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 #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 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
  96 # define fsfilt_journal_callback_set(handle, func, jcb) \
  97          journal_callback_set(handle, func, jcb)
  98 #else
  99 # error missing journal commit callback
 100 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
 101
 102 static cfs_mem_cache_t *fcb_cache;
 103
 104 struct fsfilt_cb_data {
 105         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 106         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 107         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 108         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 109         void *cb_data;                  /* MDS/OST completion function data */
 110 };
 111
 112 #ifndef ext3_find_next_bit
 113 #define ext3_find_next_bit           ext2_find_next_bit
 114 #endif
 115
 116 #ifndef ext2_find_next_bit
 117 #ifdef __LITTLE_ENDIAN
 118 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 119 #else
 120 error "Need implementation of find_next_bit on big-endian systems"
 121 #endif  /* __LITTLE_ENDIAN */
 122 #endif  /* !ext2_find_next_le_bit */
 123
 124 static char *fsfilt_ext3_get_label(struct super_block *sb)
 125 {
 126         return EXT3_SB(sb)->s_es->s_volume_name;
 127 }
 128
 129 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 130 {
 131         /* see e.g. fsfilt_ext3_write_record() */
 132         journal_t *journal;
 133         handle_t *handle;
 134         int err;
 135
 136         journal = EXT3_SB(sb)->s_journal;
 137         handle = ext3_journal_start_sb(sb, 1);
 138         if (IS_ERR(handle)) {
 139                 CERROR("can't start transaction\n");
 140                 return(PTR_ERR(handle));
 141         }
 142
 143         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 144         if (err)
 145                 goto out;
 146
 147         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 148                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 149
 150         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 151
 152 out:
 153         ext3_journal_stop(handle);
 154
 155         return(err);
 156 }
 157
 158 static char *fsfilt_ext3_uuid(struct super_block *sb)
 159 {
 160         return EXT3_SB(sb)->s_es->s_uuid;
 161 }
 162
 163 #ifdef HAVE_DISK_INODE_VERSION
 164
 165 static __u64 get_i_version(struct inode *inode)
 166 {
 167         return EXT3_I(inode)->i_fs_version;
 168 }
 169
 170 static void set_i_version(struct inode *inode, __u64 new_version)
 171 {
 172         (EXT3_I(inode))->i_fs_version = new_version;
 173 }
 174
 175 /*
 176  * Get the 64-bit version for an inode.
 177  */
 178 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 179 {
 180         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 181                get_i_version(inode), inode->i_ino);
 182         return get_i_version(inode);
 183 }
 184
 185 /*
 186  * Set the 64-bit version and return the old version.
 187  */
 188 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 189 {
 190         __u64 old_version = get_i_version(inode);
 191
 192         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 193                new_version, old_version, inode->i_ino);
 194         set_i_version(inode, new_version);
 195         /* version is set after all inode operations are finished, so we should
 196          * mark it dirty here */
 197         inode->i_sb->s_op->dirty_inode(inode);
 198         return old_version;
 199 }
 200
 201 #endif
 202
 203 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
 204 #ifdef HAVE_BLOCKS_FOR_TRUNCATE
 205 # include <ext4/truncate.h>
 206 #else
 207 static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
 208 {
 209         ext4_lblk_t needed;
 210
 211         needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
 212         if (needed < 2)
 213                 needed = 2;
 214         if (needed > EXT4_MAX_TRANS_DATA)
 215                 needed = EXT4_MAX_TRANS_DATA;
 216         return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
 217 }
 218 #endif
 219
 220 /*
 221  * We don't currently need any additional blocks for rmdir and
 222  * unlink transactions because we are storing the OST oa_id inside
 223  * the inode (which we will be changing anyways as part of this
 224  * transaction).
 225  */
 226 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 227                                int logs)
 228 {
 229         /* For updates to the last received file */
 230         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 231         journal_t *journal;
 232         void *handle;
 233
 234         if (current->journal_info) {
 235                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 236                        current->journal_info);
 237                 goto journal_start;
 238         }
 239
 240         switch(op) {
 241         case FSFILT_OP_RMDIR:
 242         case FSFILT_OP_UNLINK:
 243                 /* delete one file + create/update logs for each stripe */
 244                 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
 245                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 246                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 247                 break;
 248         case FSFILT_OP_RENAME:
 249                 /* modify additional directory */
 250                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 251                 /* no break */
 252         case FSFILT_OP_SYMLINK:
 253                 /* additional block + block bitmap + GDT for long symlink */
 254                 nblocks += 3;
 255                 /* no break */
 256         case FSFILT_OP_CREATE: {
 257                 /* no break */
 258         }
 259         case FSFILT_OP_MKDIR:
 260         case FSFILT_OP_MKNOD:
 261                 /* modify one inode + block bitmap + GDT */
 262                 nblocks += 3;
 263                 /* no break */
 264         case FSFILT_OP_LINK:
 265                 /* modify parent directory */
 266                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 267                            EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
 268                 /* create/update logs for each stripe */
 269                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 270                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 271                 break;
 272         case FSFILT_OP_SETATTR:
 273                 /* Setattr on inode */
 274                 nblocks += 1;
 275                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 276                            EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
 277                 /* quota chown log for each stripe */
 278                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 279                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 280                 break;
 281         case FSFILT_OP_CANCEL_UNLINK:
 282                 LASSERT(logs == 1);
 283
 284                 /* blocks for log header bitmap update OR
 285                  * blocks for catalog header bitmap update + unlink of logs +
 286                  * blocks for delete the inode (include blocks truncating). */
 287                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 288                           EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
 289                           ext4_blocks_for_truncate(inode) + 3;
 290                 break;
 291         default: CERROR("unknown transaction start op %d\n", op);
 292                 LBUG();
 293         }
 294
 295         LASSERT(current->journal_info == desc_private);
 296         journal = EXT3_SB(inode->i_sb)->s_journal;
 297         if (nblocks > journal->j_max_transaction_buffers) {
 298                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 299                        nblocks, op, logs, journal->j_max_transaction_buffers);
 300                 nblocks = journal->j_max_transaction_buffers;
 301         }
 302
 303  journal_start:
 304         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 305         handle = ext3_journal_start(inode, nblocks);
 306
 307         if (!IS_ERR(handle))
 308                 LASSERT(current->journal_info == handle);
 309         else
 310                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 311                        op, nblocks, PTR_ERR(handle));
 312         return handle;
 313 }
 314
 315 /*
 316  * Calculate the number of buffer credits needed to write multiple pages in
 317  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 318  * doesn't have a nice API for calculating this sort of thing in advance.
