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 super_block *sb, struct inode *inode,
 314                               void *h, int force_sync)
 315 {
 316         int rc;
 317         handle_t *handle = h;
 318
 319         LASSERT(current->journal_info == handle);
 320         if (force_sync)
 321                 handle->h_sync = 1; /* recovery likes this */
 322
 323         lock_kernel();
 324         rc = journal_stop(handle);
 325         unlock_kernel();
 326
 327         return rc;
 328 }
 329
 330 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 331                                     void **wait_handle)
 332 {
 333         unsigned long tid;
 334         transaction_t *transaction;
 335 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 336         unsigned long rtid;
 337 #endif
 338         handle_t *handle = h;
 339         journal_t *journal;
 340         int rc;
 341
 342         LASSERT(current->journal_info == handle);
 343
 344         lock_kernel();
 345         transaction = handle->h_transaction;
 346         journal = transaction->t_journal;
 347         tid = transaction->t_tid;
 348         /* we don't want to be blocked */
 349         handle->h_sync = 0;
 350         rc = journal_stop(handle);
 351         if (rc) {
 352                 CERROR("error while stopping transaction: %d\n", rc);
 353                 unlock_kernel();
 354                 return rc;
 355         }
 356 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 357         rtid = log_start_commit(journal, transaction);
 358         if (rtid != tid)
 359                 CERROR("strange race: %lu != %lu\n",
 360                        (unsigned long) tid, (unsigned long) rtid);
 361 #else
 362         log_start_commit(journal, transaction->t_tid);
 363 #endif
 364         unlock_kernel();
 365
 366         *wait_handle = (void *) tid;
 367         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 368         return 0;
 369 }
 370
 371 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 372 {
 373         tid_t tid = (tid_t)(long)h;
 374
 375         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 376         if (is_journal_aborted(EXT3_JOURNAL(inode)))
 377                 return -EIO;
 378
 379         log_wait_commit(EXT3_JOURNAL(inode), tid);
 380
 381         return 0;
 382 }
 383
 384 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 385                                struct iattr *iattr, int do_trunc)
 386 {
 387         struct inode *inode = dentry->d_inode;
 388         int rc;
 389
 390         lock_kernel();
 391
 392         /* A _really_ horrible hack to avoid removing the data stored
 393          * in the block pointers; this is really the "small" stripe MD data.
 394          * We can avoid further hackery by virtue of the MDS file size being
 395          * zero all the time (which doesn't invoke block truncate at unlink
 396          * time), so we assert we never change the MDS file size from zero. */
 397         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 398                 /* ATTR_SIZE would invoke truncate: clear it */
 399                 iattr->ia_valid &= ~ATTR_SIZE;
 400                 EXT3_I(inode)->i_disksize = inode->i_size = iattr->ia_size;
 401
 402                 /* make sure _something_ gets set - so new inode
 403                  * goes to disk (probably won't work over XFS */
 404                 if (!(iattr->ia_valid & (ATTR_MODE | ATTR_MTIME | ATTR_CTIME))){
 405                         iattr->ia_valid |= ATTR_MODE;
 406                         iattr->ia_mode = inode->i_mode;
 407                 }
 408         }
 409
 410         /* Don't allow setattr to change file type */
 411         iattr->ia_mode = (inode->i_mode & S_IFMT)|(iattr->ia_mode & ~S_IFMT);
 412
 413         /* We set these flags on the client, but have already checked perms
 414          * so don't confuse inode_change_ok. */
 415         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 416
 417         if (inode->i_op->setattr) {
 418                 rc = inode->i_op->setattr(dentry, iattr);
 419         } else {
 420                 rc = inode_change_ok(inode, iattr);
 421                 if (!rc)
 422                         rc = inode_setattr(inode, iattr);
 423         }
 424
 425         unlock_kernel();
 426
 427         return rc;
 428 }
 429
 430 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 431                                  unsigned int cmd, unsigned long arg)
 432 {
 433         int rc = 0;
 434         ENTRY;
 435
 436         if (inode->i_fop->ioctl)
 437                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 438         else
 439                 RETURN(-ENOTTY);
 440
 441         RETURN(rc);
 442 }
 443
 444 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 445                               void *lmm, int lmm_size)
 446 {
 447         int rc;
 448
 449         /* keep this when we get rid of OLD_EA (too noisy during conversion) */
 450         if (EXT3_I(inode)->i_file_acl /* || large inode EA flag */)
 451                 CWARN("setting EA on %lu/%u again... interesting\n",
 452                        inode->i_ino, inode->i_generation);
 453
 454         lock_kernel();
 455         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 456                                    XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size, 0);
 457
 458         unlock_kernel();
 459
 460         if (rc)
 461                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 462                        inode->i_ino, rc);
 463         return rc;
 464 }
 465
 466 /* Must be called with i_sem held */
 467 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size)
 468 {
 469         int rc;
 470
 471         LASSERT(down_trylock(&inode->i_sem) != 0);
 472         lock_kernel();
 473
 474         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 475                             XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size);
 476         unlock_kernel();
 477
 478         /* This gives us the MD size */
 479         if (lmm == NULL)
