lustre/lvfs/fsfilt_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/fsfilt_ext3.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/init.h>
  44 #include <linux/module.h>
  45 #include <linux/fs.h>
  46 #include <linux/jbd.h>
  47 #include <linux/slab.h>
  48 #include <linux/pagemap.h>
  49 #include <linux/quotaops.h>
  50 #include <linux/ext3_fs.h>
  51 #include <linux/ext3_jbd.h>
  52 #include <linux/version.h>
  53 #include <linux/bitops.h>
  54 #include <linux/quota.h>
  55 #include <linux/quotaio_v1.h>
  56 #include <linux/quotaio_v2.h>
  57 #include <ext3/xattr.h>
  58
  59 #include <libcfs/libcfs.h>
  60 #include <lustre_fsfilt.h>
  61 #include <obd.h>
  62 #include <lustre_quota.h>
  63 #include <linux/lustre_compat25.h>
  64 #include <linux/lprocfs_status.h>
  65
  66 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  67 #include <linux/ext3_extents.h>
  68 #endif
  69
  70 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  71 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  72 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  73 #else
  74 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  75 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  76 #endif
  77
  78 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
  79 /* for kernels 2.6.18 and later */
  80 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  81 #else
  82 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
  83 #endif
  84
  85 #define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks)
  86 #define fsfilt_ext3_journal_stop(handle)          ext3_journal_stop(handle)
  87
  88 static cfs_mem_cache_t *fcb_cache;
  89
  90 struct fsfilt_cb_data {
  91         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  92         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  93         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  94         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  95         void *cb_data;                  /* MDS/OST completion function data */
  96 };
  97
  98 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
  99 #define EXT3_XATTR_INDEX_TRUSTED        4
 100 #endif
 101
 102 static char *fsfilt_ext3_get_label(struct super_block *sb)
 103 {
 104         return EXT3_SB(sb)->s_es->s_volume_name;
 105 }
 106
 107 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 108 {
 109         /* see e.g. fsfilt_ext3_write_record() */
 110         journal_t *journal;
 111         handle_t *handle;
 112         int err;
 113
 114         journal = EXT3_SB(sb)->s_journal;
 115         handle = journal_start(journal, 1);
 116         if (IS_ERR(handle)) {
 117                 CERROR("can't start transaction\n");
 118                 return(PTR_ERR(handle));
 119         }
 120
 121         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 122         if (err)
 123                 goto out;
 124
 125         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 126                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 127
 128         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 129
 130 out:
 131         journal_stop(handle);
 132
 133         return(err);
 134 }
 135
 136 static char *fsfilt_ext3_uuid(struct super_block *sb)
 137 {
 138         return EXT3_SB(sb)->s_es->s_uuid;
 139 }
 140
 141 #ifdef HAVE_DISK_INODE_VERSION
 142 /*
 143  * Get the 64-bit version for an inode.
 144  */
 145 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 146 {
 147         return EXT3_I(inode)->i_fs_version;
 148 }
 149
 150 /*
 151  * Set the 64-bit version and return the old version.
 152  */
 153 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 154 {
 155         __u64 old_version = EXT3_I(inode)->i_fs_version;
 156
 157         (EXT3_I(inode))->i_fs_version = new_version;
 158         return old_version;
 159 }
 160
 161 #endif
 162
 163 /*
 164  * We don't currently need any additional blocks for rmdir and
 165  * unlink transactions because we are storing the OST oa_id inside
 166  * the inode (which we will be changing anyways as part of this
 167  * transaction).
 168  */
 169 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 170                                int logs)
 171 {
 172         /* For updates to the last received file */
 173         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 174         journal_t *journal;
 175         void *handle;
 176
 177         if (current->journal_info) {
 178                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 179                        current->journal_info);
 180                 goto journal_start;
 181         }
 182
 183         switch(op) {
 184         case FSFILT_OP_RMDIR:
 185         case FSFILT_OP_UNLINK:
 186                 /* delete one file + create/update logs for each stripe */
 187                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 188                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 189                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 190                 break;
 191         case FSFILT_OP_RENAME:
 192                 /* modify additional directory */
 193                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 194                 /* no break */
 195         case FSFILT_OP_SYMLINK:
 196                 /* additional block + block bitmap + GDT for long symlink */
 197                 nblocks += 3;
 198                 /* no break */
 199         case FSFILT_OP_CREATE: {
 200 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 201                 static int warned;
 202                 if (!warned) {
 203                         if (!test_opt(inode->i_sb, EXTENTS)) {
 204                                 warned = 1;
 205                         } else if (((EXT3_I(inode)->i_flags &
 206                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 207                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 208                                 CWARN("extent-mapped directory found - contact "
 209                                       "http://bugzilla.lustre.org/\n");
 210                                 warned = 1;
 211                         }
 212                 }
 213 #endif
 214                 /* no break */
 215         }
 216         case FSFILT_OP_MKDIR:
 217         case FSFILT_OP_MKNOD:
 218                 /* modify one inode + block bitmap + GDT */
 219                 nblocks += 3;
 220                 /* no break */
 221         case FSFILT_OP_LINK:
 222                 /* modify parent directory */
 223                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 224                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 225                 /* create/update logs for each stripe */
 226                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 227                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 228                 break;
 229         case FSFILT_OP_SETATTR:
 230                 /* Setattr on inode */
 231                 nblocks += 1;
 232                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 233                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 234                 /* quota chown log for each stripe */
 235                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 236                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 237                 break;
 238         case FSFILT_OP_CANCEL_UNLINK:
 239                 /* blocks for log header bitmap update OR
 240                  * blocks for catalog header bitmap update + unlink of logs */
 241                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 242                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 243                 break;
 244         case FSFILT_OP_JOIN:
 245                 /* delete 2 file(file + array id) + create 1 file (array id)
 246                  * create/update logs for each stripe */
 247                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 248
 249                 /*create array log for head file*/
 250                 nblocks += 3;
 251                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 252                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
 253                 /*update head file array */
 254                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 255                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 256                 break;
 257         default: CERROR("unknown transaction start op %d\n", op);
 258                 LBUG();
 259         }
 260
 261         LASSERT(current->journal_info == desc_private);
 262         journal = EXT3_SB(inode->i_sb)->s_journal;
 263         if (nblocks > journal->j_max_transaction_buffers) {
 264                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 265                        nblocks, op, logs, journal->j_max_transaction_buffers);
 266                 nblocks = journal->j_max_transaction_buffers;
 267         }
 268
 269  journal_start:
 270         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 271         handle = fsfilt_ext3_journal_start(inode, nblocks);
 272
 273         if (!IS_ERR(handle))
 274                 LASSERT(current->journal_info == handle);
 275         else
 276                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 277                        op, nblocks, PTR_ERR(handle));
 278         return handle;
 279 }
 280
 281 /*
 282  * Calculate the number of buffer credits needed to write multiple pages in
 283  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 284  * doesn't have a nice API for calculating this sort of thing in advance.
 285  *
 286  * See comment above ext3_writepage_trans_blocks for details.  We assume
 287  * no data journaling is being done, but it does allow for all of the pages
 288  * being non-contiguous.  If we are guaranteed contiguous pages we could
 289  * reduce the number of (d)indirect blocks a lot.
