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 &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 660
 661         if (inode->i_op->setattr) {
 662                 rc = inode->i_op->setattr(dentry, iattr);
 663         } else {
 664                 rc = inode_change_ok(inode, iattr);
 665                 if (!rc)
 666                         rc = simple_setattr(dentry, iattr);
 667         }
 668
 669  out:
 670         RETURN(rc);
 671 }
 672
 673 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 674                                  unsigned int cmd, unsigned long arg)
 675 {
 676         int rc = 0;
 677         ENTRY;
 678
 679         /* FIXME: Can't do this because of nested transaction deadlock */
 680         if (cmd == EXT3_IOC_SETFLAGS) {
 681                 /* We can't enable data journaling on OST objects, because
 682                 * this forces the transaction to be closed in order to
 683                 * flush the journal, but the caller will already have a
 684                 * compound transaction open to update the last_rcvd file,
 685                 * and this thread would deadlock trying to set the flag. */
 686                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 687                         CERROR("can't set data journal flag on file\n");
 688                         RETURN(-EPERM);
 689                 }
 690                 /* Because the MDS does not see the EXTENTS_FL set on the
 691                  * OST objects, mask this flag into all set flags.  It is
 692                  * not legal to clear this flag in any case, so we are not
 693                  * changing the functionality by doing this.  b=22911 */
 694                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 695         }
 696
 697 #ifdef HAVE_EXT4_LDISKFS
 698         /* ext4_ioctl does not have a inode argument */
 699         if (inode->i_fop->unlocked_ioctl)
 700                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 701 #else
 702         if (inode->i_fop->ioctl)
 703                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 704 #endif
 705         else
 706                 RETURN(-ENOTTY);
 707
 708         RETURN(rc);
 709 }
 710
 711 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 712                               void *lmm, int lmm_size, const char *name)
 713 {
 714         int rc;
 715
 716         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 717
 718         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 719                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 720
 721
 722         if (rc && rc != -EROFS)
 723                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 724                        inode->i_ino, rc);
 725         return rc;
 726 }
 727
 728 /* Must be called with i_mutex held */
 729 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 730                               const char *name)
 731 {
 732         int rc;
 733
 734         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 735
 736         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 737                             name, lmm, lmm_size);
 738
 739         /* This gives us the MD size */
 740         if (lmm == NULL)
 741                 return (rc == -ENODATA) ? 0 : rc;
 742
 743         if (rc < 0) {
 744                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 745                        EXT3_XATTR_INDEX_TRUSTED, name,
 746                        inode->i_ino, rc);
 747                 memset(lmm, 0, lmm_size);
 748                 return (rc == -ENODATA) ? 0 : rc;
 749         }
 750
 751         return rc;
 752 }
 753
 754 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 755 {
 756         submit_bio(rw, bio);
 757         return 0;
 758 }
 759
 760 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 761                                     loff_t *off)
 762 {
 763         struct inode *inode = file->f_dentry->d_inode;
 764         int rc = 0;
 765
 766         if (S_ISREG(inode->i_mode))
 767                 rc = file->f_op->read(file, buf, count, off);
 768         else {
 769                 const int blkbits = inode->i_sb->s_blocksize_bits;
 770                 const int blksize = inode->i_sb->s_blocksize;
 771
 772                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 773                        (unsigned long)count, inode->i_ino, *off);
 774                 while (count > 0) {
 775                         struct buffer_head *bh;
 776
 777                         bh = NULL;
 778                         if (*off < i_size_read(inode)) {
 779                                 int err = 0;
 780
 781                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 782                                                 0, &err);
 783
 784                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 785
 786                                 if (bh) {
 787                                         memcpy(buf, bh->b_data, blksize);
 788                                         brelse(bh);
 789                                 } else if (err) {
 790                                         /* XXX in theory we should just fake
 791                                          * this buffer and continue like ext3,
 792                                          * especially if this is a partial read
 793                                          */
 794                                         CERROR("error read dir %lu+%llu: %d\n",
 795                                                inode->i_ino, *off, err);
 796                                         RETURN(err);
 797                                 }
 798                         }
 799                         if (!bh) {
 800                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 801
 802                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 803                                 memset(fake, 0, sizeof(*fake));
 804                                 fake->rec_len = cpu_to_le16(blksize);
 805                         }
 806                         count -= blksize;
 807                         buf += blksize;
 808                         *off += blksize;
 809                         rc += blksize;
 810                 }
 811         }
 812
 813         return rc;
 814 }
 815
 816 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 817 static void fsfilt_ext3_cb_func(struct super_block *sb,
 818                                 struct journal_callback *jcb, int error)
 819 #else
 820 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 821 #endif
 822 {
 823         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 824
 825         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 826
 827         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 828 }
 829
 830 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 831                                       void *handle, fsfilt_cb_t cb_func,
 832                                       void *cb_data)
 833 {
 834         struct fsfilt_cb_data *fcb;
 835
 836         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 837         if (fcb == NULL)
 838                 RETURN(-ENOMEM);
 839
 840         fcb->cb_func = cb_func;
 841         fcb->cb_obd = obd;
 842         fcb->cb_last_rcvd = last_rcvd;
 843         fcb->cb_data = cb_data;
 844
 845         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 846         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 847
 848         return 0;
 849 }
 850
 851 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 852 {
 853         struct kstatfs sfs;
 854         int rc;
 855
 856         memset(&sfs, 0, sizeof(sfs));
 857         rc = ll_do_statfs(sb, &sfs);
 858         statfs_pack(osfs, &sfs);
 859         return rc;
 860 }
 861
 862 static int fsfilt_ext3_sync(struct super_block *sb)
 863 {
 864         return ext3_force_commit(sb);
 865 }
 866
 867 #ifndef EXT3_EXTENTS_FL
 868 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 869 #endif
 870
 871 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 872 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
 873 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
 874 #else
 875 # ifdef HAVE_EXT4_LDISKFS
 876 #   define fsfilt_up_truncate_sem(inode) do{ }while(0)
 877 #   define fsfilt_down_truncate_sem(inode) do{ }while(0)
 878 # else
 879 #  define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex)
 880 #  define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex)
 881 # endif
 882 #endif
 883
 884 #ifndef EXT_ASSERT
 885 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 886 #endif
 887
 888 #ifdef EXT3_EXT_HAS_NO_TREE
 889 /* for kernels 2.6.18 and later */
 890 #ifdef HAVE_EXT4_LDISKFS
 891 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 892 #else
 893 #define EXT_GENERATION(inode)           ext_generation(inode)
 894 #endif
 895 #define ext3_ext_base                   inode
