lustre/lvfs/fsfilt_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/fsfilt_ext3.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/init.h>
  44 #include <linux/module.h>
  45 #include <linux/fs.h>
  46 #include <linux/slab.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/quotaops.h>
  49 #ifdef HAVE_LINUX_EXPORTFS_H
  50 #include <linux/exportfs.h>
  51 #endif
  52 #ifdef HAVE_EXT4_LDISKFS
  53 #include <ext4/ext4.h>
  54 #include <ext4/ext4_jbd2.h>
  55 #else
  56 #include <linux/jbd.h>
  57 #include <linux/ext3_fs.h>
  58 #include <linux/ext3_jbd.h>
  59 #endif
  60 #include <linux/version.h>
  61 #include <linux/bitops.h>
  62 #include <linux/quota.h>
  63 #ifdef HAVE_QUOTAIO_H
  64 # include <linux/quotaio_v2.h>
  65 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  66 # include <quota/quotaio_v2.h>
  67 # include <quota/quota_tree.h>
  68 # define V2_DQTREEOFF    QT_TREEOFF
  69 #elif defined(HAVE_FS_QUOTAIO_V1_H)
  70 # include <quotaio_v2.h>
  71 # include <quota_tree.h>
  72 # define V2_DQTREEOFF    QT_TREEOFF
  73 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  74 #endif
  75
  76 #ifdef QFMT_VFS_V1
  77 #define QFMT_LUSTRE QFMT_VFS_V1
  78 #else
  79 #define QFMT_LUSTRE QFMT_VFS_V0
  80 #endif
  81
  82 #if defined(HAVE_EXT3_XATTR_H)
  83 #include <ext3/xattr.h>
  84 #else
  85 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  86 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  87 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  88 #endif
  89
  90 #include <libcfs/libcfs.h>
  91 #include <lustre_fsfilt.h>
  92 #include <obd.h>
  93 #include <lustre_quota.h>
  94 #include <linux/lustre_compat25.h>
  95 #include <linux/lprocfs_status.h>
  96
  97 #ifdef HAVE_EXT4_LDISKFS
  98 #include <ext4/ext4_extents.h>
  99 #else
 100 #include <linux/ext3_extents.h>
 101 #endif
 102
 103 #include "lustre_quota_fmt.h"
 104
 105 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
 106 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 107 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 108 #else
 109 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 110 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 111 #endif
 112
 113 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 114 /* for kernels 2.6.18 and later */
 115 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 116 #else
 117 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 118 #endif
 119
 120 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 121 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 122                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 123 #else
 124 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 125                ext3_ext_insert_extent(handle, inode, path, newext)
 126 #endif
 127
 128 #ifdef EXT3_DISCARD_PREALLOCATIONS
 129 #define ext3_mb_discard_inode_preallocations(inode) \
 130                  ext3_discard_preallocations(inode)
 131 #endif
 132
 133
 134 static cfs_mem_cache_t *fcb_cache;
 135
 136 struct fsfilt_cb_data {
 137         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 138         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 139         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 140         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 141         void *cb_data;                  /* MDS/OST completion function data */
 142 };
 143
 144 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 145 #define EXT3_XATTR_INDEX_TRUSTED        4
 146 #endif
 147
 148 #ifdef HAVE_EXT4_LDISKFS
 149 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 150 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 151 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 152 #else
 153 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 154 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 155 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 156 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 157 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 158 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 159 #endif
 160
 161 #ifndef ext3_find_next_bit
 162 #define ext3_find_next_bit           ext2_find_next_bit
 163 #endif
 164
 165 #ifndef ext2_find_next_bit
 166 #ifdef __LITTLE_ENDIAN
 167 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 168 #else
 169 error "Need implementation of find_next_bit on big-endian systems"
 170 #endif  /* __LITTLE_ENDIAN */
 171 #endif  /* !ext2_find_next_le_bit */
 172
 173 static char *fsfilt_ext3_get_label(struct super_block *sb)
 174 {
 175         return EXT3_SB(sb)->s_es->s_volume_name;
 176 }
 177
 178 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 179 {
 180         /* see e.g. fsfilt_ext3_write_record() */
 181         journal_t *journal;
 182         handle_t *handle;
 183         int err;
 184
 185         journal = EXT3_SB(sb)->s_journal;
 186         handle = ext3_journal_start_sb(sb, 1);
 187         if (IS_ERR(handle)) {
 188                 CERROR("can't start transaction\n");
 189                 return(PTR_ERR(handle));
 190         }
 191
 192         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 193         if (err)
 194                 goto out;
 195
 196         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 197                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 198
 199         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 200
 201 out:
 202         ext3_journal_stop(handle);
 203
 204         return(err);
 205 }
 206
 207 static char *fsfilt_ext3_uuid(struct super_block *sb)
 208 {
 209         return EXT3_SB(sb)->s_es->s_uuid;
 210 }
 211
 212 #ifdef HAVE_DISK_INODE_VERSION
 213
 214 static __u64 get_i_version(struct inode *inode)
 215 {
 216 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 217         return inode->i_version;
 218 #else
 219         return EXT3_I(inode)->i_fs_version;
 220 #endif
 221 }
 222
 223 static void set_i_version(struct inode *inode, __u64 new_version)
 224 {
 225 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 226         inode->i_version = new_version;
 227 #else
 228         (EXT3_I(inode))->i_fs_version = new_version;
 229 #endif
 230 }
 231
 232 /*
 233  * Get the 64-bit version for an inode.
 234  */
 235 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 236 {
 237         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 238                get_i_version(inode), inode->i_ino);
 239         return get_i_version(inode);
 240 }
 241
 242 /*
 243  * Set the 64-bit version and return the old version.
 244  */
 245 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 246 {
 247         __u64 old_version = get_i_version(inode);
 248
 249         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 250                new_version, old_version, inode->i_ino);
 251         set_i_version(inode, new_version);
 252         /* version is set after all inode operations are finished, so we should
 253          * mark it dirty here */
 254         inode->i_sb->s_op->dirty_inode(inode);
 255         return old_version;
 256 }
 257
 258 #endif
 259
 260 /*
 261  * We don't currently need any additional blocks for rmdir and
 262  * unlink transactions because we are storing the OST oa_id inside
 263  * the inode (which we will be changing anyways as part of this
 264  * transaction).
 265  */
 266 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 267                                int logs)
 268 {
 269         /* For updates to the last received file */
 270         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 271         journal_t *journal;
 272         void *handle;
 273
 274         if (current->journal_info) {
 275                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 276                        current->journal_info);
 277                 goto journal_start;
 278         }
 279
 280         switch(op) {
 281         case FSFILT_OP_RMDIR:
 282         case FSFILT_OP_UNLINK:
 283                 /* delete one file + create/update logs for each stripe */
 284                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 285                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 286                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 287                 break;
 288         case FSFILT_OP_RENAME:
 289                 /* modify additional directory */
 290                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 291                 /* no break */
 292         case FSFILT_OP_SYMLINK:
 293                 /* additional block + block bitmap + GDT for long symlink */
 294                 nblocks += 3;
 295                 /* no break */
 296         case FSFILT_OP_CREATE: {
 297 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 298                 static int warned;
 299                 if (!warned) {
 300                         if (!test_opt(inode->i_sb, EXTENTS)) {
 301                                 warned = 1;
 302                         } else if (((EXT3_I(inode)->i_flags &
 303                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 304                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 305                                 CWARN("extent-mapped directory found with "
 306                                       "ext3-based ldiskfs - contact "
 307                                       "http://bugzilla.lustre.org/\n");
 308                                 warned = 1;
 309                         }
 310                 }
 311 #endif
 312                 /* no break */
 313         }
 314         case FSFILT_OP_MKDIR:
 315         case FSFILT_OP_MKNOD:
 316                 /* modify one inode + block bitmap + GDT */
 317                 nblocks += 3;
 318                 /* no break */
 319         case FSFILT_OP_LINK:
 320                 /* modify parent directory */
 321                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 322                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 323                 /* create/update logs for each stripe */
 324                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 325                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 326                 break;
 327         case FSFILT_OP_SETATTR:
 328                 /* Setattr on inode */
 329                 nblocks += 1;
 330                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 331                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 332                 /* quota chown log for each stripe */
 333                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 334                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 335                 break;
 336         case FSFILT_OP_CANCEL_UNLINK:
 337                 /* blocks for log header bitmap update OR
 338                  * blocks for catalog header bitmap update + unlink of logs */
 339                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 340                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 341                 break;
 342         default: CERROR("unknown transaction start op %d\n", op);
 343                 LBUG();
 344         }
 345
 346         LASSERT(current->journal_info == desc_private);
 347         journal = EXT3_SB(inode->i_sb)->s_journal;
 348         if (nblocks > journal->j_max_transaction_buffers) {
 349                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 350                        nblocks, op, logs, journal->j_max_transaction_buffers);
 351                 nblocks = journal->j_max_transaction_buffers;
 352         }
 353
 354  journal_start:
 355         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 356         handle = ext3_journal_start(inode, nblocks);
 357
 358         if (!IS_ERR(handle))
 359                 LASSERT(current->journal_info == handle);
 360         else
 361                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 362                        op, nblocks, PTR_ERR(handle));
 363         return handle;
 364 }
 365
 366 /*
 367  * Calculate the number of buffer credits needed to write multiple pages in
 368  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 369  * doesn't have a nice API for calculating this sort of thing in advance.
