lustre/lvfs/fsfilt_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  lustre/lib/fsfilt_ext3.c
   5  *  Lustre filesystem abstraction routines
   6  *
   7  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   8  *   Author: Andreas Dilger <adilger@clusterfs.com>
   9  *
  10  *   This file is part of Lustre, http://www.lustre.org.
  11  *
  12  *   Lustre is free software; you can redistribute it and/or
  13  *   modify it under the terms of version 2 of the GNU General Public
  14  *   License as published by the Free Software Foundation.
  15  *
  16  *   Lustre is distributed in the hope that it will be useful,
  17  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  *   GNU General Public License for more details.
  20  *
  21  *   You should have received a copy of the GNU General Public License
  22  *   along with Lustre; if not, write to the Free Software
  23  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  24  */
  25
  26 #define DEBUG_SUBSYSTEM S_FILTER
  27
  28 #include <linux/init.h>
  29 #include <linux/module.h>
  30 #include <linux/fs.h>
  31 #include <linux/jbd.h>
  32 #include <linux/slab.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/ext3_fs.h>
  36 #include <linux/ext3_jbd.h>
  37 #include <linux/ext3_extents.h>
  38 #include <linux/version.h>
  39 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  40 #include <linux/ext3_xattr.h>
  41 #else
  42 #include <ext3/xattr.h>
  43 #endif
  44
  45 #include <linux/kp30.h>
  46 #include <linux/lustre_fsfilt.h>
  47 #include <linux/obd.h>
  48 #include <linux/obd_class.h>
  49 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  50 #include <linux/module.h>
  51 #include <linux/iobuf.h>
  52 #endif
  53
  54
  55 static kmem_cache_t *fcb_cache;
  56 static atomic_t fcb_cache_count = ATOMIC_INIT(0);
  57
  58 struct fsfilt_cb_data {
  59         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  60         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  61         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  62         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  63         void *cb_data;                  /* MDS/OST completion function data */
  64 };
  65
  66 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
  67 #define EXT3_XATTR_INDEX_TRUSTED        4
  68 #endif
  69 #define XATTR_LUSTRE_MDS_LOV_EA         "lov"
  70
  71 /*
  72  * We don't currently need any additional blocks for rmdir and
  73  * unlink transactions because we are storing the OST oa_id inside
  74  * the inode (which we will be changing anyways as part of this
  75  * transaction).
  76  */
  77 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
  78                                int logs)
  79 {
  80         /* For updates to the last recieved file */
  81         int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS;
  82         journal_t *journal;
  83         void *handle;
  84
  85         if (current->journal_info) {
  86                 CDEBUG(D_INODE, "increasing refcount on %p\n",
  87                        current->journal_info);
  88                 goto journal_start;
  89         }
  90
  91         /* XXX BUG 3188 -- must return to one set of opcodes */
  92         /* FIXME - cache hook */
  93         if (op & 0x20) {
  94                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
  95                 op = op & ~0x20;
  96         }
  97
  98         /* FIXME - kml */
  99         if (op & 0x10) {
 100                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
 101                 op = op & ~0x10;
 102         }
 103
 104         switch(op) {
 105         case FSFILT_OP_RMDIR:
 106         case FSFILT_OP_UNLINK:
 107                 /* delete one file + create/update logs for each stripe */
 108                 nblocks += EXT3_DELETE_TRANS_BLOCKS;
 109                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 110                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 111                 break;
 112         case FSFILT_OP_RENAME:
 113                 /* modify additional directory */
 114                 nblocks += EXT3_SINGLEDATA_TRANS_BLOCKS;
 115                 /* no break */
 116         case FSFILT_OP_SYMLINK:
 117                 /* additional block + block bitmap + GDT for long symlink */
 118                 nblocks += 3;
 119                 /* no break */
 120         case FSFILT_OP_CREATE:
 121                 /* create/update logs for each stripe */
 122                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 123                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 124                 /* no break */
 125         case FSFILT_OP_MKDIR:
 126         case FSFILT_OP_MKNOD:
 127                 /* modify one inode + block bitmap + GDT */
 128                 nblocks += 3;
 129                 /* no break */
 130         case FSFILT_OP_LINK:
 131                 /* modify parent directory */
 132                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 133                         EXT3_DATA_TRANS_BLOCKS;
 134                 break;
 135         case FSFILT_OP_SETATTR:
 136                 /* Setattr on inode */
 137                 nblocks += 1;
 138                 break;
 139         case FSFILT_OP_CANCEL_UNLINK:
 140                 /* blocks for log header bitmap update OR
 141                  * blocks for catalog header bitmap update + unlink of logs */
 142                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 143                         EXT3_DELETE_TRANS_BLOCKS * logs;
 144                 break;
 145         case FSFILT_OP_NOOP:
 146                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
 147                 break;
 148         default: CERROR("unknown transaction start op %d\n", op);
 149                 LBUG();
 150         }
 151
 152         LASSERT(current->journal_info == desc_private);
 153         journal = EXT3_SB(inode->i_sb)->s_journal;
 154         if (nblocks > journal->j_max_transaction_buffers) {
 155                 CERROR("too many credits %d for op %ux%u using %d instead\n",
 156                        nblocks, op, logs, journal->j_max_transaction_buffers);
 157                 nblocks = journal->j_max_transaction_buffers;
 158         }
 159
 160  journal_start:
 161         lock_kernel();
 162         handle = journal_start(EXT3_JOURNAL(inode), nblocks);
 163         unlock_kernel();
 164
 165         if (!IS_ERR(handle))
 166                 LASSERT(current->journal_info == handle);
 167         else
 168                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 169                        op, nblocks, PTR_ERR(handle));
 170         return handle;
 171 }
 172
 173 /*
 174  * Calculate the number of buffer credits needed to write multiple pages in
 175  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 176  * doesn't have a nice API for calculating this sort of thing in advance.