 896 #define ext3_ext_base2inode(inode)      (inode)
 897 #define EXT_DEPTH(inode)                ext_depth(inode)
 898 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 899                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 900 #else
 901 #define ext3_ext_base                   ext3_extents_tree
 902 #define ext3_ext_base2inode(tree)       (tree->inode)
 903 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 904                         ext3_ext_walk_space(tree, block, num, cb);
 905 #endif
 906
 907 #include <linux/lustre_version.h>
 908
 909 struct bpointers {
 910         unsigned long *blocks;
 911         int *created;
 912         unsigned long start;
 913         int num;
 914         int init_num;
 915         int create;
 916 };
 917
 918 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 919                                unsigned long block, int *aflags)
 920 {
 921         struct ext3_inode_info *ei = EXT3_I(inode);
 922         unsigned long bg_start;
 923         unsigned long colour;
 924         int depth;
 925
 926         if (path) {
 927                 struct ext3_extent *ex;
 928                 depth = path->p_depth;
 929
 930                 /* try to predict block placement */
 931                 if ((ex = path[depth].p_ext))
 932                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 933
 934                 /* it looks index is empty
 935                  * try to find starting from index itself */
 936                 if (path[depth].p_bh)
 937                         return path[depth].p_bh->b_blocknr;
 938         }
 939
 940         /* OK. use inode's group */
 941         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 942                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 943         colour = (current->pid % 16) *
 944                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 945         return bg_start + colour + block;
 946 }
 947
 948 #define ll_unmap_underlying_metadata(sb, blocknr) \
 949         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 950
 951 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 952 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 953                                 struct ext3_ext_path *path, unsigned long block,
 954                                 unsigned long *count, int *err)
 955 {
 956         unsigned long pblock, goal;
 957         int aflags = 0;
 958         struct inode *inode = ext3_ext_base2inode(base);
 959
 960         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 961         aflags |= 2; /* block have been already reserved */
 962         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 963         return pblock;
 964
 965 }
 966 #else
 967 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 968                                 struct ext3_ext_path *path, unsigned long block,
 969                                 unsigned long *count, int *err)
 970 {
 971         struct inode *inode = ext3_ext_base2inode(base);
 972         struct ext3_allocation_request ar;
 973         unsigned long pblock;
 974         int aflags;
 975
 976         /* find neighbour allocated blocks */
 977         ar.lleft = block;
 978         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 979         if (*err)
 980                 return 0;
 981         ar.lright = block;
 982         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 983         if (*err)
 984                 return 0;
 985
 986         /* allocate new block */
 987         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 988         ar.inode = inode;
 989         ar.logical = block;
 990         ar.len = *count;
 991         ar.flags = EXT3_MB_HINT_DATA;
 992         pblock = ext3_mb_new_blocks(handle, &ar, err);
 993         *count = ar.len;
 994         return pblock;
 995 }
 996 #endif
 997
 998 #ifdef EXT3_EXT_HAS_NO_TREE
 999 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1000                                   struct ext3_ext_path *path,
1001                                   struct ext3_ext_cache *cex,
1002 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1003                                    struct ext3_extent *ex,
1004 #endif
1005                                   void *cbdata)
1006 {
1007         struct bpointers *bp = cbdata;
1008 #else
1009 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1010                                   struct ext3_ext_path *path,
1011                                   struct ext3_ext_cache *cex
1012 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1013                                   , struct ext3_extent *ex
1014 #endif
1015                                  )
1016 {
1017         struct bpointers *bp = base->private;
1018 #endif
1019         struct inode *inode = ext3_ext_base2inode(base);
1020         struct ext3_extent nex;
1021         unsigned long pblock;
1022         unsigned long tgen;
1023         int err, i;
1024         unsigned long count;
1025         handle_t *handle;
1026
1027         i = EXT_DEPTH(base);
1028         EXT_ASSERT(path[i].p_hdr);
1029         LASSERTF(i == path->p_depth ||
1030                  EXT_GENERATION(base) != path[0].p_generation,
1031                  "base vs path extent depth:%d != %d, generation:%lu == %lu\n",
1032                  i, path->p_depth, EXT_GENERATION(base), path[0].p_generation);
1033
1034         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1035                 err = EXT_CONTINUE;
1036                 goto map;
1037         }
1038
1039         if (bp->create == 0) {
1040                 i = 0;
1041                 if (cex->ec_block < bp->start)
1042                         i = bp->start - cex->ec_block;
1043                 if (i >= cex->ec_len)
1044                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1045                                         i, cex->ec_len);
1046                 for (; i < cex->ec_len && bp->num; i++) {
1047                         *(bp->created) = 0;
1048                         bp->created++;
1049                         *(bp->blocks) = 0;
1050                         bp->blocks++;
1051                         bp->num--;
1052                         bp->start++;
1053                 }
1054
1055                 return EXT_CONTINUE;
1056         }
1057
1058         tgen = EXT_GENERATION(base);
1059         count = ext3_ext_calc_credits_for_insert(base, path);
1060         fsfilt_up_truncate_sem(inode);
1061
1062         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1063         if (IS_ERR(handle)) {
1064                 fsfilt_down_truncate_sem(inode);
1065                 return PTR_ERR(handle);
1066         }
1067
1068         fsfilt_down_truncate_sem(inode);
1069         if (tgen != EXT_GENERATION(base)) {
1070                 /* the tree has changed. so path can be invalid at moment */
1071                 ext3_journal_stop(handle);
1072                 return EXT_REPEAT;
1073         }
1074
1075 #if defined(HAVE_EXT4_LDISKFS)
1076         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1077          * protected by i_data_sem as whole. so we patch it to store
1078          * generation to path and now verify the tree hasn't changed */
1079         down_write((&EXT4_I(inode)->i_data_sem));
1080
1081         /* validate extent, make sure the extent tree does not changed */
1082         if (EXT_GENERATION(base) != path[0].p_generation) {
1083                 /* cex is invalid, try again */
1084                 up_write(&EXT4_I(inode)->i_data_sem);
1085                 ext3_journal_stop(handle);
1086                 return EXT_REPEAT;
1087         }
1088 #endif
1089
1090         count = cex->ec_len;
1091         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1092         if (!pblock)
1093                 goto out;
1094         EXT_ASSERT(count <= cex->ec_len);
1095
1096         /* insert new extent */
1097         nex.ee_block = cpu_to_le32(cex->ec_block);
1098         ext3_ext_store_pblock(&nex, pblock);
1099         nex.ee_len = cpu_to_le16(count);
1100         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1101         if (err) {
1102                 /* free data blocks we just allocated */
1103                 /* not a good idea to call discard here directly,
1104                  * but otherwise we'd need to call it every free() */
1105 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1106                 ext3_mb_discard_inode_preallocations(inode);
1107 #endif
1108                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1109                                  cpu_to_le16(nex.ee_len), 0);
1110                 goto out;
1111         }
1112
1113         /*
1114          * Putting len of the actual extent we just inserted,
1115          * we are asking ext3_ext_walk_space() to continue
1116          * scaning after that block
1117          */
1118         cex->ec_len = le16_to_cpu(nex.ee_len);
1119         cex->ec_start = ext_pblock(&nex);