 319  *
 320  * See comment above ext3_writepage_trans_blocks for details.  We assume
 321  * no data journaling is being done, but it does allow for all of the pages
 322  * being non-contiguous.  If we are guaranteed contiguous pages we could
 323  * reduce the number of (d)indirect blocks a lot.
 324  *
 325  * With N blocks per page and P pages, for each inode we have at most:
 326  * N*P indirect
 327  * min(N*P, blocksize/4 + 1) dindirect blocks
 328  * niocount tindirect
 329  *
 330  * For the entire filesystem, we have at most:
 331  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 332  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 333  * objcount inode blocks
 334  * 1 superblock
 335  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 336  *
 337  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 338  */
 339 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 340                                       int niocount, struct niobuf_local *nb)
 341 {
 342         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 343         __u64 next_indir;
 344         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 345         int nbitmaps = 0, ngdblocks;
 346         int needed = objcount + 1; /* inodes + superblock */
 347         int i, j;
 348
 349         for (i = 0, j = 0; i < objcount; i++, fso++) {
 350                 /* two or more dindirect blocks in case we cross boundary */
 351                 int ndind =
 352                         (long)((nb[j + fso->fso_bufcnt - 1].lnb_file_offset -
 353                                 nb[j].lnb_file_offset) >>
 354                                sb->s_blocksize_bits) /
 355                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 356                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 357
 358                 /* leaf, indirect, tindirect blocks for first block */
 359                 nbitmaps += blockpp + 2;
 360
 361                 j += fso->fso_bufcnt;
 362         }
 363
 364         next_indir = nb[0].lnb_file_offset +
 365                      (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 366         for (i = 1; i < niocount; i++) {
 367                 if (nb[i].lnb_file_offset >= next_indir) {
 368                         nbitmaps++;     /* additional indirect */
 369                         next_indir = nb[i].lnb_file_offset +
 370                                      (EXT3_ADDR_PER_BLOCK(sb) <<
 371                                       sb->s_blocksize_bits);
 372                 } else if (nb[i].lnb_file_offset !=
 373                            nb[i - 1].lnb_file_offset + sb->s_blocksize) {
 374                         nbitmaps++;     /* additional indirect */
 375                 }
 376                 nbitmaps += blockpp;    /* each leaf in different group? */
 377         }
 378
 379         ngdblocks = nbitmaps;
 380         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 381                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 382         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 383                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 384
 385         needed += nbitmaps + ngdblocks;
 386
 387         /* last_rcvd update */
 388         needed += EXT3_DATA_TRANS_BLOCKS(sb);
 389
 390 #if defined(CONFIG_QUOTA)
 391         /* We assume that there will be 1 bit set in s_dquot.flags for each
 392          * quota file that is active.  This is at least true for now.
 393          */
 394         needed += hweight32(sb_any_quota_loaded(sb)) *
 395                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 396 #endif
 397
 398         return needed;
 399 }
 400
 401 /* We have to start a huge journal transaction here to hold all of the
 402  * metadata for the pages being written here.  This is necessitated by
 403  * the fact that we do lots of prepare_write operations before we do
 404  * any of the matching commit_write operations, so even if we split
 405  * up to use "smaller" transactions none of them could complete until
 406  * all of them were opened.  By having a single journal transaction,
 407  * we eliminate duplicate reservations for common blocks like the
 408  * superblock and group descriptors or bitmaps.
 409  *
 410  * We will start the transaction here, but each prepare_write will
 411  * add a refcount to the transaction, and each commit_write will
 412  * remove a refcount.  The transaction will be closed when all of
 413  * the pages have been written.
 414  */
 415 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 416                                    int niocount, struct niobuf_local *nb,
 417                                    void *desc_private, int logs)
 418 {
 419         journal_t *journal;
 420         handle_t *handle;
 421         int needed;
 422         ENTRY;
 423
 424         LASSERT(current->journal_info == desc_private);
 425         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 426         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 427
 428         /* The number of blocks we could _possibly_ dirty can very large.
 429          * We reduce our request if it is absurd (and we couldn't get that
 430          * many credits for a single handle anyways).
 431          *
 432          * At some point we have to limit the size of I/Os sent at one time,
 433          * increase the size of the journal, or we have to calculate the
 434          * actual journal requirements more carefully by checking all of
 435          * the blocks instead of being maximally pessimistic.  It remains to
 436          * be seen if this is a real problem or not.
