210 {
 211 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && !defined(USE_I_FS_VERSION)
 212         inode->i_version = new_version;
 213 #else
 214         (EXT3_I(inode))->i_fs_version = new_version;
 215 #endif
 216 }
 217
 218 /*
 219  * Get the 64-bit version for an inode.
 220  */
 221 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 222 {
 223         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 224                get_i_version(inode), inode->i_ino);
 225         return get_i_version(inode);
 226 }
 227
 228 /*
 229  * Set the 64-bit version and return the old version.
 230  */
 231 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 232 {
 233         __u64 old_version = get_i_version(inode);
 234
 235         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 236                new_version, old_version, inode->i_ino);
 237         set_i_version(inode, new_version);
 238         /* version is set after all inode operations are finished, so we should
 239          * mark it dirty here */
 240         inode->i_sb->s_op->dirty_inode(inode);
 241         return old_version;
 242 }
 243
 244 #endif
 245
 246 /*
 247  * We don't currently need any additional blocks for rmdir and
 248  * unlink transactions because we are storing the OST oa_id inside
 249  * the inode (which we will be changing anyways as part of this
 250  * transaction).
 251  */
 252 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 253                                int logs)
 254 {
 255         /* For updates to the last received file */
 256         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 257         journal_t *journal;
 258         void *handle;
 259
 260         if (current->journal_info) {
 261                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 262                        current->journal_info);
 263                 goto journal_start;
 264         }
 265
 266         switch(op) {
 267         case FSFILT_OP_RMDIR:
 268         case FSFILT_OP_UNLINK:
 269                 /* delete one file + create/update logs for each stripe */
 270                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 271                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 272                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 273                 break;
 274         case FSFILT_OP_RENAME:
 275                 /* modify additional directory */
 276                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 277                 /* no break */
 278         case FSFILT_OP_SYMLINK:
 279                 /* additional block + block bitmap + GDT for long symlink */
 280                 nblocks += 3;
 281                 /* no break */
 282         case FSFILT_OP_CREATE: {
 283 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 284                 static int warned;
 285                 if (!warned) {
 286                         if (!test_opt(inode->i_sb, EXTENTS)) {
 287                                 warned = 1;
 288                         } else if (((EXT3_I(inode)->i_flags &
 289                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 290                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 291                                 CWARN("extent-mapped directory found with "
 292                                       "ext3-based ldiskfs - contact "
 293                                       "http://bugzilla.lustre.org/\n");
 294                                 warned = 1;
 295                         }
 296                 }
 297 #endif
 298                 /* no break */
 299         }
 300         case FSFILT_OP_MKDIR:
 301         case FSFILT_OP_MKNOD:
 302                 /* modify one inode + block bitmap + GDT */
 303                 nblocks += 3;
 304                 /* no break */
 305         case FSFILT_OP_LINK:
 306                 /* modify parent directory */
 307                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 308                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 309                 /* create/update logs for each stripe */
 310                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 311                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 312                 break;
 313         case FSFILT_OP_SETATTR:
 314                 /* Setattr on inode */
 315                 nblocks += 1;
 316                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 317                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 318                 /* quota chown log for each stripe */
 319                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 320                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 321                 break;
 322         case FSFILT_OP_CANCEL_UNLINK:
 323                 /* blocks for log header bitmap update OR
 324                  * blocks for catalog header bitmap update + unlink of logs */
 325                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 326                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 327                 break;
 328         case FSFILT_OP_JOIN:
 329                 /* delete 2 file(file + array id) + create 1 file (array id)
 330                  * create/update logs for each stripe */
 331                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 332
 333                 /*create array log for head file*/
 334                 nblocks += 3;
 335                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 336                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
 337                 /*update head file array */
 338                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 339                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 340                 break;
 341         default: CERROR("unknown transaction start op %d\n", op);
 342                 LBUG();
 343         }
 344
 345         LASSERT(current->journal_info == desc_private);
 346         journal = EXT3_SB(inode->i_sb)->s_journal;
 347         if (nblocks > journal->j_max_transaction_buffers) {
 348                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 349                       nblocks, op, logs, journal->j_max_transaction_buffers);
 350                 nblocks = journal->j_max_transaction_buffers;
 351         }
 352
 353  journal_start:
 354         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 355         handle = ext3_journal_start(inode, nblocks);
 356
 357         if (!IS_ERR(handle))
 358                 LASSERT(current->journal_info == handle);
 359         else
 360                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 361                        op, nblocks, PTR_ERR(handle));
 362         return handle;
 363 }
 364
 365 /*
 366  * Calculate the number of buffer credits needed to write multiple pages in
 367  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 368  * doesn't have a nice API for calculating this sort of thing in advance.
 369  *
 370  * See comment above ext3_writepage_trans_blocks for details.  We assume
 371  * no data journaling is being done, but it does allow for all of the pages
 372  * being non-contiguous.  If we are guaranteed contiguous pages we could
 373  * reduce the number of (d)indirect blocks a lot.
 374  *
 375  * With N blocks per page and P pages, for each inode we have at most:
 376  * N*P indirect
 377  * min(N*P, blocksize/4 + 1) dindirect blocks
 378  * niocount tindirect
 379  *
 380  * For the entire filesystem, we have at most:
 381  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 382  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 383  * objcount inode blocks
 384  * 1 superblock
 385  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 386  *
 387  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 388  */
 389 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 390                                       int niocount, struct niobuf_local *nb)
 391 {
 392         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 393         __u64 next_indir;
 394         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 395         int nbitmaps = 0, ngdblocks;
 396         int needed = objcount + 1; /* inodes + superblock */
 397         int i, j;
 398
 399         for (i = 0, j = 0; i < objcount; i++, fso++) {
 400                 /* two or more dindirect blocks in case we cross boundary */
 401                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 402                                     nb[j].offset) >>
 403                                    sb->s_blocksize_bits) /
 404                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 405                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 406
 407                 /* leaf, indirect, tindirect blocks for first block */
 408                 nbitmaps += blockpp + 2;
 409
 410                 j += fso->fso_bufcnt;
 411         }
 412
 413         next_indir = nb[0].offset +
 414                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 415         for (i = 1; i < niocount; i++) {
 416                 if (nb[i].offset >= next_indir) {
 417                         nbitmaps++;     /* additional indirect */
 418                         next_indir = nb[i].offset +
 419                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 420                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 421                         nbitmaps++;     /* additional indirect */
 422                 }
 423                 nbitmaps += blockpp;    /* each leaf in different group? */
 424         }
 425
 426         ngdblocks = nbitmaps;
 427         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 428                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 429         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 430                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 431
 432         needed += nbitmaps + ngdblocks;
 433
 434         /* last_rcvd update */
 435         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 436
 437 #if defined(CONFIG_QUOTA)
 438         /* We assume that there will be 1 bit set in s_dquot.flags for each
 439          * quota file that is active.  This is at least true for now.
 440          */
 441         needed += hweight32(ll_sb_any_quota_active(sb)) *
 442                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 443 #endif
 444
 445         return needed;
 446 }
 447
 448 /* We have to start a huge journal transaction here to hold all of the
 449  * metadata for the pages being written here.  This is necessitated by
 450  * the fact that we do lots of prepare_write operations before we do
 451  * any of the matching commit_write operations, so even if we split
 452  * up to use "smaller" transactions none of them could complete until
 453  * all of them were opened.  By having a single journal transaction,
 454  * we eliminate duplicate reservations for common blocks like the
 455  * superblock and group descriptors or bitmaps.
 456  *
 457  * We will start the transaction here, but each prepare_write will
 458  * add a refcount to the transaction, and each commit_write will
 459  * remove a refcount.  The transaction will be closed when all of
 460  * the pages have been written.
 461  */
 462 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 463                                    int niocount, struct niobuf_local *nb,
 464                                    void *desc_private, int logs)
 465 {
 466         journal_t *journal;
 467         handle_t *handle;
 468         int needed;
 469         ENTRY;
 470
 471         LASSERT(current->journal_info == desc_private);
 472         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 473         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 474
 475         /* The number of blocks we could _possibly_ dirty can very large.
 476          * We reduce our request if it is absurd (and we couldn't get that
 477          * many credits for a single handle anyways).
 478          *
 479          * At some point we have to limit the size of I/Os sent at one time,
 480          * increase the size of the journal, or we have to calculate the
 481          * actual journal requirements more carefully by checking all of
 482          * the blocks instead of being maximally pessimistic.  It remains to
 483          * be seen if this is a real problem or not.