1120         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1121         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1122
1123 out:
1124 #if defined(HAVE_EXT4_LDISKFS)
1125         up_write((&EXT4_I(inode)->i_data_sem));
1126 #endif
1127         ext3_journal_stop(handle);
1128 map:
1129         if (err >= 0) {
1130                 /* map blocks */
1131                 if (bp->num == 0) {
1132                         CERROR("hmm. why do we find this extent?\n");
1133                         CERROR("initial space: %lu:%u\n",
1134                                 bp->start, bp->init_num);
1135                         CERROR("current extent: %u/%u/%llu %d\n",
1136                                 cex->ec_block, cex->ec_len,
1137                                 (unsigned long long)cex->ec_start,
1138                                 cex->ec_type);
1139                 }
1140                 i = 0;
1141                 if (cex->ec_block < bp->start)
1142                         i = bp->start - cex->ec_block;
1143                 if (i >= cex->ec_len)
1144                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1145                                         i, cex->ec_len);
1146                 for (; i < cex->ec_len && bp->num; i++) {
1147                         *(bp->blocks) = cex->ec_start + i;
1148                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1149                                 *(bp->created) = 0;
1150                         } else {
1151                                 *(bp->created) = 1;
1152                                 /* unmap any possible underlying metadata from
1153                                  * the block device mapping.  bug 6998. */
1154                                 ll_unmap_underlying_metadata(inode->i_sb,
1155                                                              *(bp->blocks));
1156                         }
1157                         bp->created++;
1158                         bp->blocks++;
1159                         bp->num--;
1160                         bp->start++;
1161                 }
1162         }
1163         return err;
1164 }
1165
1166 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1167                        unsigned long num, unsigned long *blocks,
1168                        int *created, int create)
1169 {
1170 #ifdef EXT3_EXT_HAS_NO_TREE
1171         struct ext3_ext_base *base = inode;
1172 #else
1173         struct ext3_extents_tree tree;
1174         struct ext3_ext_base *base = &tree;
1175 #endif
1176         struct bpointers bp;
1177         int err;
1178
1179         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1180                block, block + num - 1, (unsigned) inode->i_ino);
1181
1182 #ifndef EXT3_EXT_HAS_NO_TREE
1183         ext3_init_tree_desc(base, inode);
1184         tree.private = &bp;
1185 #endif
1186         bp.blocks = blocks;
1187         bp.created = created;
1188         bp.start = block;
1189         bp.init_num = bp.num = num;
1190         bp.create = create;
1191
1192         fsfilt_down_truncate_sem(inode);
1193         err = fsfilt_ext3_ext_walk_space(base, block, num,
1194                                          ext3_ext_new_extent_cb, &bp);
1195         ext3_ext_invalidate_cache(base);
1196         fsfilt_up_truncate_sem(inode);
1197
1198         return err;
1199 }
1200
1201 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1202                                     int pages, unsigned long *blocks,
1203                                     int *created, int create)
1204 {
1205         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1206         int rc = 0, i = 0;
1207         struct page *fp = NULL;
1208         int clen = 0;
1209
1210         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1211                 inode->i_ino, pages, (*page)->index);
1212
1213         /* pages are sorted already. so, we just have to find
1214          * contig. space and process them properly */
1215         while (i < pages) {
1216                 if (fp == NULL) {
1217                         /* start new extent */
1218                         fp = *page++;
1219                         clen = 1;
1220                         i++;
1221                         continue;
1222                 } else if (fp->index + clen == (*page)->index) {
1223                         /* continue the extent */
1224                         page++;
1225                         clen++;
1226                         i++;
1227                         continue;
1228                 }
1229
1230                 /* process found extent */
1231                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1232                                         clen * blocks_per_page, blocks,
1233                                         created, create);
1234                 if (rc)
1235                         GOTO(cleanup, rc);
1236
1237                 /* look for next extent */
1238                 fp = NULL;
1239                 blocks += blocks_per_page * clen;
1240                 created += blocks_per_page * clen;
1241         }
1242
1243         if (fp)
1244                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1245                                         clen * blocks_per_page, blocks,
1246                                         created, create);
1247 cleanup:
1248         return rc;
1249 }
1250
1251 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1252                                unsigned long *blocks, int *created, int create);
1253 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1254                                    int pages, unsigned long *blocks,
1255                                    int *created, int create)
1256 {
1257         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1258         unsigned long *b;
1259         int rc = 0, i, *cr;
1260
1261         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1262                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1263                 if (rc) {
1264                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1265                                inode->i_ino, *b, *cr, create, rc);
1266                         break;
1267                 }
1268
1269                 b += blocks_per_page;
1270                 cr += blocks_per_page;
1271         }
1272         return rc;
1273 }
1274
1275 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1276                                 int pages, unsigned long *blocks,
1277                                 int *created, int create,
1278                                 cfs_semaphore_t *optional_sem)
1279 {
1280         int rc;
1281
1282         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1283                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1284                                                      blocks, created, create);
1285                 return rc;
1286         }
1287         if (optional_sem != NULL)
1288                 cfs_down(optional_sem);
1289         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1290                                             created, create);
1291         if (optional_sem != NULL)
1292                 cfs_up(optional_sem);
1293
1294         return rc;
1295 }
1296
1297 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1298 {
1299         unsigned long block;
1300         struct buffer_head *bh;
1301         int err, blocksize, csize, boffs, osize = size;
1302
1303         /* prevent reading after eof */
1304         cfs_lock_kernel();
1305         if (i_size_read(inode) < *offs + size) {
1306                 size = i_size_read(inode) - *offs;
1307                 cfs_unlock_kernel();
1308                 if (size < 0) {
1309                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1310                                i_size_read(inode), *offs);
1311                         return -EBADR;
1312                 } else if (size == 0) {
1313                         return 0;
1314                 }
1315         } else {
1316                 cfs_unlock_kernel();
1317         }
1318
1319         blocksize = 1 << inode->i_blkbits;
1320
1321         while (size > 0) {
1322                 block = *offs >> inode->i_blkbits;
1323                 boffs = *offs & (blocksize - 1);
1324                 csize = min(blocksize - boffs, size);
1325                 bh = ext3_bread(NULL, inode, block, 0, &err);
1326                 if (!bh) {
1327                         CERROR("can't read block: %d\n", err);
1328                         return err;
1329                 }
1330
1331                 memcpy(buf, bh->b_data + boffs, csize);
1332                 brelse(bh);
1333
1334                 *offs += csize;
1335                 buf += csize;
1336                 size -= csize;
1337         }
1338         return osize;
1339 }
1340 EXPORT_SYMBOL(fsfilt_ext3_read);
1341