 177  *
 178  * See comment above ext3_writepage_trans_blocks for details.  We assume
 179  * no data journaling is being done, but it does allow for all of the pages
 180  * being non-contiguous.  If we are guaranteed contiguous pages we could
 181  * reduce the number of (d)indirect blocks a lot.
 182  *
 183  * With N blocks per page and P pages, for each inode we have at most:
 184  * N*P indirect
 185  * min(N*P, blocksize/4 + 1) dindirect blocks
 186  * niocount tindirect
 187  *
 188  * For the entire filesystem, we have at most:
 189  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 190  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 191  * objcount inode blocks
 192  * 1 superblock
 193  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 194  *
 195  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 196  */
 197 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 198                                       int niocount, struct niobuf_local *nb)
 199 {
 200         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 201         __u64 next_indir;
 202         const int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
 203         int nbitmaps = 0, ngdblocks;
 204         int needed = objcount + 1; /* inodes + superblock */
 205         int i, j;
 206
 207         for (i = 0, j = 0; i < objcount; i++, fso++) {
 208                 /* two or more dindirect blocks in case we cross boundary */
 209                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 210                                     nb[j].offset) >>
 211                                    sb->s_blocksize_bits) /
 212                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 213                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 214
 215                 /* leaf, indirect, tindirect blocks for first block */
 216                 nbitmaps += blockpp + 2;
 217
 218                 j += fso->fso_bufcnt;
 219         }
 220
 221         next_indir = nb[0].offset +
 222                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 223         for (i = 1; i < niocount; i++) {
 224                 if (nb[i].offset >= next_indir) {
 225                         nbitmaps++;     /* additional indirect */
 226                         next_indir = nb[i].offset +
 227                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 228                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 229                         nbitmaps++;     /* additional indirect */
 230                 }
 231                 nbitmaps += blockpp;    /* each leaf in different group? */
 232         }
 233
 234         ngdblocks = nbitmaps;
 235         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 236                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 237         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 238                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 239
 240         needed += nbitmaps + ngdblocks;
 241
 242         /* last_rcvd update */
 243         needed += EXT3_DATA_TRANS_BLOCKS;
 244
 245 #if defined(CONFIG_QUOTA) && !defined(__x86_64__) /* XXX */
 246         /* We assume that there will be 1 bit set in s_dquot.flags for each
 247          * quota file that is active.  This is at least true for now.
 248          */
 249         needed += hweight32(sb_any_quota_enabled(sb)) *
 250                 EXT3_SINGLEDATA_TRANS_BLOCKS;
 251 #endif
 252
 253         return needed;
 254 }
 255
 256 /* We have to start a huge journal transaction here to hold all of the
 257  * metadata for the pages being written here.  This is necessitated by
 258  * the fact that we do lots of prepare_write operations before we do
 259  * any of the matching commit_write operations, so even if we split
 260  * up to use "smaller" transactions none of them could complete until
 261  * all of them were opened.  By having a single journal transaction,
 262  * we eliminate duplicate reservations for common blocks like the
 263  * superblock and group descriptors or bitmaps.
 264  *
 265  * We will start the transaction here, but each prepare_write will
 266  * add a refcount to the transaction, and each commit_write will
 267  * remove a refcount.  The transaction will be closed when all of
 268  * the pages have been written.
 269  */
 270 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 271                                    int niocount, struct niobuf_local *nb,
 272                                    void *desc_private, int logs)
 273 {
 274         journal_t *journal;
 275         handle_t *handle;
 276         int needed;
 277         ENTRY;
 278
 279         LASSERT(current->journal_info == desc_private);
 280         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 281         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 282
 283         /* The number of blocks we could _possibly_ dirty can very large.
 284          * We reduce our request if it is absurd (and we couldn't get that
 285          * many credits for a single handle anyways).
 286          *
 287          * At some point we have to limit the size of I/Os sent at one time,
 288          * increase the size of the journal, or we have to calculate the
 289          * actual journal requirements more carefully by checking all of
 290          * the blocks instead of being maximally pessimistic.  It remains to
 291          * be seen if this is a real problem or not.
