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 #include "lustre_quota_fmt.h"
  71
  72 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  73 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  74 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  75 #else
  76 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  77 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  78 #endif
  79
  80 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
  81 /* for kernels 2.6.18 and later */
  82 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  83 #else
  84 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
  85 #endif
  86
  87 #define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks)
  88 #define fsfilt_ext3_journal_stop(handle)          ext3_journal_stop(handle)
  89
  90 static cfs_mem_cache_t *fcb_cache;
  91
  92 struct fsfilt_cb_data {
  93         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  94         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  95         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  96         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  97         void *cb_data;                  /* MDS/OST completion function data */
  98 };
  99
 100 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 101 #define EXT3_XATTR_INDEX_TRUSTED        4
 102 #endif
 103
 104 static char *fsfilt_ext3_get_label(struct super_block *sb)
 105 {
 106         return EXT3_SB(sb)->s_es->s_volume_name;
 107 }
 108
 109 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 110 {
 111         /* see e.g. fsfilt_ext3_write_record() */
 112         journal_t *journal;
 113         handle_t *handle;
 114         int err;
 115
 116         journal = EXT3_SB(sb)->s_journal;
 117         handle = journal_start(journal, 1);
 118         if (IS_ERR(handle)) {
 119                 CERROR("can't start transaction\n");
 120                 return(PTR_ERR(handle));
 121         }
 122
 123         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 124         if (err)
 125                 goto out;
 126
 127         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 128                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 129
 130         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 131
 132 out:
 133         journal_stop(handle);
 134
 135         return(err);
 136 }
 137
 138 static char *fsfilt_ext3_uuid(struct super_block *sb)
 139 {
 140         return EXT3_SB(sb)->s_es->s_uuid;
 141 }
 142
 143 #ifdef HAVE_DISK_INODE_VERSION
 144 /*
 145  * Get the 64-bit version for an inode.
 146  */
 147 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 148 {
 149         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 150                EXT3_I(inode)->i_fs_version, inode->i_ino);
 151         return EXT3_I(inode)->i_fs_version;
 152 }
 153
 154 /*
 155  * Set the 64-bit version and return the old version.
 156  */
 157 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 158 {
 159         __u64 old_version = EXT3_I(inode)->i_fs_version;
 160
 161         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 162                new_version, old_version, inode->i_ino);
 163         (EXT3_I(inode))->i_fs_version = new_version;
 164         /* version is set after all inode operations are finished, so we should
 165          * mark it dirty here */
 166         inode->i_sb->s_op->dirty_inode(inode);
 167         return old_version;
 168 }
 169
 170 #endif
 171
 172 /*
 173  * We don't currently need any additional blocks for rmdir and
 174  * unlink transactions because we are storing the OST oa_id inside
 175  * the inode (which we will be changing anyways as part of this
 176  * transaction).
 177  */
 178 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 179                                int logs)
 180 {
 181         /* For updates to the last received file */
 182         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 183         journal_t *journal;
 184         void *handle;
 185
 186         if (current->journal_info) {
 187                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 188                        current->journal_info);
 189                 goto journal_start;
 190         }
 191
 192         switch(op) {
 193         case FSFILT_OP_RMDIR:
 194         case FSFILT_OP_UNLINK:
 195                 /* delete one file + create/update logs for each stripe */
 196                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 197                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 198                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 199                 break;
 200         case FSFILT_OP_RENAME:
 201                 /* modify additional directory */
 202                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 203                 /* no break */
 204         case FSFILT_OP_SYMLINK:
 205                 /* additional block + block bitmap + GDT for long symlink */
 206                 nblocks += 3;
 207                 /* no break */
 208         case FSFILT_OP_CREATE: {
 209 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 210                 static int warned;
 211                 if (!warned) {
 212                         if (!test_opt(inode->i_sb, EXTENTS)) {
 213                                 warned = 1;
 214                         } else if (((EXT3_I(inode)->i_flags &
 215                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 216                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 217                                 CWARN("extent-mapped directory found - contact "
 218                                       "http://bugzilla.lustre.org/\n");
 219                                 warned = 1;
 220                         }
 221                 }
 222 #endif
 223                 /* no break */
 224         }
 225         case FSFILT_OP_MKDIR:
 226         case FSFILT_OP_MKNOD:
 227                 /* modify one inode + block bitmap + GDT */
 228                 nblocks += 3;
 229                 /* no break */
 230         case FSFILT_OP_LINK:
 231                 /* modify parent directory */
 232                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 233                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 234                 /* create/update logs 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_SETATTR:
 239                 /* Setattr on inode */
 240                 nblocks += 1;
 241                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 242                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 243                 /* quota chown log for each stripe */
 244                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 245                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 246                 break;
 247         case FSFILT_OP_CANCEL_UNLINK:
 248                 /* blocks for log header bitmap update OR
 249                  * blocks for catalog header bitmap update + unlink of logs */
 250                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 251                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 252                 break;
 253         case FSFILT_OP_JOIN:
 254                 /* delete 2 file(file + array id) + create 1 file (array id)
 255                  * create/update logs for each stripe */
 256                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 257
 258                 /*create array log for head file*/
 259                 nblocks += 3;
 260                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 261                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
 262                 /*update head file array */
 263                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 264                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 265                 break;
 266         default: CERROR("unknown transaction start op %d\n", op);
 267                 LBUG();
 268         }
 269
 270         LASSERT(current->journal_info == desc_private);
 271         journal = EXT3_SB(inode->i_sb)->s_journal;
 272         if (nblocks > journal->j_max_transaction_buffers) {
 273                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 274                        nblocks, op, logs, journal->j_max_transaction_buffers);
 275                 nblocks = journal->j_max_transaction_buffers;
 276         }
 277
 278  journal_start:
 279         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 280         handle = fsfilt_ext3_journal_start(inode, nblocks);
 281
 282         if (!IS_ERR(handle))
 283                 LASSERT(current->journal_info == handle);
 284         else
 285                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 286                        op, nblocks, PTR_ERR(handle));
 287         return handle;
 288 }
 289
 290 /*
 291  * Calculate the number of buffer credits needed to write multiple pages in
 292  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 293  * doesn't have a nice API for calculating this sort of thing in advance.
 294  *
 295  * See comment above ext3_writepage_trans_blocks for details.  We assume
 296  * no data journaling is being done, but it does allow for all of the pages
 297  * being non-contiguous.  If we are guaranteed contiguous pages we could
 298  * reduce the number of (d)indirect blocks a lot.
 299  *
 300  * With N blocks per page and P pages, for each inode we have at most:
 301  * N*P indirect
 302  * min(N*P, blocksize/4 + 1) dindirect blocks
 303  * niocount tindirect
 304  *
 305  * For the entire filesystem, we have at most:
 306  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 307  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 308  * objcount inode blocks
 309  * 1 superblock
 310  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 311  *
 312  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 313  */
 314 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 315                                       int niocount, struct niobuf_local *nb)
 316 {
 317         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 318         __u64 next_indir;
 319         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 320         int nbitmaps = 0, ngdblocks;
 321         int needed = objcount + 1; /* inodes + superblock */
 322         int i, j;
 323
 324         for (i = 0, j = 0; i < objcount; i++, fso++) {
 325                 /* two or more dindirect blocks in case we cross boundary */
 326                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 327                                     nb[j].offset) >>
 328                                    sb->s_blocksize_bits) /
 329                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 330                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 331
 332                 /* leaf, indirect, tindirect blocks for first block */
 333                 nbitmaps += blockpp + 2;
 334
 335                 j += fso->fso_bufcnt;
 336         }
 337
 338         next_indir = nb[0].offset +
 339                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 340         for (i = 1; i < niocount; i++) {
 341                 if (nb[i].offset >= next_indir) {
 342                         nbitmaps++;     /* additional indirect */
 343                         next_indir = nb[i].offset +
 344                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 345                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 346                         nbitmaps++;     /* additional indirect */
 347                 }
 348                 nbitmaps += blockpp;    /* each leaf in different group? */
 349         }
 350
 351         ngdblocks = nbitmaps;
 352         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 353                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 354         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 355                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 356
 357         needed += nbitmaps + ngdblocks;
 358
 359         /* last_rcvd update */
 360         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 361
 362 #if defined(CONFIG_QUOTA)
 363         /* We assume that there will be 1 bit set in s_dquot.flags for each
 364          * quota file that is active.  This is at least true for now.
