drivers/staging/lustre/lustre/lvfs/fsfilt_ext3.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/fsfilt_ext3.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/init.h>
  44 #include <linux/module.h>
  45 #include <linux/fs.h>
  46 #include <linux/slab.h>
  47 #include <linux/pagemap.h>
  48 #include <ldiskfs/ldiskfs_config.h>
  49 #include <ext4/ext4.h>
  50 #include <ext4/ext4_jbd2.h>
  51 #include <linux/bitops.h>
  52 #include <linux/quota.h>
  53
  54 #include <linux/libcfs/libcfs.h>
  55 #include <lustre_fsfilt.h>
  56 #include <obd.h>
  57 #include <linux/lustre_compat25.h>
  58 #include <linux/lprocfs_status.h>
  59
  60 #include <ext4/ext4_extents.h>
  61
  62 #ifdef HAVE_EXT_PBLOCK /* Name changed to ext4_ext_pblock for kernel 2.6.35 */
  63 #define ext3_ext_pblock(ex) ext_pblock((ex))
  64 #endif
  65
  66 /* for kernels 2.6.18 and later */
  67 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  68
  69 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
  70                ext3_ext_insert_extent(handle, inode, path, newext, flag)
  71
  72 #define ext3_mb_discard_inode_preallocations(inode) \
  73                  ext3_discard_preallocations(inode)
  74
  75 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
  76 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
  77
  78 static struct kmem_cache *fcb_cache;
  79
  80 struct fsfilt_cb_data {
  81         struct ext4_journal_cb_entry cb_jcb; /* private data - MUST BE FIRST */
  82         fsfilt_cb_t cb_func;        /* MDS/OBD completion function */
  83         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  84         __u64 cb_last_rcvd;          /* MDS/OST last committed operation */
  85         void *cb_data;            /* MDS/OST completion function data */
  86 };
  87
  88 static char *fsfilt_ext3_get_label(struct super_block *sb)
  89 {
  90         return EXT3_SB(sb)->s_es->s_volume_name;
  91 }
  92
  93 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
  94 # include <ext4/truncate.h>
  95
  96 /*
  97  * We don't currently need any additional blocks for rmdir and
  98  * unlink transactions because we are storing the OST oa_id inside
  99  * the inode (which we will be changing anyways as part of this
 100  * transaction).
 101  */
 102 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 103                                int logs)
 104 {
 105         /* For updates to the last received file */
 106         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 107         journal_t *journal;
 108         void *handle;
 109
 110         if (current->journal_info) {
 111                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 112                        current->journal_info);
 113                 goto journal_start;
 114         }
 115
 116         switch(op) {
 117         case FSFILT_OP_UNLINK:
 118                 /* delete one file + create/update logs for each stripe */
 119                 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
 120                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 121                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 122                 break;
 123         case FSFILT_OP_CANCEL_UNLINK:
 124                 LASSERT(logs == 1);
 125
 126                 /* blocks for log header bitmap update OR
 127                  * blocks for catalog header bitmap update + unlink of logs +
 128                  * blocks for delete the inode (include blocks truncating). */
 129                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 130                           EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
 131                           ext4_blocks_for_truncate(inode) + 3;
 132                 break;
 133         default: CERROR("unknown transaction start op %d\n", op);
 134                 LBUG();
 135         }
 136
 137         LASSERT(current->journal_info == desc_private);
 138         journal = EXT3_SB(inode->i_sb)->s_journal;
 139         if (nblocks > journal->j_max_transaction_buffers) {
 140                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 141                        nblocks, op, logs, journal->j_max_transaction_buffers);
 142                 nblocks = journal->j_max_transaction_buffers;
 143         }
 144
 145  journal_start:
 146         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 147         handle = ext3_journal_start(inode, nblocks);
 148
 149         if (!IS_ERR(handle))
 150                 LASSERT(current->journal_info == handle);
 151         else
 152                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 153                        op, nblocks, PTR_ERR(handle));
 154         return handle;
 155 }
 156
 157 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 158 {