 292          */
 293         if (needed > journal->j_max_transaction_buffers) {
 294                 CERROR("want too many journal credits (%d) using %d instead\n",
 295                        needed, journal->j_max_transaction_buffers);
 296                 needed = journal->j_max_transaction_buffers;
 297         }
 298
 299         lock_kernel();
 300         handle = journal_start(journal, needed);
 301         unlock_kernel();
 302         if (IS_ERR(handle)) {
 303                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 304                        PTR_ERR(handle));
 305         } else {
 306                 LASSERT(handle->h_buffer_credits >= needed);
 307                 LASSERT(current->journal_info == handle);
 308         }
 309
 310         RETURN(handle);
 311 }
 312
 313 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 314 {
 315         int rc;
 316         handle_t *handle = h;
 317
 318         LASSERT(current->journal_info == handle);
 319         if (force_sync)
 320                 handle->h_sync = 1; /* recovery likes this */
 321
 322         lock_kernel();
 323         rc = journal_stop(handle);
 324         unlock_kernel();
 325
 326         return rc;
 327 }
 328
 329 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 330                                     void **wait_handle)
 331 {
 332         unsigned long tid;
 333         transaction_t *transaction;
 334 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 335         unsigned long rtid;
 336 #endif
 337         handle_t *handle = h;
 338         journal_t *journal;
 339         int rc;
 340
 341         LASSERT(current->journal_info == handle);
 342
 343         lock_kernel();
 344         transaction = handle->h_transaction;
 345         journal = transaction->t_journal;
 346         tid = transaction->t_tid;
 347         /* we don't want to be blocked */
 348         handle->h_sync = 0;
 349         rc = journal_stop(handle);
 350         if (rc) {
 351                 CERROR("error while stopping transaction: %d\n", rc);
 352                 unlock_kernel();
 353                 return rc;
 354         }
 355 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 356         rtid = log_start_commit(journal, transaction);
 357         if (rtid != tid)
 358                 CERROR("strange race: %lu != %lu\n",
 359                        (unsigned long) tid, (unsigned long) rtid);
 360 #else
 361         log_start_commit(journal, transaction->t_tid);
 362 #endif
 363         unlock_kernel();
 364
 365         *wait_handle = (void *) tid;
 366         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 367         return 0;
 368 }
 369
 370 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 371 {
 372         tid_t tid = (tid_t)(long)h;
 373
 374         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 375         if (is_journal_aborted(EXT3_JOURNAL(inode)))
 376                 return -EIO;
 377
 378         log_wait_commit(EXT3_JOURNAL(inode), tid);
 379
 380         return 0;
 381 }
 382
 383 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 384                                struct iattr *iattr, int do_trunc)
 385 {
 386         struct inode *inode = dentry->d_inode;
 387         int rc;
 388
 389         lock_kernel();
 390
 391         /* A _really_ horrible hack to avoid removing the data stored
 392          * in the block pointers; this is really the "small" stripe MD data.
 393          * We can avoid further hackery by virtue of the MDS file size being
 394          * zero all the time (which doesn't invoke block truncate at unlink
 395          * time), so we assert we never change the MDS file size from zero. */
 396         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 397                 /* ATTR_SIZE would invoke truncate: clear it */
 398                 iattr->ia_valid &= ~ATTR_SIZE;
 399                 EXT3_I(inode)->i_disksize = inode->i_size = iattr->ia_size;
 400
 401                 /* make sure _something_ gets set - so new inode
 402                  * goes to disk (probably won't work over XFS */
 403                 if (!(iattr->ia_valid & (ATTR_MODE | ATTR_MTIME | ATTR_CTIME))){
 404                         iattr->ia_valid |= ATTR_MODE;
 405                         iattr->ia_mode = inode->i_mode;
 406                 }
 407         }
 408
 409         /* Don't allow setattr to change file type */
 410         iattr->ia_mode = (inode->i_mode & S_IFMT)|(iattr->ia_mode & ~S_IFMT);
 411
 412         /* We set these flags on the client, but have already checked perms
 413          * so don't confuse inode_change_ok. */
 414         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 415
 416         if (inode->i_op->setattr) {
 417                 rc = inode->i_op->setattr(dentry, iattr);
 418         } else {
 419                 rc = inode_change_ok(inode, iattr);
 420                 if (!rc)
 421                         rc = inode_setattr(inode, iattr);
 422         }
 423
 424         unlock_kernel();
 425
 426         return rc;
 427 }
 428
 429 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 430                                  unsigned int cmd, unsigned long arg)
 431 {
 432         int rc = 0;
 433         ENTRY;
 434
 435         if (inode->i_fop->ioctl)
 436                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 437         else
 438                 RETURN(-ENOTTY);
 439
 440         RETURN(rc);
 441 }
 442
 443 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 444                               void *lmm, int lmm_size)
 445 {
 446         int rc;
 447
 448         /* keep this when we get rid of OLD_EA (too noisy during conversion) */
 449         if (EXT3_I(inode)->i_file_acl /* || large inode EA flag */)
 450                 CWARN("setting EA on %lu/%u again... interesting\n",
 451                        inode->i_ino, inode->i_generation);
 452
 453         lock_kernel();
 454         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 455                                    XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size, 0);
 456
 457         unlock_kernel();
 458
 459         if (rc)
 460                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 461                        inode->i_ino, rc);
 462         return rc;
 463 }
 464
 465 /* Must be called with i_sem held */
 466 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size)
 467 {
 468         int rc;
 469
 470         LASSERT(down_trylock(&inode->i_sem) != 0);
 471         lock_kernel();
 472
 473         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 474                             XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size);
 475         unlock_kernel();
 476
 477         /* This gives us the MD size */
