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  * Copyright (c) 2011 Whamcloud, Inc.
  37  */
  38 /*
  39  * This file is part of Lustre, http://www.lustre.org/
  40  * Lustre is a trademark of Sun Microsystems, Inc.
  41  *
  42  * lustre/lvfs/fsfilt_ext3.c
  43  *
  44  * Author: Andreas Dilger <adilger@clusterfs.com>
  45  */
  46
  47 #define DEBUG_SUBSYSTEM S_FILTER
  48
  49 #include <linux/init.h>
  50 #include <linux/module.h>
  51 #include <linux/fs.h>
  52 #include <linux/slab.h>
  53 #include <linux/pagemap.h>
  54 #include <linux/quotaops.h>
  55 #ifdef HAVE_LINUX_EXPORTFS_H
  56 #include <linux/exportfs.h>
  57 #endif
  58 #ifdef HAVE_EXT4_LDISKFS
  59 #include <ext4/ext4.h>
  60 #include <ext4/ext4_jbd2.h>
  61 #else
  62 #include <linux/jbd.h>
  63 #include <linux/ext3_fs.h>
  64 #include <linux/ext3_jbd.h>
  65 #endif
  66 #include <linux/version.h>
  67 #include <linux/bitops.h>
  68 #include <linux/quota.h>
  69 #ifdef HAVE_QUOTAIO_H
  70 # include <linux/quotaio_v2.h>
  71 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  72 # include <quota/quotaio_v2.h>
  73 # include <quota/quota_tree.h>
  74 # define V2_DQTREEOFF    QT_TREEOFF
  75 #elif defined(HAVE_FS_QUOTAIO_H)
  76 # include <quotaio_v2.h>
  77 # include <quota_tree.h>
  78 # define V2_DQTREEOFF    QT_TREEOFF
  79 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  80 #endif
  81
  82 #ifdef QFMT_VFS_V1
  83 #define QFMT_LUSTRE QFMT_VFS_V1
  84 #else
  85 #define QFMT_LUSTRE QFMT_VFS_V0
  86 #endif
  87
  88 #if defined(HAVE_EXT3_XATTR_H)
  89 # include <ext3/xattr.h>
  90 #elif !defined(EXT3_XATTR_INDEX_TRUSTED)
  91 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  92 /* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
  93 # define EXT3_XATTR_INDEX_TRUSTED        4
  94 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  95 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  96 #endif
  97
  98 #include <libcfs/libcfs.h>
  99 #include <lustre_fsfilt.h>
 100 #include <obd.h>
 101 #include <lustre_quota.h>
 102 #include <linux/lustre_compat25.h>
 103 #include <linux/lprocfs_status.h>
 104
 105 #ifdef HAVE_EXT4_LDISKFS
 106 #include <ext4/ext4_extents.h>
 107 #else
 108 #include <linux/ext3_extents.h>
 109 #endif
 110
 111 #include "lustre_quota_fmt.h"
 112
 113 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
 114 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
 115 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
 116 #else
 117 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 118 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 119 #endif
 120
 121 /* for kernels 2.6.18 and later */
 122 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 123
 124 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 125                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 126
 127 #define ext3_mb_discard_inode_preallocations(inode) \
 128                  ext3_discard_preallocations(inode)
 129
 130 #ifdef HAVE_EXT4_LDISKFS
 131 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 132 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 133 #else
 134 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 135 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 136 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 137 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 138 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 139 #endif
 140
 141 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 142 # define journal_callback ext4_journal_cb_entry
 143 # define fsfilt_journal_callback_set(handle, func, jcb) \
 144          ext4_journal_callback_add(handle, func, jcb)
 145 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
 146 # define fsfilt_journal_callback_set(handle, func, jcb) \
 147          jbd2_journal_callback_set(handle, func, jcb)
 148 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
 149 # define fsfilt_journal_callback_set(handle, func, jcb) \
 150          journal_callback_set(handle, func, jcb)
 151 #else
 152 # error missing journal commit callback
 153 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
 154
 155 static cfs_mem_cache_t *fcb_cache;
 156
 157 struct fsfilt_cb_data {
 158         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 159         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 160         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 161         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 162         void *cb_data;                  /* MDS/OST completion function data */
 163 };
 164
 165 #ifndef ext3_find_next_bit
 166 #define ext3_find_next_bit           ext2_find_next_bit
 167 #endif
 168
 169 #ifndef ext2_find_next_bit
 170 #ifdef __LITTLE_ENDIAN
 171 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 172 #else
 173 error "Need implementation of find_next_bit on big-endian systems"
 174 #endif  /* __LITTLE_ENDIAN */
 175 #endif  /* !ext2_find_next_le_bit */
 176
 177 static char *fsfilt_ext3_get_label(struct super_block *sb)
 178 {
 179         return EXT3_SB(sb)->s_es->s_volume_name;
 180 }
 181
 182 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 183 {
 184         /* see e.g. fsfilt_ext3_write_record() */
 185         journal_t *journal;
 186         handle_t *handle;
 187         int err;
 188
 189         journal = EXT3_SB(sb)->s_journal;
 190         handle = ext3_journal_start_sb(sb, 1);
 191         if (IS_ERR(handle)) {
 192                 CERROR("can't start transaction\n");
 193                 return(PTR_ERR(handle));
 194         }
 195
 196         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 197         if (err)
 198                 goto out;
 199
 200         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 201                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 202
 203         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 204
 205 out:
 206         ext3_journal_stop(handle);
 207
 208         return(err);
 209 }
 210
 211 static char *fsfilt_ext3_uuid(struct super_block *sb)
 212 {
 213         return EXT3_SB(sb)->s_es->s_uuid;
 214 }
 215
 216 #ifdef HAVE_DISK_INODE_VERSION
 217
 218 static __u64 get_i_version(struct inode *inode)
 219 {
 220         return EXT3_I(inode)->i_fs_version;
 221 }
 222
 223 static void set_i_version(struct inode *inode, __u64 new_version)
 224 {
 225         (EXT3_I(inode))->i_fs_version = new_version;
 226 }
 227
 228 /*
 229  * Get the 64-bit version for an inode.
 230  */
 231 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 232 {
 233         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 234                get_i_version(inode), inode->i_ino);
 235         return get_i_version(inode);
 236 }
 237
 238 /*
 239  * Set the 64-bit version and return the old version.
 240  */
 241 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 242 {
 243         __u64 old_version = get_i_version(inode);
 244
 245         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 246                new_version, old_version, inode->i_ino);
 247         set_i_version(inode, new_version);
 248         /* version is set after all inode operations are finished, so we should
 249          * mark it dirty here */
 250         inode->i_sb->s_op->dirty_inode(inode);
 251         return old_version;
 252 }
 253
 254 #endif
 255
 256 /*
 257  * We don't currently need any additional blocks for rmdir and
 258  * unlink transactions because we are storing the OST oa_id inside
 259  * the inode (which we will be changing anyways as part of this
 260  * transaction).
 261  */
 262 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 263                                int logs)
 264 {
 265         /* For updates to the last received file */
 266         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 267         journal_t *journal;
 268         void *handle;
 269
 270         if (current->journal_info) {
 271                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 272                        current->journal_info);
 273                 goto journal_start;
 274         }
 275
 276         switch(op) {
 277         case FSFILT_OP_RMDIR:
 278         case FSFILT_OP_UNLINK:
 279                 /* delete one file + create/update logs for each stripe */
 280                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 281                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 282                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 283                 break;
 284         case FSFILT_OP_RENAME:
 285                 /* modify additional directory */
 286                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 287                 /* no break */
 288         case FSFILT_OP_SYMLINK:
 289                 /* additional block + block bitmap + GDT for long symlink */
 290                 nblocks += 3;
 291                 /* no break */
 292         case FSFILT_OP_CREATE: {
 293 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 294                 static int warned;
 295                 if (!warned) {
 296                         if (!test_opt(inode->i_sb, EXTENTS)) {
 297                                 warned = 1;
 298                         } else if (((EXT3_I(inode)->i_flags &
 299                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 300                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 301                                 CWARN("extent-mapped directory found with "
 302                                       "ext3-based ldiskfs - contact "
 303                                       "http://bugs.whamcloud.com/\n");
 304                                 warned = 1;
 305                         }
 306                 }
 307 #endif
 308                 /* no break */
 309         }
 310         case FSFILT_OP_MKDIR:
 311         case FSFILT_OP_MKNOD:
 312                 /* modify one inode + block bitmap + GDT */
 313                 nblocks += 3;
 314                 /* no break */
 315         case FSFILT_OP_LINK:
 316                 /* modify parent directory */
 317                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 318                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 319                 /* create/update logs for each stripe */
 320                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 321                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 322                 break;
 323         case FSFILT_OP_SETATTR:
 324                 /* Setattr on inode */
 325                 nblocks += 1;
 326                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 327                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 328                 /* quota chown log for each stripe */
 329                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 330                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 331                 break;
 332         case FSFILT_OP_CANCEL_UNLINK:
 333                 /* blocks for log header bitmap update OR
 334                  * blocks for catalog header bitmap update + unlink of logs */
 335                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 336                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 337                 break;
 338         default: CERROR("unknown transaction start op %d\n", op);
 339                 LBUG();
 340         }
 341
 342         LASSERT(current->journal_info == desc_private);
 343         journal = EXT3_SB(inode->i_sb)->s_journal;
 344         if (nblocks > journal->j_max_transaction_buffers) {
 345                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 346                        nblocks, op, logs, journal->j_max_transaction_buffers);
 347                 nblocks = journal->j_max_transaction_buffers;
 348         }
 349
 350  journal_start:
 351         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 352         handle = ext3_journal_start(inode, nblocks);
 353
 354         if (!IS_ERR(handle))
 355                 LASSERT(current->journal_info == handle);
 356         else
 357                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 358                        op, nblocks, PTR_ERR(handle));
 359         return handle;
 360 }
 361
 362 /*
 363  * Calculate the number of buffer credits needed to write multiple pages in
 364  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 365  * doesn't have a nice API for calculating this sort of thing in advance.
