lustre/lvfs/fsfilt_ext3.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/fsfilt_ext3.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/init.h>
  44 #include <linux/module.h>
  45 #include <linux/fs.h>
  46 #include <linux/slab.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/quotaops.h>
  49 #ifdef HAVE_LINUX_EXPORTFS_H
  50 #include <linux/exportfs.h>
  51 #endif
  52 #ifdef HAVE_EXT4_LDISKFS
  53 #include <ext4/ext4.h>
  54 #include <ext4/ext4_jbd2.h>
  55 #else
  56 #include <linux/jbd.h>
  57 #include <linux/ext3_fs.h>
  58 #include <linux/ext3_jbd.h>
  59 #endif
  60 #include <linux/version.h>
  61 #include <linux/bitops.h>
  62 #include <linux/quota.h>
  63 #ifdef HAVE_QUOTAIO_H
  64 # include <linux/quotaio_v2.h>
  65 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  66 # include <quota/quotaio_v2.h>
  67 # include <quota/quota_tree.h>
  68 # define V2_DQTREEOFF    QT_TREEOFF
  69 #elif defined(HAVE_FS_QUOTAIO_H)
  70 # include <quotaio_v2.h>
  71 # include <quota_tree.h>
  72 # define V2_DQTREEOFF    QT_TREEOFF
  73 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  74 #endif
  75
  76 #ifdef QFMT_VFS_V1
  77 #define QFMT_LUSTRE QFMT_VFS_V1
  78 #else
  79 #define QFMT_LUSTRE QFMT_VFS_V0
  80 #endif
  81
  82 #if defined(HAVE_EXT3_XATTR_H)
  83 # include <ext3/xattr.h>
  84 #elif !defined(EXT3_XATTR_INDEX_TRUSTED)
  85 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  86 /* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
  87 # define EXT3_XATTR_INDEX_TRUSTED        4
  88 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  89 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  90 #endif
  91
  92 #include <libcfs/libcfs.h>
  93 #include <lustre_fsfilt.h>
  94 #include <obd.h>
  95 #include <lustre_quota.h>
  96 #include <linux/lustre_compat25.h>
  97 #include <linux/lprocfs_status.h>
  98
  99 #ifdef HAVE_EXT4_LDISKFS
 100 #include <ext4/ext4_extents.h>
 101 #else
 102 #include <linux/ext3_extents.h>
 103 #endif
 104
 105 #include "lustre_quota_fmt.h"
 106
 107 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
 108 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 109 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 110 #else
 111 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 112 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 113 #endif
 114
 115 /* for kernels 2.6.18 and later */
 116 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 117
 118 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 119                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 120
 121 #define ext3_mb_discard_inode_preallocations(inode) \
 122                  ext3_discard_preallocations(inode)
 123
 124 #ifdef HAVE_EXT4_LDISKFS
 125 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 126 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 127 #else
 128 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 129 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 130 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 131 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 132 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 133 #endif
 134
 135 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 136 # define journal_callback ext4_journal_cb_entry
 137 # define fsfilt_journal_callback_set(handle, func, jcb) \
 138          ext4_journal_callback_add(handle, func, jcb)
 139 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
 140 # define fsfilt_journal_callback_set(handle, func, jcb) \
 141          jbd2_journal_callback_set(handle, func, jcb)
 142 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
 143 # define fsfilt_journal_callback_set(handle, func, jcb) \
 144          journal_callback_set(handle, func, jcb)
 145 #else
 146 # error missing journal commit callback
 147 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
 148
 149 static cfs_mem_cache_t *fcb_cache;
 150
 151 struct fsfilt_cb_data {
 152         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 153         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 154         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 155         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 156         void *cb_data;                  /* MDS/OST completion function data */
 157 };
 158
 159 #ifndef ext3_find_next_bit
 160 #define ext3_find_next_bit           ext2_find_next_bit
 161 #endif
 162
 163 #ifndef ext2_find_next_bit
 164 #ifdef __LITTLE_ENDIAN
 165 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 166 #else
 167 error "Need implementation of find_next_bit on big-endian systems"
 168 #endif  /* __LITTLE_ENDIAN */
 169 #endif  /* !ext2_find_next_le_bit */
 170
 171 static char *fsfilt_ext3_get_label(struct super_block *sb)
 172 {
 173         return EXT3_SB(sb)->s_es->s_volume_name;
 174 }
 175
 176 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 177 {
 178         /* see e.g. fsfilt_ext3_write_record() */
 179         journal_t *journal;
 180         handle_t *handle;
 181         int err;
 182
 183         journal = EXT3_SB(sb)->s_journal;
 184         handle = ext3_journal_start_sb(sb, 1);
 185         if (IS_ERR(handle)) {
 186                 CERROR("can't start transaction\n");
 187                 return(PTR_ERR(handle));
 188         }
 189
 190         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 191         if (err)
 192                 goto out;
 193
 194         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 195                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 196
 197         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 198
 199 out:
 200         ext3_journal_stop(handle);
 201
 202         return(err);
 203 }
 204
 205 static char *fsfilt_ext3_uuid(struct super_block *sb)
 206 {
 207         return EXT3_SB(sb)->s_es->s_uuid;
 208 }
 209
 210 #ifdef HAVE_DISK_INODE_VERSION
 211
 212 static __u64 get_i_version(struct inode *inode)
 213 {
 214         return EXT3_I(inode)->i_fs_version;
 215 }
 216
 217 static void set_i_version(struct inode *inode, __u64 new_version)
 218 {
 219         (EXT3_I(inode))->i_fs_version = new_version;
 220 }
 221
 222 /*
 223  * Get the 64-bit version for an inode.
 224  */
 225 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 226 {
 227         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 228                get_i_version(inode), inode->i_ino);
 229         return get_i_version(inode);
 230 }
 231
 232 /*
 233  * Set the 64-bit version and return the old version.
 234  */
 235 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 236 {
 237         __u64 old_version = get_i_version(inode);
 238
 239         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 240                new_version, old_version, inode->i_ino);
 241         set_i_version(inode, new_version);
 242         /* version is set after all inode operations are finished, so we should
 243          * mark it dirty here */
 244         inode->i_sb->s_op->dirty_inode(inode);
 245         return old_version;
 246 }
 247
 248 #endif
 249
 250 /*
 251  * We don't currently need any additional blocks for rmdir and
 252  * unlink transactions because we are storing the OST oa_id inside
 253  * the inode (which we will be changing anyways as part of this
 254  * transaction).
 255  */
 256 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 257                                int logs)
 258 {
 259         /* For updates to the last received file */
 260         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 261         journal_t *journal;
 262         void *handle;
 263
 264         if (current->journal_info) {
 265                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 266                        current->journal_info);
 267                 goto journal_start;
 268         }
 269
 270         switch(op) {
 271         case FSFILT_OP_RMDIR:
 272         case FSFILT_OP_UNLINK:
 273                 /* delete one file + create/update logs for each stripe */
 274                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 275                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 276                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 277                 break;
 278         case FSFILT_OP_RENAME:
 279                 /* modify additional directory */
 280                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 281                 /* no break */
 282         case FSFILT_OP_SYMLINK:
 283                 /* additional block + block bitmap + GDT for long symlink */
 284                 nblocks += 3;
 285                 /* no break */
 286         case FSFILT_OP_CREATE: {
 287 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 288                 static int warned;
 289                 if (!warned) {
 290                         if (!test_opt(inode->i_sb, EXTENTS)) {
 291                                 warned = 1;
 292                         } else if (((EXT3_I(inode)->i_flags &
 293                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 294                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 295                                 CWARN("extent-mapped directory found with "
 296                                       "ext3-based ldiskfs - contact "
 297                                       "http://bugs.whamcloud.com/\n");
 298                                 warned = 1;
 299                         }
 300                 }
 301 #endif
 302                 /* no break */
 303         }
 304         case FSFILT_OP_MKDIR:
 305         case FSFILT_OP_MKNOD:
 306                 /* modify one inode + block bitmap + GDT */
 307                 nblocks += 3;
 308                 /* no break */
 309         case FSFILT_OP_LINK:
 310                 /* modify parent directory */
 311                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 312                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 313                 /* create/update logs for each stripe */
 314                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 315                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 316                 break;
 317         case FSFILT_OP_SETATTR:
 318                 /* Setattr on inode */
 319                 nblocks += 1;
 320                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 321                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 322                 /* quota chown log for each stripe */
 323                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 324                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 325                 break;
 326         case FSFILT_OP_CANCEL_UNLINK:
 327                 /* blocks for log header bitmap update OR
 328                  * blocks for catalog header bitmap update + unlink of logs */
 329                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 330                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 331                 break;
 332         default: CERROR("unknown transaction start op %d\n", op);
 333                 LBUG();
 334         }
 335
 336         LASSERT(current->journal_info == desc_private);
 337         journal = EXT3_SB(inode->i_sb)->s_journal;
 338         if (nblocks > journal->j_max_transaction_buffers) {
 339                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 340                        nblocks, op, logs, journal->j_max_transaction_buffers);
 341                 nblocks = journal->j_max_transaction_buffers;
 342         }
 343
 344  journal_start:
 345         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 346         handle = ext3_journal_start(inode, nblocks);
 347
 348         if (!IS_ERR(handle))
 349                 LASSERT(current->journal_info == handle);
 350         else
 351                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 352                        op, nblocks, PTR_ERR(handle));
 353         return handle;
 354 }
 355
 356 /*
 357  * Calculate the number of buffer credits needed to write multiple pages in
 358  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 359  * doesn't have a nice API for calculating this sort of thing in advance.
