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_V1_H
  64 # include <linux/quotaio_v1.h>
  65 # include <linux/quotaio_v2.h>
  66 #elif defined(HAVE_FS_QUOTA_QUOTAIO_V1_H)
  67 # include <quota/quotaio_v1.h>
  68 # include <quota/quotaio_v2.h>
  69 # include <quota/quota_tree.h>
  70 # define V2_DQTREEOFF    QT_TREEOFF
  71 #else
  72 # include <quotaio_v1.h>
  73 # include <quotaio_v2.h>
  74 # include <quota_tree.h>
  75 # define V2_DQTREEOFF    QT_TREEOFF
  76 #endif
  77
  78 #if defined(HAVE_EXT3_XATTR_H)
  79 #include <ext3/xattr.h>
  80 #else
  81 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  82 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  83 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  84 #endif
  85
  86 #include <libcfs/libcfs.h>
  87 #include <lustre_fsfilt.h>
  88 #include <obd.h>
  89 #include <lustre_quota.h>
  90 #include <linux/lustre_compat25.h>
  91 #include <linux/lprocfs_status.h>
  92
  93 #ifdef HAVE_EXT4_LDISKFS
  94 #include <ext4/ext4_extents.h>
  95 #else
  96 #include <linux/ext3_extents.h>
  97 #endif
  98
  99 #include "lustre_quota_fmt.h"
 100
 101 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
 102 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 103 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 104 #else
 105 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 106 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 107 #endif
 108
 109 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 110 /* for kernels 2.6.18 and later */
 111 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 112 #else
 113 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 114 #endif
 115
 116 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 117 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 118                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 119 #else
 120 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 121                ext3_ext_insert_extent(handle, inode, path, newext)
 122 #endif
 123
 124
 125 static cfs_mem_cache_t *fcb_cache;
 126
 127 struct fsfilt_cb_data {
 128         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 129         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 130         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 131         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 132         void *cb_data;                  /* MDS/OST completion function data */
 133 };
 134
 135 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 136 #define EXT3_XATTR_INDEX_TRUSTED        4
 137 #endif
 138
 139 #ifdef HAVE_EXT4_LDISKFS
 140 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 141 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 142 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 143 #else
 144 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 145 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 146 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 147 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 148 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 149 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 150 #endif
 151
 152 #ifndef ext3_find_next_bit
 153 #define ext3_find_next_bit           ext2_find_next_bit
 154 #endif
 155
 156 #ifndef ext2_find_next_bit
 157 #ifdef __LITTLE_ENDIAN
 158 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 159 #else
 160 error "Need implementation of find_next_bit on big-endian systems"
 161 #endif  /* __LITTLE_ENDIAN */
 162 #endif  /* !ext2_find_next_le_bit */
 163
 164 static char *fsfilt_ext3_get_label(struct super_block *sb)
 165 {
 166         return EXT3_SB(sb)->s_es->s_volume_name;
 167 }
 168
 169 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 170 {
 171         /* see e.g. fsfilt_ext3_write_record() */
 172         journal_t *journal;
 173         handle_t *handle;
 174         int err;
 175
 176         journal = EXT3_SB(sb)->s_journal;
 177         handle = ext3_journal_start_sb(sb, 1);
 178         if (IS_ERR(handle)) {
 179                 CERROR("can't start transaction\n");
 180                 return(PTR_ERR(handle));
 181         }
 182
 183         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 184         if (err)
 185                 goto out;
 186
 187         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 188                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 189
 190         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 191
 192 out:
 193         ext3_journal_stop(handle);
 194
 195         return(err);
 196 }
 197
 198 static char *fsfilt_ext3_uuid(struct super_block *sb)
 199 {
 200         return EXT3_SB(sb)->s_es->s_uuid;
 201 }
 202
 203 #ifdef HAVE_DISK_INODE_VERSION
 204
 205 static __u64 get_i_version(struct inode *inode)
 206 {
 207 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 208         return inode->i_version;
 209 #else
 210         return EXT3_I(inode)->i_fs_version;
 211 #endif
 212 }
 213
 214 static void set_i_version(struct inode *inode, __u64 new_version)
 215 {
 216 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 217         inode->i_version = new_version;
 218 #else
 219         (EXT3_I(inode))->i_fs_version = new_version;
 220 #endif
 221 }
 222
 223 /*
 224  * Get the 64-bit version for an inode.
 225  */
 226 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 227 {
 228         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 229                get_i_version(inode), inode->i_ino);
 230         return get_i_version(inode);
 231 }
 232
 233 /*
 234  * Set the 64-bit version and return the old version.
 235  */
 236 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 237 {
 238         __u64 old_version = get_i_version(inode);
 239
 240         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 241                new_version, old_version, inode->i_ino);
 242         set_i_version(inode, new_version);
 243         /* version is set after all inode operations are finished, so we should
 244          * mark it dirty here */
 245         inode->i_sb->s_op->dirty_inode(inode);
 246         return old_version;
 247 }
 248
 249 #endif
 250
 251 /*
 252  * We don't currently need any additional blocks for rmdir and
 253  * unlink transactions because we are storing the OST oa_id inside
 254  * the inode (which we will be changing anyways as part of this
 255  * transaction).
 256  */
 257 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 258                                int logs)
 259 {
 260         /* For updates to the last received file */
 261         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 262         journal_t *journal;
 263         void *handle;
 264
 265         if (current->journal_info) {
 266                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 267                        current->journal_info);
 268                 goto journal_start;
 269         }
 270
 271         switch(op) {
 272         case FSFILT_OP_RMDIR:
 273         case FSFILT_OP_UNLINK:
 274                 /* delete one file + create/update logs for each stripe */
 275                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 276                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 277                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 278                 break;
 279         case FSFILT_OP_RENAME:
 280                 /* modify additional directory */
 281                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 282                 /* no break */
 283         case FSFILT_OP_SYMLINK:
 284                 /* additional block + block bitmap + GDT for long symlink */
 285                 nblocks += 3;
 286                 /* no break */
 287         case FSFILT_OP_CREATE: {
 288 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 289                 static int warned;
 290                 if (!warned) {
 291                         if (!test_opt(inode->i_sb, EXTENTS)) {
 292                                 warned = 1;
 293                         } else if (((EXT3_I(inode)->i_flags &
 294                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 295                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 296                                 CWARN("extent-mapped directory found with "
 297                                       "ext3-based ldiskfs - contact "
 298                                       "http://bugzilla.lustre.org/\n");
 299                                 warned = 1;
 300                         }
 301                 }
 302 #endif
 303                 /* no break */
 304         }
 305         case FSFILT_OP_MKDIR:
 306         case FSFILT_OP_MKNOD:
 307                 /* modify one inode + block bitmap + GDT */
 308                 nblocks += 3;
 309                 /* no break */
 310         case FSFILT_OP_LINK:
 311                 /* modify parent directory */
 312                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 313                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 314                 /* create/update logs for each stripe */
 315                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 316                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 317                 break;
 318         case FSFILT_OP_SETATTR:
 319                 /* Setattr on inode */
 320                 nblocks += 1;
 321                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 322                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 323                 /* quota chown log 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_CANCEL_UNLINK:
 328                 /* blocks for log header bitmap update OR
 329                  * blocks for catalog header bitmap update + unlink of logs */
 330                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 331                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 332                 break;
 333         default: CERROR("unknown transaction start op %d\n", op);
 334                 LBUG();
 335         }
 336
 337         LASSERT(current->journal_info == desc_private);
 338         journal = EXT3_SB(inode->i_sb)->s_journal;
 339         if (nblocks > journal->j_max_transaction_buffers) {
 340                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 341                        nblocks, op, logs, journal->j_max_transaction_buffers);
 342                 nblocks = journal->j_max_transaction_buffers;
 343         }
 344
 345  journal_start:
 346         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 347         handle = ext3_journal_start(inode, nblocks);
 348
 349         if (!IS_ERR(handle))
 350                 LASSERT(current->journal_info == handle);
 351         else
 352                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 353                        op, nblocks, PTR_ERR(handle));
 354         return handle;
 355 }
 356
 357 /*
 358  * Calculate the number of buffer credits needed to write multiple pages in
 359  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 360  * doesn't have a nice API for calculating this sort of thing in advance.
 361  *
 362  * See comment above ext3_writepage_trans_blocks for details.  We assume
 363  * no data journaling is being done, but it does allow for all of the pages
 364  * being non-contiguous.  If we are guaranteed contiguous pages we could
 365  * reduce the number of (d)indirect blocks a lot.
