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 <libcfs/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 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,7))
  56 # define lock_24kernel() lock_kernel()
  57 # define unlock_24kernel() unlock_kernel()
  58 #else
  59 # define lock_24kernel() do {} while (0)
  60 # define unlock_24kernel() do {} while (0)
  61 #endif
  62
  63 static kmem_cache_t *fcb_cache;
  64 static atomic_t fcb_cache_count = ATOMIC_INIT(0);
  65
  66 struct fsfilt_cb_data {
  67         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  68         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  69         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  70         __u64 cb_last_num;              /* MDS/OST last committed operation */
  71         void *cb_data;                  /* MDS/OST completion function data */
  72 };
  73
  74 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
  75 #define EXT3_XATTR_INDEX_TRUSTED        4
  76 #endif
  77
  78 #define XATTR_LUSTRE_MDS_LOV_EA         "lov"
  79 #define XATTR_LUSTRE_MDS_MEA_EA         "mea"
  80 #define XATTR_LUSTRE_MDS_MID_EA         "mid"
  81 #define XATTR_LUSTRE_MDS_SID_EA         "sid"
  82
  83 /*
  84  * We don't currently need any additional blocks for rmdir and
  85  * unlink transactions because we are storing the OST oa_id inside
  86  * the inode (which we will be changing anyways as part of this
  87  * transaction).
  88  */
  89 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
  90                                int logs)
  91 {
  92         /* For updates to the last recieved file */
  93         int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS;
  94         journal_t *journal;
  95         void *handle;
  96
  97         if (current->journal_info) {
  98                 CDEBUG(D_INODE, "increasing refcount on %p\n",
  99                        current->journal_info);
 100                 goto journal_start;
 101         }
 102
 103         /* XXX BUG 3188 -- must return to one set of opcodes */
 104         /* FIXME - cache hook */
 105         if (op & 0x20) {
 106                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
 107                 op = op & ~0x20;
 108         }
 109
 110         /* FIXME - kml */
 111         if (op & 0x10) {
 112                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
 113                 op = op & ~0x10;
 114         }
 115
 116         switch(op) {
 117         case FSFILT_OP_RMDIR:
 118         case FSFILT_OP_UNLINK:
 119                 /* delete one file + create/update logs for each stripe */
 120                 nblocks += EXT3_DELETE_TRANS_BLOCKS;
 121                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 122                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 123                 break;
 124         case FSFILT_OP_RENAME:
 125                 /* modify additional directory */
 126                 nblocks += EXT3_SINGLEDATA_TRANS_BLOCKS;
 127                 /* no break */
 128         case FSFILT_OP_SYMLINK:
 129                 /* additional block + block bitmap + GDT for long symlink */
 130                 nblocks += 3;
 131                 /* no break */
 132         case FSFILT_OP_CREATE:
 133                 /* create/update logs for each stripe */
 134                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 135                             EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
 136                 /* no break */
 137         case FSFILT_OP_MKDIR:
 138         case FSFILT_OP_MKNOD:
 139                 /* modify one inode + block bitmap + GDT */
 140                 nblocks += 3;
 141                 /* no break */
 142         case FSFILT_OP_LINK:
 143                 /* modify parent directory */
 144                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 145                         EXT3_DATA_TRANS_BLOCKS;
 146                 break;
 147         case FSFILT_OP_SETATTR:
 148                 /* Setattr on inode */
 149                 nblocks += 1;
 150                 break;
 151         case FSFILT_OP_CANCEL_UNLINK:
 152                 /* blocks for log header bitmap update OR
 153                  * blocks for catalog header bitmap update + unlink of logs */
 154                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 155                         EXT3_DELETE_TRANS_BLOCKS * logs;
 156                 break;
 157         case FSFILT_OP_NOOP:
 158                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
 159                 break;
 160         default: CERROR("unknown transaction start op %d\n", op);
 161                 LBUG();
 162         }
 163
 164         LASSERT(current->journal_info == desc_private);
 165         journal = EXT3_SB(inode->i_sb)->s_journal;
 166         if (nblocks > journal->j_max_transaction_buffers) {
 167                 CERROR("too many credits %d for op %ux%u using %d instead\n",
 168                        nblocks, op, logs, journal->j_max_transaction_buffers);
 169                 nblocks = journal->j_max_transaction_buffers;
 170         }
 171
 172  journal_start:
 173         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 174         lock_24kernel();
 175         handle = journal_start(EXT3_JOURNAL(inode), nblocks);
 176         unlock_24kernel();
 177
 178         if (!IS_ERR(handle))
 179                 LASSERT(current->journal_info == handle);
 180         else
 181                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 182                        op, nblocks, PTR_ERR(handle));
 183         return handle;
 184 }
 185
 186 /*
 187  * Calculate the number of buffer credits needed to write multiple pages in
 188  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 189  * doesn't have a nice API for calculating this sort of thing in advance.
 190  *
 191  * See comment above ext3_writepage_trans_blocks for details.  We assume
 192  * no data journaling is being done, but it does allow for all of the pages
 193  * being non-contiguous.  If we are guaranteed contiguous pages we could
 194  * reduce the number of (d)indirect blocks a lot.
