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  *  lustre/lib/fsfilt_ext3.c
   5  *  Lustre filesystem abstraction routines
   6  *
   7  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   8  *   Author: Andreas Dilger <adilger@clusterfs.com>
   9  *
  10  *   This file is part of Lustre, http://www.lustre.org.
  11  *
  12  *   Lustre is free software; you can redistribute it and/or
  13  *   modify it under the terms of version 2 of the GNU General Public
  14  *   License as published by the Free Software Foundation.
  15  *
  16  *   Lustre is distributed in the hope that it will be useful,
  17  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  *   GNU General Public License for more details.
  20  *
  21  *   You should have received a copy of the GNU General Public License
  22  *   along with Lustre; if not, write to the Free Software
  23  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  24  */
  25
  26 #define DEBUG_SUBSYSTEM S_FILTER
  27
  28 #include <linux/init.h>
  29 #include <linux/module.h>
  30 #include <linux/fs.h>
  31 #include <linux/jbd.h>
  32 #include <linux/slab.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/ext3_fs.h>
  36 #include <linux/ext3_jbd.h>
  37 #include <linux/version.h>
  38 #include <linux/bitops.h>
  39 #include <linux/quota.h>
  40 #include <linux/quotaio_v1.h>
  41 #include <linux/quotaio_v2.h>
  42 #include <ext3/xattr.h>
  43
  44 #include <libcfs/kp30.h>
  45 #include <lustre_fsfilt.h>
  46 #include <obd.h>
  47 #include <lustre_quota.h>
  48 #include <linux/lustre_compat25.h>
  49 #include <linux/lprocfs_status.h>
  50
  51 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  52 #include <linux/ext3_extents.h>
  53 #endif
  54
  55 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  56 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  57 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  58 #else
  59 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  60 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  61 #endif
  62
  63 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
  64 /* for kernels 2.6.18 and later */
  65 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  66 #else
  67 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
  68 #endif
  69
  70 #define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks)
  71 #define fsfilt_ext3_journal_stop(handle)          ext3_journal_stop(handle)
  72
  73 static cfs_mem_cache_t *fcb_cache;
  74
  75 struct fsfilt_cb_data {
  76         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  77         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  78         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  79         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  80         void *cb_data;                  /* MDS/OST completion function data */
  81 };
  82
  83 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
  84 #define EXT3_XATTR_INDEX_TRUSTED        4
  85 #endif
  86
  87 static char *fsfilt_ext3_get_label(struct super_block *sb)
  88 {
  89         return EXT3_SB(sb)->s_es->s_volume_name;
  90 }
  91
  92 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
  93 {
  94         /* see e.g. fsfilt_ext3_write_record() */
  95         journal_t *journal;
  96         handle_t *handle;
  97         int err;
  98
  99         journal = EXT3_SB(sb)->s_journal;
 100         handle = journal_start(journal, 1);
 101         if (IS_ERR(handle)) {
 102                 CERROR("can't start transaction\n");
 103                 return(PTR_ERR(handle));
 104         }
 105
 106         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 107         if (err)
 108                 goto out;
 109
 110         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 111                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 112
 113         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 114
 115 out:
 116         journal_stop(handle);
 117
 118         return(err);
 119 }
 120
 121 static char *fsfilt_ext3_uuid(struct super_block *sb)
 122 {
 123         return EXT3_SB(sb)->s_es->s_uuid;
 124 }
 125
 126 #ifdef HAVE_DISK_INODE_VERSION
 127 /*
 128  * Get the 64-bit version for an inode.
 129  */
 130 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 131 {
 132         return EXT3_I(inode)->i_fs_version;
 133 }
 134
 135 /*
 136  * Set the 64-bit version and return the old version.
 137  */
 138 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 139 {
 140         __u64 old_version = EXT3_I(inode)->i_fs_version;
 141
 142         (EXT3_I(inode))->i_fs_version = new_version;
 143         return old_version;
 144 }
 145
 146 #endif
 147
 148 /*
 149  * We don't currently need any additional blocks for rmdir and
 150  * unlink transactions because we are storing the OST oa_id inside
 151  * the inode (which we will be changing anyways as part of this
 152  * transaction).
 153  */
 154 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 155                                int logs)
 156 {
 157         /* For updates to the last received file */
 158         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 159         journal_t *journal;
 160         void *handle;
 161
 162         if (current->journal_info) {
 163                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 164                        current->journal_info);
 165                 goto journal_start;
 166         }
 167
 168         switch(op) {
 169         case FSFILT_OP_RMDIR:
 170         case FSFILT_OP_UNLINK:
 171                 /* delete one file + create/update logs for each stripe */
 172                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 173                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 174                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 175                 break;
 176         case FSFILT_OP_RENAME:
 177                 /* modify additional directory */
 178                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 179                 /* no break */
 180         case FSFILT_OP_SYMLINK:
 181                 /* additional block + block bitmap + GDT for long symlink */
 182                 nblocks += 3;
 183                 /* no break */
 184         case FSFILT_OP_CREATE: {
 185 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 186                 static int warned;
 187                 if (!warned) {
 188                         if (!test_opt(inode->i_sb, EXTENTS)) {
 189                                 warned = 1;
 190                         } else if (((EXT3_I(inode)->i_flags &
 191                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 192                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 193                                 CWARN("extent-mapped directory found - contact "
 194                                       "CFS: support@clusterfs.com\n");
 195                                 warned = 1;
 196                         }
 197                 }
 198 #endif
 199                 /* no break */
 200         }
 201         case FSFILT_OP_MKDIR:
 202         case FSFILT_OP_MKNOD:
 203                 /* modify one inode + block bitmap + GDT */
 204                 nblocks += 3;
 205                 /* no break */
 206         case FSFILT_OP_LINK:
 207                 /* modify parent directory */
 208                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 209                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 210                 /* create/update logs for each stripe */
 211                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 212                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 213                 break;
 214         case FSFILT_OP_SETATTR:
 215                 /* Setattr on inode */
 216                 nblocks += 1;
 217                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 218                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 219                 /* quota chown log for each stripe */
 220                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 221                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 222                 break;
 223         case FSFILT_OP_CANCEL_UNLINK:
 224                 /* blocks for log header bitmap update OR
 225                  * blocks for catalog header bitmap update + unlink of logs */
 226                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 227                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 228                 break;
 229         case FSFILT_OP_JOIN:
 230                 /* delete 2 file(file + array id) + create 1 file (array id)
 231                  * create/update logs for each stripe */
 232                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 233
 234                 /*create array log for head file*/
 235                 nblocks += 3;
 236                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 237                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
 238                 /*update head file array */
 239                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 240                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 241                 break;
 242         default: CERROR("unknown transaction start op %d\n", op);
 243                 LBUG();
 244         }
 245
 246         LASSERT(current->journal_info == desc_private);
 247         journal = EXT3_SB(inode->i_sb)->s_journal;
 248         if (nblocks > journal->j_max_transaction_buffers) {
 249                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 250                        nblocks, op, logs, journal->j_max_transaction_buffers);
 251                 nblocks = journal->j_max_transaction_buffers;
 252         }
 253
 254  journal_start:
 255         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 256         handle = fsfilt_ext3_journal_start(inode, nblocks);
 257
 258         if (!IS_ERR(handle))
 259                 LASSERT(current->journal_info == handle);
 260         else
 261                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 262                        op, nblocks, PTR_ERR(handle));
 263         return handle;
 264 }
 265
 266 /*
 267  * Calculate the number of buffer credits needed to write multiple pages in
 268  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 269  * doesn't have a nice API for calculating this sort of thing in advance.
 270  *
 271  * See comment above ext3_writepage_trans_blocks for details.  We assume
 272  * no data journaling is being done, but it does allow for all of the pages
 273  * being non-contiguous.  If we are guaranteed contiguous pages we could
 274  * reduce the number of (d)indirect blocks a lot.