 437          */
 438         if (needed > journal->j_max_transaction_buffers) {
 439                 CERROR("want too many journal credits (%d) using %d instead\n",
 440                        needed, journal->j_max_transaction_buffers);
 441                 needed = journal->j_max_transaction_buffers;
 442         }
 443
 444         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 445         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 446         if (IS_ERR(handle)) {
 447                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 448                        PTR_ERR(handle));
 449         } else {
 450                 LASSERT(handle->h_buffer_credits >= needed);
 451                 LASSERT(current->journal_info == handle);
 452         }
 453
 454         RETURN(handle);
 455 }
 456
 457 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 458 {
 459        handle_t *handle = h;
 460
 461        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 462        if (nblocks == 0) {
 463                handle->h_buffer_credits = 0;
 464                CWARN("setting credits of handle %p to zero by request\n", h);
 465        }
 466
 467        if (handle->h_buffer_credits > nblocks)
 468                 return 0;
 469        if (ext3_journal_extend(handle, nblocks) == 0)
 470                 return 0;
 471
 472        ext3_mark_inode_dirty(handle, inode);
 473        return ext3_journal_restart(handle, nblocks);
 474 }
 475
 476 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 477 {
 478         int rc;
 479         handle_t *handle = h;
 480
 481         LASSERT(current->journal_info == handle);
 482         if (force_sync)
 483                 handle->h_sync = 1; /* recovery likes this */
 484
 485         rc = ext3_journal_stop(handle);
 486
 487         return rc;
 488 }
 489
 490 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 491                                     void **wait_handle)
 492 {
 493         unsigned long tid;
 494         transaction_t *transaction;
 495         handle_t *handle = h;
 496         journal_t *journal;
 497         int rc;
 498
 499         LASSERT(current->journal_info == handle);
 500
 501         transaction = handle->h_transaction;
 502         journal = transaction->t_journal;
 503         tid = transaction->t_tid;
 504         /* we don't want to be blocked */
 505         handle->h_sync = 0;
 506         rc = ext3_journal_stop(handle);
 507         if (rc) {
 508                 CERROR("error while stopping transaction: %d\n", rc);
 509                 return rc;
 510         }
 511         fsfilt_log_start_commit(journal, tid);
 512
 513         *wait_handle = (void *) tid;
 514         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 515         return 0;
 516 }
 517
 518 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 519 {
 520         journal_t *journal = EXT3_JOURNAL(inode);
 521         tid_t tid = (tid_t)(long)h;
 522
 523         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 524         if (unlikely(is_journal_aborted(journal)))
 525                 return -EIO;
 526
 527         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 528
 529         if (unlikely(is_journal_aborted(journal)))
 530                 return -EIO;
 531         return 0;
 532 }
 533
 534 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 535                                struct iattr *iattr, int do_trunc)
 536 {
 537         struct inode *inode = dentry->d_inode;
 538         int rc = 0;
 539
 540 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
 541         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
 542          * timestamps to appear to be in the far future, due old timestamp
 543          * being stored on disk as an unsigned value.  This fixes up any
 544          * bad values held by the client before storing them on disk,
 545          * and ensures any timestamp updates are correct.  LU-1042 */
 546         if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
 547                      !(iattr->ia_valid & ATTR_ATIME))) {
 548                 iattr->ia_valid |= ATTR_ATIME;
 549                 LTIME_S(iattr->ia_atime) = 0;
 550         }
 551         if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
 552                      !(iattr->ia_valid & ATTR_MTIME))) {
 553                 iattr->ia_valid |= ATTR_MTIME;
 554                 LTIME_S(iattr->ia_mtime) = 0;
 555         }
 556         if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
 557                       LTIME_S(inode->i_ctime) == 0) &&
 558                      !(iattr->ia_valid & ATTR_CTIME))) {
 559                 iattr->ia_valid |= ATTR_CTIME;
 560                 LTIME_S(iattr->ia_ctime) = 0;
 561         }
 562 #else
 563 #warning "remove old LU-221/LU-1042 workaround code"
 564 #endif
 565
 566         /* When initializating timestamps for new inodes, use the filesystem
 567          * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
 568          * that this is potentially an invalid inode.  Files with an old ctime
 569          * migrated to a newly-formatted OST with a newer s_mkfs_time will not
 570          * hit this check, since it is only for ctime == 0.  LU-1010/LU-1042 */
 571         if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
 572                 LTIME_S(iattr->ia_ctime) =
 573                         EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
 574
 575         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 576          * We are already writing the inode to disk as part of this
 577          * transaction and want to avoid a lot of extra inode writeout
 578          * later on. b=9828 */
 579         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 580                 /* ATTR_SIZE would invoke truncate: clear it */
 581                 iattr->ia_valid &= ~ATTR_SIZE;
 582                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 583                 i_size_write(inode, iattr->ia_size);
 584
 585                 if (iattr->ia_valid & ATTR_UID)
 586                         inode->i_uid = iattr->ia_uid;
 587                 if (iattr->ia_valid & ATTR_GID)
 588                         inode->i_gid = iattr->ia_gid;
 589                 if (iattr->ia_valid & ATTR_ATIME)
 590                         inode->i_atime = iattr->ia_atime;
 591                 if (iattr->ia_valid & ATTR_MTIME)
 592                         inode->i_mtime = iattr->ia_mtime;
 593                 if (iattr->ia_valid & ATTR_CTIME)
 594                         inode->i_ctime = iattr->ia_ctime;
 595                 if (iattr->ia_valid & ATTR_MODE) {
 596                         inode->i_mode = iattr->ia_mode;
 597
 598                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 599                             !cfs_capable(CFS_CAP_FSETID))
 600                                 inode->i_mode &= ~S_ISGID;
 601                 }
 602
 603                 inode->i_sb->s_op->dirty_inode(inode);
 604
 605                 goto out;
 606         }
 607
 608         /* Don't allow setattr to change file type */
 609         if (iattr->ia_valid & ATTR_MODE)
 610                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 611                                  (iattr->ia_mode & ~S_IFMT);
 612
