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         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 489          * We are already writing the inode to disk as part of this
 490          * transaction and want to avoid a lot of extra inode writeout
 491          * later on. b=9828 */
 492         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 493                 /* ATTR_SIZE would invoke truncate: clear it */
 494                 iattr->ia_valid &= ~ATTR_SIZE;
 495                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 496                 i_size_write(inode, iattr->ia_size);
 497
 498                 if (iattr->ia_valid & ATTR_UID)
 499                         inode->i_uid = iattr->ia_uid;
 500                 if (iattr->ia_valid & ATTR_GID)
 501                         inode->i_gid = iattr->ia_gid;
 502                 if (iattr->ia_valid & ATTR_ATIME)
 503                         inode->i_atime = iattr->ia_atime;
 504                 if (iattr->ia_valid & ATTR_MTIME)
 505                         inode->i_mtime = iattr->ia_mtime;
 506                 if (iattr->ia_valid & ATTR_CTIME)
 507                         inode->i_ctime = iattr->ia_ctime;
 508                 if (iattr->ia_valid & ATTR_MODE) {
 509                         inode->i_mode = iattr->ia_mode;
 510
 511                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 512                                 inode->i_mode &= ~S_ISGID;
 513                 }
 514
 515                 inode->i_sb->s_op->dirty_inode(inode);
 516
 517                 goto out;
 518         }
 519
 520         /* Don't allow setattr to change file type */
 521         if (iattr->ia_valid & ATTR_MODE)
 522                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 523                                  (iattr->ia_mode & ~S_IFMT);
 524
 525         /* We set these flags on the client, but have already checked perms
 526          * so don't confuse inode_change_ok. */
 527         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 528
 529         if (inode->i_op->setattr) {
 530                 rc = inode->i_op->setattr(dentry, iattr);
 531         } else {
 532                 rc = inode_change_ok(inode, iattr);
 533                 if (!rc)
 534                         rc = inode_setattr(inode, iattr);
 535         }
 536
 537  out:
 538         RETURN(rc);
 539 }
 540
 541 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 542                                  unsigned int cmd, unsigned long arg)
 543 {
 544         int rc = 0;
 545         ENTRY;
 546
 547         /* FIXME: Can't do this because of nested transaction deadlock */
 548         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 549                 CERROR("can't set data journal flag on file\n");
 550                 RETURN(-EPERM);
 551         }
 552
 553         if (inode->i_fop->ioctl)
 554                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 555         else
 556                 RETURN(-ENOTTY);
 557
 558         RETURN(rc);
 559 }
 560
 561 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 562                               void *lmm, int lmm_size, const char *name)
 563 {
 564         int rc;
 565
 566         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 567
 568         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 569                                    name, lmm, lmm_size, 0);
 570
 571
 572         if (rc && rc != -EROFS)
 573                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 574                        inode->i_ino, rc);
 575         return rc;
 576 }
 577
 578 /* Must be called with i_mutex held */
 579 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 580                               const char *name)
 581 {
 582         int rc;
 583
 584         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 585
 586         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 587                             name, lmm, lmm_size);
 588
 589         /* This gives us the MD size */
 590         if (lmm == NULL)
 591                 return (rc == -ENODATA) ? 0 : rc;
 592
 593         if (rc < 0) {
 594                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 595                        EXT3_XATTR_INDEX_TRUSTED, name,
 596                        inode->i_ino, rc);
 597                 memset(lmm, 0, lmm_size);
 598                 return (rc == -ENODATA) ? 0 : rc;
 599         }
 600
 601         return rc;
 602 }
 603
 604 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 605 {
 606         submit_bio(rw, bio);
 607         return 0;
 608 }
 609
 610 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 611                                     loff_t *off)
 612 {
 613         struct inode *inode = file->f_dentry->d_inode;
 614         int rc = 0;
 615
 616         if (S_ISREG(inode->i_mode))
 617                 rc = file->f_op->read(file, buf, count, off);
 618         else {
 619                 const int blkbits = inode->i_sb->s_blocksize_bits;
 620                 const int blksize = inode->i_sb->s_blocksize;
 621
 622                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 623                        count, inode->i_ino, *off);
 624                 while (count > 0) {
 625                         struct buffer_head *bh;
 626
 627                         bh = NULL;
 628                         if (*off < i_size_read(inode)) {
 629                                 int err = 0;
 630
 631                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 632                                                 0, &err);
 633
 634                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 635
 636                                 if (bh) {
 637                                         memcpy(buf, bh->b_data, blksize);
 638                                         brelse(bh);
 639                                 } else if (err) {
 640                                         /* XXX in theory we should just fake
 641                                          * this buffer and continue like ext3,
 642                                          * especially if this is a partial read
 643                                          */
 644                                         CERROR("error read dir %lu+%llu: %d\n",
 645                                                inode->i_ino, *off, err);
 646                                         RETURN(err);
 647                                 }
 648                         }
 649                         if (!bh) {
 650                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 651
 652                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 653                                 memset(fake, 0, sizeof(*fake));
 654                                 fake->rec_len = cpu_to_le16(blksize);
 655                         }
 656                         count -= blksize;
 657                         buf += blksize;
 658                         *off += blksize;
 659                         rc += blksize;
 660                 }
 661         }
 662
 663         return rc;
 664 }
 665
 666 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 667 {
 668         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 669
 670         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 671
 672         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 673 }
 674
 675 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 676                                       void *handle, fsfilt_cb_t cb_func,
 677                                       void *cb_data)
 678 {
 679         struct fsfilt_cb_data *fcb;
 680
 681         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 682         if (fcb == NULL)
 683                 RETURN(-ENOMEM);
 684
 685         fcb->cb_func = cb_func;
 686         fcb->cb_obd = obd;
 687         fcb->cb_last_rcvd = last_rcvd;
 688         fcb->cb_data = cb_data;
 689
 690         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 691         journal_callback_set(handle, fsfilt_ext3_cb_func,
 692                              (struct journal_callback *)fcb);
 693
 694         return 0;
 695 }
 696
 697 /*
 698  * We need to hack the return value for the free inode counts because
 699  * the current EA code requires one filesystem block per inode with EAs,
 700  * so it is possible to run out of blocks before we run out of inodes.
 701  *
 702  * This can be removed when the ext3 EA code is fixed.