 484          */
 485         if (needed > journal->j_max_transaction_buffers) {
 486                 CERROR("want too many journal credits (%d) using %d instead\n",
 487                        needed, journal->j_max_transaction_buffers);
 488                 needed = journal->j_max_transaction_buffers;
 489         }
 490
 491         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 492         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 493         if (IS_ERR(handle)) {
 494                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 495                        PTR_ERR(handle));
 496         } else {
 497                 LASSERT(handle->h_buffer_credits >= needed);
 498                 LASSERT(current->journal_info == handle);
 499         }
 500
 501         RETURN(handle);
 502 }
 503
 504 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 505 {
 506        handle_t *handle = h;
 507
 508        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 509        if (nblocks == 0) {
 510                handle->h_buffer_credits = 0;
 511                CWARN("setting credits of handle %p to zero by request\n", h);
 512        }
 513
 514        if (handle->h_buffer_credits > nblocks)
 515                 return 0;
 516        if (ext3_journal_extend(handle, nblocks) == 0)
 517                 return 0;
 518
 519        ext3_mark_inode_dirty(handle, inode);
 520        return ext3_journal_restart(handle, nblocks);
 521 }
 522
 523 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 524 {
 525         int rc;
 526         handle_t *handle = h;
 527
 528         LASSERT(current->journal_info == handle);
 529         if (force_sync)
 530                 handle->h_sync = 1; /* recovery likes this */
 531
 532         rc = ext3_journal_stop(handle);
 533
 534         return rc;
 535 }
 536
 537 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 538                                     void **wait_handle)
 539 {
 540         unsigned long tid;
 541         transaction_t *transaction;
 542         handle_t *handle = h;
 543         journal_t *journal;
 544         int rc;
 545
 546         LASSERT(current->journal_info == handle);
 547
 548         transaction = handle->h_transaction;
 549         journal = transaction->t_journal;
 550         tid = transaction->t_tid;
 551         /* we don't want to be blocked */
 552         handle->h_sync = 0;
 553         rc = ext3_journal_stop(handle);
 554         if (rc) {
 555                 CERROR("error while stopping transaction: %d\n", rc);
 556                 return rc;
 557         }
 558         fsfilt_log_start_commit(journal, tid);
 559
 560         *wait_handle = (void *) tid;
 561         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 562         return 0;
 563 }
 564
 565 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 566 {
 567         journal_t *journal = EXT3_JOURNAL(inode);
 568         tid_t tid = (tid_t)(long)h;
 569
 570         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 571         if (unlikely(is_journal_aborted(journal)))
 572                 return -EIO;
 573
 574         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 575
 576         if (unlikely(is_journal_aborted(journal)))
 577                 return -EIO;
 578         return 0;
 579 }
 580
 581 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 582                                struct iattr *iattr, int do_trunc)
 583 {
 584         struct inode *inode = dentry->d_inode;
 585         int rc = 0;
 586
 587         LASSERTF(!(iattr->ia_valid & ATTR_SIZE) || !S_ISDIR(inode->i_mode),
 588                  "changing i_size on directory #%lu (%p) new %llu old %llu\n",
 589                  inode->i_ino, inode, (unsigned long long)iattr->ia_size,
 590                  (unsigned long long)i_size_read(inode));
 591
 592         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 593          * We are already writing the inode to disk as part of this
 594          * transaction and want to avoid a lot of extra inode writeout
 595          * later on. b=9828 */
 596         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 597                 /* ATTR_SIZE would invoke truncate: clear it */
 598                 iattr->ia_valid &= ~ATTR_SIZE;
 599                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 600                 i_size_write(inode, iattr->ia_size);
 601
 602                 if (iattr->ia_valid & ATTR_UID)
 603                         inode->i_uid = iattr->ia_uid;
 604                 if (iattr->ia_valid & ATTR_GID)
 605                         inode->i_gid = iattr->ia_gid;
 606                 if (iattr->ia_valid & ATTR_ATIME)
 607                         inode->i_atime = iattr->ia_atime;
 608                 if (iattr->ia_valid & ATTR_MTIME)
 609                         inode->i_mtime = iattr->ia_mtime;
 610                 if (iattr->ia_valid & ATTR_CTIME)
 611                         inode->i_ctime = iattr->ia_ctime;
 612                 if (iattr->ia_valid & ATTR_MODE) {
 613                         inode->i_mode = iattr->ia_mode;
 614
 615                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 616                                 inode->i_mode &= ~S_ISGID;
 617                 }
 618
 619                 inode->i_sb->s_op->dirty_inode(inode);
 620
 621                 goto out;
 622         }
 623
 624         /* Don't allow setattr to change file type */
 625         if (iattr->ia_valid & ATTR_MODE)
 626                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 627                                  (iattr->ia_mode & ~S_IFMT);
 628
 629         /* We set these flags on the client, but have already checked perms
 630          * so don't confuse inode_change_ok. */
 631         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 632
 633         if (inode->i_op->setattr) {
 634                 rc = inode->i_op->setattr(dentry, iattr);
 635         } else {
 636                 rc = inode_change_ok(inode, iattr);
 637                 if (!rc)
 638                         rc = inode_setattr(inode, iattr);
 639         }
 640
 641  out:
 642         RETURN(rc);
 643 }
 644
 645 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 646                                  unsigned int cmd, unsigned long arg)
 647 {
 648         int rc = 0;
 649         ENTRY;
 650
 651         /* FIXME: Can't do this because of nested transaction deadlock */
 652         if (cmd == EXT3_IOC_SETFLAGS) {
 653                 /* We can't enable data journaling on OST objects, because
 654                 * this forces the transaction to be closed in order to
 655                 * flush the journal, but the caller will already have a
 656                 * compound transaction open to update the last_rcvd file,
 657                 * and this thread would deadlock trying to set the flag. */
 658                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 659                         CERROR("can't set data journal flag on file\n");
 660                         RETURN(-EPERM);
 661                 }
 662                 /* Because the MDS does not see the EXTENTS_FL set on the
 663                  * OST objects, mask this flag into all set flags.  It is
 664                  * not legal to clear this flag in any case, so we are not
 665                  * changing the functionality by doing this.  b=22911 */
 666                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 667         }
 668
 669 #ifdef HAVE_EXT4_LDISKFS
 670         /* ext4_ioctl does not have a inode argument */
 671         if (inode->i_fop->unlocked_ioctl)
 672                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 673 #else
 674         if (inode->i_fop->ioctl)
 675                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 676 #endif
 677         else
 678                 RETURN(-ENOTTY);
 679
 680         RETURN(rc);
 681 }
 682
 683 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 684                               void *lmm, int lmm_size, const char *name)
 685 {
 686         int rc;
 687
 688         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 689
 690         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 691                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 692
 693         if (rc && rc != -EROFS)
 694                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 695                        inode->i_ino, rc);
 696         return rc;
 697 }
 698
 699 /* Must be called with i_mutex held */
 700 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 701                               const char *name)
 702 {
 703         int rc;
 704
 705         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 706
 707         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 708                             name, lmm, lmm_size);
 709         /* This gives us the MD size */
 710         if (lmm == NULL)
 711                 return (rc == -ENODATA) ? 0 : rc;
 712
 713         if (rc < 0) {
 714                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 715                        EXT3_XATTR_INDEX_TRUSTED, name,
 716                        inode->i_ino, rc);
 717                 memset(lmm, 0, lmm_size);
 718                 return (rc == -ENODATA) ? 0 : rc;
 719         }
 720
 721         return rc;
 722 }
 723
 724 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 725 {
 726         submit_bio(rw, bio);
 727         return 0;
 728 }
 729
 730 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 731                                     loff_t *off)
 732 {
 733         struct inode *inode = file->f_dentry->d_inode;
 734         int rc = 0;
 735
 736         if (S_ISREG(inode->i_mode))
 737                 rc = file->f_op->read(file, buf, count, off);
 738         else {
 739                 const int blkbits = inode->i_sb->s_blocksize_bits;
 740                 const int blksize = inode->i_sb->s_blocksize;
 741
 742                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 743                        count, inode->i_ino, *off);
 744                 while (count > 0) {
 745                         struct buffer_head *bh;
 746
 747                         bh = NULL;
 748                         if (*off < i_size_read(inode)) {
 749                                 int err = 0;
 750
 751                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 752                                                 0, &err);
 753
 754                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 755
 756                                 if (bh) {
 757                                         memcpy(buf, bh->b_data, blksize);
 758                                         brelse(bh);
 759                                 } else if (err) {
 760                                         /* XXX in theory we should just fake
 761                                          * this buffer and continue like ext3,
 762                                          * especially if this is a partial read
 763                                          */
 764                                         CERROR("error read dir %lu+%llu: %d\n",
 765                                                inode->i_ino, *off, err);
 766                                         RETURN(err);
 767                                 }
 768                         }
 769                         if (!bh) {
 770                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 771
 772                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 773                                 memset(fake, 0, sizeof(*fake));
 774                                 fake->rec_len = cpu_to_le16(blksize);
 775                         }
 776                         count -= blksize;
 777                         buf += blksize;
 778                         *off += blksize;
 779                         rc += blksize;
 780                 }
 781         }
 782
 783         return rc;
 784 }
 785
 786 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 787 {
 788         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 789
 790         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 791
 792         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 793 }
 794
 795 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 796                                       void *handle, fsfilt_cb_t cb_func,
 797                                       void *cb_data)
 798 {
 799         struct fsfilt_cb_data *fcb;
 800
 801         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 802         if (fcb == NULL)
 803                 RETURN(-ENOMEM);
 804
 805         fcb->cb_func = cb_func;
 806         fcb->cb_obd = obd;
 807         fcb->cb_last_rcvd = last_rcvd;
 808         fcb->cb_data = cb_data;
 809
 810         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 811         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 812                                     (struct journal_callback *)fcb);
 813
 814         return 0;
 815 }
 816
 817 /*
 818  * We need to hack the return value for the free inode counts because
 819  * the current EA code requires one filesystem block per inode with EAs,
 820  * so it is possible to run out of blocks before we run out of inodes.
