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 (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  *
  38  * lustre/lvfs/fsfilt_ext3.c
  39  *
  40  * Author: Andreas Dilger <adilger@clusterfs.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_FILTER
  44
  45 #include <linux/init.h>
  46 #include <linux/module.h>
  47 #include <linux/fs.h>
  48 #include <linux/slab.h>
  49 #include <linux/pagemap.h>
  50 #include <linux/quotaops.h>
  51 #ifdef HAVE_LINUX_EXPORTFS_H
  52 #include <linux/exportfs.h>
  53 #endif
  54 #ifdef HAVE_EXT4_LDISKFS
  55 #include <ext4/ext4.h>
  56 #include <ext4/ext4_jbd2.h>
  57 #else
  58 #include <linux/jbd.h>
  59 #include <linux/ext3_fs.h>
  60 #include <linux/ext3_jbd.h>
  61 #endif
  62 #include <linux/version.h>
  63 #include <linux/bitops.h>
  64 #include <linux/quota.h>
  65 #ifdef HAVE_QUOTAIO_H
  66 # include <linux/quotaio_v2.h>
  67 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  68 # include <quota/quotaio_v2.h>
  69 # include <quota/quota_tree.h>
  70 # define V2_DQTREEOFF    QT_TREEOFF
  71 #elif defined(HAVE_FS_QUOTAIO_H)
  72 # include <quotaio_v2.h>
  73 # include <quota_tree.h>
  74 # define V2_DQTREEOFF    QT_TREEOFF
  75 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  76 #endif
  77
  78 #ifdef QFMT_VFS_V1
  79 #define QFMT_LUSTRE QFMT_VFS_V1
  80 #else
  81 #define QFMT_LUSTRE QFMT_VFS_V0
  82 #endif
  83
  84 #if defined(HAVE_EXT3_XATTR_H)
  85 # include <ext3/xattr.h>
  86 #elif !defined(EXT3_XATTR_INDEX_TRUSTED)
  87 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  88 /* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
  89 # define EXT3_XATTR_INDEX_TRUSTED        4
  90 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  91 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  92 #endif
  93
  94 #include <libcfs/libcfs.h>
  95 #include <lustre_fsfilt.h>
  96 #include <obd.h>
  97 #include <lustre_quota.h>
  98 #include <linux/lustre_compat25.h>
  99 #include <linux/lprocfs_status.h>
 100
 101 #ifdef HAVE_EXT4_LDISKFS
 102 #include <ext4/ext4_extents.h>
 103 #else
 104 #include <linux/ext3_extents.h>
 105 #endif
 106
 107 #include "lustre_quota_fmt.h"
 108
 109 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
 110 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 111 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 112 #else
 113 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 114 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 115 #endif
 116
 117 /* for kernels 2.6.18 and later */
 118 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 119
 120 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 121                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 122
 123 #define ext3_mb_discard_inode_preallocations(inode) \
 124                  ext3_discard_preallocations(inode)
 125
 126 #ifdef HAVE_EXT4_LDISKFS
 127 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 128 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 129 #else
 130 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 131 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 132 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 133 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 134 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 135 #endif
 136
 137 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 138 # define journal_callback ext4_journal_cb_entry
 139 # define fsfilt_journal_callback_set(handle, func, jcb) \
 140          ext4_journal_callback_add(handle, func, jcb)
 141 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
 142 # define fsfilt_journal_callback_set(handle, func, jcb) \
 143          jbd2_journal_callback_set(handle, func, jcb)
 144 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
 145 # define fsfilt_journal_callback_set(handle, func, jcb) \
 146          journal_callback_set(handle, func, jcb)
 147 #else
 148 # error missing journal commit callback
 149 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
 150
 151 static cfs_mem_cache_t *fcb_cache;
 152
 153 struct fsfilt_cb_data {
 154         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 155         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 156         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 157         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 158         void *cb_data;                  /* MDS/OST completion function data */
 159 };
 160
 161 #ifndef ext3_find_next_bit
 162 #define ext3_find_next_bit           ext2_find_next_bit
 163 #endif
 164
 165 #ifndef ext2_find_next_bit
 166 #ifdef __LITTLE_ENDIAN
 167 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 168 #else
 169 error "Need implementation of find_next_bit on big-endian systems"
 170 #endif  /* __LITTLE_ENDIAN */
 171 #endif  /* !ext2_find_next_le_bit */
 172
 173 static char *fsfilt_ext3_get_label(struct super_block *sb)
 174 {
 175         return EXT3_SB(sb)->s_es->s_volume_name;
 176 }
 177
 178 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 179 {
 180         /* see e.g. fsfilt_ext3_write_record() */
 181         journal_t *journal;
 182         handle_t *handle;
 183         int err;
 184
 185         journal = EXT3_SB(sb)->s_journal;
 186         handle = ext3_journal_start_sb(sb, 1);
 187         if (IS_ERR(handle)) {
 188                 CERROR("can't start transaction\n");
 189                 return(PTR_ERR(handle));
 190         }
 191
 192         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 193         if (err)
 194                 goto out;
 195
 196         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 197                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 198
 199         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 200
 201 out:
 202         ext3_journal_stop(handle);
 203
 204         return(err);
 205 }
 206
 207 static char *fsfilt_ext3_uuid(struct super_block *sb)
 208 {
 209         return EXT3_SB(sb)->s_es->s_uuid;
 210 }
 211
 212 #ifdef HAVE_DISK_INODE_VERSION
 213
 214 static __u64 get_i_version(struct inode *inode)
 215 {
 216         return EXT3_I(inode)->i_fs_version;
 217 }
 218
 219 static void set_i_version(struct inode *inode, __u64 new_version)
 220 {
 221         (EXT3_I(inode))->i_fs_version = new_version;
 222 }
 223
 224 /*
 225  * Get the 64-bit version for an inode.
 226  */
 227 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 228 {
 229         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 230                get_i_version(inode), inode->i_ino);
 231         return get_i_version(inode);
 232 }
 233
 234 /*
 235  * Set the 64-bit version and return the old version.
 236  */
 237 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 238 {
 239         __u64 old_version = get_i_version(inode);
 240
 241         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 242                new_version, old_version, inode->i_ino);
 243         set_i_version(inode, new_version);
 244         /* version is set after all inode operations are finished, so we should
 245          * mark it dirty here */
 246         inode->i_sb->s_op->dirty_inode(inode);
 247         return old_version;
 248 }
 249
 250 #endif
 251
 252 /*
 253  * We don't currently need any additional blocks for rmdir and
 254  * unlink transactions because we are storing the OST oa_id inside
 255  * the inode (which we will be changing anyways as part of this
 256  * transaction).
 257  */
 258 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 259                                int logs)
 260 {
 261         /* For updates to the last received file */
 262         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 263         journal_t *journal;
 264         void *handle;
 265
 266         if (current->journal_info) {
 267                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 268                        current->journal_info);
 269                 goto journal_start;
 270         }
 271
 272         switch(op) {
 273         case FSFILT_OP_RMDIR:
 274         case FSFILT_OP_UNLINK:
 275                 /* delete one file + create/update logs for each stripe */
 276                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 277                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 278                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 279                 break;
 280         case FSFILT_OP_RENAME:
 281                 /* modify additional directory */
 282                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 283                 /* no break */
 284         case FSFILT_OP_SYMLINK:
 285                 /* additional block + block bitmap + GDT for long symlink */
 286                 nblocks += 3;
 287                 /* no break */
 288         case FSFILT_OP_CREATE: {
 289 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 290                 static int warned;
 291                 if (!warned) {
 292                         if (!test_opt(inode->i_sb, EXTENTS)) {
 293                                 warned = 1;
 294                         } else if (((EXT3_I(inode)->i_flags &
 295                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 296                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 297                                 CWARN("extent-mapped directory found with "
 298                                       "ext3-based ldiskfs - contact "
 299                                       "http://bugs.whamcloud.com/\n");
 300                                 warned = 1;
 301                         }
 302                 }
 303 #endif
 304                 /* no break */
 305         }
 306         case FSFILT_OP_MKDIR:
 307         case FSFILT_OP_MKNOD:
 308                 /* modify one inode + block bitmap + GDT */
 309                 nblocks += 3;
 310                 /* no break */
 311         case FSFILT_OP_LINK:
 312                 /* modify parent directory */
 313                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 314                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 315                 /* create/update logs for each stripe */
 316                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 317                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 318                 break;
 319         case FSFILT_OP_SETATTR:
 320                 /* Setattr on inode */
 321                 nblocks += 1;
 322                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 323                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 324                 /* quota chown log for each stripe */
 325                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 326                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 327                 break;
 328         case FSFILT_OP_CANCEL_UNLINK:
 329                 /* blocks for log header bitmap update OR
 330                  * blocks for catalog header bitmap update + unlink of logs */
 331                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 332                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 333                 break;
 334         default: CERROR("unknown transaction start op %d\n", op);
 335                 LBUG();
 336         }
 337
 338         LASSERT(current->journal_info == desc_private);
 339         journal = EXT3_SB(inode->i_sb)->s_journal;
 340         if (nblocks > journal->j_max_transaction_buffers) {
 341                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 342                        nblocks, op, logs, journal->j_max_transaction_buffers);
 343                 nblocks = journal->j_max_transaction_buffers;
 344         }
 345
 346  journal_start:
 347         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 348         handle = ext3_journal_start(inode, nblocks);
 349
 350         if (!IS_ERR(handle))
 351                 LASSERT(current->journal_info == handle);
 352         else
 353                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 354                        op, nblocks, PTR_ERR(handle));
 355         return handle;
 356 }
 357
 358 /*
 359  * Calculate the number of buffer credits needed to write multiple pages in
 360  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 361  * doesn't have a nice API for calculating this sort of thing in advance.
