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 <linux/parser.h>
  43 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  44 #include <linux/ext3_xattr.h>
  45 #else
  46 #include <ext3/xattr.h>
  47 #endif
  48
  49 #include <libcfs/kp30.h>
  50 #include <lustre_fsfilt.h>
  51 #include <obd.h>
  52 #include <lustre_quota.h>
  53 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  54 #include <linux/iobuf.h>
  55 #endif
  56 #include <linux/lustre_compat25.h>
  57 #include <linux/lprocfs_status.h>
  58
  59 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  60 #include <linux/ext3_extents.h>
  61 #endif
  62
  63 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  64 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  65 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  66 #else
  67 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  68 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  69 #endif
  70
  71 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
  72 /* for kernels 2.6.18 and later */
  73 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  74 #else
  75 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
  76 #endif
  77
  78 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  79 #define fsfilt_ext3_journal_start(inode, nblocks) \
  80                                 journal_start(EXT3_JOURNAL(inode),nblocks)
  81 #define fsfilt_ext3_journal_stop(handle)          journal_stop(handle)
  82 #else
  83 #define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks)
  84 #define fsfilt_ext3_journal_stop(handle)          ext3_journal_stop(handle)
  85 #endif
  86
  87 static cfs_mem_cache_t *fcb_cache;
  88
  89 struct fsfilt_cb_data {
  90         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  91         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  92         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  93         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  94         void *cb_data;                  /* MDS/OST completion function data */
  95 };
  96
  97 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
  98 #define EXT3_XATTR_INDEX_TRUSTED        4
  99 #endif
 100
 101 static char *fsfilt_ext3_get_label(struct super_block *sb)
 102 {
 103         return EXT3_SB(sb)->s_es->s_volume_name;
 104 }
 105
 106 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 107 {
 108         /* see e.g. fsfilt_ext3_write_record() */
 109         journal_t *journal;
 110         handle_t *handle;
 111         int err;
 112
 113         journal = EXT3_SB(sb)->s_journal;
 114         lock_24kernel();
 115         handle = journal_start(journal, 1);
 116         unlock_24kernel();
 117         if (IS_ERR(handle)) {
 118                 CERROR("can't start transaction\n");
 119                 return(PTR_ERR(handle));
 120         }
 121
 122         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 123         if (err)
 124                 goto out;
 125
 126         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 127                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 128
 129         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 130
 131 out:
 132         lock_24kernel();
 133         journal_stop(handle);
 134         unlock_24kernel();
 135
 136         return(err);
 137 }
 138
 139 static char *fsfilt_ext3_uuid(struct super_block *sb)
 140 {
 141         return EXT3_SB(sb)->s_es->s_uuid;
 142 }
 143
 144 #ifdef HAVE_DISK_INODE_VERSION
 145 /*
 146  * Get the 64-bit version for an inode.
 147  */
 148 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 149 {
 150         return EXT3_I(inode)->i_fs_version;
 151 }
 152
 153 /*
 154  * Set the 64-bit version and return the old version.
 155  */
 156 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 157 {
 158         __u64 old_version = EXT3_I(inode)->i_fs_version;
 159
 160         (EXT3_I(inode))->i_fs_version = new_version;
 161         return old_version;
 162 }
 163
 164 #endif
 165
 166 /*
 167  * We don't currently need any additional blocks for rmdir and
 168  * unlink transactions because we are storing the OST oa_id inside
 169  * the inode (which we will be changing anyways as part of this
 170  * transaction).
 171  */
 172 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 173                                int logs)
 174 {
 175         /* For updates to the last received file */
 176         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 177         journal_t *journal;
 178         void *handle;
 179
 180         if (current->journal_info) {
 181                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 182                        current->journal_info);
 183                 goto journal_start;
 184         }
 185
 186         switch(op) {
 187         case FSFILT_OP_RMDIR:
 188         case FSFILT_OP_UNLINK:
 189                 /* delete one file + create/update logs for each stripe */
 190                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 191                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 192                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 193                 break;
 194         case FSFILT_OP_RENAME:
 195                 /* modify additional directory */
 196                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 197                 /* no break */
 198         case FSFILT_OP_SYMLINK:
 199                 /* additional block + block bitmap + GDT for long symlink */
 200                 nblocks += 3;
 201                 /* no break */
 202         case FSFILT_OP_CREATE: {
 203 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 204                 static int warned;
 205                 if (!warned) {
 206                         if (!test_opt(inode->i_sb, EXTENTS)) {
 207                                 warned = 1;
 208                         } else if (((EXT3_I(inode)->i_flags &
 209                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 210                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 211                                 CWARN("extent-mapped directory found - contact "
 212                                       "CFS: support@clusterfs.com\n");
 213                                 warned = 1;
 214                         }
 215                 }
 216 #endif
 217                 /* no break */
 218         }
 219         case FSFILT_OP_MKDIR:
 220         case FSFILT_OP_MKNOD:
 221                 /* modify one inode + block bitmap + GDT */
 222                 nblocks += 3;
 223                 /* no break */
 224         case FSFILT_OP_LINK:
 225                 /* modify parent directory */
 226                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 227                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 228                 /* create/update logs for each stripe */
 229                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 230                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 231                 break;
 232         case FSFILT_OP_SETATTR:
 233                 /* Setattr on inode */
 234                 nblocks += 1;
 235                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 236                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 237                 /* quota chown log for each stripe */
 238                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 239                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 240                 break;
 241         case FSFILT_OP_CANCEL_UNLINK:
 242                 /* blocks for log header bitmap update OR
 243                  * blocks for catalog header bitmap update + unlink of logs */
 244                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 245                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 246                 break;
 247         case FSFILT_OP_JOIN:
 248                 /* delete 2 file(file + array id) + create 1 file (array id)
 249                  * create/update logs for each stripe */
 250                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 251
 252                 /*create array log for head file*/
 253                 nblocks += 3;
 254                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 255                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
 256                 /*update head file array */
 257                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 258                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 259                 break;
 260         default: CERROR("unknown transaction start op %d\n", op);
 261                 LBUG();
 262         }
 263
 264         LASSERT(current->journal_info == desc_private);
 265         journal = EXT3_SB(inode->i_sb)->s_journal;
 266         if (nblocks > journal->j_max_transaction_buffers) {
 267                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 268                       nblocks, op, logs, journal->j_max_transaction_buffers);
 269                 nblocks = journal->j_max_transaction_buffers;
 270         }
 271
 272  journal_start:
 273         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 274         lock_24kernel();
 275         handle = fsfilt_ext3_journal_start(inode, nblocks);
 276         unlock_24kernel();
 277
 278         if (!IS_ERR(handle))
 279                 LASSERT(current->journal_info == handle);
 280         else
 281                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 282                        op, nblocks, PTR_ERR(handle));
 283         return handle;
 284 }
 285
 286 /*
 287  * Calculate the number of buffer credits needed to write multiple pages in
 288  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 289  * doesn't have a nice API for calculating this sort of thing in advance.
 290  *
 291  * See comment above ext3_writepage_trans_blocks for details.  We assume
 292  * no data journaling is being done, but it does allow for all of the pages
 293  * being non-contiguous.  If we are guaranteed contiguous pages we could
 294  * reduce the number of (d)indirect blocks a lot.
 295  *
 296  * With N blocks per page and P pages, for each inode we have at most:
 297  * N*P indirect
 298  * min(N*P, blocksize/4 + 1) dindirect blocks
 299  * niocount tindirect
 300  *
 301  * For the entire filesystem, we have at most:
 302  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 303  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 304  * objcount inode blocks
 305  * 1 superblock
 306  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 307  *
 308  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 309  */
 310 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 311                                       int niocount, struct niobuf_local *nb)
 312 {
 313         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 314         __u64 next_indir;
 315         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 316         int nbitmaps = 0, ngdblocks;
 317         int needed = objcount + 1; /* inodes + superblock */
 318         int i, j;
 319
 320         for (i = 0, j = 0; i < objcount; i++, fso++) {
 321                 /* two or more dindirect blocks in case we cross boundary */
 322                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 323                                     nb[j].offset) >>
 324                                    sb->s_blocksize_bits) /
 325                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 326                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 327
 328                 /* leaf, indirect, tindirect blocks for first block */
 329                 nbitmaps += blockpp + 2;
 330
 331                 j += fso->fso_bufcnt;
 332         }
 333
 334         next_indir = nb[0].offset +
 335                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 336         for (i = 1; i < niocount; i++) {
 337                 if (nb[i].offset >= next_indir) {
 338                         nbitmaps++;     /* additional indirect */
 339                         next_indir = nb[i].offset +
 340                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 341                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 342                         nbitmaps++;     /* additional indirect */
 343                 }
 344                 nbitmaps += blockpp;    /* each leaf in different group? */
 345         }
 346
 347         ngdblocks = nbitmaps;
 348         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 349                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 350         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 351                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 352
 353         needed += nbitmaps + ngdblocks;
 354
 355         /* last_rcvd update */
 356         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 357
 358 #if defined(CONFIG_QUOTA)
 359         /* We assume that there will be 1 bit set in s_dquot.flags for each
 360          * quota file that is active.  This is at least true for now.