 821  *
 822  * This can be removed when the ext3 EA code is fixed.
 823  */
 824 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 825 {
 826         struct kstatfs sfs;
 827         int rc;
 828
 829         memset(&sfs, 0, sizeof(sfs));
 830         rc = ll_do_statfs(sb,&sfs);
 831
 832         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 833                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 834                 sfs.f_ffree = sfs.f_bfree;
 835         }
 836
 837         statfs_pack(osfs, &sfs);
 838         return rc;
 839 }
 840
 841 static int fsfilt_ext3_sync(struct super_block *sb)
 842 {
 843         return ext3_force_commit(sb);
 844 }
 845
 846 #ifndef EXT3_EXTENTS_FL
 847 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 848 #endif
 849
 850 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 851 #define fsfilt_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 852 #define fsfilt_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 853 #else
 854 #ifdef HAVE_EXT4_LDISKFS
 855 #define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem));
 856 #define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem));
 857 #else
 858 #define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 859 #define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 860 #endif
 861 #endif
 862
 863 #ifndef EXT_ASSERT
 864 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 865 #endif
 866
 867 #ifdef EXT3_EXT_HAS_NO_TREE
 868 /* for kernels 2.6.18 and later */
 869 #ifdef HAVE_EXT4_LDISKFS
 870 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 871 #else
 872 #define EXT_GENERATION(inode)           ext_generation(inode)
 873 #endif
 874 #define ext3_ext_base                   inode
 875 #define ext3_ext_base2inode(inode)      (inode)
 876 #define EXT_DEPTH(inode)                ext_depth(inode)
 877 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 878                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 879 #else
 880 #define ext3_ext_base                   ext3_extents_tree
 881 #define ext3_ext_base2inode(tree)       (tree->inode)
 882 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 883                         ext3_ext_walk_space(tree, block, num, cb);
 884 #endif
 885
 886 #include <linux/lustre_version.h>
 887
 888 struct bpointers {
 889         unsigned long *blocks;
 890         int *created;
 891         unsigned long start;
 892         int num;
 893         int init_num;
 894         int create;
 895 };
 896
 897 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 898                               unsigned long block, int *aflags)
 899 {
 900         struct ext3_inode_info *ei = EXT3_I(inode);
 901         unsigned long bg_start;
 902         unsigned long colour;
 903         int depth;
 904
 905         if (path) {
 906                 struct ext3_extent *ex;
 907                 depth = path->p_depth;
 908
 909                 /* try to predict block placement */
 910                 if ((ex = path[depth].p_ext))
 911                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 912
 913                 /* it looks index is empty
 914                  * try to find starting from index itself */
 915                 if (path[depth].p_bh)
 916                         return path[depth].p_bh->b_blocknr;
 917         }
 918
 919         /* OK. use inode's group */
 920         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 921                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 922         colour = (current->pid % 16) *
 923                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 924         return bg_start + colour + block;
 925 }
 926
 927 #define ll_unmap_underlying_metadata(sb, blocknr) \
 928         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 929
 930 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 931 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 932                                 struct ext3_ext_path *path, unsigned long block,
 933                                 unsigned long *count, int *err)
 934 {
 935         unsigned long pblock, goal;
 936         int aflags = 0;
 937         struct inode *inode = ext3_ext_base2inode(base);
 938
 939         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 940         aflags |= 2; /* block have been already reserved */
 941         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 942         return pblock;
 943
 944 }
 945 #else
 946 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 947                                 struct ext3_ext_path *path, unsigned long block,
 948                                 unsigned long *count, int *err)
 949 {
 950         struct inode *inode = ext3_ext_base2inode(base);
 951         struct ext3_allocation_request ar;
 952         unsigned long pblock;
 953         int aflags;
 954
 955         /* find neighbour allocated blocks */
 956         ar.lleft = block;
 957         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 958         if (*err)
 959                 return 0;
 960         ar.lright = block;
 961         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 962         if (*err)
 963                 return 0;
 964
 965         /* allocate new block */
 966         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 967         ar.inode = inode;
 968         ar.logical = block;
 969         ar.len = *count;
 970         ar.flags = EXT3_MB_HINT_DATA;
 971         pblock = ext3_mb_new_blocks(handle, &ar, err);
 972         *count = ar.len;
 973         return pblock;
 974 }
 975 #endif
 976
 977 #ifdef EXT3_EXT_HAS_NO_TREE
 978 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 979                                   struct ext3_ext_path *path,
 980                                   struct ext3_ext_cache *cex,
 981 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 982                                   struct ext3_extent *ex,
 983 #endif
 984                                   void *cbdata)
 985 {
 986         struct bpointers *bp = cbdata;
 987 #else
 988 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 989                                   struct ext3_ext_path *path,
 990                                   struct ext3_ext_cache *cex
 991 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 992                                   , struct ext3_extent *ex
 993 #endif
 994                                  )
 995 {
 996         struct bpointers *bp = base->private;
 997 #endif
 998         struct inode *inode = ext3_ext_base2inode(base);
 999         struct ext3_extent nex;
1000         unsigned long pblock;
1001         unsigned long tgen;
1002         int err, i;
1003         unsigned long count;
1004         handle_t *handle;
1005
1006         i = EXT_DEPTH(base);
1007         EXT_ASSERT(i == path->p_depth);
1008         EXT_ASSERT(path[i].p_hdr);
1009
1010         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1011                 err = EXT_CONTINUE;
1012                 goto map;
1013         }
1014
1015         if (bp->create == 0) {
1016                 i = 0;
1017                 if (cex->ec_block < bp->start)
1018                         i = bp->start - cex->ec_block;
1019                 if (i >= cex->ec_len)
1020                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1021                                         i, cex->ec_len);
1022                 for (; i < cex->ec_len && bp->num; i++) {
1023                         *(bp->created) = 0;
1024                         bp->created++;
1025                         *(bp->blocks) = 0;
1026                         bp->blocks++;
1027                         bp->num--;
1028                         bp->start++;
1029                 }
1030
1031                 return EXT_CONTINUE;
1032         }
1033
1034         tgen = EXT_GENERATION(base);
1035         count = ext3_ext_calc_credits_for_insert(base, path);
1036         fsfilt_up_truncate_sem(inode);
1037         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1038         if (IS_ERR(handle)) {
1039                 fsfilt_down_truncate_sem(inode);
1040                 return PTR_ERR(handle);
1041         }
1042
1043         fsfilt_down_truncate_sem(inode);
1044         if (tgen != EXT_GENERATION(base)) {
1045                 /* the tree has changed. so path can be invalid at moment */
1046                 ext3_journal_stop(handle);
1047                 return EXT_REPEAT;
1048         }
1049
1050         count = cex->ec_len;
1051         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1052         if (!pblock)
1053                 goto out;
1054         EXT_ASSERT(count <= cex->ec_len);
1055
1056         /* insert new extent */
1057         nex.ee_block = cpu_to_le32(cex->ec_block);
1058         ext3_ext_store_pblock(&nex, pblock);
1059         nex.ee_len = cpu_to_le16(count);
1060         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1061         if (err) {
1062                 /* free data blocks we just allocated */
1063                 /* not a good idea to call discard here directly,
1064                  * but otherwise we'd need to call it every free() */
1065 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1066                 ext3_mb_discard_inode_preallocations(inode);
1067 #endif
1068                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1069                                  cpu_to_le16(nex.ee_len), 0);
1070                 goto out;
1071         }
1072
1073         /*
1074          * Putting len of the actual extent we just inserted,
1075          * we are asking ext3_ext_walk_space() to continue
1076          * scaning after that block
1077          */
1078         cex->ec_len = le16_to_cpu(nex.ee_len);
1079         cex->ec_start = ext_pblock(&nex);
1080         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1081         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1082
1083 out:
1084         ext3_journal_stop(handle);
1085 map:
1086         if (err >= 0) {
1087                 /* map blocks */
1088                 if (bp->num == 0) {
1089                         CERROR("hmm. why do we find this extent?\n");
1090                         CERROR("initial space: %lu:%u\n",
1091                                 bp->start, bp->init_num);