 275  *
 276  * With N blocks per page and P pages, for each inode we have at most:
 277  * N*P indirect
 278  * min(N*P, blocksize/4 + 1) dindirect blocks
 279  * niocount tindirect
 280  *
 281  * For the entire filesystem, we have at most:
 282  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 283  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 284  * objcount inode blocks
 285  * 1 superblock
 286  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 287  *
 288  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 289  */
 290 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 291                                       int niocount, struct niobuf_local *nb)
 292 {
 293         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 294         __u64 next_indir;
 295         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 296         int nbitmaps = 0, ngdblocks;
 297         int needed = objcount + 1; /* inodes + superblock */
 298         int i, j;
 299
 300         for (i = 0, j = 0; i < objcount; i++, fso++) {
 301                 /* two or more dindirect blocks in case we cross boundary */
 302                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 303                                     nb[j].offset) >>
 304                                    sb->s_blocksize_bits) /
 305                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 306                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 307
 308                 /* leaf, indirect, tindirect blocks for first block */
 309                 nbitmaps += blockpp + 2;
 310
 311                 j += fso->fso_bufcnt;
 312         }
 313
 314         next_indir = nb[0].offset +
 315                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 316         for (i = 1; i < niocount; i++) {
 317                 if (nb[i].offset >= next_indir) {
 318                         nbitmaps++;     /* additional indirect */
 319                         next_indir = nb[i].offset +
 320                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 321                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 322                         nbitmaps++;     /* additional indirect */
 323                 }
 324                 nbitmaps += blockpp;    /* each leaf in different group? */
 325         }
 326
 327         ngdblocks = nbitmaps;
 328         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 329                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 330         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 331                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 332
 333         needed += nbitmaps + ngdblocks;
 334
 335         /* last_rcvd update */
 336         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 337
 338 #if defined(CONFIG_QUOTA)
 339         /* We assume that there will be 1 bit set in s_dquot.flags for each
 340          * quota file that is active.  This is at least true for now.
 341          */
 342         needed += hweight32(sb_any_quota_enabled(sb)) *
 343                 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 344 #endif
 345
 346         return needed;
 347 }
 348
 349 /* We have to start a huge journal transaction here to hold all of the
 350  * metadata for the pages being written here.  This is necessitated by
 351  * the fact that we do lots of prepare_write operations before we do
 352  * any of the matching commit_write operations, so even if we split
 353  * up to use "smaller" transactions none of them could complete until
 354  * all of them were opened.  By having a single journal transaction,
 355  * we eliminate duplicate reservations for common blocks like the
 356  * superblock and group descriptors or bitmaps.
 357  *
 358  * We will start the transaction here, but each prepare_write will
 359  * add a refcount to the transaction, and each commit_write will
 360  * remove a refcount.  The transaction will be closed when all of
 361  * the pages have been written.
 362  */
 363 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 364                                    int niocount, struct niobuf_local *nb,
 365                                    void *desc_private, int logs)
 366 {
 367         journal_t *journal;
 368         handle_t *handle;
 369         int needed;
 370         ENTRY;
 371
 372         LASSERT(current->journal_info == desc_private);
 373         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 374         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 375
 376         /* The number of blocks we could _possibly_ dirty can very large.
 377          * We reduce our request if it is absurd (and we couldn't get that
 378          * many credits for a single handle anyways).
 379          *
 380          * At some point we have to limit the size of I/Os sent at one time,
 381          * increase the size of the journal, or we have to calculate the
 382          * actual journal requirements more carefully by checking all of
 383          * the blocks instead of being maximally pessimistic.  It remains to
 384          * be seen if this is a real problem or not.
 385          */
 386         if (needed > journal->j_max_transaction_buffers) {
 387                 CERROR("want too many journal credits (%d) using %d instead\n",
 388                        needed, journal->j_max_transaction_buffers);
 389                 needed = journal->j_max_transaction_buffers;
 390         }
 391
 392         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 393         handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
 394         if (IS_ERR(handle)) {
 395                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 396                        PTR_ERR(handle));
 397         } else {
 398                 LASSERT(handle->h_buffer_credits >= needed);
 399                 LASSERT(current->journal_info == handle);
 400         }
 401
 402         RETURN(handle);
 403 }
 404
 405 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 406 {
 407        handle_t *handle = h;
 408
 409        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 410        if (nblocks == 0) {
 411                handle->h_buffer_credits = 0;
 412                CWARN("setting credits of handle %p to zero by request\n", h);
 413        }
 414
 415        if (handle->h_buffer_credits > nblocks)
 416                 return 0;
 417        if (journal_extend(handle, nblocks) == 0)
 418                 return 0;
 419
 420        ext3_mark_inode_dirty(handle, inode);
 421        return journal_restart(handle, nblocks);
 422 }
 423
 424 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 425 {
 426         int rc;
 427         handle_t *handle = h;
 428
 429         LASSERT(current->journal_info == handle);
 430         if (force_sync)
 431                 handle->h_sync = 1; /* recovery likes this */
 432
 433         rc = fsfilt_ext3_journal_stop(handle);
 434
 435         return rc;
 436 }
 437
 438 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 439                                     void **wait_handle)
 440 {
 441         unsigned long tid;
 442         transaction_t *transaction;
 443         handle_t *handle = h;
 444         journal_t *journal;
 445         int rc;
 446
 447         LASSERT(current->journal_info == handle);
 448
 449         transaction = handle->h_transaction;
 450         journal = transaction->t_journal;
 451         tid = transaction->t_tid;
 452         /* we don't want to be blocked */
 453         handle->h_sync = 0;
 454         rc = fsfilt_ext3_journal_stop(handle);
 455         if (rc) {
 456                 CERROR("error while stopping transaction: %d\n", rc);
 457                 return rc;
 458         }
 459         log_start_commit(journal, tid);
 460
 461         *wait_handle = (void *) tid;
 462         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 463         return 0;
 464 }
 465
 466 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 467 {
 468         journal_t *journal = EXT3_JOURNAL(inode);
 469         tid_t tid = (tid_t)(long)h;
 470
 471         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 472         if (unlikely(is_journal_aborted(journal)))
 473                 return -EIO;
 474
 475         log_wait_commit(EXT3_JOURNAL(inode), tid);
 476
 477         if (unlikely(is_journal_aborted(journal)))
 478                 return -EIO;
 479         return 0;
 480 }
 481
 482 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 483                                struct iattr *iattr, int do_trunc)
 484 {
 485         struct inode *inode = dentry->d_inode;
 486         int rc = 0;
 487
 488         lock_24kernel();
 489
 490         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 491          * We are already writing the inode to disk as part of this
 492          * transaction and want to avoid a lot of extra inode writeout
 493          * later on. b=9828 */
 494         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 495                 /* ATTR_SIZE would invoke truncate: clear it */
 496                 iattr->ia_valid &= ~ATTR_SIZE;
 497                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 498                 i_size_write(inode, iattr->ia_size);
 499
 500                 if (iattr->ia_valid & ATTR_UID)
 501                         inode->i_uid = iattr->ia_uid;
 502                 if (iattr->ia_valid & ATTR_GID)
 503                         inode->i_gid = iattr->ia_gid;
 504                 if (iattr->ia_valid & ATTR_ATIME)
 505                         inode->i_atime = iattr->ia_atime;
 506                 if (iattr->ia_valid & ATTR_MTIME)
 507                         inode->i_mtime = iattr->ia_mtime;
 508                 if (iattr->ia_valid & ATTR_CTIME)
 509                         inode->i_ctime = iattr->ia_ctime;
 510                 if (iattr->ia_valid & ATTR_MODE) {
 511                         inode->i_mode = iattr->ia_mode;
 512
 513                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 514                                 inode->i_mode &= ~S_ISGID;