 366  *
 367  * With N blocks per page and P pages, for each inode we have at most:
 368  * N*P indirect
 369  * min(N*P, blocksize/4 + 1) dindirect blocks
 370  * niocount tindirect
 371  *
 372  * For the entire filesystem, we have at most:
 373  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 374  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 375  * objcount inode blocks
 376  * 1 superblock
 377  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 378  *
 379  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 380  */
 381 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 382                                       int niocount, struct niobuf_local *nb)
 383 {
 384         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 385         __u64 next_indir;
 386         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 387         int nbitmaps = 0, ngdblocks;
 388         int needed = objcount + 1; /* inodes + superblock */
 389         int i, j;
 390
 391         for (i = 0, j = 0; i < objcount; i++, fso++) {
 392                 /* two or more dindirect blocks in case we cross boundary */
 393                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 394                                     nb[j].offset) >>
 395                                    sb->s_blocksize_bits) /
 396                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 397                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 398
 399                 /* leaf, indirect, tindirect blocks for first block */
 400                 nbitmaps += blockpp + 2;
 401
 402                 j += fso->fso_bufcnt;
 403         }
 404
 405         next_indir = nb[0].offset +
 406                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 407         for (i = 1; i < niocount; i++) {
 408                 if (nb[i].offset >= next_indir) {
 409                         nbitmaps++;     /* additional indirect */
 410                         next_indir = nb[i].offset +
 411                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 412                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 413                         nbitmaps++;     /* additional indirect */
 414                 }
 415                 nbitmaps += blockpp;    /* each leaf in different group? */
 416         }
 417
 418         ngdblocks = nbitmaps;
 419         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 420                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 421         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 422                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 423
 424         needed += nbitmaps + ngdblocks;
 425
 426         /* last_rcvd update */
 427         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 428
 429 #if defined(CONFIG_QUOTA)
 430         /* We assume that there will be 1 bit set in s_dquot.flags for each
 431          * quota file that is active.  This is at least true for now.
 432          */
 433         needed += hweight32(ll_sb_any_quota_active(sb)) *
 434                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 435 #endif
 436
 437         return needed;
 438 }
 439
 440 /* We have to start a huge journal transaction here to hold all of the
 441  * metadata for the pages being written here.  This is necessitated by
 442  * the fact that we do lots of prepare_write operations before we do
 443  * any of the matching commit_write operations, so even if we split
 444  * up to use "smaller" transactions none of them could complete until
 445  * all of them were opened.  By having a single journal transaction,
 446  * we eliminate duplicate reservations for common blocks like the
 447  * superblock and group descriptors or bitmaps.
 448  *
 449  * We will start the transaction here, but each prepare_write will
 450  * add a refcount to the transaction, and each commit_write will
 451  * remove a refcount.  The transaction will be closed when all of
 452  * the pages have been written.
 453  */
 454 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 455                                    int niocount, struct niobuf_local *nb,
 456                                    void *desc_private, int logs)
 457 {
 458         journal_t *journal;
 459         handle_t *handle;
 460         int needed;
 461         ENTRY;
 462
 463         LASSERT(current->journal_info == desc_private);
 464         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 465         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 466
 467         /* The number of blocks we could _possibly_ dirty can very large.
 468          * We reduce our request if it is absurd (and we couldn't get that
 469          * many credits for a single handle anyways).
 470          *
 471          * At some point we have to limit the size of I/Os sent at one time,
 472          * increase the size of the journal, or we have to calculate the
 473          * actual journal requirements more carefully by checking all of
 474          * the blocks instead of being maximally pessimistic.  It remains to
 475          * be seen if this is a real problem or not.
 476          */
 477         if (needed > journal->j_max_transaction_buffers) {
 478                 CERROR("want too many journal credits (%d) using %d instead\n",
 479                        needed, journal->j_max_transaction_buffers);
 480                 needed = journal->j_max_transaction_buffers;
 481         }
 482
 483         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 484         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 485         if (IS_ERR(handle)) {
 486                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 487                        PTR_ERR(handle));
 488         } else {
 489                 LASSERT(handle->h_buffer_credits >= needed);
 490                 LASSERT(current->journal_info == handle);
 491         }
 492
 493         RETURN(handle);
 494 }
 495
 496 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 497 {
 498        handle_t *handle = h;
 499
 500        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 501        if (nblocks == 0) {
 502                handle->h_buffer_credits = 0;
 503                CWARN("setting credits of handle %p to zero by request\n", h);
 504        }
 505
 506        if (handle->h_buffer_credits > nblocks)
 507                 return 0;
 508        if (ext3_journal_extend(handle, nblocks) == 0)
 509                 return 0;
 510
 511        ext3_mark_inode_dirty(handle, inode);
 512        return ext3_journal_restart(handle, nblocks);
 513 }
 514
 515 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 516 {
 517         int rc;
 518         handle_t *handle = h;
 519
 520         LASSERT(current->journal_info == handle);
 521         if (force_sync)
 522                 handle->h_sync = 1; /* recovery likes this */
 523
 524         rc = ext3_journal_stop(handle);
 525
 526         return rc;
 527 }
 528
 529 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 530                                     void **wait_handle)
 531 {
 532         unsigned long tid;
 533         transaction_t *transaction;
 534         handle_t *handle = h;
 535         journal_t *journal;
 536         int rc;
 537
 538         LASSERT(current->journal_info == handle);
 539
 540         transaction = handle->h_transaction;
 541         journal = transaction->t_journal;
 542         tid = transaction->t_tid;
 543         /* we don't want to be blocked */
 544         handle->h_sync = 0;
 545         rc = ext3_journal_stop(handle);
 546         if (rc) {
 547                 CERROR("error while stopping transaction: %d\n", rc);
 548                 return rc;
 549         }
 550         fsfilt_log_start_commit(journal, tid);
 551
 552         *wait_handle = (void *) tid;
 553         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 554         return 0;
 555 }
 556
 557 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 558 {
 559         journal_t *journal = EXT3_JOURNAL(inode);
 560         tid_t tid = (tid_t)(long)h;
 561
 562         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 563         if (unlikely(is_journal_aborted(journal)))
 564                 return -EIO;
 565
 566         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 567
 568         if (unlikely(is_journal_aborted(journal)))
 569                 return -EIO;
 570         return 0;
 571 }
 572
 573 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 574                                struct iattr *iattr, int do_trunc)
 575 {
 576         struct inode *inode = dentry->d_inode;
 577         int rc = 0;
 578
 579         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 580          * We are already writing the inode to disk as part of this
 581          * transaction and want to avoid a lot of extra inode writeout
 582          * later on. b=9828 */
 583         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 584                 /* ATTR_SIZE would invoke truncate: clear it */
 585                 iattr->ia_valid &= ~ATTR_SIZE;
 586                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 587                 i_size_write(inode, iattr->ia_size);
 588
 589                 if (iattr->ia_valid & ATTR_UID)
 590                         inode->i_uid = iattr->ia_uid;
 591                 if (iattr->ia_valid & ATTR_GID)
 592                         inode->i_gid = iattr->ia_gid;
 593                 if (iattr->ia_valid & ATTR_ATIME)
 594                         inode->i_atime = iattr->ia_atime;
 595                 if (iattr->ia_valid & ATTR_MTIME)
 596                         inode->i_mtime = iattr->ia_mtime;
 597                 if (iattr->ia_valid & ATTR_CTIME)
 598                         inode->i_ctime = iattr->ia_ctime;
 599                 if (iattr->ia_valid & ATTR_MODE) {
 600                         inode->i_mode = iattr->ia_mode;
 601
 602                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 603                             !cfs_capable(CFS_CAP_FSETID))
 604                                 inode->i_mode &= ~S_ISGID;
 605                 }
 606
 607                 inode->i_sb->s_op->dirty_inode(inode);
 608
 609                 goto out;
 610         }
 611
 612         /* Don't allow setattr to change file type */
 613         if (iattr->ia_valid & ATTR_MODE)
 614                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 615                                  (iattr->ia_mode & ~S_IFMT);
 616
 617         /* We set these flags on the client, but have already checked perms
 618          * so don't confuse inode_change_ok. */
 619         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 620
 621         if (inode->i_op->setattr) {
 622                 rc = inode->i_op->setattr(dentry, iattr);
 623         } else {
 624                 rc = inode_change_ok(inode, iattr);
 625                 if (!rc)
 626                         rc = inode_setattr(inode, iattr);
 627         }
 628
 629  out:
 630         RETURN(rc);
 631 }
 632
 633 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 634                                  unsigned int cmd, unsigned long arg)
 635 {
 636         int rc = 0;
 637         ENTRY;
 638
 639         /* FIXME: Can't do this because of nested transaction deadlock */
 640         if (cmd == EXT3_IOC_SETFLAGS) {
 641                 /* We can't enable data journaling on OST objects, because
 642                 * this forces the transaction to be closed in order to
 643                 * flush the journal, but the caller will already have a
 644                 * compound transaction open to update the last_rcvd file,
 645                 * and this thread would deadlock trying to set the flag. */