 362  *
 363  * See comment above ext3_writepage_trans_blocks for details.  We assume
 364  * no data journaling is being done, but it does allow for all of the pages
 365  * being non-contiguous.  If we are guaranteed contiguous pages we could
 366  * reduce the number of (d)indirect blocks a lot.
 367  *
 368  * With N blocks per page and P pages, for each inode we have at most:
 369  * N*P indirect
 370  * min(N*P, blocksize/4 + 1) dindirect blocks
 371  * niocount tindirect
 372  *
 373  * For the entire filesystem, we have at most:
 374  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 375  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 376  * objcount inode blocks
 377  * 1 superblock
 378  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 379  *
 380  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 381  */
 382 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 383                                       int niocount, struct niobuf_local *nb)
 384 {
 385         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 386         __u64 next_indir;
 387         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 388         int nbitmaps = 0, ngdblocks;
 389         int needed = objcount + 1; /* inodes + superblock */
 390         int i, j;
 391
 392         for (i = 0, j = 0; i < objcount; i++, fso++) {
 393                 /* two or more dindirect blocks in case we cross boundary */
 394                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 395                                     nb[j].offset) >>
 396                                    sb->s_blocksize_bits) /
 397                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 398                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 399
 400                 /* leaf, indirect, tindirect blocks for first block */
 401                 nbitmaps += blockpp + 2;
 402
 403                 j += fso->fso_bufcnt;
 404         }
 405
 406         next_indir = nb[0].offset +
 407                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 408         for (i = 1; i < niocount; i++) {
 409                 if (nb[i].offset >= next_indir) {
 410                         nbitmaps++;     /* additional indirect */
 411                         next_indir = nb[i].offset +
 412                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 413                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 414                         nbitmaps++;     /* additional indirect */
 415                 }
 416                 nbitmaps += blockpp;    /* each leaf in different group? */
 417         }
 418
 419         ngdblocks = nbitmaps;
 420         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 421                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 422         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 423                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 424
 425         needed += nbitmaps + ngdblocks;
 426
 427         /* last_rcvd update */
 428         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 429
 430 #if defined(CONFIG_QUOTA)
 431         /* We assume that there will be 1 bit set in s_dquot.flags for each
 432          * quota file that is active.  This is at least true for now.
 433          */
 434         needed += hweight32(ll_sb_any_quota_active(sb)) *
 435                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 436 #endif
 437
 438         return needed;
 439 }
 440
 441 /* We have to start a huge journal transaction here to hold all of the
 442  * metadata for the pages being written here.  This is necessitated by
 443  * the fact that we do lots of prepare_write operations before we do
 444  * any of the matching commit_write operations, so even if we split
 445  * up to use "smaller" transactions none of them could complete until
 446  * all of them were opened.  By having a single journal transaction,
 447  * we eliminate duplicate reservations for common blocks like the
 448  * superblock and group descriptors or bitmaps.
 449  *
 450  * We will start the transaction here, but each prepare_write will
 451  * add a refcount to the transaction, and each commit_write will
 452  * remove a refcount.  The transaction will be closed when all of
 453  * the pages have been written.
 454  */
 455 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 456                                    int niocount, struct niobuf_local *nb,
 457                                    void *desc_private, int logs)
 458 {
 459         journal_t *journal;
 460         handle_t *handle;
 461         int needed;
 462         ENTRY;
 463
 464         LASSERT(current->journal_info == desc_private);
 465         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 466         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 467
 468         /* The number of blocks we could _possibly_ dirty can very large.
 469          * We reduce our request if it is absurd (and we couldn't get that
 470          * many credits for a single handle anyways).
 471          *
 472          * At some point we have to limit the size of I/Os sent at one time,
 473          * increase the size of the journal, or we have to calculate the
 474          * actual journal requirements more carefully by checking all of
 475          * the blocks instead of being maximally pessimistic.  It remains to
 476          * be seen if this is a real problem or not.
 477          */
 478         if (needed > journal->j_max_transaction_buffers) {
 479                 CERROR("want too many journal credits (%d) using %d instead\n",
 480                        needed, journal->j_max_transaction_buffers);
 481                 needed = journal->j_max_transaction_buffers;
 482         }
 483
 484         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 485         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 486         if (IS_ERR(handle)) {
 487                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 488                        PTR_ERR(handle));
 489         } else {
 490                 LASSERT(handle->h_buffer_credits >= needed);
 491                 LASSERT(current->journal_info == handle);
 492         }
 493
 494         RETURN(handle);
 495 }
 496
 497 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 498 {
 499        handle_t *handle = h;
 500
 501        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 502        if (nblocks == 0) {
 503                handle->h_buffer_credits = 0;
 504                CWARN("setting credits of handle %p to zero by request\n", h);
 505        }
 506
 507        if (handle->h_buffer_credits > nblocks)
 508                 return 0;
 509        if (ext3_journal_extend(handle, nblocks) == 0)
 510                 return 0;
 511
 512        ext3_mark_inode_dirty(handle, inode);
 513        return ext3_journal_restart(handle, nblocks);
 514 }
 515
 516 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 517 {
 518         int rc;
 519         handle_t *handle = h;
 520
 521         LASSERT(current->journal_info == handle);
 522         if (force_sync)
 523                 handle->h_sync = 1; /* recovery likes this */
 524
 525         rc = ext3_journal_stop(handle);
 526
 527         return rc;
 528 }
 529
 530 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 531                                     void **wait_handle)
 532 {
 533         unsigned long tid;
 534         transaction_t *transaction;
 535         handle_t *handle = h;
 536         journal_t *journal;
 537         int rc;
 538
 539         LASSERT(current->journal_info == handle);
 540
 541         transaction = handle->h_transaction;
 542         journal = transaction->t_journal;
 543         tid = transaction->t_tid;
 544         /* we don't want to be blocked */
 545         handle->h_sync = 0;
 546         rc = ext3_journal_stop(handle);
 547         if (rc) {
 548                 CERROR("error while stopping transaction: %d\n", rc);
 549                 return rc;
 550         }
 551         fsfilt_log_start_commit(journal, tid);
 552
 553         *wait_handle = (void *) tid;
 554         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 555         return 0;
 556 }
 557
 558 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 559 {
 560         journal_t *journal = EXT3_JOURNAL(inode);
 561         tid_t tid = (tid_t)(long)h;
 562
 563         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 564         if (unlikely(is_journal_aborted(journal)))
 565                 return -EIO;
 566
 567         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 568
 569         if (unlikely(is_journal_aborted(journal)))
 570                 return -EIO;
 571         return 0;
 572 }
 573
 574 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 575                                struct iattr *iattr, int do_trunc)
 576 {
 577         struct inode *inode = dentry->d_inode;
 578         int rc = 0;
 579
 580 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
 581         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
 582          * timestamps to appear to be in the far future, due old timestamp
 583          * being stored on disk as an unsigned value.  This fixes up any
 584          * bad values held by the client before storing them on disk,
 585          * and ensures any timestamp updates are correct.  LU-1042 */
 586         if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
 587                      !(iattr->ia_valid & ATTR_ATIME))) {
 588                 iattr->ia_valid |= ATTR_ATIME;
 589                 LTIME_S(iattr->ia_atime) = 0;
 590         }
 591         if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
 592                      !(iattr->ia_valid & ATTR_MTIME))) {
 593                 iattr->ia_valid |= ATTR_MTIME;
 594                 LTIME_S(iattr->ia_mtime) = 0;
 595         }
 596         if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
 597                       LTIME_S(inode->i_ctime) == 0) &&
 598                      !(iattr->ia_valid & ATTR_CTIME))) {
 599                 iattr->ia_valid |= ATTR_CTIME;
 600                 LTIME_S(iattr->ia_ctime) = 0;
 601         }
 602 #else
 603 #warning "remove old LU-221/LU-1042 workaround code"
 604 #endif
 605
 606         /* When initializating timestamps for new inodes, use the filesystem
 607          * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
 608          * that this is potentially an invalid inode.  Files with an old ctime
 609          * migrated to a newly-formatted OST with a newer s_mkfs_time will not
 610          * hit this check, since it is only for ctime == 0.  LU-1010/LU-1042 */
 611         if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
 612                 LTIME_S(iattr->ia_ctime) =
 613                         EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
 614
 615         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 616          * We are already writing the inode to disk as part of this
 617          * transaction and want to avoid a lot of extra inode writeout
 618          * later on. b=9828 */
 619         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 620                 /* ATTR_SIZE would invoke truncate: clear it */
 621                 iattr->ia_valid &= ~ATTR_SIZE;
 622                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 623                 i_size_write(inode, iattr->ia_size);
 624
 625                 if (iattr->ia_valid & ATTR_UID)
 626                         inode->i_uid = iattr->ia_uid;
 627                 if (iattr->ia_valid & ATTR_GID)
 628                         inode->i_gid = iattr->ia_gid;
 629                 if (iattr->ia_valid & ATTR_ATIME)
 630                         inode->i_atime = iattr->ia_atime;
 631                 if (iattr->ia_valid & ATTR_MTIME)
 632                         inode->i_mtime = iattr->ia_mtime;
 633                 if (iattr->ia_valid & ATTR_CTIME)
 634                         inode->i_ctime = iattr->ia_ctime;
 635                 if (iattr->ia_valid & ATTR_MODE) {
