lustre/lvfs/fsfilt_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011 Whamcloud, Inc.
  33  *
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/lvfs/fsfilt_ext3.c
  40  *
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_FILTER
  45
  46 #include <linux/init.h>
  47 #include <linux/module.h>
  48 #include <linux/fs.h>
  49 #include <linux/slab.h>
  50 #include <linux/pagemap.h>
  51 #include <linux/quotaops.h>
  52 #ifdef HAVE_LINUX_EXPORTFS_H
  53 #include <linux/exportfs.h>
  54 #endif
  55 #ifdef HAVE_EXT4_LDISKFS
  56 #include <ext4/ext4.h>
  57 #include <ext4/ext4_jbd2.h>
  58 #else
  59 #include <linux/jbd.h>
  60 #include <linux/ext3_fs.h>
  61 #include <linux/ext3_jbd.h>
  62 #endif
  63 #include <linux/version.h>
  64 #include <linux/bitops.h>
  65 #include <linux/quota.h>
  66 #ifdef HAVE_QUOTAIO_H
  67 # include <linux/quotaio_v2.h>
  68 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  69 # include <quota/quotaio_v2.h>
  70 # include <quota/quota_tree.h>
  71 # define V2_DQTREEOFF    QT_TREEOFF
  72 #elif defined(HAVE_FS_QUOTAIO_V1_H)
  73 # include <quotaio_v2.h>
  74 # include <quota_tree.h>
  75 # define V2_DQTREEOFF    QT_TREEOFF
  76 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  77 #endif
  78
  79 #ifdef QFMT_VFS_V1
  80 #define QFMT_LUSTRE QFMT_VFS_V1
  81 #else
  82 #define QFMT_LUSTRE QFMT_VFS_V0
  83 #endif
  84
  85 #if defined(HAVE_EXT3_XATTR_H)
  86 #include <ext3/xattr.h>
  87 #else
  88 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  89 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  90 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  91 #endif
  92
  93 #include <libcfs/libcfs.h>
  94 #include <lustre_fsfilt.h>
  95 #include <obd.h>
  96 #include <lustre_quota.h>
  97 #include <linux/lustre_compat25.h>
  98 #include <linux/lprocfs_status.h>
  99
 100 #ifdef HAVE_EXT4_LDISKFS
 101 #include <ext4/ext4_extents.h>
 102 #else
 103 #include <linux/ext3_extents.h>
 104 #endif
 105
 106 #include "lustre_quota_fmt.h"
 107
 108 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
 109 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 110 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 111 #else
 112 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 113 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 114 #endif
 115
 116 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 117 /* for kernels 2.6.18 and later */
 118 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 119 #else
 120 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 121 #endif
 122
 123 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 124 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 125                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 126 #else
 127 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 128                ext3_ext_insert_extent(handle, inode, path, newext)
 129 #endif
 130
 131 #ifdef EXT3_DISCARD_PREALLOCATIONS
 132 #define ext3_mb_discard_inode_preallocations(inode) \
 133                  ext3_discard_preallocations(inode)
 134 #endif
 135
 136
 137 static cfs_mem_cache_t *fcb_cache;
 138
 139 struct fsfilt_cb_data {
 140         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 141         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 142         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 143         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 144         void *cb_data;                  /* MDS/OST completion function data */
 145 };
 146
 147 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 148 #define EXT3_XATTR_INDEX_TRUSTED        4
 149 #endif
 150
 151 #ifdef HAVE_EXT4_LDISKFS
 152 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 153 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 154 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 155 #else
 156 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 157 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 158 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 159 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 160 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 161 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 162 #endif
 163
 164 #ifndef ext3_find_next_bit
 165 #define ext3_find_next_bit           ext2_find_next_bit
 166 #endif
 167
 168 #ifndef ext2_find_next_bit
 169 #ifdef __LITTLE_ENDIAN
 170 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 171 #else
 172 error "Need implementation of find_next_bit on big-endian systems"
 173 #endif  /* __LITTLE_ENDIAN */
 174 #endif  /* !ext2_find_next_le_bit */
 175
 176 static char *fsfilt_ext3_get_label(struct super_block *sb)
 177 {
 178         return EXT3_SB(sb)->s_es->s_volume_name;
 179 }
 180
 181 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 182 {
 183         /* see e.g. fsfilt_ext3_write_record() */
 184         journal_t *journal;
 185         handle_t *handle;
 186         int err;
 187
 188         journal = EXT3_SB(sb)->s_journal;
 189         handle = ext3_journal_start_sb(sb, 1);
 190         if (IS_ERR(handle)) {
 191                 CERROR("can't start transaction\n");
 192                 return(PTR_ERR(handle));
 193         }
 194
 195         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 196         if (err)
 197                 goto out;
 198
 199         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 200                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 201
 202         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 203
 204 out:
 205         ext3_journal_stop(handle);
 206
 207         return(err);
 208 }
 209
 210 static char *fsfilt_ext3_uuid(struct super_block *sb)
 211 {
 212         return EXT3_SB(sb)->s_es->s_uuid;
 213 }
 214
 215 #ifdef HAVE_DISK_INODE_VERSION
 216
 217 static __u64 get_i_version(struct inode *inode)
 218 {
 219         return EXT3_I(inode)->i_fs_version;
 220 }
 221
 222 static void set_i_version(struct inode *inode, __u64 new_version)
 223 {
 224         (EXT3_I(inode))->i_fs_version = new_version;
 225 }
 226
 227 /*
 228  * Get the 64-bit version for an inode.
 229  */
 230 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 231 {
 232         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 233                get_i_version(inode), inode->i_ino);
 234         return get_i_version(inode);
 235 }
 236
 237 /*
 238  * Set the 64-bit version and return the old version.
 239  */
 240 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 241 {
 242         __u64 old_version = get_i_version(inode);
 243
 244         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 245                new_version, old_version, inode->i_ino);
 246         set_i_version(inode, new_version);
 247         /* version is set after all inode operations are finished, so we should
 248          * mark it dirty here */
 249         inode->i_sb->s_op->dirty_inode(inode);
 250         return old_version;
 251 }
 252
 253 #endif
 254
 255 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
 256 #ifdef HAVE_BLOCKS_FOR_TRUNCATE
 257 # include <ext4/truncate.h>
 258 #else
 259 static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
 260 {
 261         ext4_lblk_t needed;
 262
 263         needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
 264         if (needed < 2)
 265                 needed = 2;
 266         if (needed > EXT4_MAX_TRANS_DATA)
 267                 needed = EXT4_MAX_TRANS_DATA;
 268         return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
 269 }
 270 #endif
 271
 272 /*
 273  * We don't currently need any additional blocks for rmdir and
 274  * unlink transactions because we are storing the OST oa_id inside
 275  * the inode (which we will be changing anyways as part of this
 276  * transaction).
 277  */
 278 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 279                                int logs)
 280 {
 281         /* For updates to the last received file */
 282         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 283         journal_t *journal;
 284         void *handle;
 285
 286         if (current->journal_info) {
 287                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 288                        current->journal_info);
 289                 goto journal_start;
 290         }
 291
 292         switch(op) {
 293         case FSFILT_OP_RMDIR:
 294         case FSFILT_OP_UNLINK:
 295                 /* delete one file + create/update logs for each stripe */
 296                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 297                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 298                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 299                 break;
 300         case FSFILT_OP_RENAME:
 301                 /* modify additional directory */
 302                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 303                 /* no break */
 304         case FSFILT_OP_SYMLINK:
 305                 /* additional block + block bitmap + GDT for long symlink */
 306                 nblocks += 3;
 307                 /* no break */
 308         case FSFILT_OP_CREATE: {
 309 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 310                 static int warned;
 311                 if (!warned) {
 312                         if (!test_opt(inode->i_sb, EXTENTS)) {
 313                                 warned = 1;
 314                         } else if (((EXT3_I(inode)->i_flags &
 315                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 316                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 317                                 CWARN("extent-mapped directory found with "
 318                                       "ext3-based ldiskfs - contact "
 319                                       "http://bugs.whamcloud.com/\n");
 320                                 warned = 1;
 321                         }
 322                 }
 323 #endif
 324                 /* no break */
 325         }
 326         case FSFILT_OP_MKDIR:
 327         case FSFILT_OP_MKNOD:
 328                 /* modify one inode + block bitmap + GDT */
 329                 nblocks += 3;
 330                 /* no break */
 331         case FSFILT_OP_LINK:
 332                 /* modify parent directory */
 333                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 334                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 335                 /* create/update logs for each stripe */
 336                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 337                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 338                 break;
 339         case FSFILT_OP_SETATTR:
 340                 /* Setattr on inode */
 341                 nblocks += 1;
 342                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 343                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 344                 /* quota chown log for each stripe */
 345                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 346                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 347                 break;
 348         case FSFILT_OP_CANCEL_UNLINK:
 349                 LASSERT(logs == 1);
 350
 351                 /* blocks for log header bitmap update OR
 352                  * blocks for catalog header bitmap update + unlink of logs +
 353                  * blocks for delete the inode (include blocks truncating). */
 354                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 355                           EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
 356                           ext4_blocks_for_truncate(inode) + 3;
 357                 break;
 358         default: CERROR("unknown transaction start op %d\n", op);
 359                 LBUG();
 360         }
 361
 362         LASSERT(current->journal_info == desc_private);
 363         journal = EXT3_SB(inode->i_sb)->s_journal;
 364         if (nblocks > journal->j_max_transaction_buffers) {
 365                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 366                        nblocks, op, logs, journal->j_max_transaction_buffers);
 367                 nblocks = journal->j_max_transaction_buffers;
 368         }
 369
 370  journal_start:
 371         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 372         handle = ext3_journal_start(inode, nblocks);
 373
 374         if (!IS_ERR(handle))
 375                 LASSERT(current->journal_info == handle);
 376         else
 377                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 378                        op, nblocks, PTR_ERR(handle));
 379         return handle;
 380 }
 381
 382 /*
 383  * Calculate the number of buffer credits needed to write multiple pages in
 384  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 385  * doesn't have a nice API for calculating this sort of thing in advance.