1092                         CERROR("current extent: %u/%u/%llu %d\n",
1093                                 cex->ec_block, cex->ec_len,
1094                                 (unsigned long long)cex->ec_start,
1095                                 cex->ec_type);
1096                 }
1097                 i = 0;
1098                 if (cex->ec_block < bp->start)
1099                         i = bp->start - cex->ec_block;
1100                 if (i >= cex->ec_len)
1101                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1102                                         i, cex->ec_len);
1103                 for (; i < cex->ec_len && bp->num; i++) {
1104                         *(bp->blocks) = cex->ec_start + i;
1105                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1106                                 *(bp->created) = 0;
1107                         } else {
1108                                 *(bp->created) = 1;
1109                                 /* unmap any possible underlying metadata from
1110                                  * the block device mapping.  bug 6998. */
1111                                 ll_unmap_underlying_metadata(inode->i_sb,
1112                                                              *(bp->blocks));
1113                         }
1114                         bp->created++;
1115                         bp->blocks++;
1116                         bp->num--;
1117                         bp->start++;
1118                 }
1119         }
1120         return err;
1121 }
1122
1123 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1124                        unsigned long num, unsigned long *blocks,
1125                        int *created, int create)
1126 {
1127 #ifdef EXT3_EXT_HAS_NO_TREE
1128         struct ext3_ext_base *base = inode;
1129 #else
1130         struct ext3_extents_tree tree;
1131         struct ext3_ext_base *base = &tree;
1132 #endif
1133         struct bpointers bp;
1134         int err;
1135
1136         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1137                block, block + num - 1, (unsigned) inode->i_ino);
1138
1139 #ifndef EXT3_EXT_HAS_NO_TREE
1140         ext3_init_tree_desc(base, inode);
1141         tree.private = &bp;
1142 #endif
1143         bp.blocks = blocks;
1144         bp.created = created;
1145         bp.start = block;
1146         bp.init_num = bp.num = num;
1147         bp.create = create;
1148
1149         fsfilt_down_truncate_sem(inode);
1150         err = fsfilt_ext3_ext_walk_space(base, block, num, ext3_ext_new_extent_cb, &bp);
1151         ext3_ext_invalidate_cache(base);
1152         fsfilt_up_truncate_sem(inode);
1153
1154         return err;
1155 }
1156
1157 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1158                                     int pages, unsigned long *blocks,
1159                                     int *created, int create)
1160 {
1161         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1162         int rc = 0, i = 0;
1163         struct page *fp = NULL;
1164         int clen = 0;
1165
1166         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1167                 inode->i_ino, pages, (*page)->index);
1168
1169         /* pages are sorted already. so, we just have to find
1170          * contig. space and process them properly */
1171         while (i < pages) {
1172                 if (fp == NULL) {
1173                         /* start new extent */
1174                         fp = *page++;
1175                         clen = 1;
1176                         i++;
1177                         continue;
1178                 } else if (fp->index + clen == (*page)->index) {
1179                         /* continue the extent */
1180                         page++;
1181                         clen++;
1182                         i++;
1183                         continue;
1184                 }
1185
1186                 /* process found extent */
1187                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1188                                         clen * blocks_per_page, blocks,
1189                                         created, create);
1190                 if (rc)
1191                         GOTO(cleanup, rc);
1192
1193                 /* look for next extent */
1194                 fp = NULL;
1195                 blocks += blocks_per_page * clen;
1196                 created += blocks_per_page * clen;
1197         }
1198
1199         if (fp)
1200                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1201                                         clen * blocks_per_page, blocks,
1202                                         created, create);
1203 cleanup:
1204         return rc;
1205 }
1206
1207 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1208                                unsigned long *blocks, int *created, int create);
1209 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1210                                    int pages, unsigned long *blocks,
1211                                    int *created, int create)
1212 {
1213         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1214         unsigned long *b;
1215         int rc = 0, i, *cr;
1216
1217         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1218                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1219                 if (rc) {
1220                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1221                                inode->i_ino, *b, *cr, create, rc);
1222                         break;
1223                 }
1224
1225                 b += blocks_per_page;
1226                 cr += blocks_per_page;
1227         }
1228         return rc;
1229 }
1230
1231 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1232                                 int pages, unsigned long *blocks,
1233                                 int *created, int create,
1234                                 struct semaphore *optional_sem)
1235 {
1236         int rc;
1237
1238         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1239                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1240                                                      blocks, created, create);
1241                 return rc;
1242         }
1243         if (optional_sem != NULL)
1244                 down(optional_sem);
1245         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1246                                             created, create);
1247         if (optional_sem != NULL)
1248                 up(optional_sem);
1249
1250         return rc;
1251 }
1252
1253 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1254                                    int size, loff_t *offs)
1255 {
1256         struct inode *inode = file->f_dentry->d_inode;
1257         unsigned long block;
1258         struct buffer_head *bh;
1259         int err, blocksize, csize, boffs;
1260
1261         /* prevent reading after eof */
1262         lock_kernel();
1263         if (i_size_read(inode) < *offs + size) {
1264                 size = i_size_read(inode) - *offs;
1265                 unlock_kernel();
1266                 if (size < 0) {
1267                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1268                                i_size_read(inode), *offs);
1269                         return -EBADR;
1270                 } else if (size == 0) {
1271                         return 0;
1272                 }
1273         } else {
1274                 unlock_kernel();
1275         }
1276
1277         blocksize = 1 << inode->i_blkbits;
1278
1279         while (size > 0) {
1280                 block = *offs >> inode->i_blkbits;
1281                 boffs = *offs & (blocksize - 1);
1282                 csize = min(blocksize - boffs, size);
1283                 bh = ext3_bread(NULL, inode, block, 0, &err);
1284                 if (!bh) {
1285                         CERROR("can't read block: %d\n", err);
1286                         return err;
1287                 }
1288
1289                 memcpy(buf, bh->b_data + boffs, csize);
1290                 brelse(bh);
1291
1292                 *offs += csize;
1293                 buf += csize;
1294                 size -= csize;
1295         }
1296         return 0;
1297 }
1298
1299 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1300                                     loff_t *offs, int force_sync)
1301 {
1302         struct buffer_head *bh = NULL;
1303         unsigned long block;
1304         struct inode *inode = file->f_dentry->d_inode;
1305         loff_t old_size = i_size_read(inode), offset = *offs;
1306         loff_t new_size = i_size_read(inode);
1307         handle_t *handle;
1308         int err = 0, block_count = 0, blocksize, size, boffs;
1309
1310         /* Determine how many transaction credits are needed */
1311         blocksize = 1 << inode->i_blkbits;
1312         block_count = (*offs & (blocksize - 1)) + bufsize;
1313         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1314
1315         handle = ext3_journal_start(inode,
1316                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1317         if (IS_ERR(handle)) {
1318                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1319                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1320                 return PTR_ERR(handle);
1321         }
1322
1323         while (bufsize > 0) {
1324                 if (bh != NULL)
1325                         brelse(bh);
1326
1327                 block = offset >> inode->i_blkbits;
1328                 boffs = offset & (blocksize - 1);
1329                 size = min(blocksize - boffs, bufsize);
1330                 bh = ext3_bread(handle, inode, block, 1, &err);
1331                 if (!bh) {
1332                         CERROR("can't read/create block: %d\n", err);
1333                         goto out;
1334                 }
1335
1336                 err = ext3_journal_get_write_access(handle, bh);
1337                 if (err) {
1338                         CERROR("journal_get_write_access() returned error %d\n",
1339                                err);
1340                         goto out;
1341                 }
1342                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1343                 memcpy(bh->b_data + boffs, buf, size);