 478         if (lmm == NULL)
 479                 return (rc == -ENODATA) ? 0 : rc;
 480
 481         if (rc < 0) {
 482                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 483                        EXT3_XATTR_INDEX_TRUSTED, XATTR_LUSTRE_MDS_LOV_EA,
 484                        inode->i_ino, rc);
 485                 memset(lmm, 0, lmm_size);
 486                 return (rc == -ENODATA) ? 0 : rc;
 487         }
 488
 489         return rc;
 490 }
 491
 492 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
 493 static int fsfilt_ext3_send_bio(struct inode *inode, struct bio *bio)
 494 {
 495         submit_bio(WRITE, bio);
 496         return 0;
 497 }
 498 #else
 499 static int fsfilt_ext3_send_bio(struct inode *inode, struct kiobuf *bio)
 500 {
 501         int rc, blocks_per_page;
 502
 503         rc = brw_kiovec(WRITE, 1, &bio, inode->i_dev,
 504                         bio->blocks, 1 << inode->i_blkbits);
 505
 506         blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 507
 508         if (rc != (1 << inode->i_blkbits) * bio->nr_pages *
 509             blocks_per_page) {
 510                 CERROR("short write?  expected %d, wrote %d\n",
 511                        (1 << inode->i_blkbits) * bio->nr_pages *
 512                        blocks_per_page, rc);
 513         }
 514
 515         return rc;
 516 }
 517 #endif
 518
 519 /* FIXME-UMKA: This should be used in 2.6.x io code later. */
 520 static struct page *fsfilt_ext3_getpage(struct inode *inode, long int index)
 521 {
 522         int rc;
 523         struct page *page;
 524
 525         page = grab_cache_page(inode->i_mapping, index);
 526         if (page == NULL)
 527                 return ERR_PTR(-ENOMEM);
 528
 529         if (PageUptodate(page)) {
 530                 unlock_page(page);
 531                 return page;
 532         }
 533
 534         rc = inode->i_mapping->a_ops->readpage(NULL, page);
 535         if (rc < 0) {
 536                 page_cache_release(page);
 537                 return ERR_PTR(rc);
 538         }
 539
 540         return page;
 541 }
 542
 543 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 544                                     loff_t *off)
 545 {
 546         struct inode *inode = file->f_dentry->d_inode;
 547         int rc = 0;
 548
 549         if (S_ISREG(inode->i_mode))
 550                 rc = file->f_op->read(file, buf, count, off);
 551         else {
 552                 const int blkbits = inode->i_sb->s_blocksize_bits;
 553                 const int blksize = inode->i_sb->s_blocksize;
 554
 555                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 556                        count, inode->i_ino, *off);
 557                 while (count > 0) {
 558                         struct buffer_head *bh;
 559
 560                         bh = NULL;
 561                         if (*off < inode->i_size) {
 562                                 int err = 0;
 563
 564                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 565                                                 0, &err);
 566
 567                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 568
 569                                 if (bh) {
 570                                         memcpy(buf, bh->b_data, blksize);
 571                                         brelse(bh);
 572                                 } else if (err) {
 573                                         /* XXX in theory we should just fake
 574                                          * this buffer and continue like ext3,
 575                                          * especially if this is a partial read
 576                                          */
 577                                         CERROR("error read dir %lu+%llu: %d\n",
 578                                                inode->i_ino, *off, err);
 579                                         RETURN(err);
 580                                 }
 581                         }
 582                         if (!bh) {
 583                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 584
 585                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 586                                 memset(fake, 0, sizeof(*fake));
 587                                 fake->rec_len = cpu_to_le32(blksize);
 588                         }
 589                         count -= blksize;
 590                         buf += blksize;
 591                         *off += blksize;
 592                         rc += blksize;
 593                 }
 594         }
 595
 596         return rc;
 597 }
 598
 599 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 600 {
 601         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 602
 603         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 604
 605         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 606         atomic_dec(&fcb_cache_count);
 607 }
 608
 609 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd,
 610                                       struct super_block *sb,
 611                                       __u64 last_rcvd,
 612                                       void *handle, fsfilt_cb_t cb_func,
 613                                       void *cb_data)
 614 {
 615         struct fsfilt_cb_data *fcb;
 616
 617         OBD_SLAB_ALLOC(fcb, fcb_cache, GFP_NOFS, sizeof *fcb);
 618         if (fcb == NULL)
 619                 RETURN(-ENOMEM);
 620
 621         atomic_inc(&fcb_cache_count);
 622         fcb->cb_func = cb_func;
 623         fcb->cb_obd = obd;
 624         fcb->cb_last_rcvd = last_rcvd;
 625         fcb->cb_data = cb_data;
 626
 627         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 628         lock_kernel();
 629         journal_callback_set(handle, fsfilt_ext3_cb_func,
 630                              (struct journal_callback *)fcb);
 631         unlock_kernel();
 632
 633         return 0;
 634 }
 635
 636 /*
 637  * We need to hack the return value for the free inode counts because
 638  * the current EA code requires one filesystem block per inode with EAs,
 639  * so it is possible to run out of blocks before we run out of inodes.
 640  *
 641  * This can be removed when the ext3 EA code is fixed.
 642  */
 643 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 644 {
 645         struct kstatfs sfs;
 646         int rc;
 647
 648         memset(&sfs, 0, sizeof(sfs));
 649
 650         rc = sb->s_op->statfs(sb, &sfs);
 651
 652         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 653                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 654                 sfs.f_ffree = sfs.f_bfree;
 655         }
 656
 657         statfs_pack(osfs, &sfs);
 658         return rc;
 659 }
 660
