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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011 Whamcloud, Inc.
  33  *
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/lvfs/fsfilt_ext3.c
  40  *
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_FILTER
  45
  46 #include <linux/init.h>
  47 #include <linux/module.h>
  48 #include <linux/fs.h>
  49 #include <linux/slab.h>
  50 #include <linux/pagemap.h>
  51 #include <linux/quotaops.h>
  52 #ifdef HAVE_LINUX_EXPORTFS_H
  53 #include <linux/exportfs.h>
  54 #endif
  55 #ifdef HAVE_EXT4_LDISKFS
  56 #include <ext4/ext4.h>
  57 #include <ext4/ext4_jbd2.h>
  58 #else
  59 #include <linux/jbd.h>
  60 #include <linux/ext3_fs.h>
  61 #include <linux/ext3_jbd.h>
  62 #endif
  63 #include <linux/version.h>
  64 #include <linux/bitops.h>
  65 #include <linux/quota.h>
  66 #ifdef HAVE_QUOTAIO_H
  67 # include <linux/quotaio_v2.h>
  68 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  69 # include <quota/quotaio_v2.h>
  70 # include <quota/quota_tree.h>
  71 # define V2_DQTREEOFF    QT_TREEOFF
  72 #elif defined(HAVE_FS_QUOTAIO_V1_H)
  73 # include <quotaio_v2.h>
  74 # include <quota_tree.h>
  75 # define V2_DQTREEOFF    QT_TREEOFF
  76 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  77 #endif
  78
  79 #ifdef QFMT_VFS_V1
  80 #define QFMT_LUSTRE QFMT_VFS_V1
  81 #else
  82 #define QFMT_LUSTRE QFMT_VFS_V0
  83 #endif
  84
  85 #if defined(HAVE_EXT3_XATTR_H)
  86 #include <ext3/xattr.h>
  87 #else
  88 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  89 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  90 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  91 #endif
  92
  93 #include <libcfs/libcfs.h>
  94 #include <lustre_fsfilt.h>
  95 #include <obd.h>
  96 #include <lustre_quota.h>
  97 #include <linux/lustre_compat25.h>
  98 #include <linux/lprocfs_status.h>
  99
 100 #ifdef HAVE_EXT4_LDISKFS
 101 #include <ext4/ext4_extents.h>
 102 #else
 103 #include <linux/ext3_extents.h>
 104 #endif
 105
 106 #include "lustre_quota_fmt.h"
 107
 108 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
 109 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 110 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 111 #else
 112 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 113 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 114 #endif
 115
 116 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 117 /* for kernels 2.6.18 and later */
 118 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 119 #else
 120 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 121 #endif
 122
 123 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 124 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 125                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 126 #else
 127 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 128                ext3_ext_insert_extent(handle, inode, path, newext)
 129 #endif
 130
 131 #ifdef EXT3_DISCARD_PREALLOCATIONS
 132 #define ext3_mb_discard_inode_preallocations(inode) \
 133                  ext3_discard_preallocations(inode)
 134 #endif
 135
 136
 137 static cfs_mem_cache_t *fcb_cache;
 138
 139 struct fsfilt_cb_data {
 140         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 141         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 142         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 143         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 144         void *cb_data;                  /* MDS/OST completion function data */
 145 };
 146
 147 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 148 #define EXT3_XATTR_INDEX_TRUSTED        4
 149 #endif
 150
 151 #ifdef HAVE_EXT4_LDISKFS
 152 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 153 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 154 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 155 #else
 156 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 157 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 158 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 159 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 160 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 161 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 162 #endif
 163
 164 #ifndef ext3_find_next_bit
 165 #define ext3_find_next_bit           ext2_find_next_bit
 166 #endif
 167
 168 #ifndef ext2_find_next_bit
 169 #ifdef __LITTLE_ENDIAN
 170 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 171 #else
 172 error "Need implementation of find_next_bit on big-endian systems"
 173 #endif  /* __LITTLE_ENDIAN */
 174 #endif  /* !ext2_find_next_le_bit */
 175
 176 static char *fsfilt_ext3_get_label(struct super_block *sb)
 177 {
 178         return EXT3_SB(sb)->s_es->s_volume_name;
 179 }
 180
 181 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 182 {
 183         /* see e.g. fsfilt_ext3_write_record() */
 184         journal_t *journal;
 185         handle_t *handle;
 186         int err;
 187
 188         journal = EXT3_SB(sb)->s_journal;
 189         handle = ext3_journal_start_sb(sb, 1);
 190         if (IS_ERR(handle)) {
 191                 CERROR("can't start transaction\n");
 192                 return(PTR_ERR(handle));
 193         }
 194
 195         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 196         if (err)
 197                 goto out;
 198
 199         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 200                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 201
 202         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 203
 204 out:
 205         ext3_journal_stop(handle);
 206
 207         return(err);
 208 }
 209
 210 static char *fsfilt_ext3_uuid(struct super_block *sb)
 211 {
 212         return EXT3_SB(sb)->s_es->s_uuid;
 213 }
 214
 215 #ifdef HAVE_DISK_INODE_VERSION
 216
 217 static __u64 get_i_version(struct inode *inode)
 218 {
 219 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 220         return inode->i_version;
 221 #else
 222         return EXT3_I(inode)->i_fs_version;
 223 #endif
 224 }
 225
 226 static void set_i_version(struct inode *inode, __u64 new_version)
 227 {
 228 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 229         inode->i_version = new_version;
 230 #else
 231         (EXT3_I(inode))->i_fs_version = new_version;
 232 #endif
 233 }
 234
 235 /*
 236  * Get the 64-bit version for an inode.
 237  */
 238 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 239 {
 240         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 241                get_i_version(inode), inode->i_ino);
 242         return get_i_version(inode);
 243 }
 244
 245 /*
 246  * Set the 64-bit version and return the old version.
 247  */
 248 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 249 {
 250         __u64 old_version = get_i_version(inode);
 251
 252         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 253                new_version, old_version, inode->i_ino);
 254         set_i_version(inode, new_version);
 255         /* version is set after all inode operations are finished, so we should
 256          * mark it dirty here */
 257         inode->i_sb->s_op->dirty_inode(inode);
 258         return old_version;
 259 }
 260
 261 #endif
 262
 263 /*
 264  * We don't currently need any additional blocks for rmdir and
 265  * unlink transactions because we are storing the OST oa_id inside
 266  * the inode (which we will be changing anyways as part of this
 267  * transaction).
 268  */
 269 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 270                                int logs)
 271 {
 272         /* For updates to the last received file */
 273         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 274         journal_t *journal;
 275         void *handle;
 276
 277         if (current->journal_info) {
 278                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 279                        current->journal_info);
 280                 goto journal_start;
 281         }
 282
 283         switch(op) {
 284         case FSFILT_OP_RMDIR:
 285         case FSFILT_OP_UNLINK:
 286                 /* delete one file + create/update logs for each stripe */
 287                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 288                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 289                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 290                 break;
 291         case FSFILT_OP_RENAME:
 292                 /* modify additional directory */
 293                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 294                 /* no break */
 295         case FSFILT_OP_SYMLINK:
 296                 /* additional block + block bitmap + GDT for long symlink */
 297                 nblocks += 3;
 298                 /* no break */
 299         case FSFILT_OP_CREATE: {
 300 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 301                 static int warned;
 302                 if (!warned) {
 303                         if (!test_opt(inode->i_sb, EXTENTS)) {
 304                                 warned = 1;
 305                         } else if (((EXT3_I(inode)->i_flags &
 306                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 307                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 308                                 CWARN("extent-mapped directory found with "
 309                                       "ext3-based ldiskfs - contact "
 310                                       "http://bugs.whamcloud.com/\n");
 311                                 warned = 1;
 312                         }
 313                 }
 314 #endif
 315                 /* no break */
 316         }
 317         case FSFILT_OP_MKDIR:
 318         case FSFILT_OP_MKNOD:
 319                 /* modify one inode + block bitmap + GDT */
 320                 nblocks += 3;
 321                 /* no break */
 322         case FSFILT_OP_LINK:
 323                 /* modify parent directory */
 324                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 325                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 326                 /* create/update logs for each stripe */
 327                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 328                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 329                 break;
 330         case FSFILT_OP_SETATTR:
 331                 /* Setattr on inode */
 332                 nblocks += 1;
 333                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 334                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 335                 /* quota chown log for each stripe */
 336                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 337                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 338                 break;
 339         case FSFILT_OP_CANCEL_UNLINK:
 340                 /* blocks for log header bitmap update OR
 341                  * blocks for catalog header bitmap update + unlink of logs */
 342                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 343                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 344                 break;
 345         default: CERROR("unknown transaction start op %d\n", op);
 346                 LBUG();
 347         }
 348
 349         LASSERT(current->journal_info == desc_private);
 350         journal = EXT3_SB(inode->i_sb)->s_journal;
 351         if (nblocks > journal->j_max_transaction_buffers) {
 352                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 353                        nblocks, op, logs, journal->j_max_transaction_buffers);
 354                 nblocks = journal->j_max_transaction_buffers;
 355         }
 356
 357  journal_start:
 358         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 359         handle = ext3_journal_start(inode, nblocks);
 360
 361         if (!IS_ERR(handle))
 362                 LASSERT(current->journal_info == handle);
 363         else
 364                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 365                        op, nblocks, PTR_ERR(handle));
 366         return handle;
 367 }
 368
 369 /*
 370  * Calculate the number of buffer credits needed to write multiple pages in
 371  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 372  * doesn't have a nice API for calculating this sort of thing in advance.