 386  *
 387  * See comment above ext3_writepage_trans_blocks for details.  We assume
 388  * no data journaling is being done, but it does allow for all of the pages
 389  * being non-contiguous.  If we are guaranteed contiguous pages we could
 390  * reduce the number of (d)indirect blocks a lot.
 391  *
 392  * With N blocks per page and P pages, for each inode we have at most:
 393  * N*P indirect
 394  * min(N*P, blocksize/4 + 1) dindirect blocks
 395  * niocount tindirect
 396  *
 397  * For the entire filesystem, we have at most:
 398  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 399  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 400  * objcount inode blocks
 401  * 1 superblock
 402  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 403  *
 404  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 405  */
 406 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 407                                       int niocount, struct niobuf_local *nb)
 408 {
 409         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 410         __u64 next_indir;
 411         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 412         int nbitmaps = 0, ngdblocks;
 413         int needed = objcount + 1; /* inodes + superblock */
 414         int i, j;
 415
 416         for (i = 0, j = 0; i < objcount; i++, fso++) {
 417                 /* two or more dindirect blocks in case we cross boundary */
 418                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 419                                     nb[j].offset) >>
 420                                    sb->s_blocksize_bits) /
 421                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 422                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 423
 424                 /* leaf, indirect, tindirect blocks for first block */
 425                 nbitmaps += blockpp + 2;
 426
 427                 j += fso->fso_bufcnt;
 428         }
 429
 430         next_indir = nb[0].offset +
 431                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 432         for (i = 1; i < niocount; i++) {
 433                 if (nb[i].offset >= next_indir) {
 434                         nbitmaps++;     /* additional indirect */
 435                         next_indir = nb[i].offset +
 436                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 437                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 438                         nbitmaps++;     /* additional indirect */
 439                 }
 440                 nbitmaps += blockpp;    /* each leaf in different group? */
 441         }
 442
 443         ngdblocks = nbitmaps;
 444         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 445                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 446         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 447                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 448
 449         needed += nbitmaps + ngdblocks;
 450
 451         /* last_rcvd update */
 452         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 453
 454 #if defined(CONFIG_QUOTA)
 455         /* We assume that there will be 1 bit set in s_dquot.flags for each
 456          * quota file that is active.  This is at least true for now.
 457          */
 458         needed += hweight32(ll_sb_any_quota_active(sb)) *
 459                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 460 #endif
 461
 462         return needed;
 463 }
 464
 465 /* We have to start a huge journal transaction here to hold all of the
 466  * metadata for the pages being written here.  This is necessitated by
 467  * the fact that we do lots of prepare_write operations before we do
 468  * any of the matching commit_write operations, so even if we split
 469  * up to use "smaller" transactions none of them could complete until
 470  * all of them were opened.  By having a single journal transaction,
 471  * we eliminate duplicate reservations for common blocks like the
 472  * superblock and group descriptors or bitmaps.
 473  *
 474  * We will start the transaction here, but each prepare_write will
 475  * add a refcount to the transaction, and each commit_write will
 476  * remove a refcount.  The transaction will be closed when all of
 477  * the pages have been written.
 478  */
 479 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 480                                    int niocount, struct niobuf_local *nb,
 481                                    void *desc_private, int logs)
 482 {
 483         journal_t *journal;
 484         handle_t *handle;
 485         int needed;
 486         ENTRY;
 487
 488         LASSERT(current->journal_info == desc_private);
 489         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 490         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 491
 492         /* The number of blocks we could _possibly_ dirty can very large.
 493          * We reduce our request if it is absurd (and we couldn't get that
 494          * many credits for a single handle anyways).
 495          *
 496          * At some point we have to limit the size of I/Os sent at one time,
 497          * increase the size of the journal, or we have to calculate the
 498          * actual journal requirements more carefully by checking all of
 499          * the blocks instead of being maximally pessimistic.  It remains to
 500          * be seen if this is a real problem or not.
 501          */
 502         if (needed > journal->j_max_transaction_buffers) {
 503                 CERROR("want too many journal credits (%d) using %d instead\n",
 504                        needed, journal->j_max_transaction_buffers);
 505                 needed = journal->j_max_transaction_buffers;
 506         }
 507
 508         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 509         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 510         if (IS_ERR(handle)) {
 511                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 512                        PTR_ERR(handle));
 513         } else {
 514                 LASSERT(handle->h_buffer_credits >= needed);
 515                 LASSERT(current->journal_info == handle);
 516         }
 517
 518         RETURN(handle);
 519 }
 520
 521 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 522 {
 523        handle_t *handle = h;
 524
 525        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 526        if (nblocks == 0) {
 527                handle->h_buffer_credits = 0;
 528                CWARN("setting credits of handle %p to zero by request\n", h);
 529        }
 530
 531        if (handle->h_buffer_credits > nblocks)
 532                 return 0;
 533        if (ext3_journal_extend(handle, nblocks) == 0)
 534                 return 0;
 535
 536        ext3_mark_inode_dirty(handle, inode);
 537        return ext3_journal_restart(handle, nblocks);
 538 }
 539
 540 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 541 {
 542         int rc;
 543         handle_t *handle = h;
 544
 545         LASSERT(current->journal_info == handle);
 546         if (force_sync)
 547                 handle->h_sync = 1; /* recovery likes this */
 548
 549         rc = ext3_journal_stop(handle);
 550
 551         return rc;
 552 }
 553
 554 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 555                                     void **wait_handle)
 556 {
 557         unsigned long tid;
 558         transaction_t *transaction;
 559         handle_t *handle = h;
 560         journal_t *journal;
 561         int rc;
 562
 563         LASSERT(current->journal_info == handle);
 564
 565         transaction = handle->h_transaction;
 566         journal = transaction->t_journal;
 567         tid = transaction->t_tid;
 568         /* we don't want to be blocked */
 569         handle->h_sync = 0;
 570         rc = ext3_journal_stop(handle);
 571         if (rc) {
 572                 CERROR("error while stopping transaction: %d\n", rc);
 573                 return rc;
 574         }
 575         fsfilt_log_start_commit(journal, tid);
 576
 577         *wait_handle = (void *) tid;
 578         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 579         return 0;
 580 }
 581
 582 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 583 {
 584         journal_t *journal = EXT3_JOURNAL(inode);
 585         tid_t tid = (tid_t)(long)h;
 586
 587         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 588         if (unlikely(is_journal_aborted(journal)))
 589                 return -EIO;
 590
 591         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 592
 593         if (unlikely(is_journal_aborted(journal)))
 594                 return -EIO;
 595         return 0;
 596 }
 597
 598 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 599                                struct iattr *iattr, int do_trunc)
 600 {
 601         struct inode *inode = dentry->d_inode;
 602         int rc = 0;
 603
 604         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 605          * We are already writing the inode to disk as part of this
 606          * transaction and want to avoid a lot of extra inode writeout
 607          * later on. b=9828 */
 608         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 609                 /* ATTR_SIZE would invoke truncate: clear it */
 610                 iattr->ia_valid &= ~ATTR_SIZE;
 611                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 612                 i_size_write(inode, iattr->ia_size);
 613
 614                 if (iattr->ia_valid & ATTR_UID)
 615                         inode->i_uid = iattr->ia_uid;
 616                 if (iattr->ia_valid & ATTR_GID)
 617                         inode->i_gid = iattr->ia_gid;
 618                 if (iattr->ia_valid & ATTR_ATIME)
 619                         inode->i_atime = iattr->ia_atime;
 620                 if (iattr->ia_valid & ATTR_MTIME)
 621                         inode->i_mtime = iattr->ia_mtime;
 622                 if (iattr->ia_valid & ATTR_CTIME)
 623                         inode->i_ctime = iattr->ia_ctime;
 624                 if (iattr->ia_valid & ATTR_MODE) {
 625                         inode->i_mode = iattr->ia_mode;
 626
 627                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 628                             !cfs_capable(CFS_CAP_FSETID))
 629                                 inode->i_mode &= ~S_ISGID;
 630                 }
 631
 632                 inode->i_sb->s_op->dirty_inode(inode);
 633
 634                 goto out;
 635         }
 636
 637         /* Don't allow setattr to change file type */
 638         if (iattr->ia_valid & ATTR_MODE)
 639                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 640                                  (iattr->ia_mode & ~S_IFMT);
 641
 642         /* We set these flags on the client, but have already checked perms
 643          * so don't confuse inode_change_ok. */
 644         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 645
 646         if (inode->i_op->setattr) {
 647                 rc = inode->i_op->setattr(dentry, iattr);
 648         } else {
 649                 rc = inode_change_ok(inode, iattr);
 650                 if (!rc)
 651                         rc = inode_setattr(inode, iattr);
 652         }
 653
 654  out:
 655         RETURN(rc);
 656 }
 657
 658 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 659                                  unsigned int cmd, unsigned long arg)
 660 {
 661         int rc = 0;
 662         ENTRY;
 663
 664         /* FIXME: Can't do this because of nested transaction deadlock */
 665         if (cmd == EXT3_IOC_SETFLAGS) {
 666                 /* We can't enable data journaling on OST objects, because
 667                 * this forces the transaction to be closed in order to