 613         /* We set these flags on the client, but have already checked perms
 614          * so don't confuse inode_change_ok. */
 615         iattr->ia_valid &= ~TIMES_SET_FLAGS;
 616
 617         if (inode->i_op->setattr) {
 618                 rc = inode->i_op->setattr(dentry, iattr);
 619         } else {
 620 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
 621                 rc = inode_change_ok(inode, iattr);
 622                 if (!rc)
 623 #endif
 624                         rc = simple_setattr(dentry, iattr);
 625         }
 626
 627  out:
 628         RETURN(rc);
 629 }
 630
 631 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 632                                  unsigned int cmd, unsigned long arg)
 633 {
 634         int rc = 0;
 635         ENTRY;
 636
 637         /* FIXME: Can't do this because of nested transaction deadlock */
 638         if (cmd == EXT3_IOC_SETFLAGS) {
 639                 /* We can't enable data journaling on OST objects, because
 640                 * this forces the transaction to be closed in order to
 641                 * flush the journal, but the caller will already have a
 642                 * compound transaction open to update the last_rcvd file,
 643                 * and this thread would deadlock trying to set the flag. */
 644                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 645                         CERROR("can't set data journal flag on file\n");
 646                         RETURN(-EPERM);
 647                 }
 648                 /* Because the MDS does not see the EXTENTS_FL set on the
 649                  * OST objects, mask this flag into all set flags.  It is
 650                  * not legal to clear this flag in any case, so we are not
 651                  * changing the functionality by doing this.  b=22911 */
 652                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 653         }
 654
 655         /* ext4_ioctl does not have a inode argument */
 656         if (inode->i_fop->unlocked_ioctl)
 657                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 658         else
 659                 RETURN(-ENOTTY);
 660
 661         RETURN(rc);
 662 }
 663
 664 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 665                               void *lmm, int lmm_size, const char *name)
 666 {
 667         int rc;
 668
 669         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 670
 671         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 672                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 673
 674         if (rc && rc != -EROFS)
 675                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 676                        inode->i_ino, rc);
 677         return rc;
 678 }
 679
 680 /* Must be called with i_mutex held */
 681 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 682                               const char *name)
 683 {
 684         int rc;
 685
 686         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 687
 688         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 689                             name, lmm, lmm_size);
 690
 691         /* This gives us the MD size */
 692         if (lmm == NULL)
 693                 return (rc == -ENODATA) ? 0 : rc;
 694
 695         if (rc < 0) {
 696                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 697                        EXT3_XATTR_INDEX_TRUSTED, name,
 698                        inode->i_ino, rc);
 699                 memset(lmm, 0, lmm_size);
 700                 return (rc == -ENODATA) ? 0 : rc;
 701         }
 702
 703         return rc;
 704 }
 705
 706 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 707 {
 708         submit_bio(rw, bio);
 709         return 0;
 710 }
 711
 712 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 713                                     loff_t *off)
 714 {
 715         struct inode *inode = file->f_dentry->d_inode;
 716         int rc = 0;
 717
 718         if (S_ISREG(inode->i_mode))
 719                 rc = file->f_op->read(file, buf, count, off);
 720         else {
 721                 const int blkbits = inode->i_sb->s_blocksize_bits;
 722                 const int blksize = inode->i_sb->s_blocksize;
 723
 724                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 725                        (unsigned long)count, inode->i_ino, *off);
 726                 while (count > 0) {
 727                         struct buffer_head *bh;
 728
 729                         bh = NULL;
 730                         if (*off < i_size_read(inode)) {
 731                                 int err = 0;
 732
 733                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 734                                                 0, &err);
 735
 736                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 737
 738                                 if (bh) {
 739                                         memcpy(buf, bh->b_data, blksize);
 740                                         brelse(bh);
 741                                 } else if (err) {
 742                                         /* XXX in theory we should just fake
 743                                          * this buffer and continue like ext3,
 744                                          * especially if this is a partial read
 745                                          */
 746                                         CERROR("error read dir %lu+%llu: %d\n",
 747                                                inode->i_ino, *off, err);
 748                                         RETURN(err);
 749                                 }
 750                         }
 751                         if (!bh) {
 752                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 753
 754                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 755                                 memset(fake, 0, sizeof(*fake));
 756                                 fake->rec_len = cpu_to_le16(blksize);
 757                         }
 758                         count -= blksize;
 759                         buf += blksize;
 760                         *off += blksize;
 761                         rc += blksize;
 762                 }
 763         }
 764
 765         return rc;
 766 }
 767
 768 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 769 static void fsfilt_ext3_cb_func(struct super_block *sb,
 770                                 struct journal_callback *jcb, int error)
 771 #else
 772 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 773 #endif
 774 {
 775         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 776
 777         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 778
 779         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 780 }
 781
 782 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 783                                       void *handle, fsfilt_cb_t cb_func,
 784                                       void *cb_data)
 785 {
 786         struct fsfilt_cb_data *fcb;
 787
 788         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 789         if (fcb == NULL)
 790                 RETURN(-ENOMEM);
 791
 792         fcb->cb_func = cb_func;
 793         fcb->cb_obd = obd;
 794         fcb->cb_last_rcvd = last_rcvd;