 366  *
 367  * See comment above ext3_writepage_trans_blocks for details.  We assume
 368  * no data journaling is being done, but it does allow for all of the pages
 369  * being non-contiguous.  If we are guaranteed contiguous pages we could
 370  * reduce the number of (d)indirect blocks a lot.
 371  *
 372  * With N blocks per page and P pages, for each inode we have at most:
 373  * N*P indirect
 374  * min(N*P, blocksize/4 + 1) dindirect blocks
 375  * niocount tindirect
 376  *
 377  * For the entire filesystem, we have at most:
 378  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 379  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 380  * objcount inode blocks
 381  * 1 superblock
 382  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 383  *
 384  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 385  */
 386 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 387                                       int niocount, struct niobuf_local *nb)
 388 {
 389         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 390         __u64 next_indir;
 391         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 392         int nbitmaps = 0, ngdblocks;
 393         int needed = objcount + 1; /* inodes + superblock */
 394         int i, j;
 395
 396         for (i = 0, j = 0; i < objcount; i++, fso++) {
 397                 /* two or more dindirect blocks in case we cross boundary */
 398                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 399                                     nb[j].offset) >>
 400                                    sb->s_blocksize_bits) /
 401                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 402                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 403
 404                 /* leaf, indirect, tindirect blocks for first block */
 405                 nbitmaps += blockpp + 2;
 406
 407                 j += fso->fso_bufcnt;
 408         }
 409
 410         next_indir = nb[0].offset +
 411                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 412         for (i = 1; i < niocount; i++) {
 413                 if (nb[i].offset >= next_indir) {
 414                         nbitmaps++;     /* additional indirect */
 415                         next_indir = nb[i].offset +
 416                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 417                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 418                         nbitmaps++;     /* additional indirect */
 419                 }
 420                 nbitmaps += blockpp;    /* each leaf in different group? */
 421         }
 422
 423         ngdblocks = nbitmaps;
 424         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 425                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 426         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 427                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 428
 429         needed += nbitmaps + ngdblocks;
 430
 431         /* last_rcvd update */
 432         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 433
 434 #if defined(CONFIG_QUOTA)
 435         /* We assume that there will be 1 bit set in s_dquot.flags for each
 436          * quota file that is active.  This is at least true for now.
 437          */
 438         needed += hweight32(ll_sb_any_quota_active(sb)) *
 439                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 440 #endif
 441
 442         return needed;
 443 }
 444
 445 /* We have to start a huge journal transaction here to hold all of the
 446  * metadata for the pages being written here.  This is necessitated by
 447  * the fact that we do lots of prepare_write operations before we do
 448  * any of the matching commit_write operations, so even if we split
 449  * up to use "smaller" transactions none of them could complete until
 450  * all of them were opened.  By having a single journal transaction,
 451  * we eliminate duplicate reservations for common blocks like the
 452  * superblock and group descriptors or bitmaps.
 453  *
 454  * We will start the transaction here, but each prepare_write will
 455  * add a refcount to the transaction, and each commit_write will
 456  * remove a refcount.  The transaction will be closed when all of
 457  * the pages have been written.
 458  */
 459 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 460                                    int niocount, struct niobuf_local *nb,
 461                                    void *desc_private, int logs)
 462 {
 463         journal_t *journal;
 464         handle_t *handle;
 465         int needed;
 466         ENTRY;
 467
 468         LASSERT(current->journal_info == desc_private);
 469         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 470         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 471
 472         /* The number of blocks we could _possibly_ dirty can very large.
 473          * We reduce our request if it is absurd (and we couldn't get that
 474          * many credits for a single handle anyways).
 475          *
 476          * At some point we have to limit the size of I/Os sent at one time,
 477          * increase the size of the journal, or we have to calculate the
 478          * actual journal requirements more carefully by checking all of
 479          * the blocks instead of being maximally pessimistic.  It remains to
 480          * be seen if this is a real problem or not.
 481          */
 482         if (needed > journal->j_max_transaction_buffers) {
 483                 CERROR("want too many journal credits (%d) using %d instead\n",
 484                        needed, journal->j_max_transaction_buffers);
 485                 needed = journal->j_max_transaction_buffers;
 486         }
 487
 488         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 489         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 490         if (IS_ERR(handle)) {
 491                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 492                        PTR_ERR(handle));
 493         } else {
 494                 LASSERT(handle->h_buffer_credits >= needed);
 495                 LASSERT(current->journal_info == handle);
 496         }
 497
 498         RETURN(handle);
 499 }
 500
 501 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 502 {
 503        handle_t *handle = h;
 504
 505        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 506        if (nblocks == 0) {
 507                handle->h_buffer_credits = 0;
 508                CWARN("setting credits of handle %p to zero by request\n", h);
 509        }
 510
 511        if (handle->h_buffer_credits > nblocks)
 512                 return 0;
 513        if (ext3_journal_extend(handle, nblocks) == 0)
 514                 return 0;
 515
 516        ext3_mark_inode_dirty(handle, inode);
 517        return ext3_journal_restart(handle, nblocks);
 518 }
 519
 520 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 521 {
 522         int rc;
 523         handle_t *handle = h;
 524
 525         LASSERT(current->journal_info == handle);
 526         if (force_sync)
 527                 handle->h_sync = 1; /* recovery likes this */
 528
 529         rc = ext3_journal_stop(handle);
 530
 531         return rc;
 532 }
 533
 534 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 535                                     void **wait_handle)
 536 {
 537         unsigned long tid;
 538         transaction_t *transaction;
 539         handle_t *handle = h;
 540         journal_t *journal;
 541         int rc;
 542
 543         LASSERT(current->journal_info == handle);
 544
 545         transaction = handle->h_transaction;
 546         journal = transaction->t_journal;
 547         tid = transaction->t_tid;
 548         /* we don't want to be blocked */
 549         handle->h_sync = 0;
 550         rc = ext3_journal_stop(handle);
 551         if (rc) {
 552                 CERROR("error while stopping transaction: %d\n", rc);
 553                 return rc;
 554         }
 555         fsfilt_log_start_commit(journal, tid);
 556
 557         *wait_handle = (void *) tid;
 558         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 559         return 0;
 560 }
 561
 562 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 563 {
 564         journal_t *journal = EXT3_JOURNAL(inode);
 565         tid_t tid = (tid_t)(long)h;
 566
 567         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 568         if (unlikely(is_journal_aborted(journal)))
 569                 return -EIO;
 570
 571         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 572
 573         if (unlikely(is_journal_aborted(journal)))
 574                 return -EIO;
 575         return 0;
 576 }
 577
 578 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 579                                struct iattr *iattr, int do_trunc)
 580 {
 581         struct inode *inode = dentry->d_inode;
 582         int rc = 0;
 583
 584 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
 585         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
 586          * timestamps to appear to be in the far future, due old timestamp
 587          * being stored on disk as an unsigned value.  This fixes up any
 588          * bad values held by the client before storing them on disk,
 589          * and ensures any timestamp updates are correct.  LU-1042 */
 590         if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
 591                      !(iattr->ia_valid & ATTR_ATIME))) {
 592                 iattr->ia_valid |= ATTR_ATIME;
 593                 LTIME_S(iattr->ia_atime) = 0;
 594         }
 595         if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
 596                      !(iattr->ia_valid & ATTR_MTIME))) {
 597                 iattr->ia_valid |= ATTR_MTIME;
 598                 LTIME_S(iattr->ia_mtime) = 0;
 599         }
 600         if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
 601                       LTIME_S(inode->i_ctime) == 0) &&
 602                      !(iattr->ia_valid & ATTR_CTIME))) {
 603                 iattr->ia_valid |= ATTR_CTIME;
 604                 LTIME_S(iattr->ia_ctime) = 0;
 605         }
 606 #else
 607 #warning "remove old LU-221/LU-1042 workaround code"
 608 #endif
 609
 610         /* When initializating timestamps for new inodes, use the filesystem
 611          * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
 612          * that this is potentially an invalid inode.  Files with an old ctime
 613          * migrated to a newly-formatted OST with a newer s_mkfs_time will not
 614          * hit this check, since it is only for ctime == 0.  LU-1010/LU-1042 */
 615         if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
 616                 LTIME_S(iattr->ia_ctime) =
 617                         EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
 618
 619         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 620          * We are already writing the inode to disk as part of this
 621          * transaction and want to avoid a lot of extra inode writeout
 622          * later on. b=9828 */
 623         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 624                 /* ATTR_SIZE would invoke truncate: clear it */
 625                 iattr->ia_valid &= ~ATTR_SIZE;
 626                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 627                 i_size_write(inode, iattr->ia_size);
 628
 629                 if (iattr->ia_valid & ATTR_UID)
 630                         inode->i_uid = iattr->ia_uid;
 631                 if (iattr->ia_valid & ATTR_GID)
 632                         inode->i_gid = iattr->ia_gid;
 633                 if (iattr->ia_valid & ATTR_ATIME)
 634                         inode->i_atime = iattr->ia_atime;
 635                 if (iattr->ia_valid & ATTR_MTIME)
 636                         inode->i_mtime = iattr->ia_mtime;
 637                 if (iattr->ia_valid & ATTR_CTIME)
 638                         inode->i_ctime = iattr->ia_ctime;
 639                 if (iattr->ia_valid & ATTR_MODE) {
