lustre/lvfs/fsfilt_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/fsfilt_ext3.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/init.h>
  44 #include <linux/module.h>
  45 #include <linux/fs.h>
  46 #include <linux/slab.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/quotaops.h>
  49 #ifdef HAVE_EXT4_LDISKFS
  50 #include <ext4/ext4.h>
  51 #include <ext4/ext4_jbd2.h>
  52 #else
  53 #include <linux/jbd.h>
  54 #include <linux/ext3_fs.h>
  55 #include <linux/ext3_jbd.h>
  56 #endif
  57 #include <linux/version.h>
  58 #include <linux/bitops.h>
  59 #include <linux/quota.h>
  60 #ifdef HAVE_QUOTAIO_V1_H
  61 # include <linux/quotaio_v1.h>
  62 # include <linux/quotaio_v2.h>
  63 #elif defined(HAVE_FS_QUOTA_QUOTAIO_V1_H)
  64 # include <quota/quotaio_v1.h>
  65 # include <quota/quotaio_v2.h>
  66 # include <quota/quota_tree.h>
  67 # define V2_DQTREEOFF    QT_TREEOFF
  68 #else
  69 # include <quotaio_v1.h>
  70 # include <quotaio_v2.h>
  71 # include <quota_tree.h>
  72 # define V2_DQTREEOFF    QT_TREEOFF
  73 #endif
  74 #include <linux/parser.h>
  75 #if defined(HAVE_EXT3_XATTR_H)
  76 #include <ext3/xattr.h>
  77 #else
  78 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  79 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  80 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  81 #endif
  82
  83 #include <libcfs/kp30.h>
  84 #include <lustre_fsfilt.h>
  85 #include <obd.h>
  86 #include <lustre_quota.h>
  87 #include <linux/lustre_compat25.h>
  88 #include <linux/lprocfs_status.h>
  89
  90 #ifdef HAVE_EXT4_LDISKFS
  91 #include <ext4/ext4_extents.h>
  92 #else
  93 #include <linux/ext3_extents.h>
  94 #endif
  95
  96 #include "lustre_quota_fmt.h"
  97
  98 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  99 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 100 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 101 #else
 102 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 103 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 104 #endif
 105
 106 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 107 /* for kernels 2.6.18 and later */
 108 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 109 #else
 110 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 111 #endif
 112
 113 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 114 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 115                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 116 #else
 117 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 118                ext3_ext_insert_extent(handle, inode, path, newext)
 119 #endif
 120
 121
 122 static cfs_mem_cache_t *fcb_cache;
 123
 124 struct fsfilt_cb_data {
 125         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 126         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 127         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 128         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 129         void *cb_data;                  /* MDS/OST completion function data */
 130 };
 131
 132 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 133 #define EXT3_XATTR_INDEX_TRUSTED        4
 134 #endif
 135 #ifndef XATTR_NO_CTIME
 136 #define XATTR_NO_CTIME 0x80
 137 #endif
 138
 139 #ifdef HAVE_EXT4_LDISKFS
 140 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 141 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 142 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 143 #else
 144 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 145 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 146 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 147 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 148 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 149 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 150 #endif
 151
 152 #ifndef ext3_find_next_bit
 153 #define ext3_find_next_bit           ext2_find_next_bit
 154 #endif
 155
 156 #ifndef ext2_find_next_bit
 157 #ifdef __LITTLE_ENDIAN
 158 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 159 #else
 160 error "Need implementation of find_next_bit on big-endian systems"
 161 #endif  /* __LITTLE_ENDIAN */
 162 #endif  /* !ext2_find_next_le_bit */
 163
 164 static char *fsfilt_ext3_get_label(struct super_block *sb)
 165 {
 166         return EXT3_SB(sb)->s_es->s_volume_name;
 167 }
 168
 169 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 170 {
 171         /* see e.g. fsfilt_ext3_write_record() */
 172         journal_t *journal;
 173         handle_t *handle;
 174         int err;
 175
 176         journal = EXT3_SB(sb)->s_journal;
 177         handle = ext3_journal_start_sb(sb, 1);
 178         if (IS_ERR(handle)) {
 179                 CERROR("can't start transaction\n");
 180                 return(PTR_ERR(handle));
 181         }
 182
 183         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 184         if (err)
 185                 goto out;
 186
 187         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 188                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 189
 190         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 191
 192 out:
 193         ext3_journal_stop(handle);
 194
 195         return(err);
 196 }
 197
 198 static char *fsfilt_ext3_uuid(struct super_block *sb)
 199 {
 200         return EXT3_SB(sb)->s_es->s_uuid;
 201 }
 202
 203 #ifdef HAVE_DISK_INODE_VERSION
 204
 205 static __u64 get_i_version(struct inode *inode)
 206 {
 207 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && !defined(USE_I_FS_VERSION)
 208         return inode->i_version;
 209 #else
 210         return EXT3_I(inode)->i_fs_version;
 211 #endif
 212 }
 213
 214 static void set_i_version(struct inode *inode, __u64 new_version)
 215 {
 216 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && !defined(USE_I_FS_VERSION)
 217         inode->i_version = new_version;
 218 #else
 219         (EXT3_I(inode))->i_fs_version = new_version;
 220 #endif
 221 }
 222
 223 /*
 224  * Get the 64-bit version for an inode.
 225  */
 226 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 227 {
 228         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 229                get_i_version(inode), inode->i_ino);
 230         return get_i_version(inode);
 231 }
 232
 233 /*
 234  * Set the 64-bit version and return the old version.
 235  */
 236 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 237 {
 238         __u64 old_version = get_i_version(inode);
 239
 240         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 241                new_version, old_version, inode->i_ino);
 242         set_i_version(inode, new_version);
 243         /* version is set after all inode operations are finished, so we should
 244          * mark it dirty here */
 245         inode->i_sb->s_op->dirty_inode(inode);
 246         return old_version;
 247 }
 248
 249 #endif
 250
 251 /*
 252  * We don't currently need any additional blocks for rmdir and
 253  * unlink transactions because we are storing the OST oa_id inside
 254  * the inode (which we will be changing anyways as part of this
 255  * transaction).
 256  */
 257 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 258                                int logs)
 259 {
 260         /* For updates to the last received file */
 261         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 262         journal_t *journal;
 263         void *handle;
 264
 265         if (current->journal_info) {
 266                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 267                        current->journal_info);
 268                 goto journal_start;
 269         }
 270
 271         switch(op) {
 272         case FSFILT_OP_RMDIR:
 273         case FSFILT_OP_UNLINK:
 274                 /* delete one file + create/update logs for each stripe */
 275                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 276                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 277                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 278                 break;
 279         case FSFILT_OP_RENAME:
 280                 /* modify additional directory */
 281                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 282                 /* no break */
 283         case FSFILT_OP_SYMLINK:
 284                 /* additional block + block bitmap + GDT for long symlink */
 285                 nblocks += 3;
 286                 /* no break */
 287         case FSFILT_OP_CREATE: {
 288 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 289                 static int warned;
 290                 if (!warned) {
 291                         if (!test_opt(inode->i_sb, EXTENTS)) {
 292                                 warned = 1;
 293                         } else if (((EXT3_I(inode)->i_flags &
 294                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 295                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 296                                 CWARN("extent-mapped directory found with "
 297                                       "ext3-based ldiskfs - contact "
 298                                       "http://bugzilla.lustre.org/\n");
 299                                 warned = 1;
 300                         }
 301                 }
 302 #endif
 303                 /* no break */
 304         }
 305         case FSFILT_OP_MKDIR:
 306         case FSFILT_OP_MKNOD:
 307                 /* modify one inode + block bitmap + GDT */
 308                 nblocks += 3;
 309                 /* no break */
 310         case FSFILT_OP_LINK:
 311                 /* modify parent directory */
 312                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 313                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 314                 /* create/update logs for each stripe */
 315                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 316                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 317                 break;
 318         case FSFILT_OP_SETATTR:
 319                 /* Setattr on inode */
 320                 nblocks += 1;
 321                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 322                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 323                 /* quota chown log for each stripe */
 324                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 325                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 326                 break;
 327         case FSFILT_OP_CANCEL_UNLINK:
 328                 /* blocks for log header bitmap update OR
 329                  * blocks for catalog header bitmap update + unlink of logs */
 330                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 331                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 332                 break;
 333         case FSFILT_OP_JOIN:
 334                 /* delete 2 file(file + array id) + create 1 file (array id)
 335                  * create/update logs for each stripe */
 336                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 337
 338                 /*create array log for head file*/
 339                 nblocks += 3;
 340                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 341                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
 342                 /*update head file array */
 343                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 344                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 345                 break;
 346         default: CERROR("unknown transaction start op %d\n", op);
 347                 LBUG();
 348         }
 349
 350         LASSERT(current->journal_info == desc_private);
 351         journal = EXT3_SB(inode->i_sb)->s_journal;
 352         if (nblocks > journal->j_max_transaction_buffers) {
 353                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 354                       nblocks, op, logs, journal->j_max_transaction_buffers);
 355                 nblocks = journal->j_max_transaction_buffers;
 356         }
 357
 358  journal_start:
 359         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 360         handle = ext3_journal_start(inode, nblocks);
 361
 362         if (!IS_ERR(handle))
 363                 LASSERT(current->journal_info == handle);
 364         else
 365                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 366                        op, nblocks, PTR_ERR(handle));
 367         return handle;
 368 }
 369
 370 /*
 371  * Calculate the number of buffer credits needed to write multiple pages in
 372  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 373  * doesn't have a nice API for calculating this sort of thing in advance.