 159         int rc;
 160         handle_t *handle = h;
 161
 162         LASSERT(current->journal_info == handle);
 163         if (force_sync)
 164                 handle->h_sync = 1; /* recovery likes this */
 165
 166         rc = ext3_journal_stop(handle);
 167
 168         return rc;
 169 }
 170
 171 #ifndef EXT3_EXTENTS_FL
 172 #define EXT3_EXTENTS_FL          0x00080000 /* Inode uses extents */
 173 #endif
 174
 175 #ifndef EXT_ASSERT
 176 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 177 #endif
 178
 179 #define EXT_GENERATION(inode)      (EXT4_I(inode)->i_ext_generation)
 180 #define ext3_ext_base              inode
 181 #define ext3_ext_base2inode(inode)      (inode)
 182 #define EXT_DEPTH(inode)                ext_depth(inode)
 183 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 184                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 185
 186 struct bpointers {
 187         unsigned long *blocks;
 188         unsigned long start;
 189         int num;
 190         int init_num;
 191         int create;
 192 };
 193
 194 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 195                                unsigned long block, int *aflags)
 196 {
 197         struct ext3_inode_info *ei = EXT3_I(inode);
 198         unsigned long bg_start;
 199         unsigned long colour;
 200         int depth;
 201
 202         if (path) {
 203                 struct ext3_extent *ex;
 204                 depth = path->p_depth;
 205
 206                 /* try to predict block placement */
 207                 if ((ex = path[depth].p_ext))
 208                         return ext4_ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 209
 210                 /* it looks index is empty
 211                  * try to find starting from index itself */
 212                 if (path[depth].p_bh)
 213                         return path[depth].p_bh->b_blocknr;
 214         }
 215
 216         /* OK. use inode's group */
 217         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 218                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 219         colour = (current->pid % 16) *
 220                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 221         return bg_start + colour + block;
 222 }
 223
 224 #define ll_unmap_underlying_metadata(sb, blocknr) \
 225         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 226
 227 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 228 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 229                                 struct ext3_ext_path *path, unsigned long block,
 230                                 unsigned long *count, int *err)
 231 {
 232         unsigned long pblock, goal;
 233         int aflags = 0;
 234         struct inode *inode = ext3_ext_base2inode(base);
 235
 236         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 237         aflags |= 2; /* block have been already reserved */
 238         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 239         return pblock;
 240
 241 }
 242 #else
 243 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 244                                 struct ext3_ext_path *path, unsigned long block,
 245                                 unsigned long *count, int *err)
 246 {
 247         struct inode *inode = ext3_ext_base2inode(base);
 248         struct ext3_allocation_request ar;
 249         unsigned long pblock;
 250         int aflags;
 251
 252         /* find neighbour allocated blocks */
 253         ar.lleft = block;
 254         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 255         if (*err)
 256                 return 0;
 257         ar.lright = block;
 258         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 259         if (*err)
 260                 return 0;
 261
 262         /* allocate new block */
 263         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 264         ar.inode = inode;
 265         ar.logical = block;
 266         ar.len = *count;
 267         ar.flags = EXT3_MB_HINT_DATA;
 268         pblock = ext3_mb_new_blocks(handle, &ar, err);
 269         *count = ar.len;
 270         return pblock;
 271 }
 272 #endif
 273
 274 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 275                                   struct ext3_ext_path *path,
 276                                   struct ext3_ext_cache *cex,
 277 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 278                                    struct ext3_extent *ex,
 279 #endif
 280                                   void *cbdata)
 281 {
 282         struct bpointers *bp = cbdata;
 283         struct inode *inode = ext3_ext_base2inode(base);
 284         struct ext3_extent nex;