 646                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 647                         CERROR("can't set data journal flag on file\n");
 648                         RETURN(-EPERM);
 649                 }
 650                 /* Because the MDS does not see the EXTENTS_FL set on the
 651                  * OST objects, mask this flag into all set flags.  It is
 652                  * not legal to clear this flag in any case, so we are not
 653                  * changing the functionality by doing this.  b=22911 */
 654                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 655         }
 656
 657 #ifdef HAVE_EXT4_LDISKFS
 658         /* ext4_ioctl does not have a inode argument */
 659         if (inode->i_fop->unlocked_ioctl)
 660                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 661 #else
 662         if (inode->i_fop->ioctl)
 663                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 664 #endif
 665         else
 666                 RETURN(-ENOTTY);
 667
 668         RETURN(rc);
 669 }
 670
 671 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 672                               void *lmm, int lmm_size, const char *name)
 673 {
 674         int rc;
 675
 676         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 677
 678         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 679                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 680
 681
 682         if (rc && rc != -EROFS)
 683                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 684                        inode->i_ino, rc);
 685         return rc;
 686 }
 687
 688 /* Must be called with i_mutex held */
 689 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 690                               const char *name)
 691 {
 692         int rc;
 693
 694         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 695
 696         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 697                             name, lmm, lmm_size);
 698
 699         /* This gives us the MD size */
 700         if (lmm == NULL)
 701                 return (rc == -ENODATA) ? 0 : rc;
 702
 703         if (rc < 0) {
 704                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 705                        EXT3_XATTR_INDEX_TRUSTED, name,
 706                        inode->i_ino, rc);
 707                 memset(lmm, 0, lmm_size);
 708                 return (rc == -ENODATA) ? 0 : rc;
 709         }
 710
 711         return rc;
 712 }
 713
 714 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 715 {
 716         submit_bio(rw, bio);
 717         return 0;
 718 }
 719
 720 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 721                                     loff_t *off)
 722 {
 723         struct inode *inode = file->f_dentry->d_inode;
 724         int rc = 0;
 725
 726         if (S_ISREG(inode->i_mode))
 727                 rc = file->f_op->read(file, buf, count, off);
 728         else {
 729                 const int blkbits = inode->i_sb->s_blocksize_bits;
 730                 const int blksize = inode->i_sb->s_blocksize;
 731
 732                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 733                        (unsigned long)count, inode->i_ino, *off);
 734                 while (count > 0) {
 735                         struct buffer_head *bh;
 736
 737                         bh = NULL;
 738                         if (*off < i_size_read(inode)) {
 739                                 int err = 0;
 740
 741                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 742                                                 0, &err);
 743
 744                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 745
 746                                 if (bh) {
 747                                         memcpy(buf, bh->b_data, blksize);
 748                                         brelse(bh);
 749                                 } else if (err) {
 750                                         /* XXX in theory we should just fake
 751                                          * this buffer and continue like ext3,
 752                                          * especially if this is a partial read
 753                                          */
 754                                         CERROR("error read dir %lu+%llu: %d\n",
 755                                                inode->i_ino, *off, err);
 756                                         RETURN(err);
 757                                 }
 758                         }
 759                         if (!bh) {
 760                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 761
 762                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 763                                 memset(fake, 0, sizeof(*fake));
 764                                 fake->rec_len = cpu_to_le16(blksize);
 765                         }
 766                         count -= blksize;
 767                         buf += blksize;
 768                         *off += blksize;
 769                         rc += blksize;
 770                 }
 771         }
 772
 773         return rc;
 774 }
 775
 776 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 777 {
 778         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 779
 780         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 781
 782         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 783 }
 784
 785 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 786                                       void *handle, fsfilt_cb_t cb_func,
 787                                       void *cb_data)
 788 {
 789         struct fsfilt_cb_data *fcb;
 790
 791         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 792         if (fcb == NULL)
 793                 RETURN(-ENOMEM);
 794
 795         fcb->cb_func = cb_func;
 796         fcb->cb_obd = obd;
 797         fcb->cb_last_rcvd = last_rcvd;
 798         fcb->cb_data = cb_data;
 799
 800         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 801         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 802                                     (struct journal_callback *)fcb);
 803
 804         return 0;
 805 }
 806
 807 /*
 808  * We need to hack the return value for the free inode counts because
 809  * the current EA code requires one filesystem block per inode with EAs,
 810  * so it is possible to run out of blocks before we run out of inodes.
 811  *
 812  * This can be removed when the ext3 EA code is fixed.
 813  */
 814 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 815 {
 816         struct kstatfs sfs;
 817         int rc;
 818
 819         memset(&sfs, 0, sizeof(sfs));
 820         rc = ll_do_statfs(sb, &sfs);
 821         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 822                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 823                 sfs.f_ffree = sfs.f_bfree;
 824         }
 825
 826         statfs_pack(osfs, &sfs);
 827         return rc;
 828 }
 829
 830 static int fsfilt_ext3_sync(struct super_block *sb)
 831 {
 832         return ext3_force_commit(sb);
 833 }
 834
 835 #ifndef EXT3_EXTENTS_FL
 836 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 837 #endif
 838
 839 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 840 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 841 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 842 #else
 843 # ifdef HAVE_EXT4_LDISKFS
 844 #  ifdef WALK_SPACE_HAS_DATA_SEM /* We only use it in fsfilt_map_nblocks() for now */
 845 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 846 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 847 #  else
 848 #   define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem))
 849 #   define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem))
 850 #  endif
 851 # else
 852 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 853 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 854 # endif
 855 #endif
 856
 857 #ifndef EXT_ASSERT
 858 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 859 #endif
 860
 861 #ifdef EXT3_EXT_HAS_NO_TREE
 862 /* for kernels 2.6.18 and later */
 863 #ifdef HAVE_EXT4_LDISKFS
 864 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 865 #else
 866 #define EXT_GENERATION(inode)           ext_generation(inode)
 867 #endif
 868 #define ext3_ext_base                   inode
 869 #define ext3_ext_base2inode(inode)      (inode)
 870 #define EXT_DEPTH(inode)                ext_depth(inode)
 871 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 872                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 873 #else
 874 #define ext3_ext_base                   ext3_extents_tree
 875 #define ext3_ext_base2inode(tree)       (tree->inode)
 876 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 877                         ext3_ext_walk_space(tree, block, num, cb);
 878 #endif
 879
 880 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 881 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 882                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 883 #else
 884 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 885                ext3_ext_insert_extent(handle, inode, path, newext)
 886 #endif
 887
 888 #include <linux/lustre_version.h>
 889
 890 struct bpointers {
 891         unsigned long *blocks;
 892         int *created;
 893         unsigned long start;
 894         int num;
 895         int init_num;
 896         int create;
 897 };
 898
 899 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 900                                unsigned long block, int *aflags)
 901 {
 902         struct ext3_inode_info *ei = EXT3_I(inode);
 903         unsigned long bg_start;
 904         unsigned long colour;
 905         int depth;
 906
 907         if (path) {
 908                 struct ext3_extent *ex;
 909                 depth = path->p_depth;
 910
 911                 /* try to predict block placement */
 912                 if ((ex = path[depth].p_ext))
 913                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 914
 915                 /* it looks index is empty
 916                  * try to find starting from index itself */
 917                 if (path[depth].p_bh)
 918                         return path[depth].p_bh->b_blocknr;
 919         }
 920
 921         /* OK. use inode's group */
 922         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 923                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 924         colour = (current->pid % 16) *
 925                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 926         return bg_start + colour + block;
 927 }
 928
 929 #define ll_unmap_underlying_metadata(sb, blocknr) \
 930         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 931
 932 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 933 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 934                                 struct ext3_ext_path *path, unsigned long block,
 935                                 unsigned long *count, int *err)
 936 {
 937         unsigned long pblock, goal;
 938         int aflags = 0;
 939         struct inode *inode = ext3_ext_base2inode(base);
 940
 941         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 942         aflags |= 2; /* block have been already reserved */
 943         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 944         return pblock;
 945
 946 }
 947 #else
 948 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 949                                 struct ext3_ext_path *path, unsigned long block,
 950                                 unsigned long *count, int *err)
 951 {