 361          */
 362         needed += hweight32(sb_any_quota_enabled(sb)) *
 363                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 364 #endif
 365
 366         return needed;
 367 }
 368
 369 /* We have to start a huge journal transaction here to hold all of the
 370  * metadata for the pages being written here.  This is necessitated by
 371  * the fact that we do lots of prepare_write operations before we do
 372  * any of the matching commit_write operations, so even if we split
 373  * up to use "smaller" transactions none of them could complete until
 374  * all of them were opened.  By having a single journal transaction,
 375  * we eliminate duplicate reservations for common blocks like the
 376  * superblock and group descriptors or bitmaps.
 377  *
 378  * We will start the transaction here, but each prepare_write will
 379  * add a refcount to the transaction, and each commit_write will
 380  * remove a refcount.  The transaction will be closed when all of
 381  * the pages have been written.
 382  */
 383 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 384                                    int niocount, struct niobuf_local *nb,
 385                                    void *desc_private, int logs)
 386 {
 387         journal_t *journal;
 388         handle_t *handle;
 389         int needed;
 390         ENTRY;
 391
 392         LASSERT(current->journal_info == desc_private);
 393         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 394         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 395
 396         /* The number of blocks we could _possibly_ dirty can very large.
 397          * We reduce our request if it is absurd (and we couldn't get that
 398          * many credits for a single handle anyways).
 399          *
 400          * At some point we have to limit the size of I/Os sent at one time,
 401          * increase the size of the journal, or we have to calculate the
 402          * actual journal requirements more carefully by checking all of
 403          * the blocks instead of being maximally pessimistic.  It remains to
 404          * be seen if this is a real problem or not.
 405          */
 406         if (needed > journal->j_max_transaction_buffers) {
 407                 CERROR("want too many journal credits (%d) using %d instead\n",
 408                        needed, journal->j_max_transaction_buffers);
 409                 needed = journal->j_max_transaction_buffers;
 410         }
 411
 412         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 413         lock_24kernel();
 414         handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
 415         unlock_24kernel();
 416         if (IS_ERR(handle)) {
 417                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 418                        PTR_ERR(handle));
 419         } else {
 420                 LASSERT(handle->h_buffer_credits >= needed);
 421                 LASSERT(current->journal_info == handle);
 422         }
 423
 424         RETURN(handle);
 425 }
 426
 427 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 428 {
 429        handle_t *handle = h;
 430
 431        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 432        if (nblocks == 0) {
 433                handle->h_buffer_credits = 0;
 434                CWARN("setting credits of handle %p to zero by request\n", h);
 435        }
 436
 437        if (handle->h_buffer_credits > nblocks)
 438                 return 0;
 439        if (journal_extend(handle, nblocks) == 0)
 440                 return 0;
 441
 442        ext3_mark_inode_dirty(handle, inode);
 443        return journal_restart(handle, nblocks);
 444 }
 445
 446 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 447 {
 448         int rc;
 449         handle_t *handle = h;
 450
 451         LASSERT(current->journal_info == handle);
 452         if (force_sync)
 453                 handle->h_sync = 1; /* recovery likes this */
 454
 455         lock_24kernel();
 456         rc = fsfilt_ext3_journal_stop(handle);
 457         unlock_24kernel();
 458
 459         return rc;
 460 }
 461
 462 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 463                                     void **wait_handle)
 464 {
 465         unsigned long tid;
 466         transaction_t *transaction;
 467 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 468         unsigned long rtid;
 469 #endif
 470         handle_t *handle = h;
 471         journal_t *journal;
 472         int rc;
 473
 474         LASSERT(current->journal_info == handle);
 475
 476         lock_24kernel();
 477         transaction = handle->h_transaction;
 478         journal = transaction->t_journal;
 479         tid = transaction->t_tid;
 480         /* we don't want to be blocked */
 481         handle->h_sync = 0;
 482         rc = fsfilt_ext3_journal_stop(handle);
 483         if (rc) {
 484                 CERROR("error while stopping transaction: %d\n", rc);
 485                 unlock_24kernel();
 486                 return rc;
 487         }
 488 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 489         rtid = log_start_commit(journal, transaction);
 490         if (rtid != tid)
 491                 CERROR("strange race: %lu != %lu\n",
 492                        (unsigned long) tid, (unsigned long) rtid);
 493 #else
 494         log_start_commit(journal, tid);
 495 #endif
 496         unlock_24kernel();
 497
 498         *wait_handle = (void *) tid;
 499         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 500         return 0;
 501 }
 502
 503 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 504 {
 505         journal_t *journal = EXT3_JOURNAL(inode);
 506         tid_t tid = (tid_t)(long)h;
 507
 508         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 509         if (unlikely(is_journal_aborted(journal)))
 510                 return -EIO;
 511
 512         log_wait_commit(EXT3_JOURNAL(inode), tid);
 513
 514         if (unlikely(is_journal_aborted(journal)))
 515                 return -EIO;
 516         return 0;
 517 }
 518
 519 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 520                                struct iattr *iattr, int do_trunc)
 521 {
 522         struct inode *inode = dentry->d_inode;
 523         int rc = 0;
 524
 525         lock_24kernel();
 526
 527         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 528          * We are already writing the inode to disk as part of this
 529          * transaction and want to avoid a lot of extra inode writeout
 530          * later on. b=9828 */
 531         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 532                 /* ATTR_SIZE would invoke truncate: clear it */
 533                 iattr->ia_valid &= ~ATTR_SIZE;
 534                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 535                 i_size_write(inode, iattr->ia_size);
 536
 537                 if (iattr->ia_valid & ATTR_UID)
 538                         inode->i_uid = iattr->ia_uid;
 539                 if (iattr->ia_valid & ATTR_GID)
 540                         inode->i_gid = iattr->ia_gid;
 541                 if (iattr->ia_valid & ATTR_ATIME)
 542                         inode->i_atime = iattr->ia_atime;
 543                 if (iattr->ia_valid & ATTR_MTIME)
 544                         inode->i_mtime = iattr->ia_mtime;
 545                 if (iattr->ia_valid & ATTR_CTIME)
 546                         inode->i_ctime = iattr->ia_ctime;
 547                 if (iattr->ia_valid & ATTR_MODE) {
 548                         inode->i_mode = iattr->ia_mode;
 549
 550                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 551                                 inode->i_mode &= ~S_ISGID;
 552                 }
 553
 554                 inode->i_sb->s_op->dirty_inode(inode);
 555
 556                 goto out;
 557         }
 558
 559         /* Don't allow setattr to change file type */
 560         if (iattr->ia_valid & ATTR_MODE)
 561                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 562                                  (iattr->ia_mode & ~S_IFMT);
 563
 564         /* We set these flags on the client, but have already checked perms
 565          * so don't confuse inode_change_ok. */
 566         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 567
 568         if (inode->i_op->setattr) {
 569                 rc = inode->i_op->setattr(dentry, iattr);
 570         } else {
 571                 rc = inode_change_ok(inode, iattr);
 572                 if (!rc)
 573                         rc = inode_setattr(inode, iattr);
 574         }
 575
 576  out:
 577         unlock_24kernel();
 578         RETURN(rc);
 579 }
 580
 581 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 582                                  unsigned int cmd, unsigned long arg)
 583 {
 584         int rc = 0;
 585         ENTRY;
 586
 587         /* FIXME: Can't do this because of nested transaction deadlock */
 588         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 589                 CERROR("can't set data journal flag on file\n");
 590                 RETURN(-EPERM);
 591         }
 592
 593         if (inode->i_fop->ioctl)
 594                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 595         else
 596                 RETURN(-ENOTTY);
 597
 598         RETURN(rc);
 599 }
 600
 601 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 602                               void *lmm, int lmm_size, const char *name)
 603 {
 604         int rc;
 605
 606         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 607
 608         lock_24kernel();
 609         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 610                                    name, lmm, lmm_size, 0);
 611
 612         unlock_24kernel();
 613
 614         if (rc && rc != -EROFS)
 615                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 616                        inode->i_ino, rc);
 617         return rc;
 618 }
 619
 620 /* Must be called with i_mutex held */
 621 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 622                               const char *name)
 623 {
 624         int rc;
 625
 626         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 627         lock_24kernel();
 628
 629         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 630                             name, lmm, lmm_size);
 631         unlock_24kernel();
 632
 633         /* This gives us the MD size */
 634         if (lmm == NULL)
 635                 return (rc == -ENODATA) ? 0 : rc;
 636
 637         if (rc < 0) {
 638                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 639                        EXT3_XATTR_INDEX_TRUSTED, name,
 640                        inode->i_ino, rc);
 641                 memset(lmm, 0, lmm_size);
 642                 return (rc == -ENODATA) ? 0 : rc;
 643         }
 644
 645         return rc;
 646 }
 647
 648 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
 649 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 650 {
 651         submit_bio(rw, bio);
 652         return 0;
 653 }
 654 #else
 655 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct kiobuf *bio)
 656 {
 657         int rc, blk_per_page;
 658
 659         rc = brw_kiovec(rw, 1, &bio, inode->i_dev,
 660                         KIOBUF_GET_BLOCKS(bio), 1 << inode->i_blkbits);
 661         /*
 662          * brw_kiovec() returns number of bytes actually written. If error
 663          * occurred after something was written, error code is returned though
 664          * kiobuf->errno. (See bug 6854.)
