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  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/fsfilt_ext3.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/init.h>
  44 #include <linux/module.h>
  45 #include <linux/fs.h>
  46 #include <linux/jbd.h>
  47 #include <linux/slab.h>
  48 #include <linux/pagemap.h>
  49 #include <linux/quotaops.h>
  50 #include <linux/ext3_fs.h>
  51 #include <linux/ext3_jbd.h>
  52 #include <linux/version.h>
  53 #include <linux/bitops.h>
  54 #include <linux/quota.h>
  55 #include <linux/quotaio_v1.h>
  56 #include <linux/quotaio_v2.h>
  57 #include <linux/parser.h>
  58 #include <ext3/xattr.h>
  59
  60 #include <libcfs/kp30.h>
  61 #include <lustre_fsfilt.h>
  62 #include <obd.h>
  63 #include <lustre_quota.h>
  64 #include <linux/lustre_compat25.h>
  65 #include <linux/lprocfs_status.h>
  66
  67 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  68 #include <linux/ext3_extents.h>
  69 #endif
  70
  71 #include "lustre_quota_fmt.h"
  72
  73 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  74 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  75 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  76 #else
  77 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  78 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  79 #endif
  80
  81 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
  82 /* for kernels 2.6.18 and later */
  83 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  84 #else
  85 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
  86 #endif
  87
  88 static cfs_mem_cache_t *fcb_cache;
  89
  90 struct fsfilt_cb_data {
  91         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  92         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  93         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  94         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  95         void *cb_data;                  /* MDS/OST completion function data */
  96 };
  97
  98 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
  99 #define EXT3_XATTR_INDEX_TRUSTED        4
 100 #endif
 101 #ifndef XATTR_NO_CTIME
 102 #define XATTR_NO_CTIME 0x80
 103 #endif
 104
 105 static char *fsfilt_ext3_get_label(struct super_block *sb)
 106 {
 107         return EXT3_SB(sb)->s_es->s_volume_name;
 108 }
 109
 110 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 111 {
 112         /* see e.g. fsfilt_ext3_write_record() */
 113         journal_t *journal;
 114         handle_t *handle;
 115         int err;
 116
 117         journal = EXT3_SB(sb)->s_journal;
 118         handle = journal_start(journal, 1);
 119         if (IS_ERR(handle)) {
 120                 CERROR("can't start transaction\n");
 121                 return(PTR_ERR(handle));
 122         }
 123
 124         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 125         if (err)
 126                 goto out;
 127
 128         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 129                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 130
 131         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 132
 133 out:
 134         journal_stop(handle);
 135
 136         return(err);
 137 }
 138
 139 static char *fsfilt_ext3_uuid(struct super_block *sb)
 140 {
 141         return EXT3_SB(sb)->s_es->s_uuid;
 142 }
 143
 144 #ifdef HAVE_DISK_INODE_VERSION
 145 /*
 146  * Get the 64-bit version for an inode.
 147  */
 148 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 149 {
 150         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 151                EXT3_I(inode)->i_fs_version, inode->i_ino);
 152         return EXT3_I(inode)->i_fs_version;
 153 }
 154
 155 /*
 156  * Set the 64-bit version and return the old version.
 157  */
 158 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 159 {
 160         __u64 old_version = EXT3_I(inode)->i_fs_version;
 161
 162         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 163                new_version, old_version, inode->i_ino);
 164         (EXT3_I(inode))->i_fs_version = new_version;
 165         /* version is set after all inode operations are finished, so we should
 166          * mark it dirty here */
 167         inode->i_sb->s_op->dirty_inode(inode);
 168         return old_version;
 169 }
 170
 171 #endif
 172
 173 /*
 174  * We don't currently need any additional blocks for rmdir and
 175  * unlink transactions because we are storing the OST oa_id inside
 176  * the inode (which we will be changing anyways as part of this
 177  * transaction).
