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 <ext4/ext4.h>
  49 #include <ext4/ext4_jbd2.h>
  50 #include <linux/version.h>
  51 #include <linux/bitops.h>
  52 #include <linux/quota.h>
  53
  54 #include <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 /* for kernels 2.6.18 and later */
  63 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  64
  65 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
  66                ext3_ext_insert_extent(handle, inode, path, newext, flag)
  67
  68 #define ext3_mb_discard_inode_preallocations(inode) \
  69                  ext3_discard_preallocations(inode)
  70
  71 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
  72 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
  73
  74 static cfs_mem_cache_t *fcb_cache;
  75
  76 struct fsfilt_cb_data {
  77         struct ext4_journal_cb_entry cb_jcb; /* private data - MUST BE FIRST */
  78         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  79         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  80         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  81         void *cb_data;                  /* MDS/OST completion function data */
  82 };
  83
  84 static char *fsfilt_ext3_get_label(struct super_block *sb)
  85 {
  86         return EXT3_SB(sb)->s_es->s_volume_name;
  87 }
  88
  89 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
  90 #ifdef HAVE_BLOCKS_FOR_TRUNCATE
  91 # include <ext4/truncate.h>
  92 #else
  93 static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
  94 {
  95         ext4_lblk_t needed;
  96
  97         needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
  98         if (needed < 2)
  99                 needed = 2;
 100         if (needed > EXT4_MAX_TRANS_DATA)
 101                 needed = EXT4_MAX_TRANS_DATA;
 102         return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
 103 }
 104 #endif
 105
 106 /*
 107  * We don't currently need any additional blocks for rmdir and
 108  * unlink transactions because we are storing the OST oa_id inside
 109  * the inode (which we will be changing anyways as part of this
 110  * transaction).
 111  */
 112 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 113                                int logs)
 114 {
 115         /* For updates to the last received file */
 116         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 117         journal_t *journal;
 118         void *handle;
 119
 120         if (current->journal_info) {
 121                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 122                        current->journal_info);
 123                 goto journal_start;
 124         }
 125
 126         switch(op) {
 127         case FSFILT_OP_UNLINK:
 128                 /* delete one file + create/update logs for each stripe */
 129                 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
 130                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 131                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 132                 break;
 133         case FSFILT_OP_CANCEL_UNLINK:
 134                 LASSERT(logs == 1);
 135
 136                 /* blocks for log header bitmap update OR
 137                  * blocks for catalog header bitmap update + unlink of logs +
 138                  * blocks for delete the inode (include blocks truncating). */
 139                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 140                           EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
 141                           ext4_blocks_for_truncate(inode) + 3;
 142                 break;
 143         default: CERROR("unknown transaction start op %d\n", op);
 144                 LBUG();
 145         }
 146
 147         LASSERT(current->journal_info == desc_private);
 148         journal = EXT3_SB(inode->i_sb)->s_journal;
 149         if (nblocks > journal->j_max_transaction_buffers) {
 150                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 151                        nblocks, op, logs, journal->j_max_transaction_buffers);
 152                 nblocks = journal->j_max_transaction_buffers;
 153         }
 154
 155  journal_start:
 156         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 157         handle = ext3_journal_start(inode, nblocks);
 158
 159         if (!IS_ERR(handle))
 160                 LASSERT(current->journal_info == handle);
 161         else
 162                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 163                        op, nblocks, PTR_ERR(handle));
 164         return handle;
 165 }
 166
 167 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 168 {
 169         int rc;
 170         handle_t *handle = h;
 171
 172         LASSERT(current->journal_info == handle);
 173         if (force_sync)
 174                 handle->h_sync = 1; /* recovery likes this */
 175
 176         rc = ext3_journal_stop(handle);
 177
 178         return rc;
 179 }
 180
 181 #ifndef EXT3_EXTENTS_FL
 182 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 183 #endif
 184
 185 #ifndef EXT_ASSERT
 186 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 187 #endif
 188
 189 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 190 #define ext3_ext_base                   inode
 191 #define ext3_ext_base2inode(inode)      (inode)
 192 #define EXT_DEPTH(inode)                ext_depth(inode)
 193 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 194                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 195
 196 struct bpointers {
 197         unsigned long *blocks;
 198         unsigned long start;