 665          */
 666
 667         blk_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
 668
 669         if (rc != (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page) {
 670                 CERROR("short write?  expected %d, wrote %d (%d)\n",
 671                        (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
 672                        rc, bio->errno);
 673         }
 674         if (bio->errno != 0) {
 675                 CERROR("IO error. Wrote %d of %d (%d)\n",
 676                        rc,
 677                        (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
 678                        bio->errno);
 679                 rc = bio->errno;
 680         }
 681
 682         return rc;
 683 }
 684 #endif
 685
 686 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 687                                     loff_t *off)
 688 {
 689         struct inode *inode = file->f_dentry->d_inode;
 690         int rc = 0;
 691
 692         if (S_ISREG(inode->i_mode))
 693                 rc = file->f_op->read(file, buf, count, off);
 694         else {
 695                 const int blkbits = inode->i_sb->s_blocksize_bits;
 696                 const int blksize = inode->i_sb->s_blocksize;
 697
 698                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 699                        count, inode->i_ino, *off);
 700                 while (count > 0) {
 701                         struct buffer_head *bh;
 702
 703                         bh = NULL;
 704                         if (*off < i_size_read(inode)) {
 705                                 int err = 0;
 706
 707                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 708                                                 0, &err);
 709
 710                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 711
 712                                 if (bh) {
 713                                         memcpy(buf, bh->b_data, blksize);
 714                                         brelse(bh);
 715                                 } else if (err) {
 716                                         /* XXX in theory we should just fake
 717                                          * this buffer and continue like ext3,
 718                                          * especially if this is a partial read
 719                                          */
 720                                         CERROR("error read dir %lu+%llu: %d\n",
 721                                                inode->i_ino, *off, err);
 722                                         RETURN(err);
 723                                 }
 724                         }
 725                         if (!bh) {
 726                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 727
 728                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 729                                 memset(fake, 0, sizeof(*fake));
 730                                 fake->rec_len = cpu_to_le16(blksize);
 731                         }
 732                         count -= blksize;
 733                         buf += blksize;
 734                         *off += blksize;
 735                         rc += blksize;
 736                 }
 737         }
 738
 739         return rc;
 740 }
 741
 742 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 743 {
 744         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 745
 746         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 747
 748         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 749 }
 750
 751 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 752                                       void *handle, fsfilt_cb_t cb_func,
 753                                       void *cb_data)
 754 {
 755         struct fsfilt_cb_data *fcb;
 756
 757         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 758         if (fcb == NULL)
 759                 RETURN(-ENOMEM);
 760
 761         fcb->cb_func = cb_func;
 762         fcb->cb_obd = obd;
 763         fcb->cb_last_rcvd = last_rcvd;
 764         fcb->cb_data = cb_data;
 765
 766         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 767         lock_24kernel();
 768         journal_callback_set(handle, fsfilt_ext3_cb_func,
 769                              (struct journal_callback *)fcb);
 770         unlock_24kernel();
 771
 772         return 0;
 773 }
 774
 775 /*
 776  * We need to hack the return value for the free inode counts because
 777  * the current EA code requires one filesystem block per inode with EAs,
 778  * so it is possible to run out of blocks before we run out of inodes.
 779  *
 780  * This can be removed when the ext3 EA code is fixed.
 781  */
 782 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 783 {
 784         struct kstatfs sfs;
 785         int rc;
 786
 787         memset(&sfs, 0, sizeof(sfs));
 788         rc = ll_do_statfs(sb,&sfs);
 789
 790         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 791                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 792                 sfs.f_ffree = sfs.f_bfree;
 793         }
 794
 795         statfs_pack(osfs, &sfs);
 796         return rc;
 797 }
 798
 799 static int fsfilt_ext3_sync(struct super_block *sb)
 800 {
 801         return ext3_force_commit(sb);
 802 }
 803
 804 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 805 #warning "kernel code has old extents/mballoc patch, disabling"
 806 #undef EXT3_MULTIBLOCK_ALLOCATOR
 807 #endif
 808 #ifndef EXT3_EXTENTS_FL
 809 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 810 #endif
 811
 812 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 813 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 814 #define ext3_up_truncate_sem(inode)  up_write(&EXT3_I(inode)->truncate_sem);
 815 #define ext3_down_truncate_sem(inode)  down_write(&EXT3_I(inode)->truncate_sem);
 816 #elif (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 817 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 818 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 819 #else
 820 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 821 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 822 #endif
 823
 824 #ifndef EXT_ASSERT
 825 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 826 #endif
 827
 828 #ifdef EXT3_EXT_HAS_NO_TREE
 829 /* for kernels 2.6.18 and later */
 830 #define ext3_ext_base                   inode
 831 #define ext3_ext_base2inode(inode)      (inode)
 832 #define EXT_DEPTH(inode)                ext_depth(inode)
 833 #define EXT_GENERATION(inode)           ext_generation(inode)
 834 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 835                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 836 #else
 837 #define ext3_ext_base                   ext3_extents_tree
 838 #define ext3_ext_base2inode(tree)       (tree->inode)
 839 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 840                         ext3_ext_walk_space(tree, block, num, cb);
 841 #endif
 842
 843 #include <linux/lustre_version.h>
 844 #if EXT3_EXT_MAGIC == 0xf301
 845 #define ee_start e_start
 846 #define ee_block e_block
 847 #define ee_len   e_num
 848 #endif
 849 #ifndef EXT3_BB_MAX_BLOCKS
 850 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 851         ext3_new_blocks(handle, inode, count, goal, err)
 852 #endif
 853
 854 struct bpointers {
 855         unsigned long *blocks;
 856         int *created;
 857         unsigned long start;
 858         int num;
 859         int init_num;
 860         int create;
 861 };
 862
 863 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 864                               unsigned long block, int *aflags)
 865 {
 866         struct ext3_inode_info *ei = EXT3_I(inode);
 867         unsigned long bg_start;
 868         unsigned long colour;
 869         int depth;
 870
 871         if (path) {
 872                 struct ext3_extent *ex;
 873                 depth = path->p_depth;
 874
 875                 /* try to predict block placement */
 876                 if ((ex = path[depth].p_ext)) {
 877 #if 0
 878                         /* This prefers to eat into a contiguous extent
 879                          * rather than find an extent that the whole
 880                          * request will fit into.  This can fragment data
 881                          * block allocation and prevents our lovely 1M I/Os
 882                          * from reaching the disk intact. */
 883                         if (ex->ee_block + ex->ee_len == block)
 884                                 *aflags |= 1;
 885 #endif
 886                         return ex->ee_start + (block - ex->ee_block);
 887                 }
 888
 889                 /* it looks index is empty
 890                  * try to find starting from index itself */
 891                 if (path[depth].p_bh)
 892                         return path[depth].p_bh->b_blocknr;
 893         }
 894
 895         /* OK. use inode's group */
 896         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 897                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 898         colour = (current->pid % 16) *
 899                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 900         return bg_start + colour + block;
 901 }
 902
 903 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 904 #include <linux/locks.h>
 905 static void ll_unmap_underlying_metadata(struct super_block *sb,
 906                                          unsigned long blocknr)
 907 {
 908         struct buffer_head *old_bh;
 909
 910         old_bh = get_hash_table(sb->s_dev, blocknr, sb->s_blocksize);
 911         if (old_bh) {
 912                 mark_buffer_clean(old_bh);
 913                 wait_on_buffer(old_bh);
 914                 clear_bit(BH_Req, &old_bh->b_state);
 915                 __brelse(old_bh);
 916         }
 917 }
 918 #else
 919 #define ll_unmap_underlying_metadata(sb, blocknr) \
