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 Whamcloud, Inc.
  33  *
  34  */
  35 /*
  36  * Copyright (c) 2011 Whamcloud, Inc.
  37  */
  38 /*
  39  * This file is part of Lustre, http://www.lustre.org/
  40  * Lustre is a trademark of Sun Microsystems, Inc.
  41  *
  42  * lustre/lvfs/fsfilt_ext3.c
  43  *
  44  * Author: Andreas Dilger <adilger@clusterfs.com>
  45  */
  46
  47 #define DEBUG_SUBSYSTEM S_FILTER
  48
  49 #include <linux/init.h>
  50 #include <linux/module.h>
  51 #include <linux/fs.h>
  52 #include <linux/slab.h>
  53 #include <linux/pagemap.h>
  54 #include <linux/quotaops.h>
  55 #ifdef HAVE_LINUX_EXPORTFS_H
  56 #include <linux/exportfs.h>
  57 #endif
  58 #ifdef HAVE_EXT4_LDISKFS
  59 #include <ext4/ext4.h>
  60 #include <ext4/ext4_jbd2.h>
  61 #else
  62 #include <linux/jbd.h>
  63 #include <linux/ext3_fs.h>
  64 #include <linux/ext3_jbd.h>
  65 #endif
  66 #include <linux/version.h>
  67 #include <linux/bitops.h>
  68 #include <linux/quota.h>
  69 #ifdef HAVE_QUOTAIO_H
  70 # include <linux/quotaio_v2.h>
  71 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  72 # include <quota/quotaio_v2.h>
  73 # include <quota/quota_tree.h>
  74 # define V2_DQTREEOFF    QT_TREEOFF
  75 #elif defined(HAVE_FS_QUOTAIO_V1_H)
  76 # include <quotaio_v2.h>
  77 # include <quota_tree.h>
  78 # define V2_DQTREEOFF    QT_TREEOFF
  79 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  80 #endif
  81
  82 #ifdef QFMT_VFS_V1
  83 #define QFMT_LUSTRE QFMT_VFS_V1
  84 #else
  85 #define QFMT_LUSTRE QFMT_VFS_V0
  86 #endif
  87
  88 #if defined(HAVE_EXT3_XATTR_H)
  89 # include <ext3/xattr.h>
  90 #elif !defined(EXT3_XATTR_INDEX_TRUSTED)
  91 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  92 /* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
  93 # define EXT3_XATTR_INDEX_TRUSTED        4
  94 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  95 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  96 #endif
  97
  98 #include <libcfs/libcfs.h>
  99 #include <lustre_fsfilt.h>
 100 #include <obd.h>
 101 #include <lustre_quota.h>
 102 #include <linux/lustre_compat25.h>
 103 #include <linux/lprocfs_status.h>
 104
 105 #ifdef HAVE_EXT4_LDISKFS
 106 #include <ext4/ext4_extents.h>
 107 #else
 108 #include <linux/ext3_extents.h>
 109 #endif
 110
 111 #include "lustre_quota_fmt.h"
 112
 113 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
 114 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 115 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 116 #else
 117 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 118 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 119 #endif
 120
 121 /* for kernels 2.6.18 and later */
 122 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 123
 124 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 125                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 126
 127 #define ext3_mb_discard_inode_preallocations(inode) \
 128                  ext3_discard_preallocations(inode)
 129
 130 #ifdef HAVE_EXT4_LDISKFS
 131 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 132 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 133 #else
 134 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 135 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 136 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 137 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 138 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 139 #endif
 140
 141 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 142 # define journal_callback ext4_journal_cb_entry
 143 # define fsfilt_journal_callback_set(handle, func, jcb) \
 144          ext4_journal_callback_add(handle, func, jcb)
 145 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
 146 # define fsfilt_journal_callback_set(handle, func, jcb) \
 147          jbd2_journal_callback_set(handle, func, jcb)
 148 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
 149 # define fsfilt_journal_callback_set(handle, func, jcb) \
 150          journal_callback_set(handle, func, jcb)
 151 #else
 152 # error missing journal commit callback
 153 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
 154
 155 static cfs_mem_cache_t *fcb_cache;
 156
 157 struct fsfilt_cb_data {
 158         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 159         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 160         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 161         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 162         void *cb_data;                  /* MDS/OST completion function data */
 163 };
 164
 165 #ifndef ext3_find_next_bit
 166 #define ext3_find_next_bit           ext2_find_next_bit
 167 #endif
 168
 169 #ifndef ext2_find_next_bit
 170 #ifdef __LITTLE_ENDIAN
 171 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 172 #else
 173 error "Need implementation of find_next_bit on big-endian systems"
 174 #endif  /* __LITTLE_ENDIAN */
 175 #endif  /* !ext2_find_next_le_bit */
 176
 177 static char *fsfilt_ext3_get_label(struct super_block *sb)
 178 {
 179         return EXT3_SB(sb)->s_es->s_volume_name;
 180 }
 181
 182 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 183 {
 184         /* see e.g. fsfilt_ext3_write_record() */
 185         journal_t *journal;
 186         handle_t *handle;
 187         int err;
 188
 189         journal = EXT3_SB(sb)->s_journal;
 190         handle = ext3_journal_start_sb(sb, 1);
 191         if (IS_ERR(handle)) {
 192                 CERROR("can't start transaction\n");
 193                 return(PTR_ERR(handle));
 194         }
 195
 196         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 197         if (err)
 198                 goto out;
 199
 200         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 201                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 202
 203         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 204
 205 out:
 206         ext3_journal_stop(handle);
 207
 208         return(err);
 209 }
 210
 211 static char *fsfilt_ext3_uuid(struct super_block *sb)
 212 {
 213         return EXT3_SB(sb)->s_es->s_uuid;
 214 }
 215
 216 #ifdef HAVE_DISK_INODE_VERSION
 217
 218 static __u64 get_i_version(struct inode *inode)
 219 {
 220         return EXT3_I(inode)->i_fs_version;
 221 }
 222
 223 static void set_i_version(struct inode *inode, __u64 new_version)
 224 {
 225         (EXT3_I(inode))->i_fs_version = new_version;
 226 }
 227
 228 /*
 229  * Get the 64-bit version for an inode.
 230  */
 231 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 232 {
 233         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 234                get_i_version(inode), inode->i_ino);
 235         return get_i_version(inode);
 236 }
 237
 238 /*
 239  * Set the 64-bit version and return the old version.
 240  */
 241 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 242 {
 243         __u64 old_version = get_i_version(inode);
 244
 245         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 246                new_version, old_version, inode->i_ino);
 247         set_i_version(inode, new_version);
 248         /* version is set after all inode operations are finished, so we should
 249          * mark it dirty here */
 250         inode->i_sb->s_op->dirty_inode(inode);
 251         return old_version;
 252 }
 253
 254 #endif
 255
 256 /*
 257  * We don't currently need any additional blocks for rmdir and
 258  * unlink transactions because we are storing the OST oa_id inside
 259  * the inode (which we will be changing anyways as part of this
 260  * transaction).
 261  */
 262 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 263                                int logs)
 264 {
 265         /* For updates to the last received file */
 266         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 267         journal_t *journal;
 268         void *handle;
 269
 270         if (current->journal_info) {
 271                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 272                        current->journal_info);
 273                 goto journal_start;
 274         }
 275
 276         switch(op) {
 277         case FSFILT_OP_RMDIR:
 278         case FSFILT_OP_UNLINK:
 279                 /* delete one file + create/update logs for each stripe */
 280                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 281                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 282                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 283                 break;
 284         case FSFILT_OP_RENAME:
 285                 /* modify additional directory */
 286                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 287                 /* no break */
 288         case FSFILT_OP_SYMLINK:
 289                 /* additional block + block bitmap + GDT for long symlink */
 290                 nblocks += 3;
 291                 /* no break */
 292         case FSFILT_OP_CREATE: {
 293 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 294                 static int warned;
 295                 if (!warned) {
 296                         if (!test_opt(inode->i_sb, EXTENTS)) {
 297                                 warned = 1;
 298                         } else if (((EXT3_I(inode)->i_flags &
 299                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 300                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 301                                 CWARN("extent-mapped directory found with "
 302                                       "ext3-based ldiskfs - contact "
 303                                       "http://bugs.whamcloud.com/\n");
 304                                 warned = 1;
 305                         }
 306                 }
 307 #endif
 308                 /* no break */
 309         }
 310         case FSFILT_OP_MKDIR:
 311         case FSFILT_OP_MKNOD:
 312                 /* modify one inode + block bitmap + GDT */
 313                 nblocks += 3;
 314                 /* no break */
 315         case FSFILT_OP_LINK:
 316                 /* modify parent directory */
 317                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 318                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 319                 /* create/update logs for each stripe */
 320                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 321                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 322                 break;
 323         case FSFILT_OP_SETATTR:
 324                 /* Setattr on inode */
 325                 nblocks += 1;
 326                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 327                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 328                 /* quota chown log for each stripe */
 329                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 330                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 331                 break;
 332         case FSFILT_OP_CANCEL_UNLINK:
 333                 /* blocks for log header bitmap update OR
 334                  * blocks for catalog header bitmap update + unlink of logs */
 335                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 336                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 337                 break;
 338         default: CERROR("unknown transaction start op %d\n", op);
 339                 LBUG();
 340         }
 341
 342         LASSERT(current->journal_info == desc_private);
 343         journal = EXT3_SB(inode->i_sb)->s_journal;
 344         if (nblocks > journal->j_max_transaction_buffers) {
 345                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 346                        nblocks, op, logs, journal->j_max_transaction_buffers);
 347                 nblocks = journal->j_max_transaction_buffers;
 348         }
 349
 350  journal_start:
 351         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 352         handle = ext3_journal_start(inode, nblocks);
 353
 354         if (!IS_ERR(handle))
 355                 LASSERT(current->journal_info == handle);
 356         else
 357                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 358                        op, nblocks, PTR_ERR(handle));
 359         return handle;
 360 }
 361
 362 /*
 363  * Calculate the number of buffer credits needed to write multiple pages in
 364  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 365  * doesn't have a nice API for calculating this sort of thing in advance.