 515                 }
 516
 517                 inode->i_sb->s_op->dirty_inode(inode);
 518
 519                 goto out;
 520         }
 521
 522         /* Don't allow setattr to change file type */
 523         if (iattr->ia_valid & ATTR_MODE)
 524                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 525                                  (iattr->ia_mode & ~S_IFMT);
 526
 527         /* We set these flags on the client, but have already checked perms
 528          * so don't confuse inode_change_ok. */
 529         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 530
 531         if (inode->i_op->setattr) {
 532                 rc = inode->i_op->setattr(dentry, iattr);
 533         } else {
 534                 rc = inode_change_ok(inode, iattr);
 535                 if (!rc)
 536                         rc = inode_setattr(inode, iattr);
 537         }
 538
 539  out:
 540         unlock_24kernel();
 541         RETURN(rc);
 542 }
 543
 544 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 545                                  unsigned int cmd, unsigned long arg)
 546 {
 547         int rc = 0;
 548         ENTRY;
 549
 550         /* FIXME: Can't do this because of nested transaction deadlock */
 551         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 552                 CERROR("can't set data journal flag on file\n");
 553                 RETURN(-EPERM);
 554         }
 555
 556         if (inode->i_fop->ioctl)
 557                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 558         else
 559                 RETURN(-ENOTTY);
 560
 561         RETURN(rc);
 562 }
 563
 564 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 565                               void *lmm, int lmm_size, const char *name)
 566 {
 567         int rc;
 568
 569         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 570
 571         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 572                                    name, lmm, lmm_size, 0);
 573
 574
 575         if (rc && rc != -EROFS)
 576                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 577                        inode->i_ino, rc);
 578         return rc;
 579 }
 580
 581 /* Must be called with i_mutex held */
 582 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 583                               const char *name)
 584 {
 585         int rc;
 586
 587         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 588
 589         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 590                             name, lmm, lmm_size);
 591
 592         /* This gives us the MD size */
 593         if (lmm == NULL)
 594                 return (rc == -ENODATA) ? 0 : rc;
 595
 596         if (rc < 0) {
 597                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 598                        EXT3_XATTR_INDEX_TRUSTED, name,
 599                        inode->i_ino, rc);
 600                 memset(lmm, 0, lmm_size);
 601                 return (rc == -ENODATA) ? 0 : rc;
 602         }
 603
 604         return rc;
 605 }
 606
 607 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 608 {
 609         submit_bio(rw, bio);
 610         return 0;
 611 }
 612
 613 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 614                                     loff_t *off)
 615 {
 616         struct inode *inode = file->f_dentry->d_inode;
 617         int rc = 0;
 618
 619         if (S_ISREG(inode->i_mode))
 620                 rc = file->f_op->read(file, buf, count, off);
 621         else {
 622                 const int blkbits = inode->i_sb->s_blocksize_bits;
 623                 const int blksize = inode->i_sb->s_blocksize;
 624
 625                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 626                        count, inode->i_ino, *off);
 627                 while (count > 0) {
 628                         struct buffer_head *bh;
 629
 630                         bh = NULL;
 631                         if (*off < i_size_read(inode)) {
 632                                 int err = 0;
 633
 634                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 635                                                 0, &err);
 636
 637                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 638
 639                                 if (bh) {
 640                                         memcpy(buf, bh->b_data, blksize);
 641                                         brelse(bh);
 642                                 } else if (err) {
 643                                         /* XXX in theory we should just fake
 644                                          * this buffer and continue like ext3,
 645                                          * especially if this is a partial read
 646                                          */
 647                                         CERROR("error read dir %lu+%llu: %d\n",
 648                                                inode->i_ino, *off, err);
 649                                         RETURN(err);
 650                                 }
 651                         }
 652                         if (!bh) {
 653                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 654
 655                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 656                                 memset(fake, 0, sizeof(*fake));
 657                                 fake->rec_len = cpu_to_le16(blksize);
 658                         }
 659                         count -= blksize;
 660                         buf += blksize;
 661                         *off += blksize;
 662                         rc += blksize;
 663                 }
 664         }
 665
 666         return rc;
 667 }
 668
 669 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 670 {
 671         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 672
 673         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 674
 675         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 676 }
 677
 678 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 679                                       void *handle, fsfilt_cb_t cb_func,
 680                                       void *cb_data)
 681 {
 682         struct fsfilt_cb_data *fcb;
 683
 684         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 685         if (fcb == NULL)
 686                 RETURN(-ENOMEM);
 687
 688         fcb->cb_func = cb_func;
 689         fcb->cb_obd = obd;
 690         fcb->cb_last_rcvd = last_rcvd;
 691         fcb->cb_data = cb_data;
 692
 693         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 694         journal_callback_set(handle, fsfilt_ext3_cb_func,
 695                              (struct journal_callback *)fcb);
 696
 697         return 0;
 698 }
 699
 700 /*
 701  * We need to hack the return value for the free inode counts because
 702  * the current EA code requires one filesystem block per inode with EAs,
 703  * so it is possible to run out of blocks before we run out of inodes.
 704  *
 705  * This can be removed when the ext3 EA code is fixed.
 706  */
 707 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 708 {
 709         struct kstatfs sfs;
 710         int rc;
 711
 712         memset(&sfs, 0, sizeof(sfs));
 713
 714         rc = ll_do_statfs(sb, &sfs);
 715
 716         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 717                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 718                 sfs.f_ffree = sfs.f_bfree;
 719         }
 720
 721         statfs_pack(osfs, &sfs);
 722         return rc;
 723 }
 724
 725 static int fsfilt_ext3_sync(struct super_block *sb)
 726 {
 727         return ext3_force_commit(sb);
 728 }
 729
 730 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 731 #warning "kernel code has old extents/mballoc patch, disabling"
 732 #undef EXT3_MULTIBLOCK_ALLOCATOR
 733 #endif
 734 #ifndef EXT3_EXTENTS_FL
 735 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 736 #endif
 737
 738 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 739 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 740 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 741 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 742 #else
 743 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 744 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 745 #endif
 746
 747 #ifndef EXT_ASSERT
 748 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 749 #endif
 750
 751 #ifdef EXT3_EXT_HAS_NO_TREE
 752 /* for kernels 2.6.18 and later */
 753 #define ext3_ext_base                   inode
 754 #define ext3_ext_base2inode(inode)      (inode)
 755 #define EXT_DEPTH(inode)                ext_depth(inode)
 756 #define EXT_GENERATION(inode)           ext_generation(inode)
 757 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 758                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 759 #else
 760 #define ext3_ext_base                   ext3_extents_tree
 761 #define ext3_ext_base2inode(tree)       (tree->inode)
 762 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 763                         ext3_ext_walk_space(tree, block, num, cb);
 764 #endif
 765
 766 #include <linux/lustre_version.h>
 767 #if EXT3_EXT_MAGIC == 0xf301
 768 #define ee_start e_start
 769 #define ee_block e_block
 770 #define ee_len   e_num
 771 #endif
 772 #ifndef EXT3_BB_MAX_BLOCKS
 773 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 774         ext3_new_blocks(handle, inode, count, goal, err)
 775 #endif
 776
 777 struct bpointers {
 778         unsigned long *blocks;
 779         int *created;
 780         unsigned long start;
 781         int num;
 782         int init_num;
 783         int create;
 784 };
 785
 786 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 787                               unsigned long block, int *aflags)
 788 {
 789         struct ext3_inode_info *ei = EXT3_I(inode);
 790         unsigned long bg_start;
 791         unsigned long colour;
