lustre/lvfs/fsfilt_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/fsfilt_ext3.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/init.h>
  44 #include <linux/module.h>
  45 #include <linux/fs.h>
  46 #include <linux/slab.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/quotaops.h>
  49 #ifdef HAVE_LINUX_EXPORTFS_H
  50 #include <linux/exportfs.h>
  51 #endif
  52 #ifdef HAVE_EXT4_LDISKFS
  53 #include <ext4/ext4.h>
  54 #include <ext4/ext4_jbd2.h>
  55 #else
  56 #include <linux/jbd.h>
  57 #include <linux/ext3_fs.h>
  58 #include <linux/ext3_jbd.h>
  59 #endif
  60 #include <linux/version.h>
  61 #include <linux/bitops.h>
  62 #include <linux/quota.h>
  63 #ifdef HAVE_QUOTAIO_H
  64 # include <linux/quotaio_v2.h>
  65 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  66 # include <quota/quotaio_v2.h>
  67 # include <quota/quota_tree.h>
  68 # define V2_DQTREEOFF    QT_TREEOFF
  69 #elif defined(HAVE_FS_QUOTAIO_V1_H)
  70 # include <quotaio_v2.h>
  71 # include <quota_tree.h>
  72 # define V2_DQTREEOFF    QT_TREEOFF
  73 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  74 #endif
  75
  76 #ifdef QFMT_VFS_V1
  77 #define QFMT_LUSTRE QFMT_VFS_V1
  78 #else
  79 #define QFMT_LUSTRE QFMT_VFS_V0
  80 #endif
  81
  82 #if defined(HAVE_EXT3_XATTR_H)
  83 #include <ext3/xattr.h>
  84 #else
  85 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  86 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  87 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  88 #endif
  89
  90 #include <libcfs/libcfs.h>
  91 #include <lustre_fsfilt.h>
  92 #include <obd.h>
  93 #include <lustre_quota.h>
  94 #include <linux/lustre_compat25.h>
  95 #include <linux/lprocfs_status.h>
  96
  97 #ifdef HAVE_EXT4_LDISKFS
  98 #include <ext4/ext4_extents.h>
  99 #else
 100 #include <linux/ext3_extents.h>
 101 #endif
 102
 103 #include "lustre_quota_fmt.h"
 104
 105 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
 106 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 107 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 108 #else
 109 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 110 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 111 #endif
 112
 113 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 114 /* for kernels 2.6.18 and later */
 115 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 116 #else
 117 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 118 #endif
 119
 120 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 121 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 122                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 123 #else
 124 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 125                ext3_ext_insert_extent(handle, inode, path, newext)
 126 #endif
 127
 128 #ifdef EXT3_DISCARD_PREALLOCATIONS
 129 #define ext3_mb_discard_inode_preallocations(inode) \
 130                  ext3_discard_preallocations(inode)
 131 #endif
 132
 133
 134 static cfs_mem_cache_t *fcb_cache;
 135
 136 struct fsfilt_cb_data {
 137         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 138         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 139         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 140         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 141         void *cb_data;                  /* MDS/OST completion function data */
 142 };
 143
 144 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 145 #define EXT3_XATTR_INDEX_TRUSTED        4
 146 #endif
 147
 148 #ifdef HAVE_EXT4_LDISKFS
 149 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 150 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 151 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 152 #else
 153 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 154 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 155 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 156 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 157 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 158 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 159 #endif
 160
 161 #ifndef ext3_find_next_bit
 162 #define ext3_find_next_bit           ext2_find_next_bit
 163 #endif
 164
 165 #ifndef ext2_find_next_bit
 166 #ifdef __LITTLE_ENDIAN
 167 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 168 #else
 169 error "Need implementation of find_next_bit on big-endian systems"
 170 #endif  /* __LITTLE_ENDIAN */
 171 #endif  /* !ext2_find_next_le_bit */
 172
 173 static char *fsfilt_ext3_get_label(struct super_block *sb)
 174 {
 175         return EXT3_SB(sb)->s_es->s_volume_name;
 176 }
 177
 178 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 179 {
 180         /* see e.g. fsfilt_ext3_write_record() */
 181         journal_t *journal;
 182         handle_t *handle;
 183         int err;
 184
 185         journal = EXT3_SB(sb)->s_journal;
 186         handle = ext3_journal_start_sb(sb, 1);
 187         if (IS_ERR(handle)) {
 188                 CERROR("can't start transaction\n");
 189                 return(PTR_ERR(handle));
 190         }
 191
 192         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 193         if (err)
 194                 goto out;
 195
 196         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 197                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 198
 199         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 200
 201 out:
 202         ext3_journal_stop(handle);
 203
 204         return(err);
 205 }
 206
 207 static char *fsfilt_ext3_uuid(struct super_block *sb)
 208 {
 209         return EXT3_SB(sb)->s_es->s_uuid;
 210 }
 211
 212 #ifdef HAVE_DISK_INODE_VERSION
 213
 214 static __u64 get_i_version(struct inode *inode)
 215 {
 216 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 217         return inode->i_version;
 218 #else
 219         return EXT3_I(inode)->i_fs_version;
 220 #endif
 221 }
 222
 223 static void set_i_version(struct inode *inode, __u64 new_version)
 224 {
 225 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 226         inode->i_version = new_version;
 227 #else
 228         (EXT3_I(inode))->i_fs_version = new_version;
 229 #endif
 230 }
 231
 232 /*
 233  * Get the 64-bit version for an inode.
 234  */
 235 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 236 {
 237         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 238                get_i_version(inode), inode->i_ino);
 239         return get_i_version(inode);
 240 }
 241
 242 /*
 243  * Set the 64-bit version and return the old version.
 244  */
 245 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 246 {
 247         __u64 old_version = get_i_version(inode);
 248
 249         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 250                new_version, old_version, inode->i_ino);
 251         set_i_version(inode, new_version);
 252         /* version is set after all inode operations are finished, so we should
 253          * mark it dirty here */
 254         inode->i_sb->s_op->dirty_inode(inode);
 255         return old_version;
 256 }
 257
 258 #endif
 259
 260 /*
 261  * We don't currently need any additional blocks for rmdir and
 262  * unlink transactions because we are storing the OST oa_id inside
 263  * the inode (which we will be changing anyways as part of this
 264  * transaction).
 265  */
 266 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 267                                int logs)
 268 {
 269         /* For updates to the last received file */
 270         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 271         journal_t *journal;
 272         void *handle;
 273
 274         if (current->journal_info) {
 275                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 276                        current->journal_info);
 277                 goto journal_start;
 278         }
 279
 280         switch(op) {
 281         case FSFILT_OP_RMDIR:
 282         case FSFILT_OP_UNLINK:
 283                 /* delete one file + create/update logs for each stripe */
 284                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 285                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 286                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 287                 break;
 288         case FSFILT_OP_RENAME:
 289                 /* modify additional directory */
 290                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 291                 /* no break */
 292         case FSFILT_OP_SYMLINK:
 293                 /* additional block + block bitmap + GDT for long symlink */
 294                 nblocks += 3;
 295                 /* no break */
 296         case FSFILT_OP_CREATE: {
 297 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 298                 static int warned;
 299                 if (!warned) {
 300                         if (!test_opt(inode->i_sb, EXTENTS)) {
 301                                 warned = 1;
 302                         } else if (((EXT3_I(inode)->i_flags &
 303                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 304                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 305                                 CWARN("extent-mapped directory found with "
 306                                       "ext3-based ldiskfs - contact "
 307                                       "http://bugzilla.lustre.org/\n");
 308                                 warned = 1;
 309                         }
 310                 }
 311 #endif
 312                 /* no break */
 313         }
 314         case FSFILT_OP_MKDIR:
 315         case FSFILT_OP_MKNOD:
 316                 /* modify one inode + block bitmap + GDT */
 317                 nblocks += 3;
 318                 /* no break */
 319         case FSFILT_OP_LINK:
 320                 /* modify parent directory */
 321                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 322                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 323                 /* create/update logs for each stripe */
 324                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 325                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 326                 break;
 327         case FSFILT_OP_SETATTR:
 328                 /* Setattr on inode */
 329                 nblocks += 1;
 330                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 331                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 332                 /* quota chown log for each stripe */
 333                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 334                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 335                 break;
 336         case FSFILT_OP_CANCEL_UNLINK:
 337                 /* blocks for log header bitmap update OR
 338                  * blocks for catalog header bitmap update + unlink of logs */
 339                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 340                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 341                 break;
 342         default: CERROR("unknown transaction start op %d\n", op);
 343                 LBUG();
 344         }
 345
 346         LASSERT(current->journal_info == desc_private);
 347         journal = EXT3_SB(inode->i_sb)->s_journal;
 348         if (nblocks > journal->j_max_transaction_buffers) {
 349                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 350                        nblocks, op, logs, journal->j_max_transaction_buffers);
 351                 nblocks = journal->j_max_transaction_buffers;
 352         }
 353
 354  journal_start:
 355         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 356         handle = ext3_journal_start(inode, nblocks);
 357
 358         if (!IS_ERR(handle))
 359                 LASSERT(current->journal_info == handle);
 360         else
 361                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 362                        op, nblocks, PTR_ERR(handle));
 363         return handle;
 364 }
 365
 366 /*
 367  * Calculate the number of buffer credits needed to write multiple pages in
 368  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 369  * doesn't have a nice API for calculating this sort of thing in advance.