 952         struct inode *inode = ext3_ext_base2inode(base);
 953         struct ext3_allocation_request ar;
 954         unsigned long pblock;
 955         int aflags;
 956
 957         /* find neighbour allocated blocks */
 958         ar.lleft = block;
 959         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 960         if (*err)
 961                 return 0;
 962         ar.lright = block;
 963         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 964         if (*err)
 965                 return 0;
 966
 967         /* allocate new block */
 968         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 969         ar.inode = inode;
 970         ar.logical = block;
 971         ar.len = *count;
 972         ar.flags = EXT3_MB_HINT_DATA;
 973         pblock = ext3_mb_new_blocks(handle, &ar, err);
 974         *count = ar.len;
 975         return pblock;
 976 }
 977 #endif
 978
 979 #ifdef EXT3_EXT_HAS_NO_TREE
 980 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 981                                   struct ext3_ext_path *path,
 982                                   struct ext3_ext_cache *cex,
 983 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 984                                    struct ext3_extent *ex,
 985 #endif
 986                                   void *cbdata)
 987 {
 988         struct bpointers *bp = cbdata;
 989 #else
 990 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 991                                   struct ext3_ext_path *path,
 992                                   struct ext3_ext_cache *cex
 993 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 994                                   , struct ext3_extent *ex
 995 #endif
 996                                  )
 997 {
 998         struct bpointers *bp = base->private;
 999 #endif
1000         struct inode *inode = ext3_ext_base2inode(base);
1001         struct ext3_extent nex;
1002         unsigned long pblock;
1003         unsigned long tgen;
1004         int err, i;
1005         unsigned long count;
1006         handle_t *handle;
1007
1008         i = EXT_DEPTH(base);
1009         EXT_ASSERT(i == path->p_depth);
1010         EXT_ASSERT(path[i].p_hdr);
1011
1012         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1013                 err = EXT_CONTINUE;
1014                 goto map;
1015         }
1016
1017         if (bp->create == 0) {
1018                 i = 0;
1019                 if (cex->ec_block < bp->start)
1020                         i = bp->start - cex->ec_block;
1021                 if (i >= cex->ec_len)
1022                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1023                                         i, cex->ec_len);
1024                 for (; i < cex->ec_len && bp->num; i++) {
1025                         *(bp->created) = 0;
1026                         bp->created++;
1027                         *(bp->blocks) = 0;
1028                         bp->blocks++;
1029                         bp->num--;
1030                         bp->start++;
1031                 }
1032
1033                 return EXT_CONTINUE;
1034         }
1035
1036         tgen = EXT_GENERATION(base);
1037         count = ext3_ext_calc_credits_for_insert(base, path);
1038         fsfilt_up_truncate_sem(inode);
1039
1040         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1041         if (IS_ERR(handle)) {
1042                 fsfilt_down_truncate_sem(inode);
1043                 return PTR_ERR(handle);
1044         }
1045
1046         fsfilt_down_truncate_sem(inode);
1047         if (tgen != EXT_GENERATION(base)) {
1048                 /* the tree has changed. so path can be invalid at moment */
1049                 ext3_journal_stop(handle);
1050                 return EXT_REPEAT;
1051         }
1052
1053         count = cex->ec_len;
1054         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1055         if (!pblock)
1056                 goto out;
1057         EXT_ASSERT(count <= cex->ec_len);
1058
1059         /* insert new extent */
1060         nex.ee_block = cpu_to_le32(cex->ec_block);
1061         ext3_ext_store_pblock(&nex, pblock);
1062         nex.ee_len = cpu_to_le16(count);
1063         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1064         if (err) {
1065                 /* free data blocks we just allocated */
1066                 /* not a good idea to call discard here directly,
1067                  * but otherwise we'd need to call it every free() */
1068 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1069                 ext3_mb_discard_inode_preallocations(inode);
1070 #endif
1071                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1072                                  cpu_to_le16(nex.ee_len), 0);
1073                 goto out;
1074         }
1075
1076         /*
1077          * Putting len of the actual extent we just inserted,
1078          * we are asking ext3_ext_walk_space() to continue
1079          * scaning after that block
1080          */
1081         cex->ec_len = le16_to_cpu(nex.ee_len);
1082         cex->ec_start = ext_pblock(&nex);
1083         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1084         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1085
1086 out:
1087         ext3_journal_stop(handle);
1088 map:
1089         if (err >= 0) {
1090                 /* map blocks */
1091                 if (bp->num == 0) {
1092                         CERROR("hmm. why do we find this extent?\n");
1093                         CERROR("initial space: %lu:%u\n",
1094                                 bp->start, bp->init_num);
1095                         CERROR("current extent: %u/%u/%llu %d\n",
1096                                 cex->ec_block, cex->ec_len,
1097                                 (unsigned long long)cex->ec_start,
1098                                 cex->ec_type);
1099                 }
1100                 i = 0;
1101                 if (cex->ec_block < bp->start)
1102                         i = bp->start - cex->ec_block;
1103                 if (i >= cex->ec_len)
1104                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1105                                         i, cex->ec_len);
1106                 for (; i < cex->ec_len && bp->num; i++) {
1107                         *(bp->blocks) = cex->ec_start + i;
1108                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1109                                 *(bp->created) = 0;
1110                         } else {
1111                                 *(bp->created) = 1;
1112                                 /* unmap any possible underlying metadata from
1113                                  * the block device mapping.  bug 6998. */
1114                                 ll_unmap_underlying_metadata(inode->i_sb,
1115                                                              *(bp->blocks));
1116                         }
1117                         bp->created++;
1118                         bp->blocks++;
1119                         bp->num--;
1120                         bp->start++;
1121                 }
1122         }
1123         return err;
1124 }
1125
1126 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1127                        unsigned long num, unsigned long *blocks,
1128                        int *created, int create)
1129 {
1130 #ifdef EXT3_EXT_HAS_NO_TREE
1131         struct ext3_ext_base *base = inode;
1132 #else
1133         struct ext3_extents_tree tree;
1134         struct ext3_ext_base *base = &tree;
1135 #endif
1136         struct bpointers bp;
1137         int err;
1138
1139         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1140                block, block + num - 1, (unsigned) inode->i_ino);
1141
1142 #ifndef EXT3_EXT_HAS_NO_TREE
1143         ext3_init_tree_desc(base, inode);
1144         tree.private = &bp;
1145 #endif
1146         bp.blocks = blocks;
1147         bp.created = created;
1148         bp.start = block;
1149         bp.init_num = bp.num = num;
1150         bp.create = create;
1151
1152         fsfilt_down_truncate_sem(inode);
1153         err = fsfilt_ext3_ext_walk_space(base, block, num,
1154                                          ext3_ext_new_extent_cb, &bp);
1155         ext3_ext_invalidate_cache(base);
1156         fsfilt_up_truncate_sem(inode);
1157
1158         return err;
1159 }
1160
1161 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1162                                     int pages, unsigned long *blocks,
1163                                     int *created, int create)
1164 {
1165         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1166         int rc = 0, i = 0;
1167         struct page *fp = NULL;
1168         int clen = 0;
1169
1170         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1171                 inode->i_ino, pages, (*page)->index);
1172
1173         /* pages are sorted already. so, we just have to find
1174          * contig. space and process them properly */
1175         while (i < pages) {
1176                 if (fp == NULL) {
1177                         /* start new extent */
1178                         fp = *page++;
1179                         clen = 1;
1180                         i++;
1181                         continue;
1182                 } else if (fp->index + clen == (*page)->index) {
1183                         /* continue the extent */
1184                         page++;
1185                         clen++;
1186                         i++;
1187                         continue;
1188                 }
1189
1190                 /* process found extent */
1191                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1192                                         clen * blocks_per_page, blocks,
1193                                         created, create);
1194                 if (rc)
1195                         GOTO(cleanup, rc);
1196
1197                 /* look for next extent */
1198                 fp = NULL;
1199                 blocks += blocks_per_page * clen;
1200                 created += blocks_per_page * clen;
1201         }
1202
1203         if (fp)
1204                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1205                                         clen * blocks_per_page, blocks,
1206                                         created, create);
1207 cleanup:
1208         return rc;
1209 }
1210
1211 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1212                                unsigned long *blocks, int *created, int create);
1213 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1214                                    int pages, unsigned long *blocks,
1215                                    int *created, int create)
1216 {
1217         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1218         unsigned long *b;
1219         int rc = 0, i, *cr;
1220
1221         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1222                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1223                 if (rc) {
1224                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1225                                inode->i_ino, *b, *cr, create, rc);
1226                         break;
1227                 }
1228
1229                 b += blocks_per_page;
1230                 cr += blocks_per_page;
1231         }
1232         return rc;
1233 }
1234
1235 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1236                                 int pages, unsigned long *blocks,
1237                                 int *created, int create,
1238                                 cfs_semaphore_t *optional_sem)
1239 {
1240         int rc;
1241
