lustre/lvfs/fsfilt_ext3.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  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 #include <ext4/ext4.h>
  53 #include <ext4/ext4_jbd2.h>
  54 #include <linux/version.h>
  55 #include <linux/bitops.h>
  56 #include <linux/quota.h>
  57 #ifdef HAVE_QUOTAIO_H
  58 # include <linux/quotaio_v2.h>
  59 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  60 # include <quota/quotaio_v2.h>
  61 # include <quota/quota_tree.h>
  62 # define V2_DQTREEOFF    QT_TREEOFF
  63 #elif defined(HAVE_FS_QUOTAIO_H)
  64 # include <quotaio_v2.h>
  65 # include <quota_tree.h>
  66 # define V2_DQTREEOFF    QT_TREEOFF
  67 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  68 #endif
  69
  70 #ifdef QFMT_VFS_V1
  71 #define QFMT_LUSTRE QFMT_VFS_V1
  72 #else
  73 #define QFMT_LUSTRE QFMT_VFS_V0
  74 #endif
  75
  76 #if defined(HAVE_EXT3_XATTR_H)
  77 # include <ext3/xattr.h>
  78 #elif !defined(EXT3_XATTR_INDEX_TRUSTED)
  79 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  80 /* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
  81 # define EXT3_XATTR_INDEX_TRUSTED        4
  82 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  83 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  84 #endif
  85
  86 #include <libcfs/libcfs.h>
  87 #include <lustre_fsfilt.h>
  88 #include <obd.h>
  89 #include <lustre_quota.h>
  90 #include <linux/lustre_compat25.h>
  91 #include <linux/lprocfs_status.h>
  92
  93 #include <ext4/ext4_extents.h>
  94
  95 #include "lustre_quota_fmt.h"
  96
  97 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  98 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  99 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 100 #else
 101 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 102 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 103 #endif
 104
 105 /* for kernels 2.6.18 and later */
 106 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 107
 108 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 109                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 110
 111 #define ext3_mb_discard_inode_preallocations(inode) \
 112                  ext3_discard_preallocations(inode)
 113
 114 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 115 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 116
 117 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 118 # define journal_callback ext4_journal_cb_entry
 119 # define fsfilt_journal_callback_set(handle, func, jcb) \
 120          ext4_journal_callback_add(handle, func, jcb)
 121 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
 122 # define fsfilt_journal_callback_set(handle, func, jcb) \
 123          jbd2_journal_callback_set(handle, func, jcb)
 124 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
 125 # define fsfilt_journal_callback_set(handle, func, jcb) \
 126          journal_callback_set(handle, func, jcb)
 127 #else
 128 # error missing journal commit callback
 129 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
 130
 131 static cfs_mem_cache_t *fcb_cache;
 132
 133 struct fsfilt_cb_data {
 134         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 135         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 136         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 137         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 138         void *cb_data;                  /* MDS/OST completion function data */
 139 };
 140
 141 #ifndef ext3_find_next_bit
 142 #define ext3_find_next_bit           ext2_find_next_bit
 143 #endif
 144
 145 #ifndef ext2_find_next_bit
 146 #ifdef __LITTLE_ENDIAN
 147 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 148 #else
 149 error "Need implementation of find_next_bit on big-endian systems"
 150 #endif  /* __LITTLE_ENDIAN */
 151 #endif  /* !ext2_find_next_le_bit */
 152
 153 static char *fsfilt_ext3_get_label(struct super_block *sb)
 154 {
 155         return EXT3_SB(sb)->s_es->s_volume_name;
 156 }
 157
 158 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 159 {
 160         /* see e.g. fsfilt_ext3_write_record() */
 161         journal_t *journal;
 162         handle_t *handle;
 163         int err;
 164
 165         journal = EXT3_SB(sb)->s_journal;
 166         handle = ext3_journal_start_sb(sb, 1);
 167         if (IS_ERR(handle)) {
 168                 CERROR("can't start transaction\n");
 169                 return(PTR_ERR(handle));
 170         }
 171
 172         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 173         if (err)
 174                 goto out;
 175
 176         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 177                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 178
 179         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 180
 181 out:
 182         ext3_journal_stop(handle);
 183
 184         return(err);
 185 }
 186
 187 static char *fsfilt_ext3_uuid(struct super_block *sb)
 188 {
 189         return EXT3_SB(sb)->s_es->s_uuid;
 190 }
 191
 192 #ifdef HAVE_DISK_INODE_VERSION
 193
 194 static __u64 get_i_version(struct inode *inode)
 195 {
 196         return EXT3_I(inode)->i_fs_version;
 197 }
 198
 199 static void set_i_version(struct inode *inode, __u64 new_version)
 200 {
 201         (EXT3_I(inode))->i_fs_version = new_version;
 202 }
 203
 204 /*
 205  * Get the 64-bit version for an inode.
 206  */
 207 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 208 {
 209         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 210                get_i_version(inode), inode->i_ino);
 211         return get_i_version(inode);
 212 }
 213
 214 /*
 215  * Set the 64-bit version and return the old version.
 216  */
 217 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 218 {
 219         __u64 old_version = get_i_version(inode);
 220
 221         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 222                new_version, old_version, inode->i_ino);
 223         set_i_version(inode, new_version);
 224         /* version is set after all inode operations are finished, so we should
 225          * mark it dirty here */
 226         inode->i_sb->s_op->dirty_inode(inode);
 227         return old_version;
 228 }
 229
 230 #endif
 231
 232 /*
 233  * We don't currently need any additional blocks for rmdir and
 234  * unlink transactions because we are storing the OST oa_id inside
 235  * the inode (which we will be changing anyways as part of this
 236  * transaction).
 237  */
 238 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 239                                int logs)
 240 {
 241         /* For updates to the last received file */
 242         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 243         journal_t *journal;
 244         void *handle;
 245
 246         if (current->journal_info) {
 247                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 248                        current->journal_info);
 249                 goto journal_start;
 250         }
 251
 252         switch(op) {
 253         case FSFILT_OP_RMDIR:
 254         case FSFILT_OP_UNLINK:
 255                 /* delete one file + create/update logs for each stripe */
 256                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 257                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 258                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 259                 break;
 260         case FSFILT_OP_RENAME:
 261                 /* modify additional directory */
 262                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 263                 /* no break */
 264         case FSFILT_OP_SYMLINK:
 265                 /* additional block + block bitmap + GDT for long symlink */
 266                 nblocks += 3;
 267                 /* no break */
 268         case FSFILT_OP_CREATE: {
 269                 /* no break */
 270         }
 271         case FSFILT_OP_MKDIR:
 272         case FSFILT_OP_MKNOD:
 273                 /* modify one inode + block bitmap + GDT */
 274                 nblocks += 3;
 275                 /* no break */
 276         case FSFILT_OP_LINK:
 277                 /* modify parent directory */
 278                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 279                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 280                 /* create/update logs for each stripe */
 281                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 282                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 283                 break;
 284         case FSFILT_OP_SETATTR:
 285                 /* Setattr on inode */
 286                 nblocks += 1;
 287                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 288                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 289                 /* quota chown log for each stripe */
 290                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 291                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 292                 break;
 293         case FSFILT_OP_CANCEL_UNLINK:
 294                 /* blocks for log header bitmap update OR
 295                  * blocks for catalog header bitmap update + unlink of logs */
 296                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 297                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 298                 break;
 299         default: CERROR("unknown transaction start op %d\n", op);
 300                 LBUG();
 301         }
 302
 303         LASSERT(current->journal_info == desc_private);
 304         journal = EXT3_SB(inode->i_sb)->s_journal;
 305         if (nblocks > journal->j_max_transaction_buffers) {
 306                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 307                        nblocks, op, logs, journal->j_max_transaction_buffers);
 308                 nblocks = journal->j_max_transaction_buffers;
 309         }
 310
 311  journal_start:
 312         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 313         handle = ext3_journal_start(inode, nblocks);
 314
 315         if (!IS_ERR(handle))
 316                 LASSERT(current->journal_info == handle);
 317         else
 318                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 319                        op, nblocks, PTR_ERR(handle));
 320         return handle;
 321 }
 322
 323 /*
 324  * Calculate the number of buffer credits needed to write multiple pages in
 325  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 326  * doesn't have a nice API for calculating this sort of thing in advance.
 327  *
 328  * See comment above ext3_writepage_trans_blocks for details.  We assume
 329  * no data journaling is being done, but it does allow for all of the pages
 330  * being non-contiguous.  If we are guaranteed contiguous pages we could
 331  * reduce the number of (d)indirect blocks a lot.