 373  *
 374  * See comment above ext3_writepage_trans_blocks for details.  We assume
 375  * no data journaling is being done, but it does allow for all of the pages
 376  * being non-contiguous.  If we are guaranteed contiguous pages we could
 377  * reduce the number of (d)indirect blocks a lot.
 378  *
 379  * With N blocks per page and P pages, for each inode we have at most:
 380  * N*P indirect
 381  * min(N*P, blocksize/4 + 1) dindirect blocks
 382  * niocount tindirect
 383  *
 384  * For the entire filesystem, we have at most:
 385  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 386  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 387  * objcount inode blocks
 388  * 1 superblock
 389  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 390  *
 391  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 392  */
 393 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 394                                       int niocount, struct niobuf_local *nb)
 395 {
 396         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 397         __u64 next_indir;
 398         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 399         int nbitmaps = 0, ngdblocks;
 400         int needed = objcount + 1; /* inodes + superblock */
 401         int i, j;
 402
 403         for (i = 0, j = 0; i < objcount; i++, fso++) {
 404                 /* two or more dindirect blocks in case we cross boundary */
 405                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 406                                     nb[j].offset) >>
 407                                    sb->s_blocksize_bits) /
 408                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 409                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 410
 411                 /* leaf, indirect, tindirect blocks for first block */
 412                 nbitmaps += blockpp + 2;
 413
 414                 j += fso->fso_bufcnt;
 415         }
 416
 417         next_indir = nb[0].offset +
 418                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 419         for (i = 1; i < niocount; i++) {
 420                 if (nb[i].offset >= next_indir) {
 421                         nbitmaps++;     /* additional indirect */
 422                         next_indir = nb[i].offset +
 423                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 424                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 425                         nbitmaps++;     /* additional indirect */
 426                 }
 427                 nbitmaps += blockpp;    /* each leaf in different group? */
 428         }
 429
 430         ngdblocks = nbitmaps;
 431         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 432                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 433         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 434                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 435
 436         needed += nbitmaps + ngdblocks;
 437
 438         /* last_rcvd update */
 439         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 440
 441 #if defined(CONFIG_QUOTA)
 442         /* We assume that there will be 1 bit set in s_dquot.flags for each
 443          * quota file that is active.  This is at least true for now.
 444          */
 445         needed += hweight32(ll_sb_any_quota_active(sb)) *
 446                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 447 #endif
 448
 449         return needed;
 450 }
 451
 452 /* We have to start a huge journal transaction here to hold all of the
 453  * metadata for the pages being written here.  This is necessitated by
 454  * the fact that we do lots of prepare_write operations before we do
 455  * any of the matching commit_write operations, so even if we split
 456  * up to use "smaller" transactions none of them could complete until
 457  * all of them were opened.  By having a single journal transaction,
 458  * we eliminate duplicate reservations for common blocks like the
 459  * superblock and group descriptors or bitmaps.
 460  *
 461  * We will start the transaction here, but each prepare_write will
 462  * add a refcount to the transaction, and each commit_write will
 463  * remove a refcount.  The transaction will be closed when all of
 464  * the pages have been written.
 465  */
 466 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 467                                    int niocount, struct niobuf_local *nb,
 468                                    void *desc_private, int logs)
 469 {
 470         journal_t *journal;
 471         handle_t *handle;
 472         int needed;
 473         ENTRY;
 474
 475         LASSERT(current->journal_info == desc_private);
 476         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 477         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 478
 479         /* The number of blocks we could _possibly_ dirty can very large.
 480          * We reduce our request if it is absurd (and we couldn't get that
 481          * many credits for a single handle anyways).
 482          *
 483          * At some point we have to limit the size of I/Os sent at one time,
 484          * increase the size of the journal, or we have to calculate the
 485          * actual journal requirements more carefully by checking all of
 486          * the blocks instead of being maximally pessimistic.  It remains to
 487          * be seen if this is a real problem or not.
 488          */
 489         if (needed > journal->j_max_transaction_buffers) {
 490                 CERROR("want too many journal credits (%d) using %d instead\n",
 491                        needed, journal->j_max_transaction_buffers);
 492                 needed = journal->j_max_transaction_buffers;
 493         }
 494
 495         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 496         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 497         if (IS_ERR(handle)) {
 498                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 499                        PTR_ERR(handle));
 500         } else {
 501                 LASSERT(handle->h_buffer_credits >= needed);
 502                 LASSERT(current->journal_info == handle);
 503         }
 504
 505         RETURN(handle);
 506 }
 507
 508 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 509 {
 510        handle_t *handle = h;
 511
 512        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 513        if (nblocks == 0) {
 514                handle->h_buffer_credits = 0;
 515                CWARN("setting credits of handle %p to zero by request\n", h);
 516        }
 517
 518        if (handle->h_buffer_credits > nblocks)
 519                 return 0;
 520        if (ext3_journal_extend(handle, nblocks) == 0)
 521                 return 0;
 522
 523        ext3_mark_inode_dirty(handle, inode);
 524        return ext3_journal_restart(handle, nblocks);
 525 }
 526
 527 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 528 {
 529         int rc;
 530         handle_t *handle = h;
 531
 532         LASSERT(current->journal_info == handle);
 533         if (force_sync)
 534                 handle->h_sync = 1; /* recovery likes this */
 535
 536         rc = ext3_journal_stop(handle);
 537
 538         return rc;
 539 }
 540
 541 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 542                                     void **wait_handle)
 543 {
 544         unsigned long tid;
 545         transaction_t *transaction;
 546         handle_t *handle = h;
 547         journal_t *journal;
 548         int rc;
 549
 550         LASSERT(current->journal_info == handle);
 551
 552         transaction = handle->h_transaction;
 553         journal = transaction->t_journal;
 554         tid = transaction->t_tid;
 555         /* we don't want to be blocked */
 556         handle->h_sync = 0;
 557         rc = ext3_journal_stop(handle);
 558         if (rc) {
 559                 CERROR("error while stopping transaction: %d\n", rc);
 560                 return rc;
 561         }
 562         fsfilt_log_start_commit(journal, tid);
 563
 564         *wait_handle = (void *) tid;
 565         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 566         return 0;
 567 }
 568
 569 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 570 {
 571         journal_t *journal = EXT3_JOURNAL(inode);
 572         tid_t tid = (tid_t)(long)h;
 573
 574         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 575         if (unlikely(is_journal_aborted(journal)))
 576                 return -EIO;
 577
 578         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 579
 580         if (unlikely(is_journal_aborted(journal)))
 581                 return -EIO;
 582         return 0;
 583 }
 584
 585 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 586                                struct iattr *iattr, int do_trunc)
 587 {
 588         struct inode *inode = dentry->d_inode;
 589         int rc = 0;
 590
 591         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 592          * We are already writing the inode to disk as part of this
 593          * transaction and want to avoid a lot of extra inode writeout
 594          * later on. b=9828 */
 595         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 596                 /* ATTR_SIZE would invoke truncate: clear it */
 597                 iattr->ia_valid &= ~ATTR_SIZE;
 598                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 599                 i_size_write(inode, iattr->ia_size);
 600
 601                 if (iattr->ia_valid & ATTR_UID)
 602                         inode->i_uid = iattr->ia_uid;
 603                 if (iattr->ia_valid & ATTR_GID)
 604                         inode->i_gid = iattr->ia_gid;
 605                 if (iattr->ia_valid & ATTR_ATIME)
 606                         inode->i_atime = iattr->ia_atime;
 607                 if (iattr->ia_valid & ATTR_MTIME)
 608                         inode->i_mtime = iattr->ia_mtime;
 609                 if (iattr->ia_valid & ATTR_CTIME)
 610                         inode->i_ctime = iattr->ia_ctime;
 611                 if (iattr->ia_valid & ATTR_MODE) {
 612                         inode->i_mode = iattr->ia_mode;
 613
 614                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 615                             !cfs_capable(CFS_CAP_FSETID))
 616                                 inode->i_mode &= ~S_ISGID;
 617                 }
 618
 619                 inode->i_sb->s_op->dirty_inode(inode);
 620
 621                 goto out;
 622         }
 623
 624         /* Don't allow setattr to change file type */
 625         if (iattr->ia_valid & ATTR_MODE)
 626                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 627                                  (iattr->ia_mode & ~S_IFMT);
 628
 629         /* We set these flags on the client, but have already checked perms
 630          * so don't confuse inode_change_ok. */
 631         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 632
 633         if (inode->i_op->setattr) {
 634                 rc = inode->i_op->setattr(dentry, iattr);
 635         } else {
 636                 rc = inode_change_ok(inode, iattr);
 637                 if (!rc)
 638                         rc = inode_setattr(inode, iattr);
 639         }
 640
 641  out:
 642         RETURN(rc);
 643 }
 644
 645 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 646                                  unsigned int cmd, unsigned long arg)
 647 {
 648         int rc = 0;
 649         ENTRY;
 650
 651         /* FIXME: Can't do this because of nested transaction deadlock */
 652         if (cmd == EXT3_IOC_SETFLAGS) {
 653                 /* We can't enable data journaling on OST objects, because
 654                 * this forces the transaction to be closed in order to
 655                 * flush the journal, but the caller will already have a
