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, Intel Corporation.
  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 #include <ext4/ext4.h>
  53 #include <ext4/ext4_jbd2.h>
  54 #include <linux/version.h>
  55 #include <linux/bitops.h>
  56 #include <linux/quota.h>
  57
  58 #if defined(HAVE_EXT3_XATTR_H)
  59 # include <ext3/xattr.h>
  60 #elif !defined(EXT3_XATTR_INDEX_TRUSTED)
  61 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  62 /* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
  63 # define EXT3_XATTR_INDEX_TRUSTED        4
  64 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  65 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  66 #endif
  67
  68 #include <libcfs/libcfs.h>
  69 #include <lustre_fsfilt.h>
  70 #include <obd.h>
  71 #include <linux/lustre_compat25.h>
  72 #include <linux/lprocfs_status.h>
  73
  74 #include <ext4/ext4_extents.h>
  75
  76 /* for kernels 2.6.18 and later */
  77 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  78
  79 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
  80                ext3_ext_insert_extent(handle, inode, path, newext, flag)
  81
  82 #define ext3_mb_discard_inode_preallocations(inode) \
  83                  ext3_discard_preallocations(inode)
  84
  85 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
  86 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
  87
  88 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
  89 # define journal_callback ext4_journal_cb_entry
  90 # define fsfilt_journal_callback_set(handle, func, jcb) \
  91          ext4_journal_callback_add(handle, func, jcb)
  92 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
  93 # define fsfilt_journal_callback_set(handle, func, jcb) \
  94          jbd2_journal_callback_set(handle, func, jcb)
  95 #else
  96 # error missing journal commit callback
  97 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
  98
  99 static cfs_mem_cache_t *fcb_cache;
 100
 101 struct fsfilt_cb_data {
 102         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 103         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 104         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 105         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 106         void *cb_data;                  /* MDS/OST completion function data */
 107 };
 108
 109 #ifndef ext3_find_next_bit
 110 #define ext3_find_next_bit           ext2_find_next_bit
 111 #endif
 112
 113 #ifndef ext2_find_next_bit
 114 #ifdef __LITTLE_ENDIAN
 115 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 116 #else
 117 error "Need implementation of find_next_bit on big-endian systems"
 118 #endif  /* __LITTLE_ENDIAN */
 119 #endif  /* !ext2_find_next_le_bit */
 120
 121 static char *fsfilt_ext3_get_label(struct super_block *sb)
 122 {
 123         return EXT3_SB(sb)->s_es->s_volume_name;
 124 }
 125
 126 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 127 {
 128         /* see e.g. fsfilt_ext3_write_record() */
 129         journal_t *journal;
 130         handle_t *handle;
 131         int err;
 132
 133         journal = EXT3_SB(sb)->s_journal;
 134         handle = ext3_journal_start_sb(sb, 1);
 135         if (IS_ERR(handle)) {
 136                 CERROR("can't start transaction\n");
 137                 return(PTR_ERR(handle));
 138         }
 139
 140         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 141         if (err)
 142                 goto out;
 143
 144         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 145                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 146
 147         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 148
 149 out:
 150         ext3_journal_stop(handle);
 151
 152         return(err);
 153 }
 154
 155 static char *fsfilt_ext3_uuid(struct super_block *sb)
 156 {
 157         return EXT3_SB(sb)->s_es->s_uuid;
 158 }
 159
 160 #ifdef HAVE_DISK_INODE_VERSION
 161
 162 static __u64 get_i_version(struct inode *inode)
 163 {
 164         return EXT3_I(inode)->i_fs_version;
 165 }
 166
 167 static void set_i_version(struct inode *inode, __u64 new_version)
 168 {
 169         (EXT3_I(inode))->i_fs_version = new_version;
 170 }
 171
 172 /*
 173  * Get the 64-bit version for an inode.
 174  */
 175 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 176 {
 177         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 178                get_i_version(inode), inode->i_ino);
 179         return get_i_version(inode);
 180 }
 181
 182 /*
 183  * Set the 64-bit version and return the old version.
 184  */
 185 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 186 {
 187         __u64 old_version = get_i_version(inode);
 188
 189         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 190                new_version, old_version, inode->i_ino);
 191         set_i_version(inode, new_version);
 192         /* version is set after all inode operations are finished, so we should
 193          * mark it dirty here */
 194         inode->i_sb->s_op->dirty_inode(inode);
 195         return old_version;
 196 }
 197
 198 #endif
 199
 200 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
 201 #ifdef HAVE_BLOCKS_FOR_TRUNCATE
 202 # include <ext4/truncate.h>
 203 #else
 204 static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
 205 {
 206         ext4_lblk_t needed;
 207
 208         needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
 209         if (needed < 2)
 210                 needed = 2;
 211         if (needed > EXT4_MAX_TRANS_DATA)
 212                 needed = EXT4_MAX_TRANS_DATA;
 213         return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
 214 }
 215 #endif
 216
 217 /*
 218  * We don't currently need any additional blocks for rmdir and
 219  * unlink transactions because we are storing the OST oa_id inside
 220  * the inode (which we will be changing anyways as part of this
 221  * transaction).