 178  */
 179 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 180                                int logs)
 181 {
 182         /* For updates to the last received file */
 183         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 184         journal_t *journal;
 185         void *handle;
 186
 187         if (current->journal_info) {
 188                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 189                        current->journal_info);
 190                 goto journal_start;
 191         }
 192
 193         switch(op) {
 194         case FSFILT_OP_RMDIR:
 195         case FSFILT_OP_UNLINK:
 196                 /* delete one file + create/update logs for each stripe */
 197                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 198                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 199                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 200                 break;
 201         case FSFILT_OP_RENAME:
 202                 /* modify additional directory */
 203                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 204                 /* no break */
 205         case FSFILT_OP_SYMLINK:
 206                 /* additional block + block bitmap + GDT for long symlink */
 207                 nblocks += 3;
 208                 /* no break */
 209         case FSFILT_OP_CREATE: {
 210 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 211                 static int warned;
 212                 if (!warned) {
 213                         if (!test_opt(inode->i_sb, EXTENTS)) {
 214                                 warned = 1;
 215                         } else if (((EXT3_I(inode)->i_flags &
 216                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 217                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 218                                 CWARN("extent-mapped directory found - contact "
 219                                       "http://bugzilla.lustre.org/\n");
 220                                 warned = 1;
 221                         }
 222                 }
 223 #endif
 224                 /* no break */
 225         }
 226         case FSFILT_OP_MKDIR:
 227         case FSFILT_OP_MKNOD:
 228                 /* modify one inode + block bitmap + GDT */
 229                 nblocks += 3;
 230                 /* no break */
 231         case FSFILT_OP_LINK:
 232                 /* modify parent directory */
 233                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 234                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 235                 /* create/update logs for each stripe */
 236                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 237                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 238                 break;
 239         case FSFILT_OP_SETATTR:
 240                 /* Setattr on inode */
 241                 nblocks += 1;
 242                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 243                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 244                 /* quota chown log for each stripe */
 245                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 246                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 247                 break;
 248         case FSFILT_OP_CANCEL_UNLINK:
 249                 /* blocks for log header bitmap update OR
 250                  * blocks for catalog header bitmap update + unlink of logs */
 251                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 252                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 253                 break;
 254         case FSFILT_OP_JOIN:
 255                 /* delete 2 file(file + array id) + create 1 file (array id)
 256                  * create/update logs for each stripe */
 257                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 258
 259                 /*create array log for head file*/
 260                 nblocks += 3;
 261                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 262                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
 263                 /*update head file array */
 264                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 265                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 266                 break;
 267         default: CERROR("unknown transaction start op %d\n", op);
 268                 LBUG();
 269         }
 270
 271         LASSERT(current->journal_info == desc_private);
 272         journal = EXT3_SB(inode->i_sb)->s_journal;
 273         if (nblocks > journal->j_max_transaction_buffers) {
 274                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 275                       nblocks, op, logs, journal->j_max_transaction_buffers);
 276                 nblocks = journal->j_max_transaction_buffers;
 277         }
 278
 279  journal_start:
 280         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 281         handle = ext3_journal_start(inode, nblocks);
 282
 283         if (!IS_ERR(handle))
 284                 LASSERT(current->journal_info == handle);
 285         else
 286                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 287                        op, nblocks, PTR_ERR(handle));
 288         return handle;
 289 }
 290
 291 /*
 292  * Calculate the number of buffer credits needed to write multiple pages in
 293  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 294  * doesn't have a nice API for calculating this sort of thing in advance.
 295  *
 296  * See comment above ext3_writepage_trans_blocks for details.  We assume
 297  * no data journaling is being done, but it does allow for all of the pages
 298  * being non-contiguous.  If we are guaranteed contiguous pages we could
 299  * reduce the number of (d)indirect blocks a lot.