 636                         inode->i_mode = iattr->ia_mode;
 637
 638                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 639                             !cfs_capable(CFS_CAP_FSETID))
 640                                 inode->i_mode &= ~S_ISGID;
 641                 }
 642
 643                 inode->i_sb->s_op->dirty_inode(inode);
 644
 645                 goto out;
 646         }
 647
 648         /* Don't allow setattr to change file type */
 649         if (iattr->ia_valid & ATTR_MODE)
 650                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 651                                  (iattr->ia_mode & ~S_IFMT);
 652
 653         /* We set these flags on the client, but have already checked perms
 654          * so don't confuse inode_change_ok. */
 655         iattr->ia_valid &= ~TIMES_SET_FLAGS;
 656
 657         if (inode->i_op->setattr) {
 658                 rc = inode->i_op->setattr(dentry, iattr);
 659         } else {
 660 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
 661                 rc = inode_change_ok(inode, iattr);
 662                 if (!rc)
 663 #endif
 664                         rc = simple_setattr(dentry, iattr);
 665         }
 666
 667  out:
 668         RETURN(rc);
 669 }
 670
 671 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 672                                  unsigned int cmd, unsigned long arg)
 673 {
 674         int rc = 0;
 675         ENTRY;
 676
 677         /* FIXME: Can't do this because of nested transaction deadlock */
 678         if (cmd == EXT3_IOC_SETFLAGS) {
 679                 /* We can't enable data journaling on OST objects, because
 680                 * this forces the transaction to be closed in order to
 681                 * flush the journal, but the caller will already have a
 682                 * compound transaction open to update the last_rcvd file,
 683                 * and this thread would deadlock trying to set the flag. */
 684                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 685                         CERROR("can't set data journal flag on file\n");
 686                         RETURN(-EPERM);
 687                 }
 688                 /* Because the MDS does not see the EXTENTS_FL set on the
 689                  * OST objects, mask this flag into all set flags.  It is
 690                  * not legal to clear this flag in any case, so we are not
 691                  * changing the functionality by doing this.  b=22911 */
 692                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 693         }
 694
 695 #ifdef HAVE_EXT4_LDISKFS
 696         /* ext4_ioctl does not have a inode argument */
 697         if (inode->i_fop->unlocked_ioctl)
 698                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 699 #else
 700         if (inode->i_fop->ioctl)
 701                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 702 #endif
 703         else
 704                 RETURN(-ENOTTY);
 705
 706         RETURN(rc);
 707 }
 708
 709 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 710                               void *lmm, int lmm_size, const char *name)
 711 {
 712         int rc;
 713
 714         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 715
 716         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 717                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 718
 719
 720         if (rc && rc != -EROFS)
 721                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 722                        inode->i_ino, rc);
 723         return rc;
 724 }
 725
 726 /* Must be called with i_mutex held */
 727 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 728                               const char *name)
 729 {
 730         int rc;
 731
 732         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 733
 734         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 735                             name, lmm, lmm_size);
 736
 737         /* This gives us the MD size */
 738         if (lmm == NULL)
 739                 return (rc == -ENODATA) ? 0 : rc;
 740
 741         if (rc < 0) {
 742                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 743                        EXT3_XATTR_INDEX_TRUSTED, name,
 744                        inode->i_ino, rc);
 745                 memset(lmm, 0, lmm_size);
 746                 return (rc == -ENODATA) ? 0 : rc;
 747         }
 748
 749         return rc;
 750 }
 751
 752 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 753 {
 754         submit_bio(rw, bio);
 755         return 0;
 756 }
 757
 758 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 759                                     loff_t *off)
 760 {
 761         struct inode *inode = file->f_dentry->d_inode;
 762         int rc = 0;
 763
 764         if (S_ISREG(inode->i_mode))
 765                 rc = file->f_op->read(file, buf, count, off);
 766         else {
 767                 const int blkbits = inode->i_sb->s_blocksize_bits;
 768                 const int blksize = inode->i_sb->s_blocksize;
 769
 770                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 771                        (unsigned long)count, inode->i_ino, *off);
 772                 while (count > 0) {
 773                         struct buffer_head *bh;
 774
 775                         bh = NULL;
 776                         if (*off < i_size_read(inode)) {
 777                                 int err = 0;
 778
 779                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 780                                                 0, &err);
 781
 782                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 783
 784                                 if (bh) {
 785                                         memcpy(buf, bh->b_data, blksize);
 786                                         brelse(bh);
 787                                 } else if (err) {
 788                                         /* XXX in theory we should just fake
 789                                          * this buffer and continue like ext3,
 790                                          * especially if this is a partial read
 791                                          */
 792                                         CERROR("error read dir %lu+%llu: %d\n",
 793                                                inode->i_ino, *off, err);
 794                                         RETURN(err);
 795                                 }
 796                         }
 797                         if (!bh) {
 798                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 799
 800                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 801                                 memset(fake, 0, sizeof(*fake));
 802                                 fake->rec_len = cpu_to_le16(blksize);
 803                         }
 804                         count -= blksize;
 805                         buf += blksize;
 806                         *off += blksize;
 807                         rc += blksize;
 808                 }
 809         }
 810
 811         return rc;
 812 }
 813
 814 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 815 static void fsfilt_ext3_cb_func(struct super_block *sb,
 816                                 struct journal_callback *jcb, int error)
 817 #else
 818 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 819 #endif
 820 {
 821         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 822
 823         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 824
 825         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 826 }
 827
 828 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 829                                       void *handle, fsfilt_cb_t cb_func,
 830                                       void *cb_data)
 831 {
 832         struct fsfilt_cb_data *fcb;
 833
 834         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 835         if (fcb == NULL)
 836                 RETURN(-ENOMEM);
 837
 838         fcb->cb_func = cb_func;
 839         fcb->cb_obd = obd;
 840         fcb->cb_last_rcvd = last_rcvd;
 841         fcb->cb_data = cb_data;
 842
 843         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 844         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 845
 846         return 0;
 847 }
 848
 849 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 850 {
 851         struct kstatfs sfs;
 852         int rc;
 853
 854         memset(&sfs, 0, sizeof(sfs));
 855         rc = ll_do_statfs(sb, &sfs);
 856         statfs_pack(osfs, &sfs);
 857         return rc;
 858 }
 859
 860 static int fsfilt_ext3_sync(struct super_block *sb)
 861 {
 862         return ext3_force_commit(sb);
 863 }
 864
 865 #ifndef EXT3_EXTENTS_FL
 866 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 867 #endif
 868
 869 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 870 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 871 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 872 #else
 873 # ifdef HAVE_EXT4_LDISKFS
 874 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 875 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 876 # else
 877 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 878 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 879 # endif
 880 #endif
 881
 882 #ifndef EXT_ASSERT
 883 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 884 #endif
 885
 886 #ifdef EXT3_EXT_HAS_NO_TREE
 887 /* for kernels 2.6.18 and later */
 888 #ifdef HAVE_EXT4_LDISKFS
 889 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 890 #else
 891 #define EXT_GENERATION(inode)           ext_generation(inode)
 892 #endif
 893 #define ext3_ext_base                   inode
 894 #define ext3_ext_base2inode(inode)      (inode)
 895 #define EXT_DEPTH(inode)                ext_depth(inode)
 896 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 897                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 898 #else
 899 #define ext3_ext_base                   ext3_extents_tree
 900 #define ext3_ext_base2inode(tree)       (tree->inode)
 901 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 902                         ext3_ext_walk_space(tree, block, num, cb);
 903 #endif
 904
 905 #include <linux/lustre_version.h>
 906
 907 struct bpointers {
 908         unsigned long *blocks;
 909         int *created;
 910         unsigned long start;
 911         int num;
 912         int init_num;
 913         int create;
 914 };
 915
 916 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 917                                unsigned long block, int *aflags)
 918 {
 919         struct ext3_inode_info *ei = EXT3_I(inode);
 920         unsigned long bg_start;
 921         unsigned long colour;
 922         int depth;
 923
 924         if (path) {
 925                 struct ext3_extent *ex;
 926                 depth = path->p_depth;
 927
 928                 /* try to predict block placement */
 929                 if ((ex = path[depth].p_ext))
 930                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 931
 932                 /* it looks index is empty
 933                  * try to find starting from index itself */
 934                 if (path[depth].p_bh)
 935                         return path[depth].p_bh->b_blocknr;
 936         }
 937
 938         /* OK. use inode's group */
 939         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 940                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 941         colour = (current->pid % 16) *
 942                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 943         return bg_start + colour + block;
 944 }
 945
 946 #define ll_unmap_underlying_metadata(sb, blocknr) \
 947         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 948
 949 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 950 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 951                                 struct ext3_ext_path *path, unsigned long block,
 952                                 unsigned long *count, int *err)