1242         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1243                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1244                                                      blocks, created, create);
1245                 return rc;
1246         }
1247         if (optional_sem != NULL)
1248                 cfs_down(optional_sem);
1249         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1250                                             created, create);
1251         if (optional_sem != NULL)
1252                 cfs_up(optional_sem);
1253
1254         return rc;
1255 }
1256
1257 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1258 {
1259         unsigned long block;
1260         struct buffer_head *bh;
1261         int err, blocksize, csize, boffs, osize = size;
1262
1263         /* prevent reading after eof */
1264         cfs_lock_kernel();
1265         if (i_size_read(inode) < *offs + size) {
1266                 size = i_size_read(inode) - *offs;
1267                 cfs_unlock_kernel();
1268                 if (size < 0) {
1269                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1270                                i_size_read(inode), *offs);
1271                         return -EBADR;
1272                 } else if (size == 0) {
1273                         return 0;
1274                 }
1275         } else {
1276                 cfs_unlock_kernel();
1277         }
1278
1279         blocksize = 1 << inode->i_blkbits;
1280
1281         while (size > 0) {
1282                 block = *offs >> inode->i_blkbits;
1283                 boffs = *offs & (blocksize - 1);
1284                 csize = min(blocksize - boffs, size);
1285                 bh = ext3_bread(NULL, inode, block, 0, &err);
1286                 if (!bh) {
1287                         CERROR("can't read block: %d\n", err);
1288                         return err;
1289                 }
1290
1291                 memcpy(buf, bh->b_data + boffs, csize);
1292                 brelse(bh);
1293
1294                 *offs += csize;
1295                 buf += csize;
1296                 size -= csize;
1297         }
1298         return osize;
1299 }
1300 EXPORT_SYMBOL(fsfilt_ext3_read);
1301
1302 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1303                                    int size, loff_t *offs)
1304 {
1305         int rc;
1306         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1307         if (rc > 0)
1308                 rc = 0;
1309         return rc;
1310 }
1311
1312 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1313                                 loff_t *offs, handle_t *handle)
1314 {
1315         struct buffer_head *bh = NULL;
1316         loff_t old_size = i_size_read(inode), offset = *offs;
1317         loff_t new_size = i_size_read(inode);
1318         unsigned long block;
1319         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1320
1321         while (bufsize > 0) {
1322                 if (bh != NULL)
1323                         brelse(bh);
1324
1325                 block = offset >> inode->i_blkbits;
1326                 boffs = offset & (blocksize - 1);
1327                 size = min(blocksize - boffs, bufsize);
1328                 bh = ext3_bread(handle, inode, block, 1, &err);
1329                 if (!bh) {
1330                         CERROR("can't read/create block: %d\n", err);
1331                         break;
1332                 }
1333
1334                 err = ext3_journal_get_write_access(handle, bh);
1335                 if (err) {
1336                         CERROR("journal_get_write_access() returned error %d\n",
1337                                err);
1338                         break;
1339                 }
1340                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1341                 memcpy(bh->b_data + boffs, buf, size);
1342                 err = ext3_journal_dirty_metadata(handle, bh);
1343                 if (err) {
1344                         CERROR("journal_dirty_metadata() returned error %d\n",
1345                                err);
1346                         break;
1347                 }
1348                 if (offset + size > new_size)
1349                         new_size = offset + size;
1350                 offset += size;
1351                 bufsize -= size;
1352                 buf += size;
1353         }
1354         if (bh)
1355                 brelse(bh);
1356
1357         /* correct in-core and on-disk sizes */
1358         if (new_size > i_size_read(inode)) {
1359                 cfs_lock_kernel();
1360                 if (new_size > i_size_read(inode))
1361                         i_size_write(inode, new_size);
1362                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1363                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1364                 if (i_size_read(inode) > old_size)
1365                         mark_inode_dirty(inode);
1366                 cfs_unlock_kernel();
1367         }
1368
1369         if (err == 0)
1370                 *offs = offset;
1371         return err;
1372 }
1373 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1374
1375 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1376                                     loff_t *offs, int force_sync)
1377 {
1378         struct inode *inode = file->f_dentry->d_inode;
1379         handle_t *handle;
1380         int err, block_count = 0, blocksize;
1381
1382         /* Determine how many transaction credits are needed */
1383         blocksize = 1 << inode->i_blkbits;
1384         block_count = (*offs & (blocksize - 1)) + bufsize;
1385         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1386
1387         handle = ext3_journal_start(inode,
1388                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1389         if (IS_ERR(handle)) {
1390                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1391                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1392                 return PTR_ERR(handle);
1393         }
1394
1395         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1396
1397         if (!err && force_sync)
1398                 handle->h_sync = 1; /* recovery likes this */
1399
1400         ext3_journal_stop(handle);
1401
1402         return err;
1403 }
1404
1405 static int fsfilt_ext3_setup(struct super_block *sb)
1406 {
1407 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1408      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1409         struct ext3_sb_info *sbi = EXT3_SB(sb);
1410 #if 0
1411         sbi->dx_lock = fsfilt_ext3_dx_lock;
1412         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1413 #endif
1414 #endif
1415 #ifdef S_PDIROPS
1416         CWARN("Enabling PDIROPS\n");
1417         set_opt(sbi->s_mount_opt, PDIROPS);
1418         sb->s_flags |= S_PDIROPS;
1419 #endif
1420         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1421                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1422 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1423         /* enable journaled quota support */
1424         /* kfreed in ext3_put_super() */
1425         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1426         if (!sbi->s_qf_names[USRQUOTA])
1427                 return -ENOMEM;
1428         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1429         if (!sbi->s_qf_names[GRPQUOTA]) {
1430                 kfree(sbi->s_qf_names[USRQUOTA]);
1431                 sbi->s_qf_names[USRQUOTA] = NULL;
1432                 return -ENOMEM;
1433         }
1434         sbi->s_jquota_fmt = QFMT_VFS_V0;
1435 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1436         set_opt(sbi->s_mount_opt, QUOTA);
1437 #endif
1438 #endif
1439         return 0;
1440 }
1441
1442 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1443    objects. Logs is number of logfiles to update */
1444 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1445 {
1446         if ( !fso ) {
1447                 switch(op) {
1448                 case FSFILT_OP_CREATE:
1449                                  /* directory leaf, index & indirect & EA*/
1450                         return 4 + 3 * logs;
1451                 case FSFILT_OP_UNLINK:
1452                         return 3 * logs;
1453                 }
1454         } else {
1455                 int i;
1456                 int needed = 0;
1457                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1458                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1459                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1460                 for (i = 0; i < op; i++, fso++) {
1461                         int nblocks = fso->fso_bufcnt * blockpp;
1462                         int ndindirect = min(nblocks, addrpp + 1);
1463                         int nindir = nblocks + ndindirect + 1;
1464
1465                         needed += nindir;
1466                 }
1467                 return needed + 3 * logs;
1468         }
1469
1470         return 0;
1471 }
1472
1473 #ifdef HAVE_QUOTA_SUPPORT
1474 #define DQINFO_COPY(out, in)                    \
1475 do {                                            \
1476         Q_COPY(out, in, dqi_bgrace);            \
1477         Q_COPY(out, in, dqi_igrace);            \
1478         Q_COPY(out, in, dqi_flags);             \
1479         Q_COPY(out, in, dqi_valid);             \
1480 } while (0)
1481
1482 #define DQBLK_COPY(out, in)                     \
1483 do {                                            \
1484         Q_COPY(out, in, dqb_bhardlimit);        \
1485         Q_COPY(out, in, dqb_bsoftlimit);        \
1486         Q_COPY(out, in, dqb_curspace);          \
1487         Q_COPY(out, in, dqb_ihardlimit);        \
1488         Q_COPY(out, in, dqb_isoftlimit);        \
1489         Q_COPY(out, in, dqb_curinodes);         \
1490         Q_COPY(out, in, dqb_btime);             \
1491         Q_COPY(out, in, dqb_itime);             \
1492         Q_COPY(out, in, dqb_valid);             \
1493 } while (0)
1494
1495 static int fsfilt_ext3_quotactl(struct super_block *sb,
1496                                 struct obd_quotactl *oqc)
1497 {
1498         int i, rc = 0, error = 0;
1499         const struct quotactl_ops *qcop;
1500         struct if_dqinfo *info;
1501         struct if_dqblk *dqblk;
1502         ENTRY;
1503
1504         if (!sb->s_qcop)
1505                 RETURN(-ENOSYS);
1506
1507         OBD_ALLOC_PTR(info);
1508         if (!info)
1509                 RETURN(-ENOMEM);
1510         OBD_ALLOC_PTR(dqblk);
1511         if (!dqblk) {
1512                 OBD_FREE_PTR(info);
1513                 RETURN(-ENOMEM);
1514         }
1515
1516         DQINFO_COPY(info, &oqc->qc_dqinfo);
1517         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1518
1519         qcop = sb->s_qcop;
1520         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1521                 for (i = 0; i < MAXQUOTAS; i++) {
1522                         if (!Q_TYPESET(oqc, i))
1523                                 continue;
1524
1525                         if (oqc->qc_cmd == Q_QUOTAON) {
1526                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1527
1528                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1529
1530                                 rc = ll_quota_on(sb, i, QFMT_VFS_V0,
1531                                                  name[i], 0);
1532                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1533                                 rc = ll_quota_off(sb, i, 0);
1534                         }
1535
1536                         if (rc == -EBUSY)
1537                                 error = rc;
1538                         else if (rc)
1539                                 GOTO(out, rc);
1540                 }
1541                 GOTO(out, rc ?: error);