 920         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 921 #endif
 922
 923 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 924 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 925                                 struct ext3_ext_path *path, unsigned long block,
 926                                 unsigned long *count, int *err)
 927 {
 928         unsigned long pblock, goal;
 929         int aflags = 0;
 930         struct inode *inode = ext3_ext_base2inode(base);
 931
 932         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 933         aflags |= 2; /* block have been already reserved */
 934         lock_24kernel();
 935         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 936         unlock_24kernel();
 937         return pblock;
 938
 939 }
 940 #else
 941 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 942                                 struct ext3_ext_path *path, unsigned long block,
 943                                 unsigned long *count, int *err)
 944 {
 945         struct inode *inode = ext3_ext_base2inode(base);
 946         struct ext3_allocation_request ar;
 947         unsigned long pblock;
 948         int aflags;
 949
 950         /* find neighbour allocated blocks */
 951         ar.lleft = block;
 952         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 953         if (*err)
 954                 return 0;
 955         ar.lright = block;
 956         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 957         if (*err)
 958                 return 0;
 959
 960         /* allocate new block */
 961         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 962         ar.inode = inode;
 963         ar.logical = block;
 964         ar.len = *count;
 965         ar.flags = EXT3_MB_HINT_DATA;
 966         pblock = ext3_mb_new_blocks(handle, &ar, err);
 967         *count = ar.len;
 968         return pblock;
 969 }
 970 #endif
 971
 972 #ifdef EXT3_EXT_HAS_NO_TREE
 973 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 974                                   struct ext3_ext_path *path,
 975                                   struct ext3_ext_cache *cex,
 976 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 977                                   struct ext3_extent *ex,
 978 #endif
 979                                   void *cbdata)
 980 {
 981         struct bpointers *bp = cbdata;
 982 #else
 983 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 984                                   struct ext3_ext_path *path,
 985                                   struct ext3_ext_cache *cex
 986 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 987                                   , struct ext3_extent *ex
 988 #endif
 989                                  )
 990 {
 991         struct bpointers *bp = base->private;
 992 #endif
 993         struct inode *inode = ext3_ext_base2inode(base);
 994         struct ext3_extent nex;
 995         unsigned long pblock;
 996         unsigned long tgen;
 997         int err, i;
 998         unsigned long count;
 999         handle_t *handle;
1000
1001         i = EXT_DEPTH(base);
1002         EXT_ASSERT(i == path->p_depth);
1003         EXT_ASSERT(path[i].p_hdr);
1004
1005         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1006                 err = EXT_CONTINUE;
1007                 goto map;
1008         }
1009
1010         if (bp->create == 0) {
1011                 i = 0;
1012                 if (cex->ec_block < bp->start)
1013                         i = bp->start - cex->ec_block;
1014                 if (i >= cex->ec_len)
1015                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1016                                         i, cex->ec_len);
1017                 for (; i < cex->ec_len && bp->num; i++) {
1018                         *(bp->created) = 0;
1019                         bp->created++;
1020                         *(bp->blocks) = 0;
1021                         bp->blocks++;
1022                         bp->num--;
1023                         bp->start++;
1024                 }
1025
1026                 return EXT_CONTINUE;
1027         }
1028
1029         tgen = EXT_GENERATION(base);
1030         count = ext3_ext_calc_credits_for_insert(base, path);
1031         ext3_up_truncate_sem(inode);
1032
1033         lock_24kernel();
1034         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1035         unlock_24kernel();
1036         if (IS_ERR(handle)) {
1037                 ext3_down_truncate_sem(inode);
1038                 return PTR_ERR(handle);
1039         }
1040
1041         ext3_down_truncate_sem(inode);
1042         if (tgen != EXT_GENERATION(base)) {
1043                 /* the tree has changed. so path can be invalid at moment */
1044                 lock_24kernel();
1045                 fsfilt_ext3_journal_stop(handle);
1046                 unlock_24kernel();
1047                 return EXT_REPEAT;
1048         }
1049
1050         count = cex->ec_len;
1051         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1052         if (!pblock)
1053                 goto out;
1054         EXT_ASSERT(count <= cex->ec_len);
1055
1056         /* insert new extent */
1057         nex.ee_block = cex->ec_block;
1058         nex.ee_start = pblock;
1059         nex.ee_len = count;
1060         err = ext3_ext_insert_extent(handle, base, path, &nex);
1061         if (err) {
1062                 CERROR("can't insert extent: %d\n", err);
1063                 /* XXX: export ext3_free_blocks() */
1064                 /*ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);*/
1065                 goto out;
1066         }
1067
1068         /*
1069          * Putting len of the actual extent we just inserted,
1070          * we are asking ext3_ext_walk_space() to continue
1071          * scaning after that block
1072          */
1073         cex->ec_len = nex.ee_len;
1074         cex->ec_start = nex.ee_start;
1075         BUG_ON(nex.ee_len == 0);
1076         BUG_ON(nex.ee_block != cex->ec_block);
1077
1078 out:
1079         lock_24kernel();
1080         fsfilt_ext3_journal_stop(handle);
1081         unlock_24kernel();
1082 map:
1083         if (err >= 0) {
1084                 /* map blocks */
1085                 if (bp->num == 0) {
1086                         CERROR("hmm. why do we find this extent?\n");
1087                         CERROR("initial space: %lu:%u\n",
1088                                 bp->start, bp->init_num);
1089                         CERROR("current extent: %u/%u/%u %d\n",
1090                                 cex->ec_block, cex->ec_len,
1091                                 cex->ec_start, cex->ec_type);
1092                 }
1093                 i = 0;
1094                 if (cex->ec_block < bp->start)
1095                         i = bp->start - cex->ec_block;
1096                 if (i >= cex->ec_len)
1097                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1098                                         i, cex->ec_len);
1099                 for (; i < cex->ec_len && bp->num; i++) {
1100                         *(bp->blocks) = cex->ec_start + i;
1101                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1102                                 *(bp->created) = 0;
1103                         } else {
1104                                 *(bp->created) = 1;
1105                                 /* unmap any possible underlying metadata from
1106                                  * the block device mapping.  bug 6998. */
1107                                 ll_unmap_underlying_metadata(inode->i_sb,
1108                                                              *(bp->blocks));
1109                         }
1110                         bp->created++;
1111                         bp->blocks++;
1112                         bp->num--;
1113                         bp->start++;
1114                 }
1115         }
1116         return err;
1117 }
1118
1119 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1120                        unsigned long num, unsigned long *blocks,
1121                        int *created, int create)
1122 {
1123 #ifdef EXT3_EXT_HAS_NO_TREE
1124         struct ext3_ext_base *base = inode;
1125 #else
1126         struct ext3_extents_tree tree;
1127         struct ext3_ext_base *base = &tree;
1128 #endif
1129         struct bpointers bp;
1130         int err;
1131
1132         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1133                block, block + num - 1, (unsigned) inode->i_ino);
1134
1135 #ifndef EXT3_EXT_HAS_NO_TREE
1136         ext3_init_tree_desc(base, inode);
1137         tree.private = &bp;
1138 #endif
1139         bp.blocks = blocks;
1140         bp.created = created;
1141         bp.start = block;
1142         bp.init_num = bp.num = num;
1143         bp.create = create;
1144
1145         ext3_down_truncate_sem(inode);
1146         err = fsfilt_ext3_ext_walk_space(base, block, num, ext3_ext_new_extent_cb, &bp);
1147         ext3_ext_invalidate_cache(base);
1148         ext3_up_truncate_sem(inode);
1149
1150         return err;
1151 }
1152
1153 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1154                                     int pages, unsigned long *blocks,
1155                                     int *created, int create)
1156 {
1157         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1158         int rc = 0, i = 0;
1159         struct page *fp = NULL;
1160         int clen = 0;
1161
1162         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1163                 inode->i_ino, pages, (*page)->index);
1164
1165         /* pages are sorted already. so, we just have to find
1166          * contig. space and process them properly */
1167         while (i < pages) {
1168                 if (fp == NULL) {
1169                         /* start new extent */
1170                         fp = *page++;
1171                         clen = 1;
1172                         i++;
1173                         continue;