 332  *
 333  * With N blocks per page and P pages, for each inode we have at most:
 334  * N*P indirect
 335  * min(N*P, blocksize/4 + 1) dindirect blocks
 336  * niocount tindirect
 337  *
 338  * For the entire filesystem, we have at most:
 339  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 340  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 341  * objcount inode blocks
 342  * 1 superblock
 343  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 344  *
 345  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 346  */
 347 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 348                                       int niocount, struct niobuf_local *nb)
 349 {
 350         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 351         __u64 next_indir;
 352         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 353         int nbitmaps = 0, ngdblocks;
 354         int needed = objcount + 1; /* inodes + superblock */
 355         int i, j;
 356
 357         for (i = 0, j = 0; i < objcount; i++, fso++) {
 358                 /* two or more dindirect blocks in case we cross boundary */
 359                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 360                                     nb[j].offset) >>
 361                                    sb->s_blocksize_bits) /
 362                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 363                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 364
 365                 /* leaf, indirect, tindirect blocks for first block */
 366                 nbitmaps += blockpp + 2;
 367
 368                 j += fso->fso_bufcnt;
 369         }
 370
 371         next_indir = nb[0].offset +
 372                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 373         for (i = 1; i < niocount; i++) {
 374                 if (nb[i].offset >= next_indir) {
 375                         nbitmaps++;     /* additional indirect */
 376                         next_indir = nb[i].offset +
 377                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 378                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 379                         nbitmaps++;     /* additional indirect */
 380                 }
 381                 nbitmaps += blockpp;    /* each leaf in different group? */
 382         }
 383
 384         ngdblocks = nbitmaps;
 385         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 386                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 387         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 388                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 389
 390         needed += nbitmaps + ngdblocks;
 391
 392         /* last_rcvd update */
 393         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 394
 395 #if defined(CONFIG_QUOTA)
 396         /* We assume that there will be 1 bit set in s_dquot.flags for each
 397          * quota file that is active.  This is at least true for now.
 398          */
 399         needed += hweight32(ll_sb_any_quota_active(sb)) *
 400                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 401 #endif
 402
 403         return needed;
 404 }
 405
 406 /* We have to start a huge journal transaction here to hold all of the
 407  * metadata for the pages being written here.  This is necessitated by
 408  * the fact that we do lots of prepare_write operations before we do
 409  * any of the matching commit_write operations, so even if we split
 410  * up to use "smaller" transactions none of them could complete until
 411  * all of them were opened.  By having a single journal transaction,
 412  * we eliminate duplicate reservations for common blocks like the
 413  * superblock and group descriptors or bitmaps.
 414  *
 415  * We will start the transaction here, but each prepare_write will
 416  * add a refcount to the transaction, and each commit_write will
 417  * remove a refcount.  The transaction will be closed when all of
 418  * the pages have been written.
 419  */
 420 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 421                                    int niocount, struct niobuf_local *nb,
 422                                    void *desc_private, int logs)
 423 {
 424         journal_t *journal;
 425         handle_t *handle;
 426         int needed;
 427         ENTRY;
 428
 429         LASSERT(current->journal_info == desc_private);
 430         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 431         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 432
 433         /* The number of blocks we could _possibly_ dirty can very large.
 434          * We reduce our request if it is absurd (and we couldn't get that
 435          * many credits for a single handle anyways).
 436          *
 437          * At some point we have to limit the size of I/Os sent at one time,
 438          * increase the size of the journal, or we have to calculate the
 439          * actual journal requirements more carefully by checking all of
 440          * the blocks instead of being maximally pessimistic.  It remains to
 441          * be seen if this is a real problem or not.
 442          */
 443         if (needed > journal->j_max_transaction_buffers) {
 444                 CERROR("want too many journal credits (%d) using %d instead\n",
 445                        needed, journal->j_max_transaction_buffers);
 446                 needed = journal->j_max_transaction_buffers;
 447         }
 448
 449         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 450         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 451         if (IS_ERR(handle)) {
 452                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 453                        PTR_ERR(handle));
 454         } else {
 455                 LASSERT(handle->h_buffer_credits >= needed);
 456                 LASSERT(current->journal_info == handle);
 457         }
 458
 459         RETURN(handle);
 460 }
 461
 462 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 463 {
 464        handle_t *handle = h;
 465
 466        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 467        if (nblocks == 0) {
 468                handle->h_buffer_credits = 0;
 469                CWARN("setting credits of handle %p to zero by request\n", h);
 470        }
 471
 472        if (handle->h_buffer_credits > nblocks)
 473                 return 0;
 474        if (ext3_journal_extend(handle, nblocks) == 0)
 475                 return 0;
 476
 477        ext3_mark_inode_dirty(handle, inode);
 478        return ext3_journal_restart(handle, nblocks);
 479 }
 480
 481 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 482 {
 483         int rc;
 484         handle_t *handle = h;
 485
 486         LASSERT(current->journal_info == handle);
 487         if (force_sync)
 488                 handle->h_sync = 1; /* recovery likes this */
 489
 490         rc = ext3_journal_stop(handle);
 491
 492         return rc;
 493 }
 494
 495 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 496                                     void **wait_handle)
 497 {
 498         unsigned long tid;
 499         transaction_t *transaction;
 500         handle_t *handle = h;
 501         journal_t *journal;
 502         int rc;
 503
 504         LASSERT(current->journal_info == handle);
 505
 506         transaction = handle->h_transaction;
 507         journal = transaction->t_journal;
 508         tid = transaction->t_tid;
 509         /* we don't want to be blocked */
 510         handle->h_sync = 0;
 511         rc = ext3_journal_stop(handle);
 512         if (rc) {
 513                 CERROR("error while stopping transaction: %d\n", rc);
 514                 return rc;
 515         }
 516         fsfilt_log_start_commit(journal, tid);
 517
 518         *wait_handle = (void *) tid;
 519         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 520         return 0;
 521 }
 522
 523 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 524 {
 525         journal_t *journal = EXT3_JOURNAL(inode);
 526         tid_t tid = (tid_t)(long)h;
 527
 528         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 529         if (unlikely(is_journal_aborted(journal)))
 530                 return -EIO;
 531
 532         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 533
 534         if (unlikely(is_journal_aborted(journal)))
 535                 return -EIO;
 536         return 0;
 537 }
 538
 539 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 540                                struct iattr *iattr, int do_trunc)
 541 {
 542         struct inode *inode = dentry->d_inode;
 543         int rc = 0;
 544
 545 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
 546         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
 547          * timestamps to appear to be in the far future, due old timestamp
 548          * being stored on disk as an unsigned value.  This fixes up any
 549          * bad values held by the client before storing them on disk,
 550          * and ensures any timestamp updates are correct.  LU-1042 */
 551         if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
 552                      !(iattr->ia_valid & ATTR_ATIME))) {
 553                 iattr->ia_valid |= ATTR_ATIME;
 554                 LTIME_S(iattr->ia_atime) = 0;
 555         }
 556         if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
 557                      !(iattr->ia_valid & ATTR_MTIME))) {
 558                 iattr->ia_valid |= ATTR_MTIME;
 559                 LTIME_S(iattr->ia_mtime) = 0;
 560         }
 561         if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
 562                       LTIME_S(inode->i_ctime) == 0) &&
 563                      !(iattr->ia_valid & ATTR_CTIME))) {
 564                 iattr->ia_valid |= ATTR_CTIME;
 565                 LTIME_S(iattr->ia_ctime) = 0;
 566         }
 567 #else
 568 #warning "remove old LU-221/LU-1042 workaround code"
 569 #endif
 570
 571         /* When initializating timestamps for new inodes, use the filesystem
 572          * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
 573          * that this is potentially an invalid inode.  Files with an old ctime
 574          * migrated to a newly-formatted OST with a newer s_mkfs_time will not
 575          * hit this check, since it is only for ctime == 0.  LU-1010/LU-1042 */
 576         if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
 577                 LTIME_S(iattr->ia_ctime) =
 578                         EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
 579
 580         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 581          * We are already writing the inode to disk as part of this
 582          * transaction and want to avoid a lot of extra inode writeout
 583          * later on. b=9828 */
 584         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 585                 /* ATTR_SIZE would invoke truncate: clear it */
 586                 iattr->ia_valid &= ~ATTR_SIZE;
 587                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 588                 i_size_write(inode, iattr->ia_size);
 589
 590                 if (iattr->ia_valid & ATTR_UID)
 591                         inode->i_uid = iattr->ia_uid;
 592                 if (iattr->ia_valid & ATTR_GID)
 593                         inode->i_gid = iattr->ia_gid;
 594                 if (iattr->ia_valid & ATTR_ATIME)
 595                         inode->i_atime = iattr->ia_atime;
 596                 if (iattr->ia_valid & ATTR_MTIME)
 597                         inode->i_mtime = iattr->ia_mtime;
 598                 if (iattr->ia_valid & ATTR_CTIME)
 599                         inode->i_ctime = iattr->ia_ctime;