 365          */
 366         needed += hweight32(sb_any_quota_enabled(sb)) *
 367                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 368 #endif
 369
 370         return needed;
 371 }
 372
 373 /* We have to start a huge journal transaction here to hold all of the
 374  * metadata for the pages being written here.  This is necessitated by
 375  * the fact that we do lots of prepare_write operations before we do
 376  * any of the matching commit_write operations, so even if we split
 377  * up to use "smaller" transactions none of them could complete until
 378  * all of them were opened.  By having a single journal transaction,
 379  * we eliminate duplicate reservations for common blocks like the
 380  * superblock and group descriptors or bitmaps.
 381  *
 382  * We will start the transaction here, but each prepare_write will
 383  * add a refcount to the transaction, and each commit_write will
 384  * remove a refcount.  The transaction will be closed when all of
 385  * the pages have been written.
 386  */
 387 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 388                                    int niocount, struct niobuf_local *nb,
 389                                    void *desc_private, int logs)
 390 {
 391         journal_t *journal;
 392         handle_t *handle;
 393         int needed;
 394         ENTRY;
 395
 396         LASSERT(current->journal_info == desc_private);
 397         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 398         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 399
 400         /* The number of blocks we could _possibly_ dirty can very large.
 401          * We reduce our request if it is absurd (and we couldn't get that
 402          * many credits for a single handle anyways).
 403          *
 404          * At some point we have to limit the size of I/Os sent at one time,
 405          * increase the size of the journal, or we have to calculate the
 406          * actual journal requirements more carefully by checking all of
 407          * the blocks instead of being maximally pessimistic.  It remains to
 408          * be seen if this is a real problem or not.
 409          */
 410         if (needed > journal->j_max_transaction_buffers) {
 411                 CERROR("want too many journal credits (%d) using %d instead\n",
 412                        needed, journal->j_max_transaction_buffers);
 413                 needed = journal->j_max_transaction_buffers;
 414         }
 415
 416         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 417         handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
 418         if (IS_ERR(handle)) {
 419                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 420                        PTR_ERR(handle));
 421         } else {
 422                 LASSERT(handle->h_buffer_credits >= needed);
 423                 LASSERT(current->journal_info == handle);
 424         }
 425
 426         RETURN(handle);
 427 }
 428
 429 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 430 {
 431        handle_t *handle = h;
 432
 433        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 434        if (nblocks == 0) {
 435                handle->h_buffer_credits = 0;
 436                CWARN("setting credits of handle %p to zero by request\n", h);
 437        }
 438
 439        if (handle->h_buffer_credits > nblocks)
 440                 return 0;
 441        if (journal_extend(handle, nblocks) == 0)
 442                 return 0;
 443
 444        ext3_mark_inode_dirty(handle, inode);
 445        return journal_restart(handle, nblocks);
 446 }
 447
 448 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 449 {
 450         int rc;
 451         handle_t *handle = h;
 452
 453         LASSERT(current->journal_info == handle);
 454         if (force_sync)
 455                 handle->h_sync = 1; /* recovery likes this */
 456
 457         rc = fsfilt_ext3_journal_stop(handle);
 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         handle_t *handle = h;
 468         journal_t *journal;
 469         int rc;
 470
 471         LASSERT(current->journal_info == handle);
 472
 473         transaction = handle->h_transaction;
 474         journal = transaction->t_journal;
 475         tid = transaction->t_tid;
 476         /* we don't want to be blocked */
 477         handle->h_sync = 0;
 478         rc = fsfilt_ext3_journal_stop(handle);
 479         if (rc) {
 480                 CERROR("error while stopping transaction: %d\n", rc);
 481                 return rc;
 482         }
 483         log_start_commit(journal, tid);
 484
 485         *wait_handle = (void *) tid;
 486         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 487         return 0;
 488 }
 489
 490 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 491 {
 492         journal_t *journal = EXT3_JOURNAL(inode);
 493         tid_t tid = (tid_t)(long)h;
 494
 495         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 496         if (unlikely(is_journal_aborted(journal)))
 497                 return -EIO;
 498
 499         log_wait_commit(EXT3_JOURNAL(inode), tid);
 500
 501         if (unlikely(is_journal_aborted(journal)))
 502                 return -EIO;
 503         return 0;
 504 }
 505
 506 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 507                                struct iattr *iattr, int do_trunc)
 508 {
 509         struct inode *inode = dentry->d_inode;
 510         int rc = 0;
 511
 512         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 513          * We are already writing the inode to disk as part of this
 514          * transaction and want to avoid a lot of extra inode writeout
 515          * later on. b=9828 */
 516         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 517                 /* ATTR_SIZE would invoke truncate: clear it */
 518                 iattr->ia_valid &= ~ATTR_SIZE;
 519                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 520                 i_size_write(inode, iattr->ia_size);
 521
 522                 if (iattr->ia_valid & ATTR_UID)
 523                         inode->i_uid = iattr->ia_uid;
 524                 if (iattr->ia_valid & ATTR_GID)
 525                         inode->i_gid = iattr->ia_gid;
 526                 if (iattr->ia_valid & ATTR_ATIME)
 527                         inode->i_atime = iattr->ia_atime;
 528                 if (iattr->ia_valid & ATTR_MTIME)
 529                         inode->i_mtime = iattr->ia_mtime;
 530                 if (iattr->ia_valid & ATTR_CTIME)
 531                         inode->i_ctime = iattr->ia_ctime;
 532                 if (iattr->ia_valid & ATTR_MODE) {
 533                         inode->i_mode = iattr->ia_mode;
 534
 535                         if (!in_group_p(inode->i_gid) &&
 536                             !cfs_capable(CFS_CAP_FSETID))
 537                                 inode->i_mode &= ~S_ISGID;
 538                 }
 539
 540                 inode->i_sb->s_op->dirty_inode(inode);
 541
 542                 goto out;
 543         }
 544
 545         /* Don't allow setattr to change file type */
 546         if (iattr->ia_valid & ATTR_MODE)
 547                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 548                                  (iattr->ia_mode & ~S_IFMT);
 549
 550         /* We set these flags on the client, but have already checked perms
 551          * so don't confuse inode_change_ok. */
 552         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 553
 554         if (inode->i_op->setattr) {
 555                 rc = inode->i_op->setattr(dentry, iattr);
 556         } else {
 557                 rc = inode_change_ok(inode, iattr);
 558                 if (!rc)
 559                         rc = inode_setattr(inode, iattr);
 560         }
 561
 562  out:
 563         RETURN(rc);
 564 }
 565
 566 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 567                                  unsigned int cmd, unsigned long arg)
 568 {
 569         int rc = 0;
 570         ENTRY;
 571
 572         /* FIXME: Can't do this because of nested transaction deadlock */
 573         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 574                 CERROR("can't set data journal flag on file\n");
 575                 RETURN(-EPERM);
 576         }
 577
 578         if (inode->i_fop->ioctl)
 579                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 580         else
 581                 RETURN(-ENOTTY);
 582
 583         RETURN(rc);
 584 }
 585
 586 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 587                               void *lmm, int lmm_size, const char *name)
 588 {
 589         int rc;
 590
 591         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 592
 593         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 594                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 595
 596
 597         if (rc && rc != -EROFS)
 598                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 599                        inode->i_ino, rc);
 600         return rc;
 601 }
 602
 603 /* Must be called with i_mutex held */
 604 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 605                               const char *name)
 606 {
 607         int rc;
 608
 609         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 610
 611         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 612                             name, lmm, lmm_size);
 613
 614         /* This gives us the MD size */
 615         if (lmm == NULL)
 616                 return (rc == -ENODATA) ? 0 : rc;
 617
 618         if (rc < 0) {
 619                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 620                        EXT3_XATTR_INDEX_TRUSTED, name,
 621                        inode->i_ino, rc);
 622                 memset(lmm, 0, lmm_size);
 623                 return (rc == -ENODATA) ? 0 : rc;
 624         }
 625
 626         return rc;
 627 }
 628
 629 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 630 {