 370  *
 371  * See comment above ext3_writepage_trans_blocks for details.  We assume
 372  * no data journaling is being done, but it does allow for all of the pages
 373  * being non-contiguous.  If we are guaranteed contiguous pages we could
 374  * reduce the number of (d)indirect blocks a lot.
 375  *
 376  * With N blocks per page and P pages, for each inode we have at most:
 377  * N*P indirect
 378  * min(N*P, blocksize/4 + 1) dindirect blocks
 379  * niocount tindirect
 380  *
 381  * For the entire filesystem, we have at most:
 382  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 383  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 384  * objcount inode blocks
 385  * 1 superblock
 386  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 387  *
 388  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 389  */
 390 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 391                                       int niocount, struct niobuf_local *nb)
 392 {
 393         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 394         __u64 next_indir;
 395         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 396         int nbitmaps = 0, ngdblocks;
 397         int needed = objcount + 1; /* inodes + superblock */
 398         int i, j;
 399
 400         for (i = 0, j = 0; i < objcount; i++, fso++) {
 401                 /* two or more dindirect blocks in case we cross boundary */
 402                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 403                                     nb[j].offset) >>
 404                                    sb->s_blocksize_bits) /
 405                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 406                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 407
 408                 /* leaf, indirect, tindirect blocks for first block */
 409                 nbitmaps += blockpp + 2;
 410
 411                 j += fso->fso_bufcnt;
 412         }
 413
 414         next_indir = nb[0].offset +
 415                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 416         for (i = 1; i < niocount; i++) {
 417                 if (nb[i].offset >= next_indir) {
 418                         nbitmaps++;     /* additional indirect */
 419                         next_indir = nb[i].offset +
 420                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 421                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 422                         nbitmaps++;     /* additional indirect */
 423                 }
 424                 nbitmaps += blockpp;    /* each leaf in different group? */
 425         }
 426
 427         ngdblocks = nbitmaps;
 428         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 429                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 430         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 431                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 432
 433         needed += nbitmaps + ngdblocks;
 434
 435         /* last_rcvd update */
 436         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 437
 438 #if defined(CONFIG_QUOTA)
 439         /* We assume that there will be 1 bit set in s_dquot.flags for each
 440          * quota file that is active.  This is at least true for now.
 441          */
 442         needed += hweight32(ll_sb_any_quota_active(sb)) *
 443                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 444 #endif
 445
 446         return needed;
 447 }
 448
 449 /* We have to start a huge journal transaction here to hold all of the
 450  * metadata for the pages being written here.  This is necessitated by
 451  * the fact that we do lots of prepare_write operations before we do
 452  * any of the matching commit_write operations, so even if we split
 453  * up to use "smaller" transactions none of them could complete until
 454  * all of them were opened.  By having a single journal transaction,
 455  * we eliminate duplicate reservations for common blocks like the
 456  * superblock and group descriptors or bitmaps.
 457  *
 458  * We will start the transaction here, but each prepare_write will
 459  * add a refcount to the transaction, and each commit_write will
 460  * remove a refcount.  The transaction will be closed when all of
 461  * the pages have been written.
 462  */
 463 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 464                                    int niocount, struct niobuf_local *nb,
 465                                    void *desc_private, int logs)
 466 {
 467         journal_t *journal;
 468         handle_t *handle;
 469         int needed;
 470         ENTRY;
 471
 472         LASSERT(current->journal_info == desc_private);
 473         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 474         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 475
 476         /* The number of blocks we could _possibly_ dirty can very large.
 477          * We reduce our request if it is absurd (and we couldn't get that
 478          * many credits for a single handle anyways).
 479          *
 480          * At some point we have to limit the size of I/Os sent at one time,
 481          * increase the size of the journal, or we have to calculate the
 482          * actual journal requirements more carefully by checking all of
 483          * the blocks instead of being maximally pessimistic.  It remains to
 484          * be seen if this is a real problem or not.
 485          */
 486         if (needed > journal->j_max_transaction_buffers) {
 487                 CERROR("want too many journal credits (%d) using %d instead\n",
 488                        needed, journal->j_max_transaction_buffers);
 489                 needed = journal->j_max_transaction_buffers;
 490         }
 491
 492         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 493         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 494         if (IS_ERR(handle)) {
 495                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 496                        PTR_ERR(handle));
 497         } else {
 498                 LASSERT(handle->h_buffer_credits >= needed);
 499                 LASSERT(current->journal_info == handle);
 500         }
 501
 502         RETURN(handle);
 503 }
 504
 505 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 506 {
 507        handle_t *handle = h;
 508
 509        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 510        if (nblocks == 0) {
 511                handle->h_buffer_credits = 0;
 512                CWARN("setting credits of handle %p to zero by request\n", h);
 513        }
 514
 515        if (handle->h_buffer_credits > nblocks)
 516                 return 0;
 517        if (ext3_journal_extend(handle, nblocks) == 0)
 518                 return 0;
 519
 520        ext3_mark_inode_dirty(handle, inode);
 521        return ext3_journal_restart(handle, nblocks);
 522 }
 523
 524 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 525 {
 526         int rc;
 527         handle_t *handle = h;
 528
 529         LASSERT(current->journal_info == handle);
 530         if (force_sync)
 531                 handle->h_sync = 1; /* recovery likes this */
 532
 533         rc = ext3_journal_stop(handle);
 534
 535         return rc;
 536 }
 537
 538 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 539                                     void **wait_handle)
 540 {
 541         unsigned long tid;
 542         transaction_t *transaction;
 543         handle_t *handle = h;
 544         journal_t *journal;
 545         int rc;
 546
 547         LASSERT(current->journal_info == handle);
 548
 549         transaction = handle->h_transaction;
 550         journal = transaction->t_journal;
 551         tid = transaction->t_tid;
 552         /* we don't want to be blocked */
 553         handle->h_sync = 0;
 554         rc = ext3_journal_stop(handle);
 555         if (rc) {
 556                 CERROR("error while stopping transaction: %d\n", rc);
 557                 return rc;
 558         }
 559         fsfilt_log_start_commit(journal, tid);
 560
 561         *wait_handle = (void *) tid;
 562         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 563         return 0;
 564 }
 565
 566 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 567 {
 568         journal_t *journal = EXT3_JOURNAL(inode);
 569         tid_t tid = (tid_t)(long)h;
 570
 571         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 572         if (unlikely(is_journal_aborted(journal)))
 573                 return -EIO;
 574
 575         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 576
 577         if (unlikely(is_journal_aborted(journal)))
 578                 return -EIO;
 579         return 0;
 580 }
 581
 582 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 583                                struct iattr *iattr, int do_trunc)
 584 {
 585         struct inode *inode = dentry->d_inode;
 586         int rc = 0;
 587
 588         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 589          * We are already writing the inode to disk as part of this
 590          * transaction and want to avoid a lot of extra inode writeout
 591          * later on. b=9828 */
 592         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 593                 /* ATTR_SIZE would invoke truncate: clear it */
 594                 iattr->ia_valid &= ~ATTR_SIZE;
 595                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 596                 i_size_write(inode, iattr->ia_size);
 597
 598                 if (iattr->ia_valid & ATTR_UID)
 599                         inode->i_uid = iattr->ia_uid;
 600                 if (iattr->ia_valid & ATTR_GID)
 601                         inode->i_gid = iattr->ia_gid;
 602                 if (iattr->ia_valid & ATTR_ATIME)
 603                         inode->i_atime = iattr->ia_atime;
 604                 if (iattr->ia_valid & ATTR_MTIME)
 605                         inode->i_mtime = iattr->ia_mtime;
 606                 if (iattr->ia_valid & ATTR_CTIME)
 607                         inode->i_ctime = iattr->ia_ctime;
 608                 if (iattr->ia_valid & ATTR_MODE) {
 609                         inode->i_mode = iattr->ia_mode;
 610
 611                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 612                             !cfs_capable(CFS_CAP_FSETID))
 613                                 inode->i_mode &= ~S_ISGID;
 614                 }
 615
 616                 inode->i_sb->s_op->dirty_inode(inode);
 617
 618                 goto out;
 619         }
 620
 621         /* Don't allow setattr to change file type */
 622         if (iattr->ia_valid & ATTR_MODE)
 623                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 624                                  (iattr->ia_mode & ~S_IFMT);
 625
 626         /* We set these flags on the client, but have already checked perms
 627          * so don't confuse inode_change_ok. */
 628         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 629