 703  */
 704 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 705 {
 706         struct kstatfs sfs;
 707         int rc;
 708
 709         memset(&sfs, 0, sizeof(sfs));
 710
 711         rc = ll_do_statfs(sb, &sfs);
 712
 713         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 714                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 715                 sfs.f_ffree = sfs.f_bfree;
 716         }
 717
 718         statfs_pack(osfs, &sfs);
 719         return rc;
 720 }
 721
 722 static int fsfilt_ext3_sync(struct super_block *sb)
 723 {
 724         return ext3_force_commit(sb);
 725 }
 726
 727 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 728 #warning "kernel code has old extents/mballoc patch, disabling"
 729 #undef EXT3_MULTIBLOCK_ALLOCATOR
 730 #endif
 731 #ifndef EXT3_EXTENTS_FL
 732 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 733 #endif
 734
 735 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 736 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 737 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 738 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 739 #else
 740 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 741 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 742 #endif
 743
 744 #ifndef EXT_ASSERT
 745 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 746 #endif
 747
 748 #ifdef EXT3_EXT_HAS_NO_TREE
 749 /* for kernels 2.6.18 and later */
 750 #define ext3_ext_base                   inode
 751 #define ext3_ext_base2inode(inode)      (inode)
 752 #define EXT_DEPTH(inode)                ext_depth(inode)
 753 #define EXT_GENERATION(inode)           ext_generation(inode)
 754 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 755                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 756 #else
 757 #define ext3_ext_base                   ext3_extents_tree
 758 #define ext3_ext_base2inode(tree)       (tree->inode)
 759 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 760                         ext3_ext_walk_space(tree, block, num, cb);
 761 #endif
 762
 763 #include <linux/lustre_version.h>
 764 #if EXT3_EXT_MAGIC == 0xf301
 765 #define ee_start e_start
 766 #define ee_block e_block
 767 #define ee_len   e_num
 768 #endif
 769 #ifndef EXT3_BB_MAX_BLOCKS
 770 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 771         ext3_new_blocks(handle, inode, count, goal, err)
 772 #endif
 773
 774 struct bpointers {
 775         unsigned long *blocks;
 776         int *created;
 777         unsigned long start;
 778         int num;
 779         int init_num;
 780         int create;
 781 };
 782
 783 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 784                               unsigned long block, int *aflags)
 785 {
 786         struct ext3_inode_info *ei = EXT3_I(inode);
 787         unsigned long bg_start;
 788         unsigned long colour;
 789         int depth;
 790
 791         if (path) {
 792                 struct ext3_extent *ex;
 793                 depth = path->p_depth;
 794
 795                 /* try to predict block placement */
 796                 if ((ex = path[depth].p_ext)) {
 797 #if 0
 798                         /* This prefers to eat into a contiguous extent
 799                          * rather than find an extent that the whole
 800                          * request will fit into.  This can fragment data
 801                          * block allocation and prevents our lovely 1M I/Os
 802                          * from reaching the disk intact. */
 803                         if (ex->ee_block + ex->ee_len == block)
 804                                 *aflags |= 1;
 805 #endif
 806                         return ex->ee_start + (block - ex->ee_block);
 807                 }
 808
 809                 /* it looks index is empty
 810                  * try to find starting from index itself */
 811                 if (path[depth].p_bh)
 812                         return path[depth].p_bh->b_blocknr;
 813         }
 814
 815         /* OK. use inode's group */
 816         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 817                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 818         colour = (current->pid % 16) *
 819                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 820         return bg_start + colour + block;
 821 }
 822
 823 #define ll_unmap_underlying_metadata(sb, blocknr) \
 824         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 825
 826 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 827 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 828                                 struct ext3_ext_path *path, unsigned long block,
 829                                 unsigned long *count, int *err)
 830 {
 831         unsigned long pblock, goal;
 832         int aflags = 0;
 833         struct inode *inode = ext3_ext_base2inode(base);
 834
 835         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 836         aflags |= 2; /* block have been already reserved */
 837         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 838         return pblock;
 839
 840 }
 841 #else
 842 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 843                                 struct ext3_ext_path *path, unsigned long block,
 844                                 unsigned long *count, int *err)
 845 {
 846         struct inode *inode = ext3_ext_base2inode(base);
 847         struct ext3_allocation_request ar;
 848         unsigned long pblock;
 849         int aflags;
 850
 851         /* find neighbour allocated blocks */
 852         ar.lleft = block;
 853         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 854         if (*err)
 855                 return 0;
 856         ar.lright = block;
 857         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 858         if (*err)
 859                 return 0;
 860
 861         /* allocate new block */
 862         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 863         ar.inode = inode;
 864         ar.logical = block;
 865         ar.len = *count;
 866         ar.flags = EXT3_MB_HINT_DATA;
 867         pblock = ext3_mb_new_blocks(handle, &ar, err);
 868         *count = ar.len;
 869         return pblock;
 870
 871 }
 872 #endif
 873
 874 #ifdef EXT3_EXT_HAS_NO_TREE
 875 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 876                                   struct ext3_ext_path *path,
 877                                   struct ext3_ext_cache *cex,
 878                                   void *cbdata)
 879 {
 880         struct bpointers *bp = cbdata;
 881 #else
 882 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 883                                   struct ext3_ext_path *path,
 884                                   struct ext3_ext_cache *cex)
 885 {
 886         struct bpointers *bp = base->private;
 887 #endif
 888         struct inode *inode = ext3_ext_base2inode(base);
 889         struct ext3_extent nex;
 890         unsigned long pblock;
 891         unsigned long tgen;
 892         int err, i;
 893         unsigned long count;
 894         handle_t *handle;
 895
 896         i = EXT_DEPTH(base);
 897         EXT_ASSERT(i == path->p_depth);
 898         EXT_ASSERT(path[i].p_hdr);
 899
 900         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 901                 err = EXT_CONTINUE;
 902                 goto map;
 903         }
 904
 905         if (bp->create == 0) {
 906                 i = 0;
 907                 if (cex->ec_block < bp->start)
 908                         i = bp->start - cex->ec_block;