 656                 * compound transaction open to update the last_rcvd file,
 657                 * and this thread would deadlock trying to set the flag. */
 658                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 659                         CERROR("can't set data journal flag on file\n");
 660                         RETURN(-EPERM);
 661                 }
 662                 /* Because the MDS does not see the EXTENTS_FL set on the
 663                  * OST objects, mask this flag into all set flags.  It is
 664                  * not legal to clear this flag in any case, so we are not
 665                  * changing the functionality by doing this.  b=22911 */
 666                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 667         }
 668
 669 #ifdef HAVE_EXT4_LDISKFS
 670         /* ext4_ioctl does not have a inode argument */
 671         if (inode->i_fop->unlocked_ioctl)
 672                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 673 #else
 674         if (inode->i_fop->ioctl)
 675                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 676 #endif
 677         else
 678                 RETURN(-ENOTTY);
 679
 680         RETURN(rc);
 681 }
 682
 683 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 684                               void *lmm, int lmm_size, const char *name)
 685 {
 686         int rc;
 687
 688         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 689
 690         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 691                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 692
 693
 694         if (rc && rc != -EROFS)
 695                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 696                        inode->i_ino, rc);
 697         return rc;
 698 }
 699
 700 /* Must be called with i_mutex held */
 701 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 702                               const char *name)
 703 {
 704         int rc;
 705
 706         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 707
 708         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 709                             name, lmm, lmm_size);
 710
 711         /* This gives us the MD size */
 712         if (lmm == NULL)
 713                 return (rc == -ENODATA) ? 0 : rc;
 714
 715         if (rc < 0) {
 716                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 717                        EXT3_XATTR_INDEX_TRUSTED, name,
 718                        inode->i_ino, rc);
 719                 memset(lmm, 0, lmm_size);
 720                 return (rc == -ENODATA) ? 0 : rc;
 721         }
 722
 723         return rc;
 724 }
 725
 726 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 727 {
 728         submit_bio(rw, bio);
 729         return 0;
 730 }
 731
 732 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 733                                     loff_t *off)
 734 {
 735         struct inode *inode = file->f_dentry->d_inode;
 736         int rc = 0;
 737
 738         if (S_ISREG(inode->i_mode))
 739                 rc = file->f_op->read(file, buf, count, off);
 740         else {
 741                 const int blkbits = inode->i_sb->s_blocksize_bits;
 742                 const int blksize = inode->i_sb->s_blocksize;
 743
 744                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 745                        (unsigned long)count, inode->i_ino, *off);
 746                 while (count > 0) {
 747                         struct buffer_head *bh;
 748
 749                         bh = NULL;
 750                         if (*off < i_size_read(inode)) {
 751                                 int err = 0;
 752
 753                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 754                                                 0, &err);
 755
 756                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 757
 758                                 if (bh) {
 759                                         memcpy(buf, bh->b_data, blksize);
 760                                         brelse(bh);
 761                                 } else if (err) {
 762                                         /* XXX in theory we should just fake
 763                                          * this buffer and continue like ext3,
 764                                          * especially if this is a partial read
 765                                          */
 766                                         CERROR("error read dir %lu+%llu: %d\n",
 767                                                inode->i_ino, *off, err);
 768                                         RETURN(err);
 769                                 }
 770                         }
 771                         if (!bh) {
 772                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 773
 774                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 775                                 memset(fake, 0, sizeof(*fake));
 776                                 fake->rec_len = cpu_to_le16(blksize);
 777                         }
 778                         count -= blksize;
 779                         buf += blksize;
 780                         *off += blksize;
 781                         rc += blksize;
 782                 }
 783         }
 784
 785         return rc;
 786 }
 787
 788 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 789 {
 790         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 791
 792         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 793
 794         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 795 }
 796
 797 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 798                                       void *handle, fsfilt_cb_t cb_func,
 799                                       void *cb_data)
 800 {
 801         struct fsfilt_cb_data *fcb;
 802
 803         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 804         if (fcb == NULL)
 805                 RETURN(-ENOMEM);
 806
 807         fcb->cb_func = cb_func;
 808         fcb->cb_obd = obd;
 809         fcb->cb_last_rcvd = last_rcvd;
 810         fcb->cb_data = cb_data;
 811
 812         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 813         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 814                                     (struct journal_callback *)fcb);
 815
 816         return 0;
 817 }
 818
 819 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 820 {
 821         struct kstatfs sfs;
 822         int rc;
 823
 824         memset(&sfs, 0, sizeof(sfs));
 825         rc = ll_do_statfs(sb, &sfs);
 826         statfs_pack(osfs, &sfs);
 827         return rc;
 828 }
 829
 830 static int fsfilt_ext3_sync(struct super_block *sb)
 831 {
 832         return ext3_force_commit(sb);
 833 }
 834
 835 #ifndef EXT3_EXTENTS_FL
 836 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 837 #endif
 838
 839 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 840 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 841 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 842 #else
 843 # ifdef HAVE_EXT4_LDISKFS
 844 #  ifdef WALK_SPACE_HAS_DATA_SEM /* We only use it in fsfilt_map_nblocks() for now */
 845 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 846 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 847 #  else
 848 #   define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem))
 849 #   define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem))
 850 #  endif
 851 # else
 852 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 853 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 854 # endif
 855 #endif
 856
 857 #ifndef EXT_ASSERT
 858 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 859 #endif
 860
 861 #ifdef EXT3_EXT_HAS_NO_TREE
 862 /* for kernels 2.6.18 and later */
 863 #ifdef HAVE_EXT4_LDISKFS
 864 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 865 #else
 866 #define EXT_GENERATION(inode)           ext_generation(inode)
 867 #endif
 868 #define ext3_ext_base                   inode
 869 #define ext3_ext_base2inode(inode)      (inode)
 870 #define EXT_DEPTH(inode)                ext_depth(inode)
 871 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 872                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 873 #else
 874 #define ext3_ext_base                   ext3_extents_tree
 875 #define ext3_ext_base2inode(tree)       (tree->inode)
 876 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 877                         ext3_ext_walk_space(tree, block, num, cb);
 878 #endif
 879
 880 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
 881 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 882                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 883 #else
 884 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 885                ext3_ext_insert_extent(handle, inode, path, newext)
 886 #endif
 887
 888 #include <linux/lustre_version.h>
 889
 890 struct bpointers {
 891         unsigned long *blocks;
 892         int *created;
 893         unsigned long start;
 894         int num;
 895         int init_num;
 896         int create;
 897 };
 898
 899 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 900                                unsigned long block, int *aflags)
 901 {
 902         struct ext3_inode_info *ei = EXT3_I(inode);
 903         unsigned long bg_start;
 904         unsigned long colour;
 905         int depth;
 906
 907         if (path) {
 908                 struct ext3_extent *ex;
 909                 depth = path->p_depth;
 910
 911                 /* try to predict block placement */
 912                 if ((ex = path[depth].p_ext))
 913                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 914
 915                 /* it looks index is empty
 916                  * try to find starting from index itself */
 917                 if (path[depth].p_bh)
 918                         return path[depth].p_bh->b_blocknr;
 919         }
 920
 921         /* OK. use inode's group */
 922         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 923                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 924         colour = (current->pid % 16) *
 925                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 926         return bg_start + colour + block;
 927 }
 928
 929 #define ll_unmap_underlying_metadata(sb, blocknr) \
 930         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 931
 932 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 933 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 934                                 struct ext3_ext_path *path, unsigned long block,
 935                                 unsigned long *count, int *err)
 936 {
 937         unsigned long pblock, goal;
 938         int aflags = 0;
 939         struct inode *inode = ext3_ext_base2inode(base);
 940
 941         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 942         aflags |= 2; /* block have been already reserved */
 943         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 944         return pblock;
 945
 946 }
 947 #else
 948 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 949                                 struct ext3_ext_path *path, unsigned long block,
 950                                 unsigned long *count, int *err)
 951 {
 952         struct inode *inode = ext3_ext_base2inode(base);
 953         struct ext3_allocation_request ar;
 954         unsigned long pblock;
 955         int aflags;
 956
 957         /* find neighbour allocated blocks */
 958         ar.lleft = block;
 959         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 960         if (*err)
 961                 return 0;
 962         ar.lright = block;
 963         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 964         if (*err)
 965                 return 0;
 966