 360  *
 361  * See comment above ext3_writepage_trans_blocks for details.  We assume
 362  * no data journaling is being done, but it does allow for all of the pages
 363  * being non-contiguous.  If we are guaranteed contiguous pages we could
 364  * reduce the number of (d)indirect blocks a lot.
 365  *
 366  * With N blocks per page and P pages, for each inode we have at most:
 367  * N*P indirect
 368  * min(N*P, blocksize/4 + 1) dindirect blocks
 369  * niocount tindirect
 370  *
 371  * For the entire filesystem, we have at most:
 372  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 373  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 374  * objcount inode blocks
 375  * 1 superblock
 376  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 377  *
 378  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 379  */
 380 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 381                                       int niocount, struct niobuf_local *nb)
 382 {
 383         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 384         __u64 next_indir;
 385         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 386         int nbitmaps = 0, ngdblocks;
 387         int needed = objcount + 1; /* inodes + superblock */
 388         int i, j;
 389
 390         for (i = 0, j = 0; i < objcount; i++, fso++) {
 391                 /* two or more dindirect blocks in case we cross boundary */
 392                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 393                                     nb[j].offset) >>
 394                                    sb->s_blocksize_bits) /
 395                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 396                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 397
 398                 /* leaf, indirect, tindirect blocks for first block */
 399                 nbitmaps += blockpp + 2;
 400
 401                 j += fso->fso_bufcnt;
 402         }
 403
 404         next_indir = nb[0].offset +
 405                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 406         for (i = 1; i < niocount; i++) {
 407                 if (nb[i].offset >= next_indir) {
 408                         nbitmaps++;     /* additional indirect */
 409                         next_indir = nb[i].offset +
 410                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 411                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 412                         nbitmaps++;     /* additional indirect */
 413                 }
 414                 nbitmaps += blockpp;    /* each leaf in different group? */
 415         }
 416
 417         ngdblocks = nbitmaps;
 418         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 419                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 420         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 421                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 422
 423         needed += nbitmaps + ngdblocks;
 424
 425         /* last_rcvd update */
 426         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 427
 428 #if defined(CONFIG_QUOTA)
 429         /* We assume that there will be 1 bit set in s_dquot.flags for each
 430          * quota file that is active.  This is at least true for now.
 431          */
 432         needed += hweight32(ll_sb_any_quota_active(sb)) *
 433                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 434 #endif
 435
 436         return needed;
 437 }
 438
 439 /* We have to start a huge journal transaction here to hold all of the
 440  * metadata for the pages being written here.  This is necessitated by
 441  * the fact that we do lots of prepare_write operations before we do
 442  * any of the matching commit_write operations, so even if we split
 443  * up to use "smaller" transactions none of them could complete until
 444  * all of them were opened.  By having a single journal transaction,
 445  * we eliminate duplicate reservations for common blocks like the
 446  * superblock and group descriptors or bitmaps.
 447  *
 448  * We will start the transaction here, but each prepare_write will
 449  * add a refcount to the transaction, and each commit_write will
 450  * remove a refcount.  The transaction will be closed when all of
 451  * the pages have been written.
 452  */
 453 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 454                                    int niocount, struct niobuf_local *nb,
 455                                    void *desc_private, int logs)
 456 {
 457         journal_t *journal;
 458         handle_t *handle;
 459         int needed;
 460         ENTRY;
 461
 462         LASSERT(current->journal_info == desc_private);
 463         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 464         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 465
 466         /* The number of blocks we could _possibly_ dirty can very large.
 467          * We reduce our request if it is absurd (and we couldn't get that
 468          * many credits for a single handle anyways).
 469          *
 470          * At some point we have to limit the size of I/Os sent at one time,
 471          * increase the size of the journal, or we have to calculate the
 472          * actual journal requirements more carefully by checking all of
 473          * the blocks instead of being maximally pessimistic.  It remains to
 474          * be seen if this is a real problem or not.
 475          */
 476         if (needed > journal->j_max_transaction_buffers) {
 477                 CERROR("want too many journal credits (%d) using %d instead\n",
 478                        needed, journal->j_max_transaction_buffers);
 479                 needed = journal->j_max_transaction_buffers;
 480         }
 481
 482         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 483         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 484         if (IS_ERR(handle)) {
 485                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 486                        PTR_ERR(handle));
 487         } else {
 488                 LASSERT(handle->h_buffer_credits >= needed);
 489                 LASSERT(current->journal_info == handle);
 490         }
 491
 492         RETURN(handle);
 493 }
 494
 495 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 496 {
 497        handle_t *handle = h;
 498
 499        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 500        if (nblocks == 0) {
 501                handle->h_buffer_credits = 0;
 502                CWARN("setting credits of handle %p to zero by request\n", h);
 503        }
 504
 505        if (handle->h_buffer_credits > nblocks)
 506                 return 0;
 507        if (ext3_journal_extend(handle, nblocks) == 0)
 508                 return 0;
 509
 510        ext3_mark_inode_dirty(handle, inode);
 511        return ext3_journal_restart(handle, nblocks);
 512 }
 513
 514 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 515 {
 516         int rc;
 517         handle_t *handle = h;
 518
 519         LASSERT(current->journal_info == handle);
 520         if (force_sync)
 521                 handle->h_sync = 1; /* recovery likes this */
 522
 523         rc = ext3_journal_stop(handle);
 524
 525         return rc;
 526 }
 527
 528 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 529                                     void **wait_handle)
 530 {
 531         unsigned long tid;
 532         transaction_t *transaction;
 533         handle_t *handle = h;
 534         journal_t *journal;
 535         int rc;
 536
 537         LASSERT(current->journal_info == handle);
 538
 539         transaction = handle->h_transaction;
 540         journal = transaction->t_journal;
 541         tid = transaction->t_tid;
 542         /* we don't want to be blocked */
 543         handle->h_sync = 0;
 544         rc = ext3_journal_stop(handle);
 545         if (rc) {
 546                 CERROR("error while stopping transaction: %d\n", rc);
 547                 return rc;
 548         }
 549         fsfilt_log_start_commit(journal, tid);
 550
 551         *wait_handle = (void *) tid;
 552         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 553         return 0;
 554 }
 555
 556 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 557 {
 558         journal_t *journal = EXT3_JOURNAL(inode);
 559         tid_t tid = (tid_t)(long)h;
 560
 561         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 562         if (unlikely(is_journal_aborted(journal)))
 563                 return -EIO;
 564
 565         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 566
 567         if (unlikely(is_journal_aborted(journal)))
 568                 return -EIO;
 569         return 0;
 570 }
 571
 572 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 573                                struct iattr *iattr, int do_trunc)
 574 {
 575         struct inode *inode = dentry->d_inode;
 576         int rc = 0;
 577
 578 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
 579         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
 580          * timestamps to appear to be in the far future, due old timestamp
 581          * being stored on disk as an unsigned value.  This fixes up any
 582          * bad values held by the client before storing them on disk,
 583          * and ensures any timestamp updates are correct.  LU-1042 */
 584         if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
 585                      !(iattr->ia_valid & ATTR_ATIME))) {
 586                 iattr->ia_valid |= ATTR_ATIME;
 587                 LTIME_S(iattr->ia_atime) = 0;
 588         }
 589         if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
 590                      !(iattr->ia_valid & ATTR_MTIME))) {
 591                 iattr->ia_valid |= ATTR_MTIME;
 592                 LTIME_S(iattr->ia_mtime) = 0;
 593         }
 594         if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
 595                       LTIME_S(inode->i_ctime) == 0) &&
 596                      !(iattr->ia_valid & ATTR_CTIME))) {
 597                 iattr->ia_valid |= ATTR_CTIME;
 598                 LTIME_S(iattr->ia_ctime) = 0;
 599         }
 600 #else
 601 #warning "remove old LU-221/LU-1042 workaround code"
 602 #endif
 603
 604         /* When initializating timestamps for new inodes, use the filesystem
 605          * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
 606          * that this is potentially an invalid inode.  Files with an old ctime
 607          * migrated to a newly-formatted OST with a newer s_mkfs_time will not
 608          * hit this check, since it is only for ctime == 0.  LU-1010/LU-1042 */
 609         if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
 610                 LTIME_S(iattr->ia_ctime) =
 611                         EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
 612
 613         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 614          * We are already writing the inode to disk as part of this
 615          * transaction and want to avoid a lot of extra inode writeout
 616          * later on. b=9828 */
 617         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 618                 /* ATTR_SIZE would invoke truncate: clear it */
 619                 iattr->ia_valid &= ~ATTR_SIZE;
 620                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 621                 i_size_write(inode, iattr->ia_size);
 622
 623                 if (iattr->ia_valid & ATTR_UID)
 624                         inode->i_uid = iattr->ia_uid;
 625                 if (iattr->ia_valid & ATTR_GID)
 626                         inode->i_gid = iattr->ia_gid;
 627                 if (iattr->ia_valid & ATTR_ATIME)
 628                         inode->i_atime = iattr->ia_atime;
 629                 if (iattr->ia_valid & ATTR_MTIME)
 630                         inode->i_mtime = iattr->ia_mtime;
 631                 if (iattr->ia_valid & ATTR_CTIME)
 632                         inode->i_ctime = iattr->ia_ctime;