 300  *
 301  * With N blocks per page and P pages, for each inode we have at most:
 302  * N*P indirect
 303  * min(N*P, blocksize/4 + 1) dindirect blocks
 304  * niocount tindirect
 305  *
 306  * For the entire filesystem, we have at most:
 307  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 308  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 309  * objcount inode blocks
 310  * 1 superblock
 311  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 312  *
 313  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 314  */
 315 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 316                                       int niocount, struct niobuf_local *nb)
 317 {
 318         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 319         __u64 next_indir;
 320         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 321         int nbitmaps = 0, ngdblocks;
 322         int needed = objcount + 1; /* inodes + superblock */
 323         int i, j;
 324
 325         for (i = 0, j = 0; i < objcount; i++, fso++) {
 326                 /* two or more dindirect blocks in case we cross boundary */
 327                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 328                                     nb[j].offset) >>
 329                                    sb->s_blocksize_bits) /
 330                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 331                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 332
 333                 /* leaf, indirect, tindirect blocks for first block */
 334                 nbitmaps += blockpp + 2;
 335
 336                 j += fso->fso_bufcnt;
 337         }
 338
 339         next_indir = nb[0].offset +
 340                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 341         for (i = 1; i < niocount; i++) {
 342                 if (nb[i].offset >= next_indir) {
 343                         nbitmaps++;     /* additional indirect */
 344                         next_indir = nb[i].offset +
 345                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 346                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 347                         nbitmaps++;     /* additional indirect */
 348                 }
 349                 nbitmaps += blockpp;    /* each leaf in different group? */
 350         }
 351
 352         ngdblocks = nbitmaps;
 353         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 354                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 355         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 356                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 357
 358         needed += nbitmaps + ngdblocks;
 359
 360         /* last_rcvd update */
 361         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 362
 363 #if defined(CONFIG_QUOTA)
 364         /* We assume that there will be 1 bit set in s_dquot.flags for each
 365          * quota file that is active.  This is at least true for now.
 366          */
 367         needed += hweight32(sb_any_quota_enabled(sb)) *
 368                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 369 #endif
 370
 371         return needed;
 372 }
 373
 374 /* We have to start a huge journal transaction here to hold all of the
 375  * metadata for the pages being written here.  This is necessitated by
 376  * the fact that we do lots of prepare_write operations before we do
 377  * any of the matching commit_write operations, so even if we split
 378  * up to use "smaller" transactions none of them could complete until
 379  * all of them were opened.  By having a single journal transaction,
 380  * we eliminate duplicate reservations for common blocks like the
 381  * superblock and group descriptors or bitmaps.
 382  *
 383  * We will start the transaction here, but each prepare_write will
 384  * add a refcount to the transaction, and each commit_write will
 385  * remove a refcount.  The transaction will be closed when all of
 386  * the pages have been written.
 387  */
 388 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 389                                    int niocount, struct niobuf_local *nb,
 390                                    void *desc_private, int logs)
 391 {
 392         journal_t *journal;
 393         handle_t *handle;
 394         int needed;
 395         ENTRY;
 396
 397         LASSERT(current->journal_info == desc_private);
 398         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 399         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 400
 401         /* The number of blocks we could _possibly_ dirty can very large.
 402          * We reduce our request if it is absurd (and we couldn't get that
 403          * many credits for a single handle anyways).
 404          *
 405          * At some point we have to limit the size of I/Os sent at one time,
 406          * increase the size of the journal, or we have to calculate the
 407          * actual journal requirements more carefully by checking all of
 408          * the blocks instead of being maximally pessimistic.  It remains to
 409          * be seen if this is a real problem or not.