 661 static int fsfilt_ext3_sync(struct super_block *sb)
 662 {
 663         return ext3_force_commit(sb);
 664 }
 665
 666 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 667 struct bpointers {
 668         unsigned long *blocks;
 669         int *created;
 670         unsigned long start;
 671         int num;
 672         int init_num;
 673 };
 674
 675 static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
 676                                   struct ext3_ext_path *path,
 677                                   struct ext3_extent *newex, int exist)
 678 {
 679         struct inode *inode = tree->inode;
 680         struct bpointers *bp = tree->private;
 681         int count, err, goal;
 682         unsigned long pblock;
 683         unsigned long tgen;
 684         loff_t new_i_size;
 685         handle_t *handle;
 686         int i;
 687
 688         i = EXT_DEPTH(tree);
 689         EXT_ASSERT(i == path->p_depth);
 690         EXT_ASSERT(path[i].p_hdr);
 691
 692         if (exist) {
 693                 err = EXT_CONTINUE;
 694                 goto map;
 695         }
 696
 697         tgen = EXT_GENERATION(tree);
 698         count = ext3_ext_calc_credits_for_insert(tree, path);
 699         up_write(&EXT3_I(inode)->truncate_sem);
 700
 701         handle = ext3_journal_start(inode, count + EXT3_ALLOC_NEEDED + 1);
 702         if (IS_ERR(handle)) {
 703                 down_write(&EXT3_I(inode)->truncate_sem);
 704                 return PTR_ERR(handle);
 705         }
 706
 707         if (tgen != EXT_GENERATION(tree)) {
 708                 /* the tree has changed. so path can be invalid at moment */
 709                 ext3_journal_stop(handle, inode);
 710                 down_write(&EXT3_I(inode)->truncate_sem);
 711                 return EXT_REPEAT;
 712         }
 713
 714         down_write(&EXT3_I(inode)->truncate_sem);
 715         goal = ext3_ext_find_goal(inode, path, newex->e_block);
 716         count = newex->e_num;
 717         pblock = ext3_new_blocks(handle, inode, &count, goal, &err);
 718         if (!pblock)
 719                 goto out;
 720         EXT_ASSERT(count <= newex->e_num);
 721
 722         /* insert new extent */
 723         newex->e_start = pblock;
 724         newex->e_num = count;
 725         err = ext3_ext_insert_extent(handle, tree, path, newex);
 726         if (err)
 727                 goto out;
 728
 729         /* correct on-disk inode size */
 730         if (newex->e_num > 0) {
 731                 new_i_size = (loff_t) newex->e_block + newex->e_num;
 732                 new_i_size = new_i_size << inode->i_blkbits;
 733                 if (new_i_size > EXT3_I(inode)->i_disksize) {
 734                         EXT3_I(inode)->i_disksize = new_i_size;
 735                         err = ext3_mark_inode_dirty(handle, inode);
 736                 }
 737         }
 738
 739 out:
 740         ext3_journal_stop(handle, inode);
 741 map:
 742         if (err >= 0) {
 743                 /* map blocks */
 744                 if (bp->num == 0) {
 745                         CERROR("hmm. why do we find this extent?\n");
 746                         CERROR("initial space: %lu:%u\n",
 747                                 bp->start, bp->init_num);
 748                         CERROR("current extent: %u/%u/%u %d\n",
 749                                 newex->e_block, newex->e_num,
 750                                 newex->e_start, exist);
 751                 }
 752                 i = 0;
 753                 if (newex->e_block < bp->start)
 754                         i = bp->start - newex->e_block;
 755                 if (i >= newex->e_num)
 756                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 757                                         i, newex->e_num);
 758                 for (; i < newex->e_num && bp->num; i++) {
 759                         *(bp->created) = (exist == 0 ? 1 : 0);
 760                         bp->created++;
 761                         *(bp->blocks) = newex->e_start + i;
 762                         bp->blocks++;
 763                         bp->num--;
 764                 }
 765         }
 766         return err;
 767 }
 768
 769 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
 770                        unsigned long num, unsigned long *blocks,
 771                        int *created, int create)
 772 {
 773         struct ext3_extents_tree tree;
 774         struct bpointers bp;
 775         int err;
 776
 777         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
 778                 block, block + num, (unsigned) inode->i_ino);
 779
 780         ext3_init_tree_desc(&tree, inode);
 781         tree.private = &bp;
 782         bp.blocks = blocks;
 783         bp.created = created;
 784         bp.start = block;
 785         bp.init_num = bp.num = num;
 786
 787         down_write(&EXT3_I(inode)->truncate_sem);
 788         err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
 789         ext3_ext_invalidate_cache(&tree);
 790         up_write(&EXT3_I(inode)->truncate_sem);
 791
 792         return err;
 793 }
 794
 795 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
 796                                     int pages, unsigned long *blocks,
 797                                     int *created, int create)
 798 {
 799         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 800         int rc = 0, i = 0;
 801         struct page *fp = NULL;
 802         int clen = 0;
 803
 804         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
 805                 inode->i_ino, pages, (*page)->index);
 806
 807         /* pages are sorted already. so, we just have to find
 808          * contig. space and process them properly */
 809         while (i < pages) {
 810                 if (fp == NULL) {
 811                         /* start new extent */
 812                         fp = *page++;
 813                         clen = 1;
 814                         i++;
 815                         continue;
 816                 } else if (fp->index + clen == (*page)->index) {
 817                         /* continue the extent */
 818                         page++;
 819                         clen++;
 820                         i++;
 821                         continue;
 822                 }
 823
 824                 /* process found extent */
 825                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 826                                         clen * blocks_per_page, blocks,
 827                                         created, create);
 828                 if (rc)
 829                         GOTO(cleanup, rc);
 830
 831                 /* look for next extent */
 832                 fp = NULL;
 833                 blocks += blocks_per_page * clen;
 834                 created += blocks_per_page * clen;