 195  *
 196  * With N blocks per page and P pages, for each inode we have at most:
 197  * N*P indirect
 198  * min(N*P, blocksize/4 + 1) dindirect blocks
 199  * niocount tindirect
 200  *
 201  * For the entire filesystem, we have at most:
 202  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 203  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 204  * objcount inode blocks
 205  * 1 superblock
 206  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 207  *
 208  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 209  */
 210 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 211                                       int niocount, struct niobuf_local *nb)
 212 {
 213         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 214         __u64 next_indir;
 215         const int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
 216         int nbitmaps = 0, ngdblocks;
 217         int needed = objcount + 1; /* inodes + superblock */
 218         int i, j;
 219
 220         for (i = 0, j = 0; i < objcount; i++, fso++) {
 221                 /* two or more dindirect blocks in case we cross boundary */
 222                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 223                                     nb[j].offset) >>
 224                                    sb->s_blocksize_bits) /
 225                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 226                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 227
 228                 /* leaf, indirect, tindirect blocks for first block */
 229                 nbitmaps += blockpp + 2;
 230
 231                 j += fso->fso_bufcnt;
 232         }
 233
 234         next_indir = nb[0].offset +
 235                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 236         for (i = 1; i < niocount; i++) {
 237                 if (nb[i].offset >= next_indir) {
 238                         nbitmaps++;     /* additional indirect */
 239                         next_indir = nb[i].offset +
 240                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 241                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 242                         nbitmaps++;     /* additional indirect */
 243                 }
 244                 nbitmaps += blockpp;    /* each leaf in different group? */
 245         }
 246
 247         ngdblocks = nbitmaps;
 248         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 249                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 250         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 251                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 252
 253         needed += nbitmaps + ngdblocks;
 254
 255         /* last_rcvd update */
 256         needed += EXT3_DATA_TRANS_BLOCKS;
 257
 258 #if defined(CONFIG_QUOTA) && !defined(__x86_64__) /* XXX */
 259         /* We assume that there will be 1 bit set in s_dquot.flags for each
 260          * quota file that is active.  This is at least true for now.
 261          */
 262         needed += hweight32(sb_any_quota_enabled(sb)) *
 263                 EXT3_SINGLEDATA_TRANS_BLOCKS;
 264 #endif
 265
 266         return needed;
 267 }
 268
 269 /* We have to start a huge journal transaction here to hold all of the
 270  * metadata for the pages being written here.  This is necessitated by
 271  * the fact that we do lots of prepare_write operations before we do
 272  * any of the matching commit_write operations, so even if we split
 273  * up to use "smaller" transactions none of them could complete until
 274  * all of them were opened.  By having a single journal transaction,
 275  * we eliminate duplicate reservations for common blocks like the
 276  * superblock and group descriptors or bitmaps.
 277  *
 278  * We will start the transaction here, but each prepare_write will
 279  * add a refcount to the transaction, and each commit_write will
 280  * remove a refcount.  The transaction will be closed when all of
 281  * the pages have been written.
 282  */
 283 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 284                                    int niocount, struct niobuf_local *nb,
 285                                    void *desc_private, int logs)
 286 {
 287         journal_t *journal;
 288         handle_t *handle;
 289         int needed;
 290         ENTRY;
 291
 292         LASSERT(current->journal_info == desc_private);
 293         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 294         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 295
 296         /* The number of blocks we could _possibly_ dirty can very large.
 297          * We reduce our request if it is absurd (and we couldn't get that
 298          * many credits for a single handle anyways).
 299          *
 300          * At some point we have to limit the size of I/Os sent at one time,
 301          * increase the size of the journal, or we have to calculate the
 302          * actual journal requirements more carefully by checking all of
 303          * the blocks instead of being maximally pessimistic.  It remains to
 304          * be seen if this is a real problem or not.
 305          */
 306         if (needed > journal->j_max_transaction_buffers) {
 307                 CERROR("want too many journal credits (%d) using %d instead\n",
 308                        needed, journal->j_max_transaction_buffers);
 309                 needed = journal->j_max_transaction_buffers;
 310         }
 311
 312         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 313         lock_24kernel();
 314         handle = journal_start(journal, needed);
 315         unlock_24kernel();
 316         if (IS_ERR(handle)) {
 317                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 318                        PTR_ERR(handle));
 319         } else {
 320                 LASSERT(handle->h_buffer_credits >= needed);
 321                 LASSERT(current->journal_info == handle);
 322         }
 323
 324         RETURN(handle);
 325 }
 326
 327 static int fsfilt_ext3_commit(struct super_block *sb, struct inode *inode,
 328                               void *h, int force_sync)
 329 {
 330         int rc;
 331         handle_t *handle = h;
 332
 333         LASSERT(current->journal_info == handle);
 334         if (force_sync)
 335                 handle->h_sync = 1; /* recovery likes this */
 336
 337         lock_24kernel();
 338         rc = journal_stop(handle);
 339         unlock_24kernel();
 340
 341         return rc;
 342 }
 343
 344 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 345                                     void **wait_handle)
 346 {
 347         unsigned long tid;
 348         transaction_t *transaction;
 349 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 350         unsigned long rtid;
 351 #endif
 352         handle_t *handle = h;
 353         journal_t *journal;
 354         int rc;
 355
 356         LASSERT(current->journal_info == handle);
 357
 358         lock_kernel();
 359         transaction = handle->h_transaction;
 360         journal = transaction->t_journal;
 361         tid = transaction->t_tid;
 362         /* we don't want to be blocked */
 363         handle->h_sync = 0;
 364         rc = journal_stop(handle);
 365         if (rc) {
 366                 CERROR("error while stopping transaction: %d\n", rc);
 367                 unlock_kernel();
 368                 return rc;
 369         }
 370 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 371         rtid = log_start_commit(journal, transaction);
 372         if (rtid != tid)
 373                 CERROR("strange race: %lu != %lu\n",
 374                        (unsigned long) tid, (unsigned long) rtid);
 375 #else
 376         log_start_commit(journal, transaction->t_tid);
 377 #endif
 378         unlock_kernel();
 379
 380         *wait_handle = (void *) tid;
 381         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 382         return 0;
 383 }
 384
 385 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 386 {
 387         tid_t tid = (tid_t)(long)h;
 388
 389         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 390         if (is_journal_aborted(EXT3_JOURNAL(inode)))
 391                 return -EIO;
 392
 393         log_wait_commit(EXT3_JOURNAL(inode), tid);
 394
 395         return 0;
 396 }
 397
 398 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 399                                struct iattr *iattr, int do_trunc)
 400 {
 401         struct inode *inode = dentry->d_inode;
 402         int rc;
 403
 404         lock_kernel();
 405
 406         /* A _really_ horrible hack to avoid removing the data stored
 407          * in the block pointers; this is really the "small" stripe MD data.