 290  *
 291  * With N blocks per page and P pages, for each inode we have at most:
 292  * N*P indirect
 293  * min(N*P, blocksize/4 + 1) dindirect blocks
 294  * niocount tindirect
 295  *
 296  * For the entire filesystem, we have at most:
 297  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 298  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 299  * objcount inode blocks
 300  * 1 superblock
 301  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 302  *
 303  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 304  */
 305 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 306                                       int niocount, struct niobuf_local *nb)
 307 {
 308         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 309         __u64 next_indir;
 310         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 311         int nbitmaps = 0, ngdblocks;
 312         int needed = objcount + 1; /* inodes + superblock */
 313         int i, j;
 314
 315         for (i = 0, j = 0; i < objcount; i++, fso++) {
 316                 /* two or more dindirect blocks in case we cross boundary */
 317                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 318                                     nb[j].offset) >>
 319                                    sb->s_blocksize_bits) /
 320                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 321                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 322
 323                 /* leaf, indirect, tindirect blocks for first block */
 324                 nbitmaps += blockpp + 2;
 325
 326                 j += fso->fso_bufcnt;
 327         }
 328
 329         next_indir = nb[0].offset +
 330                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 331         for (i = 1; i < niocount; i++) {
 332                 if (nb[i].offset >= next_indir) {
 333                         nbitmaps++;     /* additional indirect */
 334                         next_indir = nb[i].offset +
 335                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 336                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 337                         nbitmaps++;     /* additional indirect */
 338                 }
 339                 nbitmaps += blockpp;    /* each leaf in different group? */
 340         }
 341
 342         ngdblocks = nbitmaps;
 343         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 344                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 345         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 346                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 347
 348         needed += nbitmaps + ngdblocks;
 349
 350         /* last_rcvd update */
 351         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 352
 353 #if defined(CONFIG_QUOTA)
 354         /* We assume that there will be 1 bit set in s_dquot.flags for each
 355          * quota file that is active.  This is at least true for now.
 356          */
 357         needed += hweight32(sb_any_quota_enabled(sb)) *
 358                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 359 #endif
 360
 361         return needed;
 362 }
 363
 364 /* We have to start a huge journal transaction here to hold all of the
 365  * metadata for the pages being written here.  This is necessitated by
 366  * the fact that we do lots of prepare_write operations before we do
 367  * any of the matching commit_write operations, so even if we split
 368  * up to use "smaller" transactions none of them could complete until
 369  * all of them were opened.  By having a single journal transaction,
 370  * we eliminate duplicate reservations for common blocks like the
 371  * superblock and group descriptors or bitmaps.
 372  *
 373  * We will start the transaction here, but each prepare_write will
 374  * add a refcount to the transaction, and each commit_write will
 375  * remove a refcount.  The transaction will be closed when all of
 376  * the pages have been written.
 377  */
 378 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 379                                    int niocount, struct niobuf_local *nb,
 380                                    void *desc_private, int logs)
 381 {
 382         journal_t *journal;
 383         handle_t *handle;
 384         int needed;
 385         ENTRY;
 386
 387         LASSERT(current->journal_info == desc_private);
 388         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 389         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 390
 391         /* The number of blocks we could _possibly_ dirty can very large.
 392          * We reduce our request if it is absurd (and we couldn't get that
 393          * many credits for a single handle anyways).
 394          *
 395          * At some point we have to limit the size of I/Os sent at one time,
 396          * increase the size of the journal, or we have to calculate the
 397          * actual journal requirements more carefully by checking all of
 398          * the blocks instead of being maximally pessimistic.  It remains to
 399          * be seen if this is a real problem or not.
 400          */
 401         if (needed > journal->j_max_transaction_buffers) {
 402                 CERROR("want too many journal credits (%d) using %d instead\n",
 403                        needed, journal->j_max_transaction_buffers);
 404                 needed = journal->j_max_transaction_buffers;
 405         }
 406
 407         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 408         handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
 409         if (IS_ERR(handle)) {
 410                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 411                        PTR_ERR(handle));
 412         } else {
 413                 LASSERT(handle->h_buffer_credits >= needed);
 414                 LASSERT(current->journal_info == handle);
 415         }
 416
 417         RETURN(handle);
 418 }
 419
 420 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 421 {
 422        handle_t *handle = h;
 423
 424        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 425        if (nblocks == 0) {
 426                handle->h_buffer_credits = 0;
 427                CWARN("setting credits of handle %p to zero by request\n", h);
 428        }
 429
 430        if (handle->h_buffer_credits > nblocks)
 431                 return 0;
 432        if (journal_extend(handle, nblocks) == 0)
 433                 return 0;
 434
 435        ext3_mark_inode_dirty(handle, inode);
 436        return journal_restart(handle, nblocks);
 437 }
 438
 439 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 440 {
 441         int rc;
 442         handle_t *handle = h;
 443
 444         LASSERT(current->journal_info == handle);
 445         if (force_sync)
 446                 handle->h_sync = 1; /* recovery likes this */
 447
 448         rc = fsfilt_ext3_journal_stop(handle);
 449
 450         return rc;
 451 }
 452
 453 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 454                                     void **wait_handle)
 455 {
 456         unsigned long tid;
 457         transaction_t *transaction;
 458         handle_t *handle = h;
 459         journal_t *journal;
 460         int rc;
 461
 462         LASSERT(current->journal_info == handle);
 463
 464         transaction = handle->h_transaction;
 465         journal = transaction->t_journal;
 466         tid = transaction->t_tid;
 467         /* we don't want to be blocked */
 468         handle->h_sync = 0;
 469         rc = fsfilt_ext3_journal_stop(handle);
 470         if (rc) {
 471                 CERROR("error while stopping transaction: %d\n", rc);
 472                 return rc;
 473         }
 474         log_start_commit(journal, tid);
 475
 476         *wait_handle = (void *) tid;
 477         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 478         return 0;
 479 }
 480
 481 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 482 {
 483         journal_t *journal = EXT3_JOURNAL(inode);
 484         tid_t tid = (tid_t)(long)h;
 485
 486         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 487         if (unlikely(is_journal_aborted(journal)))
 488                 return -EIO;
 489
 490         log_wait_commit(EXT3_JOURNAL(inode), tid);
 491
 492         if (unlikely(is_journal_aborted(journal)))
 493                 return -EIO;
 494         return 0;
 495 }
 496
 497 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 498                                struct iattr *iattr, int do_trunc)
 499 {
 500         struct inode *inode = dentry->d_inode;
 501         int rc = 0;
 502
 503         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 504          * We are already writing the inode to disk as part of this
 505          * transaction and want to avoid a lot of extra inode writeout
 506          * later on. b=9828 */
 507         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 508                 /* ATTR_SIZE would invoke truncate: clear it */
 509                 iattr->ia_valid &= ~ATTR_SIZE;
 510                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 511                 i_size_write(inode, iattr->ia_size);
 512
 513                 if (iattr->ia_valid & ATTR_UID)
 514                         inode->i_uid = iattr->ia_uid;
 515                 if (iattr->ia_valid & ATTR_GID)
 516                         inode->i_gid = iattr->ia_gid;
 517                 if (iattr->ia_valid & ATTR_ATIME)
 518                         inode->i_atime = iattr->ia_atime;
 519                 if (iattr->ia_valid & ATTR_MTIME)
 520                         inode->i_mtime = iattr->ia_mtime;
 521                 if (iattr->ia_valid & ATTR_CTIME)
 522                         inode->i_ctime = iattr->ia_ctime;
 523                 if (iattr->ia_valid & ATTR_MODE) {
 524                         inode->i_mode = iattr->ia_mode;
 525
 526                         if (!in_group_p(inode->i_gid) &&
 527                             !cfs_capable(CFS_CAP_FSETID))
 528                                 inode->i_mode &= ~S_ISGID;