 366  *
 367  * See comment above ext3_writepage_trans_blocks for details.  We assume
 368  * no data journaling is being done, but it does allow for all of the pages
 369  * being non-contiguous.  If we are guaranteed contiguous pages we could
 370  * reduce the number of (d)indirect blocks a lot.
 371  *
 372  * With N blocks per page and P pages, for each inode we have at most:
 373  * N*P indirect
 374  * min(N*P, blocksize/4 + 1) dindirect blocks
 375  * niocount tindirect
 376  *
 377  * For the entire filesystem, we have at most:
 378  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 379  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 380  * objcount inode blocks
 381  * 1 superblock
 382  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 383  *
 384  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 385  */
 386 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 387                                       int niocount, struct niobuf_local *nb)
 388 {
 389         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 390         __u64 next_indir;
 391         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 392         int nbitmaps = 0, ngdblocks;
 393         int needed = objcount + 1; /* inodes + superblock */
 394         int i, j;
 395
 396         for (i = 0, j = 0; i < objcount; i++, fso++) {
 397                 /* two or more dindirect blocks in case we cross boundary */
 398                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 399                                     nb[j].offset) >>
 400                                    sb->s_blocksize_bits) /
 401                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 402                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 403
 404                 /* leaf, indirect, tindirect blocks for first block */
 405                 nbitmaps += blockpp + 2;
 406
 407                 j += fso->fso_bufcnt;
 408         }
 409
 410         next_indir = nb[0].offset +
 411                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 412         for (i = 1; i < niocount; i++) {
 413                 if (nb[i].offset >= next_indir) {
 414                         nbitmaps++;     /* additional indirect */
 415                         next_indir = nb[i].offset +
 416                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 417                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 418                         nbitmaps++;     /* additional indirect */
 419                 }
 420                 nbitmaps += blockpp;    /* each leaf in different group? */
 421         }
 422
 423         ngdblocks = nbitmaps;
 424         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 425                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 426         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 427                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 428
 429         needed += nbitmaps + ngdblocks;
 430
 431         /* last_rcvd update */
 432         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 433
 434 #if defined(CONFIG_QUOTA)
 435         /* We assume that there will be 1 bit set in s_dquot.flags for each
 436          * quota file that is active.  This is at least true for now.
 437          */
 438         needed += hweight32(ll_sb_any_quota_active(sb)) *
 439                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 440 #endif
 441
 442         return needed;
 443 }
 444
 445 /* We have to start a huge journal transaction here to hold all of the
 446  * metadata for the pages being written here.  This is necessitated by
 447  * the fact that we do lots of prepare_write operations before we do
 448  * any of the matching commit_write operations, so even if we split
 449  * up to use "smaller" transactions none of them could complete until
 450  * all of them were opened.  By having a single journal transaction,
 451  * we eliminate duplicate reservations for common blocks like the
 452  * superblock and group descriptors or bitmaps.
 453  *
 454  * We will start the transaction here, but each prepare_write will
 455  * add a refcount to the transaction, and each commit_write will
 456  * remove a refcount.  The transaction will be closed when all of
 457  * the pages have been written.
 458  */
 459 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 460                                    int niocount, struct niobuf_local *nb,
 461                                    void *desc_private, int logs)
 462 {
 463         journal_t *journal;
 464         handle_t *handle;
 465         int needed;
 466         ENTRY;
 467
 468         LASSERT(current->journal_info == desc_private);
 469         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 470         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 471
 472         /* The number of blocks we could _possibly_ dirty can very large.
 473          * We reduce our request if it is absurd (and we couldn't get that
 474          * many credits for a single handle anyways).
 475          *
 476          * At some point we have to limit the size of I/Os sent at one time,
 477          * increase the size of the journal, or we have to calculate the
 478          * actual journal requirements more carefully by checking all of
 479          * the blocks instead of being maximally pessimistic.  It remains to
 480          * be seen if this is a real problem or not.
 481          */
 482         if (needed > journal->j_max_transaction_buffers) {
 483                 CERROR("want too many journal credits (%d) using %d instead\n",
 484                        needed, journal->j_max_transaction_buffers);
 485                 needed = journal->j_max_transaction_buffers;
 486         }
 487
 488         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 489         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 490         if (IS_ERR(handle)) {
 491                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 492                        PTR_ERR(handle));
 493         } else {
 494                 LASSERT(handle->h_buffer_credits >= needed);
 495                 LASSERT(current->journal_info == handle);
 496         }
 497
 498         RETURN(handle);
 499 }
 500
 501 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 502 {
 503        handle_t *handle = h;
 504
 505        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 506        if (nblocks == 0) {
 507                handle->h_buffer_credits = 0;
 508                CWARN("setting credits of handle %p to zero by request\n", h);
 509        }
 510
 511        if (handle->h_buffer_credits > nblocks)
 512                 return 0;
 513        if (ext3_journal_extend(handle, nblocks) == 0)
 514                 return 0;
 515
 516        ext3_mark_inode_dirty(handle, inode);
 517        return ext3_journal_restart(handle, nblocks);
 518 }
 519
 520 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 521 {
 522         int rc;
 523         handle_t *handle = h;
 524
 525         LASSERT(current->journal_info == handle);
 526         if (force_sync)
 527                 handle->h_sync = 1; /* recovery likes this */
 528
 529         rc = ext3_journal_stop(handle);
 530
 531         return rc;
 532 }
 533
 534 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 535                                     void **wait_handle)
 536 {
 537         unsigned long tid;
 538         transaction_t *transaction;
 539         handle_t *handle = h;
 540         journal_t *journal;
 541         int rc;
 542
 543         LASSERT(current->journal_info == handle);
 544
 545         transaction = handle->h_transaction;
 546         journal = transaction->t_journal;
 547         tid = transaction->t_tid;
 548         /* we don't want to be blocked */
 549         handle->h_sync = 0;
 550         rc = ext3_journal_stop(handle);
 551         if (rc) {
 552                 CERROR("error while stopping transaction: %d\n", rc);
 553                 return rc;
 554         }
 555         fsfilt_log_start_commit(journal, tid);
 556
 557         *wait_handle = (void *) tid;
 558         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 559         return 0;
 560 }
 561
 562 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 563 {
 564         journal_t *journal = EXT3_JOURNAL(inode);
 565         tid_t tid = (tid_t)(long)h;
 566
 567         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 568         if (unlikely(is_journal_aborted(journal)))
 569                 return -EIO;
 570
 571         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 572
 573         if (unlikely(is_journal_aborted(journal)))
 574                 return -EIO;
 575         return 0;
 576 }
 577
 578 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 579                                struct iattr *iattr, int do_trunc)
 580 {
 581         struct inode *inode = dentry->d_inode;
 582         int rc = 0;
 583
 584         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 585          * We are already writing the inode to disk as part of this
 586          * transaction and want to avoid a lot of extra inode writeout
 587          * later on. b=9828 */
 588         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 589                 /* ATTR_SIZE would invoke truncate: clear it */
 590                 iattr->ia_valid &= ~ATTR_SIZE;
 591                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 592                 i_size_write(inode, iattr->ia_size);
 593
 594                 if (iattr->ia_valid & ATTR_UID)
 595                         inode->i_uid = iattr->ia_uid;
 596                 if (iattr->ia_valid & ATTR_GID)
 597                         inode->i_gid = iattr->ia_gid;
 598                 if (iattr->ia_valid & ATTR_ATIME)
 599                         inode->i_atime = iattr->ia_atime;
 600                 if (iattr->ia_valid & ATTR_MTIME)
 601                         inode->i_mtime = iattr->ia_mtime;
 602                 if (iattr->ia_valid & ATTR_CTIME)
 603                         inode->i_ctime = iattr->ia_ctime;
 604                 if (iattr->ia_valid & ATTR_MODE) {
 605                         inode->i_mode = iattr->ia_mode;
 606
 607                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 608                             !cfs_capable(CFS_CAP_FSETID))
 609                                 inode->i_mode &= ~S_ISGID;
 610                 }
 611
 612                 inode->i_sb->s_op->dirty_inode(inode);
 613
 614                 goto out;
 615         }
 616
 617         /* Don't allow setattr to change file type */
 618         if (iattr->ia_valid & ATTR_MODE)
 619                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 620                                  (iattr->ia_mode & ~S_IFMT);
 621
 622         /* We set these flags on the client, but have already checked perms
 623          * so don't confuse inode_change_ok. */
 624         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 625
 626         if (inode->i_op->setattr) {
 627                 rc = inode->i_op->setattr(dentry, iattr);
 628         } else {
 629                 rc = inode_change_ok(inode, iattr);
 630                 if (!rc)
 631                         rc = inode_setattr(inode, iattr);
 632         }
 633
 634  out:
 635         RETURN(rc);
 636 }
 637
 638 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 639                                  unsigned int cmd, unsigned long arg)
 640 {
 641         int rc = 0;
 642         ENTRY;
 643
 644         /* FIXME: Can't do this because of nested transaction deadlock */
 645         if (cmd == EXT3_IOC_SETFLAGS) {
 646                 /* We can't enable data journaling on OST objects, because