 795         fcb->cb_data = cb_data;
 796
 797         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 798         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 799
 800         return 0;
 801 }
 802
 803 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 804 {
 805         struct kstatfs sfs;
 806         int rc;
 807
 808         memset(&sfs, 0, sizeof(sfs));
 809         rc = sb->s_op->statfs(sb->s_root, &sfs);
 810         statfs_pack(osfs, &sfs);
 811         return rc;
 812 }
 813
 814 static int fsfilt_ext3_sync(struct super_block *sb)
 815 {
 816         return ext3_force_commit(sb);
 817 }
 818
 819 #ifndef EXT3_EXTENTS_FL
 820 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 821 #endif
 822
 823 #ifndef EXT_ASSERT
 824 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 825 #endif
 826
 827 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 828 #define ext3_ext_base                   inode
 829 #define ext3_ext_base2inode(inode)      (inode)
 830 #define EXT_DEPTH(inode)                ext_depth(inode)
 831 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 832                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 833
 834 struct bpointers {
 835         unsigned long *blocks;
 836         int *created;
 837         unsigned long start;
 838         int num;
 839         int init_num;
 840         int create;
 841 };
 842
 843 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 844                                unsigned long block, int *aflags)
 845 {
 846         struct ext3_inode_info *ei = EXT3_I(inode);
 847         unsigned long bg_start;
 848         unsigned long colour;
 849         int depth;
 850
 851         if (path) {
 852                 struct ext3_extent *ex;
 853                 depth = path->p_depth;
 854
 855                 /* try to predict block placement */
 856                 if ((ex = path[depth].p_ext))
 857                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 858
 859                 /* it looks index is empty
 860                  * try to find starting from index itself */
 861                 if (path[depth].p_bh)
 862                         return path[depth].p_bh->b_blocknr;
 863         }
 864
 865         /* OK. use inode's group */
 866         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 867                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 868         colour = (current->pid % 16) *
 869                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 870         return bg_start + colour + block;
 871 }
 872
 873 #define ll_unmap_underlying_metadata(sb, blocknr) \
 874         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 875
 876 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 877 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 878                                 struct ext3_ext_path *path, unsigned long block,
 879                                 unsigned long *count, int *err)
 880 {
 881         unsigned long pblock, goal;
 882         int aflags = 0;
 883         struct inode *inode = ext3_ext_base2inode(base);
 884
 885         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 886         aflags |= 2; /* block have been already reserved */
 887         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 888         return pblock;
 889
 890 }
 891 #else
 892 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 893                                 struct ext3_ext_path *path, unsigned long block,
 894                                 unsigned long *count, int *err)
 895 {
 896         struct inode *inode = ext3_ext_base2inode(base);
 897         struct ext3_allocation_request ar;
 898         unsigned long pblock;
 899         int aflags;
 900
 901         /* find neighbour allocated blocks */
 902         ar.lleft = block;
 903         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 904         if (*err)
 905                 return 0;
 906         ar.lright = block;
 907         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 908         if (*err)
 909                 return 0;
 910
 911         /* allocate new block */
 912         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 913         ar.inode = inode;
 914         ar.logical = block;
 915         ar.len = *count;
 916         ar.flags = EXT3_MB_HINT_DATA;
 917         pblock = ext3_mb_new_blocks(handle, &ar, err);
 918         *count = ar.len;
 919         return pblock;
 920 }
 921 #endif
 922
 923 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 924                                   struct ext3_ext_path *path,
 925                                   struct ext3_ext_cache *cex,
 926 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 927                                    struct ext3_extent *ex,
 928 #endif
 929                                   void *cbdata)
 930 {
 931         struct bpointers *bp = cbdata;
 932         struct inode *inode = ext3_ext_base2inode(base);
 933         struct ext3_extent nex;
 934         unsigned long pblock;
 935         unsigned long tgen;
 936         int err, i;
 937         unsigned long count;
 938         handle_t *handle;
 939
 940         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 941                 err = EXT_CONTINUE;
 942                 goto map;
 943         }
 944
 945         if (bp->create == 0) {
 946                 i = 0;
 947                 if (cex->ec_block < bp->start)
 948                         i = bp->start - cex->ec_block;
 949                 if (i >= cex->ec_len)
 950                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 951                                         i, cex->ec_len);
 952                 for (; i < cex->ec_len && bp->num; i++) {
 953                         *(bp->created) = 0;
 954                         bp->created++;
 955                         *(bp->blocks) = 0;
 956                         bp->blocks++;
 957                         bp->num--;
 958                         bp->start++;
 959                 }
 960
 961                 return EXT_CONTINUE;
 962         }
 963
 964         tgen = EXT_GENERATION(base);
 965         count = ext3_ext_calc_credits_for_insert(base, path);
 966
 967         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 968         if (IS_ERR(handle)) {
 969                 return PTR_ERR(handle);
 970         }
 971
 972         if (tgen != EXT_GENERATION(base)) {
 973                 /* the tree has changed. so path can be invalid at moment */
 974                 ext3_journal_stop(handle);
 975                 return EXT_REPEAT;
 976         }
 977
 978         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
 979          * protected by i_data_sem as whole. so we patch it to store
 980          * generation to path and now verify the tree hasn't changed */
 981         down_write((&EXT4_I(inode)->i_data_sem));
 982
 983         /* validate extent, make sure the extent tree does not changed */
 984         if (EXT_GENERATION(base) != path[0].p_generation) {
 985                 /* cex is invalid, try again */
 986                 up_write(&EXT4_I(inode)->i_data_sem);
 987                 ext3_journal_stop(handle);
 988                 return EXT_REPEAT;
 989         }
 990
 991         count = cex->ec_len;