 285         unsigned long pblock;
 286         unsigned long tgen;
 287         int err, i;
 288         unsigned long count;
 289         handle_t *handle;
 290
 291 #ifdef EXT3_EXT_CACHE_EXTENT
 292         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT)
 293 #else
 294         if ((cex->ec_len != 0) && (cex->ec_start != 0))
 295 #endif
 296                                                    {
 297                 err = EXT_CONTINUE;
 298                 goto map;
 299         }
 300
 301         if (bp->create == 0) {
 302                 i = 0;
 303                 if (cex->ec_block < bp->start)
 304                         i = bp->start - cex->ec_block;
 305                 if (i >= cex->ec_len)
 306                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 307                                         i, cex->ec_len);
 308                 for (; i < cex->ec_len && bp->num; i++) {
 309                         *(bp->blocks) = 0;
 310                         bp->blocks++;
 311                         bp->num--;
 312                         bp->start++;
 313                 }
 314
 315                 return EXT_CONTINUE;
 316         }
 317
 318         tgen = EXT_GENERATION(base);
 319         count = ext3_ext_calc_credits_for_insert(base, path);
 320
 321         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 322         if (IS_ERR(handle)) {
 323                 return PTR_ERR(handle);
 324         }
 325
 326         if (tgen != EXT_GENERATION(base)) {
 327                 /* the tree has changed. so path can be invalid at moment */
 328                 ext3_journal_stop(handle);
 329                 return EXT_REPEAT;
 330         }
 331
 332         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
 333          * protected by i_data_sem as whole. so we patch it to store
 334          * generation to path and now verify the tree hasn't changed */
 335         down_write((&EXT4_I(inode)->i_data_sem));
 336
 337         /* validate extent, make sure the extent tree does not changed */
 338         if (EXT_GENERATION(base) != path[0].p_generation) {
 339                 /* cex is invalid, try again */
 340                 up_write(&EXT4_I(inode)->i_data_sem);
 341                 ext3_journal_stop(handle);
 342                 return EXT_REPEAT;
 343         }
 344
 345         count = cex->ec_len;
 346         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 347         if (!pblock)
 348                 goto out;
 349         EXT_ASSERT(count <= cex->ec_len);
 350
 351         /* insert new extent */
 352         nex.ee_block = cpu_to_le32(cex->ec_block);
 353         ext3_ext_store_pblock(&nex, pblock);
 354         nex.ee_len = cpu_to_le16(count);
 355         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
 356         if (err) {
 357                 /* free data blocks we just allocated */
 358                 /* not a good idea to call discard here directly,
 359                  * but otherwise we'd need to call it every free() */
 360 #ifdef EXT3_MB_HINT_GROUP_ALLOC
 361                 ext3_mb_discard_inode_preallocations(inode);
 362 #endif
 363 #ifdef HAVE_EXT_FREE_BLOCK_WITH_BUFFER_HEAD /* Introduced in 2.6.32-rc7 */
 364                 ext3_free_blocks(handle, inode, NULL, ext4_ext_pblock(&nex),
 365                                  cpu_to_le16(nex.ee_len), 0);
 366 #else
 367                 ext3_free_blocks(handle, inode, ext4_ext_pblock(&nex),
 368                                  cpu_to_le16(nex.ee_len), 0);
 369 #endif
 370                 goto out;
 371         }
 372
 373         /*
 374          * Putting len of the actual extent we just inserted,
 375          * we are asking ext3_ext_walk_space() to continue
 376          * scaning after that block
 377          */
 378         cex->ec_len = le16_to_cpu(nex.ee_len);
 379         cex->ec_start = ext4_ext_pblock(&nex);
 380         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
 381         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
 382
 383 out:
 384         up_write((&EXT4_I(inode)->i_data_sem));
 385         ext3_journal_stop(handle);
 386 map:
 387         if (err >= 0) {
 388                 /* map blocks */
 389                 if (bp->num == 0) {
 390                         CERROR("hmm. why do we find this extent?\n");
 391                         CERROR("initial space: %lu:%u\n",
 392                                 bp->start, bp->init_num);
 393 #ifdef EXT3_EXT_CACHE_EXTENT
 394                         CERROR("current extent: %u/%u/%llu %d\n",
 395                                 cex->ec_block, cex->ec_len,
 396                                 (unsigned long long)cex->ec_start,
 397                                 cex->ec_type);
 398 #else