 631         submit_bio(rw, bio);
 632         return 0;
 633 }
 634
 635 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 636                                     loff_t *off)
 637 {
 638         struct inode *inode = file->f_dentry->d_inode;
 639         int rc = 0;
 640
 641         if (S_ISREG(inode->i_mode))
 642                 rc = file->f_op->read(file, buf, count, off);
 643         else {
 644                 const int blkbits = inode->i_sb->s_blocksize_bits;
 645                 const int blksize = inode->i_sb->s_blocksize;
 646
 647                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 648                        count, inode->i_ino, *off);
 649                 while (count > 0) {
 650                         struct buffer_head *bh;
 651
 652                         bh = NULL;
 653                         if (*off < i_size_read(inode)) {
 654                                 int err = 0;
 655
 656                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 657                                                 0, &err);
 658
 659                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 660
 661                                 if (bh) {
 662                                         memcpy(buf, bh->b_data, blksize);
 663                                         brelse(bh);
 664                                 } else if (err) {
 665                                         /* XXX in theory we should just fake
 666                                          * this buffer and continue like ext3,
 667                                          * especially if this is a partial read
 668                                          */
 669                                         CERROR("error read dir %lu+%llu: %d\n",
 670                                                inode->i_ino, *off, err);
 671                                         RETURN(err);
 672                                 }
 673                         }
 674                         if (!bh) {
 675                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 676
 677                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 678                                 memset(fake, 0, sizeof(*fake));
 679                                 fake->rec_len = cpu_to_le16(blksize);
 680                         }
 681                         count -= blksize;
 682                         buf += blksize;
 683                         *off += blksize;
 684                         rc += blksize;
 685                 }
 686         }
 687
 688         return rc;
 689 }
 690
 691 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 692 {
 693         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 694
 695         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 696
 697         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 698 }
 699
 700 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 701                                       void *handle, fsfilt_cb_t cb_func,
 702                                       void *cb_data)
 703 {
 704         struct fsfilt_cb_data *fcb;
 705
 706         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 707         if (fcb == NULL)
 708                 RETURN(-ENOMEM);
 709
 710         fcb->cb_func = cb_func;
 711         fcb->cb_obd = obd;
 712         fcb->cb_last_rcvd = last_rcvd;
 713         fcb->cb_data = cb_data;
 714
 715         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 716         journal_callback_set(handle, fsfilt_ext3_cb_func,
 717                              (struct journal_callback *)fcb);
 718
 719         return 0;
 720 }
 721
 722 /*
 723  * We need to hack the return value for the free inode counts because
 724  * the current EA code requires one filesystem block per inode with EAs,
 725  * so it is possible to run out of blocks before we run out of inodes.
 726  *
 727  * This can be removed when the ext3 EA code is fixed.
 728  */
 729 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 730 {
 731         struct kstatfs sfs;
 732         int rc;
 733
 734         memset(&sfs, 0, sizeof(sfs));
 735         rc = ll_do_statfs(sb, &sfs);
 736         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 737                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 738                 sfs.f_ffree = sfs.f_bfree;
 739         }
 740
 741         statfs_pack(osfs, &sfs);
 742         return rc;
 743 }
 744
 745 static int fsfilt_ext3_sync(struct super_block *sb)
 746 {
 747         return ext3_force_commit(sb);
 748 }
 749
 750 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 751 #warning "kernel code has old extents/mballoc patch, disabling"
 752 #undef EXT3_MULTIBLOCK_ALLOCATOR
 753 #endif
 754 #ifndef EXT3_EXTENTS_FL
 755 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 756 #endif
 757
 758 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 759 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 760 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 761 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 762 #else
 763 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 764 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 765 #endif
 766
 767 #ifndef EXT_ASSERT
 768 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 769 #endif
 770
 771 #ifdef EXT3_EXT_HAS_NO_TREE
 772 /* for kernels 2.6.18 and later */
 773 #define ext3_ext_base                   inode
 774 #define ext3_ext_base2inode(inode)      (inode)
 775 #define EXT_DEPTH(inode)                ext_depth(inode)
 776 #define EXT_GENERATION(inode)           ext_generation(inode)
 777 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 778                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 779 #else
 780 #define ext3_ext_base                   ext3_extents_tree
 781 #define ext3_ext_base2inode(tree)       (tree->inode)
 782 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 783                         ext3_ext_walk_space(tree, block, num, cb);
 784 #endif
 785
 786 #include <linux/lustre_version.h>
 787 #if EXT3_EXT_MAGIC == 0xf301
 788 #define ee_start e_start
 789 #define ee_block e_block
 790 #define ee_len   e_num
 791 #endif
 792 #ifndef EXT3_BB_MAX_BLOCKS
 793 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 794         ext3_new_blocks(handle, inode, count, goal, err)
 795 #endif
 796
 797 struct bpointers {
 798         unsigned long *blocks;
 799         int *created;
 800         unsigned long start;
 801         int num;
 802         int init_num;
 803         int create;
 804 };
 805
 806 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 807                               unsigned long block, int *aflags)
 808 {
 809         struct ext3_inode_info *ei = EXT3_I(inode);
 810         unsigned long bg_start;
 811         unsigned long colour;
 812         int depth;
 813
 814         if (path) {
 815                 struct ext3_extent *ex;
 816                 depth = path->p_depth;
 817
 818                 /* try to predict block placement */
 819                 if ((ex = path[depth].p_ext)) {
 820 #if 0
 821                         /* This prefers to eat into a contiguous extent
 822                          * rather than find an extent that the whole
 823                          * request will fit into.  This can fragment data
 824                          * block allocation and prevents our lovely 1M I/Os
 825                          * from reaching the disk intact. */
 826                         if (ex->ee_block + ex->ee_len == block)
 827                                 *aflags |= 1;
 828 #endif
 829                         return ex->ee_start + (block - ex->ee_block);
 830                 }
 831
 832                 /* it looks index is empty
 833                  * try to find starting from index itself */
 834                 if (path[depth].p_bh)
 835                         return path[depth].p_bh->b_blocknr;
 836         }
 837
 838         /* OK. use inode's group */
 839         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 840                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 841         colour = (current->pid % 16) *
 842                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 843         return bg_start + colour + block;
 844 }
 845
 846 #define ll_unmap_underlying_metadata(sb, blocknr) \
 847         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 848
 849 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 850 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 851                                 struct ext3_ext_path *path, unsigned long block,
 852                                 unsigned long *count, int *err)
 853 {
 854         unsigned long pblock, goal;
 855         int aflags = 0;
 856         struct inode *inode = ext3_ext_base2inode(base);
 857
 858         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 859         aflags |= 2; /* block have been already reserved */
 860         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 861         return pblock;
 862
 863 }
 864 #else
 865 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 866                                 struct ext3_ext_path *path, unsigned long block,
 867                                 unsigned long *count, int *err)
 868 {
 869         struct inode *inode = ext3_ext_base2inode(base);
 870         struct ext3_allocation_request ar;
 871         unsigned long pblock;
 872         int aflags;
 873
 874         /* find neighbour allocated blocks */
 875         ar.lleft = block;
 876         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 877         if (*err)
 878                 return 0;
 879         ar.lright = block;
 880         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 881         if (*err)
 882                 return 0;
 883
 884         /* allocate new block */
 885         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 886         ar.inode = inode;
 887         ar.logical = block;
 888         ar.len = *count;
 889         ar.flags = EXT3_MB_HINT_DATA;