 835         }
 836
 837         if (fp)
 838                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 839                                         clen * blocks_per_page, blocks,
 840                                         created, create);
 841 cleanup:
 842         return rc;
 843 }
 844 #endif
 845
 846 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
 847                                unsigned long *blocks, int *created, int create);
 848 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
 849                                    int pages, unsigned long *blocks,
 850                                    int *created, int create)
 851 {
 852         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 853         unsigned long *b;
 854         int rc = 0, i, *cr;
 855
 856         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
 857                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
 858                 if (rc) {
 859                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
 860                                inode->i_ino, *b, *cr, create, rc);
 861                         break;
 862                 }
 863
 864                 b += blocks_per_page;
 865                 cr += blocks_per_page;
 866         }
 867         return rc;
 868 }
 869
 870 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
 871                                 int pages, unsigned long *blocks,
 872                                 int *created, int create,
 873                                 struct semaphore *optional_sem)
 874 {
 875         int rc;
 876 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 877         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
 878                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
 879                                                      blocks, created, create);
 880                 return rc;
 881         }
 882 #endif
 883         if (optional_sem != NULL)
 884                 down(optional_sem);
 885         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
 886                                             created, create);
 887         if (optional_sem != NULL)
 888                 up(optional_sem);
 889
 890         return rc;
 891 }
 892
 893 extern int ext3_prep_san_write(struct inode *inode, long *blocks,
 894                                int nblocks, loff_t newsize);
 895 static int fsfilt_ext3_prep_san_write(struct inode *inode, long *blocks,
 896                                       int nblocks, loff_t newsize)
 897 {
 898         return ext3_prep_san_write(inode, blocks, nblocks, newsize);
 899 }
 900
 901 static int fsfilt_ext3_read_record(struct file * file, void *buf,
 902                                    int size, loff_t *offs)
 903 {
 904         struct inode *inode = file->f_dentry->d_inode;
 905         unsigned long block;
 906         struct buffer_head *bh;
 907         int err, blocksize, csize, boffs;
 908
 909         /* prevent reading after eof */
 910         lock_kernel();
 911         if (inode->i_size < *offs + size) {
 912                 size = inode->i_size - *offs;
 913                 unlock_kernel();
 914                 if (size < 0) {
 915                         CERROR("size %llu is too short for read %u@%llu\n",
 916                                inode->i_size, size, *offs);
 917                         return -EIO;
 918                 } else if (size == 0) {
 919                         return 0;
 920                 }
 921         } else {
 922                 unlock_kernel();
 923         }
 924
 925         blocksize = 1 << inode->i_blkbits;
 926
 927         while (size > 0) {
 928                 block = *offs >> inode->i_blkbits;
 929                 boffs = *offs & (blocksize - 1);
 930                 csize = min(blocksize - boffs, size);
 931                 bh = ext3_bread(NULL, inode, block, 0, &err);
 932                 if (!bh) {
 933                         CERROR("can't read block: %d\n", err);
 934                         return err;
 935                 }
 936
 937                 memcpy(buf, bh->b_data + boffs, csize);
 938                 brelse(bh);
 939
 940                 *offs += csize;
 941                 buf += csize;
 942                 size -= csize;
 943         }
 944         return 0;
 945 }
 946
 947 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
 948                                     loff_t *offs, int force_sync)
 949 {
 950         struct buffer_head *bh = NULL;
 951         unsigned long block;
 952         struct inode *inode = file->f_dentry->d_inode;
 953         loff_t old_size = inode->i_size, offset = *offs;
 954         loff_t new_size = inode->i_size;
 955         journal_t *journal;
 956         handle_t *handle;
 957         int err, block_count = 0, blocksize, size, boffs;
 958
 959         /* Determine how many transaction credits are needed */
 960         blocksize = 1 << inode->i_blkbits;
 961         block_count = (*offs & (blocksize - 1)) + bufsize;
 962         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
 963
 964         journal = EXT3_SB(inode->i_sb)->s_journal;
 965         lock_kernel();
 966         handle = journal_start(journal,
 967                                block_count * EXT3_DATA_TRANS_BLOCKS + 2);
 968         unlock_kernel();
 969         if (IS_ERR(handle)) {
 970                 CERROR("can't start transaction\n");
 971                 return PTR_ERR(handle);
 972         }
 973
 974         while (bufsize > 0) {
 975                 if (bh != NULL)
 976                         brelse(bh);
 977
 978                 block = offset >> inode->i_blkbits;
 979                 boffs = offset & (blocksize - 1);
 980                 size = min(blocksize - boffs, bufsize);
 981                 bh = ext3_bread(handle, inode, block, 1, &err);
 982                 if (!bh) {
 983                         CERROR("can't read/create block: %d\n", err);
 984                         goto out;
 985                 }
 986
 987                 err = ext3_journal_get_write_access(handle, bh);
 988                 if (err) {
 989                         CERROR("journal_get_write_access() returned error %d\n",
 990                                err);
 991                         goto out;
 992                 }
 993                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
 994                 memcpy(bh->b_data + boffs, buf, size);
 995                 err = ext3_journal_dirty_metadata(handle, bh);
 996                 if (err) {
 997                         CERROR("journal_dirty_metadata() returned error %d\n",
 998                                err);
 999                         goto out;
1000                 }
1001                 if (offset + size > new_size)