 953 {
 954         unsigned long pblock, goal;
 955         int aflags = 0;
 956         struct inode *inode = ext3_ext_base2inode(base);
 957
 958         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 959         aflags |= 2; /* block have been already reserved */
 960         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 961         return pblock;
 962
 963 }
 964 #else
 965 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 966                                 struct ext3_ext_path *path, unsigned long block,
 967                                 unsigned long *count, int *err)
 968 {
 969         struct inode *inode = ext3_ext_base2inode(base);
 970         struct ext3_allocation_request ar;
 971         unsigned long pblock;
 972         int aflags;
 973
 974         /* find neighbour allocated blocks */
 975         ar.lleft = block;
 976         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 977         if (*err)
 978                 return 0;
 979         ar.lright = block;
 980         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 981         if (*err)
 982                 return 0;
 983
 984         /* allocate new block */
 985         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 986         ar.inode = inode;
 987         ar.logical = block;
 988         ar.len = *count;
 989         ar.flags = EXT3_MB_HINT_DATA;
 990         pblock = ext3_mb_new_blocks(handle, &ar, err);
 991         *count = ar.len;
 992         return pblock;
 993 }
 994 #endif
 995
 996 #ifdef EXT3_EXT_HAS_NO_TREE
 997 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 998                                   struct ext3_ext_path *path,
 999                                   struct ext3_ext_cache *cex,
1000 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1001                                    struct ext3_extent *ex,
1002 #endif
1003                                   void *cbdata)
1004 {
1005         struct bpointers *bp = cbdata;
1006 #else
1007 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1008                                   struct ext3_ext_path *path,
1009                                   struct ext3_ext_cache *cex
1010 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1011                                   , struct ext3_extent *ex
1012 #endif
1013                                  )
1014 {
1015         struct bpointers *bp = base->private;
1016 #endif
1017         struct inode *inode = ext3_ext_base2inode(base);
1018         struct ext3_extent nex;
1019         unsigned long pblock;
1020         unsigned long tgen;
1021         int err, i;
1022         unsigned long count;
1023         handle_t *handle;
1024
1025         i = EXT_DEPTH(base);
1026         EXT_ASSERT(path[i].p_hdr);
1027         LASSERTF(i == path->p_depth ||
1028                  EXT_GENERATION(base) != path[0].p_generation,
1029                  "base vs path extent depth:%d != %d, generation:%lu == %lu\n",
1030                  i, path->p_depth, EXT_GENERATION(base), path[0].p_generation);
1031
1032         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1033                 err = EXT_CONTINUE;
1034                 goto map;
1035         }
1036
1037         if (bp->create == 0) {
1038                 i = 0;
1039                 if (cex->ec_block < bp->start)
1040                         i = bp->start - cex->ec_block;
1041                 if (i >= cex->ec_len)
1042                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1043                                         i, cex->ec_len);
1044                 for (; i < cex->ec_len && bp->num; i++) {
1045                         *(bp->created) = 0;
1046                         bp->created++;
1047                         *(bp->blocks) = 0;
1048                         bp->blocks++;
1049                         bp->num--;
1050                         bp->start++;
1051                 }
1052
1053                 return EXT_CONTINUE;
1054         }
1055
1056         tgen = EXT_GENERATION(base);
1057         count = ext3_ext_calc_credits_for_insert(base, path);
1058         fsfilt_up_truncate_sem(inode);
1059
1060         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1061         if (IS_ERR(handle)) {
1062                 fsfilt_down_truncate_sem(inode);
1063                 return PTR_ERR(handle);
1064         }
1065
1066         fsfilt_down_truncate_sem(inode);
1067         if (tgen != EXT_GENERATION(base)) {
1068                 /* the tree has changed. so path can be invalid at moment */
1069                 ext3_journal_stop(handle);
1070                 return EXT_REPEAT;
1071         }
1072
1073 #if defined(HAVE_EXT4_LDISKFS)
1074         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1075          * protected by i_data_sem as whole. so we patch it to store
1076          * generation to path and now verify the tree hasn't changed */
1077         down_write((&EXT4_I(inode)->i_data_sem));
1078
1079         /* validate extent, make sure the extent tree does not changed */
1080         if (EXT_GENERATION(base) != path[0].p_generation) {
1081                 /* cex is invalid, try again */
1082                 up_write(&EXT4_I(inode)->i_data_sem);
1083                 ext3_journal_stop(handle);
1084                 return EXT_REPEAT;
1085         }
1086 #endif
1087
1088         count = cex->ec_len;
1089         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1090         if (!pblock)
1091                 goto out;
1092         EXT_ASSERT(count <= cex->ec_len);
1093
1094         /* insert new extent */
1095         nex.ee_block = cpu_to_le32(cex->ec_block);
1096         ext3_ext_store_pblock(&nex, pblock);
1097         nex.ee_len = cpu_to_le16(count);
1098         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1099         if (err) {
1100                 /* free data blocks we just allocated */
1101                 /* not a good idea to call discard here directly,
1102                  * but otherwise we'd need to call it every free() */
1103 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1104                 ext3_mb_discard_inode_preallocations(inode);
1105 #endif
1106                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1107                                  cpu_to_le16(nex.ee_len), 0);
1108                 goto out;
1109         }
1110
1111         /*
1112          * Putting len of the actual extent we just inserted,
1113          * we are asking ext3_ext_walk_space() to continue
1114          * scaning after that block
1115          */
1116         cex->ec_len = le16_to_cpu(nex.ee_len);
1117         cex->ec_start = ext_pblock(&nex);
1118         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1119         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1120
1121 out:
1122 #if defined(HAVE_EXT4_LDISKFS)
1123         up_write((&EXT4_I(inode)->i_data_sem));
1124 #endif
1125         ext3_journal_stop(handle);
1126 map:
1127         if (err >= 0) {
1128                 /* map blocks */
1129                 if (bp->num == 0) {
1130                         CERROR("hmm. why do we find this extent?\n");
1131                         CERROR("initial space: %lu:%u\n",
1132                                 bp->start, bp->init_num);
1133                         CERROR("current extent: %u/%u/%llu %d\n",
1134                                 cex->ec_block, cex->ec_len,
1135                                 (unsigned long long)cex->ec_start,
1136                                 cex->ec_type);
1137                 }
1138                 i = 0;
1139                 if (cex->ec_block < bp->start)
1140                         i = bp->start - cex->ec_block;
1141                 if (i >= cex->ec_len)
1142                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1143                                         i, cex->ec_len);
1144                 for (; i < cex->ec_len && bp->num; i++) {
1145                         *(bp->blocks) = cex->ec_start + i;
1146                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1147                                 *(bp->created) = 0;
1148                         } else {
1149                                 *(bp->created) = 1;
1150                                 /* unmap any possible underlying metadata from
1151                                  * the block device mapping.  bug 6998. */
1152                                 ll_unmap_underlying_metadata(inode->i_sb,
1153                                                              *(bp->blocks));
1154                         }
1155                         bp->created++;
1156                         bp->blocks++;
1157                         bp->num--;
1158                         bp->start++;
1159                 }
1160         }
1161         return err;
1162 }
1163
1164 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1165                        unsigned long num, unsigned long *blocks,
1166                        int *created, int create)
1167 {
1168 #ifdef EXT3_EXT_HAS_NO_TREE
1169         struct ext3_ext_base *base = inode;
1170 #else
1171         struct ext3_extents_tree tree;
1172         struct ext3_ext_base *base = &tree;
1173 #endif
1174         struct bpointers bp;
1175         int err;
1176
1177         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1178                block, block + num - 1, (unsigned) inode->i_ino);
1179
1180 #ifndef EXT3_EXT_HAS_NO_TREE
1181         ext3_init_tree_desc(base, inode);
1182         tree.private = &bp;
1183 #endif
1184         bp.blocks = blocks;
1185         bp.created = created;
1186         bp.start = block;
1187         bp.init_num = bp.num = num;
1188         bp.create = create;
1189
1190         fsfilt_down_truncate_sem(inode);
1191         err = fsfilt_ext3_ext_walk_space(base, block, num,
1192                                          ext3_ext_new_extent_cb, &bp);
1193         ext3_ext_invalidate_cache(base);
1194         fsfilt_up_truncate_sem(inode);
1195
1196         return err;
1197 }
1198
1199 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1200                                     int pages, unsigned long *blocks,
1201                                     int *created, int create)
1202 {
1203         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1204         int rc = 0, i = 0;
1205         struct page *fp = NULL;
1206         int clen = 0;
1207
1208         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1209                 inode->i_ino, pages, (*page)->index);
1210
1211         /* pages are sorted already. so, we just have to find
1212          * contig. space and process them properly */
1213         while (i < pages) {
1214                 if (fp == NULL) {
1215                         /* start new extent */
1216                         fp = *page++;
1217                         clen = 1;
1218                         i++;
1219                         continue;
1220                 } else if (fp->index + clen == (*page)->index) {
1221                         /* continue the extent */
1222                         page++;
1223                         clen++;
1224                         i++;
1225                         continue;
1226                 }
1227
1228                 /* process found extent */
1229                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1230                                         clen * blocks_per_page, blocks,
1231                                         created, create);
1232                 if (rc)
1233                         GOTO(cleanup, rc);
1234
1235                 /* look for next extent */
1236                 fp = NULL;
1237                 blocks += blocks_per_page * clen;
1238                 created += blocks_per_page * clen;
1239         }
1240
1241         if (fp)
1242                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1243                                         clen * blocks_per_page, blocks,
1244                                         created, create);
1245 cleanup:
1246         return rc;
1247 }
1248