 668                 * flush the journal, but the caller will already have a
 669                 * compound transaction open to update the last_rcvd file,
 670                 * and this thread would deadlock trying to set the flag. */
 671                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 672                         CERROR("can't set data journal flag on file\n");
 673                         RETURN(-EPERM);
 674                 }
 675                 /* Because the MDS does not see the EXTENTS_FL set on the
 676                  * OST objects, mask this flag into all set flags.  It is
 677                  * not legal to clear this flag in any case, so we are not
 678                  * changing the functionality by doing this.  b=22911 */
 679                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 680         }
 681
 682 #ifdef HAVE_EXT4_LDISKFS
 683         /* ext4_ioctl does not have a inode argument */
 684         if (inode->i_fop->unlocked_ioctl)
 685                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 686 #else
 687         if (inode->i_fop->ioctl)
 688                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 689 #endif
 690         else
 691                 RETURN(-ENOTTY);
 692
 693         RETURN(rc);
 694 }
 695
 696 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 697                               void *lmm, int lmm_size, const char *name)
 698 {
 699         int rc;
 700
 701         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 702
 703         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 704                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 705
 706
 707         if (rc && rc != -EROFS)
 708                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 709                        inode->i_ino, rc);
 710         return rc;
 711 }
 712
 713 /* Must be called with i_mutex held */
 714 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 715                               const char *name)
 716 {
 717         int rc;
 718
 719         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 720
 721         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 722                             name, lmm, lmm_size);
 723
 724         /* This gives us the MD size */
 725         if (lmm == NULL)
 726                 return (rc == -ENODATA) ? 0 : rc;
 727
 728         if (rc < 0) {
 729                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 730                        EXT3_XATTR_INDEX_TRUSTED, name,
 731                        inode->i_ino, rc);
 732                 memset(lmm, 0, lmm_size);
 733                 return (rc == -ENODATA) ? 0 : rc;
 734         }
 735
 736         return rc;
 737 }
 738
 739 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 740 {
 741         submit_bio(rw, bio);
 742         return 0;
 743 }
 744
 745 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 746                                     loff_t *off)
 747 {
 748         struct inode *inode = file->f_dentry->d_inode;
 749         int rc = 0;
 750
 751         if (S_ISREG(inode->i_mode))
 752                 rc = file->f_op->read(file, buf, count, off);
 753         else {
 754                 const int blkbits = inode->i_sb->s_blocksize_bits;
 755                 const int blksize = inode->i_sb->s_blocksize;
 756
 757                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 758                        (unsigned long)count, inode->i_ino, *off);
 759                 while (count > 0) {
 760                         struct buffer_head *bh;
 761
 762                         bh = NULL;
 763                         if (*off < i_size_read(inode)) {
 764                                 int err = 0;
 765
 766                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 767                                                 0, &err);
 768
 769                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 770
 771                                 if (bh) {
 772                                         memcpy(buf, bh->b_data, blksize);
 773                                         brelse(bh);
 774                                 } else if (err) {
 775                                         /* XXX in theory we should just fake
 776                                          * this buffer and continue like ext3,
 777                                          * especially if this is a partial read
 778                                          */
 779                                         CERROR("error read dir %lu+%llu: %d\n",
 780                                                inode->i_ino, *off, err);
 781                                         RETURN(err);
 782                                 }
 783                         }
 784                         if (!bh) {
 785                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 786
 787                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 788                                 memset(fake, 0, sizeof(*fake));
 789                                 fake->rec_len = cpu_to_le16(blksize);
 790                         }
 791                         count -= blksize;
 792                         buf += blksize;
 793                         *off += blksize;
 794                         rc += blksize;
 795                 }
 796         }
 797
 798         return rc;
 799 }
 800
 801 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 802 {
 803         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 804
 805         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 806
 807         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 808 }
 809
 810 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 811                                       void *handle, fsfilt_cb_t cb_func,
 812                                       void *cb_data)
 813 {
 814         struct fsfilt_cb_data *fcb;
 815
 816         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 817         if (fcb == NULL)
 818                 RETURN(-ENOMEM);
 819
 820         fcb->cb_func = cb_func;
 821         fcb->cb_obd = obd;
 822         fcb->cb_last_rcvd = last_rcvd;
 823         fcb->cb_data = cb_data;
 824
 825         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 826         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 827                                     (struct journal_callback *)fcb);
 828
 829         return 0;
 830 }
 831
 832 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 833 {
 834         struct kstatfs sfs;
 835         int rc;
 836
 837         memset(&sfs, 0, sizeof(sfs));
 838         rc = ll_do_statfs(sb, &sfs);
 839         statfs_pack(osfs, &sfs);
 840         return rc;
 841 }
 842
 843 static int fsfilt_ext3_sync(struct super_block *sb)
 844 {
 845         return ext3_force_commit(sb);
 846 }
 847
 848 #ifndef EXT3_EXTENTS_FL
 849 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 850 #endif
 851
 852 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 853 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 854 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 855 #else
 856 # ifdef HAVE_EXT4_LDISKFS
 857 #  ifdef WALK_SPACE_HAS_DATA_SEM /* We only use it in fsfilt_map_nblocks() for now */
 858 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 859 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 860 #  else
 861 #   define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem))
 862 #   define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem))
 863 #  endif
 864 # else
 865 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 866 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 867 # endif
 868 #endif
 869
 870 #ifndef EXT_ASSERT
 871 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 872 #endif
 873
 874 #ifdef EXT3_EXT_HAS_NO_TREE
 875 /* for kernels 2.6.18 and later */
 876 #ifdef HAVE_EXT4_LDISKFS
 877 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 878 #else
 879 #define EXT_GENERATION(inode)           ext_generation(inode)
 880 #endif
 881 #define ext3_ext_base                   inode
 882 #define ext3_ext_base2inode(inode)      (inode)
 883 #define EXT_DEPTH(inode)                ext_depth(inode)
 884 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 885                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 886 #else
 887 #define ext3_ext_base                   ext3_extents_tree
 888 #define ext3_ext_base2inode(tree)       (tree->inode)
 889 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 890                         ext3_ext_walk_space(tree, block, num, cb);
 891 #endif
 892
 893 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 894 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 895                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 896 #else
 897 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 898                ext3_ext_insert_extent(handle, inode, path, newext)
 899 #endif
 900
 901 #include <linux/lustre_version.h>
 902
 903 struct bpointers {
 904         unsigned long *blocks;
 905         int *created;
 906         unsigned long start;
 907         int num;
 908         int init_num;
 909         int create;
 910 };
 911
 912 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 913                                unsigned long block, int *aflags)
 914 {
 915         struct ext3_inode_info *ei = EXT3_I(inode);
 916         unsigned long bg_start;
 917         unsigned long colour;
 918         int depth;
 919
 920         if (path) {
 921                 struct ext3_extent *ex;
 922                 depth = path->p_depth;
 923
 924                 /* try to predict block placement */
 925                 if ((ex = path[depth].p_ext))
 926                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 927
 928                 /* it looks index is empty
 929                  * try to find starting from index itself */
 930                 if (path[depth].p_bh)
 931                         return path[depth].p_bh->b_blocknr;
 932         }
 933
 934         /* OK. use inode's group */
 935         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 936                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 937         colour = (current->pid % 16) *
 938                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 939         return bg_start + colour + block;
 940 }
 941
 942 #define ll_unmap_underlying_metadata(sb, blocknr) \
 943         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 944
 945 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 946 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 947                                 struct ext3_ext_path *path, unsigned long block,
 948                                 unsigned long *count, int *err)
 949 {
 950         unsigned long pblock, goal;
 951         int aflags = 0;
 952         struct inode *inode = ext3_ext_base2inode(base);
 953
 954         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 955         aflags |= 2; /* block have been already reserved */
 956         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 957         return pblock;
 958
 959 }
 960 #else
 961 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 962                                 struct ext3_ext_path *path, unsigned long block,
 963                                 unsigned long *count, int *err)
 964 {
 965         struct inode *inode = ext3_ext_base2inode(base);
 966         struct ext3_allocation_request ar;
 967         unsigned long pblock;
 968         int aflags;
 969
 970         /* find neighbour allocated blocks */
 971         ar.lleft = block;
 972         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 973         if (*err)