 647                 * this forces the transaction to be closed in order to
 648                 * flush the journal, but the caller will already have a
 649                 * compound transaction open to update the last_rcvd file,
 650                 * and this thread would deadlock trying to set the flag. */
 651                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 652                         CERROR("can't set data journal flag on file\n");
 653                         RETURN(-EPERM);
 654                 }
 655                 /* Because the MDS does not see the EXTENTS_FL set on the
 656                  * OST objects, mask this flag into all set flags.  It is
 657                  * not legal to clear this flag in any case, so we are not
 658                  * changing the functionality by doing this.  b=22911 */
 659                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 660         }
 661
 662 #ifdef HAVE_EXT4_LDISKFS
 663         /* ext4_ioctl does not have a inode argument */
 664         if (inode->i_fop->unlocked_ioctl)
 665                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 666 #else
 667         if (inode->i_fop->ioctl)
 668                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 669 #endif
 670         else
 671                 RETURN(-ENOTTY);
 672
 673         RETURN(rc);
 674 }
 675
 676 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 677                               void *lmm, int lmm_size, const char *name)
 678 {
 679         int rc;
 680
 681         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 682
 683         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 684                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 685
 686
 687         if (rc && rc != -EROFS)
 688                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 689                        inode->i_ino, rc);
 690         return rc;
 691 }
 692
 693 /* Must be called with i_mutex held */
 694 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 695                               const char *name)
 696 {
 697         int rc;
 698
 699         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 700
 701         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 702                             name, lmm, lmm_size);
 703
 704         /* This gives us the MD size */
 705         if (lmm == NULL)
 706                 return (rc == -ENODATA) ? 0 : rc;
 707
 708         if (rc < 0) {
 709                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 710                        EXT3_XATTR_INDEX_TRUSTED, name,
 711                        inode->i_ino, rc);
 712                 memset(lmm, 0, lmm_size);
 713                 return (rc == -ENODATA) ? 0 : rc;
 714         }
 715
 716         return rc;
 717 }
 718
 719 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 720 {
 721         submit_bio(rw, bio);
 722         return 0;
 723 }
 724
 725 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 726                                     loff_t *off)
 727 {
 728         struct inode *inode = file->f_dentry->d_inode;
 729         int rc = 0;
 730
 731         if (S_ISREG(inode->i_mode))
 732                 rc = file->f_op->read(file, buf, count, off);
 733         else {
 734                 const int blkbits = inode->i_sb->s_blocksize_bits;
 735                 const int blksize = inode->i_sb->s_blocksize;
 736
 737                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 738                        (unsigned long)count, inode->i_ino, *off);
 739                 while (count > 0) {
 740                         struct buffer_head *bh;
 741
 742                         bh = NULL;
 743                         if (*off < i_size_read(inode)) {
 744                                 int err = 0;
 745
 746                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 747                                                 0, &err);
 748
 749                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 750
 751                                 if (bh) {
 752                                         memcpy(buf, bh->b_data, blksize);
 753                                         brelse(bh);
 754                                 } else if (err) {
 755                                         /* XXX in theory we should just fake
 756                                          * this buffer and continue like ext3,
 757                                          * especially if this is a partial read
 758                                          */
 759                                         CERROR("error read dir %lu+%llu: %d\n",
 760                                                inode->i_ino, *off, err);
 761                                         RETURN(err);
 762                                 }
 763                         }
 764                         if (!bh) {
 765                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 766
 767                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 768                                 memset(fake, 0, sizeof(*fake));
 769                                 fake->rec_len = cpu_to_le16(blksize);
 770                         }
 771                         count -= blksize;
 772                         buf += blksize;
 773                         *off += blksize;
 774                         rc += blksize;
 775                 }
 776         }
 777
 778         return rc;
 779 }
 780
 781 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 782 static void fsfilt_ext3_cb_func(struct super_block *sb,
 783                                 struct journal_callback *jcb, int error)
 784 #else
 785 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 786 #endif
 787 {
 788         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 789
 790         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 791
 792         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 793 }
 794
 795 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 796                                       void *handle, fsfilt_cb_t cb_func,
 797                                       void *cb_data)
 798 {
 799         struct fsfilt_cb_data *fcb;
 800
 801         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 802         if (fcb == NULL)
 803                 RETURN(-ENOMEM);
 804
 805         fcb->cb_func = cb_func;
 806         fcb->cb_obd = obd;
 807         fcb->cb_last_rcvd = last_rcvd;
 808         fcb->cb_data = cb_data;
 809
 810         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 811         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 812
 813         return 0;
 814 }
 815
 816 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 817 {
 818         struct kstatfs sfs;
 819         int rc;
 820
 821         memset(&sfs, 0, sizeof(sfs));
 822         rc = ll_do_statfs(sb, &sfs);
 823         statfs_pack(osfs, &sfs);
 824         return rc;
 825 }
 826
 827 static int fsfilt_ext3_sync(struct super_block *sb)
 828 {
 829         return ext3_force_commit(sb);
 830 }
 831
 832 #ifndef EXT3_EXTENTS_FL
 833 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 834 #endif
 835
 836 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 837 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 838 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 839 #else
 840 # ifdef HAVE_EXT4_LDISKFS
 841 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 842 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 843 # else
 844 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 845 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 846 # endif
 847 #endif
 848
 849 #ifndef EXT_ASSERT
 850 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 851 #endif
 852
 853 #ifdef EXT3_EXT_HAS_NO_TREE
 854 /* for kernels 2.6.18 and later */
 855 #ifdef HAVE_EXT4_LDISKFS
 856 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 857 #else
 858 #define EXT_GENERATION(inode)           ext_generation(inode)
 859 #endif
 860 #define ext3_ext_base                   inode
 861 #define ext3_ext_base2inode(inode)      (inode)
 862 #define EXT_DEPTH(inode)                ext_depth(inode)
 863 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 864                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 865 #else
 866 #define ext3_ext_base                   ext3_extents_tree
 867 #define ext3_ext_base2inode(tree)       (tree->inode)
 868 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 869                         ext3_ext_walk_space(tree, block, num, cb);
 870 #endif
 871
 872 #include <linux/lustre_version.h>
 873
 874 struct bpointers {
 875         unsigned long *blocks;
 876         int *created;
 877         unsigned long start;
 878         int num;
 879         int init_num;
 880         int create;
 881 };
 882
 883 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 884                                unsigned long block, int *aflags)
 885 {
 886         struct ext3_inode_info *ei = EXT3_I(inode);
 887         unsigned long bg_start;
 888         unsigned long colour;
 889         int depth;
 890
 891         if (path) {
 892                 struct ext3_extent *ex;
 893                 depth = path->p_depth;
 894
 895                 /* try to predict block placement */
 896                 if ((ex = path[depth].p_ext))
 897                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 898
 899                 /* it looks index is empty
 900                  * try to find starting from index itself */
 901                 if (path[depth].p_bh)
 902                         return path[depth].p_bh->b_blocknr;
 903         }
 904
 905         /* OK. use inode's group */
 906         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 907                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 908         colour = (current->pid % 16) *
 909                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 910         return bg_start + colour + block;
 911 }
 912
 913 #define ll_unmap_underlying_metadata(sb, blocknr) \
 914         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 915
 916 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 917 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 918                                 struct ext3_ext_path *path, unsigned long block,
 919                                 unsigned long *count, int *err)
 920 {
 921         unsigned long pblock, goal;
 922         int aflags = 0;
 923         struct inode *inode = ext3_ext_base2inode(base);
 924
 925         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 926         aflags |= 2; /* block have been already reserved */
 927         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 928         return pblock;
 929
 930 }
 931 #else
 932 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 933                                 struct ext3_ext_path *path, unsigned long block,
 934                                 unsigned long *count, int *err)
 935 {
 936         struct inode *inode = ext3_ext_base2inode(base);
 937         struct ext3_allocation_request ar;
 938         unsigned long pblock;
 939         int aflags;
 940
 941         /* find neighbour allocated blocks */
 942         ar.lleft = block;
 943         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 944         if (*err)
 945                 return 0;
 946         ar.lright = block;
 947         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 948         if (*err)
 949                 return 0;
 950
 951         /* allocate new block */
 952         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 953         ar.inode = inode;
 954         ar.logical = block;
 955         ar.len = *count;
 956         ar.flags = EXT3_MB_HINT_DATA;