 374  *
 375  * See comment above ext3_writepage_trans_blocks for details.  We assume
 376  * no data journaling is being done, but it does allow for all of the pages
 377  * being non-contiguous.  If we are guaranteed contiguous pages we could
 378  * reduce the number of (d)indirect blocks a lot.
 379  *
 380  * With N blocks per page and P pages, for each inode we have at most:
 381  * N*P indirect
 382  * min(N*P, blocksize/4 + 1) dindirect blocks
 383  * niocount tindirect
 384  *
 385  * For the entire filesystem, we have at most:
 386  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 387  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 388  * objcount inode blocks
 389  * 1 superblock
 390  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 391  *
 392  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 393  */
 394 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 395                                       int niocount, struct niobuf_local *nb)
 396 {
 397         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 398         __u64 next_indir;
 399         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 400         int nbitmaps = 0, ngdblocks;
 401         int needed = objcount + 1; /* inodes + superblock */
 402         int i, j;
 403
 404         for (i = 0, j = 0; i < objcount; i++, fso++) {
 405                 /* two or more dindirect blocks in case we cross boundary */
 406                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 407                                     nb[j].offset) >>
 408                                    sb->s_blocksize_bits) /
 409                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 410                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 411
 412                 /* leaf, indirect, tindirect blocks for first block */
 413                 nbitmaps += blockpp + 2;
 414
 415                 j += fso->fso_bufcnt;
 416         }
 417
 418         next_indir = nb[0].offset +
 419                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 420         for (i = 1; i < niocount; i++) {
 421                 if (nb[i].offset >= next_indir) {
 422                         nbitmaps++;     /* additional indirect */
 423                         next_indir = nb[i].offset +
 424                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 425                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 426                         nbitmaps++;     /* additional indirect */
 427                 }
 428                 nbitmaps += blockpp;    /* each leaf in different group? */
 429         }
 430
 431         ngdblocks = nbitmaps;
 432         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 433                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 434         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 435                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 436
 437         needed += nbitmaps + ngdblocks;
 438
 439         /* last_rcvd update */
 440         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 441
 442 #if defined(CONFIG_QUOTA)
 443         /* We assume that there will be 1 bit set in s_dquot.flags for each
 444          * quota file that is active.  This is at least true for now.
 445          */
 446         needed += hweight32(ll_sb_any_quota_active(sb)) *
 447                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 448 #endif
 449
 450         return needed;
 451 }
 452
 453 /* We have to start a huge journal transaction here to hold all of the
 454  * metadata for the pages being written here.  This is necessitated by
 455  * the fact that we do lots of prepare_write operations before we do
 456  * any of the matching commit_write operations, so even if we split
 457  * up to use "smaller" transactions none of them could complete until
 458  * all of them were opened.  By having a single journal transaction,
 459  * we eliminate duplicate reservations for common blocks like the
 460  * superblock and group descriptors or bitmaps.
 461  *
 462  * We will start the transaction here, but each prepare_write will
 463  * add a refcount to the transaction, and each commit_write will
 464  * remove a refcount.  The transaction will be closed when all of
 465  * the pages have been written.
 466  */
 467 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 468                                    int niocount, struct niobuf_local *nb,
 469                                    void *desc_private, int logs)
 470 {
 471         journal_t *journal;
 472         handle_t *handle;
 473         int needed;
 474         ENTRY;
 475
 476         LASSERT(current->journal_info == desc_private);
 477         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 478         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 479
 480         /* The number of blocks we could _possibly_ dirty can very large.
 481          * We reduce our request if it is absurd (and we couldn't get that
 482          * many credits for a single handle anyways).
 483          *
 484          * At some point we have to limit the size of I/Os sent at one time,
 485          * increase the size of the journal, or we have to calculate the
 486          * actual journal requirements more carefully by checking all of
 487          * the blocks instead of being maximally pessimistic.  It remains to
 488          * be seen if this is a real problem or not.
 489          */
 490         if (needed > journal->j_max_transaction_buffers) {
 491                 CERROR("want too many journal credits (%d) using %d instead\n",
 492                        needed, journal->j_max_transaction_buffers);
 493                 needed = journal->j_max_transaction_buffers;
 494         }
 495
 496         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 497         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 498         if (IS_ERR(handle)) {
 499                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 500                        PTR_ERR(handle));
 501         } else {
 502                 LASSERT(handle->h_buffer_credits >= needed);
 503                 LASSERT(current->journal_info == handle);
 504         }
 505
 506         RETURN(handle);
 507 }
 508
 509 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 510 {
 511        handle_t *handle = h;
 512
 513        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 514        if (nblocks == 0) {
 515                handle->h_buffer_credits = 0;
 516                CWARN("setting credits of handle %p to zero by request\n", h);
 517        }
 518
 519        if (handle->h_buffer_credits > nblocks)
 520                 return 0;
 521        if (ext3_journal_extend(handle, nblocks) == 0)
 522                 return 0;
 523
 524        ext3_mark_inode_dirty(handle, inode);
 525        return ext3_journal_restart(handle, nblocks);
 526 }
 527
 528 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 529 {
 530         int rc;
 531         handle_t *handle = h;
 532
 533         LASSERT(current->journal_info == handle);
 534         if (force_sync)
 535                 handle->h_sync = 1; /* recovery likes this */
 536
 537         rc = ext3_journal_stop(handle);
 538
 539         return rc;
 540 }
 541
 542 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 543                                     void **wait_handle)
 544 {
 545         unsigned long tid;
 546         transaction_t *transaction;
 547         handle_t *handle = h;
 548         journal_t *journal;
 549         int rc;
 550
 551         LASSERT(current->journal_info == handle);
 552
 553         transaction = handle->h_transaction;
 554         journal = transaction->t_journal;
 555         tid = transaction->t_tid;
 556         /* we don't want to be blocked */
 557         handle->h_sync = 0;
 558         rc = ext3_journal_stop(handle);
 559         if (rc) {
 560                 CERROR("error while stopping transaction: %d\n", rc);
 561                 return rc;
 562         }
 563         fsfilt_log_start_commit(journal, tid);
 564
 565         *wait_handle = (void *) tid;
 566         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 567         return 0;
 568 }
 569
 570 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 571 {
 572         journal_t *journal = EXT3_JOURNAL(inode);
 573         tid_t tid = (tid_t)(long)h;
 574
 575         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 576         if (unlikely(is_journal_aborted(journal)))
 577                 return -EIO;
 578
 579         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 580
 581         if (unlikely(is_journal_aborted(journal)))
 582                 return -EIO;
 583         return 0;
 584 }
 585
 586 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 587                                struct iattr *iattr, int do_trunc)
 588 {
 589         struct inode *inode = dentry->d_inode;
 590         int rc = 0;
 591
 592         LASSERTF(!(iattr->ia_valid & ATTR_SIZE) || !S_ISDIR(inode->i_mode),
 593                  "changing i_size on directory #%lu (%p) new %llu old %llu\n",
 594                  inode->i_ino, inode, (unsigned long long)iattr->ia_size,
 595                  (unsigned long long)i_size_read(inode));
 596
 597         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 598          * We are already writing the inode to disk as part of this
 599          * transaction and want to avoid a lot of extra inode writeout
 600          * later on. b=9828 */
 601         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 602                 /* ATTR_SIZE would invoke truncate: clear it */
 603                 iattr->ia_valid &= ~ATTR_SIZE;
 604                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 605                 i_size_write(inode, iattr->ia_size);
 606
 607                 if (iattr->ia_valid & ATTR_UID)
 608                         inode->i_uid = iattr->ia_uid;
 609                 if (iattr->ia_valid & ATTR_GID)
 610                         inode->i_gid = iattr->ia_gid;
 611                 if (iattr->ia_valid & ATTR_ATIME)
 612                         inode->i_atime = iattr->ia_atime;
 613                 if (iattr->ia_valid & ATTR_MTIME)
 614                         inode->i_mtime = iattr->ia_mtime;
 615                 if (iattr->ia_valid & ATTR_CTIME)
 616                         inode->i_ctime = iattr->ia_ctime;
 617                 if (iattr->ia_valid & ATTR_MODE) {
 618                         inode->i_mode = iattr->ia_mode;
 619
 620                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 621                                 inode->i_mode &= ~S_ISGID;
 622                 }
 623
 624                 inode->i_sb->s_op->dirty_inode(inode);
 625
 626                 goto out;
 627         }
 628
 629         /* Don't allow setattr to change file type */
 630         if (iattr->ia_valid & ATTR_MODE)
 631                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 632                                  (iattr->ia_mode & ~S_IFMT);
 633
 634         /* We set these flags on the client, but have already checked perms
 635          * so don't confuse inode_change_ok. */
 636         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 637
 638         if (inode->i_op->setattr) {
 639                 rc = inode->i_op->setattr(dentry, iattr);
 640         } else {
 641                 rc = inode_change_ok(inode, iattr);
 642                 if (!rc)
 643                         rc = inode_setattr(inode, iattr);
 644         }
 645
 646  out:
 647         RETURN(rc);
 648 }
 649
 650 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 651                                  unsigned int cmd, unsigned long arg)
 652 {
 653         int rc = 0;