 633                 if (iattr->ia_valid & ATTR_MODE) {
 634                         inode->i_mode = iattr->ia_mode;
 635
 636                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 637                             !cfs_capable(CFS_CAP_FSETID))
 638                                 inode->i_mode &= ~S_ISGID;
 639                 }
 640
 641                 inode->i_sb->s_op->dirty_inode(inode);
 642
 643                 goto out;
 644         }
 645
 646         /* Don't allow setattr to change file type */
 647         if (iattr->ia_valid & ATTR_MODE)
 648                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 649                                  (iattr->ia_mode & ~S_IFMT);
 650
 651         /* We set these flags on the client, but have already checked perms
 652          * so don't confuse inode_change_ok. */
 653         iattr->ia_valid &= ~TIMES_SET_FLAGS;
 654
 655         if (inode->i_op->setattr) {
 656                 rc = inode->i_op->setattr(dentry, iattr);
 657         } else {
 658 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
 659                 rc = inode_change_ok(inode, iattr);
 660                 if (!rc)
 661 #endif
 662                         rc = simple_setattr(dentry, iattr);
 663         }
 664
 665  out:
 666         RETURN(rc);
 667 }
 668
 669 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 670                                  unsigned int cmd, unsigned long arg)
 671 {
 672         int rc = 0;
 673         ENTRY;
 674
 675         /* FIXME: Can't do this because of nested transaction deadlock */
 676         if (cmd == EXT3_IOC_SETFLAGS) {
 677                 /* We can't enable data journaling on OST objects, because
 678                 * this forces the transaction to be closed in order to
 679                 * flush the journal, but the caller will already have a
 680                 * compound transaction open to update the last_rcvd file,
 681                 * and this thread would deadlock trying to set the flag. */
 682                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 683                         CERROR("can't set data journal flag on file\n");
 684                         RETURN(-EPERM);
 685                 }
 686                 /* Because the MDS does not see the EXTENTS_FL set on the
 687                  * OST objects, mask this flag into all set flags.  It is
 688                  * not legal to clear this flag in any case, so we are not
 689                  * changing the functionality by doing this.  b=22911 */
 690                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 691         }
 692
 693 #ifdef HAVE_EXT4_LDISKFS
 694         /* ext4_ioctl does not have a inode argument */
 695         if (inode->i_fop->unlocked_ioctl)
 696                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 697 #else
 698         if (inode->i_fop->ioctl)
 699                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 700 #endif
 701         else
 702                 RETURN(-ENOTTY);
 703
 704         RETURN(rc);
 705 }
 706
 707 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 708                               void *lmm, int lmm_size, const char *name)
 709 {
 710         int rc;
 711
 712         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 713
 714         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 715                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 716
 717
 718         if (rc && rc != -EROFS)
 719                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 720                        inode->i_ino, rc);
 721         return rc;
 722 }
 723
 724 /* Must be called with i_mutex held */
 725 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 726                               const char *name)
 727 {
 728         int rc;
 729
 730         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 731
 732         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 733                             name, lmm, lmm_size);
 734
 735         /* This gives us the MD size */
 736         if (lmm == NULL)
 737                 return (rc == -ENODATA) ? 0 : rc;
 738
 739         if (rc < 0) {
 740                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 741                        EXT3_XATTR_INDEX_TRUSTED, name,
 742                        inode->i_ino, rc);
 743                 memset(lmm, 0, lmm_size);
 744                 return (rc == -ENODATA) ? 0 : rc;
 745         }
 746
 747         return rc;
 748 }
 749
 750 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 751 {
 752         submit_bio(rw, bio);
 753         return 0;
 754 }
 755
 756 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 757                                     loff_t *off)
 758 {
 759         struct inode *inode = file->f_dentry->d_inode;
 760         int rc = 0;
 761
 762         if (S_ISREG(inode->i_mode))
 763                 rc = file->f_op->read(file, buf, count, off);
 764         else {
 765                 const int blkbits = inode->i_sb->s_blocksize_bits;
 766                 const int blksize = inode->i_sb->s_blocksize;
 767
 768                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 769                        (unsigned long)count, inode->i_ino, *off);
 770                 while (count > 0) {
 771                         struct buffer_head *bh;
 772
 773                         bh = NULL;
 774                         if (*off < i_size_read(inode)) {
 775                                 int err = 0;
 776
 777                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 778                                                 0, &err);
 779
 780                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 781
 782                                 if (bh) {
 783                                         memcpy(buf, bh->b_data, blksize);
 784                                         brelse(bh);
 785                                 } else if (err) {
 786                                         /* XXX in theory we should just fake
 787                                          * this buffer and continue like ext3,
 788                                          * especially if this is a partial read
 789                                          */
 790                                         CERROR("error read dir %lu+%llu: %d\n",
 791                                                inode->i_ino, *off, err);
 792                                         RETURN(err);
 793                                 }
 794                         }
 795                         if (!bh) {
 796                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 797
 798                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 799                                 memset(fake, 0, sizeof(*fake));
 800                                 fake->rec_len = cpu_to_le16(blksize);
 801                         }
 802                         count -= blksize;
 803                         buf += blksize;
 804                         *off += blksize;
 805                         rc += blksize;
 806                 }
 807         }
 808
 809         return rc;
 810 }
 811
 812 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 813 static void fsfilt_ext3_cb_func(struct super_block *sb,
 814                                 struct journal_callback *jcb, int error)
 815 #else
 816 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 817 #endif
 818 {
 819         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 820
 821         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 822
 823         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 824 }
 825
 826 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 827                                       void *handle, fsfilt_cb_t cb_func,
 828                                       void *cb_data)
 829 {
 830         struct fsfilt_cb_data *fcb;
 831
 832         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 833         if (fcb == NULL)
 834                 RETURN(-ENOMEM);
 835
 836         fcb->cb_func = cb_func;
 837         fcb->cb_obd = obd;
 838         fcb->cb_last_rcvd = last_rcvd;
 839         fcb->cb_data = cb_data;
 840
 841         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 842         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 843
 844         return 0;
 845 }
 846
 847 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 848 {
 849         struct kstatfs sfs;
 850         int rc;
 851
 852         memset(&sfs, 0, sizeof(sfs));
 853         rc = ll_do_statfs(sb, &sfs);
 854         statfs_pack(osfs, &sfs);
 855         return rc;
 856 }
 857
 858 static int fsfilt_ext3_sync(struct super_block *sb)
 859 {
 860         return ext3_force_commit(sb);
 861 }
 862
 863 #ifndef EXT3_EXTENTS_FL
 864 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 865 #endif
 866
 867 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 868 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 869 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 870 #else
 871 # ifdef HAVE_EXT4_LDISKFS
 872 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 873 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 874 # else
 875 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 876 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 877 # endif
 878 #endif
 879
 880 #ifndef EXT_ASSERT
 881 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 882 #endif
 883
 884 #ifdef EXT3_EXT_HAS_NO_TREE
 885 /* for kernels 2.6.18 and later */
 886 #ifdef HAVE_EXT4_LDISKFS
 887 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 888 #else
 889 #define EXT_GENERATION(inode)           ext_generation(inode)
 890 #endif
 891 #define ext3_ext_base                   inode
 892 #define ext3_ext_base2inode(inode)      (inode)
 893 #define EXT_DEPTH(inode)                ext_depth(inode)
 894 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 895                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 896 #else
 897 #define ext3_ext_base                   ext3_extents_tree
 898 #define ext3_ext_base2inode(tree)       (tree->inode)
 899 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 900                         ext3_ext_walk_space(tree, block, num, cb);
 901 #endif
 902
 903 struct bpointers {
 904         unsigned long *blocks;
 905         int *created;
 906         unsigned long start;
 907         int num;
 908         int init_num;
 909         int create;
 910 };
 911
 912 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 913                                unsigned long block, int *aflags)
 914 {
 915         struct ext3_inode_info *ei = EXT3_I(inode);
 916         unsigned long bg_start;
 917         unsigned long colour;
 918         int depth;
 919
 920         if (path) {
 921                 struct ext3_extent *ex;
 922                 depth = path->p_depth;
 923
 924                 /* try to predict block placement */
 925                 if ((ex = path[depth].p_ext))
 926                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 927
 928                 /* it looks index is empty
 929                  * try to find starting from index itself */
 930                 if (path[depth].p_bh)
 931                         return path[depth].p_bh->b_blocknr;
 932         }
 933
 934         /* OK. use inode's group */
 935         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 936                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 937         colour = (current->pid % 16) *
 938                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 939         return bg_start + colour + block;
 940 }
 941
 942 #define ll_unmap_underlying_metadata(sb, blocknr) \
 943         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 944
 945 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 946 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 947                                 struct ext3_ext_path *path, unsigned long block,
 948                                 unsigned long *count, int *err)