 967         /* allocate new block */
 968         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 969         ar.inode = inode;
 970         ar.logical = block;
 971         ar.len = *count;
 972         ar.flags = EXT3_MB_HINT_DATA;
 973         pblock = ext3_mb_new_blocks(handle, &ar, err);
 974         *count = ar.len;
 975         return pblock;
 976 }
 977 #endif
 978
 979 #ifdef EXT3_EXT_HAS_NO_TREE
 980 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 981                                   struct ext3_ext_path *path,
 982                                   struct ext3_ext_cache *cex,
 983 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 984                                    struct ext3_extent *ex,
 985 #endif
 986                                   void *cbdata)
 987 {
 988         struct bpointers *bp = cbdata;
 989 #else
 990 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 991                                   struct ext3_ext_path *path,
 992                                   struct ext3_ext_cache *cex
 993 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 994                                   , struct ext3_extent *ex
 995 #endif
 996                                  )
 997 {
 998         struct bpointers *bp = base->private;
 999 #endif
1000         struct inode *inode = ext3_ext_base2inode(base);
1001         struct ext3_extent nex;
1002 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1003         struct ext4_ext_path *tmppath = NULL;
1004         struct ext4_extent *tmpex;
1005 #endif
1006         unsigned long pblock;
1007         unsigned long tgen;
1008         int err, i, depth;
1009         unsigned long count;
1010         handle_t *handle;
1011
1012         i = depth = EXT_DEPTH(base);
1013         EXT_ASSERT(i == path->p_depth);
1014         EXT_ASSERT(path[i].p_hdr);
1015
1016         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1017                 err = EXT_CONTINUE;
1018                 goto map;
1019         }
1020
1021         if (bp->create == 0) {
1022                 i = 0;
1023                 if (cex->ec_block < bp->start)
1024                         i = bp->start - cex->ec_block;
1025                 if (i >= cex->ec_len)
1026                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1027                                         i, cex->ec_len);
1028                 for (; i < cex->ec_len && bp->num; i++) {
1029                         *(bp->created) = 0;
1030                         bp->created++;
1031                         *(bp->blocks) = 0;
1032                         bp->blocks++;
1033                         bp->num--;
1034                         bp->start++;
1035                 }
1036
1037                 return EXT_CONTINUE;
1038         }
1039
1040         tgen = EXT_GENERATION(base);
1041         count = ext3_ext_calc_credits_for_insert(base, path);
1042         fsfilt_up_truncate_sem(inode);
1043
1044         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1045         if (IS_ERR(handle)) {
1046                 fsfilt_down_truncate_sem(inode);
1047                 return PTR_ERR(handle);
1048         }
1049
1050         fsfilt_down_truncate_sem(inode);
1051         if (tgen != EXT_GENERATION(base)) {
1052                 /* the tree has changed. so path can be invalid at moment */
1053                 ext3_journal_stop(handle);
1054                 return EXT_REPEAT;
1055         }
1056
1057 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1058         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1059          * protected by i_data_sem, we need revalidate extent to be created */
1060         down_write((&EXT4_I(inode)->i_data_sem));
1061
1062         /* validate extent, make sure the extent tree does not changed */
1063         tmppath = ext4_ext_find_extent(inode, cex->ec_block, NULL);
1064         if (IS_ERR(tmppath)) {
1065                 up_write(&EXT4_I(inode)->i_data_sem);
1066                 ext3_journal_stop(handle);
1067                 return PTR_ERR(tmppath);
1068         }
1069         tmpex = tmppath[depth].p_ext;
1070         if (tmpex != ex) {
1071                 /* cex is invalid, try again */
1072                 ext4_ext_drop_refs(tmppath);
1073                 kfree(tmppath);
1074                 up_write(&EXT4_I(inode)->i_data_sem);
1075                 ext3_journal_stop(handle);
1076                 return EXT_REPEAT;
1077         }
1078 #endif
1079
1080         count = cex->ec_len;
1081         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1082         if (!pblock)
1083                 goto out;
1084         EXT_ASSERT(count <= cex->ec_len);
1085
1086         /* insert new extent */
1087         nex.ee_block = cpu_to_le32(cex->ec_block);
1088         ext3_ext_store_pblock(&nex, pblock);
1089         nex.ee_len = cpu_to_le16(count);
1090         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1091         if (err) {
1092                 /* free data blocks we just allocated */
1093                 /* not a good idea to call discard here directly,
1094                  * but otherwise we'd need to call it every free() */
1095 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1096                 ext3_mb_discard_inode_preallocations(inode);
1097 #endif
1098                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1099                                  cpu_to_le16(nex.ee_len), 0);
1100                 goto out;
1101         }
1102
1103         /*
1104          * Putting len of the actual extent we just inserted,
1105          * we are asking ext3_ext_walk_space() to continue
1106          * scaning after that block
1107          */
1108         cex->ec_len = le16_to_cpu(nex.ee_len);
1109         cex->ec_start = ext_pblock(&nex);
1110         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1111         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1112
1113 out:
1114 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1115         ext4_ext_drop_refs(tmppath);
1116         kfree(tmppath);
1117         up_write((&EXT4_I(inode)->i_data_sem));
1118 #endif
1119         ext3_journal_stop(handle);
1120 map:
1121         if (err >= 0) {
1122                 /* map blocks */
1123                 if (bp->num == 0) {
1124                         CERROR("hmm. why do we find this extent?\n");
1125                         CERROR("initial space: %lu:%u\n",
1126                                 bp->start, bp->init_num);
1127                         CERROR("current extent: %u/%u/%llu %d\n",
1128                                 cex->ec_block, cex->ec_len,
1129                                 (unsigned long long)cex->ec_start,
1130                                 cex->ec_type);
1131                 }
1132                 i = 0;
1133                 if (cex->ec_block < bp->start)
1134                         i = bp->start - cex->ec_block;
1135                 if (i >= cex->ec_len)
1136                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1137                                         i, cex->ec_len);
1138                 for (; i < cex->ec_len && bp->num; i++) {
1139                         *(bp->blocks) = cex->ec_start + i;
1140                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1141                                 *(bp->created) = 0;
1142                         } else {
1143                                 *(bp->created) = 1;
1144                                 /* unmap any possible underlying metadata from
1145                                  * the block device mapping.  bug 6998. */
1146                                 ll_unmap_underlying_metadata(inode->i_sb,
1147                                                              *(bp->blocks));
1148                         }
1149                         bp->created++;
1150                         bp->blocks++;
1151                         bp->num--;
1152                         bp->start++;
1153                 }
1154         }
1155         return err;
1156 }
1157
1158 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1159                        unsigned long num, unsigned long *blocks,
1160                        int *created, int create)
1161 {
1162 #ifdef EXT3_EXT_HAS_NO_TREE
1163         struct ext3_ext_base *base = inode;
1164 #else
1165         struct ext3_extents_tree tree;
1166         struct ext3_ext_base *base = &tree;
1167 #endif
1168         struct bpointers bp;
1169         int err;
1170
1171         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1172                block, block + num - 1, (unsigned) inode->i_ino);
1173
1174 #ifndef EXT3_EXT_HAS_NO_TREE
1175         ext3_init_tree_desc(base, inode);
1176         tree.private = &bp;
1177 #endif
1178         bp.blocks = blocks;
1179         bp.created = created;
1180         bp.start = block;
1181         bp.init_num = bp.num = num;
1182         bp.create = create;
1183
1184         fsfilt_down_truncate_sem(inode);
1185         err = fsfilt_ext3_ext_walk_space(base, block, num,
1186                                          ext3_ext_new_extent_cb, &bp);
1187         ext3_ext_invalidate_cache(base);
1188         fsfilt_up_truncate_sem(inode);
1189
1190         return err;
1191 }
1192
1193 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1194                                     int pages, unsigned long *blocks,
1195                                     int *created, int create)
1196 {
1197         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1198         int rc = 0, i = 0;
1199         struct page *fp = NULL;
1200         int clen = 0;
1201
1202         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1203                 inode->i_ino, pages, (*page)->index);
1204
1205         /* pages are sorted already. so, we just have to find
1206          * contig. space and process them properly */
1207         while (i < pages) {
1208                 if (fp == NULL) {
1209                         /* start new extent */
1210                         fp = *page++;
1211                         clen = 1;
1212                         i++;
1213                         continue;
1214                 } else if (fp->index + clen == (*page)->index) {
1215                         /* continue the extent */
1216                         page++;
1217                         clen++;
1218                         i++;
1219                         continue;
1220                 }
1221
1222                 /* process found extent */
1223                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1224                                         clen * blocks_per_page, blocks,
1225                                         created, create);
1226                 if (rc)
1227                         GOTO(cleanup, rc);
1228
1229                 /* look for next extent */
1230                 fp = NULL;
1231                 blocks += blocks_per_page * clen;
1232                 created += blocks_per_page * clen;
1233         }
1234
1235         if (fp)
1236                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1237                                         clen * blocks_per_page, blocks,
1238                                         created, create);
1239 cleanup:
1240         return rc;
1241 }
1242
1243 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1244                                unsigned long *blocks, int *created, int create);
1245 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1246                                    int pages, unsigned long *blocks,
1247                                    int *created, int create)
1248 {
1249         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1250         unsigned long *b;
1251         int rc = 0, i, *cr;
1252
1253         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1254                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1255                 if (rc) {