 792         int depth;
 793
 794         if (path) {
 795                 struct ext3_extent *ex;
 796                 depth = path->p_depth;
 797
 798                 /* try to predict block placement */
 799                 if ((ex = path[depth].p_ext)) {
 800 #if 0
 801                         /* This prefers to eat into a contiguous extent
 802                          * rather than find an extent that the whole
 803                          * request will fit into.  This can fragment data
 804                          * block allocation and prevents our lovely 1M I/Os
 805                          * from reaching the disk intact. */
 806                         if (ex->ee_block + ex->ee_len == block)
 807                                 *aflags |= 1;
 808 #endif
 809                         return ex->ee_start + (block - ex->ee_block);
 810                 }
 811
 812                 /* it looks index is empty
 813                  * try to find starting from index itself */
 814                 if (path[depth].p_bh)
 815                         return path[depth].p_bh->b_blocknr;
 816         }
 817
 818         /* OK. use inode's group */
 819         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 820                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 821         colour = (current->pid % 16) *
 822                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 823         return bg_start + colour + block;
 824 }
 825
 826 #define ll_unmap_underlying_metadata(sb, blocknr) \
 827         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 828
 829 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 830 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 831                                 struct ext3_ext_path *path, unsigned long block,
 832                                 unsigned long *count, int *err)
 833 {
 834         unsigned long pblock, goal;
 835         int aflags = 0;
 836         struct inode *inode = ext3_ext_base2inode(base);
 837
 838         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 839         aflags |= 2; /* block have been already reserved */
 840         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 841         return pblock;
 842
 843 }
 844 #else
 845 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 846                                 struct ext3_ext_path *path, unsigned long block,
 847                                 unsigned long *count, int *err)
 848 {
 849         struct inode *inode = ext3_ext_base2inode(base);
 850         struct ext3_allocation_request ar;
 851         unsigned long pblock;
 852         int aflags;
 853
 854         /* find neighbour allocated blocks */
 855         ar.lleft = block;
 856         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 857         if (*err)
 858                 return 0;
 859         ar.lright = block;
 860         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 861         if (*err)
 862                 return 0;
 863
 864         /* allocate new block */
 865         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 866         ar.inode = inode;
 867         ar.logical = block;
 868         ar.len = *count;
 869         ar.flags = EXT3_MB_HINT_DATA;
 870         pblock = ext3_mb_new_blocks(handle, &ar, err);
 871         *count = ar.len;
 872         return pblock;
 873
 874 }
 875 #endif
 876
 877 #ifdef EXT3_EXT_HAS_NO_TREE
 878 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 879                                   struct ext3_ext_path *path,
 880                                   struct ext3_ext_cache *cex,
 881                                   void *cbdata)
 882 {
 883         struct bpointers *bp = cbdata;
 884 #else
 885 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 886                                   struct ext3_ext_path *path,
 887                                   struct ext3_ext_cache *cex)
 888 {
 889         struct bpointers *bp = base->private;
 890 #endif
 891         struct inode *inode = ext3_ext_base2inode(base);
 892         struct ext3_extent nex;
 893         unsigned long pblock;
 894         unsigned long tgen;
 895         int err, i;
 896         unsigned long count;
 897         handle_t *handle;
 898
 899         i = EXT_DEPTH(base);
 900         EXT_ASSERT(i == path->p_depth);
 901         EXT_ASSERT(path[i].p_hdr);
 902
 903         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 904                 err = EXT_CONTINUE;
 905                 goto map;
 906         }
 907
 908         if (bp->create == 0) {
 909                 i = 0;
 910                 if (cex->ec_block < bp->start)
 911                         i = bp->start - cex->ec_block;
 912                 if (i >= cex->ec_len)
 913                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 914                                         i, cex->ec_len);
 915                 for (; i < cex->ec_len && bp->num; i++) {
 916                         *(bp->created) = 0;
 917                         bp->created++;
 918                         *(bp->blocks) = 0;
 919                         bp->blocks++;
 920                         bp->num--;
 921                         bp->start++;
 922                 }
 923
 924                 return EXT_CONTINUE;
 925         }
 926
 927         tgen = EXT_GENERATION(base);
 928         count = ext3_ext_calc_credits_for_insert(base, path);
 929         ext3_up_truncate_sem(inode);
 930
 931         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 932         if (IS_ERR(handle)) {
 933                 ext3_down_truncate_sem(inode);
 934                 return PTR_ERR(handle);
 935         }
 936
 937         ext3_down_truncate_sem(inode);
 938         if (tgen != EXT_GENERATION(base)) {
 939                 /* the tree has changed. so path can be invalid at moment */
 940                 fsfilt_ext3_journal_stop(handle);
 941                 return EXT_REPEAT;
 942         }
 943
 944         count = cex->ec_len;
 945         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 946         if (!pblock)
 947                 goto out;
 948         EXT_ASSERT(count <= cex->ec_len);
 949
 950         /* insert new extent */
 951         nex.ee_block = cex->ec_block;
 952         nex.ee_start = pblock;
 953         nex.ee_len = count;
 954         err = ext3_ext_insert_extent(handle, base, path, &nex);
 955         if (err) {
 956                 CERROR("can't insert extent: %d\n", err);
 957                 /* XXX: export ext3_free_blocks() */
 958                 /*ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);*/
 959                 goto out;
 960         }
 961
 962         /*
 963          * Putting len of the actual extent we just inserted,
 964          * we are asking ext3_ext_walk_space() to continue
 965          * scaning after that block
 966          */
 967         cex->ec_len = nex.ee_len;
 968         cex->ec_start = nex.ee_start;
 969         BUG_ON(nex.ee_len == 0);
 970         BUG_ON(nex.ee_block != cex->ec_block);
 971
 972 out:
 973         fsfilt_ext3_journal_stop(handle);
 974 map:
 975         if (err >= 0) {
 976                 /* map blocks */
 977                 if (bp->num == 0) {
 978                         CERROR("hmm. why do we find this extent?\n");
 979                         CERROR("initial space: %lu:%u\n",
 980                                 bp->start, bp->init_num);
 981                         CERROR("current extent: %u/%u/%u %d\n",
 982                                 cex->ec_block, cex->ec_len,
 983                                 cex->ec_start, cex->ec_type);
 984                 }
 985                 i = 0;
 986                 if (cex->ec_block < bp->start)
 987                         i = bp->start - cex->ec_block;
 988                 if (i >= cex->ec_len)
 989                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 990                                         i, cex->ec_len);
 991                 for (; i < cex->ec_len && bp->num; i++) {
 992                         *(bp->blocks) = cex->ec_start + i;
 993                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 994                                 *(bp->created) = 0;
 995                         } else {
 996                                 *(bp->created) = 1;
 997                                 /* unmap any possible underlying metadata from
 998                                  * the block device mapping.  bug 6998. */
 999                                 ll_unmap_underlying_metadata(inode->i_sb,
1000                                                              *(bp->blocks));
1001                         }
1002                         bp->created++;
1003                         bp->blocks++;
1004                         bp->num--;
1005                         bp->start++;
1006                 }
1007         }
1008         return err;
1009 }
1010
1011 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1012                        unsigned long num, unsigned long *blocks,
1013                        int *created, int create)
1014 {
1015 #ifdef EXT3_EXT_HAS_NO_TREE
1016         struct ext3_ext_base *base = inode;
1017 #else
1018         struct ext3_extents_tree tree;
1019         struct ext3_ext_base *base = &tree;
1020 #endif
1021         struct bpointers bp;
1022         int err;
1023
1024         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1025                block, block + num - 1, (unsigned) inode->i_ino);
1026
1027 #ifndef EXT3_EXT_HAS_NO_TREE
1028         ext3_init_tree_desc(base, inode);
1029         tree.private = &bp;
1030 #endif
1031         bp.blocks = blocks;
1032         bp.created = created;
1033         bp.start = block;
1034         bp.init_num = bp.num = num;
1035         bp.create = create;
1036
1037         ext3_down_truncate_sem(inode);
1038         err = fsfilt_ext3_ext_walk_space(base, block, num,
1039                                          ext3_ext_new_extent_cb, &bp);
1040         ext3_ext_invalidate_cache(base);
1041         ext3_up_truncate_sem(inode);