 410          */
 411         if (needed > journal->j_max_transaction_buffers) {
 412                 CERROR("want too many journal credits (%d) using %d instead\n",
 413                        needed, journal->j_max_transaction_buffers);
 414                 needed = journal->j_max_transaction_buffers;
 415         }
 416
 417         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 418         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 419         if (IS_ERR(handle)) {
 420                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 421                        PTR_ERR(handle));
 422         } else {
 423                 LASSERT(handle->h_buffer_credits >= needed);
 424                 LASSERT(current->journal_info == handle);
 425         }
 426
 427         RETURN(handle);
 428 }
 429
 430 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 431 {
 432        handle_t *handle = h;
 433
 434        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 435        if (nblocks == 0) {
 436                handle->h_buffer_credits = 0;
 437                CWARN("setting credits of handle %p to zero by request\n", h);
 438        }
 439
 440        if (handle->h_buffer_credits > nblocks)
 441                 return 0;
 442        if (journal_extend(handle, nblocks) == 0)
 443                 return 0;
 444
 445        ext3_mark_inode_dirty(handle, inode);
 446        return journal_restart(handle, nblocks);
 447 }
 448
 449 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 450 {
 451         int rc;
 452         handle_t *handle = h;
 453
 454         LASSERT(current->journal_info == handle);
 455         if (force_sync)
 456                 handle->h_sync = 1; /* recovery likes this */
 457
 458         rc = ext3_journal_stop(handle);
 459
 460         return rc;
 461 }
 462
 463 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 464                                     void **wait_handle)
 465 {
 466         unsigned long tid;
 467         transaction_t *transaction;
 468         handle_t *handle = h;
 469         journal_t *journal;
 470         int rc;
 471
 472         LASSERT(current->journal_info == handle);
 473
 474         transaction = handle->h_transaction;
 475         journal = transaction->t_journal;
 476         tid = transaction->t_tid;
 477         /* we don't want to be blocked */
 478         handle->h_sync = 0;
 479         rc = ext3_journal_stop(handle);
 480         if (rc) {
 481                 CERROR("error while stopping transaction: %d\n", rc);
 482                 return rc;
 483         }
 484         log_start_commit(journal, tid);
 485
 486         *wait_handle = (void *) tid;
 487         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 488         return 0;
 489 }
 490
 491 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 492 {
 493         journal_t *journal = EXT3_JOURNAL(inode);
 494         tid_t tid = (tid_t)(long)h;
 495
 496         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 497         if (unlikely(is_journal_aborted(journal)))
 498                 return -EIO;
 499
 500         log_wait_commit(EXT3_JOURNAL(inode), tid);
 501
 502         if (unlikely(is_journal_aborted(journal)))
 503                 return -EIO;
 504         return 0;
 505 }
 506
 507 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 508                                struct iattr *iattr, int do_trunc)
 509 {
 510         struct inode *inode = dentry->d_inode;
 511         int rc = 0;
 512
 513         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 514          * We are already writing the inode to disk as part of this
 515          * transaction and want to avoid a lot of extra inode writeout
 516          * later on. b=9828 */
 517         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 518                 /* ATTR_SIZE would invoke truncate: clear it */
 519                 iattr->ia_valid &= ~ATTR_SIZE;
 520                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 521                 i_size_write(inode, iattr->ia_size);
 522
 523                 if (iattr->ia_valid & ATTR_UID)
 524                         inode->i_uid = iattr->ia_uid;
 525                 if (iattr->ia_valid & ATTR_GID)
 526                         inode->i_gid = iattr->ia_gid;
 527                 if (iattr->ia_valid & ATTR_ATIME)
 528                         inode->i_atime = iattr->ia_atime;
 529                 if (iattr->ia_valid & ATTR_MTIME)
 530                         inode->i_mtime = iattr->ia_mtime;
 531                 if (iattr->ia_valid & ATTR_CTIME)
 532                         inode->i_ctime = iattr->ia_ctime;
 533                 if (iattr->ia_valid & ATTR_MODE) {
 534                         inode->i_mode = iattr->ia_mode;
 535
 536                         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 537                                 inode->i_mode &= ~S_ISGID;
 538                 }
 539
 540                 inode->i_sb->s_op->dirty_inode(inode);
 541
 542                 goto out;
 543         }
 544
 545         /* Don't allow setattr to change file type */
 546         if (iattr->ia_valid & ATTR_MODE)
 547                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 548                                  (iattr->ia_mode & ~S_IFMT);
 549
 550         /* We set these flags on the client, but have already checked perms
 551          * so don't confuse inode_change_ok. */
 552         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 553
 554         if (inode->i_op->setattr) {
 555                 rc = inode->i_op->setattr(dentry, iattr);
 556         } else {
 557                 rc = inode_change_ok(inode, iattr);
 558                 if (!rc)
 559                         rc = inode_setattr(inode, iattr);
 560         }
 561
 562  out:
 563         RETURN(rc);
 564 }
 565
 566 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 567                                  unsigned int cmd, unsigned long arg)