 600                 if (iattr->ia_valid & ATTR_MODE) {
 601                         inode->i_mode = iattr->ia_mode;
 602
 603                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 604                             !cfs_capable(CFS_CAP_FSETID))
 605                                 inode->i_mode &= ~S_ISGID;
 606                 }
 607
 608                 inode->i_sb->s_op->dirty_inode(inode);
 609
 610                 goto out;
 611         }
 612
 613         /* Don't allow setattr to change file type */
 614         if (iattr->ia_valid & ATTR_MODE)
 615                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 616                                  (iattr->ia_mode & ~S_IFMT);
 617
 618         /* We set these flags on the client, but have already checked perms
 619          * so don't confuse inode_change_ok. */
 620         iattr->ia_valid &= ~TIMES_SET_FLAGS;
 621
 622         if (inode->i_op->setattr) {
 623                 rc = inode->i_op->setattr(dentry, iattr);
 624         } else {
 625 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
 626                 rc = inode_change_ok(inode, iattr);
 627                 if (!rc)
 628 #endif
 629                         rc = simple_setattr(dentry, iattr);
 630         }
 631
 632  out:
 633         RETURN(rc);
 634 }
 635
 636 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 637                                  unsigned int cmd, unsigned long arg)
 638 {
 639         int rc = 0;
 640         ENTRY;
 641
 642         /* FIXME: Can't do this because of nested transaction deadlock */
 643         if (cmd == EXT3_IOC_SETFLAGS) {
 644                 /* We can't enable data journaling on OST objects, because
 645                 * this forces the transaction to be closed in order to
 646                 * flush the journal, but the caller will already have a
 647                 * compound transaction open to update the last_rcvd file,
 648                 * and this thread would deadlock trying to set the flag. */
 649                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 650                         CERROR("can't set data journal flag on file\n");
 651                         RETURN(-EPERM);
 652                 }
 653                 /* Because the MDS does not see the EXTENTS_FL set on the
 654                  * OST objects, mask this flag into all set flags.  It is
 655                  * not legal to clear this flag in any case, so we are not
 656                  * changing the functionality by doing this.  b=22911 */
 657                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 658         }
 659
 660         /* ext4_ioctl does not have a inode argument */
 661         if (inode->i_fop->unlocked_ioctl)
 662                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 663         else
 664                 RETURN(-ENOTTY);
 665
 666         RETURN(rc);
 667 }
 668
 669 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 670                               void *lmm, int lmm_size, const char *name)
 671 {
 672         int rc;
 673
 674         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 675
 676         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 677                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 678
 679         if (rc && rc != -EROFS)
 680                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 681                        inode->i_ino, rc);
 682         return rc;
 683 }
 684
 685 /* Must be called with i_mutex held */
 686 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 687                               const char *name)
 688 {
 689         int rc;
 690
 691         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 692
 693         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 694                             name, lmm, lmm_size);
 695
 696         /* This gives us the MD size */
 697         if (lmm == NULL)
 698                 return (rc == -ENODATA) ? 0 : rc;
 699
 700         if (rc < 0) {
 701                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 702                        EXT3_XATTR_INDEX_TRUSTED, name,
 703                        inode->i_ino, rc);
 704                 memset(lmm, 0, lmm_size);
 705                 return (rc == -ENODATA) ? 0 : rc;
 706         }
 707
 708         return rc;
 709 }
 710
 711 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 712 {
 713         submit_bio(rw, bio);
 714         return 0;
 715 }
 716
 717 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 718                                     loff_t *off)
 719 {
 720         struct inode *inode = file->f_dentry->d_inode;
 721         int rc = 0;
 722
 723         if (S_ISREG(inode->i_mode))
 724                 rc = file->f_op->read(file, buf, count, off);
 725         else {
 726                 const int blkbits = inode->i_sb->s_blocksize_bits;
 727                 const int blksize = inode->i_sb->s_blocksize;
 728
 729                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 730                        (unsigned long)count, inode->i_ino, *off);
 731                 while (count > 0) {
 732                         struct buffer_head *bh;
 733
 734                         bh = NULL;
 735                         if (*off < i_size_read(inode)) {
 736                                 int err = 0;
 737
 738                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 739                                                 0, &err);
 740
 741                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 742
 743                                 if (bh) {
 744                                         memcpy(buf, bh->b_data, blksize);
 745                                         brelse(bh);
 746                                 } else if (err) {
 747                                         /* XXX in theory we should just fake
 748                                          * this buffer and continue like ext3,
 749                                          * especially if this is a partial read
 750                                          */
 751                                         CERROR("error read dir %lu+%llu: %d\n",
 752                                                inode->i_ino, *off, err);
 753                                         RETURN(err);
 754                                 }
 755                         }
 756                         if (!bh) {
 757                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 758
 759                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 760                                 memset(fake, 0, sizeof(*fake));
 761                                 fake->rec_len = cpu_to_le16(blksize);
 762                         }
 763                         count -= blksize;
 764                         buf += blksize;
 765                         *off += blksize;
 766                         rc += blksize;
 767                 }
 768         }
 769
 770         return rc;
 771 }
 772
 773 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 774 static void fsfilt_ext3_cb_func(struct super_block *sb,
 775                                 struct journal_callback *jcb, int error)
 776 #else
 777 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 778 #endif
 779 {
 780         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 781
 782         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 783
 784         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 785 }
 786
 787 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 788                                       void *handle, fsfilt_cb_t cb_func,
 789                                       void *cb_data)
 790 {
 791         struct fsfilt_cb_data *fcb;
 792
 793         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 794         if (fcb == NULL)
 795                 RETURN(-ENOMEM);
 796
 797         fcb->cb_func = cb_func;
 798         fcb->cb_obd = obd;
 799         fcb->cb_last_rcvd = last_rcvd;
 800         fcb->cb_data = cb_data;
 801
 802         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 803         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 804
 805         return 0;
 806 }
 807
 808 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 809 {
 810         struct kstatfs sfs;
 811         int rc;
 812
 813         memset(&sfs, 0, sizeof(sfs));
 814         rc = ll_do_statfs(sb, &sfs);
 815         statfs_pack(osfs, &sfs);
 816         return rc;
 817 }
 818
 819 static int fsfilt_ext3_sync(struct super_block *sb)
 820 {
 821         return ext3_force_commit(sb);
 822 }
 823
 824 #ifndef EXT3_EXTENTS_FL
 825 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 826 #endif
 827
 828 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 829 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 830 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 831 #else
 832 # define fsfilt_up_truncate_sem(inode) do{ }while(0)
 833 # define fsfilt_down_truncate_sem(inode) do{ }while(0)
 834 #endif
 835
 836 #ifndef EXT_ASSERT
 837 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 838 #endif
 839
 840 #ifdef EXT3_EXT_HAS_NO_TREE
 841 /* for kernels 2.6.18 and later */
 842 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 843 #define ext3_ext_base                   inode
 844 #define ext3_ext_base2inode(inode)      (inode)
 845 #define EXT_DEPTH(inode)                ext_depth(inode)
 846 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 847                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 848 #else
 849 #define ext3_ext_base                   ext3_extents_tree
 850 #define ext3_ext_base2inode(tree)       (tree->inode)
 851 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 852                         ext3_ext_walk_space(tree, block, num, cb);
 853 #endif
 854
 855 struct bpointers {
 856         unsigned long *blocks;
 857         int *created;
 858         unsigned long start;
 859         int num;
 860         int init_num;
 861         int create;
 862 };
 863
 864 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 865                                unsigned long block, int *aflags)
 866 {
 867         struct ext3_inode_info *ei = EXT3_I(inode);
 868         unsigned long bg_start;
 869         unsigned long colour;
 870         int depth;
 871
 872         if (path) {
 873                 struct ext3_extent *ex;
 874                 depth = path->p_depth;
 875
 876                 /* try to predict block placement */
 877                 if ((ex = path[depth].p_ext))
 878                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 879
 880                 /* it looks index is empty
 881                  * try to find starting from index itself */
 882                 if (path[depth].p_bh)
 883                         return path[depth].p_bh->b_blocknr;
 884         }
 885
 886         /* OK. use inode's group */
 887         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 888                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 889         colour = (current->pid % 16) *
 890                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 891         return bg_start + colour + block;
 892 }
 893
 894 #define ll_unmap_underlying_metadata(sb, blocknr) \
 895         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 896
 897 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 898 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 899                                 struct ext3_ext_path *path, unsigned long block,
 900                                 unsigned long *count, int *err)
 901 {
 902         unsigned long pblock, goal;