 890         pblock = ext3_mb_new_blocks(handle, &ar, err);
 891         *count = ar.len;
 892         return pblock;
 893 }
 894 #endif
 895
 896 #ifdef EXT3_EXT_HAS_NO_TREE
 897 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 898                                   struct ext3_ext_path *path,
 899                                   struct ext3_ext_cache *cex,
 900 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 901                                    struct ext3_extent *ex,
 902 #endif
 903                                   void *cbdata)
 904 {
 905         struct bpointers *bp = cbdata;
 906 #else
 907 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 908                                   struct ext3_ext_path *path,
 909                                   struct ext3_ext_cache *cex
 910 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 911                                   , struct ext3_extent *ex
 912 #endif
 913                                  )
 914 {
 915         struct bpointers *bp = base->private;
 916 #endif
 917         struct inode *inode = ext3_ext_base2inode(base);
 918         struct ext3_extent nex;
 919         unsigned long pblock;
 920         unsigned long tgen;
 921         int err, i;
 922         unsigned long count;
 923         handle_t *handle;
 924
 925         i = EXT_DEPTH(base);
 926         EXT_ASSERT(i == path->p_depth);
 927         EXT_ASSERT(path[i].p_hdr);
 928
 929         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 930                 err = EXT_CONTINUE;
 931                 goto map;
 932         }
 933
 934         if (bp->create == 0) {
 935                 i = 0;
 936                 if (cex->ec_block < bp->start)
 937                         i = bp->start - cex->ec_block;
 938                 if (i >= cex->ec_len)
 939                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 940                                         i, cex->ec_len);
 941                 for (; i < cex->ec_len && bp->num; i++) {
 942                         *(bp->created) = 0;
 943                         bp->created++;
 944                         *(bp->blocks) = 0;
 945                         bp->blocks++;
 946                         bp->num--;
 947                         bp->start++;
 948                 }
 949
 950                 return EXT_CONTINUE;
 951         }
 952
 953         tgen = EXT_GENERATION(base);
 954         count = ext3_ext_calc_credits_for_insert(base, path);
 955         ext3_up_truncate_sem(inode);
 956
 957         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 958         if (IS_ERR(handle)) {
 959                 ext3_down_truncate_sem(inode);
 960                 return PTR_ERR(handle);
 961         }
 962
 963         ext3_down_truncate_sem(inode);
 964         if (tgen != EXT_GENERATION(base)) {
 965                 /* the tree has changed. so path can be invalid at moment */
 966                 fsfilt_ext3_journal_stop(handle);
 967                 return EXT_REPEAT;
 968         }
 969
 970         count = cex->ec_len;
 971         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 972         if (!pblock)
 973                 goto out;
 974         EXT_ASSERT(count <= cex->ec_len);
 975
 976         /* insert new extent */
 977         nex.ee_block = cex->ec_block;
 978         nex.ee_start = pblock;
 979         nex.ee_len = count;
 980         err = ext3_ext_insert_extent(handle, base, path, &nex);
 981         if (err) {
 982                 /* free data blocks we just allocated */
 983                 /* not a good idea to call discard here directly,
 984                  * but otherwise we'd need to call it every free() */
 985 #ifdef EXT3_MB_HINT_GROUP_ALLOC
 986                 ext3_mb_discard_inode_preallocations(inode);
 987 #endif
 988                 ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);
 989                 goto out;
 990         }
 991
 992         /*
 993          * Putting len of the actual extent we just inserted,
 994          * we are asking ext3_ext_walk_space() to continue
 995          * scaning after that block
 996          */
 997         cex->ec_len = nex.ee_len;
 998         cex->ec_start = nex.ee_start;
 999         BUG_ON(nex.ee_len == 0);
1000         BUG_ON(nex.ee_block != cex->ec_block);
1001
1002 out:
1003         fsfilt_ext3_journal_stop(handle);
1004 map:
1005         if (err >= 0) {
1006                 /* map blocks */
1007                 if (bp->num == 0) {
1008                         CERROR("hmm. why do we find this extent?\n");
1009                         CERROR("initial space: %lu:%u\n",
1010                                 bp->start, bp->init_num);
1011                         CERROR("current extent: %u/%u/%u %d\n",
1012                                 cex->ec_block, cex->ec_len,
1013                                 cex->ec_start, cex->ec_type);
1014                 }
1015                 i = 0;
1016                 if (cex->ec_block < bp->start)
1017                         i = bp->start - cex->ec_block;
1018                 if (i >= cex->ec_len)
1019                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1020                                         i, cex->ec_len);
1021                 for (; i < cex->ec_len && bp->num; i++) {
1022                         *(bp->blocks) = cex->ec_start + i;
1023                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1024                                 *(bp->created) = 0;
1025                         } else {
1026                                 *(bp->created) = 1;
1027                                 /* unmap any possible underlying metadata from
1028                                  * the block device mapping.  bug 6998. */
1029                                 ll_unmap_underlying_metadata(inode->i_sb,
1030                                                              *(bp->blocks));
1031                         }
1032                         bp->created++;
1033                         bp->blocks++;
1034                         bp->num--;
1035                         bp->start++;
1036                 }
1037         }
1038         return err;
1039 }
1040
1041 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1042                        unsigned long num, unsigned long *blocks,
1043                        int *created, int create)
1044 {
1045 #ifdef EXT3_EXT_HAS_NO_TREE
1046         struct ext3_ext_base *base = inode;
1047 #else
1048         struct ext3_extents_tree tree;
1049         struct ext3_ext_base *base = &tree;
1050 #endif
1051         struct bpointers bp;
1052         int err;
1053
1054         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1055                block, block + num - 1, (unsigned) inode->i_ino);
1056
1057 #ifndef EXT3_EXT_HAS_NO_TREE
1058         ext3_init_tree_desc(base, inode);
1059         tree.private = &bp;
1060 #endif
1061         bp.blocks = blocks;
1062         bp.created = created;
1063         bp.start = block;
1064         bp.init_num = bp.num = num;
1065         bp.create = create;
1066
1067         ext3_down_truncate_sem(inode);
1068         err = fsfilt_ext3_ext_walk_space(base, block, num,
1069                                          ext3_ext_new_extent_cb, &bp);
1070         ext3_ext_invalidate_cache(base);
1071         ext3_up_truncate_sem(inode);
1072
1073         return err;
1074 }
1075
1076 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1077                                     int pages, unsigned long *blocks,
1078                                     int *created, int create)
1079 {
1080         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1081         int rc = 0, i = 0;
1082         struct page *fp = NULL;
1083         int clen = 0;
1084
1085         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1086                 inode->i_ino, pages, (*page)->index);
1087
1088         /* pages are sorted already. so, we just have to find
1089          * contig. space and process them properly */
1090         while (i < pages) {
1091                 if (fp == NULL) {
1092                         /* start new extent */
1093                         fp = *page++;
1094                         clen = 1;
1095                         i++;
1096                         continue;
1097                 } else if (fp->index + clen == (*page)->index) {
1098                         /* continue the extent */
1099                         page++;
1100                         clen++;
1101                         i++;
1102                         continue;
1103                 }
1104
1105                 /* process found extent */
1106                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1107                                         clen * blocks_per_page, blocks,
1108                                         created, create);
1109                 if (rc)
1110                         GOTO(cleanup, rc);
1111
1112                 /* look for next extent */
1113                 fp = NULL;
1114                 blocks += blocks_per_page * clen;
1115                 created += blocks_per_page * clen;
1116         }
1117
1118         if (fp)
1119                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1120                                         clen * blocks_per_page, blocks,
1121                                         created, create);
1122 cleanup:
1123         return rc;
1124 }
1125 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1126
1127 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1128                                unsigned long *blocks, int *created, int create);
1129 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1130                                    int pages, unsigned long *blocks,
1131                                    int *created, int create)
1132 {
1133         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1134         unsigned long *b;
1135         int rc = 0, i, *cr;
1136
1137         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1138                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1139                 if (rc) {