 654         ENTRY;
 655
 656         /* FIXME: Can't do this because of nested transaction deadlock */
 657         if (cmd == EXT3_IOC_SETFLAGS) {
 658                 /* We can't enable data journaling on OST objects, because
 659                 * this forces the transaction to be closed in order to
 660                 * flush the journal, but the caller will already have a
 661                 * compound transaction open to update the last_rcvd file,
 662                 * and this thread would deadlock trying to set the flag. */
 663                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 664                         CERROR("can't set data journal flag on file\n");
 665                         RETURN(-EPERM);
 666                 }
 667                 /* Because the MDS does not see the EXTENTS_FL set on the
 668                  * OST objects, mask this flag into all set flags.  It is
 669                  * not legal to clear this flag in any case, so we are not
 670                  * changing the functionality by doing this.  b=22911 */
 671                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 672         }
 673
 674 #ifdef HAVE_EXT4_LDISKFS
 675         /* ext4_ioctl does not have a inode argument */
 676         if (inode->i_fop->unlocked_ioctl)
 677                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 678 #else
 679         if (inode->i_fop->ioctl)
 680                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 681 #endif
 682         else
 683                 RETURN(-ENOTTY);
 684
 685         RETURN(rc);
 686 }
 687
 688 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 689                               void *lmm, int lmm_size, const char *name)
 690 {
 691         int rc;
 692
 693         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 694
 695         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 696                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 697
 698         if (rc && rc != -EROFS)
 699                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 700                        inode->i_ino, rc);
 701         return rc;
 702 }
 703
 704 /* Must be called with i_mutex held */
 705 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 706                               const char *name)
 707 {
 708         int rc;
 709
 710         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 711
 712         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 713                             name, lmm, lmm_size);
 714         /* This gives us the MD size */
 715         if (lmm == NULL)
 716                 return (rc == -ENODATA) ? 0 : rc;
 717
 718         if (rc < 0) {
 719                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 720                        EXT3_XATTR_INDEX_TRUSTED, name,
 721                        inode->i_ino, rc);
 722                 memset(lmm, 0, lmm_size);
 723                 return (rc == -ENODATA) ? 0 : rc;
 724         }
 725
 726         return rc;
 727 }
 728
 729 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 730 {
 731         submit_bio(rw, bio);
 732         return 0;
 733 }
 734
 735 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 736                                     loff_t *off)
 737 {
 738         struct inode *inode = file->f_dentry->d_inode;
 739         int rc = 0;
 740
 741         if (S_ISREG(inode->i_mode))
 742                 rc = file->f_op->read(file, buf, count, off);
 743         else {
 744                 const int blkbits = inode->i_sb->s_blocksize_bits;
 745                 const int blksize = inode->i_sb->s_blocksize;
 746
 747                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 748                        (unsigned long)count, inode->i_ino, *off);
 749                 while (count > 0) {
 750                         struct buffer_head *bh;
 751
 752                         bh = NULL;
 753                         if (*off < i_size_read(inode)) {
 754                                 int err = 0;
 755
 756                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 757                                                 0, &err);
 758
 759                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 760
 761                                 if (bh) {
 762                                         memcpy(buf, bh->b_data, blksize);
 763                                         brelse(bh);
 764                                 } else if (err) {
 765                                         /* XXX in theory we should just fake
 766                                          * this buffer and continue like ext3,
 767                                          * especially if this is a partial read
 768                                          */
 769                                         CERROR("error read dir %lu+%llu: %d\n",
 770                                                inode->i_ino, *off, err);
 771                                         RETURN(err);
 772                                 }
 773                         }
 774                         if (!bh) {
 775                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 776
 777                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 778                                 memset(fake, 0, sizeof(*fake));
 779                                 fake->rec_len = cpu_to_le16(blksize);
 780                         }
 781                         count -= blksize;
 782                         buf += blksize;
 783                         *off += blksize;
 784                         rc += blksize;
 785                 }
 786         }
 787
 788         return rc;
 789 }
 790
 791 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 792 {
 793         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 794
 795         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 796
 797         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 798 }
 799
 800 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 801                                       void *handle, fsfilt_cb_t cb_func,
 802                                       void *cb_data)
 803 {
 804         struct fsfilt_cb_data *fcb;
 805
 806         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 807         if (fcb == NULL)
 808                 RETURN(-ENOMEM);
 809
 810         fcb->cb_func = cb_func;
 811         fcb->cb_obd = obd;
 812         fcb->cb_last_rcvd = last_rcvd;
 813         fcb->cb_data = cb_data;
 814
 815         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 816         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 817                                     (struct journal_callback *)fcb);
 818
 819         return 0;
 820 }
 821
 822 /*
 823  * We need to hack the return value for the free inode counts because
 824  * the current EA code requires one filesystem block per inode with EAs,
 825  * so it is possible to run out of blocks before we run out of inodes.
 826  *
 827  * This can be removed when the ext3 EA code is fixed.
 828  */
 829 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 830 {
 831         struct kstatfs sfs;
 832         int rc;
 833
 834         memset(&sfs, 0, sizeof(sfs));
 835         rc = ll_do_statfs(sb,&sfs);
 836
 837         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 838                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 839                 sfs.f_ffree = sfs.f_bfree;
 840         }
 841
 842         statfs_pack(osfs, &sfs);
 843         return rc;
 844 }
 845
 846 static int fsfilt_ext3_sync(struct super_block *sb)
 847 {
 848         return ext3_force_commit(sb);
 849 }
 850
 851 #ifndef EXT3_EXTENTS_FL
 852 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 853 #endif
 854
 855 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 856 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 857 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 858 # define fsfilt_up_truncate_sem_in_cb(inode) do { } while (0)
 859 # define fsfilt_down_truncate_sem_in_cb(inode) do { } while (0)
 860 #else
 861 # ifdef HAVE_EXT4_LDISKFS
 862 #  ifdef WALK_SPACE_HAS_DATA_SEM /* We only use it in fsfilt_map_nblocks() for now */
 863 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 864 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 865 #   define fsfilt_up_truncate_sem_in_cb(inode) up_write((&EXT4_I(inode)->i_data_sem))
 866 #   define fsfilt_down_truncate_sem_in_cb(inode) down_write((&EXT4_I(inode)->i_data_sem))
 867 #  else
 868 #   define fsfilt_up_truncate_sem(inode) up_write((&LDISKFS_I(inode)->i_data_sem));
 869 #   define fsfilt_down_truncate_sem(inode) down_write((&LDISKFS_I(inode)->i_data_sem));
 870 #   define fsfilt_up_truncate_sem_in_cb(inode) do { } while (0)
 871 #   define fsfilt_down_truncate_sem_in_cb(inode) do { } while (0)
 872 #  endif
 873 # else
 874 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&LDISKFS_I(inode)->truncate_mutex);
 875 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&LDISKFS_I(inode)->truncate_mutex);
 876 #  define fsfilt_up_truncate_sem_in_cb(inode) do { } while (0)
 877 #  define fsfilt_down_truncate_sem_in_cb(inode) do { } while (0)
 878 # endif
 879 #endif
 880
 881 #ifndef EXT_ASSERT
 882 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 883 #endif
 884
 885 #ifdef EXT3_EXT_HAS_NO_TREE
 886 /* for kernels 2.6.18 and later */
 887 #ifdef HAVE_EXT4_LDISKFS
 888 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 889 #else
 890 #define EXT_GENERATION(inode)           ext_generation(inode)
 891 #endif
 892 #define ext3_ext_base                   inode
 893 #define ext3_ext_base2inode(inode)      (inode)
 894 #define EXT_DEPTH(inode)                ext_depth(inode)
 895 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 896                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 897 #else
 898 #define ext3_ext_base                   ext3_extents_tree
 899 #define ext3_ext_base2inode(tree)       (tree->inode)
 900 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 901                         ext3_ext_walk_space(tree, block, num, cb);
 902 #endif
 903
 904 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 905 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 906                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 907 #else
 908 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 909                ext3_ext_insert_extent(handle, inode, path, newext)
 910 #endif
 911
 912 #include <linux/lustre_version.h>
 913
 914 struct bpointers {
 915         unsigned long *blocks;
 916         int *created;
 917         unsigned long start;
 918         int num;
 919         int init_num;
 920         int create;
 921 };
 922
 923 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 924                               unsigned long block, int *aflags)
 925 {
 926         struct ext3_inode_info *ei = EXT3_I(inode);
 927         unsigned long bg_start;
 928         unsigned long colour;
 929         int depth;
 930
 931         if (path) {
 932                 struct ext3_extent *ex;
 933                 depth = path->p_depth;
 934
 935                 /* try to predict block placement */
 936                 if ((ex = path[depth].p_ext))
 937                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 938
 939                 /* it looks index is empty
 940                  * try to find starting from index itself */
 941                 if (path[depth].p_bh)
 942                         return path[depth].p_bh->b_blocknr;
 943         }
 944
 945         /* OK. use inode's group */
 946         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 947                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 948         colour = (current->pid % 16) *