 199         int num;
 200         int init_num;
 201         int create;
 202 };
 203
 204 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 205                                unsigned long block, int *aflags)
 206 {
 207         struct ext3_inode_info *ei = EXT3_I(inode);
 208         unsigned long bg_start;
 209         unsigned long colour;
 210         int depth;
 211
 212         if (path) {
 213                 struct ext3_extent *ex;
 214                 depth = path->p_depth;
 215
 216                 /* try to predict block placement */
 217                 if ((ex = path[depth].p_ext))
 218                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 219
 220                 /* it looks index is empty
 221                  * try to find starting from index itself */
 222                 if (path[depth].p_bh)
 223                         return path[depth].p_bh->b_blocknr;
 224         }
 225
 226         /* OK. use inode's group */
 227         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 228                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 229         colour = (current->pid % 16) *
 230                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 231         return bg_start + colour + block;
 232 }
 233
 234 #define ll_unmap_underlying_metadata(sb, blocknr) \
 235         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 236
 237 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 238 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 239                                 struct ext3_ext_path *path, unsigned long block,
 240                                 unsigned long *count, int *err)
 241 {
 242         unsigned long pblock, goal;
 243         int aflags = 0;
 244         struct inode *inode = ext3_ext_base2inode(base);
 245
 246         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 247         aflags |= 2; /* block have been already reserved */
 248         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 249         return pblock;
 250
 251 }
 252 #else
 253 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 254                                 struct ext3_ext_path *path, unsigned long block,
 255                                 unsigned long *count, int *err)
 256 {
 257         struct inode *inode = ext3_ext_base2inode(base);
 258         struct ext3_allocation_request ar;
 259         unsigned long pblock;
 260         int aflags;
 261
 262         /* find neighbour allocated blocks */
 263         ar.lleft = block;
 264         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 265         if (*err)
 266                 return 0;
 267         ar.lright = block;
 268         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 269         if (*err)
 270                 return 0;
 271
 272         /* allocate new block */
 273         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 274         ar.inode = inode;
 275         ar.logical = block;
 276         ar.len = *count;
 277         ar.flags = EXT3_MB_HINT_DATA;
 278         pblock = ext3_mb_new_blocks(handle, &ar, err);
 279         *count = ar.len;
 280         return pblock;
 281 }
 282 #endif
 283
 284 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 285                                   struct ext3_ext_path *path,
 286                                   struct ext3_ext_cache *cex,
 287 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 288                                    struct ext3_extent *ex,
 289 #endif
 290                                   void *cbdata)
 291 {
 292         struct bpointers *bp = cbdata;
 293         struct inode *inode = ext3_ext_base2inode(base);
 294         struct ext3_extent nex;
 295         unsigned long pblock;
 296         unsigned long tgen;
 297         int err, i;
 298         unsigned long count;
 299         handle_t *handle;
 300
 301 #ifdef EXT3_EXT_CACHE_EXTENT
 302         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT)
 303 #else
 304         if ((cex->ec_len != 0) && (cex->ec_start != 0))
 305 #endif
 306                                                    {
 307                 err = EXT_CONTINUE;
 308                 goto map;
 309         }
 310
 311         if (bp->create == 0) {
 312                 i = 0;
 313                 if (cex->ec_block < bp->start)
 314                         i = bp->start - cex->ec_block;
 315                 if (i >= cex->ec_len)
 316                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 317                                         i, cex->ec_len);
 318                 for (; i < cex->ec_len && bp->num; i++) {
 319                         *(bp->blocks) = 0;
 320                         bp->blocks++;
 321                         bp->num--;
 322                         bp->start++;
 323                 }
 324
 325                 return EXT_CONTINUE;
 326         }
 327
 328         tgen = EXT_GENERATION(base);
 329         count = ext3_ext_calc_credits_for_insert(base, path);
 330
 331         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 332         if (IS_ERR(handle)) {
 333                 return PTR_ERR(handle);
 334         }
 335
 336         if (tgen != EXT_GENERATION(base)) {
 337                 /* the tree has changed. so path can be invalid at moment */
 338                 ext3_journal_stop(handle);
 339                 return EXT_REPEAT;
 340         }
 341
 342         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
 343          * protected by i_data_sem as whole. so we patch it to store
 344          * generation to path and now verify the tree hasn't changed */