1042
1043         return err;
1044 }
1045
1046 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1047                                     int pages, unsigned long *blocks,
1048                                     int *created, int create)
1049 {
1050         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1051         int rc = 0, i = 0;
1052         struct page *fp = NULL;
1053         int clen = 0;
1054
1055         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1056                 inode->i_ino, pages, (*page)->index);
1057
1058         /* pages are sorted already. so, we just have to find
1059          * contig. space and process them properly */
1060         while (i < pages) {
1061                 if (fp == NULL) {
1062                         /* start new extent */
1063                         fp = *page++;
1064                         clen = 1;
1065                         i++;
1066                         continue;
1067                 } else if (fp->index + clen == (*page)->index) {
1068                         /* continue the extent */
1069                         page++;
1070                         clen++;
1071                         i++;
1072                         continue;
1073                 }
1074
1075                 /* process found extent */
1076                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1077                                         clen * blocks_per_page, blocks,
1078                                         created, create);
1079                 if (rc)
1080                         GOTO(cleanup, rc);
1081
1082                 /* look for next extent */
1083                 fp = NULL;
1084                 blocks += blocks_per_page * clen;
1085                 created += blocks_per_page * clen;
1086         }
1087
1088         if (fp)
1089                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1090                                         clen * blocks_per_page, blocks,
1091                                         created, create);
1092 cleanup:
1093         return rc;
1094 }
1095 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1096
1097 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1098                                unsigned long *blocks, int *created, int create);
1099 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1100                                    int pages, unsigned long *blocks,
1101                                    int *created, int create)
1102 {
1103         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1104         unsigned long *b;
1105         int rc = 0, i, *cr;
1106
1107         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1108                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1109                 if (rc) {
1110                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1111                                inode->i_ino, *b, *cr, create, rc);
1112                         break;
1113                 }
1114
1115                 b += blocks_per_page;
1116                 cr += blocks_per_page;
1117         }
1118         return rc;
1119 }
1120
1121 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1122                                 int pages, unsigned long *blocks,
1123                                 int *created, int create,
1124                                 struct semaphore *optional_sem)
1125 {
1126         int rc;
1127 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1128         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1129                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1130                                                      blocks, created, create);
1131                 return rc;
1132         }
1133 #endif
1134         if (optional_sem != NULL)
1135                 down(optional_sem);
1136         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1137                                             created, create);
1138         if (optional_sem != NULL)
1139                 up(optional_sem);
1140
1141         return rc;
1142 }
1143
1144 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1145 {
1146         unsigned long block;
1147         struct buffer_head *bh;
1148         int err, blocksize, csize, boffs, osize = size;
1149
1150         /* prevent reading after eof */
1151         lock_kernel();
1152         if (i_size_read(inode) < *offs + size) {
1153                 size = i_size_read(inode) - *offs;
1154                 unlock_kernel();
1155                 if (size < 0) {
1156                         CERROR("size %llu is too short for read %u@%llu\n",
1157                                i_size_read(inode), size, *offs);
1158                         return -EIO;
1159                 } else if (size == 0) {
1160                         return 0;
1161                 }
1162         } else {
1163                 unlock_kernel();
1164         }
1165
1166         blocksize = 1 << inode->i_blkbits;
1167
1168         while (size > 0) {
1169                 block = *offs >> inode->i_blkbits;
1170                 boffs = *offs & (blocksize - 1);
1171                 csize = min(blocksize - boffs, size);
1172                 bh = ext3_bread(NULL, inode, block, 0, &err);
1173                 if (!bh) {
1174                         CERROR("can't read block: %d\n", err);
1175                         return err;
1176                 }
1177
1178                 memcpy(buf, bh->b_data + boffs, csize);
1179                 brelse(bh);
1180
1181                 *offs += csize;
1182                 buf += csize;
1183                 size -= csize;
1184         }
1185         return osize;
1186 }
1187 EXPORT_SYMBOL(fsfilt_ext3_read);
1188
1189 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1190                                    int size, loff_t *offs)
1191 {
1192         int rc;
1193         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1194         if (rc > 0)
1195                 rc = 0;
1196         return rc;
1197 }
1198
1199 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1200                                 loff_t *offs, handle_t *handle)
1201 {
1202         struct buffer_head *bh = NULL;
1203         loff_t old_size = i_size_read(inode), offset = *offs;
1204         loff_t new_size = i_size_read(inode);
1205         unsigned long block;
1206         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1207
1208         while (bufsize > 0) {
1209                 if (bh != NULL)
1210                         brelse(bh);
1211
1212                 block = offset >> inode->i_blkbits;
1213                 boffs = offset & (blocksize - 1);
1214                 size = min(blocksize - boffs, bufsize);
1215                 bh = ext3_bread(handle, inode, block, 1, &err);
1216                 if (!bh) {
1217                         CERROR("can't read/create block: %d\n", err);
1218                         break;
1219                 }
1220
1221                 err = ext3_journal_get_write_access(handle, bh);
1222                 if (err) {
1223                         CERROR("journal_get_write_access() returned error %d\n",
1224                                err);
1225                         break;
1226                 }
1227                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1228                 memcpy(bh->b_data + boffs, buf, size);
1229                 err = ext3_journal_dirty_metadata(handle, bh);
1230                 if (err) {
1231                         CERROR("journal_dirty_metadata() returned error %d\n",
1232                                err);
1233                         break;
1234                 }
1235                 if (offset + size > new_size)
1236                         new_size = offset + size;
1237                 offset += size;
1238                 bufsize -= size;
1239                 buf += size;
1240         }
1241         if (bh)
1242                 brelse(bh);
1243
1244         /* correct in-core and on-disk sizes */
1245         if (new_size > i_size_read(inode)) {
1246                 lock_kernel();
1247                 if (new_size > i_size_read(inode))
1248                         i_size_write(inode, new_size);
1249                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1250                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1251                 if (i_size_read(inode) > old_size)
1252                         mark_inode_dirty(inode);
1253                 unlock_kernel();
1254         }
1255
1256         if (err == 0)
1257                 *offs = offset;
1258         return err;
1259 }
1260 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1261
1262 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1263                                     loff_t *offs, int force_sync)
1264 {
1265         struct inode *inode = file->f_dentry->d_inode;
1266         handle_t *handle;
1267         int err, block_count = 0, blocksize;
1268
1269         /* Determine how many transaction credits are needed */
1270         blocksize = 1 << inode->i_blkbits;
1271         block_count = (*offs & (blocksize - 1)) + bufsize;
1272         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1273
1274         handle = fsfilt_ext3_journal_start(inode,
1275                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1276         if (IS_ERR(handle)) {
1277                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1278                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1279                 return PTR_ERR(handle);
1280         }
1281
1282         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1283
1284         if (!err && force_sync)
1285                 handle->h_sync = 1; /* recovery likes this */
1286
1287         fsfilt_ext3_journal_stop(handle);
1288
1289         return err;
1290 }
1291
1292 static int fsfilt_ext3_setup(struct super_block *sb)
1293 {
1294 #if 0
1295         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1296         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1297 #endif