 222  */
 223 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 224                                int logs)
 225 {
 226         /* For updates to the last received file */
 227         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 228         journal_t *journal;
 229         void *handle;
 230
 231         if (current->journal_info) {
 232                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 233                        current->journal_info);
 234                 goto journal_start;
 235         }
 236
 237         switch(op) {
 238         case FSFILT_OP_RMDIR:
 239         case FSFILT_OP_UNLINK:
 240                 /* delete one file + create/update logs for each stripe */
 241                 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
 242                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 243                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 244                 break;
 245         case FSFILT_OP_RENAME:
 246                 /* modify additional directory */
 247                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 248                 /* no break */
 249         case FSFILT_OP_SYMLINK:
 250                 /* additional block + block bitmap + GDT for long symlink */
 251                 nblocks += 3;
 252                 /* no break */
 253         case FSFILT_OP_CREATE: {
 254                 /* no break */
 255         }
 256         case FSFILT_OP_MKDIR:
 257         case FSFILT_OP_MKNOD:
 258                 /* modify one inode + block bitmap + GDT */
 259                 nblocks += 3;
 260                 /* no break */
 261         case FSFILT_OP_LINK:
 262                 /* modify parent directory */
 263                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 264                            EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
 265                 /* create/update logs for each stripe */
 266                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 267                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 268                 break;
 269         case FSFILT_OP_SETATTR:
 270                 /* Setattr on inode */
 271                 nblocks += 1;
 272                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 273                            EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
 274                 /* quota chown log for each stripe */
 275                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 276                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 277                 break;
 278         case FSFILT_OP_CANCEL_UNLINK:
 279                 LASSERT(logs == 1);
 280
 281                 /* blocks for log header bitmap update OR
 282                  * blocks for catalog header bitmap update + unlink of logs +
 283                  * blocks for delete the inode (include blocks truncating). */
 284                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 285                           EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
 286                           ext4_blocks_for_truncate(inode) + 3;
 287                 break;
 288         default: CERROR("unknown transaction start op %d\n", op);
 289                 LBUG();
 290         }
 291
 292         LASSERT(current->journal_info == desc_private);
 293         journal = EXT3_SB(inode->i_sb)->s_journal;
 294         if (nblocks > journal->j_max_transaction_buffers) {
 295                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 296                        nblocks, op, logs, journal->j_max_transaction_buffers);
 297                 nblocks = journal->j_max_transaction_buffers;
 298         }
 299
 300  journal_start:
 301         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 302         handle = ext3_journal_start(inode, nblocks);
 303
 304         if (!IS_ERR(handle))
 305                 LASSERT(current->journal_info == handle);
 306         else
 307                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 308                        op, nblocks, PTR_ERR(handle));
 309         return handle;
 310 }
 311
 312 /*
 313  * Calculate the number of buffer credits needed to write multiple pages in
 314  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 315  * doesn't have a nice API for calculating this sort of thing in advance.
 316  *
 317  * See comment above ext3_writepage_trans_blocks for details.  We assume
 318  * no data journaling is being done, but it does allow for all of the pages
 319  * being non-contiguous.  If we are guaranteed contiguous pages we could
 320  * reduce the number of (d)indirect blocks a lot.
 321  *
 322  * With N blocks per page and P pages, for each inode we have at most:
 323  * N*P indirect
 324  * min(N*P, blocksize/4 + 1) dindirect blocks
 325  * niocount tindirect
 326  *
 327  * For the entire filesystem, we have at most:
 328  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 329  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 330  * objcount inode blocks
 331  * 1 superblock
 332  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 333  *
 334  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 335  */
 336 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 337                                       int niocount, struct niobuf_local *nb)
 338 {
 339         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 340         __u64 next_indir;
 341         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 342         int nbitmaps = 0, ngdblocks;
 343         int needed = objcount + 1; /* inodes + superblock */
 344         int i, j;
 345
 346         for (i = 0, j = 0; i < objcount; i++, fso++) {
 347                 /* two or more dindirect blocks in case we cross boundary */
 348                 int ndind =
 349                         (long)((nb[j + fso->fso_bufcnt - 1].lnb_file_offset -
 350                                 nb[j].lnb_file_offset) >>
 351                                sb->s_blocksize_bits) /
 352                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 353                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 354
 355                 /* leaf, indirect, tindirect blocks for first block */
 356                 nbitmaps += blockpp + 2;
 357
 358                 j += fso->fso_bufcnt;
 359         }
 360
 361         next_indir = nb[0].lnb_file_offset +
 362                      (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 363         for (i = 1; i < niocount; i++) {
 364                 if (nb[i].lnb_file_offset >= next_indir) {
 365                         nbitmaps++;     /* additional indirect */
 366                         next_indir = nb[i].lnb_file_offset +
 367                                      (EXT3_ADDR_PER_BLOCK(sb) <<
 368                                       sb->s_blocksize_bits);
 369                 } else if (nb[i].lnb_file_offset !=
 370                            nb[i - 1].lnb_file_offset + sb->s_blocksize) {
 371                         nbitmaps++;     /* additional indirect */
 372                 }
 373                 nbitmaps += blockpp;    /* each leaf in different group? */
 374         }
 375
 376         ngdblocks = nbitmaps;
 377         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 378                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 379         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 380                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 381
 382         needed += nbitmaps + ngdblocks;
 383
 384         /* last_rcvd update */
 385         needed += EXT3_DATA_TRANS_BLOCKS(sb);
 386
 387 #if defined(CONFIG_QUOTA)
 388         /* We assume that there will be 1 bit set in s_dquot.flags for each
 389          * quota file that is active.  This is at least true for now.
 390          */
 391         needed += hweight32(sb_any_quota_loaded(sb)) *
 392                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 393 #endif
 394
 395         return needed;
 396 }
 397
 398 /* We have to start a huge journal transaction here to hold all of the
 399  * metadata for the pages being written here.  This is necessitated by
 400  * the fact that we do lots of prepare_write operations before we do
 401  * any of the matching commit_write operations, so even if we split
 402  * up to use "smaller" transactions none of them could complete until
 403  * all of them were opened.  By having a single journal transaction,
 404  * we eliminate duplicate reservations for common blocks like the
 405  * superblock and group descriptors or bitmaps.
 406  *
 407  * We will start the transaction here, but each prepare_write will
 408  * add a refcount to the transaction, and each commit_write will
 409  * remove a refcount.  The transaction will be closed when all of
 410  * the pages have been written.