 345         down_write((&EXT4_I(inode)->i_data_sem));
 346
 347         /* validate extent, make sure the extent tree does not changed */
 348         if (EXT_GENERATION(base) != path[0].p_generation) {
 349                 /* cex is invalid, try again */
 350                 up_write(&EXT4_I(inode)->i_data_sem);
 351                 ext3_journal_stop(handle);
 352                 return EXT_REPEAT;
 353         }
 354
 355         count = cex->ec_len;
 356         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 357         if (!pblock)
 358                 goto out;
 359         EXT_ASSERT(count <= cex->ec_len);
 360
 361         /* insert new extent */
 362         nex.ee_block = cpu_to_le32(cex->ec_block);
 363         ext3_ext_store_pblock(&nex, pblock);
 364         nex.ee_len = cpu_to_le16(count);
 365         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
 366         if (err) {
 367                 /* free data blocks we just allocated */
 368                 /* not a good idea to call discard here directly,
 369                  * but otherwise we'd need to call it every free() */
 370 #ifdef EXT3_MB_HINT_GROUP_ALLOC
 371                 ext3_mb_discard_inode_preallocations(inode);
 372 #endif
 373                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
 374                                  cpu_to_le16(nex.ee_len), 0);
 375                 goto out;
 376         }
 377
 378         /*
 379          * Putting len of the actual extent we just inserted,
 380          * we are asking ext3_ext_walk_space() to continue
 381          * scaning after that block
 382          */
 383         cex->ec_len = le16_to_cpu(nex.ee_len);
 384         cex->ec_start = ext_pblock(&nex);
 385         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
 386         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
 387
 388 out:
 389         up_write((&EXT4_I(inode)->i_data_sem));
 390         ext3_journal_stop(handle);
 391 map:
 392         if (err >= 0) {
 393                 /* map blocks */
 394                 if (bp->num == 0) {
 395                         CERROR("hmm. why do we find this extent?\n");
 396                         CERROR("initial space: %lu:%u\n",
 397                                 bp->start, bp->init_num);
 398 #ifdef EXT3_EXT_CACHE_EXTENT
 399                         CERROR("current extent: %u/%u/%llu %d\n",
 400                                 cex->ec_block, cex->ec_len,
 401                                 (unsigned long long)cex->ec_start,
 402                                 cex->ec_type);
 403 #else
 404                         CERROR("current extent: %u/%u/%llu\n",
 405                                 cex->ec_block, cex->ec_len,
 406                                 (unsigned long long)cex->ec_start);
 407 #endif
 408                 }
 409                 i = 0;
 410                 if (cex->ec_block < bp->start)
 411                         i = bp->start - cex->ec_block;
 412                 if (i >= cex->ec_len)
 413                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 414                                         i, cex->ec_len);
 415                 for (; i < cex->ec_len && bp->num; i++) {
 416                         *(bp->blocks) = cex->ec_start + i;
 417 #ifdef EXT3_EXT_CACHE_EXTENT
 418                         if (cex->ec_type != EXT3_EXT_CACHE_EXTENT)
 419 #else
 420                         if ((cex->ec_len == 0) || (cex->ec_start == 0))
 421 #endif
 422                                                                         {
 423                                 /* unmap any possible underlying metadata from
 424                                  * the block device mapping.  bug 6998. */
 425                                 ll_unmap_underlying_metadata(inode->i_sb,
 426                                                              *(bp->blocks));
 427                         }
 428                         bp->blocks++;
 429                         bp->num--;
 430                         bp->start++;
 431                 }
 432         }
 433         return err;
 434 }
 435
 436 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
 437                        unsigned long num, unsigned long *blocks,
 438                        int create)
 439 {
 440         struct ext3_ext_base *base = inode;
 441         struct bpointers bp;
 442         int err;
 443
 444         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
 445                block, block + num - 1, (unsigned) inode->i_ino);
 446
 447         bp.blocks = blocks;
 448         bp.start = block;
 449         bp.init_num = bp.num = num;
 450         bp.create = create;
 451
 452         err = fsfilt_ext3_ext_walk_space(base, block, num,
 453                                          ext3_ext_new_extent_cb, &bp);
 454         ext3_ext_invalidate_cache(base);
 455
 456         return err;
 457 }
 458
 459 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
 460                                     int pages, unsigned long *blocks,
 461                                     int create)
 462 {
 463         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
 464         int rc = 0, i = 0;
 465         struct page *fp = NULL;
 466         int clen = 0;
 467
 468         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
 469                 inode->i_ino, pages, (*page)->index);
 470
 471         /* pages are sorted already. so, we just have to find
 472          * contig. space and process them properly */
 473         while (i < pages) {
 474                 if (fp == NULL) {
 475                         /* start new extent */