1298 #ifdef S_PDIROPS
1299         CWARN("Enabling PDIROPS\n");
1300         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1301         sb->s_flags |= S_PDIROPS;
1302 #endif
1303         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1304                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1305 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13)) && HAVE_QUOTA_SUPPORT
1306         set_opt(EXT3_SB(sb)->s_mount_opt, QUOTA);
1307 #endif
1308         return 0;
1309 }
1310
1311 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1312    objects. Logs is number of logfiles to update */
1313 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1314 {
1315         if ( !fso ) {
1316                 switch(op) {
1317                 case FSFILT_OP_CREATE:
1318                                  /* directory leaf, index & indirect & EA*/
1319                         return 4 + 3 * logs;
1320                 case FSFILT_OP_UNLINK:
1321                         return 3 * logs;
1322                 }
1323         } else {
1324                 int i;
1325                 int needed = 0;
1326                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1327                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1328                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1329                 for (i = 0; i < op; i++, fso++) {
1330                         int nblocks = fso->fso_bufcnt * blockpp;
1331                         int ndindirect = min(nblocks, addrpp + 1);
1332                         int nindir = nblocks + ndindirect + 1;
1333
1334                         needed += nindir;
1335                 }
1336                 return needed + 3 * logs;
1337         }
1338
1339         return 0;
1340 }
1341
1342 static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
1343
1344 #define DQINFO_COPY(out, in)                    \
1345 do {                                            \
1346         Q_COPY(out, in, dqi_bgrace);            \
1347         Q_COPY(out, in, dqi_igrace);            \
1348         Q_COPY(out, in, dqi_flags);             \
1349         Q_COPY(out, in, dqi_valid);             \
1350 } while (0)
1351
1352 #define DQBLK_COPY(out, in)                     \
1353 do {                                            \
1354         Q_COPY(out, in, dqb_bhardlimit);        \
1355         Q_COPY(out, in, dqb_bsoftlimit);        \
1356         Q_COPY(out, in, dqb_curspace);          \
1357         Q_COPY(out, in, dqb_ihardlimit);        \
1358         Q_COPY(out, in, dqb_isoftlimit);        \
1359         Q_COPY(out, in, dqb_curinodes);         \
1360         Q_COPY(out, in, dqb_btime);             \
1361         Q_COPY(out, in, dqb_itime);             \
1362         Q_COPY(out, in, dqb_valid);             \
1363 } while (0)
1364
1365
1366
1367 static int fsfilt_ext3_quotactl(struct super_block *sb,
1368                                 struct obd_quotactl *oqc)
1369 {
1370         int i, rc = 0, error = 0;
1371         struct quotactl_ops *qcop;
1372         struct if_dqinfo *info;
1373         struct if_dqblk *dqblk;
1374         ENTRY;
1375
1376         if (!sb->s_qcop)
1377                 RETURN(-ENOSYS);
1378
1379         OBD_ALLOC_PTR(info);
1380         if (!info)
1381                 RETURN(-ENOMEM);
1382         OBD_ALLOC_PTR(dqblk);
1383         if (!dqblk) {
1384                 OBD_FREE_PTR(info);
1385                 RETURN(-ENOMEM);
1386         }
1387
1388         DQINFO_COPY(info, &oqc->qc_dqinfo);
1389         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1390
1391         qcop = sb->s_qcop;
1392         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1393                 for (i = 0; i < MAXQUOTAS; i++) {
1394                         if (!Q_TYPESET(oqc, i))
1395                                 continue;
1396
1397                         if (oqc->qc_cmd == Q_QUOTAON) {
1398                                 if (!qcop->quota_on)
1399                                         GOTO(out, rc = -ENOSYS);
1400                                 rc = qcop->quota_on(sb, i, oqc->qc_id,
1401                                                     (char *)op_quotafile[i]);
1402                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1403                                 if (!qcop->quota_off)
1404                                         GOTO(out, rc = -ENOSYS);
1405                                 rc = qcop->quota_off(sb, i);
1406                         }
1407
1408                         if (rc == -EBUSY)
1409                                 error = rc;
1410                         else if (rc)
1411                                 GOTO(out, rc);
1412                 }
1413                 GOTO(out, rc ?: error);
1414         }
1415
1416         switch (oqc->qc_cmd) {
1417         case Q_GETOINFO:
1418         case Q_GETINFO:
1419                 if (!qcop->get_info)
1420                         GOTO(out, rc = -ENOSYS);
1421                 rc = qcop->get_info(sb, oqc->qc_type, info);
1422                 break;
1423         case Q_SETQUOTA:
1424         case Q_INITQUOTA:
1425                 if (!qcop->set_dqblk)
1426                         GOTO(out, rc = -ENOSYS);
1427                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1428                 break;
1429         case Q_GETOQUOTA:
1430         case Q_GETQUOTA:
1431                 if (!qcop->get_dqblk)
1432                         GOTO(out, rc = -ENOSYS);
1433                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1434                 break;
1435         case Q_SYNC:
1436                 if (!sb->s_qcop->quota_sync)
1437                         GOTO(out, rc = -ENOSYS);
1438                 qcop->quota_sync(sb, oqc->qc_type);
1439                 break;
1440         default:
1441                 CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
1442                 LBUG();
1443         }
1444 out:
1445         DQINFO_COPY(&oqc->qc_dqinfo, info);
1446         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1447
1448         OBD_FREE_PTR(info);
1449         OBD_FREE_PTR(dqblk);
1450
1451         if (rc)
1452                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1453                                 "failed: %d\n",
1454                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1455         RETURN(rc);
1456 }
1457
1458 struct chk_dqblk{
1459         struct hlist_node       dqb_hash;        /* quotacheck hash */
1460         struct list_head        dqb_list;        /* in list also */
1461         qid_t                   dqb_id;          /* uid/gid */
1462         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1463         __u32                   dqb_bhardlimit;  /* block hard limit */
1464         __u32                   dqb_bsoftlimit;  /* block soft limit */
1465         qsize_t                 dqb_curspace;    /* current space */
1466         __u32                   dqb_ihardlimit;  /* inode hard limit */
1467         __u32                   dqb_isoftlimit;  /* inode soft limit */
1468         __u32                   dqb_curinodes;   /* current inodes */
1469         __u64                   dqb_btime;       /* block grace time */
1470         __u64                   dqb_itime;       /* inode grace time */
1471         __u32                   dqb_valid;       /* flag for above fields */
1472 };
1473
1474 static inline unsigned int chkquot_hash(qid_t id, int type)
1475                                         __attribute__((__const__));
1476
1477 static inline unsigned int chkquot_hash(qid_t id, int type)
1478 {
1479         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1480 }
1481
1482 static inline struct chk_dqblk *
1483 find_chkquot(struct hlist_head *head, qid_t id, int type)
1484 {
1485         struct hlist_node *node;
1486         struct chk_dqblk *cdqb;
1487
1488         hlist_for_each(node, head) {
1489                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1490                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1491                         return cdqb;
1492         }
1493
1494         return NULL;
1495 }
1496
1497 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1498 {
1499         struct chk_dqblk *cdqb;
1500
1501         OBD_ALLOC_PTR(cdqb);
1502         if (cdqb) {
1503                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1504                 INIT_LIST_HEAD(&cdqb->dqb_list);
1505                 cdqb->dqb_id = id;
1506                 cdqb->dqb_type = type;
1507         }
1508
1509         return cdqb;
1510 }
1511
1512 static struct chk_dqblk *
1513 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1514       qid_t id, int type, int first_check)
1515 {
1516         struct hlist_head *head = hash + chkquot_hash(id, type);
1517         struct if_dqblk dqb;
1518         struct chk_dqblk *cdqb;
1519         int rc;
1520
1521         cdqb = find_chkquot(head, id, type);
1522         if (cdqb)
1523                 return cdqb;
1524
1525         cdqb = alloc_chkquot(id, type);
1526         if (!cdqb)
1527                 return NULL;
1528
1529         if (!first_check) {
1530                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1531                 if (rc) {
1532                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1533                                id, type, rc);
1534                 } else {
1535                         DQBLK_COPY(cdqb, &dqb);
1536                         cdqb->dqb_curspace = 0;
1537                         cdqb->dqb_curinodes = 0;
1538                 }
1539         }
1540
1541         hlist_add_head(&cdqb->dqb_hash, head);
1542         list_add_tail(&cdqb->dqb_list, list);
1543
1544         return cdqb;
1545 }
1546
1547 static inline int quota_onoff(struct super_block *sb, int cmd, int type)
1548 {
1549         struct obd_quotactl *oqctl;
1550         int rc;
1551
1552         OBD_ALLOC_PTR(oqctl);
1553         if (!oqctl)
1554                 RETURN(-ENOMEM);
1555
1556         oqctl->qc_cmd = cmd;
1557         oqctl->qc_id = QFMT_LDISKFS;