 992         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 993         if (!pblock)
 994                 goto out;
 995         EXT_ASSERT(count <= cex->ec_len);
 996
 997         /* insert new extent */
 998         nex.ee_block = cpu_to_le32(cex->ec_block);
 999         ext3_ext_store_pblock(&nex, pblock);
1000         nex.ee_len = cpu_to_le16(count);
1001         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1002         if (err) {
1003                 /* free data blocks we just allocated */
1004                 /* not a good idea to call discard here directly,
1005                  * but otherwise we'd need to call it every free() */
1006 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1007                 ext3_mb_discard_inode_preallocations(inode);
1008 #endif
1009                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1010                                  cpu_to_le16(nex.ee_len), 0);
1011                 goto out;
1012         }
1013
1014         /*
1015          * Putting len of the actual extent we just inserted,
1016          * we are asking ext3_ext_walk_space() to continue
1017          * scaning after that block
1018          */
1019         cex->ec_len = le16_to_cpu(nex.ee_len);
1020         cex->ec_start = ext_pblock(&nex);
1021         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1022         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1023
1024 out:
1025         up_write((&EXT4_I(inode)->i_data_sem));
1026         ext3_journal_stop(handle);
1027 map:
1028         if (err >= 0) {
1029                 /* map blocks */
1030                 if (bp->num == 0) {
1031                         CERROR("hmm. why do we find this extent?\n");
1032                         CERROR("initial space: %lu:%u\n",
1033                                 bp->start, bp->init_num);
1034                         CERROR("current extent: %u/%u/%llu %d\n",
1035                                 cex->ec_block, cex->ec_len,
1036                                 (unsigned long long)cex->ec_start,
1037                                 cex->ec_type);
1038                 }
1039                 i = 0;
1040                 if (cex->ec_block < bp->start)
1041                         i = bp->start - cex->ec_block;
1042                 if (i >= cex->ec_len)
1043                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1044                                         i, cex->ec_len);
1045                 for (; i < cex->ec_len && bp->num; i++) {
1046                         *(bp->blocks) = cex->ec_start + i;
1047                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1048                                 *(bp->created) = 0;
1049                         } else {
1050                                 *(bp->created) = 1;
1051                                 /* unmap any possible underlying metadata from
1052                                  * the block device mapping.  bug 6998. */
1053                                 ll_unmap_underlying_metadata(inode->i_sb,
1054                                                              *(bp->blocks));
1055                         }
1056                         bp->created++;
1057                         bp->blocks++;
1058                         bp->num--;
1059                         bp->start++;
1060                 }
1061         }
1062         return err;
1063 }
1064
1065 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1066                        unsigned long num, unsigned long *blocks,
1067                        int *created, int create)
1068 {
1069         struct ext3_ext_base *base = inode;
1070         struct bpointers bp;
1071         int err;
1072
1073         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1074                block, block + num - 1, (unsigned) inode->i_ino);
1075
1076         bp.blocks = blocks;
1077         bp.created = created;
1078         bp.start = block;
1079         bp.init_num = bp.num = num;
1080         bp.create = create;
1081
1082         err = fsfilt_ext3_ext_walk_space(base, block, num,
1083                                          ext3_ext_new_extent_cb, &bp);
1084         ext3_ext_invalidate_cache(base);
1085
1086         return err;
1087 }
1088
1089 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1090                                     int pages, unsigned long *blocks,
1091                                     int *created, int create)
1092 {
1093         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1094         int rc = 0, i = 0;
1095         struct page *fp = NULL;
1096         int clen = 0;
1097
1098         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1099                 inode->i_ino, pages, (*page)->index);
1100
1101         /* pages are sorted already. so, we just have to find
1102          * contig. space and process them properly */
1103         while (i < pages) {
1104                 if (fp == NULL) {
1105                         /* start new extent */
1106                         fp = *page++;
1107                         clen = 1;
1108                         i++;
1109                         continue;
1110                 } else if (fp->index + clen == (*page)->index) {
1111                         /* continue the extent */
1112                         page++;
1113                         clen++;
1114                         i++;
1115                         continue;
1116                 }
1117
1118                 /* process found extent */
1119                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1120                                         clen * blocks_per_page, blocks,
1121                                         created, create);
1122                 if (rc)
1123                         GOTO(cleanup, rc);
1124
1125                 /* look for next extent */
1126                 fp = NULL;
1127                 blocks += blocks_per_page * clen;
1128                 created += blocks_per_page * clen;
1129         }
1130
1131         if (fp)
1132                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1133                                         clen * blocks_per_page, blocks,
1134                                         created, create);
1135 cleanup:
1136         return rc;
1137 }
1138
1139 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1140                                unsigned long *blocks, int *created, int create);
1141 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1142                                    int pages, unsigned long *blocks,
1143                                    int *created, int create)
1144 {
1145         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1146         unsigned long *b;
1147         int rc = 0, i, *cr;
1148
1149         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1150                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1151                 if (rc) {
1152                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1153                                inode->i_ino, *b, *cr, create, rc);
1154                         break;
1155                 }
1156
1157                 b += blocks_per_page;
1158                 cr += blocks_per_page;
1159         }
1160         return rc;
1161 }
1162
1163 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1164                                 int pages, unsigned long *blocks,