 476                         fp = *page++;
 477                         clen = 1;
 478                         i++;
 479                         continue;
 480                 } else if (fp->index + clen == (*page)->index) {
 481                         /* continue the extent */
 482                         page++;
 483                         clen++;
 484                         i++;
 485                         continue;
 486                 }
 487
 488                 /* process found extent */
 489                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 490                                         clen * blocks_per_page, blocks,
 491                                         create);
 492                 if (rc)
 493                         GOTO(cleanup, rc);
 494
 495                 /* look for next extent */
 496                 fp = NULL;
 497                 blocks += blocks_per_page * clen;
 498         }
 499
 500         if (fp)
 501                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
 502                                         clen * blocks_per_page, blocks,
 503                                         create);
 504 cleanup:
 505         return rc;
 506 }
 507
 508 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
 509                                    int pages, unsigned long *blocks,
 510                                    int create)
 511 {
 512         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
 513         unsigned long *b;
 514         int rc = 0, i;
 515
 516         for (i = 0, b = blocks; i < pages; i++, page++) {
 517                 rc = ext3_map_inode_page(inode, *page, b, create);
 518                 if (rc) {
 519                         CERROR("ino %lu, blk %lu create %d: rc %d\n",
 520                                inode->i_ino, *b, create, rc);
 521                         break;
 522                 }
 523
 524                 b += blocks_per_page;
 525         }
 526         return rc;
 527 }
 528
 529 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
 530                                 int pages, unsigned long *blocks,
 531                                 int create, struct mutex *optional_mutex)
 532 {
 533         int rc;
 534
 535         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
 536                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
 537                                                      blocks, create);
 538                 return rc;
 539         }
 540         if (optional_mutex != NULL)
 541                 mutex_lock(optional_mutex);
 542         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks, create);
 543         if (optional_mutex != NULL)
 544                 mutex_unlock(optional_mutex);
 545
 546         return rc;
 547 }
 548
 549 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
 550 {
 551         unsigned long block;
 552         struct buffer_head *bh;
 553         int err, blocksize, csize, boffs, osize = size;
 554
 555         /* prevent reading after eof */
 556         spin_lock(&inode->i_lock);
 557         if (i_size_read(inode) < *offs + size) {
 558                 size = i_size_read(inode) - *offs;
 559                 spin_unlock(&inode->i_lock);
 560                 if (size < 0) {
 561                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
 562                                i_size_read(inode), *offs);
 563                         return -EBADR;
 564                 } else if (size == 0) {
 565                         return 0;
 566                 }
 567         } else {
 568                 spin_unlock(&inode->i_lock);
 569         }
 570
 571         blocksize = 1 << inode->i_blkbits;
 572
 573         while (size > 0) {
 574                 block = *offs >> inode->i_blkbits;
 575                 boffs = *offs & (blocksize - 1);
 576                 csize = min(blocksize - boffs, size);
 577                 bh = ext3_bread(NULL, inode, block, 0, &err);
 578                 if (!bh) {
 579                         CERROR("can't read block: %d\n", err);
 580                         return err;
 581                 }
 582
 583                 memcpy(buf, bh->b_data + boffs, csize);
 584                 brelse(bh);
 585
 586                 *offs += csize;
 587                 buf += csize;
 588                 size -= csize;
 589         }
 590         return osize;
 591 }
 592 EXPORT_SYMBOL(fsfilt_ext3_read);
 593
 594 static int fsfilt_ext3_read_record(struct file * file, void *buf,
 595                                    int size, loff_t *offs)
 596 {
 597         int rc;
 598         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
 599         if (rc > 0)
 600                 rc = 0;
 601         return rc;
 602 }
 603
 604 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
 605                                 loff_t *offs, handle_t *handle)
 606 {
 607         struct buffer_head *bh = NULL;
 608         loff_t old_size = i_size_read(inode), offset = *offs;
 609         loff_t new_size = i_size_read(inode);
 610         unsigned long block;
 611         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
 612
 613         while (bufsize > 0) {
 614                 if (bh != NULL)
 615                         brelse(bh);
 616
 617                 block = offset >> inode->i_blkbits;
 618                 boffs = offset & (blocksize - 1);
 619                 size = min(blocksize - boffs, bufsize);
 620                 bh = ext3_bread(handle, inode, block, 1, &err);
 621                 if (!bh) {
 622                         CERROR("can't read/create block: %d\n", err);