1344                 err = ext3_journal_dirty_metadata(handle, bh);
1345                 if (err) {
1346                         CERROR("journal_dirty_metadata() returned error %d\n",
1347                                err);
1348                         goto out;
1349                 }
1350                 if (offset + size > new_size)
1351                         new_size = offset + size;
1352                 offset += size;
1353                 bufsize -= size;
1354                 buf += size;
1355         }
1356
1357         if (force_sync)
1358                 handle->h_sync = 1; /* recovery likes this */
1359 out:
1360         if (bh)
1361                 brelse(bh);
1362
1363         /* correct in-core and on-disk sizes */
1364         if (new_size > i_size_read(inode)) {
1365                 lock_kernel();
1366                 if (new_size > i_size_read(inode))
1367                         i_size_write(inode, new_size);
1368                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1369                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1370                 if (i_size_read(inode) > old_size)
1371                         mark_inode_dirty(inode);
1372                 unlock_kernel();
1373         }
1374
1375         ext3_journal_stop(handle);
1376
1377         if (err == 0)
1378                 *offs = offset;
1379         return err;
1380 }
1381
1382 static int fsfilt_ext3_setup(struct super_block *sb)
1383 {
1384 #if !defined(S_PDIROPS) && (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1385         struct ext3_sb_info *sbi = EXT3_SB(sb);
1386 #endif
1387 #if 0
1388         sbi->dx_lock = fsfilt_ext3_dx_lock;
1389         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1390 #endif
1391 #ifdef S_PDIROPS
1392         CWARN("Enabling PDIROPS\n");
1393         set_opt(sbi->s_mount_opt, PDIROPS);
1394         sb->s_flags |= S_PDIROPS;
1395 #endif
1396         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1397                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1398 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1399         /* enable journaled quota support */
1400         /* kfreed in ext3_put_super() */
1401         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1402         if (!sbi->s_qf_names[USRQUOTA])
1403                 return -ENOMEM;
1404         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1405         if (!sbi->s_qf_names[GRPQUOTA]) {
1406                 kfree(sbi->s_qf_names[USRQUOTA]);
1407                 sbi->s_qf_names[USRQUOTA] = NULL;
1408                 return -ENOMEM;
1409         }
1410         sbi->s_jquota_fmt = QFMT_VFS_V0;
1411 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1412         set_opt(sbi->s_mount_opt, QUOTA);
1413 #endif
1414 #endif
1415         return 0;
1416 }
1417
1418 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1419    objects. Logs is number of logfiles to update */
1420 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1421 {
1422         if ( !fso ) {
1423                 switch(op) {
1424                 case FSFILT_OP_CREATE:
1425                                  /* directory leaf, index & indirect & EA*/
1426                         return 4 + 3 * logs;
1427                 case FSFILT_OP_UNLINK:
1428                         return 3 * logs;
1429                 }
1430         } else {
1431                 int i;
1432                 int needed = 0;
1433                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1434                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1435                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1436                 for (i = 0; i < op; i++, fso++) {
1437                         int nblocks = fso->fso_bufcnt * blockpp;
1438                         int ndindirect = min(nblocks, addrpp + 1);
1439                         int nindir = nblocks + ndindirect + 1;
1440
1441                         needed += nindir;
1442                 }
1443                 return needed + 3 * logs;
1444         }
1445
1446         return 0;
1447 }
1448
1449 #ifdef HAVE_QUOTA_SUPPORT
1450 #define DQINFO_COPY(out, in)                    \
1451 do {                                            \
1452         Q_COPY(out, in, dqi_bgrace);            \
1453         Q_COPY(out, in, dqi_igrace);            \
1454         Q_COPY(out, in, dqi_flags);             \
1455         Q_COPY(out, in, dqi_valid);             \
1456 } while (0)
1457
1458 #define DQBLK_COPY(out, in)                     \
1459 do {                                            \
1460         Q_COPY(out, in, dqb_bhardlimit);        \
1461         Q_COPY(out, in, dqb_bsoftlimit);        \
1462         Q_COPY(out, in, dqb_curspace);          \
1463         Q_COPY(out, in, dqb_ihardlimit);        \
1464         Q_COPY(out, in, dqb_isoftlimit);        \
1465         Q_COPY(out, in, dqb_curinodes);         \
1466         Q_COPY(out, in, dqb_btime);             \
1467         Q_COPY(out, in, dqb_itime);             \
1468         Q_COPY(out, in, dqb_valid);             \
1469 } while (0)
1470
1471
1472
1473 static int fsfilt_ext3_quotactl(struct super_block *sb,
1474                                 struct obd_quotactl *oqc)
1475 {
1476         int i, rc = 0, error = 0;
1477         struct quotactl_ops *qcop;
1478         struct if_dqinfo *info;
1479         struct if_dqblk *dqblk;
1480         ENTRY;
1481
1482         if (!sb->s_qcop)
1483                 RETURN(-ENOSYS);
1484
1485         OBD_ALLOC_PTR(info);
1486         if (!info)
1487                 RETURN(-ENOMEM);
1488         OBD_ALLOC_PTR(dqblk);
1489         if (!dqblk) {
1490                 OBD_FREE_PTR(info);
1491                 RETURN(-ENOMEM);
1492         }
1493
1494         DQINFO_COPY(info, &oqc->qc_dqinfo);
1495         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1496
1497         qcop = sb->s_qcop;
1498         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1499                 for (i = 0; i < MAXQUOTAS; i++) {
1500                         if (!Q_TYPESET(oqc, i))
1501                                 continue;
1502
1503                         if (oqc->qc_cmd == Q_QUOTAON) {
1504                                 lustre_quota_version_t qfmt = oqc->qc_id;
1505                                 char *name[][MAXQUOTAS] = LUSTRE_OPQFILES_NAMES;
1506
1507                                 rc = ll_quota_on(sb, i, QFMT_VFS_V0,
1508                                                  name[qfmt][i], 0);
1509 #ifdef HAVE_QUOTA64
1510                                 if (rc == -ENOENT || rc == -EINVAL) {
1511                                         /* see bug 13904 */
1512                                         rc = lustre_slave_quota_convert(qfmt, i);
1513                                         if (!rc)
1514                                                 rc = ll_quota_on(sb, i,
1515                                                               QFMT_VFS_V0,
1516                                                               name[qfmt][i], 0);
1517                                         else if (rc == -ESTALE)
1518                                                 rc = -ENOENT;
1519                                 }
1520 #endif
1521                         } else if (oqc->qc_cmd == Q_QUOTAOFF)
1522                                 rc = ll_quota_off(sb, i, 0);
1523
1524                         if (rc == -EBUSY)
1525                                 error = rc;
1526                         else if (rc)
1527                                 GOTO(out, rc);
1528                 }
1529                 GOTO(out, rc ?: error);
1530         }
1531
1532         switch (oqc->qc_cmd) {
1533         case Q_GETOINFO:
1534         case Q_GETINFO:
1535                 if (!qcop->get_info)
1536                         GOTO(out, rc = -ENOSYS);
1537                 rc = qcop->get_info(sb, oqc->qc_type, info);
1538                 break;
1539         case Q_SETQUOTA:
1540         case Q_INITQUOTA:
1541                 if (!qcop->set_dqblk)
1542                         GOTO(out, rc = -ENOSYS);
1543                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1544                 break;
1545         case Q_GETOQUOTA:
1546         case Q_GETQUOTA:
1547                 if (!qcop->get_dqblk)
1548                         GOTO(out, rc = -ENOSYS);
1549                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1550                 if (!rc)
1551                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1552                 break;
1553         case Q_SYNC:
1554                 if (!sb->s_qcop->quota_sync)
1555                         GOTO(out, rc = -ENOSYS);
1556                 qcop->quota_sync(sb, oqc->qc_type);
1557                 break;
1558         case Q_FINVALIDATE:
1559                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1560                 for (i = 0; i < MAXQUOTAS; i++) {
1561                         struct file *fp;
1562                         lustre_quota_version_t qfmt = oqc->qc_id;
1563                         char *name[][MAXQUOTAS] = LUSTRE_OPQFILES_NAMES;
1564
1565                         if (!Q_TYPESET(oqc, i))
1566                                 continue;
1567
1568                         fp = filp_open(name[qfmt][i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1569                         if (IS_ERR(fp)) {
1570                                 rc = PTR_ERR(fp);
1571                                 CERROR("error invalidating operational quota file"
1572                                        " %s (rc:%d)\n", name[qfmt][i], rc);
1573                         } else {
1574                                 filp_close(fp, 0);
1575                         }
1576
1577                 }
1578                 break;
1579         default:
1580                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1581                 LBUG();
1582         }
1583 out:
1584         DQINFO_COPY(&oqc->qc_dqinfo, info);
1585         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1586
1587         OBD_FREE_PTR(info);
1588         OBD_FREE_PTR(dqblk);
1589
1590         if (rc)
1591                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1592                                 "failed: %d\n",