 640                         inode->i_mode = iattr->ia_mode;
 641
 642                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 643                             !cfs_capable(CFS_CAP_FSETID))
 644                                 inode->i_mode &= ~S_ISGID;
 645                 }
 646
 647                 inode->i_sb->s_op->dirty_inode(inode);
 648
 649                 goto out;
 650         }
 651
 652         /* Don't allow setattr to change file type */
 653         if (iattr->ia_valid & ATTR_MODE)
 654                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 655                                  (iattr->ia_mode & ~S_IFMT);
 656
 657         /* We set these flags on the client, but have already checked perms
 658          * so don't confuse inode_change_ok. */
 659         iattr->ia_valid &= ~TIMES_SET_FLAGS;
 660
 661         if (inode->i_op->setattr) {
 662                 rc = inode->i_op->setattr(dentry, iattr);
 663         } else {
 664 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
 665                 rc = inode_change_ok(inode, iattr);
 666                 if (!rc)
 667 #endif
 668                         rc = simple_setattr(dentry, iattr);
 669         }
 670
 671  out:
 672         RETURN(rc);
 673 }
 674
 675 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 676                                  unsigned int cmd, unsigned long arg)
 677 {
 678         int rc = 0;
 679         ENTRY;
 680
 681         /* FIXME: Can't do this because of nested transaction deadlock */
 682         if (cmd == EXT3_IOC_SETFLAGS) {
 683                 /* We can't enable data journaling on OST objects, because
 684                 * this forces the transaction to be closed in order to
 685                 * flush the journal, but the caller will already have a
 686                 * compound transaction open to update the last_rcvd file,
 687                 * and this thread would deadlock trying to set the flag. */
 688                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 689                         CERROR("can't set data journal flag on file\n");
 690                         RETURN(-EPERM);
 691                 }
 692                 /* Because the MDS does not see the EXTENTS_FL set on the
 693                  * OST objects, mask this flag into all set flags.  It is
 694                  * not legal to clear this flag in any case, so we are not
 695                  * changing the functionality by doing this.  b=22911 */
 696                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 697         }
 698
 699 #ifdef HAVE_EXT4_LDISKFS
 700         /* ext4_ioctl does not have a inode argument */
 701         if (inode->i_fop->unlocked_ioctl)
 702                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 703 #else
 704         if (inode->i_fop->ioctl)
 705                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 706 #endif
 707         else
 708                 RETURN(-ENOTTY);
 709
 710         RETURN(rc);
 711 }
 712
 713 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 714                               void *lmm, int lmm_size, const char *name)
 715 {
 716         int rc;
 717
 718         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 719
 720         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 721                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 722
 723
 724         if (rc && rc != -EROFS)
 725                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 726                        inode->i_ino, rc);
 727         return rc;
 728 }
 729
 730 /* Must be called with i_mutex held */
 731 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 732                               const char *name)
 733 {
 734         int rc;
 735
 736         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 737
 738         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 739                             name, lmm, lmm_size);
 740
 741         /* This gives us the MD size */
 742         if (lmm == NULL)
 743                 return (rc == -ENODATA) ? 0 : rc;
 744
 745         if (rc < 0) {
 746                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 747                        EXT3_XATTR_INDEX_TRUSTED, name,
 748                        inode->i_ino, rc);
 749                 memset(lmm, 0, lmm_size);
 750                 return (rc == -ENODATA) ? 0 : rc;
 751         }
 752
 753         return rc;
 754 }
 755
 756 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 757 {
 758         submit_bio(rw, bio);
 759         return 0;
 760 }
 761
 762 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 763                                     loff_t *off)
 764 {
 765         struct inode *inode = file->f_dentry->d_inode;
 766         int rc = 0;
 767
 768         if (S_ISREG(inode->i_mode))
 769                 rc = file->f_op->read(file, buf, count, off);
 770         else {
 771                 const int blkbits = inode->i_sb->s_blocksize_bits;
 772                 const int blksize = inode->i_sb->s_blocksize;
 773
 774                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 775                        (unsigned long)count, inode->i_ino, *off);
 776                 while (count > 0) {
 777                         struct buffer_head *bh;
 778
 779                         bh = NULL;
 780                         if (*off < i_size_read(inode)) {
 781                                 int err = 0;
 782
 783                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 784                                                 0, &err);
 785
 786                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 787
 788                                 if (bh) {
 789                                         memcpy(buf, bh->b_data, blksize);
 790                                         brelse(bh);
 791                                 } else if (err) {
 792                                         /* XXX in theory we should just fake
 793                                          * this buffer and continue like ext3,
 794                                          * especially if this is a partial read
 795                                          */
 796                                         CERROR("error read dir %lu+%llu: %d\n",
 797                                                inode->i_ino, *off, err);
 798                                         RETURN(err);
 799                                 }
 800                         }
 801                         if (!bh) {
 802                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 803
 804                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 805                                 memset(fake, 0, sizeof(*fake));
 806                                 fake->rec_len = cpu_to_le16(blksize);
 807                         }
 808                         count -= blksize;
 809                         buf += blksize;
 810                         *off += blksize;
 811                         rc += blksize;
 812                 }
 813         }
 814
 815         return rc;
 816 }
 817
 818 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 819 static void fsfilt_ext3_cb_func(struct super_block *sb,
 820                                 struct journal_callback *jcb, int error)
 821 #else
 822 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 823 #endif
 824 {
 825         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 826
 827         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 828
 829         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 830 }
 831
 832 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 833                                       void *handle, fsfilt_cb_t cb_func,
 834                                       void *cb_data)
 835 {
 836         struct fsfilt_cb_data *fcb;
 837
 838         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 839         if (fcb == NULL)
 840                 RETURN(-ENOMEM);
 841
 842         fcb->cb_func = cb_func;
 843         fcb->cb_obd = obd;
 844         fcb->cb_last_rcvd = last_rcvd;
 845         fcb->cb_data = cb_data;
 846
 847         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 848         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 849
 850         return 0;
 851 }
 852
 853 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 854 {
 855         struct kstatfs sfs;
 856         int rc;
 857
 858         memset(&sfs, 0, sizeof(sfs));
 859         rc = ll_do_statfs(sb, &sfs);
 860         statfs_pack(osfs, &sfs);
 861         return rc;
 862 }
 863
 864 static int fsfilt_ext3_sync(struct super_block *sb)
 865 {
 866         return ext3_force_commit(sb);
 867 }
 868
 869 #ifndef EXT3_EXTENTS_FL
 870 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 871 #endif
 872
 873 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 874 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 875 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 876 #else
 877 # ifdef HAVE_EXT4_LDISKFS
 878 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 879 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 880 # else
 881 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 882 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 883 # endif
 884 #endif
 885
 886 #ifndef EXT_ASSERT
 887 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 888 #endif
 889
 890 #ifdef EXT3_EXT_HAS_NO_TREE
 891 /* for kernels 2.6.18 and later */
 892 #ifdef HAVE_EXT4_LDISKFS
 893 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 894 #else
 895 #define EXT_GENERATION(inode)           ext_generation(inode)
 896 #endif
 897 #define ext3_ext_base                   inode
 898 #define ext3_ext_base2inode(inode)      (inode)
 899 #define EXT_DEPTH(inode)                ext_depth(inode)
 900 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 901                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 902 #else
 903 #define ext3_ext_base                   ext3_extents_tree
 904 #define ext3_ext_base2inode(tree)       (tree->inode)
 905 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 906                         ext3_ext_walk_space(tree, block, num, cb);
 907 #endif
 908
 909 #include <linux/lustre_version.h>
 910
 911 struct bpointers {
 912         unsigned long *blocks;
 913         int *created;
 914         unsigned long start;
 915         int num;
 916         int init_num;
 917         int create;
 918 };
 919
 920 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 921                                unsigned long block, int *aflags)
 922 {
 923         struct ext3_inode_info *ei = EXT3_I(inode);
 924         unsigned long bg_start;
 925         unsigned long colour;
 926         int depth;
 927
 928         if (path) {
 929                 struct ext3_extent *ex;
 930                 depth = path->p_depth;
 931
 932                 /* try to predict block placement */
 933                 if ((ex = path[depth].p_ext))
 934                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 935
 936                 /* it looks index is empty
 937                  * try to find starting from index itself */
 938                 if (path[depth].p_bh)
 939                         return path[depth].p_bh->b_blocknr;
 940         }
 941
 942         /* OK. use inode's group */
 943         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 944                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 945         colour = (current->pid % 16) *
 946                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 947         return bg_start + colour + block;
 948 }
 949
 950 #define ll_unmap_underlying_metadata(sb, blocknr) \
 951         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 952
 953 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 954 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 955                                 struct ext3_ext_path *path, unsigned long block,
 956                                 unsigned long *count, int *err)