 909                 if (i >= cex->ec_len)
 910                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 911                                         i, cex->ec_len);
 912                 for (; i < cex->ec_len && bp->num; i++) {
 913                         *(bp->created) = 0;
 914                         bp->created++;
 915                         *(bp->blocks) = 0;
 916                         bp->blocks++;
 917                         bp->num--;
 918                         bp->start++;
 919                 }
 920
 921                 return EXT_CONTINUE;
 922         }
 923
 924         tgen = EXT_GENERATION(base);
 925         count = ext3_ext_calc_credits_for_insert(base, path);
 926         ext3_up_truncate_sem(inode);
 927
 928         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 929         if (IS_ERR(handle)) {
 930                 ext3_down_truncate_sem(inode);
 931                 return PTR_ERR(handle);
 932         }
 933
 934         ext3_down_truncate_sem(inode);
 935         if (tgen != EXT_GENERATION(base)) {
 936                 /* the tree has changed. so path can be invalid at moment */
 937                 fsfilt_ext3_journal_stop(handle);
 938                 return EXT_REPEAT;
 939         }
 940
 941         count = cex->ec_len;
 942         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 943         if (!pblock)
 944                 goto out;
 945         EXT_ASSERT(count <= cex->ec_len);
 946
 947         /* insert new extent */
 948         nex.ee_block = cex->ec_block;
 949         nex.ee_start = pblock;
 950         nex.ee_len = count;
 951         err = ext3_ext_insert_extent(handle, base, path, &nex);
 952         if (err) {
 953                 CERROR("can't insert extent: %d\n", err);
 954                 /* XXX: export ext3_free_blocks() */
 955                 /*ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);*/
 956                 goto out;
 957         }
 958
 959         /*
 960          * Putting len of the actual extent we just inserted,
 961          * we are asking ext3_ext_walk_space() to continue
 962          * scaning after that block
 963          */
 964         cex->ec_len = nex.ee_len;
 965         cex->ec_start = nex.ee_start;
 966         BUG_ON(nex.ee_len == 0);
 967         BUG_ON(nex.ee_block != cex->ec_block);
 968
 969 out:
 970         fsfilt_ext3_journal_stop(handle);
 971 map:
 972         if (err >= 0) {
 973                 /* map blocks */
 974                 if (bp->num == 0) {
 975                         CERROR("hmm. why do we find this extent?\n");
 976                         CERROR("initial space: %lu:%u\n",
 977                                 bp->start, bp->init_num);
 978                         CERROR("current extent: %u/%u/%u %d\n",
 979                                 cex->ec_block, cex->ec_len,
 980                                 cex->ec_start, cex->ec_type);
 981                 }
 982                 i = 0;
 983                 if (cex->ec_block < bp->start)
 984                         i = bp->start - cex->ec_block;
 985                 if (i >= cex->ec_len)
 986                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 987                                         i, cex->ec_len);
 988                 for (; i < cex->ec_len && bp->num; i++) {
 989                         *(bp->blocks) = cex->ec_start + i;
 990                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 991                                 *(bp->created) = 0;
 992                         } else {
 993                                 *(bp->created) = 1;
 994                                 /* unmap any possible underlying metadata from
 995                                  * the block device mapping.  bug 6998. */
 996                                 ll_unmap_underlying_metadata(inode->i_sb,
 997                                                              *(bp->blocks));
 998                         }
 999                         bp->created++;
1000                         bp->blocks++;
1001                         bp->num--;
1002                         bp->start++;
1003                 }
1004         }
1005         return err;
1006 }
1007
1008 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1009                        unsigned long num, unsigned long *blocks,
1010                        int *created, int create)
1011 {
1012 #ifdef EXT3_EXT_HAS_NO_TREE
1013         struct ext3_ext_base *base = inode;
1014 #else
1015         struct ext3_extents_tree tree;
1016         struct ext3_ext_base *base = &tree;
1017 #endif
1018         struct bpointers bp;
1019         int err;
1020
1021         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1022                block, block + num - 1, (unsigned) inode->i_ino);
1023
1024 #ifndef EXT3_EXT_HAS_NO_TREE
1025         ext3_init_tree_desc(base, inode);
1026         tree.private = &bp;
1027 #endif
1028         bp.blocks = blocks;
1029         bp.created = created;
1030         bp.start = block;
1031         bp.init_num = bp.num = num;
1032         bp.create = create;
1033
1034         ext3_down_truncate_sem(inode);
1035         err = fsfilt_ext3_ext_walk_space(base, block, num,
1036                                          ext3_ext_new_extent_cb, &bp);
1037         ext3_ext_invalidate_cache(base);
1038         ext3_up_truncate_sem(inode);
1039
1040         return err;
1041 }
1042
1043 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1044                                     int pages, unsigned long *blocks,
1045                                     int *created, int create)
1046 {
1047         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1048         int rc = 0, i = 0;
1049         struct page *fp = NULL;
1050         int clen = 0;
1051
1052         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1053                 inode->i_ino, pages, (*page)->index);
1054
1055         /* pages are sorted already. so, we just have to find
1056          * contig. space and process them properly */
1057         while (i < pages) {
1058                 if (fp == NULL) {
1059                         /* start new extent */
1060                         fp = *page++;
1061                         clen = 1;
1062                         i++;
1063                         continue;
1064                 } else if (fp->index + clen == (*page)->index) {
1065                         /* continue the extent */
1066                         page++;
1067                         clen++;
1068                         i++;
1069                         continue;
1070                 }
1071
1072                 /* process found extent */
1073                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1074                                         clen * blocks_per_page, blocks,
1075                                         created, create);
1076                 if (rc)
1077                         GOTO(cleanup, rc);
1078
1079                 /* look for next extent */
1080                 fp = NULL;
1081                 blocks += blocks_per_page * clen;
1082                 created += blocks_per_page * clen;
1083         }
1084
1085         if (fp)
1086                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1087                                         clen * blocks_per_page, blocks,
1088                                         created, create);
1089 cleanup:
1090         return rc;
1091 }
1092 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1093
1094 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1095                                unsigned long *blocks, int *created, int create);