 949                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 950         return bg_start + colour + block;
 951 }
 952
 953 #define ll_unmap_underlying_metadata(sb, blocknr) \
 954         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 955
 956 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 957 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 958                                 struct ext3_ext_path *path, unsigned long block,
 959                                 unsigned long *count, int *err)
 960 {
 961         unsigned long pblock, goal;
 962         int aflags = 0;
 963         struct inode *inode = ext3_ext_base2inode(base);
 964
 965         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 966         aflags |= 2; /* block have been already reserved */
 967         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 968         return pblock;
 969
 970 }
 971 #else
 972 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 973                                 struct ext3_ext_path *path, unsigned long block,
 974                                 unsigned long *count, int *err)
 975 {
 976         struct inode *inode = ext3_ext_base2inode(base);
 977         struct ext3_allocation_request ar;
 978         unsigned long pblock;
 979         int aflags;
 980
 981         /* find neighbour allocated blocks */
 982         ar.lleft = block;
 983         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 984         if (*err)
 985                 return 0;
 986         ar.lright = block;
 987         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 988         if (*err)
 989                 return 0;
 990
 991         /* allocate new block */
 992         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 993         ar.inode = inode;
 994         ar.logical = block;
 995         ar.len = *count;
 996         ar.flags = EXT3_MB_HINT_DATA;
 997         pblock = ext3_mb_new_blocks(handle, &ar, err);
 998         *count = ar.len;
 999         return pblock;
1000 }
1001 #endif
1002
1003 #ifdef EXT3_EXT_HAS_NO_TREE
1004 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1005                                   struct ext3_ext_path *path,
1006                                   struct ext3_ext_cache *cex,
1007 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1008                                   struct ext3_extent *ex,
1009 #endif
1010                                   void *cbdata)
1011 {
1012         struct bpointers *bp = cbdata;
1013 #else
1014 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1015                                   struct ext3_ext_path *path,
1016                                   struct ext3_ext_cache *cex
1017 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1018                                   , struct ext3_extent *ex
1019 #endif
1020                                  )
1021 {
1022         struct bpointers *bp = base->private;
1023 #endif
1024         struct inode *inode = ext3_ext_base2inode(base);
1025         struct ext3_extent nex;
1026         unsigned long pblock;
1027         unsigned long tgen;
1028         int err, i;
1029         unsigned long count;
1030         handle_t *handle;
1031
1032         i = EXT_DEPTH(base);
1033         EXT_ASSERT(i == path->p_depth);
1034         EXT_ASSERT(path[i].p_hdr);
1035
1036         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1037                 err = EXT_CONTINUE;
1038                 goto map;
1039         }
1040
1041         if (bp->create == 0) {
1042                 i = 0;
1043                 if (cex->ec_block < bp->start)
1044                         i = bp->start - cex->ec_block;
1045                 if (i >= cex->ec_len)
1046                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1047                                         i, cex->ec_len);
1048                 for (; i < cex->ec_len && bp->num; i++) {
1049                         *(bp->created) = 0;
1050                         bp->created++;
1051                         *(bp->blocks) = 0;
1052                         bp->blocks++;
1053                         bp->num--;
1054                         bp->start++;
1055                 }
1056
1057                 return EXT_CONTINUE;
1058         }
1059
1060         tgen = EXT_GENERATION(base);
1061         count = ext3_ext_calc_credits_for_insert(base, path);
1062         fsfilt_up_truncate_sem(inode);
1063         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1064         if (IS_ERR(handle)) {
1065                 fsfilt_down_truncate_sem(inode);
1066                 return PTR_ERR(handle);
1067         }
1068
1069         fsfilt_down_truncate_sem(inode);
1070         if (tgen != EXT_GENERATION(base)) {
1071                 /* the tree has changed. so path can be invalid at moment */
1072                 ext3_journal_stop(handle);
1073                 return EXT_REPEAT;
1074         }
1075
1076         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1077          * protected by i_data_sem, we need do it ourselves, since we create
1078          * file system blocks */
1079         fsfilt_down_truncate_sem_in_cb(inode);
1080
1081         count = cex->ec_len;
1082         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1083         if (!pblock)
1084                 goto out;
1085         EXT_ASSERT(count <= cex->ec_len);
1086
1087         /* insert new extent */
1088         nex.ee_block = cpu_to_le32(cex->ec_block);
1089         ext3_ext_store_pblock(&nex, pblock);
1090         nex.ee_len = cpu_to_le16(count);
1091         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1092         if (err) {
1093                 /* free data blocks we just allocated */
1094                 /* not a good idea to call discard here directly,
1095                  * but otherwise we'd need to call it every free() */
1096 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1097                 ext3_mb_discard_inode_preallocations(inode);
1098 #endif
1099                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1100                                  cpu_to_le16(nex.ee_len), 0);
1101                 goto out;
1102         }
1103
1104         /*
1105          * Putting len of the actual extent we just inserted,
1106          * we are asking ext3_ext_walk_space() to continue
1107          * scaning after that block
1108          */
1109         cex->ec_len = le16_to_cpu(nex.ee_len);
1110         cex->ec_start = ext_pblock(&nex);
1111         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1112         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1113
1114 out:
1115         fsfilt_up_truncate_sem_in_cb(inode);
1116         ext3_journal_stop(handle);
1117 map:
1118         if (err >= 0) {
1119                 /* map blocks */
1120                 if (bp->num == 0) {
1121                         CERROR("hmm. why do we find this extent?\n");
1122                         CERROR("initial space: %lu:%u\n",
1123                                 bp->start, bp->init_num);
1124                         CERROR("current extent: %u/%u/%llu %d\n",
1125                                 cex->ec_block, cex->ec_len,
1126                                 (unsigned long long)cex->ec_start,
1127                                 cex->ec_type);
1128                 }
1129                 i = 0;
1130                 if (cex->ec_block < bp->start)
1131                         i = bp->start - cex->ec_block;
1132                 if (i >= cex->ec_len)
1133                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1134                                         i, cex->ec_len);
1135                 for (; i < cex->ec_len && bp->num; i++) {
1136                         *(bp->blocks) = cex->ec_start + i;
1137                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1138                                 *(bp->created) = 0;
1139                         } else {
1140                                 *(bp->created) = 1;
1141                                 /* unmap any possible underlying metadata from
1142                                  * the block device mapping.  bug 6998. */
1143                                 ll_unmap_underlying_metadata(inode->i_sb,
1144                                                              *(bp->blocks));
1145                         }
1146                         bp->created++;
1147                         bp->blocks++;
1148                         bp->num--;
1149                         bp->start++;
1150                 }
1151         }
1152         return err;
1153 }
1154
1155 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1156                        unsigned long num, unsigned long *blocks,
1157                        int *created, int create)
1158 {
1159 #ifdef EXT3_EXT_HAS_NO_TREE
1160         struct ext3_ext_base *base = inode;
1161 #else
1162         struct ext3_extents_tree tree;
1163         struct ext3_ext_base *base = &tree;
1164 #endif
1165         struct bpointers bp;
1166         int err;
1167
1168         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1169                block, block + num - 1, (unsigned) inode->i_ino);
1170
1171 #ifndef EXT3_EXT_HAS_NO_TREE
1172         ext3_init_tree_desc(base, inode);
1173         tree.private = &bp;
1174 #endif
1175         bp.blocks = blocks;
1176         bp.created = created;
1177         bp.start = block;
1178         bp.init_num = bp.num = num;
1179         bp.create = create;
1180
1181         fsfilt_down_truncate_sem(inode);
1182         err = fsfilt_ext3_ext_walk_space(base, block, num, ext3_ext_new_extent_cb, &bp);
1183         ext3_ext_invalidate_cache(base);
1184         fsfilt_up_truncate_sem(inode);
1185
1186         return err;
1187 }
1188
1189 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1190                                     int pages, unsigned long *blocks,
1191                                     int *created, int create)
1192 {
1193         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1194         int rc = 0, i = 0;
1195         struct page *fp = NULL;
1196         int clen = 0;
1197
1198         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1199                 inode->i_ino, pages, (*page)->index);
1200
1201         /* pages are sorted already. so, we just have to find
1202          * contig. space and process them properly */
1203         while (i < pages) {
1204                 if (fp == NULL) {
1205                         /* start new extent */
1206                         fp = *page++;
1207                         clen = 1;
1208                         i++;
1209                         continue;
1210                 } else if (fp->index + clen == (*page)->index) {
1211                         /* continue the extent */
1212                         page++;
1213                         clen++;
1214                         i++;
1215                         continue;
1216                 }
1217
1218                 /* process found extent */
1219                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1220                                         clen * blocks_per_page, blocks,
1221                                         created, create);
1222                 if (rc)
1223                         GOTO(cleanup, rc);
1224
1225                 /* look for next extent */
1226                 fp = NULL;
1227                 blocks += blocks_per_page * clen;
1228                 created += blocks_per_page * clen;
1229         }
1230
1231         if (fp)
1232                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1233                                         clen * blocks_per_page, blocks,
1234                                         created, create);
1235 cleanup:
1236         return rc;
1237 }
1238