 411  */
 412 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 413                                    int niocount, struct niobuf_local *nb,
 414                                    void *desc_private, int logs)
 415 {
 416         journal_t *journal;
 417         handle_t *handle;
 418         int needed;
 419         ENTRY;
 420
 421         LASSERT(current->journal_info == desc_private);
 422         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 423         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 424
 425         /* The number of blocks we could _possibly_ dirty can very large.
 426          * We reduce our request if it is absurd (and we couldn't get that
 427          * many credits for a single handle anyways).
 428          *
 429          * At some point we have to limit the size of I/Os sent at one time,
 430          * increase the size of the journal, or we have to calculate the
 431          * actual journal requirements more carefully by checking all of
 432          * the blocks instead of being maximally pessimistic.  It remains to
 433          * be seen if this is a real problem or not.
 434          */
 435         if (needed > journal->j_max_transaction_buffers) {
 436                 CERROR("want too many journal credits (%d) using %d instead\n",
 437                        needed, journal->j_max_transaction_buffers);
 438                 needed = journal->j_max_transaction_buffers;
 439         }
 440
 441         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 442         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 443         if (IS_ERR(handle)) {
 444                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 445                        PTR_ERR(handle));
 446         } else {
 447                 LASSERT(handle->h_buffer_credits >= needed);
 448                 LASSERT(current->journal_info == handle);
 449         }
 450
 451         RETURN(handle);
 452 }
 453
 454 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 455 {
 456        handle_t *handle = h;
 457
 458        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 459        if (nblocks == 0) {
 460                handle->h_buffer_credits = 0;
 461                CWARN("setting credits of handle %p to zero by request\n", h);
 462        }
 463
 464        if (handle->h_buffer_credits > nblocks)
 465                 return 0;
 466        if (ext3_journal_extend(handle, nblocks) == 0)
 467                 return 0;
 468
 469        ext3_mark_inode_dirty(handle, inode);
 470        return ext3_journal_restart(handle, nblocks);
 471 }
 472
 473 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 474 {
 475         int rc;
 476         handle_t *handle = h;
 477
 478         LASSERT(current->journal_info == handle);
 479         if (force_sync)
 480                 handle->h_sync = 1; /* recovery likes this */
 481
 482         rc = ext3_journal_stop(handle);
 483
 484         return rc;
 485 }
 486
 487 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 488                                     void **wait_handle)
 489 {
 490         unsigned long tid;
 491         transaction_t *transaction;
 492         handle_t *handle = h;
 493         journal_t *journal;
 494         int rc;
 495
 496         LASSERT(current->journal_info == handle);
 497
 498         transaction = handle->h_transaction;
 499         journal = transaction->t_journal;
 500         tid = transaction->t_tid;
 501         /* we don't want to be blocked */
 502         handle->h_sync = 0;
 503         rc = ext3_journal_stop(handle);
 504         if (rc) {
 505                 CERROR("error while stopping transaction: %d\n", rc);
 506                 return rc;
 507         }
 508         fsfilt_log_start_commit(journal, tid);
 509
 510         *wait_handle = (void *) tid;
 511         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 512         return 0;
 513 }
 514
 515 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 516 {
 517         journal_t *journal = EXT3_JOURNAL(inode);
 518         tid_t tid = (tid_t)(long)h;
 519
 520         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 521         if (unlikely(is_journal_aborted(journal)))
 522                 return -EIO;
 523
 524         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 525
 526         if (unlikely(is_journal_aborted(journal)))
 527                 return -EIO;
 528         return 0;
 529 }
 530
 531 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 532                                struct iattr *iattr, int do_trunc)
 533 {
 534         struct inode *inode = dentry->d_inode;
 535         int rc = 0;
 536
 537 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
 538         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
 539          * timestamps to appear to be in the far future, due old timestamp
 540          * being stored on disk as an unsigned value.  This fixes up any
 541          * bad values held by the client before storing them on disk,
 542          * and ensures any timestamp updates are correct.  LU-1042 */
 543         if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
 544                      !(iattr->ia_valid & ATTR_ATIME))) {
 545                 iattr->ia_valid |= ATTR_ATIME;
 546                 LTIME_S(iattr->ia_atime) = 0;
 547         }
 548         if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
 549                      !(iattr->ia_valid & ATTR_MTIME))) {
 550                 iattr->ia_valid |= ATTR_MTIME;
 551                 LTIME_S(iattr->ia_mtime) = 0;
 552         }
 553         if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
 554                       LTIME_S(inode->i_ctime) == 0) &&
 555                      !(iattr->ia_valid & ATTR_CTIME))) {
 556                 iattr->ia_valid |= ATTR_CTIME;
 557                 LTIME_S(iattr->ia_ctime) = 0;
 558         }
 559 #else
 560 #warning "remove old LU-221/LU-1042 workaround code"
 561 #endif
 562
 563         /* When initializating timestamps for new inodes, use the filesystem
 564          * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
 565          * that this is potentially an invalid inode.  Files with an old ctime
 566          * migrated to a newly-formatted OST with a newer s_mkfs_time will not
 567          * hit this check, since it is only for ctime == 0.  LU-1010/LU-1042 */
 568         if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
 569                 LTIME_S(iattr->ia_ctime) =
 570                         EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
 571
 572         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 573          * We are already writing the inode to disk as part of this
 574          * transaction and want to avoid a lot of extra inode writeout
 575          * later on. b=9828 */
 576         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 577                 /* ATTR_SIZE would invoke truncate: clear it */
 578                 iattr->ia_valid &= ~ATTR_SIZE;
 579                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 580                 i_size_write(inode, iattr->ia_size);
 581
 582                 if (iattr->ia_valid & ATTR_UID)
 583                         inode->i_uid = iattr->ia_uid;
 584                 if (iattr->ia_valid & ATTR_GID)
 585                         inode->i_gid = iattr->ia_gid;
 586                 if (iattr->ia_valid & ATTR_ATIME)
 587                         inode->i_atime = iattr->ia_atime;