1256                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1257                                inode->i_ino, *b, *cr, create, rc);
1258                         break;
1259                 }
1260
1261                 b += blocks_per_page;
1262                 cr += blocks_per_page;
1263         }
1264         return rc;
1265 }
1266
1267 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1268                                 int pages, unsigned long *blocks,
1269                                 int *created, int create,
1270                                 cfs_semaphore_t *optional_sem)
1271 {
1272         int rc;
1273
1274         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1275                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1276                                                      blocks, created, create);
1277                 return rc;
1278         }
1279         if (optional_sem != NULL)
1280                 cfs_down(optional_sem);
1281         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1282                                             created, create);
1283         if (optional_sem != NULL)
1284                 cfs_up(optional_sem);
1285
1286         return rc;
1287 }
1288
1289 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1290 {
1291         unsigned long block;
1292         struct buffer_head *bh;
1293         int err, blocksize, csize, boffs, osize = size;
1294
1295         /* prevent reading after eof */
1296         cfs_lock_kernel();
1297         if (i_size_read(inode) < *offs + size) {
1298                 size = i_size_read(inode) - *offs;
1299                 cfs_unlock_kernel();
1300                 if (size < 0) {
1301                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1302                                i_size_read(inode), *offs);
1303                         return -EBADR;
1304                 } else if (size == 0) {
1305                         return 0;
1306                 }
1307         } else {
1308                 cfs_unlock_kernel();
1309         }
1310
1311         blocksize = 1 << inode->i_blkbits;
1312
1313         while (size > 0) {
1314                 block = *offs >> inode->i_blkbits;
1315                 boffs = *offs & (blocksize - 1);
1316                 csize = min(blocksize - boffs, size);
1317                 bh = ext3_bread(NULL, inode, block, 0, &err);
1318                 if (!bh) {
1319                         CERROR("can't read block: %d\n", err);
1320                         return err;
1321                 }
1322
1323                 memcpy(buf, bh->b_data + boffs, csize);
1324                 brelse(bh);
1325
1326                 *offs += csize;
1327                 buf += csize;
1328                 size -= csize;
1329         }
1330         return osize;
1331 }
1332 EXPORT_SYMBOL(fsfilt_ext3_read);
1333
1334 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1335                                    int size, loff_t *offs)
1336 {
1337         int rc;
1338         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1339         if (rc > 0)
1340                 rc = 0;
1341         return rc;
1342 }
1343
1344 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1345                                 loff_t *offs, handle_t *handle)
1346 {
1347         struct buffer_head *bh = NULL;
1348         loff_t old_size = i_size_read(inode), offset = *offs;
1349         loff_t new_size = i_size_read(inode);
1350         unsigned long block;
1351         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1352
1353         while (bufsize > 0) {
1354                 if (bh != NULL)
1355                         brelse(bh);
1356
1357                 block = offset >> inode->i_blkbits;
1358                 boffs = offset & (blocksize - 1);
1359                 size = min(blocksize - boffs, bufsize);
1360                 bh = ext3_bread(handle, inode, block, 1, &err);
1361                 if (!bh) {
1362                         CERROR("can't read/create block: %d\n", err);
1363                         break;
1364                 }
1365
1366                 err = ext3_journal_get_write_access(handle, bh);
1367                 if (err) {
1368                         CERROR("journal_get_write_access() returned error %d\n",
1369                                err);
1370                         break;
1371                 }
1372                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1373                 memcpy(bh->b_data + boffs, buf, size);
1374                 err = ext3_journal_dirty_metadata(handle, bh);
1375                 if (err) {
1376                         CERROR("journal_dirty_metadata() returned error %d\n",
1377                                err);
1378                         break;
1379                 }
1380                 if (offset + size > new_size)
1381                         new_size = offset + size;
1382                 offset += size;
1383                 bufsize -= size;
1384                 buf += size;
1385         }
1386         if (bh)
1387                 brelse(bh);
1388
1389         /* correct in-core and on-disk sizes */
1390         if (new_size > i_size_read(inode)) {
1391                 cfs_lock_kernel();
1392                 if (new_size > i_size_read(inode))
1393                         i_size_write(inode, new_size);
1394                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1395                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1396                 if (i_size_read(inode) > old_size)
1397                         mark_inode_dirty(inode);
1398                 cfs_unlock_kernel();
1399         }
1400
1401         if (err == 0)
1402                 *offs = offset;
1403         return err;
1404 }
1405 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1406
1407 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1408                                     loff_t *offs, int force_sync)
1409 {
1410         struct inode *inode = file->f_dentry->d_inode;
1411         handle_t *handle;
1412         int err, block_count = 0, blocksize;
1413
1414         /* Determine how many transaction credits are needed */
1415         blocksize = 1 << inode->i_blkbits;
1416         block_count = (*offs & (blocksize - 1)) + bufsize;
1417         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1418
1419         handle = ext3_journal_start(inode,
1420                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1421         if (IS_ERR(handle)) {
1422                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1423                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1424                 return PTR_ERR(handle);
1425         }
1426
1427         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1428
1429         if (!err && force_sync)
1430                 handle->h_sync = 1; /* recovery likes this */
1431
1432         ext3_journal_stop(handle);
1433
1434         return err;
1435 }
1436
1437 static int fsfilt_ext3_setup(struct super_block *sb)
1438 {
1439 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1440      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1441         struct ext3_sb_info *sbi = EXT3_SB(sb);
1442 #if 0
1443         sbi->dx_lock = fsfilt_ext3_dx_lock;
1444         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1445 #endif
1446 #endif
1447 #ifdef S_PDIROPS
1448         CWARN("Enabling PDIROPS\n");
1449         set_opt(sbi->s_mount_opt, PDIROPS);
1450         sb->s_flags |= S_PDIROPS;
1451 #endif
1452         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1453                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1454 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1455         /* enable journaled quota support */
1456         /* kfreed in ext3_put_super() */
1457         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1458         if (!sbi->s_qf_names[USRQUOTA])
1459                 return -ENOMEM;
1460         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1461         if (!sbi->s_qf_names[GRPQUOTA]) {
1462                 kfree(sbi->s_qf_names[USRQUOTA]);
1463                 sbi->s_qf_names[USRQUOTA] = NULL;
1464                 return -ENOMEM;
1465         }
1466         sbi->s_jquota_fmt = QFMT_LUSTRE;
1467 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1468         set_opt(sbi->s_mount_opt, QUOTA);
1469 #endif
1470 #endif
1471         return 0;
1472 }
1473
1474 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1475    objects. Logs is number of logfiles to update */
1476 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1477 {
1478         if ( !fso ) {
1479                 switch(op) {
1480                 case FSFILT_OP_CREATE:
1481                                  /* directory leaf, index & indirect & EA*/
1482                         return 4 + 3 * logs;
1483                 case FSFILT_OP_UNLINK:
1484                         return 3 * logs;
1485                 }
1486         } else {
1487                 int i;
1488                 int needed = 0;
1489                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1490                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1491                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1492                 for (i = 0; i < op; i++, fso++) {
1493                         int nblocks = fso->fso_bufcnt * blockpp;
1494                         int ndindirect = min(nblocks, addrpp + 1);
1495                         int nindir = nblocks + ndindirect + 1;
1496
1497                         needed += nindir;
1498                 }
1499                 return needed + 3 * logs;
1500         }
1501
1502         return 0;
1503 }
1504
1505 #ifdef HAVE_QUOTA_SUPPORT
1506 #define DQINFO_COPY(out, in)                    \
1507 do {                                            \
1508         Q_COPY(out, in, dqi_bgrace);            \
1509         Q_COPY(out, in, dqi_igrace);            \
1510         Q_COPY(out, in, dqi_flags);             \
1511         Q_COPY(out, in, dqi_valid);             \
1512 } while (0)
1513
1514 #define DQBLK_COPY(out, in)                     \
1515 do {                                            \
1516         Q_COPY(out, in, dqb_bhardlimit);        \
1517         Q_COPY(out, in, dqb_bsoftlimit);        \
1518         Q_COPY(out, in, dqb_curspace);          \
1519         Q_COPY(out, in, dqb_ihardlimit);        \
1520         Q_COPY(out, in, dqb_isoftlimit);        \
1521         Q_COPY(out, in, dqb_curinodes);         \
1522         Q_COPY(out, in, dqb_btime);             \
1523         Q_COPY(out, in, dqb_itime);             \
1524         Q_COPY(out, in, dqb_valid);             \
1525 } while (0)
1526
1527 static int fsfilt_ext3_quotactl(struct super_block *sb,
1528                                 struct obd_quotactl *oqc)
1529 {
1530         int i, rc = 0, error = 0;
1531         const struct quotactl_ops *qcop;
1532         struct if_dqinfo *info;
1533         struct if_dqblk *dqblk;
1534         ENTRY;
1535
1536         if (!sb->s_qcop)
1537                 RETURN(-ENOSYS);
1538
1539         OBD_ALLOC_PTR(info);
1540         if (!info)
1541                 RETURN(-ENOMEM);
1542         OBD_ALLOC_PTR(dqblk);
1543         if (!dqblk) {
1544                 OBD_FREE_PTR(info);
1545                 RETURN(-ENOMEM);
1546         }
1547
1548         DQINFO_COPY(info, &oqc->qc_dqinfo);
1549         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1550
1551         qcop = sb->s_qcop;
1552         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1553                 for (i = 0; i < MAXQUOTAS; i++) {
1554                         if (!Q_TYPESET(oqc, i))
1555                                 continue;
1556
1557                         if (oqc->qc_cmd == Q_QUOTAON) {
1558                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1559
1560                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1561