 630         if (inode->i_op->setattr) {
 631                 rc = inode->i_op->setattr(dentry, iattr);
 632         } else {
 633                 rc = inode_change_ok(inode, iattr);
 634                 if (!rc)
 635                         rc = inode_setattr(inode, iattr);
 636         }
 637
 638  out:
 639         RETURN(rc);
 640 }
 641
 642 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 643                                  unsigned int cmd, unsigned long arg)
 644 {
 645         int rc = 0;
 646         ENTRY;
 647
 648         /* FIXME: Can't do this because of nested transaction deadlock */
 649         if (cmd == EXT3_IOC_SETFLAGS) {
 650                 /* We can't enable data journaling on OST objects, because
 651                 * this forces the transaction to be closed in order to
 652                 * flush the journal, but the caller will already have a
 653                 * compound transaction open to update the last_rcvd file,
 654                 * and this thread would deadlock trying to set the flag. */
 655                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 656                         CERROR("can't set data journal flag on file\n");
 657                         RETURN(-EPERM);
 658                 }
 659                 /* Because the MDS does not see the EXTENTS_FL set on the
 660                  * OST objects, mask this flag into all set flags.  It is
 661                  * not legal to clear this flag in any case, so we are not
 662                  * changing the functionality by doing this.  b=22911 */
 663                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 664         }
 665
 666 #ifdef HAVE_EXT4_LDISKFS
 667         /* ext4_ioctl does not have a inode argument */
 668         if (inode->i_fop->unlocked_ioctl)
 669                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 670 #else
 671         if (inode->i_fop->ioctl)
 672                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 673 #endif
 674         else
 675                 RETURN(-ENOTTY);
 676
 677         RETURN(rc);
 678 }
 679
 680 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 681                               void *lmm, int lmm_size, const char *name)
 682 {
 683         int rc;
 684
 685         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 686
 687         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 688                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 689
 690
 691         if (rc && rc != -EROFS)
 692                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 693                        inode->i_ino, rc);
 694         return rc;
 695 }
 696
 697 /* Must be called with i_mutex held */
 698 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 699                               const char *name)
 700 {
 701         int rc;
 702
 703         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 704
 705         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 706                             name, lmm, lmm_size);
 707
 708         /* This gives us the MD size */
 709         if (lmm == NULL)
 710                 return (rc == -ENODATA) ? 0 : rc;
 711
 712         if (rc < 0) {
 713                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 714                        EXT3_XATTR_INDEX_TRUSTED, name,
 715                        inode->i_ino, rc);
 716                 memset(lmm, 0, lmm_size);
 717                 return (rc == -ENODATA) ? 0 : rc;
 718         }
 719
 720         return rc;
 721 }
 722
 723 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 724 {
 725         submit_bio(rw, bio);
 726         return 0;
 727 }
 728
 729 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 730                                     loff_t *off)
 731 {
 732         struct inode *inode = file->f_dentry->d_inode;
 733         int rc = 0;
 734
 735         if (S_ISREG(inode->i_mode))
 736                 rc = file->f_op->read(file, buf, count, off);
 737         else {
 738                 const int blkbits = inode->i_sb->s_blocksize_bits;
 739                 const int blksize = inode->i_sb->s_blocksize;
 740
 741                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 742                        (unsigned long)count, inode->i_ino, *off);
 743                 while (count > 0) {
 744                         struct buffer_head *bh;
 745
 746                         bh = NULL;
 747                         if (*off < i_size_read(inode)) {
 748                                 int err = 0;
 749
 750                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 751                                                 0, &err);
 752
 753                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 754
 755                                 if (bh) {
 756                                         memcpy(buf, bh->b_data, blksize);
 757                                         brelse(bh);
 758                                 } else if (err) {
 759                                         /* XXX in theory we should just fake
 760                                          * this buffer and continue like ext3,
 761                                          * especially if this is a partial read
 762                                          */
 763                                         CERROR("error read dir %lu+%llu: %d\n",
 764                                                inode->i_ino, *off, err);
 765                                         RETURN(err);
 766                                 }
 767                         }
 768                         if (!bh) {
 769                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 770
 771                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 772                                 memset(fake, 0, sizeof(*fake));
 773                                 fake->rec_len = cpu_to_le16(blksize);
 774                         }
 775                         count -= blksize;
 776                         buf += blksize;
 777                         *off += blksize;
 778                         rc += blksize;
 779                 }
 780         }
 781
 782         return rc;
 783 }
 784
 785 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 786 {
 787         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 788
 789         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 790
 791         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 792 }
 793
 794 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 795                                       void *handle, fsfilt_cb_t cb_func,
 796                                       void *cb_data)
 797 {
 798         struct fsfilt_cb_data *fcb;
 799
 800         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 801         if (fcb == NULL)
 802                 RETURN(-ENOMEM);
 803
 804         fcb->cb_func = cb_func;
 805         fcb->cb_obd = obd;
 806         fcb->cb_last_rcvd = last_rcvd;
 807         fcb->cb_data = cb_data;
 808
 809         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 810         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 811                                     (struct journal_callback *)fcb);
 812
 813         return 0;
 814 }
 815
 816 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 817 {
 818         struct kstatfs sfs;
 819         int rc;
 820
 821         memset(&sfs, 0, sizeof(sfs));
 822         rc = ll_do_statfs(sb, &sfs);
 823         statfs_pack(osfs, &sfs);
 824         return rc;
 825 }
 826
 827 static int fsfilt_ext3_sync(struct super_block *sb)
 828 {
 829         return ext3_force_commit(sb);
 830 }
 831
 832 #ifndef EXT3_EXTENTS_FL
 833 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 834 #endif
 835
 836 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 837 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 838 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 839 #else
 840 # ifdef HAVE_EXT4_LDISKFS
 841 #  ifdef WALK_SPACE_HAS_DATA_SEM /* We only use it in fsfilt_map_nblocks() for now */
 842 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 843 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 844 #  else
 845 #   define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem))
 846 #   define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem))
 847 #  endif
 848 # else
 849 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 850 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 851 # endif
 852 #endif
 853
 854 #ifndef EXT_ASSERT
 855 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 856 #endif
 857
 858 #ifdef EXT3_EXT_HAS_NO_TREE
 859 /* for kernels 2.6.18 and later */
 860 #ifdef HAVE_EXT4_LDISKFS
 861 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 862 #else
 863 #define EXT_GENERATION(inode)           ext_generation(inode)
 864 #endif
 865 #define ext3_ext_base                   inode
 866 #define ext3_ext_base2inode(inode)      (inode)
 867 #define EXT_DEPTH(inode)                ext_depth(inode)
 868 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 869                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 870 #else
 871 #define ext3_ext_base                   ext3_extents_tree
 872 #define ext3_ext_base2inode(tree)       (tree->inode)
 873 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 874                         ext3_ext_walk_space(tree, block, num, cb);
 875 #endif
 876
 877 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 878 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 879                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 880 #else
 881 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 882                ext3_ext_insert_extent(handle, inode, path, newext)
 883 #endif
 884
 885 #include <linux/lustre_version.h>
 886
 887 struct bpointers {
 888         unsigned long *blocks;
 889         int *created;
 890         unsigned long start;
 891         int num;
 892         int init_num;
 893         int create;
 894 };
 895
 896 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 897                                unsigned long block, int *aflags)
 898 {
 899         struct ext3_inode_info *ei = EXT3_I(inode);
 900         unsigned long bg_start;
 901         unsigned long colour;
 902         int depth;
 903
 904         if (path) {
 905                 struct ext3_extent *ex;
 906                 depth = path->p_depth;
 907
 908                 /* try to predict block placement */
 909                 if ((ex = path[depth].p_ext))
 910                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 911
 912                 /* it looks index is empty
 913                  * try to find starting from index itself */
 914                 if (path[depth].p_bh)
 915                         return path[depth].p_bh->b_blocknr;
 916         }
 917
 918         /* OK. use inode's group */
 919         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 920                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 921         colour = (current->pid % 16) *