 974                 return 0;
 975         ar.lright = block;
 976         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 977         if (*err)
 978                 return 0;
 979
 980         /* allocate new block */
 981         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 982         ar.inode = inode;
 983         ar.logical = block;
 984         ar.len = *count;
 985         ar.flags = EXT3_MB_HINT_DATA;
 986         pblock = ext3_mb_new_blocks(handle, &ar, err);
 987         *count = ar.len;
 988         return pblock;
 989 }
 990 #endif
 991
 992 #ifdef EXT3_EXT_HAS_NO_TREE
 993 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 994                                   struct ext3_ext_path *path,
 995                                   struct ext3_ext_cache *cex,
 996 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 997                                    struct ext3_extent *ex,
 998 #endif
 999                                   void *cbdata)
1000 {
1001         struct bpointers *bp = cbdata;
1002 #else
1003 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1004                                   struct ext3_ext_path *path,
1005                                   struct ext3_ext_cache *cex
1006 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1007                                   , struct ext3_extent *ex
1008 #endif
1009                                  )
1010 {
1011         struct bpointers *bp = base->private;
1012 #endif
1013         struct inode *inode = ext3_ext_base2inode(base);
1014         struct ext3_extent nex;
1015         unsigned long pblock;
1016         unsigned long tgen;
1017         int err, i;
1018         unsigned long count;
1019         handle_t *handle;
1020
1021         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1022                 err = EXT_CONTINUE;
1023                 goto map;
1024         }
1025
1026         if (bp->create == 0) {
1027                 i = 0;
1028                 if (cex->ec_block < bp->start)
1029                         i = bp->start - cex->ec_block;
1030                 if (i >= cex->ec_len)
1031                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1032                                         i, cex->ec_len);
1033                 for (; i < cex->ec_len && bp->num; i++) {
1034                         *(bp->created) = 0;
1035                         bp->created++;
1036                         *(bp->blocks) = 0;
1037                         bp->blocks++;
1038                         bp->num--;
1039                         bp->start++;
1040                 }
1041
1042                 return EXT_CONTINUE;
1043         }
1044
1045         tgen = EXT_GENERATION(base);
1046         count = ext3_ext_calc_credits_for_insert(base, path);
1047         fsfilt_up_truncate_sem(inode);
1048
1049         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1050         if (IS_ERR(handle)) {
1051                 fsfilt_down_truncate_sem(inode);
1052                 return PTR_ERR(handle);
1053         }
1054
1055         fsfilt_down_truncate_sem(inode);
1056         if (tgen != EXT_GENERATION(base)) {
1057                 /* the tree has changed. so path can be invalid at moment */
1058                 ext3_journal_stop(handle);
1059                 return EXT_REPEAT;
1060         }
1061
1062 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1063         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1064          * protected by i_data_sem as whole. so we patch it to store
1065          * generation to path and now verify the tree hasn't changed */
1066         down_write((&EXT4_I(inode)->i_data_sem));
1067
1068         /* validate extent, make sure the extent tree does not changed */
1069         if (EXT_GENERATION(base) != path[0].p_generation) {
1070                 /* cex is invalid, try again */
1071                 up_write(&EXT4_I(inode)->i_data_sem);
1072                 ext3_journal_stop(handle);
1073                 return EXT_REPEAT;
1074         }
1075 #endif
1076
1077         count = cex->ec_len;
1078         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1079         if (!pblock)
1080                 goto out;
1081         EXT_ASSERT(count <= cex->ec_len);
1082
1083         /* insert new extent */
1084         nex.ee_block = cpu_to_le32(cex->ec_block);
1085         ext3_ext_store_pblock(&nex, pblock);
1086         nex.ee_len = cpu_to_le16(count);
1087         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1088         if (err) {
1089                 /* free data blocks we just allocated */
1090                 /* not a good idea to call discard here directly,
1091                  * but otherwise we'd need to call it every free() */
1092 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1093                 ext3_mb_discard_inode_preallocations(inode);
1094 #endif
1095                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1096                                  cpu_to_le16(nex.ee_len), 0);
1097                 goto out;
1098         }
1099
1100         /*
1101          * Putting len of the actual extent we just inserted,
1102          * we are asking ext3_ext_walk_space() to continue
1103          * scaning after that block
1104          */
1105         cex->ec_len = le16_to_cpu(nex.ee_len);
1106         cex->ec_start = ext_pblock(&nex);
1107         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1108         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1109
1110 out:
1111 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1112         up_write((&EXT4_I(inode)->i_data_sem));
1113 #endif
1114         ext3_journal_stop(handle);
1115 map:
1116         if (err >= 0) {
1117                 /* map blocks */
1118                 if (bp->num == 0) {
1119                         CERROR("hmm. why do we find this extent?\n");
1120                         CERROR("initial space: %lu:%u\n",
1121                                 bp->start, bp->init_num);
1122                         CERROR("current extent: %u/%u/%llu %d\n",
1123                                 cex->ec_block, cex->ec_len,
1124                                 (unsigned long long)cex->ec_start,
1125                                 cex->ec_type);
1126                 }
1127                 i = 0;
1128                 if (cex->ec_block < bp->start)
1129                         i = bp->start - cex->ec_block;
1130                 if (i >= cex->ec_len)
1131                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1132                                         i, cex->ec_len);
1133                 for (; i < cex->ec_len && bp->num; i++) {
1134                         *(bp->blocks) = cex->ec_start + i;
1135                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1136                                 *(bp->created) = 0;
1137                         } else {
1138                                 *(bp->created) = 1;
1139                                 /* unmap any possible underlying metadata from
1140                                  * the block device mapping.  bug 6998. */
1141                                 ll_unmap_underlying_metadata(inode->i_sb,
1142                                                              *(bp->blocks));
1143                         }
1144                         bp->created++;
1145                         bp->blocks++;
1146                         bp->num--;
1147                         bp->start++;
1148                 }
1149         }
1150         return err;
1151 }
1152
1153 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1154                        unsigned long num, unsigned long *blocks,
1155                        int *created, int create)
1156 {
1157 #ifdef EXT3_EXT_HAS_NO_TREE
1158         struct ext3_ext_base *base = inode;
1159 #else
1160         struct ext3_extents_tree tree;
1161         struct ext3_ext_base *base = &tree;
1162 #endif
1163         struct bpointers bp;
1164         int err;
1165
1166         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1167                block, block + num - 1, (unsigned) inode->i_ino);
1168
1169 #ifndef EXT3_EXT_HAS_NO_TREE
1170         ext3_init_tree_desc(base, inode);
1171         tree.private = &bp;
1172 #endif
1173         bp.blocks = blocks;
1174         bp.created = created;
1175         bp.start = block;
1176         bp.init_num = bp.num = num;
1177         bp.create = create;
1178
1179         fsfilt_down_truncate_sem(inode);
1180         err = fsfilt_ext3_ext_walk_space(base, block, num,
1181                                          ext3_ext_new_extent_cb, &bp);
1182         ext3_ext_invalidate_cache(base);
1183         fsfilt_up_truncate_sem(inode);
1184
1185         return err;
1186 }
1187
1188 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1189                                     int pages, unsigned long *blocks,
1190                                     int *created, int create)
1191 {
1192         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1193         int rc = 0, i = 0;
1194         struct page *fp = NULL;
1195         int clen = 0;
1196
1197         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1198                 inode->i_ino, pages, (*page)->index);
1199
1200         /* pages are sorted already. so, we just have to find
1201          * contig. space and process them properly */
1202         while (i < pages) {
1203                 if (fp == NULL) {
1204                         /* start new extent */
1205                         fp = *page++;
1206                         clen = 1;
1207                         i++;
1208                         continue;
1209                 } else if (fp->index + clen == (*page)->index) {
1210                         /* continue the extent */
1211                         page++;
1212                         clen++;
1213                         i++;
1214                         continue;
1215                 }
1216
1217                 /* process found extent */
1218                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1219                                         clen * blocks_per_page, blocks,
1220                                         created, create);
1221                 if (rc)
1222                         GOTO(cleanup, rc);
1223
1224                 /* look for next extent */
1225                 fp = NULL;
1226                 blocks += blocks_per_page * clen;
1227                 created += blocks_per_page * clen;
1228         }
1229
1230         if (fp)
1231                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1232                                         clen * blocks_per_page, blocks,
1233                                         created, create);
1234 cleanup:
1235         return rc;
1236 }
1237
1238 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1239                                unsigned long *blocks, int *created, int create);
1240 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1241                                    int pages, unsigned long *blocks,
1242                                    int *created, int create)
1243 {
1244         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1245         unsigned long *b;
1246         int rc = 0, i, *cr;
1247
1248         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1249                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1250                 if (rc) {
1251                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1252                                inode->i_ino, *b, *cr, create, rc);
1253                         break;
1254                 }
1255
1256                 b += blocks_per_page;
1257                 cr += blocks_per_page;
1258         }
1259         return rc;
1260 }
1261
1262 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1263                                 int pages, unsigned long *blocks,