 957 {
 958         unsigned long pblock, goal;
 959         int aflags = 0;
 960         struct inode *inode = ext3_ext_base2inode(base);
 961
 962         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 963         aflags |= 2; /* block have been already reserved */
 964         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 965         return pblock;
 966
 967 }
 968 #else
 969 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 970                                 struct ext3_ext_path *path, unsigned long block,
 971                                 unsigned long *count, int *err)
 972 {
 973         struct inode *inode = ext3_ext_base2inode(base);
 974         struct ext3_allocation_request ar;
 975         unsigned long pblock;
 976         int aflags;
 977
 978         /* find neighbour allocated blocks */
 979         ar.lleft = block;
 980         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 981         if (*err)
 982                 return 0;
 983         ar.lright = block;
 984         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 985         if (*err)
 986                 return 0;
 987
 988         /* allocate new block */
 989         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 990         ar.inode = inode;
 991         ar.logical = block;
 992         ar.len = *count;
 993         ar.flags = EXT3_MB_HINT_DATA;
 994         pblock = ext3_mb_new_blocks(handle, &ar, err);
 995         *count = ar.len;
 996         return pblock;
 997 }
 998 #endif
 999
1000 #ifdef EXT3_EXT_HAS_NO_TREE
1001 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1002                                   struct ext3_ext_path *path,
1003                                   struct ext3_ext_cache *cex,
1004 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1005                                    struct ext3_extent *ex,
1006 #endif
1007                                   void *cbdata)
1008 {
1009         struct bpointers *bp = cbdata;
1010 #else
1011 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1012                                   struct ext3_ext_path *path,
1013                                   struct ext3_ext_cache *cex
1014 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1015                                   , struct ext3_extent *ex
1016 #endif
1017                                  )
1018 {
1019         struct bpointers *bp = base->private;
1020 #endif
1021         struct inode *inode = ext3_ext_base2inode(base);
1022         struct ext3_extent nex;
1023         unsigned long pblock;
1024         unsigned long tgen;
1025         int err, i;
1026         unsigned long count;
1027         handle_t *handle;
1028
1029         i = EXT_DEPTH(base);
1030         EXT_ASSERT(path[i].p_hdr);
1031         LASSERTF(i == path->p_depth ||
1032                  EXT_GENERATION(base) != path[0].p_generation,
1033                  "base vs path extent depth:%d != %d, generation:%lu == %lu\n",
1034                  i, path->p_depth, EXT_GENERATION(base), path[0].p_generation);
1035
1036         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1037                 err = EXT_CONTINUE;
1038                 goto map;
1039         }
1040
1041         if (bp->create == 0) {
1042                 i = 0;
1043                 if (cex->ec_block < bp->start)
1044                         i = bp->start - cex->ec_block;
1045                 if (i >= cex->ec_len)
1046                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1047                                         i, cex->ec_len);
1048                 for (; i < cex->ec_len && bp->num; i++) {
1049                         *(bp->created) = 0;
1050                         bp->created++;
1051                         *(bp->blocks) = 0;
1052                         bp->blocks++;
1053                         bp->num--;
1054                         bp->start++;
1055                 }
1056
1057                 return EXT_CONTINUE;
1058         }
1059
1060         tgen = EXT_GENERATION(base);
1061         count = ext3_ext_calc_credits_for_insert(base, path);
1062         fsfilt_up_truncate_sem(inode);
1063
1064         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1065         if (IS_ERR(handle)) {
1066                 fsfilt_down_truncate_sem(inode);
1067                 return PTR_ERR(handle);
1068         }
1069
1070         fsfilt_down_truncate_sem(inode);
1071         if (tgen != EXT_GENERATION(base)) {
1072                 /* the tree has changed. so path can be invalid at moment */
1073                 ext3_journal_stop(handle);
1074                 return EXT_REPEAT;
1075         }
1076
1077 #if defined(HAVE_EXT4_LDISKFS)
1078         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1079          * protected by i_data_sem as whole. so we patch it to store
1080          * generation to path and now verify the tree hasn't changed */
1081         down_write((&EXT4_I(inode)->i_data_sem));
1082
1083         /* validate extent, make sure the extent tree does not changed */
1084         if (EXT_GENERATION(base) != path[0].p_generation) {
1085                 /* cex is invalid, try again */
1086                 up_write(&EXT4_I(inode)->i_data_sem);
1087                 ext3_journal_stop(handle);
1088                 return EXT_REPEAT;
1089         }
1090 #endif
1091
1092         count = cex->ec_len;
1093         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1094         if (!pblock)
1095                 goto out;
1096         EXT_ASSERT(count <= cex->ec_len);
1097
1098         /* insert new extent */
1099         nex.ee_block = cpu_to_le32(cex->ec_block);
1100         ext3_ext_store_pblock(&nex, pblock);
1101         nex.ee_len = cpu_to_le16(count);
1102         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1103         if (err) {
1104                 /* free data blocks we just allocated */
1105                 /* not a good idea to call discard here directly,
1106                  * but otherwise we'd need to call it every free() */
1107 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1108                 ext3_mb_discard_inode_preallocations(inode);
1109 #endif
1110                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1111                                  cpu_to_le16(nex.ee_len), 0);
1112                 goto out;
1113         }
1114
1115         /*
1116          * Putting len of the actual extent we just inserted,
1117          * we are asking ext3_ext_walk_space() to continue
1118          * scaning after that block
1119          */
1120         cex->ec_len = le16_to_cpu(nex.ee_len);
1121         cex->ec_start = ext_pblock(&nex);
1122         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1123         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1124
1125 out:
1126 #if defined(HAVE_EXT4_LDISKFS)
1127         up_write((&EXT4_I(inode)->i_data_sem));
1128 #endif
1129         ext3_journal_stop(handle);
1130 map:
1131         if (err >= 0) {
1132                 /* map blocks */
1133                 if (bp->num == 0) {
1134                         CERROR("hmm. why do we find this extent?\n");
1135                         CERROR("initial space: %lu:%u\n",
1136                                 bp->start, bp->init_num);
1137                         CERROR("current extent: %u/%u/%llu %d\n",
1138                                 cex->ec_block, cex->ec_len,
1139                                 (unsigned long long)cex->ec_start,
1140                                 cex->ec_type);
1141                 }
1142                 i = 0;
1143                 if (cex->ec_block < bp->start)
1144                         i = bp->start - cex->ec_block;
1145                 if (i >= cex->ec_len)
1146                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1147                                         i, cex->ec_len);
1148                 for (; i < cex->ec_len && bp->num; i++) {
1149                         *(bp->blocks) = cex->ec_start + i;
1150                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1151                                 *(bp->created) = 0;
1152                         } else {
1153                                 *(bp->created) = 1;
1154                                 /* unmap any possible underlying metadata from
1155                                  * the block device mapping.  bug 6998. */
1156                                 ll_unmap_underlying_metadata(inode->i_sb,
1157                                                              *(bp->blocks));
1158                         }
1159                         bp->created++;
1160                         bp->blocks++;
1161                         bp->num--;
1162                         bp->start++;
1163                 }
1164         }
1165         return err;
1166 }
1167
1168 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1169                        unsigned long num, unsigned long *blocks,
1170                        int *created, int create)
1171 {
1172 #ifdef EXT3_EXT_HAS_NO_TREE
1173         struct ext3_ext_base *base = inode;
1174 #else
1175         struct ext3_extents_tree tree;
1176         struct ext3_ext_base *base = &tree;
1177 #endif
1178         struct bpointers bp;
1179         int err;
1180
1181         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1182                block, block + num - 1, (unsigned) inode->i_ino);
1183
1184 #ifndef EXT3_EXT_HAS_NO_TREE
1185         ext3_init_tree_desc(base, inode);
1186         tree.private = &bp;
1187 #endif
1188         bp.blocks = blocks;
1189         bp.created = created;
1190         bp.start = block;
1191         bp.init_num = bp.num = num;
1192         bp.create = create;
1193
1194         fsfilt_down_truncate_sem(inode);
1195         err = fsfilt_ext3_ext_walk_space(base, block, num,
1196                                          ext3_ext_new_extent_cb, &bp);
1197         ext3_ext_invalidate_cache(base);
1198         fsfilt_up_truncate_sem(inode);
1199
1200         return err;
1201 }
1202
1203 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1204                                     int pages, unsigned long *blocks,
1205                                     int *created, int create)
1206 {
1207         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1208         int rc = 0, i = 0;
1209         struct page *fp = NULL;
1210         int clen = 0;
1211
1212         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1213                 inode->i_ino, pages, (*page)->index);
1214
1215         /* pages are sorted already. so, we just have to find
1216          * contig. space and process them properly */
1217         while (i < pages) {
1218                 if (fp == NULL) {
1219                         /* start new extent */
1220                         fp = *page++;
1221                         clen = 1;
1222                         i++;
1223                         continue;
1224                 } else if (fp->index + clen == (*page)->index) {
1225                         /* continue the extent */
1226                         page++;
1227                         clen++;
1228                         i++;
1229                         continue;
1230                 }
1231
1232                 /* process found extent */
1233                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1234                                         clen * blocks_per_page, blocks,
1235                                         created, create);
1236                 if (rc)
1237                         GOTO(cleanup, rc);
1238
1239                 /* look for next extent */
1240                 fp = NULL;
1241                 blocks += blocks_per_page * clen;
1242                 created += blocks_per_page * clen;
1243         }
1244
1245         if (fp)
1246                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1247                                         clen * blocks_per_page, blocks,
1248                                         created, create);
1249 cleanup:
1250         return rc;
1251 }
1252