 408          * We can avoid further hackery by virtue of the MDS file size being
 409          * zero all the time (which doesn't invoke block truncate at unlink
 410          * time), so we assert we never change the MDS file size from zero. */
 411         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 412                 /* ATTR_SIZE would invoke truncate: clear it */
 413                 iattr->ia_valid &= ~ATTR_SIZE;
 414                 EXT3_I(inode)->i_disksize = inode->i_size = iattr->ia_size;
 415
 416                 /* make sure _something_ gets set - so new inode
 417                  * goes to disk (probably won't work over XFS */
 418                 if (!(iattr->ia_valid & (ATTR_MODE | ATTR_MTIME | ATTR_CTIME))){
 419                         iattr->ia_valid |= ATTR_MODE;
 420                         iattr->ia_mode = inode->i_mode;
 421                 }
 422         }
 423
 424         /* Don't allow setattr to change file type */
 425         iattr->ia_mode = (inode->i_mode & S_IFMT)|(iattr->ia_mode & ~S_IFMT);
 426
 427         /* We set these flags on the client, but have already checked perms
 428          * so don't confuse inode_change_ok. */
 429         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 430
 431         if (inode->i_op->setattr) {
 432                 rc = inode->i_op->setattr(dentry, iattr);
 433         } else {
 434                 rc = inode_change_ok(inode, iattr);
 435                 if (!rc)
 436                         rc = inode_setattr(inode, iattr);
 437         }
 438
 439         unlock_kernel();
 440
 441         return rc;
 442 }
 443
 444 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 445                                  unsigned int cmd, unsigned long arg)
 446 {
 447         int rc = 0;
 448         ENTRY;
 449
 450         if (inode->i_fop->ioctl)
 451                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 452         else
 453                 RETURN(-ENOTTY);
 454
 455         RETURN(rc);
 456 }
 457
 458 static int fsfilt_ext3_set_xattr(struct inode * inode, void *handle, char *name,
 459                                  void *buffer, int buffer_size)
 460 {
 461         int rc = 0;
 462
 463         lock_kernel();
 464
 465         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 466                                    name, buffer, buffer_size, 0);
 467         unlock_kernel();
 468         if (rc)
 469                 CERROR("set xattr %s from inode %lu: rc %d\n",
 470                        name,  inode->i_ino, rc);
 471         return rc;
 472 }
 473
 474 static int fsfilt_ext3_get_xattr(struct inode *inode, char *name,
 475                                  void *buffer, int buffer_size)
 476 {
 477         int rc = 0;
 478
 479         lock_kernel();
 480
 481         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 482                             name, buffer, buffer_size);
 483         unlock_kernel();
 484
 485         if (buffer == NULL)
 486                 return (rc == -ENODATA) ? 0 : rc;
 487         if (rc < 0) {
 488                 CDEBUG(D_INFO, "error getting EA %s from inode %lu: rc %d\n",
 489                        name,  inode->i_ino, rc);
 490                 memset(buffer, 0, buffer_size);
 491                 return (rc == -ENODATA) ? 0 : rc;
 492         }
 493
 494         return rc;
 495 }
 496
 497 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 498                               void *lmm, int lmm_size,
 499                               enum ea_type type)
 500 {
 501         int rc;
 502
 503         switch(type) {
 504         case EA_LOV:
 505                 rc = fsfilt_ext3_set_xattr(inode, handle,
 506                                            XATTR_LUSTRE_MDS_LOV_EA,
 507                                            lmm, lmm_size);
 508                 break;
 509         case EA_MEA:
 510                 rc = fsfilt_ext3_set_xattr(inode, handle,
 511                                            XATTR_LUSTRE_MDS_MEA_EA,
 512                                            lmm, lmm_size);
 513                 break;
 514         case EA_SID:
 515                 rc = fsfilt_ext3_set_xattr(inode, handle,
 516                                            XATTR_LUSTRE_MDS_SID_EA,
 517                                            lmm, lmm_size);
 518                 break;
 519         case EA_MID:
 520                 rc = fsfilt_ext3_set_xattr(inode, handle,
 521                                            XATTR_LUSTRE_MDS_MID_EA,
 522                                            lmm, lmm_size);
 523                 break;
 524         default:
 525                 return -EINVAL;
 526         }
 527
 528         return rc;
 529 }
 530
 531 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm,
 532                               int lmm_size, enum ea_type type)
 533 {
 534         int rc;
 535
 536         switch (type) {
 537         case EA_LOV:
 538                 rc = fsfilt_ext3_get_xattr(inode,
 539                                            XATTR_LUSTRE_MDS_LOV_EA,
 540                                            lmm, lmm_size);
 541                 break;
 542         case EA_MEA:
 543                 rc = fsfilt_ext3_get_xattr(inode,
 544                                            XATTR_LUSTRE_MDS_MEA_EA,
 545                                            lmm, lmm_size);
 546                 break;
 547         case EA_SID:
 548                 rc = fsfilt_ext3_get_xattr(inode,
 549                                            XATTR_LUSTRE_MDS_SID_EA,
 550                                            lmm, lmm_size);
 551                 break;
 552         case EA_MID:
 553                 rc = fsfilt_ext3_get_xattr(inode,
 554                                            XATTR_LUSTRE_MDS_MID_EA,
 555                                            lmm, lmm_size);
 556                 break;
 557         default:
 558                 return -EINVAL;
 559         }
 560
 561         return rc;
 562 }
 563
 564 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, void *bio)
 565 {
 566         int rc = 0;
 567 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
 568         submit_bio(rw, (struct bio *)bio);
 569 #else
 570         struct bio *b = (struct kiobuf *)bio;
 571         int blocks_per_page;
 572
 573         rc = brw_kiovec(rw, 1, &b, inode->i_dev,
 574                         b->blocks, 1 << inode->i_blkbits);
 575
 576         blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 577
 578         if (rc != (1 << inode->i_blkbits) * b->nr_pages * blocks_per_page) {
 579                 CERROR("short write?  expected %d, wrote %d\n",
 580                        (1 << inode->i_blkbits) * b->nr_pages *
 581                        blocks_per_page, rc);
 582         }
 583 #endif
 584         return rc;
 585 }
 586
 587 static struct page *fsfilt_ext3_getpage(struct inode *inode, long int index)
 588 {
 589         int rc;
 590         struct page *page;
 591