1140                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1141                                inode->i_ino, *b, *cr, create, rc);
1142                         break;
1143                 }
1144
1145                 b += blocks_per_page;
1146                 cr += blocks_per_page;
1147         }
1148         return rc;
1149 }
1150
1151 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1152                                 int pages, unsigned long *blocks,
1153                                 int *created, int create,
1154                                 struct semaphore *optional_sem)
1155 {
1156         int rc;
1157 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1158         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1159                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1160                                                      blocks, created, create);
1161                 return rc;
1162         }
1163 #endif
1164         if (optional_sem != NULL)
1165                 down(optional_sem);
1166         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1167                                             created, create);
1168         if (optional_sem != NULL)
1169                 up(optional_sem);
1170
1171         return rc;
1172 }
1173
1174 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1175 {
1176         unsigned long block;
1177         struct buffer_head *bh;
1178         int err, blocksize, csize, boffs, osize = size;
1179
1180         /* prevent reading after eof */
1181         lock_kernel();
1182         if (i_size_read(inode) < *offs + size) {
1183                 size = i_size_read(inode) - *offs;
1184                 unlock_kernel();
1185                 if (size < 0) {
1186                         CERROR("size %llu is too short for read %u@%llu\n",
1187                                i_size_read(inode), size, *offs);
1188                         return -EIO;
1189                 } else if (size == 0) {
1190                         return 0;
1191                 }
1192         } else {
1193                 unlock_kernel();
1194         }
1195
1196         blocksize = 1 << inode->i_blkbits;
1197
1198         while (size > 0) {
1199                 block = *offs >> inode->i_blkbits;
1200                 boffs = *offs & (blocksize - 1);
1201                 csize = min(blocksize - boffs, size);
1202                 bh = ext3_bread(NULL, inode, block, 0, &err);
1203                 if (!bh) {
1204                         CERROR("can't read block: %d\n", err);
1205                         return err;
1206                 }
1207
1208                 memcpy(buf, bh->b_data + boffs, csize);
1209                 brelse(bh);
1210
1211                 *offs += csize;
1212                 buf += csize;
1213                 size -= csize;
1214         }
1215         return osize;
1216 }
1217 EXPORT_SYMBOL(fsfilt_ext3_read);
1218
1219 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1220                                    int size, loff_t *offs)
1221 {
1222         int rc;
1223         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1224         if (rc > 0)
1225                 rc = 0;
1226         return rc;
1227 }
1228
1229 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1230                                 loff_t *offs, handle_t *handle)
1231 {
1232         struct buffer_head *bh = NULL;
1233         loff_t old_size = i_size_read(inode), offset = *offs;
1234         loff_t new_size = i_size_read(inode);
1235         unsigned long block;
1236         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1237
1238         while (bufsize > 0) {
1239                 if (bh != NULL)
1240                         brelse(bh);
1241
1242                 block = offset >> inode->i_blkbits;
1243                 boffs = offset & (blocksize - 1);
1244                 size = min(blocksize - boffs, bufsize);
1245                 bh = ext3_bread(handle, inode, block, 1, &err);
1246                 if (!bh) {
1247                         CERROR("can't read/create block: %d\n", err);
1248                         break;
1249                 }
1250
1251                 err = ext3_journal_get_write_access(handle, bh);
1252                 if (err) {
1253                         CERROR("journal_get_write_access() returned error %d\n",
1254                                err);
1255                         break;
1256                 }
1257                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1258                 memcpy(bh->b_data + boffs, buf, size);
1259                 err = ext3_journal_dirty_metadata(handle, bh);
1260                 if (err) {
1261                         CERROR("journal_dirty_metadata() returned error %d\n",
1262                                err);
1263                         break;
1264                 }
1265                 if (offset + size > new_size)
1266                         new_size = offset + size;
1267                 offset += size;
1268                 bufsize -= size;
1269                 buf += size;
1270         }
1271         if (bh)
1272                 brelse(bh);
1273
1274         /* correct in-core and on-disk sizes */
1275         if (new_size > i_size_read(inode)) {
1276                 lock_kernel();
1277                 if (new_size > i_size_read(inode))
1278                         i_size_write(inode, new_size);
1279                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1280                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1281                 if (i_size_read(inode) > old_size)
1282                         mark_inode_dirty(inode);
1283                 unlock_kernel();
1284         }
1285
1286         if (err == 0)
1287                 *offs = offset;
1288         return err;
1289 }
1290 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1291
1292 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1293                                     loff_t *offs, int force_sync)
1294 {
1295         struct inode *inode = file->f_dentry->d_inode;
1296         handle_t *handle;
1297         int err, block_count = 0, blocksize;
1298
1299         /* Determine how many transaction credits are needed */
1300         blocksize = 1 << inode->i_blkbits;
1301         block_count = (*offs & (blocksize - 1)) + bufsize;
1302         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1303
1304         handle = fsfilt_ext3_journal_start(inode,
1305                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1306         if (IS_ERR(handle)) {
1307                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1308                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1309                 return PTR_ERR(handle);
1310         }
1311
1312         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1313
1314         if (!err && force_sync)
1315                 handle->h_sync = 1; /* recovery likes this */
1316
1317         fsfilt_ext3_journal_stop(handle);
1318
1319         return err;
1320 }
1321
1322 static int fsfilt_ext3_setup(struct super_block *sb)
1323 {
1324 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1325      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1326         struct ext3_sb_info *sbi = EXT3_SB(sb);
1327 #if 0
1328         sbi->dx_lock = fsfilt_ext3_dx_lock;
1329         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1330 #endif
1331 #endif
1332 #ifdef S_PDIROPS
1333         CWARN("Enabling PDIROPS\n");
1334         set_opt(sbi->s_mount_opt, PDIROPS);
1335         sb->s_flags |= S_PDIROPS;
1336 #endif
1337         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1338                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1339 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1340         /* enable journaled quota support */
1341         /* kfreed in ext3_put_super() */
1342         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1343         if (!sbi->s_qf_names[USRQUOTA])
1344                 return -ENOMEM;
1345         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1346         if (!sbi->s_qf_names[GRPQUOTA]) {
1347                 kfree(sbi->s_qf_names[USRQUOTA]);
1348                 sbi->s_qf_names[USRQUOTA] = NULL;
1349                 return -ENOMEM;
1350         }
1351         sbi->s_jquota_fmt = QFMT_VFS_V0;
1352 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1353         set_opt(sbi->s_mount_opt, QUOTA);
1354 #endif
1355 #endif
1356         return 0;
1357 }
1358
1359 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1360    objects. Logs is number of logfiles to update */
1361 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1362 {
1363         if ( !fso ) {
1364                 switch(op) {
1365                 case FSFILT_OP_CREATE:
1366                                  /* directory leaf, index & indirect & EA*/
1367                         return 4 + 3 * logs;
1368                 case FSFILT_OP_UNLINK:
1369                         return 3 * logs;
1370                 }
1371         } else {
1372                 int i;
1373                 int needed = 0;
1374                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1375                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1376                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1377                 for (i = 0; i < op; i++, fso++) {
1378                         int nblocks = fso->fso_bufcnt * blockpp;
1379                         int ndindirect = min(nblocks, addrpp + 1);
1380                         int nindir = nblocks + ndindirect + 1;
1381
1382                         needed += nindir;
1383                 }
1384                 return needed + 3 * logs;
1385         }
1386
1387         return 0;
1388 }
1389
1390 #ifdef HAVE_QUOTA_SUPPORT
1391 #define DQINFO_COPY(out, in)                    \
1392 do {                                            \
1393         Q_COPY(out, in, dqi_bgrace);            \
1394         Q_COPY(out, in, dqi_igrace);            \
1395         Q_COPY(out, in, dqi_flags);             \
1396         Q_COPY(out, in, dqi_valid);             \