1253 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1254                                unsigned long *blocks, int *created, int create);
1255 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1256                                    int pages, unsigned long *blocks,
1257                                    int *created, int create)
1258 {
1259         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1260         unsigned long *b;
1261         int rc = 0, i, *cr;
1262
1263         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1264                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1265                 if (rc) {
1266                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1267                                inode->i_ino, *b, *cr, create, rc);
1268                         break;
1269                 }
1270
1271                 b += blocks_per_page;
1272                 cr += blocks_per_page;
1273         }
1274         return rc;
1275 }
1276
1277 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1278                                 int pages, unsigned long *blocks,
1279                                 int *created, int create,
1280                                 cfs_semaphore_t *optional_sem)
1281 {
1282         int rc;
1283
1284         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1285                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1286                                                      blocks, created, create);
1287                 return rc;
1288         }
1289         if (optional_sem != NULL)
1290                 cfs_down(optional_sem);
1291         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1292                                             created, create);
1293         if (optional_sem != NULL)
1294                 cfs_up(optional_sem);
1295
1296         return rc;
1297 }
1298
1299 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1300 {
1301         unsigned long block;
1302         struct buffer_head *bh;
1303         int err, blocksize, csize, boffs, osize = size;
1304
1305         /* prevent reading after eof */
1306         cfs_lock_kernel();
1307         if (i_size_read(inode) < *offs + size) {
1308                 size = i_size_read(inode) - *offs;
1309                 cfs_unlock_kernel();
1310                 if (size < 0) {
1311                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1312                                i_size_read(inode), *offs);
1313                         return -EBADR;
1314                 } else if (size == 0) {
1315                         return 0;
1316                 }
1317         } else {
1318                 cfs_unlock_kernel();
1319         }
1320
1321         blocksize = 1 << inode->i_blkbits;
1322
1323         while (size > 0) {
1324                 block = *offs >> inode->i_blkbits;
1325                 boffs = *offs & (blocksize - 1);
1326                 csize = min(blocksize - boffs, size);
1327                 bh = ext3_bread(NULL, inode, block, 0, &err);
1328                 if (!bh) {
1329                         CERROR("can't read block: %d\n", err);
1330                         return err;
1331                 }
1332
1333                 memcpy(buf, bh->b_data + boffs, csize);
1334                 brelse(bh);
1335
1336                 *offs += csize;
1337                 buf += csize;
1338                 size -= csize;
1339         }
1340         return osize;
1341 }
1342 EXPORT_SYMBOL(fsfilt_ext3_read);
1343
1344 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1345                                    int size, loff_t *offs)
1346 {
1347         int rc;
1348         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1349         if (rc > 0)
1350                 rc = 0;
1351         return rc;
1352 }
1353
1354 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1355                                 loff_t *offs, handle_t *handle)
1356 {
1357         struct buffer_head *bh = NULL;
1358         loff_t old_size = i_size_read(inode), offset = *offs;
1359         loff_t new_size = i_size_read(inode);
1360         unsigned long block;
1361         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1362
1363         while (bufsize > 0) {
1364                 if (bh != NULL)
1365                         brelse(bh);
1366
1367                 block = offset >> inode->i_blkbits;
1368                 boffs = offset & (blocksize - 1);
1369                 size = min(blocksize - boffs, bufsize);
1370                 bh = ext3_bread(handle, inode, block, 1, &err);
1371                 if (!bh) {
1372                         CERROR("can't read/create block: %d\n", err);
1373                         break;
1374                 }
1375
1376                 err = ext3_journal_get_write_access(handle, bh);
1377                 if (err) {
1378                         CERROR("journal_get_write_access() returned error %d\n",
1379                                err);
1380                         break;
1381                 }
1382                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1383                 memcpy(bh->b_data + boffs, buf, size);
1384                 err = ext3_journal_dirty_metadata(handle, bh);
1385                 if (err) {
1386                         CERROR("journal_dirty_metadata() returned error %d\n",
1387                                err);
1388                         break;
1389                 }
1390                 if (offset + size > new_size)
1391                         new_size = offset + size;
1392                 offset += size;
1393                 bufsize -= size;
1394                 buf += size;
1395         }
1396         if (bh)
1397                 brelse(bh);
1398
1399         /* correct in-core and on-disk sizes */
1400         if (new_size > i_size_read(inode)) {
1401                 cfs_lock_kernel();
1402                 if (new_size > i_size_read(inode))
1403                         i_size_write(inode, new_size);
1404                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1405                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1406                 if (i_size_read(inode) > old_size)
1407                         mark_inode_dirty(inode);
1408                 cfs_unlock_kernel();
1409         }
1410
1411         if (err == 0)
1412                 *offs = offset;
1413         return err;
1414 }
1415 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1416
1417 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1418                                     loff_t *offs, int force_sync)
1419 {
1420         struct inode *inode = file->f_dentry->d_inode;
1421         handle_t *handle;
1422         int err, block_count = 0, blocksize;
1423
1424         /* Determine how many transaction credits are needed */
1425         blocksize = 1 << inode->i_blkbits;
1426         block_count = (*offs & (blocksize - 1)) + bufsize;
1427         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1428
1429         handle = ext3_journal_start(inode,
1430                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1431         if (IS_ERR(handle)) {
1432                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1433                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1434                 return PTR_ERR(handle);
1435         }
1436
1437         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1438
1439         if (!err && force_sync)
1440                 handle->h_sync = 1; /* recovery likes this */
1441
1442         ext3_journal_stop(handle);
1443
1444         return err;
1445 }
1446
1447 static int fsfilt_ext3_setup(struct super_block *sb)
1448 {
1449 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1450      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1451         struct ext3_sb_info *sbi = EXT3_SB(sb);
1452 #if 0
1453         sbi->dx_lock = fsfilt_ext3_dx_lock;
1454         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1455 #endif
1456 #endif
1457 #ifdef S_PDIROPS
1458         CWARN("Enabling PDIROPS\n");
1459         set_opt(sbi->s_mount_opt, PDIROPS);
1460         sb->s_flags |= S_PDIROPS;
1461 #endif
1462         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1463                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1464 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1465         /* enable journaled quota support */
1466         /* kfreed in ext3_put_super() */
1467         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1468         if (!sbi->s_qf_names[USRQUOTA])
1469                 return -ENOMEM;
1470         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1471         if (!sbi->s_qf_names[GRPQUOTA]) {
1472                 kfree(sbi->s_qf_names[USRQUOTA]);
1473                 sbi->s_qf_names[USRQUOTA] = NULL;
1474                 return -ENOMEM;
1475         }
1476         sbi->s_jquota_fmt = QFMT_LUSTRE;
1477 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1478         set_opt(sbi->s_mount_opt, QUOTA);
1479 #endif
1480 #endif
1481         return 0;
1482 }
1483
1484 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1485    objects. Logs is number of logfiles to update */
1486 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1487 {
1488         if ( !fso ) {
1489                 switch(op) {
1490                 case FSFILT_OP_CREATE:
1491                                  /* directory leaf, index & indirect & EA*/
1492                         return 4 + 3 * logs;
1493                 case FSFILT_OP_UNLINK:
1494                         return 3 * logs;
1495                 }
1496         } else {
1497                 int i;
1498                 int needed = 0;
1499                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1500                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1501                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1502                 for (i = 0; i < op; i++, fso++) {
1503                         int nblocks = fso->fso_bufcnt * blockpp;
1504                         int ndindirect = min(nblocks, addrpp + 1);
1505                         int nindir = nblocks + ndindirect + 1;
1506
1507                         needed += nindir;
1508                 }
1509                 return needed + 3 * logs;
1510         }
1511
1512         return 0;
1513 }
1514
1515 #ifdef HAVE_QUOTA_SUPPORT
1516 #define DQINFO_COPY(out, in)                    \
1517 do {                                            \
1518         Q_COPY(out, in, dqi_bgrace);            \
1519         Q_COPY(out, in, dqi_igrace);            \
1520         Q_COPY(out, in, dqi_flags);             \
1521         Q_COPY(out, in, dqi_valid);             \
1522 } while (0)
1523
1524 #define DQBLK_COPY(out, in)                     \
1525 do {                                            \
1526         Q_COPY(out, in, dqb_bhardlimit);        \
1527         Q_COPY(out, in, dqb_bsoftlimit);        \
1528         Q_COPY(out, in, dqb_curspace);          \
1529         Q_COPY(out, in, dqb_ihardlimit);        \
1530         Q_COPY(out, in, dqb_isoftlimit);        \
1531         Q_COPY(out, in, dqb_curinodes);         \
1532         Q_COPY(out, in, dqb_btime);             \
1533         Q_COPY(out, in, dqb_itime);             \
1534         Q_COPY(out, in, dqb_valid);             \
1535 } while (0)
1536
1537 static int fsfilt_ext3_quotactl(struct super_block *sb,
1538                                 struct obd_quotactl *oqc)
1539 {
1540         int i, rc = 0, error = 0;
1541         const struct quotactl_ops *qcop;
1542         struct if_dqinfo *info;
1543         struct if_dqblk *dqblk;
1544         ENTRY;
1545
1546         if (!sb->s_qcop)
1547                 RETURN(-ENOSYS);
1548
1549         OBD_ALLOC_PTR(info);
1550         if (!info)
1551                 RETURN(-ENOMEM);
1552         OBD_ALLOC_PTR(dqblk);
1553         if (!dqblk) {
1554                 OBD_FREE_PTR(info);
1555                 RETURN(-ENOMEM);
1556         }
1557
1558         DQINFO_COPY(info, &oqc->qc_dqinfo);
1559         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1560
1561         qcop = sb->s_qcop;