 529                 }
 530
 531                 inode->i_sb->s_op->dirty_inode(inode);
 532
 533                 goto out;
 534         }
 535
 536         /* Don't allow setattr to change file type */
 537         if (iattr->ia_valid & ATTR_MODE)
 538                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 539                                  (iattr->ia_mode & ~S_IFMT);
 540
 541         /* We set these flags on the client, but have already checked perms
 542          * so don't confuse inode_change_ok. */
 543         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 544
 545         if (inode->i_op->setattr) {
 546                 rc = inode->i_op->setattr(dentry, iattr);
 547         } else {
 548                 rc = inode_change_ok(inode, iattr);
 549                 if (!rc)
 550                         rc = inode_setattr(inode, iattr);
 551         }
 552
 553  out:
 554         RETURN(rc);
 555 }
 556
 557 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 558                                  unsigned int cmd, unsigned long arg)
 559 {
 560         int rc = 0;
 561         ENTRY;
 562
 563         /* FIXME: Can't do this because of nested transaction deadlock */
 564         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 565                 CERROR("can't set data journal flag on file\n");
 566                 RETURN(-EPERM);
 567         }
 568
 569         if (inode->i_fop->ioctl)
 570                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 571         else
 572                 RETURN(-ENOTTY);
 573
 574         RETURN(rc);
 575 }
 576
 577 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 578                               void *lmm, int lmm_size, const char *name)
 579 {
 580         int rc;
 581
 582         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 583
 584         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 585                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 586
 587
 588         if (rc && rc != -EROFS)
 589                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 590                        inode->i_ino, rc);
 591         return rc;
 592 }
 593
 594 /* Must be called with i_mutex held */
 595 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 596                               const char *name)
 597 {
 598         int rc;
 599
 600         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 601
 602         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 603                             name, lmm, lmm_size);
 604
 605         /* This gives us the MD size */
 606         if (lmm == NULL)
 607                 return (rc == -ENODATA) ? 0 : rc;
 608
 609         if (rc < 0) {
 610                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 611                        EXT3_XATTR_INDEX_TRUSTED, name,
 612                        inode->i_ino, rc);
 613                 memset(lmm, 0, lmm_size);
 614                 return (rc == -ENODATA) ? 0 : rc;
 615         }
 616
 617         return rc;
 618 }
 619
 620 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 621 {
 622         submit_bio(rw, bio);
 623         return 0;
 624 }
 625
 626 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 627                                     loff_t *off)
 628 {
 629         struct inode *inode = file->f_dentry->d_inode;
 630         int rc = 0;
 631
 632         if (S_ISREG(inode->i_mode))
 633                 rc = file->f_op->read(file, buf, count, off);
 634         else {
 635                 const int blkbits = inode->i_sb->s_blocksize_bits;
 636                 const int blksize = inode->i_sb->s_blocksize;
 637
 638                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 639                        count, inode->i_ino, *off);
 640                 while (count > 0) {
 641                         struct buffer_head *bh;
 642
 643                         bh = NULL;
 644                         if (*off < i_size_read(inode)) {
 645                                 int err = 0;
 646
 647                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 648                                                 0, &err);
 649
 650                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 651
 652                                 if (bh) {
 653                                         memcpy(buf, bh->b_data, blksize);
 654                                         brelse(bh);
 655                                 } else if (err) {
 656                                         /* XXX in theory we should just fake
 657                                          * this buffer and continue like ext3,
 658                                          * especially if this is a partial read
 659                                          */
 660                                         CERROR("error read dir %lu+%llu: %d\n",
 661                                                inode->i_ino, *off, err);
 662                                         RETURN(err);
 663                                 }
 664                         }
 665                         if (!bh) {
 666                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 667
 668                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 669                                 memset(fake, 0, sizeof(*fake));
 670                                 fake->rec_len = cpu_to_le16(blksize);
 671                         }
 672                         count -= blksize;
 673                         buf += blksize;
 674                         *off += blksize;
 675                         rc += blksize;
 676                 }
 677         }
 678
 679         return rc;
 680 }
 681
 682 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 683 {
 684         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 685
 686         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 687
 688         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 689 }
 690
 691 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 692                                       void *handle, fsfilt_cb_t cb_func,
 693                                       void *cb_data)
 694 {
 695         struct fsfilt_cb_data *fcb;
 696
 697         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 698         if (fcb == NULL)
 699                 RETURN(-ENOMEM);
 700
 701         fcb->cb_func = cb_func;
 702         fcb->cb_obd = obd;
 703         fcb->cb_last_rcvd = last_rcvd;
 704         fcb->cb_data = cb_data;
 705
 706         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 707         journal_callback_set(handle, fsfilt_ext3_cb_func,
 708                              (struct journal_callback *)fcb);
 709
 710         return 0;
 711 }
 712
 713 /*
 714  * We need to hack the return value for the free inode counts because
 715  * the current EA code requires one filesystem block per inode with EAs,
 716  * so it is possible to run out of blocks before we run out of inodes.
 717  *
 718  * This can be removed when the ext3 EA code is fixed.
 719  */
 720 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 721 {
 722         struct kstatfs sfs;
 723         int rc;
 724
 725         memset(&sfs, 0, sizeof(sfs));
 726
 727         rc = ll_do_statfs(sb, &sfs);
 728
 729         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 730                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 731                 sfs.f_ffree = sfs.f_bfree;
 732         }
 733
 734         statfs_pack(osfs, &sfs);
 735         return rc;
 736 }
 737
 738 static int fsfilt_ext3_sync(struct super_block *sb)
 739 {
 740         return ext3_force_commit(sb);
 741 }
 742
 743 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 744 #warning "kernel code has old extents/mballoc patch, disabling"
 745 #undef EXT3_MULTIBLOCK_ALLOCATOR
 746 #endif
 747 #ifndef EXT3_EXTENTS_FL
 748 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 749 #endif
 750
 751 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 752 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 753 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 754 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 755 #else
 756 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 757 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 758 #endif
 759
 760 #ifndef EXT_ASSERT
 761 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 762 #endif
 763
 764 #ifdef EXT3_EXT_HAS_NO_TREE
 765 /* for kernels 2.6.18 and later */
 766 #define ext3_ext_base                   inode
 767 #define ext3_ext_base2inode(inode)      (inode)
 768 #define EXT_DEPTH(inode)                ext_depth(inode)
 769 #define EXT_GENERATION(inode)           ext_generation(inode)
 770 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 771                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 772 #else
 773 #define ext3_ext_base                   ext3_extents_tree
 774 #define ext3_ext_base2inode(tree)       (tree->inode)
 775 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 776                         ext3_ext_walk_space(tree, block, num, cb);
 777 #endif
 778
 779 #include <linux/lustre_version.h>
 780 #if EXT3_EXT_MAGIC == 0xf301
 781 #define ee_start e_start
 782 #define ee_block e_block
 783 #define ee_len   e_num
 784 #endif
 785 #ifndef EXT3_BB_MAX_BLOCKS
 786 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 787         ext3_new_blocks(handle, inode, count, goal, err)
 788 #endif
 789
 790 struct bpointers {
 791         unsigned long *blocks;
 792         int *created;
 793         unsigned long start;
 794         int num;
 795         int init_num;
 796         int create;
 797 };
 798
 799 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 800                               unsigned long block, int *aflags)
 801 {
 802         struct ext3_inode_info *ei = EXT3_I(inode);
 803         unsigned long bg_start;
 804         unsigned long colour;
 805         int depth;
 806
 807         if (path) {
 808                 struct ext3_extent *ex;
 809                 depth = path->p_depth;