 568 {
 569         int rc = 0;
 570         ENTRY;
 571
 572         /* FIXME: Can't do this because of nested transaction deadlock */
 573         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 574                 CERROR("can't set data journal flag on file\n");
 575                 RETURN(-EPERM);
 576         }
 577
 578         if (inode->i_fop->ioctl)
 579                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 580         else
 581                 RETURN(-ENOTTY);
 582
 583         RETURN(rc);
 584 }
 585
 586 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 587                               void *lmm, int lmm_size, const char *name)
 588 {
 589         int rc;
 590
 591         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 592
 593         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 594                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 595
 596         if (rc && rc != -EROFS)
 597                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 598                        inode->i_ino, rc);
 599         return rc;
 600 }
 601
 602 /* Must be called with i_mutex held */
 603 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 604                               const char *name)
 605 {
 606         int rc;
 607
 608         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 609
 610         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 611                             name, lmm, lmm_size);
 612         /* This gives us the MD size */
 613         if (lmm == NULL)
 614                 return (rc == -ENODATA) ? 0 : rc;
 615
 616         if (rc < 0) {
 617                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 618                        EXT3_XATTR_INDEX_TRUSTED, name,
 619                        inode->i_ino, rc);
 620                 memset(lmm, 0, lmm_size);
 621                 return (rc == -ENODATA) ? 0 : rc;
 622         }
 623
 624         return rc;
 625 }
 626
 627 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 628 {
 629         submit_bio(rw, bio);
 630         return 0;
 631 }
 632
 633 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 634                                     loff_t *off)
 635 {
 636         struct inode *inode = file->f_dentry->d_inode;
 637         int rc = 0;
 638
 639         if (S_ISREG(inode->i_mode))
 640                 rc = file->f_op->read(file, buf, count, off);
 641         else {
 642                 const int blkbits = inode->i_sb->s_blocksize_bits;
 643                 const int blksize = inode->i_sb->s_blocksize;
 644
 645                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 646                        count, inode->i_ino, *off);
 647                 while (count > 0) {
 648                         struct buffer_head *bh;
 649
 650                         bh = NULL;
 651                         if (*off < i_size_read(inode)) {
 652                                 int err = 0;
 653
 654                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 655                                                 0, &err);
 656
 657                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 658
 659                                 if (bh) {
 660                                         memcpy(buf, bh->b_data, blksize);
 661                                         brelse(bh);
 662                                 } else if (err) {
 663                                         /* XXX in theory we should just fake
 664                                          * this buffer and continue like ext3,
 665                                          * especially if this is a partial read
 666                                          */
 667                                         CERROR("error read dir %lu+%llu: %d\n",
 668                                                inode->i_ino, *off, err);
 669                                         RETURN(err);
 670                                 }
 671                         }
 672                         if (!bh) {
 673                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 674
 675                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 676                                 memset(fake, 0, sizeof(*fake));
 677                                 fake->rec_len = cpu_to_le16(blksize);
 678                         }
 679                         count -= blksize;
 680                         buf += blksize;
 681                         *off += blksize;
 682                         rc += blksize;
 683                 }
 684         }
 685
 686         return rc;
 687 }
 688
 689 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 690 {
 691         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 692
 693         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 694
 695         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 696 }
 697
 698 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 699                                       void *handle, fsfilt_cb_t cb_func,
 700                                       void *cb_data)
 701 {
 702         struct fsfilt_cb_data *fcb;
 703
 704         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 705         if (fcb == NULL)
 706                 RETURN(-ENOMEM);
 707
 708         fcb->cb_func = cb_func;
 709         fcb->cb_obd = obd;
 710         fcb->cb_last_rcvd = last_rcvd;
 711         fcb->cb_data = cb_data;
 712
 713         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 714         journal_callback_set(handle, fsfilt_ext3_cb_func,
 715                              (struct journal_callback *)fcb);
 716
 717         return 0;
 718 }
 719
 720 /*
 721  * We need to hack the return value for the free inode counts because
 722  * the current EA code requires one filesystem block per inode with EAs,
 723  * so it is possible to run out of blocks before we run out of inodes.