 588                 if (iattr->ia_valid & ATTR_MTIME)
 589                         inode->i_mtime = iattr->ia_mtime;
 590                 if (iattr->ia_valid & ATTR_CTIME)
 591                         inode->i_ctime = iattr->ia_ctime;
 592                 if (iattr->ia_valid & ATTR_MODE) {
 593                         inode->i_mode = iattr->ia_mode;
 594
 595                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 596                             !cfs_capable(CFS_CAP_FSETID))
 597                                 inode->i_mode &= ~S_ISGID;
 598                 }
 599
 600                 inode->i_sb->s_op->dirty_inode(inode);
 601
 602                 goto out;
 603         }
 604
 605         /* Don't allow setattr to change file type */
 606         if (iattr->ia_valid & ATTR_MODE)
 607                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 608                                  (iattr->ia_mode & ~S_IFMT);
 609
 610         /* We set these flags on the client, but have already checked perms
 611          * so don't confuse inode_change_ok. */
 612         iattr->ia_valid &= ~TIMES_SET_FLAGS;
 613
 614         if (inode->i_op->setattr) {
 615                 rc = inode->i_op->setattr(dentry, iattr);
 616         } else {
 617 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
 618                 rc = inode_change_ok(inode, iattr);
 619                 if (!rc)
 620 #endif
 621                         rc = simple_setattr(dentry, iattr);
 622         }
 623
 624  out:
 625         RETURN(rc);
 626 }
 627
 628 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 629                                  unsigned int cmd, unsigned long arg)
 630 {
 631         int rc = 0;
 632         ENTRY;
 633
 634         /* FIXME: Can't do this because of nested transaction deadlock */
 635         if (cmd == EXT3_IOC_SETFLAGS) {
 636                 /* We can't enable data journaling on OST objects, because
 637                 * this forces the transaction to be closed in order to
 638                 * flush the journal, but the caller will already have a
 639                 * compound transaction open to update the last_rcvd file,
 640                 * and this thread would deadlock trying to set the flag. */
 641                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 642                         CERROR("can't set data journal flag on file\n");
 643                         RETURN(-EPERM);
 644                 }
 645                 /* Because the MDS does not see the EXTENTS_FL set on the
 646                  * OST objects, mask this flag into all set flags.  It is
 647                  * not legal to clear this flag in any case, so we are not
 648                  * changing the functionality by doing this.  b=22911 */
 649                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 650         }
 651
 652         /* ext4_ioctl does not have a inode argument */
 653         if (inode->i_fop->unlocked_ioctl)
 654                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 655         else
 656                 RETURN(-ENOTTY);
 657
 658         RETURN(rc);
 659 }
 660
 661 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 662                               void *lmm, int lmm_size, const char *name)
 663 {
 664         int rc;
 665
 666         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 667
 668         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 669                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 670
 671         if (rc && rc != -EROFS)
 672                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 673                        inode->i_ino, rc);
 674         return rc;
 675 }
 676
 677 /* Must be called with i_mutex held */
 678 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 679                               const char *name)
 680 {
 681         int rc;
 682
 683         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 684
 685         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 686                             name, lmm, lmm_size);
 687
 688         /* This gives us the MD size */
 689         if (lmm == NULL)
 690                 return (rc == -ENODATA) ? 0 : rc;
 691
 692         if (rc < 0) {
 693                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 694                        EXT3_XATTR_INDEX_TRUSTED, name,
 695                        inode->i_ino, rc);
 696                 memset(lmm, 0, lmm_size);
 697                 return (rc == -ENODATA) ? 0 : rc;
 698         }
 699
 700         return rc;
 701 }
 702
 703 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 704 {
 705         submit_bio(rw, bio);
 706         return 0;
 707 }
 708
 709 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 710                                     loff_t *off)
 711 {
 712         struct inode *inode = file->f_dentry->d_inode;
 713         int rc = 0;
 714
 715         if (S_ISREG(inode->i_mode))
 716                 rc = file->f_op->read(file, buf, count, off);
 717         else {
 718                 const int blkbits = inode->i_sb->s_blocksize_bits;
 719                 const int blksize = inode->i_sb->s_blocksize;
 720
 721                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 722                        (unsigned long)count, inode->i_ino, *off);
 723                 while (count > 0) {
 724                         struct buffer_head *bh;
 725
 726                         bh = NULL;
 727                         if (*off < i_size_read(inode)) {
 728                                 int err = 0;
 729
 730                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 731                                                 0, &err);
 732
 733                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 734
 735                                 if (bh) {
 736                                         memcpy(buf, bh->b_data, blksize);
 737                                         brelse(bh);
 738                                 } else if (err) {
 739                                         /* XXX in theory we should just fake
 740                                          * this buffer and continue like ext3,
 741                                          * especially if this is a partial read
 742                                          */
 743                                         CERROR("error read dir %lu+%llu: %d\n",
 744                                                inode->i_ino, *off, err);
 745                                         RETURN(err);
 746                                 }
 747                         }
 748                         if (!bh) {
 749                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 750
 751                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 752                                 memset(fake, 0, sizeof(*fake));
 753                                 fake->rec_len = cpu_to_le16(blksize);
 754                         }
 755                         count -= blksize;
 756                         buf += blksize;
 757                         *off += blksize;
 758                         rc += blksize;
 759                 }
 760         }
 761
 762         return rc;
 763 }
 764
 765 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 766 static void fsfilt_ext3_cb_func(struct super_block *sb,
 767                                 struct journal_callback *jcb, int error)
 768 #else