1249 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1250                                unsigned long *blocks, int *created, int create);
1251 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1252                                    int pages, unsigned long *blocks,
1253                                    int *created, int create)
1254 {
1255         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1256         unsigned long *b;
1257         int rc = 0, i, *cr;
1258
1259         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1260                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1261                 if (rc) {
1262                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1263                                inode->i_ino, *b, *cr, create, rc);
1264                         break;
1265                 }
1266
1267                 b += blocks_per_page;
1268                 cr += blocks_per_page;
1269         }
1270         return rc;
1271 }
1272
1273 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1274                                 int pages, unsigned long *blocks,
1275                                 int *created, int create,
1276                                 cfs_semaphore_t *optional_sem)
1277 {
1278         int rc;
1279
1280         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1281                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1282                                                      blocks, created, create);
1283                 return rc;
1284         }
1285         if (optional_sem != NULL)
1286                 cfs_down(optional_sem);
1287         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1288                                             created, create);
1289         if (optional_sem != NULL)
1290                 cfs_up(optional_sem);
1291
1292         return rc;
1293 }
1294
1295 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1296 {
1297         unsigned long block;
1298         struct buffer_head *bh;
1299         int err, blocksize, csize, boffs, osize = size;
1300
1301         /* prevent reading after eof */
1302         cfs_lock_kernel();
1303         if (i_size_read(inode) < *offs + size) {
1304                 size = i_size_read(inode) - *offs;
1305                 cfs_unlock_kernel();
1306                 if (size < 0) {
1307                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1308                                i_size_read(inode), *offs);
1309                         return -EBADR;
1310                 } else if (size == 0) {
1311                         return 0;
1312                 }
1313         } else {
1314                 cfs_unlock_kernel();
1315         }
1316
1317         blocksize = 1 << inode->i_blkbits;
1318
1319         while (size > 0) {
1320                 block = *offs >> inode->i_blkbits;
1321                 boffs = *offs & (blocksize - 1);
1322                 csize = min(blocksize - boffs, size);
1323                 bh = ext3_bread(NULL, inode, block, 0, &err);
1324                 if (!bh) {
1325                         CERROR("can't read block: %d\n", err);
1326                         return err;
1327                 }
1328
1329                 memcpy(buf, bh->b_data + boffs, csize);
1330                 brelse(bh);
1331
1332                 *offs += csize;
1333                 buf += csize;
1334                 size -= csize;
1335         }
1336         return osize;
1337 }
1338 EXPORT_SYMBOL(fsfilt_ext3_read);
1339
1340 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1341                                    int size, loff_t *offs)
1342 {
1343         int rc;
1344         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1345         if (rc > 0)
1346                 rc = 0;
1347         return rc;
1348 }
1349
1350 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1351                                 loff_t *offs, handle_t *handle)
1352 {
1353         struct buffer_head *bh = NULL;
1354         loff_t old_size = i_size_read(inode), offset = *offs;
1355         loff_t new_size = i_size_read(inode);
1356         unsigned long block;
1357         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1358
1359         while (bufsize > 0) {
1360                 if (bh != NULL)
1361                         brelse(bh);
1362
1363                 block = offset >> inode->i_blkbits;
1364                 boffs = offset & (blocksize - 1);
1365                 size = min(blocksize - boffs, bufsize);
1366                 bh = ext3_bread(handle, inode, block, 1, &err);
1367                 if (!bh) {
1368                         CERROR("can't read/create block: %d\n", err);
1369                         break;
1370                 }
1371
1372                 err = ext3_journal_get_write_access(handle, bh);
1373                 if (err) {
1374                         CERROR("journal_get_write_access() returned error %d\n",
1375                                err);
1376                         break;
1377                 }
1378                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1379                 memcpy(bh->b_data + boffs, buf, size);
1380                 err = ext3_journal_dirty_metadata(handle, bh);
1381                 if (err) {
1382                         CERROR("journal_dirty_metadata() returned error %d\n",
1383                                err);
1384                         break;
1385                 }
1386                 if (offset + size > new_size)
1387                         new_size = offset + size;
1388                 offset += size;
1389                 bufsize -= size;
1390                 buf += size;
1391         }
1392         if (bh)
1393                 brelse(bh);
1394
1395         /* correct in-core and on-disk sizes */
1396         if (new_size > i_size_read(inode)) {
1397                 cfs_lock_kernel();
1398                 if (new_size > i_size_read(inode))
1399                         i_size_write(inode, new_size);
1400                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1401                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1402                 if (i_size_read(inode) > old_size)
1403                         mark_inode_dirty(inode);
1404                 cfs_unlock_kernel();
1405         }
1406
1407         if (err == 0)
1408                 *offs = offset;
1409         return err;
1410 }
1411 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1412
1413 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1414                                     loff_t *offs, int force_sync)
1415 {
1416         struct inode *inode = file->f_dentry->d_inode;
1417         handle_t *handle;
1418         int err, block_count = 0, blocksize;
1419
1420         /* Determine how many transaction credits are needed */
1421         blocksize = 1 << inode->i_blkbits;
1422         block_count = (*offs & (blocksize - 1)) + bufsize;
1423         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1424
1425         handle = ext3_journal_start(inode,
1426                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1427         if (IS_ERR(handle)) {
1428                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1429                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1430                 return PTR_ERR(handle);
1431         }
1432
1433         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1434
1435         if (!err && force_sync)
1436                 handle->h_sync = 1; /* recovery likes this */
1437
1438         ext3_journal_stop(handle);
1439
1440         return err;
1441 }
1442
1443 static int fsfilt_ext3_setup(struct super_block *sb)
1444 {
1445 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1446      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1447         struct ext3_sb_info *sbi = EXT3_SB(sb);
1448 #if 0
1449         sbi->dx_lock = fsfilt_ext3_dx_lock;
1450         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1451 #endif
1452 #endif
1453 #ifdef S_PDIROPS
1454         CWARN("Enabling PDIROPS\n");
1455         set_opt(sbi->s_mount_opt, PDIROPS);
1456         sb->s_flags |= S_PDIROPS;
1457 #endif
1458         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1459                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1460 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1461         /* enable journaled quota support */
1462         /* kfreed in ext3_put_super() */
1463         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1464         if (!sbi->s_qf_names[USRQUOTA])
1465                 return -ENOMEM;
1466         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1467         if (!sbi->s_qf_names[GRPQUOTA]) {
1468                 kfree(sbi->s_qf_names[USRQUOTA]);
1469                 sbi->s_qf_names[USRQUOTA] = NULL;
1470                 return -ENOMEM;
1471         }
1472         sbi->s_jquota_fmt = QFMT_LUSTRE;
1473 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1474         set_opt(sbi->s_mount_opt, QUOTA);
1475 #endif
1476 #endif
1477         return 0;
1478 }
1479
1480 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1481    objects. Logs is number of logfiles to update */
1482 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1483 {
1484         if ( !fso ) {
1485                 switch(op) {
1486                 case FSFILT_OP_CREATE:
1487                                  /* directory leaf, index & indirect & EA*/
1488                         return 4 + 3 * logs;
1489                 case FSFILT_OP_UNLINK:
1490                         return 3 * logs;
1491                 }
1492         } else {
1493                 int i;
1494                 int needed = 0;
1495                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1496                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1497                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1498                 for (i = 0; i < op; i++, fso++) {
1499                         int nblocks = fso->fso_bufcnt * blockpp;
1500                         int ndindirect = min(nblocks, addrpp + 1);
1501                         int nindir = nblocks + ndindirect + 1;
1502
1503                         needed += nindir;
1504                 }
1505                 return needed + 3 * logs;
1506         }
1507
1508         return 0;
1509 }
1510
1511 #ifdef HAVE_QUOTA_SUPPORT
1512 #define DQINFO_COPY(out, in)                    \
1513 do {                                            \
1514         Q_COPY(out, in, dqi_bgrace);            \
1515         Q_COPY(out, in, dqi_igrace);            \
1516         Q_COPY(out, in, dqi_flags);             \
1517         Q_COPY(out, in, dqi_valid);             \
1518 } while (0)
1519
1520 #define DQBLK_COPY(out, in)                     \
1521 do {                                            \
1522         Q_COPY(out, in, dqb_bhardlimit);        \
1523         Q_COPY(out, in, dqb_bsoftlimit);        \
1524         Q_COPY(out, in, dqb_curspace);          \
1525         Q_COPY(out, in, dqb_ihardlimit);        \
1526         Q_COPY(out, in, dqb_isoftlimit);        \
1527         Q_COPY(out, in, dqb_curinodes);         \
1528         Q_COPY(out, in, dqb_btime);             \
1529         Q_COPY(out, in, dqb_itime);             \
1530         Q_COPY(out, in, dqb_valid);             \
1531 } while (0)
1532
1533 static int fsfilt_ext3_quotactl(struct super_block *sb,
1534                                 struct obd_quotactl *oqc)
1535 {
1536         int i, rc = 0, error = 0;
1537         const struct quotactl_ops *qcop;
1538         struct if_dqinfo *info;
1539         struct if_dqblk *dqblk;
1540         ENTRY;
1541
1542         if (!sb->s_qcop)
1543                 RETURN(-ENOSYS);
1544
1545         OBD_ALLOC_PTR(info);
1546         if (!info)
1547                 RETURN(-ENOMEM);
1548         OBD_ALLOC_PTR(dqblk);
1549         if (!dqblk) {
1550                 OBD_FREE_PTR(info);
1551                 RETURN(-ENOMEM);
1552         }
1553
1554         DQINFO_COPY(info, &oqc->qc_dqinfo);
1555         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1556
1557         qcop = sb->s_qcop;