 724  *
 725  * This can be removed when the ext3 EA code is fixed.
 726  */
 727 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 728 {
 729         struct kstatfs sfs;
 730         int rc;
 731
 732         memset(&sfs, 0, sizeof(sfs));
 733         rc = ll_do_statfs(sb,&sfs);
 734
 735         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 736                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 737                 sfs.f_ffree = sfs.f_bfree;
 738         }
 739
 740         statfs_pack(osfs, &sfs);
 741         return rc;
 742 }
 743
 744 static int fsfilt_ext3_sync(struct super_block *sb)
 745 {
 746         return ext3_force_commit(sb);
 747 }
 748
 749 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 750 #warning "kernel code has old extents/mballoc patch, disabling"
 751 #undef EXT3_MULTIBLOCK_ALLOCATOR
 752 #endif
 753 #ifndef EXT3_EXTENTS_FL
 754 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 755 #endif
 756
 757 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 758 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 759 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 760 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 761 #else
 762 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 763 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 764 #endif
 765
 766 #ifndef EXT_ASSERT
 767 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 768 #endif
 769
 770 #ifdef EXT3_EXT_HAS_NO_TREE
 771 /* for kernels 2.6.18 and later */
 772 #define ext3_ext_base                   inode
 773 #define ext3_ext_base2inode(inode)      (inode)
 774 #define EXT_DEPTH(inode)                ext_depth(inode)
 775 #define EXT_GENERATION(inode)           ext_generation(inode)
 776 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 777                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 778 #else
 779 #define ext3_ext_base                   ext3_extents_tree
 780 #define ext3_ext_base2inode(tree)       (tree->inode)
 781 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 782                         ext3_ext_walk_space(tree, block, num, cb);
 783 #endif
 784
 785 #include <linux/lustre_version.h>
 786 #if EXT3_EXT_MAGIC == 0xf301
 787 #define ee_start e_start
 788 #define ee_block e_block
 789 #define ee_len   e_num
 790 #endif
 791 #ifndef EXT3_BB_MAX_BLOCKS
 792 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 793         ext3_new_blocks(handle, inode, count, goal, err)
 794 #endif
 795
 796 struct bpointers {
 797         unsigned long *blocks;
 798         int *created;
 799         unsigned long start;
 800         int num;
 801         int init_num;
 802         int create;
 803 };
 804
 805 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 806                               unsigned long block, int *aflags)
 807 {
 808         struct ext3_inode_info *ei = EXT3_I(inode);
 809         unsigned long bg_start;
 810         unsigned long colour;
 811         int depth;
 812
 813         if (path) {
 814                 struct ext3_extent *ex;
 815                 depth = path->p_depth;
 816
 817                 /* try to predict block placement */
 818                 if ((ex = path[depth].p_ext)) {
 819 #if 0
 820                         /* This prefers to eat into a contiguous extent
 821                          * rather than find an extent that the whole
 822                          * request will fit into.  This can fragment data
 823                          * block allocation and prevents our lovely 1M I/Os
 824                          * from reaching the disk intact. */
 825                         if (ex->ee_block + ex->ee_len == block)
 826                                 *aflags |= 1;
 827 #endif
 828                         return ex->ee_start + (block - ex->ee_block);
 829                 }
 830
 831                 /* it looks index is empty
 832                  * try to find starting from index itself */
 833                 if (path[depth].p_bh)
 834                         return path[depth].p_bh->b_blocknr;
 835         }
 836
 837         /* OK. use inode's group */
 838         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 839                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 840         colour = (current->pid % 16) *