1342 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1343                                    int size, loff_t *offs)
1344 {
1345         int rc;
1346         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1347         if (rc > 0)
1348                 rc = 0;
1349         return rc;
1350 }
1351
1352 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1353                                 loff_t *offs, handle_t *handle)
1354 {
1355         struct buffer_head *bh = NULL;
1356         loff_t old_size = i_size_read(inode), offset = *offs;
1357         loff_t new_size = i_size_read(inode);
1358         unsigned long block;
1359         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1360
1361         while (bufsize > 0) {
1362                 if (bh != NULL)
1363                         brelse(bh);
1364
1365                 block = offset >> inode->i_blkbits;
1366                 boffs = offset & (blocksize - 1);
1367                 size = min(blocksize - boffs, bufsize);
1368                 bh = ext3_bread(handle, inode, block, 1, &err);
1369                 if (!bh) {
1370                         CERROR("can't read/create block: %d\n", err);
1371                         break;
1372                 }
1373
1374                 err = ext3_journal_get_write_access(handle, bh);
1375                 if (err) {
1376                         CERROR("journal_get_write_access() returned error %d\n",
1377                                err);
1378                         break;
1379                 }
1380                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1381                 memcpy(bh->b_data + boffs, buf, size);
1382                 err = ext3_journal_dirty_metadata(handle, bh);
1383                 if (err) {
1384                         CERROR("journal_dirty_metadata() returned error %d\n",
1385                                err);
1386                         break;
1387                 }
1388                 if (offset + size > new_size)
1389                         new_size = offset + size;
1390                 offset += size;
1391                 bufsize -= size;
1392                 buf += size;
1393         }
1394         if (bh)
1395                 brelse(bh);
1396
1397         /* correct in-core and on-disk sizes */
1398         if (new_size > i_size_read(inode)) {
1399                 cfs_lock_kernel();
1400                 if (new_size > i_size_read(inode))
1401                         i_size_write(inode, new_size);
1402                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1403                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1404                 if (i_size_read(inode) > old_size)
1405                         mark_inode_dirty(inode);
1406                 cfs_unlock_kernel();
1407         }
1408
1409         if (err == 0)
1410                 *offs = offset;
1411         return err;
1412 }
1413 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1414
1415 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1416                                     loff_t *offs, int force_sync)
1417 {
1418         struct inode *inode = file->f_dentry->d_inode;
1419         handle_t *handle;
1420         int err, block_count = 0, blocksize;
1421
1422         /* Determine how many transaction credits are needed */
1423         blocksize = 1 << inode->i_blkbits;
1424         block_count = (*offs & (blocksize - 1)) + bufsize;
1425         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1426
1427         handle = ext3_journal_start(inode,
1428                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1429         if (IS_ERR(handle)) {
1430                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1431                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1432                 return PTR_ERR(handle);
1433         }
1434
1435         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1436
1437         if (!err && force_sync)
1438                 handle->h_sync = 1; /* recovery likes this */
1439
1440         ext3_journal_stop(handle);
1441
1442         return err;
1443 }
1444
1445 static int fsfilt_ext3_setup(struct super_block *sb)
1446 {
1447 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1448      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1449         struct ext3_sb_info *sbi = EXT3_SB(sb);
1450 #if 0
1451         sbi->dx_lock = fsfilt_ext3_dx_lock;
1452         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1453 #endif
1454 #endif
1455 #ifdef S_PDIROPS
1456         CWARN("Enabling PDIROPS\n");
1457         set_opt(sbi->s_mount_opt, PDIROPS);
1458         sb->s_flags |= S_PDIROPS;
1459 #endif
1460         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1461                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1462 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1463         /* enable journaled quota support */
1464         /* kfreed in ext3_put_super() */
1465         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1466         if (!sbi->s_qf_names[USRQUOTA])
1467                 return -ENOMEM;
1468         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1469         if (!sbi->s_qf_names[GRPQUOTA]) {
1470                 kfree(sbi->s_qf_names[USRQUOTA]);
1471                 sbi->s_qf_names[USRQUOTA] = NULL;
1472                 return -ENOMEM;
1473         }
1474         sbi->s_jquota_fmt = QFMT_LUSTRE;
1475 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1476         set_opt(sbi->s_mount_opt, QUOTA);
1477 #endif
1478 #endif
1479         return 0;
1480 }
1481
1482 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1483    objects. Logs is number of logfiles to update */
1484 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1485 {
1486         if ( !fso ) {
1487                 switch(op) {
1488                 case FSFILT_OP_CREATE:
1489                                  /* directory leaf, index & indirect & EA*/
1490                         return 4 + 3 * logs;
1491                 case FSFILT_OP_UNLINK:
1492                         return 3 * logs;
1493                 }
1494         } else {
1495                 int i;
1496                 int needed = 0;
1497                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1498                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1499                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1500                 for (i = 0; i < op; i++, fso++) {
1501                         int nblocks = fso->fso_bufcnt * blockpp;
1502                         int ndindirect = min(nblocks, addrpp + 1);
1503                         int nindir = nblocks + ndindirect + 1;
1504
1505                         needed += nindir;
1506                 }
1507                 return needed + 3 * logs;
1508         }
1509
1510         return 0;
1511 }
1512
1513 #ifdef HAVE_QUOTA_SUPPORT
1514 #define DQINFO_COPY(out, in)                    \
1515 do {                                            \
1516         Q_COPY(out, in, dqi_bgrace);            \
1517         Q_COPY(out, in, dqi_igrace);            \
1518         Q_COPY(out, in, dqi_flags);             \
1519         Q_COPY(out, in, dqi_valid);             \
1520 } while (0)
1521
1522 #define DQBLK_COPY(out, in)                     \
1523 do {                                            \
1524         Q_COPY(out, in, dqb_bhardlimit);        \
1525         Q_COPY(out, in, dqb_bsoftlimit);        \
1526         Q_COPY(out, in, dqb_curspace);          \
1527         Q_COPY(out, in, dqb_ihardlimit);        \
1528         Q_COPY(out, in, dqb_isoftlimit);        \
1529         Q_COPY(out, in, dqb_curinodes);         \
1530         Q_COPY(out, in, dqb_btime);             \
1531         Q_COPY(out, in, dqb_itime);             \
1532         Q_COPY(out, in, dqb_valid);             \
1533 } while (0)
1534
1535 static int fsfilt_ext3_quotactl(struct super_block *sb,
1536                                 struct obd_quotactl *oqc)
1537 {
1538         int i, rc = 0, error = 0;
1539         const struct quotactl_ops *qcop;
1540         struct if_dqinfo *info;
1541         struct if_dqblk *dqblk;
1542         ENTRY;
1543
1544         if (!sb->s_qcop)
1545                 RETURN(-ENOSYS);
1546
1547         OBD_ALLOC_PTR(info);
1548         if (!info)
1549                 RETURN(-ENOMEM);
1550         OBD_ALLOC_PTR(dqblk);
1551         if (!dqblk) {
1552                 OBD_FREE_PTR(info);
1553                 RETURN(-ENOMEM);
1554         }
1555
1556         DQINFO_COPY(info, &oqc->qc_dqinfo);
1557         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1558
1559         qcop = sb->s_qcop;
1560         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1561                 for (i = 0; i < MAXQUOTAS; i++) {
1562                         if (!Q_TYPESET(oqc, i))
1563                                 continue;
1564
1565                         if (oqc->qc_cmd == Q_QUOTAON) {
1566                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1567