 370  *
 371  * See comment above ext3_writepage_trans_blocks for details.  We assume
 372  * no data journaling is being done, but it does allow for all of the pages
 373  * being non-contiguous.  If we are guaranteed contiguous pages we could
 374  * reduce the number of (d)indirect blocks a lot.
 375  *
 376  * With N blocks per page and P pages, for each inode we have at most:
 377  * N*P indirect
 378  * min(N*P, blocksize/4 + 1) dindirect blocks
 379  * niocount tindirect
 380  *
 381  * For the entire filesystem, we have at most:
 382  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 383  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 384  * objcount inode blocks
 385  * 1 superblock
 386  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 387  *
 388  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 389  */
 390 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 391                                       int niocount, struct niobuf_local *nb)
 392 {
 393         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 394         __u64 next_indir;
 395         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 396         int nbitmaps = 0, ngdblocks;
 397         int needed = objcount + 1; /* inodes + superblock */
 398         int i, j;
 399
 400         for (i = 0, j = 0; i < objcount; i++, fso++) {
 401                 /* two or more dindirect blocks in case we cross boundary */
 402                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 403                                     nb[j].offset) >>
 404                                    sb->s_blocksize_bits) /
 405                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 406                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 407
 408                 /* leaf, indirect, tindirect blocks for first block */
 409                 nbitmaps += blockpp + 2;
 410
 411                 j += fso->fso_bufcnt;
 412         }
 413
 414         next_indir = nb[0].offset +
 415                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 416         for (i = 1; i < niocount; i++) {
 417                 if (nb[i].offset >= next_indir) {
 418                         nbitmaps++;     /* additional indirect */
 419                         next_indir = nb[i].offset +
 420                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 421                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 422                         nbitmaps++;     /* additional indirect */
 423                 }
 424                 nbitmaps += blockpp;    /* each leaf in different group? */
 425         }
 426
 427         ngdblocks = nbitmaps;
 428         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 429                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 430         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 431                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 432
 433         needed += nbitmaps + ngdblocks;
 434
 435         /* last_rcvd update */
 436         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 437
 438 #if defined(CONFIG_QUOTA)
 439         /* We assume that there will be 1 bit set in s_dquot.flags for each
 440          * quota file that is active.  This is at least true for now.
 441          */
 442         needed += hweight32(ll_sb_any_quota_active(sb)) *
 443                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 444 #endif
 445
 446         return needed;
 447 }
 448
 449 /* We have to start a huge journal transaction here to hold all of the
 450  * metadata for the pages being written here.  This is necessitated by
 451  * the fact that we do lots of prepare_write operations before we do
 452  * any of the matching commit_write operations, so even if we split
 453  * up to use "smaller" transactions none of them could complete until
 454  * all of them were opened.  By having a single journal transaction,
 455  * we eliminate duplicate reservations for common blocks like the
 456  * superblock and group descriptors or bitmaps.
 457  *
 458  * We will start the transaction here, but each prepare_write will
 459  * add a refcount to the transaction, and each commit_write will
 460  * remove a refcount.  The transaction will be closed when all of
 461  * the pages have been written.
 462  */
 463 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 464                                    int niocount, struct niobuf_local *nb,
 465                                    void *desc_private, int logs)
 466 {
 467         journal_t *journal;
 468         handle_t *handle;
 469         int needed;
 470         ENTRY;
 471
 472         LASSERT(current->journal_info == desc_private);
 473         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 474         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 475
 476         /* The number of blocks we could _possibly_ dirty can very large.
 477          * We reduce our request if it is absurd (and we couldn't get that
 478          * many credits for a single handle anyways).
 479          *
 480          * At some point we have to limit the size of I/Os sent at one time,
 481          * increase the size of the journal, or we have to calculate the
 482          * actual journal requirements more carefully by checking all of
 483          * the blocks instead of being maximally pessimistic.  It remains to
 484          * be seen if this is a real problem or not.
 485          */
 486         if (needed > journal->j_max_transaction_buffers) {
 487                 CERROR("want too many journal credits (%d) using %d instead\n",
 488                        needed, journal->j_max_transaction_buffers);
 489                 needed = journal->j_max_transaction_buffers;
 490         }
 491
 492         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 493         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 494         if (IS_ERR(handle)) {
 495                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 496                        PTR_ERR(handle));
 497         } else {
 498                 LASSERT(handle->h_buffer_credits >= needed);
 499                 LASSERT(current->journal_info == handle);
 500         }
 501
 502         RETURN(handle);
 503 }
 504
 505 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 506 {
 507        handle_t *handle = h;
 508
 509        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 510        if (nblocks == 0) {
 511                handle->h_buffer_credits = 0;
 512                CWARN("setting credits of handle %p to zero by request\n", h);
 513        }
 514
 515        if (handle->h_buffer_credits > nblocks)
 516                 return 0;
 517        if (ext3_journal_extend(handle, nblocks) == 0)
 518                 return 0;
 519
 520        ext3_mark_inode_dirty(handle, inode);
 521        return ext3_journal_restart(handle, nblocks);
 522 }
 523
 524 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 525 {
 526         int rc;
 527         handle_t *handle = h;
 528
 529         LASSERT(current->journal_info == handle);
 530         if (force_sync)
 531                 handle->h_sync = 1; /* recovery likes this */
 532
 533         rc = ext3_journal_stop(handle);
 534
 535         return rc;
 536 }
 537
 538 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 539                                     void **wait_handle)
 540 {
 541         unsigned long tid;
 542         transaction_t *transaction;
 543         handle_t *handle = h;
 544         journal_t *journal;
 545         int rc;
 546
 547         LASSERT(current->journal_info == handle);
 548
 549         transaction = handle->h_transaction;
 550         journal = transaction->t_journal;
 551         tid = transaction->t_tid;
 552         /* we don't want to be blocked */
 553         handle->h_sync = 0;
 554         rc = ext3_journal_stop(handle);
 555         if (rc) {
 556                 CERROR("error while stopping transaction: %d\n", rc);
 557                 return rc;
 558         }
 559         fsfilt_log_start_commit(journal, tid);
 560
 561         *wait_handle = (void *) tid;
 562         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 563         return 0;
 564 }
 565
 566 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 567 {
 568         journal_t *journal = EXT3_JOURNAL(inode);
 569         tid_t tid = (tid_t)(long)h;
 570
 571         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 572         if (unlikely(is_journal_aborted(journal)))
 573                 return -EIO;
 574
 575         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 576
 577         if (unlikely(is_journal_aborted(journal)))
 578                 return -EIO;
 579         return 0;
 580 }
 581
 582 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 583                                struct iattr *iattr, int do_trunc)
 584 {
 585         struct inode *inode = dentry->d_inode;
 586         int rc = 0;
 587
 588         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 589          * We are already writing the inode to disk as part of this
 590          * transaction and want to avoid a lot of extra inode writeout
 591          * later on. b=9828 */
 592         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 593                 /* ATTR_SIZE would invoke truncate: clear it */
 594                 iattr->ia_valid &= ~ATTR_SIZE;
 595                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 596                 i_size_write(inode, iattr->ia_size);
 597
 598                 if (iattr->ia_valid & ATTR_UID)
 599                         inode->i_uid = iattr->ia_uid;
 600                 if (iattr->ia_valid & ATTR_GID)
 601                         inode->i_gid = iattr->ia_gid;
 602                 if (iattr->ia_valid & ATTR_ATIME)
 603                         inode->i_atime = iattr->ia_atime;
 604                 if (iattr->ia_valid & ATTR_MTIME)
 605                         inode->i_mtime = iattr->ia_mtime;
 606                 if (iattr->ia_valid & ATTR_CTIME)
 607                         inode->i_ctime = iattr->ia_ctime;
 608                 if (iattr->ia_valid & ATTR_MODE) {
 609                         inode->i_mode = iattr->ia_mode;
 610
 611                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 612                             !cfs_capable(CFS_CAP_FSETID))
 613                                 inode->i_mode &= ~S_ISGID;
 614                 }
 615
 616                 inode->i_sb->s_op->dirty_inode(inode);
 617
 618                 goto out;
 619         }
 620
 621         /* Don't allow setattr to change file type */
 622         if (iattr->ia_valid & ATTR_MODE)
 623                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 624                                  (iattr->ia_mode & ~S_IFMT);
 625
 626         /* We set these flags on the client, but have already checked perms
 627          * so don't confuse inode_change_ok. */
 628         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 629
 630         if (inode->i_op->setattr) {
 631                 rc = inode->i_op->setattr(dentry, iattr);
 632         } else {
 633                 rc = inode_change_ok(inode, iattr);
 634                 if (!rc)
 635                         rc = inode_setattr(inode, iattr);
 636         }
 637
 638  out:
 639         RETURN(rc);
 640 }
 641
 642 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 643                                  unsigned int cmd, unsigned long arg)
 644 {
 645         int rc = 0;
 646         ENTRY;
 647
 648         /* FIXME: Can't do this because of nested transaction deadlock */
 649         if (cmd == EXT3_IOC_SETFLAGS) {
 650                 /* We can't enable data journaling on OST objects, because
 651                 * this forces the transaction to be closed in order to
 652                 * flush the journal, but the caller will already have a