1239 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1240                                unsigned long *blocks, int *created, int create);
1241 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1242                                    int pages, unsigned long *blocks,
1243                                    int *created, int create)
1244 {
1245         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1246         unsigned long *b;
1247         int rc = 0, i, *cr;
1248
1249         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1250                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1251                 if (rc) {
1252                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1253                                inode->i_ino, *b, *cr, create, rc);
1254                         break;
1255                 }
1256
1257                 b += blocks_per_page;
1258                 cr += blocks_per_page;
1259         }
1260         return rc;
1261 }
1262
1263 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1264                                 int pages, unsigned long *blocks,
1265                                 int *created, int create,
1266                                 struct semaphore *optional_sem)
1267 {
1268         int rc;
1269
1270         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1271                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1272                                                      blocks, created, create);
1273                 return rc;
1274         }
1275         if (optional_sem != NULL)
1276                 down(optional_sem);
1277         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1278                                             created, create);
1279         if (optional_sem != NULL)
1280                 up(optional_sem);
1281
1282         return rc;
1283 }
1284
1285 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1286                                    int size, loff_t *offs)
1287 {
1288         struct inode *inode = file->f_dentry->d_inode;
1289         unsigned long block;
1290         struct buffer_head *bh;
1291         int err, blocksize, csize, boffs;
1292
1293         /* prevent reading after eof */
1294         lock_kernel();
1295         if (i_size_read(inode) < *offs + size) {
1296                 size = i_size_read(inode) - *offs;
1297                 unlock_kernel();
1298                 if (size < 0) {
1299                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1300                                i_size_read(inode), *offs);
1301                         return -EBADR;
1302                 } else if (size == 0) {
1303                         return 0;
1304                 }
1305         } else {
1306                 unlock_kernel();
1307         }
1308
1309         blocksize = 1 << inode->i_blkbits;
1310
1311         while (size > 0) {
1312                 block = *offs >> inode->i_blkbits;
1313                 boffs = *offs & (blocksize - 1);
1314                 csize = min(blocksize - boffs, size);
1315                 bh = ext3_bread(NULL, inode, block, 0, &err);
1316                 if (!bh) {
1317                         CERROR("can't read block: %d\n", err);
1318                         return err;
1319                 }
1320
1321                 memcpy(buf, bh->b_data + boffs, csize);
1322                 brelse(bh);
1323
1324                 *offs += csize;
1325                 buf += csize;
1326                 size -= csize;
1327         }
1328         return 0;
1329 }
1330
1331 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1332                                     loff_t *offs, int force_sync)
1333 {
1334         struct buffer_head *bh = NULL;
1335         unsigned long block;
1336         struct inode *inode = file->f_dentry->d_inode;
1337         loff_t old_size = i_size_read(inode), offset = *offs;
1338         loff_t new_size = i_size_read(inode);
1339         handle_t *handle;
1340         int err = 0, block_count = 0, blocksize, size, boffs;
1341
1342         /* Determine how many transaction credits are needed */
1343         blocksize = 1 << inode->i_blkbits;
1344         block_count = (*offs & (blocksize - 1)) + bufsize;
1345         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1346
1347         handle = ext3_journal_start(inode,
1348                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1349         if (IS_ERR(handle)) {
1350                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1351                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1352                 return PTR_ERR(handle);
1353         }
1354
1355         while (bufsize > 0) {
1356                 if (bh != NULL)
1357                         brelse(bh);
1358
1359                 block = offset >> inode->i_blkbits;
1360                 boffs = offset & (blocksize - 1);
1361                 size = min(blocksize - boffs, bufsize);
1362                 bh = ext3_bread(handle, inode, block, 1, &err);
1363                 if (!bh) {
1364                         CERROR("can't read/create block: %d\n", err);
1365                         goto out;
1366                 }
1367
1368                 err = ext3_journal_get_write_access(handle, bh);
1369                 if (err) {
1370                         CERROR("journal_get_write_access() returned error %d\n",
1371                                err);
1372                         goto out;
1373                 }
1374                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1375                 memcpy(bh->b_data + boffs, buf, size);
1376                 err = ext3_journal_dirty_metadata(handle, bh);
1377                 if (err) {
1378                         CERROR("journal_dirty_metadata() returned error %d\n",
1379                                err);
1380                         goto out;
1381                 }
1382                 if (offset + size > new_size)
1383                         new_size = offset + size;
1384                 offset += size;
1385                 bufsize -= size;
1386                 buf += size;
1387         }
1388
1389         if (force_sync)
1390                 handle->h_sync = 1; /* recovery likes this */
1391 out:
1392         if (bh)
1393                 brelse(bh);
1394
1395         /* correct in-core and on-disk sizes */
1396         if (new_size > i_size_read(inode)) {
1397                 lock_kernel();
1398                 if (new_size > i_size_read(inode))
1399                         i_size_write(inode, new_size);
1400                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1401                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1402                 if (i_size_read(inode) > old_size)
1403                         mark_inode_dirty(inode);
1404                 unlock_kernel();
1405         }
1406
1407         ext3_journal_stop(handle);
1408
1409         if (err == 0)
1410                 *offs = offset;
1411         return err;
1412 }
1413
1414 static int fsfilt_ext3_setup(struct super_block *sb)
1415 {
1416 #if !defined(S_PDIROPS) && (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1417         struct ext3_sb_info *sbi = EXT3_SB(sb);
1418 #endif
1419 #if 0
1420         sbi->dx_lock = fsfilt_ext3_dx_lock;
1421         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1422 #endif
1423 #ifdef S_PDIROPS
1424         CWARN("Enabling PDIROPS\n");
1425         set_opt(sbi->s_mount_opt, PDIROPS);
1426         sb->s_flags |= S_PDIROPS;
1427 #endif
1428         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1429                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1430 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1431         /* enable journaled quota support */
1432         /* kfreed in ext3_put_super() */
1433         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1434         if (!sbi->s_qf_names[USRQUOTA])
1435                 return -ENOMEM;
1436         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1437         if (!sbi->s_qf_names[GRPQUOTA]) {
1438                 kfree(sbi->s_qf_names[USRQUOTA]);
1439                 sbi->s_qf_names[USRQUOTA] = NULL;
1440                 return -ENOMEM;
1441         }
1442         sbi->s_jquota_fmt = QFMT_VFS_V0;
1443 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1444         set_opt(sbi->s_mount_opt, QUOTA);
1445 #endif
1446 #endif
1447         return 0;
1448 }
1449
1450 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1451    objects. Logs is number of logfiles to update */
1452 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1453 {
1454         if ( !fso ) {
1455                 switch(op) {
1456                 case FSFILT_OP_CREATE:
1457                                  /* directory leaf, index & indirect & EA*/
1458                         return 4 + 3 * logs;
1459                 case FSFILT_OP_UNLINK:
1460                         return 3 * logs;
1461                 }
1462         } else {
1463                 int i;
1464                 int needed = 0;
1465                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1466                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1467                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1468                 for (i = 0; i < op; i++, fso++) {
1469                         int nblocks = fso->fso_bufcnt * blockpp;
1470                         int ndindirect = min(nblocks, addrpp + 1);
1471                         int nindir = nblocks + ndindirect + 1;
1472
1473                         needed += nindir;
1474                 }
1475                 return needed + 3 * logs;
1476         }
1477
1478         return 0;
1479 }
1480
1481 #ifdef HAVE_QUOTA_SUPPORT
1482 #define DQINFO_COPY(out, in)                    \
1483 do {                                            \
1484         Q_COPY(out, in, dqi_bgrace);            \
1485         Q_COPY(out, in, dqi_igrace);            \
1486         Q_COPY(out, in, dqi_flags);             \
1487         Q_COPY(out, in, dqi_valid);             \
1488 } while (0)
1489
1490 #define DQBLK_COPY(out, in)                     \
1491 do {                                            \
1492         Q_COPY(out, in, dqb_bhardlimit);        \
1493         Q_COPY(out, in, dqb_bsoftlimit);        \
1494         Q_COPY(out, in, dqb_curspace);          \
1495         Q_COPY(out, in, dqb_ihardlimit);        \
1496         Q_COPY(out, in, dqb_isoftlimit);        \
1497         Q_COPY(out, in, dqb_curinodes);         \
1498         Q_COPY(out, in, dqb_btime);             \
1499         Q_COPY(out, in, dqb_itime);             \
1500         Q_COPY(out, in, dqb_valid);             \
1501 } while (0)
1502
1503
1504
1505 static int fsfilt_ext3_quotactl(struct super_block *sb,
1506                                 struct obd_quotactl *oqc)
1507 {
1508         int i, rc = 0, error = 0;
1509         const struct quotactl_ops *qcop;
1510         struct if_dqinfo *info;
1511         struct if_dqblk *dqblk;
1512         ENTRY;
1513
1514         if (!sb->s_qcop)
1515                 RETURN(-ENOSYS);
1516
1517         OBD_ALLOC_PTR(info);
1518         if (!info)
1519                 RETURN(-ENOMEM);
1520         OBD_ALLOC_PTR(dqblk);
1521         if (!dqblk) {
1522                 OBD_FREE_PTR(info);
1523                 RETURN(-ENOMEM);
1524         }
1525
1526         DQINFO_COPY(info, &oqc->qc_dqinfo);
1527         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1528
1529         qcop = sb->s_qcop;
1530         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1531                 for (i = 0; i < MAXQUOTAS; i++) {
1532                         if (!Q_TYPESET(oqc, i))
1533                                 continue;
1534
1535                         if (oqc->qc_cmd == Q_QUOTAON) {