1562         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1563                 for (i = 0; i < MAXQUOTAS; i++) {
1564                         if (!Q_TYPESET(oqc, i))
1565                                 continue;
1566
1567                         if (oqc->qc_cmd == Q_QUOTAON) {
1568                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1569
1570                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1571
1572                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1573                                                  name[i], 0);
1574                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1575                                 rc = ll_quota_off(sb, i, 0);
1576                         }
1577
1578                         if (rc == -EBUSY)
1579                                 error = rc;
1580                         else if (rc)
1581                                 GOTO(out, rc);
1582                 }
1583                 GOTO(out, rc ?: error);
1584         }
1585
1586         switch (oqc->qc_cmd) {
1587         case Q_GETOINFO:
1588         case Q_GETINFO:
1589                 if (!qcop->get_info)
1590                         GOTO(out, rc = -ENOSYS);
1591                 rc = qcop->get_info(sb, oqc->qc_type, info);
1592                 break;
1593         case Q_SETQUOTA:
1594         case Q_INITQUOTA:
1595                 if (!qcop->set_dqblk)
1596                         GOTO(out, rc = -ENOSYS);
1597                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1598                 break;
1599         case Q_GETOQUOTA:
1600         case Q_GETQUOTA:
1601                 if (!qcop->get_dqblk)
1602                         GOTO(out, rc = -ENOSYS);
1603                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1604                 if (!rc)
1605                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1606                 break;
1607         case Q_SYNC:
1608                 if (!sb->s_qcop->quota_sync)
1609                         GOTO(out, rc = -ENOSYS);
1610                 qcop->quota_sync(sb, oqc->qc_type);
1611                 break;
1612         case Q_FINVALIDATE:
1613                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1614                 for (i = 0; i < MAXQUOTAS; i++) {
1615                         struct file *fp;
1616                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1617
1618                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1619
1620                         if (!Q_TYPESET(oqc, i))
1621                                 continue;
1622
1623                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1624                         if (IS_ERR(fp)) {
1625                                 rc = PTR_ERR(fp);
1626                                 CERROR("error invalidating operational quota file"
1627                                        " %s (rc:%d)\n", name[i], rc);
1628                         } else {
1629                                 filp_close(fp, 0);
1630                         }
1631
1632                 }
1633                 break;
1634         default:
1635                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1636                 LBUG();
1637         }
1638 out:
1639         DQINFO_COPY(&oqc->qc_dqinfo, info);
1640         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1641
1642         OBD_FREE_PTR(info);
1643         OBD_FREE_PTR(dqblk);
1644
1645         if (rc)
1646                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1647                                 "failed: %d\n",
1648                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1649         RETURN(rc);
1650 }
1651
1652 struct chk_dqblk{
1653         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1654         cfs_list_t              dqb_list;        /** in list also */
1655         qid_t                   dqb_id;          /** uid/gid */
1656         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1657         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1658         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1659         qsize_t                 dqb_curspace;    /** current space */
1660         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1661         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1662         qsize_t                 dqb_curinodes;   /** current inodes */
1663         obd_time                dqb_btime;       /** block grace time */
1664         obd_time                dqb_itime;       /** inode grace time */
1665         __u32                   dqb_valid;       /** flag for above fields */
1666 };
1667
1668 static inline unsigned int chkquot_hash(qid_t id, int type)
1669                                         __attribute__((__const__));
1670
1671 static inline unsigned int chkquot_hash(qid_t id, int type)
1672 {
1673         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1674 }
1675
1676 static inline struct chk_dqblk *
1677 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1678 {
1679         cfs_hlist_node_t *node;
1680         struct chk_dqblk *cdqb;
1681
1682         cfs_hlist_for_each(node, head) {
1683                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1684                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1685                         return cdqb;
1686         }
1687
1688         return NULL;
1689 }
1690
1691 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1692 {
1693         struct chk_dqblk *cdqb;
1694
1695         OBD_ALLOC_PTR(cdqb);
1696         if (cdqb) {
1697                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1698                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1699                 cdqb->dqb_id = id;
1700                 cdqb->dqb_type = type;
1701         }
1702
1703         return cdqb;
1704 }
1705
1706 static struct chk_dqblk *
1707 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1708       cfs_list_t *list, qid_t id, int type, int first_check)
1709 {
1710         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1711         struct if_dqblk dqb;
1712         struct chk_dqblk *cdqb;
1713         int rc;
1714
1715         cdqb = find_chkquot(head, id, type);
1716         if (cdqb)
1717                 return cdqb;
1718
1719         cdqb = alloc_chkquot(id, type);
1720         if (!cdqb)
1721                 return NULL;
1722
1723         if (!first_check) {
1724                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1725                 if (rc) {
1726                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1727                                id, type, rc);
1728                 } else {
1729                         DQBLK_COPY(cdqb, &dqb);
1730                         cdqb->dqb_curspace = 0;
1731                         cdqb->dqb_curinodes = 0;
1732                 }
1733         }
1734
1735         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1736         cfs_list_add_tail(&cdqb->dqb_list, list);
1737
1738         return cdqb;
1739 }
1740
1741 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1742 {
1743         struct obd_quotactl *oqctl;
1744         int rc;
1745
1746         OBD_ALLOC_PTR(oqctl);
1747         if (!oqctl)
1748                 RETURN(-ENOMEM);
1749
1750         oqctl->qc_cmd = cmd;
1751         oqctl->qc_id = qfmt;
1752         oqctl->qc_type = type;
1753         rc = fsfilt_ext3_quotactl(sb, oqctl);
1754
1755         OBD_FREE_PTR(oqctl);
1756         return rc;
1757 }
1758
1759 static inline int read_old_dqinfo(struct super_block *sb, int type,
1760                                   struct if_dqinfo *dqinfo)
1761 {
1762         struct obd_quotactl *oqctl;
1763         int rc;
1764         ENTRY;
1765
1766         OBD_ALLOC_PTR(oqctl);
1767         if (!oqctl)
1768                 RETURN(-ENOMEM);
1769
1770         oqctl->qc_cmd = Q_GETINFO;
1771         oqctl->qc_type = type;
1772         rc = fsfilt_ext3_quotactl(sb, oqctl);
1773         if (!rc)
1774                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1775
1776         OBD_FREE_PTR(oqctl);
1777         RETURN(rc);
1778 }
1779
1780 #ifndef HAVE_EXT4_LDISKFS
1781 static inline struct ext3_group_desc *
1782 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1783 {
1784         unsigned long desc_block, desc;
1785         struct ext3_group_desc *gdp;
1786
1787         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1788         desc = group % EXT3_DESC_PER_BLOCK(sb);
1789         gdp = (struct ext3_group_desc *)
1790               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1791
1792         return gdp + desc;
1793 }
1794
1795 static inline struct buffer_head *
1796 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1797 {
1798         struct ext3_group_desc *desc;
1799         struct buffer_head *bh;
1800
1801         desc = get_group_desc(sb, group, NULL);
1802         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1803         return bh;
1804 }
1805
1806 static __u32 ext3_itable_unused_count(struct super_block *sb,
1807                                struct ext3_group_desc *bg) {
1808        return le16_to_cpu(bg->bg_itable_unused);
1809 }
1810 #else
1811 #define get_group_desc ext3_get_group_desc
1812 #endif
1813
1814 struct qchk_ctxt {
1815         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1816         cfs_list_t              qckt_list;                 /* quotacheck list */
1817         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1818         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1819 };
1820
1821 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1822                            struct obd_quotactl *oqc)
1823 {
1824         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1825         loff_t size = 0;
1826         qid_t qid[MAXQUOTAS];
1827         int cnt, i, rc = 0;
1828
1829         if (!inode)
1830                 return 0;
1831
1832         qid[USRQUOTA] = inode->i_uid;
1833         qid[GRPQUOTA] = inode->i_gid;
1834
1835         if (S_ISDIR(inode->i_mode) ||
1836             S_ISREG(inode->i_mode) ||
1837             S_ISLNK(inode->i_mode))
1838                 size = inode_get_bytes(inode);
1839
1840         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1841                 if (!Q_TYPESET(oqc, cnt))
1842                         continue;
1843
1844                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1845                                 &qctxt->qckt_list, qid[cnt], cnt,
1846                                 qctxt->qckt_first_check[cnt]);
1847                 if (!cdqb[cnt]) {
1848                         rc = -ENOMEM;
1849                         break;
1850                 }
1851
1852                 cdqb[cnt]->dqb_curspace += size;
1853                 cdqb[cnt]->dqb_curinodes++;
1854         }
1855
1856         if (rc) {
1857                 for (i = 0; i < cnt; i++) {
1858                         if (!Q_TYPESET(oqc, i))
1859                                 continue;
1860                         LASSERT(cdqb[i]);
1861                         cdqb[i]->dqb_curspace -= size;
1862                         cdqb[i]->dqb_curinodes--;
1863                 }
1864         }
1865
1866         return rc;
1867 }
1868
1869 /* write dqinfo struct in a new quota file */
1870 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1871 {
1872         struct v2_disk_dqinfo dqinfo;
1873         __u32 blocks = V2_DQTREEOFF + 1;
1874         loff_t offset = V2_DQINFOOFF;
1875
1876         if (info) {
1877                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1878                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1879                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1880                                                ~DQF_INFO_DIRTY);
1881         } else {
1882                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1883                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1884                 dqinfo.dqi_flags = 0;
1885         }
1886