1593                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1594         RETURN(rc);
1595 }
1596
1597 struct chk_dqblk{
1598         struct hlist_node       dqb_hash;        /* quotacheck hash */
1599         struct list_head        dqb_list;        /* in list also */
1600         qid_t                   dqb_id;          /* uid/gid */
1601         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1602         qsize_t                 dqb_bhardlimit;  /* block hard limit */
1603         qsize_t                 dqb_bsoftlimit;  /* block soft limit */
1604         qsize_t                 dqb_curspace;    /* current space */
1605         qsize_t                 dqb_ihardlimit;  /* inode hard limit */
1606         qsize_t                 dqb_isoftlimit;  /* inode soft limit */
1607         qsize_t                 dqb_curinodes;   /* current inodes */
1608         __u64                   dqb_btime;       /* block grace time */
1609         __u64                   dqb_itime;       /* inode grace time */
1610         __u32                   dqb_valid;       /* flag for above fields */
1611 };
1612
1613 static inline unsigned int chkquot_hash(qid_t id, int type)
1614                                         __attribute__((__const__));
1615
1616 static inline unsigned int chkquot_hash(qid_t id, int type)
1617 {
1618         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1619 }
1620
1621 static inline struct chk_dqblk *
1622 find_chkquot(struct hlist_head *head, qid_t id, int type)
1623 {
1624         struct hlist_node *node;
1625         struct chk_dqblk *cdqb;
1626
1627         hlist_for_each(node, head) {
1628                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1629                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1630                         return cdqb;
1631         }
1632
1633         return NULL;
1634 }
1635
1636 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1637 {
1638         struct chk_dqblk *cdqb;
1639
1640         OBD_ALLOC_PTR(cdqb);
1641         if (cdqb) {
1642                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1643                 INIT_LIST_HEAD(&cdqb->dqb_list);
1644                 cdqb->dqb_id = id;
1645                 cdqb->dqb_type = type;
1646         }
1647
1648         return cdqb;
1649 }
1650
1651 static struct chk_dqblk *
1652 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1653       qid_t id, int type, int first_check)
1654 {
1655         struct hlist_head *head = hash + chkquot_hash(id, type);
1656         struct if_dqblk dqb;
1657         struct chk_dqblk *cdqb;
1658         int rc;
1659
1660         cdqb = find_chkquot(head, id, type);
1661         if (cdqb)
1662                 return cdqb;
1663
1664         cdqb = alloc_chkquot(id, type);
1665         if (!cdqb)
1666                 return NULL;
1667
1668         if (!first_check) {
1669                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1670                 if (rc) {
1671                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1672                                id, type, rc);
1673                 } else {
1674                         DQBLK_COPY(cdqb, &dqb);
1675                         cdqb->dqb_curspace = 0;
1676                         cdqb->dqb_curinodes = 0;
1677                 }
1678         }
1679
1680         hlist_add_head(&cdqb->dqb_hash, head);
1681         list_add_tail(&cdqb->dqb_list, list);
1682
1683         return cdqb;
1684 }
1685
1686 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1687 {
1688         struct obd_quotactl *oqctl;
1689         int rc;
1690
1691         OBD_ALLOC_PTR(oqctl);
1692         if (!oqctl)
1693                 RETURN(-ENOMEM);
1694
1695         oqctl->qc_cmd = cmd;
1696         oqctl->qc_id = qfmt;
1697         oqctl->qc_type = type;
1698         rc = fsfilt_ext3_quotactl(sb, oqctl);
1699
1700         OBD_FREE_PTR(oqctl);
1701         return rc;
1702 }
1703
1704 static inline int read_old_dqinfo(struct super_block *sb, int type,
1705                                   struct if_dqinfo *dqinfo)
1706 {
1707         struct obd_quotactl *oqctl;
1708         int rc;
1709         ENTRY;
1710
1711         OBD_ALLOC_PTR(oqctl);
1712         if (!oqctl)
1713                 RETURN(-ENOMEM);
1714
1715         oqctl->qc_cmd = Q_GETINFO;
1716         oqctl->qc_type = type;
1717         rc = fsfilt_ext3_quotactl(sb, oqctl);
1718         if (!rc)
1719                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1720
1721         OBD_FREE_PTR(oqctl);
1722         RETURN(rc);
1723 }
1724
1725 #ifndef HAVE_EXT4_LDISKFS
1726 static inline struct ext3_group_desc *
1727 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1728 {
1729         unsigned long desc_block, desc;
1730         struct ext3_group_desc *gdp;
1731
1732         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1733         desc = group % EXT3_DESC_PER_BLOCK(sb);
1734         gdp = (struct ext3_group_desc *)
1735               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1736
1737         return gdp + desc;
1738 }
1739
1740 static inline struct buffer_head *
1741 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1742 {
1743         struct ext3_group_desc *desc;
1744         struct buffer_head *bh;
1745
1746         desc = get_group_desc(sb, group, NULL);
1747         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1748         return bh;
1749 }
1750
1751 static __u32 ext3_itable_unused_count(struct super_block *sb,
1752                                struct ext3_group_desc *bg) {
1753        return le16_to_cpu(bg->bg_itable_unused);
1754 }
1755 #else
1756 #define get_group_desc ext3_get_group_desc
1757 #endif
1758
1759 struct qchk_ctxt {
1760         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1761         struct list_head        qckt_list;                   /* quotacheck list */
1762         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1763         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1764 };
1765
1766 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1767                            struct obd_quotactl *oqc)
1768 {
1769         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1770         loff_t size = 0;
1771         qid_t qid[MAXQUOTAS];
1772         int cnt, i, rc = 0;
1773
1774         if (!inode)
1775                 return 0;
1776
1777         qid[USRQUOTA] = inode->i_uid;
1778         qid[GRPQUOTA] = inode->i_gid;
1779
1780         if (S_ISDIR(inode->i_mode) ||
1781             S_ISREG(inode->i_mode) ||
1782             S_ISLNK(inode->i_mode))
1783                 size = inode_get_bytes(inode);
1784
1785         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1786                 if (!Q_TYPESET(oqc, cnt))
1787                         continue;
1788
1789                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1790                                 &qctxt->qckt_list, qid[cnt], cnt,
1791                                 qctxt->qckt_first_check[cnt]);
1792                 if (!cdqb[cnt]) {
1793                         rc = -ENOMEM;
1794                         break;
1795                 }
1796
1797                 cdqb[cnt]->dqb_curspace += size;
1798                 cdqb[cnt]->dqb_curinodes++;
1799         }
1800
1801         if (rc) {
1802                 for (i = 0; i < cnt; i++) {
1803                         if (!Q_TYPESET(oqc, i))
1804                                 continue;
1805                         LASSERT(cdqb[i]);
1806                         cdqb[i]->dqb_curspace -= size;
1807                         cdqb[i]->dqb_curinodes--;
1808                 }
1809         }
1810
1811         return rc;
1812 }
1813
1814 static int v2_write_dqheader(struct file *f, int type)
1815 {
1816         static const __u32 quota_magics[] = V2_INITQMAGICS;
1817         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V1;
1818         struct v2_disk_dqheader dqhead;
1819         loff_t offset = 0;
1820
1821         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1822         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1823
1824         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1825         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1826
1827         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1828 }
1829
1830 /* write dqinfo struct in a new quota file */
1831 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1832 {
1833         struct v2_disk_dqinfo dqinfo;
1834         __u32 blocks = V2_DQTREEOFF + 1;
1835         loff_t offset = V2_DQINFOOFF;
1836
1837         if (info) {
1838                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1839                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1840                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1841                                                ~DQF_INFO_DIRTY);
1842         } else {
1843                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1844                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1845                 dqinfo.dqi_flags = 0;
1846         }
1847
1848         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1849         dqinfo.dqi_free_blk = 0;
1850         dqinfo.dqi_free_entry = 0;
1851
1852         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1853 }
1854
1855 #ifdef HAVE_QUOTA64
1856 static int v3_write_dqheader(struct file *f, int type)
1857 {
1858         static const __u32 quota_magics[] = V2_INITQMAGICS;
1859         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1860         struct v2_disk_dqheader dqhead;
1861         loff_t offset = 0;
1862
1863         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1864         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1865
1866         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1867         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1868
1869         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1870 }
1871
1872 /* write dqinfo struct in a new quota file */
1873 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1874 {
1875         return v2_write_dqinfo(f, type, info);
1876 }
1877 #endif
1878
1879 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1880                                   struct obd_quotactl *oqc)