 623                         break;
 624                 }
 625
 626                 err = ext3_journal_get_write_access(handle, bh);
 627                 if (err) {
 628                         CERROR("journal_get_write_access() returned error %d\n",
 629                                err);
 630                         break;
 631                 }
 632                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
 633                 memcpy(bh->b_data + boffs, buf, size);
 634                 err = ext3_journal_dirty_metadata(handle, bh);
 635                 if (err) {
 636                         CERROR("journal_dirty_metadata() returned error %d\n",
 637                                err);
 638                         break;
 639                 }
 640                 if (offset + size > new_size)
 641                         new_size = offset + size;
 642                 offset += size;
 643                 bufsize -= size;
 644                 buf += size;
 645         }
 646         if (bh)
 647                 brelse(bh);
 648
 649         /* correct in-core and on-disk sizes */
 650         if (new_size > i_size_read(inode)) {
 651                 spin_lock(&inode->i_lock);
 652                 if (new_size > i_size_read(inode))
 653                         i_size_write(inode, new_size);
 654                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
 655                         EXT3_I(inode)->i_disksize = i_size_read(inode);
 656                 if (i_size_read(inode) > old_size) {
 657                         spin_unlock(&inode->i_lock);
 658                         mark_inode_dirty(inode);
 659                 } else {
 660                         spin_unlock(&inode->i_lock);
 661                 }
 662         }
 663
 664         if (err == 0)
 665                 *offs = offset;
 666         return err;
 667 }
 668 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
 669
 670 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
 671                                     loff_t *offs, int force_sync)
 672 {
 673         struct inode *inode = file->f_dentry->d_inode;
 674         handle_t *handle;
 675         int err, block_count = 0, blocksize;
 676
 677         /* Determine how many transaction credits are needed */
 678         blocksize = 1 << inode->i_blkbits;
 679         block_count = (*offs & (blocksize - 1)) + bufsize;
 680         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
 681
 682         handle = ext3_journal_start(inode,
 683                         block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
 684         if (IS_ERR(handle)) {
 685                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
 686                        block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
 687                        bufsize);
 688                 return PTR_ERR(handle);
 689         }
 690
 691         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
 692
 693         if (!err && force_sync)
 694                 handle->h_sync = 1; /* recovery likes this */
 695
 696         ext3_journal_stop(handle);
 697
 698         return err;
 699 }
 700
 701 static int fsfilt_ext3_setup(struct super_block *sb)
 702 {
 703         if (!EXT3_HAS_COMPAT_FEATURE(sb,
 704                                 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 705                 CERROR("ext3 mounted without journal\n");
 706                 return -EINVAL;
 707         }
 708
 709 #ifdef S_PDIROPS
 710         CWARN("Enabling PDIROPS\n");
 711         set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
 712         sb->s_flags |= S_PDIROPS;
 713 #endif
 714         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
 715                 CWARN("filesystem doesn't have dir_index feature enabled\n");
 716         return 0;
 717 }
 718 static struct fsfilt_operations fsfilt_ext3_ops = {
 719         .fs_type                = "ext3",
 720         .fs_owner               = THIS_MODULE,
 721         .fs_getlabel            = fsfilt_ext3_get_label,
 722         .fs_start               = fsfilt_ext3_start,
 723         .fs_commit              = fsfilt_ext3_commit,
 724         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
 725         .fs_write_record        = fsfilt_ext3_write_record,
 726         .fs_read_record         = fsfilt_ext3_read_record,
 727         .fs_setup               = fsfilt_ext3_setup,
 728 };
 729
 730 static int __init fsfilt_ext3_init(void)
 731 {
 732         int rc;
 733
 734         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
 735                                          sizeof(struct fsfilt_cb_data), 0, 0);
 736         if (!fcb_cache) {
 737                 CERROR("error allocating fsfilt journal callback cache\n");
 738                 GOTO(out, rc = -ENOMEM);
 739         }
 740
 741         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
 742
 743         if (rc) {
 744                 int err = cfs_mem_cache_destroy(fcb_cache);
 745                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
 746         }
 747 out:
 748         return rc;
 749 }
 750
 751 static void __exit fsfilt_ext3_exit(void)
 752 {
 753         int rc;
 754
 755         fsfilt_unregister_ops(&fsfilt_ext3_ops);
 756         rc = cfs_mem_cache_destroy(fcb_cache);
 757         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
 758 }
 759
 760 module_init(fsfilt_ext3_init);
 761 module_exit(fsfilt_ext3_exit);
 762
 763 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 764 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
 765 MODULE_LICENSE("GPL");