1096 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1097                                    int pages, unsigned long *blocks,
1098                                    int *created, int create)
1099 {
1100         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1101         unsigned long *b;
1102         int rc = 0, i, *cr;
1103
1104         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1105                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1106                 if (rc) {
1107                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1108                                inode->i_ino, *b, *cr, create, rc);
1109                         break;
1110                 }
1111
1112                 b += blocks_per_page;
1113                 cr += blocks_per_page;
1114         }
1115         return rc;
1116 }
1117
1118 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1119                                 int pages, unsigned long *blocks,
1120                                 int *created, int create,
1121                                 struct semaphore *optional_sem)
1122 {
1123         int rc;
1124 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1125         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1126                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1127                                                      blocks, created, create);
1128                 return rc;
1129         }
1130 #endif
1131         if (optional_sem != NULL)
1132                 down(optional_sem);
1133         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1134                                             created, create);
1135         if (optional_sem != NULL)
1136                 up(optional_sem);
1137
1138         return rc;
1139 }
1140
1141 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1142 {
1143         unsigned long block;
1144         struct buffer_head *bh;
1145         int err, blocksize, csize, boffs, osize = size;
1146
1147         /* prevent reading after eof */
1148         lock_kernel();
1149         if (i_size_read(inode) < *offs + size) {
1150                 size = i_size_read(inode) - *offs;
1151                 unlock_kernel();
1152                 if (size < 0) {
1153                         CERROR("size %llu is too short for read %u@%llu\n",
1154                                i_size_read(inode), size, *offs);
1155                         return -EIO;
1156                 } else if (size == 0) {
1157                         return 0;
1158                 }
1159         } else {
1160                 unlock_kernel();
1161         }
1162
1163         blocksize = 1 << inode->i_blkbits;
1164
1165         while (size > 0) {
1166                 block = *offs >> inode->i_blkbits;
1167                 boffs = *offs & (blocksize - 1);
1168                 csize = min(blocksize - boffs, size);
1169                 bh = ext3_bread(NULL, inode, block, 0, &err);
1170                 if (!bh) {
1171                         CERROR("can't read block: %d\n", err);
1172                         return err;
1173                 }
1174
1175                 memcpy(buf, bh->b_data + boffs, csize);
1176                 brelse(bh);
1177
1178                 *offs += csize;
1179                 buf += csize;
1180                 size -= csize;
1181         }
1182         return osize;
1183 }
1184 EXPORT_SYMBOL(fsfilt_ext3_read);
1185
1186 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1187                                    int size, loff_t *offs)
1188 {
1189         int rc;
1190         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1191         if (rc > 0)
1192                 rc = 0;
1193         return rc;
1194 }
1195
1196 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1197                                 loff_t *offs, handle_t *handle)
1198 {
1199         struct buffer_head *bh = NULL;
1200         loff_t old_size = i_size_read(inode), offset = *offs;
1201         loff_t new_size = i_size_read(inode);
1202         unsigned long block;
1203         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1204
1205         while (bufsize > 0) {
1206                 if (bh != NULL)
1207                         brelse(bh);
1208
1209                 block = offset >> inode->i_blkbits;
1210                 boffs = offset & (blocksize - 1);
1211                 size = min(blocksize - boffs, bufsize);
1212                 bh = ext3_bread(handle, inode, block, 1, &err);
1213                 if (!bh) {
1214                         CERROR("can't read/create block: %d\n", err);
1215                         break;
1216                 }
1217
1218                 err = ext3_journal_get_write_access(handle, bh);
1219                 if (err) {
1220                         CERROR("journal_get_write_access() returned error %d\n",
1221                                err);
1222                         break;
1223                 }
1224                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1225                 memcpy(bh->b_data + boffs, buf, size);
1226                 err = ext3_journal_dirty_metadata(handle, bh);
1227                 if (err) {
1228                         CERROR("journal_dirty_metadata() returned error %d\n",
1229                                err);
1230                         break;
1231                 }
1232                 if (offset + size > new_size)
1233                         new_size = offset + size;
1234                 offset += size;
1235                 bufsize -= size;
1236                 buf += size;
1237         }
1238         if (bh)
1239                 brelse(bh);
1240
1241         /* correct in-core and on-disk sizes */
1242         if (new_size > i_size_read(inode)) {
1243                 lock_kernel();
1244                 if (new_size > i_size_read(inode))
1245                         i_size_write(inode, new_size);
1246                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1247                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1248                 if (i_size_read(inode) > old_size)
1249                         mark_inode_dirty(inode);
1250                 unlock_kernel();
1251         }
1252
1253         if (err == 0)
1254                 *offs = offset;
1255         return err;
1256 }
1257 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1258
1259 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1260                                     loff_t *offs, int force_sync)
1261 {
1262         struct inode *inode = file->f_dentry->d_inode;
1263         handle_t *handle;
1264         int err, block_count = 0, blocksize;
1265
1266         /* Determine how many transaction credits are needed */
1267         blocksize = 1 << inode->i_blkbits;
1268         block_count = (*offs & (blocksize - 1)) + bufsize;
1269         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1270
1271         handle = fsfilt_ext3_journal_start(inode,
1272                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1273         if (IS_ERR(handle)) {
1274                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1275                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1276                 return PTR_ERR(handle);
1277         }
1278
1279         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1280
1281         if (!err && force_sync)
1282                 handle->h_sync = 1; /* recovery likes this */
1283
1284         fsfilt_ext3_journal_stop(handle);
1285
1286         return err;
1287 }
1288