 810
 811                 /* try to predict block placement */
 812                 if ((ex = path[depth].p_ext)) {
 813 #if 0
 814                         /* This prefers to eat into a contiguous extent
 815                          * rather than find an extent that the whole
 816                          * request will fit into.  This can fragment data
 817                          * block allocation and prevents our lovely 1M I/Os
 818                          * from reaching the disk intact. */
 819                         if (ex->ee_block + ex->ee_len == block)
 820                                 *aflags |= 1;
 821 #endif
 822                         return ex->ee_start + (block - ex->ee_block);
 823                 }
 824
 825                 /* it looks index is empty
 826                  * try to find starting from index itself */
 827                 if (path[depth].p_bh)
 828                         return path[depth].p_bh->b_blocknr;
 829         }
 830
 831         /* OK. use inode's group */
 832         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 833                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 834         colour = (current->pid % 16) *
 835                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 836         return bg_start + colour + block;
 837 }
 838
 839 #define ll_unmap_underlying_metadata(sb, blocknr) \
 840         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 841
 842 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 843 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 844                                 struct ext3_ext_path *path, unsigned long block,
 845                                 unsigned long *count, int *err)
 846 {
 847         unsigned long pblock, goal;
 848         int aflags = 0;
 849         struct inode *inode = ext3_ext_base2inode(base);
 850
 851         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 852         aflags |= 2; /* block have been already reserved */
 853         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 854         return pblock;
 855
 856 }
 857 #else
 858 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 859                                 struct ext3_ext_path *path, unsigned long block,
 860                                 unsigned long *count, int *err)
 861 {
 862         struct inode *inode = ext3_ext_base2inode(base);
 863         struct ext3_allocation_request ar;
 864         unsigned long pblock;
 865         int aflags;
 866
 867         /* find neighbour allocated blocks */
 868         ar.lleft = block;
 869         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 870         if (*err)
 871                 return 0;
 872         ar.lright = block;
 873         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 874         if (*err)
 875                 return 0;
 876
 877         /* allocate new block */
 878         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 879         ar.inode = inode;
 880         ar.logical = block;
 881         ar.len = *count;
 882         ar.flags = EXT3_MB_HINT_DATA;
 883         pblock = ext3_mb_new_blocks(handle, &ar, err);
 884         *count = ar.len;
 885         return pblock;
 886
 887 }
 888 #endif
 889
 890 #ifdef EXT3_EXT_HAS_NO_TREE
 891 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 892                                   struct ext3_ext_path *path,
 893                                   struct ext3_ext_cache *cex,
 894 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 895                                    struct ext3_extent *ex,
 896 #endif
 897                                   void *cbdata)
 898 {
 899         struct bpointers *bp = cbdata;
 900 #else
 901 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 902                                   struct ext3_ext_path *path,
 903                                   struct ext3_ext_cache *cex
 904 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 905                                   , struct ext3_extent *ex
 906 #endif
 907                                  )
 908 {
 909         struct bpointers *bp = base->private;
 910 #endif
 911         struct inode *inode = ext3_ext_base2inode(base);
 912         struct ext3_extent nex;
 913         unsigned long pblock;
 914         unsigned long tgen;
 915         int err, i;
 916         unsigned long count;
 917         handle_t *handle;
 918
 919         i = EXT_DEPTH(base);
 920         EXT_ASSERT(i == path->p_depth);
 921         EXT_ASSERT(path[i].p_hdr);
 922
 923         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 924                 err = EXT_CONTINUE;
 925                 goto map;
 926         }
 927
 928         if (bp->create == 0) {
 929                 i = 0;
 930                 if (cex->ec_block < bp->start)
 931                         i = bp->start - cex->ec_block;
 932                 if (i >= cex->ec_len)
 933                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 934                                         i, cex->ec_len);
 935                 for (; i < cex->ec_len && bp->num; i++) {
 936                         *(bp->created) = 0;
 937                         bp->created++;
 938                         *(bp->blocks) = 0;
 939                         bp->blocks++;
 940                         bp->num--;
 941                         bp->start++;
 942                 }
 943
 944                 return EXT_CONTINUE;
 945         }
 946
 947         tgen = EXT_GENERATION(base);
 948         count = ext3_ext_calc_credits_for_insert(base, path);
 949         ext3_up_truncate_sem(inode);
 950
 951         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 952         if (IS_ERR(handle)) {
 953                 ext3_down_truncate_sem(inode);
 954                 return PTR_ERR(handle);
 955         }
 956
 957         ext3_down_truncate_sem(inode);
 958         if (tgen != EXT_GENERATION(base)) {
 959                 /* the tree has changed. so path can be invalid at moment */
 960                 fsfilt_ext3_journal_stop(handle);
 961                 return EXT_REPEAT;
 962         }
 963
 964         count = cex->ec_len;
 965         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 966         if (!pblock)
 967                 goto out;
 968         EXT_ASSERT(count <= cex->ec_len);
 969
 970         /* insert new extent */
 971         nex.ee_block = cex->ec_block;
 972         nex.ee_start = pblock;
 973         nex.ee_len = count;
 974         err = ext3_ext_insert_extent(handle, base, path, &nex);
 975         if (err) {
 976                 /* free data blocks we just allocated */
 977                 /* not a good idea to call discard here directly,
 978                  * but otherwise we'd need to call it every free() */
 979 #ifdef EXT3_MB_HINT_GROUP_ALLOC
 980                 ext3_mb_discard_inode_preallocations(inode);
 981 #endif
 982                 ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);
 983                 goto out;
 984         }
 985
 986         /*
 987          * Putting len of the actual extent we just inserted,
 988          * we are asking ext3_ext_walk_space() to continue
 989          * scaning after that block
 990          */
 991         cex->ec_len = nex.ee_len;
 992         cex->ec_start = nex.ee_start;
 993         BUG_ON(nex.ee_len == 0);
 994         BUG_ON(nex.ee_block != cex->ec_block);
 995
 996 out:
 997         fsfilt_ext3_journal_stop(handle);
 998 map:
 999         if (err >= 0) {
1000                 /* map blocks */
1001                 if (bp->num == 0) {
1002                         CERROR("hmm. why do we find this extent?\n");
1003                         CERROR("initial space: %lu:%u\n",
1004                                 bp->start, bp->init_num);
1005                         CERROR("current extent: %u/%u/%u %d\n",
1006                                 cex->ec_block, cex->ec_len,
1007                                 cex->ec_start, cex->ec_type);
1008                 }
1009                 i = 0;
1010                 if (cex->ec_block < bp->start)
1011                         i = bp->start - cex->ec_block;
1012                 if (i >= cex->ec_len)
1013                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1014                                         i, cex->ec_len);
1015                 for (; i < cex->ec_len && bp->num; i++) {
1016                         *(bp->blocks) = cex->ec_start + i;
1017                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1018                                 *(bp->created) = 0;
1019                         } else {
1020                                 *(bp->created) = 1;
1021                                 /* unmap any possible underlying metadata from
1022                                  * the block device mapping.  bug 6998. */
1023                                 ll_unmap_underlying_metadata(inode->i_sb,
1024                                                              *(bp->blocks));
1025                         }
1026                         bp->created++;
1027                         bp->blocks++;
1028                         bp->num--;
1029                         bp->start++;
1030                 }
1031         }
1032         return err;
1033 }
1034
1035 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1036                        unsigned long num, unsigned long *blocks,
1037                        int *created, int create)
1038 {
1039 #ifdef EXT3_EXT_HAS_NO_TREE
1040         struct ext3_ext_base *base = inode;
1041 #else
1042         struct ext3_extents_tree tree;
1043         struct ext3_ext_base *base = &tree;
1044 #endif
1045         struct bpointers bp;
1046         int err;
1047
1048         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1049                block, block + num - 1, (unsigned) inode->i_ino);
1050
1051 #ifndef EXT3_EXT_HAS_NO_TREE
1052         ext3_init_tree_desc(base, inode);
1053         tree.private = &bp;
1054 #endif
1055         bp.blocks = blocks;
1056         bp.created = created;
1057         bp.start = block;
1058         bp.init_num = bp.num = num;
1059         bp.create = create;