 922                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 923         return bg_start + colour + block;
 924 }
 925
 926 #define ll_unmap_underlying_metadata(sb, blocknr) \
 927         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 928
 929 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 930 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 931                                 struct ext3_ext_path *path, unsigned long block,
 932                                 unsigned long *count, int *err)
 933 {
 934         unsigned long pblock, goal;
 935         int aflags = 0;
 936         struct inode *inode = ext3_ext_base2inode(base);
 937
 938         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 939         aflags |= 2; /* block have been already reserved */
 940         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 941         return pblock;
 942
 943 }
 944 #else
 945 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 946                                 struct ext3_ext_path *path, unsigned long block,
 947                                 unsigned long *count, int *err)
 948 {
 949         struct inode *inode = ext3_ext_base2inode(base);
 950         struct ext3_allocation_request ar;
 951         unsigned long pblock;
 952         int aflags;
 953
 954         /* find neighbour allocated blocks */
 955         ar.lleft = block;
 956         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 957         if (*err)
 958                 return 0;
 959         ar.lright = block;
 960         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 961         if (*err)
 962                 return 0;
 963
 964         /* allocate new block */
 965         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 966         ar.inode = inode;
 967         ar.logical = block;
 968         ar.len = *count;
 969         ar.flags = EXT3_MB_HINT_DATA;
 970         pblock = ext3_mb_new_blocks(handle, &ar, err);
 971         *count = ar.len;
 972         return pblock;
 973 }
 974 #endif
 975
 976 #ifdef EXT3_EXT_HAS_NO_TREE
 977 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 978                                   struct ext3_ext_path *path,
 979                                   struct ext3_ext_cache *cex,
 980 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 981                                    struct ext3_extent *ex,
 982 #endif
 983                                   void *cbdata)
 984 {
 985         struct bpointers *bp = cbdata;
 986 #else
 987 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 988                                   struct ext3_ext_path *path,
 989                                   struct ext3_ext_cache *cex
 990 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 991                                   , struct ext3_extent *ex
 992 #endif
 993                                  )
 994 {
 995         struct bpointers *bp = base->private;
 996 #endif
 997         struct inode *inode = ext3_ext_base2inode(base);
 998         struct ext3_extent nex;
 999         unsigned long pblock;
1000         unsigned long tgen;
1001         int err, i;
1002         unsigned long count;
1003         handle_t *handle;
1004
1005         i = EXT_DEPTH(base);
1006         EXT_ASSERT(i == path->p_depth);
1007         EXT_ASSERT(path[i].p_hdr);
1008
1009         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1010                 err = EXT_CONTINUE;
1011                 goto map;
1012         }
1013
1014         if (bp->create == 0) {
1015                 i = 0;
1016                 if (cex->ec_block < bp->start)
1017                         i = bp->start - cex->ec_block;
1018                 if (i >= cex->ec_len)
1019                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1020                                         i, cex->ec_len);
1021                 for (; i < cex->ec_len && bp->num; i++) {
1022                         *(bp->created) = 0;
1023                         bp->created++;
1024                         *(bp->blocks) = 0;
1025                         bp->blocks++;
1026                         bp->num--;
1027                         bp->start++;
1028                 }
1029
1030                 return EXT_CONTINUE;
1031         }
1032
1033         tgen = EXT_GENERATION(base);
1034         count = ext3_ext_calc_credits_for_insert(base, path);
1035         fsfilt_up_truncate_sem(inode);
1036
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,
1151                                          ext3_ext_new_extent_cb, &bp);
1152         ext3_ext_invalidate_cache(base);
1153         fsfilt_up_truncate_sem(inode);
1154
1155         return err;
1156 }
1157
1158 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1159                                     int pages, unsigned long *blocks,
1160                                     int *created, int create)
1161 {
1162         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1163         int rc = 0, i = 0;
1164         struct page *fp = NULL;
1165         int clen = 0;
1166
1167         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1168                 inode->i_ino, pages, (*page)->index);
1169
1170         /* pages are sorted already. so, we just have to find
1171          * contig. space and process them properly */
1172         while (i < pages) {
1173                 if (fp == NULL) {
1174                         /* start new extent */
1175                         fp = *page++;
1176                         clen = 1;
1177                         i++;
1178                         continue;
1179                 } else if (fp->index + clen == (*page)->index) {
1180                         /* continue the extent */
1181                         page++;
1182                         clen++;
1183                         i++;
1184                         continue;
1185                 }
1186
1187                 /* process found extent */
1188                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1189                                         clen * blocks_per_page, blocks,
1190                                         created, create);
1191                 if (rc)
1192                         GOTO(cleanup, rc);
1193
1194                 /* look for next extent */
1195                 fp = NULL;
1196                 blocks += blocks_per_page * clen;
1197                 created += blocks_per_page * clen;
1198         }
1199
1200         if (fp)
1201                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1202                                         clen * blocks_per_page, blocks,
1203                                         created, create);
1204 cleanup:
1205         return rc;
1206 }
1207
1208 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1209                                unsigned long *blocks, int *created, int create);
1210 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1211                                    int pages, unsigned long *blocks,
1212                                    int *created, int create)
1213 {
1214         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1215         unsigned long *b;
1216         int rc = 0, i, *cr;
1217
1218         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1219                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1220                 if (rc) {
1221                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1222                                inode->i_ino, *b, *cr, create, rc);
1223                         break;
1224                 }
1225
1226                 b += blocks_per_page;
1227                 cr += blocks_per_page;
1228         }
1229         return rc;
1230 }
1231
1232 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1233                                 int pages, unsigned long *blocks,
1234                                 int *created, int create,
1235                                 cfs_semaphore_t *optional_sem)
1236 {
1237         int rc;
1238
1239         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1240                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1241                                                      blocks, created, create);
1242                 return rc;
1243         }
1244         if (optional_sem != NULL)
1245                 cfs_down(optional_sem);
1246         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1247                                             created, create);
1248         if (optional_sem != NULL)
1249                 cfs_up(optional_sem);
1250
1251         return rc;
1252 }
1253
1254 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1255 {
1256         unsigned long block;
1257         struct buffer_head *bh;
1258         int err, blocksize, csize, boffs, osize = size;
1259
1260         /* prevent reading after eof */
1261         cfs_lock_kernel();
1262         if (i_size_read(inode) < *offs + size) {
1263                 size = i_size_read(inode) - *offs;
1264                 cfs_unlock_kernel();
1265                 if (size < 0) {
1266                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1267                                i_size_read(inode), *offs);
1268                         return -EBADR;
1269                 } else if (size == 0) {
1270                         return 0;
1271                 }
1272         } else {
1273                 cfs_unlock_kernel();
1274         }
1275
1276         blocksize = 1 << inode->i_blkbits;
1277
1278         while (size > 0) {
1279                 block = *offs >> inode->i_blkbits;
1280                 boffs = *offs & (blocksize - 1);
1281                 csize = min(blocksize - boffs, size);
1282                 bh = ext3_bread(NULL, inode, block, 0, &err);
1283                 if (!bh) {
1284                         CERROR("can't read block: %d\n", err);
1285                         return err;
1286                 }
1287
1288                 memcpy(buf, bh->b_data + boffs, csize);
1289                 brelse(bh);
1290
1291                 *offs += csize;
1292                 buf += csize;
1293                 size -= csize;
1294         }
1295         return osize;
1296 }
1297 EXPORT_SYMBOL(fsfilt_ext3_read);
1298
1299 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1300                                    int size, loff_t *offs)
1301 {
1302         int rc;
1303         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1304         if (rc > 0)
1305                 rc = 0;
1306         return rc;
1307 }
1308
1309 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1310                                 loff_t *offs, handle_t *handle)
1311 {
1312         struct buffer_head *bh = NULL;
1313         loff_t old_size = i_size_read(inode), offset = *offs;
1314         loff_t new_size = i_size_read(inode);
1315         unsigned long block;
1316         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1317
1318         while (bufsize > 0) {