 841                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 842         return bg_start + colour + block;
 843 }
 844
 845 #define ll_unmap_underlying_metadata(sb, blocknr) \
 846         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 847
 848 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 849 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 850                                 struct ext3_ext_path *path, unsigned long block,
 851                                 unsigned long *count, int *err)
 852 {
 853         unsigned long pblock, goal;
 854         int aflags = 0;
 855         struct inode *inode = ext3_ext_base2inode(base);
 856
 857         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 858         aflags |= 2; /* block have been already reserved */
 859         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 860         return pblock;
 861
 862 }
 863 #else
 864 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 865                                 struct ext3_ext_path *path, unsigned long block,
 866                                 unsigned long *count, int *err)
 867 {
 868         struct inode *inode = ext3_ext_base2inode(base);
 869         struct ext3_allocation_request ar;
 870         unsigned long pblock;
 871         int aflags;
 872
 873         /* find neighbour allocated blocks */
 874         ar.lleft = block;
 875         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 876         if (*err)
 877                 return 0;
 878         ar.lright = block;
 879         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 880         if (*err)
 881                 return 0;
 882
 883         /* allocate new block */
 884         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 885         ar.inode = inode;
 886         ar.logical = block;
 887         ar.len = *count;
 888         ar.flags = EXT3_MB_HINT_DATA;
 889         pblock = ext3_mb_new_blocks(handle, &ar, err);
 890         *count = ar.len;
 891         return pblock;
 892 }
 893 #endif
 894
 895 #ifdef EXT3_EXT_HAS_NO_TREE
 896 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 897                                   struct ext3_ext_path *path,
 898                                   struct ext3_ext_cache *cex,
 899 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 900                                   struct ext3_extent *ex,
 901 #endif
 902                                   void *cbdata)
 903 {
 904         struct bpointers *bp = cbdata;
 905 #else
 906 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 907                                   struct ext3_ext_path *path,
 908                                   struct ext3_ext_cache *cex
 909 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 910                                   , struct ext3_extent *ex
 911 #endif
 912                                  )
 913 {
 914         struct bpointers *bp = base->private;
 915 #endif
 916         struct inode *inode = ext3_ext_base2inode(base);
 917         struct ext3_extent nex;
 918         unsigned long pblock;
 919         unsigned long tgen;
 920         int err, i;
 921         unsigned long count;
 922         handle_t *handle;
 923
 924         i = EXT_DEPTH(base);
 925         EXT_ASSERT(i == path->p_depth);
 926         EXT_ASSERT(path[i].p_hdr);
 927
 928         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 929                 err = EXT_CONTINUE;
 930                 goto map;
 931         }
 932
 933         if (bp->create == 0) {
 934                 i = 0;
 935                 if (cex->ec_block < bp->start)
 936                         i = bp->start - cex->ec_block;
 937                 if (i >= cex->ec_len)
 938                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 939                                         i, cex->ec_len);
 940                 for (; i < cex->ec_len && bp->num; i++) {
 941                         *(bp->created) = 0;
 942                         bp->created++;
 943                         *(bp->blocks) = 0;
 944                         bp->blocks++;
 945                         bp->num--;
 946                         bp->start++;
 947                 }
 948
 949                 return EXT_CONTINUE;
 950         }
 951
 952         tgen = EXT_GENERATION(base);
 953         count = ext3_ext_calc_credits_for_insert(base, path);