1264                                 int *created, int create,
1265                                 cfs_semaphore_t *optional_sem)
1266 {
1267         int rc;
1268
1269         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1270                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1271                                                      blocks, created, create);
1272                 return rc;
1273         }
1274         if (optional_sem != NULL)
1275                 cfs_down(optional_sem);
1276         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1277                                             created, create);
1278         if (optional_sem != NULL)
1279                 cfs_up(optional_sem);
1280
1281         return rc;
1282 }
1283
1284 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1285 {
1286         unsigned long block;
1287         struct buffer_head *bh;
1288         int err, blocksize, csize, boffs, osize = size;
1289
1290         /* prevent reading after eof */
1291         cfs_lock_kernel();
1292         if (i_size_read(inode) < *offs + size) {
1293                 size = i_size_read(inode) - *offs;
1294                 cfs_unlock_kernel();
1295                 if (size < 0) {
1296                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1297                                i_size_read(inode), *offs);
1298                         return -EBADR;
1299                 } else if (size == 0) {
1300                         return 0;
1301                 }
1302         } else {
1303                 cfs_unlock_kernel();
1304         }
1305
1306         blocksize = 1 << inode->i_blkbits;
1307
1308         while (size > 0) {
1309                 block = *offs >> inode->i_blkbits;
1310                 boffs = *offs & (blocksize - 1);
1311                 csize = min(blocksize - boffs, size);
1312                 bh = ext3_bread(NULL, inode, block, 0, &err);
1313                 if (!bh) {
1314                         CERROR("can't read block: %d\n", err);
1315                         return err;
1316                 }
1317
1318                 memcpy(buf, bh->b_data + boffs, csize);
1319                 brelse(bh);
1320
1321                 *offs += csize;
1322                 buf += csize;
1323                 size -= csize;
1324         }
1325         return osize;
1326 }
1327 EXPORT_SYMBOL(fsfilt_ext3_read);
1328
1329 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1330                                    int size, loff_t *offs)
1331 {
1332         int rc;
1333         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1334         if (rc > 0)
1335                 rc = 0;
1336         return rc;
1337 }
1338
1339 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1340                                 loff_t *offs, handle_t *handle)
1341 {
1342         struct buffer_head *bh = NULL;
1343         loff_t old_size = i_size_read(inode), offset = *offs;
1344         loff_t new_size = i_size_read(inode);
1345         unsigned long block;
1346         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1347
1348         while (bufsize > 0) {
1349                 if (bh != NULL)
1350                         brelse(bh);
1351
1352                 block = offset >> inode->i_blkbits;
1353                 boffs = offset & (blocksize - 1);
1354                 size = min(blocksize - boffs, bufsize);
1355                 bh = ext3_bread(handle, inode, block, 1, &err);
1356                 if (!bh) {
1357                         CERROR("can't read/create block: %d\n", err);
1358                         break;
1359                 }
1360
1361                 err = ext3_journal_get_write_access(handle, bh);
1362                 if (err) {
1363                         CERROR("journal_get_write_access() returned error %d\n",
1364                                err);
1365                         break;
1366                 }
1367                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1368                 memcpy(bh->b_data + boffs, buf, size);
1369                 err = ext3_journal_dirty_metadata(handle, bh);
1370                 if (err) {
1371                         CERROR("journal_dirty_metadata() returned error %d\n",
1372                                err);
1373                         break;
1374                 }
1375                 if (offset + size > new_size)
1376                         new_size = offset + size;
1377                 offset += size;
1378                 bufsize -= size;
1379                 buf += size;
1380         }
1381         if (bh)
1382                 brelse(bh);
1383
1384         /* correct in-core and on-disk sizes */
1385         if (new_size > i_size_read(inode)) {
1386                 cfs_lock_kernel();
1387                 if (new_size > i_size_read(inode))
1388                         i_size_write(inode, new_size);
1389                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1390                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1391                 if (i_size_read(inode) > old_size)
1392                         mark_inode_dirty(inode);
1393                 cfs_unlock_kernel();
1394         }
1395
1396         if (err == 0)
1397                 *offs = offset;
1398         return err;
1399 }
1400 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1401
1402 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1403                                     loff_t *offs, int force_sync)
1404 {
1405         struct inode *inode = file->f_dentry->d_inode;
1406         handle_t *handle;
1407         int err, block_count = 0, blocksize;
1408
1409         /* Determine how many transaction credits are needed */
1410         blocksize = 1 << inode->i_blkbits;
1411         block_count = (*offs & (blocksize - 1)) + bufsize;
1412         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1413
1414         handle = ext3_journal_start(inode,
1415                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1416         if (IS_ERR(handle)) {
1417                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1418                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1419                 return PTR_ERR(handle);
1420         }
1421
1422         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1423
1424         if (!err && force_sync)
1425                 handle->h_sync = 1; /* recovery likes this */
1426
1427         ext3_journal_stop(handle);
1428
1429         return err;
1430 }
1431
1432 static int fsfilt_ext3_setup(struct super_block *sb)
1433 {
1434 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1435      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1436         struct ext3_sb_info *sbi = EXT3_SB(sb);
1437 #if 0
1438         sbi->dx_lock = fsfilt_ext3_dx_lock;
1439         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1440 #endif
1441 #endif
1442 #ifdef S_PDIROPS
1443         CWARN("Enabling PDIROPS\n");
1444         set_opt(sbi->s_mount_opt, PDIROPS);
1445         sb->s_flags |= S_PDIROPS;
1446 #endif
1447         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1448                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1449 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1450         /* enable journaled quota support */
1451         /* kfreed in ext3_put_super() */
1452         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1453         if (!sbi->s_qf_names[USRQUOTA])
1454                 return -ENOMEM;
1455         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1456         if (!sbi->s_qf_names[GRPQUOTA]) {
1457                 kfree(sbi->s_qf_names[USRQUOTA]);
1458                 sbi->s_qf_names[USRQUOTA] = NULL;
1459                 return -ENOMEM;
1460         }
1461         sbi->s_jquota_fmt = QFMT_LUSTRE;
1462 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1463         set_opt(sbi->s_mount_opt, QUOTA);
1464 #endif
1465 #endif
1466         return 0;
1467 }
1468
1469 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1470    objects. Logs is number of logfiles to update */
1471 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1472 {
1473         if ( !fso ) {
1474                 switch(op) {
1475                 case FSFILT_OP_CREATE:
1476                                  /* directory leaf, index & indirect & EA*/
1477                         return 4 + 3 * logs;
1478                 case FSFILT_OP_UNLINK:
1479                         return 3 * logs;
1480                 }
1481         } else {
1482                 int i;
1483                 int needed = 0;
1484                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1485                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1486                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1487                 for (i = 0; i < op; i++, fso++) {
1488                         int nblocks = fso->fso_bufcnt * blockpp;
1489                         int ndindirect = min(nblocks, addrpp + 1);
1490                         int nindir = nblocks + ndindirect + 1;
1491
1492                         needed += nindir;
1493                 }
1494                 return needed + 3 * logs;
1495         }
1496
1497         return 0;
1498 }
1499
1500 #ifdef HAVE_QUOTA_SUPPORT
1501 #define DQINFO_COPY(out, in)                    \
1502 do {                                            \
1503         Q_COPY(out, in, dqi_bgrace);            \
1504         Q_COPY(out, in, dqi_igrace);            \
1505         Q_COPY(out, in, dqi_flags);             \
1506         Q_COPY(out, in, dqi_valid);             \
1507 } while (0)
1508
1509 #define DQBLK_COPY(out, in)                     \
1510 do {                                            \
1511         Q_COPY(out, in, dqb_bhardlimit);        \
1512         Q_COPY(out, in, dqb_bsoftlimit);        \
1513         Q_COPY(out, in, dqb_curspace);          \
1514         Q_COPY(out, in, dqb_ihardlimit);        \
1515         Q_COPY(out, in, dqb_isoftlimit);        \
1516         Q_COPY(out, in, dqb_curinodes);         \
1517         Q_COPY(out, in, dqb_btime);             \
1518         Q_COPY(out, in, dqb_itime);             \
1519         Q_COPY(out, in, dqb_valid);             \
1520 } while (0)
1521
1522 static int fsfilt_ext3_quotactl(struct super_block *sb,
1523                                 struct obd_quotactl *oqc)
1524 {
1525         int i, rc = 0, error = 0;
1526         const struct quotactl_ops *qcop;
1527         struct if_dqinfo *info;
1528         struct if_dqblk *dqblk;
1529         ENTRY;
1530
1531         if (!sb->s_qcop)
1532                 RETURN(-ENOSYS);
1533
1534         OBD_ALLOC_PTR(info);
1535         if (!info)
1536                 RETURN(-ENOMEM);
1537         OBD_ALLOC_PTR(dqblk);
1538         if (!dqblk) {
1539                 OBD_FREE_PTR(info);
1540                 RETURN(-ENOMEM);
1541         }
1542
1543         DQINFO_COPY(info, &oqc->qc_dqinfo);
1544         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1545
1546         qcop = sb->s_qcop;
1547         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1548                 for (i = 0; i < MAXQUOTAS; i++) {
1549                         if (!Q_TYPESET(oqc, i))
1550                                 continue;
1551
1552                         if (oqc->qc_cmd == Q_QUOTAON) {
1553                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1554
1555                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1556
1557                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1558                                                  name[i], 0);
1559                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1560                                 rc = ll_quota_off(sb, i, 0);
1561                         }
1562
1563                         if (rc == -EBUSY)