 903         int aflags = 0;
 904         struct inode *inode = ext3_ext_base2inode(base);
 905
 906         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 907         aflags |= 2; /* block have been already reserved */
 908         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 909         return pblock;
 910
 911 }
 912 #else
 913 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 914                                 struct ext3_ext_path *path, unsigned long block,
 915                                 unsigned long *count, int *err)
 916 {
 917         struct inode *inode = ext3_ext_base2inode(base);
 918         struct ext3_allocation_request ar;
 919         unsigned long pblock;
 920         int aflags;
 921
 922         /* find neighbour allocated blocks */
 923         ar.lleft = block;
 924         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 925         if (*err)
 926                 return 0;
 927         ar.lright = block;
 928         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 929         if (*err)
 930                 return 0;
 931
 932         /* allocate new block */
 933         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 934         ar.inode = inode;
 935         ar.logical = block;
 936         ar.len = *count;
 937         ar.flags = EXT3_MB_HINT_DATA;
 938         pblock = ext3_mb_new_blocks(handle, &ar, err);
 939         *count = ar.len;
 940         return pblock;
 941 }
 942 #endif
 943
 944 #ifdef EXT3_EXT_HAS_NO_TREE
 945 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 946                                   struct ext3_ext_path *path,
 947                                   struct ext3_ext_cache *cex,
 948 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 949                                    struct ext3_extent *ex,
 950 #endif
 951                                   void *cbdata)
 952 {
 953         struct bpointers *bp = cbdata;
 954 #else
 955 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 956                                   struct ext3_ext_path *path,
 957                                   struct ext3_ext_cache *cex
 958 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 959                                   , struct ext3_extent *ex
 960 #endif
 961                                  )
 962 {
 963         struct bpointers *bp = base->private;
 964 #endif
 965         struct inode *inode = ext3_ext_base2inode(base);
 966         struct ext3_extent nex;
 967         unsigned long pblock;
 968         unsigned long tgen;
 969         int err, i;
 970         unsigned long count;
 971         handle_t *handle;
 972
 973         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 974                 err = EXT_CONTINUE;
 975                 goto map;
 976         }
 977
 978         if (bp->create == 0) {
 979                 i = 0;
 980                 if (cex->ec_block < bp->start)
 981                         i = bp->start - cex->ec_block;
 982                 if (i >= cex->ec_len)
 983                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 984                                         i, cex->ec_len);
 985                 for (; i < cex->ec_len && bp->num; i++) {
 986                         *(bp->created) = 0;
 987                         bp->created++;
 988                         *(bp->blocks) = 0;
 989                         bp->blocks++;
 990                         bp->num--;
 991                         bp->start++;
 992                 }
 993
 994                 return EXT_CONTINUE;
 995         }
 996
 997         tgen = EXT_GENERATION(base);
 998         count = ext3_ext_calc_credits_for_insert(base, path);
 999         fsfilt_up_truncate_sem(inode);
1000
1001         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1002         if (IS_ERR(handle)) {
1003                 fsfilt_down_truncate_sem(inode);
1004                 return PTR_ERR(handle);
1005         }
1006
1007         fsfilt_down_truncate_sem(inode);
1008         if (tgen != EXT_GENERATION(base)) {
1009                 /* the tree has changed. so path can be invalid at moment */
1010                 ext3_journal_stop(handle);
1011                 return EXT_REPEAT;
1012         }
1013
1014         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1015          * protected by i_data_sem as whole. so we patch it to store
1016          * generation to path and now verify the tree hasn't changed */
1017         down_write((&EXT4_I(inode)->i_data_sem));
1018
1019         /* validate extent, make sure the extent tree does not changed */
1020         if (EXT_GENERATION(base) != path[0].p_generation) {
1021                 /* cex is invalid, try again */
1022                 up_write(&EXT4_I(inode)->i_data_sem);
1023                 ext3_journal_stop(handle);
1024                 return EXT_REPEAT;
1025         }
1026
1027         count = cex->ec_len;
1028         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1029         if (!pblock)
1030                 goto out;
1031         EXT_ASSERT(count <= cex->ec_len);
1032
1033         /* insert new extent */
1034         nex.ee_block = cpu_to_le32(cex->ec_block);
1035         ext3_ext_store_pblock(&nex, pblock);
1036         nex.ee_len = cpu_to_le16(count);
1037         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1038         if (err) {
1039                 /* free data blocks we just allocated */
1040                 /* not a good idea to call discard here directly,
1041                  * but otherwise we'd need to call it every free() */
1042 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1043                 ext3_mb_discard_inode_preallocations(inode);
1044 #endif
1045                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1046                                  cpu_to_le16(nex.ee_len), 0);
1047                 goto out;
1048         }
1049
1050         /*
1051          * Putting len of the actual extent we just inserted,
1052          * we are asking ext3_ext_walk_space() to continue
1053          * scaning after that block
1054          */
1055         cex->ec_len = le16_to_cpu(nex.ee_len);
1056         cex->ec_start = ext_pblock(&nex);
1057         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1058         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1059
1060 out:
1061         up_write((&EXT4_I(inode)->i_data_sem));
1062         ext3_journal_stop(handle);
1063 map:
1064         if (err >= 0) {
1065                 /* map blocks */
1066                 if (bp->num == 0) {
1067                         CERROR("hmm. why do we find this extent?\n");
1068                         CERROR("initial space: %lu:%u\n",
1069                                 bp->start, bp->init_num);
1070                         CERROR("current extent: %u/%u/%llu %d\n",
1071                                 cex->ec_block, cex->ec_len,
1072                                 (unsigned long long)cex->ec_start,
1073                                 cex->ec_type);
1074                 }
1075                 i = 0;
1076                 if (cex->ec_block < bp->start)
1077                         i = bp->start - cex->ec_block;
1078                 if (i >= cex->ec_len)
1079                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1080                                         i, cex->ec_len);
1081                 for (; i < cex->ec_len && bp->num; i++) {
1082                         *(bp->blocks) = cex->ec_start + i;
1083                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1084                                 *(bp->created) = 0;
1085                         } else {
1086                                 *(bp->created) = 1;
1087                                 /* unmap any possible underlying metadata from
1088                                  * the block device mapping.  bug 6998. */
1089                                 ll_unmap_underlying_metadata(inode->i_sb,
1090                                                              *(bp->blocks));
1091                         }
1092                         bp->created++;
1093                         bp->blocks++;
1094                         bp->num--;
1095                         bp->start++;
1096                 }
1097         }
1098         return err;
1099 }
1100
1101 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1102                        unsigned long num, unsigned long *blocks,
1103                        int *created, int create)
1104 {
1105 #ifdef EXT3_EXT_HAS_NO_TREE
1106         struct ext3_ext_base *base = inode;
1107 #else
1108         struct ext3_extents_tree tree;
1109         struct ext3_ext_base *base = &tree;
1110 #endif
1111         struct bpointers bp;
1112         int err;
1113
1114         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1115                block, block + num - 1, (unsigned) inode->i_ino);
1116
1117 #ifndef EXT3_EXT_HAS_NO_TREE
1118         ext3_init_tree_desc(base, inode);
1119         tree.private = &bp;
1120 #endif
1121         bp.blocks = blocks;
1122         bp.created = created;
1123         bp.start = block;
1124         bp.init_num = bp.num = num;
1125         bp.create = create;
1126
1127         fsfilt_down_truncate_sem(inode);
1128         err = fsfilt_ext3_ext_walk_space(base, block, num,
1129                                          ext3_ext_new_extent_cb, &bp);
1130         ext3_ext_invalidate_cache(base);
1131         fsfilt_up_truncate_sem(inode);
1132
1133         return err;
1134 }
1135
1136 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1137                                     int pages, unsigned long *blocks,
1138                                     int *created, int create)
1139 {
1140         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1141         int rc = 0, i = 0;
1142         struct page *fp = NULL;
1143         int clen = 0;
1144
1145         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1146                 inode->i_ino, pages, (*page)->index);
1147
1148         /* pages are sorted already. so, we just have to find
1149          * contig. space and process them properly */
1150         while (i < pages) {
1151                 if (fp == NULL) {
1152                         /* start new extent */
1153                         fp = *page++;
1154                         clen = 1;
1155                         i++;
1156                         continue;
1157                 } else if (fp->index + clen == (*page)->index) {
1158                         /* continue the extent */
1159                         page++;
1160                         clen++;
1161                         i++;
1162                         continue;
1163                 }
1164
1165                 /* process found extent */
1166                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1167                                         clen * blocks_per_page, blocks,
1168                                         created, create);
1169                 if (rc)
1170                         GOTO(cleanup, rc);
1171
1172                 /* look for next extent */
1173                 fp = NULL;
1174                 blocks += blocks_per_page * clen;
1175                 created += blocks_per_page * clen;
1176         }
1177
1178         if (fp)
1179                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1180                                         clen * blocks_per_page, blocks,
1181                                         created, create);
1182 cleanup:
1183         return rc;
1184 }
1185
1186 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1187                                unsigned long *blocks, int *created, int create);