1289 static int fsfilt_ext3_setup(struct super_block *sb)
1290 {
1291 #if 0
1292         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1293         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1294 #endif
1295 #ifdef S_PDIROPS
1296         CWARN("Enabling PDIROPS\n");
1297         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1298         sb->s_flags |= S_PDIROPS;
1299 #endif
1300         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1301                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1302 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13)) && HAVE_QUOTA_SUPPORT
1303         set_opt(EXT3_SB(sb)->s_mount_opt, QUOTA);
1304 #endif
1305         return 0;
1306 }
1307
1308 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1309    objects. Logs is number of logfiles to update */
1310 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1311 {
1312         if ( !fso ) {
1313                 switch(op) {
1314                 case FSFILT_OP_CREATE:
1315                                  /* directory leaf, index & indirect & EA*/
1316                         return 4 + 3 * logs;
1317                 case FSFILT_OP_UNLINK:
1318                         return 3 * logs;
1319                 }
1320         } else {
1321                 int i;
1322                 int needed = 0;
1323                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1324                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1325                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1326                 for (i = 0; i < op; i++, fso++) {
1327                         int nblocks = fso->fso_bufcnt * blockpp;
1328                         int ndindirect = min(nblocks, addrpp + 1);
1329                         int nindir = nblocks + ndindirect + 1;
1330
1331                         needed += nindir;
1332                 }
1333                 return needed + 3 * logs;
1334         }
1335
1336         return 0;
1337 }
1338
1339 static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
1340
1341 #define DQINFO_COPY(out, in)                    \
1342 do {                                            \
1343         Q_COPY(out, in, dqi_bgrace);            \
1344         Q_COPY(out, in, dqi_igrace);            \
1345         Q_COPY(out, in, dqi_flags);             \
1346         Q_COPY(out, in, dqi_valid);             \
1347 } while (0)
1348
1349 #define DQBLK_COPY(out, in)                     \
1350 do {                                            \
1351         Q_COPY(out, in, dqb_bhardlimit);        \
1352         Q_COPY(out, in, dqb_bsoftlimit);        \
1353         Q_COPY(out, in, dqb_curspace);          \
1354         Q_COPY(out, in, dqb_ihardlimit);        \
1355         Q_COPY(out, in, dqb_isoftlimit);        \
1356         Q_COPY(out, in, dqb_curinodes);         \
1357         Q_COPY(out, in, dqb_btime);             \
1358         Q_COPY(out, in, dqb_itime);             \
1359         Q_COPY(out, in, dqb_valid);             \
1360 } while (0)
1361
1362
1363
1364 static int fsfilt_ext3_quotactl(struct super_block *sb,
1365                                 struct obd_quotactl *oqc)
1366 {
1367         int i, rc = 0, error = 0;
1368         struct quotactl_ops *qcop;
1369         struct if_dqinfo *info;
1370         struct if_dqblk *dqblk;
1371         ENTRY;
1372
1373         if (!sb->s_qcop)
1374                 RETURN(-ENOSYS);
1375
1376         OBD_ALLOC_PTR(info);
1377         if (!info)
1378                 RETURN(-ENOMEM);
1379         OBD_ALLOC_PTR(dqblk);
1380         if (!dqblk) {
1381                 OBD_FREE_PTR(info);
1382                 RETURN(-ENOMEM);
1383         }
1384
1385         DQINFO_COPY(info, &oqc->qc_dqinfo);
1386         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1387
1388         qcop = sb->s_qcop;
1389         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1390                 for (i = 0; i < MAXQUOTAS; i++) {
1391                         if (!Q_TYPESET(oqc, i))
1392                                 continue;
1393
1394                         if (oqc->qc_cmd == Q_QUOTAON) {
1395                                 if (!qcop->quota_on)
1396                                         GOTO(out, rc = -ENOSYS);
1397                                 rc = qcop->quota_on(sb, i, oqc->qc_id,
1398                                                     (char *)op_quotafile[i]);
1399                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1400                                 if (!qcop->quota_off)
1401                                         GOTO(out, rc = -ENOSYS);
1402                                 rc = qcop->quota_off(sb, i);
1403                         }
1404
1405                         if (rc == -EBUSY)
1406                                 error = rc;
1407                         else if (rc)
1408                                 GOTO(out, rc);
1409                 }
1410                 GOTO(out, rc ?: error);
1411         }
1412
1413         switch (oqc->qc_cmd) {
1414         case Q_GETOINFO:
1415         case Q_GETINFO:
1416                 if (!qcop->get_info)
1417                         GOTO(out, rc = -ENOSYS);
1418                 rc = qcop->get_info(sb, oqc->qc_type, info);
1419                 break;
1420         case Q_SETQUOTA:
1421         case Q_INITQUOTA:
1422                 if (!qcop->set_dqblk)
1423                         GOTO(out, rc = -ENOSYS);
1424                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1425                 break;
1426         case Q_GETOQUOTA:
1427         case Q_GETQUOTA:
1428                 if (!qcop->get_dqblk)
1429                         GOTO(out, rc = -ENOSYS);
1430                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1431                 break;
1432         case Q_SYNC:
1433                 if (!sb->s_qcop->quota_sync)
1434                         GOTO(out, rc = -ENOSYS);
1435                 qcop->quota_sync(sb, oqc->qc_type);
1436                 break;
1437         default:
1438                 CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
1439                 LBUG();
1440         }
1441 out:
1442         DQINFO_COPY(&oqc->qc_dqinfo, info);
1443         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1444
1445         OBD_FREE_PTR(info);
1446         OBD_FREE_PTR(dqblk);
1447
1448         if (rc)
1449                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1450                                 "failed: %d\n",
1451                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1452         RETURN(rc);
1453 }
1454
1455 struct chk_dqblk{
1456         struct hlist_node       dqb_hash;        /* quotacheck hash */
1457         struct list_head        dqb_list;        /* in list also */
1458         qid_t                   dqb_id;          /* uid/gid */
1459         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1460         __u32                   dqb_bhardlimit;  /* block hard limit */
1461         __u32                   dqb_bsoftlimit;  /* block soft limit */
1462         qsize_t                 dqb_curspace;    /* current space */
1463         __u32                   dqb_ihardlimit;  /* inode hard limit */
1464         __u32                   dqb_isoftlimit;  /* inode soft limit */
1465         __u32                   dqb_curinodes;   /* current inodes */
1466         __u64                   dqb_btime;       /* block grace time */
1467         __u64                   dqb_itime;       /* inode grace time */
1468         __u32                   dqb_valid;       /* flag for above fields */
1469 };
1470