1881 {
1882         int i, rc = 0;
1883         ENTRY;
1884
1885         for (i = 0; i < MAXQUOTAS; i++) {
1886                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1887                                          NULL : &qctxt->qckt_dqinfo[i];
1888                 struct file *file;
1889                 const char *name[][MAXQUOTAS] = LUSTRE_OPQFILES_NAMES;
1890                 int (*write_dqheader)(struct file *, int);
1891                 int (*write_dqinfo)(struct file *, int, struct if_dqinfo *);
1892
1893                 if (!Q_TYPESET(oqc, i))
1894                         continue;
1895
1896                 file = filp_open(name[oqc->qc_id][i],
1897                                  O_RDWR | O_CREAT | O_TRUNC, 0644);
1898                 if (IS_ERR(file)) {
1899                         rc = PTR_ERR(file);
1900                         CERROR("can't create %s file: rc = %d\n",
1901                                name[oqc->qc_id][i], rc);
1902                         GOTO(out, rc);
1903                 }
1904
1905                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1906                         CERROR("file %s is not regular", name[oqc->qc_id][i]);
1907                         filp_close(file, 0);
1908                         GOTO(out, rc = -EINVAL);
1909                 }
1910
1911                 DQUOT_DROP(file->f_dentry->d_inode);
1912
1913                 switch (oqc->qc_id) {
1914                 case LUSTRE_QUOTA_V1 : write_dqheader = v2_write_dqheader;
1915                                        write_dqinfo   = v2_write_dqinfo;
1916                                        break;
1917 #ifdef HAVE_QUOTA64
1918                 case LUSTRE_QUOTA_V2 : write_dqheader = v3_write_dqheader;
1919                                        write_dqinfo   = v3_write_dqinfo;
1920                                        break;
1921 #endif
1922                 default              : CERROR("unknown quota format!\n");
1923                                        LBUG();
1924                 }
1925
1926                 rc = (*write_dqheader)(file, i);
1927                 if (rc) {
1928                         filp_close(file, 0);
1929                         GOTO(out, rc);
1930                 }
1931
1932                 rc = (*write_dqinfo)(file, i, info);
1933                 filp_close(file, 0);
1934                 if (rc)
1935                         GOTO(out, rc);
1936         }
1937
1938 out:
1939         RETURN(rc);
1940 }
1941
1942
1943 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1944                           struct chk_dqblk *cdqb)
1945 {
1946         struct obd_quotactl *oqc;
1947         long now;
1948         int rc;
1949         ENTRY;
1950
1951         OBD_ALLOC_PTR(oqc);
1952         if (!oqc)
1953                 RETURN(-ENOMEM);
1954
1955         now = CURRENT_SECONDS;
1956
1957         if (cdqb->dqb_bsoftlimit &&
1958             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1959             !cdqb->dqb_btime)
1960                 cdqb->dqb_btime =
1961                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1962
1963         if (cdqb->dqb_isoftlimit &&
1964             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1965             !cdqb->dqb_itime)
1966                 cdqb->dqb_itime =
1967                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1968
1969         cdqb->dqb_valid = QIF_ALL;
1970
1971         oqc->qc_cmd = Q_SETQUOTA;
1972         oqc->qc_type = cdqb->dqb_type;
1973         oqc->qc_id = cdqb->dqb_id;
1974         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1975
1976         rc = fsfilt_ext3_quotactl(sb, oqc);
1977         OBD_FREE_PTR(oqc);
1978         RETURN(rc);
1979 }
1980
1981 static int prune_chkquots(struct super_block *sb,
1982                           struct qchk_ctxt *qctxt, int error)
1983 {
1984         struct chk_dqblk *cdqb, *tmp;
1985         int rc;
1986
1987         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1988                 if (!error) {
1989                         rc = commit_chkquot(sb, qctxt, cdqb);
1990                         if (rc)
1991                                 error = rc;
1992                 }
1993                 hlist_del_init(&cdqb->dqb_hash);
1994                 list_del(&cdqb->dqb_list);
1995                 OBD_FREE_PTR(cdqb);
1996         }
1997
1998         return error;
1999 }
2000
2001 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2002 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2003 #endif
2004
2005 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2006                                   struct obd_quotactl *oqc)
2007 {
2008         struct ext3_sb_info *sbi = EXT3_SB(sb);
2009         int i, group, uninit_feat = 0;
2010         struct qchk_ctxt *qctxt;
2011         struct buffer_head *bitmap_bh = NULL;
2012         unsigned long ino, inode_inuse;
2013         struct inode *inode;
2014         int rc = 0;
2015         ENTRY;
2016
2017         /* turn on quota and read dqinfo if existed */
2018         OBD_ALLOC_PTR(qctxt);
2019         if (!qctxt) {
2020                 oqc->qc_stat = -ENOMEM;
2021                 RETURN(-ENOMEM);
2022         }
2023
2024         for (i = 0; i < NR_DQHASH; i++)
2025                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2026         INIT_LIST_HEAD(&qctxt->qckt_list);
2027
2028         for (i = 0; i < MAXQUOTAS; i++) {
2029                 if (!Q_TYPESET(oqc, i))
2030                         continue;
2031
2032                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2033                 if (!rc || rc == -EBUSY) {
2034                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2035                         if (rc)
2036                                 GOTO(out, rc);
2037                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2038                         qctxt->qckt_first_check[i] = 1;
2039                 } else if (rc) {
2040                         GOTO(out, rc);
2041                 }
2042         }
2043         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2044                 /* This filesystem supports the uninit group feature */
2045                 uninit_feat = 1;
2046
2047         /* number of inodes that have been allocated */
2048         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2049                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2050
2051         /* check quota and update in hash */
2052         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2053              group++) {
2054                 unsigned long used_count = sbi->s_inodes_per_group;
2055
2056                 if (uninit_feat) {
2057                         struct ext3_group_desc *desc;
2058                         desc = get_group_desc(sb, group, NULL);
2059                         if (!desc)
2060                                 GOTO(out, -EIO);
2061
2062                         /* we don't really need to take the group lock here,
2063                          * but it may be useful if one day we support online
2064                          * quotacheck */
2065                         spin_lock(sb_bgl_lock(sbi, group));
2066                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2067                                 /* no inode in use in this group, just skip it */
2068                                 spin_unlock(sb_bgl_lock(sbi, group));
2069                                 continue;
2070                         }
2071                         used_count -= ext3_itable_unused_count(sb, desc);
2072                         spin_unlock(sb_bgl_lock(sbi, group));
2073                 }
2074
2075                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2076                 if (!bitmap_bh) {
2077                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2078                                sb->s_id, group);
2079                         GOTO(out, -EIO);
2080                 }
2081
2082                 i = 0;
2083                 while (i < used_count &&
2084                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2085                                                used_count, i)) < used_count) {
2086                         inode_inuse--;
2087                         i++;
2088                         ino = i + group * sbi->s_inodes_per_group;
2089                         if (ino < sbi->s_first_ino)
2090                                 continue;
2091 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2092                         inode = ext3_iget(sb, ino);
2093 #else
2094                         inode = iget(sb, ino);
2095 #endif
2096                         if (!inode || IS_ERR(inode))
2097                                 continue;
2098
2099                         rc = add_inode_quota(inode, qctxt, oqc);
2100                         iput(inode);
2101                         if (rc) {
2102                                 brelse(bitmap_bh);
2103                                 GOTO(out, rc);
2104                         }
2105                 }
2106                 brelse(bitmap_bh);
2107         }
2108
2109         /* read old quota limits from old quota file. (only for the user
2110          * has limits but hasn't file) */
2111 #ifdef HAVE_QUOTA_SUPPORT
2112         for (i = 0; i < MAXQUOTAS; i++) {
2113                 struct list_head id_list;
2114                 struct dquot_id *dqid, *tmp;
2115
2116                 if (!Q_TYPESET(oqc, i))
2117                         continue;
2118
2119                 if (qctxt->qckt_first_check[i])
2120                         continue;
2121
2122
2123                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2124                 INIT_LIST_HEAD(&id_list);
2125 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2126                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2127 #else
2128                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2129 #endif
2130                 if (rc)
2131                         CERROR("read old limits failed. (rc:%d)\n", rc);
2132
2133                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2134                         list_del_init(&dqid->di_link);
2135
2136                         if (!rc)