 653                 * compound transaction open to update the last_rcvd file,
 654                 * and this thread would deadlock trying to set the flag. */
 655                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 656                         CERROR("can't set data journal flag on file\n");
 657                         RETURN(-EPERM);
 658                 }
 659                 /* Because the MDS does not see the EXTENTS_FL set on the
 660                  * OST objects, mask this flag into all set flags.  It is
 661                  * not legal to clear this flag in any case, so we are not
 662                  * changing the functionality by doing this.  b=22911 */
 663                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 664         }
 665
 666 #ifdef HAVE_EXT4_LDISKFS
 667         /* ext4_ioctl does not have a inode argument */
 668         if (inode->i_fop->unlocked_ioctl)
 669                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 670 #else
 671         if (inode->i_fop->ioctl)
 672                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 673 #endif
 674         else
 675                 RETURN(-ENOTTY);
 676
 677         RETURN(rc);
 678 }
 679
 680 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 681                               void *lmm, int lmm_size, const char *name)
 682 {
 683         int rc;
 684
 685         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 686
 687         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 688                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 689
 690
 691         if (rc && rc != -EROFS)
 692                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 693                        inode->i_ino, rc);
 694         return rc;
 695 }
 696
 697 /* Must be called with i_mutex held */
 698 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 699                               const char *name)
 700 {
 701         int rc;
 702
 703         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 704
 705         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 706                             name, lmm, lmm_size);
 707
 708         /* This gives us the MD size */
 709         if (lmm == NULL)
 710                 return (rc == -ENODATA) ? 0 : rc;
 711
 712         if (rc < 0) {
 713                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 714                        EXT3_XATTR_INDEX_TRUSTED, name,
 715                        inode->i_ino, rc);
 716                 memset(lmm, 0, lmm_size);
 717                 return (rc == -ENODATA) ? 0 : rc;
 718         }
 719
 720         return rc;
 721 }
 722
 723 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 724 {
 725         submit_bio(rw, bio);
 726         return 0;
 727 }
 728
 729 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 730                                     loff_t *off)
 731 {
 732         struct inode *inode = file->f_dentry->d_inode;
 733         int rc = 0;
 734
 735         if (S_ISREG(inode->i_mode))
 736                 rc = file->f_op->read(file, buf, count, off);
 737         else {
 738                 const int blkbits = inode->i_sb->s_blocksize_bits;
 739                 const int blksize = inode->i_sb->s_blocksize;
 740
 741                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 742                        (unsigned long)count, inode->i_ino, *off);
 743                 while (count > 0) {
 744                         struct buffer_head *bh;
 745
 746                         bh = NULL;
 747                         if (*off < i_size_read(inode)) {
 748                                 int err = 0;
 749
 750                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 751                                                 0, &err);
 752
 753                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 754
 755                                 if (bh) {
 756                                         memcpy(buf, bh->b_data, blksize);
 757                                         brelse(bh);
 758                                 } else if (err) {
 759                                         /* XXX in theory we should just fake
 760                                          * this buffer and continue like ext3,
 761                                          * especially if this is a partial read
 762                                          */
 763                                         CERROR("error read dir %lu+%llu: %d\n",
 764                                                inode->i_ino, *off, err);
 765                                         RETURN(err);
 766                                 }
 767                         }
 768                         if (!bh) {
 769                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 770
 771                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 772                                 memset(fake, 0, sizeof(*fake));
 773                                 fake->rec_len = cpu_to_le16(blksize);
 774                         }
 775                         count -= blksize;
 776                         buf += blksize;
 777                         *off += blksize;
 778                         rc += blksize;
 779                 }
 780         }
 781
 782         return rc;
 783 }
 784
 785 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 786 {
 787         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 788
 789         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 790
 791         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 792 }
 793
 794 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 795                                       void *handle, fsfilt_cb_t cb_func,
 796                                       void *cb_data)
 797 {
 798         struct fsfilt_cb_data *fcb;
 799
 800         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 801         if (fcb == NULL)
 802                 RETURN(-ENOMEM);
 803
 804         fcb->cb_func = cb_func;
 805         fcb->cb_obd = obd;
 806         fcb->cb_last_rcvd = last_rcvd;
 807         fcb->cb_data = cb_data;
 808
 809         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 810         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 811                                     (struct journal_callback *)fcb);
 812
 813         return 0;
 814 }
 815
 816 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 817 {
 818         struct kstatfs sfs;
 819         int rc;
 820
 821         memset(&sfs, 0, sizeof(sfs));
 822         rc = ll_do_statfs(sb, &sfs);
 823         statfs_pack(osfs, &sfs);
 824         return rc;
 825 }
 826
 827 static int fsfilt_ext3_sync(struct super_block *sb)
 828 {
 829         return ext3_force_commit(sb);
 830 }
 831
 832 #ifndef EXT3_EXTENTS_FL
 833 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 834 #endif
 835
 836 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 837 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 838 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 839 #else
 840 # ifdef HAVE_EXT4_LDISKFS
 841 #  ifdef WALK_SPACE_HAS_DATA_SEM /* We only use it in fsfilt_map_nblocks() for now */
 842 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 843 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 844 #  else
 845 #   define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem))
 846 #   define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem))
 847 #  endif
 848 # else
 849 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 850 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 851 # endif
 852 #endif
 853
 854 #ifndef EXT_ASSERT
 855 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 856 #endif
 857
 858 #ifdef EXT3_EXT_HAS_NO_TREE
 859 /* for kernels 2.6.18 and later */
 860 #ifdef HAVE_EXT4_LDISKFS
 861 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 862 #else
 863 #define EXT_GENERATION(inode)           ext_generation(inode)
 864 #endif
 865 #define ext3_ext_base                   inode
 866 #define ext3_ext_base2inode(inode)      (inode)
 867 #define EXT_DEPTH(inode)                ext_depth(inode)
 868 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 869                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 870 #else
 871 #define ext3_ext_base                   ext3_extents_tree
 872 #define ext3_ext_base2inode(tree)       (tree->inode)
 873 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 874                         ext3_ext_walk_space(tree, block, num, cb);
 875 #endif
 876
 877 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 878 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 879                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 880 #else
 881 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 882                ext3_ext_insert_extent(handle, inode, path, newext)
 883 #endif
 884
 885 #include <linux/lustre_version.h>
 886
 887 struct bpointers {
 888         unsigned long *blocks;
 889         int *created;
 890         unsigned long start;
 891         int num;
 892         int init_num;
 893         int create;
 894 };
 895
 896 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 897                                unsigned long block, int *aflags)
 898 {
 899         struct ext3_inode_info *ei = EXT3_I(inode);
 900         unsigned long bg_start;
 901         unsigned long colour;
 902         int depth;
 903
 904         if (path) {
 905                 struct ext3_extent *ex;
 906                 depth = path->p_depth;
 907
 908                 /* try to predict block placement */
 909                 if ((ex = path[depth].p_ext))
 910                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 911
 912                 /* it looks index is empty
 913                  * try to find starting from index itself */
 914                 if (path[depth].p_bh)
 915                         return path[depth].p_bh->b_blocknr;
 916         }
 917
 918         /* OK. use inode's group */
 919         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 920                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 921         colour = (current->pid % 16) *
 922                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 923         return bg_start + colour + block;
 924 }
 925
 926 #define ll_unmap_underlying_metadata(sb, blocknr) \
 927         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 928
 929 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 930 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 931                                 struct ext3_ext_path *path, unsigned long block,
 932                                 unsigned long *count, int *err)
 933 {
 934         unsigned long pblock, goal;
 935         int aflags = 0;
 936         struct inode *inode = ext3_ext_base2inode(base);
 937
 938         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 939         aflags |= 2; /* block have been already reserved */
 940         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 941         return pblock;
 942
 943 }
 944 #else
 945 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 946                                 struct ext3_ext_path *path, unsigned long block,
 947                                 unsigned long *count, int *err)
 948 {
 949         struct inode *inode = ext3_ext_base2inode(base);
 950         struct ext3_allocation_request ar;
 951         unsigned long pblock;
 952         int aflags;
 953
 954         /* find neighbour allocated blocks */
 955         ar.lleft = block;
 956         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 957         if (*err)
 958                 return 0;
 959         ar.lright = block;
 960         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 961         if (*err)
 962                 return 0;
 963
 964         /* allocate new block */