1542         }
1543
1544         switch (oqc->qc_cmd) {
1545         case Q_GETOINFO:
1546         case Q_GETINFO:
1547                 if (!qcop->get_info)
1548                         GOTO(out, rc = -ENOSYS);
1549                 rc = qcop->get_info(sb, oqc->qc_type, info);
1550                 break;
1551         case Q_SETQUOTA:
1552         case Q_INITQUOTA:
1553                 if (!qcop->set_dqblk)
1554                         GOTO(out, rc = -ENOSYS);
1555                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1556                 break;
1557         case Q_GETOQUOTA:
1558         case Q_GETQUOTA:
1559                 if (!qcop->get_dqblk)
1560                         GOTO(out, rc = -ENOSYS);
1561                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1562                 if (!rc)
1563                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1564                 break;
1565         case Q_SYNC:
1566                 if (!sb->s_qcop->quota_sync)
1567                         GOTO(out, rc = -ENOSYS);
1568                 qcop->quota_sync(sb, oqc->qc_type);
1569                 break;
1570         case Q_FINVALIDATE:
1571                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1572                 for (i = 0; i < MAXQUOTAS; i++) {
1573                         struct file *fp;
1574                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1575
1576                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1577
1578                         if (!Q_TYPESET(oqc, i))
1579                                 continue;
1580
1581                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1582                         if (IS_ERR(fp)) {
1583                                 rc = PTR_ERR(fp);
1584                                 CERROR("error invalidating operational quota file"
1585                                        " %s (rc:%d)\n", name[i], rc);
1586                         } else {
1587                                 filp_close(fp, 0);
1588                         }
1589
1590                 }
1591                 break;
1592         default:
1593                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1594                 LBUG();
1595         }
1596 out:
1597         DQINFO_COPY(&oqc->qc_dqinfo, info);
1598         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1599
1600         OBD_FREE_PTR(info);
1601         OBD_FREE_PTR(dqblk);
1602
1603         if (rc)
1604                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1605                                 "failed: %d\n",
1606                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1607         RETURN(rc);
1608 }
1609
1610 struct chk_dqblk{
1611         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1612         cfs_list_t              dqb_list;        /** in list also */
1613         qid_t                   dqb_id;          /** uid/gid */
1614         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1615         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1616         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1617         qsize_t                 dqb_curspace;    /** current space */
1618         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1619         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1620         qsize_t                 dqb_curinodes;   /** current inodes */
1621         obd_time                dqb_btime;       /** block grace time */
1622         obd_time                dqb_itime;       /** inode grace time */
1623         __u32                   dqb_valid;       /** flag for above fields */
1624 };
1625
1626 static inline unsigned int chkquot_hash(qid_t id, int type)
1627                                         __attribute__((__const__));
1628
1629 static inline unsigned int chkquot_hash(qid_t id, int type)
1630 {
1631         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1632 }
1633
1634 static inline struct chk_dqblk *
1635 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1636 {
1637         cfs_hlist_node_t *node;
1638         struct chk_dqblk *cdqb;
1639
1640         cfs_hlist_for_each(node, head) {
1641                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1642                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1643                         return cdqb;
1644         }
1645
1646         return NULL;
1647 }
1648
1649 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1650 {
1651         struct chk_dqblk *cdqb;
1652
1653         OBD_ALLOC_PTR(cdqb);
1654         if (cdqb) {
1655                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1656                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1657                 cdqb->dqb_id = id;
1658                 cdqb->dqb_type = type;
1659         }
1660
1661         return cdqb;
1662 }
1663
1664 static struct chk_dqblk *
1665 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1666       cfs_list_t *list, qid_t id, int type, int first_check)
1667 {
1668         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1669         struct if_dqblk dqb;
1670         struct chk_dqblk *cdqb;
1671         int rc;
1672
1673         cdqb = find_chkquot(head, id, type);
1674         if (cdqb)
1675                 return cdqb;
1676
1677         cdqb = alloc_chkquot(id, type);
1678         if (!cdqb)
1679                 return NULL;
1680
1681         if (!first_check) {
1682                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1683                 if (rc) {
1684                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1685                                id, type, rc);
1686                 } else {
1687                         DQBLK_COPY(cdqb, &dqb);
1688                         cdqb->dqb_curspace = 0;
1689                         cdqb->dqb_curinodes = 0;
1690                 }
1691         }
1692
1693         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1694         cfs_list_add_tail(&cdqb->dqb_list, list);
1695
1696         return cdqb;
1697 }
1698
1699 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1700 {
1701         struct obd_quotactl *oqctl;
1702         int rc;
1703
1704         OBD_ALLOC_PTR(oqctl);
1705         if (!oqctl)
1706                 RETURN(-ENOMEM);
1707
1708         oqctl->qc_cmd = cmd;
1709         oqctl->qc_id = qfmt;
1710         oqctl->qc_type = type;
1711         rc = fsfilt_ext3_quotactl(sb, oqctl);
1712
1713         OBD_FREE_PTR(oqctl);
1714         return rc;
1715 }
1716
1717 static inline int read_old_dqinfo(struct super_block *sb, int type,
1718                                   struct if_dqinfo *dqinfo)
1719 {
1720         struct obd_quotactl *oqctl;
1721         int rc;
1722         ENTRY;
1723
1724         OBD_ALLOC_PTR(oqctl);
1725         if (!oqctl)
1726                 RETURN(-ENOMEM);
1727
1728         oqctl->qc_cmd = Q_GETINFO;
1729         oqctl->qc_type = type;
1730         rc = fsfilt_ext3_quotactl(sb, oqctl);
1731         if (!rc)
1732                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1733
1734         OBD_FREE_PTR(oqctl);
1735         RETURN(rc);
1736 }
1737
1738 #ifndef HAVE_EXT4_LDISKFS
1739 static inline struct ext3_group_desc *
1740 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1741 {
1742         unsigned long desc_block, desc;
1743         struct ext3_group_desc *gdp;
1744
1745         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1746         desc = group % EXT3_DESC_PER_BLOCK(sb);
1747         gdp = (struct ext3_group_desc *)
1748               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1749
1750         return gdp + desc;
1751 }
1752
1753 static inline struct buffer_head *
1754 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1755 {
1756         struct ext3_group_desc *desc;
1757         struct buffer_head *bh;
1758
1759         desc = get_group_desc(sb, group, NULL);
1760         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1761         return bh;
1762 }
1763
1764 static __u32 ext3_itable_unused_count(struct super_block *sb,
1765                                struct ext3_group_desc *bg) {
1766        return le16_to_cpu(bg->bg_itable_unused);
1767 }
1768 #else
1769 #define get_group_desc ext3_get_group_desc
1770 #endif
1771
1772 struct qchk_ctxt {
1773         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1774         cfs_list_t              qckt_list;                 /* quotacheck list */
1775         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1776         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1777 };
1778
1779 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1780                            struct obd_quotactl *oqc)
1781 {
1782         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1783         loff_t size = 0;
1784         qid_t qid[MAXQUOTAS];
1785         int cnt, i, rc = 0;
1786
1787         if (!inode)
1788                 return 0;
1789
1790         qid[USRQUOTA] = inode->i_uid;
1791         qid[GRPQUOTA] = inode->i_gid;
1792
1793         if (S_ISDIR(inode->i_mode) ||
1794             S_ISREG(inode->i_mode) ||
1795             S_ISLNK(inode->i_mode))
1796                 size = inode_get_bytes(inode);
1797
1798         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1799                 if (!Q_TYPESET(oqc, cnt))
1800                         continue;
1801
1802                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1803                                 &qctxt->qckt_list, qid[cnt], cnt,
1804                                 qctxt->qckt_first_check[cnt]);
1805                 if (!cdqb[cnt]) {
1806                         rc = -ENOMEM;
1807                         break;
1808                 }
1809
1810                 cdqb[cnt]->dqb_curspace += size;
1811                 cdqb[cnt]->dqb_curinodes++;
1812         }
1813
1814         if (rc) {
1815                 for (i = 0; i < cnt; i++) {
1816                         if (!Q_TYPESET(oqc, i))
1817                                 continue;
1818                         LASSERT(cdqb[i]);
1819                         cdqb[i]->dqb_curspace -= size;
1820                         cdqb[i]->dqb_curinodes--;
1821                 }
1822         }
1823
1824         return rc;
1825 }
1826
1827 /* write dqinfo struct in a new quota file */
1828 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1829 {
1830         struct v2_disk_dqinfo dqinfo;
1831         __u32 blocks = V2_DQTREEOFF + 1;
1832         loff_t offset = V2_DQINFOOFF;
1833
1834         if (info) {
1835                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1836                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1837                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1838                                                ~DQF_INFO_DIRTY);
1839         } else {
1840                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1841                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1842                 dqinfo.dqi_flags = 0;
1843         }
1844
1845         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1846         dqinfo.dqi_free_blk = 0;
1847         dqinfo.dqi_free_entry = 0;
1848
1849         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1850 }
1851
1852 static int v3_write_dqheader(struct file *f, int type)
1853 {
1854         static const __u32 quota_magics[] = V2_INITQMAGICS;
1855         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1856         struct v2_disk_dqheader dqhead;
1857         loff_t offset = 0;
1858
1859         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1860         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1861