1568                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1569
1570                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1571                                                  name[i], 0);
1572                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1573                                 rc = ll_quota_off(sb, i, 0);
1574                         }
1575
1576                         if (rc == -EBUSY)
1577                                 error = rc;
1578                         else if (rc)
1579                                 GOTO(out, rc);
1580                 }
1581                 GOTO(out, rc ?: error);
1582         }
1583
1584         switch (oqc->qc_cmd) {
1585         case Q_GETOINFO:
1586         case Q_GETINFO:
1587                 if (!qcop->get_info)
1588                         GOTO(out, rc = -ENOSYS);
1589                 rc = qcop->get_info(sb, oqc->qc_type, info);
1590                 break;
1591         case Q_SETQUOTA:
1592         case Q_INITQUOTA:
1593                 if (!qcop->set_dqblk)
1594                         GOTO(out, rc = -ENOSYS);
1595                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1596                 break;
1597         case Q_GETOQUOTA:
1598         case Q_GETQUOTA:
1599                 if (!qcop->get_dqblk)
1600                         GOTO(out, rc = -ENOSYS);
1601                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1602                 if (!rc)
1603                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1604                 break;
1605         case Q_SYNC:
1606                 if (!sb->s_qcop->quota_sync)
1607                         GOTO(out, rc = -ENOSYS);
1608                 qcop->quota_sync(sb, oqc->qc_type);
1609                 break;
1610         case Q_FINVALIDATE:
1611                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1612                 for (i = 0; i < MAXQUOTAS; i++) {
1613                         struct file *fp;
1614                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1615
1616                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1617
1618                         if (!Q_TYPESET(oqc, i))
1619                                 continue;
1620
1621                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1622                         if (IS_ERR(fp)) {
1623                                 rc = PTR_ERR(fp);
1624                                 CERROR("error invalidating operational quota file"
1625                                        " %s (rc:%d)\n", name[i], rc);
1626                         } else {
1627                                 filp_close(fp, 0);
1628                         }
1629
1630                 }
1631                 break;
1632         default:
1633                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1634                 LBUG();
1635         }
1636 out:
1637         DQINFO_COPY(&oqc->qc_dqinfo, info);
1638         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1639
1640         OBD_FREE_PTR(info);
1641         OBD_FREE_PTR(dqblk);
1642
1643         if (rc)
1644                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1645                                 "failed: %d\n",
1646                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1647         RETURN(rc);
1648 }
1649
1650 struct chk_dqblk{
1651         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1652         cfs_list_t              dqb_list;        /** in list also */
1653         qid_t                   dqb_id;          /** uid/gid */
1654         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1655         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1656         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1657         qsize_t                 dqb_curspace;    /** current space */
1658         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1659         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1660         qsize_t                 dqb_curinodes;   /** current inodes */
1661         obd_time                dqb_btime;       /** block grace time */
1662         obd_time                dqb_itime;       /** inode grace time */
1663         __u32                   dqb_valid;       /** flag for above fields */
1664 };
1665
1666 static inline unsigned int chkquot_hash(qid_t id, int type)
1667                                         __attribute__((__const__));
1668
1669 static inline unsigned int chkquot_hash(qid_t id, int type)
1670 {
1671         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1672 }
1673
1674 static inline struct chk_dqblk *
1675 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1676 {
1677         cfs_hlist_node_t *node;
1678         struct chk_dqblk *cdqb;
1679
1680         cfs_hlist_for_each(node, head) {
1681                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1682                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1683                         return cdqb;
1684         }
1685
1686         return NULL;
1687 }
1688
1689 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1690 {
1691         struct chk_dqblk *cdqb;
1692
1693         OBD_ALLOC_PTR(cdqb);
1694         if (cdqb) {
1695                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1696                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1697                 cdqb->dqb_id = id;
1698                 cdqb->dqb_type = type;
1699         }
1700
1701         return cdqb;
1702 }
1703
1704 static struct chk_dqblk *
1705 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1706       cfs_list_t *list, qid_t id, int type, int first_check)
1707 {
1708         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1709         struct if_dqblk dqb;
1710         struct chk_dqblk *cdqb;
1711         int rc;
1712
1713         cdqb = find_chkquot(head, id, type);
1714         if (cdqb)
1715                 return cdqb;
1716
1717         cdqb = alloc_chkquot(id, type);
1718         if (!cdqb)
1719                 return NULL;
1720
1721         if (!first_check) {
1722                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1723                 if (rc) {
1724                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1725                                id, type, rc);
1726                 } else {
1727                         DQBLK_COPY(cdqb, &dqb);
1728                         cdqb->dqb_curspace = 0;
1729                         cdqb->dqb_curinodes = 0;
1730                 }
1731         }
1732
1733         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1734         cfs_list_add_tail(&cdqb->dqb_list, list);
1735
1736         return cdqb;
1737 }
1738
1739 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1740 {
1741         struct obd_quotactl *oqctl;
1742         int rc;
1743
1744         OBD_ALLOC_PTR(oqctl);
1745         if (!oqctl)
1746                 RETURN(-ENOMEM);
1747
1748         oqctl->qc_cmd = cmd;
1749         oqctl->qc_id = qfmt;
1750         oqctl->qc_type = type;
1751         rc = fsfilt_ext3_quotactl(sb, oqctl);
1752
1753         OBD_FREE_PTR(oqctl);
1754         return rc;
1755 }
1756
1757 static inline int read_old_dqinfo(struct super_block *sb, int type,
1758                                   struct if_dqinfo *dqinfo)
1759 {
1760         struct obd_quotactl *oqctl;
1761         int rc;
1762         ENTRY;
1763
1764         OBD_ALLOC_PTR(oqctl);
1765         if (!oqctl)
1766                 RETURN(-ENOMEM);
1767
1768         oqctl->qc_cmd = Q_GETINFO;
1769         oqctl->qc_type = type;
1770         rc = fsfilt_ext3_quotactl(sb, oqctl);
1771         if (!rc)
1772                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1773
1774         OBD_FREE_PTR(oqctl);
1775         RETURN(rc);
1776 }
1777
1778 #ifndef HAVE_EXT4_LDISKFS
1779 static inline struct ext3_group_desc *
1780 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1781 {
1782         unsigned long desc_block, desc;
1783         struct ext3_group_desc *gdp;
1784
1785         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1786         desc = group % EXT3_DESC_PER_BLOCK(sb);
1787         gdp = (struct ext3_group_desc *)
1788               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1789
1790         return gdp + desc;
1791 }
1792
1793 static inline struct buffer_head *
1794 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1795 {
1796         struct ext3_group_desc *desc;
1797         struct buffer_head *bh;
1798
1799         desc = get_group_desc(sb, group, NULL);
1800         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1801         return bh;
1802 }
1803
1804 static __u32 ext3_itable_unused_count(struct super_block *sb,
1805                                struct ext3_group_desc *bg) {
1806        return le16_to_cpu(bg->bg_itable_unused);
1807 }
1808 #else
1809 #define get_group_desc ext3_get_group_desc
1810 #endif
1811
1812 struct qchk_ctxt {