 965         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 966         ar.inode = inode;
 967         ar.logical = block;
 968         ar.len = *count;
 969         ar.flags = EXT3_MB_HINT_DATA;
 970         pblock = ext3_mb_new_blocks(handle, &ar, err);
 971         *count = ar.len;
 972         return pblock;
 973 }
 974 #endif
 975
 976 #ifdef EXT3_EXT_HAS_NO_TREE
 977 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 978                                   struct ext3_ext_path *path,
 979                                   struct ext3_ext_cache *cex,
 980 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 981                                    struct ext3_extent *ex,
 982 #endif
 983                                   void *cbdata)
 984 {
 985         struct bpointers *bp = cbdata;
 986 #else
 987 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 988                                   struct ext3_ext_path *path,
 989                                   struct ext3_ext_cache *cex
 990 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 991                                   , struct ext3_extent *ex
 992 #endif
 993                                  )
 994 {
 995         struct bpointers *bp = base->private;
 996 #endif
 997         struct inode *inode = ext3_ext_base2inode(base);
 998         struct ext3_extent nex;
 999 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1000         struct ext4_ext_path *tmppath = NULL;
1001         struct ext4_extent *tmpex;
1002 #endif
1003         unsigned long pblock;
1004         unsigned long tgen;
1005         int err, i, depth;
1006         unsigned long count;
1007         handle_t *handle;
1008
1009         i = depth = EXT_DEPTH(base);
1010         EXT_ASSERT(i == path->p_depth);
1011         EXT_ASSERT(path[i].p_hdr);
1012
1013         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1014                 err = EXT_CONTINUE;
1015                 goto map;
1016         }
1017
1018         if (bp->create == 0) {
1019                 i = 0;
1020                 if (cex->ec_block < bp->start)
1021                         i = bp->start - cex->ec_block;
1022                 if (i >= cex->ec_len)
1023                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1024                                         i, cex->ec_len);
1025                 for (; i < cex->ec_len && bp->num; i++) {
1026                         *(bp->created) = 0;
1027                         bp->created++;
1028                         *(bp->blocks) = 0;
1029                         bp->blocks++;
1030                         bp->num--;
1031                         bp->start++;
1032                 }
1033
1034                 return EXT_CONTINUE;
1035         }
1036
1037         tgen = EXT_GENERATION(base);
1038         count = ext3_ext_calc_credits_for_insert(base, path);
1039         fsfilt_up_truncate_sem(inode);
1040
1041         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1042         if (IS_ERR(handle)) {
1043                 fsfilt_down_truncate_sem(inode);
1044                 return PTR_ERR(handle);
1045         }
1046
1047         fsfilt_down_truncate_sem(inode);
1048         if (tgen != EXT_GENERATION(base)) {
1049                 /* the tree has changed. so path can be invalid at moment */
1050                 ext3_journal_stop(handle);
1051                 return EXT_REPEAT;
1052         }
1053
1054 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1055         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1056          * protected by i_data_sem, we need revalidate extent to be created */
1057         down_write((&EXT4_I(inode)->i_data_sem));
1058
1059         /* validate extent, make sure the extent tree does not changed */
1060         tmppath = ext4_ext_find_extent(inode, cex->ec_block, NULL);
1061         if (IS_ERR(tmppath)) {
1062                 up_write(&EXT4_I(inode)->i_data_sem);
1063                 ext3_journal_stop(handle);
1064                 return PTR_ERR(tmppath);
1065         }
1066         tmpex = tmppath[depth].p_ext;
1067         if (tmpex != ex) {
1068                 /* cex is invalid, try again */
1069                 ext4_ext_drop_refs(tmppath);
1070                 kfree(tmppath);
1071                 up_write(&EXT4_I(inode)->i_data_sem);
1072                 ext3_journal_stop(handle);
1073                 return EXT_REPEAT;
1074         }
1075 #endif
1076
1077         count = cex->ec_len;
1078         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1079         if (!pblock)
1080                 goto out;
1081         EXT_ASSERT(count <= cex->ec_len);
1082
1083         /* insert new extent */
1084         nex.ee_block = cpu_to_le32(cex->ec_block);
1085         ext3_ext_store_pblock(&nex, pblock);
1086         nex.ee_len = cpu_to_le16(count);
1087         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1088         if (err) {
1089                 /* free data blocks we just allocated */
1090                 /* not a good idea to call discard here directly,
1091                  * but otherwise we'd need to call it every free() */
1092 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1093                 ext3_mb_discard_inode_preallocations(inode);
1094 #endif
1095                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1096                                  cpu_to_le16(nex.ee_len), 0);
1097                 goto out;
1098         }
1099
1100         /*
1101          * Putting len of the actual extent we just inserted,
1102          * we are asking ext3_ext_walk_space() to continue
1103          * scaning after that block
1104          */
1105         cex->ec_len = le16_to_cpu(nex.ee_len);
1106         cex->ec_start = ext_pblock(&nex);
1107         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1108         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1109
1110 out:
1111 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1112         ext4_ext_drop_refs(tmppath);
1113         kfree(tmppath);
1114         up_write((&EXT4_I(inode)->i_data_sem));
1115 #endif
1116         ext3_journal_stop(handle);
1117 map:
1118         if (err >= 0) {
1119                 /* map blocks */
1120                 if (bp->num == 0) {
1121                         CERROR("hmm. why do we find this extent?\n");
1122                         CERROR("initial space: %lu:%u\n",
1123                                 bp->start, bp->init_num);
1124                         CERROR("current extent: %u/%u/%llu %d\n",
1125                                 cex->ec_block, cex->ec_len,
1126                                 (unsigned long long)cex->ec_start,
1127                                 cex->ec_type);
1128                 }
1129                 i = 0;
1130                 if (cex->ec_block < bp->start)
1131                         i = bp->start - cex->ec_block;
1132                 if (i >= cex->ec_len)
1133                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1134                                         i, cex->ec_len);
1135                 for (; i < cex->ec_len && bp->num; i++) {
1136                         *(bp->blocks) = cex->ec_start + i;
1137                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1138                                 *(bp->created) = 0;
1139                         } else {
1140                                 *(bp->created) = 1;
1141                                 /* unmap any possible underlying metadata from
1142                                  * the block device mapping.  bug 6998. */
1143                                 ll_unmap_underlying_metadata(inode->i_sb,
1144                                                              *(bp->blocks));
1145                         }
1146                         bp->created++;
1147                         bp->blocks++;
1148                         bp->num--;
1149                         bp->start++;
1150                 }
1151         }
1152         return err;
1153 }
1154
1155 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1156                        unsigned long num, unsigned long *blocks,
1157                        int *created, int create)
1158 {
1159 #ifdef EXT3_EXT_HAS_NO_TREE
1160         struct ext3_ext_base *base = inode;
1161 #else
1162         struct ext3_extents_tree tree;
1163         struct ext3_ext_base *base = &tree;
1164 #endif
1165         struct bpointers bp;
1166         int err;
1167
1168         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1169                block, block + num - 1, (unsigned) inode->i_ino);
1170
1171 #ifndef EXT3_EXT_HAS_NO_TREE
1172         ext3_init_tree_desc(base, inode);
1173         tree.private = &bp;
1174 #endif
1175         bp.blocks = blocks;
1176         bp.created = created;
1177         bp.start = block;
1178         bp.init_num = bp.num = num;
1179         bp.create = create;
1180
1181         fsfilt_down_truncate_sem(inode);
1182         err = fsfilt_ext3_ext_walk_space(base, block, num,
1183                                          ext3_ext_new_extent_cb, &bp);
1184         ext3_ext_invalidate_cache(base);
1185         fsfilt_up_truncate_sem(inode);
1186
1187         return err;
1188 }
1189
1190 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1191                                     int pages, unsigned long *blocks,
1192                                     int *created, int create)
1193 {
1194         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1195         int rc = 0, i = 0;
1196         struct page *fp = NULL;
1197         int clen = 0;
1198
1199         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1200                 inode->i_ino, pages, (*page)->index);
1201
1202         /* pages are sorted already. so, we just have to find
1203          * contig. space and process them properly */
1204         while (i < pages) {
1205                 if (fp == NULL) {
1206                         /* start new extent */
1207                         fp = *page++;
1208                         clen = 1;
1209                         i++;
1210                         continue;
1211                 } else if (fp->index + clen == (*page)->index) {
1212                         /* continue the extent */
1213                         page++;
1214                         clen++;
1215                         i++;
1216                         continue;
1217                 }
1218
1219                 /* process found extent */
1220                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1221                                         clen * blocks_per_page, blocks,
1222                                         created, create);
1223                 if (rc)
1224                         GOTO(cleanup, rc);
1225
1226                 /* look for next extent */
1227                 fp = NULL;
1228                 blocks += blocks_per_page * clen;
1229                 created += blocks_per_page * clen;
1230         }
1231
1232         if (fp)
1233                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1234                                         clen * blocks_per_page, blocks,
1235                                         created, create);
1236 cleanup:
1237         return rc;
1238 }
1239
1240 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1241                                unsigned long *blocks, int *created, int create);
1242 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1243                                    int pages, unsigned long *blocks,
1244                                    int *created, int create)
1245 {
1246         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1247         unsigned long *b;
1248         int rc = 0, i, *cr;
1249
1250         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1251                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1252                 if (rc) {