 949 {
 950         unsigned long pblock, goal;
 951         int aflags = 0;
 952         struct inode *inode = ext3_ext_base2inode(base);
 953
 954         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 955         aflags |= 2; /* block have been already reserved */
 956         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 957         return pblock;
 958
 959 }
 960 #else
 961 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 962                                 struct ext3_ext_path *path, unsigned long block,
 963                                 unsigned long *count, int *err)
 964 {
 965         struct inode *inode = ext3_ext_base2inode(base);
 966         struct ext3_allocation_request ar;
 967         unsigned long pblock;
 968         int aflags;
 969
 970         /* find neighbour allocated blocks */
 971         ar.lleft = block;
 972         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 973         if (*err)
 974                 return 0;
 975         ar.lright = block;
 976         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 977         if (*err)
 978                 return 0;
 979
 980         /* allocate new block */
 981         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 982         ar.inode = inode;
 983         ar.logical = block;
 984         ar.len = *count;
 985         ar.flags = EXT3_MB_HINT_DATA;
 986         pblock = ext3_mb_new_blocks(handle, &ar, err);
 987         *count = ar.len;
 988         return pblock;
 989 }
 990 #endif
 991
 992 #ifdef EXT3_EXT_HAS_NO_TREE
 993 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 994                                   struct ext3_ext_path *path,
 995                                   struct ext3_ext_cache *cex,
 996 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 997                                    struct ext3_extent *ex,
 998 #endif
 999                                   void *cbdata)
1000 {
1001         struct bpointers *bp = cbdata;
1002 #else
1003 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1004                                   struct ext3_ext_path *path,
1005                                   struct ext3_ext_cache *cex
1006 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1007                                   , struct ext3_extent *ex
1008 #endif
1009                                  )
1010 {
1011         struct bpointers *bp = base->private;
1012 #endif
1013         struct inode *inode = ext3_ext_base2inode(base);
1014         struct ext3_extent nex;
1015         unsigned long pblock;
1016         unsigned long tgen;
1017         int err, i;
1018         unsigned long count;
1019         handle_t *handle;
1020
1021         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1022                 err = EXT_CONTINUE;
1023                 goto map;
1024         }
1025
1026         if (bp->create == 0) {
1027                 i = 0;
1028                 if (cex->ec_block < bp->start)
1029                         i = bp->start - cex->ec_block;
1030                 if (i >= cex->ec_len)
1031                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1032                                         i, cex->ec_len);
1033                 for (; i < cex->ec_len && bp->num; i++) {
1034                         *(bp->created) = 0;
1035                         bp->created++;
1036                         *(bp->blocks) = 0;
1037                         bp->blocks++;
1038                         bp->num--;
1039                         bp->start++;
1040                 }
1041
1042                 return EXT_CONTINUE;
1043         }
1044
1045         tgen = EXT_GENERATION(base);
1046         count = ext3_ext_calc_credits_for_insert(base, path);
1047         fsfilt_up_truncate_sem(inode);
1048
1049         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1050         if (IS_ERR(handle)) {
1051                 fsfilt_down_truncate_sem(inode);
1052                 return PTR_ERR(handle);
1053         }
1054
1055         fsfilt_down_truncate_sem(inode);
1056         if (tgen != EXT_GENERATION(base)) {
1057                 /* the tree has changed. so path can be invalid at moment */
1058                 ext3_journal_stop(handle);
1059                 return EXT_REPEAT;
1060         }
1061
1062 #if defined(HAVE_EXT4_LDISKFS)
1063         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1064          * protected by i_data_sem as whole. so we patch it to store
1065          * generation to path and now verify the tree hasn't changed */
1066         down_write((&EXT4_I(inode)->i_data_sem));
1067
1068         /* validate extent, make sure the extent tree does not changed */
1069         if (EXT_GENERATION(base) != path[0].p_generation) {
1070                 /* cex is invalid, try again */
1071                 up_write(&EXT4_I(inode)->i_data_sem);
1072                 ext3_journal_stop(handle);
1073                 return EXT_REPEAT;
1074         }
1075 #endif
1076
1077         count = cex->ec_len;
1078         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1079         if (!pblock)
1080                 goto out;
1081         EXT_ASSERT(count <= cex->ec_len);
1082
1083         /* insert new extent */
1084         nex.ee_block = cpu_to_le32(cex->ec_block);
1085         ext3_ext_store_pblock(&nex, pblock);
1086         nex.ee_len = cpu_to_le16(count);
1087         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1088         if (err) {
1089                 /* free data blocks we just allocated */
1090                 /* not a good idea to call discard here directly,
1091                  * but otherwise we'd need to call it every free() */
1092 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1093                 ext3_mb_discard_inode_preallocations(inode);
1094 #endif
1095                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1096                                  cpu_to_le16(nex.ee_len), 0);
1097                 goto out;
1098         }
1099
1100         /*
1101          * Putting len of the actual extent we just inserted,
1102          * we are asking ext3_ext_walk_space() to continue
1103          * scaning after that block
1104          */
1105         cex->ec_len = le16_to_cpu(nex.ee_len);
1106         cex->ec_start = ext_pblock(&nex);
1107         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1108         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1109
1110 out:
1111 #if defined(HAVE_EXT4_LDISKFS)
1112         up_write((&EXT4_I(inode)->i_data_sem));
1113 #endif
1114         ext3_journal_stop(handle);
1115 map:
1116         if (err >= 0) {
1117                 /* map blocks */
1118                 if (bp->num == 0) {
1119                         CERROR("hmm. why do we find this extent?\n");
1120                         CERROR("initial space: %lu:%u\n",
1121                                 bp->start, bp->init_num);
1122                         CERROR("current extent: %u/%u/%llu %d\n",
1123                                 cex->ec_block, cex->ec_len,
1124                                 (unsigned long long)cex->ec_start,
1125                                 cex->ec_type);
1126                 }
1127                 i = 0;
1128                 if (cex->ec_block < bp->start)
1129                         i = bp->start - cex->ec_block;
1130                 if (i >= cex->ec_len)
1131                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1132                                         i, cex->ec_len);
1133                 for (; i < cex->ec_len && bp->num; i++) {
1134                         *(bp->blocks) = cex->ec_start + i;
1135                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1136                                 *(bp->created) = 0;
1137                         } else {
1138                                 *(bp->created) = 1;
1139                                 /* unmap any possible underlying metadata from
1140                                  * the block device mapping.  bug 6998. */
1141                                 ll_unmap_underlying_metadata(inode->i_sb,
1142                                                              *(bp->blocks));
1143                         }
1144                         bp->created++;
1145                         bp->blocks++;
1146                         bp->num--;
1147                         bp->start++;
1148                 }
1149         }
1150         return err;
1151 }
1152
1153 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1154                        unsigned long num, unsigned long *blocks,
1155                        int *created, int create)
1156 {
1157 #ifdef EXT3_EXT_HAS_NO_TREE
1158         struct ext3_ext_base *base = inode;
1159 #else
1160         struct ext3_extents_tree tree;
1161         struct ext3_ext_base *base = &tree;
1162 #endif
1163         struct bpointers bp;
1164         int err;
1165
1166         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1167                block, block + num - 1, (unsigned) inode->i_ino);
1168
1169 #ifndef EXT3_EXT_HAS_NO_TREE
1170         ext3_init_tree_desc(base, inode);
1171         tree.private = &bp;
1172 #endif
1173         bp.blocks = blocks;
1174         bp.created = created;
1175         bp.start = block;
1176         bp.init_num = bp.num = num;
1177         bp.create = create;
1178
1179         fsfilt_down_truncate_sem(inode);
1180         err = fsfilt_ext3_ext_walk_space(base, block, num,
1181                                          ext3_ext_new_extent_cb, &bp);
1182         ext3_ext_invalidate_cache(base);
1183         fsfilt_up_truncate_sem(inode);
1184
1185         return err;
1186 }
1187
1188 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1189                                     int pages, unsigned long *blocks,
1190                                     int *created, int create)
1191 {
1192         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1193         int rc = 0, i = 0;
1194         struct page *fp = NULL;
1195         int clen = 0;
1196
1197         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1198                 inode->i_ino, pages, (*page)->index);
1199
1200         /* pages are sorted already. so, we just have to find
1201          * contig. space and process them properly */
1202         while (i < pages) {
1203                 if (fp == NULL) {
1204                         /* start new extent */
1205                         fp = *page++;
1206                         clen = 1;
1207                         i++;
1208                         continue;
1209                 } else if (fp->index + clen == (*page)->index) {
1210                         /* continue the extent */
1211                         page++;
1212                         clen++;
1213                         i++;
1214                         continue;
1215                 }
1216
1217                 /* process found extent */
1218                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1219                                         clen * blocks_per_page, blocks,
1220                                         created, create);
1221                 if (rc)
1222                         GOTO(cleanup, rc);
1223
1224                 /* look for next extent */
1225                 fp = NULL;
1226                 blocks += blocks_per_page * clen;
1227                 created += blocks_per_page * clen;
1228         }
1229
1230         if (fp)
1231                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1232                                         clen * blocks_per_page, blocks,
1233                                         created, create);
1234 cleanup:
1235         return rc;
1236 }
1237
1238 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1239                                unsigned long *blocks, int *created, int create);
1240 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1241                                    int pages, unsigned long *blocks,
1242                                    int *created, int create)
1243 {
1244         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1245         unsigned long *b;