1188 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1189                                    int pages, unsigned long *blocks,
1190                                    int *created, int create)
1191 {
1192         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1193         unsigned long *b;
1194         int rc = 0, i, *cr;
1195
1196         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1197                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1198                 if (rc) {
1199                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1200                                inode->i_ino, *b, *cr, create, rc);
1201                         break;
1202                 }
1203
1204                 b += blocks_per_page;
1205                 cr += blocks_per_page;
1206         }
1207         return rc;
1208 }
1209
1210 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1211                                 int pages, unsigned long *blocks,
1212                                 int *created, int create,
1213                                 cfs_mutex_t *optional_mutex)
1214 {
1215         int rc;
1216
1217         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1218                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1219                                                      blocks, created, create);
1220                 return rc;
1221         }
1222         if (optional_mutex != NULL)
1223                 cfs_mutex_lock(optional_mutex);
1224         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1225                                             created, create);
1226         if (optional_mutex != NULL)
1227                 cfs_mutex_unlock(optional_mutex);
1228
1229         return rc;
1230 }
1231
1232 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1233 {
1234         unsigned long block;
1235         struct buffer_head *bh;
1236         int err, blocksize, csize, boffs, osize = size;
1237
1238         /* prevent reading after eof */
1239         spin_lock(&inode->i_lock);
1240         if (i_size_read(inode) < *offs + size) {
1241                 size = i_size_read(inode) - *offs;
1242                 spin_unlock(&inode->i_lock);
1243                 if (size < 0) {
1244                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1245                                i_size_read(inode), *offs);
1246                         return -EBADR;
1247                 } else if (size == 0) {
1248                         return 0;
1249                 }
1250         } else {
1251                 spin_unlock(&inode->i_lock);
1252         }
1253
1254         blocksize = 1 << inode->i_blkbits;
1255
1256         while (size > 0) {
1257                 block = *offs >> inode->i_blkbits;
1258                 boffs = *offs & (blocksize - 1);
1259                 csize = min(blocksize - boffs, size);
1260                 bh = ext3_bread(NULL, inode, block, 0, &err);
1261                 if (!bh) {
1262                         CERROR("can't read block: %d\n", err);
1263                         return err;
1264                 }
1265
1266                 memcpy(buf, bh->b_data + boffs, csize);
1267                 brelse(bh);
1268
1269                 *offs += csize;
1270                 buf += csize;
1271                 size -= csize;
1272         }
1273         return osize;
1274 }
1275 EXPORT_SYMBOL(fsfilt_ext3_read);
1276
1277 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1278                                    int size, loff_t *offs)
1279 {
1280         int rc;
1281         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1282         if (rc > 0)
1283                 rc = 0;
1284         return rc;
1285 }
1286
1287 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1288                                 loff_t *offs, handle_t *handle)
1289 {
1290         struct buffer_head *bh = NULL;
1291         loff_t old_size = i_size_read(inode), offset = *offs;
1292         loff_t new_size = i_size_read(inode);
1293         unsigned long block;
1294         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1295
1296         while (bufsize > 0) {
1297                 if (bh != NULL)
1298                         brelse(bh);
1299
1300                 block = offset >> inode->i_blkbits;
1301                 boffs = offset & (blocksize - 1);
1302                 size = min(blocksize - boffs, bufsize);
1303                 bh = ext3_bread(handle, inode, block, 1, &err);
1304                 if (!bh) {
1305                         CERROR("can't read/create block: %d\n", err);
1306                         break;
1307                 }
1308
1309                 err = ext3_journal_get_write_access(handle, bh);
1310                 if (err) {
1311                         CERROR("journal_get_write_access() returned error %d\n",
1312                                err);
1313                         break;
1314                 }
1315                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1316                 memcpy(bh->b_data + boffs, buf, size);
1317                 err = ext3_journal_dirty_metadata(handle, bh);
1318                 if (err) {
1319                         CERROR("journal_dirty_metadata() returned error %d\n",
1320                                err);
1321                         break;
1322                 }
1323                 if (offset + size > new_size)
1324                         new_size = offset + size;
1325                 offset += size;
1326                 bufsize -= size;
1327                 buf += size;
1328         }
1329         if (bh)
1330                 brelse(bh);
1331
1332         /* correct in-core and on-disk sizes */
1333         if (new_size > i_size_read(inode)) {
1334                 spin_lock(&inode->i_lock);
1335                 if (new_size > i_size_read(inode))
1336                         i_size_write(inode, new_size);
1337                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1338                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1339                 if (i_size_read(inode) > old_size) {
1340                         spin_unlock(&inode->i_lock);
1341                         mark_inode_dirty(inode);
1342                 } else {
1343                         spin_unlock(&inode->i_lock);
1344                 }
1345         }
1346
1347         if (err == 0)
1348                 *offs = offset;
1349         return err;
1350 }
1351 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1352
1353 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1354                                     loff_t *offs, int force_sync)
1355 {
1356         struct inode *inode = file->f_dentry->d_inode;
1357         handle_t *handle;
1358         int err, block_count = 0, blocksize;
1359
1360         /* Determine how many transaction credits are needed */
1361         blocksize = 1 << inode->i_blkbits;
1362         block_count = (*offs & (blocksize - 1)) + bufsize;
1363         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1364
1365         handle = ext3_journal_start(inode,
1366                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1367         if (IS_ERR(handle)) {
1368                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1369                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1370                 return PTR_ERR(handle);
1371         }
1372
1373         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1374
1375         if (!err && force_sync)
1376                 handle->h_sync = 1; /* recovery likes this */
1377
1378         ext3_journal_stop(handle);
1379
1380         return err;
1381 }
1382
1383 static int fsfilt_ext3_setup(struct super_block *sb)
1384 {
1385 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1386      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1387         struct ext3_sb_info *sbi = EXT3_SB(sb);
1388 #if 0
1389         sbi->dx_lock = fsfilt_ext3_dx_lock;
1390         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1391 #endif
1392 #endif
1393         if (!EXT3_HAS_COMPAT_FEATURE(sb,
1394                                 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1395                 CERROR("ext3 mounted without journal\n");
1396                 return -EINVAL;
1397         }
1398
1399 #ifdef S_PDIROPS
1400         CWARN("Enabling PDIROPS\n");
1401         set_opt(sbi->s_mount_opt, PDIROPS);
1402         sb->s_flags |= S_PDIROPS;
1403 #endif
1404         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1405                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1406 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1407         /* enable journaled quota support */
1408         /* kfreed in ext3_put_super() */
1409         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1410         if (!sbi->s_qf_names[USRQUOTA])
1411                 return -ENOMEM;
1412         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1413         if (!sbi->s_qf_names[GRPQUOTA]) {
1414                 kfree(sbi->s_qf_names[USRQUOTA]);
1415                 sbi->s_qf_names[USRQUOTA] = NULL;
1416                 return -ENOMEM;
1417         }
1418         sbi->s_jquota_fmt = QFMT_LUSTRE;
1419 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1420         set_opt(sbi->s_mount_opt, QUOTA);
1421 #endif
1422 #endif
1423         return 0;
1424 }
1425
1426 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1427    objects. Logs is number of logfiles to update */
1428 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1429 {
1430         if ( !fso ) {
1431                 switch(op) {
1432                 case FSFILT_OP_CREATE:
1433                                  /* directory leaf, index & indirect & EA*/
1434                         return 4 + 3 * logs;
1435                 case FSFILT_OP_UNLINK:
1436                         return 3 * logs;
1437                 }
1438         } else {
1439                 int i;
1440                 int needed = 0;
1441                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1442                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1443                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1444                 for (i = 0; i < op; i++, fso++) {
1445                         int nblocks = fso->fso_bufcnt * blockpp;
1446                         int ndindirect = min(nblocks, addrpp + 1);
1447                         int nindir = nblocks + ndindirect + 1;
1448
1449                         needed += nindir;
1450                 }
1451                 return needed + 3 * logs;
1452         }
1453
1454         return 0;
1455 }
1456
1457 #ifdef HAVE_QUOTA_SUPPORT
1458 #define DQINFO_COPY(out, in)                    \
1459 do {                                            \
1460         Q_COPY(out, in, dqi_bgrace);            \
1461         Q_COPY(out, in, dqi_igrace);            \
1462         Q_COPY(out, in, dqi_flags);             \
1463         Q_COPY(out, in, dqi_valid);             \
1464 } while (0)
1465
1466 #define DQBLK_COPY(out, in)                     \
1467 do {                                            \
1468         Q_COPY(out, in, dqb_bhardlimit);        \
1469         Q_COPY(out, in, dqb_bsoftlimit);        \
1470         Q_COPY(out, in, dqb_curspace);          \
1471         Q_COPY(out, in, dqb_ihardlimit);        \
1472         Q_COPY(out, in, dqb_isoftlimit);        \
1473         Q_COPY(out, in, dqb_curinodes);         \
1474         Q_COPY(out, in, dqb_btime);             \
1475         Q_COPY(out, in, dqb_itime);             \
1476         Q_COPY(out, in, dqb_valid);             \
1477 } while (0)
1478
1479 static int fsfilt_ext3_quotactl(struct super_block *sb,
1480                                 struct obd_quotactl *oqc)
1481 {