1060
1061         ext3_down_truncate_sem(inode);
1062         err = fsfilt_ext3_ext_walk_space(base, block, num,
1063                                          ext3_ext_new_extent_cb, &bp);
1064         ext3_ext_invalidate_cache(base);
1065         ext3_up_truncate_sem(inode);
1066
1067         return err;
1068 }
1069
1070 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1071                                     int pages, unsigned long *blocks,
1072                                     int *created, int create)
1073 {
1074         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1075         int rc = 0, i = 0;
1076         struct page *fp = NULL;
1077         int clen = 0;
1078
1079         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1080                 inode->i_ino, pages, (*page)->index);
1081
1082         /* pages are sorted already. so, we just have to find
1083          * contig. space and process them properly */
1084         while (i < pages) {
1085                 if (fp == NULL) {
1086                         /* start new extent */
1087                         fp = *page++;
1088                         clen = 1;
1089                         i++;
1090                         continue;
1091                 } else if (fp->index + clen == (*page)->index) {
1092                         /* continue the extent */
1093                         page++;
1094                         clen++;
1095                         i++;
1096                         continue;
1097                 }
1098
1099                 /* process found extent */
1100                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1101                                         clen * blocks_per_page, blocks,
1102                                         created, create);
1103                 if (rc)
1104                         GOTO(cleanup, rc);
1105
1106                 /* look for next extent */
1107                 fp = NULL;
1108                 blocks += blocks_per_page * clen;
1109                 created += blocks_per_page * clen;
1110         }
1111
1112         if (fp)
1113                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1114                                         clen * blocks_per_page, blocks,
1115                                         created, create);
1116 cleanup:
1117         return rc;
1118 }
1119 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1120
1121 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1122                                unsigned long *blocks, int *created, int create);
1123 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1124                                    int pages, unsigned long *blocks,
1125                                    int *created, int create)
1126 {
1127         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1128         unsigned long *b;
1129         int rc = 0, i, *cr;
1130
1131         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1132                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1133                 if (rc) {
1134                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1135                                inode->i_ino, *b, *cr, create, rc);
1136                         break;
1137                 }
1138
1139                 b += blocks_per_page;
1140                 cr += blocks_per_page;
1141         }
1142         return rc;
1143 }
1144
1145 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1146                                 int pages, unsigned long *blocks,
1147                                 int *created, int create,
1148                                 struct semaphore *optional_sem)
1149 {
1150         int rc;
1151 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1152         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1153                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1154                                                      blocks, created, create);
1155                 return rc;
1156         }
1157 #endif
1158         if (optional_sem != NULL)
1159                 down(optional_sem);
1160         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1161                                             created, create);
1162         if (optional_sem != NULL)
1163                 up(optional_sem);
1164
1165         return rc;
1166 }
1167
1168 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1169 {
1170         unsigned long block;
1171         struct buffer_head *bh;
1172         int err, blocksize, csize, boffs, osize = size;
1173
1174         /* prevent reading after eof */
1175         lock_kernel();
1176         if (i_size_read(inode) < *offs + size) {
1177                 size = i_size_read(inode) - *offs;
1178                 unlock_kernel();
1179                 if (size < 0) {
1180                         CERROR("size %llu is too short for read %u@%llu\n",
1181                                i_size_read(inode), size, *offs);
1182                         return -EIO;
1183                 } else if (size == 0) {
1184                         return 0;
1185                 }
1186         } else {
1187                 unlock_kernel();
1188         }
1189
1190         blocksize = 1 << inode->i_blkbits;
1191
1192         while (size > 0) {
1193                 block = *offs >> inode->i_blkbits;
1194                 boffs = *offs & (blocksize - 1);
1195                 csize = min(blocksize - boffs, size);
1196                 bh = ext3_bread(NULL, inode, block, 0, &err);
1197                 if (!bh) {
1198                         CERROR("can't read block: %d\n", err);
1199                         return err;
1200                 }
1201
1202                 memcpy(buf, bh->b_data + boffs, csize);
1203                 brelse(bh);
1204
1205                 *offs += csize;
1206                 buf += csize;
1207                 size -= csize;
1208         }
1209         return osize;
1210 }
1211 EXPORT_SYMBOL(fsfilt_ext3_read);
1212
1213 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1214                                    int size, loff_t *offs)
1215 {
1216         int rc;
1217         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1218         if (rc > 0)
1219                 rc = 0;
1220         return rc;
1221 }
1222
1223 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1224                                 loff_t *offs, handle_t *handle)
1225 {
1226         struct buffer_head *bh = NULL;
1227         loff_t old_size = i_size_read(inode), offset = *offs;
1228         loff_t new_size = i_size_read(inode);
1229         unsigned long block;
1230         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1231
1232         while (bufsize > 0) {
1233                 if (bh != NULL)
1234                         brelse(bh);
1235
1236                 block = offset >> inode->i_blkbits;
1237                 boffs = offset & (blocksize - 1);
1238                 size = min(blocksize - boffs, bufsize);
1239                 bh = ext3_bread(handle, inode, block, 1, &err);
1240                 if (!bh) {
1241                         CERROR("can't read/create block: %d\n", err);
1242                         break;
1243                 }
1244
1245                 err = ext3_journal_get_write_access(handle, bh);
1246                 if (err) {
1247                         CERROR("journal_get_write_access() returned error %d\n",
1248                                err);
1249                         break;
1250                 }
1251                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1252                 memcpy(bh->b_data + boffs, buf, size);
1253                 err = ext3_journal_dirty_metadata(handle, bh);
1254                 if (err) {
1255                         CERROR("journal_dirty_metadata() returned error %d\n",
1256                                err);
1257                         break;
1258                 }
1259                 if (offset + size > new_size)
1260                         new_size = offset + size;
1261                 offset += size;
1262                 bufsize -= size;
1263                 buf += size;
1264         }
1265         if (bh)
1266                 brelse(bh);
1267
1268         /* correct in-core and on-disk sizes */
1269         if (new_size > i_size_read(inode)) {
1270                 lock_kernel();
1271                 if (new_size > i_size_read(inode))
1272                         i_size_write(inode, new_size);
1273                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1274                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1275                 if (i_size_read(inode) > old_size)
1276                         mark_inode_dirty(inode);
1277                 unlock_kernel();
1278         }
1279
1280         if (err == 0)
1281                 *offs = offset;
1282         return err;
1283 }
1284 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1285
1286 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1287                                     loff_t *offs, int force_sync)
1288 {
1289         struct inode *inode = file->f_dentry->d_inode;
1290         handle_t *handle;
1291         int err, block_count = 0, blocksize;
1292
1293         /* Determine how many transaction credits are needed */
1294         blocksize = 1 << inode->i_blkbits;
1295         block_count = (*offs & (blocksize - 1)) + bufsize;
1296         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1297
1298         handle = fsfilt_ext3_journal_start(inode,
1299                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1300         if (IS_ERR(handle)) {
1301                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1302                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1303                 return PTR_ERR(handle);
1304         }
1305
1306         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1307
1308         if (!err && force_sync)
1309                 handle->h_sync = 1; /* recovery likes this */
1310
1311         fsfilt_ext3_journal_stop(handle);
1312
1313         return err;
1314 }
1315