1562                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1563                                                  name[i], 0);
1564                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1565                                 rc = ll_quota_off(sb, i, 0);
1566                         }
1567
1568                         if (rc == -EBUSY)
1569                                 error = rc;
1570                         else if (rc)
1571                                 GOTO(out, rc);
1572                 }
1573                 GOTO(out, rc ?: error);
1574         }
1575
1576         switch (oqc->qc_cmd) {
1577         case Q_GETOINFO:
1578         case Q_GETINFO:
1579                 if (!qcop->get_info)
1580                         GOTO(out, rc = -ENOSYS);
1581                 rc = qcop->get_info(sb, oqc->qc_type, info);
1582                 break;
1583         case Q_SETQUOTA:
1584         case Q_INITQUOTA:
1585                 if (!qcop->set_dqblk)
1586                         GOTO(out, rc = -ENOSYS);
1587                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1588                 break;
1589         case Q_GETOQUOTA:
1590         case Q_GETQUOTA:
1591                 if (!qcop->get_dqblk)
1592                         GOTO(out, rc = -ENOSYS);
1593                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1594                 if (!rc)
1595                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1596                 break;
1597         case Q_SYNC:
1598                 if (!sb->s_qcop->quota_sync)
1599                         GOTO(out, rc = -ENOSYS);
1600                 qcop->quota_sync(sb, oqc->qc_type);
1601                 break;
1602         case Q_FINVALIDATE:
1603                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1604                 for (i = 0; i < MAXQUOTAS; i++) {
1605                         struct file *fp;
1606                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1607
1608                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1609
1610                         if (!Q_TYPESET(oqc, i))
1611                                 continue;
1612
1613                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1614                         if (IS_ERR(fp)) {
1615                                 rc = PTR_ERR(fp);
1616                                 CERROR("error invalidating operational quota file"
1617                                        " %s (rc:%d)\n", name[i], rc);
1618                         } else {
1619                                 filp_close(fp, 0);
1620                         }
1621
1622                 }
1623                 break;
1624         default:
1625                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1626                 LBUG();
1627         }
1628 out:
1629         DQINFO_COPY(&oqc->qc_dqinfo, info);
1630         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1631
1632         OBD_FREE_PTR(info);
1633         OBD_FREE_PTR(dqblk);
1634
1635         if (rc)
1636                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1637                                 "failed: %d\n",
1638                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1639         RETURN(rc);
1640 }
1641
1642 struct chk_dqblk{
1643         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1644         cfs_list_t              dqb_list;        /** in list also */
1645         qid_t                   dqb_id;          /** uid/gid */
1646         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1647         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1648         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1649         qsize_t                 dqb_curspace;    /** current space */
1650         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1651         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1652         qsize_t                 dqb_curinodes;   /** current inodes */
1653         obd_time                dqb_btime;       /** block grace time */
1654         obd_time                dqb_itime;       /** inode grace time */
1655         __u32                   dqb_valid;       /** flag for above fields */
1656 };
1657
1658 static inline unsigned int chkquot_hash(qid_t id, int type)
1659                                         __attribute__((__const__));
1660
1661 static inline unsigned int chkquot_hash(qid_t id, int type)
1662 {
1663         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1664 }
1665
1666 static inline struct chk_dqblk *
1667 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1668 {
1669         cfs_hlist_node_t *node;
1670         struct chk_dqblk *cdqb;
1671
1672         cfs_hlist_for_each(node, head) {
1673                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1674                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1675                         return cdqb;
1676         }
1677
1678         return NULL;
1679 }
1680
1681 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1682 {
1683         struct chk_dqblk *cdqb;
1684
1685         OBD_ALLOC_PTR(cdqb);
1686         if (cdqb) {
1687                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1688                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1689                 cdqb->dqb_id = id;
1690                 cdqb->dqb_type = type;
1691         }
1692
1693         return cdqb;
1694 }
1695
1696 static struct chk_dqblk *
1697 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1698       cfs_list_t *list, qid_t id, int type, int first_check)
1699 {
1700         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1701         struct if_dqblk dqb;
1702         struct chk_dqblk *cdqb;
1703         int rc;
1704
1705         cdqb = find_chkquot(head, id, type);
1706         if (cdqb)
1707                 return cdqb;
1708
1709         cdqb = alloc_chkquot(id, type);
1710         if (!cdqb)
1711                 return NULL;
1712
1713         if (!first_check) {
1714                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1715                 if (rc) {
1716                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1717                                id, type, rc);
1718                 } else {
1719                         DQBLK_COPY(cdqb, &dqb);
1720                         cdqb->dqb_curspace = 0;
1721                         cdqb->dqb_curinodes = 0;
1722                 }
1723         }
1724
1725         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1726         cfs_list_add_tail(&cdqb->dqb_list, list);
1727
1728         return cdqb;
1729 }
1730
1731 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1732 {
1733         struct obd_quotactl *oqctl;
1734         int rc;
1735
1736         OBD_ALLOC_PTR(oqctl);
1737         if (!oqctl)
1738                 RETURN(-ENOMEM);
1739
1740         oqctl->qc_cmd = cmd;
1741         oqctl->qc_id = qfmt;
1742         oqctl->qc_type = type;
1743         rc = fsfilt_ext3_quotactl(sb, oqctl);
1744
1745         OBD_FREE_PTR(oqctl);
1746         return rc;
1747 }
1748
1749 static inline int read_old_dqinfo(struct super_block *sb, int type,
1750                                   struct if_dqinfo *dqinfo)
1751 {
1752         struct obd_quotactl *oqctl;
1753         int rc;
1754         ENTRY;
1755
1756         OBD_ALLOC_PTR(oqctl);
1757         if (!oqctl)
1758                 RETURN(-ENOMEM);
1759
1760         oqctl->qc_cmd = Q_GETINFO;
1761         oqctl->qc_type = type;
1762         rc = fsfilt_ext3_quotactl(sb, oqctl);
1763         if (!rc)
1764                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1765
1766         OBD_FREE_PTR(oqctl);
1767         RETURN(rc);
1768 }
1769
1770 #ifndef HAVE_EXT4_LDISKFS
1771 static inline struct ext3_group_desc *
1772 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1773 {
1774         unsigned long desc_block, desc;
1775         struct ext3_group_desc *gdp;
1776
1777         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1778         desc = group % EXT3_DESC_PER_BLOCK(sb);
1779         gdp = (struct ext3_group_desc *)
1780               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1781
1782         return gdp + desc;
1783 }
1784
1785 static inline struct buffer_head *
1786 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1787 {
1788         struct ext3_group_desc *desc;
1789         struct buffer_head *bh;
1790
1791         desc = get_group_desc(sb, group, NULL);
1792         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1793         return bh;
1794 }
1795
1796 static __u32 ext3_itable_unused_count(struct super_block *sb,
1797                                struct ext3_group_desc *bg) {
1798        return le16_to_cpu(bg->bg_itable_unused);
1799 }
1800 #else
1801 #define get_group_desc ext3_get_group_desc
1802 #endif
1803
1804 struct qchk_ctxt {
1805         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1806         cfs_list_t              qckt_list;                 /* quotacheck list */
1807         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1808         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1809 };
1810
1811 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1812                            struct obd_quotactl *oqc)
1813 {
1814         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1815         loff_t size = 0;
1816         qid_t qid[MAXQUOTAS];
1817         int cnt, i, rc = 0;
1818
1819         if (!inode)
1820                 return 0;
1821
1822         qid[USRQUOTA] = inode->i_uid;
1823         qid[GRPQUOTA] = inode->i_gid;
1824
1825         if (S_ISDIR(inode->i_mode) ||
1826             S_ISREG(inode->i_mode) ||
1827             S_ISLNK(inode->i_mode))
1828                 size = inode_get_bytes(inode);
1829
1830         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1831                 if (!Q_TYPESET(oqc, cnt))
1832                         continue;
1833
1834                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1835                                 &qctxt->qckt_list, qid[cnt], cnt,
1836                                 qctxt->qckt_first_check[cnt]);
1837                 if (!cdqb[cnt]) {
1838                         rc = -ENOMEM;
1839                         break;
1840                 }
1841
1842                 cdqb[cnt]->dqb_curspace += size;
1843                 cdqb[cnt]->dqb_curinodes++;
1844         }
1845
1846         if (rc) {
1847                 for (i = 0; i < cnt; i++) {
1848                         if (!Q_TYPESET(oqc, i))
1849                                 continue;
1850                         LASSERT(cdqb[i]);
1851                         cdqb[i]->dqb_curspace -= size;
1852                         cdqb[i]->dqb_curinodes--;
1853                 }
1854         }
1855
1856         return rc;
1857 }
1858
1859 /* write dqinfo struct in a new quota file */
1860 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1861 {
1862         struct v2_disk_dqinfo dqinfo;
1863         __u32 blocks = V2_DQTREEOFF + 1;
1864         loff_t offset = V2_DQINFOOFF;
1865
1866         if (info) {
1867                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1868                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1869                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1870                                                ~DQF_INFO_DIRTY);
1871         } else {
1872                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1873                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1874                 dqinfo.dqi_flags = 0;
1875         }
1876
1877         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1878         dqinfo.dqi_free_blk = 0;
1879         dqinfo.dqi_free_entry = 0;
1880
1881         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1882 }
1883
1884 static int v3_write_dqheader(struct file *f, int type)
1885 {
1886         static const __u32 quota_magics[] = V2_INITQMAGICS;