1397 } while (0)
1398
1399 #define DQBLK_COPY(out, in)                     \
1400 do {                                            \
1401         Q_COPY(out, in, dqb_bhardlimit);        \
1402         Q_COPY(out, in, dqb_bsoftlimit);        \
1403         Q_COPY(out, in, dqb_curspace);          \
1404         Q_COPY(out, in, dqb_ihardlimit);        \
1405         Q_COPY(out, in, dqb_isoftlimit);        \
1406         Q_COPY(out, in, dqb_curinodes);         \
1407         Q_COPY(out, in, dqb_btime);             \
1408         Q_COPY(out, in, dqb_itime);             \
1409         Q_COPY(out, in, dqb_valid);             \
1410 } while (0)
1411
1412 static int fsfilt_ext3_quotactl(struct super_block *sb,
1413                                 struct obd_quotactl *oqc)
1414 {
1415         int i, rc = 0, error = 0;
1416         struct quotactl_ops *qcop;
1417         struct if_dqinfo *info;
1418         struct if_dqblk *dqblk;
1419         ENTRY;
1420
1421         if (!sb->s_qcop)
1422                 RETURN(-ENOSYS);
1423
1424         OBD_ALLOC_PTR(info);
1425         if (!info)
1426                 RETURN(-ENOMEM);
1427         OBD_ALLOC_PTR(dqblk);
1428         if (!dqblk) {
1429                 OBD_FREE_PTR(info);
1430                 RETURN(-ENOMEM);
1431         }
1432
1433         DQINFO_COPY(info, &oqc->qc_dqinfo);
1434         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1435
1436         qcop = sb->s_qcop;
1437         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1438                 for (i = 0; i < MAXQUOTAS; i++) {
1439                         if (!Q_TYPESET(oqc, i))
1440                                 continue;
1441
1442                         if (oqc->qc_cmd == Q_QUOTAON) {
1443                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1444
1445                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1446
1447                                 if (!qcop->quota_on)
1448                                         GOTO(out, rc = -ENOSYS);
1449
1450                                 rc = qcop->quota_on(sb, i, QFMT_VFS_V0,
1451                                                     name[i]);
1452                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1453                                 if (!qcop->quota_off)
1454                                         GOTO(out, rc = -ENOSYS);
1455                                 rc = qcop->quota_off(sb, i);
1456                         }
1457
1458                         if (rc == -EBUSY)
1459                                 error = rc;
1460                         else if (rc)
1461                                 GOTO(out, rc);
1462                 }
1463                 GOTO(out, rc ?: error);
1464         }
1465
1466         switch (oqc->qc_cmd) {
1467         case Q_GETOINFO:
1468         case Q_GETINFO:
1469                 if (!qcop->get_info)
1470                         GOTO(out, rc = -ENOSYS);
1471                 rc = qcop->get_info(sb, oqc->qc_type, info);
1472                 break;
1473         case Q_SETQUOTA:
1474         case Q_INITQUOTA:
1475                 if (!qcop->set_dqblk)
1476                         GOTO(out, rc = -ENOSYS);
1477                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1478                 break;
1479         case Q_GETOQUOTA:
1480         case Q_GETQUOTA:
1481                 if (!qcop->get_dqblk)
1482                         GOTO(out, rc = -ENOSYS);
1483                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1484                 if (!rc)
1485                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1486                 break;
1487         case Q_SYNC:
1488                 if (!sb->s_qcop->quota_sync)
1489                         GOTO(out, rc = -ENOSYS);
1490                 qcop->quota_sync(sb, oqc->qc_type);
1491                 break;
1492         case Q_FINVALIDATE:
1493                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1494                 for (i = 0; i < MAXQUOTAS; i++) {
1495                         struct file *fp;
1496                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1497
1498                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1499
1500                         if (!Q_TYPESET(oqc, i))
1501                                 continue;
1502
1503                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1504                         if (IS_ERR(fp)) {
1505                                 rc = PTR_ERR(fp);
1506                                 CERROR("error invalidating operational quota file"
1507                                        " %s (rc:%d)\n", name[i], rc);
1508                         } else {
1509                                 filp_close(fp, 0);
1510                         }
1511
1512                 }
1513                 break;
1514         default:
1515                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1516                 LBUG();
1517         }
1518 out:
1519         DQINFO_COPY(&oqc->qc_dqinfo, info);
1520         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1521
1522         OBD_FREE_PTR(info);
1523         OBD_FREE_PTR(dqblk);
1524
1525         if (rc)
1526                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1527                                 "failed: %d\n",
1528                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1529         RETURN(rc);
1530 }
1531
1532 struct chk_dqblk{
1533         struct hlist_node       dqb_hash;        /** quotacheck hash */
1534         struct list_head        dqb_list;        /** in list also */
1535         qid_t                   dqb_id;          /** uid/gid */
1536         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1537         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1538         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1539         qsize_t                 dqb_curspace;    /** current space */
1540         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1541         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1542         qsize_t                 dqb_curinodes;   /** current inodes */
1543         __u64                   dqb_btime;       /** block grace time */
1544         __u64                   dqb_itime;       /** inode grace time */
1545         __u32                   dqb_valid;       /** flag for above fields */
1546 };
1547
1548 static inline unsigned int chkquot_hash(qid_t id, int type)
1549                                         __attribute__((__const__));
1550
1551 static inline unsigned int chkquot_hash(qid_t id, int type)
1552 {
1553         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1554 }
1555
1556 static inline struct chk_dqblk *
1557 find_chkquot(struct hlist_head *head, qid_t id, int type)
1558 {
1559         struct hlist_node *node;
1560         struct chk_dqblk *cdqb;
1561
1562         hlist_for_each(node, head) {
1563                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1564                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1565                         return cdqb;
1566         }
1567
1568         return NULL;
1569 }
1570
1571 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1572 {
1573         struct chk_dqblk *cdqb;
1574
1575         OBD_ALLOC_PTR(cdqb);
1576         if (cdqb) {
1577                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1578                 INIT_LIST_HEAD(&cdqb->dqb_list);
1579                 cdqb->dqb_id = id;
1580                 cdqb->dqb_type = type;
1581         }
1582
1583         return cdqb;
1584 }
1585
1586 static struct chk_dqblk *
1587 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1588       qid_t id, int type, int first_check)
1589 {
1590         struct hlist_head *head = hash + chkquot_hash(id, type);
1591         struct if_dqblk dqb;
1592         struct chk_dqblk *cdqb;
1593         int rc;
1594
1595         cdqb = find_chkquot(head, id, type);
1596         if (cdqb)
1597                 return cdqb;
1598
1599         cdqb = alloc_chkquot(id, type);
1600         if (!cdqb)
1601                 return NULL;
1602
1603         if (!first_check) {
1604                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1605                 if (rc) {
1606                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1607                                id, type, rc);
1608                 } else {
1609                         DQBLK_COPY(cdqb, &dqb);
1610                         cdqb->dqb_curspace = 0;
1611                         cdqb->dqb_curinodes = 0;
1612                 }
1613         }
1614
1615         hlist_add_head(&cdqb->dqb_hash, head);
1616         list_add_tail(&cdqb->dqb_list, list);
1617
1618         return cdqb;
1619 }
1620
1621 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1622 {
1623         struct obd_quotactl *oqctl;
1624         int rc;
1625
1626         OBD_ALLOC_PTR(oqctl);
1627         if (!oqctl)
1628                 RETURN(-ENOMEM);
1629
1630         oqctl->qc_cmd = cmd;
1631         oqctl->qc_id = qfmt;
1632         oqctl->qc_type = type;
1633         rc = fsfilt_ext3_quotactl(sb, oqctl);
1634
1635         OBD_FREE_PTR(oqctl);
1636         return rc;
1637 }
1638
1639 static inline int read_old_dqinfo(struct super_block *sb, int type,
1640                                   struct if_dqinfo *dqinfo)
1641 {
1642         struct obd_quotactl *oqctl;
1643         int rc;
1644         ENTRY;
1645
1646         OBD_ALLOC_PTR(oqctl);
1647         if (!oqctl)
1648                 RETURN(-ENOMEM);
1649
1650         oqctl->qc_cmd = Q_GETINFO;
1651         oqctl->qc_type = type;
1652         rc = fsfilt_ext3_quotactl(sb, oqctl);
1653         if (!rc)
1654                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1655
1656         OBD_FREE_PTR(oqctl);
1657         RETURN(rc);
1658 }
1659
1660 static inline struct ext3_group_desc *
1661 get_group_desc(struct super_block *sb, int group)
1662 {
1663         unsigned long desc_block, desc;
1664         struct ext3_group_desc *gdp;
1665