 957         pblock = ext3_mb_new_blocks(handle, &ar, err);
 958         *count = ar.len;
 959         return pblock;
 960 }
 961 #endif
 962
 963 #ifdef EXT3_EXT_HAS_NO_TREE
 964 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 965                                   struct ext3_ext_path *path,
 966                                   struct ext3_ext_cache *cex,
 967 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 968                                    struct ext3_extent *ex,
 969 #endif
 970                                   void *cbdata)
 971 {
 972         struct bpointers *bp = cbdata;
 973 #else
 974 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 975                                   struct ext3_ext_path *path,
 976                                   struct ext3_ext_cache *cex
 977 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 978                                   , struct ext3_extent *ex
 979 #endif
 980                                  )
 981 {
 982         struct bpointers *bp = base->private;
 983 #endif
 984         struct inode *inode = ext3_ext_base2inode(base);
 985         struct ext3_extent nex;
 986         unsigned long pblock;
 987         unsigned long tgen;
 988         int err, i;
 989         unsigned long count;
 990         handle_t *handle;
 991
 992         i = EXT_DEPTH(base);
 993         EXT_ASSERT(path[i].p_hdr);
 994         LASSERTF(i == path->p_depth ||
 995                  EXT_GENERATION(base) != path[0].p_generation,
 996                  "base vs path extent depth:%d != %d, generation:%lu == %lu\n",
 997                  i, path->p_depth, EXT_GENERATION(base), path[0].p_generation);
 998
 999         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1000                 err = EXT_CONTINUE;
1001                 goto map;
1002         }
1003
1004         if (bp->create == 0) {
1005                 i = 0;
1006                 if (cex->ec_block < bp->start)
1007                         i = bp->start - cex->ec_block;
1008                 if (i >= cex->ec_len)
1009                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1010                                         i, cex->ec_len);
1011                 for (; i < cex->ec_len && bp->num; i++) {
1012                         *(bp->created) = 0;
1013                         bp->created++;
1014                         *(bp->blocks) = 0;
1015                         bp->blocks++;
1016                         bp->num--;
1017                         bp->start++;
1018                 }
1019
1020                 return EXT_CONTINUE;
1021         }
1022
1023         tgen = EXT_GENERATION(base);
1024         count = ext3_ext_calc_credits_for_insert(base, path);
1025         fsfilt_up_truncate_sem(inode);
1026
1027         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1028         if (IS_ERR(handle)) {
1029                 fsfilt_down_truncate_sem(inode);
1030                 return PTR_ERR(handle);
1031         }
1032
1033         fsfilt_down_truncate_sem(inode);
1034         if (tgen != EXT_GENERATION(base)) {
1035                 /* the tree has changed. so path can be invalid at moment */
1036                 ext3_journal_stop(handle);
1037                 return EXT_REPEAT;
1038         }
1039
1040 #if defined(HAVE_EXT4_LDISKFS)
1041         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1042          * protected by i_data_sem as whole. so we patch it to store
1043          * generation to path and now verify the tree hasn't changed */
1044         down_write((&EXT4_I(inode)->i_data_sem));
1045
1046         /* validate extent, make sure the extent tree does not changed */
1047         if (EXT_GENERATION(base) != path[0].p_generation) {
1048                 /* cex is invalid, try again */
1049                 up_write(&EXT4_I(inode)->i_data_sem);
1050                 ext3_journal_stop(handle);
1051                 return EXT_REPEAT;
1052         }
1053 #endif
1054
1055         count = cex->ec_len;
1056         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1057         if (!pblock)
1058                 goto out;
1059         EXT_ASSERT(count <= cex->ec_len);
1060
1061         /* insert new extent */
1062         nex.ee_block = cpu_to_le32(cex->ec_block);
1063         ext3_ext_store_pblock(&nex, pblock);
1064         nex.ee_len = cpu_to_le16(count);
1065         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1066         if (err) {
1067                 /* free data blocks we just allocated */
1068                 /* not a good idea to call discard here directly,
1069                  * but otherwise we'd need to call it every free() */
1070 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1071                 ext3_mb_discard_inode_preallocations(inode);
1072 #endif
1073                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1074                                  cpu_to_le16(nex.ee_len), 0);
1075                 goto out;
1076         }
1077
1078         /*
1079          * Putting len of the actual extent we just inserted,
1080          * we are asking ext3_ext_walk_space() to continue
1081          * scaning after that block
1082          */
1083         cex->ec_len = le16_to_cpu(nex.ee_len);
1084         cex->ec_start = ext_pblock(&nex);
1085         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1086         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1087
1088 out:
1089 #if defined(HAVE_EXT4_LDISKFS)
1090         up_write((&EXT4_I(inode)->i_data_sem));
1091 #endif
1092         ext3_journal_stop(handle);
1093 map:
1094         if (err >= 0) {
1095                 /* map blocks */
1096                 if (bp->num == 0) {
1097                         CERROR("hmm. why do we find this extent?\n");
1098                         CERROR("initial space: %lu:%u\n",
1099                                 bp->start, bp->init_num);
1100                         CERROR("current extent: %u/%u/%llu %d\n",
1101                                 cex->ec_block, cex->ec_len,
1102                                 (unsigned long long)cex->ec_start,
1103                                 cex->ec_type);
1104                 }
1105                 i = 0;
1106                 if (cex->ec_block < bp->start)
1107                         i = bp->start - cex->ec_block;
1108                 if (i >= cex->ec_len)
1109                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1110                                         i, cex->ec_len);
1111                 for (; i < cex->ec_len && bp->num; i++) {
1112                         *(bp->blocks) = cex->ec_start + i;
1113                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1114                                 *(bp->created) = 0;
1115                         } else {
1116                                 *(bp->created) = 1;
1117                                 /* unmap any possible underlying metadata from
1118                                  * the block device mapping.  bug 6998. */
1119                                 ll_unmap_underlying_metadata(inode->i_sb,
1120                                                              *(bp->blocks));
1121                         }
1122                         bp->created++;
1123                         bp->blocks++;
1124                         bp->num--;
1125                         bp->start++;
1126                 }
1127         }
1128         return err;
1129 }
1130
1131 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1132                        unsigned long num, unsigned long *blocks,
1133                        int *created, int create)
1134 {
1135 #ifdef EXT3_EXT_HAS_NO_TREE
1136         struct ext3_ext_base *base = inode;
1137 #else
1138         struct ext3_extents_tree tree;
1139         struct ext3_ext_base *base = &tree;
1140 #endif
1141         struct bpointers bp;
1142         int err;
1143
1144         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1145                block, block + num - 1, (unsigned) inode->i_ino);
1146
1147 #ifndef EXT3_EXT_HAS_NO_TREE
1148         ext3_init_tree_desc(base, inode);
1149         tree.private = &bp;
1150 #endif
1151         bp.blocks = blocks;
1152         bp.created = created;
1153         bp.start = block;
1154         bp.init_num = bp.num = num;
1155         bp.create = create;
1156
1157         fsfilt_down_truncate_sem(inode);
1158         err = fsfilt_ext3_ext_walk_space(base, block, num,
1159                                          ext3_ext_new_extent_cb, &bp);
1160         ext3_ext_invalidate_cache(base);
1161         fsfilt_up_truncate_sem(inode);
1162
1163         return err;
1164 }
1165
1166 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1167                                     int pages, unsigned long *blocks,
1168                                     int *created, int create)
1169 {
1170         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1171         int rc = 0, i = 0;
1172         struct page *fp = NULL;
1173         int clen = 0;
1174
1175         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1176                 inode->i_ino, pages, (*page)->index);
1177
1178         /* pages are sorted already. so, we just have to find
1179          * contig. space and process them properly */
1180         while (i < pages) {
1181                 if (fp == NULL) {
1182                         /* start new extent */
1183                         fp = *page++;
1184                         clen = 1;
1185                         i++;
1186                         continue;
1187                 } else if (fp->index + clen == (*page)->index) {
1188                         /* continue the extent */
1189                         page++;
1190                         clen++;
1191                         i++;
1192                         continue;
1193                 }
1194
1195                 /* process found extent */
1196                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1197                                         clen * blocks_per_page, blocks,
1198                                         created, create);
1199                 if (rc)
1200                         GOTO(cleanup, rc);
1201
1202                 /* look for next extent */
1203                 fp = NULL;
1204                 blocks += blocks_per_page * clen;
1205                 created += blocks_per_page * clen;
1206         }
1207
1208         if (fp)
1209                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1210                                         clen * blocks_per_page, blocks,
1211                                         created, create);
1212 cleanup:
1213         return rc;
1214 }
1215
1216 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1217                                unsigned long *blocks, int *created, int create);
1218 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1219                                    int pages, unsigned long *blocks,
1220                                    int *created, int create)
1221 {
1222         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1223         unsigned long *b;
1224         int rc = 0, i, *cr;
1225
1226         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1227                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1228                 if (rc) {
1229                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1230                                inode->i_ino, *b, *cr, create, rc);
1231                         break;
1232                 }
1233
1234                 b += blocks_per_page;
1235                 cr += blocks_per_page;
1236         }
1237         return rc;
1238 }
1239
1240 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1241                                 int pages, unsigned long *blocks,
1242                                 int *created, int create,