 480                 return (rc == -ENODATA) ? 0 : rc;
 481
 482         if (rc < 0) {
 483                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 484                        EXT3_XATTR_INDEX_TRUSTED, XATTR_LUSTRE_MDS_LOV_EA,
 485                        inode->i_ino, rc);
 486                 memset(lmm, 0, lmm_size);
 487                 return (rc == -ENODATA) ? 0 : rc;
 488         }
 489
 490         return rc;
 491 }
 492
 493 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
 494 static int fsfilt_ext3_send_bio(struct inode *inode, struct bio *bio)
 495 {
 496         submit_bio(WRITE, bio);
 497         return 0;
 498 }
 499 #else
 500 static int fsfilt_ext3_send_bio(struct inode *inode, struct kiobuf *bio)
 501 {
 502         int rc, blocks_per_page;
 503
 504         rc = brw_kiovec(WRITE, 1, &bio, inode->i_dev,
 505                         bio->blocks, 1 << inode->i_blkbits);
 506
 507         blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 508
 509         if (rc != (1 << inode->i_blkbits) * bio->nr_pages *
 510             blocks_per_page) {
 511                 CERROR("short write?  expected %d, wrote %d\n",
 512                        (1 << inode->i_blkbits) * bio->nr_pages *
 513                        blocks_per_page, rc);
 514         }
 515
 516         return rc;
 517 }
 518 #endif
 519
 520 /* FIXME-UMKA: This should be used in 2.6.x io code later. */
 521 static struct page *fsfilt_ext3_getpage(struct inode *inode, long int index)
 522 {
 523         int rc;
 524         struct page *page;
 525
 526         page = grab_cache_page(inode->i_mapping, index);
 527         if (page == NULL)
 528                 return ERR_PTR(-ENOMEM);
 529
 530         if (PageUptodate(page)) {
 531                 unlock_page(page);
 532                 return page;
 533         }
 534
 535         rc = inode->i_mapping->a_ops->readpage(NULL, page);
 536         if (rc < 0) {
 537                 page_cache_release(page);
 538                 return ERR_PTR(rc);
 539         }
 540
 541         return page;
 542 }
 543
 544 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 545                                     loff_t *off)
 546 {
 547         struct inode *inode = file->f_dentry->d_inode;
 548         int rc = 0;
 549
 550         if (S_ISREG(inode->i_mode))
 551                 rc = file->f_op->read(file, buf, count, off);
 552         else {
 553                 const int blkbits = inode->i_sb->s_blocksize_bits;
 554                 const int blksize = inode->i_sb->s_blocksize;
 555
 556                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 557                        count, inode->i_ino, *off);
 558                 while (count > 0) {
 559                         struct buffer_head *bh;
 560
 561                         bh = NULL;
 562                         if (*off < inode->i_size) {
 563                                 int err = 0;
 564
 565                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 566                                                 0, &err);
 567
 568                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 569
 570                                 if (bh) {
 571                                         memcpy(buf, bh->b_data, blksize);
 572                                         brelse(bh);
 573                                 } else if (err) {
 574                                         /* XXX in theory we should just fake
 575                                          * this buffer and continue like ext3,
 576                                          * especially if this is a partial read
 577                                          */
 578                                         CERROR("error read dir %lu+%llu: %d\n",
 579                                                inode->i_ino, *off, err);
 580                                         RETURN(err);
 581                                 }
 582                         }
 583                         if (!bh) {
 584                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 585
 586                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 587                                 memset(fake, 0, sizeof(*fake));
 588                                 fake->rec_len = cpu_to_le32(blksize);
 589                         }
 590                         count -= blksize;
 591                         buf += blksize;
 592                         *off += blksize;
 593                         rc += blksize;
 594                 }
 595         }
 596
 597         return rc;
 598 }
 599
 600 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 601 {
 602         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 603
 604         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 605
 606         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 607         atomic_dec(&fcb_cache_count);
 608 }
 609
 610 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd,
 611                                       struct super_block *sb,
 612                                       __u64 last_rcvd,
 613                                       void *handle, fsfilt_cb_t cb_func,
 614                                       void *cb_data)
 615 {
 616         struct fsfilt_cb_data *fcb;
 617
 618         OBD_SLAB_ALLOC(fcb, fcb_cache, GFP_NOFS, sizeof *fcb);
 619         if (fcb == NULL)
 620                 RETURN(-ENOMEM);
 621
 622         atomic_inc(&fcb_cache_count);
 623         fcb->cb_func = cb_func;
 624         fcb->cb_obd = obd;
 625         fcb->cb_last_rcvd = last_rcvd;
 626         fcb->cb_data = cb_data;
 627
 628         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 629         lock_kernel();
 630         journal_callback_set(handle, fsfilt_ext3_cb_func,
 631                              (struct journal_callback *)fcb);
 632         unlock_kernel();
 633
 634         return 0;
 635 }
 636
 637 /*
 638  * We need to hack the return value for the free inode counts because
 639  * the current EA code requires one filesystem block per inode with EAs,
 640  * so it is possible to run out of blocks before we run out of inodes.