 769 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 770 #endif
 771 {
 772         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 773
 774         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 775
 776         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 777 }
 778
 779 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 780                                       void *handle, fsfilt_cb_t cb_func,
 781                                       void *cb_data)
 782 {
 783         struct fsfilt_cb_data *fcb;
 784
 785         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 786         if (fcb == NULL)
 787                 RETURN(-ENOMEM);
 788
 789         fcb->cb_func = cb_func;
 790         fcb->cb_obd = obd;
 791         fcb->cb_last_rcvd = last_rcvd;
 792         fcb->cb_data = cb_data;
 793
 794         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 795         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 796
 797         return 0;
 798 }
 799
 800 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 801 {
 802         struct kstatfs sfs;
 803         int rc;
 804
 805         memset(&sfs, 0, sizeof(sfs));
 806         rc = sb->s_op->statfs(sb->s_root, &sfs);
 807         statfs_pack(osfs, &sfs);
 808         return rc;
 809 }
 810
 811 static int fsfilt_ext3_sync(struct super_block *sb)
 812 {
 813         return ext3_force_commit(sb);
 814 }
 815
 816 #ifndef EXT3_EXTENTS_FL
 817 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 818 #endif
 819
 820 #ifndef EXT_ASSERT
 821 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 822 #endif
 823
 824 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 825 #define ext3_ext_base                   inode
 826 #define ext3_ext_base2inode(inode)      (inode)
 827 #define EXT_DEPTH(inode)                ext_depth(inode)
 828 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 829                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 830
 831 struct bpointers {
 832         unsigned long *blocks;
 833         int *created;
 834         unsigned long start;
 835         int num;
 836         int init_num;
 837         int create;
 838 };
 839
 840 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 841                                unsigned long block, int *aflags)
 842 {
 843         struct ext3_inode_info *ei = EXT3_I(inode);
 844         unsigned long bg_start;
 845         unsigned long colour;
 846         int depth;
 847
 848         if (path) {
 849                 struct ext3_extent *ex;
 850                 depth = path->p_depth;
 851
 852                 /* try to predict block placement */
 853                 if ((ex = path[depth].p_ext))
 854                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 855
 856                 /* it looks index is empty
 857                  * try to find starting from index itself */
 858                 if (path[depth].p_bh)
 859                         return path[depth].p_bh->b_blocknr;
 860         }
 861
 862         /* OK. use inode's group */
 863         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 864                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 865         colour = (current->pid % 16) *
 866                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 867         return bg_start + colour + block;
 868 }
 869
 870 #define ll_unmap_underlying_metadata(sb, blocknr) \
 871         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 872
 873 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 874 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 875                                 struct ext3_ext_path *path, unsigned long block,
 876                                 unsigned long *count, int *err)
 877 {
 878         unsigned long pblock, goal;
 879         int aflags = 0;
 880         struct inode *inode = ext3_ext_base2inode(base);
 881
 882         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 883         aflags |= 2; /* block have been already reserved */
 884         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 885         return pblock;
 886
 887 }
 888 #else
 889 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 890                                 struct ext3_ext_path *path, unsigned long block,
 891                                 unsigned long *count, int *err)
 892 {
 893         struct inode *inode = ext3_ext_base2inode(base);
 894         struct ext3_allocation_request ar;
 895         unsigned long pblock;
 896         int aflags;
 897
 898         /* find neighbour allocated blocks */
 899         ar.lleft = block;
 900         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 901         if (*err)
 902                 return 0;
 903         ar.lright = block;
 904         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 905         if (*err)
 906                 return 0;
 907
 908         /* allocate new block */
 909         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 910         ar.inode = inode;
 911         ar.logical = block;
 912         ar.len = *count;
 913         ar.flags = EXT3_MB_HINT_DATA;
 914         pblock = ext3_mb_new_blocks(handle, &ar, err);
 915         *count = ar.len;
 916         return pblock;
 917 }
 918 #endif
 919
 920 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 921                                   struct ext3_ext_path *path,
 922                                   struct ext3_ext_cache *cex,
 923 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 924                                    struct ext3_extent *ex,
 925 #endif
 926                                   void *cbdata)
 927 {
 928         struct bpointers *bp = cbdata;
 929         struct inode *inode = ext3_ext_base2inode(base);
 930         struct ext3_extent nex;
 931         unsigned long pblock;
 932         unsigned long tgen;
 933         int err, i;
 934         unsigned long count;
 935         handle_t *handle;
 936
 937         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 938                 err = EXT_CONTINUE;
 939                 goto map;
 940         }
 941
 942         if (bp->create == 0) {
 943                 i = 0;
 944                 if (cex->ec_block < bp->start)
 945                         i = bp->start - cex->ec_block;
 946                 if (i >= cex->ec_len)
 947                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 948                                         i, cex->ec_len);
 949                 for (; i < cex->ec_len && bp->num; i++) {
 950                         *(bp->created) = 0;
 951                         bp->created++;
 952                         *(bp->blocks) = 0;
 953                         bp->blocks++;
 954                         bp->num--;
 955                         bp->start++;
 956                 }
 957
 958                 return EXT_CONTINUE;
 959         }
 960
 961         tgen = EXT_GENERATION(base);
 962         count = ext3_ext_calc_credits_for_insert(base, path);
 963
 964         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 965         if (IS_ERR(handle)) {
 966                 return PTR_ERR(handle);
 967         }
 968
 969         if (tgen != EXT_GENERATION(base)) {
 970                 /* the tree has changed. so path can be invalid at moment */
 971                 ext3_journal_stop(handle);
 972                 return EXT_REPEAT;
 973         }
 974
 975         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
 976          * protected by i_data_sem as whole. so we patch it to store