1243                                 cfs_semaphore_t *optional_sem)
1244 {
1245         int rc;
1246
1247         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1248                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1249                                                      blocks, created, create);
1250                 return rc;
1251         }
1252         if (optional_sem != NULL)
1253                 cfs_down(optional_sem);
1254         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1255                                             created, create);
1256         if (optional_sem != NULL)
1257                 cfs_up(optional_sem);
1258
1259         return rc;
1260 }
1261
1262 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1263 {
1264         unsigned long block;
1265         struct buffer_head *bh;
1266         int err, blocksize, csize, boffs, osize = size;
1267
1268         /* prevent reading after eof */
1269         cfs_lock_kernel();
1270         if (i_size_read(inode) < *offs + size) {
1271                 size = i_size_read(inode) - *offs;
1272                 cfs_unlock_kernel();
1273                 if (size < 0) {
1274                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1275                                i_size_read(inode), *offs);
1276                         return -EBADR;
1277                 } else if (size == 0) {
1278                         return 0;
1279                 }
1280         } else {
1281                 cfs_unlock_kernel();
1282         }
1283
1284         blocksize = 1 << inode->i_blkbits;
1285
1286         while (size > 0) {
1287                 block = *offs >> inode->i_blkbits;
1288                 boffs = *offs & (blocksize - 1);
1289                 csize = min(blocksize - boffs, size);
1290                 bh = ext3_bread(NULL, inode, block, 0, &err);
1291                 if (!bh) {
1292                         CERROR("can't read block: %d\n", err);
1293                         return err;
1294                 }
1295
1296                 memcpy(buf, bh->b_data + boffs, csize);
1297                 brelse(bh);
1298
1299                 *offs += csize;
1300                 buf += csize;
1301                 size -= csize;
1302         }
1303         return osize;
1304 }
1305 EXPORT_SYMBOL(fsfilt_ext3_read);
1306
1307 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1308                                    int size, loff_t *offs)
1309 {
1310         int rc;
1311         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1312         if (rc > 0)
1313                 rc = 0;
1314         return rc;
1315 }
1316
1317 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1318                                 loff_t *offs, handle_t *handle)
1319 {
1320         struct buffer_head *bh = NULL;
1321         loff_t old_size = i_size_read(inode), offset = *offs;
1322         loff_t new_size = i_size_read(inode);
1323         unsigned long block;
1324         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1325
1326         while (bufsize > 0) {
1327                 if (bh != NULL)
1328                         brelse(bh);
1329
1330                 block = offset >> inode->i_blkbits;
1331                 boffs = offset & (blocksize - 1);
1332                 size = min(blocksize - boffs, bufsize);
1333                 bh = ext3_bread(handle, inode, block, 1, &err);
1334                 if (!bh) {
1335                         CERROR("can't read/create block: %d\n", err);
1336                         break;
1337                 }
1338
1339                 err = ext3_journal_get_write_access(handle, bh);
1340                 if (err) {
1341                         CERROR("journal_get_write_access() returned error %d\n",
1342                                err);
1343                         break;
1344                 }
1345                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1346                 memcpy(bh->b_data + boffs, buf, size);
1347                 err = ext3_journal_dirty_metadata(handle, bh);
1348                 if (err) {
1349                         CERROR("journal_dirty_metadata() returned error %d\n",
1350                                err);
1351                         break;
1352                 }
1353                 if (offset + size > new_size)
1354                         new_size = offset + size;
1355                 offset += size;
1356                 bufsize -= size;
1357                 buf += size;
1358         }
1359         if (bh)
1360                 brelse(bh);
1361
1362         /* correct in-core and on-disk sizes */
1363         if (new_size > i_size_read(inode)) {
1364                 cfs_lock_kernel();
1365                 if (new_size > i_size_read(inode))
1366                         i_size_write(inode, new_size);
1367                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1368                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1369                 if (i_size_read(inode) > old_size)
1370                         mark_inode_dirty(inode);
1371                 cfs_unlock_kernel();
1372         }
1373
1374         if (err == 0)
1375                 *offs = offset;
1376         return err;
1377 }
1378 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1379
1380 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1381                                     loff_t *offs, int force_sync)
1382 {
1383         struct inode *inode = file->f_dentry->d_inode;
1384         handle_t *handle;
1385         int err, block_count = 0, blocksize;
1386
1387         /* Determine how many transaction credits are needed */
1388         blocksize = 1 << inode->i_blkbits;
1389         block_count = (*offs & (blocksize - 1)) + bufsize;
1390         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1391
1392         handle = ext3_journal_start(inode,
1393                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1394         if (IS_ERR(handle)) {
1395                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1396                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1397                 return PTR_ERR(handle);
1398         }
1399
1400         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1401
1402         if (!err && force_sync)
1403                 handle->h_sync = 1; /* recovery likes this */
1404
1405         ext3_journal_stop(handle);
1406
1407         return err;
1408 }
1409
1410 static int fsfilt_ext3_setup(struct super_block *sb)
1411 {
1412 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1413      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1414         struct ext3_sb_info *sbi = EXT3_SB(sb);
1415 #if 0
1416         sbi->dx_lock = fsfilt_ext3_dx_lock;
1417         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1418 #endif
1419 #endif
1420 #ifdef S_PDIROPS
1421         CWARN("Enabling PDIROPS\n");
1422         set_opt(sbi->s_mount_opt, PDIROPS);
1423         sb->s_flags |= S_PDIROPS;
1424 #endif
1425         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1426                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1427 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1428         /* enable journaled quota support */
1429         /* kfreed in ext3_put_super() */
1430         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1431         if (!sbi->s_qf_names[USRQUOTA])
1432                 return -ENOMEM;
1433         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1434         if (!sbi->s_qf_names[GRPQUOTA]) {
1435                 kfree(sbi->s_qf_names[USRQUOTA]);
1436                 sbi->s_qf_names[USRQUOTA] = NULL;
1437                 return -ENOMEM;
1438         }
1439         sbi->s_jquota_fmt = QFMT_LUSTRE;
1440 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1441         set_opt(sbi->s_mount_opt, QUOTA);
1442 #endif
1443 #endif
1444         return 0;
1445 }
1446
1447 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1448    objects. Logs is number of logfiles to update */
1449 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1450 {
1451         if ( !fso ) {
1452                 switch(op) {
1453                 case FSFILT_OP_CREATE:
1454                                  /* directory leaf, index & indirect & EA*/
1455                         return 4 + 3 * logs;
1456                 case FSFILT_OP_UNLINK:
1457                         return 3 * logs;
1458                 }
1459         } else {
1460                 int i;
1461                 int needed = 0;
1462                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1463                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1464                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1465                 for (i = 0; i < op; i++, fso++) {
1466                         int nblocks = fso->fso_bufcnt * blockpp;
1467                         int ndindirect = min(nblocks, addrpp + 1);
1468                         int nindir = nblocks + ndindirect + 1;
1469
1470                         needed += nindir;
1471                 }
1472                 return needed + 3 * logs;
1473         }
1474
1475         return 0;
1476 }
1477
1478 #ifdef HAVE_QUOTA_SUPPORT
1479 #define DQINFO_COPY(out, in)                    \
1480 do {                                            \
1481         Q_COPY(out, in, dqi_bgrace);            \
1482         Q_COPY(out, in, dqi_igrace);            \
1483         Q_COPY(out, in, dqi_flags);             \
1484         Q_COPY(out, in, dqi_valid);             \
1485 } while (0)
1486
1487 #define DQBLK_COPY(out, in)                     \
1488 do {                                            \
1489         Q_COPY(out, in, dqb_bhardlimit);        \
1490         Q_COPY(out, in, dqb_bsoftlimit);        \
1491         Q_COPY(out, in, dqb_curspace);          \
1492         Q_COPY(out, in, dqb_ihardlimit);        \
1493         Q_COPY(out, in, dqb_isoftlimit);        \
1494         Q_COPY(out, in, dqb_curinodes);         \
1495         Q_COPY(out, in, dqb_btime);             \
1496         Q_COPY(out, in, dqb_itime);             \
1497         Q_COPY(out, in, dqb_valid);             \
1498 } while (0)
1499
1500 static int fsfilt_ext3_quotactl(struct super_block *sb,
1501                                 struct obd_quotactl *oqc)
1502 {
1503         int i, rc = 0, error = 0;
1504         const struct quotactl_ops *qcop;
1505         struct if_dqinfo *info;
1506         struct if_dqblk *dqblk;
1507         ENTRY;
1508
1509         if (!sb->s_qcop)
1510                 RETURN(-ENOSYS);
1511
1512         OBD_ALLOC_PTR(info);
1513         if (!info)
1514                 RETURN(-ENOMEM);
1515         OBD_ALLOC_PTR(dqblk);
1516         if (!dqblk) {
1517                 OBD_FREE_PTR(info);
1518                 RETURN(-ENOMEM);
1519         }
1520
1521         DQINFO_COPY(info, &oqc->qc_dqinfo);
1522         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1523
1524         qcop = sb->s_qcop;
1525         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1526                 for (i = 0; i < MAXQUOTAS; i++) {
1527                         if (!Q_TYPESET(oqc, i))
1528                                 continue;
1529
1530                         if (oqc->qc_cmd == Q_QUOTAON) {
1531                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1532
1533                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1534
1535                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1536                                                  name[i], 0);
1537                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1538                                 rc = ll_quota_off(sb, i, 0);
1539                         }
1540
1541                         if (rc == -EBUSY)
1542                                 error = rc;