1316 static int fsfilt_ext3_setup(struct super_block *sb)
1317 {
1318 #if 0
1319         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1320         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1321 #endif
1322 #ifdef S_PDIROPS
1323         CWARN("Enabling PDIROPS\n");
1324         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1325         sb->s_flags |= S_PDIROPS;
1326 #endif
1327         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1328                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1329 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13)) && defined(HAVE_QUOTA_SUPPORT)
1330         set_opt(EXT3_SB(sb)->s_mount_opt, QUOTA);
1331 #endif
1332         return 0;
1333 }
1334
1335 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1336    objects. Logs is number of logfiles to update */
1337 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1338 {
1339         if ( !fso ) {
1340                 switch(op) {
1341                 case FSFILT_OP_CREATE:
1342                                  /* directory leaf, index & indirect & EA*/
1343                         return 4 + 3 * logs;
1344                 case FSFILT_OP_UNLINK:
1345                         return 3 * logs;
1346                 }
1347         } else {
1348                 int i;
1349                 int needed = 0;
1350                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1351                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1352                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1353                 for (i = 0; i < op; i++, fso++) {
1354                         int nblocks = fso->fso_bufcnt * blockpp;
1355                         int ndindirect = min(nblocks, addrpp + 1);
1356                         int nindir = nblocks + ndindirect + 1;
1357
1358                         needed += nindir;
1359                 }
1360                 return needed + 3 * logs;
1361         }
1362
1363         return 0;
1364 }
1365
1366 static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
1367
1368 #define DQINFO_COPY(out, in)                    \
1369 do {                                            \
1370         Q_COPY(out, in, dqi_bgrace);            \
1371         Q_COPY(out, in, dqi_igrace);            \
1372         Q_COPY(out, in, dqi_flags);             \
1373         Q_COPY(out, in, dqi_valid);             \
1374 } while (0)
1375
1376 #define DQBLK_COPY(out, in)                     \
1377 do {                                            \
1378         Q_COPY(out, in, dqb_bhardlimit);        \
1379         Q_COPY(out, in, dqb_bsoftlimit);        \
1380         Q_COPY(out, in, dqb_curspace);          \
1381         Q_COPY(out, in, dqb_ihardlimit);        \
1382         Q_COPY(out, in, dqb_isoftlimit);        \
1383         Q_COPY(out, in, dqb_curinodes);         \
1384         Q_COPY(out, in, dqb_btime);             \
1385         Q_COPY(out, in, dqb_itime);             \
1386         Q_COPY(out, in, dqb_valid);             \
1387 } while (0)
1388
1389
1390
1391 static int fsfilt_ext3_quotactl(struct super_block *sb,
1392                                 struct obd_quotactl *oqc)
1393 {
1394         int i, rc = 0, error = 0;
1395         struct quotactl_ops *qcop;
1396         struct if_dqinfo *info;
1397         struct if_dqblk *dqblk;
1398         ENTRY;
1399
1400         if (!sb->s_qcop)
1401                 RETURN(-ENOSYS);
1402
1403         OBD_ALLOC_PTR(info);
1404         if (!info)
1405                 RETURN(-ENOMEM);
1406         OBD_ALLOC_PTR(dqblk);
1407         if (!dqblk) {
1408                 OBD_FREE_PTR(info);
1409                 RETURN(-ENOMEM);
1410         }
1411
1412         DQINFO_COPY(info, &oqc->qc_dqinfo);
1413         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1414
1415         qcop = sb->s_qcop;
1416         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1417                 for (i = 0; i < MAXQUOTAS; i++) {
1418                         if (!Q_TYPESET(oqc, i))
1419                                 continue;
1420
1421                         if (oqc->qc_cmd == Q_QUOTAON) {
1422                                 if (!qcop->quota_on)
1423                                         GOTO(out, rc = -ENOSYS);
1424                                 rc = qcop->quota_on(sb, i, oqc->qc_id,
1425                                                     (char *)op_quotafile[i]);
1426                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1427                                 if (!qcop->quota_off)
1428                                         GOTO(out, rc = -ENOSYS);
1429                                 rc = qcop->quota_off(sb, i);
1430                         }
1431
1432                         if (rc == -EBUSY)
1433                                 error = rc;
1434                         else if (rc)
1435                                 GOTO(out, rc);
1436                 }
1437                 GOTO(out, rc ?: error);
1438         }
1439
1440         switch (oqc->qc_cmd) {
1441         case Q_GETOINFO:
1442         case Q_GETINFO:
1443                 if (!qcop->get_info)
1444                         GOTO(out, rc = -ENOSYS);
1445                 rc = qcop->get_info(sb, oqc->qc_type, info);
1446                 break;
1447         case Q_SETQUOTA:
1448         case Q_INITQUOTA:
1449                 if (!qcop->set_dqblk)
1450                         GOTO(out, rc = -ENOSYS);
1451                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1452                 break;
1453         case Q_GETOQUOTA:
1454         case Q_GETQUOTA:
1455                 if (!qcop->get_dqblk)
1456                         GOTO(out, rc = -ENOSYS);
1457                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1458                 break;
1459         case Q_SYNC:
1460                 if (!sb->s_qcop->quota_sync)
1461                         GOTO(out, rc = -ENOSYS);
1462                 qcop->quota_sync(sb, oqc->qc_type);
1463                 break;
1464         default:
1465                 CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
1466                 LBUG();
1467         }
1468 out:
1469         DQINFO_COPY(&oqc->qc_dqinfo, info);
1470         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1471
1472         OBD_FREE_PTR(info);
1473         OBD_FREE_PTR(dqblk);
1474
1475         if (rc)
1476                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1477                                 "failed: %d\n",
1478                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1479         RETURN(rc);
1480 }
1481
1482 struct chk_dqblk{
1483         struct hlist_node       dqb_hash;        /* quotacheck hash */
1484         struct list_head        dqb_list;        /* in list also */
1485         qid_t                   dqb_id;          /* uid/gid */
1486         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1487         __u32                   dqb_bhardlimit;  /* block hard limit */
1488         __u32                   dqb_bsoftlimit;  /* block soft limit */
1489         qsize_t                 dqb_curspace;    /* current space */
1490         __u32                   dqb_ihardlimit;  /* inode hard limit */
1491         __u32                   dqb_isoftlimit;  /* inode soft limit */
1492         __u32                   dqb_curinodes;   /* current inodes */
1493         __u64                   dqb_btime;       /* block grace time */
1494         __u64                   dqb_itime;       /* inode grace time */
1495         __u32                   dqb_valid;       /* flag for above fields */
1496 };
1497
1498 static inline unsigned int chkquot_hash(qid_t id, int type)
1499                                         __attribute__((__const__));
1500
1501 static inline unsigned int chkquot_hash(qid_t id, int type)
1502 {
1503         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1504 }
1505
1506 static inline struct chk_dqblk *
1507 find_chkquot(struct hlist_head *head, qid_t id, int type)
1508 {
1509         struct hlist_node *node;
1510         struct chk_dqblk *cdqb;
1511
1512         hlist_for_each(node, head) {
1513                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1514                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1515                         return cdqb;
1516         }
1517
1518         return NULL;
1519 }
1520
1521 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1522 {
1523         struct chk_dqblk *cdqb;
1524
1525         OBD_ALLOC_PTR(cdqb);
1526         if (cdqb) {
1527                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1528                 INIT_LIST_HEAD(&cdqb->dqb_list);
1529                 cdqb->dqb_id = id;
1530                 cdqb->dqb_type = type;
1531         }
1532
1533         return cdqb;
1534 }
1535
1536 static struct chk_dqblk *
1537 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1538       qid_t id, int type, int first_check)
1539 {
1540         struct hlist_head *head = hash + chkquot_hash(id, type);
1541         struct if_dqblk dqb;
1542         struct chk_dqblk *cdqb;
1543         int rc;
1544
1545         cdqb = find_chkquot(head, id, type);
1546         if (cdqb)
1547                 return cdqb;
1548
1549         cdqb = alloc_chkquot(id, type);
1550         if (!cdqb)
1551                 return NULL;
1552
1553         if (!first_check) {
1554                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1555                 if (rc) {
1556                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1557                                id, type, rc);
1558                 } else {
1559                         DQBLK_COPY(cdqb, &dqb);
1560                         cdqb->dqb_curspace = 0;
1561                         cdqb->dqb_curinodes = 0;
1562                 }
1563         }
1564
1565         hlist_add_head(&cdqb->dqb_hash, head);
1566         list_add_tail(&cdqb->dqb_list, list);
1567
1568         return cdqb;
1569 }
1570
1571 static inline int quota_onoff(struct super_block *sb, int cmd, int type)
1572 {
1573         struct obd_quotactl *oqctl;
1574         int rc;
1575
1576         OBD_ALLOC_PTR(oqctl);