1558         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1559                 for (i = 0; i < MAXQUOTAS; i++) {
1560                         if (!Q_TYPESET(oqc, i))
1561                                 continue;
1562
1563                         if (oqc->qc_cmd == Q_QUOTAON) {
1564                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1565
1566                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1567
1568                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1569                                                  name[i], 0);
1570                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1571                                 rc = ll_quota_off(sb, i, 0);
1572                         }
1573
1574                         if (rc == -EBUSY)
1575                                 error = rc;
1576                         else if (rc)
1577                                 GOTO(out, rc);
1578                 }
1579                 GOTO(out, rc ?: error);
1580         }
1581
1582         switch (oqc->qc_cmd) {
1583         case Q_GETOINFO:
1584         case Q_GETINFO:
1585                 if (!qcop->get_info)
1586                         GOTO(out, rc = -ENOSYS);
1587                 rc = qcop->get_info(sb, oqc->qc_type, info);
1588                 break;
1589         case Q_SETQUOTA:
1590         case Q_INITQUOTA:
1591                 if (!qcop->set_dqblk)
1592                         GOTO(out, rc = -ENOSYS);
1593                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1594                 break;
1595         case Q_GETOQUOTA:
1596         case Q_GETQUOTA:
1597                 if (!qcop->get_dqblk)
1598                         GOTO(out, rc = -ENOSYS);
1599                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1600                 if (!rc)
1601                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1602                 break;
1603         case Q_SYNC:
1604                 if (!sb->s_qcop->quota_sync)
1605                         GOTO(out, rc = -ENOSYS);
1606                 qcop->quota_sync(sb, oqc->qc_type);
1607                 break;
1608         case Q_FINVALIDATE:
1609                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1610                 for (i = 0; i < MAXQUOTAS; i++) {
1611                         struct file *fp;
1612                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1613
1614                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1615
1616                         if (!Q_TYPESET(oqc, i))
1617                                 continue;
1618
1619                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1620                         if (IS_ERR(fp)) {
1621                                 rc = PTR_ERR(fp);
1622                                 CERROR("error invalidating operational quota file"
1623                                        " %s (rc:%d)\n", name[i], rc);
1624                         } else {
1625                                 filp_close(fp, 0);
1626                         }
1627
1628                 }
1629                 break;
1630         default:
1631                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1632                 LBUG();
1633         }
1634 out:
1635         DQINFO_COPY(&oqc->qc_dqinfo, info);
1636         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1637
1638         OBD_FREE_PTR(info);
1639         OBD_FREE_PTR(dqblk);
1640
1641         if (rc)
1642                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1643                                 "failed: %d\n",
1644                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1645         RETURN(rc);
1646 }
1647
1648 struct chk_dqblk{
1649         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1650         cfs_list_t              dqb_list;        /** in list also */
1651         qid_t                   dqb_id;          /** uid/gid */
1652         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1653         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1654         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1655         qsize_t                 dqb_curspace;    /** current space */
1656         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1657         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1658         qsize_t                 dqb_curinodes;   /** current inodes */
1659         obd_time                dqb_btime;       /** block grace time */
1660         obd_time                dqb_itime;       /** inode grace time */
1661         __u32                   dqb_valid;       /** flag for above fields */
1662 };
1663
1664 static inline unsigned int chkquot_hash(qid_t id, int type)
1665                                         __attribute__((__const__));
1666
1667 static inline unsigned int chkquot_hash(qid_t id, int type)
1668 {
1669         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1670 }
1671
1672 static inline struct chk_dqblk *
1673 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1674 {
1675         cfs_hlist_node_t *node;
1676         struct chk_dqblk *cdqb;
1677
1678         cfs_hlist_for_each(node, head) {
1679                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1680                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1681                         return cdqb;
1682         }
1683
1684         return NULL;
1685 }
1686
1687 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1688 {
1689         struct chk_dqblk *cdqb;
1690
1691         OBD_ALLOC_PTR(cdqb);
1692         if (cdqb) {
1693                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1694                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1695                 cdqb->dqb_id = id;
1696                 cdqb->dqb_type = type;
1697         }
1698
1699         return cdqb;
1700 }
1701
1702 static struct chk_dqblk *
1703 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1704       cfs_list_t *list, qid_t id, int type, int first_check)
1705 {
1706         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1707         struct if_dqblk dqb;
1708         struct chk_dqblk *cdqb;
1709         int rc;
1710
1711         cdqb = find_chkquot(head, id, type);
1712         if (cdqb)
1713                 return cdqb;
1714
1715         cdqb = alloc_chkquot(id, type);
1716         if (!cdqb)
1717                 return NULL;
1718
1719         if (!first_check) {
1720                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1721                 if (rc) {
1722                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1723                                id, type, rc);
1724                 } else {
1725                         DQBLK_COPY(cdqb, &dqb);
1726                         cdqb->dqb_curspace = 0;
1727                         cdqb->dqb_curinodes = 0;
1728                 }
1729         }
1730
1731         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1732         cfs_list_add_tail(&cdqb->dqb_list, list);
1733
1734         return cdqb;
1735 }
1736
1737 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1738 {
1739         struct obd_quotactl *oqctl;
1740         int rc;
1741
1742         OBD_ALLOC_PTR(oqctl);
1743         if (!oqctl)
1744                 RETURN(-ENOMEM);
1745
1746         oqctl->qc_cmd = cmd;
1747         oqctl->qc_id = qfmt;
1748         oqctl->qc_type = type;
1749         rc = fsfilt_ext3_quotactl(sb, oqctl);
1750
1751         OBD_FREE_PTR(oqctl);
1752         return rc;
1753 }
1754
1755 static inline int read_old_dqinfo(struct super_block *sb, int type,
1756                                   struct if_dqinfo *dqinfo)
1757 {
1758         struct obd_quotactl *oqctl;
1759         int rc;
1760         ENTRY;
1761
1762         OBD_ALLOC_PTR(oqctl);
1763         if (!oqctl)
1764                 RETURN(-ENOMEM);
1765
1766         oqctl->qc_cmd = Q_GETINFO;
1767         oqctl->qc_type = type;
1768         rc = fsfilt_ext3_quotactl(sb, oqctl);
1769         if (!rc)
1770                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1771
1772         OBD_FREE_PTR(oqctl);
1773         RETURN(rc);
1774 }
1775
1776 #ifndef HAVE_EXT4_LDISKFS
1777 static inline struct ext3_group_desc *
1778 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1779 {
1780         unsigned long desc_block, desc;
1781         struct ext3_group_desc *gdp;
1782
1783         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1784         desc = group % EXT3_DESC_PER_BLOCK(sb);
1785         gdp = (struct ext3_group_desc *)
1786               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1787
1788         return gdp + desc;
1789 }
1790
1791 static inline struct buffer_head *
1792 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1793 {
1794         struct ext3_group_desc *desc;
1795         struct buffer_head *bh;
1796
1797         desc = get_group_desc(sb, group, NULL);
1798         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1799         return bh;
1800 }
1801
1802 static __u32 ext3_itable_unused_count(struct super_block *sb,
1803                                struct ext3_group_desc *bg) {
1804        return le16_to_cpu(bg->bg_itable_unused);
1805 }
1806 #else
1807 #define get_group_desc ext3_get_group_desc
1808 #endif
1809
1810 struct qchk_ctxt {
1811         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1812         cfs_list_t              qckt_list;                 /* quotacheck list */
1813         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1814         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1815 };
1816
1817 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1818                            struct obd_quotactl *oqc)
1819 {
1820         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1821         loff_t size = 0;
1822         qid_t qid[MAXQUOTAS];
1823         int cnt, i, rc = 0;
1824
1825         if (!inode)
1826                 return 0;
1827
1828         qid[USRQUOTA] = inode->i_uid;
1829         qid[GRPQUOTA] = inode->i_gid;
1830
1831         if (S_ISDIR(inode->i_mode) ||
1832             S_ISREG(inode->i_mode) ||
1833             S_ISLNK(inode->i_mode))
1834                 size = inode_get_bytes(inode);
1835
1836         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1837                 if (!Q_TYPESET(oqc, cnt))
1838                         continue;
1839
1840                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1841                                 &qctxt->qckt_list, qid[cnt], cnt,
1842                                 qctxt->qckt_first_check[cnt]);
1843                 if (!cdqb[cnt]) {
1844                         rc = -ENOMEM;
1845                         break;
1846                 }
1847
1848                 cdqb[cnt]->dqb_curspace += size;
1849                 cdqb[cnt]->dqb_curinodes++;
1850         }
1851
1852         if (rc) {
1853                 for (i = 0; i < cnt; i++) {
1854                         if (!Q_TYPESET(oqc, i))
1855                                 continue;
1856                         LASSERT(cdqb[i]);
1857                         cdqb[i]->dqb_curspace -= size;
1858                         cdqb[i]->dqb_curinodes--;
1859                 }
1860         }
1861
1862         return rc;
1863 }
1864
1865 /* write dqinfo struct in a new quota file */
1866 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1867 {
1868         struct v2_disk_dqinfo dqinfo;
1869         __u32 blocks = V2_DQTREEOFF + 1;
1870         loff_t offset = V2_DQINFOOFF;
1871
1872         if (info) {
1873                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1874                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1875                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1876                                                ~DQF_INFO_DIRTY);
1877         } else {
1878                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1879                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1880                 dqinfo.dqi_flags = 0;
1881         }
1882