1543                         else if (rc)
1544                                 GOTO(out, rc);
1545                 }
1546                 GOTO(out, rc ?: error);
1547         }
1548
1549         switch (oqc->qc_cmd) {
1550         case Q_GETOINFO:
1551         case Q_GETINFO:
1552                 if (!qcop->get_info)
1553                         GOTO(out, rc = -ENOSYS);
1554                 rc = qcop->get_info(sb, oqc->qc_type, info);
1555                 break;
1556         case Q_SETQUOTA:
1557         case Q_INITQUOTA:
1558                 if (!qcop->set_dqblk)
1559                         GOTO(out, rc = -ENOSYS);
1560                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1561                 break;
1562         case Q_GETOQUOTA:
1563         case Q_GETQUOTA:
1564                 if (!qcop->get_dqblk)
1565                         GOTO(out, rc = -ENOSYS);
1566                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1567                 if (!rc)
1568                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1569                 break;
1570         case Q_SYNC:
1571                 if (!sb->s_qcop->quota_sync)
1572                         GOTO(out, rc = -ENOSYS);
1573                 qcop->quota_sync(sb, oqc->qc_type);
1574                 break;
1575         case Q_FINVALIDATE:
1576                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1577                 for (i = 0; i < MAXQUOTAS; i++) {
1578                         struct file *fp;
1579                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1580
1581                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1582
1583                         if (!Q_TYPESET(oqc, i))
1584                                 continue;
1585
1586                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1587                         if (IS_ERR(fp)) {
1588                                 rc = PTR_ERR(fp);
1589                                 CERROR("error invalidating operational quota file"
1590                                        " %s (rc:%d)\n", name[i], rc);
1591                         } else {
1592                                 filp_close(fp, 0);
1593                         }
1594
1595                 }
1596                 break;
1597         default:
1598                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1599                 LBUG();
1600         }
1601 out:
1602         DQINFO_COPY(&oqc->qc_dqinfo, info);
1603         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1604
1605         OBD_FREE_PTR(info);
1606         OBD_FREE_PTR(dqblk);
1607
1608         if (rc)
1609                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1610                                 "failed: %d\n",
1611                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1612         RETURN(rc);
1613 }
1614
1615 struct chk_dqblk{
1616         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1617         cfs_list_t              dqb_list;        /** in list also */
1618         qid_t                   dqb_id;          /** uid/gid */
1619         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1620         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1621         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1622         qsize_t                 dqb_curspace;    /** current space */
1623         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1624         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1625         qsize_t                 dqb_curinodes;   /** current inodes */
1626         obd_time                dqb_btime;       /** block grace time */
1627         obd_time                dqb_itime;       /** inode grace time */
1628         __u32                   dqb_valid;       /** flag for above fields */
1629 };
1630
1631 static inline unsigned int chkquot_hash(qid_t id, int type)
1632                                         __attribute__((__const__));
1633
1634 static inline unsigned int chkquot_hash(qid_t id, int type)
1635 {
1636         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1637 }
1638
1639 static inline struct chk_dqblk *
1640 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1641 {
1642         cfs_hlist_node_t *node;
1643         struct chk_dqblk *cdqb;
1644
1645         cfs_hlist_for_each(node, head) {
1646                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1647                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1648                         return cdqb;
1649         }
1650
1651         return NULL;
1652 }
1653
1654 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1655 {
1656         struct chk_dqblk *cdqb;
1657
1658         OBD_ALLOC_PTR(cdqb);
1659         if (cdqb) {
1660                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1661                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1662                 cdqb->dqb_id = id;
1663                 cdqb->dqb_type = type;
1664         }
1665
1666         return cdqb;
1667 }
1668
1669 static struct chk_dqblk *
1670 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1671       cfs_list_t *list, qid_t id, int type, int first_check)
1672 {
1673         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1674         struct if_dqblk dqb;
1675         struct chk_dqblk *cdqb;
1676         int rc;
1677
1678         cdqb = find_chkquot(head, id, type);
1679         if (cdqb)
1680                 return cdqb;
1681
1682         cdqb = alloc_chkquot(id, type);
1683         if (!cdqb)
1684                 return NULL;
1685
1686         if (!first_check) {
1687                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1688                 if (rc) {
1689                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1690                                id, type, rc);
1691                 } else {
1692                         DQBLK_COPY(cdqb, &dqb);
1693                         cdqb->dqb_curspace = 0;
1694                         cdqb->dqb_curinodes = 0;
1695                 }
1696         }
1697
1698         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1699         cfs_list_add_tail(&cdqb->dqb_list, list);
1700
1701         return cdqb;
1702 }
1703
1704 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1705 {
1706         struct obd_quotactl *oqctl;
1707         int rc;
1708
1709         OBD_ALLOC_PTR(oqctl);
1710         if (!oqctl)
1711                 RETURN(-ENOMEM);
1712
1713         oqctl->qc_cmd = cmd;
1714         oqctl->qc_id = qfmt;
1715         oqctl->qc_type = type;
1716         rc = fsfilt_ext3_quotactl(sb, oqctl);
1717
1718         OBD_FREE_PTR(oqctl);
1719         return rc;
1720 }
1721
1722 static inline int read_old_dqinfo(struct super_block *sb, int type,
1723                                   struct if_dqinfo *dqinfo)
1724 {
1725         struct obd_quotactl *oqctl;
1726         int rc;
1727         ENTRY;
1728
1729         OBD_ALLOC_PTR(oqctl);
1730         if (!oqctl)
1731                 RETURN(-ENOMEM);
1732
1733         oqctl->qc_cmd = Q_GETINFO;
1734         oqctl->qc_type = type;
1735         rc = fsfilt_ext3_quotactl(sb, oqctl);
1736         if (!rc)
1737                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1738
1739         OBD_FREE_PTR(oqctl);
1740         RETURN(rc);
1741 }
1742
1743 #ifndef HAVE_EXT4_LDISKFS
1744 static inline struct ext3_group_desc *
1745 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1746 {
1747         unsigned long desc_block, desc;
1748         struct ext3_group_desc *gdp;
1749
1750         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1751         desc = group % EXT3_DESC_PER_BLOCK(sb);
1752         gdp = (struct ext3_group_desc *)
1753               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1754
1755         return gdp + desc;
1756 }
1757
1758 static inline struct buffer_head *
1759 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1760 {
1761         struct ext3_group_desc *desc;
1762         struct buffer_head *bh;
1763
1764         desc = get_group_desc(sb, group, NULL);
1765         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1766         return bh;
1767 }
1768
1769 static __u32 ext3_itable_unused_count(struct super_block *sb,
1770                                struct ext3_group_desc *bg) {
1771        return le16_to_cpu(bg->bg_itable_unused);
1772 }
1773 #else
1774 #define get_group_desc ext3_get_group_desc
1775 #endif
1776
1777 struct qchk_ctxt {
1778         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1779         cfs_list_t              qckt_list;                 /* quotacheck list */
1780         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1781         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1782 };
1783
1784 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1785                            struct obd_quotactl *oqc)
1786 {
1787         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1788         loff_t size = 0;
1789         qid_t qid[MAXQUOTAS];
1790         int cnt, i, rc = 0;
1791
1792         if (!inode)
1793                 return 0;
1794
1795         qid[USRQUOTA] = inode->i_uid;
1796         qid[GRPQUOTA] = inode->i_gid;
1797
1798         if (S_ISDIR(inode->i_mode) ||
1799             S_ISREG(inode->i_mode) ||
1800             S_ISLNK(inode->i_mode))
1801                 size = inode_get_bytes(inode);
1802
1803         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1804                 if (!Q_TYPESET(oqc, cnt))
1805                         continue;
1806
1807                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1808                                 &qctxt->qckt_list, qid[cnt], cnt,
1809                                 qctxt->qckt_first_check[cnt]);
1810                 if (!cdqb[cnt]) {
1811                         rc = -ENOMEM;
1812                         break;
1813                 }
1814
1815                 cdqb[cnt]->dqb_curspace += size;
1816                 cdqb[cnt]->dqb_curinodes++;
1817         }
1818
1819         if (rc) {
1820                 for (i = 0; i < cnt; i++) {
1821                         if (!Q_TYPESET(oqc, i))
1822                                 continue;
1823                         LASSERT(cdqb[i]);
1824                         cdqb[i]->dqb_curspace -= size;
1825                         cdqb[i]->dqb_curinodes--;
1826                 }
1827         }
1828
1829         return rc;
1830 }
1831
1832 /* write dqinfo struct in a new quota file */
1833 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1834 {
1835         struct v2_disk_dqinfo dqinfo;
1836         __u32 blocks = V2_DQTREEOFF + 1;
1837         loff_t offset = V2_DQINFOOFF;
1838
1839         if (info) {
1840                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1841                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1842                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1843                                                ~DQF_INFO_DIRTY);
1844         } else {
1845                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1846                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1847                 dqinfo.dqi_flags = 0;
1848         }
1849
1850         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1851         dqinfo.dqi_free_blk = 0;
1852         dqinfo.dqi_free_entry = 0;
1853
1854         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1855 }
1856
1857 static int v3_write_dqheader(struct file *f, int type)
1858 {
1859         static const __u32 quota_magics[] = V2_INITQMAGICS;
1860         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1861         struct v2_disk_dqheader dqhead;
1862         loff_t offset = 0;
1863
1864         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1865         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1866