1564                                 error = rc;
1565                         else if (rc)
1566                                 GOTO(out, rc);
1567                 }
1568                 GOTO(out, rc ?: error);
1569         }
1570
1571         switch (oqc->qc_cmd) {
1572         case Q_GETOINFO:
1573         case Q_GETINFO:
1574                 if (!qcop->get_info)
1575                         GOTO(out, rc = -ENOSYS);
1576                 rc = qcop->get_info(sb, oqc->qc_type, info);
1577                 break;
1578         case Q_SETQUOTA:
1579         case Q_INITQUOTA:
1580                 if (!qcop->set_dqblk)
1581                         GOTO(out, rc = -ENOSYS);
1582                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1583                 break;
1584         case Q_GETOQUOTA:
1585         case Q_GETQUOTA:
1586                 if (!qcop->get_dqblk)
1587                         GOTO(out, rc = -ENOSYS);
1588                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1589                 if (!rc)
1590                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1591                 break;
1592         case Q_SYNC:
1593                 if (!sb->s_qcop->quota_sync)
1594                         GOTO(out, rc = -ENOSYS);
1595                 qcop->quota_sync(sb, oqc->qc_type);
1596                 break;
1597         case Q_FINVALIDATE:
1598                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1599                 for (i = 0; i < MAXQUOTAS; i++) {
1600                         struct file *fp;
1601                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1602
1603                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1604
1605                         if (!Q_TYPESET(oqc, i))
1606                                 continue;
1607
1608                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1609                         if (IS_ERR(fp)) {
1610                                 rc = PTR_ERR(fp);
1611                                 CERROR("error invalidating operational quota file"
1612                                        " %s (rc:%d)\n", name[i], rc);
1613                         } else {
1614                                 filp_close(fp, 0);
1615                         }
1616
1617                 }
1618                 break;
1619         default:
1620                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1621                 LBUG();
1622         }
1623 out:
1624         DQINFO_COPY(&oqc->qc_dqinfo, info);
1625         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1626
1627         OBD_FREE_PTR(info);
1628         OBD_FREE_PTR(dqblk);
1629
1630         if (rc)
1631                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1632                                 "failed: %d\n",
1633                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1634         RETURN(rc);
1635 }
1636
1637 struct chk_dqblk{
1638         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1639         cfs_list_t              dqb_list;        /** in list also */
1640         qid_t                   dqb_id;          /** uid/gid */
1641         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1642         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1643         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1644         qsize_t                 dqb_curspace;    /** current space */
1645         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1646         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1647         qsize_t                 dqb_curinodes;   /** current inodes */
1648         obd_time                dqb_btime;       /** block grace time */
1649         obd_time                dqb_itime;       /** inode grace time */
1650         __u32                   dqb_valid;       /** flag for above fields */
1651 };
1652
1653 static inline unsigned int chkquot_hash(qid_t id, int type)
1654                                         __attribute__((__const__));
1655
1656 static inline unsigned int chkquot_hash(qid_t id, int type)
1657 {
1658         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1659 }
1660
1661 static inline struct chk_dqblk *
1662 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1663 {
1664         cfs_hlist_node_t *node;
1665         struct chk_dqblk *cdqb;
1666
1667         cfs_hlist_for_each(node, head) {
1668                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1669                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1670                         return cdqb;
1671         }
1672
1673         return NULL;
1674 }
1675
1676 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1677 {
1678         struct chk_dqblk *cdqb;
1679
1680         OBD_ALLOC_PTR(cdqb);
1681         if (cdqb) {
1682                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1683                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1684                 cdqb->dqb_id = id;
1685                 cdqb->dqb_type = type;
1686         }
1687
1688         return cdqb;
1689 }
1690
1691 static struct chk_dqblk *
1692 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1693       cfs_list_t *list, qid_t id, int type, int first_check)
1694 {
1695         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1696         struct if_dqblk dqb;
1697         struct chk_dqblk *cdqb;
1698         int rc;
1699
1700         cdqb = find_chkquot(head, id, type);
1701         if (cdqb)
1702                 return cdqb;
1703
1704         cdqb = alloc_chkquot(id, type);
1705         if (!cdqb)
1706                 return NULL;
1707
1708         if (!first_check) {
1709                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1710                 if (rc) {
1711                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1712                                id, type, rc);
1713                 } else {
1714                         DQBLK_COPY(cdqb, &dqb);
1715                         cdqb->dqb_curspace = 0;
1716                         cdqb->dqb_curinodes = 0;
1717                 }
1718         }
1719
1720         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1721         cfs_list_add_tail(&cdqb->dqb_list, list);
1722
1723         return cdqb;
1724 }
1725
1726 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1727 {
1728         struct obd_quotactl *oqctl;
1729         int rc;
1730
1731         OBD_ALLOC_PTR(oqctl);
1732         if (!oqctl)
1733                 RETURN(-ENOMEM);
1734
1735         oqctl->qc_cmd = cmd;
1736         oqctl->qc_id = qfmt;
1737         oqctl->qc_type = type;
1738         rc = fsfilt_ext3_quotactl(sb, oqctl);
1739
1740         OBD_FREE_PTR(oqctl);
1741         return rc;
1742 }
1743
1744 static inline int read_old_dqinfo(struct super_block *sb, int type,
1745                                   struct if_dqinfo *dqinfo)
1746 {
1747         struct obd_quotactl *oqctl;
1748         int rc;
1749         ENTRY;
1750
1751         OBD_ALLOC_PTR(oqctl);
1752         if (!oqctl)
1753                 RETURN(-ENOMEM);
1754
1755         oqctl->qc_cmd = Q_GETINFO;
1756         oqctl->qc_type = type;
1757         rc = fsfilt_ext3_quotactl(sb, oqctl);
1758         if (!rc)
1759                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1760
1761         OBD_FREE_PTR(oqctl);
1762         RETURN(rc);
1763 }
1764
1765 #ifndef HAVE_EXT4_LDISKFS
1766 static inline struct ext3_group_desc *
1767 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1768 {
1769         unsigned long desc_block, desc;
1770         struct ext3_group_desc *gdp;
1771
1772         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1773         desc = group % EXT3_DESC_PER_BLOCK(sb);
1774         gdp = (struct ext3_group_desc *)
1775               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1776
1777         return gdp + desc;
1778 }
1779
1780 static inline struct buffer_head *
1781 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1782 {
1783         struct ext3_group_desc *desc;
1784         struct buffer_head *bh;
1785
1786         desc = get_group_desc(sb, group, NULL);
1787         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1788         return bh;
1789 }
1790
1791 static __u32 ext3_itable_unused_count(struct super_block *sb,
1792                                struct ext3_group_desc *bg) {
1793        return le16_to_cpu(bg->bg_itable_unused);
1794 }
1795 #else
1796 #define get_group_desc ext3_get_group_desc
1797 #endif
1798
1799 struct qchk_ctxt {
1800         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1801         cfs_list_t              qckt_list;                 /* quotacheck list */
1802         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1803         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1804 };
1805
1806 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1807                            struct obd_quotactl *oqc)
1808 {
1809         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1810         loff_t size = 0;
1811         qid_t qid[MAXQUOTAS];
1812         int cnt, i, rc = 0;
1813
1814         if (!inode)
1815                 return 0;
1816
1817         qid[USRQUOTA] = inode->i_uid;
1818         qid[GRPQUOTA] = inode->i_gid;
1819
1820         if (S_ISDIR(inode->i_mode) ||
1821             S_ISREG(inode->i_mode) ||
1822             S_ISLNK(inode->i_mode))
1823                 size = inode_get_bytes(inode);
1824
1825         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1826                 if (!Q_TYPESET(oqc, cnt))
1827                         continue;
1828
1829                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1830                                 &qctxt->qckt_list, qid[cnt], cnt,
1831                                 qctxt->qckt_first_check[cnt]);
1832                 if (!cdqb[cnt]) {
1833                         rc = -ENOMEM;
1834                         break;
1835                 }
1836
1837                 cdqb[cnt]->dqb_curspace += size;
1838                 cdqb[cnt]->dqb_curinodes++;
1839         }
1840
1841         if (rc) {
1842                 for (i = 0; i < cnt; i++) {
1843                         if (!Q_TYPESET(oqc, i))
1844                                 continue;
1845                         LASSERT(cdqb[i]);
1846                         cdqb[i]->dqb_curspace -= size;
1847                         cdqb[i]->dqb_curinodes--;
1848                 }
1849         }
1850
1851         return rc;
1852 }
1853
1854 /* write dqinfo struct in a new quota file */
1855 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1856 {
1857         struct v2_disk_dqinfo dqinfo;
1858         __u32 blocks = V2_DQTREEOFF + 1;
1859         loff_t offset = V2_DQINFOOFF;
1860
1861         if (info) {
1862                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1863                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1864                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1865                                                ~DQF_INFO_DIRTY);
1866         } else {
1867                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1868                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1869                 dqinfo.dqi_flags = 0;
1870         }
1871
1872         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1873         dqinfo.dqi_free_blk = 0;
1874         dqinfo.dqi_free_entry = 0;
1875
1876         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1877 }
1878
1879 static int v3_write_dqheader(struct file *f, int type)
1880 {
1881         static const __u32 quota_magics[] = V2_INITQMAGICS;
1882         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1883         struct v2_disk_dqheader dqhead;
1884         loff_t offset = 0;
1885