 641  *
 642  * This can be removed when the ext3 EA code is fixed.
 643  */
 644 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 645 {
 646         struct kstatfs sfs;
 647         int rc;
 648
 649         memset(&sfs, 0, sizeof(sfs));
 650
 651         rc = sb->s_op->statfs(sb, &sfs);
 652
 653         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 654                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 655                 sfs.f_ffree = sfs.f_bfree;
 656         }
 657
 658         statfs_pack(osfs, &sfs);
 659         return rc;
 660 }
 661
 662 static int fsfilt_ext3_sync(struct super_block *sb)
 663 {
 664         return ext3_force_commit(sb);
 665 }
 666
 667 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 668 struct bpointers {
 669         unsigned long *blocks;
 670         int *created;
 671         unsigned long start;
 672         int num;
 673         int init_num;
 674         int create;
 675 };
 676
 677 static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
 678                                   struct ext3_ext_path *path,
 679                                   struct ext3_extent *newex, int exist)
 680 {
 681         struct inode *inode = tree->inode;
 682         struct bpointers *bp = tree->private;
 683         int count, err, goal;
 684         unsigned long pblock;
 685         unsigned long tgen;
 686         loff_t new_i_size;
 687         handle_t *handle;
 688         int i;
 689
 690         i = EXT_DEPTH(tree);
 691         EXT_ASSERT(i == path->p_depth);
 692         EXT_ASSERT(path[i].p_hdr);
 693
 694         if (exist) {
 695                 err = EXT_CONTINUE;
 696                 goto map;
 697         }
 698
 699         if (bp->create == 0) {
 700                 i = 0;
 701                 if (newex->e_block < bp->start)
 702                         i = bp->start - newex->e_block;
 703                 if (i >= newex->e_num)
 704                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 705                                         i, newex->e_num);
 706                 for (; i < newex->e_num && bp->num; i++) {
 707                         *(bp->created) = 0;
 708                         bp->created++;
 709                         *(bp->blocks) = 0;
 710                         bp->blocks++;
 711                         bp->num--;
 712                         bp->start++;
 713                 }
 714
 715                 return EXT_CONTINUE;
 716         }
 717
 718         tgen = EXT_GENERATION(tree);
 719         count = ext3_ext_calc_credits_for_insert(tree, path);
 720         up_write(&EXT3_I(inode)->truncate_sem);
 721
 722         handle = ext3_journal_start(inode, count + EXT3_ALLOC_NEEDED + 1);
 723         if (IS_ERR(handle)) {
 724                 down_write(&EXT3_I(inode)->truncate_sem);
 725                 return PTR_ERR(handle);
 726         }
 727
 728         if (tgen != EXT_GENERATION(tree)) {
 729                 /* the tree has changed. so path can be invalid at moment */
 730                 ext3_journal_stop(handle, inode);
 731                 down_write(&EXT3_I(inode)->truncate_sem);
 732                 return EXT_REPEAT;
 733         }
 734
 735         down_write(&EXT3_I(inode)->truncate_sem);
 736         goal = ext3_ext_find_goal(inode, path, newex->e_block);
 737         count = newex->e_num;
 738         pblock = ext3_new_blocks(handle, inode, &count, goal, &err);
 739         if (!pblock)
 740                 goto out;
 741         EXT_ASSERT(count <= newex->e_num);
 742
 743         /* insert new extent */
 744         newex->e_start = pblock;
 745         newex->e_num = count;
 746         err = ext3_ext_insert_extent(handle, tree, path, newex);
 747         if (err)
 748                 goto out;
 749
 750         /* correct on-disk inode size */
 751         if (newex->e_num > 0) {
 752                 new_i_size = (loff_t) newex->e_block + newex->e_num;
 753                 new_i_size = new_i_size << inode->i_blkbits;
 754                 if (new_i_size > EXT3_I(inode)->i_disksize) {
 755                         EXT3_I(inode)->i_disksize = new_i_size;
 756                         err = ext3_mark_inode_dirty(handle, inode);
 757                 }
 758         }
 759
 760 out:
 761         ext3_journal_stop(handle, inode);
 762 map:
 763         if (err >= 0) {
 764                 /* map blocks */
 765                 if (bp->num == 0) {
 766                         CERROR("hmm. why do we find this extent?\n");
 767                         CERROR("initial space: %lu:%u\n",
 768                                 bp->start, bp->init_num);
 769                         CERROR("current extent: %u/%u/%u %d\n",
 770                                 newex->e_block, newex->e_num,
 771                                 newex->e_start, exist);
 772                 }
 773                 i = 0;
 774                 if (newex->e_block < bp->start)
 775                         i = bp->start - newex->e_block;
 776                 if (i >= newex->e_num)