1165                                 int *created, int create,
1166                                 cfs_mutex_t *optional_mutex)
1167 {
1168         int rc;
1169
1170         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1171                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1172                                                      blocks, created, create);
1173                 return rc;
1174         }
1175         if (optional_mutex != NULL)
1176                 cfs_mutex_lock(optional_mutex);
1177         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1178                                             created, create);
1179         if (optional_mutex != NULL)
1180                 cfs_mutex_unlock(optional_mutex);
1181
1182         return rc;
1183 }
1184
1185 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1186 {
1187         unsigned long block;
1188         struct buffer_head *bh;
1189         int err, blocksize, csize, boffs, osize = size;
1190
1191         /* prevent reading after eof */
1192         spin_lock(&inode->i_lock);
1193         if (i_size_read(inode) < *offs + size) {
1194                 size = i_size_read(inode) - *offs;
1195                 spin_unlock(&inode->i_lock);
1196                 if (size < 0) {
1197                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1198                                i_size_read(inode), *offs);
1199                         return -EBADR;
1200                 } else if (size == 0) {
1201                         return 0;
1202                 }
1203         } else {
1204                 spin_unlock(&inode->i_lock);
1205         }
1206
1207         blocksize = 1 << inode->i_blkbits;
1208
1209         while (size > 0) {
1210                 block = *offs >> inode->i_blkbits;
1211                 boffs = *offs & (blocksize - 1);
1212                 csize = min(blocksize - boffs, size);
1213                 bh = ext3_bread(NULL, inode, block, 0, &err);
1214                 if (!bh) {
1215                         CERROR("can't read block: %d\n", err);
1216                         return err;
1217                 }
1218
1219                 memcpy(buf, bh->b_data + boffs, csize);
1220                 brelse(bh);
1221
1222                 *offs += csize;
1223                 buf += csize;
1224                 size -= csize;
1225         }
1226         return osize;
1227 }
1228 EXPORT_SYMBOL(fsfilt_ext3_read);
1229
1230 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1231                                    int size, loff_t *offs)
1232 {
1233         int rc;
1234         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1235         if (rc > 0)
1236                 rc = 0;
1237         return rc;
1238 }
1239
1240 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1241                                 loff_t *offs, handle_t *handle)
1242 {
1243         struct buffer_head *bh = NULL;
1244         loff_t old_size = i_size_read(inode), offset = *offs;
1245         loff_t new_size = i_size_read(inode);
1246         unsigned long block;
1247         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1248
1249         while (bufsize > 0) {
1250                 if (bh != NULL)
1251                         brelse(bh);
1252
1253                 block = offset >> inode->i_blkbits;
1254                 boffs = offset & (blocksize - 1);
1255                 size = min(blocksize - boffs, bufsize);
1256                 bh = ext3_bread(handle, inode, block, 1, &err);
1257                 if (!bh) {
1258                         CERROR("can't read/create block: %d\n", err);
1259                         break;
1260                 }
1261
1262                 err = ext3_journal_get_write_access(handle, bh);
1263                 if (err) {
1264                         CERROR("journal_get_write_access() returned error %d\n",
1265                                err);
1266                         break;
1267                 }
1268                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1269                 memcpy(bh->b_data + boffs, buf, size);
1270                 err = ext3_journal_dirty_metadata(handle, bh);
1271                 if (err) {
1272                         CERROR("journal_dirty_metadata() returned error %d\n",
1273                                err);
1274                         break;
1275                 }
1276                 if (offset + size > new_size)
1277                         new_size = offset + size;
1278                 offset += size;
1279                 bufsize -= size;
1280                 buf += size;
1281         }
1282         if (bh)
1283                 brelse(bh);
1284
1285         /* correct in-core and on-disk sizes */
1286         if (new_size > i_size_read(inode)) {
1287                 spin_lock(&inode->i_lock);
1288                 if (new_size > i_size_read(inode))
1289                         i_size_write(inode, new_size);
1290                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1291                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1292                 if (i_size_read(inode) > old_size) {
1293                         spin_unlock(&inode->i_lock);
1294                         mark_inode_dirty(inode);
1295                 } else {
1296                         spin_unlock(&inode->i_lock);
1297                 }
1298         }
1299
1300         if (err == 0)
1301                 *offs = offset;
1302         return err;
1303 }
1304 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1305
1306 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1307                                     loff_t *offs, int force_sync)
1308 {
1309         struct inode *inode = file->f_dentry->d_inode;
1310         handle_t *handle;
1311         int err, block_count = 0, blocksize;
1312
1313         /* Determine how many transaction credits are needed */
1314         blocksize = 1 << inode->i_blkbits;
1315         block_count = (*offs & (blocksize - 1)) + bufsize;
1316         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1317
1318         handle = ext3_journal_start(inode,
1319                         block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1320         if (IS_ERR(handle)) {
1321                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1322                        block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
1323                        bufsize);
1324                 return PTR_ERR(handle);
1325         }
1326
1327         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1328
1329         if (!err && force_sync)
1330                 handle->h_sync = 1; /* recovery likes this */
1331
1332         ext3_journal_stop(handle);
1333
1334         return err;
1335 }
1336
1337 static int fsfilt_ext3_setup(struct super_block *sb)
1338 {
1339         if (!EXT3_HAS_COMPAT_FEATURE(sb,
1340                                 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1341                 CERROR("ext3 mounted without journal\n");
1342                 return -EINVAL;
1343         }
1344
1345 #ifdef S_PDIROPS
1346         CWARN("Enabling PDIROPS\n");
1347         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1348         sb->s_flags |= S_PDIROPS;
1349 #endif