1887         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1888         dqinfo.dqi_free_blk = 0;
1889         dqinfo.dqi_free_entry = 0;
1890
1891         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1892 }
1893
1894 static int v3_write_dqheader(struct file *f, int type)
1895 {
1896         static const __u32 quota_magics[] = V2_INITQMAGICS;
1897         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1898         struct v2_disk_dqheader dqhead;
1899         loff_t offset = 0;
1900
1901         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1902         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1903
1904         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1905         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1906
1907         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1908 }
1909
1910 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1911                                   struct obd_quotactl *oqc)
1912 {
1913         int i, rc = 0;
1914         ENTRY;
1915
1916         for (i = 0; i < MAXQUOTAS; i++) {
1917                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1918                                          NULL : &qctxt->qckt_dqinfo[i];
1919                 struct file *file;
1920                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1921
1922                 if (!Q_TYPESET(oqc, i))
1923                         continue;
1924
1925                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1926
1927                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1928                 if (IS_ERR(file)) {
1929                         rc = PTR_ERR(file);
1930                         CERROR("can't create %s file: rc = %d\n",
1931                                name[i], rc);
1932                         GOTO(out, rc);
1933                 }
1934
1935                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1936                         CERROR("file %s is not regular", name[i]);
1937                         filp_close(file, 0);
1938                         GOTO(out, rc = -EINVAL);
1939                 }
1940
1941                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1942
1943                 rc = v3_write_dqheader(file, i);
1944                 if (rc) {
1945                         filp_close(file, 0);
1946                         GOTO(out, rc);
1947                 }
1948
1949                 rc = v3_write_dqinfo(file, i, info);
1950                 filp_close(file, 0);
1951                 if (rc)
1952                         GOTO(out, rc);
1953         }
1954
1955 out:
1956         RETURN(rc);
1957 }
1958
1959
1960 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1961                           struct chk_dqblk *cdqb)
1962 {
1963         struct obd_quotactl *oqc;
1964         long now;
1965         int rc;
1966         ENTRY;
1967
1968         OBD_ALLOC_PTR(oqc);
1969         if (!oqc)
1970                 RETURN(-ENOMEM);
1971
1972         now = cfs_time_current_sec();
1973
1974         if (cdqb->dqb_bsoftlimit &&
1975             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1976             !cdqb->dqb_btime)
1977                 cdqb->dqb_btime =
1978                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1979
1980         if (cdqb->dqb_isoftlimit &&
1981             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1982             !cdqb->dqb_itime)
1983                 cdqb->dqb_itime =
1984                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1985
1986         cdqb->dqb_valid = QIF_ALL;
1987
1988         oqc->qc_cmd = Q_SETQUOTA;
1989         oqc->qc_type = cdqb->dqb_type;
1990         oqc->qc_id = cdqb->dqb_id;
1991         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1992
1993         rc = fsfilt_ext3_quotactl(sb, oqc);
1994         OBD_FREE_PTR(oqc);
1995         RETURN(rc);
1996 }
1997
1998 static int prune_chkquots(struct super_block *sb,
1999                           struct qchk_ctxt *qctxt, int error)
2000 {
2001         struct chk_dqblk *cdqb, *tmp;
2002         int rc;
2003
2004         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
2005                 if (!error) {
2006                         rc = commit_chkquot(sb, qctxt, cdqb);
2007                         if (rc)
2008                                 error = rc;
2009                 }
2010                 cfs_hlist_del_init(&cdqb->dqb_hash);
2011                 cfs_list_del(&cdqb->dqb_list);
2012                 OBD_FREE_PTR(cdqb);
2013         }
2014
2015         return error;
2016 }
2017
2018 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2019 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2020 #endif
2021
2022 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2023                                   struct obd_quotactl *oqc)
2024 {
2025         struct ext3_sb_info *sbi = EXT3_SB(sb);
2026         int i, group, uninit_feat = 0;
2027         struct qchk_ctxt *qctxt;
2028         struct buffer_head *bitmap_bh = NULL;
2029         unsigned long ino, inode_inuse;
2030         struct inode *inode;
2031         int rc = 0;
2032         ENTRY;
2033
2034         /* turn on quota and read dqinfo if existed */
2035         OBD_ALLOC_PTR(qctxt);
2036         if (!qctxt) {
2037                 oqc->qc_stat = -ENOMEM;
2038                 RETURN(-ENOMEM);
2039         }
2040
2041         for (i = 0; i < NR_DQHASH; i++)
2042                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2043         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2044
2045         for (i = 0; i < MAXQUOTAS; i++) {
2046                 if (!Q_TYPESET(oqc, i))
2047                         continue;
2048
2049                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2050                 if (!rc || rc == -EBUSY) {
2051                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2052                         if (rc)
2053                                 GOTO(out, rc);
2054                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2055                         qctxt->qckt_first_check[i] = 1;
2056                 } else if (rc) {
2057                         GOTO(out, rc);
2058                 }
2059         }
2060         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2061                 /* This filesystem supports the uninit group feature */
2062                 uninit_feat = 1;
2063
2064         /* number of inodes that have been allocated */
2065         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2066                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2067
2068         /* check quota and update in hash */
2069         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2070              group++) {
2071                 unsigned long used_count = sbi->s_inodes_per_group;
2072
2073                 if (uninit_feat) {
2074                         struct ext3_group_desc *desc;
2075                         desc = get_group_desc(sb, group, NULL);
2076                         if (!desc)
2077                                 GOTO(out, -EIO);
2078
2079                         /* we don't really need to take the group lock here,
2080                          * but it may be useful if one day we support online
2081                          * quotacheck */
2082 #ifdef HAVE_EXT4_LDISKFS
2083                         ext4_lock_group(sb, group);
2084 #else
2085                         spin_lock(sb_bgl_lock(sbi, group));
2086 #endif
2087                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2088                                 /* no inode in use in this group, just skip it */
2089 #ifdef HAVE_EXT4_LDISKFS
2090                                 ext3_unlock_group(sb, group);
2091 #else
2092                                 spin_unlock(sb_bgl_lock(sbi, group));
2093 #endif
2094                                 continue;
2095                         }
2096
2097                         used_count -= ext3_itable_unused_count(sb, desc);
2098 #ifdef HAVE_EXT4_LDISKFS
2099                         ext3_unlock_group(sb, group);
2100 #else
2101                         spin_unlock(sb_bgl_lock(sbi, group));
2102 #endif
2103                 }
2104
2105                 ino = group * sbi->s_inodes_per_group + 1;
2106                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2107                 if (!bitmap_bh) {
2108                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2109                                sb->s_id, group);
2110                         GOTO(out, -EIO);
2111                 }
2112
2113                 i = 0;
2114                 while (i < used_count &&
2115                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2116                                                used_count, i)) < used_count) {
2117                         inode_inuse--;
2118                         i++;
2119                         ino = i + group * sbi->s_inodes_per_group;
2120                         if (ino < sbi->s_first_ino)
2121                                 continue;
2122 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2123                         inode = ext3_iget(sb, ino);
2124 #else
2125                         inode = iget(sb, ino);
2126 #endif
2127                         if (!inode || IS_ERR(inode))
2128                                 continue;
2129
2130                         rc = add_inode_quota(inode, qctxt, oqc);
2131                         iput(inode);
2132                         if (rc) {
2133                                 brelse(bitmap_bh);
2134                                 GOTO(out, rc);
2135                         }
2136                 }
2137                 brelse(bitmap_bh);
2138         }
2139
2140         /* read old quota limits from old quota file. (only for the user
2141          * has limits but hasn't file) */
2142 #ifdef HAVE_QUOTA_SUPPORT
2143         for (i = 0; i < MAXQUOTAS; i++) {
2144                 cfs_list_t id_list;
2145                 struct dquot_id *dqid, *tmp;
2146
2147                 if (!Q_TYPESET(oqc, i))
2148                         continue;
2149
2150                 if (qctxt->qckt_first_check[i])
2151                         continue;
2152
2153
2154                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2155                 CFS_INIT_LIST_HEAD(&id_list);
2156 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2157                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2158 #else
2159                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2160 #endif
2161                 if (rc)
2162                         CERROR("read old limits failed. (rc:%d)\n", rc);
2163
2164                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2165                         cfs_list_del_init(&dqid->di_link);
2166
2167                         if (!rc)
2168                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2169                                       dqid->di_id, i,
2170                                       qctxt->qckt_first_check[i]);
2171                         OBD_FREE_PTR(dqid);
2172                 }
2173         }
2174 #endif
2175         /* turn off quota cause we are to dump chk_dqblk to files */
2176         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2177
2178         rc = create_new_quota_files(qctxt, oqc);
2179         if (rc)
2180                 GOTO(out, rc);
2181
2182         /* we use vfs functions to set dqblk, so turn quota on */
2183         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2184 out:
2185         /* dump and free chk_dqblk */
2186         rc = prune_chkquots(sb, qctxt, rc);
2187         OBD_FREE_PTR(qctxt);
2188
2189         /* turn off quota, `lfs quotacheck` will turn on when all
2190          * nodes quotacheck finish. */
2191         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2192
2193         oqc->qc_stat = rc;
2194         if (rc)
2195                 CERROR("quotacheck failed: rc = %d\n", rc);
2196
2197         RETURN(rc);
2198 }
2199