 777                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 778                                         i, newex->e_num);
 779                 for (; i < newex->e_num && bp->num; i++) {
 780                         *(bp->created) = (exist == 0 ? 1 : 0);
 781                         bp->created++;
 782                         *(bp->blocks) = newex->e_start + i;
 783                         bp->blocks++;
 784                         bp->num--;
 785                 }
 786         }
 787         return err;
 788 }
 789
 790 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
 791                        unsigned long num, unsigned long *blocks,
 792                        int *created, int create)
 793 {
 794         struct ext3_extents_tree tree;
 795         struct bpointers bp;
 796         int err;
 797
 798         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
 799                 block, block + num, (unsigned) inode->i_ino);
 800
 801         ext3_init_tree_desc(&tree, inode);
 802         tree.private = &bp;
 803         bp.blocks = blocks;
 804         bp.created = created;
 805         bp.start = block;
 806         bp.init_num = bp.num = num;
 807         bp.create = create;
 808
 809         down_write(&EXT3_I(inode)->truncate_sem);
 810         err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
 811         ext3_ext_invalidate_cache(&tree);
 812         up_write(&EXT3_I(inode)->truncate_sem);
 813
 814         return err;
 815 }
 816
 817 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
 818                                     int pages, unsigned long *blocks,
 819                                     int *created, int create)
 820 {
 821         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 822         int rc = 0, i = 0;
 823         struct page *fp = NULL;
 824         int clen = 0;
 825
 826         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
 827                 inode->i_ino, pages, (*page)->index);
 828
 829         /* pages are sorted already. so, we just have to find
 830          * contig. space and process them properly */
 831         while (i < pages) {
 832                 if (fp == NULL) {
 833                         /* start new extent */
 834                         fp = *page++;
 835                         clen = 1;
 836                         i++;
 837                         continue;
 838                 } else if (fp->index + clen == (*page)->index) {
 839                         /* continue the extent */
 840                         page++;
 841                         clen++;
 842                         i++;
 843                         continue;
 844                 }
 845
 846                 /* process found extent */
 847                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 848                                         clen * blocks_per_page, blocks,
 849                                         created, create);
 850                 if (rc)
 851                         GOTO(cleanup, rc);
 852
 853                 /* look for next extent */
 854                 fp = NULL;
 855                 blocks += blocks_per_page * clen;
 856                 created += blocks_per_page * clen;
 857         }
 858
 859         if (fp)
 860                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 861                                         clen * blocks_per_page, blocks,
 862                                         created, create);
 863 cleanup:
 864         return rc;
 865 }
 866 #endif
 867
 868 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
 869                                unsigned long *blocks, int *created, int create);
 870 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
 871                                    int pages, unsigned long *blocks,
 872                                    int *created, int create)
 873 {
 874         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 875         unsigned long *b;
 876         int rc = 0, i, *cr;
 877
 878         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
 879                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
 880                 if (rc) {
 881                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
 882                                inode->i_ino, *b, *cr, create, rc);
 883                         break;
 884                 }
 885
 886                 b += blocks_per_page;
 887                 cr += blocks_per_page;
 888         }
 889         return rc;
 890 }
 891
 892 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
 893                                 int pages, unsigned long *blocks,
 894                                 int *created, int create,
 895                                 struct semaphore *optional_sem)
 896 {
 897         int rc;
 898 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 899         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
 900                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
 901                                                      blocks, created, create);
 902                 return rc;