1482         int i, rc = 0, error = 0;
1483         const struct quotactl_ops *qcop;
1484         struct if_dqinfo *info;
1485         struct if_dqblk *dqblk;
1486         ENTRY;
1487
1488         if (!sb->s_qcop)
1489                 RETURN(-ENOSYS);
1490
1491         OBD_ALLOC_PTR(info);
1492         if (!info)
1493                 RETURN(-ENOMEM);
1494         OBD_ALLOC_PTR(dqblk);
1495         if (!dqblk) {
1496                 OBD_FREE_PTR(info);
1497                 RETURN(-ENOMEM);
1498         }
1499
1500         DQINFO_COPY(info, &oqc->qc_dqinfo);
1501         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1502
1503         qcop = sb->s_qcop;
1504         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1505                 for (i = 0; i < MAXQUOTAS; i++) {
1506                         if (!Q_TYPESET(oqc, i))
1507                                 continue;
1508
1509                         if (oqc->qc_cmd == Q_QUOTAON) {
1510                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1511
1512                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1513
1514                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1515                                                  name[i], 0);
1516                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1517                                 rc = ll_quota_off(sb, i, 0);
1518                         }
1519
1520                         if (rc == -EBUSY)
1521                                 error = rc;
1522                         else if (rc)
1523                                 GOTO(out, rc);
1524                 }
1525                 GOTO(out, rc ?: error);
1526         }
1527
1528         switch (oqc->qc_cmd) {
1529         case Q_GETOINFO:
1530         case Q_GETINFO:
1531                 if (!qcop->get_info)
1532                         GOTO(out, rc = -ENOSYS);
1533                 rc = qcop->get_info(sb, oqc->qc_type, info);
1534                 break;
1535         case Q_SETQUOTA:
1536         case Q_INITQUOTA:
1537                 if (!qcop->set_dqblk)
1538                         GOTO(out, rc = -ENOSYS);
1539                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1540                 break;
1541         case Q_GETOQUOTA:
1542         case Q_GETQUOTA:
1543                 if (!qcop->get_dqblk)
1544                         GOTO(out, rc = -ENOSYS);
1545                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1546                 if (!rc)
1547                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1548                 break;
1549         case Q_SYNC:
1550                 if (!sb->s_qcop->quota_sync)
1551                         GOTO(out, rc = -ENOSYS);
1552                 qcop->quota_sync(sb, oqc->qc_type);
1553                 break;
1554         case Q_FINVALIDATE:
1555                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1556                 for (i = 0; i < MAXQUOTAS; i++) {
1557                         struct file *fp;
1558                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1559
1560                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1561
1562                         if (!Q_TYPESET(oqc, i))
1563                                 continue;
1564
1565                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1566                         if (IS_ERR(fp)) {
1567                                 rc = PTR_ERR(fp);
1568                                 CERROR("error invalidating operational quota file"
1569                                        " %s (rc:%d)\n", name[i], rc);
1570                         } else {
1571                                 filp_close(fp, 0);
1572                         }
1573
1574                 }
1575                 break;
1576         default:
1577                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1578                 LBUG();
1579         }
1580 out:
1581         DQINFO_COPY(&oqc->qc_dqinfo, info);
1582         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1583
1584         OBD_FREE_PTR(info);
1585         OBD_FREE_PTR(dqblk);
1586
1587         if (rc)
1588                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1589                                 "failed: %d\n",
1590                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1591         RETURN(rc);
1592 }
1593
1594 struct chk_dqblk{
1595         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1596         cfs_list_t              dqb_list;        /** in list also */
1597         qid_t                   dqb_id;          /** uid/gid */
1598         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1599         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1600         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1601         qsize_t                 dqb_curspace;    /** current space */
1602         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1603         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1604         qsize_t                 dqb_curinodes;   /** current inodes */
1605         obd_time                dqb_btime;       /** block grace time */
1606         obd_time                dqb_itime;       /** inode grace time */
1607         __u32                   dqb_valid;       /** flag for above fields */
1608 };
1609
1610 static inline unsigned int chkquot_hash(qid_t id, int type)
1611                                         __attribute__((__const__));
1612
1613 static inline unsigned int chkquot_hash(qid_t id, int type)
1614 {
1615         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1616 }
1617
1618 static inline struct chk_dqblk *
1619 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1620 {
1621         cfs_hlist_node_t *node;
1622         struct chk_dqblk *cdqb;
1623
1624         cfs_hlist_for_each(node, head) {
1625                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1626                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1627                         return cdqb;
1628         }
1629
1630         return NULL;
1631 }
1632
1633 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1634 {
1635         struct chk_dqblk *cdqb;
1636
1637         OBD_ALLOC_PTR(cdqb);
1638         if (cdqb) {
1639                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1640                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1641                 cdqb->dqb_id = id;
1642                 cdqb->dqb_type = type;
1643         }
1644
1645         return cdqb;
1646 }
1647
1648 static struct chk_dqblk *
1649 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1650       cfs_list_t *list, qid_t id, int type, int first_check)
1651 {
1652         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1653         struct if_dqblk dqb;
1654         struct chk_dqblk *cdqb;
1655         int rc;
1656
1657         cdqb = find_chkquot(head, id, type);
1658         if (cdqb)
1659                 return cdqb;
1660
1661         cdqb = alloc_chkquot(id, type);
1662         if (!cdqb)
1663                 return NULL;
1664
1665         if (!first_check) {
1666                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1667                 if (rc) {
1668                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1669                                id, type, rc);
1670                 } else {
1671                         DQBLK_COPY(cdqb, &dqb);
1672                         cdqb->dqb_curspace = 0;
1673                         cdqb->dqb_curinodes = 0;
1674                 }
1675         }
1676
1677         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1678         cfs_list_add_tail(&cdqb->dqb_list, list);
1679
1680         return cdqb;
1681 }
1682
1683 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1684 {
1685         struct obd_quotactl *oqctl;
1686         int rc;
1687
1688         OBD_ALLOC_PTR(oqctl);
1689         if (!oqctl)
1690                 RETURN(-ENOMEM);
1691
1692         oqctl->qc_cmd = cmd;
1693         oqctl->qc_id = qfmt;
1694         oqctl->qc_type = type;
1695         rc = fsfilt_ext3_quotactl(sb, oqctl);
1696
1697         OBD_FREE_PTR(oqctl);
1698         return rc;
1699 }
1700
1701 static inline int read_old_dqinfo(struct super_block *sb, int type,
1702                                   struct if_dqinfo *dqinfo)
1703 {
1704         struct obd_quotactl *oqctl;
1705         int rc;
1706         ENTRY;
1707
1708         OBD_ALLOC_PTR(oqctl);
1709         if (!oqctl)
1710                 RETURN(-ENOMEM);
1711
1712         oqctl->qc_cmd = Q_GETINFO;
1713         oqctl->qc_type = type;
1714         rc = fsfilt_ext3_quotactl(sb, oqctl);
1715         if (!rc)
1716                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1717
1718         OBD_FREE_PTR(oqctl);
1719         RETURN(rc);
1720 }
1721
1722 struct qchk_ctxt {
1723         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1724         cfs_list_t              qckt_list;                 /* quotacheck list */
1725         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1726         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1727 };
1728
1729 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1730                            struct obd_quotactl *oqc)
1731 {
1732         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1733         loff_t size = 0;
1734         qid_t qid[MAXQUOTAS];
1735         int cnt, i, rc = 0;
1736
1737         if (!inode)
1738                 return 0;
1739
1740         qid[USRQUOTA] = inode->i_uid;
1741         qid[GRPQUOTA] = inode->i_gid;
1742
1743         if (S_ISDIR(inode->i_mode) ||
1744             S_ISREG(inode->i_mode) ||
1745             S_ISLNK(inode->i_mode))
1746                 size = inode_get_bytes(inode);
1747
1748         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1749                 if (!Q_TYPESET(oqc, cnt))
1750                         continue;
1751
1752                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1753                                 &qctxt->qckt_list, qid[cnt], cnt,
1754                                 qctxt->qckt_first_check[cnt]);
1755                 if (!cdqb[cnt]) {
1756                         rc = -ENOMEM;
1757                         break;
1758                 }
1759
1760                 cdqb[cnt]->dqb_curspace += size;
1761                 cdqb[cnt]->dqb_curinodes++;
1762         }
1763
1764         if (rc) {
1765                 for (i = 0; i < cnt; i++) {
1766                         if (!Q_TYPESET(oqc, i))
1767                                 continue;
1768                         LASSERT(cdqb[i]);
1769                         cdqb[i]->dqb_curspace -= size;
1770                         cdqb[i]->dqb_curinodes--;
1771                 }
1772         }
1773
1774         return rc;
1775 }
1776
1777 /* write dqinfo struct in a new quota file */
1778 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1779 {
1780         struct v2_disk_dqinfo dqinfo;
1781         __u32 blocks = V2_DQTREEOFF + 1;
1782         loff_t offset = V2_DQINFOOFF;
1783
1784         if (info) {
1785                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1786                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1787                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1788                                                ~DQF_INFO_DIRTY);
1789         } else {
1790                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1791                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1792                 dqinfo.dqi_flags = 0;
1793         }
1794
1795         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1796         dqinfo.dqi_free_blk = 0;