1887         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1888         struct v2_disk_dqheader dqhead;
1889         loff_t offset = 0;
1890
1891         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1892         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1893
1894         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1895         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1896
1897         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1898 }
1899
1900 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1901                                   struct obd_quotactl *oqc)
1902 {
1903         int i, rc = 0;
1904         ENTRY;
1905
1906         for (i = 0; i < MAXQUOTAS; i++) {
1907                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1908                                          NULL : &qctxt->qckt_dqinfo[i];
1909                 struct file *file;
1910                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1911
1912                 if (!Q_TYPESET(oqc, i))
1913                         continue;
1914
1915                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1916
1917                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1918                 if (IS_ERR(file)) {
1919                         rc = PTR_ERR(file);
1920                         CERROR("can't create %s file: rc = %d\n",
1921                                name[i], rc);
1922                         GOTO(out, rc);
1923                 }
1924
1925                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1926                         CERROR("file %s is not regular", name[i]);
1927                         filp_close(file, 0);
1928                         GOTO(out, rc = -EINVAL);
1929                 }
1930
1931                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1932
1933                 rc = v3_write_dqheader(file, i);
1934                 if (rc) {
1935                         filp_close(file, 0);
1936                         GOTO(out, rc);
1937                 }
1938
1939                 rc = v3_write_dqinfo(file, i, info);
1940                 filp_close(file, 0);
1941                 if (rc)
1942                         GOTO(out, rc);
1943         }
1944
1945 out:
1946         RETURN(rc);
1947 }
1948
1949
1950 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1951                           struct chk_dqblk *cdqb)
1952 {
1953         struct obd_quotactl *oqc;
1954         long now;
1955         int rc;
1956         ENTRY;
1957
1958         OBD_ALLOC_PTR(oqc);
1959         if (!oqc)
1960                 RETURN(-ENOMEM);
1961
1962         now = cfs_time_current_sec();
1963
1964         if (cdqb->dqb_bsoftlimit &&
1965             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1966             !cdqb->dqb_btime)
1967                 cdqb->dqb_btime =
1968                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1969
1970         if (cdqb->dqb_isoftlimit &&
1971             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1972             !cdqb->dqb_itime)
1973                 cdqb->dqb_itime =
1974                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1975
1976         cdqb->dqb_valid = QIF_ALL;
1977
1978         oqc->qc_cmd = Q_SETQUOTA;
1979         oqc->qc_type = cdqb->dqb_type;
1980         oqc->qc_id = cdqb->dqb_id;
1981         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1982
1983         rc = fsfilt_ext3_quotactl(sb, oqc);
1984         OBD_FREE_PTR(oqc);
1985         RETURN(rc);
1986 }
1987
1988 static int prune_chkquots(struct super_block *sb,
1989                           struct qchk_ctxt *qctxt, int error)
1990 {
1991         struct chk_dqblk *cdqb, *tmp;
1992         int rc;
1993
1994         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1995                 if (!error) {
1996                         rc = commit_chkquot(sb, qctxt, cdqb);
1997                         if (rc)
1998                                 error = rc;
1999                 }
2000                 cfs_hlist_del_init(&cdqb->dqb_hash);
2001                 cfs_list_del(&cdqb->dqb_list);
2002                 OBD_FREE_PTR(cdqb);
2003         }
2004
2005         return error;
2006 }
2007
2008 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2009 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2010 #endif
2011
2012 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2013                                   struct obd_quotactl *oqc)
2014 {
2015         struct ext3_sb_info *sbi = EXT3_SB(sb);
2016         int i, group, uninit_feat = 0;
2017         struct qchk_ctxt *qctxt;
2018         struct buffer_head *bitmap_bh = NULL;
2019         unsigned long ino, inode_inuse;
2020         struct inode *inode;
2021         int rc = 0;
2022         ENTRY;
2023
2024         /* turn on quota and read dqinfo if existed */
2025         OBD_ALLOC_PTR(qctxt);
2026         if (!qctxt) {
2027                 oqc->qc_stat = -ENOMEM;
2028                 RETURN(-ENOMEM);
2029         }
2030
2031         for (i = 0; i < NR_DQHASH; i++)
2032                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2033         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2034
2035         for (i = 0; i < MAXQUOTAS; i++) {
2036                 if (!Q_TYPESET(oqc, i))
2037                         continue;
2038
2039                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2040                 if (!rc || rc == -EBUSY) {
2041                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2042                         if (rc)
2043                                 GOTO(out, rc);
2044                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2045                         qctxt->qckt_first_check[i] = 1;
2046                 } else if (rc) {
2047                         GOTO(out, rc);
2048                 }
2049         }
2050         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2051                 /* This filesystem supports the uninit group feature */
2052                 uninit_feat = 1;
2053
2054         /* number of inodes that have been allocated */
2055         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2056                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2057
2058         /* check quota and update in hash */
2059         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2060              group++) {
2061                 unsigned long used_count = sbi->s_inodes_per_group;
2062
2063                 if (uninit_feat) {
2064                         struct ext3_group_desc *desc;
2065                         desc = get_group_desc(sb, group, NULL);
2066                         if (!desc)
2067                                 GOTO(out, -EIO);
2068
2069                         /* we don't really need to take the group lock here,
2070                          * but it may be useful if one day we support online
2071                          * quotacheck */
2072 #ifdef HAVE_EXT4_LDISKFS
2073                         ext4_lock_group(sb, group);
2074 #else
2075                         spin_lock(sb_bgl_lock(sbi, group));
2076 #endif
2077                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2078                                 /* no inode in use in this group, just skip it */
2079 #ifdef HAVE_EXT4_LDISKFS
2080                                 ext3_unlock_group(sb, group);
2081 #else
2082                                 spin_unlock(sb_bgl_lock(sbi, group));
2083 #endif
2084                                 continue;
2085                         }
2086
2087                         used_count -= ext3_itable_unused_count(sb, desc);
2088 #ifdef HAVE_EXT4_LDISKFS
2089                         ext3_unlock_group(sb, group);
2090 #else
2091                         spin_unlock(sb_bgl_lock(sbi, group));
2092 #endif
2093                 }
2094
2095                 ino = group * sbi->s_inodes_per_group + 1;
2096                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2097                 if (!bitmap_bh) {
2098                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2099                                sb->s_id, group);
2100                         GOTO(out, -EIO);
2101                 }
2102
2103                 i = 0;
2104                 while (i < used_count &&
2105                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2106                                                used_count, i)) < used_count) {
2107                         inode_inuse--;
2108                         i++;
2109                         ino = i + group * sbi->s_inodes_per_group;
2110                         if (ino < sbi->s_first_ino)
2111                                 continue;
2112 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2113                         inode = ext3_iget(sb, ino);
2114 #else
2115                         inode = iget(sb, ino);
2116 #endif
2117                         if (!inode || IS_ERR(inode))
2118                                 continue;
2119
2120                         rc = add_inode_quota(inode, qctxt, oqc);
2121                         iput(inode);
2122                         if (rc) {
2123                                 brelse(bitmap_bh);
2124                                 GOTO(out, rc);
2125                         }
2126                 }
2127                 brelse(bitmap_bh);
2128         }
2129
2130         /* read old quota limits from old quota file. (only for the user
2131          * has limits but hasn't file) */
2132 #ifdef HAVE_QUOTA_SUPPORT
2133         for (i = 0; i < MAXQUOTAS; i++) {
2134                 cfs_list_t id_list;
2135                 struct dquot_id *dqid, *tmp;
2136
2137                 if (!Q_TYPESET(oqc, i))
2138                         continue;
2139
2140                 if (qctxt->qckt_first_check[i])
2141                         continue;
2142
2143
2144                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2145                 CFS_INIT_LIST_HEAD(&id_list);
2146 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2147                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2148 #else
2149                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2150 #endif
2151                 if (rc)
2152                         CERROR("read old limits failed. (rc:%d)\n", rc);
2153
2154                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2155                         cfs_list_del_init(&dqid->di_link);
2156
2157                         if (!rc)
2158                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2159                                       dqid->di_id, i,
2160                                       qctxt->qckt_first_check[i]);
2161                         OBD_FREE_PTR(dqid);
2162                 }
2163         }
2164 #endif
2165         /* turn off quota cause we are to dump chk_dqblk to files */
2166         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2167
2168         rc = create_new_quota_files(qctxt, oqc);
2169         if (rc)
2170                 GOTO(out, rc);
2171
2172         /* we use vfs functions to set dqblk, so turn quota on */
2173         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2174 out:
2175         /* dump and free chk_dqblk */
2176         rc = prune_chkquots(sb, qctxt, rc);
2177         OBD_FREE_PTR(qctxt);
2178
2179         /* turn off quota, `lfs quotacheck` will turn on when all
2180          * nodes quotacheck finish. */
2181         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2182
2183         oqc->qc_stat = rc;
2184         if (rc)
2185                 CERROR("quotacheck failed: rc = %d\n", rc);
2186
2187         RETURN(rc);
2188 }
2189
2190 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2191                                  int cmd)
2192 {
2193         int rc = 0;
2194         ENTRY;
2195
2196         if (lqi->qi_files[type] == NULL) {