1536                                 lustre_quota_version_t qfmt = oqc->qc_id;
1537                                 char *name[][MAXQUOTAS] = LUSTRE_OPQFILES_NAMES;
1538
1539                                 rc = ll_quota_on(sb, i, QFMT_VFS_V0,
1540                                                  name[qfmt][i], 0);
1541 #ifdef HAVE_QUOTA64
1542                                 if (rc == -ENOENT || rc == -EINVAL) {
1543                                         /* see bug 13904 */
1544                                         rc = lustre_slave_quota_convert(qfmt, i);
1545                                         if (!rc)
1546                                                 rc = ll_quota_on(sb, i,
1547                                                               QFMT_VFS_V0,
1548                                                               name[qfmt][i], 0);
1549                                         else if (rc == -ESTALE)
1550                                                 rc = -ENOENT;
1551                                 }
1552 #endif
1553                         } else if (oqc->qc_cmd == Q_QUOTAOFF)
1554                                 rc = ll_quota_off(sb, i, 0);
1555
1556                         if (rc == -EBUSY)
1557                                 error = rc;
1558                         else if (rc)
1559                                 GOTO(out, rc);
1560                 }
1561                 GOTO(out, rc ?: error);
1562         }
1563
1564         switch (oqc->qc_cmd) {
1565         case Q_GETOINFO:
1566         case Q_GETINFO:
1567                 if (!qcop->get_info)
1568                         GOTO(out, rc = -ENOSYS);
1569                 rc = qcop->get_info(sb, oqc->qc_type, info);
1570                 break;
1571         case Q_SETQUOTA:
1572         case Q_INITQUOTA:
1573                 if (!qcop->set_dqblk)
1574                         GOTO(out, rc = -ENOSYS);
1575                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1576                 break;
1577         case Q_GETOQUOTA:
1578         case Q_GETQUOTA:
1579                 if (!qcop->get_dqblk)
1580                         GOTO(out, rc = -ENOSYS);
1581                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1582                 if (!rc)
1583                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1584                 break;
1585         case Q_SYNC:
1586                 if (!sb->s_qcop->quota_sync)
1587                         GOTO(out, rc = -ENOSYS);
1588                 qcop->quota_sync(sb, oqc->qc_type);
1589                 break;
1590         case Q_FINVALIDATE:
1591                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1592                 for (i = 0; i < MAXQUOTAS; i++) {
1593                         struct file *fp;
1594                         lustre_quota_version_t qfmt = oqc->qc_id;
1595                         char *name[][MAXQUOTAS] = LUSTRE_OPQFILES_NAMES;
1596
1597                         if (!Q_TYPESET(oqc, i))
1598                                 continue;
1599
1600                         fp = filp_open(name[qfmt][i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1601                         if (IS_ERR(fp)) {
1602                                 rc = PTR_ERR(fp);
1603                                 CERROR("error invalidating operational quota file"
1604                                        " %s (rc:%d)\n", name[qfmt][i], rc);
1605                         } else {
1606                                 filp_close(fp, 0);
1607                         }
1608
1609                 }
1610                 break;
1611         default:
1612                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1613                 LBUG();
1614         }
1615 out:
1616         DQINFO_COPY(&oqc->qc_dqinfo, info);
1617         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1618
1619         OBD_FREE_PTR(info);
1620         OBD_FREE_PTR(dqblk);
1621
1622         if (rc)
1623                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1624                                 "failed: %d\n",
1625                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1626         RETURN(rc);
1627 }
1628
1629 struct chk_dqblk{
1630         struct hlist_node       dqb_hash;        /* quotacheck hash */
1631         struct list_head        dqb_list;        /* in list also */
1632         qid_t                   dqb_id;          /* uid/gid */
1633         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1634         qsize_t                 dqb_bhardlimit;  /* block hard limit */
1635         qsize_t                 dqb_bsoftlimit;  /* block soft limit */
1636         qsize_t                 dqb_curspace;    /* current space */
1637         qsize_t                 dqb_ihardlimit;  /* inode hard limit */
1638         qsize_t                 dqb_isoftlimit;  /* inode soft limit */
1639         qsize_t                 dqb_curinodes;   /* current inodes */
1640         __u64                   dqb_btime;       /* block grace time */
1641         __u64                   dqb_itime;       /* inode grace time */
1642         __u32                   dqb_valid;       /* flag for above fields */
1643 };
1644
1645 static inline unsigned int chkquot_hash(qid_t id, int type)
1646                                         __attribute__((__const__));
1647
1648 static inline unsigned int chkquot_hash(qid_t id, int type)
1649 {
1650         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1651 }
1652
1653 static inline struct chk_dqblk *
1654 find_chkquot(struct hlist_head *head, qid_t id, int type)
1655 {
1656         struct hlist_node *node;
1657         struct chk_dqblk *cdqb;
1658
1659         hlist_for_each(node, head) {
1660                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1661                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1662                         return cdqb;
1663         }
1664
1665         return NULL;
1666 }
1667
1668 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1669 {
1670         struct chk_dqblk *cdqb;
1671
1672         OBD_ALLOC_PTR(cdqb);
1673         if (cdqb) {
1674                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1675                 INIT_LIST_HEAD(&cdqb->dqb_list);
1676                 cdqb->dqb_id = id;
1677                 cdqb->dqb_type = type;
1678         }
1679
1680         return cdqb;
1681 }
1682
1683 static struct chk_dqblk *
1684 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1685       qid_t id, int type, int first_check)
1686 {
1687         struct hlist_head *head = hash + chkquot_hash(id, type);
1688         struct if_dqblk dqb;
1689         struct chk_dqblk *cdqb;
1690         int rc;
1691
1692         cdqb = find_chkquot(head, id, type);
1693         if (cdqb)
1694                 return cdqb;
1695
1696         cdqb = alloc_chkquot(id, type);
1697         if (!cdqb)
1698                 return NULL;
1699
1700         if (!first_check) {
1701                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1702                 if (rc) {
1703                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1704                                id, type, rc);
1705                 } else {
1706                         DQBLK_COPY(cdqb, &dqb);
1707                         cdqb->dqb_curspace = 0;
1708                         cdqb->dqb_curinodes = 0;
1709                 }
1710         }
1711
1712         hlist_add_head(&cdqb->dqb_hash, head);
1713         list_add_tail(&cdqb->dqb_list, list);
1714
1715         return cdqb;
1716 }
1717
1718 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1719 {
1720         struct obd_quotactl *oqctl;
1721         int rc;
1722
1723         OBD_ALLOC_PTR(oqctl);
1724         if (!oqctl)
1725                 RETURN(-ENOMEM);
1726
1727         oqctl->qc_cmd = cmd;
1728         oqctl->qc_id = qfmt;
1729         oqctl->qc_type = type;
1730         rc = fsfilt_ext3_quotactl(sb, oqctl);
1731
1732         OBD_FREE_PTR(oqctl);
1733         return rc;
1734 }
1735
1736 static inline int read_old_dqinfo(struct super_block *sb, int type,
1737                                   struct if_dqinfo *dqinfo)
1738 {
1739         struct obd_quotactl *oqctl;
1740         int rc;
1741         ENTRY;
1742
1743         OBD_ALLOC_PTR(oqctl);
1744         if (!oqctl)
1745                 RETURN(-ENOMEM);
1746
1747         oqctl->qc_cmd = Q_GETINFO;
1748         oqctl->qc_type = type;
1749         rc = fsfilt_ext3_quotactl(sb, oqctl);
1750         if (!rc)
1751                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1752
1753         OBD_FREE_PTR(oqctl);
1754         RETURN(rc);
1755 }
1756
1757 #ifndef HAVE_EXT4_LDISKFS
1758 static inline struct ext3_group_desc *
1759 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1760 {
1761         unsigned long desc_block, desc;
1762         struct ext3_group_desc *gdp;
1763
1764         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1765         desc = group % EXT3_DESC_PER_BLOCK(sb);
1766         gdp = (struct ext3_group_desc *)
1767               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1768
1769         return gdp + desc;
1770 }
1771
1772 static inline struct buffer_head *
1773 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1774 {
1775         struct ext3_group_desc *desc;
1776         struct buffer_head *bh;
1777
1778         desc = get_group_desc(sb, group, NULL);
1779         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1780         return bh;
1781 }
1782
1783 static __u32 ext3_itable_unused_count(struct super_block *sb,
1784                                struct ext3_group_desc *bg) {
1785        return le16_to_cpu(bg->bg_itable_unused);
1786 }
1787 #else
1788 #define get_group_desc ext3_get_group_desc
1789 #endif
1790
1791 struct qchk_ctxt {
1792         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1793         struct list_head        qckt_list;                   /* quotacheck list */
1794         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1795         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1796 };
1797
1798 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1799                            struct obd_quotactl *oqc)
1800 {
1801         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1802         loff_t size = 0;
1803         qid_t qid[MAXQUOTAS];
1804         int cnt, i, rc = 0;
1805
1806         if (!inode)
1807                 return 0;
1808
1809         qid[USRQUOTA] = inode->i_uid;
1810         qid[GRPQUOTA] = inode->i_gid;
1811
1812         if (S_ISDIR(inode->i_mode) ||
1813             S_ISREG(inode->i_mode) ||
1814             S_ISLNK(inode->i_mode))
1815                 size = inode_get_bytes(inode);
1816
1817         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1818                 if (!Q_TYPESET(oqc, cnt))
1819                         continue;
1820
1821                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1822                                 &qctxt->qckt_list, qid[cnt], cnt,
1823                                 qctxt->qckt_first_check[cnt]);
1824                 if (!cdqb[cnt]) {
1825                         rc = -ENOMEM;
1826                         break;
1827                 }
1828
1829                 cdqb[cnt]->dqb_curspace += size;
1830                 cdqb[cnt]->dqb_curinodes++;
1831         }
1832
1833         if (rc) {
1834                 for (i = 0; i < cnt; i++) {
1835                         if (!Q_TYPESET(oqc, i))
1836                                 continue;
1837                         LASSERT(cdqb[i]);
1838                         cdqb[i]->dqb_curspace -= size;
1839                         cdqb[i]->dqb_curinodes--;
1840                 }
1841         }
1842
1843         return rc;
1844 }
1845
1846 static int v2_write_dqheader(struct file *f, int type)
1847 {