1002                         new_size = offset + size;
1003                 offset += size;
1004                 bufsize -= size;
1005                 buf += size;
1006         }
1007
1008         if (force_sync)
1009                 handle->h_sync = 1; /* recovery likes this */
1010 out:
1011         if (bh)
1012                 brelse(bh);
1013
1014         /* correct in-core and on-disk sizes */
1015         if (new_size > inode->i_size) {
1016                 lock_kernel();
1017                 if (new_size > inode->i_size)
1018                         inode->i_size = new_size;
1019                 if (inode->i_size > EXT3_I(inode)->i_disksize)
1020                         EXT3_I(inode)->i_disksize = inode->i_size;
1021                 if (inode->i_size > old_size)
1022                         mark_inode_dirty(inode);
1023                 unlock_kernel();
1024         }
1025
1026         lock_kernel();
1027         journal_stop(handle);
1028         unlock_kernel();
1029
1030         if (err == 0)
1031                 *offs = offset;
1032         return err;
1033 }
1034
1035 static int fsfilt_ext3_setup(struct obd_device *obd, struct super_block *sb)
1036 {
1037 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1038         struct mds_obd *mds = &obd->u.mds;
1039 #endif
1040 #if 0
1041         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1042         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1043 #endif
1044 #ifdef S_PDIROPS
1045         CWARN("Enabling PDIROPS\n");
1046         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1047         sb->s_flags |= S_PDIROPS;
1048 #endif
1049         /* setup mdsnum in underlying fs */
1050 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1051         if (mds->mds_lmv_obd) {
1052                 struct ext3_sb_info *sbi = EXT3_SB(sb);
1053                 struct ext3_super_block *es = sbi->s_es;
1054                 handle_t *handle;
1055                 int err;
1056
1057                 if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_MDSNUM)) {
1058                         CWARN("%s: set mdsnum %d in ext3fs\n",
1059                                         obd->obd_name, mds->mds_num);
1060                         lock_kernel();
1061                         handle = journal_start(sbi->s_journal, 1);
1062                         unlock_kernel();
1063                         LASSERT(!IS_ERR(handle));
1064                         err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1065                         LASSERT(err == 0);
1066                         EXT3_SET_INCOMPAT_FEATURE(sb,
1067                                                 EXT3_FEATURE_INCOMPAT_MDSNUM);
1068                         es->s_mdsnum = mds->mds_num;
1069                         err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1070                         LASSERT(err == 0);
1071                         lock_kernel();
1072                         journal_stop(handle);
1073                         unlock_kernel();
1074                 } else {
1075                         CWARN("%s: mdsnum initialized to %u in ext3fs\n",
1076                                 obd->obd_name, es->s_mdsnum);
1077                 }
1078                 sbi->s_mdsnum = es->s_mdsnum;
1079         }
1080 #endif
1081         return 0;
1082 }
1083
1084 extern int ext3_add_dir_entry(struct dentry *dentry);
1085 extern int ext3_del_dir_entry(struct dentry *dentry);
1086
1087 static int fsfilt_ext3_add_dir_entry(struct obd_device *obd,
1088                                  struct dentry *parent,
1089                                  char *name, int namelen,
1090                                  unsigned long ino,
1091                                  unsigned long generation,
1092                                  unsigned mds)
1093 {
1094 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1095         struct dentry *dentry;
1096         int err;
1097         dentry = ll_lookup_one_len(name, parent, namelen);
1098         if (IS_ERR(dentry))
1099                 RETURN(PTR_ERR(dentry));
1100         if (dentry->d_inode != NULL) {
1101                 CERROR("dentry %*s(0x%p) found\n", dentry->d_name.len,
1102                        dentry->d_name.name, dentry);
1103                 l_dput(dentry);
1104                 RETURN(-EEXIST);
1105         }
1106
1107         /* mds_reint_rename() may use this method to add dir entry
1108          * that points onto local inode. and we don't want to find
1109          * it cross-ref by subsequent lookups */
1110         d_drop(dentry);
1111
1112         dentry->d_flags |= DCACHE_CROSS_REF;
1113         dentry->d_inum = ino;
1114         dentry->d_mdsnum = mds;
1115         dentry->d_generation = generation;
1116         lock_kernel();
1117         err = ext3_add_dir_entry(dentry);
1118         unlock_kernel();
1119
1120         l_dput(dentry);
1121
1122         RETURN(err);
1123 #else
1124 #error "rebuild kernel and lustre with ext3-mds-num patch!"
1125         LASSERT(0);
1126 #endif
1127 }
1128
1129 static int fsfilt_ext3_del_dir_entry(struct obd_device *obd,
1130                                  struct dentry *dentry)
1131 {
1132 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1133         int err;
1134         lock_kernel();
1135         err = ext3_del_dir_entry(dentry);
1136         unlock_kernel();
1137         if (err == 0)
1138                 d_drop(dentry);
1139         return err;
1140 #else
1141 #error "rebuild kernel and lustre with ext3-mds-num patch!"
1142         LASSERT(0);
1143 #endif
1144 }
1145
1146 static int fsfilt_ext3_set_xattr(struct inode * inode, void *handle, char *name,
1147                                  void *buffer, int buffer_size)
1148 {
1149         int rc = 0;
1150
1151         lock_kernel();
1152
1153         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
1154                                    name, buffer, buffer_size, 0);
1155         unlock_kernel();
1156         if (rc)
1157                 CERROR("set xattr %s from inode %lu: rc %d\n",
1158                        name,  inode->i_ino, rc);
1159         return rc;
1160 }
1161
1162 static int fsfilt_ext3_get_xattr(struct inode *inode, char *name,
1163                                  void *buffer, int buffer_size)
1164 {
1165         int rc = 0;
1166
1167         lock_kernel();
1168
1169         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
1170                             name, buffer, buffer_size);
1171         unlock_kernel();
1172
1173         if (buffer == NULL)
1174                 return (rc == -ENODATA) ? 0 : rc;
1175         if (rc < 0) {
1176                 CDEBUG(D_INFO, "error getting EA %s from inode %lu: rc %d\n",
1177                        name,  inode->i_ino, rc);