1319                 if (bh != NULL)
1320                         brelse(bh);
1321
1322                 block = offset >> inode->i_blkbits;
1323                 boffs = offset & (blocksize - 1);
1324                 size = min(blocksize - boffs, bufsize);
1325                 bh = ext3_bread(handle, inode, block, 1, &err);
1326                 if (!bh) {
1327                         CERROR("can't read/create block: %d\n", err);
1328                         break;
1329                 }
1330
1331                 err = ext3_journal_get_write_access(handle, bh);
1332                 if (err) {
1333                         CERROR("journal_get_write_access() returned error %d\n",
1334                                err);
1335                         break;
1336                 }
1337                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1338                 memcpy(bh->b_data + boffs, buf, size);
1339                 err = ext3_journal_dirty_metadata(handle, bh);
1340                 if (err) {
1341                         CERROR("journal_dirty_metadata() returned error %d\n",
1342                                err);
1343                         break;
1344                 }
1345                 if (offset + size > new_size)
1346                         new_size = offset + size;
1347                 offset += size;
1348                 bufsize -= size;
1349                 buf += size;
1350         }
1351         if (bh)
1352                 brelse(bh);
1353
1354         /* correct in-core and on-disk sizes */
1355         if (new_size > i_size_read(inode)) {
1356                 cfs_lock_kernel();
1357                 if (new_size > i_size_read(inode))
1358                         i_size_write(inode, new_size);
1359                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1360                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1361                 if (i_size_read(inode) > old_size)
1362                         mark_inode_dirty(inode);
1363                 cfs_unlock_kernel();
1364         }
1365
1366         if (err == 0)
1367                 *offs = offset;
1368         return err;
1369 }
1370 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1371
1372 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1373                                     loff_t *offs, int force_sync)
1374 {
1375         struct inode *inode = file->f_dentry->d_inode;
1376         handle_t *handle;
1377         int err, block_count = 0, blocksize;
1378
1379         /* Determine how many transaction credits are needed */
1380         blocksize = 1 << inode->i_blkbits;
1381         block_count = (*offs & (blocksize - 1)) + bufsize;
1382         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1383
1384         handle = ext3_journal_start(inode,
1385                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1386         if (IS_ERR(handle)) {
1387                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1388                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1389                 return PTR_ERR(handle);
1390         }
1391
1392         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1393
1394         if (!err && force_sync)
1395                 handle->h_sync = 1; /* recovery likes this */
1396
1397         ext3_journal_stop(handle);
1398
1399         return err;
1400 }
1401
1402 static int fsfilt_ext3_setup(struct super_block *sb)
1403 {
1404 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1405      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1406         struct ext3_sb_info *sbi = EXT3_SB(sb);
1407 #if 0
1408         sbi->dx_lock = fsfilt_ext3_dx_lock;
1409         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1410 #endif
1411 #endif
1412 #ifdef S_PDIROPS
1413         CWARN("Enabling PDIROPS\n");
1414         set_opt(sbi->s_mount_opt, PDIROPS);
1415         sb->s_flags |= S_PDIROPS;
1416 #endif
1417         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1418                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1419 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1420         /* enable journaled quota support */
1421         /* kfreed in ext3_put_super() */
1422         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1423         if (!sbi->s_qf_names[USRQUOTA])
1424                 return -ENOMEM;
1425         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1426         if (!sbi->s_qf_names[GRPQUOTA]) {
1427                 kfree(sbi->s_qf_names[USRQUOTA]);
1428                 sbi->s_qf_names[USRQUOTA] = NULL;
1429                 return -ENOMEM;
1430         }
1431         sbi->s_jquota_fmt = QFMT_LUSTRE;
1432 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1433         set_opt(sbi->s_mount_opt, QUOTA);
1434 #endif
1435 #endif
1436         return 0;
1437 }
1438
1439 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1440    objects. Logs is number of logfiles to update */
1441 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1442 {
1443         if ( !fso ) {
1444                 switch(op) {
1445                 case FSFILT_OP_CREATE:
1446                                  /* directory leaf, index & indirect & EA*/
1447                         return 4 + 3 * logs;
1448                 case FSFILT_OP_UNLINK:
1449                         return 3 * logs;
1450                 }
1451         } else {
1452                 int i;
1453                 int needed = 0;
1454                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1455                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1456                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1457                 for (i = 0; i < op; i++, fso++) {
1458                         int nblocks = fso->fso_bufcnt * blockpp;
1459                         int ndindirect = min(nblocks, addrpp + 1);
1460                         int nindir = nblocks + ndindirect + 1;
1461
1462                         needed += nindir;
1463                 }
1464                 return needed + 3 * logs;
1465         }
1466
1467         return 0;
1468 }
1469
1470 #ifdef HAVE_QUOTA_SUPPORT
1471 #define DQINFO_COPY(out, in)                    \
1472 do {                                            \
1473         Q_COPY(out, in, dqi_bgrace);            \
1474         Q_COPY(out, in, dqi_igrace);            \
1475         Q_COPY(out, in, dqi_flags);             \
1476         Q_COPY(out, in, dqi_valid);             \
1477 } while (0)
1478
1479 #define DQBLK_COPY(out, in)                     \
1480 do {                                            \
1481         Q_COPY(out, in, dqb_bhardlimit);        \
1482         Q_COPY(out, in, dqb_bsoftlimit);        \
1483         Q_COPY(out, in, dqb_curspace);          \
1484         Q_COPY(out, in, dqb_ihardlimit);        \
1485         Q_COPY(out, in, dqb_isoftlimit);        \
1486         Q_COPY(out, in, dqb_curinodes);         \
1487         Q_COPY(out, in, dqb_btime);             \
1488         Q_COPY(out, in, dqb_itime);             \
1489         Q_COPY(out, in, dqb_valid);             \
1490 } while (0)
1491
1492 static int fsfilt_ext3_quotactl(struct super_block *sb,
1493                                 struct obd_quotactl *oqc)
1494 {
1495         int i, rc = 0, error = 0;
1496         const struct quotactl_ops *qcop;
1497         struct if_dqinfo *info;
1498         struct if_dqblk *dqblk;
1499         ENTRY;
1500
1501         if (!sb->s_qcop)
1502                 RETURN(-ENOSYS);
1503
1504         OBD_ALLOC_PTR(info);
1505         if (!info)
1506                 RETURN(-ENOMEM);
1507         OBD_ALLOC_PTR(dqblk);
1508         if (!dqblk) {
1509                 OBD_FREE_PTR(info);
1510                 RETURN(-ENOMEM);
1511         }
1512
1513         DQINFO_COPY(info, &oqc->qc_dqinfo);
1514         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1515
1516         qcop = sb->s_qcop;
1517         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1518                 for (i = 0; i < MAXQUOTAS; i++) {
1519                         if (!Q_TYPESET(oqc, i))
1520                                 continue;
1521
1522                         if (oqc->qc_cmd == Q_QUOTAON) {
1523                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1524
1525                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1526
1527                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1528                                                  name[i], 0);
1529                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1530                                 rc = ll_quota_off(sb, i, 0);
1531                         }
1532
1533                         if (rc == -EBUSY)
1534                                 error = rc;
1535                         else if (rc)
1536                                 GOTO(out, rc);
1537                 }
1538                 GOTO(out, rc ?: error);
1539         }
1540
1541         switch (oqc->qc_cmd) {
1542         case Q_GETOINFO:
1543         case Q_GETINFO:
1544                 if (!qcop->get_info)
1545                         GOTO(out, rc = -ENOSYS);
1546                 rc = qcop->get_info(sb, oqc->qc_type, info);
1547                 break;
1548         case Q_SETQUOTA:
1549         case Q_INITQUOTA:
1550                 if (!qcop->set_dqblk)
1551                         GOTO(out, rc = -ENOSYS);
1552                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1553                 break;
1554         case Q_GETOQUOTA:
1555         case Q_GETQUOTA:
1556                 if (!qcop->get_dqblk)
1557                         GOTO(out, rc = -ENOSYS);
1558                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1559                 if (!rc)
1560                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1561                 break;
1562         case Q_SYNC:
1563                 if (!sb->s_qcop->quota_sync)
1564                         GOTO(out, rc = -ENOSYS);
1565                 qcop->quota_sync(sb, oqc->qc_type);
1566                 break;
1567         case Q_FINVALIDATE:
1568                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1569                 for (i = 0; i < MAXQUOTAS; i++) {
1570                         struct file *fp;
1571                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1572
1573                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1574
1575                         if (!Q_TYPESET(oqc, i))
1576                                 continue;
1577
1578                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1579                         if (IS_ERR(fp)) {
1580                                 rc = PTR_ERR(fp);
1581                                 CERROR("error invalidating operational quota file"
1582                                        " %s (rc:%d)\n", name[i], rc);
1583                         } else {
1584                                 filp_close(fp, 0);
1585                         }
1586
1587                 }
1588                 break;
1589         default:
1590                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1591                 LBUG();
1592         }
1593 out:
1594         DQINFO_COPY(&oqc->qc_dqinfo, info);
1595         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1596
1597         OBD_FREE_PTR(info);
1598         OBD_FREE_PTR(dqblk);
1599
1600         if (rc)
1601                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1602                                 "failed: %d\n",
1603                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1604         RETURN(rc);
1605 }
1606
1607 struct chk_dqblk{
1608         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1609         cfs_list_t              dqb_list;        /** in list also */
1610         qid_t                   dqb_id;          /** uid/gid */
1611         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1612         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1613         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1614         qsize_t                 dqb_curspace;    /** current space */
1615         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1616         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1617         qsize_t                 dqb_curinodes;   /** current inodes */
1618         obd_time                dqb_btime;       /** block grace time */
1619         obd_time                dqb_itime;       /** inode grace time */
1620         __u32                   dqb_valid;       /** flag for above fields */
1621 };
1622
1623 static inline unsigned int chkquot_hash(qid_t id, int type)
1624                                         __attribute__((__const__));
1625
1626 static inline unsigned int chkquot_hash(qid_t id, int type)
1627 {
1628         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1629 }
1630
1631 static inline struct chk_dqblk *
1632 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1633 {
1634         cfs_hlist_node_t *node;
1635         struct chk_dqblk *cdqb;
1636
1637         cfs_hlist_for_each(node, head) {
1638                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1639                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1640                         return cdqb;
1641         }
1642
1643         return NULL;
1644 }
1645
1646 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1647 {
1648         struct chk_dqblk *cdqb;
1649
1650         OBD_ALLOC_PTR(cdqb);
1651         if (cdqb) {
1652                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1653                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1654                 cdqb->dqb_id = id;
1655                 cdqb->dqb_type = type;
1656         }
1657
1658         return cdqb;
1659 }
1660
1661 static struct chk_dqblk *
1662 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1663       cfs_list_t *list, qid_t id, int type, int first_check)
1664 {
1665         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1666         struct if_dqblk dqb;
1667         struct chk_dqblk *cdqb;
1668         int rc;
1669
1670         cdqb = find_chkquot(head, id, type);
1671         if (cdqb)
1672                 return cdqb;
1673
1674         cdqb = alloc_chkquot(id, type);
1675         if (!cdqb)
1676                 return NULL;
1677
1678         if (!first_check) {
1679                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1680                 if (rc) {
1681                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1682                                id, type, rc);
1683                 } else {
1684                         DQBLK_COPY(cdqb, &dqb);
1685                         cdqb->dqb_curspace = 0;
1686                         cdqb->dqb_curinodes = 0;
1687                 }
1688         }
1689
1690         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1691         cfs_list_add_tail(&cdqb->dqb_list, list);
1692
1693         return cdqb;
1694 }
1695
1696 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1697 {
1698         struct obd_quotactl *oqctl;
1699         int rc;
1700
1701         OBD_ALLOC_PTR(oqctl);
1702         if (!oqctl)
1703                 RETURN(-ENOMEM);
1704
1705         oqctl->qc_cmd = cmd;
1706         oqctl->qc_id = qfmt;
1707         oqctl->qc_type = type;
1708         rc = fsfilt_ext3_quotactl(sb, oqctl);
1709
1710         OBD_FREE_PTR(oqctl);
1711         return rc;
1712 }
1713
1714 static inline int read_old_dqinfo(struct super_block *sb, int type,
1715                                   struct if_dqinfo *dqinfo)
1716 {
1717         struct obd_quotactl *oqctl;
1718         int rc;
1719         ENTRY;
1720
1721         OBD_ALLOC_PTR(oqctl);
1722         if (!oqctl)
1723                 RETURN(-ENOMEM);
1724
1725         oqctl->qc_cmd = Q_GETINFO;
1726         oqctl->qc_type = type;
1727         rc = fsfilt_ext3_quotactl(sb, oqctl);
1728         if (!rc)
1729                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1730
1731         OBD_FREE_PTR(oqctl);
1732         RETURN(rc);
1733 }
1734
1735 #ifndef HAVE_EXT4_LDISKFS
1736 static inline struct ext3_group_desc *
1737 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1738 {
1739         unsigned long desc_block, desc;
1740         struct ext3_group_desc *gdp;
1741
1742         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1743         desc = group % EXT3_DESC_PER_BLOCK(sb);
1744         gdp = (struct ext3_group_desc *)
1745               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1746
1747         return gdp + desc;
1748 }
1749
1750 static inline struct buffer_head *
1751 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1752 {
1753         struct ext3_group_desc *desc;
1754         struct buffer_head *bh;
1755
1756         desc = get_group_desc(sb, group, NULL);
1757         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1758         return bh;
1759 }
1760
1761 static __u32 ext3_itable_unused_count(struct super_block *sb,
1762                                struct ext3_group_desc *bg) {
1763        return le16_to_cpu(bg->bg_itable_unused);
1764 }
1765 #else
1766 #define get_group_desc ext3_get_group_desc
1767 #endif
1768
1769 struct qchk_ctxt {
1770         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1771         cfs_list_t              qckt_list;                 /* quotacheck list */
1772         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1773         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1774 };
1775
1776 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1777                            struct obd_quotactl *oqc)
1778 {
1779         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1780         loff_t size = 0;
1781         qid_t qid[MAXQUOTAS];
1782         int cnt, i, rc = 0;
1783
1784         if (!inode)
1785                 return 0;
1786
1787         qid[USRQUOTA] = inode->i_uid;
1788         qid[GRPQUOTA] = inode->i_gid;
1789
1790         if (S_ISDIR(inode->i_mode) ||
1791             S_ISREG(inode->i_mode) ||
1792             S_ISLNK(inode->i_mode))
1793                 size = inode_get_bytes(inode);
1794
1795         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1796                 if (!Q_TYPESET(oqc, cnt))
1797                         continue;
1798
1799                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1800                                 &qctxt->qckt_list, qid[cnt], cnt,
1801                                 qctxt->qckt_first_check[cnt]);
1802                 if (!cdqb[cnt]) {
1803                         rc = -ENOMEM;
1804                         break;
1805                 }
1806
1807                 cdqb[cnt]->dqb_curspace += size;
1808                 cdqb[cnt]->dqb_curinodes++;
1809         }
1810
1811         if (rc) {
1812                 for (i = 0; i < cnt; i++) {
1813                         if (!Q_TYPESET(oqc, i))
1814                                 continue;
1815                         LASSERT(cdqb[i]);
1816                         cdqb[i]->dqb_curspace -= size;
1817                         cdqb[i]->dqb_curinodes--;
1818                 }
1819         }
1820
1821         return rc;
1822 }
1823
1824 /* write dqinfo struct in a new quota file */
1825 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1826 {
1827         struct v2_disk_dqinfo dqinfo;
1828         __u32 blocks = V2_DQTREEOFF + 1;
1829         loff_t offset = V2_DQINFOOFF;
1830
1831         if (info) {
1832                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1833                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1834                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1835                                                ~DQF_INFO_DIRTY);
1836         } else {
1837                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1838                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1839                 dqinfo.dqi_flags = 0;
1840         }
1841
1842         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1843         dqinfo.dqi_free_blk = 0;
1844         dqinfo.dqi_free_entry = 0;
1845
1846         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1847 }
1848
1849 static int v3_write_dqheader(struct file *f, int type)
1850 {
1851         static const __u32 quota_magics[] = V2_INITQMAGICS;
1852         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1853         struct v2_disk_dqheader dqhead;
1854         loff_t offset = 0;
1855
1856         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1857         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1858
1859         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1860         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1861
1862         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1863 }
1864
1865 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1866                                   struct obd_quotactl *oqc)
1867 {
1868         int i, rc = 0;
1869         ENTRY;
1870
1871         for (i = 0; i < MAXQUOTAS; i++) {
1872                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1873                                          NULL : &qctxt->qckt_dqinfo[i];
1874                 struct file *file;
1875                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1876
1877                 if (!Q_TYPESET(oqc, i))
1878                         continue;
1879
1880                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1881
1882                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1883                 if (IS_ERR(file)) {
1884                         rc = PTR_ERR(file);
1885                         CERROR("can't create %s file: rc = %d\n",
1886                                name[i], rc);
1887                         GOTO(out, rc);
1888                 }
1889
1890                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1891                         CERROR("file %s is not regular", name[i]);
1892                         filp_close(file, 0);
1893                         GOTO(out, rc = -EINVAL);