1848         static const __u32 quota_magics[] = V2_INITQMAGICS;
1849         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V1;
1850         struct v2_disk_dqheader dqhead;
1851         loff_t offset = 0;
1852
1853         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1854         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1855
1856         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1857         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1858
1859         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1860 }
1861
1862 /* write dqinfo struct in a new quota file */
1863 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1864 {
1865         struct v2_disk_dqinfo dqinfo;
1866         __u32 blocks = V2_DQTREEOFF + 1;
1867         loff_t offset = V2_DQINFOOFF;
1868
1869         if (info) {
1870                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1871                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1872                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1873                                                ~DQF_INFO_DIRTY);
1874         } else {
1875                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1876                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1877                 dqinfo.dqi_flags = 0;
1878         }
1879
1880         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1881         dqinfo.dqi_free_blk = 0;
1882         dqinfo.dqi_free_entry = 0;
1883
1884         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1885 }
1886
1887 #ifdef HAVE_QUOTA64
1888 static int v3_write_dqheader(struct file *f, int type)
1889 {
1890         static const __u32 quota_magics[] = V2_INITQMAGICS;
1891         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1892         struct v2_disk_dqheader dqhead;
1893         loff_t offset = 0;
1894
1895         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1896         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1897
1898         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1899         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1900
1901         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1902 }
1903
1904 /* write dqinfo struct in a new quota file */
1905 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1906 {
1907         return v2_write_dqinfo(f, type, info);
1908 }
1909 #endif
1910
1911 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1912                                   struct obd_quotactl *oqc)
1913 {
1914         int i, rc = 0;
1915         ENTRY;
1916
1917         for (i = 0; i < MAXQUOTAS; i++) {
1918                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1919                                          NULL : &qctxt->qckt_dqinfo[i];
1920                 struct file *file;
1921                 const char *name[][MAXQUOTAS] = LUSTRE_OPQFILES_NAMES;
1922                 int (*write_dqheader)(struct file *, int);
1923                 int (*write_dqinfo)(struct file *, int, struct if_dqinfo *);
1924
1925                 if (!Q_TYPESET(oqc, i))
1926                         continue;
1927
1928                 file = filp_open(name[oqc->qc_id][i],
1929                                  O_RDWR | O_CREAT | O_TRUNC, 0644);
1930                 if (IS_ERR(file)) {
1931                         rc = PTR_ERR(file);
1932                         CERROR("can't create %s file: rc = %d\n",
1933                                name[oqc->qc_id][i], rc);
1934                         GOTO(out, rc);
1935                 }
1936
1937                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1938                         CERROR("file %s is not regular", name[oqc->qc_id][i]);
1939                         filp_close(file, 0);
1940                         GOTO(out, rc = -EINVAL);
1941                 }
1942
1943                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1944
1945                 switch (oqc->qc_id) {
1946                 case LUSTRE_QUOTA_V1 : write_dqheader = v2_write_dqheader;
1947                                        write_dqinfo   = v2_write_dqinfo;
1948                                        break;
1949 #ifdef HAVE_QUOTA64
1950                 case LUSTRE_QUOTA_V2 : write_dqheader = v3_write_dqheader;
1951                                        write_dqinfo   = v3_write_dqinfo;
1952                                        break;
1953 #endif
1954                 default              : CERROR("unknown quota format!\n");
1955                                        LBUG();
1956                 }
1957
1958                 rc = (*write_dqheader)(file, i);
1959                 if (rc) {
1960                         filp_close(file, 0);
1961                         GOTO(out, rc);
1962                 }
1963
1964                 rc = (*write_dqinfo)(file, i, info);
1965                 filp_close(file, 0);
1966                 if (rc)
1967                         GOTO(out, rc);
1968         }
1969
1970 out:
1971         RETURN(rc);
1972 }
1973
1974
1975 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1976                           struct chk_dqblk *cdqb)
1977 {
1978         struct obd_quotactl *oqc;
1979         long now;
1980         int rc;
1981         ENTRY;
1982
1983         OBD_ALLOC_PTR(oqc);
1984         if (!oqc)
1985                 RETURN(-ENOMEM);
1986
1987         now = CURRENT_SECONDS;
1988
1989         if (cdqb->dqb_bsoftlimit &&
1990             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1991             !cdqb->dqb_btime)
1992                 cdqb->dqb_btime =
1993                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1994
1995         if (cdqb->dqb_isoftlimit &&
1996             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1997             !cdqb->dqb_itime)
1998                 cdqb->dqb_itime =
1999                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
2000
2001         cdqb->dqb_valid = QIF_ALL;
2002
2003         oqc->qc_cmd = Q_SETQUOTA;
2004         oqc->qc_type = cdqb->dqb_type;
2005         oqc->qc_id = cdqb->dqb_id;
2006         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
2007
2008         rc = fsfilt_ext3_quotactl(sb, oqc);
2009         OBD_FREE_PTR(oqc);
2010         RETURN(rc);
2011 }
2012
2013 static int prune_chkquots(struct super_block *sb,
2014                           struct qchk_ctxt *qctxt, int error)
2015 {
2016         struct chk_dqblk *cdqb, *tmp;
2017         int rc;
2018
2019         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
2020                 if (!error) {
2021                         rc = commit_chkquot(sb, qctxt, cdqb);
2022                         if (rc)
2023                                 error = rc;
2024                 }
2025                 hlist_del_init(&cdqb->dqb_hash);
2026                 list_del(&cdqb->dqb_list);
2027                 OBD_FREE_PTR(cdqb);
2028         }
2029
2030         return error;
2031 }
2032
2033 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2034 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2035 #endif
2036
2037 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2038                                   struct obd_quotactl *oqc)
2039 {
2040         struct ext3_sb_info *sbi = EXT3_SB(sb);
2041         int i, group, uninit_feat = 0;
2042         struct qchk_ctxt *qctxt;
2043         struct buffer_head *bitmap_bh = NULL;
2044         unsigned long ino, inode_inuse;
2045         struct inode *inode;
2046         int rc = 0;
2047         ENTRY;
2048
2049         /* turn on quota and read dqinfo if existed */
2050         OBD_ALLOC_PTR(qctxt);
2051         if (!qctxt) {
2052                 oqc->qc_stat = -ENOMEM;
2053                 RETURN(-ENOMEM);
2054         }
2055
2056         for (i = 0; i < NR_DQHASH; i++)
2057                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2058         INIT_LIST_HEAD(&qctxt->qckt_list);
2059
2060         for (i = 0; i < MAXQUOTAS; i++) {
2061                 if (!Q_TYPESET(oqc, i))
2062                         continue;
2063
2064                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2065                 if (!rc || rc == -EBUSY) {
2066                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2067                         if (rc)
2068                                 GOTO(out, rc);
2069                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2070                         qctxt->qckt_first_check[i] = 1;
2071                 } else if (rc) {
2072                         GOTO(out, rc);
2073                 }
2074         }
2075         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2076                 /* This filesystem supports the uninit group feature */
2077                 uninit_feat = 1;
2078
2079         /* number of inodes that have been allocated */
2080         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2081 #ifdef EXT3_SBI_WITH_COUNTER64
2082                       percpu_counter64_sum(&sbi->s_freeinodes_counter);
2083 #else
2084                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2085 #endif
2086
2087         /* check quota and update in hash */
2088         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2089              group++) {
2090                 unsigned long used_count = sbi->s_inodes_per_group;
2091
2092                 if (uninit_feat) {
2093                         struct ext3_group_desc *desc;
2094                         desc = get_group_desc(sb, group, NULL);
2095                         if (!desc)
2096                                 GOTO(out, -EIO);
2097
2098                         /* we don't really need to take the group lock here,
2099                          * but it may be useful if one day we support online
2100                          * quotacheck */
2101 #ifdef HAVE_EXT4_LDISKFS
2102                         ext4_lock_group(sb, group);
2103 #else
2104                         spin_lock(sb_bgl_lock(sbi, group));
2105 #endif
2106                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2107                                 /* no inode in use in this group, just skip it */
2108 #ifdef HAVE_EXT4_LDISKFS
2109                                 ext3_unlock_group(sb, group);
2110 #else
2111                                 spin_unlock(sb_bgl_lock(sbi, group));
2112 #endif
2113                                 continue;
2114                         }
2115
2116                         used_count -= ext3_itable_unused_count(sb, desc);
2117 #ifdef HAVE_EXT4_LDISKFS
2118                         ext3_unlock_group(sb, group);
2119 #else
2120                         spin_unlock(sb_bgl_lock(sbi, group));
2121 #endif
2122                 }
2123
2124                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2125                 if (!bitmap_bh) {
2126                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2127                                sb->s_id, group);
2128                         GOTO(out, -EIO);
2129                 }
2130
2131                 i = 0;
2132                 while (i < used_count &&
2133                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2134                                                used_count, i)) < used_count) {
2135                         inode_inuse--;
2136                         i++;
2137                         ino = i + group * sbi->s_inodes_per_group;
2138                         if (ino < sbi->s_first_ino)
2139                                 continue;
2140 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2141                         inode = ext3_iget(sb, ino);
2142 #else
2143                         inode = iget(sb, ino);
2144 #endif
2145                         if (!inode || IS_ERR(inode))
2146                                 continue;
2147