1174                 } else if (fp->index + clen == (*page)->index) {
1175                         /* continue the extent */
1176                         page++;
1177                         clen++;
1178                         i++;
1179                         continue;
1180                 }
1181
1182                 /* process found extent */
1183                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1184                                         clen * blocks_per_page, blocks,
1185                                         created, create);
1186                 if (rc)
1187                         GOTO(cleanup, rc);
1188
1189                 /* look for next extent */
1190                 fp = NULL;
1191                 blocks += blocks_per_page * clen;
1192                 created += blocks_per_page * clen;
1193         }
1194
1195         if (fp)
1196                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1197                                         clen * blocks_per_page, blocks,
1198                                         created, create);
1199 cleanup:
1200         return rc;
1201 }
1202 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1203
1204 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1205                                unsigned long *blocks, int *created, int create);
1206 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1207                                    int pages, unsigned long *blocks,
1208                                    int *created, int create)
1209 {
1210         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1211         unsigned long *b;
1212         int rc = 0, i, *cr;
1213
1214         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1215                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1216                 if (rc) {
1217                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1218                                inode->i_ino, *b, *cr, create, rc);
1219                         break;
1220                 }
1221
1222                 b += blocks_per_page;
1223                 cr += blocks_per_page;
1224         }
1225         return rc;
1226 }
1227
1228 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1229                                 int pages, unsigned long *blocks,
1230                                 int *created, int create,
1231                                 struct semaphore *optional_sem)
1232 {
1233         int rc;
1234 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1235         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1236                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1237                                                      blocks, created, create);
1238                 return rc;
1239         }
1240 #endif
1241         if (optional_sem != NULL)
1242                 down(optional_sem);
1243         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1244                                             created, create);
1245         if (optional_sem != NULL)
1246                 up(optional_sem);
1247
1248         return rc;
1249 }
1250
1251 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1252                                    int size, loff_t *offs)
1253 {
1254         struct inode *inode = file->f_dentry->d_inode;
1255         unsigned long block;
1256         struct buffer_head *bh;
1257         int err, blocksize, csize, boffs;
1258
1259         /* prevent reading after eof */
1260         lock_kernel();
1261         if (i_size_read(inode) < *offs + size) {
1262                 size = i_size_read(inode) - *offs;
1263                 unlock_kernel();
1264                 if (size < 0) {
1265                         CERROR("size %llu is too short for read %u@%llu\n",
1266                                i_size_read(inode), size, *offs);
1267                         return -EIO;
1268                 } else if (size == 0) {
1269                         return 0;
1270                 }
1271         } else {
1272                 unlock_kernel();
1273         }
1274
1275         blocksize = 1 << inode->i_blkbits;
1276
1277         while (size > 0) {
1278                 block = *offs >> inode->i_blkbits;
1279                 boffs = *offs & (blocksize - 1);
1280                 csize = min(blocksize - boffs, size);
1281                 bh = ext3_bread(NULL, inode, block, 0, &err);
1282                 if (!bh) {
1283                         CERROR("can't read block: %d\n", err);
1284                         return err;
1285                 }
1286
1287                 memcpy(buf, bh->b_data + boffs, csize);
1288                 brelse(bh);
1289
1290                 *offs += csize;
1291                 buf += csize;
1292                 size -= csize;
1293         }
1294         return 0;
1295 }
1296
1297 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1298                                     loff_t *offs, int force_sync)
1299 {
1300         struct buffer_head *bh = NULL;
1301         unsigned long block;
1302         struct inode *inode = file->f_dentry->d_inode;
1303         loff_t old_size = i_size_read(inode), offset = *offs;
1304         loff_t new_size = i_size_read(inode);
1305         handle_t *handle;
1306         int err = 0, block_count = 0, blocksize, size, boffs;
1307
1308         /* Determine how many transaction credits are needed */
1309         blocksize = 1 << inode->i_blkbits;
1310         block_count = (*offs & (blocksize - 1)) + bufsize;
1311         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1312
1313         lock_24kernel();
1314         handle = fsfilt_ext3_journal_start(inode,
1315                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1316         unlock_24kernel();
1317         if (IS_ERR(handle)) {
1318                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1319                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1320                 return PTR_ERR(handle);
1321         }
1322
1323         while (bufsize > 0) {
1324                 if (bh != NULL)
1325                         brelse(bh);
1326
1327                 block = offset >> inode->i_blkbits;
1328                 boffs = offset & (blocksize - 1);
1329                 size = min(blocksize - boffs, bufsize);
1330                 bh = ext3_bread(handle, inode, block, 1, &err);
1331                 if (!bh) {
1332                         CERROR("can't read/create block: %d\n", err);
1333                         goto out;
1334                 }
1335
1336                 err = ext3_journal_get_write_access(handle, bh);
1337                 if (err) {
1338                         CERROR("journal_get_write_access() returned error %d\n",
1339                                err);
1340                         goto out;
1341                 }
1342                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1343                 memcpy(bh->b_data + boffs, buf, size);
1344                 err = ext3_journal_dirty_metadata(handle, bh);
1345                 if (err) {
1346                         CERROR("journal_dirty_metadata() returned error %d\n",
1347                                err);
1348                         goto out;
1349                 }
1350                 if (offset + size > new_size)
1351                         new_size = offset + size;
1352                 offset += size;
1353                 bufsize -= size;
1354                 buf += size;
1355         }
1356
1357         if (force_sync)
1358                 handle->h_sync = 1; /* recovery likes this */
1359 out:
1360         if (bh)
1361                 brelse(bh);
1362
1363         /* correct in-core and on-disk sizes */
1364         if (new_size > i_size_read(inode)) {
1365                 lock_kernel();
1366                 if (new_size > i_size_read(inode))
1367                         i_size_write(inode, new_size);
1368                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1369                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1370                 if (i_size_read(inode) > old_size)
1371                         mark_inode_dirty(inode);
1372                 unlock_kernel();
1373         }
1374
1375         lock_24kernel();
1376         fsfilt_ext3_journal_stop(handle);
1377         unlock_24kernel();
1378
1379         if (err == 0)
1380                 *offs = offset;
1381         return err;
1382 }
1383
1384 static int fsfilt_ext3_setup(struct super_block *sb)
1385 {
1386         struct ext3_sb_info *sbi = EXT3_SB(sb);
1387 #if 0
1388         sbi->dx_lock = fsfilt_ext3_dx_lock;
1389         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1390 #endif
1391 #ifdef S_PDIROPS
1392         CWARN("Enabling PDIROPS\n");
1393         set_opt(sbi->s_mount_opt, PDIROPS);
1394         sb->s_flags |= S_PDIROPS;
1395 #endif
1396         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1397                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1398 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && HAVE_QUOTA_SUPPORT
1399         /* enable journaled quota support */
1400         /* kfreed in ext3_put_super() */
1401         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user", GFP_KERNEL);
1402         if (!sbi->s_qf_names[USRQUOTA])
1403                 return -ENOMEM;
1404         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group", GFP_KERNEL);
1405         if (!sbi->s_qf_names[GRPQUOTA]) {
1406                 kfree(sbi->s_qf_names[USRQUOTA]);
1407                 sbi->s_qf_names[USRQUOTA] = NULL;
1408                 return -ENOMEM;
1409         }
1410         sbi->s_jquota_fmt = QFMT_VFS_V0;
1411 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1412         set_opt(sbi->s_mount_opt, QUOTA);
1413 #endif
1414 #endif
1415         return 0;
1416 }
1417
1418 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1419    objects. Logs is number of logfiles to update */