 903         }
 904 #endif
 905         if (optional_sem != NULL)
 906                 down(optional_sem);
 907         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
 908                                             created, create);
 909         if (optional_sem != NULL)
 910                 up(optional_sem);
 911
 912         return rc;
 913 }
 914
 915 extern int ext3_prep_san_write(struct inode *inode, long *blocks,
 916                                int nblocks, loff_t newsize);
 917 static int fsfilt_ext3_prep_san_write(struct inode *inode, long *blocks,
 918                                       int nblocks, loff_t newsize)
 919 {
 920         return ext3_prep_san_write(inode, blocks, nblocks, newsize);
 921 }
 922
 923 static int fsfilt_ext3_read_record(struct file * file, void *buf,
 924                                    int size, loff_t *offs)
 925 {
 926         struct inode *inode = file->f_dentry->d_inode;
 927         unsigned long block;
 928         struct buffer_head *bh;
 929         int err, blocksize, csize, boffs;
 930
 931         /* prevent reading after eof */
 932         lock_kernel();
 933         if (inode->i_size < *offs + size) {
 934                 size = inode->i_size - *offs;
 935                 unlock_kernel();
 936                 if (size < 0) {
 937                         CERROR("size %llu is too short for read %u@%llu\n",
 938                                inode->i_size, size, *offs);
 939                         return -EIO;
 940                 } else if (size == 0) {
 941                         return 0;
 942                 }
 943         } else {
 944                 unlock_kernel();
 945         }
 946
 947         blocksize = 1 << inode->i_blkbits;
 948
 949         while (size > 0) {
 950                 block = *offs >> inode->i_blkbits;
 951                 boffs = *offs & (blocksize - 1);
 952                 csize = min(blocksize - boffs, size);
 953                 bh = ext3_bread(NULL, inode, block, 0, &err);
 954                 if (!bh) {
 955                         CERROR("can't read block: %d\n", err);
 956                         return err;
 957                 }
 958
 959                 memcpy(buf, bh->b_data + boffs, csize);
 960                 brelse(bh);
 961
 962                 *offs += csize;
 963                 buf += csize;
 964                 size -= csize;
 965         }
 966         return 0;
 967 }
 968
 969 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
 970                                     loff_t *offs, int force_sync)
 971 {
 972         struct buffer_head *bh = NULL;
 973         unsigned long block;
 974         struct inode *inode = file->f_dentry->d_inode;
 975         loff_t old_size = inode->i_size, offset = *offs;
 976         loff_t new_size = inode->i_size;
 977         journal_t *journal;
 978         handle_t *handle;
 979         int err, block_count = 0, blocksize, size, boffs;
 980
 981         /* Determine how many transaction credits are needed */
 982         blocksize = 1 << inode->i_blkbits;
 983         block_count = (*offs & (blocksize - 1)) + bufsize;
 984         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
 985
 986         journal = EXT3_SB(inode->i_sb)->s_journal;
 987         lock_kernel();
 988         handle = journal_start(journal,
 989                                block_count * EXT3_DATA_TRANS_BLOCKS + 2);
 990         unlock_kernel();
 991         if (IS_ERR(handle)) {
 992                 CERROR("can't start transaction\n");
 993                 return PTR_ERR(handle);
 994         }
 995
 996         while (bufsize > 0) {
 997                 if (bh != NULL)
 998                         brelse(bh);
 999
1000                 block = offset >> inode->i_blkbits;
1001                 boffs = offset & (blocksize - 1);
1002                 size = min(blocksize - boffs, bufsize);
1003                 bh = ext3_bread(handle, inode, block, 1, &err);
1004                 if (!bh) {
1005                         CERROR("can't read/create block: %d\n", err);
1006                         goto out;
1007                 }
1008
1009                 err = ext3_journal_get_write_access(handle, bh);
1010                 if (err) {
1011                         CERROR("journal_get_write_access() returned error %d\n",
1012                                err);
1013                         goto out;
1014                 }
1015                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1016                 memcpy(bh->b_data + boffs, buf, size);
1017                 err = ext3_journal_dirty_metadata(handle, bh);
1018                 if (err) {
1019                         CERROR("journal_dirty_metadata() returned error %d\n",
1020                                err);
1021                         goto out;
1022                 }
1023                 if (offset + size > new_size)
1024                         new_size = offset + size;
1025                 offset += size;
1026                 bufsize -= size;
1027                 buf += size;
1028         }
1029
1030         if (force_sync)