1178                 memset(buffer, 0, buffer_size);
1179                 return (rc == -ENODATA) ? 0 : rc;
1180         }
1181
1182         return rc;
1183 }
1184
1185 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1186    objects. Logs is number of logfiles to update */
1187 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1188 {
1189         if ( !fso ) {
1190                 switch(op) {
1191                 case FSFILT_OP_CREATE:
1192                                  /* directory leaf, index & indirect & EA*/
1193                         return 4 + 3 * logs;
1194                 case FSFILT_OP_UNLINK:
1195                         return 3 * logs;
1196                 }
1197         } else {
1198                 int i;
1199                 int needed = 0;
1200                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1201                 int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
1202                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1203                 for (i = 0; i < op; i++, fso++) {
1204                         int nblocks = fso->fso_bufcnt * blockpp;
1205                         int ndindirect = min(nblocks, addrpp + 1);
1206                         int nindir = nblocks + ndindirect + 1;
1207
1208                         needed += nindir;
1209                 }
1210                 return needed + 3 * logs;
1211         }
1212
1213         return 0;
1214 }
1215
1216
1217 #define EXTENTS_EA "write_extents"
1218 #define EXTENTS_EA_SIZE 64
1219
1220 static int fsfilt_ext3_insert_extents_ea(struct inode *inode,
1221                                       unsigned long from,
1222                                       unsigned long num)
1223 {
1224         int rc = 0;
1225
1226         rc = ext3_ext_in_ea_alloc_space(inode, EXT3_XATTR_INDEX_TRUSTED,
1227                                         EXTENTS_EA, from, num);
1228         return rc;
1229 }
1230
1231 static int fsfilt_ext3_remove_extents_ea(struct inode *inode,
1232                                          unsigned long from,
1233                                          unsigned long num)
1234 {
1235         int rc = 0;
1236
1237         rc = ext3_ext_in_ea_remove_space(inode, EXT3_XATTR_INDEX_TRUSTED,
1238                                          EXTENTS_EA, from, num);
1239         return rc;
1240 }
1241
1242 extern int ext3_init_tree_in_ea(struct inode *inode, int name_index,
1243                                 const char *eaname, int size);
1244
1245 static int fsfilt_ext3_init_extents_ea(struct inode *inode)
1246 {
1247         int rc = 0;
1248
1249         rc = ext3_init_tree_in_ea(inode, EXT3_XATTR_INDEX_TRUSTED,
1250                                   EXTENTS_EA, 64);
1251         return rc;
1252 }
1253
1254 static int fsfilt_ext3_get_inode_write_extents(struct inode *inode,
1255                                          char **pbuf, int *size)
1256 {
1257         int rc = 0;
1258
1259         rc = ext3_ext_in_ea_get_extents(inode, EXT3_XATTR_INDEX_TRUSTED,
1260                                         EXTENTS_EA,  pbuf, size);
1261         return rc;
1262 }
1263
1264 static int fsfilt_ext3_get_write_extents_num(struct inode *inode, int *size)
1265 {
1266         int rc = 0;
1267
1268         rc = ext3_ext_in_ea_get_extents_num(inode, EXT3_XATTR_INDEX_TRUSTED,
1269                                             EXTENTS_EA, size);
1270         return rc;
1271 }
1272
1273 static struct fsfilt_operations fsfilt_ext3_ops = {
1274         .fs_type                = "ext3",
1275         .fs_owner               = THIS_MODULE,
1276         .fs_start               = fsfilt_ext3_start,
1277         .fs_brw_start           = fsfilt_ext3_brw_start,
1278         .fs_commit              = fsfilt_ext3_commit,
1279         .fs_commit_async        = fsfilt_ext3_commit_async,
1280         .fs_commit_wait         = fsfilt_ext3_commit_wait,
1281         .fs_setattr             = fsfilt_ext3_setattr,
1282         .fs_iocontrol           = fsfilt_ext3_iocontrol,
1283         .fs_set_md              = fsfilt_ext3_set_md,
1284         .fs_get_md              = fsfilt_ext3_get_md,
1285         .fs_readpage            = fsfilt_ext3_readpage,
1286         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
1287         .fs_statfs              = fsfilt_ext3_statfs,
1288         .fs_sync                = fsfilt_ext3_sync,
1289         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
1290         .fs_prep_san_write      = fsfilt_ext3_prep_san_write,
1291         .fs_write_record        = fsfilt_ext3_write_record,
1292         .fs_read_record         = fsfilt_ext3_read_record,
1293         .fs_setup               = fsfilt_ext3_setup,
1294         .fs_getpage             = fsfilt_ext3_getpage,
1295         .fs_send_bio            = fsfilt_ext3_send_bio,
1296         .fs_set_xattr           = fsfilt_ext3_set_xattr,
1297         .fs_get_xattr           = fsfilt_ext3_get_xattr,
1298         .fs_get_op_len          = fsfilt_ext3_get_op_len,
1299         .fs_add_dir_entry       = fsfilt_ext3_add_dir_entry,
1300         .fs_del_dir_entry       = fsfilt_ext3_del_dir_entry,
1301         .fs_init_extents_ea         = fsfilt_ext3_init_extents_ea,
1302         .fs_insert_extents_ea       = fsfilt_ext3_insert_extents_ea,
1303         .fs_remove_extents_ea       = fsfilt_ext3_remove_extents_ea,
1304         .fs_get_inode_write_extents = fsfilt_ext3_get_inode_write_extents,
1305         .fs_get_write_extents_num   = fsfilt_ext3_get_write_extents_num,
1306 };
1307
1308 static int __init fsfilt_ext3_init(void)
1309 {
1310         int rc;
1311
1312         fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
1313                                       sizeof(struct fsfilt_cb_data), 0,
1314                                       0, NULL, NULL);
1315         if (!fcb_cache) {
1316                 CERROR("error allocating fsfilt journal callback cache\n");
1317                 GOTO(out, rc = -ENOMEM);
1318         }
1319
1320         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
1321
1322         if (rc)
1323                 kmem_cache_destroy(fcb_cache);
1324 out:
1325         return rc;
1326 }
1327
1328 static void __exit fsfilt_ext3_exit(void)
1329 {
1330         int rc;
1331
1332         fsfilt_unregister_ops(&fsfilt_ext3_ops);
1333         rc = kmem_cache_destroy(fcb_cache);
1334
1335         if (rc || atomic_read(&fcb_cache_count)) {
1336                 CERROR("can't free fsfilt callback cache: count %d, rc = %d\n",
1337                        atomic_read(&fcb_cache_count), rc);
1338         }
1339 }
1340
1341 module_init(fsfilt_ext3_init);
1342 module_exit(fsfilt_ext3_exit);
1343
1344 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1345 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
1346 MODULE_LICENSE("GPL");