1420 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1421 {
1422         if ( !fso ) {
1423                 switch(op) {
1424                 case FSFILT_OP_CREATE:
1425                                  /* directory leaf, index & indirect & EA*/
1426                         return 4 + 3 * logs;
1427                 case FSFILT_OP_UNLINK:
1428                         return 3 * logs;
1429                 }
1430         } else {
1431                 int i;
1432                 int needed = 0;
1433                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1434                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1435                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1436                 for (i = 0; i < op; i++, fso++) {
1437                         int nblocks = fso->fso_bufcnt * blockpp;
1438                         int ndindirect = min(nblocks, addrpp + 1);
1439                         int nindir = nblocks + ndindirect + 1;
1440
1441                         needed += nindir;
1442                 }
1443                 return needed + 3 * logs;
1444         }
1445
1446         return 0;
1447 }
1448
1449 static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
1450
1451 #define DQINFO_COPY(out, in)                    \
1452 do {                                            \
1453         Q_COPY(out, in, dqi_bgrace);            \
1454         Q_COPY(out, in, dqi_igrace);            \
1455         Q_COPY(out, in, dqi_flags);             \
1456         Q_COPY(out, in, dqi_valid);             \
1457 } while (0)
1458
1459 #define DQBLK_COPY(out, in)                     \
1460 do {                                            \
1461         Q_COPY(out, in, dqb_bhardlimit);        \
1462         Q_COPY(out, in, dqb_bsoftlimit);        \
1463         Q_COPY(out, in, dqb_curspace);          \
1464         Q_COPY(out, in, dqb_ihardlimit);        \
1465         Q_COPY(out, in, dqb_isoftlimit);        \
1466         Q_COPY(out, in, dqb_curinodes);         \
1467         Q_COPY(out, in, dqb_btime);             \
1468         Q_COPY(out, in, dqb_itime);             \
1469         Q_COPY(out, in, dqb_valid);             \
1470 } while (0)
1471
1472
1473
1474 static int fsfilt_ext3_quotactl(struct super_block *sb,
1475                                 struct obd_quotactl *oqc)
1476 {
1477         int i, rc = 0, error = 0;
1478         struct quotactl_ops *qcop;
1479         struct if_dqinfo *info;
1480         struct if_dqblk *dqblk;
1481         ENTRY;
1482
1483         if (!sb->s_qcop)
1484                 RETURN(-ENOSYS);
1485
1486         OBD_ALLOC_PTR(info);
1487         if (!info)
1488                 RETURN(-ENOMEM);
1489         OBD_ALLOC_PTR(dqblk);
1490         if (!dqblk) {
1491                 OBD_FREE_PTR(info);
1492                 RETURN(-ENOMEM);
1493         }
1494
1495         DQINFO_COPY(info, &oqc->qc_dqinfo);
1496         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1497
1498         qcop = sb->s_qcop;
1499         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1500                 for (i = 0; i < MAXQUOTAS; i++) {
1501                         if (!Q_TYPESET(oqc, i))
1502                                 continue;
1503
1504                         if (oqc->qc_cmd == Q_QUOTAON) {
1505                                 if (!qcop->quota_on)
1506                                         GOTO(out, rc = -ENOSYS);
1507                                 rc = qcop->quota_on(sb, i, oqc->qc_id,
1508                                                     (char *)op_quotafile[i]);
1509                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1510                                 if (!qcop->quota_off)
1511                                         GOTO(out, rc = -ENOSYS);
1512                                 rc = qcop->quota_off(sb, i);
1513                         }
1514
1515                         if (rc == -EBUSY)
1516                                 error = rc;
1517                         else if (rc)
1518                                 GOTO(out, rc);
1519                 }
1520                 GOTO(out, rc ?: error);
1521         }
1522
1523         switch (oqc->qc_cmd) {
1524         case Q_GETOINFO:
1525         case Q_GETINFO:
1526                 if (!qcop->get_info)
1527                         GOTO(out, rc = -ENOSYS);
1528                 rc = qcop->get_info(sb, oqc->qc_type, info);
1529                 break;
1530         case Q_SETQUOTA:
1531         case Q_INITQUOTA:
1532                 if (!qcop->set_dqblk)
1533                         GOTO(out, rc = -ENOSYS);
1534                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1535                 break;
1536         case Q_GETOQUOTA:
1537         case Q_GETQUOTA:
1538                 if (!qcop->get_dqblk)
1539                         GOTO(out, rc = -ENOSYS);
1540                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1541                 if (!rc)
1542                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1543                 break;
1544         case Q_SYNC:
1545                 if (!sb->s_qcop->quota_sync)
1546                         GOTO(out, rc = -ENOSYS);
1547                 qcop->quota_sync(sb, oqc->qc_type);
1548                 break;
1549         default:
1550                 CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
1551                 LBUG();
1552         }
1553 out:
1554         DQINFO_COPY(&oqc->qc_dqinfo, info);
1555         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1556
1557         OBD_FREE_PTR(info);
1558         OBD_FREE_PTR(dqblk);
1559
1560         if (rc)
1561                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1562                                 "failed: %d\n",
1563                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1564         RETURN(rc);
1565 }
1566
1567 struct chk_dqblk{
1568         struct hlist_node       dqb_hash;        /* quotacheck hash */
1569         struct list_head        dqb_list;        /* in list also */
1570         qid_t                   dqb_id;          /* uid/gid */
1571         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1572         qsize_t                 dqb_bhardlimit;  /* block hard limit */
1573         qsize_t                 dqb_bsoftlimit;  /* block soft limit */
1574         qsize_t                 dqb_curspace;    /* current space */
1575         qsize_t                 dqb_ihardlimit;  /* inode hard limit */
1576         qsize_t                 dqb_isoftlimit;  /* inode soft limit */
1577         qsize_t                 dqb_curinodes;   /* current inodes */
1578         __u64                   dqb_btime;       /* block grace time */
1579         __u64                   dqb_itime;       /* inode grace time */
1580         __u32                   dqb_valid;       /* flag for above fields */
1581 };
1582
1583 static inline unsigned int chkquot_hash(qid_t id, int type)
1584                                         __attribute__((__const__));
1585
1586 static inline unsigned int chkquot_hash(qid_t id, int type)
1587 {
1588         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1589 }
1590
1591 static inline struct chk_dqblk *
1592 find_chkquot(struct hlist_head *head, qid_t id, int type)
1593 {
1594         struct hlist_node *node;
1595         struct chk_dqblk *cdqb;
1596
1597         hlist_for_each(node, head) {
1598                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1599                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1600                         return cdqb;
1601         }
1602
1603         return NULL;
1604 }
1605
1606 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1607 {
1608         struct chk_dqblk *cdqb;
1609
1610         OBD_ALLOC_PTR(cdqb);
1611         if (cdqb) {
1612                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1613                 INIT_LIST_HEAD(&cdqb->dqb_list);
1614                 cdqb->dqb_id = id;
1615                 cdqb->dqb_type = type;
1616         }
1617
1618         return cdqb;
1619 }
1620
1621 static struct chk_dqblk *
1622 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1623       qid_t id, int type, int first_check)
1624 {
1625         struct hlist_head *head = hash + chkquot_hash(id, type);
1626         struct if_dqblk dqb;
1627         struct chk_dqblk *cdqb;
1628         int rc;
1629
1630         cdqb = find_chkquot(head, id, type);
1631         if (cdqb)
1632                 return cdqb;
1633
1634         cdqb = alloc_chkquot(id, type);
1635         if (!cdqb)
1636                 return NULL;
1637
1638         if (!first_check) {
1639                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1640                 if (rc) {
1641                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1642                                id, type, rc);
1643                 } else {
1644                         DQBLK_COPY(cdqb, &dqb);
1645                         cdqb->dqb_curspace = 0;
1646                         cdqb->dqb_curinodes = 0;
1647                 }
1648         }
1649
1650         hlist_add_head(&cdqb->dqb_hash, head);
1651         list_add_tail(&cdqb->dqb_list, list);
1652
1653         return cdqb;
1654 }
1655
1656 static inline int quota_onoff(struct super_block *sb, int cmd, int type)
1657 {
1658         struct obd_quotactl *oqctl;
1659         int rc;
1660
1661         OBD_ALLOC_PTR(oqctl);
1662         if (!oqctl)
1663                 RETURN(-ENOMEM);
1664
1665         oqctl->qc_cmd = cmd;
1666         oqctl->qc_id = QFMT_LDISKFS;
1667         oqctl->qc_type = type;
1668         rc = fsfilt_ext3_quotactl(sb, oqctl);
1669
1670         OBD_FREE_PTR(oqctl);
1671         return rc;
1672 }
1673
1674 static inline int read_old_dqinfo(struct super_block *sb, int type,
1675                                   struct if_dqinfo *dqinfo)
1676 {
1677         struct obd_quotactl *oqctl;
1678         int rc;
1679         ENTRY;
1680
1681         OBD_ALLOC_PTR(oqctl);