1558         oqctl->qc_type = type;
1559         rc = fsfilt_ext3_quotactl(sb, oqctl);
1560
1561         OBD_FREE_PTR(oqctl);
1562         return rc;
1563 }
1564
1565 static inline int read_old_dqinfo(struct super_block *sb, int type,
1566                                   struct if_dqinfo *dqinfo)
1567 {
1568         struct obd_quotactl *oqctl;
1569         int rc;
1570         ENTRY;
1571
1572         OBD_ALLOC_PTR(oqctl);
1573         if (!oqctl)
1574                 RETURN(-ENOMEM);
1575
1576         oqctl->qc_cmd = Q_GETINFO;
1577         oqctl->qc_type = type;
1578         rc = fsfilt_ext3_quotactl(sb, oqctl);
1579         if (!rc)
1580                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1581
1582         OBD_FREE_PTR(oqctl);
1583         RETURN(rc);
1584 }
1585
1586 static inline struct ext3_group_desc *
1587 get_group_desc(struct super_block *sb, int group)
1588 {
1589         unsigned long desc_block, desc;
1590         struct ext3_group_desc *gdp;
1591
1592         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1593         desc = group % EXT3_DESC_PER_BLOCK(sb);
1594         gdp = (struct ext3_group_desc *)
1595               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1596
1597         return gdp + desc;
1598 }
1599
1600 static inline struct buffer_head *
1601 read_inode_bitmap(struct super_block *sb, unsigned long group)
1602 {
1603         struct ext3_group_desc *desc;
1604         struct buffer_head *bh;
1605
1606         desc = get_group_desc(sb, group);
1607         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1608
1609         return bh;
1610 }
1611
1612 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1613                                      struct buffer_head *bitmap_bh,
1614                                      int index, unsigned long ino)
1615 {
1616         struct inode *inode = NULL;
1617
1618         if (ext3_test_bit(index, bitmap_bh->b_data))
1619                 inode = iget(sb, ino);
1620
1621         return inode;
1622 }
1623
1624 struct qchk_ctxt {
1625         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1626         struct list_head        qckt_list;                   /* quotacheck list */
1627         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1628         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1629 };
1630
1631 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1632                            struct obd_quotactl *oqc)
1633 {
1634         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1635         loff_t size = 0;
1636         qid_t qid[MAXQUOTAS];
1637         int cnt, i, rc = 0;
1638
1639         if (!inode)
1640                 return 0;
1641
1642         qid[USRQUOTA] = inode->i_uid;
1643         qid[GRPQUOTA] = inode->i_gid;
1644
1645         if (S_ISDIR(inode->i_mode) ||
1646             S_ISREG(inode->i_mode) ||
1647             S_ISLNK(inode->i_mode))
1648                 size = inode_get_bytes(inode);
1649
1650         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1651                 if (!Q_TYPESET(oqc, cnt))
1652                         continue;
1653
1654                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1655                                 &qctxt->qckt_list, qid[cnt], cnt,
1656                                 qctxt->qckt_first_check[cnt]);
1657                 if (!cdqb[cnt]) {
1658                         rc = -ENOMEM;
1659                         break;
1660                 }
1661
1662                 cdqb[cnt]->dqb_curspace += size;
1663                 cdqb[cnt]->dqb_curinodes++;
1664         }
1665
1666         if (rc) {
1667                 for (i = 0; i < cnt; i++) {
1668                         if (!Q_TYPESET(oqc, i))
1669                                 continue;
1670                         LASSERT(cdqb[i]);
1671                         cdqb[i]->dqb_curspace -= size;
1672                         cdqb[i]->dqb_curinodes--;
1673                 }
1674         }
1675
1676         return rc;
1677 }
1678
1679 static int v2_write_dqheader(struct file *f, int type)
1680 {
1681         static const __u32 quota_magics[] = V2_INITQMAGICS;
1682         static const __u32 quota_versions[] = V2_INITQVERSIONS;
1683         struct v2_disk_dqheader dqhead;
1684         loff_t offset = 0;
1685
1686         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1687         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1688
1689         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1690         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1691
1692         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1693 }
1694
1695 /* write dqinfo struct in a new quota file */
1696 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1697 {
1698         struct v2_disk_dqinfo dqinfo;
1699         __u32 blocks = V2_DQTREEOFF + 1;
1700         loff_t offset = V2_DQINFOOFF;
1701
1702         if (info) {
1703                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1704                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1705                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1706                                                ~DQF_INFO_DIRTY);
1707         } else {
1708                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1709                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1710                 dqinfo.dqi_flags = 0;
1711         }
1712
1713         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1714         dqinfo.dqi_free_blk = 0;
1715         dqinfo.dqi_free_entry = 0;
1716
1717         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1718 }
1719
1720 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1721                                   struct obd_quotactl *oqc)
1722 {
1723         int i, rc = 0;
1724         ENTRY;
1725
1726         for (i = 0; i < MAXQUOTAS; i++) {
1727                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1728                                          NULL : &qctxt->qckt_dqinfo[i];
1729                 struct file *file;
1730
1731                 if (!Q_TYPESET(oqc, i))
1732                         continue;
1733
1734                 file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
1735                                  0644);
1736                 if (IS_ERR(file)) {
1737                         rc = PTR_ERR(file);
1738                         CERROR("can't create %s file: rc = %d\n",
1739                                op_quotafile[i], rc);
1740                         GOTO(out, rc);
1741                 }
1742
1743                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1744                         CERROR("file %s is not regular", op_quotafile[i]);
1745                         filp_close(file, 0);
1746                         GOTO(out, rc = -EINVAL);
1747                 }
1748
1749                 rc = v2_write_dqheader(file, i);
1750                 if (rc) {
1751                         filp_close(file, 0);
1752                         GOTO(out, rc);
1753                 }
1754
1755                 rc = v2_write_dqinfo(file, i, info);
1756                 filp_close(file, 0);
1757                 if (rc)
1758                         GOTO(out, rc);
1759         }
1760
1761 out:
1762         RETURN(rc);
1763 }
1764
1765
1766 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1767                           struct chk_dqblk *cdqb)
1768 {
1769         struct obd_quotactl *oqc;
1770         long now;
1771         int rc;
1772         ENTRY;
1773
1774         OBD_ALLOC_PTR(oqc);
1775         if (!oqc)
1776                 RETURN(-ENOMEM);
1777
1778         now = CURRENT_SECONDS;
1779
1780         if (cdqb->dqb_bsoftlimit &&
1781             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1782             !cdqb->dqb_btime)
1783                 cdqb->dqb_btime =
1784                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1785
1786         if (cdqb->dqb_isoftlimit &&
1787             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1788             !cdqb->dqb_itime)
1789                 cdqb->dqb_itime =
1790                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1791
1792         cdqb->dqb_valid = QIF_ALL;
1793
1794         oqc->qc_cmd = Q_SETQUOTA;
1795         oqc->qc_type = cdqb->dqb_type;
1796         oqc->qc_id = cdqb->dqb_id;
1797         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1798
1799         rc = fsfilt_ext3_quotactl(sb, oqc);
1800         OBD_FREE_PTR(oqc);
1801         RETURN(rc);
1802 }
1803
1804 static int prune_chkquots(struct super_block *sb,
1805                           struct qchk_ctxt *qctxt, int error)
1806 {
1807         struct chk_dqblk *cdqb, *tmp;
1808         int rc;
1809
1810         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1811                 if (!error) {
1812                         rc = commit_chkquot(sb, qctxt, cdqb);
1813                         if (rc)
1814                                 error = rc;
1815                 }
1816                 hlist_del_init(&cdqb->dqb_hash);
1817                 list_del(&cdqb->dqb_list);
1818                 OBD_FREE_PTR(cdqb);
1819         }
1820
1821         return error;
1822 }
1823
1824 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1825                                   struct obd_quotactl *oqc)
1826 {
1827         struct ext3_sb_info *sbi = EXT3_SB(sb);
1828         int i, group;
1829         struct qchk_ctxt *qctxt;
1830         struct buffer_head *bitmap_bh = NULL;
1831         unsigned long ino;
1832         struct inode *inode;
1833         int rc = 0;
1834         ENTRY;
1835
1836         /* turn on quota and read dqinfo if existed */
1837         OBD_ALLOC_PTR(qctxt);
1838         if (!qctxt) {
1839                 oqc->qc_stat = -ENOMEM;
1840                 RETURN(-ENOMEM);
1841         }
1842
1843         for (i = 0; i < NR_DQHASH; i++)
1844                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1845         INIT_LIST_HEAD(&qctxt->qckt_list);
1846