 592         page = grab_cache_page(inode->i_mapping, index);
 593         if (page == NULL)
 594                 return ERR_PTR(-ENOMEM);
 595
 596         if (PageUptodate(page)) {
 597                 unlock_page(page);
 598                 return page;
 599         }
 600
 601         rc = inode->i_mapping->a_ops->readpage(NULL, page);
 602         if (rc < 0) {
 603                 page_cache_release(page);
 604                 return ERR_PTR(rc);
 605         }
 606
 607         return page;
 608 }
 609
 610 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 611                                     loff_t *off)
 612 {
 613         struct inode *inode = file->f_dentry->d_inode;
 614         int rc = 0;
 615
 616         if (S_ISREG(inode->i_mode))
 617                 rc = file->f_op->read(file, buf, count, off);
 618         else {
 619                 const int blkbits = inode->i_sb->s_blocksize_bits;
 620                 const int blksize = inode->i_sb->s_blocksize;
 621
 622                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 623                        count, inode->i_ino, *off);
 624                 while (count > 0) {
 625                         struct buffer_head *bh;
 626
 627                         bh = NULL;
 628                         if (*off < inode->i_size) {
 629                                 int err = 0;
 630
 631                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 632                                                 0, &err);
 633
 634                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 635
 636                                 if (bh) {
 637                                         memcpy(buf, bh->b_data, blksize);
 638                                         brelse(bh);
 639                                 } else if (err) {
 640                                         /* XXX in theory we should just fake
 641                                          * this buffer and continue like ext3,
 642                                          * especially if this is a partial read
 643                                          */
 644                                         CERROR("error read dir %lu+%llu: %d\n",
 645                                                inode->i_ino, *off, err);
 646                                         RETURN(err);
 647                                 }
 648                         }
 649                         if (!bh) {
 650                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 651
 652                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 653                                 memset(fake, 0, sizeof(*fake));
 654                                 fake->rec_len = cpu_to_le32(blksize);
 655                         }
 656                         count -= blksize;
 657                         buf += blksize;
 658                         *off += blksize;
 659                         rc += blksize;
 660                 }
 661         }
 662
 663         return rc;
 664 }
 665
 666 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 667 {
 668         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 669
 670         fcb->cb_func(fcb->cb_obd, fcb->cb_last_num, fcb->cb_data, error);
 671
 672         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 673         atomic_dec(&fcb_cache_count);
 674 }
 675
 676 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd,
 677                                       struct super_block *sb,
 678                                       __u64 last_num, void *handle,
 679                                       fsfilt_cb_t cb_func,
 680                                       void *cb_data)
 681 {
 682         struct fsfilt_cb_data *fcb;
 683
 684         OBD_SLAB_ALLOC(fcb, fcb_cache, GFP_NOFS, sizeof *fcb);
 685         if (fcb == NULL)
 686                 RETURN(-ENOMEM);
 687
 688         atomic_inc(&fcb_cache_count);
 689         fcb->cb_func = cb_func;
 690         fcb->cb_obd = obd;
 691         fcb->cb_last_num = last_num;
 692         fcb->cb_data = cb_data;
 693
 694         CDEBUG(D_EXT2, "set callback for last_num: "LPD64"\n", last_num);
 695         lock_kernel();
 696         journal_callback_set(handle, fsfilt_ext3_cb_func,
 697                              (struct journal_callback *)fcb);
 698         unlock_kernel();
 699         return 0;
 700 }
 701
 702 /*
 703  * We need to hack the return value for the free inode counts because
 704  * the current EA code requires one filesystem block per inode with EAs,
 705  * so it is possible to run out of blocks before we run out of inodes.
 706  *
 707  * This can be removed when the ext3 EA code is fixed.
 708  */
 709 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 710 {
 711         struct kstatfs sfs;
 712         int rc;
 713
 714         memset(&sfs, 0, sizeof(sfs));
 715
 716         rc = sb->s_op->statfs(sb, &sfs);
 717
 718         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 719                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 720                 sfs.f_ffree = sfs.f_bfree;
 721         }
 722
 723         statfs_pack(osfs, &sfs);
 724         return rc;
 725 }
 726
 727 static int fsfilt_ext3_sync(struct super_block *sb)
 728 {
 729         return ext3_force_commit(sb);
 730 }
 731
 732 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 733 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 734 #define ext3_up_truncate_sem(inode)  up_write(&EXT3_I(inode)->truncate_sem);
 735 #define ext3_down_truncate_sem(inode)  down_write(&EXT3_I(inode)->truncate_sem);
 736 #else
 737 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 738 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 739 #endif
 740
 741 #include <linux/lustre_version.h>
 742 #if EXT3_EXT_MAGIC == 0xf301
 743 #define ee_start e_start
 744 #define ee_block e_block
 745 #define ee_len   e_num
 746 #endif
 747 #ifndef EXT3_BB_MAX_BLOCKS
 748 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 749         ext3_new_blocks(handle, inode, count, goal, err)
 750 #endif
 751
 752 struct bpointers {
 753         unsigned long *blocks;
 754         int *created;
 755         unsigned long start;
 756         int num;
 757         int init_num;
 758         int create;
 759 };
 760 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 761                                 unsigned long block, int *aflags)
 762 {
 763         struct ext3_inode_info *ei = EXT3_I(inode);
 764         unsigned long bg_start;
 765         unsigned long colour;
 766         int depth;
 767
 768         if (path) {
 769                 struct ext3_extent *ex;
 770                 depth = path->p_depth;
 771
 772                 /* try to predict block placement */
 773                 if ((ex = path[depth].p_ext)) {
 774                         if (ex->ee_block + ex->ee_len == block)
 775                                 *aflags |= 1;
 776                         return ex->ee_start + (block - ex->ee_block);