1350         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1351                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1352         return 0;
1353 }
1354
1355 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1356    objects. Logs is number of logfiles to update */
1357 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1358 {
1359         if ( !fso ) {
1360                 switch(op) {
1361                 case FSFILT_OP_CREATE:
1362                                  /* directory leaf, index & indirect & EA*/
1363                         return 4 + 3 * logs;
1364                 case FSFILT_OP_UNLINK:
1365                         return 3 * logs;
1366                 }
1367         } else {
1368                 int i;
1369                 int needed = 0;
1370                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1371                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1372                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1373                 for (i = 0; i < op; i++, fso++) {
1374                         int nblocks = fso->fso_bufcnt * blockpp;
1375                         int ndindirect = min(nblocks, addrpp + 1);
1376                         int nindir = nblocks + ndindirect + 1;
1377
1378                         needed += nindir;
1379                 }
1380                 return needed + 3 * logs;
1381         }
1382
1383         return 0;
1384 }
1385
1386 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
1387 {
1388         return (EXT3_SB(sb)->journal_bdev);
1389 }
1390 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
1391
1392 static int ll_decode_fh_accept(void *context, struct dentry *de)
1393 {
1394         return 1;
1395 }
1396
1397 #ifdef HAVE_EXPORTFS_DECODE_FH
1398 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
1399          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
1400                             acceptable, context)
1401 #else
1402 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
1403          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
1404                                      type, acceptable, context)
1405 # define FILEID_INO32_GEN 1
1406 extern struct export_operations export_op_default;
1407 #endif
1408
1409 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
1410                                       struct fsfilt_fid *fid, int ignore_gen)
1411 {
1412         struct inode  *inode;
1413         struct dentry *result;
1414
1415         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
1416                                        ll_decode_fh_accept, NULL);
1417         if (IS_ERR(result)) {
1418                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
1419                        fid->ino, fid->gen, PTR_ERR(result));
1420                 return result;
1421         }
1422
1423         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
1424                fid->ino, fid->gen);
1425         inode = result->d_inode;
1426         if (inode == NULL)
1427                 goto err_out;
1428
1429         if (inode->i_nlink == 0 &&
1430             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
1431                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
1432                               " ctime -- this may indicate disk "
1433                               "corruption (inode: %lu, link: %lu, "
1434                               "count: %d)\n", inode->i_ino,
1435                               (unsigned long)inode->i_nlink,
1436                               atomic_read(&inode->i_count));
1437                 goto err_out;
1438         }
1439         if (fid->gen && inode->i_generation != fid->gen) {
1440                 /* we didn't find the right inode.. */
1441                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
1442                        "count: %d, generation %u/%u\n",
1443                        inode->i_ino, (unsigned long)inode->i_nlink,
1444                        atomic_read(&inode->i_count), inode->i_generation,
1445                        fid->gen);
1446                 goto err_out;
1447         }
1448
1449         return result;
1450 err_out:
1451         l_dput(result);
1452         return ERR_PTR(-ENOENT);
1453 }
1454
1455 static struct fsfilt_operations fsfilt_ext3_ops = {
1456         .fs_type                = "ext3",
1457         .fs_owner               = THIS_MODULE,
1458         .fs_getlabel            = fsfilt_ext3_get_label,
1459         .fs_setlabel            = fsfilt_ext3_set_label,
1460         .fs_uuid                = fsfilt_ext3_uuid,
1461         .fs_start               = fsfilt_ext3_start,
1462         .fs_brw_start           = fsfilt_ext3_brw_start,
1463         .fs_extend              = fsfilt_ext3_extend,
1464         .fs_commit              = fsfilt_ext3_commit,
1465         .fs_commit_async        = fsfilt_ext3_commit_async,
1466         .fs_commit_wait         = fsfilt_ext3_commit_wait,
1467         .fs_setattr             = fsfilt_ext3_setattr,
1468         .fs_iocontrol           = fsfilt_ext3_iocontrol,
1469         .fs_set_md              = fsfilt_ext3_set_md,
1470         .fs_get_md              = fsfilt_ext3_get_md,
1471         .fs_readpage            = fsfilt_ext3_readpage,
1472         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
1473         .fs_statfs              = fsfilt_ext3_statfs,
1474         .fs_sync                = fsfilt_ext3_sync,
1475         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
1476         .fs_write_record        = fsfilt_ext3_write_record,
1477         .fs_read_record         = fsfilt_ext3_read_record,
1478         .fs_setup               = fsfilt_ext3_setup,
1479         .fs_send_bio            = fsfilt_ext3_send_bio,
1480         .fs_get_op_len          = fsfilt_ext3_get_op_len,
1481 #ifdef HAVE_DISK_INODE_VERSION
1482         .fs_get_version         = fsfilt_ext3_get_version,
1483         .fs_set_version         = fsfilt_ext3_set_version,
1484 #endif
1485         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
1486         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
1487 };
1488
1489 static int __init fsfilt_ext3_init(void)
1490 {
1491         int rc;
1492
1493         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
1494                                          sizeof(struct fsfilt_cb_data), 0, 0);
1495         if (!fcb_cache) {
1496                 CERROR("error allocating fsfilt journal callback cache\n");
1497                 GOTO(out, rc = -ENOMEM);
1498         }
1499
1500         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
1501
1502         if (rc) {
1503                 int err = cfs_mem_cache_destroy(fcb_cache);
1504                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
1505         }
1506 out:
1507         return rc;
1508 }
1509
1510 static void __exit fsfilt_ext3_exit(void)
1511 {
1512         int rc;
1513
1514         fsfilt_unregister_ops(&fsfilt_ext3_ops);
1515         rc = cfs_mem_cache_destroy(fcb_cache);
1516         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
1517 }
1518
1519 module_init(fsfilt_ext3_init);
1520 module_exit(fsfilt_ext3_exit);
1521
1522 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1523 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
1524 MODULE_LICENSE("GPL");