1253                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1254                                inode->i_ino, *b, *cr, create, rc);
1255                         break;
1256                 }
1257
1258                 b += blocks_per_page;
1259                 cr += blocks_per_page;
1260         }
1261         return rc;
1262 }
1263
1264 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1265                                 int pages, unsigned long *blocks,
1266                                 int *created, int create,
1267                                 cfs_semaphore_t *optional_sem)
1268 {
1269         int rc;
1270
1271         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1272                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1273                                                      blocks, created, create);
1274                 return rc;
1275         }
1276         if (optional_sem != NULL)
1277                 cfs_down(optional_sem);
1278         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1279                                             created, create);
1280         if (optional_sem != NULL)
1281                 cfs_up(optional_sem);
1282
1283         return rc;
1284 }
1285
1286 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1287 {
1288         unsigned long block;
1289         struct buffer_head *bh;
1290         int err, blocksize, csize, boffs, osize = size;
1291
1292         /* prevent reading after eof */
1293         cfs_lock_kernel();
1294         if (i_size_read(inode) < *offs + size) {
1295                 size = i_size_read(inode) - *offs;
1296                 cfs_unlock_kernel();
1297                 if (size < 0) {
1298                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1299                                i_size_read(inode), *offs);
1300                         return -EBADR;
1301                 } else if (size == 0) {
1302                         return 0;
1303                 }
1304         } else {
1305                 cfs_unlock_kernel();
1306         }
1307
1308         blocksize = 1 << inode->i_blkbits;
1309
1310         while (size > 0) {
1311                 block = *offs >> inode->i_blkbits;
1312                 boffs = *offs & (blocksize - 1);
1313                 csize = min(blocksize - boffs, size);
1314                 bh = ext3_bread(NULL, inode, block, 0, &err);
1315                 if (!bh) {
1316                         CERROR("can't read block: %d\n", err);
1317                         return err;
1318                 }
1319
1320                 memcpy(buf, bh->b_data + boffs, csize);
1321                 brelse(bh);
1322
1323                 *offs += csize;
1324                 buf += csize;
1325                 size -= csize;
1326         }
1327         return osize;
1328 }
1329 EXPORT_SYMBOL(fsfilt_ext3_read);
1330
1331 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1332                                    int size, loff_t *offs)
1333 {
1334         int rc;
1335         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1336         if (rc > 0)
1337                 rc = 0;
1338         return rc;
1339 }
1340
1341 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1342                                 loff_t *offs, handle_t *handle)
1343 {
1344         struct buffer_head *bh = NULL;
1345         loff_t old_size = i_size_read(inode), offset = *offs;
1346         loff_t new_size = i_size_read(inode);
1347         unsigned long block;
1348         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1349
1350         while (bufsize > 0) {
1351                 if (bh != NULL)
1352                         brelse(bh);
1353
1354                 block = offset >> inode->i_blkbits;
1355                 boffs = offset & (blocksize - 1);
1356                 size = min(blocksize - boffs, bufsize);
1357                 bh = ext3_bread(handle, inode, block, 1, &err);
1358                 if (!bh) {
1359                         CERROR("can't read/create block: %d\n", err);
1360                         break;
1361                 }
1362
1363                 err = ext3_journal_get_write_access(handle, bh);
1364                 if (err) {
1365                         CERROR("journal_get_write_access() returned error %d\n",
1366                                err);
1367                         break;
1368                 }
1369                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1370                 memcpy(bh->b_data + boffs, buf, size);
1371                 err = ext3_journal_dirty_metadata(handle, bh);
1372                 if (err) {
1373                         CERROR("journal_dirty_metadata() returned error %d\n",
1374                                err);
1375                         break;
1376                 }
1377                 if (offset + size > new_size)
1378                         new_size = offset + size;
1379                 offset += size;
1380                 bufsize -= size;
1381                 buf += size;
1382         }
1383         if (bh)
1384                 brelse(bh);
1385
1386         /* correct in-core and on-disk sizes */
1387         if (new_size > i_size_read(inode)) {
1388                 cfs_lock_kernel();
1389                 if (new_size > i_size_read(inode))
1390                         i_size_write(inode, new_size);
1391                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1392                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1393                 if (i_size_read(inode) > old_size)
1394                         mark_inode_dirty(inode);
1395                 cfs_unlock_kernel();
1396         }
1397
1398         if (err == 0)
1399                 *offs = offset;
1400         return err;
1401 }
1402 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1403
1404 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1405                                     loff_t *offs, int force_sync)
1406 {
1407         struct inode *inode = file->f_dentry->d_inode;
1408         handle_t *handle;
1409         int err, block_count = 0, blocksize;
1410
1411         /* Determine how many transaction credits are needed */
1412         blocksize = 1 << inode->i_blkbits;
1413         block_count = (*offs & (blocksize - 1)) + bufsize;
1414         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1415
1416         handle = ext3_journal_start(inode,
1417                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1418         if (IS_ERR(handle)) {
1419                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1420                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1421                 return PTR_ERR(handle);
1422         }
1423
1424         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1425
1426         if (!err && force_sync)
1427                 handle->h_sync = 1; /* recovery likes this */
1428
1429         ext3_journal_stop(handle);
1430
1431         return err;
1432 }
1433
1434 static int fsfilt_ext3_setup(struct super_block *sb)
1435 {
1436 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1437      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1438         struct ext3_sb_info *sbi = EXT3_SB(sb);
1439 #if 0
1440         sbi->dx_lock = fsfilt_ext3_dx_lock;
1441         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1442 #endif
1443 #endif
1444 #ifdef S_PDIROPS
1445         CWARN("Enabling PDIROPS\n");
1446         set_opt(sbi->s_mount_opt, PDIROPS);
1447         sb->s_flags |= S_PDIROPS;
1448 #endif
1449         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1450                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1451 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1452         /* enable journaled quota support */
1453         /* kfreed in ext3_put_super() */
1454         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1455         if (!sbi->s_qf_names[USRQUOTA])
1456                 return -ENOMEM;
1457         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1458         if (!sbi->s_qf_names[GRPQUOTA]) {
1459                 kfree(sbi->s_qf_names[USRQUOTA]);
1460                 sbi->s_qf_names[USRQUOTA] = NULL;
1461                 return -ENOMEM;
1462         }
1463         sbi->s_jquota_fmt = QFMT_LUSTRE;
1464 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1465         set_opt(sbi->s_mount_opt, QUOTA);
1466 #endif
1467 #endif
1468         return 0;
1469 }
1470
1471 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1472    objects. Logs is number of logfiles to update */
1473 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1474 {
1475         if ( !fso ) {
1476                 switch(op) {
1477                 case FSFILT_OP_CREATE:
1478                                  /* directory leaf, index & indirect & EA*/
1479                         return 4 + 3 * logs;
1480                 case FSFILT_OP_UNLINK:
1481                         return 3 * logs;
1482                 }
1483         } else {
1484                 int i;
1485                 int needed = 0;
1486                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1487                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1488                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1489                 for (i = 0; i < op; i++, fso++) {
1490                         int nblocks = fso->fso_bufcnt * blockpp;
1491                         int ndindirect = min(nblocks, addrpp + 1);
1492                         int nindir = nblocks + ndindirect + 1;
1493
1494                         needed += nindir;
1495                 }
1496                 return needed + 3 * logs;
1497         }
1498
1499         return 0;
1500 }
1501
1502 #ifdef HAVE_QUOTA_SUPPORT
1503 #define DQINFO_COPY(out, in)                    \
1504 do {                                            \
1505         Q_COPY(out, in, dqi_bgrace);            \
1506         Q_COPY(out, in, dqi_igrace);            \
1507         Q_COPY(out, in, dqi_flags);             \
1508         Q_COPY(out, in, dqi_valid);             \
1509 } while (0)
1510
1511 #define DQBLK_COPY(out, in)                     \
1512 do {                                            \
1513         Q_COPY(out, in, dqb_bhardlimit);        \
1514         Q_COPY(out, in, dqb_bsoftlimit);        \
1515         Q_COPY(out, in, dqb_curspace);          \
1516         Q_COPY(out, in, dqb_ihardlimit);        \
1517         Q_COPY(out, in, dqb_isoftlimit);        \
1518         Q_COPY(out, in, dqb_curinodes);         \
1519         Q_COPY(out, in, dqb_btime);             \
1520         Q_COPY(out, in, dqb_itime);             \
1521         Q_COPY(out, in, dqb_valid);             \
1522 } while (0)
1523
1524 static int fsfilt_ext3_quotactl(struct super_block *sb,
1525                                 struct obd_quotactl *oqc)
1526 {
1527         int i, rc = 0, error = 0;
1528         const struct quotactl_ops *qcop;
1529         struct if_dqinfo *info;
1530         struct if_dqblk *dqblk;
1531         ENTRY;
1532
1533         if (!sb->s_qcop)
1534                 RETURN(-ENOSYS);
1535
1536         OBD_ALLOC_PTR(info);
1537         if (!info)
1538                 RETURN(-ENOMEM);
1539         OBD_ALLOC_PTR(dqblk);
1540         if (!dqblk) {
1541                 OBD_FREE_PTR(info);
1542                 RETURN(-ENOMEM);
1543         }
1544
1545         DQINFO_COPY(info, &oqc->qc_dqinfo);
1546         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1547
1548         qcop = sb->s_qcop;
1549         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1550                 for (i = 0; i < MAXQUOTAS; i++) {
1551                         if (!Q_TYPESET(oqc, i))
1552                                 continue;
1553
1554                         if (oqc->qc_cmd == Q_QUOTAON) {
1555                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1556
1557                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1558