1666         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1667         desc = group % EXT3_DESC_PER_BLOCK(sb);
1668         gdp = (struct ext3_group_desc *)
1669               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1670
1671         return gdp + desc;
1672 }
1673
1674 static inline struct buffer_head *
1675 read_inode_bitmap(struct super_block *sb, unsigned long group)
1676 {
1677         struct ext3_group_desc *desc;
1678         struct buffer_head *bh;
1679
1680         desc = get_group_desc(sb, group);
1681         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1682
1683         return bh;
1684 }
1685
1686 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1687                                      struct buffer_head *bitmap_bh,
1688                                      int index, unsigned long ino)
1689 {
1690         struct inode *inode = NULL;
1691
1692         if (ext3_test_bit(index, bitmap_bh->b_data))
1693                 inode = iget(sb, ino);
1694
1695         return inode;
1696 }
1697
1698 struct qchk_ctxt {
1699         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1700         struct list_head        qckt_list;                   /* quotacheck list */
1701         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1702         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1703 };
1704
1705 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1706                            struct obd_quotactl *oqc)
1707 {
1708         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1709         loff_t size = 0;
1710         qid_t qid[MAXQUOTAS];
1711         int cnt, i, rc = 0;
1712
1713         if (!inode)
1714                 return 0;
1715
1716         qid[USRQUOTA] = inode->i_uid;
1717         qid[GRPQUOTA] = inode->i_gid;
1718
1719         if (S_ISDIR(inode->i_mode) ||
1720             S_ISREG(inode->i_mode) ||
1721             S_ISLNK(inode->i_mode))
1722                 size = inode_get_bytes(inode);
1723
1724         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1725                 if (!Q_TYPESET(oqc, cnt))
1726                         continue;
1727
1728                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1729                                 &qctxt->qckt_list, qid[cnt], cnt,
1730                                 qctxt->qckt_first_check[cnt]);
1731                 if (!cdqb[cnt]) {
1732                         rc = -ENOMEM;
1733                         break;
1734                 }
1735
1736                 cdqb[cnt]->dqb_curspace += size;
1737                 cdqb[cnt]->dqb_curinodes++;
1738         }
1739
1740         if (rc) {
1741                 for (i = 0; i < cnt; i++) {
1742                         if (!Q_TYPESET(oqc, i))
1743                                 continue;
1744                         LASSERT(cdqb[i]);
1745                         cdqb[i]->dqb_curspace -= size;
1746                         cdqb[i]->dqb_curinodes--;
1747                 }
1748         }
1749
1750         return rc;
1751 }
1752
1753 /* write dqinfo struct in a new quota file */
1754 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1755 {
1756         struct v2_disk_dqinfo dqinfo;
1757         __u32 blocks = V2_DQTREEOFF + 1;
1758         loff_t offset = V2_DQINFOOFF;
1759
1760         if (info) {
1761                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1762                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1763                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1764                                                ~DQF_INFO_DIRTY);
1765         } else {
1766                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1767                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1768                 dqinfo.dqi_flags = 0;
1769         }
1770
1771         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1772         dqinfo.dqi_free_blk = 0;
1773         dqinfo.dqi_free_entry = 0;
1774
1775         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1776 }
1777
1778 static int v3_write_dqheader(struct file *f, int type)
1779 {
1780         static const __u32 quota_magics[] = V2_INITQMAGICS;
1781         static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1782         struct v2_disk_dqheader dqhead;
1783         loff_t offset = 0;
1784
1785         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1786         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1787
1788         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1789         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1790
1791         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1792 }
1793
1794 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1795                                   struct obd_quotactl *oqc)
1796 {
1797         int i, rc = 0;
1798         ENTRY;
1799
1800         for (i = 0; i < MAXQUOTAS; i++) {
1801                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1802                                          NULL : &qctxt->qckt_dqinfo[i];
1803                 struct file *file;
1804                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1805
1806                 if (!Q_TYPESET(oqc, i))
1807                         continue;
1808
1809                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1810
1811                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1812                 if (IS_ERR(file)) {
1813                         rc = PTR_ERR(file);
1814                         CERROR("can't create %s file: rc = %d\n",
1815                                name[i], rc);
1816                         GOTO(out, rc);
1817                 }
1818
1819                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1820                         CERROR("file %s is not regular", name[i]);
1821                         filp_close(file, 0);
1822                         GOTO(out, rc = -EINVAL);
1823                 }
1824
1825                 DQUOT_DROP(file->f_dentry->d_inode);
1826
1827                 rc = v3_write_dqheader(file, i);
1828                 if (rc) {
1829                         filp_close(file, 0);
1830                         GOTO(out, rc);
1831                 }
1832
1833                 rc = v3_write_dqinfo(file, i, info);
1834                 filp_close(file, 0);
1835                 if (rc)
1836                         GOTO(out, rc);
1837         }
1838
1839 out:
1840         RETURN(rc);
1841 }
1842
1843
1844 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1845                           struct chk_dqblk *cdqb)
1846 {
1847         struct obd_quotactl *oqc;
1848         long now;
1849         int rc;
1850         ENTRY;
1851
1852         OBD_ALLOC_PTR(oqc);
1853         if (!oqc)
1854                 RETURN(-ENOMEM);
1855
1856         now = cfs_time_current_sec();
1857
1858         if (cdqb->dqb_bsoftlimit &&
1859             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1860             !cdqb->dqb_btime)
1861                 cdqb->dqb_btime =
1862                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1863
1864         if (cdqb->dqb_isoftlimit &&
1865             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1866             !cdqb->dqb_itime)
1867                 cdqb->dqb_itime =
1868                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1869
1870         cdqb->dqb_valid = QIF_ALL;
1871
1872         oqc->qc_cmd = Q_SETQUOTA;
1873         oqc->qc_type = cdqb->dqb_type;
1874         oqc->qc_id = cdqb->dqb_id;
1875         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1876
1877         rc = fsfilt_ext3_quotactl(sb, oqc);
1878         OBD_FREE_PTR(oqc);
1879         RETURN(rc);
1880 }
1881
1882 static int prune_chkquots(struct super_block *sb,
1883                           struct qchk_ctxt *qctxt, int error)
1884 {
1885         struct chk_dqblk *cdqb, *tmp;
1886         int rc;
1887
1888         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1889                 if (!error) {
1890                         rc = commit_chkquot(sb, qctxt, cdqb);
1891                         if (rc)
1892                                 error = rc;
1893                 }
1894                 hlist_del_init(&cdqb->dqb_hash);
1895                 list_del(&cdqb->dqb_list);
1896                 OBD_FREE_PTR(cdqb);
1897         }
1898
1899         return error;
1900 }
1901
1902 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1903                                   struct obd_quotactl *oqc)
1904 {
1905         struct ext3_sb_info *sbi = EXT3_SB(sb);
1906         int i, group;
1907         struct qchk_ctxt *qctxt;
1908         struct buffer_head *bitmap_bh = NULL;
1909         unsigned long ino;
1910         struct inode *inode;
1911         int rc = 0;
1912         ENTRY;
1913
1914         /* turn on quota and read dqinfo if existed */
1915         OBD_ALLOC_PTR(qctxt);
1916         if (!qctxt) {
1917                 oqc->qc_stat = -ENOMEM;
1918                 RETURN(-ENOMEM);
1919         }
1920
1921         for (i = 0; i < NR_DQHASH; i++)
1922                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1923         INIT_LIST_HEAD(&qctxt->qckt_list);
1924
1925         for (i = 0; i < MAXQUOTAS; i++) {
1926                 if (!Q_TYPESET(oqc, i))
1927                         continue;
1928
1929                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1930                 if (!rc || rc == -EBUSY) {
1931                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1932                         if (rc)
1933                                 GOTO(out, rc);
1934                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1935                         qctxt->qckt_first_check[i] = 1;
1936                 } else if (rc) {
1937                         GOTO(out, rc);
1938                 }
1939         }
1940
1941         /* check quota and update in hash */
1942         for (group = 0; group < sbi->s_groups_count; group++) {
1943                 ino = group * sbi->s_inodes_per_group + 1;
1944                 bitmap_bh = read_inode_bitmap(sb, group);