1862         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1863         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1864
1865         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1866 }
1867
1868 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1869                                   struct obd_quotactl *oqc)
1870 {
1871         int i, rc = 0;
1872         ENTRY;
1873
1874         for (i = 0; i < MAXQUOTAS; i++) {
1875                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1876                                          NULL : &qctxt->qckt_dqinfo[i];
1877                 struct file *file;
1878                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1879
1880                 if (!Q_TYPESET(oqc, i))
1881                         continue;
1882
1883                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1884
1885                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1886                 if (IS_ERR(file)) {
1887                         rc = PTR_ERR(file);
1888                         CERROR("can't create %s file: rc = %d\n",
1889                                name[i], rc);
1890                         GOTO(out, rc);
1891                 }
1892
1893                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1894                         CERROR("file %s is not regular", name[i]);
1895                         filp_close(file, 0);
1896                         GOTO(out, rc = -EINVAL);
1897                 }
1898
1899                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1900
1901                 rc = v3_write_dqheader(file, i);
1902                 if (rc) {
1903                         filp_close(file, 0);
1904                         GOTO(out, rc);
1905                 }
1906
1907                 rc = v3_write_dqinfo(file, i, info);
1908                 filp_close(file, 0);
1909                 if (rc)
1910                         GOTO(out, rc);
1911         }
1912
1913 out:
1914         RETURN(rc);
1915 }
1916
1917
1918 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1919                           struct chk_dqblk *cdqb)
1920 {
1921         struct obd_quotactl *oqc;
1922         long now;
1923         int rc;
1924         ENTRY;
1925
1926         OBD_ALLOC_PTR(oqc);
1927         if (!oqc)
1928                 RETURN(-ENOMEM);
1929
1930         now = cfs_time_current_sec();
1931
1932         if (cdqb->dqb_bsoftlimit &&
1933             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1934             !cdqb->dqb_btime)
1935                 cdqb->dqb_btime =
1936                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1937
1938         if (cdqb->dqb_isoftlimit &&
1939             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1940             !cdqb->dqb_itime)
1941                 cdqb->dqb_itime =
1942                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1943
1944         cdqb->dqb_valid = QIF_ALL;
1945
1946         oqc->qc_cmd = Q_SETQUOTA;
1947         oqc->qc_type = cdqb->dqb_type;
1948         oqc->qc_id = cdqb->dqb_id;
1949         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1950
1951         rc = fsfilt_ext3_quotactl(sb, oqc);
1952         OBD_FREE_PTR(oqc);
1953         RETURN(rc);
1954 }
1955
1956 static int prune_chkquots(struct super_block *sb,
1957                           struct qchk_ctxt *qctxt, int error)
1958 {
1959         struct chk_dqblk *cdqb, *tmp;
1960         int rc;
1961
1962         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1963                 if (!error) {
1964                         rc = commit_chkquot(sb, qctxt, cdqb);
1965                         if (rc)
1966                                 error = rc;
1967                 }
1968                 cfs_hlist_del_init(&cdqb->dqb_hash);
1969                 cfs_list_del(&cdqb->dqb_list);
1970                 OBD_FREE_PTR(cdqb);
1971         }
1972
1973         return error;
1974 }
1975
1976 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1977 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1978 #endif
1979
1980 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1981                                   struct obd_quotactl *oqc)
1982 {
1983         struct ext3_sb_info *sbi = EXT3_SB(sb);
1984         int i, group, uninit_feat = 0;
1985         struct qchk_ctxt *qctxt;
1986         struct buffer_head *bitmap_bh = NULL;
1987         unsigned long ino, inode_inuse;
1988         struct inode *inode;
1989         int rc = 0;
1990         ENTRY;
1991
1992         /* turn on quota and read dqinfo if existed */
1993         OBD_ALLOC_PTR(qctxt);
1994         if (!qctxt) {
1995                 oqc->qc_stat = -ENOMEM;
1996                 RETURN(-ENOMEM);
1997         }
1998
1999         for (i = 0; i < NR_DQHASH; i++)
2000                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2001         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2002
2003         for (i = 0; i < MAXQUOTAS; i++) {
2004                 if (!Q_TYPESET(oqc, i))
2005                         continue;
2006
2007                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2008                 if (!rc || rc == -EBUSY) {
2009                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2010                         if (rc)
2011                                 GOTO(out, rc);
2012                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2013                         qctxt->qckt_first_check[i] = 1;
2014                 } else if (rc) {
2015                         GOTO(out, rc);
2016                 }
2017         }
2018         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2019                 /* This filesystem supports the uninit group feature */
2020                 uninit_feat = 1;
2021
2022         /* number of inodes that have been allocated */
2023         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2024                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2025
2026         /* check quota and update in hash */
2027         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2028              group++) {
2029                 unsigned long used_count = sbi->s_inodes_per_group;
2030
2031                 if (uninit_feat) {
2032                         struct ext3_group_desc *desc;
2033                         desc = get_group_desc(sb, group, NULL);
2034                         if (!desc)
2035                                 GOTO(out, -EIO);
2036
2037                         /* we don't really need to take the group lock here,
2038                          * but it may be useful if one day we support online
2039                          * quotacheck */
2040 #ifdef HAVE_EXT4_LDISKFS
2041                         ext4_lock_group(sb, group);
2042 #else
2043                         spin_lock(sb_bgl_lock(sbi, group));
2044 #endif
2045                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2046                                 /* no inode in use in this group, just skip it */
2047 #ifdef HAVE_EXT4_LDISKFS
2048                                 ext3_unlock_group(sb, group);
2049 #else
2050                                 spin_unlock(sb_bgl_lock(sbi, group));
2051 #endif
2052                                 continue;
2053                         }
2054
2055                         used_count -= ext3_itable_unused_count(sb, desc);
2056 #ifdef HAVE_EXT4_LDISKFS
2057                         ext3_unlock_group(sb, group);
2058 #else
2059                         spin_unlock(sb_bgl_lock(sbi, group));
2060 #endif
2061                 }
2062
2063                 ino = group * sbi->s_inodes_per_group + 1;
2064                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2065                 if (!bitmap_bh) {
2066                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2067                                sb->s_id, group);
2068                         GOTO(out, -EIO);
2069                 }
2070
2071                 i = 0;
2072                 while (i < used_count &&
2073                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2074                                                used_count, i)) < used_count) {
2075                         inode_inuse--;
2076                         i++;
2077                         ino = i + group * sbi->s_inodes_per_group;
2078                         if (ino < sbi->s_first_ino)
2079                                 continue;
2080 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2081                         inode = ext3_iget(sb, ino);
2082 #else
2083                         inode = iget(sb, ino);
2084 #endif
2085                         if (!inode || IS_ERR(inode))
2086                                 continue;
2087
2088                         rc = add_inode_quota(inode, qctxt, oqc);
2089                         iput(inode);
2090                         if (rc) {
2091                                 brelse(bitmap_bh);
2092                                 GOTO(out, rc);
2093                         }
2094                 }
2095                 brelse(bitmap_bh);
2096         }
2097
2098         /* read old quota limits from old quota file. (only for the user
2099          * has limits but hasn't file) */
2100 #ifdef HAVE_QUOTA_SUPPORT
2101         for (i = 0; i < MAXQUOTAS; i++) {
2102                 cfs_list_t id_list;
2103                 struct dquot_id *dqid, *tmp;
2104
2105                 if (!Q_TYPESET(oqc, i))
2106                         continue;
2107
2108                 if (qctxt->qckt_first_check[i])
2109                         continue;
2110
2111
2112                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2113                 CFS_INIT_LIST_HEAD(&id_list);
2114 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2115                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2116 #else
2117                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2118 #endif
2119                 if (rc)
2120                         CERROR("read old limits failed. (rc:%d)\n", rc);
2121
2122                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2123                         cfs_list_del_init(&dqid->di_link);
2124
2125                         if (!rc)
2126                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2127                                       dqid->di_id, i,
2128                                       qctxt->qckt_first_check[i]);
2129                         OBD_FREE_PTR(dqid);
2130                 }
2131         }
2132 #endif
2133         /* turn off quota cause we are to dump chk_dqblk to files */
2134         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2135
2136         rc = create_new_quota_files(qctxt, oqc);
2137         if (rc)
2138                 GOTO(out, rc);
2139
2140         /* we use vfs functions to set dqblk, so turn quota on */
2141         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2142 out:
2143         /* dump and free chk_dqblk */
2144         rc = prune_chkquots(sb, qctxt, rc);
2145         OBD_FREE_PTR(qctxt);
2146
2147         /* turn off quota, `lfs quotacheck` will turn on when all
2148          * nodes quotacheck finish. */
2149         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2150
2151         oqc->qc_stat = rc;
2152         if (rc)
2153                 CERROR("quotacheck failed: rc = %d\n", rc);
2154
2155         RETURN(rc);
2156 }
2157
2158 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2159                                  int cmd)
2160 {
2161         int rc = 0;
2162         ENTRY;
2163
2164         if (lqi->qi_files[type] == NULL) {
2165                 CERROR("operate qinfo before it's enabled!\n");
2166                 RETURN(-ESRCH);
2167         }
2168
2169         switch (cmd) {
2170         case QFILE_CHK:
2171                 rc = lustre_check_quota_file(lqi, type);
2172                 break;
2173         case QFILE_RD_INFO:
2174                 rc = lustre_read_quota_info(lqi, type);
2175                 break;
2176         case QFILE_WR_INFO:
2177                 rc = lustre_write_quota_info(lqi, type);
2178                 break;
2179         case QFILE_INIT_INFO:
2180                 rc = lustre_init_quota_info(lqi, type);
2181                 break;
2182         case QFILE_CONVERT:
2183                 rc = -ENOTSUPP;
2184                 CERROR("quota CONVERT command is not supported\n");
2185                 break;
2186         default:
2187                 rc = -ENOTSUPP;
2188                 CERROR("Unsupported admin quota file cmd %d\n"
2189                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2190                        cmd);
2191                 break;
2192         }
2193         RETURN(rc);
2194 }
2195
2196 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2197                             cfs_list_t *list)
2198 {
2199         return lustre_get_qids(file, inode, type, list);
2200 }
2201
2202 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2203 {
2204         int rc = 0;
2205         ENTRY;
2206
2207         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2208                 CERROR("operate dquot before it's enabled!\n");
2209                 RETURN(-ESRCH);
2210         }
2211
2212         switch (cmd) {
2213         case QFILE_RD_DQUOT:
2214                 rc = lustre_read_dquot(dquot);
2215                 break;
2216         case QFILE_WR_DQUOT:
2217                 if (dquot->dq_dqb.dqb_ihardlimit ||
2218                     dquot->dq_dqb.dqb_isoftlimit ||
2219                     dquot->dq_dqb.dqb_bhardlimit ||
2220                     dquot->dq_dqb.dqb_bsoftlimit)
2221                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2222                 else
2223                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2224
2225                 rc = lustre_commit_dquot(dquot);
2226                 if (rc >= 0)
2227                         rc = 0;
2228                 break;
2229         default:
2230                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2231                 LBUG();
2232                 break;
2233         }
2234         RETURN(rc);
2235 }
2236
2237 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2238                                 struct inode *inode, int frags)
2239 {
2240 #ifdef EXT3_EXT_HAS_NO_TREE
2241         struct ext3_ext_base *base = inode;
2242 #else
2243         struct ext3_extents_tree tree;
2244         struct ext3_ext_base *base = &tree;
2245
2246         ext3_init_tree_desc(base, inode);
2247 #endif
2248         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2249          * metablocks at maxium b=16542 */
2250         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2251         return 0;
2252 }
2253
2254 #endif
2255
2256 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2257 {
2258         return (EXT3_SB(sb)->journal_bdev);
2259 }
2260 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2261
2262 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2263                           char *buf, int count, loff_t pos)
2264 {
2265         loff_t p = pos;
2266         int rc;
2267
2268         if (!f && !inode) {
2269                 CERROR("lustre_read_quota failed for no quota file!\n");
2270                 libcfs_debug_dumpstack(NULL);
2271                 return -EINVAL;
2272         }
2273
2274         /* Support for both adm and op quota files must be provided */
2275         if (f) {
2276                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2277                 rc = rc < 0 ? rc : p - pos;
2278         } else {
2279                 struct super_block *sb = inode->i_sb;
2280                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2281         }
2282         return rc;
2283 }
2284
2285 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2286 {
2287         loff_t p = pos;
2288         int rc;
2289
2290         /* Only adm quota files are supported, op updates are handled by vfs */
2291         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2292         rc = rc < 0 ? rc : p - pos;
2293
2294         return rc;
2295 }
2296
2297 void *lustre_quota_journal_start(struct inode *inode, int delete)
2298 {
2299         handle_t *handle;
2300         unsigned block_count;
2301
2302         if (delete) {
2303                 /* each indirect block (+4) may become free, attaching to the
2304                  * header list of free blocks (+1); the data block (+1) may
2305                  * become a free block (+0) or a block with free dqentries (+0) */
2306                 block_count = (4 + 1) + 1;
2307                 handle = ext3_journal_start(inode,
2308                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2309         } else {
2310                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2311                  * freeblk reusage with a header update (+1); dqentry is either reused
2312                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2313                  * a new block allocation (+1) with a header update (+1)              */
2314                 block_count = (4 + 1) + 3;
2315                 handle = ext3_journal_start(inode,
2316                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2317
2318         }
2319
2320         return handle;
2321 }
2322
2323 void lustre_quota_journal_stop(void *handle)
2324 {
2325         ext3_journal_stop((handle_t *)handle);
2326 }
2327
2328 static int ll_decode_fh_accept(void *context, struct dentry *de)
2329 {
2330         return 1;
2331 }
2332
2333 #ifdef HAVE_EXPORTFS_DECODE_FH
2334 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2335          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2336                             acceptable, context)
2337 #else
2338 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2339          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2340                                      type, acceptable, context)
2341 # define FILEID_INO32_GEN 1
2342 extern struct export_operations export_op_default;
2343 #endif
2344
2345 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2346                                       struct fsfilt_fid *fid, int ignore_gen)
2347 {
2348         struct inode  *inode;
2349         struct dentry *result;
2350
2351         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2352                                        ll_decode_fh_accept, NULL);
2353         if (IS_ERR(result)) {
2354                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2355                        fid->ino, fid->gen, PTR_ERR(result));
2356                 return result;
2357         }
2358
2359         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2360                fid->ino, fid->gen);
2361         inode = result->d_inode;
2362         if (inode == NULL)
2363                 goto err_out;
2364
2365         if (inode->i_nlink == 0 &&
2366             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2367                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2368                               " ctime -- this may indicate disk "
2369                               "corruption (inode: %lu, link: %lu, "
2370                               "count: %d)\n", inode->i_ino,
2371                               (unsigned long)inode->i_nlink,
2372                               atomic_read(&inode->i_count));
2373                 goto err_out;
2374         }
2375         if (fid->gen && inode->i_generation != fid->gen) {
2376                 /* we didn't find the right inode.. */
2377                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2378                        "count: %d, generation %u/%u\n",
2379                        inode->i_ino, (unsigned long)inode->i_nlink,
2380                        atomic_read(&inode->i_count), inode->i_generation,
2381                        fid->gen);
2382                 goto err_out;
2383         }
2384
2385         return result;
2386 err_out:
2387         l_dput(result);
2388         return ERR_PTR(-ENOENT);
2389 }
2390
2391 static struct fsfilt_operations fsfilt_ext3_ops = {
2392         .fs_type                = "ext3",
2393         .fs_owner               = THIS_MODULE,
2394         .fs_getlabel            = fsfilt_ext3_get_label,
2395         .fs_setlabel            = fsfilt_ext3_set_label,
2396         .fs_uuid                = fsfilt_ext3_uuid,
2397         .fs_start               = fsfilt_ext3_start,
2398         .fs_brw_start           = fsfilt_ext3_brw_start,
2399         .fs_extend              = fsfilt_ext3_extend,
2400         .fs_commit              = fsfilt_ext3_commit,
2401         .fs_commit_async        = fsfilt_ext3_commit_async,
2402         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2403         .fs_setattr             = fsfilt_ext3_setattr,
2404         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2405         .fs_set_md              = fsfilt_ext3_set_md,
2406         .fs_get_md              = fsfilt_ext3_get_md,
2407         .fs_readpage            = fsfilt_ext3_readpage,
2408         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2409         .fs_statfs              = fsfilt_ext3_statfs,
2410         .fs_sync                = fsfilt_ext3_sync,
2411         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2412         .fs_write_record        = fsfilt_ext3_write_record,
2413         .fs_read_record         = fsfilt_ext3_read_record,
2414         .fs_setup               = fsfilt_ext3_setup,
2415         .fs_send_bio            = fsfilt_ext3_send_bio,
2416         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2417 #ifdef HAVE_DISK_INODE_VERSION
2418         .fs_get_version         = fsfilt_ext3_get_version,
2419         .fs_set_version         = fsfilt_ext3_set_version,
2420 #endif
2421 #ifdef HAVE_QUOTA_SUPPORT
2422         .fs_quotactl            = fsfilt_ext3_quotactl,
2423         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2424         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2425         .fs_qids                = fsfilt_ext3_qids,
2426         .fs_dquot               = fsfilt_ext3_dquot,
2427         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2428 #endif
2429         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2430         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2431 };
2432
2433 static int __init fsfilt_ext3_init(void)
2434 {
2435         int rc;
2436
2437         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2438                                          sizeof(struct fsfilt_cb_data), 0, 0);
2439         if (!fcb_cache) {
2440                 CERROR("error allocating fsfilt journal callback cache\n");
2441                 GOTO(out, rc = -ENOMEM);
2442         }
2443
2444         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2445
2446         if (rc) {
2447                 int err = cfs_mem_cache_destroy(fcb_cache);
2448                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2449         }
2450 out:
2451         return rc;
2452 }
2453
2454 static void __exit fsfilt_ext3_exit(void)
2455 {
2456         int rc;
2457
2458         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2459         rc = cfs_mem_cache_destroy(fcb_cache);
2460         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2461 }
2462
2463 module_init(fsfilt_ext3_init);
2464 module_exit(fsfilt_ext3_exit);
2465
2466 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2467 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2468 MODULE_LICENSE("GPL");