1246         int rc = 0, i, *cr;
1247
1248         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1249                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1250                 if (rc) {
1251                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1252                                inode->i_ino, *b, *cr, create, rc);
1253                         break;
1254                 }
1255
1256                 b += blocks_per_page;
1257                 cr += blocks_per_page;
1258         }
1259         return rc;
1260 }
1261
1262 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1263                                 int pages, unsigned long *blocks,
1264                                 int *created, int create,
1265                                 cfs_mutex_t *optional_mutex)
1266 {
1267         int rc;
1268
1269         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1270                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1271                                                      blocks, created, create);
1272                 return rc;
1273         }
1274         if (optional_mutex != NULL)
1275                 cfs_mutex_lock(optional_mutex);
1276         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1277                                             created, create);
1278         if (optional_mutex != NULL)
1279                 cfs_mutex_unlock(optional_mutex);
1280
1281         return rc;
1282 }
1283
1284 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1285 {
1286         unsigned long block;
1287         struct buffer_head *bh;
1288         int err, blocksize, csize, boffs, osize = size;
1289
1290         /* prevent reading after eof */
1291         cfs_lock_kernel();
1292         if (i_size_read(inode) < *offs + size) {
1293                 size = i_size_read(inode) - *offs;
1294                 cfs_unlock_kernel();
1295                 if (size < 0) {
1296                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1297                                i_size_read(inode), *offs);
1298                         return -EBADR;
1299                 } else if (size == 0) {
1300                         return 0;
1301                 }
1302         } else {
1303                 cfs_unlock_kernel();
1304         }
1305
1306         blocksize = 1 << inode->i_blkbits;
1307
1308         while (size > 0) {
1309                 block = *offs >> inode->i_blkbits;
1310                 boffs = *offs & (blocksize - 1);
1311                 csize = min(blocksize - boffs, size);
1312                 bh = ext3_bread(NULL, inode, block, 0, &err);
1313                 if (!bh) {
1314                         CERROR("can't read block: %d\n", err);
1315                         return err;
1316                 }
1317
1318                 memcpy(buf, bh->b_data + boffs, csize);
1319                 brelse(bh);
1320
1321                 *offs += csize;
1322                 buf += csize;
1323                 size -= csize;
1324         }
1325         return osize;
1326 }
1327 EXPORT_SYMBOL(fsfilt_ext3_read);
1328
1329 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1330                                    int size, loff_t *offs)
1331 {
1332         int rc;
1333         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1334         if (rc > 0)
1335                 rc = 0;
1336         return rc;
1337 }
1338
1339 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1340                                 loff_t *offs, handle_t *handle)
1341 {
1342         struct buffer_head *bh = NULL;
1343         loff_t old_size = i_size_read(inode), offset = *offs;
1344         loff_t new_size = i_size_read(inode);
1345         unsigned long block;
1346         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1347
1348         while (bufsize > 0) {
1349                 if (bh != NULL)
1350                         brelse(bh);
1351
1352                 block = offset >> inode->i_blkbits;
1353                 boffs = offset & (blocksize - 1);
1354                 size = min(blocksize - boffs, bufsize);
1355                 bh = ext3_bread(handle, inode, block, 1, &err);
1356                 if (!bh) {
1357                         CERROR("can't read/create block: %d\n", err);
1358                         break;
1359                 }
1360
1361                 err = ext3_journal_get_write_access(handle, bh);
1362                 if (err) {
1363                         CERROR("journal_get_write_access() returned error %d\n",
1364                                err);
1365                         break;
1366                 }
1367                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1368                 memcpy(bh->b_data + boffs, buf, size);
1369                 err = ext3_journal_dirty_metadata(handle, bh);
1370                 if (err) {
1371                         CERROR("journal_dirty_metadata() returned error %d\n",
1372                                err);
1373                         break;
1374                 }
1375                 if (offset + size > new_size)
1376                         new_size = offset + size;
1377                 offset += size;
1378                 bufsize -= size;
1379                 buf += size;
1380         }
1381         if (bh)
1382                 brelse(bh);
1383
1384         /* correct in-core and on-disk sizes */
1385         if (new_size > i_size_read(inode)) {
1386                 cfs_lock_kernel();
1387                 if (new_size > i_size_read(inode))
1388                         i_size_write(inode, new_size);
1389                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1390                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1391                 if (i_size_read(inode) > old_size)
1392                         mark_inode_dirty(inode);
1393                 cfs_unlock_kernel();
1394         }
1395
1396         if (err == 0)
1397                 *offs = offset;
1398         return err;
1399 }
1400 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1401
1402 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1403                                     loff_t *offs, int force_sync)
1404 {
1405         struct inode *inode = file->f_dentry->d_inode;
1406         handle_t *handle;
1407         int err, block_count = 0, blocksize;
1408
1409         /* Determine how many transaction credits are needed */
1410         blocksize = 1 << inode->i_blkbits;
1411         block_count = (*offs & (blocksize - 1)) + bufsize;
1412         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1413
1414         handle = ext3_journal_start(inode,
1415                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1416         if (IS_ERR(handle)) {
1417                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1418                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1419                 return PTR_ERR(handle);
1420         }
1421
1422         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1423
1424         if (!err && force_sync)
1425                 handle->h_sync = 1; /* recovery likes this */
1426
1427         ext3_journal_stop(handle);
1428
1429         return err;
1430 }
1431
1432 static int fsfilt_ext3_setup(struct super_block *sb)
1433 {
1434 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1435      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1436         struct ext3_sb_info *sbi = EXT3_SB(sb);
1437 #if 0
1438         sbi->dx_lock = fsfilt_ext3_dx_lock;
1439         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1440 #endif
1441 #endif
1442         if (!EXT3_HAS_COMPAT_FEATURE(sb,
1443                                 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1444                 CERROR("ext3 mounted without journal\n");
1445                 return -EINVAL;
1446         }
1447
1448 #ifdef S_PDIROPS
1449         CWARN("Enabling PDIROPS\n");
1450         set_opt(sbi->s_mount_opt, PDIROPS);
1451         sb->s_flags |= S_PDIROPS;
1452 #endif
1453         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1454                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1455 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1456         /* enable journaled quota support */
1457         /* kfreed in ext3_put_super() */
1458         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1459         if (!sbi->s_qf_names[USRQUOTA])
1460                 return -ENOMEM;
1461         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1462         if (!sbi->s_qf_names[GRPQUOTA]) {
1463                 kfree(sbi->s_qf_names[USRQUOTA]);
1464                 sbi->s_qf_names[USRQUOTA] = NULL;
1465                 return -ENOMEM;
1466         }
1467         sbi->s_jquota_fmt = QFMT_LUSTRE;
1468 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1469         set_opt(sbi->s_mount_opt, QUOTA);
1470 #endif
1471 #endif
1472         return 0;
1473 }
1474
1475 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1476    objects. Logs is number of logfiles to update */
1477 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1478 {
1479         if ( !fso ) {
1480                 switch(op) {
1481                 case FSFILT_OP_CREATE:
1482                                  /* directory leaf, index & indirect & EA*/
1483                         return 4 + 3 * logs;
1484                 case FSFILT_OP_UNLINK:
1485                         return 3 * logs;
1486                 }
1487         } else {
1488                 int i;
1489                 int needed = 0;
1490                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1491                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1492                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1493                 for (i = 0; i < op; i++, fso++) {
1494                         int nblocks = fso->fso_bufcnt * blockpp;
1495                         int ndindirect = min(nblocks, addrpp + 1);
1496                         int nindir = nblocks + ndindirect + 1;
1497
1498                         needed += nindir;
1499                 }
1500                 return needed + 3 * logs;
1501         }
1502
1503         return 0;
1504 }
1505
1506 #ifdef HAVE_QUOTA_SUPPORT
1507 #define DQINFO_COPY(out, in)                    \
1508 do {                                            \
1509         Q_COPY(out, in, dqi_bgrace);            \
1510         Q_COPY(out, in, dqi_igrace);            \
1511         Q_COPY(out, in, dqi_flags);             \
1512         Q_COPY(out, in, dqi_valid);             \
1513 } while (0)
1514
1515 #define DQBLK_COPY(out, in)                     \
1516 do {                                            \
1517         Q_COPY(out, in, dqb_bhardlimit);        \
1518         Q_COPY(out, in, dqb_bsoftlimit);        \
1519         Q_COPY(out, in, dqb_curspace);          \
1520         Q_COPY(out, in, dqb_ihardlimit);        \
1521         Q_COPY(out, in, dqb_isoftlimit);        \
1522         Q_COPY(out, in, dqb_curinodes);         \
1523         Q_COPY(out, in, dqb_btime);             \
1524         Q_COPY(out, in, dqb_itime);             \
1525         Q_COPY(out, in, dqb_valid);             \
1526 } while (0)
1527
1528 static int fsfilt_ext3_quotactl(struct super_block *sb,
1529                                 struct obd_quotactl *oqc)
1530 {
1531         int i, rc = 0, error = 0;
1532         const struct quotactl_ops *qcop;
1533         struct if_dqinfo *info;
1534         struct if_dqblk *dqblk;
1535         ENTRY;
1536
1537         if (!sb->s_qcop)
1538                 RETURN(-ENOSYS);
1539
1540         OBD_ALLOC_PTR(info);
1541         if (!info)
1542                 RETURN(-ENOMEM);
1543         OBD_ALLOC_PTR(dqblk);
1544         if (!dqblk) {
1545                 OBD_FREE_PTR(info);
1546                 RETURN(-ENOMEM);
1547         }
1548
1549         DQINFO_COPY(info, &oqc->qc_dqinfo);
1550         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1551
1552         qcop = sb->s_qcop;