1847         for (i = 0; i < MAXQUOTAS; i++) {
1848                 if (!Q_TYPESET(oqc, i))
1849                         continue;
1850
1851                 rc = quota_onoff(sb, Q_QUOTAON, i);
1852                 if (!rc || rc == -EBUSY) {
1853                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1854                         if (rc)
1855                                 GOTO(out, rc);
1856                 } else if (rc == -ENOENT) {
1857                         qctxt->qckt_first_check[i] = 1;
1858                 } else if (rc) {
1859                         GOTO(out, rc);
1860                 }
1861         }
1862
1863         /* check quota and update in hash */
1864         for (group = 0; group < sbi->s_groups_count; group++) {
1865                 ino = group * sbi->s_inodes_per_group + 1;
1866                 bitmap_bh = read_inode_bitmap(sb, group);
1867                 if (!bitmap_bh) {
1868                         CERROR("read_inode_bitmap group %d failed", group);
1869                         GOTO(out, rc = -EIO);
1870                 }
1871
1872                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1873                         if (ino < sbi->s_first_ino)
1874                                 continue;
1875
1876                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1877                         rc = add_inode_quota(inode, qctxt, oqc);
1878                         iput(inode);
1879                         if (rc) {
1880                                 brelse(bitmap_bh);
1881                                 GOTO(out, rc);
1882                         }
1883                 }
1884
1885                 brelse(bitmap_bh);
1886         }
1887
1888         /* read old quota limits from old quota file. (only for the user
1889          * has limits but hasn't file) */
1890 #ifdef HAVE_QUOTA_SUPPORT
1891         for (i = 0; i < MAXQUOTAS; i++) {
1892                 struct list_head id_list;
1893                 struct dquot_id *dqid, *tmp;
1894
1895                 if (!Q_TYPESET(oqc, i))
1896                         continue;
1897
1898                 if (qctxt->qckt_first_check[i])
1899                         continue;
1900
1901
1902                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1903                 INIT_LIST_HEAD(&id_list);
1904 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1905                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1906 #else
1907                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1908 #endif
1909                 if (rc)
1910                         CERROR("read old limits failed. (rc:%d)\n", rc);
1911
1912                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1913                         list_del_init(&dqid->di_link);
1914
1915                         if (!rc)
1916                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1917                                       dqid->di_id, i,
1918                                       qctxt->qckt_first_check[i]);
1919                         kfree(dqid);
1920                 }
1921         }
1922 #endif
1923         /* turn off quota cause we are to dump chk_dqblk to files */
1924         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1925
1926         rc = create_new_quota_files(qctxt, oqc);
1927         if (rc)
1928                 GOTO(out, rc);
1929
1930         /* we use vfs functions to set dqblk, so turn quota on */
1931         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
1932 out:
1933         /* dump and free chk_dqblk */
1934         rc = prune_chkquots(sb, qctxt, rc);
1935         OBD_FREE_PTR(qctxt);
1936
1937         /* turn off quota, `lfs quotacheck` will turn on when all
1938          * nodes quotacheck finish. */
1939         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1940
1941         oqc->qc_stat = rc;
1942         if (rc)
1943                 CERROR("quotacheck failed: rc = %d\n", rc);
1944
1945         RETURN(rc);
1946 }
1947
1948 #ifdef HAVE_QUOTA_SUPPORT
1949 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
1950                                  int cmd)
1951 {
1952         int rc = 0;
1953         ENTRY;
1954
1955         if (lqi->qi_files[type] == NULL) {
1956                 CERROR("operate qinfo before it's enabled!\n");
1957                 RETURN(-EIO);
1958         }
1959
1960         switch (cmd) {
1961         case QFILE_CHK:
1962                 rc = lustre_check_quota_file(lqi, type);
1963                 break;
1964         case QFILE_RD_INFO:
1965                 rc = lustre_read_quota_info(lqi, type);
1966                 break;
1967         case QFILE_WR_INFO:
1968                 rc = lustre_write_quota_info(lqi, type);
1969                 break;
1970         case QFILE_INIT_INFO:
1971                 rc = lustre_init_quota_info(lqi, type);
1972                 break;
1973         default:
1974                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1975                 LBUG();
1976                 break;
1977         }
1978         RETURN(rc);
1979 }
1980
1981 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
1982                             struct list_head *list)
1983 {
1984         return lustre_get_qids(file, inode, type, list);
1985 }
1986
1987 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
1988 {
1989         int rc = 0;
1990         ENTRY;
1991
1992         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
1993                 CERROR("operate dquot before it's enabled!\n");
1994                 RETURN(-EIO);
1995         }
1996
1997         switch (cmd) {
1998         case QFILE_RD_DQUOT:
1999                 rc = lustre_read_dquot(dquot);
2000                 break;
2001         case QFILE_WR_DQUOT:
2002                 if (dquot->dq_dqb.dqb_ihardlimit ||
2003                     dquot->dq_dqb.dqb_isoftlimit ||
2004                     dquot->dq_dqb.dqb_bhardlimit ||
2005                     dquot->dq_dqb.dqb_bsoftlimit)
2006                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2007                 else
2008                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2009
2010                 rc = lustre_commit_dquot(dquot);
2011                 if (rc >= 0)
2012                         rc = 0;
2013                 break;
2014         default:
2015                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2016                 LBUG();
2017                 break;
2018         }
2019         RETURN(rc);
2020 }
2021 #endif
2022
2023 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2024 {
2025         return (EXT3_SB(sb)->journal_bdev);
2026 }
2027 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2028
2029 static struct fsfilt_operations fsfilt_ext3_ops = {
2030         .fs_type                = "ext3",
2031         .fs_owner               = THIS_MODULE,
2032         .fs_getlabel            = fsfilt_ext3_get_label,
2033         .fs_setlabel            = fsfilt_ext3_set_label,
2034         .fs_uuid                = fsfilt_ext3_uuid,
2035         .fs_start               = fsfilt_ext3_start,
2036         .fs_brw_start           = fsfilt_ext3_brw_start,
2037         .fs_extend              = fsfilt_ext3_extend,
2038         .fs_commit              = fsfilt_ext3_commit,
2039         .fs_commit_async        = fsfilt_ext3_commit_async,
2040         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2041         .fs_setattr             = fsfilt_ext3_setattr,
2042         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2043         .fs_set_md              = fsfilt_ext3_set_md,
2044         .fs_get_md              = fsfilt_ext3_get_md,
2045         .fs_readpage            = fsfilt_ext3_readpage,
2046         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2047         .fs_statfs              = fsfilt_ext3_statfs,
2048         .fs_sync                = fsfilt_ext3_sync,
2049         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2050         .fs_write_record        = fsfilt_ext3_write_record,
2051         .fs_read_record         = fsfilt_ext3_read_record,
2052         .fs_setup               = fsfilt_ext3_setup,
2053         .fs_send_bio            = fsfilt_ext3_send_bio,
2054         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2055         .fs_quotactl            = fsfilt_ext3_quotactl,
2056         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2057 #ifdef HAVE_DISK_INODE_VERSION
2058         .fs_get_version         = fsfilt_ext3_get_version,
2059         .fs_set_version         = fsfilt_ext3_set_version,
2060 #endif
2061 #ifdef HAVE_QUOTA_SUPPORT
2062         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2063         .fs_qids                = fsfilt_ext3_qids,
2064         .fs_dquot               = fsfilt_ext3_dquot,
2065 #endif
2066         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2067 };
2068
2069 static int __init fsfilt_ext3_init(void)
2070 {
2071         int rc;
2072
2073         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2074                                          sizeof(struct fsfilt_cb_data), 0, 0);
2075         if (!fcb_cache) {
2076                 CERROR("error allocating fsfilt journal callback cache\n");
2077                 GOTO(out, rc = -ENOMEM);
2078         }
2079
2080         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2081
2082         if (rc) {
2083                 int err = cfs_mem_cache_destroy(fcb_cache);
2084                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2085         }
2086 out:
2087         return rc;
2088 }
2089
2090 static void __exit fsfilt_ext3_exit(void)
2091 {
2092         int rc;
2093
2094         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2095         rc = cfs_mem_cache_destroy(fcb_cache);
2096         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2097 }
2098
2099 module_init(fsfilt_ext3_init);
2100 module_exit(fsfilt_ext3_exit);
2101
2102 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2103 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2104 MODULE_LICENSE("GPL");