2137                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2138                                       dqid->di_id, i,
2139                                       qctxt->qckt_first_check[i]);
2140                         OBD_FREE_PTR(dqid);
2141                 }
2142         }
2143 #endif
2144         /* turn off quota cause we are to dump chk_dqblk to files */
2145         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2146
2147         rc = create_new_quota_files(qctxt, oqc);
2148         if (rc)
2149                 GOTO(out, rc);
2150
2151         /* we use vfs functions to set dqblk, so turn quota on */
2152         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2153 out:
2154         /* dump and free chk_dqblk */
2155         rc = prune_chkquots(sb, qctxt, rc);
2156         OBD_FREE_PTR(qctxt);
2157
2158         /* turn off quota, `lfs quotacheck` will turn on when all
2159          * nodes quotacheck finish. */
2160         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2161
2162         oqc->qc_stat = rc;
2163         if (rc)
2164                 CERROR("quotacheck failed: rc = %d\n", rc);
2165
2166         RETURN(rc);
2167 }
2168
2169 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2170                                  int cmd)
2171 {
2172         int rc = 0;
2173         ENTRY;
2174
2175         if (lqi->qi_files[type] == NULL) {
2176                 CERROR("operate qinfo before it's enabled!\n");
2177                 RETURN(-EIO);
2178         }
2179
2180         switch (cmd) {
2181         case QFILE_CHK:
2182                 rc = lustre_check_quota_file(lqi, type);
2183                 break;
2184         case QFILE_RD_INFO:
2185                 rc = lustre_read_quota_info(lqi, type);
2186                 break;
2187         case QFILE_WR_INFO:
2188                 rc = lustre_write_quota_info(lqi, type);
2189                 break;
2190         case QFILE_INIT_INFO:
2191                 rc = lustre_init_quota_info(lqi, type);
2192                 break;
2193         case QFILE_CONVERT:
2194                 rc = lustre_quota_convert(lqi, type);
2195                 break;
2196         default:
2197                 rc = -ENOTSUPP;
2198                 CERROR("Unsupported admin quota file cmd %d\n"
2199                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2200                        cmd);
2201                 break;
2202         }
2203         RETURN(rc);
2204 }
2205
2206 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2207                             struct list_head *list)
2208 {
2209         return lustre_get_qids(file, inode, type, list);
2210 }
2211
2212 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2213 {
2214         int rc = 0;
2215         ENTRY;
2216
2217         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2218                 CERROR("operate dquot before it's enabled!\n");
2219                 RETURN(-EIO);
2220         }
2221
2222         switch (cmd) {
2223         case QFILE_RD_DQUOT:
2224                 rc = lustre_read_dquot(dquot);
2225                 break;
2226         case QFILE_WR_DQUOT:
2227                 if (dquot->dq_dqb.dqb_ihardlimit ||
2228                     dquot->dq_dqb.dqb_isoftlimit ||
2229                     dquot->dq_dqb.dqb_bhardlimit ||
2230                     dquot->dq_dqb.dqb_bsoftlimit)
2231                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2232                 else
2233                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2234
2235                 rc = lustre_commit_dquot(dquot);
2236                 if (rc >= 0)
2237                         rc = 0;
2238                 break;
2239         default:
2240                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2241                 LBUG();
2242                 break;
2243         }
2244         RETURN(rc);
2245 }
2246
2247 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2248                                 struct inode *inode, int frags)
2249 {
2250 #ifdef EXT3_EXT_HAS_NO_TREE
2251         struct ext3_ext_base *base = inode;
2252 #else
2253         struct ext3_extents_tree tree;
2254         struct ext3_ext_base *base = &tree;
2255
2256         ext3_init_tree_desc(base, inode);
2257 #endif
2258         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2259          * metablocks at maxium b=16542 */
2260         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2261         return 0;
2262 }
2263
2264 #endif
2265
2266 static lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2267 {
2268         return (EXT3_SB(sb)->journal_bdev);
2269 }
2270
2271 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2272                           char *buf, int count, loff_t pos)
2273 {
2274         loff_t p = pos;
2275         int rc;
2276
2277         /* Support for both adm and op quota files must be provided */
2278         if (f) {
2279                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2280                 rc = rc < 0 ? rc : p - pos;
2281         } else {
2282                 struct super_block *sb = inode->i_sb;
2283                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2284         }
2285         return rc;
2286 }
2287
2288 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2289 {
2290         loff_t p = pos;
2291         int rc;
2292
2293         /* Only adm quota files are supported, op updates are handled by vfs */
2294         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2295         rc = rc < 0 ? rc : p - pos;
2296
2297         return rc;
2298 }
2299
2300 void *lustre_quota_journal_start(struct inode *inode, int delete)
2301 {
2302         handle_t *handle;
2303         unsigned block_count;
2304
2305         if (delete) {
2306                 /* each indirect block (+4) may become free, attaching to the
2307                  * header list of free blocks (+1); the data block (+1) may
2308                  * become a free block (+0) or a block with free dqentries (+0) */
2309                 block_count = (4 + 1) + 1;
2310                 handle = ext3_journal_start(inode,
2311                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2312         } else {
2313                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2314                  * freeblk reusage with a header update (+1); dqentry is either reused
2315                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2316                  * a new block allocation (+1) with a header update (+1)              */
2317                 block_count = (4 + 1) + 3;
2318                 handle = ext3_journal_start(inode,
2319                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2320
2321         }
2322
2323         return handle;
2324 }
2325
2326 void lustre_quota_journal_stop(void *handle)
2327 {
2328         ext3_journal_stop((handle_t *)handle);
2329 }
2330
2331 static struct fsfilt_operations fsfilt_ext3_ops = {
2332         .fs_type                = "ext3",
2333         .fs_owner               = THIS_MODULE,
2334         .fs_getlabel            = fsfilt_ext3_get_label,
2335         .fs_setlabel            = fsfilt_ext3_set_label,
2336         .fs_uuid                = fsfilt_ext3_uuid,
2337         .fs_start               = fsfilt_ext3_start,
2338         .fs_brw_start           = fsfilt_ext3_brw_start,
2339         .fs_extend              = fsfilt_ext3_extend,
2340         .fs_commit              = fsfilt_ext3_commit,
2341         .fs_commit_async        = fsfilt_ext3_commit_async,
2342         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2343         .fs_setattr             = fsfilt_ext3_setattr,
2344         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2345         .fs_set_md              = fsfilt_ext3_set_md,
2346         .fs_get_md              = fsfilt_ext3_get_md,
2347         .fs_readpage            = fsfilt_ext3_readpage,
2348         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2349         .fs_statfs              = fsfilt_ext3_statfs,
2350         .fs_sync                = fsfilt_ext3_sync,
2351         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2352         .fs_write_record        = fsfilt_ext3_write_record,
2353         .fs_read_record         = fsfilt_ext3_read_record,
2354         .fs_setup               = fsfilt_ext3_setup,
2355         .fs_send_bio            = fsfilt_ext3_send_bio,
2356         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2357 #ifdef HAVE_DISK_INODE_VERSION
2358         .fs_get_version         = fsfilt_ext3_get_version,
2359         .fs_set_version         = fsfilt_ext3_set_version,
2360 #endif
2361 #ifdef HAVE_QUOTA_SUPPORT
2362         .fs_quotactl            = fsfilt_ext3_quotactl,
2363         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2364         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2365         .fs_qids                = fsfilt_ext3_qids,
2366         .fs_dquot               = fsfilt_ext3_dquot,
2367         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2368 #endif
2369         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2370 };
2371
2372 static int __init fsfilt_ext3_init(void)
2373 {
2374         int rc;
2375
2376         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2377                                          sizeof(struct fsfilt_cb_data), 0, 0);
2378         if (!fcb_cache) {
2379                 CERROR("error allocating fsfilt journal callback cache\n");
2380                 GOTO(out, rc = -ENOMEM);
2381         }
2382
2383         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2384
2385         if (rc) {
2386                 int err = cfs_mem_cache_destroy(fcb_cache);
2387                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2388         }
2389 out:
2390         return rc;
2391 }
2392
2393 static void __exit fsfilt_ext3_exit(void)
2394 {
2395         int rc;
2396
2397         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2398         rc = cfs_mem_cache_destroy(fcb_cache);
2399         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2400 }
2401
2402 module_init(fsfilt_ext3_init);
2403 module_exit(fsfilt_ext3_exit);
2404
2405 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2406 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2407 MODULE_LICENSE("GPL");