1813         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1814         cfs_list_t              qckt_list;                 /* quotacheck list */
1815         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1816         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1817 };
1818
1819 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1820                            struct obd_quotactl *oqc)
1821 {
1822         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1823         loff_t size = 0;
1824         qid_t qid[MAXQUOTAS];
1825         int cnt, i, rc = 0;
1826
1827         if (!inode)
1828                 return 0;
1829
1830         qid[USRQUOTA] = inode->i_uid;
1831         qid[GRPQUOTA] = inode->i_gid;
1832
1833         if (S_ISDIR(inode->i_mode) ||
1834             S_ISREG(inode->i_mode) ||
1835             S_ISLNK(inode->i_mode))
1836                 size = inode_get_bytes(inode);
1837
1838         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1839                 if (!Q_TYPESET(oqc, cnt))
1840                         continue;
1841
1842                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1843                                 &qctxt->qckt_list, qid[cnt], cnt,
1844                                 qctxt->qckt_first_check[cnt]);
1845                 if (!cdqb[cnt]) {
1846                         rc = -ENOMEM;
1847                         break;
1848                 }
1849
1850                 cdqb[cnt]->dqb_curspace += size;
1851                 cdqb[cnt]->dqb_curinodes++;
1852         }
1853
1854         if (rc) {
1855                 for (i = 0; i < cnt; i++) {
1856                         if (!Q_TYPESET(oqc, i))
1857                                 continue;
1858                         LASSERT(cdqb[i]);
1859                         cdqb[i]->dqb_curspace -= size;
1860                         cdqb[i]->dqb_curinodes--;
1861                 }
1862         }
1863
1864         return rc;
1865 }
1866
1867 /* write dqinfo struct in a new quota file */
1868 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1869 {
1870         struct v2_disk_dqinfo dqinfo;
1871         __u32 blocks = V2_DQTREEOFF + 1;
1872         loff_t offset = V2_DQINFOOFF;
1873
1874         if (info) {
1875                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1876                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1877                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1878                                                ~DQF_INFO_DIRTY);
1879         } else {
1880                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1881                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1882                 dqinfo.dqi_flags = 0;
1883         }
1884
1885         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1886         dqinfo.dqi_free_blk = 0;
1887         dqinfo.dqi_free_entry = 0;
1888
1889         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1890 }
1891
1892 static int v3_write_dqheader(struct file *f, int type)
1893 {
1894         static const __u32 quota_magics[] = V2_INITQMAGICS;
1895         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1896         struct v2_disk_dqheader dqhead;
1897         loff_t offset = 0;
1898
1899         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1900         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1901
1902         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1903         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1904
1905         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1906 }
1907
1908 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1909                                   struct obd_quotactl *oqc)
1910 {
1911         int i, rc = 0;
1912         ENTRY;
1913
1914         for (i = 0; i < MAXQUOTAS; i++) {
1915                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1916                                          NULL : &qctxt->qckt_dqinfo[i];
1917                 struct file *file;
1918                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1919
1920                 if (!Q_TYPESET(oqc, i))
1921                         continue;
1922
1923                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1924
1925                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1926                 if (IS_ERR(file)) {
1927                         rc = PTR_ERR(file);
1928                         CERROR("can't create %s file: rc = %d\n",
1929                                name[i], rc);
1930                         GOTO(out, rc);
1931                 }
1932
1933                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1934                         CERROR("file %s is not regular", name[i]);
1935                         filp_close(file, 0);
1936                         GOTO(out, rc = -EINVAL);
1937                 }
1938
1939                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1940
1941                 rc = v3_write_dqheader(file, i);
1942                 if (rc) {
1943                         filp_close(file, 0);
1944                         GOTO(out, rc);
1945                 }
1946
1947                 rc = v3_write_dqinfo(file, i, info);
1948                 filp_close(file, 0);
1949                 if (rc)
1950                         GOTO(out, rc);
1951         }
1952
1953 out:
1954         RETURN(rc);
1955 }
1956
1957
1958 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1959                           struct chk_dqblk *cdqb)
1960 {
1961         struct obd_quotactl *oqc;
1962         long now;
1963         int rc;
1964         ENTRY;
1965
1966         OBD_ALLOC_PTR(oqc);
1967         if (!oqc)
1968                 RETURN(-ENOMEM);
1969
1970         now = cfs_time_current_sec();
1971
1972         if (cdqb->dqb_bsoftlimit &&
1973             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1974             !cdqb->dqb_btime)
1975                 cdqb->dqb_btime =
1976                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1977
1978         if (cdqb->dqb_isoftlimit &&
1979             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1980             !cdqb->dqb_itime)
1981                 cdqb->dqb_itime =
1982                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1983
1984         cdqb->dqb_valid = QIF_ALL;
1985
1986         oqc->qc_cmd = Q_SETQUOTA;
1987         oqc->qc_type = cdqb->dqb_type;
1988         oqc->qc_id = cdqb->dqb_id;
1989         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1990
1991         rc = fsfilt_ext3_quotactl(sb, oqc);
1992         OBD_FREE_PTR(oqc);
1993         RETURN(rc);
1994 }
1995
1996 static int prune_chkquots(struct super_block *sb,
1997                           struct qchk_ctxt *qctxt, int error)
1998 {
1999         struct chk_dqblk *cdqb, *tmp;
2000         int rc;
2001
2002         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
2003                 if (!error) {
2004                         rc = commit_chkquot(sb, qctxt, cdqb);
2005                         if (rc)
2006                                 error = rc;
2007                 }
2008                 cfs_hlist_del_init(&cdqb->dqb_hash);
2009                 cfs_list_del(&cdqb->dqb_list);
2010                 OBD_FREE_PTR(cdqb);
2011         }
2012
2013         return error;
2014 }
2015
2016 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2017 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2018 #endif
2019
2020 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2021                                   struct obd_quotactl *oqc)
2022 {
2023         struct ext3_sb_info *sbi = EXT3_SB(sb);
2024         int i, group, uninit_feat = 0;
2025         struct qchk_ctxt *qctxt;
2026         struct buffer_head *bitmap_bh = NULL;
2027         unsigned long ino, inode_inuse;
2028         struct inode *inode;
2029         int rc = 0;
2030         ENTRY;
2031
2032         /* turn on quota and read dqinfo if existed */
2033         OBD_ALLOC_PTR(qctxt);
2034         if (!qctxt) {
2035                 oqc->qc_stat = -ENOMEM;
2036                 RETURN(-ENOMEM);
2037         }
2038
2039         for (i = 0; i < NR_DQHASH; i++)
2040                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2041         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2042
2043         for (i = 0; i < MAXQUOTAS; i++) {
2044                 if (!Q_TYPESET(oqc, i))
2045                         continue;
2046
2047                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2048                 if (!rc || rc == -EBUSY) {
2049                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2050                         if (rc)
2051                                 GOTO(out, rc);
2052                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2053                         qctxt->qckt_first_check[i] = 1;
2054                 } else if (rc) {
2055                         GOTO(out, rc);
2056                 }
2057         }