1894                 }
1895
1896                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1897
1898                 rc = v3_write_dqheader(file, i);
1899                 if (rc) {
1900                         filp_close(file, 0);
1901                         GOTO(out, rc);
1902                 }
1903
1904                 rc = v3_write_dqinfo(file, i, info);
1905                 filp_close(file, 0);
1906                 if (rc)
1907                         GOTO(out, rc);
1908         }
1909
1910 out:
1911         RETURN(rc);
1912 }
1913
1914
1915 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1916                           struct chk_dqblk *cdqb)
1917 {
1918         struct obd_quotactl *oqc;
1919         long now;
1920         int rc;
1921         ENTRY;
1922
1923         OBD_ALLOC_PTR(oqc);
1924         if (!oqc)
1925                 RETURN(-ENOMEM);
1926
1927         now = cfs_time_current_sec();
1928
1929         if (cdqb->dqb_bsoftlimit &&
1930             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1931             !cdqb->dqb_btime)
1932                 cdqb->dqb_btime =
1933                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1934
1935         if (cdqb->dqb_isoftlimit &&
1936             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1937             !cdqb->dqb_itime)
1938                 cdqb->dqb_itime =
1939                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1940
1941         cdqb->dqb_valid = QIF_ALL;
1942
1943         oqc->qc_cmd = Q_SETQUOTA;
1944         oqc->qc_type = cdqb->dqb_type;
1945         oqc->qc_id = cdqb->dqb_id;
1946         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1947
1948         rc = fsfilt_ext3_quotactl(sb, oqc);
1949         OBD_FREE_PTR(oqc);
1950         RETURN(rc);
1951 }
1952
1953 static int prune_chkquots(struct super_block *sb,
1954                           struct qchk_ctxt *qctxt, int error)
1955 {
1956         struct chk_dqblk *cdqb, *tmp;
1957         int rc;
1958
1959         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1960                 if (!error) {
1961                         rc = commit_chkquot(sb, qctxt, cdqb);
1962                         if (rc)
1963                                 error = rc;
1964                 }
1965                 cfs_hlist_del_init(&cdqb->dqb_hash);
1966                 cfs_list_del(&cdqb->dqb_list);
1967                 OBD_FREE_PTR(cdqb);
1968         }
1969
1970         return error;
1971 }
1972
1973 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1974 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1975 #endif
1976
1977 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1978                                   struct obd_quotactl *oqc)
1979 {
1980         struct ext3_sb_info *sbi = EXT3_SB(sb);
1981         int i, group, uninit_feat = 0;
1982         struct qchk_ctxt *qctxt;
1983         struct buffer_head *bitmap_bh = NULL;
1984         unsigned long ino, inode_inuse;
1985         struct inode *inode;
1986         int rc = 0;
1987         ENTRY;
1988
1989         /* turn on quota and read dqinfo if existed */
1990         OBD_ALLOC_PTR(qctxt);
1991         if (!qctxt) {
1992                 oqc->qc_stat = -ENOMEM;
1993                 RETURN(-ENOMEM);
1994         }
1995
1996         for (i = 0; i < NR_DQHASH; i++)
1997                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1998         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
1999
2000         for (i = 0; i < MAXQUOTAS; i++) {
2001                 if (!Q_TYPESET(oqc, i))
2002                         continue;
2003
2004                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2005                 if (!rc || rc == -EBUSY) {
2006                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2007                         if (rc)
2008                                 GOTO(out, rc);
2009                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2010                         qctxt->qckt_first_check[i] = 1;
2011                 } else if (rc) {
2012                         GOTO(out, rc);
2013                 }
2014         }
2015         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2016                 /* This filesystem supports the uninit group feature */
2017                 uninit_feat = 1;
2018
2019         /* number of inodes that have been allocated */
2020         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2021                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2022
2023         /* check quota and update in hash */
2024         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2025              group++) {
2026                 unsigned long used_count = sbi->s_inodes_per_group;
2027
2028                 if (uninit_feat) {
2029                         struct ext3_group_desc *desc;
2030                         desc = get_group_desc(sb, group, NULL);
2031                         if (!desc)
2032                                 GOTO(out, -EIO);
2033
2034                         /* we don't really need to take the group lock here,
2035                          * but it may be useful if one day we support online
2036                          * quotacheck */
2037 #ifdef HAVE_EXT4_LDISKFS
2038                         ext4_lock_group(sb, group);
2039 #else
2040                         spin_lock(sb_bgl_lock(sbi, group));
2041 #endif
2042                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2043                                 /* no inode in use in this group, just skip it */
2044 #ifdef HAVE_EXT4_LDISKFS
2045                                 ext3_unlock_group(sb, group);
2046 #else
2047                                 spin_unlock(sb_bgl_lock(sbi, group));
2048 #endif
2049                                 continue;
2050                         }
2051
2052                         used_count -= ext3_itable_unused_count(sb, desc);
2053 #ifdef HAVE_EXT4_LDISKFS
2054                         ext3_unlock_group(sb, group);
2055 #else
2056                         spin_unlock(sb_bgl_lock(sbi, group));
2057 #endif
2058                 }
2059
2060                 ino = group * sbi->s_inodes_per_group + 1;
2061                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2062                 if (!bitmap_bh) {
2063                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2064                                sb->s_id, group);
2065                         GOTO(out, -EIO);
2066                 }
2067
2068                 i = 0;
2069                 while (i < used_count &&
2070                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2071                                                used_count, i)) < used_count) {
2072                         inode_inuse--;
2073                         i++;
2074                         ino = i + group * sbi->s_inodes_per_group;
2075                         if (ino < sbi->s_first_ino)
2076                                 continue;
2077 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2078                         inode = ext3_iget(sb, ino);
2079 #else
2080                         inode = iget(sb, ino);
2081 #endif
2082                         if (!inode || IS_ERR(inode))
2083                                 continue;
2084
2085                         rc = add_inode_quota(inode, qctxt, oqc);
2086                         iput(inode);
2087                         if (rc) {
2088                                 brelse(bitmap_bh);
2089                                 GOTO(out, rc);
2090                         }
2091                 }
2092                 brelse(bitmap_bh);
2093         }
2094
2095         /* read old quota limits from old quota file. (only for the user
2096          * has limits but hasn't file) */
2097 #ifdef HAVE_QUOTA_SUPPORT
2098         for (i = 0; i < MAXQUOTAS; i++) {
2099                 cfs_list_t id_list;
2100                 struct dquot_id *dqid, *tmp;
2101
2102                 if (!Q_TYPESET(oqc, i))
2103                         continue;
2104
2105                 if (qctxt->qckt_first_check[i])
2106                         continue;
2107
2108
2109                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2110                 CFS_INIT_LIST_HEAD(&id_list);
2111 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2112                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2113 #else
2114                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2115 #endif
2116                 if (rc)
2117                         CERROR("read old limits failed. (rc:%d)\n", rc);
2118
2119                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2120                         cfs_list_del_init(&dqid->di_link);
2121
2122                         if (!rc)
2123                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2124                                       dqid->di_id, i,
2125                                       qctxt->qckt_first_check[i]);
2126                         OBD_FREE_PTR(dqid);
2127                 }
2128         }
2129 #endif
2130         /* turn off quota cause we are to dump chk_dqblk to files */
2131         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2132
2133         rc = create_new_quota_files(qctxt, oqc);
2134         if (rc)
2135                 GOTO(out, rc);
2136
2137         /* we use vfs functions to set dqblk, so turn quota on */
2138         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2139 out:
2140         /* dump and free chk_dqblk */
2141         rc = prune_chkquots(sb, qctxt, rc);
2142         OBD_FREE_PTR(qctxt);
2143
2144         /* turn off quota, `lfs quotacheck` will turn on when all
2145          * nodes quotacheck finish. */
2146         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2147
2148         oqc->qc_stat = rc;
2149         if (rc)
2150                 CERROR("quotacheck failed: rc = %d\n", rc);
2151
2152         RETURN(rc);
2153 }
2154
2155 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2156                                  int cmd)
2157 {
2158         int rc = 0;
2159         ENTRY;
2160
2161         if (lqi->qi_files[type] == NULL) {
2162                 CERROR("operate qinfo before it's enabled!\n");
2163                 RETURN(-ESRCH);
2164         }
2165
2166         switch (cmd) {
2167         case QFILE_CHK:
2168                 rc = lustre_check_quota_file(lqi, type);
2169                 break;
2170         case QFILE_RD_INFO:
2171                 rc = lustre_read_quota_info(lqi, type);
2172                 break;
2173         case QFILE_WR_INFO:
2174                 rc = lustre_write_quota_info(lqi, type);
2175                 break;
2176         case QFILE_INIT_INFO:
2177                 rc = lustre_init_quota_info(lqi, type);
2178                 break;
2179         case QFILE_CONVERT:
2180                 rc = -ENOTSUPP;
2181                 CERROR("quota CONVERT command is not supported\n");
2182                 break;
2183         default:
2184                 rc = -ENOTSUPP;