 977          * generation to path and now verify the tree hasn't changed */
 978         down_write((&EXT4_I(inode)->i_data_sem));
 979
 980         /* validate extent, make sure the extent tree does not changed */
 981         if (EXT_GENERATION(base) != path[0].p_generation) {
 982                 /* cex is invalid, try again */
 983                 up_write(&EXT4_I(inode)->i_data_sem);
 984                 ext3_journal_stop(handle);
 985                 return EXT_REPEAT;
 986         }
 987
 988         count = cex->ec_len;
 989         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 990         if (!pblock)
 991                 goto out;
 992         EXT_ASSERT(count <= cex->ec_len);
 993
 994         /* insert new extent */
 995         nex.ee_block = cpu_to_le32(cex->ec_block);
 996         ext3_ext_store_pblock(&nex, pblock);
 997         nex.ee_len = cpu_to_le16(count);
 998         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
 999         if (err) {
1000                 /* free data blocks we just allocated */
1001                 /* not a good idea to call discard here directly,
1002                  * but otherwise we'd need to call it every free() */
1003 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1004                 ext3_mb_discard_inode_preallocations(inode);
1005 #endif
1006                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1007                                  cpu_to_le16(nex.ee_len), 0);
1008                 goto out;
1009         }
1010
1011         /*
1012          * Putting len of the actual extent we just inserted,
1013          * we are asking ext3_ext_walk_space() to continue
1014          * scaning after that block
1015          */
1016         cex->ec_len = le16_to_cpu(nex.ee_len);
1017         cex->ec_start = ext_pblock(&nex);
1018         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1019         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1020
1021 out:
1022         up_write((&EXT4_I(inode)->i_data_sem));
1023         ext3_journal_stop(handle);
1024 map:
1025         if (err >= 0) {
1026                 /* map blocks */
1027                 if (bp->num == 0) {
1028                         CERROR("hmm. why do we find this extent?\n");
1029                         CERROR("initial space: %lu:%u\n",
1030                                 bp->start, bp->init_num);
1031                         CERROR("current extent: %u/%u/%llu %d\n",
1032                                 cex->ec_block, cex->ec_len,
1033                                 (unsigned long long)cex->ec_start,
1034                                 cex->ec_type);
1035                 }
1036                 i = 0;
1037                 if (cex->ec_block < bp->start)
1038                         i = bp->start - cex->ec_block;
1039                 if (i >= cex->ec_len)
1040                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1041                                         i, cex->ec_len);
1042                 for (; i < cex->ec_len && bp->num; i++) {
1043                         *(bp->blocks) = cex->ec_start + i;
1044                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1045                                 *(bp->created) = 0;
1046                         } else {
1047                                 *(bp->created) = 1;
1048                                 /* unmap any possible underlying metadata from
1049                                  * the block device mapping.  bug 6998. */
1050                                 ll_unmap_underlying_metadata(inode->i_sb,
1051                                                              *(bp->blocks));
1052                         }
1053                         bp->created++;
1054                         bp->blocks++;
1055                         bp->num--;
1056                         bp->start++;
1057                 }
1058         }
1059         return err;
1060 }
1061
1062 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1063                        unsigned long num, unsigned long *blocks,
1064                        int *created, int create)
1065 {
1066         struct ext3_ext_base *base = inode;
1067         struct bpointers bp;
1068         int err;
1069
1070         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1071                block, block + num - 1, (unsigned) inode->i_ino);
1072
1073         bp.blocks = blocks;
1074         bp.created = created;
1075         bp.start = block;
1076         bp.init_num = bp.num = num;
1077         bp.create = create;
1078
1079         err = fsfilt_ext3_ext_walk_space(base, block, num,
1080                                          ext3_ext_new_extent_cb, &bp);
1081         ext3_ext_invalidate_cache(base);
1082
1083         return err;
1084 }
1085
1086 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1087                                     int pages, unsigned long *blocks,
1088                                     int *created, int create)
1089 {
1090         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1091         int rc = 0, i = 0;
1092         struct page *fp = NULL;
1093         int clen = 0;
1094
1095         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1096                 inode->i_ino, pages, (*page)->index);
1097
1098         /* pages are sorted already. so, we just have to find
1099          * contig. space and process them properly */
1100         while (i < pages) {
1101                 if (fp == NULL) {
1102                         /* start new extent */
1103                         fp = *page++;
1104                         clen = 1;
1105                         i++;
1106                         continue;
1107                 } else if (fp->index + clen == (*page)->index) {
1108                         /* continue the extent */
1109                         page++;
1110                         clen++;
1111                         i++;
1112                         continue;
1113                 }
1114
1115                 /* process found extent */
1116                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1117                                         clen * blocks_per_page, blocks,
1118                                         created, create);
1119                 if (rc)
1120                         GOTO(cleanup, rc);
1121
1122                 /* look for next extent */
1123                 fp = NULL;
1124                 blocks += blocks_per_page * clen;
1125                 created += blocks_per_page * clen;
1126         }
1127
1128         if (fp)
1129                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1130                                         clen * blocks_per_page, blocks,
1131                                         created, create);
1132 cleanup:
1133         return rc;
1134 }
1135
1136 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1137                                unsigned long *blocks, int *created, int create);
1138 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1139                                    int pages, unsigned long *blocks,
1140                                    int *created, int create)
1141 {
1142         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1143         unsigned long *b;
1144         int rc = 0, i, *cr;
1145
1146         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1147                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1148                 if (rc) {
