lustre/lvfs/fsfilt_ext3.c

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