1945                 if (!bitmap_bh) {
1946                         CERROR("read_inode_bitmap group %d failed", group);
1947                         GOTO(out, rc = -EIO);
1948                 }
1949
1950                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1951                         if (ino < sbi->s_first_ino)
1952                                 continue;
1953
1954                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1955                         rc = add_inode_quota(inode, qctxt, oqc);
1956                         iput(inode);
1957                         if (rc) {
1958                                 brelse(bitmap_bh);
1959                                 GOTO(out, rc);
1960                         }
1961                 }
1962
1963                 brelse(bitmap_bh);
1964         }
1965
1966         /* read old quota limits from old quota file. (only for the user
1967          * has limits but hasn't file) */
1968 #ifdef HAVE_QUOTA_SUPPORT
1969         for (i = 0; i < MAXQUOTAS; i++) {
1970                 struct list_head id_list;
1971                 struct dquot_id *dqid, *tmp;
1972
1973                 if (!Q_TYPESET(oqc, i))
1974                         continue;
1975
1976                 if (qctxt->qckt_first_check[i])
1977                         continue;
1978
1979
1980                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1981                 INIT_LIST_HEAD(&id_list);
1982 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1983                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1984 #else
1985                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1986 #endif
1987                 if (rc)
1988                         CERROR("read old limits failed. (rc:%d)\n", rc);
1989
1990                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1991                         list_del_init(&dqid->di_link);
1992
1993                         if (!rc)
1994                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1995                                       dqid->di_id, i,
1996                                       qctxt->qckt_first_check[i]);
1997                         kfree(dqid);
1998                 }
1999         }
2000 #endif
2001         /* turn off quota cause we are to dump chk_dqblk to files */
2002         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2003
2004         rc = create_new_quota_files(qctxt, oqc);
2005         if (rc)
2006                 GOTO(out, rc);
2007
2008         /* we use vfs functions to set dqblk, so turn quota on */
2009         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2010 out:
2011         /* dump and free chk_dqblk */
2012         rc = prune_chkquots(sb, qctxt, rc);
2013         OBD_FREE_PTR(qctxt);
2014
2015         /* turn off quota, `lfs quotacheck` will turn on when all
2016          * nodes quotacheck finish. */
2017         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2018
2019         oqc->qc_stat = rc;
2020         if (rc)
2021                 CERROR("quotacheck failed: rc = %d\n", rc);
2022
2023         RETURN(rc);
2024 }
2025
2026 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2027                                  int cmd)
2028 {
2029         int rc = 0;
2030         ENTRY;
2031
2032         if (lqi->qi_files[type] == NULL) {
2033                 CERROR("operate qinfo before it's enabled!\n");
2034                 RETURN(-EIO);
2035         }
2036
2037         switch (cmd) {
2038         case QFILE_CHK:
2039                 rc = lustre_check_quota_file(lqi, type);
2040                 break;
2041         case QFILE_RD_INFO:
2042                 rc = lustre_read_quota_info(lqi, type);
2043                 break;
2044         case QFILE_WR_INFO:
2045                 rc = lustre_write_quota_info(lqi, type);
2046                 break;
2047         case QFILE_INIT_INFO:
2048                 rc = lustre_init_quota_info(lqi, type);
2049                 break;
2050         case QFILE_CONVERT:
2051                 rc = -ENOTSUPP;
2052                 CERROR("quota CONVERT command is not supported\n");
2053                 break;
2054         default:
2055                 rc = -ENOTSUPP;
2056                 CERROR("Unsupported admin quota file cmd %d\n"
2057                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2058                        cmd);
2059                 break;
2060         }
2061         RETURN(rc);
2062 }
2063
2064 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2065                             struct list_head *list)
2066 {
2067         return lustre_get_qids(file, inode, type, list);
2068 }
2069
2070 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2071 {
2072         int rc = 0;
2073         ENTRY;
2074
2075         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2076                 CERROR("operate dquot before it's enabled!\n");
2077                 RETURN(-EIO);
2078         }
2079
2080         switch (cmd) {
2081         case QFILE_RD_DQUOT:
2082                 rc = lustre_read_dquot(dquot);
2083                 break;
2084         case QFILE_WR_DQUOT:
2085                 if (dquot->dq_dqb.dqb_ihardlimit ||
2086                     dquot->dq_dqb.dqb_isoftlimit ||
2087                     dquot->dq_dqb.dqb_bhardlimit ||
2088                     dquot->dq_dqb.dqb_bsoftlimit)
2089                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2090                 else
2091                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2092
2093                 rc = lustre_commit_dquot(dquot);
2094                 if (rc >= 0)
2095                         rc = 0;
2096                 break;
2097         default:
2098                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2099                 LBUG();
2100                 break;
2101         }
2102         RETURN(rc);
2103 }
2104
2105 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2106                                 struct inode *inode, int frags)
2107 {
2108 #ifdef EXT3_EXT_HAS_NO_TREE
2109         struct ext3_ext_base *base = inode;
2110 #else
2111         struct ext3_extents_tree tree;
2112         struct ext3_ext_base *base = &tree;
2113
2114         ext3_init_tree_desc(base, inode);
2115 #endif
2116         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2117          * metablocks at maxium b=16542 */
2118         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2119         return 0;
2120 }
2121
2122 #endif
2123
2124 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2125 {
2126         return (EXT3_SB(sb)->journal_bdev);
2127 }
2128 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2129
2130 static struct fsfilt_operations fsfilt_ext3_ops = {
2131         .fs_type                = "ext3",
2132         .fs_owner               = THIS_MODULE,
2133         .fs_getlabel            = fsfilt_ext3_get_label,
2134         .fs_setlabel            = fsfilt_ext3_set_label,
2135         .fs_uuid                = fsfilt_ext3_uuid,
2136         .fs_start               = fsfilt_ext3_start,
2137         .fs_brw_start           = fsfilt_ext3_brw_start,
2138         .fs_extend              = fsfilt_ext3_extend,
2139         .fs_commit              = fsfilt_ext3_commit,
2140         .fs_commit_async        = fsfilt_ext3_commit_async,
2141         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2142         .fs_setattr             = fsfilt_ext3_setattr,
2143         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2144         .fs_set_md              = fsfilt_ext3_set_md,
2145         .fs_get_md              = fsfilt_ext3_get_md,
2146         .fs_readpage            = fsfilt_ext3_readpage,
2147         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2148         .fs_statfs              = fsfilt_ext3_statfs,
2149         .fs_sync                = fsfilt_ext3_sync,
2150         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2151         .fs_write_record        = fsfilt_ext3_write_record,
2152         .fs_read_record         = fsfilt_ext3_read_record,
2153         .fs_setup               = fsfilt_ext3_setup,
2154         .fs_send_bio            = fsfilt_ext3_send_bio,
2155         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2156 #ifdef HAVE_DISK_INODE_VERSION
2157         .fs_get_version         = fsfilt_ext3_get_version,
2158         .fs_set_version         = fsfilt_ext3_set_version,
2159 #endif
2160 #ifdef HAVE_QUOTA_SUPPORT
2161         .fs_quotactl            = fsfilt_ext3_quotactl,
2162         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2163         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2164         .fs_qids                = fsfilt_ext3_qids,
2165         .fs_dquot               = fsfilt_ext3_dquot,
2166         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2167 #endif
2168         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2169 };
2170
2171 static int __init fsfilt_ext3_init(void)
2172 {
2173         int rc;
2174
2175         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2176                                          sizeof(struct fsfilt_cb_data), 0, 0);
2177         if (!fcb_cache) {
2178                 CERROR("error allocating fsfilt journal callback cache\n");
2179                 GOTO(out, rc = -ENOMEM);
2180         }
2181
2182         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2183
2184         if (rc) {
2185                 int err = cfs_mem_cache_destroy(fcb_cache);
2186                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2187         }
2188 out:
2189         return rc;
2190 }
2191
2192 static void __exit fsfilt_ext3_exit(void)
2193 {
2194         int rc;
2195
2196         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2197         rc = cfs_mem_cache_destroy(fcb_cache);
2198         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2199 }
2200
2201 module_init(fsfilt_ext3_init);
2202 module_exit(fsfilt_ext3_exit);
2203
2204 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2205 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2206 MODULE_LICENSE("GPL");