1886         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1887         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1888
1889         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1890         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1891
1892         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1893 }
1894
1895 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1896                                   struct obd_quotactl *oqc)
1897 {
1898         int i, rc = 0;
1899         ENTRY;
1900
1901         for (i = 0; i < MAXQUOTAS; i++) {
1902                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1903                                          NULL : &qctxt->qckt_dqinfo[i];
1904                 struct file *file;
1905                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1906
1907                 if (!Q_TYPESET(oqc, i))
1908                         continue;
1909
1910                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1911
1912                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1913                 if (IS_ERR(file)) {
1914                         rc = PTR_ERR(file);
1915                         CERROR("can't create %s file: rc = %d\n",
1916                                name[i], rc);
1917                         GOTO(out, rc);
1918                 }
1919
1920                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1921                         CERROR("file %s is not regular", name[i]);
1922                         filp_close(file, 0);
1923                         GOTO(out, rc = -EINVAL);
1924                 }
1925
1926                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1927
1928                 rc = v3_write_dqheader(file, i);
1929                 if (rc) {
1930                         filp_close(file, 0);
1931                         GOTO(out, rc);
1932                 }
1933
1934                 rc = v3_write_dqinfo(file, i, info);
1935                 filp_close(file, 0);
1936                 if (rc)
1937                         GOTO(out, rc);
1938         }
1939
1940 out:
1941         RETURN(rc);
1942 }
1943
1944
1945 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1946                           struct chk_dqblk *cdqb)
1947 {
1948         struct obd_quotactl *oqc;
1949         long now;
1950         int rc;
1951         ENTRY;
1952
1953         OBD_ALLOC_PTR(oqc);
1954         if (!oqc)
1955                 RETURN(-ENOMEM);
1956
1957         now = cfs_time_current_sec();
1958
1959         if (cdqb->dqb_bsoftlimit &&
1960             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1961             !cdqb->dqb_btime)
1962                 cdqb->dqb_btime =
1963                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1964
1965         if (cdqb->dqb_isoftlimit &&
1966             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1967             !cdqb->dqb_itime)
1968                 cdqb->dqb_itime =
1969                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1970
1971         cdqb->dqb_valid = QIF_ALL;
1972
1973         oqc->qc_cmd = Q_SETQUOTA;
1974         oqc->qc_type = cdqb->dqb_type;
1975         oqc->qc_id = cdqb->dqb_id;
1976         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1977
1978         rc = fsfilt_ext3_quotactl(sb, oqc);
1979         OBD_FREE_PTR(oqc);
1980         RETURN(rc);
1981 }
1982
1983 static int prune_chkquots(struct super_block *sb,
1984                           struct qchk_ctxt *qctxt, int error)
1985 {
1986         struct chk_dqblk *cdqb, *tmp;
1987         int rc;
1988
1989         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1990                 if (!error) {
1991                         rc = commit_chkquot(sb, qctxt, cdqb);
1992                         if (rc)
1993                                 error = rc;
1994                 }
1995                 cfs_hlist_del_init(&cdqb->dqb_hash);
1996                 cfs_list_del(&cdqb->dqb_list);
1997                 OBD_FREE_PTR(cdqb);
1998         }
1999
2000         return error;
2001 }
2002
2003 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2004 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2005 #endif
2006
2007 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2008                                   struct obd_quotactl *oqc)
2009 {
2010         struct ext3_sb_info *sbi = EXT3_SB(sb);
2011         int i, group, uninit_feat = 0;
2012         struct qchk_ctxt *qctxt;
2013         struct buffer_head *bitmap_bh = NULL;
2014         unsigned long ino, inode_inuse;
2015         struct inode *inode;
2016         int rc = 0;
2017         ENTRY;
2018
2019         /* turn on quota and read dqinfo if existed */
2020         OBD_ALLOC_PTR(qctxt);
2021         if (!qctxt) {
2022                 oqc->qc_stat = -ENOMEM;
2023                 RETURN(-ENOMEM);
2024         }
2025
2026         for (i = 0; i < NR_DQHASH; i++)
2027                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2028         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2029
2030         for (i = 0; i < MAXQUOTAS; i++) {
2031                 if (!Q_TYPESET(oqc, i))
2032                         continue;
2033
2034                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2035                 if (!rc || rc == -EBUSY) {
2036                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2037                         if (rc)
2038                                 GOTO(out, rc);
2039                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2040                         qctxt->qckt_first_check[i] = 1;
2041                 } else if (rc) {
2042                         GOTO(out, rc);
2043                 }
2044         }
2045         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2046                 /* This filesystem supports the uninit group feature */
2047                 uninit_feat = 1;
2048
2049         /* number of inodes that have been allocated */
2050         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2051                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2052
2053         /* check quota and update in hash */
2054         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2055              group++) {
2056                 unsigned long used_count = sbi->s_inodes_per_group;
2057
2058                 if (uninit_feat) {
2059                         struct ext3_group_desc *desc;
2060                         desc = get_group_desc(sb, group, NULL);
2061                         if (!desc)
2062                                 GOTO(out, -EIO);
2063
2064                         /* we don't really need to take the group lock here,
2065                          * but it may be useful if one day we support online
2066                          * quotacheck */
2067 #ifdef HAVE_EXT4_LDISKFS
2068                         ext4_lock_group(sb, group);
2069 #else
2070                         spin_lock(sb_bgl_lock(sbi, group));
2071 #endif
2072                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2073                                 /* no inode in use in this group, just skip it */
2074 #ifdef HAVE_EXT4_LDISKFS
2075                                 ext3_unlock_group(sb, group);
2076 #else
2077                                 spin_unlock(sb_bgl_lock(sbi, group));
2078 #endif
2079                                 continue;
2080                         }
2081
2082                         used_count -= ext3_itable_unused_count(sb, desc);
2083 #ifdef HAVE_EXT4_LDISKFS
2084                         ext3_unlock_group(sb, group);
2085 #else
2086                         spin_unlock(sb_bgl_lock(sbi, group));
2087 #endif
2088                 }
2089
2090                 ino = group * sbi->s_inodes_per_group + 1;
2091                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2092                 if (!bitmap_bh) {
2093                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2094                                sb->s_id, group);
2095                         GOTO(out, -EIO);
2096                 }
2097
2098                 i = 0;
2099                 while (i < used_count &&
2100                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2101                                                used_count, i)) < used_count) {
2102                         inode_inuse--;
2103                         i++;
2104                         ino = i + group * sbi->s_inodes_per_group;
2105                         if (ino < sbi->s_first_ino)
2106                                 continue;
2107 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2108                         inode = ext3_iget(sb, ino);
2109 #else
2110                         inode = iget(sb, ino);
2111 #endif
2112                         if (!inode || IS_ERR(inode))
2113                                 continue;
2114
2115                         rc = add_inode_quota(inode, qctxt, oqc);
2116                         iput(inode);
2117                         if (rc) {
2118                                 brelse(bitmap_bh);
2119                                 GOTO(out, rc);
2120                         }
2121                 }
2122                 brelse(bitmap_bh);
2123         }
2124
2125         /* read old quota limits from old quota file. (only for the user
2126          * has limits but hasn't file) */
2127 #ifdef HAVE_QUOTA_SUPPORT
2128         for (i = 0; i < MAXQUOTAS; i++) {
2129                 cfs_list_t id_list;
2130                 struct dquot_id *dqid, *tmp;
2131
2132                 if (!Q_TYPESET(oqc, i))
2133                         continue;
2134
2135                 if (qctxt->qckt_first_check[i])
2136                         continue;
2137
2138
2139                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2140                 CFS_INIT_LIST_HEAD(&id_list);
2141 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2142                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2143 #else
2144                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2145 #endif
2146                 if (rc)
2147                         CERROR("read old limits failed. (rc:%d)\n", rc);
2148
2149                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2150                         cfs_list_del_init(&dqid->di_link);
2151
2152                         if (!rc)
2153                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2154                                       dqid->di_id, i,
2155                                       qctxt->qckt_first_check[i]);
2156                         OBD_FREE_PTR(dqid);
2157                 }
2158         }
2159 #endif
2160         /* turn off quota cause we are to dump chk_dqblk to files */
2161         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2162
2163         rc = create_new_quota_files(qctxt, oqc);
2164         if (rc)
2165                 GOTO(out, rc);
2166
2167         /* we use vfs functions to set dqblk, so turn quota on */
2168         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2169 out:
2170         /* dump and free chk_dqblk */
2171         rc = prune_chkquots(sb, qctxt, rc);
2172         OBD_FREE_PTR(qctxt);
2173
2174         /* turn off quota, `lfs quotacheck` will turn on when all
2175          * nodes quotacheck finish. */
2176         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2177
2178         oqc->qc_stat = rc;
2179         if (rc)
2180                 CERROR("quotacheck failed: rc = %d\n", rc);
2181
2182         RETURN(rc);
2183 }
2184
2185 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2186                                  int cmd)
2187 {
2188         int rc = 0;
2189         ENTRY;
2190
2191         if (lqi->qi_files[type] == NULL) {
2192                 CERROR("operate qinfo before it's enabled!\n");
2193                 RETURN(-ESRCH);