 777                 }
 778
 779                 /* it looks index is empty
 780                  * try to find starting from index itself */
 781                 if (path[depth].p_bh)
 782                         return path[depth].p_bh->b_blocknr;
 783         }
 784
 785         /* OK. use inode's group */
 786         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 787                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 788         colour = (current->pid % 16) *
 789                         (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 790         return bg_start + colour + block;
 791 }
 792
 793 static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
 794                                   struct ext3_ext_path *path,
 795                                   struct ext3_extent *newex, int exist)
 796 {
 797         struct inode *inode = tree->inode;
 798         struct bpointers *bp = tree->private;
 799         int count, err, goal;
 800         unsigned long pblock;
 801         unsigned long tgen;
 802         loff_t new_i_size;
 803         handle_t *handle;
 804         int i, aflags = 0;
 805
 806         i = EXT_DEPTH(tree);
 807         EXT_ASSERT(i == path->p_depth);
 808         EXT_ASSERT(path[i].p_hdr);
 809
 810         if (exist) {
 811                 err = EXT_CONTINUE;
 812                 goto map;
 813         }
 814
 815         if (bp->create == 0) {
 816                 i = 0;
 817                 if (newex->ee_block < bp->start)
 818                         i = bp->start - newex->ee_block;
 819                 if (i >= newex->ee_len)
 820                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 821                                         i, newex->ee_len);
 822                 for (; i < newex->ee_len && bp->num; i++) {
 823                         *(bp->created) = 0;
 824                         bp->created++;
 825                         *(bp->blocks) = 0;
 826                         bp->blocks++;
 827                         bp->num--;
 828                         bp->start++;
 829                 }
 830
 831                 return EXT_CONTINUE;
 832         }
 833         tgen = EXT_GENERATION(tree);
 834         count = ext3_ext_calc_credits_for_insert(tree, path);
 835         ext3_up_truncate_sem(inode);
 836         lock_kernel();
 837         handle = journal_start(EXT3_JOURNAL(inode), count + EXT3_ALLOC_NEEDED + 1);
 838         unlock_kernel();
 839         if (IS_ERR(handle)) {
 840                 ext3_down_truncate_sem(inode);
 841                 return PTR_ERR(handle);
 842         }
 843
 844         if (tgen != EXT_GENERATION(tree)) {
 845                 /* the tree has changed. so path can be invalid at moment */
 846                 lock_kernel();
 847                 journal_stop(handle);
 848                 unlock_kernel();
 849                 ext3_down_truncate_sem(inode);
 850                 return EXT_REPEAT;
 851         }
 852         ext3_down_truncate_sem(inode);
 853         count = newex->ee_len;
 854         goal = ext3_ext_find_goal(inode, path, newex->ee_block, &aflags);
 855         aflags |= 2; /* block have been already reserved */
 856         pblock = ext3_mb_new_blocks(handle, inode, goal, &count, aflags, &err);
 857         if (!pblock)
 858                 goto out;
 859         EXT_ASSERT(count <= newex->ee_len);
 860
 861         /* insert new extent */
 862         newex->ee_start = pblock;
 863         newex->ee_len = count;
 864         err = ext3_ext_insert_extent(handle, tree, path, newex);
 865         if (err)
 866                 goto out;
 867
 868         /* correct on-disk inode size */
 869         if (newex->ee_len > 0) {
 870                 new_i_size = (loff_t) newex->ee_block + newex->ee_len;
 871                 new_i_size = new_i_size << inode->i_blkbits;
 872                 if (new_i_size > EXT3_I(inode)->i_disksize) {
 873                         EXT3_I(inode)->i_disksize = new_i_size;
 874                         err = ext3_mark_inode_dirty(handle, inode);
 875                 }
 876         }
 877 out:
 878         lock_24kernel();
 879         journal_stop(handle);
 880         unlock_24kernel();
 881 map:
 882         if (err >= 0) {
 883                 /* map blocks */
 884                 if (bp->num == 0) {
 885                         CERROR("hmm. why do we find this extent?\n");
 886                         CERROR("initial space: %lu:%u\n",
 887                                 bp->start, bp->init_num);
 888                         CERROR("current extent: %u/%u/%u %d\n",
 889                                 newex->ee_block, newex->ee_len,
 890                                 newex->ee_start, exist);
 891                 }
 892                 i = 0;
 893                 if (newex->ee_block < bp->start)
 894                         i = bp->start - newex->ee_block;
 895                 if (i >= newex->ee_len)
 896                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 897                                         i, newex->ee_len);
 898                 for (; i < newex->ee_len && bp->num; i++) {
 899                         *(bp->created) = (exist == 0 ? 1 : 0);
 900                         bp->created++;
 901                         *(bp->blocks) = newex->ee_start + i;
 902                         bp->blocks++;
 903                         bp->num--;
 904                         bp->start++;
 905                 }
 906         }
 907         return err;
 908 }
 909
 910 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
 911                        unsigned long num, unsigned long *blocks,
 912                        int *created, int create)
 913 {
 914         struct ext3_extents_tree tree;
 915         struct bpointers bp;
 916         int err;
 917
 918         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
 919                 block, block + num, (unsigned) inode->i_ino);
 920
 921         ext3_init_tree_desc(&tree, inode);
 922         tree.private = &bp;
 923         bp.blocks = blocks;
 924         bp.created = created;
 925         bp.start = block;
 926         bp.init_num = bp.num = num;
 927         bp.create = create;
 928
 929         ext3_down_truncate_sem(inode);
 930         err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
 931         ext3_ext_invalidate_cache(&tree);
 932         ext3_up_truncate_sem(inode);
 933         return err;
 934 }
 935
 936 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
 937                                     int pages, unsigned long *blocks,
 938                                     int *created, int create)
 939 {
 940         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 941         int rc = 0, i = 0;
 942         struct page *fp = NULL;
 943         int clen = 0;
 944
 945         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
 946                 inode->i_ino, pages, (*page)->index);
 947
 948         /* pages are sorted already. so, we just have to find