 399                         CERROR("current extent: %u/%u/%llu\n",
 400                                 cex->ec_block, cex->ec_len,
 401                                 (unsigned long long)cex->ec_start);
 402 #endif
 403                 }
 404                 i = 0;
 405                 if (cex->ec_block < bp->start)
 406                         i = bp->start - cex->ec_block;
 407                 if (i >= cex->ec_len)
 408                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 409                                         i, cex->ec_len);
 410                 for (; i < cex->ec_len && bp->num; i++) {
 411                         *(bp->blocks) = cex->ec_start + i;
 412 #ifdef EXT3_EXT_CACHE_EXTENT
 413                         if (cex->ec_type != EXT3_EXT_CACHE_EXTENT)
 414 #else
 415                         if ((cex->ec_len == 0) || (cex->ec_start == 0))
 416 #endif
 417                                                                         {
 418                                 /* unmap any possible underlying metadata from
 419                                  * the block device mapping.  bug 6998. */
 420                                 ll_unmap_underlying_metadata(inode->i_sb,
 421                                                              *(bp->blocks));
 422                         }
 423                         bp->blocks++;
 424                         bp->num--;
 425                         bp->start++;
 426                 }
 427         }
 428         return err;
 429 }
 430
 431 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
 432                        unsigned long num, unsigned long *blocks,
 433                        int create)
 434 {
 435         struct ext3_ext_base *base = inode;
 436         struct bpointers bp;
 437         int err;
 438
 439         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
 440                block, block + num - 1, (unsigned) inode->i_ino);
 441
 442         bp.blocks = blocks;
 443         bp.start = block;
 444         bp.init_num = bp.num = num;
 445         bp.create = create;
 446
 447         err = fsfilt_ext3_ext_walk_space(base, block, num,
 448                                          ext3_ext_new_extent_cb, &bp);
 449         ext3_ext_invalidate_cache(base);
 450
 451         return err;
 452 }
 453
 454 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
 455                                     int pages, unsigned long *blocks,
 456                                     int create)
 457 {
 458         int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
 459         int rc = 0, i = 0;
 460         struct page *fp = NULL;
 461         int clen = 0;
 462
 463         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
 464                 inode->i_ino, pages, (*page)->index);
 465
 466         /* pages are sorted already. so, we just have to find
 467          * contig. space and process them properly */
 468         while (i < pages) {
 469                 if (fp == NULL) {
 470                         /* start new extent */
 471                         fp = *page++;
 472                         clen = 1;
 473                         i++;
 474                         continue;
 475                 } else if (fp->index + clen == (*page)->index) {
 476                         /* continue the extent */
 477                         page++;
 478                         clen++;
 479                         i++;
 480                         continue;
 481                 }
 482
 483                 /* process found extent */
 484                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 485                                         clen * blocks_per_page, blocks,
 486                                         create);
 487                 if (rc)
 488                         GOTO(cleanup, rc);
 489
 490                 /* look for next extent */
 491                 fp = NULL;
 492                 blocks += blocks_per_page * clen;
 493         }
 494
 495         if (fp)
 496                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 497                                         clen * blocks_per_page, blocks,
 498                                         create);
 499 cleanup:
 500         return rc;
 501 }
 502
 503 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
 504                                    int pages, unsigned long *blocks,
 505                                    int create)
 506 {
 507         int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
 508         unsigned long *b;
 509         int rc = 0, i;
 510
 511         for (i = 0, b = blocks; i < pages; i++, page++) {
 512                 rc = ext3_map_inode_page(inode, *page, b, create);
 513                 if (rc) {