1867         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1868         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1869
1870         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1871 }
1872
1873 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1874                                   struct obd_quotactl *oqc)
1875 {
1876         int i, rc = 0;
1877         ENTRY;
1878
1879         for (i = 0; i < MAXQUOTAS; i++) {
1880                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1881                                          NULL : &qctxt->qckt_dqinfo[i];
1882                 struct file *file;
1883                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1884
1885                 if (!Q_TYPESET(oqc, i))
1886                         continue;
1887
1888                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1889
1890                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1891                 if (IS_ERR(file)) {
1892                         rc = PTR_ERR(file);
1893                         CERROR("can't create %s file: rc = %d\n",
1894                                name[i], rc);
1895                         GOTO(out, rc);
1896                 }
1897
1898                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1899                         CERROR("file %s is not regular", name[i]);
1900                         filp_close(file, 0);
1901                         GOTO(out, rc = -EINVAL);
1902                 }
1903
1904                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1905
1906                 rc = v3_write_dqheader(file, i);
1907                 if (rc) {
1908                         filp_close(file, 0);
1909                         GOTO(out, rc);
1910                 }
1911
1912                 rc = v3_write_dqinfo(file, i, info);
1913                 filp_close(file, 0);
1914                 if (rc)
1915                         GOTO(out, rc);
1916         }
1917
1918 out:
1919         RETURN(rc);
1920 }
1921
1922
1923 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1924                           struct chk_dqblk *cdqb)
1925 {
1926         struct obd_quotactl *oqc;
1927         long now;
1928         int rc;
1929         ENTRY;
1930
1931         OBD_ALLOC_PTR(oqc);
1932         if (!oqc)
1933                 RETURN(-ENOMEM);
1934
1935         now = cfs_time_current_sec();
1936
1937         if (cdqb->dqb_bsoftlimit &&
1938             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1939             !cdqb->dqb_btime)
1940                 cdqb->dqb_btime =
1941                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1942
1943         if (cdqb->dqb_isoftlimit &&
1944             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1945             !cdqb->dqb_itime)
1946                 cdqb->dqb_itime =
1947                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1948
1949         cdqb->dqb_valid = QIF_ALL;
1950
1951         oqc->qc_cmd = Q_SETQUOTA;
1952         oqc->qc_type = cdqb->dqb_type;
1953         oqc->qc_id = cdqb->dqb_id;
1954         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1955
1956         rc = fsfilt_ext3_quotactl(sb, oqc);
1957         OBD_FREE_PTR(oqc);
1958         RETURN(rc);
1959 }
1960
1961 static int prune_chkquots(struct super_block *sb,
1962                           struct qchk_ctxt *qctxt, int error)
1963 {
1964         struct chk_dqblk *cdqb, *tmp;
1965         int rc;
1966
1967         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1968                 if (!error) {
1969                         rc = commit_chkquot(sb, qctxt, cdqb);
1970                         if (rc)
1971                                 error = rc;
1972                 }
1973                 cfs_hlist_del_init(&cdqb->dqb_hash);
1974                 cfs_list_del(&cdqb->dqb_list);
1975                 OBD_FREE_PTR(cdqb);
1976         }
1977
1978         return error;
1979 }
1980
1981 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1982 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1983 #endif
1984
1985 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1986                                   struct obd_quotactl *oqc)
1987 {
1988         struct ext3_sb_info *sbi = EXT3_SB(sb);
1989         int i, group, uninit_feat = 0;
1990         struct qchk_ctxt *qctxt;
1991         struct buffer_head *bitmap_bh = NULL;
1992         unsigned long ino, inode_inuse;
1993         struct inode *inode;
1994         int rc = 0;
1995         ENTRY;
1996
1997         /* turn on quota and read dqinfo if existed */
1998         OBD_ALLOC_PTR(qctxt);
1999         if (!qctxt) {
2000                 oqc->qc_stat = -ENOMEM;
2001                 RETURN(-ENOMEM);
2002         }
2003
2004         for (i = 0; i < NR_DQHASH; i++)
2005                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2006         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2007
2008         for (i = 0; i < MAXQUOTAS; i++) {
2009                 if (!Q_TYPESET(oqc, i))
2010                         continue;
2011
2012                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2013                 if (!rc || rc == -EBUSY) {
2014                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2015                         if (rc)
2016                                 GOTO(out, rc);
2017                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2018                         qctxt->qckt_first_check[i] = 1;
2019                 } else if (rc) {
2020                         GOTO(out, rc);
2021                 }
2022         }
2023         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2024                 /* This filesystem supports the uninit group feature */
2025                 uninit_feat = 1;
2026
2027         /* number of inodes that have been allocated */
2028         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2029                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2030
2031         /* check quota and update in hash */
2032         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2033              group++) {
2034                 unsigned long used_count = sbi->s_inodes_per_group;
2035
2036                 if (uninit_feat) {
2037                         struct ext3_group_desc *desc;
2038                         desc = get_group_desc(sb, group, NULL);
2039                         if (!desc)
2040                                 GOTO(out, -EIO);
2041
2042                         /* we don't really need to take the group lock here,
2043                          * but it may be useful if one day we support online
2044                          * quotacheck */
2045 #ifdef HAVE_EXT4_LDISKFS
2046                         ext4_lock_group(sb, group);
2047 #else
2048                         spin_lock(sb_bgl_lock(sbi, group));
2049 #endif
2050                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2051                                 /* no inode in use in this group, just skip it */
2052 #ifdef HAVE_EXT4_LDISKFS
2053                                 ext3_unlock_group(sb, group);
2054 #else
2055                                 spin_unlock(sb_bgl_lock(sbi, group));
2056 #endif
2057                                 continue;
2058                         }
2059
2060                         used_count -= ext3_itable_unused_count(sb, desc);
2061 #ifdef HAVE_EXT4_LDISKFS
2062                         ext3_unlock_group(sb, group);
2063 #else
2064                         spin_unlock(sb_bgl_lock(sbi, group));
2065 #endif
2066                 }
2067
2068                 ino = group * sbi->s_inodes_per_group + 1;
2069                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2070                 if (!bitmap_bh) {
2071                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2072                                sb->s_id, group);
2073                         GOTO(out, -EIO);
2074                 }
2075
2076                 i = 0;
2077                 while (i < used_count &&
2078                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2079                                                used_count, i)) < used_count) {
2080                         inode_inuse--;
2081                         i++;
2082                         ino = i + group * sbi->s_inodes_per_group;
2083                         if (ino < sbi->s_first_ino)
2084                                 continue;
2085 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2086                         inode = ext3_iget(sb, ino);
2087 #else
2088                         inode = iget(sb, ino);
2089 #endif
2090                         if (!inode || IS_ERR(inode))
2091                                 continue;
2092
2093                         rc = add_inode_quota(inode, qctxt, oqc);
2094                         iput(inode);
2095                         if (rc) {
2096                                 brelse(bitmap_bh);
2097                                 GOTO(out, rc);
2098                         }
2099                 }
2100                 brelse(bitmap_bh);
2101         }
2102
2103         /* read old quota limits from old quota file. (only for the user
2104          * has limits but hasn't file) */
2105 #ifdef HAVE_QUOTA_SUPPORT
2106         for (i = 0; i < MAXQUOTAS; i++) {
2107                 cfs_list_t id_list;
2108                 struct dquot_id *dqid, *tmp;
2109
2110                 if (!Q_TYPESET(oqc, i))
2111                         continue;
2112
2113                 if (qctxt->qckt_first_check[i])
2114                         continue;
2115
2116
2117                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2118                 CFS_INIT_LIST_HEAD(&id_list);
2119 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2120                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2121 #else
2122                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2123 #endif
2124                 if (rc)
2125                         CERROR("read old limits failed. (rc:%d)\n", rc);
2126
2127                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2128                         cfs_list_del_init(&dqid->di_link);
2129
2130                         if (!rc)
2131                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2132                                       dqid->di_id, i,
2133                                       qctxt->qckt_first_check[i]);
2134                         OBD_FREE_PTR(dqid);
2135                 }
2136         }
2137 #endif
2138         /* turn off quota cause we are to dump chk_dqblk to files */
2139         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2140
2141         rc = create_new_quota_files(qctxt, oqc);
2142         if (rc)
2143                 GOTO(out, rc);
2144
2145         /* we use vfs functions to set dqblk, so turn quota on */
2146         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2147 out:
2148         /* dump and free chk_dqblk */
2149         rc = prune_chkquots(sb, qctxt, rc);
2150         OBD_FREE_PTR(qctxt);
2151
2152         /* turn off quota, `lfs quotacheck` will turn on when all
2153          * nodes quotacheck finish. */
2154         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2155
2156         oqc->qc_stat = rc;
2157         if (rc)
2158                 CERROR("quotacheck failed: rc = %d\n", rc);
2159
2160         RETURN(rc);
2161 }
2162
2163 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2164                                  int cmd)
2165 {
2166         int rc = 0;
2167         ENTRY;
2168
2169         if (lqi->qi_files[type] == NULL) {
2170                 CERROR("operate qinfo before it's enabled!\n");
2171                 RETURN(-ESRCH);
2172         }
2173