1883         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1884         dqinfo.dqi_free_blk = 0;
1885         dqinfo.dqi_free_entry = 0;
1886
1887         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1888 }
1889
1890 static int v3_write_dqheader(struct file *f, int type)
1891 {
1892         static const __u32 quota_magics[] = V2_INITQMAGICS;
1893         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1894         struct v2_disk_dqheader dqhead;
1895         loff_t offset = 0;
1896
1897         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1898         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1899
1900         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1901         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1902
1903         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1904 }
1905
1906 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1907                                   struct obd_quotactl *oqc)
1908 {
1909         int i, rc = 0;
1910         ENTRY;
1911
1912         for (i = 0; i < MAXQUOTAS; i++) {
1913                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1914                                          NULL : &qctxt->qckt_dqinfo[i];
1915                 struct file *file;
1916                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1917
1918                 if (!Q_TYPESET(oqc, i))
1919                         continue;
1920
1921                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1922
1923                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1924                 if (IS_ERR(file)) {
1925                         rc = PTR_ERR(file);
1926                         CERROR("can't create %s file: rc = %d\n",
1927                                name[i], rc);
1928                         GOTO(out, rc);
1929                 }
1930
1931                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1932                         CERROR("file %s is not regular", name[i]);
1933                         filp_close(file, 0);
1934                         GOTO(out, rc = -EINVAL);
1935                 }
1936
1937                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1938
1939                 rc = v3_write_dqheader(file, i);
1940                 if (rc) {
1941                         filp_close(file, 0);
1942                         GOTO(out, rc);
1943                 }
1944
1945                 rc = v3_write_dqinfo(file, i, info);
1946                 filp_close(file, 0);
1947                 if (rc)
1948                         GOTO(out, rc);
1949         }
1950
1951 out:
1952         RETURN(rc);
1953 }
1954
1955
1956 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1957                           struct chk_dqblk *cdqb)
1958 {
1959         struct obd_quotactl *oqc;
1960         long now;
1961         int rc;
1962         ENTRY;
1963
1964         OBD_ALLOC_PTR(oqc);
1965         if (!oqc)
1966                 RETURN(-ENOMEM);
1967
1968         now = cfs_time_current_sec();
1969
1970         if (cdqb->dqb_bsoftlimit &&
1971             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1972             !cdqb->dqb_btime)
1973                 cdqb->dqb_btime =
1974                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1975
1976         if (cdqb->dqb_isoftlimit &&
1977             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1978             !cdqb->dqb_itime)
1979                 cdqb->dqb_itime =
1980                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1981
1982         cdqb->dqb_valid = QIF_ALL;
1983
1984         oqc->qc_cmd = Q_SETQUOTA;
1985         oqc->qc_type = cdqb->dqb_type;
1986         oqc->qc_id = cdqb->dqb_id;
1987         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1988
1989         rc = fsfilt_ext3_quotactl(sb, oqc);
1990         OBD_FREE_PTR(oqc);
1991         RETURN(rc);
1992 }
1993
1994 static int prune_chkquots(struct super_block *sb,
1995                           struct qchk_ctxt *qctxt, int error)
1996 {
1997         struct chk_dqblk *cdqb, *tmp;
1998         int rc;
1999
2000         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
2001                 if (!error) {
2002                         rc = commit_chkquot(sb, qctxt, cdqb);
2003                         if (rc)
2004                                 error = rc;
2005                 }
2006                 cfs_hlist_del_init(&cdqb->dqb_hash);
2007                 cfs_list_del(&cdqb->dqb_list);
2008                 OBD_FREE_PTR(cdqb);
2009         }
2010
2011         return error;
2012 }
2013
2014 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2015 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2016 #endif
2017
2018 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2019                                   struct obd_quotactl *oqc)
2020 {
2021         struct ext3_sb_info *sbi = EXT3_SB(sb);
2022         int i, group, uninit_feat = 0;
2023         struct qchk_ctxt *qctxt;
2024         struct buffer_head *bitmap_bh = NULL;
2025         unsigned long ino, inode_inuse;
2026         struct inode *inode;
2027         int rc = 0;
2028         ENTRY;
2029
2030         /* turn on quota and read dqinfo if existed */
2031         OBD_ALLOC_PTR(qctxt);
2032         if (!qctxt) {
2033                 oqc->qc_stat = -ENOMEM;
2034                 RETURN(-ENOMEM);
2035         }
2036
2037         for (i = 0; i < NR_DQHASH; i++)
2038                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2039         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2040
2041         for (i = 0; i < MAXQUOTAS; i++) {
2042                 if (!Q_TYPESET(oqc, i))
2043                         continue;
2044
2045                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2046                 if (!rc || rc == -EBUSY) {
2047                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2048                         if (rc)
2049                                 GOTO(out, rc);
2050                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2051                         qctxt->qckt_first_check[i] = 1;
2052                 } else if (rc) {
2053                         GOTO(out, rc);
2054                 }
2055         }
2056         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2057                 /* This filesystem supports the uninit group feature */
2058                 uninit_feat = 1;
2059
2060         /* number of inodes that have been allocated */
2061         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2062                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2063
2064         /* check quota and update in hash */
2065         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2066              group++) {
2067                 unsigned long used_count = sbi->s_inodes_per_group;
2068
2069                 if (uninit_feat) {
2070                         struct ext3_group_desc *desc;
2071                         desc = get_group_desc(sb, group, NULL);
2072                         if (!desc)
2073                                 GOTO(out, -EIO);
2074
2075                         /* we don't really need to take the group lock here,
2076                          * but it may be useful if one day we support online
2077                          * quotacheck */
2078 #ifdef HAVE_EXT4_LDISKFS
2079                         ext4_lock_group(sb, group);
2080 #else
2081                         spin_lock(sb_bgl_lock(sbi, group));
2082 #endif
2083                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2084                                 /* no inode in use in this group, just skip it */
2085 #ifdef HAVE_EXT4_LDISKFS
2086                                 ext3_unlock_group(sb, group);
2087 #else
2088                                 spin_unlock(sb_bgl_lock(sbi, group));
2089 #endif
2090                                 continue;
2091                         }
2092
2093                         used_count -= ext3_itable_unused_count(sb, desc);
2094 #ifdef HAVE_EXT4_LDISKFS
2095                         ext3_unlock_group(sb, group);
2096 #else
2097                         spin_unlock(sb_bgl_lock(sbi, group));
2098 #endif
2099                 }
2100
2101                 ino = group * sbi->s_inodes_per_group + 1;
2102                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2103                 if (!bitmap_bh) {
2104                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2105                                sb->s_id, group);
2106                         GOTO(out, -EIO);
2107                 }
2108
2109                 i = 0;
2110                 while (i < used_count &&
2111                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2112                                                used_count, i)) < used_count) {
2113                         inode_inuse--;
2114                         i++;
2115                         ino = i + group * sbi->s_inodes_per_group;
2116                         if (ino < sbi->s_first_ino)
2117                                 continue;
2118 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2119                         inode = ext3_iget(sb, ino);
2120 #else
2121                         inode = iget(sb, ino);
2122 #endif
2123                         if (!inode || IS_ERR(inode))
2124                                 continue;
2125
2126                         rc = add_inode_quota(inode, qctxt, oqc);
2127                         iput(inode);
2128                         if (rc) {
2129                                 brelse(bitmap_bh);
2130                                 GOTO(out, rc);
2131                         }
2132                 }
2133                 brelse(bitmap_bh);
2134         }
2135
2136         /* read old quota limits from old quota file. (only for the user
2137          * has limits but hasn't file) */
2138 #ifdef HAVE_QUOTA_SUPPORT
2139         for (i = 0; i < MAXQUOTAS; i++) {
2140                 cfs_list_t id_list;
2141                 struct dquot_id *dqid, *tmp;
2142
2143                 if (!Q_TYPESET(oqc, i))
2144                         continue;
2145
2146                 if (qctxt->qckt_first_check[i])
2147                         continue;
2148
2149
2150                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2151                 CFS_INIT_LIST_HEAD(&id_list);
2152 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2153                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2154 #else
2155                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2156 #endif
2157                 if (rc)
2158                         CERROR("read old limits failed. (rc:%d)\n", rc);
2159
2160                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2161                         cfs_list_del_init(&dqid->di_link);
2162
2163                         if (!rc)
2164                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2165                                       dqid->di_id, i,
2166                                       qctxt->qckt_first_check[i]);
2167                         OBD_FREE_PTR(dqid);
2168                 }
2169         }
2170 #endif
2171         /* turn off quota cause we are to dump chk_dqblk to files */
2172         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2173
2174         rc = create_new_quota_files(qctxt, oqc);
2175         if (rc)
2176                 GOTO(out, rc);
2177
2178         /* we use vfs functions to set dqblk, so turn quota on */
2179         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2180 out:
2181         /* dump and free chk_dqblk */
2182         rc = prune_chkquots(sb, qctxt, rc);
2183         OBD_FREE_PTR(qctxt);
2184
2185         /* turn off quota, `lfs quotacheck` will turn on when all
2186          * nodes quotacheck finish. */
2187         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2188
2189         oqc->qc_stat = rc;
2190         if (rc)
2191                 CERROR("quotacheck failed: rc = %d\n", rc);
2192
2193         RETURN(rc);
2194 }
2195