1471 static inline unsigned int chkquot_hash(qid_t id, int type)
1472                                         __attribute__((__const__));
1473
1474 static inline unsigned int chkquot_hash(qid_t id, int type)
1475 {
1476         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1477 }
1478
1479 static inline struct chk_dqblk *
1480 find_chkquot(struct hlist_head *head, qid_t id, int type)
1481 {
1482         struct hlist_node *node;
1483         struct chk_dqblk *cdqb;
1484
1485         hlist_for_each(node, head) {
1486                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1487                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1488                         return cdqb;
1489         }
1490
1491         return NULL;
1492 }
1493
1494 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1495 {
1496         struct chk_dqblk *cdqb;
1497
1498         OBD_ALLOC_PTR(cdqb);
1499         if (cdqb) {
1500                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1501                 INIT_LIST_HEAD(&cdqb->dqb_list);
1502                 cdqb->dqb_id = id;
1503                 cdqb->dqb_type = type;
1504         }
1505
1506         return cdqb;
1507 }
1508
1509 static struct chk_dqblk *
1510 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1511       qid_t id, int type, int first_check)
1512 {
1513         struct hlist_head *head = hash + chkquot_hash(id, type);
1514         struct if_dqblk dqb;
1515         struct chk_dqblk *cdqb;
1516         int rc;
1517
1518         cdqb = find_chkquot(head, id, type);
1519         if (cdqb)
1520                 return cdqb;
1521
1522         cdqb = alloc_chkquot(id, type);
1523         if (!cdqb)
1524                 return NULL;
1525
1526         if (!first_check) {
1527                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1528                 if (rc) {
1529                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1530                                id, type, rc);
1531                 } else {
1532                         DQBLK_COPY(cdqb, &dqb);
1533                         cdqb->dqb_curspace = 0;
1534                         cdqb->dqb_curinodes = 0;
1535                 }
1536         }
1537
1538         hlist_add_head(&cdqb->dqb_hash, head);
1539         list_add_tail(&cdqb->dqb_list, list);
1540
1541         return cdqb;
1542 }
1543
1544 static inline int quota_onoff(struct super_block *sb, int cmd, int type)
1545 {
1546         struct obd_quotactl *oqctl;
1547         int rc;
1548
1549         OBD_ALLOC_PTR(oqctl);
1550         if (!oqctl)
1551                 RETURN(-ENOMEM);
1552
1553         oqctl->qc_cmd = cmd;
1554         oqctl->qc_id = QFMT_LDISKFS;
1555         oqctl->qc_type = type;
1556         rc = fsfilt_ext3_quotactl(sb, oqctl);
1557
1558         OBD_FREE_PTR(oqctl);
1559         return rc;
1560 }
1561
1562 static inline int read_old_dqinfo(struct super_block *sb, int type,
1563                                   struct if_dqinfo *dqinfo)
1564 {
1565         struct obd_quotactl *oqctl;
1566         int rc;
1567         ENTRY;
1568
1569         OBD_ALLOC_PTR(oqctl);
1570         if (!oqctl)
1571                 RETURN(-ENOMEM);
1572
1573         oqctl->qc_cmd = Q_GETINFO;
1574         oqctl->qc_type = type;
1575         rc = fsfilt_ext3_quotactl(sb, oqctl);
1576         if (!rc)
1577                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1578
1579         OBD_FREE_PTR(oqctl);
1580         RETURN(rc);
1581 }
1582
1583 static inline struct ext3_group_desc *
1584 get_group_desc(struct super_block *sb, int group)
1585 {
1586         unsigned long desc_block, desc;
1587         struct ext3_group_desc *gdp;
1588
1589         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1590         desc = group % EXT3_DESC_PER_BLOCK(sb);
1591         gdp = (struct ext3_group_desc *)
1592               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1593
1594         return gdp + desc;
1595 }
1596
1597 static inline struct buffer_head *
1598 read_inode_bitmap(struct super_block *sb, unsigned long group)
1599 {
1600         struct ext3_group_desc *desc;
1601         struct buffer_head *bh;
1602
1603         desc = get_group_desc(sb, group);
1604         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1605
1606         return bh;
1607 }
1608
1609 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1610                                      struct buffer_head *bitmap_bh,
1611                                      int index, unsigned long ino)
1612 {
1613         struct inode *inode = NULL;
1614
1615         if (ext3_test_bit(index, bitmap_bh->b_data))
1616                 inode = iget(sb, ino);
1617
1618         return inode;
1619 }
1620
1621 struct qchk_ctxt {
1622         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1623         struct list_head        qckt_list;                   /* quotacheck list */
1624         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1625         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1626 };
1627
1628 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1629                            struct obd_quotactl *oqc)
1630 {
1631         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1632         loff_t size = 0;
1633         qid_t qid[MAXQUOTAS];
1634         int cnt, i, rc = 0;
1635
1636         if (!inode)
1637                 return 0;
1638
1639         qid[USRQUOTA] = inode->i_uid;
1640         qid[GRPQUOTA] = inode->i_gid;
1641
1642         if (S_ISDIR(inode->i_mode) ||
1643             S_ISREG(inode->i_mode) ||
1644             S_ISLNK(inode->i_mode))
1645                 size = inode_get_bytes(inode);
1646
1647         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1648                 if (!Q_TYPESET(oqc, cnt))
1649                         continue;
1650
1651                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1652                                 &qctxt->qckt_list, qid[cnt], cnt,
1653                                 qctxt->qckt_first_check[cnt]);
1654                 if (!cdqb[cnt]) {
1655                         rc = -ENOMEM;
1656                         break;
1657                 }
1658
1659                 cdqb[cnt]->dqb_curspace += size;
1660                 cdqb[cnt]->dqb_curinodes++;
1661         }
1662
1663         if (rc) {
1664                 for (i = 0; i < cnt; i++) {
1665                         if (!Q_TYPESET(oqc, i))
1666                                 continue;
1667                         LASSERT(cdqb[i]);
1668                         cdqb[i]->dqb_curspace -= size;
1669                         cdqb[i]->dqb_curinodes--;
1670                 }
1671         }
1672
1673         return rc;
1674 }
1675
1676 static int v2_write_dqheader(struct file *f, int type)
1677 {
1678         static const __u32 quota_magics[] = V2_INITQMAGICS;
1679         static const __u32 quota_versions[] = V2_INITQVERSIONS;
1680         struct v2_disk_dqheader dqhead;
1681         loff_t offset = 0;
1682
1683         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1684         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1685
1686         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1687         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1688
1689         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1690 }
1691
1692 /* write dqinfo struct in a new quota file */
1693 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1694 {
1695         struct v2_disk_dqinfo dqinfo;
1696         __u32 blocks = V2_DQTREEOFF + 1;
1697         loff_t offset = V2_DQINFOOFF;
1698
1699         if (info) {