2058         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2059                 /* This filesystem supports the uninit group feature */
2060                 uninit_feat = 1;
2061
2062         /* number of inodes that have been allocated */
2063         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2064                       percpu_counter_sum(&sbi->s_freeinodes_counter);
2065
2066         /* check quota and update in hash */
2067         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2068              group++) {
2069                 unsigned long used_count = sbi->s_inodes_per_group;
2070
2071                 if (uninit_feat) {
2072                         struct ext3_group_desc *desc;
2073                         desc = get_group_desc(sb, group, NULL);
2074                         if (!desc)
2075                                 GOTO(out, -EIO);
2076
2077                         /* we don't really need to take the group lock here,
2078                          * but it may be useful if one day we support online
2079                          * quotacheck */
2080 #ifdef HAVE_EXT4_LDISKFS
2081                         ext4_lock_group(sb, group);
2082 #else
2083                         spin_lock(sb_bgl_lock(sbi, group));
2084 #endif
2085                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2086                                 /* no inode in use in this group, just skip it */
2087 #ifdef HAVE_EXT4_LDISKFS
2088                                 ext3_unlock_group(sb, group);
2089 #else
2090                                 spin_unlock(sb_bgl_lock(sbi, group));
2091 #endif
2092                                 continue;
2093                         }
2094
2095                         used_count -= ext3_itable_unused_count(sb, desc);
2096 #ifdef HAVE_EXT4_LDISKFS
2097                         ext3_unlock_group(sb, group);
2098 #else
2099                         spin_unlock(sb_bgl_lock(sbi, group));
2100 #endif
2101                 }
2102
2103                 ino = group * sbi->s_inodes_per_group + 1;
2104                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2105                 if (!bitmap_bh) {
2106                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2107                                sb->s_id, group);
2108                         GOTO(out, -EIO);
2109                 }
2110
2111                 i = 0;
2112                 while (i < used_count &&
2113                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2114                                                used_count, i)) < used_count) {
2115                         inode_inuse--;
2116                         i++;
2117                         ino = i + group * sbi->s_inodes_per_group;
2118                         if (ino < sbi->s_first_ino)
2119                                 continue;
2120 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2121                         inode = ext3_iget(sb, ino);
2122 #else
2123                         inode = iget(sb, ino);
2124 #endif
2125                         if (!inode || IS_ERR(inode))
2126                                 continue;
2127
2128                         rc = add_inode_quota(inode, qctxt, oqc);
2129                         iput(inode);
2130                         if (rc) {
2131                                 brelse(bitmap_bh);
2132                                 GOTO(out, rc);
2133                         }
2134                 }
2135                 brelse(bitmap_bh);
2136         }
2137
2138         /* read old quota limits from old quota file. (only for the user
2139          * has limits but hasn't file) */
2140 #ifdef HAVE_QUOTA_SUPPORT
2141         for (i = 0; i < MAXQUOTAS; i++) {
2142                 cfs_list_t id_list;
2143                 struct dquot_id *dqid, *tmp;
2144
2145                 if (!Q_TYPESET(oqc, i))
2146                         continue;
2147
2148                 if (qctxt->qckt_first_check[i])
2149                         continue;
2150
2151
2152                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2153                 CFS_INIT_LIST_HEAD(&id_list);
2154 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2155                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2156 #else
2157                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2158 #endif
2159                 if (rc)
2160                         CERROR("read old limits failed. (rc:%d)\n", rc);
2161
2162                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2163                         cfs_list_del_init(&dqid->di_link);
2164
2165                         if (!rc)
2166                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2167                                       dqid->di_id, i,
2168                                       qctxt->qckt_first_check[i]);
2169                         OBD_FREE_PTR(dqid);
2170                 }
2171         }
2172 #endif
2173         /* turn off quota cause we are to dump chk_dqblk to files */
2174         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2175
2176         rc = create_new_quota_files(qctxt, oqc);
2177         if (rc)
2178                 GOTO(out, rc);
2179
2180         /* we use vfs functions to set dqblk, so turn quota on */
2181         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2182 out:
2183         /* dump and free chk_dqblk */
2184         rc = prune_chkquots(sb, qctxt, rc);
2185         OBD_FREE_PTR(qctxt);
2186
2187         /* turn off quota, `lfs quotacheck` will turn on when all
2188          * nodes quotacheck finish. */
2189         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2190
2191         oqc->qc_stat = rc;
2192         if (rc)
2193                 CERROR("quotacheck failed: rc = %d\n", rc);
2194
2195         RETURN(rc);
2196 }
2197
2198 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2199                                  int cmd)
2200 {
2201         int rc = 0;
2202         ENTRY;
2203
2204         if (lqi->qi_files[type] == NULL) {
2205                 CERROR("operate qinfo before it's enabled!\n");
2206                 RETURN(-ESRCH);
2207         }
2208
2209         switch (cmd) {
2210         case QFILE_CHK:
2211                 rc = lustre_check_quota_file(lqi, type);
2212                 break;
2213         case QFILE_RD_INFO:
2214                 rc = lustre_read_quota_info(lqi, type);
2215                 break;
2216         case QFILE_WR_INFO:
2217                 rc = lustre_write_quota_info(lqi, type);
2218                 break;
2219         case QFILE_INIT_INFO:
2220                 rc = lustre_init_quota_info(lqi, type);
2221                 break;
2222         case QFILE_CONVERT:
2223                 rc = -ENOTSUPP;
2224                 CERROR("quota CONVERT command is not supported\n");
2225                 break;
2226         default:
2227                 rc = -ENOTSUPP;
2228                 CERROR("Unsupported admin quota file cmd %d\n"
2229                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2230                        cmd);
2231                 break;
2232         }
2233         RETURN(rc);
2234 }
2235
2236 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2237                             cfs_list_t *list)
2238 {
2239         return lustre_get_qids(file, inode, type, list);
2240 }
2241
2242 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2243 {
2244         int rc = 0;
2245         ENTRY;
2246
2247         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2248                 CERROR("operate dquot before it's enabled!\n");
2249                 RETURN(-ESRCH);
2250         }
2251
2252         switch (cmd) {
2253         case QFILE_RD_DQUOT:
2254                 rc = lustre_read_dquot(dquot);
2255                 break;
2256         case QFILE_WR_DQUOT:
2257                 if (dquot->dq_dqb.dqb_ihardlimit ||
2258                     dquot->dq_dqb.dqb_isoftlimit ||
2259                     dquot->dq_dqb.dqb_bhardlimit ||
2260                     dquot->dq_dqb.dqb_bsoftlimit)
2261                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2262                 else
2263                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2264
2265                 rc = lustre_commit_dquot(dquot);
2266                 if (rc >= 0)
2267                         rc = 0;
2268                 break;
2269         default:
2270                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2271                 LBUG();
2272                 break;
2273         }
2274         RETURN(rc);
2275 }
2276
2277 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2278                                 struct inode *inode, int frags)
2279 {
2280 #ifdef EXT3_EXT_HAS_NO_TREE
2281         struct ext3_ext_base *base = inode;
2282 #else
2283         struct ext3_extents_tree tree;
2284         struct ext3_ext_base *base = &tree;
2285
2286         ext3_init_tree_desc(base, inode);
2287 #endif