1553         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1554                 for (i = 0; i < MAXQUOTAS; i++) {
1555                         if (!Q_TYPESET(oqc, i))
1556                                 continue;
1557
1558                         if (oqc->qc_cmd == Q_QUOTAON) {
1559                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1560
1561                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1562
1563                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1564                                                  name[i], 0);
1565                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1566                                 rc = ll_quota_off(sb, i, 0);
1567                         }
1568
1569                         if (rc == -EBUSY)
1570                                 error = rc;
1571                         else if (rc)
1572                                 GOTO(out, rc);
1573                 }
1574                 GOTO(out, rc ?: error);
1575         }
1576
1577         switch (oqc->qc_cmd) {
1578         case Q_GETOINFO:
1579         case Q_GETINFO:
1580                 if (!qcop->get_info)
1581                         GOTO(out, rc = -ENOSYS);
1582                 rc = qcop->get_info(sb, oqc->qc_type, info);
1583                 break;
1584         case Q_SETQUOTA:
1585         case Q_INITQUOTA:
1586                 if (!qcop->set_dqblk)
1587                         GOTO(out, rc = -ENOSYS);
1588                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1589                 break;
1590         case Q_GETOQUOTA:
1591         case Q_GETQUOTA:
1592                 if (!qcop->get_dqblk)
1593                         GOTO(out, rc = -ENOSYS);
1594                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1595                 if (!rc)
1596                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1597                 break;
1598         case Q_SYNC:
1599                 if (!sb->s_qcop->quota_sync)
1600                         GOTO(out, rc = -ENOSYS);
1601                 qcop->quota_sync(sb, oqc->qc_type);
1602                 break;
1603         case Q_FINVALIDATE:
1604                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1605                 for (i = 0; i < MAXQUOTAS; i++) {
1606                         struct file *fp;
1607                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1608
1609                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1610
1611                         if (!Q_TYPESET(oqc, i))
1612                                 continue;
1613
1614                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1615                         if (IS_ERR(fp)) {
1616                                 rc = PTR_ERR(fp);
1617                                 CERROR("error invalidating operational quota file"
1618                                        " %s (rc:%d)\n", name[i], rc);
1619                         } else {
1620                                 filp_close(fp, 0);
1621                         }
1622
1623                 }
1624                 break;
1625         default:
1626                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1627                 LBUG();
1628         }
1629 out:
1630         DQINFO_COPY(&oqc->qc_dqinfo, info);
1631         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1632
1633         OBD_FREE_PTR(info);
1634         OBD_FREE_PTR(dqblk);
1635
1636         if (rc)
1637                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1638                                 "failed: %d\n",
1639                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1640         RETURN(rc);
1641 }
1642
1643 struct chk_dqblk{
1644         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1645         cfs_list_t              dqb_list;        /** in list also */
1646         qid_t                   dqb_id;          /** uid/gid */
1647         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1648         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1649         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1650         qsize_t                 dqb_curspace;    /** current space */
1651         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1652         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1653         qsize_t                 dqb_curinodes;   /** current inodes */
1654         obd_time                dqb_btime;       /** block grace time */
1655         obd_time                dqb_itime;       /** inode grace time */
1656         __u32                   dqb_valid;       /** flag for above fields */
1657 };
1658
1659 static inline unsigned int chkquot_hash(qid_t id, int type)
1660                                         __attribute__((__const__));
1661
1662 static inline unsigned int chkquot_hash(qid_t id, int type)
1663 {
1664         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1665 }
1666
1667 static inline struct chk_dqblk *
1668 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1669 {
1670         cfs_hlist_node_t *node;
1671         struct chk_dqblk *cdqb;
1672
1673         cfs_hlist_for_each(node, head) {
1674                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1675                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1676                         return cdqb;
1677         }
1678
1679         return NULL;
1680 }
1681
1682 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1683 {
1684         struct chk_dqblk *cdqb;
1685
1686         OBD_ALLOC_PTR(cdqb);
1687         if (cdqb) {
1688                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1689                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1690                 cdqb->dqb_id = id;
1691                 cdqb->dqb_type = type;
1692         }
1693
1694         return cdqb;
1695 }
1696
1697 static struct chk_dqblk *
1698 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1699       cfs_list_t *list, qid_t id, int type, int first_check)
1700 {
1701         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1702         struct if_dqblk dqb;
1703         struct chk_dqblk *cdqb;
1704         int rc;
1705
1706         cdqb = find_chkquot(head, id, type);
1707         if (cdqb)
1708                 return cdqb;
1709
1710         cdqb = alloc_chkquot(id, type);
1711         if (!cdqb)
1712                 return NULL;
1713
1714         if (!first_check) {
1715                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1716                 if (rc) {
1717                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1718                                id, type, rc);
1719                 } else {
1720                         DQBLK_COPY(cdqb, &dqb);
1721                         cdqb->dqb_curspace = 0;
1722                         cdqb->dqb_curinodes = 0;
1723                 }
1724         }
1725
1726         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1727         cfs_list_add_tail(&cdqb->dqb_list, list);
1728
1729         return cdqb;
1730 }
1731
1732 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1733 {
1734         struct obd_quotactl *oqctl;
1735         int rc;
1736
1737         OBD_ALLOC_PTR(oqctl);
1738         if (!oqctl)
1739                 RETURN(-ENOMEM);
1740
1741         oqctl->qc_cmd = cmd;
1742         oqctl->qc_id = qfmt;
1743         oqctl->qc_type = type;
1744         rc = fsfilt_ext3_quotactl(sb, oqctl);
1745
1746         OBD_FREE_PTR(oqctl);
1747         return rc;
1748 }
1749
1750 static inline int read_old_dqinfo(struct super_block *sb, int type,
1751                                   struct if_dqinfo *dqinfo)
1752 {
1753         struct obd_quotactl *oqctl;
1754         int rc;
1755         ENTRY;
1756
1757         OBD_ALLOC_PTR(oqctl);
1758         if (!oqctl)
1759                 RETURN(-ENOMEM);
1760
1761         oqctl->qc_cmd = Q_GETINFO;
1762         oqctl->qc_type = type;
1763         rc = fsfilt_ext3_quotactl(sb, oqctl);
1764         if (!rc)
1765                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1766
1767         OBD_FREE_PTR(oqctl);
1768         RETURN(rc);
1769 }
1770
1771 #ifndef HAVE_EXT4_LDISKFS
1772 static inline struct ext3_group_desc *
1773 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1774 {
1775         unsigned long desc_block, desc;
1776         struct ext3_group_desc *gdp;
1777
1778         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1779         desc = group % EXT3_DESC_PER_BLOCK(sb);
1780         gdp = (struct ext3_group_desc *)
1781               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1782
1783         return gdp + desc;
1784 }
1785
1786 static inline struct buffer_head *
1787 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1788 {
1789         struct ext3_group_desc *desc;
1790         struct buffer_head *bh;
1791
1792         desc = get_group_desc(sb, group, NULL);
1793         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1794         return bh;
1795 }
1796
1797 static __u32 ext3_itable_unused_count(struct super_block *sb,
1798                                struct ext3_group_desc *bg) {
1799        return le16_to_cpu(bg->bg_itable_unused);
1800 }
1801 #else
1802 #define get_group_desc ext3_get_group_desc
1803 #endif
1804
1805 struct qchk_ctxt {
1806         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1807         cfs_list_t              qckt_list;                 /* quotacheck list */
1808         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1809         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1810 };
1811
1812 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1813                            struct obd_quotactl *oqc)
1814 {
1815         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1816         loff_t size = 0;
1817         qid_t qid[MAXQUOTAS];
1818         int cnt, i, rc = 0;
1819
1820         if (!inode)
1821                 return 0;
1822
1823         qid[USRQUOTA] = inode->i_uid;
1824         qid[GRPQUOTA] = inode->i_gid;
1825
1826         if (S_ISDIR(inode->i_mode) ||
1827             S_ISREG(inode->i_mode) ||
1828             S_ISLNK(inode->i_mode))
1829                 size = inode_get_bytes(inode);
1830
1831         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1832                 if (!Q_TYPESET(oqc, cnt))
1833                         continue;
1834
1835                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1836                                 &qctxt->qckt_list, qid[cnt], cnt,
1837                                 qctxt->qckt_first_check[cnt]);
1838                 if (!cdqb[cnt]) {
1839                         rc = -ENOMEM;
1840                         break;
1841                 }
1842
1843                 cdqb[cnt]->dqb_curspace += size;
1844                 cdqb[cnt]->dqb_curinodes++;
1845         }
1846
1847         if (rc) {
1848                 for (i = 0; i < cnt; i++) {
1849                         if (!Q_TYPESET(oqc, i))
1850                                 continue;
1851                         LASSERT(cdqb[i]);
1852                         cdqb[i]->dqb_curspace -= size;
1853                         cdqb[i]->dqb_curinodes--;
1854                 }
1855         }
1856
1857         return rc;
1858 }
1859
1860 /* write dqinfo struct in a new quota file */
1861 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1862 {
1863         struct v2_disk_dqinfo dqinfo;
1864         __u32 blocks = V2_DQTREEOFF + 1;
1865         loff_t offset = V2_DQINFOOFF;
1866
1867         if (info) {
1868                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1869                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1870                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1871                                                ~DQF_INFO_DIRTY);
1872         } else {
1873                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1874                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1875                 dqinfo.dqi_flags = 0;
1876         }
1877
1878         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1879         dqinfo.dqi_free_blk = 0;