 514                         CERROR("ino %lu, blk %lu create %d: rc %d\n",
 515                                inode->i_ino, *b, create, rc);
 516                         break;
 517                 }
 518
 519                 b += blocks_per_page;
 520         }
 521         return rc;
 522 }
 523
 524 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
 525                                 int pages, unsigned long *blocks,
 526                                 int create, struct mutex *optional_mutex)
 527 {
 528         int rc;
 529
 530         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
 531                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
 532                                                      blocks, create);
 533                 return rc;
 534         }
 535         if (optional_mutex != NULL)
 536                 mutex_lock(optional_mutex);
 537         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks, create);
 538         if (optional_mutex != NULL)
 539                 mutex_unlock(optional_mutex);
 540
 541         return rc;
 542 }
 543
 544 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
 545 {
 546         unsigned long block;
 547         struct buffer_head *bh;
 548         int err, blocksize, csize, boffs, osize = size;
 549
 550         /* prevent reading after eof */
 551         spin_lock(&inode->i_lock);
 552         if (i_size_read(inode) < *offs + size) {
 553                 size = i_size_read(inode) - *offs;
 554                 spin_unlock(&inode->i_lock);
 555                 if (size < 0) {
 556                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
 557                                i_size_read(inode), *offs);
 558                         return -EBADR;
 559                 } else if (size == 0) {
 560                         return 0;
 561                 }
 562         } else {
 563                 spin_unlock(&inode->i_lock);
 564         }
 565
 566         blocksize = 1 << inode->i_blkbits;
 567
 568         while (size > 0) {
 569                 block = *offs >> inode->i_blkbits;
 570                 boffs = *offs & (blocksize - 1);
 571                 csize = min(blocksize - boffs, size);
 572                 bh = ext3_bread(NULL, inode, block, 0, &err);
 573                 if (!bh) {
 574                         CERROR("can't read block: %d\n", err);
 575                         return err;
 576                 }
 577
 578                 memcpy(buf, bh->b_data + boffs, csize);
 579                 brelse(bh);
 580
 581                 *offs += csize;
 582                 buf += csize;
 583                 size -= csize;
 584         }
 585         return osize;
 586 }
 587 EXPORT_SYMBOL(fsfilt_ext3_read);
 588
 589 static int fsfilt_ext3_read_record(struct file * file, void *buf,
 590                                    int size, loff_t *offs)
 591 {
 592         int rc;
 593         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
 594         if (rc > 0)
 595                 rc = 0;
 596         return rc;
 597 }
 598
 599 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
 600                                 loff_t *offs, handle_t *handle)
 601 {
 602         struct buffer_head *bh = NULL;
 603         loff_t old_size = i_size_read(inode), offset = *offs;
 604         loff_t new_size = i_size_read(inode);
 605         unsigned long block;
 606         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
 607
 608         while (bufsize > 0) {
 609                 if (bh != NULL)
 610                         brelse(bh);
 611
 612                 block = offset >> inode->i_blkbits;
 613                 boffs = offset & (blocksize - 1);
 614                 size = min(blocksize - boffs, bufsize);
 615                 bh = ext3_bread(handle, inode, block, 1, &err);
 616                 if (!bh) {
 617                         CERROR("can't read/create block: %d\n", err);
 618                         break;
 619                 }
 620
 621                 err = ext3_journal_get_write_access(handle, bh);
 622                 if (err) {
 623                         CERROR("journal_get_write_access() returned error %d\n",
 624                                err);
 625                         break;
 626                 }
 627                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
 628                 memcpy(bh->b_data + boffs, buf, size);
 629                 err = ext3_journal_dirty_metadata(handle, bh);
 630                 if (err) {
 631                         CERROR("journal_dirty_metadata() returned error %d\n",
 632                                err);
 633                         break;
 634                 }
 635                 if (offset + size > new_size)