1700                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1701                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1702                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1703                                                ~DQF_INFO_DIRTY);
1704         } else {
1705                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1706                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1707                 dqinfo.dqi_flags = 0;
1708         }
1709
1710         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1711         dqinfo.dqi_free_blk = 0;
1712         dqinfo.dqi_free_entry = 0;
1713
1714         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1715 }
1716
1717 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1718                                   struct obd_quotactl *oqc)
1719 {
1720         int i, rc = 0;
1721         ENTRY;
1722
1723         for (i = 0; i < MAXQUOTAS; i++) {
1724                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1725                                          NULL : &qctxt->qckt_dqinfo[i];
1726                 struct file *file;
1727
1728                 if (!Q_TYPESET(oqc, i))
1729                         continue;
1730
1731                 file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
1732                                  0644);
1733                 if (IS_ERR(file)) {
1734                         rc = PTR_ERR(file);
1735                         CERROR("can't create %s file: rc = %d\n",
1736                                op_quotafile[i], rc);
1737                         GOTO(out, rc);
1738                 }
1739
1740                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1741                         CERROR("file %s is not regular", op_quotafile[i]);
1742                         filp_close(file, 0);
1743                         GOTO(out, rc = -EINVAL);
1744                 }
1745
1746                 rc = v2_write_dqheader(file, i);
1747                 if (rc) {
1748                         filp_close(file, 0);
1749                         GOTO(out, rc);
1750                 }
1751
1752                 rc = v2_write_dqinfo(file, i, info);
1753                 filp_close(file, 0);
1754                 if (rc)
1755                         GOTO(out, rc);
1756         }
1757
1758 out:
1759         RETURN(rc);
1760 }
1761
1762
1763 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1764                           struct chk_dqblk *cdqb)
1765 {
1766         struct obd_quotactl *oqc;
1767         long now;
1768         int rc;
1769         ENTRY;
1770
1771         OBD_ALLOC_PTR(oqc);
1772         if (!oqc)
1773                 RETURN(-ENOMEM);
1774
1775         now = CURRENT_SECONDS;
1776
1777         if (cdqb->dqb_bsoftlimit &&
1778             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1779             !cdqb->dqb_btime)
1780                 cdqb->dqb_btime =
1781                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1782
1783         if (cdqb->dqb_isoftlimit &&
1784             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1785             !cdqb->dqb_itime)
1786                 cdqb->dqb_itime =
1787                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1788
1789         cdqb->dqb_valid = QIF_ALL;
1790
1791         oqc->qc_cmd = Q_SETQUOTA;
1792         oqc->qc_type = cdqb->dqb_type;
1793         oqc->qc_id = cdqb->dqb_id;
1794         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1795
1796         rc = fsfilt_ext3_quotactl(sb, oqc);
1797         OBD_FREE_PTR(oqc);
1798         RETURN(rc);
1799 }
1800
1801 static int prune_chkquots(struct super_block *sb,
1802                           struct qchk_ctxt *qctxt, int error)
1803 {
1804         struct chk_dqblk *cdqb, *tmp;
1805         int rc;
1806
1807         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1808                 if (!error) {
1809                         rc = commit_chkquot(sb, qctxt, cdqb);
1810                         if (rc)
1811                                 error = rc;
1812                 }
1813                 hlist_del_init(&cdqb->dqb_hash);
1814                 list_del(&cdqb->dqb_list);
1815                 OBD_FREE_PTR(cdqb);
1816         }
1817
1818         return error;
1819 }
1820
1821 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1822                                   struct obd_quotactl *oqc)
1823 {
1824         struct ext3_sb_info *sbi = EXT3_SB(sb);
1825         int i, group;
1826         struct qchk_ctxt *qctxt;
1827         struct buffer_head *bitmap_bh = NULL;
1828         unsigned long ino;
1829         struct inode *inode;
1830         int rc = 0;
1831         ENTRY;
1832
1833         /* turn on quota and read dqinfo if existed */
1834         OBD_ALLOC_PTR(qctxt);
1835         if (!qctxt) {
1836                 oqc->qc_stat = -ENOMEM;
1837                 RETURN(-ENOMEM);
1838         }
1839
1840         for (i = 0; i < NR_DQHASH; i++)
1841                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1842         INIT_LIST_HEAD(&qctxt->qckt_list);
1843
1844         for (i = 0; i < MAXQUOTAS; i++) {
1845                 if (!Q_TYPESET(oqc, i))
1846                         continue;
1847
1848                 rc = quota_onoff(sb, Q_QUOTAON, i);
1849                 if (!rc || rc == -EBUSY) {
1850                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1851                         if (rc)
1852                                 GOTO(out, rc);
1853                 } else if (rc == -ENOENT) {
1854                         qctxt->qckt_first_check[i] = 1;
1855                 } else if (rc) {
1856                         GOTO(out, rc);
1857                 }
1858         }
1859
1860         /* check quota and update in hash */
1861         for (group = 0; group < sbi->s_groups_count; group++) {
1862                 ino = group * sbi->s_inodes_per_group + 1;
1863                 bitmap_bh = read_inode_bitmap(sb, group);
1864                 if (!bitmap_bh) {
1865                         CERROR("read_inode_bitmap group %d failed", group);
1866                         GOTO(out, rc = -EIO);
1867                 }
1868
1869                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1870                         if (ino < sbi->s_first_ino)
1871                                 continue;
1872
1873                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1874                         rc = add_inode_quota(inode, qctxt, oqc);
1875                         iput(inode);
1876                         if (rc) {
1877                                 brelse(bitmap_bh);
1878                                 GOTO(out, rc);
1879                         }
1880                 }
1881
1882                 brelse(bitmap_bh);
1883         }
1884
1885         /* read old quota limits from old quota file. (only for the user
1886          * has limits but hasn't file) */
1887 #ifdef HAVE_QUOTA_SUPPORT
1888         for (i = 0; i < MAXQUOTAS; i++) {
1889                 struct list_head id_list;
1890                 struct dquot_id *dqid, *tmp;
1891
1892                 if (!Q_TYPESET(oqc, i))
1893                         continue;
1894
1895                 if (qctxt->qckt_first_check[i])
1896                         continue;
1897
1898
1899                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1900                 INIT_LIST_HEAD(&id_list);
1901 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1902                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1903 #else