1031                 handle->h_sync = 1; /* recovery likes this */
1032 out:
1033         if (bh)
1034                 brelse(bh);
1035
1036         /* correct in-core and on-disk sizes */
1037         if (new_size > inode->i_size) {
1038                 lock_kernel();
1039                 if (new_size > inode->i_size)
1040                         inode->i_size = new_size;
1041                 if (inode->i_size > EXT3_I(inode)->i_disksize)
1042                         EXT3_I(inode)->i_disksize = inode->i_size;
1043                 if (inode->i_size > old_size)
1044                         mark_inode_dirty(inode);
1045                 unlock_kernel();
1046         }
1047
1048         lock_kernel();
1049         journal_stop(handle);
1050         unlock_kernel();
1051
1052         if (err == 0)
1053                 *offs = offset;
1054         return err;
1055 }
1056
1057 static int fsfilt_ext3_setup(struct obd_device *obd, struct super_block *sb)
1058 {
1059 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1060         struct mds_obd *mds = &obd->u.mds;
1061 #endif
1062 #if 0
1063         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1064         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1065 #endif
1066 #ifdef S_PDIROPS
1067         CWARN("Enabling PDIROPS\n");
1068         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1069         sb->s_flags |= S_PDIROPS;
1070 #endif
1071         /* setup mdsnum in underlying fs */
1072 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1073         if (mds->mds_lmv_obd) {
1074                 struct ext3_sb_info *sbi = EXT3_SB(sb);
1075                 struct ext3_super_block *es = sbi->s_es;
1076                 handle_t *handle;
1077                 int err;
1078
1079                 if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_MDSNUM)) {
1080                         CWARN("%s: set mdsnum %d in ext3fs\n",
1081                                         obd->obd_name, mds->mds_num);
1082                         lock_kernel();
1083                         handle = journal_start(sbi->s_journal, 1);
1084                         unlock_kernel();
1085                         LASSERT(!IS_ERR(handle));
1086                         err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1087                         LASSERT(err == 0);
1088                         EXT3_SET_INCOMPAT_FEATURE(sb,
1089                                                 EXT3_FEATURE_INCOMPAT_MDSNUM);
1090                         es->s_mdsnum = mds->mds_num;
1091                         err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1092                         LASSERT(err == 0);
1093                         lock_kernel();
1094                         journal_stop(handle);
1095                         unlock_kernel();
1096                 } else {
1097                         CWARN("%s: mdsnum initialized to %u in ext3fs\n",
1098                                 obd->obd_name, es->s_mdsnum);
1099                 }
1100                 sbi->s_mdsnum = es->s_mdsnum;
1101         }
1102 #endif
1103         return 0;
1104 }
1105
1106 extern int ext3_add_dir_entry(struct dentry *dentry);
1107 extern int ext3_del_dir_entry(struct dentry *dentry);
1108
1109 static int fsfilt_ext3_add_dir_entry(struct obd_device *obd,
1110                                  struct dentry *parent,
1111                                  char *name, int namelen,
1112                                  unsigned long ino,
1113                                  unsigned long generation,
1114                                  unsigned mds)
1115 {
1116 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1117         struct dentry *dentry;
1118         int err;
1119         dentry = ll_lookup_one_len(name, parent, namelen);
1120         if (IS_ERR(dentry)) {
1121                 CERROR("can't lookup %*s in %lu/%lu: %d\n", dentry->d_name.len,
1122                        dentry->d_name.name, dentry->d_inode->i_ino,
1123                        (unsigned long) dentry->d_inode->i_generation,
1124                        (int) PTR_ERR(dentry));
1125                 RETURN(PTR_ERR(dentry));
1126         }
1127         if (dentry->d_inode != NULL) {
1128                 CERROR("dentry %*s(0x%p) found\n", dentry->d_name.len,
1129                        dentry->d_name.name, dentry);
1130                 l_dput(dentry);
1131                 RETURN(-EEXIST);
1132         }
1133
1134         /* mds_reint_rename() may use this method to add dir entry
1135          * that points onto local inode. and we don't want to find
1136          * it cross-ref by subsequent lookups */
1137         d_drop(dentry);
1138
1139         dentry->d_flags |= DCACHE_CROSS_REF;
1140         dentry->d_inum = ino;
1141         dentry->d_mdsnum = mds;
1142         dentry->d_generation = generation;
1143         lock_kernel();
1144         err = ext3_add_dir_entry(dentry);
1145         unlock_kernel();
1146
1147         l_dput(dentry);
1148
1149         RETURN(err);
1150 #else
1151 #error "rebuild kernel and lustre with ext3-mds-num patch!"