1577         if (!oqctl)
1578                 RETURN(-ENOMEM);
1579
1580         oqctl->qc_cmd = cmd;
1581         oqctl->qc_id = QFMT_LDISKFS;
1582         oqctl->qc_type = type;
1583         rc = fsfilt_ext3_quotactl(sb, oqctl);
1584
1585         OBD_FREE_PTR(oqctl);
1586         return rc;
1587 }
1588
1589 static inline int read_old_dqinfo(struct super_block *sb, int type,
1590                                   struct if_dqinfo *dqinfo)
1591 {
1592         struct obd_quotactl *oqctl;
1593         int rc;
1594         ENTRY;
1595
1596         OBD_ALLOC_PTR(oqctl);
1597         if (!oqctl)
1598                 RETURN(-ENOMEM);
1599
1600         oqctl->qc_cmd = Q_GETINFO;
1601         oqctl->qc_type = type;
1602         rc = fsfilt_ext3_quotactl(sb, oqctl);
1603         if (!rc)
1604                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1605
1606         OBD_FREE_PTR(oqctl);
1607         RETURN(rc);
1608 }
1609
1610 static inline struct ext3_group_desc *
1611 get_group_desc(struct super_block *sb, int group)
1612 {
1613         unsigned long desc_block, desc;
1614         struct ext3_group_desc *gdp;
1615
1616         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1617         desc = group % EXT3_DESC_PER_BLOCK(sb);
1618         gdp = (struct ext3_group_desc *)
1619               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1620
1621         return gdp + desc;
1622 }
1623
1624 static inline struct buffer_head *
1625 read_inode_bitmap(struct super_block *sb, unsigned long group)
1626 {
1627         struct ext3_group_desc *desc;
1628         struct buffer_head *bh;
1629
1630         desc = get_group_desc(sb, group);
1631         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1632
1633         return bh;
1634 }
1635
1636 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1637                                      struct buffer_head *bitmap_bh,
1638                                      int index, unsigned long ino)
1639 {
1640         struct inode *inode = NULL;
1641
1642         if (ext3_test_bit(index, bitmap_bh->b_data))
1643                 inode = iget(sb, ino);
1644
1645         return inode;
1646 }
1647
1648 struct qchk_ctxt {
1649         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1650         struct list_head        qckt_list;                   /* quotacheck list */
1651         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1652         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1653 };
1654
1655 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1656                            struct obd_quotactl *oqc)
1657 {
1658         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1659         loff_t size = 0;
1660         qid_t qid[MAXQUOTAS];
1661         int cnt, i, rc = 0;
1662
1663         if (!inode)
1664                 return 0;
1665
1666         qid[USRQUOTA] = inode->i_uid;
1667         qid[GRPQUOTA] = inode->i_gid;
1668
1669         if (S_ISDIR(inode->i_mode) ||
1670             S_ISREG(inode->i_mode) ||
1671             S_ISLNK(inode->i_mode))
1672                 size = inode_get_bytes(inode);
1673
1674         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1675                 if (!Q_TYPESET(oqc, cnt))
1676                         continue;
1677
1678                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1679                                 &qctxt->qckt_list, qid[cnt], cnt,
1680                                 qctxt->qckt_first_check[cnt]);
1681                 if (!cdqb[cnt]) {
1682                         rc = -ENOMEM;
1683                         break;
1684                 }
1685
1686                 cdqb[cnt]->dqb_curspace += size;
1687                 cdqb[cnt]->dqb_curinodes++;
1688         }
1689
1690         if (rc) {
1691                 for (i = 0; i < cnt; i++) {
1692                         if (!Q_TYPESET(oqc, i))
1693                                 continue;
1694                         LASSERT(cdqb[i]);
1695                         cdqb[i]->dqb_curspace -= size;
1696                         cdqb[i]->dqb_curinodes--;
1697                 }
1698         }
1699
1700         return rc;
1701 }
1702
1703 static int v2_write_dqheader(struct file *f, int type)
1704 {
1705         static const __u32 quota_magics[] = V2_INITQMAGICS;
1706         static const __u32 quota_versions[] = V2_INITQVERSIONS;
1707         struct v2_disk_dqheader dqhead;
1708         loff_t offset = 0;
1709
1710         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1711         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1712
1713         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1714         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1715
1716         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1717 }
1718
1719 /* write dqinfo struct in a new quota file */
1720 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1721 {
1722         struct v2_disk_dqinfo dqinfo;
1723         __u32 blocks = V2_DQTREEOFF + 1;
1724         loff_t offset = V2_DQINFOOFF;
1725
1726         if (info) {
1727                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1728                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1729                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1730                                                ~DQF_INFO_DIRTY);
1731         } else {
1732                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1733                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1734                 dqinfo.dqi_flags = 0;
1735         }
1736
1737         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1738         dqinfo.dqi_free_blk = 0;
1739         dqinfo.dqi_free_entry = 0;
1740
1741         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1742 }
1743
1744 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1745                                   struct obd_quotactl *oqc)
1746 {
1747         int i, rc = 0;
1748         ENTRY;
1749
1750         for (i = 0; i < MAXQUOTAS; i++) {
1751                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1752                                          NULL : &qctxt->qckt_dqinfo[i];
1753                 struct file *file;
1754
1755                 if (!Q_TYPESET(oqc, i))
1756                         continue;
1757
1758                 file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
1759                                  0644);
1760                 if (IS_ERR(file)) {
1761                         rc = PTR_ERR(file);
1762                         CERROR("can't create %s file: rc = %d\n",
1763                                op_quotafile[i], rc);
1764                         GOTO(out, rc);
1765                 }
1766
1767                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1768                         CERROR("file %s is not regular", op_quotafile[i]);
1769                         filp_close(file, 0);
1770                         GOTO(out, rc = -EINVAL);
1771                 }
1772
1773                 rc = v2_write_dqheader(file, i);
1774                 if (rc) {
1775                         filp_close(file, 0);
1776                         GOTO(out, rc);
1777                 }
1778
1779                 rc = v2_write_dqinfo(file, i, info);
1780                 filp_close(file, 0);
1781                 if (rc)
1782                         GOTO(out, rc);
1783         }
1784
1785 out:
1786         RETURN(rc);
1787 }
1788
1789
1790 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1791                           struct chk_dqblk *cdqb)
1792 {
1793         struct obd_quotactl *oqc;
1794         long now;
1795         int rc;
1796         ENTRY;
1797
1798         OBD_ALLOC_PTR(oqc);
1799         if (!oqc)
1800                 RETURN(-ENOMEM);
1801
1802         now = cfs_time_current_sec();
1803
1804         if (cdqb->dqb_bsoftlimit &&
1805             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1806             !cdqb->dqb_btime)
1807                 cdqb->dqb_btime =
1808                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1809
1810         if (cdqb->dqb_isoftlimit &&
1811             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1812             !cdqb->dqb_itime)
1813                 cdqb->dqb_itime =
1814                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1815
1816         cdqb->dqb_valid = QIF_ALL;
1817
1818         oqc->qc_cmd = Q_SETQUOTA;
1819         oqc->qc_type = cdqb->dqb_type;
1820         oqc->qc_id = cdqb->dqb_id;
1821         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1822
1823         rc = fsfilt_ext3_quotactl(sb, oqc);
1824         OBD_FREE_PTR(oqc);
1825         RETURN(rc);
1826 }
1827
1828 static int prune_chkquots(struct super_block *sb,
1829                           struct qchk_ctxt *qctxt, int error)
1830 {
1831         struct chk_dqblk *cdqb, *tmp;
1832         int rc;
1833
1834         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1835                 if (!error) {
1836                         rc = commit_chkquot(sb, qctxt, cdqb);
1837                         if (rc)
1838                                 error = rc;
1839                 }
1840                 hlist_del_init(&cdqb->dqb_hash);
1841                 list_del(&cdqb->dqb_list);
1842                 OBD_FREE_PTR(cdqb);
1843         }
1844
1845         return error;
1846 }
1847
1848 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1849                                   struct obd_quotactl *oqc)
1850 {
1851         struct ext3_sb_info *sbi = EXT3_SB(sb);
1852         int i, group;
1853         struct qchk_ctxt *qctxt;
1854         struct buffer_head *bitmap_bh = NULL;
1855         unsigned long ino;
1856         struct inode *inode;
1857         int rc = 0;
1858         ENTRY;
1859
1860         /* turn on quota and read dqinfo if existed */
1861         OBD_ALLOC_PTR(qctxt);
1862         if (!qctxt) {
1863                 oqc->qc_stat = -ENOMEM;
1864                 RETURN(-ENOMEM);
1865         }
1866
1867         for (i = 0; i < NR_DQHASH; i++)