1904                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1905 #endif
1906                 if (rc)
1907                         CERROR("read old limits failed. (rc:%d)\n", rc);
1908
1909                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1910                         list_del_init(&dqid->di_link);
1911
1912                         if (!rc)
1913                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1914                                       dqid->di_id, i,
1915                                       qctxt->qckt_first_check[i]);
1916                         kfree(dqid);
1917                 }
1918         }
1919 #endif
1920         /* turn off quota cause we are to dump chk_dqblk to files */
1921         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1922
1923         rc = create_new_quota_files(qctxt, oqc);
1924         if (rc)
1925                 GOTO(out, rc);
1926
1927         /* we use vfs functions to set dqblk, so turn quota on */
1928         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
1929 out:
1930         /* dump and free chk_dqblk */
1931         rc = prune_chkquots(sb, qctxt, rc);
1932         OBD_FREE_PTR(qctxt);
1933
1934         /* turn off quota, `lfs quotacheck` will turn on when all
1935          * nodes quotacheck finish. */
1936         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1937
1938         oqc->qc_stat = rc;
1939         if (rc)
1940                 CERROR("quotacheck failed: rc = %d\n", rc);
1941
1942         RETURN(rc);
1943 }
1944
1945 #ifdef HAVE_QUOTA_SUPPORT
1946 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
1947                                  int cmd)
1948 {
1949         int rc = 0;
1950         ENTRY;
1951
1952         if (lqi->qi_files[type] == NULL) {
1953                 CERROR("operate qinfo before it's enabled!\n");
1954                 RETURN(-EIO);
1955         }
1956
1957         switch (cmd) {
1958         case QFILE_CHK:
1959                 rc = lustre_check_quota_file(lqi, type);
1960                 break;
1961         case QFILE_RD_INFO:
1962                 rc = lustre_read_quota_info(lqi, type);
1963                 break;
1964         case QFILE_WR_INFO:
1965                 rc = lustre_write_quota_info(lqi, type);
1966                 break;
1967         case QFILE_INIT_INFO:
1968                 rc = lustre_init_quota_info(lqi, type);
1969                 break;
1970         default:
1971                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1972                 LBUG();
1973                 break;
1974         }
1975         RETURN(rc);
1976 }
1977
1978 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
1979                             struct list_head *list)
1980 {
1981         return lustre_get_qids(file, inode, type, list);
1982 }
1983
1984 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
1985 {
1986         int rc = 0;
1987         ENTRY;
1988
1989         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
1990                 CERROR("operate dquot before it's enabled!\n");
1991                 RETURN(-EIO);
1992         }
1993
1994         switch (cmd) {
1995         case QFILE_RD_DQUOT:
1996                 rc = lustre_read_dquot(dquot);
1997                 break;
1998         case QFILE_WR_DQUOT:
1999                 if (dquot->dq_dqb.dqb_ihardlimit ||
2000                     dquot->dq_dqb.dqb_isoftlimit ||
2001                     dquot->dq_dqb.dqb_bhardlimit ||
2002                     dquot->dq_dqb.dqb_bsoftlimit)
2003                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2004                 else
2005                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2006
2007                 rc = lustre_commit_dquot(dquot);
2008                 if (rc >= 0)
2009                         rc = 0;
2010                 break;
2011         default:
2012                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2013                 LBUG();
2014                 break;
2015         }
2016         RETURN(rc);
2017 }
2018 #endif
2019
2020 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2021 {
2022         return (EXT3_SB(sb)->journal_bdev);
2023 }
2024 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2025
2026 static struct fsfilt_operations fsfilt_ext3_ops = {
2027         .fs_type                = "ext3",
2028         .fs_owner               = THIS_MODULE,
2029         .fs_getlabel            = fsfilt_ext3_get_label,
2030         .fs_setlabel            = fsfilt_ext3_set_label,
2031         .fs_uuid                = fsfilt_ext3_uuid,
2032         .fs_start               = fsfilt_ext3_start,
2033         .fs_brw_start           = fsfilt_ext3_brw_start,
2034         .fs_extend              = fsfilt_ext3_extend,
2035         .fs_commit              = fsfilt_ext3_commit,
2036         .fs_commit_async        = fsfilt_ext3_commit_async,
2037         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2038         .fs_setattr             = fsfilt_ext3_setattr,
2039         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2040         .fs_set_md              = fsfilt_ext3_set_md,
2041         .fs_get_md              = fsfilt_ext3_get_md,
2042         .fs_readpage            = fsfilt_ext3_readpage,
2043         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2044         .fs_statfs              = fsfilt_ext3_statfs,
2045         .fs_sync                = fsfilt_ext3_sync,
2046         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2047         .fs_write_record        = fsfilt_ext3_write_record,
2048         .fs_read_record         = fsfilt_ext3_read_record,
2049         .fs_setup               = fsfilt_ext3_setup,
2050         .fs_send_bio            = fsfilt_ext3_send_bio,
2051         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2052         .fs_quotactl            = fsfilt_ext3_quotactl,
2053         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2054 #ifdef HAVE_DISK_INODE_VERSION
2055         .fs_get_version         = fsfilt_ext3_get_version,
2056         .fs_set_version         = fsfilt_ext3_set_version,
2057 #endif
2058 #ifdef HAVE_QUOTA_SUPPORT
2059         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2060         .fs_qids                = fsfilt_ext3_qids,
2061         .fs_dquot               = fsfilt_ext3_dquot,
2062 #endif
2063         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2064 };
2065
2066 static int __init fsfilt_ext3_init(void)
2067 {
2068         int rc;
2069
2070         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2071                                          sizeof(struct fsfilt_cb_data), 0, 0);
2072         if (!fcb_cache) {
2073                 CERROR("error allocating fsfilt journal callback cache\n");
2074                 GOTO(out, rc = -ENOMEM);
2075         }
2076
2077         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2078
2079         if (rc) {
2080                 int err = cfs_mem_cache_destroy(fcb_cache);
2081                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2082         }
2083 out:
2084         return rc;
2085 }
2086
2087 static void __exit fsfilt_ext3_exit(void)
2088 {
2089         int rc;
2090
2091         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2092         rc = cfs_mem_cache_destroy(fcb_cache);
2093         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2094 }
2095
2096 module_init(fsfilt_ext3_init);
2097 module_exit(fsfilt_ext3_exit);
2098
2099 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2100 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2101 MODULE_LICENSE("GPL");