1880         dqinfo.dqi_free_entry = 0;
1881
1882         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1883 }
1884
1885 static int v3_write_dqheader(struct file *f, int type)
1886 {
1887         static const __u32 quota_magics[] = V2_INITQMAGICS;
1888         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1889         struct v2_disk_dqheader dqhead;
1890         loff_t offset = 0;
1891
1892         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1893         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1894
1895         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1896         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1897
1898         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1899 }
1900
1901 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1902                                   struct obd_quotactl *oqc)
1903 {
1904         int i, rc = 0;
1905         ENTRY;
1906
1907         for (i = 0; i < MAXQUOTAS; i++) {
1908                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1909                                          NULL : &qctxt->qckt_dqinfo[i];
1910                 struct file *file;
1911                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1912
1913                 if (!Q_TYPESET(oqc, i))
1914                         continue;
1915
1916                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1917
1918                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1919                 if (IS_ERR(file)) {
1920                         rc = PTR_ERR(file);
1921                         CERROR("can't create %s file: rc = %d\n",
1922                                name[i], rc);
1923                         GOTO(out, rc);
1924                 }
1925
1926                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1927                         CERROR("file %s is not regular", name[i]);
1928                         filp_close(file, 0);
1929                         GOTO(out, rc = -EINVAL);
1930                 }
1931
1932                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1933
1934                 rc = v3_write_dqheader(file, i);
1935                 if (rc) {
1936                         filp_close(file, 0);
1937                         GOTO(out, rc);
1938                 }
1939
1940                 rc = v3_write_dqinfo(file, i, info);
1941                 filp_close(file, 0);
1942                 if (rc)
1943                         GOTO(out, rc);
1944         }
1945
1946 out:
1947         RETURN(rc);
1948 }
1949
1950
1951 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1952                           struct chk_dqblk *cdqb)
1953 {
1954         struct obd_quotactl *oqc;
1955         long now;
1956         int rc;
1957         ENTRY;
1958
1959         OBD_ALLOC_PTR(oqc);
1960         if (!oqc)
1961                 RETURN(-ENOMEM);
1962
1963         now = cfs_time_current_sec();
1964
1965         if (cdqb->dqb_bsoftlimit &&
1966             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1967             !cdqb->dqb_btime)
1968                 cdqb->dqb_btime =
1969                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1970
1971         if (cdqb->dqb_isoftlimit &&
1972             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1973             !cdqb->dqb_itime)
1974                 cdqb->dqb_itime =
1975                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1976
1977         cdqb->dqb_valid = QIF_ALL;
1978
1979         oqc->qc_cmd = Q_SETQUOTA;
1980         oqc->qc_type = cdqb->dqb_type;
1981         oqc->qc_id = cdqb->dqb_id;
1982         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1983
1984         rc = fsfilt_ext3_quotactl(sb, oqc);
1985         OBD_FREE_PTR(oqc);
1986         RETURN(rc);
1987 }
1988
1989 static int prune_chkquots(struct super_block *sb,
1990                           struct qchk_ctxt *qctxt, int error)
1991 {
1992         struct chk_dqblk *cdqb, *tmp;
1993         int rc;
1994
1995         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1996                 if (!error) {
1997                         rc = commit_chkquot(sb, qctxt, cdqb);
1998                         if (rc)
1999                                 error = rc;
2000                 }
2001                 cfs_hlist_del_init(&cdqb->dqb_hash);
2002                 cfs_list_del(&cdqb->dqb_list);
2003                 OBD_FREE_PTR(cdqb);
2004         }
2005
2006         return error;
2007 }
2008
2009 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2010 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2011 #endif
2012
2013 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2014                                   struct obd_quotactl *oqc)
2015 {
2016         struct ext3_sb_info *sbi = EXT3_SB(sb);
2017         int i, group, uninit_feat = 0;
2018         struct qchk_ctxt *qctxt;
2019         struct buffer_head *bitmap_bh = NULL;
2020         unsigned long ino, inode_inuse;
2021         struct inode *inode;
2022         int rc = 0;
2023         ENTRY;
2024
2025         /* turn on quota and read dqinfo if existed */
2026         OBD_ALLOC_PTR(qctxt);
2027         if (!qctxt) {
2028                 oqc->qc_stat = -ENOMEM;
2029                 RETURN(-ENOMEM);
2030         }
2031
2032         for (i = 0; i < NR_DQHASH; i++)
2033                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2034         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2035
2036         for (i = 0; i < MAXQUOTAS; i++) {
2037                 if (!Q_TYPESET(oqc, i))
2038                         continue;
2039
2040                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2041                 if (!rc || rc == -EBUSY) {
2042                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2043                         if (rc)
2044                                 GOTO(out, rc);
2045                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2046                         qctxt->qckt_first_check[i] = 1;
2047                 } else if (rc) {
2048                         GOTO(out, rc);
2049                 }
2050         }
2051         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2052                 /* This filesystem supports the uninit group feature */
2053                 uninit_feat = 1;
2054
2055         /* number of inodes that have been allocated */
2056         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2057                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2058
2059         /* check quota and update in hash */
2060         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2061              group++) {
2062                 unsigned long used_count = sbi->s_inodes_per_group;
2063
2064                 if (uninit_feat) {
2065                         struct ext3_group_desc *desc;
2066                         desc = get_group_desc(sb, group, NULL);
2067                         if (!desc)
2068                                 GOTO(out, -EIO);
2069
2070                         /* we don't really need to take the group lock here,
2071                          * but it may be useful if one day we support online
2072                          * quotacheck */
2073 #ifdef HAVE_EXT4_LDISKFS
2074                         ext4_lock_group(sb, group);
2075 #else
2076                         spin_lock(sb_bgl_lock(sbi, group));
2077 #endif
2078                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2079                                 /* no inode in use in this group, just skip it */
2080 #ifdef HAVE_EXT4_LDISKFS
2081                                 ext3_unlock_group(sb, group);
2082 #else
2083                                 spin_unlock(sb_bgl_lock(sbi, group));
2084 #endif
2085                                 continue;
2086                         }
2087
2088                         used_count -= ext3_itable_unused_count(sb, desc);
2089 #ifdef HAVE_EXT4_LDISKFS
2090                         ext3_unlock_group(sb, group);
2091 #else
2092                         spin_unlock(sb_bgl_lock(sbi, group));
2093 #endif
2094                 }
2095
2096                 ino = group * sbi->s_inodes_per_group + 1;
2097                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2098                 if (!bitmap_bh) {
2099                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2100                                sb->s_id, group);
2101                         GOTO(out, -EIO);
2102                 }
2103
2104                 i = 0;
2105                 while (i < used_count &&
2106                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2107                                                used_count, i)) < used_count) {
2108                         inode_inuse--;
2109                         i++;
2110                         ino = i + group * sbi->s_inodes_per_group;
2111                         if (ino < sbi->s_first_ino)
2112                                 continue;
2113 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2114                         inode = ext3_iget(sb, ino);
2115 #else
2116                         inode = iget(sb, ino);
2117 #endif
2118                         if (!inode || IS_ERR(inode))
2119                                 continue;
2120
2121                         rc = add_inode_quota(inode, qctxt, oqc);
2122                         iput(inode);
2123                         if (rc) {
2124                                 brelse(bitmap_bh);
2125                                 GOTO(out, rc);
2126                         }
2127                 }
2128                 brelse(bitmap_bh);
2129         }
2130
2131         /* read old quota limits from old quota file. (only for the user
2132          * has limits but hasn't file) */
2133 #ifdef HAVE_QUOTA_SUPPORT
2134         for (i = 0; i < MAXQUOTAS; i++) {
2135                 cfs_list_t id_list;
2136                 struct dquot_id *dqid, *tmp;
2137
2138                 if (!Q_TYPESET(oqc, i))
2139                         continue;
2140
2141                 if (qctxt->qckt_first_check[i])
2142                         continue;
2143
2144
2145                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2146                 CFS_INIT_LIST_HEAD(&id_list);
2147 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2148                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2149 #else
2150                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2151 #endif
2152                 if (rc)
2153                         CERROR("read old limits failed. (rc:%d)\n", rc);
2154
2155                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2156                         cfs_list_del_init(&dqid->di_link);
2157
2158                         if (!rc)
2159                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2160                                       dqid->di_id, i,
2161                                       qctxt->qckt_first_check[i]);
2162                         OBD_FREE_PTR(dqid);
2163                 }
2164         }
2165 #endif
2166         /* turn off quota cause we are to dump chk_dqblk to files */
2167         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2168
2169         rc = create_new_quota_files(qctxt, oqc);
2170         if (rc)
2171                 GOTO(out, rc);
2172
2173         /* we use vfs functions to set dqblk, so turn quota on */
2174         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2175 out:
2176         /* dump and free chk_dqblk */
2177         rc = prune_chkquots(sb, qctxt, rc);
2178         OBD_FREE_PTR(qctxt);
2179
2180         /* turn off quota, `lfs quotacheck` will turn on when all
2181          * nodes quotacheck finish. */
2182         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2183
2184         oqc->qc_stat = rc;
2185         if (rc)
2186                 CERROR("quotacheck failed: rc = %d\n", rc);
2187
2188         RETURN(rc);
2189 }
2190
2191 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2192                                  int cmd)