1149                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1150                                inode->i_ino, *b, *cr, create, rc);
1151                         break;
1152                 }
1153
1154                 b += blocks_per_page;
1155                 cr += blocks_per_page;
1156         }
1157         return rc;
1158 }
1159
1160 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1161                                 int pages, unsigned long *blocks,
1162                                 int *created, int create,
1163                                 struct mutex *optional_mutex)
1164 {
1165         int rc;
1166
1167         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1168                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1169                                                      blocks, created, create);
1170                 return rc;
1171         }
1172         if (optional_mutex != NULL)
1173                 mutex_lock(optional_mutex);
1174         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1175                                             created, create);
1176         if (optional_mutex != NULL)
1177                 mutex_unlock(optional_mutex);
1178
1179         return rc;
1180 }
1181
1182 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1183 {
1184         unsigned long block;
1185         struct buffer_head *bh;
1186         int err, blocksize, csize, boffs, osize = size;
1187
1188         /* prevent reading after eof */
1189         spin_lock(&inode->i_lock);
1190         if (i_size_read(inode) < *offs + size) {
1191                 size = i_size_read(inode) - *offs;
1192                 spin_unlock(&inode->i_lock);
1193                 if (size < 0) {
1194                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1195                                i_size_read(inode), *offs);
1196                         return -EBADR;
1197                 } else if (size == 0) {
1198                         return 0;
1199                 }
1200         } else {
1201                 spin_unlock(&inode->i_lock);
1202         }
1203
1204         blocksize = 1 << inode->i_blkbits;
1205
1206         while (size > 0) {
1207                 block = *offs >> inode->i_blkbits;
1208                 boffs = *offs & (blocksize - 1);
1209                 csize = min(blocksize - boffs, size);
1210                 bh = ext3_bread(NULL, inode, block, 0, &err);
1211                 if (!bh) {
1212                         CERROR("can't read block: %d\n", err);
1213                         return err;
1214                 }
1215
1216                 memcpy(buf, bh->b_data + boffs, csize);
1217                 brelse(bh);
1218
1219                 *offs += csize;
1220                 buf += csize;
1221                 size -= csize;
1222         }
1223         return osize;
1224 }
1225 EXPORT_SYMBOL(fsfilt_ext3_read);
1226
1227 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1228                                    int size, loff_t *offs)
1229 {
1230         int rc;
1231         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1232         if (rc > 0)
1233                 rc = 0;
1234         return rc;
1235 }
1236
1237 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1238                                 loff_t *offs, handle_t *handle)
1239 {
1240         struct buffer_head *bh = NULL;
1241         loff_t old_size = i_size_read(inode), offset = *offs;
1242         loff_t new_size = i_size_read(inode);
1243         unsigned long block;
1244         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1245
1246         while (bufsize > 0) {
1247                 if (bh != NULL)
1248                         brelse(bh);
1249
1250                 block = offset >> inode->i_blkbits;
1251                 boffs = offset & (blocksize - 1);
1252                 size = min(blocksize - boffs, bufsize);
1253                 bh = ext3_bread(handle, inode, block, 1, &err);
1254                 if (!bh) {
1255                         CERROR("can't read/create block: %d\n", err);
1256                         break;
1257                 }
1258
1259                 err = ext3_journal_get_write_access(handle, bh);
1260                 if (err) {
1261                         CERROR("journal_get_write_access() returned error %d\n",
1262                                err);
1263                         break;
1264                 }
1265                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1266                 memcpy(bh->b_data + boffs, buf, size);
1267                 err = ext3_journal_dirty_metadata(handle, bh);
1268                 if (err) {
1269                         CERROR("journal_dirty_metadata() returned error %d\n",
1270                                err);
1271                         break;
1272                 }
1273                 if (offset + size > new_size)
1274                         new_size = offset + size;
1275                 offset += size;
1276                 bufsize -= size;
1277                 buf += size;
1278         }
1279         if (bh)
1280                 brelse(bh);
1281
1282         /* correct in-core and on-disk sizes */
1283         if (new_size > i_size_read(inode)) {
1284                 spin_lock(&inode->i_lock);
1285                 if (new_size > i_size_read(inode))
1286                         i_size_write(inode, new_size);
1287                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1288                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1289                 if (i_size_read(inode) > old_size) {
1290                         spin_unlock(&inode->i_lock);
1291                         mark_inode_dirty(inode);
1292                 } else {
1293                         spin_unlock(&inode->i_lock);
1294                 }
1295         }
1296
1297         if (err == 0)
1298                 *offs = offset;
1299         return err;
1300 }
1301 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1302
1303 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1304                                     loff_t *offs, int force_sync)
1305 {
1306         struct inode *inode = file->f_dentry->d_inode;
1307         handle_t *handle;
1308         int err, block_count = 0, blocksize;
1309
1310         /* Determine how many transaction credits are needed */
1311         blocksize = 1 << inode->i_blkbits;
1312         block_count = (*offs & (blocksize - 1)) + bufsize;
1313         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1314
1315         handle = ext3_journal_start(inode,
1316                         block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1317         if (IS_ERR(handle)) {
1318                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1319                        block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
1320                        bufsize);
1321                 return PTR_ERR(handle);
1322         }
1323
1324         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1325
1326         if (!err && force_sync)
1327                 handle->h_sync = 1; /* recovery likes this */
1328
1329         ext3_journal_stop(handle);
1330
1331         return err;
1332 }
1333
1334 static int fsfilt_ext3_setup(struct super_block *sb)
1335 {
1336         if (!EXT3_HAS_COMPAT_FEATURE(sb,
1337                                 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1338                 CERROR("ext3 mounted without journal\n");
1339                 return -EINVAL;
1340         }
1341