2200 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2201                                  int cmd)
2202 {
2203         int rc = 0;
2204         ENTRY;
2205
2206         if (lqi->qi_files[type] == NULL) {
2207                 CERROR("operate qinfo before it's enabled!\n");
2208                 RETURN(-ESRCH);
2209         }
2210
2211         switch (cmd) {
2212         case QFILE_CHK:
2213                 rc = lustre_check_quota_file(lqi, type);
2214                 break;
2215         case QFILE_RD_INFO:
2216                 rc = lustre_read_quota_info(lqi, type);
2217                 break;
2218         case QFILE_WR_INFO:
2219                 rc = lustre_write_quota_info(lqi, type);
2220                 break;
2221         case QFILE_INIT_INFO:
2222                 rc = lustre_init_quota_info(lqi, type);
2223                 break;
2224         case QFILE_CONVERT:
2225                 rc = -ENOTSUPP;
2226                 CERROR("quota CONVERT command is not supported\n");
2227                 break;
2228         default:
2229                 rc = -ENOTSUPP;
2230                 CERROR("Unsupported admin quota file cmd %d\n"
2231                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2232                        cmd);
2233                 break;
2234         }
2235         RETURN(rc);
2236 }
2237
2238 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2239                             cfs_list_t *list)
2240 {
2241         return lustre_get_qids(file, inode, type, list);
2242 }
2243
2244 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2245 {
2246         int rc = 0;
2247         ENTRY;
2248
2249         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2250                 CERROR("operate dquot before it's enabled!\n");
2251                 RETURN(-ESRCH);
2252         }
2253
2254         switch (cmd) {
2255         case QFILE_RD_DQUOT:
2256                 rc = lustre_read_dquot(dquot);
2257                 break;
2258         case QFILE_WR_DQUOT:
2259                 if (dquot->dq_dqb.dqb_ihardlimit ||
2260                     dquot->dq_dqb.dqb_isoftlimit ||
2261                     dquot->dq_dqb.dqb_bhardlimit ||
2262                     dquot->dq_dqb.dqb_bsoftlimit)
2263                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2264                 else
2265                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2266
2267                 rc = lustre_commit_dquot(dquot);
2268                 if (rc >= 0)
2269                         rc = 0;
2270                 break;
2271         default:
2272                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2273                 LBUG();
2274                 break;
2275         }
2276         RETURN(rc);
2277 }
2278
2279 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2280                                 struct inode *inode, int frags)
2281 {
2282 #ifdef EXT3_EXT_HAS_NO_TREE
2283         struct ext3_ext_base *base = inode;
2284 #else
2285         struct ext3_extents_tree tree;
2286         struct ext3_ext_base *base = &tree;
2287
2288         ext3_init_tree_desc(base, inode);
2289 #endif
2290         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2291          * metablocks at maxium b=16542 */
2292         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2293         return 0;
2294 }
2295
2296 #endif
2297
2298 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2299 {
2300         return (EXT3_SB(sb)->journal_bdev);
2301 }
2302 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2303
2304 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2305                           char *buf, int count, loff_t pos)
2306 {
2307         loff_t p = pos;
2308         int rc;
2309
2310         if (!f && !inode) {
2311                 CERROR("lustre_read_quota failed for no quota file!\n");
2312                 libcfs_debug_dumpstack(NULL);
2313                 return -EINVAL;
2314         }
2315
2316         /* Support for both adm and op quota files must be provided */
2317         if (f) {
2318                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2319                 rc = rc < 0 ? rc : p - pos;
2320         } else {
2321                 struct super_block *sb = inode->i_sb;
2322                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2323         }
2324         return rc;
2325 }
2326
2327 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2328 {
2329         loff_t p = pos;
2330         int rc;
2331
2332         /* Only adm quota files are supported, op updates are handled by vfs */
2333         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2334         rc = rc < 0 ? rc : p - pos;
2335
2336         return rc;
2337 }
2338
2339 void *lustre_quota_journal_start(struct inode *inode, int delete)
2340 {
2341         handle_t *handle;
2342         unsigned block_count;
2343
2344         if (delete) {
2345                 /* each indirect block (+4) may become free, attaching to the
2346                  * header list of free blocks (+1); the data block (+1) may
2347                  * become a free block (+0) or a block with free dqentries (+0) */
2348                 block_count = (4 + 1) + 1;
2349                 handle = ext3_journal_start(inode,
2350                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2351         } else {
2352                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2353                  * freeblk reusage with a header update (+1); dqentry is either reused
2354                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2355                  * a new block allocation (+1) with a header update (+1)              */
2356                 block_count = (4 + 1) + 3;
2357                 handle = ext3_journal_start(inode,
2358                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2359
2360         }
2361
2362         return handle;
2363 }
2364
2365 void lustre_quota_journal_stop(void *handle)
2366 {
2367         ext3_journal_stop((handle_t *)handle);
2368 }
2369
2370 static int ll_decode_fh_accept(void *context, struct dentry *de)
2371 {
2372         return 1;
2373 }
2374
2375 #ifdef HAVE_EXPORTFS_DECODE_FH
2376 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2377          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2378                             acceptable, context)
2379 #else
2380 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2381          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2382                                      type, acceptable, context)
2383 # define FILEID_INO32_GEN 1
2384 extern struct export_operations export_op_default;
2385 #endif
2386
2387 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2388                                       struct fsfilt_fid *fid, int ignore_gen)
2389 {
2390         struct inode  *inode;
2391         struct dentry *result;
2392
2393         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2394                                        ll_decode_fh_accept, NULL);
2395         if (IS_ERR(result)) {
2396                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2397                        fid->ino, fid->gen, PTR_ERR(result));
2398                 return result;
2399         }
2400
2401         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2402                fid->ino, fid->gen);
2403         inode = result->d_inode;
2404         if (inode == NULL)
2405                 goto err_out;
2406
2407         if (inode->i_nlink == 0 &&
2408             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2409                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2410                               " ctime -- this may indicate disk "
2411                               "corruption (inode: %lu, link: %lu, "
2412                               "count: %d)\n", inode->i_ino,
2413                               (unsigned long)inode->i_nlink,
2414                               atomic_read(&inode->i_count));
2415                 goto err_out;
2416         }
2417         if (fid->gen && inode->i_generation != fid->gen) {
2418                 /* we didn't find the right inode.. */
2419                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2420                        "count: %d, generation %u/%u\n",
2421                        inode->i_ino, (unsigned long)inode->i_nlink,
2422                        atomic_read(&inode->i_count), inode->i_generation,
2423                        fid->gen);
2424                 goto err_out;
2425         }
2426
2427         return result;
2428 err_out:
2429         l_dput(result);
2430         return ERR_PTR(-ENOENT);
2431 }
2432
2433 static struct fsfilt_operations fsfilt_ext3_ops = {
2434         .fs_type                = "ext3",
2435         .fs_owner               = THIS_MODULE,
2436         .fs_getlabel            = fsfilt_ext3_get_label,
2437         .fs_setlabel            = fsfilt_ext3_set_label,
2438         .fs_uuid                = fsfilt_ext3_uuid,
2439         .fs_start               = fsfilt_ext3_start,
2440         .fs_brw_start           = fsfilt_ext3_brw_start,
2441         .fs_extend              = fsfilt_ext3_extend,
2442         .fs_commit              = fsfilt_ext3_commit,
2443         .fs_commit_async        = fsfilt_ext3_commit_async,
2444         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2445         .fs_setattr             = fsfilt_ext3_setattr,
2446         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2447         .fs_set_md              = fsfilt_ext3_set_md,
2448         .fs_get_md              = fsfilt_ext3_get_md,
2449         .fs_readpage            = fsfilt_ext3_readpage,
2450         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2451         .fs_statfs              = fsfilt_ext3_statfs,
2452         .fs_sync                = fsfilt_ext3_sync,
2453         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2454         .fs_write_record        = fsfilt_ext3_write_record,
2455         .fs_read_record         = fsfilt_ext3_read_record,
2456         .fs_setup               = fsfilt_ext3_setup,
2457         .fs_send_bio            = fsfilt_ext3_send_bio,
2458         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2459 #ifdef HAVE_DISK_INODE_VERSION
2460         .fs_get_version         = fsfilt_ext3_get_version,
2461         .fs_set_version         = fsfilt_ext3_set_version,
2462 #endif
2463 #ifdef HAVE_QUOTA_SUPPORT
2464         .fs_quotactl            = fsfilt_ext3_quotactl,
2465         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2466         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2467         .fs_qids                = fsfilt_ext3_qids,
2468         .fs_dquot               = fsfilt_ext3_dquot,
2469         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2470 #endif
2471         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2472         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2473 };
2474
2475 static int __init fsfilt_ext3_init(void)
2476 {
2477         int rc;
2478
2479         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2480                                          sizeof(struct fsfilt_cb_data), 0, 0);
2481         if (!fcb_cache) {
2482                 CERROR("error allocating fsfilt journal callback cache\n");
2483                 GOTO(out, rc = -ENOMEM);
2484         }
2485
2486         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2487
2488         if (rc) {
2489                 int err = cfs_mem_cache_destroy(fcb_cache);
2490                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2491         }
2492 out:
2493         return rc;
2494 }
2495
2496 static void __exit fsfilt_ext3_exit(void)
2497 {
2498         int rc;
2499
2500         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2501         rc = cfs_mem_cache_destroy(fcb_cache);
2502         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2503 }
2504
2505 module_init(fsfilt_ext3_init);
2506 module_exit(fsfilt_ext3_exit);
2507
2508 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2509 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2510 MODULE_LICENSE("GPL");