2174         switch (cmd) {
2175         case QFILE_CHK:
2176                 rc = lustre_check_quota_file(lqi, type);
2177                 break;
2178         case QFILE_RD_INFO:
2179                 rc = lustre_read_quota_info(lqi, type);
2180                 break;
2181         case QFILE_WR_INFO:
2182                 rc = lustre_write_quota_info(lqi, type);
2183                 break;
2184         case QFILE_INIT_INFO:
2185                 rc = lustre_init_quota_info(lqi, type);
2186                 break;
2187         case QFILE_CONVERT:
2188                 rc = -ENOTSUPP;
2189                 CERROR("quota CONVERT command is not supported\n");
2190                 break;
2191         default:
2192                 rc = -ENOTSUPP;
2193                 CERROR("Unsupported admin quota file cmd %d\n"
2194                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2195                        cmd);
2196                 break;
2197         }
2198         RETURN(rc);
2199 }
2200
2201 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2202                             cfs_list_t *list)
2203 {
2204         return lustre_get_qids(file, inode, type, list);
2205 }
2206
2207 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2208 {
2209         int rc = 0;
2210         ENTRY;
2211
2212         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2213                 CERROR("operate dquot before it's enabled!\n");
2214                 RETURN(-ESRCH);
2215         }
2216
2217         switch (cmd) {
2218         case QFILE_RD_DQUOT:
2219                 rc = lustre_read_dquot(dquot);
2220                 break;
2221         case QFILE_WR_DQUOT:
2222                 if (dquot->dq_dqb.dqb_ihardlimit ||
2223                     dquot->dq_dqb.dqb_isoftlimit ||
2224                     dquot->dq_dqb.dqb_bhardlimit ||
2225                     dquot->dq_dqb.dqb_bsoftlimit)
2226                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2227                 else
2228                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2229
2230                 rc = lustre_commit_dquot(dquot);
2231                 if (rc >= 0)
2232                         rc = 0;
2233                 break;
2234         default:
2235                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2236                 LBUG();
2237                 break;
2238         }
2239         RETURN(rc);
2240 }
2241
2242 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2243                                 struct inode *inode, int frags)
2244 {
2245 #ifdef EXT3_EXT_HAS_NO_TREE
2246         struct ext3_ext_base *base = inode;
2247 #else
2248         struct ext3_extents_tree tree;
2249         struct ext3_ext_base *base = &tree;
2250
2251         ext3_init_tree_desc(base, inode);
2252 #endif
2253         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2254          * metablocks at maxium b=16542 */
2255         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2256         return 0;
2257 }
2258
2259 #endif
2260
2261 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2262 {
2263         return (EXT3_SB(sb)->journal_bdev);
2264 }
2265 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2266
2267 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2268                           char *buf, int count, loff_t pos)
2269 {
2270         loff_t p = pos;
2271         int rc;
2272
2273         if (!f && !inode) {
2274                 CERROR("lustre_read_quota failed for no quota file!\n");
2275                 libcfs_debug_dumpstack(NULL);
2276                 return -EINVAL;
2277         }
2278
2279         /* Support for both adm and op quota files must be provided */
2280         if (f) {
2281                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2282                 rc = rc < 0 ? rc : p - pos;
2283         } else {
2284                 struct super_block *sb = inode->i_sb;
2285                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2286         }
2287         return rc;
2288 }
2289
2290 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2291 {
2292         loff_t p = pos;
2293         int rc;
2294
2295         /* Only adm quota files are supported, op updates are handled by vfs */
2296         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2297         rc = rc < 0 ? rc : p - pos;
2298
2299         return rc;
2300 }
2301
2302 void *lustre_quota_journal_start(struct inode *inode, int delete)
2303 {
2304         handle_t *handle;
2305         unsigned block_count;
2306
2307         if (delete) {
2308                 /* each indirect block (+4) may become free, attaching to the
2309                  * header list of free blocks (+1); the data block (+1) may
2310                  * become a free block (+0) or a block with free dqentries (+0) */
2311                 block_count = (4 + 1) + 1;
2312                 handle = ext3_journal_start(inode,
2313                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2314         } else {
2315                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2316                  * freeblk reusage with a header update (+1); dqentry is either reused
2317                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2318                  * a new block allocation (+1) with a header update (+1)              */
2319                 block_count = (4 + 1) + 3;
2320                 handle = ext3_journal_start(inode,
2321                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2322
2323         }
2324
2325         return handle;
2326 }
2327
2328 void lustre_quota_journal_stop(void *handle)
2329 {
2330         ext3_journal_stop((handle_t *)handle);
2331 }
2332
2333 static int ll_decode_fh_accept(void *context, struct dentry *de)
2334 {
2335         return 1;
2336 }
2337
2338 #ifdef HAVE_EXPORTFS_DECODE_FH
2339 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2340          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2341                             acceptable, context)
2342 #else
2343 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2344          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2345                                      type, acceptable, context)
2346 # define FILEID_INO32_GEN 1
2347 extern struct export_operations export_op_default;
2348 #endif
2349
2350 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2351                                       struct fsfilt_fid *fid, int ignore_gen)
2352 {
2353         struct inode  *inode;
2354         struct dentry *result;
2355
2356         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2357                                        ll_decode_fh_accept, NULL);
2358         if (IS_ERR(result)) {
2359                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2360                        fid->ino, fid->gen, PTR_ERR(result));
2361                 return result;
2362         }
2363
2364         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2365                fid->ino, fid->gen);
2366         inode = result->d_inode;
2367         if (inode == NULL)
2368                 goto err_out;
2369
2370         if (inode->i_nlink == 0 &&
2371             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2372                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2373                               " ctime -- this may indicate disk "
2374                               "corruption (inode: %lu, link: %lu, "
2375                               "count: %d)\n", inode->i_ino,
2376                               (unsigned long)inode->i_nlink,
2377                               atomic_read(&inode->i_count));
2378                 goto err_out;
2379         }
2380         if (fid->gen && inode->i_generation != fid->gen) {
2381                 /* we didn't find the right inode.. */
2382                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2383                        "count: %d, generation %u/%u\n",
2384                        inode->i_ino, (unsigned long)inode->i_nlink,
2385                        atomic_read(&inode->i_count), inode->i_generation,
2386                        fid->gen);
2387                 goto err_out;
2388         }
2389
2390         return result;
2391 err_out:
2392         l_dput(result);
2393         return ERR_PTR(-ENOENT);
2394 }
2395
2396 static struct fsfilt_operations fsfilt_ext3_ops = {
2397         .fs_type                = "ext3",
2398         .fs_owner               = THIS_MODULE,
2399         .fs_getlabel            = fsfilt_ext3_get_label,
2400         .fs_setlabel            = fsfilt_ext3_set_label,
2401         .fs_uuid                = fsfilt_ext3_uuid,
2402         .fs_start               = fsfilt_ext3_start,
2403         .fs_brw_start           = fsfilt_ext3_brw_start,
2404         .fs_extend              = fsfilt_ext3_extend,
2405         .fs_commit              = fsfilt_ext3_commit,
2406         .fs_commit_async        = fsfilt_ext3_commit_async,
2407         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2408         .fs_setattr             = fsfilt_ext3_setattr,
2409         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2410         .fs_set_md              = fsfilt_ext3_set_md,
2411         .fs_get_md              = fsfilt_ext3_get_md,
2412         .fs_readpage            = fsfilt_ext3_readpage,
2413         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2414         .fs_statfs              = fsfilt_ext3_statfs,
2415         .fs_sync                = fsfilt_ext3_sync,
2416         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2417         .fs_write_record        = fsfilt_ext3_write_record,
2418         .fs_read_record         = fsfilt_ext3_read_record,
2419         .fs_setup               = fsfilt_ext3_setup,
2420         .fs_send_bio            = fsfilt_ext3_send_bio,
2421         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2422 #ifdef HAVE_DISK_INODE_VERSION
2423         .fs_get_version         = fsfilt_ext3_get_version,
2424         .fs_set_version         = fsfilt_ext3_set_version,
2425 #endif
2426 #ifdef HAVE_QUOTA_SUPPORT
2427         .fs_quotactl            = fsfilt_ext3_quotactl,
2428         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2429         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2430         .fs_qids                = fsfilt_ext3_qids,
2431         .fs_dquot               = fsfilt_ext3_dquot,
2432         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2433 #endif
2434         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2435         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2436 };
2437
2438 static int __init fsfilt_ext3_init(void)
2439 {
2440         int rc;
2441
2442         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2443                                          sizeof(struct fsfilt_cb_data), 0, 0);
2444         if (!fcb_cache) {
2445                 CERROR("error allocating fsfilt journal callback cache\n");
2446                 GOTO(out, rc = -ENOMEM);
2447         }
2448
2449         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2450
2451         if (rc) {
2452                 int err = cfs_mem_cache_destroy(fcb_cache);
2453                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2454         }
2455 out:
2456         return rc;
2457 }
2458
2459 static void __exit fsfilt_ext3_exit(void)
2460 {
2461         int rc;
2462
2463         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2464         rc = cfs_mem_cache_destroy(fcb_cache);
2465         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2466 }
2467
2468 module_init(fsfilt_ext3_init);
2469 module_exit(fsfilt_ext3_exit);
2470
2471 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2472 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2473 MODULE_LICENSE("GPL");