 949          * contig. space and process them properly */
 950         while (i < pages) {
 951                 if (fp == NULL) {
 952                         /* start new extent */
 953                         fp = *page++;
 954                         clen = 1;
 955                         i++;
 956                         continue;
 957                 } else if (fp->index + clen == (*page)->index) {
 958                         /* continue the extent */
 959                         page++;
 960                         clen++;
 961                         i++;
 962                         continue;
 963                 }
 964
 965                 /* process found extent */
 966                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 967                                         clen * blocks_per_page, blocks,
 968                                         created, create);
 969                 if (rc)
 970                         GOTO(cleanup, rc);
 971
 972                 /* look for next extent */
 973                 fp = NULL;
 974                 blocks += blocks_per_page * clen;
 975                 created += blocks_per_page * clen;
 976         }
 977
 978         if (fp)
 979                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 980                                         clen * blocks_per_page, blocks,
 981                                         created, create);
 982 cleanup:
 983         return rc;
 984 }
 985 #endif
 986
 987 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
 988                                unsigned long *blocks, int *created, int create);
 989 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
 990                                    int pages, unsigned long *blocks,
 991                                    int *created, int create)
 992 {
 993         int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
 994         unsigned long *b;
 995         int rc = 0, i, *cr;
 996
 997         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
 998                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
 999                 if (rc) {
1000                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1001                                inode->i_ino, *b, *cr, create, rc);
1002                         break;
1003                 }
1004
1005                 b += blocks_per_page;
1006                 cr += blocks_per_page;
1007         }
1008         return rc;
1009 }
1010
1011 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1012                                 int pages, unsigned long *blocks,
1013                                 int *created, int create,
1014                                 struct semaphore *optional_sem)
1015 {
1016         int rc;
1017 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1018         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1019                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1020                                                      blocks, created, create);
1021                 return rc;
1022         }
1023 #endif
1024         if (optional_sem != NULL)
1025                 down(optional_sem);
1026         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1027                                             created, create);
1028         if (optional_sem != NULL)
1029                 up(optional_sem);
1030
1031         return rc;
1032 }
1033
1034 extern int ext3_prep_san_write(struct inode *inode, long *blocks,
1035                                int nblocks, loff_t newsize);
1036 static int fsfilt_ext3_prep_san_write(struct inode *inode, long *blocks,
1037                                       int nblocks, loff_t newsize)
1038 {
1039         return ext3_prep_san_write(inode, blocks, nblocks, newsize);
1040 }
1041
1042 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1043                                    int size, loff_t *offs)
1044 {
1045         struct inode *inode = file->f_dentry->d_inode;
1046         unsigned long block;
1047         struct buffer_head *bh;
1048         int err, blocksize, csize, boffs;
1049
1050         /* prevent reading after eof */
1051         lock_kernel();
1052         if (inode->i_size < *offs + size) {
1053                 size = inode->i_size - *offs;
1054                 unlock_kernel();
1055                 if (size < 0) {
1056                         CERROR("size %llu is too short for read %u@%llu\n",
1057                                inode->i_size, size, *offs);
1058                         return -EIO;
1059                 } else if (size == 0) {
1060                         return 0;
1061                 }
1062         } else {
1063                 unlock_kernel();
1064         }
1065
1066         blocksize = 1 << inode->i_blkbits;
1067
1068         while (size > 0) {
1069                 block = *offs >> inode->i_blkbits;
1070                 boffs = *offs & (blocksize - 1);
1071                 csize = min(blocksize - boffs, size);
1072                 bh = ext3_bread(NULL, inode, block, 0, &err);
1073                 if (!bh) {
1074                         CERROR("can't read block: %d\n", err);
1075                         return err;
1076                 }
1077
1078                 memcpy(buf, bh->b_data + boffs, csize);
1079                 brelse(bh);
1080
1081                 *offs += csize;
1082                 buf += csize;
1083                 size -= csize;
1084         }
1085         return 0;
1086 }
1087
1088 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1089                                     loff_t *offs, int force_sync)
1090 {
1091         struct buffer_head *bh = NULL;
1092         unsigned long block;
1093         struct inode *inode = file->f_dentry->d_inode;
1094         loff_t old_size = inode->i_size, offset = *offs;
1095         loff_t new_size = inode->i_size;
1096         journal_t *journal;
1097         handle_t *handle;
1098         int err = 0, block_count = 0, blocksize, size, boffs;
1099
1100         /* Determine how many transaction credits are needed */
1101         blocksize = 1 << inode->i_blkbits;
1102         block_count = (*offs & (blocksize - 1)) + bufsize;
1103         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1104
1105         journal = EXT3_SB(inode->i_sb)->s_journal;
1106         lock_24kernel();
1107         handle = journal_start(journal,
1108                                block_count * EXT3_DATA_TRANS_BLOCKS + 2);
1109         unlock_24kernel();
1110         if (IS_ERR(handle)) {
1111                 CERROR("can't start transaction\n");
1112                 return PTR_ERR(handle);
1113         }
1114
1115         while (bufsize > 0) {
1116                 if (bh != NULL)
1117                         brelse(bh);
1118
1119                 block = offset >> inode->i_blkbits;
1120                 boffs = offset & (blocksize - 1);
1121                 size = min(blocksize - boffs, bufsize);
1122                 bh = ext3_bread(handle, inode, block, 1, &err);