1152         LASSERT(0);
1153 #endif
1154 }
1155
1156 static int fsfilt_ext3_del_dir_entry(struct obd_device *obd,
1157                                  struct dentry *dentry)
1158 {
1159 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1160         int err;
1161         lock_kernel();
1162         err = ext3_del_dir_entry(dentry);
1163         unlock_kernel();
1164         if (err == 0)
1165                 d_drop(dentry);
1166         return err;
1167 #else
1168 #error "rebuild kernel and lustre with ext3-mds-num patch!"
1169         LASSERT(0);
1170 #endif
1171 }
1172
1173 static int fsfilt_ext3_set_xattr(struct inode * inode, void *handle, char *name,
1174                                  void *buffer, int buffer_size)
1175 {
1176         int rc = 0;
1177
1178         lock_kernel();
1179
1180         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
1181                                    name, buffer, buffer_size, 0);
1182         unlock_kernel();
1183         if (rc)
1184                 CERROR("set xattr %s from inode %lu: rc %d\n",
1185                        name,  inode->i_ino, rc);
1186         return rc;
1187 }
1188
1189 static int fsfilt_ext3_get_xattr(struct inode *inode, char *name,
1190                                  void *buffer, int buffer_size)
1191 {
1192         int rc = 0;
1193
1194         lock_kernel();
1195
1196         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
1197                             name, buffer, buffer_size);
1198         unlock_kernel();
1199
1200         if (buffer == NULL)
1201                 return (rc == -ENODATA) ? 0 : rc;
1202         if (rc < 0) {
1203                 CDEBUG(D_INFO, "error getting EA %s from inode %lu: rc %d\n",
1204                        name,  inode->i_ino, rc);
1205                 memset(buffer, 0, buffer_size);
1206                 return (rc == -ENODATA) ? 0 : rc;
1207         }
1208
1209         return rc;
1210 }
1211
1212 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1213    objects. Logs is number of logfiles to update */
1214 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1215 {
1216         if ( !fso ) {
1217                 switch(op) {
1218                 case FSFILT_OP_CREATE:
1219                                  /* directory leaf, index & indirect & EA*/
1220                         return 4 + 3 * logs;
1221                 case FSFILT_OP_UNLINK:
1222                         return 3 * logs;
1223                 }
1224         } else {
1225                 int i;
1226                 int needed = 0;
1227                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1228                 int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
1229                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1230                 for (i = 0; i < op; i++, fso++) {
1231                         int nblocks = fso->fso_bufcnt * blockpp;
1232                         int ndindirect = min(nblocks, addrpp + 1);
1233                         int nindir = nblocks + ndindirect + 1;
1234
1235                         needed += nindir;
1236                 }
1237                 return needed + 3 * logs;
1238         }
1239
1240         return 0;
1241 }
1242
1243
1244 #define EXTENTS_EA "write_extents"
1245 #define EXTENTS_EA_SIZE 64
1246
1247 static int fsfilt_ext3_insert_extents_ea(struct inode *inode,
1248                                       unsigned long from,
1249                                       unsigned long num)
1250 {
1251         int rc = 0;
1252
1253         rc = ext3_ext_in_ea_alloc_space(inode, EXT3_XATTR_INDEX_TRUSTED,
1254                                         EXTENTS_EA, from, num);
1255         return rc;
1256 }
1257
1258 static int fsfilt_ext3_remove_extents_ea(struct inode *inode,
1259                                          unsigned long from,
1260                                          unsigned long num)
1261 {
1262         int rc = 0;
1263
1264         rc = ext3_ext_in_ea_remove_space(inode, EXT3_XATTR_INDEX_TRUSTED,
1265                                          EXTENTS_EA, from, num);
1266         return rc;
1267 }
1268
1269 extern int ext3_init_tree_in_ea(struct inode *inode, int name_index,
1270                                 const char *eaname, int size);
1271