1682         if (!oqctl)
1683                 RETURN(-ENOMEM);
1684
1685         oqctl->qc_cmd = Q_GETINFO;
1686         oqctl->qc_type = type;
1687         rc = fsfilt_ext3_quotactl(sb, oqctl);
1688         if (!rc)
1689                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1690
1691         OBD_FREE_PTR(oqctl);
1692         RETURN(rc);
1693 }
1694
1695 static inline struct ext3_group_desc *
1696 get_group_desc(struct super_block *sb, int group)
1697 {
1698         unsigned long desc_block, desc;
1699         struct ext3_group_desc *gdp;
1700
1701         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1702         desc = group % EXT3_DESC_PER_BLOCK(sb);
1703         gdp = (struct ext3_group_desc *)
1704               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1705
1706         return gdp + desc;
1707 }
1708
1709 static inline struct buffer_head *
1710 read_inode_bitmap(struct super_block *sb, unsigned long group)
1711 {
1712         struct ext3_group_desc *desc;
1713         struct buffer_head *bh;
1714
1715         desc = get_group_desc(sb, group);
1716         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1717
1718         return bh;
1719 }
1720
1721 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1722                                      struct buffer_head *bitmap_bh,
1723                                      int index, unsigned long ino)
1724 {
1725         struct inode *inode = NULL;
1726
1727         if (ext3_test_bit(index, bitmap_bh->b_data))
1728                 inode = iget(sb, ino);
1729
1730         return inode;
1731 }
1732
1733 struct qchk_ctxt {
1734         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1735         struct list_head        qckt_list;                   /* quotacheck list */
1736         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1737         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1738 };
1739
1740 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1741                            struct obd_quotactl *oqc)
1742 {
1743         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1744         loff_t size = 0;
1745         qid_t qid[MAXQUOTAS];
1746         int cnt, i, rc = 0;
1747
1748         if (!inode)
1749                 return 0;
1750
1751         qid[USRQUOTA] = inode->i_uid;
1752         qid[GRPQUOTA] = inode->i_gid;
1753
1754         if (S_ISDIR(inode->i_mode) ||
1755             S_ISREG(inode->i_mode) ||
1756             S_ISLNK(inode->i_mode))
1757                 size = inode_get_bytes(inode);
1758
1759         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1760                 if (!Q_TYPESET(oqc, cnt))
1761                         continue;
1762
1763                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1764                                 &qctxt->qckt_list, qid[cnt], cnt,
1765                                 qctxt->qckt_first_check[cnt]);
1766                 if (!cdqb[cnt]) {
1767                         rc = -ENOMEM;
1768                         break;
1769                 }
1770
1771                 cdqb[cnt]->dqb_curspace += size;
1772                 cdqb[cnt]->dqb_curinodes++;
1773         }
1774
1775         if (rc) {
1776                 for (i = 0; i < cnt; i++) {
1777                         if (!Q_TYPESET(oqc, i))
1778                                 continue;
1779                         LASSERT(cdqb[i]);
1780                         cdqb[i]->dqb_curspace -= size;
1781                         cdqb[i]->dqb_curinodes--;
1782                 }
1783         }
1784
1785         return rc;
1786 }
1787
1788 static int v2_write_dqheader(struct file *f, int type)
1789 {
1790         static const __u32 quota_magics[] = V2_INITQMAGICS;
1791         static const __u32 quota_versions[] = V2_INITQVERSIONS;
1792         struct v2_disk_dqheader dqhead;
1793         loff_t offset = 0;
1794
1795         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1796         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1797
1798         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1799         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1800
1801         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1802 }
1803
1804 /* write dqinfo struct in a new quota file */
1805 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1806 {
1807         struct v2_disk_dqinfo dqinfo;
1808         __u32 blocks = V2_DQTREEOFF + 1;
1809         loff_t offset = V2_DQINFOOFF;
1810
1811         if (info) {
1812                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1813                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1814                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1815                                                ~DQF_INFO_DIRTY);
1816         } else {
1817                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1818                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1819                 dqinfo.dqi_flags = 0;
1820         }
1821
1822         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1823         dqinfo.dqi_free_blk = 0;
1824         dqinfo.dqi_free_entry = 0;
1825
1826         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1827 }
1828
1829 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1830                                   struct obd_quotactl *oqc)
1831 {
1832         int i, rc = 0;
1833         ENTRY;
1834
1835         for (i = 0; i < MAXQUOTAS; i++) {
1836                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1837                                          NULL : &qctxt->qckt_dqinfo[i];
1838                 struct file *file;
1839
1840                 if (!Q_TYPESET(oqc, i))
1841                         continue;
1842
1843                 file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
1844                                  0644);
1845                 if (IS_ERR(file)) {
1846                         rc = PTR_ERR(file);
1847                         CERROR("can't create %s file: rc = %d\n",
1848                                op_quotafile[i], rc);
1849                         GOTO(out, rc);
1850                 }
1851
1852                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1853                         CERROR("file %s is not regular", op_quotafile[i]);
1854                         filp_close(file, 0);
1855                         GOTO(out, rc = -EINVAL);
1856                 }
1857
1858                 DQUOT_DROP(file->f_dentry->d_inode);
1859
1860                 rc = v2_write_dqheader(file, i);
1861                 if (rc) {
1862                         filp_close(file, 0);
1863                         GOTO(out, rc);
1864                 }
1865
1866                 rc = v2_write_dqinfo(file, i, info);
1867                 filp_close(file, 0);
1868                 if (rc)
1869                         GOTO(out, rc);
1870         }
1871
1872 out:
1873         RETURN(rc);
1874 }
1875
1876
1877 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1878                           struct chk_dqblk *cdqb)
1879 {
1880         struct obd_quotactl *oqc;
1881         long now;
1882         int rc;
1883         ENTRY;
1884
1885         OBD_ALLOC_PTR(oqc);
1886         if (!oqc)
1887                 RETURN(-ENOMEM);
1888
1889         now = CURRENT_SECONDS;
1890
1891         if (cdqb->dqb_bsoftlimit &&
1892             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1893             !cdqb->dqb_btime)
1894                 cdqb->dqb_btime =
1895                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1896
1897         if (cdqb->dqb_isoftlimit &&
1898             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1899             !cdqb->dqb_itime)
1900                 cdqb->dqb_itime =
1901                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1902
1903         cdqb->dqb_valid = QIF_ALL;
1904
1905         oqc->qc_cmd = Q_SETQUOTA;
1906         oqc->qc_type = cdqb->dqb_type;
1907         oqc->qc_id = cdqb->dqb_id;
1908         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1909
1910         rc = fsfilt_ext3_quotactl(sb, oqc);
1911         OBD_FREE_PTR(oqc);
1912         RETURN(rc);
1913 }
1914
1915 static int prune_chkquots(struct super_block *sb,
1916                           struct qchk_ctxt *qctxt, int error)
1917 {
1918         struct chk_dqblk *cdqb, *tmp;
1919         int rc;
1920
1921         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1922                 if (!error) {
1923                         rc = commit_chkquot(sb, qctxt, cdqb);
1924                         if (rc)
1925                                 error = rc;
1926                 }
1927                 hlist_del_init(&cdqb->dqb_hash);
1928                 list_del(&cdqb->dqb_list);
1929                 OBD_FREE_PTR(cdqb);
1930         }
1931
1932         return error;
1933 }
1934
1935 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1936                                   struct obd_quotactl *oqc)
1937 {
1938         struct ext3_sb_info *sbi = EXT3_SB(sb);
1939         int i, group;
1940         struct qchk_ctxt *qctxt;
1941         struct buffer_head *bitmap_bh = NULL;
1942         unsigned long ino;
1943         struct inode *inode;
1944         int rc = 0;
1945         ENTRY;
1946
1947         /* turn on quota and read dqinfo if existed */
1948         OBD_ALLOC_PTR(qctxt);
1949         if (!qctxt) {
1950                 oqc->qc_stat = -ENOMEM;
1951                 RETURN(-ENOMEM);
1952         }
1953
1954         for (i = 0; i < NR_DQHASH; i++)
1955                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1956         INIT_LIST_HEAD(&qctxt->qckt_list);
1957
1958         for (i = 0; i < MAXQUOTAS; i++) {
1959                 if (!Q_TYPESET(oqc, i))
1960                         continue;
1961
1962                 rc = quota_onoff(sb, Q_QUOTAON, i);
1963                 if (!rc || rc == -EBUSY) {
1964                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1965                         if (rc)