1123                 if (!bh) {
1124                         CERROR("can't read/create block: %d\n", err);
1125                         goto out;
1126                 }
1127
1128                 err = ext3_journal_get_write_access(handle, bh);
1129                 if (err) {
1130                         CERROR("journal_get_write_access() returned error %d\n",
1131                                err);
1132                         goto out;
1133                 }
1134                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1135                 memcpy(bh->b_data + boffs, buf, size);
1136                 err = ext3_journal_dirty_metadata(handle, bh);
1137                 if (err) {
1138                         CERROR("journal_dirty_metadata() returned error %d\n",
1139                                err);
1140                         goto out;
1141                 }
1142                 if (offset + size > new_size)
1143                         new_size = offset + size;
1144                 offset += size;
1145                 bufsize -= size;
1146                 buf += size;
1147         }
1148
1149         if (force_sync)
1150                 handle->h_sync = 1; /* recovery likes this */
1151 out:
1152         if (bh)
1153                 brelse(bh);
1154
1155         /* correct in-core and on-disk sizes */
1156         if (new_size > inode->i_size) {
1157                 lock_kernel();
1158                 if (new_size > inode->i_size)
1159                         inode->i_size = new_size;
1160                 if (inode->i_size > EXT3_I(inode)->i_disksize)
1161                         EXT3_I(inode)->i_disksize = inode->i_size;
1162                 if (inode->i_size > old_size)
1163                         mark_inode_dirty(inode);
1164                 unlock_kernel();
1165         }
1166
1167         lock_24kernel();
1168         journal_stop(handle);
1169         unlock_24kernel();
1170
1171         if (err == 0)
1172                 *offs = offset;
1173         return err;
1174 }
1175
1176 static int fsfilt_ext3_setup(struct obd_device *obd, struct super_block *sb)
1177 {
1178 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1179         struct mds_obd *mds = &obd->u.mds;
1180 #endif
1181 #if 0
1182         EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1183         EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1184 #endif
1185 #ifdef S_PDIROPS
1186         CWARN("Enabling PDIROPS\n");
1187         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1188         sb->s_flags |= S_PDIROPS;
1189 #endif
1190         /* setup mdsnum in underlying fs */
1191 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1192         if (mds->mds_md_obd) {
1193                 struct ext3_sb_info *sbi = EXT3_SB(sb);
1194                 struct ext3_super_block *es = sbi->s_es;
1195                 handle_t *handle;
1196                 int err;
1197
1198                 if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_MDSNUM)) {
1199                         CWARN("%s: set mdsnum %d in ext3\n",
1200                               obd->obd_name, mds->mds_num);
1201                         lock_kernel();
1202                         handle = journal_start(sbi->s_journal, 1);
1203                         unlock_kernel();
1204                         LASSERT(!IS_ERR(handle));
1205                         err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1206                         LASSERT(err == 0);
1207                         EXT3_SET_INCOMPAT_FEATURE(sb,
1208                                                 EXT3_FEATURE_INCOMPAT_MDSNUM);
1209                         es->s_mdsnum = mds->mds_num;
1210                         err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1211                         LASSERT(err == 0);
1212                         lock_kernel();
1213                         journal_stop(handle);
1214                         unlock_kernel();
1215                 } else {
1216                         CWARN("%s: mdsnum initialized to %u in ext3fs\n",
1217                                 obd->obd_name, es->s_mdsnum);
1218                 }
1219                 sbi->s_mdsnum = es->s_mdsnum;
1220         }
1221 #endif
1222         return 0;
1223 }
1224
1225 extern int ext3_add_dir_entry(struct dentry *dentry);
1226 extern int ext3_del_dir_entry(struct dentry *dentry);
1227
1228 static int fsfilt_ext3_add_dir_entry(struct obd_device *obd,
1229                                      struct dentry *parent,
1230                                      char *name, int namelen,
1231                                      unsigned long ino,
1232                                      unsigned long generation,
1233                                      unsigned long mds,
1234                                      unsigned long fid)
1235 {
1236 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1237         struct dentry *dentry;
1238         int err;
1239         LASSERT(ino != 0);
1240         LASSERT(namelen != 0);
1241         dentry = ll_lookup_one_len(name, parent, namelen);
1242         if (IS_ERR(dentry)) {
1243                 CERROR("can't lookup %*s in %lu/%lu: %d\n", dentry->d_name.len,
1244                        dentry->d_name.name, dentry->d_inode->i_ino,
1245                        (unsigned long) dentry->d_inode->i_generation,
1246                        (int) PTR_ERR(dentry));
1247                 RETURN(PTR_ERR(dentry));
1248         }
1249         if (dentry->d_inode != NULL || dentry->d_flags & DCACHE_CROSS_REF) {
1250                 CERROR("dentry %*s(0x%p) found\n", dentry->d_name.len,
1251                        dentry->d_name.name, dentry);
1252                 l_dput(dentry);
1253                 RETURN(-EEXIST);
1254         }
1255
1256         /* mds_reint_rename() may use this method to add dir entry
1257          * that points onto local inode. and we don't want to find
1258          * it cross-ref by subsequent lookups */
1259         d_drop(dentry);
1260
1261         dentry->d_flags |= DCACHE_CROSS_REF;
1262         dentry->d_inum = ino;
1263         dentry->d_mdsnum = mds;
1264         dentry->d_generation = generation;
1265         dentry->d_fid = fid;
1266         lock_kernel();
1267         err = ext3_add_dir_entry(dentry);
1268         unlock_kernel();
1269
1270         l_dput(dentry);
1271
1272         RETURN(err);
1273 #else
1274 #error "rebuild kernel and lustre with ext3-mds-num patch!"
1275         LASSERT(0);
1276 #endif
1277 }
1278
1279 static int fsfilt_ext3_del_dir_entry(struct obd_device *obd,
1280                                  struct dentry *dentry)
1281 {
1282 #ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
1283         int err;
1284         lock_kernel();
1285         err = ext3_del_dir_entry(dentry);
1286         unlock_kernel();
1287         if (err == 0)
1288                 d_drop(dentry);
1289         return err;
1290 #else
1291 #error "rebuild kernel and lustre with ext3-mds-num patch!"