2197                 CERROR("operate qinfo before it's enabled!\n");
2198                 RETURN(-ESRCH);
2199         }
2200
2201         switch (cmd) {
2202         case QFILE_CHK:
2203                 rc = lustre_check_quota_file(lqi, type);
2204                 break;
2205         case QFILE_RD_INFO:
2206                 rc = lustre_read_quota_info(lqi, type);
2207                 break;
2208         case QFILE_WR_INFO:
2209                 rc = lustre_write_quota_info(lqi, type);
2210                 break;
2211         case QFILE_INIT_INFO:
2212                 rc = lustre_init_quota_info(lqi, type);
2213                 break;
2214         case QFILE_CONVERT:
2215                 rc = -ENOTSUPP;
2216                 CERROR("quota CONVERT command is not supported\n");
2217                 break;
2218         default:
2219                 rc = -ENOTSUPP;
2220                 CERROR("Unsupported admin quota file cmd %d\n"
2221                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2222                        cmd);
2223                 break;
2224         }
2225         RETURN(rc);
2226 }
2227
2228 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2229                             cfs_list_t *list)
2230 {
2231         return lustre_get_qids(file, inode, type, list);
2232 }
2233
2234 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2235 {
2236         int rc = 0;
2237         ENTRY;
2238
2239         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2240                 CERROR("operate dquot before it's enabled!\n");
2241                 RETURN(-ESRCH);
2242         }
2243
2244         switch (cmd) {
2245         case QFILE_RD_DQUOT:
2246                 rc = lustre_read_dquot(dquot);
2247                 break;
2248         case QFILE_WR_DQUOT:
2249                 if (dquot->dq_dqb.dqb_ihardlimit ||
2250                     dquot->dq_dqb.dqb_isoftlimit ||
2251                     dquot->dq_dqb.dqb_bhardlimit ||
2252                     dquot->dq_dqb.dqb_bsoftlimit)
2253                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2254                 else
2255                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2256
2257                 rc = lustre_commit_dquot(dquot);
2258                 if (rc >= 0)
2259                         rc = 0;
2260                 break;
2261         default:
2262                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2263                 LBUG();
2264                 break;
2265         }
2266         RETURN(rc);
2267 }
2268
2269 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2270                                 struct inode *inode, int frags)
2271 {
2272 #ifdef EXT3_EXT_HAS_NO_TREE
2273         struct ext3_ext_base *base = inode;
2274 #else
2275         struct ext3_extents_tree tree;
2276         struct ext3_ext_base *base = &tree;
2277
2278         ext3_init_tree_desc(base, inode);
2279 #endif
2280         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2281          * metablocks at maxium b=16542 */
2282         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2283         return 0;
2284 }
2285
2286 #endif
2287
2288 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2289 {
2290         return (EXT3_SB(sb)->journal_bdev);
2291 }
2292 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2293
2294 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2295                           char *buf, int count, loff_t pos)
2296 {
2297         loff_t p = pos;
2298         int rc;
2299
2300         if (!f && !inode) {
2301                 CERROR("lustre_read_quota failed for no quota file!\n");
2302                 libcfs_debug_dumpstack(NULL);
2303                 return -EINVAL;
2304         }
2305
2306         /* Support for both adm and op quota files must be provided */
2307         if (f) {
2308                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2309                 rc = rc < 0 ? rc : p - pos;
2310         } else {
2311                 struct super_block *sb = inode->i_sb;
2312                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2313         }
2314         return rc;
2315 }
2316
2317 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2318 {
2319         loff_t p = pos;
2320         int rc;
2321
2322         /* Only adm quota files are supported, op updates are handled by vfs */
2323         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2324         rc = rc < 0 ? rc : p - pos;
2325
2326         return rc;
2327 }
2328
2329 void *lustre_quota_journal_start(struct inode *inode, int delete)
2330 {
2331         handle_t *handle;
2332         unsigned block_count;
2333
2334         if (delete) {
2335                 /* each indirect block (+4) may become free, attaching to the
2336                  * header list of free blocks (+1); the data block (+1) may
2337                  * become a free block (+0) or a block with free dqentries (+0) */
2338                 block_count = (4 + 1) + 1;
2339                 handle = ext3_journal_start(inode,
2340                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2341         } else {
2342                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2343                  * freeblk reusage with a header update (+1); dqentry is either reused
2344                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2345                  * a new block allocation (+1) with a header update (+1)              */
2346                 block_count = (4 + 1) + 3;
2347                 handle = ext3_journal_start(inode,
2348                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2349
2350         }
2351
2352         return handle;
2353 }
2354
2355 void lustre_quota_journal_stop(void *handle)
2356 {
2357         ext3_journal_stop((handle_t *)handle);
2358 }
2359
2360 static int ll_decode_fh_accept(void *context, struct dentry *de)
2361 {
2362         return 1;
2363 }
2364
2365 #ifdef HAVE_EXPORTFS_DECODE_FH
2366 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2367          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2368                             acceptable, context)
2369 #else
2370 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2371          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2372                                      type, acceptable, context)
2373 # define FILEID_INO32_GEN 1
2374 extern struct export_operations export_op_default;
2375 #endif
2376
2377 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2378                                       struct fsfilt_fid *fid, int ignore_gen)
2379 {
2380         struct inode  *inode;
2381         struct dentry *result;
2382
2383         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2384                                        ll_decode_fh_accept, NULL);
2385         if (IS_ERR(result)) {
2386                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2387                        fid->ino, fid->gen, PTR_ERR(result));
2388                 return result;
2389         }
2390
2391         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2392                fid->ino, fid->gen);
2393         inode = result->d_inode;
2394         if (inode == NULL)
2395                 goto err_out;
2396
2397         if (inode->i_nlink == 0 &&
2398             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2399                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2400                               " ctime -- this may indicate disk "
2401                               "corruption (inode: %lu, link: %lu, "
2402                               "count: %d)\n", inode->i_ino,
2403                               (unsigned long)inode->i_nlink,
2404                               atomic_read(&inode->i_count));
2405                 goto err_out;
2406         }
2407         if (fid->gen && inode->i_generation != fid->gen) {
2408                 /* we didn't find the right inode.. */
2409                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2410                        "count: %d, generation %u/%u\n",
2411                        inode->i_ino, (unsigned long)inode->i_nlink,
2412                        atomic_read(&inode->i_count), inode->i_generation,
2413                        fid->gen);
2414                 goto err_out;
2415         }
2416
2417         return result;
2418 err_out:
2419         l_dput(result);
2420         return ERR_PTR(-ENOENT);
2421 }
2422
2423 static struct fsfilt_operations fsfilt_ext3_ops = {
2424         .fs_type                = "ext3",
2425         .fs_owner               = THIS_MODULE,
2426         .fs_getlabel            = fsfilt_ext3_get_label,
2427         .fs_setlabel            = fsfilt_ext3_set_label,
2428         .fs_uuid                = fsfilt_ext3_uuid,
2429         .fs_start               = fsfilt_ext3_start,
2430         .fs_brw_start           = fsfilt_ext3_brw_start,
2431         .fs_extend              = fsfilt_ext3_extend,
2432         .fs_commit              = fsfilt_ext3_commit,
2433         .fs_commit_async        = fsfilt_ext3_commit_async,
2434         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2435         .fs_setattr             = fsfilt_ext3_setattr,
2436         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2437         .fs_set_md              = fsfilt_ext3_set_md,
2438         .fs_get_md              = fsfilt_ext3_get_md,
2439         .fs_readpage            = fsfilt_ext3_readpage,
2440         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2441         .fs_statfs              = fsfilt_ext3_statfs,
2442         .fs_sync                = fsfilt_ext3_sync,
2443         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2444         .fs_write_record        = fsfilt_ext3_write_record,
2445         .fs_read_record         = fsfilt_ext3_read_record,
2446         .fs_setup               = fsfilt_ext3_setup,
2447         .fs_send_bio            = fsfilt_ext3_send_bio,
2448         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2449 #ifdef HAVE_DISK_INODE_VERSION
2450         .fs_get_version         = fsfilt_ext3_get_version,
2451         .fs_set_version         = fsfilt_ext3_set_version,
2452 #endif
2453 #ifdef HAVE_QUOTA_SUPPORT
2454         .fs_quotactl            = fsfilt_ext3_quotactl,
2455         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2456         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2457         .fs_qids                = fsfilt_ext3_qids,
2458         .fs_dquot               = fsfilt_ext3_dquot,
2459         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2460 #endif
2461         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2462         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2463 };
2464
2465 static int __init fsfilt_ext3_init(void)
2466 {
2467         int rc;
2468
2469         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2470                                          sizeof(struct fsfilt_cb_data), 0, 0);
2471         if (!fcb_cache) {
2472                 CERROR("error allocating fsfilt journal callback cache\n");
2473                 GOTO(out, rc = -ENOMEM);
2474         }
2475
2476         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2477
2478         if (rc) {
2479                 int err = cfs_mem_cache_destroy(fcb_cache);
2480                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2481         }
2482 out:
2483         return rc;
2484 }
2485
2486 static void __exit fsfilt_ext3_exit(void)
2487 {
2488         int rc;
2489
2490         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2491         rc = cfs_mem_cache_destroy(fcb_cache);
2492         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2493 }
2494
2495 module_init(fsfilt_ext3_init);
2496 module_exit(fsfilt_ext3_exit);
2497
2498 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2499 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2500 MODULE_LICENSE("GPL");