1342 #ifdef S_PDIROPS
1343         CWARN("Enabling PDIROPS\n");
1344         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1345         sb->s_flags |= S_PDIROPS;
1346 #endif
1347         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1348                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1349         return 0;
1350 }
1351
1352 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1353    objects. Logs is number of logfiles to update */
1354 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1355 {
1356         if ( !fso ) {
1357                 switch(op) {
1358                 case FSFILT_OP_CREATE:
1359                                  /* directory leaf, index & indirect & EA*/
1360                         return 4 + 3 * logs;
1361                 case FSFILT_OP_UNLINK:
1362                         return 3 * logs;
1363                 }
1364         } else {
1365                 int i;
1366                 int needed = 0;
1367                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1368                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1369                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1370                 for (i = 0; i < op; i++, fso++) {
1371                         int nblocks = fso->fso_bufcnt * blockpp;
1372                         int ndindirect = min(nblocks, addrpp + 1);
1373                         int nindir = nblocks + ndindirect + 1;
1374
1375                         needed += nindir;
1376                 }
1377                 return needed + 3 * logs;
1378         }
1379
1380         return 0;
1381 }
1382
1383 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
1384 {
1385         return (EXT3_SB(sb)->journal_bdev);
1386 }
1387 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
1388
1389 static int ll_decode_fh_accept(void *context, struct dentry *de)
1390 {
1391         return 1;
1392 }
1393
1394 #ifdef HAVE_EXPORTFS_DECODE_FH
1395 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
1396          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
1397                             acceptable, context)
1398 #else
1399 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
1400          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
1401                                      type, acceptable, context)
1402 # define FILEID_INO32_GEN 1
1403 extern struct export_operations export_op_default;
1404 #endif
1405
1406 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
1407                                       struct fsfilt_fid *fid, int ignore_gen)
1408 {
1409         struct inode  *inode;
1410         struct dentry *result;
1411
1412         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
1413                                        ll_decode_fh_accept, NULL);
1414         if (IS_ERR(result)) {
1415                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
1416                        fid->ino, fid->gen, PTR_ERR(result));
1417                 return result;
1418         }
1419
1420         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
1421                fid->ino, fid->gen);
1422         inode = result->d_inode;
1423         if (inode == NULL)
1424                 goto err_out;
1425
1426         if (inode->i_nlink == 0 &&
1427             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
1428                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
1429                               " ctime -- this may indicate disk "
1430                               "corruption (inode: %lu, link: %lu, "
1431                               "count: %d)\n", inode->i_ino,
1432                               (unsigned long)inode->i_nlink,
1433                               atomic_read(&inode->i_count));
1434                 goto err_out;
1435         }
1436         if (fid->gen && inode->i_generation != fid->gen) {
1437                 /* we didn't find the right inode.. */
1438                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
1439                        "count: %d, generation %u/%u\n",
1440                        inode->i_ino, (unsigned long)inode->i_nlink,
1441                        atomic_read(&inode->i_count), inode->i_generation,
1442                        fid->gen);
1443                 goto err_out;
1444         }
1445
1446         return result;
1447 err_out:
1448         l_dput(result);
1449         return ERR_PTR(-ENOENT);
1450 }
1451
1452 static struct fsfilt_operations fsfilt_ext3_ops = {
1453         .fs_type                = "ext3",
1454         .fs_owner               = THIS_MODULE,
1455         .fs_getlabel            = fsfilt_ext3_get_label,
1456         .fs_setlabel            = fsfilt_ext3_set_label,
1457         .fs_uuid                = fsfilt_ext3_uuid,
1458         .fs_start               = fsfilt_ext3_start,
1459         .fs_brw_start           = fsfilt_ext3_brw_start,
1460         .fs_extend              = fsfilt_ext3_extend,
1461         .fs_commit              = fsfilt_ext3_commit,
1462         .fs_commit_async        = fsfilt_ext3_commit_async,
1463         .fs_commit_wait         = fsfilt_ext3_commit_wait,
1464         .fs_setattr             = fsfilt_ext3_setattr,
1465         .fs_iocontrol           = fsfilt_ext3_iocontrol,
1466         .fs_set_md              = fsfilt_ext3_set_md,
1467         .fs_get_md              = fsfilt_ext3_get_md,
1468         .fs_readpage            = fsfilt_ext3_readpage,
1469         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
1470         .fs_statfs              = fsfilt_ext3_statfs,
1471         .fs_sync                = fsfilt_ext3_sync,
1472         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
1473         .fs_write_record        = fsfilt_ext3_write_record,
1474         .fs_read_record         = fsfilt_ext3_read_record,
1475         .fs_setup               = fsfilt_ext3_setup,
1476         .fs_send_bio            = fsfilt_ext3_send_bio,
1477         .fs_get_op_len          = fsfilt_ext3_get_op_len,
1478 #ifdef HAVE_DISK_INODE_VERSION
1479         .fs_get_version         = fsfilt_ext3_get_version,
1480         .fs_set_version         = fsfilt_ext3_set_version,
1481 #endif
1482         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
1483         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
1484 };
1485
1486 static int __init fsfilt_ext3_init(void)
1487 {
1488         int rc;
1489
1490         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
1491                                          sizeof(struct fsfilt_cb_data), 0, 0);
1492         if (!fcb_cache) {
1493                 CERROR("error allocating fsfilt journal callback cache\n");
1494                 GOTO(out, rc = -ENOMEM);
1495         }
1496
1497         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
1498
1499         if (rc) {
1500                 int err = cfs_mem_cache_destroy(fcb_cache);
1501                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
1502         }
1503 out:
1504         return rc;
1505 }
1506
1507 static void __exit fsfilt_ext3_exit(void)
1508 {
1509         int rc;
1510
1511         fsfilt_unregister_ops(&fsfilt_ext3_ops);
1512         rc = cfs_mem_cache_destroy(fcb_cache);
1513         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
1514 }
1515
1516 module_init(fsfilt_ext3_init);
1517 module_exit(fsfilt_ext3_exit);
1518
1519 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1520 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
1521 MODULE_LICENSE("GPL");