2196 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2197                                  int cmd)
2198 {
2199         int rc = 0;
2200         ENTRY;
2201
2202         if (lqi->qi_files[type] == NULL) {
2203                 CERROR("operate qinfo before it's enabled!\n");
2204                 RETURN(-ESRCH);
2205         }
2206
2207         switch (cmd) {
2208         case QFILE_CHK:
2209                 rc = lustre_check_quota_file(lqi, type);
2210                 break;
2211         case QFILE_RD_INFO:
2212                 rc = lustre_read_quota_info(lqi, type);
2213                 break;
2214         case QFILE_WR_INFO:
2215                 rc = lustre_write_quota_info(lqi, type);
2216                 break;
2217         case QFILE_INIT_INFO:
2218                 rc = lustre_init_quota_info(lqi, type);
2219                 break;
2220         case QFILE_CONVERT:
2221                 rc = -ENOTSUPP;
2222                 CERROR("quota CONVERT command is not supported\n");
2223                 break;
2224         default:
2225                 rc = -ENOTSUPP;
2226                 CERROR("Unsupported admin quota file cmd %d\n"
2227                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2228                        cmd);
2229                 break;
2230         }
2231         RETURN(rc);
2232 }
2233
2234 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2235                             cfs_list_t *list)
2236 {
2237         return lustre_get_qids(file, inode, type, list);
2238 }
2239
2240 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2241 {
2242         int rc = 0;
2243         ENTRY;
2244
2245         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2246                 CERROR("operate dquot before it's enabled!\n");
2247                 RETURN(-ESRCH);
2248         }
2249
2250         switch (cmd) {
2251         case QFILE_RD_DQUOT:
2252                 rc = lustre_read_dquot(dquot);
2253                 break;
2254         case QFILE_WR_DQUOT:
2255                 if (dquot->dq_dqb.dqb_ihardlimit ||
2256                     dquot->dq_dqb.dqb_isoftlimit ||
2257                     dquot->dq_dqb.dqb_bhardlimit ||
2258                     dquot->dq_dqb.dqb_bsoftlimit)
2259                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2260                 else
2261                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2262
2263                 rc = lustre_commit_dquot(dquot);
2264                 if (rc >= 0)
2265                         rc = 0;
2266                 break;
2267         default:
2268                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2269                 LBUG();
2270                 break;
2271         }
2272         RETURN(rc);
2273 }
2274
2275 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2276                                 struct inode *inode, int frags)
2277 {
2278 #ifdef EXT3_EXT_HAS_NO_TREE
2279         struct ext3_ext_base *base = inode;
2280 #else
2281         struct ext3_extents_tree tree;
2282         struct ext3_ext_base *base = &tree;
2283
2284         ext3_init_tree_desc(base, inode);
2285 #endif
2286         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2287          * metablocks at maxium b=16542 */
2288         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2289         return 0;
2290 }
2291
2292 #endif
2293
2294 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2295 {
2296         return (EXT3_SB(sb)->journal_bdev);
2297 }
2298 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2299
2300 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2301                           char *buf, int count, loff_t pos)
2302 {
2303         loff_t p = pos;
2304         int rc;
2305
2306         if (!f && !inode) {
2307                 CERROR("lustre_read_quota failed for no quota file!\n");
2308                 libcfs_debug_dumpstack(NULL);
2309                 return -EINVAL;
2310         }
2311
2312         /* Support for both adm and op quota files must be provided */
2313         if (f) {
2314                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2315                 rc = rc < 0 ? rc : p - pos;
2316         } else {
2317                 struct super_block *sb = inode->i_sb;
2318                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2319         }
2320         return rc;
2321 }
2322
2323 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2324 {
2325         loff_t p = pos;
2326         int rc;
2327
2328         /* Only adm quota files are supported, op updates are handled by vfs */
2329         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2330         rc = rc < 0 ? rc : p - pos;
2331
2332         return rc;
2333 }
2334
2335 void *lustre_quota_journal_start(struct inode *inode, int delete)
2336 {
2337         handle_t *handle;
2338         unsigned block_count;
2339
2340         if (delete) {
2341                 /* each indirect block (+4) may become free, attaching to the
2342                  * header list of free blocks (+1); the data block (+1) may
2343                  * become a free block (+0) or a block with free dqentries (+0) */
2344                 block_count = (4 + 1) + 1;
2345                 handle = ext3_journal_start(inode,
2346                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2347         } else {
2348                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2349                  * freeblk reusage with a header update (+1); dqentry is either reused
2350                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2351                  * a new block allocation (+1) with a header update (+1)              */
2352                 block_count = (4 + 1) + 3;
2353                 handle = ext3_journal_start(inode,
2354                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2355
2356         }
2357
2358         return handle;
2359 }
2360
2361 void lustre_quota_journal_stop(void *handle)
2362 {
2363         ext3_journal_stop((handle_t *)handle);
2364 }
2365
2366 static int ll_decode_fh_accept(void *context, struct dentry *de)
2367 {
2368         return 1;
2369 }
2370
2371 #ifdef HAVE_EXPORTFS_DECODE_FH
2372 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2373          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2374                             acceptable, context)
2375 #else
2376 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2377          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2378                                      type, acceptable, context)
2379 # define FILEID_INO32_GEN 1
2380 extern struct export_operations export_op_default;
2381 #endif
2382
2383 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2384                                       struct fsfilt_fid *fid, int ignore_gen)
2385 {
2386         struct inode  *inode;
2387         struct dentry *result;
2388
2389         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2390                                        ll_decode_fh_accept, NULL);
2391         if (IS_ERR(result)) {
2392                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2393                        fid->ino, fid->gen, PTR_ERR(result));
2394                 return result;
2395         }
2396
2397         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2398                fid->ino, fid->gen);
2399         inode = result->d_inode;
2400         if (inode == NULL)
2401                 goto err_out;
2402
2403         if (inode->i_nlink == 0 &&
2404             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2405                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2406                               " ctime -- this may indicate disk "
2407                               "corruption (inode: %lu, link: %lu, "
2408                               "count: %d)\n", inode->i_ino,
2409                               (unsigned long)inode->i_nlink,
2410                               atomic_read(&inode->i_count));
2411                 goto err_out;
2412         }
2413         if (fid->gen && inode->i_generation != fid->gen) {
2414                 /* we didn't find the right inode.. */
2415                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2416                        "count: %d, generation %u/%u\n",
2417                        inode->i_ino, (unsigned long)inode->i_nlink,
2418                        atomic_read(&inode->i_count), inode->i_generation,
2419                        fid->gen);
2420                 goto err_out;
2421         }
2422
2423         return result;
2424 err_out:
2425         l_dput(result);
2426         return ERR_PTR(-ENOENT);
2427 }
2428
2429 static struct fsfilt_operations fsfilt_ext3_ops = {
2430         .fs_type                = "ext3",
2431         .fs_owner               = THIS_MODULE,
2432         .fs_getlabel            = fsfilt_ext3_get_label,
2433         .fs_setlabel            = fsfilt_ext3_set_label,
2434         .fs_uuid                = fsfilt_ext3_uuid,
2435         .fs_start               = fsfilt_ext3_start,
2436         .fs_brw_start           = fsfilt_ext3_brw_start,
2437         .fs_extend              = fsfilt_ext3_extend,
2438         .fs_commit              = fsfilt_ext3_commit,
2439         .fs_commit_async        = fsfilt_ext3_commit_async,
2440         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2441         .fs_setattr             = fsfilt_ext3_setattr,
2442         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2443         .fs_set_md              = fsfilt_ext3_set_md,
2444         .fs_get_md              = fsfilt_ext3_get_md,
2445         .fs_readpage            = fsfilt_ext3_readpage,
2446         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2447         .fs_statfs              = fsfilt_ext3_statfs,
2448         .fs_sync                = fsfilt_ext3_sync,
2449         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2450         .fs_write_record        = fsfilt_ext3_write_record,
2451         .fs_read_record         = fsfilt_ext3_read_record,
2452         .fs_setup               = fsfilt_ext3_setup,
2453         .fs_send_bio            = fsfilt_ext3_send_bio,
2454         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2455 #ifdef HAVE_DISK_INODE_VERSION
2456         .fs_get_version         = fsfilt_ext3_get_version,
2457         .fs_set_version         = fsfilt_ext3_set_version,
2458 #endif
2459 #ifdef HAVE_QUOTA_SUPPORT
2460         .fs_quotactl            = fsfilt_ext3_quotactl,
2461         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2462         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2463         .fs_qids                = fsfilt_ext3_qids,
2464         .fs_dquot               = fsfilt_ext3_dquot,
2465         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2466 #endif
2467         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2468         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2469 };
2470
2471 static int __init fsfilt_ext3_init(void)
2472 {
2473         int rc;
2474
2475         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2476                                          sizeof(struct fsfilt_cb_data), 0, 0);
2477         if (!fcb_cache) {
2478                 CERROR("error allocating fsfilt journal callback cache\n");
2479                 GOTO(out, rc = -ENOMEM);
2480         }
2481
2482         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2483
2484         if (rc) {
2485                 int err = cfs_mem_cache_destroy(fcb_cache);
2486                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2487         }
2488 out:
2489         return rc;
2490 }
2491
2492 static void __exit fsfilt_ext3_exit(void)
2493 {
2494         int rc;
2495
2496         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2497         rc = cfs_mem_cache_destroy(fcb_cache);
2498         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2499 }
2500
2501 module_init(fsfilt_ext3_init);
2502 module_exit(fsfilt_ext3_exit);
2503
2504 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2505 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2506 MODULE_LICENSE("GPL");