1272 static int fsfilt_ext3_init_extents_ea(struct inode *inode)
1273 {
1274         int rc = 0;
1275
1276         rc = ext3_init_tree_in_ea(inode, EXT3_XATTR_INDEX_TRUSTED,
1277                                   EXTENTS_EA, 64);
1278         return rc;
1279 }
1280
1281 static int fsfilt_ext3_get_inode_write_extents(struct inode *inode,
1282                                          char **pbuf, int *size)
1283 {
1284         int rc = 0;
1285
1286         rc = ext3_ext_in_ea_get_extents(inode, EXT3_XATTR_INDEX_TRUSTED,
1287                                         EXTENTS_EA,  pbuf, size);
1288         return rc;
1289 }
1290
1291 static int fsfilt_ext3_get_write_extents_num(struct inode *inode, int *size)
1292 {
1293         int rc = 0;
1294
1295         rc = ext3_ext_in_ea_get_extents_num(inode, EXT3_XATTR_INDEX_TRUSTED,
1296                                             EXTENTS_EA, size);
1297         return rc;
1298 }
1299
1300 static struct fsfilt_operations fsfilt_ext3_ops = {
1301         .fs_type                = "ext3",
1302         .fs_owner               = THIS_MODULE,
1303         .fs_start               = fsfilt_ext3_start,
1304         .fs_brw_start           = fsfilt_ext3_brw_start,
1305         .fs_commit              = fsfilt_ext3_commit,
1306         .fs_commit_async        = fsfilt_ext3_commit_async,
1307         .fs_commit_wait         = fsfilt_ext3_commit_wait,
1308         .fs_setattr             = fsfilt_ext3_setattr,
1309         .fs_iocontrol           = fsfilt_ext3_iocontrol,
1310         .fs_set_md              = fsfilt_ext3_set_md,
1311         .fs_get_md              = fsfilt_ext3_get_md,
1312         .fs_readpage            = fsfilt_ext3_readpage,
1313         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
1314         .fs_statfs              = fsfilt_ext3_statfs,
1315         .fs_sync                = fsfilt_ext3_sync,
1316         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
1317         .fs_prep_san_write      = fsfilt_ext3_prep_san_write,
1318         .fs_write_record        = fsfilt_ext3_write_record,
1319         .fs_read_record         = fsfilt_ext3_read_record,
1320         .fs_setup               = fsfilt_ext3_setup,
1321         .fs_getpage             = fsfilt_ext3_getpage,
1322         .fs_send_bio            = fsfilt_ext3_send_bio,
1323         .fs_set_xattr           = fsfilt_ext3_set_xattr,
1324         .fs_get_xattr           = fsfilt_ext3_get_xattr,
1325         .fs_get_op_len          = fsfilt_ext3_get_op_len,
1326         .fs_add_dir_entry       = fsfilt_ext3_add_dir_entry,
1327         .fs_del_dir_entry       = fsfilt_ext3_del_dir_entry,
1328         .fs_init_extents_ea         = fsfilt_ext3_init_extents_ea,
1329         .fs_insert_extents_ea       = fsfilt_ext3_insert_extents_ea,
1330         .fs_remove_extents_ea       = fsfilt_ext3_remove_extents_ea,
1331         .fs_get_inode_write_extents = fsfilt_ext3_get_inode_write_extents,
1332         .fs_get_write_extents_num   = fsfilt_ext3_get_write_extents_num,
1333 };
1334
1335 static int __init fsfilt_ext3_init(void)
1336 {
1337         int rc;
1338
1339         fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
1340                                       sizeof(struct fsfilt_cb_data), 0,
1341                                       0, NULL, NULL);
1342         if (!fcb_cache) {
1343                 CERROR("error allocating fsfilt journal callback cache\n");
1344                 GOTO(out, rc = -ENOMEM);
1345         }
1346
1347         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
1348
1349         if (rc)
1350                 kmem_cache_destroy(fcb_cache);
1351 out:
1352         return rc;
1353 }
1354
1355 static void __exit fsfilt_ext3_exit(void)
1356 {
1357         int rc;
1358
1359         fsfilt_unregister_ops(&fsfilt_ext3_ops);
1360         rc = kmem_cache_destroy(fcb_cache);
1361
1362         if (rc || atomic_read(&fcb_cache_count)) {
1363                 CERROR("can't free fsfilt callback cache: count %d, rc = %d\n",
1364                        atomic_read(&fcb_cache_count), rc);
1365         }
1366 }
1367
1368 module_init(fsfilt_ext3_init);
1369 module_exit(fsfilt_ext3_exit);
1370
1371 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1372 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
1373 MODULE_LICENSE("GPL");