1559                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1560                                                  name[i], 0);
1561                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1562                                 rc = ll_quota_off(sb, i, 0);
1563                         }
1564
1565                         if (rc == -EBUSY)
1566                                 error = rc;
1567                         else if (rc)
1568                                 GOTO(out, rc);
1569                 }
1570                 GOTO(out, rc ?: error);
1571         }
1572
1573         switch (oqc->qc_cmd) {
1574         case Q_GETOINFO:
1575         case Q_GETINFO:
1576                 if (!qcop->get_info)
1577                         GOTO(out, rc = -ENOSYS);
1578                 rc = qcop->get_info(sb, oqc->qc_type, info);
1579                 break;
1580         case Q_SETQUOTA:
1581         case Q_INITQUOTA:
1582                 if (!qcop->set_dqblk)
1583                         GOTO(out, rc = -ENOSYS);
1584                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1585                 break;
1586         case Q_GETOQUOTA:
1587         case Q_GETQUOTA:
1588                 if (!qcop->get_dqblk)
1589                         GOTO(out, rc = -ENOSYS);
1590                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1591                 if (!rc)
1592                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1593                 break;
1594         case Q_SYNC:
1595                 if (!sb->s_qcop->quota_sync)
1596                         GOTO(out, rc = -ENOSYS);
1597                 qcop->quota_sync(sb, oqc->qc_type);
1598                 break;
1599         case Q_FINVALIDATE:
1600                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1601                 for (i = 0; i < MAXQUOTAS; i++) {
1602                         struct file *fp;
1603                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1604
1605                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1606
1607                         if (!Q_TYPESET(oqc, i))
1608                                 continue;
1609
1610                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1611                         if (IS_ERR(fp)) {
1612                                 rc = PTR_ERR(fp);
1613                                 CERROR("error invalidating operational quota file"
1614                                        " %s (rc:%d)\n", name[i], rc);
1615                         } else {
1616                                 filp_close(fp, 0);
1617                         }
1618
1619                 }
1620                 break;
1621         default:
1622                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1623                 LBUG();
1624         }
1625 out:
1626         DQINFO_COPY(&oqc->qc_dqinfo, info);
1627         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1628
1629         OBD_FREE_PTR(info);
1630         OBD_FREE_PTR(dqblk);
1631
1632         if (rc)
1633                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1634                                 "failed: %d\n",
1635                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1636         RETURN(rc);
1637 }
1638
1639 struct chk_dqblk{
1640         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1641         cfs_list_t              dqb_list;        /** in list also */
1642         qid_t                   dqb_id;          /** uid/gid */
1643         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1644         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1645         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1646         qsize_t                 dqb_curspace;    /** current space */
1647         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1648         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1649         qsize_t                 dqb_curinodes;   /** current inodes */
1650         obd_time                dqb_btime;       /** block grace time */
1651         obd_time                dqb_itime;       /** inode grace time */
1652         __u32                   dqb_valid;       /** flag for above fields */
1653 };
1654
1655 static inline unsigned int chkquot_hash(qid_t id, int type)
1656                                         __attribute__((__const__));
1657
1658 static inline unsigned int chkquot_hash(qid_t id, int type)
1659 {
1660         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1661 }
1662
1663 static inline struct chk_dqblk *
1664 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1665 {
1666         cfs_hlist_node_t *node;
1667         struct chk_dqblk *cdqb;
1668
1669         cfs_hlist_for_each(node, head) {
1670                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1671                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1672                         return cdqb;
1673         }
1674
1675         return NULL;
1676 }
1677
1678 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1679 {
1680         struct chk_dqblk *cdqb;
1681
1682         OBD_ALLOC_PTR(cdqb);
1683         if (cdqb) {
1684                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1685                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1686                 cdqb->dqb_id = id;
1687                 cdqb->dqb_type = type;
1688         }
1689
1690         return cdqb;
1691 }
1692
1693 static struct chk_dqblk *
1694 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1695       cfs_list_t *list, qid_t id, int type, int first_check)
1696 {
1697         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1698         struct if_dqblk dqb;
1699         struct chk_dqblk *cdqb;
1700         int rc;
1701
1702         cdqb = find_chkquot(head, id, type);
1703         if (cdqb)
1704                 return cdqb;
1705
1706         cdqb = alloc_chkquot(id, type);
1707         if (!cdqb)
1708                 return NULL;
1709
1710         if (!first_check) {
1711                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1712                 if (rc) {
1713                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1714                                id, type, rc);
1715                 } else {
1716                         DQBLK_COPY(cdqb, &dqb);
1717                         cdqb->dqb_curspace = 0;
1718                         cdqb->dqb_curinodes = 0;
1719                 }
1720         }
1721
1722         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1723         cfs_list_add_tail(&cdqb->dqb_list, list);
1724
1725         return cdqb;
1726 }
1727
1728 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1729 {
1730         struct obd_quotactl *oqctl;
1731         int rc;
1732
1733         OBD_ALLOC_PTR(oqctl);
1734         if (!oqctl)
1735                 RETURN(-ENOMEM);
1736
1737         oqctl->qc_cmd = cmd;
1738         oqctl->qc_id = qfmt;
1739         oqctl->qc_type = type;
1740         rc = fsfilt_ext3_quotactl(sb, oqctl);
1741
1742         OBD_FREE_PTR(oqctl);
1743         return rc;
1744 }
1745
1746 static inline int read_old_dqinfo(struct super_block *sb, int type,
1747                                   struct if_dqinfo *dqinfo)
1748 {
1749         struct obd_quotactl *oqctl;
1750         int rc;
1751         ENTRY;
1752
1753         OBD_ALLOC_PTR(oqctl);
1754         if (!oqctl)
1755                 RETURN(-ENOMEM);
1756
1757         oqctl->qc_cmd = Q_GETINFO;
1758         oqctl->qc_type = type;
1759         rc = fsfilt_ext3_quotactl(sb, oqctl);
1760         if (!rc)
1761                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1762
1763         OBD_FREE_PTR(oqctl);
1764         RETURN(rc);
1765 }
1766
1767 #ifndef HAVE_EXT4_LDISKFS
1768 static inline struct ext3_group_desc *
1769 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1770 {
1771         unsigned long desc_block, desc;
1772         struct ext3_group_desc *gdp;
1773
1774         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1775         desc = group % EXT3_DESC_PER_BLOCK(sb);
1776         gdp = (struct ext3_group_desc *)
1777               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1778
1779         return gdp + desc;
1780 }
1781
1782 static inline struct buffer_head *
1783 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1784 {
1785         struct ext3_group_desc *desc;
1786         struct buffer_head *bh;
1787
1788         desc = get_group_desc(sb, group, NULL);
1789         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1790         return bh;
1791 }
1792
1793 static __u32 ext3_itable_unused_count(struct super_block *sb,
1794                                struct ext3_group_desc *bg) {
1795        return le16_to_cpu(bg->bg_itable_unused);
1796 }
1797 #else
1798 #define get_group_desc ext3_get_group_desc
1799 #endif
1800
1801 struct qchk_ctxt {
1802         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1803         cfs_list_t              qckt_list;                 /* quotacheck list */
1804         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1805         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1806 };
1807
1808 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1809                            struct obd_quotactl *oqc)
1810 {
1811         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1812         loff_t size = 0;
1813         qid_t qid[MAXQUOTAS];
1814         int cnt, i, rc = 0;
1815
1816         if (!inode)
1817                 return 0;
1818
1819         qid[USRQUOTA] = inode->i_uid;
1820         qid[GRPQUOTA] = inode->i_gid;
1821
1822         if (S_ISDIR(inode->i_mode) ||
1823             S_ISREG(inode->i_mode) ||
1824             S_ISLNK(inode->i_mode))
1825                 size = inode_get_bytes(inode);
1826
1827         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1828                 if (!Q_TYPESET(oqc, cnt))
1829                         continue;
1830
1831                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1832                                 &qctxt->qckt_list, qid[cnt], cnt,
1833                                 qctxt->qckt_first_check[cnt]);
1834                 if (!cdqb[cnt]) {
1835                         rc = -ENOMEM;
1836                         break;
1837                 }
1838
1839                 cdqb[cnt]->dqb_curspace += size;
1840                 cdqb[cnt]->dqb_curinodes++;
1841         }
1842
1843         if (rc) {
1844                 for (i = 0; i < cnt; i++) {
1845                         if (!Q_TYPESET(oqc, i))
1846                                 continue;
1847                         LASSERT(cdqb[i]);
1848                         cdqb[i]->dqb_curspace -= size;
1849                         cdqb[i]->dqb_curinodes--;
1850                 }
1851         }
1852
1853         return rc;
1854 }
1855
1856 /* write dqinfo struct in a new quota file */
1857 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1858 {
1859         struct v2_disk_dqinfo dqinfo;
1860         __u32 blocks = V2_DQTREEOFF + 1;
1861         loff_t offset = V2_DQINFOOFF;
1862
1863         if (info) {
1864                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1865                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1866                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1867                                                ~DQF_INFO_DIRTY);
1868         } else {
1869                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1870                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1871                 dqinfo.dqi_flags = 0;
1872         }
1873
1874         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1875         dqinfo.dqi_free_blk = 0;
1876         dqinfo.dqi_free_entry = 0;
1877
1878         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1879 }
1880
1881 static int v3_write_dqheader(struct file *f, int type)
1882 {
1883         static const __u32 quota_magics[] = V2_INITQMAGICS;