1966                                 GOTO(out, rc);
1967                 } else if (rc == -ENOENT) {
1968                         qctxt->qckt_first_check[i] = 1;
1969                 } else if (rc) {
1970                         GOTO(out, rc);
1971                 }
1972         }
1973
1974         /* check quota and update in hash */
1975         for (group = 0; group < sbi->s_groups_count; group++) {
1976                 ino = group * sbi->s_inodes_per_group + 1;
1977                 bitmap_bh = read_inode_bitmap(sb, group);
1978                 if (!bitmap_bh) {
1979                         CERROR("read_inode_bitmap group %d failed", group);
1980                         GOTO(out, -EIO);
1981                 }
1982
1983                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1984                         if (ino < sbi->s_first_ino)
1985                                 continue;
1986
1987                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1988                         rc = add_inode_quota(inode, qctxt, oqc);
1989                         iput(inode);
1990                         if (rc) {
1991                                 brelse(bitmap_bh);
1992                                 GOTO(out, rc);
1993                         }
1994                 }
1995
1996                 brelse(bitmap_bh);
1997         }
1998
1999         /* read old quota limits from old quota file. (only for the user
2000          * has limits but hasn't file) */
2001 #ifdef HAVE_QUOTA_SUPPORT
2002         for (i = 0; i < MAXQUOTAS; i++) {
2003                 struct list_head id_list;
2004                 struct dquot_id *dqid, *tmp;
2005
2006                 if (!Q_TYPESET(oqc, i))
2007                         continue;
2008
2009                 if (qctxt->qckt_first_check[i])
2010                         continue;
2011
2012
2013                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2014                 INIT_LIST_HEAD(&id_list);
2015 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2016                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2017 #else
2018                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2019 #endif
2020                 if (rc)
2021                         CERROR("read old limits failed. (rc:%d)\n", rc);
2022
2023                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2024                         list_del_init(&dqid->di_link);
2025
2026                         if (!rc)
2027                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2028                                       dqid->di_id, i,
2029                                       qctxt->qckt_first_check[i]);
2030                         kfree(dqid);
2031                 }
2032         }
2033 #endif
2034         /* turn off quota cause we are to dump chk_dqblk to files */
2035         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
2036
2037         rc = create_new_quota_files(qctxt, oqc);
2038         if (rc)
2039                 GOTO(out, rc);
2040
2041         /* we use vfs functions to set dqblk, so turn quota on */
2042         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
2043 out:
2044         /* dump and free chk_dqblk */
2045         rc = prune_chkquots(sb, qctxt, rc);
2046         OBD_FREE_PTR(qctxt);
2047
2048         /* turn off quota, `lfs quotacheck` will turn on when all
2049          * nodes quotacheck finish. */
2050         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
2051
2052         oqc->qc_stat = rc;
2053         if (rc)
2054                 CERROR("quotacheck failed: rc = %d\n", rc);
2055
2056         RETURN(rc);
2057 }
2058
2059 #ifdef HAVE_QUOTA_SUPPORT
2060 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2061                                  int cmd)
2062 {
2063         int rc = 0;
2064         ENTRY;
2065
2066         if (lqi->qi_files[type] == NULL) {
2067                 CERROR("operate qinfo before it's enabled!\n");
2068                 RETURN(-EIO);
2069         }
2070
2071         switch (cmd) {
2072         case QFILE_CHK:
2073                 rc = lustre_check_quota_file(lqi, type);
2074                 break;
2075         case QFILE_RD_INFO:
2076                 rc = lustre_read_quota_info(lqi, type);
2077                 break;
2078         case QFILE_WR_INFO:
2079                 rc = lustre_write_quota_info(lqi, type);
2080                 break;
2081         case QFILE_INIT_INFO:
2082                 rc = lustre_init_quota_info(lqi, type);
2083                 break;
2084         case QFILE_CONVERT:
2085                 rc = lustre_quota_convert(lqi, type);
2086                 break;
2087         default:
2088                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2089                 LBUG();
2090                 break;
2091         }
2092         RETURN(rc);
2093 }
2094
2095 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2096                             struct list_head *list)
2097 {
2098         return lustre_get_qids(file, inode, type, list);
2099 }
2100
2101 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2102 {
2103         int rc = 0;
2104         ENTRY;
2105
2106         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2107                 CERROR("operate dquot before it's enabled!\n");
2108                 RETURN(-EIO);
2109         }
2110
2111         switch (cmd) {
2112         case QFILE_RD_DQUOT:
2113                 rc = lustre_read_dquot(dquot);
2114                 break;
2115         case QFILE_WR_DQUOT:
2116                 if (dquot->dq_dqb.dqb_ihardlimit ||
2117                     dquot->dq_dqb.dqb_isoftlimit ||
2118                     dquot->dq_dqb.dqb_bhardlimit ||
2119                     dquot->dq_dqb.dqb_bsoftlimit)
2120                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2121                 else
2122                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2123
2124                 rc = lustre_commit_dquot(dquot);
2125                 if (rc >= 0)
2126                         rc = 0;
2127                 break;
2128         default:
2129                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2130                 LBUG();
2131                 break;
2132         }
2133         RETURN(rc);
2134 }
2135 #endif
2136
2137 static lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2138 {
2139 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2140         return (EXT3_SB(sb)->journal_bdev);
2141 #else
2142         return kdev_t_to_nr(EXT3_SB(sb)->s_journal->j_dev);
2143 #endif
2144 }
2145
2146 static struct fsfilt_operations fsfilt_ext3_ops = {
2147         .fs_type                = "ext3",
2148         .fs_owner               = THIS_MODULE,
2149         .fs_getlabel            = fsfilt_ext3_get_label,
2150         .fs_setlabel            = fsfilt_ext3_set_label,
2151         .fs_uuid                = fsfilt_ext3_uuid,
2152         .fs_start               = fsfilt_ext3_start,
2153         .fs_brw_start           = fsfilt_ext3_brw_start,
2154         .fs_extend              = fsfilt_ext3_extend,
2155         .fs_commit              = fsfilt_ext3_commit,
2156         .fs_commit_async        = fsfilt_ext3_commit_async,
2157         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2158         .fs_setattr             = fsfilt_ext3_setattr,
2159         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2160         .fs_set_md              = fsfilt_ext3_set_md,
2161         .fs_get_md              = fsfilt_ext3_get_md,
2162         .fs_readpage            = fsfilt_ext3_readpage,
2163         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2164         .fs_statfs              = fsfilt_ext3_statfs,
2165         .fs_sync                = fsfilt_ext3_sync,
2166         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2167         .fs_write_record        = fsfilt_ext3_write_record,
2168         .fs_read_record         = fsfilt_ext3_read_record,
2169         .fs_setup               = fsfilt_ext3_setup,
2170         .fs_send_bio            = fsfilt_ext3_send_bio,
2171         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2172         .fs_quotactl            = fsfilt_ext3_quotactl,
2173         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2174 #ifdef HAVE_DISK_INODE_VERSION
2175         .fs_get_version         = fsfilt_ext3_get_version,
2176         .fs_set_version         = fsfilt_ext3_set_version,
2177 #endif
2178 #ifdef HAVE_QUOTA_SUPPORT
2179         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2180         .fs_qids                = fsfilt_ext3_qids,
2181         .fs_dquot               = fsfilt_ext3_dquot,
2182 #endif
2183         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2184 };
2185
2186 static int __init fsfilt_ext3_init(void)
2187 {
2188         int rc;
2189
2190         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2191                                          sizeof(struct fsfilt_cb_data), 0, 0);
2192         if (!fcb_cache) {
2193                 CERROR("error allocating fsfilt journal callback cache\n");
2194                 GOTO(out, rc = -ENOMEM);
2195         }
2196
2197         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2198
2199         if (rc) {
2200                 int err = cfs_mem_cache_destroy(fcb_cache);
2201                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2202         }
2203 out:
2204         return rc;
2205 }
2206
2207 static void __exit fsfilt_ext3_exit(void)
2208 {
2209         int rc;
2210
2211         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2212         rc = cfs_mem_cache_destroy(fcb_cache);
2213         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2214 }
2215
2216 module_init(fsfilt_ext3_init);
2217 module_exit(fsfilt_ext3_exit);
2218
2219 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2220 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2221 MODULE_LICENSE("GPL");