2185                 CERROR("Unsupported admin quota file cmd %d\n"
2186                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2187                        cmd);
2188                 break;
2189         }
2190         RETURN(rc);
2191 }
2192
2193 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2194                             cfs_list_t *list)
2195 {
2196         return lustre_get_qids(file, inode, type, list);
2197 }
2198
2199 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2200 {
2201         int rc = 0;
2202         ENTRY;
2203
2204         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2205                 CERROR("operate dquot before it's enabled!\n");
2206                 RETURN(-ESRCH);
2207         }
2208
2209         switch (cmd) {
2210         case QFILE_RD_DQUOT:
2211                 rc = lustre_read_dquot(dquot);
2212                 break;
2213         case QFILE_WR_DQUOT:
2214                 if (dquot->dq_dqb.dqb_ihardlimit ||
2215                     dquot->dq_dqb.dqb_isoftlimit ||
2216                     dquot->dq_dqb.dqb_bhardlimit ||
2217                     dquot->dq_dqb.dqb_bsoftlimit)
2218                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2219                 else
2220                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2221
2222                 rc = lustre_commit_dquot(dquot);
2223                 if (rc >= 0)
2224                         rc = 0;
2225                 break;
2226         default:
2227                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2228                 LBUG();
2229                 break;
2230         }
2231         RETURN(rc);
2232 }
2233
2234 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2235                                 struct inode *inode, int frags)
2236 {
2237 #ifdef EXT3_EXT_HAS_NO_TREE
2238         struct ext3_ext_base *base = inode;
2239 #else
2240         struct ext3_extents_tree tree;
2241         struct ext3_ext_base *base = &tree;
2242
2243         ext3_init_tree_desc(base, inode);
2244 #endif
2245         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2246          * metablocks at maxium b=16542 */
2247         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2248         return 0;
2249 }
2250
2251 #endif
2252
2253 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2254 {
2255         return (EXT3_SB(sb)->journal_bdev);
2256 }
2257 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2258
2259 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2260                           char *buf, int count, loff_t pos)
2261 {
2262         loff_t p = pos;
2263         int rc;
2264
2265         if (!f && !inode) {
2266                 CERROR("lustre_read_quota failed for no quota file!\n");
2267                 libcfs_debug_dumpstack(NULL);
2268                 return -EINVAL;
2269         }
2270
2271         /* Support for both adm and op quota files must be provided */
2272         if (f) {
2273                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2274                 rc = rc < 0 ? rc : p - pos;
2275         } else {
2276                 struct super_block *sb = inode->i_sb;
2277                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2278         }
2279         return rc;
2280 }
2281
2282 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2283 {
2284         loff_t p = pos;
2285         int rc;
2286
2287         /* Only adm quota files are supported, op updates are handled by vfs */
2288         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2289         rc = rc < 0 ? rc : p - pos;
2290
2291         return rc;
2292 }
2293
2294 void *lustre_quota_journal_start(struct inode *inode, int delete)
2295 {
2296         handle_t *handle;
2297         unsigned block_count;
2298
2299         if (delete) {
2300                 /* each indirect block (+4) may become free, attaching to the
2301                  * header list of free blocks (+1); the data block (+1) may
2302                  * become a free block (+0) or a block with free dqentries (+0) */
2303                 block_count = (4 + 1) + 1;
2304                 handle = ext3_journal_start(inode,
2305                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2306         } else {
2307                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2308                  * freeblk reusage with a header update (+1); dqentry is either reused
2309                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2310                  * a new block allocation (+1) with a header update (+1)              */
2311                 block_count = (4 + 1) + 3;
2312                 handle = ext3_journal_start(inode,
2313                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2314
2315         }
2316
2317         return handle;
2318 }
2319
2320 void lustre_quota_journal_stop(void *handle)
2321 {
2322         ext3_journal_stop((handle_t *)handle);
2323 }
2324
2325 static int ll_decode_fh_accept(void *context, struct dentry *de)
2326 {
2327         return 1;
2328 }
2329
2330 #ifdef HAVE_EXPORTFS_DECODE_FH
2331 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2332          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2333                             acceptable, context)
2334 #else
2335 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2336          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2337                                      type, acceptable, context)
2338 # define FILEID_INO32_GEN 1
2339 extern struct export_operations export_op_default;
2340 #endif
2341
2342 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2343                                       struct fsfilt_fid *fid, int ignore_gen)
2344 {
2345         struct inode  *inode;
2346         struct dentry *result;
2347
2348         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2349                                        ll_decode_fh_accept, NULL);
2350         if (IS_ERR(result)) {
2351                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2352                        fid->ino, fid->gen, PTR_ERR(result));
2353                 return result;
2354         }
2355
2356         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2357                fid->ino, fid->gen);
2358         inode = result->d_inode;
2359         if (inode == NULL)
2360                 goto err_out;
2361
2362         if (inode->i_nlink == 0 &&
2363             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2364                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2365                               " ctime -- this may indicate disk "
2366                               "corruption (inode: %lu, link: %lu, "
2367                               "count: %d)\n", inode->i_ino,
2368                               (unsigned long)inode->i_nlink,
2369                               atomic_read(&inode->i_count));
2370                 goto err_out;
2371         }
2372         if (fid->gen && inode->i_generation != fid->gen) {
2373                 /* we didn't find the right inode.. */
2374                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2375                        "count: %d, generation %u/%u\n",
2376                        inode->i_ino, (unsigned long)inode->i_nlink,
2377                        atomic_read(&inode->i_count), inode->i_generation,
2378                        fid->gen);
2379                 goto err_out;
2380         }
2381
2382         return result;
2383 err_out:
2384         l_dput(result);
2385         return ERR_PTR(-ENOENT);
2386 }
2387
2388 static struct fsfilt_operations fsfilt_ext3_ops = {
2389         .fs_type                = "ext3",
2390         .fs_owner               = THIS_MODULE,
2391         .fs_getlabel            = fsfilt_ext3_get_label,
2392         .fs_setlabel            = fsfilt_ext3_set_label,
2393         .fs_uuid                = fsfilt_ext3_uuid,
2394         .fs_start               = fsfilt_ext3_start,
2395         .fs_brw_start           = fsfilt_ext3_brw_start,
2396         .fs_extend              = fsfilt_ext3_extend,
2397         .fs_commit              = fsfilt_ext3_commit,
2398         .fs_commit_async        = fsfilt_ext3_commit_async,
2399         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2400         .fs_setattr             = fsfilt_ext3_setattr,
2401         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2402         .fs_set_md              = fsfilt_ext3_set_md,
2403         .fs_get_md              = fsfilt_ext3_get_md,
2404         .fs_readpage            = fsfilt_ext3_readpage,
2405         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2406         .fs_statfs              = fsfilt_ext3_statfs,
2407         .fs_sync                = fsfilt_ext3_sync,
2408         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2409         .fs_write_record        = fsfilt_ext3_write_record,
2410         .fs_read_record         = fsfilt_ext3_read_record,
2411         .fs_setup               = fsfilt_ext3_setup,
2412         .fs_send_bio            = fsfilt_ext3_send_bio,
2413         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2414 #ifdef HAVE_DISK_INODE_VERSION
2415         .fs_get_version         = fsfilt_ext3_get_version,
2416         .fs_set_version         = fsfilt_ext3_set_version,
2417 #endif
2418 #ifdef HAVE_QUOTA_SUPPORT
2419         .fs_quotactl            = fsfilt_ext3_quotactl,
2420         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2421         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2422         .fs_qids                = fsfilt_ext3_qids,
2423         .fs_dquot               = fsfilt_ext3_dquot,
2424         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2425 #endif
2426         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2427         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2428 };
2429
2430 static int __init fsfilt_ext3_init(void)
2431 {
2432         int rc;
2433
2434         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2435                                          sizeof(struct fsfilt_cb_data), 0, 0);
2436         if (!fcb_cache) {
2437                 CERROR("error allocating fsfilt journal callback cache\n");
2438                 GOTO(out, rc = -ENOMEM);
2439         }
2440
2441         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2442
2443         if (rc) {
2444                 int err = cfs_mem_cache_destroy(fcb_cache);
2445                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2446         }
2447 out:
2448         return rc;
2449 }
2450
2451 static void __exit fsfilt_ext3_exit(void)
2452 {
2453         int rc;
2454
2455         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2456         rc = cfs_mem_cache_destroy(fcb_cache);
2457         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2458 }
2459
2460 module_init(fsfilt_ext3_init);
2461 module_exit(fsfilt_ext3_exit);
2462
2463 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2464 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2465 MODULE_LICENSE("GPL");