1884         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1885         struct v2_disk_dqheader dqhead;
1886         loff_t offset = 0;
1887
1888         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1889         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1890
1891         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1892         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1893
1894         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1895 }
1896
1897 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1898                                   struct obd_quotactl *oqc)
1899 {
1900         int i, rc = 0;
1901         ENTRY;
1902
1903         for (i = 0; i < MAXQUOTAS; i++) {
1904                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1905                                          NULL : &qctxt->qckt_dqinfo[i];
1906                 struct file *file;
1907                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1908
1909                 if (!Q_TYPESET(oqc, i))
1910                         continue;
1911
1912                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1913
1914                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1915                 if (IS_ERR(file)) {
1916                         rc = PTR_ERR(file);
1917                         CERROR("can't create %s file: rc = %d\n",
1918                                name[i], rc);
1919                         GOTO(out, rc);
1920                 }
1921
1922                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1923                         CERROR("file %s is not regular", name[i]);
1924                         filp_close(file, 0);
1925                         GOTO(out, rc = -EINVAL);
1926                 }
1927
1928                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1929
1930                 rc = v3_write_dqheader(file, i);
1931                 if (rc) {
1932                         filp_close(file, 0);
1933                         GOTO(out, rc);
1934                 }
1935
1936                 rc = v3_write_dqinfo(file, i, info);
1937                 filp_close(file, 0);
1938                 if (rc)
1939                         GOTO(out, rc);
1940         }
1941
1942 out:
1943         RETURN(rc);
1944 }
1945
1946
1947 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1948                           struct chk_dqblk *cdqb)
1949 {
1950         struct obd_quotactl *oqc;
1951         long now;
1952         int rc;
1953         ENTRY;
1954
1955         OBD_ALLOC_PTR(oqc);
1956         if (!oqc)
1957                 RETURN(-ENOMEM);
1958
1959         now = cfs_time_current_sec();
1960
1961         if (cdqb->dqb_bsoftlimit &&
1962             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1963             !cdqb->dqb_btime)
1964                 cdqb->dqb_btime =
1965                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1966
1967         if (cdqb->dqb_isoftlimit &&
1968             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1969             !cdqb->dqb_itime)
1970                 cdqb->dqb_itime =
1971                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1972
1973         cdqb->dqb_valid = QIF_ALL;
1974
1975         oqc->qc_cmd = Q_SETQUOTA;
1976         oqc->qc_type = cdqb->dqb_type;
1977         oqc->qc_id = cdqb->dqb_id;
1978         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1979
1980         rc = fsfilt_ext3_quotactl(sb, oqc);
1981         OBD_FREE_PTR(oqc);
1982         RETURN(rc);
1983 }
1984
1985 static int prune_chkquots(struct super_block *sb,
1986                           struct qchk_ctxt *qctxt, int error)
1987 {
1988         struct chk_dqblk *cdqb, *tmp;
1989         int rc;
1990
1991         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1992                 if (!error) {
1993                         rc = commit_chkquot(sb, qctxt, cdqb);
1994                         if (rc)
1995                                 error = rc;
1996                 }
1997                 cfs_hlist_del_init(&cdqb->dqb_hash);
1998                 cfs_list_del(&cdqb->dqb_list);
1999                 OBD_FREE_PTR(cdqb);
2000         }
2001
2002         return error;
2003 }
2004
2005 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2006 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2007 #endif
2008
2009 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2010                                   struct obd_quotactl *oqc)
2011 {
2012         struct ext3_sb_info *sbi = EXT3_SB(sb);
2013         int i, group, uninit_feat = 0;
2014         struct qchk_ctxt *qctxt;
2015         struct buffer_head *bitmap_bh = NULL;
2016         unsigned long ino, inode_inuse;
2017         struct inode *inode;
2018         int rc = 0;
2019         ENTRY;
2020
2021         /* turn on quota and read dqinfo if existed */
2022         OBD_ALLOC_PTR(qctxt);
2023         if (!qctxt) {
2024                 oqc->qc_stat = -ENOMEM;
2025                 RETURN(-ENOMEM);
2026         }
2027
2028         for (i = 0; i < NR_DQHASH; i++)
2029                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2030         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2031
2032         for (i = 0; i < MAXQUOTAS; i++) {
2033                 if (!Q_TYPESET(oqc, i))
2034                         continue;
2035
2036                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2037                 if (!rc || rc == -EBUSY) {
2038                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2039                         if (rc)
2040                                 GOTO(out, rc);
2041                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2042                         qctxt->qckt_first_check[i] = 1;
2043                 } else if (rc) {
2044                         GOTO(out, rc);
2045                 }
2046         }
2047         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2048                 /* This filesystem supports the uninit group feature */
2049                 uninit_feat = 1;
2050
2051         /* number of inodes that have been allocated */
2052         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2053                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2054
2055         /* check quota and update in hash */
2056         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2057              group++) {
2058                 unsigned long used_count = sbi->s_inodes_per_group;
2059
2060                 if (uninit_feat) {
2061                         struct ext3_group_desc *desc;
2062                         desc = get_group_desc(sb, group, NULL);
2063                         if (!desc)
2064                                 GOTO(out, -EIO);
2065
2066                         /* we don't really need to take the group lock here,
2067                          * but it may be useful if one day we support online
2068                          * quotacheck */
2069 #ifdef HAVE_EXT4_LDISKFS
2070                         ext4_lock_group(sb, group);
2071 #else
2072                         spin_lock(sb_bgl_lock(sbi, group));
2073 #endif
2074                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2075                                 /* no inode in use in this group, just skip it */
2076 #ifdef HAVE_EXT4_LDISKFS
2077                                 ext3_unlock_group(sb, group);
2078 #else
2079                                 spin_unlock(sb_bgl_lock(sbi, group));
2080 #endif
2081                                 continue;
2082                         }
2083
2084                         used_count -= ext3_itable_unused_count(sb, desc);
2085 #ifdef HAVE_EXT4_LDISKFS
2086                         ext3_unlock_group(sb, group);
2087 #else
2088                         spin_unlock(sb_bgl_lock(sbi, group));
2089 #endif
2090                 }
2091
2092                 ino = group * sbi->s_inodes_per_group + 1;
2093                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2094                 if (!bitmap_bh) {
2095                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2096                                sb->s_id, group);
2097                         GOTO(out, -EIO);
2098                 }
2099
2100                 i = 0;
2101                 while (i < used_count &&
2102                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2103                                                used_count, i)) < used_count) {
2104                         inode_inuse--;
2105                         i++;
2106                         ino = i + group * sbi->s_inodes_per_group;
2107                         if (ino < sbi->s_first_ino)
2108                                 continue;
2109 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2110                         inode = ext3_iget(sb, ino);
2111 #else
2112                         inode = iget(sb, ino);
2113 #endif
2114                         if (!inode || IS_ERR(inode))
2115                                 continue;
2116
2117                         rc = add_inode_quota(inode, qctxt, oqc);
2118                         iput(inode);
2119                         if (rc) {
2120                                 brelse(bitmap_bh);
2121                                 GOTO(out, rc);
2122                         }
2123                 }
2124                 brelse(bitmap_bh);
2125         }
2126
2127         /* read old quota limits from old quota file. (only for the user
2128          * has limits but hasn't file) */
2129 #ifdef HAVE_QUOTA_SUPPORT
2130         for (i = 0; i < MAXQUOTAS; i++) {
2131                 cfs_list_t id_list;
2132                 struct dquot_id *dqid, *tmp;
2133
2134                 if (!Q_TYPESET(oqc, i))
2135                         continue;
2136
2137                 if (qctxt->qckt_first_check[i])
2138                         continue;
2139
2140
2141                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2142                 CFS_INIT_LIST_HEAD(&id_list);
2143 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2144                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2145 #else
2146                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2147 #endif
2148                 if (rc)
2149                         CERROR("read old limits failed. (rc:%d)\n", rc);
2150
2151                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2152                         cfs_list_del_init(&dqid->di_link);
2153
2154                         if (!rc)
2155                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2156                                       dqid->di_id, i,
2157                                       qctxt->qckt_first_check[i]);
2158                         OBD_FREE_PTR(dqid);
2159                 }
2160         }
2161 #endif
2162         /* turn off quota cause we are to dump chk_dqblk to files */
2163         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2164
2165         rc = create_new_quota_files(qctxt, oqc);
2166         if (rc)
2167                 GOTO(out, rc);
2168
2169         /* we use vfs functions to set dqblk, so turn quota on */
2170         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2171 out:
2172         /* dump and free chk_dqblk */
2173         rc = prune_chkquots(sb, qctxt, rc);
2174         OBD_FREE_PTR(qctxt);
2175
2176         /* turn off quota, `lfs quotacheck` will turn on when all
2177          * nodes quotacheck finish. */
2178         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2179
2180         oqc->qc_stat = rc;
2181         if (rc)
2182                 CERROR("quotacheck failed: rc = %d\n", rc);
2183
2184         RETURN(rc);
2185 }
2186
2187 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2188                                  int cmd)
2189 {
2190         int rc = 0;
2191         ENTRY;
2192
2193         if (lqi->qi_files[type] == NULL) {