1797         dqinfo.dqi_free_entry = 0;
1798
1799         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1800 }
1801
1802 static int v3_write_dqheader(struct file *f, int type)
1803 {
1804         static const __u32 quota_magics[] = V2_INITQMAGICS;
1805         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1806         struct v2_disk_dqheader dqhead;
1807         loff_t offset = 0;
1808
1809         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1810         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1811
1812         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1813         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1814
1815         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1816 }
1817
1818 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1819                                   struct obd_quotactl *oqc)
1820 {
1821         int i, rc = 0;
1822         ENTRY;
1823
1824         for (i = 0; i < MAXQUOTAS; i++) {
1825                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1826                                          NULL : &qctxt->qckt_dqinfo[i];
1827                 struct file *file;
1828                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1829
1830                 if (!Q_TYPESET(oqc, i))
1831                         continue;
1832
1833                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1834
1835                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1836                 if (IS_ERR(file)) {
1837                         rc = PTR_ERR(file);
1838                         CERROR("can't create %s file: rc = %d\n",
1839                                name[i], rc);
1840                         GOTO(out, rc);
1841                 }
1842
1843                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1844                         CERROR("file %s is not regular", name[i]);
1845                         filp_close(file, 0);
1846                         GOTO(out, rc = -EINVAL);
1847                 }
1848
1849                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1850
1851                 rc = v3_write_dqheader(file, i);
1852                 if (rc) {
1853                         filp_close(file, 0);
1854                         GOTO(out, rc);
1855                 }
1856
1857                 rc = v3_write_dqinfo(file, i, info);
1858                 filp_close(file, 0);
1859                 if (rc)
1860                         GOTO(out, rc);
1861         }
1862
1863 out:
1864         RETURN(rc);
1865 }
1866
1867
1868 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1869                           struct chk_dqblk *cdqb)
1870 {
1871         struct obd_quotactl *oqc;
1872         long now;
1873         int rc;
1874         ENTRY;
1875
1876         OBD_ALLOC_PTR(oqc);
1877         if (!oqc)
1878                 RETURN(-ENOMEM);
1879
1880         now = cfs_time_current_sec();
1881
1882         if (cdqb->dqb_bsoftlimit &&
1883             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1884             !cdqb->dqb_btime)
1885                 cdqb->dqb_btime =
1886                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1887
1888         if (cdqb->dqb_isoftlimit &&
1889             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1890             !cdqb->dqb_itime)
1891                 cdqb->dqb_itime =
1892                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1893
1894         cdqb->dqb_valid = QIF_ALL;
1895
1896         oqc->qc_cmd = Q_SETQUOTA;
1897         oqc->qc_type = cdqb->dqb_type;
1898         oqc->qc_id = cdqb->dqb_id;
1899         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1900
1901         rc = fsfilt_ext3_quotactl(sb, oqc);
1902         OBD_FREE_PTR(oqc);
1903         RETURN(rc);
1904 }
1905
1906 static int prune_chkquots(struct super_block *sb,
1907                           struct qchk_ctxt *qctxt, int error)
1908 {
1909         struct chk_dqblk *cdqb, *tmp;
1910         int rc;
1911
1912         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1913                 if (!error) {
1914                         rc = commit_chkquot(sb, qctxt, cdqb);
1915                         if (rc)
1916                                 error = rc;
1917                 }
1918                 cfs_hlist_del_init(&cdqb->dqb_hash);
1919                 cfs_list_del(&cdqb->dqb_list);
1920                 OBD_FREE_PTR(cdqb);
1921         }
1922
1923         return error;
1924 }
1925
1926 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1927 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1928 #endif
1929
1930 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1931                                   struct obd_quotactl *oqc)
1932 {
1933         struct ext3_sb_info *sbi = EXT3_SB(sb);
1934         int i, group, uninit_feat = 0;
1935         struct qchk_ctxt *qctxt;
1936         struct buffer_head *bitmap_bh = NULL;
1937         unsigned long ino, inode_inuse;
1938         struct inode *inode;
1939         int rc = 0;
1940         ENTRY;
1941
1942         /* turn on quota and read dqinfo if existed */
1943         OBD_ALLOC_PTR(qctxt);
1944         if (!qctxt) {
1945                 oqc->qc_stat = -ENOMEM;
1946                 RETURN(-ENOMEM);
1947         }
1948
1949         for (i = 0; i < NR_DQHASH; i++)
1950                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1951         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
1952
1953         for (i = 0; i < MAXQUOTAS; i++) {
1954                 if (!Q_TYPESET(oqc, i))
1955                         continue;
1956
1957                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1958                 if (!rc || rc == -EBUSY) {
1959                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1960                         if (rc)
1961                                 GOTO(out, rc);
1962                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1963                         qctxt->qckt_first_check[i] = 1;
1964                 } else if (rc) {
1965                         GOTO(out, rc);
1966                 }
1967         }
1968         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
1969                 /* This filesystem supports the uninit group feature */
1970                 uninit_feat = 1;
1971
1972         /* number of inodes that have been allocated */
1973         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
1974                       percpu_counter_sum(&sbi->s_freeinodes_counter);
1975
1976         /* check quota and update in hash */
1977         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
1978              group++) {
1979                 unsigned long used_count = sbi->s_inodes_per_group;
1980
1981                 if (uninit_feat) {
1982                         struct ext3_group_desc *desc;
1983                         desc = ext3_get_group_desc(sb, group, NULL);
1984                         if (!desc)
1985                                 GOTO(out, -EIO);
1986
1987                         /* we don't really need to take the group lock here,
1988                          * but it may be useful if one day we support online
1989                          * quotacheck */
1990                         ext4_lock_group(sb, group);
1991                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
1992                                 /* no inode in use in this group, just skip it */
1993                                 ext3_unlock_group(sb, group);
1994                                 continue;
1995                         }
1996
1997                         used_count -= ext3_itable_unused_count(sb, desc);
1998                         ext3_unlock_group(sb, group);
1999                 }
2000
2001                 ino = group * sbi->s_inodes_per_group + 1;
2002                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2003                 if (!bitmap_bh) {
2004                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2005                                sb->s_id, group);
2006                         GOTO(out, -EIO);
2007                 }
2008
2009                 i = 0;
2010                 while (i < used_count &&
2011                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2012                                                used_count, i)) < used_count) {
2013                         inode_inuse--;
2014                         i++;
2015                         ino = i + group * sbi->s_inodes_per_group;
2016                         if (ino < sbi->s_first_ino)
2017                                 continue;
2018                         inode = ext3_iget(sb, ino);
2019                         if (!inode || IS_ERR(inode))
2020                                 continue;
2021
2022                         rc = add_inode_quota(inode, qctxt, oqc);
2023                         iput(inode);
2024                         if (rc) {
2025                                 brelse(bitmap_bh);
2026                                 GOTO(out, rc);
2027                         }
2028                 }
2029                 brelse(bitmap_bh);
2030         }
2031
2032         /* read old quota limits from old quota file. (only for the user
2033          * has limits but hasn't file) */
2034 #ifdef HAVE_QUOTA_SUPPORT
2035         for (i = 0; i < MAXQUOTAS; i++) {
2036                 cfs_list_t id_list;
2037                 struct dquot_id *dqid, *tmp;
2038
2039                 if (!Q_TYPESET(oqc, i))
2040                         continue;
2041
2042                 if (qctxt->qckt_first_check[i])
2043                         continue;
2044
2045
2046                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2047                 CFS_INIT_LIST_HEAD(&id_list);
2048                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2049                 if (rc)
2050                         CERROR("read old limits failed. (rc:%d)\n", rc);
2051
2052                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2053                         cfs_list_del_init(&dqid->di_link);
2054
2055                         if (!rc)
2056                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2057                                       dqid->di_id, i,
2058                                       qctxt->qckt_first_check[i]);
2059                         OBD_FREE_PTR(dqid);
2060                 }
2061         }
2062 #endif
2063         /* turn off quota cause we are to dump chk_dqblk to files */
2064         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2065
2066         rc = create_new_quota_files(qctxt, oqc);
2067         if (rc)
2068                 GOTO(out, rc);
2069
2070         /* we use vfs functions to set dqblk, so turn quota on */
2071         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2072 out:
2073         /* dump and free chk_dqblk */
2074         rc = prune_chkquots(sb, qctxt, rc);
2075         OBD_FREE_PTR(qctxt);
2076
2077         /* turn off quota, `lfs quotacheck` will turn on when all
2078          * nodes quotacheck finish. */
2079         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2080
2081         oqc->qc_stat = rc;
2082         if (rc)
2083                 CERROR("quotacheck failed: rc = %d\n", rc);
2084
2085         RETURN(rc);
2086 }
2087
2088 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2089                                  int cmd)
2090 {
2091         int rc = 0;
2092         ENTRY;
2093
2094         if (lqi->qi_files[type] == NULL) {
2095                 CERROR("operate qinfo before it's enabled!\n");
2096                 RETURN(-ESRCH);
2097         }
2098
2099         switch (cmd) {
2100         case QFILE_CHK:
2101                 rc = lustre_check_quota_file(lqi, type);
2102                 break;
2103         case QFILE_RD_INFO:
2104                 rc = lustre_read_quota_info(lqi, type);
2105                 break;
2106         case QFILE_WR_INFO:
2107                 rc = lustre_write_quota_info(lqi, type);
2108                 break;
2109         case QFILE_INIT_INFO:
2110                 rc = lustre_init_quota_info(lqi, type);
2111                 break;
2112         case QFILE_CONVERT:
2113                 rc = -ENOTSUPP;
2114                 CERROR("quota CONVERT command is not supported\n");
2115                 break;
2116         default:
2117                 rc = -ENOTSUPP;
2118                 CERROR("Unsupported admin quota file cmd %d\n"
2119                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2120                        cmd);
2121                 break;
2122         }
2123         RETURN(rc);
2124 }
2125
2126 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2127                             cfs_list_t *list)
2128 {
2129         return lustre_get_qids(file, inode, type, list);
2130 }
2131
2132 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2133 {
2134         int rc = 0;
2135         ENTRY;
2136
2137         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2138                 CERROR("operate dquot before it's enabled!\n");
2139                 RETURN(-ESRCH);
2140         }
2141
2142         switch (cmd) {
2143         case QFILE_RD_DQUOT:
2144                 rc = lustre_read_dquot(dquot);
2145                 break;
2146         case QFILE_WR_DQUOT:
2147                 if (dquot->dq_dqb.dqb_ihardlimit ||
2148                     dquot->dq_dqb.dqb_isoftlimit ||
2149                     dquot->dq_dqb.dqb_bhardlimit ||
2150                     dquot->dq_dqb.dqb_bsoftlimit)
2151                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2152                 else
2153                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2154
2155                 rc = lustre_commit_dquot(dquot);
2156                 if (rc >= 0)
2157                         rc = 0;
2158                 break;
2159         default:
2160                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2161                 LBUG();
2162                 break;
2163         }
2164         RETURN(rc);
2165 }
2166
2167 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2168                                 struct inode *inode, int frags)
2169 {
2170 #ifdef EXT3_EXT_HAS_NO_TREE
2171         struct ext3_ext_base *base = inode;
2172 #else
2173         struct ext3_extents_tree tree;
2174         struct ext3_ext_base *base = &tree;
2175
2176         ext3_init_tree_desc(base, inode);
2177 #endif
2178         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2179          * metablocks at maxium b=16542 */
2180         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2181         return 0;
2182 }
2183
2184 #endif
2185
2186 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2187 {
2188         return (EXT3_SB(sb)->journal_bdev);
2189 }
2190 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2191
2192 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2193                           char *buf, int count, loff_t pos)
2194 {
2195         loff_t p = pos;
2196         int rc;
2197
2198         if (!f && !inode) {
2199                 CERROR("lustre_read_quota failed for no quota file!\n");
2200                 libcfs_debug_dumpstack(NULL);
2201                 return -EINVAL;
2202         }
2203
2204         /* Support for both adm and op quota files must be provided */
2205         if (f) {
2206                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2207                 rc = rc < 0 ? rc : p - pos;
2208         } else {
2209                 struct super_block *sb = inode->i_sb;
2210                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2211         }
2212         return rc;
2213 }
2214
2215 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2216 {
2217         loff_t p = pos;
2218         int rc;
2219
2220         /* Only adm quota files are supported, op updates are handled by vfs */
2221         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2222         rc = rc < 0 ? rc : p - pos;
2223
2224         return rc;
2225 }
2226
2227 void *lustre_quota_journal_start(struct inode *inode, int delete)
2228 {
2229         handle_t *handle;
2230         unsigned block_count;
2231
2232         if (delete) {
2233                 /* each indirect block (+4) may become free, attaching to the
2234                  * header list of free blocks (+1); the data block (+1) may
2235                  * become a free block (+0) or a block with free dqentries (+0) */
2236                 block_count = (4 + 1) + 1;
2237                 handle = ext3_journal_start(inode,
2238                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2239         } else {
2240                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2241                  * freeblk reusage with a header update (+1); dqentry is either reused
2242                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2243                  * a new block allocation (+1) with a header update (+1)              */
2244                 block_count = (4 + 1) + 3;
2245                 handle = ext3_journal_start(inode,
2246                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2247
2248         }
2249
2250         return handle;
2251 }
2252
2253 void lustre_quota_journal_stop(void *handle)
2254 {
2255         ext3_journal_stop((handle_t *)handle);
2256 }
2257
2258 static int ll_decode_fh_accept(void *context, struct dentry *de)
2259 {
2260         return 1;
2261 }
2262
2263 #ifdef HAVE_EXPORTFS_DECODE_FH
2264 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2265          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2266                             acceptable, context)
2267 #else
2268 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2269          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2270                                      type, acceptable, context)
2271 # define FILEID_INO32_GEN 1
2272 extern struct export_operations export_op_default;
2273 #endif
2274
2275 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2276                                       struct fsfilt_fid *fid, int ignore_gen)
2277 {
2278         struct inode  *inode;
2279         struct dentry *result;
2280
2281         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2282                                        ll_decode_fh_accept, NULL);
2283         if (IS_ERR(result)) {
2284                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2285                        fid->ino, fid->gen, PTR_ERR(result));
2286                 return result;
2287         }
2288
2289         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2290                fid->ino, fid->gen);
2291         inode = result->d_inode;
2292         if (inode == NULL)
2293                 goto err_out;
2294
2295         if (inode->i_nlink == 0 &&
2296             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2297                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2298                               " ctime -- this may indicate disk "
2299                               "corruption (inode: %lu, link: %lu, "
2300                               "count: %d)\n", inode->i_ino,
2301                               (unsigned long)inode->i_nlink,
2302                               atomic_read(&inode->i_count));
2303                 goto err_out;
2304         }
2305         if (fid->gen && inode->i_generation != fid->gen) {
2306                 /* we didn't find the right inode.. */
2307                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2308                        "count: %d, generation %u/%u\n",
2309                        inode->i_ino, (unsigned long)inode->i_nlink,
2310                        atomic_read(&inode->i_count), inode->i_generation,
2311                        fid->gen);
2312                 goto err_out;
2313         }
2314
2315         return result;
2316 err_out:
2317         l_dput(result);
2318         return ERR_PTR(-ENOENT);
2319 }
2320
2321 static struct fsfilt_operations fsfilt_ext3_ops = {
2322         .fs_type                = "ext3",
2323         .fs_owner               = THIS_MODULE,
2324         .fs_getlabel            = fsfilt_ext3_get_label,
2325         .fs_setlabel            = fsfilt_ext3_set_label,
2326         .fs_uuid                = fsfilt_ext3_uuid,
2327         .fs_start               = fsfilt_ext3_start,
2328         .fs_brw_start           = fsfilt_ext3_brw_start,
2329         .fs_extend              = fsfilt_ext3_extend,
2330         .fs_commit              = fsfilt_ext3_commit,
2331         .fs_commit_async        = fsfilt_ext3_commit_async,
2332         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2333         .fs_setattr             = fsfilt_ext3_setattr,
2334         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2335         .fs_set_md              = fsfilt_ext3_set_md,
2336         .fs_get_md              = fsfilt_ext3_get_md,
2337         .fs_readpage            = fsfilt_ext3_readpage,
2338         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2339         .fs_statfs              = fsfilt_ext3_statfs,
2340         .fs_sync                = fsfilt_ext3_sync,
2341         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2342         .fs_write_record        = fsfilt_ext3_write_record,
2343         .fs_read_record         = fsfilt_ext3_read_record,
2344         .fs_setup               = fsfilt_ext3_setup,
2345         .fs_send_bio            = fsfilt_ext3_send_bio,
2346         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2347 #ifdef HAVE_DISK_INODE_VERSION
2348         .fs_get_version         = fsfilt_ext3_get_version,
2349         .fs_set_version         = fsfilt_ext3_set_version,
2350 #endif
2351 #ifdef HAVE_QUOTA_SUPPORT
2352         .fs_quotactl            = fsfilt_ext3_quotactl,
2353         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2354         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2355         .fs_qids                = fsfilt_ext3_qids,
2356         .fs_dquot               = fsfilt_ext3_dquot,
2357         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2358 #endif
2359         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2360         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2361 };
2362
2363 static int __init fsfilt_ext3_init(void)
2364 {
2365         int rc;
2366
2367         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2368                                          sizeof(struct fsfilt_cb_data), 0, 0);
2369         if (!fcb_cache) {
2370                 CERROR("error allocating fsfilt journal callback cache\n");
2371                 GOTO(out, rc = -ENOMEM);
2372         }
2373
2374         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2375
2376         if (rc) {
2377                 int err = cfs_mem_cache_destroy(fcb_cache);
2378                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2379         }
2380 out:
2381         return rc;
2382 }
2383
2384 static void __exit fsfilt_ext3_exit(void)
2385 {
2386         int rc;
2387
2388         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2389         rc = cfs_mem_cache_destroy(fcb_cache);
2390         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2391 }
2392
2393 module_init(fsfilt_ext3_init);
2394 module_exit(fsfilt_ext3_exit);
2395
2396 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2397 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2398 MODULE_LICENSE("GPL");