1292         LASSERT(0);
1293 #endif
1294 }
1295
1296 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1297    objects. Logs is number of logfiles to update */
1298 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1299 {
1300         if ( !fso ) {
1301                 switch(op) {
1302                 case FSFILT_OP_CREATE:
1303                                  /* directory leaf, index & indirect & EA*/
1304                         return 4 + 3 * logs;
1305                 case FSFILT_OP_UNLINK:
1306                         return 3 * logs;
1307                 }
1308         } else {
1309                 int i;
1310                 int needed = 0;
1311                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1312                 int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
1313                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1314                 for (i = 0; i < op; i++, fso++) {
1315                         int nblocks = fso->fso_bufcnt * blockpp;
1316                         int ndindirect = min(nblocks, addrpp + 1);
1317                         int nindir = nblocks + ndindirect + 1;
1318
1319                         needed += nindir;
1320                 }
1321                 return needed + 3 * logs;
1322         }
1323
1324         return 0;
1325 }
1326
1327
1328 #define EXTENTS_EA "write_extents"
1329 #define EXTENTS_EA_SIZE 64
1330
1331 int ext3_ext_in_ea_alloc_space(struct inode *, int, const char *, unsigned long, unsigned long);
1332 int ext3_ext_in_ea_remove_space(struct inode *, int, const char *, unsigned long, unsigned long);
1333 int ext3_ext_in_ea_get_extents(struct inode *, int, const char *, char **, int *);
1334 int ext3_ext_in_ea_get_extents_num(struct inode *, int, const char *, int *);
1335
1336 static int fsfilt_ext3_insert_extents_ea(struct inode *inode,
1337                                       unsigned long from,
1338                                       unsigned long num)
1339 {
1340         int rc = 0;
1341
1342         rc = ext3_ext_in_ea_alloc_space(inode, EXT3_XATTR_INDEX_TRUSTED,
1343                                         EXTENTS_EA, from, num);
1344         return rc;
1345 }
1346
1347 static int fsfilt_ext3_remove_extents_ea(struct inode *inode,
1348                                          unsigned long from,
1349                                          unsigned long num)
1350 {
1351         int rc = 0;
1352
1353         rc = ext3_ext_in_ea_remove_space(inode, EXT3_XATTR_INDEX_TRUSTED,
1354                                          EXTENTS_EA, from, num);
1355         return rc;
1356 }
1357
1358 extern int ext3_init_tree_in_ea(struct inode *inode, int name_index,
1359                                 const char *eaname, int size);
1360
1361 static int fsfilt_ext3_init_extents_ea(struct inode *inode)
1362 {
1363         int rc = 0;
1364
1365         rc = ext3_init_tree_in_ea(inode, EXT3_XATTR_INDEX_TRUSTED,
1366                                   EXTENTS_EA, 64);
1367         return rc;
1368 }
1369
1370 static int fsfilt_ext3_get_inode_write_extents(struct inode *inode,
1371                                          char **pbuf, int *size)
1372 {
1373         int rc = 0;
1374
1375         rc = ext3_ext_in_ea_get_extents(inode, EXT3_XATTR_INDEX_TRUSTED,
1376                                         EXTENTS_EA,  pbuf, size);
1377         return rc;
1378 }
1379
1380 static int fsfilt_ext3_get_write_extents_num(struct inode *inode, int *size)
1381 {
1382         int rc = 0;
1383
1384         rc = ext3_ext_in_ea_get_extents_num(inode, EXT3_XATTR_INDEX_TRUSTED,
1385                                             EXTENTS_EA, size);
1386         return rc;
1387 }
1388
1389 static struct fsfilt_operations fsfilt_ext3_ops = {
1390         .fs_type                    = "ext3",
1391         .fs_owner                   = THIS_MODULE,
1392         .fs_start                   = fsfilt_ext3_start,
1393         .fs_brw_start               = fsfilt_ext3_brw_start,
1394         .fs_commit                  = fsfilt_ext3_commit,
1395         .fs_commit_async            = fsfilt_ext3_commit_async,
1396         .fs_commit_wait             = fsfilt_ext3_commit_wait,
1397         .fs_setattr                 = fsfilt_ext3_setattr,
1398         .fs_iocontrol               = fsfilt_ext3_iocontrol,
1399         .fs_set_md                  = fsfilt_ext3_set_md,
1400         .fs_get_md                  = fsfilt_ext3_get_md,
1401         .fs_readpage                = fsfilt_ext3_readpage,
1402         .fs_add_journal_cb          = fsfilt_ext3_add_journal_cb,
1403         .fs_statfs                  = fsfilt_ext3_statfs,
1404         .fs_sync                    = fsfilt_ext3_sync,
1405         .fs_map_inode_pages         = fsfilt_ext3_map_inode_pages,
1406         .fs_prep_san_write          = fsfilt_ext3_prep_san_write,
1407         .fs_write_record            = fsfilt_ext3_write_record,
1408         .fs_read_record             = fsfilt_ext3_read_record,
1409         .fs_setup                   = fsfilt_ext3_setup,
1410         .fs_getpage                 = fsfilt_ext3_getpage,
1411         .fs_send_bio                = fsfilt_ext3_send_bio,
1412         .fs_set_xattr               = fsfilt_ext3_set_xattr,
1413         .fs_get_xattr               = fsfilt_ext3_get_xattr,
1414         .fs_get_op_len              = fsfilt_ext3_get_op_len,
1415         .fs_add_dir_entry           = fsfilt_ext3_add_dir_entry,
1416         .fs_del_dir_entry           = fsfilt_ext3_del_dir_entry,
1417         .fs_init_extents_ea         = fsfilt_ext3_init_extents_ea,
1418         .fs_insert_extents_ea       = fsfilt_ext3_insert_extents_ea,
1419         .fs_remove_extents_ea       = fsfilt_ext3_remove_extents_ea,
1420         .fs_get_inode_write_extents = fsfilt_ext3_get_inode_write_extents,
1421         .fs_get_write_extents_num   = fsfilt_ext3_get_write_extents_num,
1422 };
1423
1424 static int __init fsfilt_ext3_init(void)
1425 {
1426         int rc;
1427
1428         fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
1429                                       sizeof(struct fsfilt_cb_data), 0,
1430                                       0, NULL, NULL);
1431         if (!fcb_cache) {
1432                 CERROR("error allocating fsfilt journal callback cache\n");
1433                 GOTO(out, rc = -ENOMEM);
1434         }
1435
1436         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
1437
1438         if (rc)
1439                 kmem_cache_destroy(fcb_cache);
1440 out:
1441         return rc;
1442 }
1443
1444 static void __exit fsfilt_ext3_exit(void)
1445 {
1446         fsfilt_unregister_ops(&fsfilt_ext3_ops);
1447         LASSERTF(kmem_cache_destroy(fcb_cache) == 0,
1448                  "can't free fsfilt callback cache: count %d\n",
1449                  atomic_read(&fcb_cache_count));
1450 }
1451
1452 module_init(fsfilt_ext3_init);
1453 module_exit(fsfilt_ext3_exit);
1454
1455 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1456 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
1457 MODULE_LICENSE("GPL");