2193 {
2194         int rc = 0;
2195         ENTRY;
2196
2197         if (lqi->qi_files[type] == NULL) {
2198                 CERROR("operate qinfo before it's enabled!\n");
2199                 RETURN(-ESRCH);
2200         }
2201
2202         switch (cmd) {
2203         case QFILE_CHK:
2204                 rc = lustre_check_quota_file(lqi, type);
2205                 break;
2206         case QFILE_RD_INFO:
2207                 rc = lustre_read_quota_info(lqi, type);
2208                 break;
2209         case QFILE_WR_INFO:
2210                 rc = lustre_write_quota_info(lqi, type);
2211                 break;
2212         case QFILE_INIT_INFO:
2213                 rc = lustre_init_quota_info(lqi, type);
2214                 break;
2215         case QFILE_CONVERT:
2216                 rc = -ENOTSUPP;
2217                 CERROR("quota CONVERT command is not supported\n");
2218                 break;
2219         default:
2220                 rc = -ENOTSUPP;
2221                 CERROR("Unsupported admin quota file cmd %d\n"
2222                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2223                        cmd);
2224                 break;
2225         }
2226         RETURN(rc);
2227 }
2228
2229 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2230                             cfs_list_t *list)
2231 {
2232         return lustre_get_qids(file, inode, type, list);
2233 }
2234
2235 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2236 {
2237         int rc = 0;
2238         ENTRY;
2239
2240         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2241                 CERROR("operate dquot before it's enabled!\n");
2242                 RETURN(-ESRCH);
2243         }
2244
2245         switch (cmd) {
2246         case QFILE_RD_DQUOT:
2247                 rc = lustre_read_dquot(dquot);
2248                 break;
2249         case QFILE_WR_DQUOT:
2250                 if (dquot->dq_dqb.dqb_ihardlimit ||
2251                     dquot->dq_dqb.dqb_isoftlimit ||
2252                     dquot->dq_dqb.dqb_bhardlimit ||
2253                     dquot->dq_dqb.dqb_bsoftlimit)
2254                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2255                 else
2256                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2257
2258                 rc = lustre_commit_dquot(dquot);
2259                 if (rc >= 0)
2260                         rc = 0;
2261                 break;
2262         default:
2263                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2264                 LBUG();
2265                 break;
2266         }
2267         RETURN(rc);
2268 }
2269
2270 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2271                                 struct inode *inode, int frags)
2272 {
2273 #ifdef EXT3_EXT_HAS_NO_TREE
2274         struct ext3_ext_base *base = inode;
2275 #else
2276         struct ext3_extents_tree tree;
2277         struct ext3_ext_base *base = &tree;
2278
2279         ext3_init_tree_desc(base, inode);
2280 #endif
2281         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2282          * metablocks at maxium b=16542 */
2283         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2284         return 0;
2285 }
2286
2287 #endif
2288
2289 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2290 {
2291         return (EXT3_SB(sb)->journal_bdev);
2292 }
2293 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2294
2295 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2296                           char *buf, int count, loff_t pos)
2297 {
2298         loff_t p = pos;
2299         int rc;
2300
2301         if (!f && !inode) {
2302                 CERROR("lustre_read_quota failed for no quota file!\n");
2303                 libcfs_debug_dumpstack(NULL);
2304                 return -EINVAL;
2305         }
2306
2307         /* Support for both adm and op quota files must be provided */
2308         if (f) {
2309                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2310                 rc = rc < 0 ? rc : p - pos;
2311         } else {
2312                 struct super_block *sb = inode->i_sb;
2313                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2314         }
2315         return rc;
2316 }
2317
2318 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2319 {
2320         loff_t p = pos;
2321         int rc;
2322
2323         /* Only adm quota files are supported, op updates are handled by vfs */
2324         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2325         rc = rc < 0 ? rc : p - pos;
2326
2327         return rc;
2328 }
2329
2330 void *lustre_quota_journal_start(struct inode *inode, int delete)
2331 {
2332         handle_t *handle;
2333         unsigned block_count;
2334
2335         if (delete) {
2336                 /* each indirect block (+4) may become free, attaching to the
2337                  * header list of free blocks (+1); the data block (+1) may
2338                  * become a free block (+0) or a block with free dqentries (+0) */
2339                 block_count = (4 + 1) + 1;
2340                 handle = ext3_journal_start(inode,
2341                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2342         } else {
2343                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2344                  * freeblk reusage with a header update (+1); dqentry is either reused
2345                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2346                  * a new block allocation (+1) with a header update (+1)              */
2347                 block_count = (4 + 1) + 3;
2348                 handle = ext3_journal_start(inode,
2349                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2350
2351         }
2352
2353         return handle;
2354 }
2355
2356 void lustre_quota_journal_stop(void *handle)
2357 {
2358         ext3_journal_stop((handle_t *)handle);
2359 }
2360
2361 static int ll_decode_fh_accept(void *context, struct dentry *de)
2362 {
2363         return 1;
2364 }
2365
2366 #ifdef HAVE_EXPORTFS_DECODE_FH
2367 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2368          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2369                             acceptable, context)
2370 #else
2371 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2372          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2373                                      type, acceptable, context)
2374 # define FILEID_INO32_GEN 1
2375 extern struct export_operations export_op_default;
2376 #endif
2377
2378 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2379                                       struct fsfilt_fid *fid, int ignore_gen)
2380 {
2381         struct inode  *inode;
2382         struct dentry *result;
2383
2384         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2385                                        ll_decode_fh_accept, NULL);
2386         if (IS_ERR(result)) {
2387                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2388                        fid->ino, fid->gen, PTR_ERR(result));
2389                 return result;
2390         }
2391
2392         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2393                fid->ino, fid->gen);
2394         inode = result->d_inode;
2395         if (inode == NULL)
2396                 goto err_out;
2397
2398         if (inode->i_nlink == 0 &&
2399             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2400                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2401                               " ctime -- this may indicate disk "
2402                               "corruption (inode: %lu, link: %lu, "
2403                               "count: %d)\n", inode->i_ino,
2404                               (unsigned long)inode->i_nlink,
2405                               atomic_read(&inode->i_count));
2406                 goto err_out;
2407         }
2408         if (fid->gen && inode->i_generation != fid->gen) {
2409                 /* we didn't find the right inode.. */
2410                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2411                        "count: %d, generation %u/%u\n",
2412                        inode->i_ino, (unsigned long)inode->i_nlink,
2413                        atomic_read(&inode->i_count), inode->i_generation,
2414                        fid->gen);
2415                 goto err_out;
2416         }
2417
2418         return result;
2419 err_out:
2420         l_dput(result);
2421         return ERR_PTR(-ENOENT);
2422 }
2423
2424 static struct fsfilt_operations fsfilt_ext3_ops = {
2425         .fs_type                = "ext3",
2426         .fs_owner               = THIS_MODULE,
2427         .fs_getlabel            = fsfilt_ext3_get_label,
2428         .fs_setlabel            = fsfilt_ext3_set_label,
2429         .fs_uuid                = fsfilt_ext3_uuid,
2430         .fs_start               = fsfilt_ext3_start,
2431         .fs_brw_start           = fsfilt_ext3_brw_start,
2432         .fs_extend              = fsfilt_ext3_extend,
2433         .fs_commit              = fsfilt_ext3_commit,
2434         .fs_commit_async        = fsfilt_ext3_commit_async,
2435         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2436         .fs_setattr             = fsfilt_ext3_setattr,
2437         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2438         .fs_set_md              = fsfilt_ext3_set_md,
2439         .fs_get_md              = fsfilt_ext3_get_md,
2440         .fs_readpage            = fsfilt_ext3_readpage,
2441         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2442         .fs_statfs              = fsfilt_ext3_statfs,
2443         .fs_sync                = fsfilt_ext3_sync,
2444         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2445         .fs_write_record        = fsfilt_ext3_write_record,
2446         .fs_read_record         = fsfilt_ext3_read_record,
2447         .fs_setup               = fsfilt_ext3_setup,
2448         .fs_send_bio            = fsfilt_ext3_send_bio,
2449         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2450 #ifdef HAVE_DISK_INODE_VERSION
2451         .fs_get_version         = fsfilt_ext3_get_version,
2452         .fs_set_version         = fsfilt_ext3_set_version,
2453 #endif
2454 #ifdef HAVE_QUOTA_SUPPORT
2455         .fs_quotactl            = fsfilt_ext3_quotactl,
2456         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2457         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2458         .fs_qids                = fsfilt_ext3_qids,
2459         .fs_dquot               = fsfilt_ext3_dquot,
2460         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2461 #endif
2462         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2463         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2464 };
2465
2466 static int __init fsfilt_ext3_init(void)
2467 {
2468         int rc;
2469
2470         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2471                                          sizeof(struct fsfilt_cb_data), 0, 0);
2472         if (!fcb_cache) {
2473                 CERROR("error allocating fsfilt journal callback cache\n");
2474                 GOTO(out, rc = -ENOMEM);
2475         }
2476
2477         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2478
2479         if (rc) {
2480                 int err = cfs_mem_cache_destroy(fcb_cache);
2481                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2482         }
2483 out:
2484         return rc;
2485 }
2486
2487 static void __exit fsfilt_ext3_exit(void)
2488 {
2489         int rc;
2490
2491         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2492         rc = cfs_mem_cache_destroy(fcb_cache);
2493         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2494 }
2495
2496 module_init(fsfilt_ext3_init);
2497 module_exit(fsfilt_ext3_exit);
2498
2499 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2500 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2501 MODULE_LICENSE("GPL");