2194                 CERROR("operate qinfo before it's enabled!\n");
2195                 RETURN(-ESRCH);
2196         }
2197
2198         switch (cmd) {
2199         case QFILE_CHK:
2200                 rc = lustre_check_quota_file(lqi, type);
2201                 break;
2202         case QFILE_RD_INFO:
2203                 rc = lustre_read_quota_info(lqi, type);
2204                 break;
2205         case QFILE_WR_INFO:
2206                 rc = lustre_write_quota_info(lqi, type);
2207                 break;
2208         case QFILE_INIT_INFO:
2209                 rc = lustre_init_quota_info(lqi, type);
2210                 break;
2211         case QFILE_CONVERT:
2212                 rc = -ENOTSUPP;
2213                 CERROR("quota CONVERT command is not supported\n");
2214                 break;
2215         default:
2216                 rc = -ENOTSUPP;
2217                 CERROR("Unsupported admin quota file cmd %d\n"
2218                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2219                        cmd);
2220                 break;
2221         }
2222         RETURN(rc);
2223 }
2224
2225 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2226                             cfs_list_t *list)
2227 {
2228         return lustre_get_qids(file, inode, type, list);
2229 }
2230
2231 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2232 {
2233         int rc = 0;
2234         ENTRY;
2235
2236         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2237                 CERROR("operate dquot before it's enabled!\n");
2238                 RETURN(-ESRCH);
2239         }
2240
2241         switch (cmd) {
2242         case QFILE_RD_DQUOT:
2243                 rc = lustre_read_dquot(dquot);
2244                 break;
2245         case QFILE_WR_DQUOT:
2246                 if (dquot->dq_dqb.dqb_ihardlimit ||
2247                     dquot->dq_dqb.dqb_isoftlimit ||
2248                     dquot->dq_dqb.dqb_bhardlimit ||
2249                     dquot->dq_dqb.dqb_bsoftlimit)
2250                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2251                 else
2252                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2253
2254                 rc = lustre_commit_dquot(dquot);
2255                 if (rc >= 0)
2256                         rc = 0;
2257                 break;
2258         default:
2259                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2260                 LBUG();
2261                 break;
2262         }
2263         RETURN(rc);
2264 }
2265
2266 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2267                                 struct inode *inode, int frags)
2268 {
2269 #ifdef EXT3_EXT_HAS_NO_TREE
2270         struct ext3_ext_base *base = inode;
2271 #else
2272         struct ext3_extents_tree tree;
2273         struct ext3_ext_base *base = &tree;
2274
2275         ext3_init_tree_desc(base, inode);
2276 #endif
2277         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2278          * metablocks at maxium b=16542 */
2279         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2280         return 0;
2281 }
2282
2283 #endif
2284
2285 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2286 {
2287         return (EXT3_SB(sb)->journal_bdev);
2288 }
2289 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2290
2291 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2292                           char *buf, int count, loff_t pos)
2293 {
2294         loff_t p = pos;
2295         int rc;
2296
2297         if (!f && !inode) {
2298                 CERROR("lustre_read_quota failed for no quota file!\n");
2299                 libcfs_debug_dumpstack(NULL);
2300                 return -EINVAL;
2301         }
2302
2303         /* Support for both adm and op quota files must be provided */
2304         if (f) {
2305                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2306                 rc = rc < 0 ? rc : p - pos;
2307         } else {
2308                 struct super_block *sb = inode->i_sb;
2309                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2310         }
2311         return rc;
2312 }
2313
2314 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2315 {
2316         loff_t p = pos;
2317         int rc;
2318
2319         /* Only adm quota files are supported, op updates are handled by vfs */
2320         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2321         rc = rc < 0 ? rc : p - pos;
2322
2323         return rc;
2324 }
2325
2326 void *lustre_quota_journal_start(struct inode *inode, int delete)
2327 {
2328         handle_t *handle;
2329         unsigned block_count;
2330
2331         if (delete) {
2332                 /* each indirect block (+4) may become free, attaching to the
2333                  * header list of free blocks (+1); the data block (+1) may
2334                  * become a free block (+0) or a block with free dqentries (+0) */
2335                 block_count = (4 + 1) + 1;
2336                 handle = ext3_journal_start(inode,
2337                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2338         } else {
2339                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2340                  * freeblk reusage with a header update (+1); dqentry is either reused
2341                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2342                  * a new block allocation (+1) with a header update (+1)              */
2343                 block_count = (4 + 1) + 3;
2344                 handle = ext3_journal_start(inode,
2345                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2346
2347         }
2348
2349         return handle;
2350 }
2351
2352 void lustre_quota_journal_stop(void *handle)
2353 {
2354         ext3_journal_stop((handle_t *)handle);
2355 }
2356
2357 static int ll_decode_fh_accept(void *context, struct dentry *de)
2358 {
2359         return 1;
2360 }
2361
2362 #ifdef HAVE_EXPORTFS_DECODE_FH
2363 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2364          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2365                             acceptable, context)
2366 #else
2367 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2368          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2369                                      type, acceptable, context)
2370 # define FILEID_INO32_GEN 1
2371 extern struct export_operations export_op_default;
2372 #endif
2373
2374 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2375                                       struct fsfilt_fid *fid, int ignore_gen)
2376 {
2377         struct inode  *inode;
2378         struct dentry *result;
2379
2380         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2381                                        ll_decode_fh_accept, NULL);
2382         if (IS_ERR(result)) {
2383                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2384                        fid->ino, fid->gen, PTR_ERR(result));
2385                 return result;
2386         }
2387
2388         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2389                fid->ino, fid->gen);
2390         inode = result->d_inode;
2391         if (inode == NULL)
2392                 goto err_out;
2393
2394         if (inode->i_nlink == 0 &&
2395             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2396                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2397                               " ctime -- this may indicate disk "
2398                               "corruption (inode: %lu, link: %lu, "
2399                               "count: %d)\n", inode->i_ino,
2400                               (unsigned long)inode->i_nlink,
2401                               atomic_read(&inode->i_count));
2402                 goto err_out;
2403         }
2404         if (fid->gen && inode->i_generation != fid->gen) {
2405                 /* we didn't find the right inode.. */
2406                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2407                        "count: %d, generation %u/%u\n",
2408                        inode->i_ino, (unsigned long)inode->i_nlink,
2409                        atomic_read(&inode->i_count), inode->i_generation,
2410                        fid->gen);
2411                 goto err_out;
2412         }
2413
2414         return result;
2415 err_out:
2416         l_dput(result);
2417         return ERR_PTR(-ENOENT);
2418 }
2419
2420 static struct fsfilt_operations fsfilt_ext3_ops = {
2421         .fs_type                = "ext3",
2422         .fs_owner               = THIS_MODULE,
2423         .fs_getlabel            = fsfilt_ext3_get_label,
2424         .fs_setlabel            = fsfilt_ext3_set_label,
2425         .fs_uuid                = fsfilt_ext3_uuid,
2426         .fs_start               = fsfilt_ext3_start,
2427         .fs_brw_start           = fsfilt_ext3_brw_start,
2428         .fs_extend              = fsfilt_ext3_extend,
2429         .fs_commit              = fsfilt_ext3_commit,
2430         .fs_commit_async        = fsfilt_ext3_commit_async,
2431         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2432         .fs_setattr             = fsfilt_ext3_setattr,
2433         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2434         .fs_set_md              = fsfilt_ext3_set_md,
2435         .fs_get_md              = fsfilt_ext3_get_md,
2436         .fs_readpage            = fsfilt_ext3_readpage,
2437         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2438         .fs_statfs              = fsfilt_ext3_statfs,
2439         .fs_sync                = fsfilt_ext3_sync,
2440         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2441         .fs_write_record        = fsfilt_ext3_write_record,
2442         .fs_read_record         = fsfilt_ext3_read_record,
2443         .fs_setup               = fsfilt_ext3_setup,
2444         .fs_send_bio            = fsfilt_ext3_send_bio,
2445         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2446 #ifdef HAVE_DISK_INODE_VERSION
2447         .fs_get_version         = fsfilt_ext3_get_version,
2448         .fs_set_version         = fsfilt_ext3_set_version,
2449 #endif
2450 #ifdef HAVE_QUOTA_SUPPORT
2451         .fs_quotactl            = fsfilt_ext3_quotactl,
2452         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2453         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2454         .fs_qids                = fsfilt_ext3_qids,
2455         .fs_dquot               = fsfilt_ext3_dquot,
2456         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2457 #endif
2458         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2459         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2460 };
2461
2462 static int __init fsfilt_ext3_init(void)
2463 {
2464         int rc;
2465
2466         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2467                                          sizeof(struct fsfilt_cb_data), 0, 0);
2468         if (!fcb_cache) {
2469                 CERROR("error allocating fsfilt journal callback cache\n");
2470                 GOTO(out, rc = -ENOMEM);
2471         }
2472
2473         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2474
2475         if (rc) {
2476                 int err = cfs_mem_cache_destroy(fcb_cache);
2477                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2478         }
2479 out:
2480         return rc;
2481 }
2482
2483 static void __exit fsfilt_ext3_exit(void)
2484 {
2485         int rc;
2486
2487         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2488         rc = cfs_mem_cache_destroy(fcb_cache);
2489         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2490 }
2491
2492 module_init(fsfilt_ext3_init);
2493 module_exit(fsfilt_ext3_exit);
2494
2495 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2496 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2497 MODULE_LICENSE("GPL");