2288         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2289          * metablocks at maxium b=16542 */
2290         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2291         return 0;
2292 }
2293
2294 #endif
2295
2296 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2297 {
2298         return (EXT3_SB(sb)->journal_bdev);
2299 }
2300 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2301
2302 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2303                           char *buf, int count, loff_t pos)
2304 {
2305         loff_t p = pos;
2306         int rc;
2307
2308         if (!f && !inode) {
2309                 CERROR("lustre_read_quota failed for no quota file!\n");
2310                 libcfs_debug_dumpstack(NULL);
2311                 return -EINVAL;
2312         }
2313
2314         /* Support for both adm and op quota files must be provided */
2315         if (f) {
2316                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2317                 rc = rc < 0 ? rc : p - pos;
2318         } else {
2319                 struct super_block *sb = inode->i_sb;
2320                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2321         }
2322         return rc;
2323 }
2324
2325 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2326 {
2327         loff_t p = pos;
2328         int rc;
2329
2330         /* Only adm quota files are supported, op updates are handled by vfs */
2331         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2332         rc = rc < 0 ? rc : p - pos;
2333
2334         return rc;
2335 }
2336
2337 void *lustre_quota_journal_start(struct inode *inode, int delete)
2338 {
2339         handle_t *handle;
2340         unsigned block_count;
2341
2342         if (delete) {
2343                 /* each indirect block (+4) may become free, attaching to the
2344                  * header list of free blocks (+1); the data block (+1) may
2345                  * become a free block (+0) or a block with free dqentries (+0) */
2346                 block_count = (4 + 1) + 1;
2347                 handle = ext3_journal_start(inode,
2348                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2349         } else {
2350                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2351                  * freeblk reusage with a header update (+1); dqentry is either reused
2352                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2353                  * a new block allocation (+1) with a header update (+1)              */
2354                 block_count = (4 + 1) + 3;
2355                 handle = ext3_journal_start(inode,
2356                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2357
2358         }
2359
2360         return handle;
2361 }
2362
2363 void lustre_quota_journal_stop(void *handle)
2364 {
2365         ext3_journal_stop((handle_t *)handle);
2366 }
2367
2368 static int ll_decode_fh_accept(void *context, struct dentry *de)
2369 {
2370         return 1;
2371 }
2372
2373 #ifdef HAVE_EXPORTFS_DECODE_FH
2374 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2375          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2376                             acceptable, context)
2377 #else
2378 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2379          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2380                                      type, acceptable, context)
2381 # define FILEID_INO32_GEN 1
2382 extern struct export_operations export_op_default;
2383 #endif
2384
2385 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2386                                       struct fsfilt_fid *fid, int ignore_gen)
2387 {
2388         struct inode  *inode;
2389         struct dentry *result;
2390
2391         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2392                                        ll_decode_fh_accept, NULL);
2393         if (IS_ERR(result)) {
2394                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2395                        fid->ino, fid->gen, PTR_ERR(result));
2396                 return result;
2397         }
2398
2399         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2400                fid->ino, fid->gen);
2401         inode = result->d_inode;
2402         if (inode == NULL)
2403                 goto err_out;
2404
2405         if (inode->i_nlink == 0 &&
2406             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2407                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2408                               " ctime -- this may indicate disk "
2409                               "corruption (inode: %lu, link: %lu, "
2410                               "count: %d)\n", inode->i_ino,
2411                               (unsigned long)inode->i_nlink,
2412                               atomic_read(&inode->i_count));
2413                 goto err_out;
2414         }
2415         if (fid->gen && inode->i_generation != fid->gen) {
2416                 /* we didn't find the right inode.. */
2417                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2418                        "count: %d, generation %u/%u\n",
2419                        inode->i_ino, (unsigned long)inode->i_nlink,
2420                        atomic_read(&inode->i_count), inode->i_generation,
2421                        fid->gen);
2422                 goto err_out;
2423         }
2424
2425         return result;
2426 err_out:
2427         l_dput(result);
2428         return ERR_PTR(-ENOENT);
2429 }
2430
2431 static struct fsfilt_operations fsfilt_ext3_ops = {
2432         .fs_type                = "ext3",
2433         .fs_owner               = THIS_MODULE,
2434         .fs_getlabel            = fsfilt_ext3_get_label,
2435         .fs_setlabel            = fsfilt_ext3_set_label,
2436         .fs_uuid                = fsfilt_ext3_uuid,
2437         .fs_start               = fsfilt_ext3_start,
2438         .fs_brw_start           = fsfilt_ext3_brw_start,
2439         .fs_extend              = fsfilt_ext3_extend,
2440         .fs_commit              = fsfilt_ext3_commit,
2441         .fs_commit_async        = fsfilt_ext3_commit_async,
2442         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2443         .fs_setattr             = fsfilt_ext3_setattr,
2444         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2445         .fs_set_md              = fsfilt_ext3_set_md,
2446         .fs_get_md              = fsfilt_ext3_get_md,
2447         .fs_readpage            = fsfilt_ext3_readpage,
2448         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2449         .fs_statfs              = fsfilt_ext3_statfs,
2450         .fs_sync                = fsfilt_ext3_sync,
2451         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2452         .fs_write_record        = fsfilt_ext3_write_record,
2453         .fs_read_record         = fsfilt_ext3_read_record,
2454         .fs_setup               = fsfilt_ext3_setup,
2455         .fs_send_bio            = fsfilt_ext3_send_bio,
2456         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2457 #ifdef HAVE_DISK_INODE_VERSION
2458         .fs_get_version         = fsfilt_ext3_get_version,
2459         .fs_set_version         = fsfilt_ext3_set_version,
2460 #endif
2461 #ifdef HAVE_QUOTA_SUPPORT
2462         .fs_quotactl            = fsfilt_ext3_quotactl,
2463         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2464         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2465         .fs_qids                = fsfilt_ext3_qids,
2466         .fs_dquot               = fsfilt_ext3_dquot,
2467         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2468 #endif
2469         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2470         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2471 };
2472
2473 static int __init fsfilt_ext3_init(void)
2474 {
2475         int rc;
2476
2477         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2478                                          sizeof(struct fsfilt_cb_data), 0, 0);
2479         if (!fcb_cache) {
2480                 CERROR("error allocating fsfilt journal callback cache\n");
2481                 GOTO(out, rc = -ENOMEM);
2482         }
2483
2484         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2485
2486         if (rc) {
2487                 int err = cfs_mem_cache_destroy(fcb_cache);
2488                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2489         }
2490 out:
2491         return rc;
2492 }
2493
2494 static void __exit fsfilt_ext3_exit(void)
2495 {
2496         int rc;
2497
2498         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2499         rc = cfs_mem_cache_destroy(fcb_cache);
2500         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2501 }
2502
2503 module_init(fsfilt_ext3_init);
2504 module_exit(fsfilt_ext3_exit);
2505
2506 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2507 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2508 MODULE_LICENSE("GPL");