2194         }
2195
2196         switch (cmd) {
2197         case QFILE_CHK:
2198                 rc = lustre_check_quota_file(lqi, type);
2199                 break;
2200         case QFILE_RD_INFO:
2201                 rc = lustre_read_quota_info(lqi, type);
2202                 break;
2203         case QFILE_WR_INFO:
2204                 rc = lustre_write_quota_info(lqi, type);
2205                 break;
2206         case QFILE_INIT_INFO:
2207                 rc = lustre_init_quota_info(lqi, type);
2208                 break;
2209         case QFILE_CONVERT:
2210                 rc = -ENOTSUPP;
2211                 CERROR("quota CONVERT command is not supported\n");
2212                 break;
2213         default:
2214                 rc = -ENOTSUPP;
2215                 CERROR("Unsupported admin quota file cmd %d\n"
2216                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2217                        cmd);
2218                 break;
2219         }
2220         RETURN(rc);
2221 }
2222
2223 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2224                             cfs_list_t *list)
2225 {
2226         return lustre_get_qids(file, inode, type, list);
2227 }
2228
2229 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2230 {
2231         int rc = 0;
2232         ENTRY;
2233
2234         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2235                 CERROR("operate dquot before it's enabled!\n");
2236                 RETURN(-ESRCH);
2237         }
2238
2239         switch (cmd) {
2240         case QFILE_RD_DQUOT:
2241                 rc = lustre_read_dquot(dquot);
2242                 break;
2243         case QFILE_WR_DQUOT:
2244                 if (dquot->dq_dqb.dqb_ihardlimit ||
2245                     dquot->dq_dqb.dqb_isoftlimit ||
2246                     dquot->dq_dqb.dqb_bhardlimit ||
2247                     dquot->dq_dqb.dqb_bsoftlimit)
2248                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2249                 else
2250                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2251
2252                 rc = lustre_commit_dquot(dquot);
2253                 if (rc >= 0)
2254                         rc = 0;
2255                 break;
2256         default:
2257                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2258                 LBUG();
2259                 break;
2260         }
2261         RETURN(rc);
2262 }
2263
2264 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2265                                 struct inode *inode, int frags)
2266 {
2267 #ifdef EXT3_EXT_HAS_NO_TREE
2268         struct ext3_ext_base *base = inode;
2269 #else
2270         struct ext3_extents_tree tree;
2271         struct ext3_ext_base *base = &tree;
2272
2273         ext3_init_tree_desc(base, inode);
2274 #endif
2275         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2276          * metablocks at maxium b=16542 */
2277         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2278         return 0;
2279 }
2280
2281 #endif
2282
2283 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2284 {
2285         return (EXT3_SB(sb)->journal_bdev);
2286 }
2287 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2288
2289 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2290                           char *buf, int count, loff_t pos)
2291 {
2292         loff_t p = pos;
2293         int rc;
2294
2295         if (!f && !inode) {
2296                 CERROR("lustre_read_quota failed for no quota file!\n");
2297                 libcfs_debug_dumpstack(NULL);
2298                 return -EINVAL;
2299         }
2300
2301         /* Support for both adm and op quota files must be provided */
2302         if (f) {
2303                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2304                 rc = rc < 0 ? rc : p - pos;
2305         } else {
2306                 struct super_block *sb = inode->i_sb;
2307                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2308         }
2309         return rc;
2310 }
2311
2312 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2313 {
2314         loff_t p = pos;
2315         int rc;
2316
2317         /* Only adm quota files are supported, op updates are handled by vfs */
2318         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2319         rc = rc < 0 ? rc : p - pos;
2320
2321         return rc;
2322 }
2323
2324 void *lustre_quota_journal_start(struct inode *inode, int delete)
2325 {
2326         handle_t *handle;
2327         unsigned block_count;
2328
2329         if (delete) {
2330                 /* each indirect block (+4) may become free, attaching to the
2331                  * header list of free blocks (+1); the data block (+1) may
2332                  * become a free block (+0) or a block with free dqentries (+0) */
2333                 block_count = (4 + 1) + 1;
2334                 handle = ext3_journal_start(inode,
2335                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2336         } else {
2337                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2338                  * freeblk reusage with a header update (+1); dqentry is either reused
2339                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2340                  * a new block allocation (+1) with a header update (+1)              */
2341                 block_count = (4 + 1) + 3;
2342                 handle = ext3_journal_start(inode,
2343                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2344
2345         }
2346
2347         return handle;
2348 }
2349
2350 void lustre_quota_journal_stop(void *handle)
2351 {
2352         ext3_journal_stop((handle_t *)handle);
2353 }
2354
2355 static int ll_decode_fh_accept(void *context, struct dentry *de)
2356 {
2357         return 1;
2358 }
2359
2360 #ifdef HAVE_EXPORTFS_DECODE_FH
2361 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2362          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2363                             acceptable, context)
2364 #else
2365 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2366          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2367                                      type, acceptable, context)
2368 # define FILEID_INO32_GEN 1
2369 extern struct export_operations export_op_default;
2370 #endif
2371
2372 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2373                                       struct fsfilt_fid *fid, int ignore_gen)
2374 {
2375         struct inode  *inode;
2376         struct dentry *result;
2377
2378         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2379                                        ll_decode_fh_accept, NULL);
2380         if (IS_ERR(result)) {
2381                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2382                        fid->ino, fid->gen, PTR_ERR(result));
2383                 return result;
2384         }
2385
2386         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2387                fid->ino, fid->gen);
2388         inode = result->d_inode;
2389         if (inode == NULL)
2390                 goto err_out;
2391
2392         if (inode->i_nlink == 0 &&
2393             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2394                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2395                               " ctime -- this may indicate disk "
2396                               "corruption (inode: %lu, link: %lu, "
2397                               "count: %d)\n", inode->i_ino,
2398                               (unsigned long)inode->i_nlink,
2399                               atomic_read(&inode->i_count));
2400                 goto err_out;
2401         }
2402         if (fid->gen && inode->i_generation != fid->gen) {
2403                 /* we didn't find the right inode.. */
2404                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2405                        "count: %d, generation %u/%u\n",
2406                        inode->i_ino, (unsigned long)inode->i_nlink,
2407                        atomic_read(&inode->i_count), inode->i_generation,
2408                        fid->gen);
2409                 goto err_out;
2410         }
2411
2412         return result;
2413 err_out:
2414         l_dput(result);
2415         return ERR_PTR(-ENOENT);
2416 }
2417
2418 static struct fsfilt_operations fsfilt_ext3_ops = {
2419         .fs_type                = "ext3",
2420         .fs_owner               = THIS_MODULE,
2421         .fs_getlabel            = fsfilt_ext3_get_label,
2422         .fs_setlabel            = fsfilt_ext3_set_label,
2423         .fs_uuid                = fsfilt_ext3_uuid,
2424         .fs_start               = fsfilt_ext3_start,
2425         .fs_brw_start           = fsfilt_ext3_brw_start,
2426         .fs_extend              = fsfilt_ext3_extend,
2427         .fs_commit              = fsfilt_ext3_commit,
2428         .fs_commit_async        = fsfilt_ext3_commit_async,
2429         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2430         .fs_setattr             = fsfilt_ext3_setattr,
2431         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2432         .fs_set_md              = fsfilt_ext3_set_md,
2433         .fs_get_md              = fsfilt_ext3_get_md,
2434         .fs_readpage            = fsfilt_ext3_readpage,
2435         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2436         .fs_statfs              = fsfilt_ext3_statfs,
2437         .fs_sync                = fsfilt_ext3_sync,
2438         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2439         .fs_write_record        = fsfilt_ext3_write_record,
2440         .fs_read_record         = fsfilt_ext3_read_record,
2441         .fs_setup               = fsfilt_ext3_setup,
2442         .fs_send_bio            = fsfilt_ext3_send_bio,
2443         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2444 #ifdef HAVE_DISK_INODE_VERSION
2445         .fs_get_version         = fsfilt_ext3_get_version,
2446         .fs_set_version         = fsfilt_ext3_set_version,
2447 #endif
2448 #ifdef HAVE_QUOTA_SUPPORT
2449         .fs_quotactl            = fsfilt_ext3_quotactl,
2450         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2451         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2452         .fs_qids                = fsfilt_ext3_qids,
2453         .fs_dquot               = fsfilt_ext3_dquot,
2454         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2455 #endif
2456         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2457         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2458 };
2459
2460 static int __init fsfilt_ext3_init(void)
2461 {
2462         int rc;
2463
2464         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2465                                          sizeof(struct fsfilt_cb_data), 0, 0);
2466         if (!fcb_cache) {
2467                 CERROR("error allocating fsfilt journal callback cache\n");
2468                 GOTO(out, rc = -ENOMEM);
2469         }
2470
2471         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2472
2473         if (rc) {
2474                 int err = cfs_mem_cache_destroy(fcb_cache);
2475                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2476         }
2477 out:
2478         return rc;
2479 }
2480
2481 static void __exit fsfilt_ext3_exit(void)
2482 {
2483         int rc;
2484
2485         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2486         rc = cfs_mem_cache_destroy(fcb_cache);
2487         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2488 }
2489
2490 module_init(fsfilt_ext3_init);
2491 module_exit(fsfilt_ext3_exit);
2492
2493 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2494 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2495 MODULE_LICENSE("GPL");