 636                         new_size = offset + size;
 637                 offset += size;
 638                 bufsize -= size;
 639                 buf += size;
 640         }
 641         if (bh)
 642                 brelse(bh);
 643
 644         /* correct in-core and on-disk sizes */
 645         if (new_size > i_size_read(inode)) {
 646                 spin_lock(&inode->i_lock);
 647                 if (new_size > i_size_read(inode))
 648                         i_size_write(inode, new_size);
 649                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
 650                         EXT3_I(inode)->i_disksize = i_size_read(inode);
 651                 if (i_size_read(inode) > old_size) {
 652                         spin_unlock(&inode->i_lock);
 653                         mark_inode_dirty(inode);
 654                 } else {
 655                         spin_unlock(&inode->i_lock);
 656                 }
 657         }
 658
 659         if (err == 0)
 660                 *offs = offset;
 661         return err;
 662 }
 663 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
 664
 665 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
 666                                     loff_t *offs, int force_sync)
 667 {
 668         struct inode *inode = file->f_dentry->d_inode;
 669         handle_t *handle;
 670         int err, block_count = 0, blocksize;
 671
 672         /* Determine how many transaction credits are needed */
 673         blocksize = 1 << inode->i_blkbits;
 674         block_count = (*offs & (blocksize - 1)) + bufsize;
 675         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
 676
 677         handle = ext3_journal_start(inode,
 678                         block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
 679         if (IS_ERR(handle)) {
 680                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
 681                        block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
 682                        bufsize);
 683                 return PTR_ERR(handle);
 684         }
 685
 686         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
 687
 688         if (!err && force_sync)
 689                 handle->h_sync = 1; /* recovery likes this */
 690
 691         ext3_journal_stop(handle);
 692
 693         return err;
 694 }
 695
 696 static int fsfilt_ext3_setup(struct super_block *sb)
 697 {
 698         if (!EXT3_HAS_COMPAT_FEATURE(sb,
 699                                 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 700                 CERROR("ext3 mounted without journal\n");
 701                 return -EINVAL;
 702         }
 703
 704 #ifdef S_PDIROPS
 705         CWARN("Enabling PDIROPS\n");
 706         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
 707         sb->s_flags |= S_PDIROPS;
 708 #endif
 709         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
 710                 CWARN("filesystem doesn't have dir_index feature enabled\n");
 711         return 0;
 712 }
 713 static struct fsfilt_operations fsfilt_ext3_ops = {
 714         .fs_type                = "ext3",
 715         .fs_owner              = THIS_MODULE,
 716         .fs_getlabel        = fsfilt_ext3_get_label,
 717         .fs_start              = fsfilt_ext3_start,
 718         .fs_commit            = fsfilt_ext3_commit,
 719         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
 720         .fs_write_record        = fsfilt_ext3_write_record,
 721         .fs_read_record  = fsfilt_ext3_read_record,
 722         .fs_setup              = fsfilt_ext3_setup,
 723 };
 724
 725 static int __init fsfilt_ext3_init(void)
 726 {
 727         int rc;
 728
 729         fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
 730                                          sizeof(struct fsfilt_cb_data), 0, 0);
 731         if (!fcb_cache) {
 732                 CERROR("error allocating fsfilt journal callback cache\n");
 733                 GOTO(out, rc = -ENOMEM);
 734         }
 735
 736         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
 737
 738         if (rc) {
 739                 int err = kmem_cache_destroy(fcb_cache);
 740                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
 741         }
 742 out:
 743         return rc;
 744 }
 745
 746 static void __exit fsfilt_ext3_exit(void)
 747 {
 748         int rc;
 749
 750         fsfilt_unregister_ops(&fsfilt_ext3_ops);
 751         rc = kmem_cache_destroy(fcb_cache);
 752         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
 753 }
 754
 755 module_init(fsfilt_ext3_init);
 756 module_exit(fsfilt_ext3_exit);
 757
 758 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 759 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
 760 MODULE_LICENSE("GPL");