2148                         rc = add_inode_quota(inode, qctxt, oqc);
2149                         iput(inode);
2150                         if (rc) {
2151                                 brelse(bitmap_bh);
2152                                 GOTO(out, rc);
2153                         }
2154                 }
2155                 brelse(bitmap_bh);
2156         }
2157
2158         /* read old quota limits from old quota file. (only for the user
2159          * has limits but hasn't file) */
2160 #ifdef HAVE_QUOTA_SUPPORT
2161         for (i = 0; i < MAXQUOTAS; i++) {
2162                 struct list_head id_list;
2163                 struct dquot_id *dqid, *tmp;
2164
2165                 if (!Q_TYPESET(oqc, i))
2166                         continue;
2167
2168                 if (qctxt->qckt_first_check[i])
2169                         continue;
2170
2171
2172                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2173                 INIT_LIST_HEAD(&id_list);
2174 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2175                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2176 #else
2177                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2178 #endif
2179                 if (rc)
2180                         CERROR("read old limits failed. (rc:%d)\n", rc);
2181
2182                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2183                         list_del_init(&dqid->di_link);
2184
2185                         if (!rc)
2186                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2187                                       dqid->di_id, i,
2188                                       qctxt->qckt_first_check[i]);
2189                         OBD_FREE_PTR(dqid);
2190                 }
2191         }
2192 #endif
2193         /* turn off quota cause we are to dump chk_dqblk to files */
2194         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2195
2196         rc = create_new_quota_files(qctxt, oqc);
2197         if (rc)
2198                 GOTO(out, rc);
2199
2200         /* we use vfs functions to set dqblk, so turn quota on */
2201         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2202 out:
2203         /* dump and free chk_dqblk */
2204         rc = prune_chkquots(sb, qctxt, rc);
2205         OBD_FREE_PTR(qctxt);
2206
2207         /* turn off quota, `lfs quotacheck` will turn on when all
2208          * nodes quotacheck finish. */
2209         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2210
2211         oqc->qc_stat = rc;
2212         if (rc)
2213                 CERROR("quotacheck failed: rc = %d\n", rc);
2214
2215         RETURN(rc);
2216 }
2217
2218 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2219                                  int cmd)
2220 {
2221         int rc = 0;
2222         ENTRY;
2223
2224         if (lqi->qi_files[type] == NULL) {
2225                 CERROR("operate qinfo before it's enabled!\n");
2226                 RETURN(-EIO);
2227         }
2228
2229         switch (cmd) {
2230         case QFILE_CHK:
2231                 rc = lustre_check_quota_file(lqi, type);
2232                 break;
2233         case QFILE_RD_INFO:
2234                 rc = lustre_read_quota_info(lqi, type);
2235                 break;
2236         case QFILE_WR_INFO:
2237                 rc = lustre_write_quota_info(lqi, type);
2238                 break;
2239         case QFILE_INIT_INFO:
2240                 rc = lustre_init_quota_info(lqi, type);
2241                 break;
2242         case QFILE_CONVERT:
2243                 rc = lustre_quota_convert(lqi, type);
2244                 break;
2245         default:
2246                 rc = -ENOTSUPP;
2247                 CERROR("Unsupported admin quota file cmd %d\n"
2248                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2249                        cmd);
2250                 break;
2251         }
2252         RETURN(rc);
2253 }
2254
2255 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2256                             struct list_head *list)
2257 {
2258         return lustre_get_qids(file, inode, type, list);
2259 }
2260
2261 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2262 {
2263         int rc = 0;
2264         ENTRY;
2265
2266         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2267                 CERROR("operate dquot before it's enabled!\n");
2268                 RETURN(-EIO);
2269         }
2270
2271         switch (cmd) {
2272         case QFILE_RD_DQUOT:
2273                 rc = lustre_read_dquot(dquot);
2274                 break;
2275         case QFILE_WR_DQUOT:
2276                 if (dquot->dq_dqb.dqb_ihardlimit ||
2277                     dquot->dq_dqb.dqb_isoftlimit ||
2278                     dquot->dq_dqb.dqb_bhardlimit ||
2279                     dquot->dq_dqb.dqb_bsoftlimit)
2280                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2281                 else
2282                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2283
2284                 rc = lustre_commit_dquot(dquot);
2285                 if (rc >= 0)
2286                         rc = 0;
2287                 break;
2288         default:
2289                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2290                 LBUG();
2291                 break;
2292         }
2293         RETURN(rc);
2294 }
2295
2296 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2297                                 struct inode *inode, int frags)
2298 {
2299 #ifdef EXT3_EXT_HAS_NO_TREE
2300         struct ext3_ext_base *base = inode;
2301 #else
2302         struct ext3_extents_tree tree;
2303         struct ext3_ext_base *base = &tree;
2304
2305         ext3_init_tree_desc(base, inode);
2306 #endif
2307         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2308          * metablocks at maxium b=16542 */
2309         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2310         return 0;
2311 }
2312
2313 #endif
2314
2315 static lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2316 {
2317         return (EXT3_SB(sb)->journal_bdev);
2318 }
2319
2320 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2321                           char *buf, int count, loff_t pos)
2322 {
2323         loff_t p = pos;
2324         int rc;
2325
2326         /* Support for both adm and op quota files must be provided */
2327         if (f) {
2328                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2329                 rc = rc < 0 ? rc : p - pos;
2330         } else {
2331                 struct super_block *sb = inode->i_sb;
2332                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2333         }
2334         return rc;
2335 }
2336
2337 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2338 {
2339         loff_t p = pos;
2340         int rc;
2341
2342         /* Only adm quota files are supported, op updates are handled by vfs */
2343         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2344         rc = rc < 0 ? rc : p - pos;
2345
2346         return rc;
2347 }
2348
2349 void *lustre_quota_journal_start(struct inode *inode, int delete)
2350 {
2351         handle_t *handle;
2352         unsigned block_count;
2353
2354         if (delete) {
2355                 /* each indirect block (+4) may become free, attaching to the
2356                  * header list of free blocks (+1); the data block (+1) may
2357                  * become a free block (+0) or a block with free dqentries (+0) */
2358                 block_count = (4 + 1) + 1;
2359                 handle = ext3_journal_start(inode,
2360                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2361         } else {
2362                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2363                  * freeblk reusage with a header update (+1); dqentry is either reused
2364                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2365                  * a new block allocation (+1) with a header update (+1)              */
2366                 block_count = (4 + 1) + 3;
2367                 handle = ext3_journal_start(inode,
2368                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2369
2370         }
2371
2372         return handle;
2373 }
2374
2375 void lustre_quota_journal_stop(void *handle)
2376 {
2377         ext3_journal_stop((handle_t *)handle);
2378 }
2379
2380 static struct fsfilt_operations fsfilt_ext3_ops = {
2381         .fs_type                = "ext3",
2382         .fs_owner               = THIS_MODULE,
2383         .fs_getlabel            = fsfilt_ext3_get_label,
2384         .fs_setlabel            = fsfilt_ext3_set_label,
2385         .fs_uuid                = fsfilt_ext3_uuid,
2386         .fs_start               = fsfilt_ext3_start,
2387         .fs_brw_start           = fsfilt_ext3_brw_start,
2388         .fs_extend              = fsfilt_ext3_extend,
2389         .fs_commit              = fsfilt_ext3_commit,
2390         .fs_commit_async        = fsfilt_ext3_commit_async,
2391         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2392         .fs_setattr             = fsfilt_ext3_setattr,
2393         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2394         .fs_set_md              = fsfilt_ext3_set_md,
2395         .fs_get_md              = fsfilt_ext3_get_md,
2396         .fs_readpage            = fsfilt_ext3_readpage,
2397         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2398         .fs_statfs              = fsfilt_ext3_statfs,
2399         .fs_sync                = fsfilt_ext3_sync,
2400         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2401         .fs_write_record        = fsfilt_ext3_write_record,
2402         .fs_read_record         = fsfilt_ext3_read_record,
2403         .fs_setup               = fsfilt_ext3_setup,
2404         .fs_send_bio            = fsfilt_ext3_send_bio,
2405         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2406 #ifdef HAVE_DISK_INODE_VERSION
2407         .fs_get_version         = fsfilt_ext3_get_version,
2408         .fs_set_version         = fsfilt_ext3_set_version,
2409 #endif
2410 #ifdef HAVE_QUOTA_SUPPORT
2411         .fs_quotactl            = fsfilt_ext3_quotactl,
2412         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2413         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2414         .fs_qids                = fsfilt_ext3_qids,
2415         .fs_dquot               = fsfilt_ext3_dquot,
2416         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2417 #endif
2418         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2419 };
2420
2421 static int __init fsfilt_ext3_init(void)
2422 {
2423         int rc;
2424
2425         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2426                                          sizeof(struct fsfilt_cb_data), 0, 0);
2427         if (!fcb_cache) {
2428                 CERROR("error allocating fsfilt journal callback cache\n");
2429                 GOTO(out, rc = -ENOMEM);
2430         }
2431
2432         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2433
2434         if (rc) {
2435                 int err = cfs_mem_cache_destroy(fcb_cache);
2436                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2437         }
2438 out:
2439         return rc;
2440 }
2441
2442 static void __exit fsfilt_ext3_exit(void)
2443 {
2444         int rc;
2445
2446         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2447         rc = cfs_mem_cache_destroy(fcb_cache);
2448         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2449 }
2450
2451 module_init(fsfilt_ext3_init);
2452 module_exit(fsfilt_ext3_exit);
2453
2454 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2455 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2456 MODULE_LICENSE("GPL");