1868                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1869         INIT_LIST_HEAD(&qctxt->qckt_list);
1870
1871         for (i = 0; i < MAXQUOTAS; i++) {
1872                 if (!Q_TYPESET(oqc, i))
1873                         continue;
1874
1875                 rc = quota_onoff(sb, Q_QUOTAON, i);
1876                 if (!rc || rc == -EBUSY) {
1877                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1878                         if (rc)
1879                                 GOTO(out, rc);
1880                 } else if (rc == -ENOENT) {
1881                         qctxt->qckt_first_check[i] = 1;
1882                 } else if (rc) {
1883                         GOTO(out, rc);
1884                 }
1885         }
1886
1887         /* check quota and update in hash */
1888         for (group = 0; group < sbi->s_groups_count; group++) {
1889                 ino = group * sbi->s_inodes_per_group + 1;
1890                 bitmap_bh = read_inode_bitmap(sb, group);
1891                 if (!bitmap_bh) {
1892                         CERROR("read_inode_bitmap group %d failed", group);
1893                         GOTO(out, rc = -EIO);
1894                 }
1895
1896                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1897                         if (ino < sbi->s_first_ino)
1898                                 continue;
1899
1900                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1901                         rc = add_inode_quota(inode, qctxt, oqc);
1902                         iput(inode);
1903                         if (rc) {
1904                                 brelse(bitmap_bh);
1905                                 GOTO(out, rc);
1906                         }
1907                 }
1908
1909                 brelse(bitmap_bh);
1910         }
1911
1912         /* read old quota limits from old quota file. (only for the user
1913          * has limits but hasn't file) */
1914 #ifdef HAVE_QUOTA_SUPPORT
1915         for (i = 0; i < MAXQUOTAS; i++) {
1916                 struct list_head id_list;
1917                 struct dquot_id *dqid, *tmp;
1918
1919                 if (!Q_TYPESET(oqc, i))
1920                         continue;
1921
1922                 if (qctxt->qckt_first_check[i])
1923                         continue;
1924
1925
1926                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1927                 INIT_LIST_HEAD(&id_list);
1928 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1929                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1930 #else
1931                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1932 #endif
1933                 if (rc)
1934                         CERROR("read old limits failed. (rc:%d)\n", rc);
1935
1936                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1937                         list_del_init(&dqid->di_link);
1938
1939                         if (!rc)
1940                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1941                                       dqid->di_id, i,
1942                                       qctxt->qckt_first_check[i]);
1943                         kfree(dqid);
1944                 }
1945         }
1946 #endif
1947         /* turn off quota cause we are to dump chk_dqblk to files */
1948         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1949
1950         rc = create_new_quota_files(qctxt, oqc);
1951         if (rc)
1952                 GOTO(out, rc);
1953
1954         /* we use vfs functions to set dqblk, so turn quota on */
1955         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
1956 out:
1957         /* dump and free chk_dqblk */
1958         rc = prune_chkquots(sb, qctxt, rc);
1959         OBD_FREE_PTR(qctxt);
1960
1961         /* turn off quota, `lfs quotacheck` will turn on when all
1962          * nodes quotacheck finish. */
1963         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1964
1965         oqc->qc_stat = rc;
1966         if (rc)
1967                 CERROR("quotacheck failed: rc = %d\n", rc);
1968
1969         RETURN(rc);
1970 }
1971
1972 #ifdef HAVE_QUOTA_SUPPORT
1973 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
1974                                  int cmd)
1975 {
1976         int rc = 0;
1977         ENTRY;
1978
1979         if (lqi->qi_files[type] == NULL) {
1980                 CERROR("operate qinfo before it's enabled!\n");
1981                 RETURN(-EIO);
1982         }
1983
1984         switch (cmd) {
1985         case QFILE_CHK:
1986                 rc = lustre_check_quota_file(lqi, type);
1987                 break;
1988         case QFILE_RD_INFO:
1989                 rc = lustre_read_quota_info(lqi, type);
1990                 break;
1991         case QFILE_WR_INFO:
1992                 rc = lustre_write_quota_info(lqi, type);
1993                 break;
1994         case QFILE_INIT_INFO:
1995                 rc = lustre_init_quota_info(lqi, type);
1996                 break;
1997         default:
1998                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1999                 LBUG();
2000                 break;
2001         }
2002         RETURN(rc);
2003 }
2004
2005 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2006                             struct list_head *list)
2007 {
2008         return lustre_get_qids(file, inode, type, list);
2009 }
2010
2011 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2012 {
2013         int rc = 0;
2014         ENTRY;
2015
2016         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2017                 CERROR("operate dquot before it's enabled!\n");
2018                 RETURN(-EIO);
2019         }
2020
2021         switch (cmd) {
2022         case QFILE_RD_DQUOT:
2023                 rc = lustre_read_dquot(dquot);
2024                 break;
2025         case QFILE_WR_DQUOT:
2026                 if (dquot->dq_dqb.dqb_ihardlimit ||
2027                     dquot->dq_dqb.dqb_isoftlimit ||
2028                     dquot->dq_dqb.dqb_bhardlimit ||
2029                     dquot->dq_dqb.dqb_bsoftlimit)
2030                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2031                 else
2032                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2033
2034                 rc = lustre_commit_dquot(dquot);
2035                 if (rc >= 0)
2036                         rc = 0;
2037                 break;
2038         default:
2039                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2040                 LBUG();
2041                 break;
2042         }
2043         RETURN(rc);
2044 }
2045 #endif
2046
2047 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2048 {
2049         return (EXT3_SB(sb)->journal_bdev);
2050 }
2051 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2052
2053 static struct fsfilt_operations fsfilt_ext3_ops = {
2054         .fs_type                = "ext3",
2055         .fs_owner               = THIS_MODULE,
2056         .fs_getlabel            = fsfilt_ext3_get_label,
2057         .fs_setlabel            = fsfilt_ext3_set_label,
2058         .fs_uuid                = fsfilt_ext3_uuid,
2059         .fs_start               = fsfilt_ext3_start,
2060         .fs_brw_start           = fsfilt_ext3_brw_start,
2061         .fs_extend              = fsfilt_ext3_extend,
2062         .fs_commit              = fsfilt_ext3_commit,
2063         .fs_commit_async        = fsfilt_ext3_commit_async,
2064         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2065         .fs_setattr             = fsfilt_ext3_setattr,
2066         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2067         .fs_set_md              = fsfilt_ext3_set_md,
2068         .fs_get_md              = fsfilt_ext3_get_md,
2069         .fs_readpage            = fsfilt_ext3_readpage,
2070         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2071         .fs_statfs              = fsfilt_ext3_statfs,
2072         .fs_sync                = fsfilt_ext3_sync,
2073         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2074         .fs_write_record        = fsfilt_ext3_write_record,
2075         .fs_read_record         = fsfilt_ext3_read_record,
2076         .fs_setup               = fsfilt_ext3_setup,
2077         .fs_send_bio            = fsfilt_ext3_send_bio,
2078         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2079         .fs_quotactl            = fsfilt_ext3_quotactl,
2080         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2081 #ifdef HAVE_DISK_INODE_VERSION
2082         .fs_get_version         = fsfilt_ext3_get_version,
2083         .fs_set_version         = fsfilt_ext3_set_version,
2084 #endif
2085 #ifdef HAVE_QUOTA_SUPPORT
2086         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2087         .fs_qids                = fsfilt_ext3_qids,
2088         .fs_dquot               = fsfilt_ext3_dquot,
2089 #endif
2090         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2091 };
2092
2093 static int __init fsfilt_ext3_init(void)
2094 {
2095         int rc;
2096
2097         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2098                                          sizeof(struct fsfilt_cb_data), 0, 0);
2099         if (!fcb_cache) {
2100                 CERROR("error allocating fsfilt journal callback cache\n");
2101                 GOTO(out, rc = -ENOMEM);
2102         }
2103
2104         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2105
2106         if (rc) {
2107                 int err = cfs_mem_cache_destroy(fcb_cache);
2108                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2109         }
2110 out:
2111         return rc;
2112 }
2113
2114 static void __exit fsfilt_ext3_exit(void)
2115 {
2116         int rc;
2117
2118         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2119         rc = cfs_mem_cache_destroy(fcb_cache);
2120         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2121 }
2122
2123 module_init(fsfilt_ext3_init);
2124 module_exit(fsfilt_ext3_exit);
2125
2126 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2127 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2128 MODULE_LICENSE("GPL");