 954         ext3_up_truncate_sem(inode);
 955
 956         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 957         if (IS_ERR(handle)) {
 958                 ext3_down_truncate_sem(inode);
 959                 return PTR_ERR(handle);
 960         }
 961
 962         ext3_down_truncate_sem(inode);
 963         if (tgen != EXT_GENERATION(base)) {
 964                 /* the tree has changed. so path can be invalid at moment */
 965                 ext3_journal_stop(handle);
 966                 return EXT_REPEAT;
 967         }
 968
 969         count = cex->ec_len;
 970         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 971         if (!pblock)
 972                 goto out;
 973         EXT_ASSERT(count <= cex->ec_len);
 974
 975         /* insert new extent */
 976         nex.ee_block = cex->ec_block;
 977         nex.ee_start = pblock;
 978         nex.ee_len = count;
 979         err = ext3_ext_insert_extent(handle, base, path, &nex);
 980         if (err) {
 981                 /* free data blocks we just allocated */
 982                 /* not a good idea to call discard here directly,
 983                  * but otherwise we'd need to call it every free() */
 984 #ifdef EXT3_MB_HINT_GROUP_ALLOC
 985                 ext3_mb_discard_inode_preallocations(inode);
 986 #endif
 987                 ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);
 988                 goto out;
 989         }
 990
 991         /*
 992          * Putting len of the actual extent we just inserted,
 993          * we are asking ext3_ext_walk_space() to continue
 994          * scaning after that block
 995          */
 996         cex->ec_len = nex.ee_len;
 997         cex->ec_start = nex.ee_start;
 998         BUG_ON(nex.ee_len == 0);
 999         BUG_ON(nex.ee_block != cex->ec_block);
1000
1001 out:
1002         ext3_journal_stop(handle);
1003 map:
1004         if (err >= 0) {
1005                 /* map blocks */
1006                 if (bp->num == 0) {
1007                         CERROR("hmm. why do we find this extent?\n");
1008                         CERROR("initial space: %lu:%u\n",
1009                                 bp->start, bp->init_num);
1010                         CERROR("current extent: %u/%u/%u %d\n",
1011                                 cex->ec_block, cex->ec_len,
1012                                 cex->ec_start, cex->ec_type);
1013                 }
1014                 i = 0;
1015                 if (cex->ec_block < bp->start)
1016                         i = bp->start - cex->ec_block;
1017                 if (i >= cex->ec_len)
1018                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1019                                         i, cex->ec_len);
1020                 for (; i < cex->ec_len && bp->num; i++) {
1021                         *(bp->blocks) = cex->ec_start + i;
1022                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1023                                 *(bp->created) = 0;
1024                         } else {
1025                                 *(bp->created) = 1;
1026                                 /* unmap any possible underlying metadata from
1027                                  * the block device mapping.  bug 6998. */
1028                                 ll_unmap_underlying_metadata(inode->i_sb,
1029                                                              *(bp->blocks));
1030                         }
1031                         bp->created++;
1032                         bp->blocks++;
1033                         bp->num--;
1034                         bp->start++;
1035                 }
1036         }
1037         return err;
1038 }
1039
1040 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1041                        unsigned long num, unsigned long *blocks,
1042                        int *created, int create)
1043 {
1044 #ifdef EXT3_EXT_HAS_NO_TREE
1045         struct ext3_ext_base *base = inode;
1046 #else
1047         struct ext3_extents_tree tree;
1048         struct ext3_ext_base *base = &tree;
1049 #endif
1050         struct bpointers bp;
1051         int err;
1052
1053         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1054                block, block + num - 1, (unsigned) inode->i_ino);
1055
1056 #ifndef EXT3_EXT_HAS_NO_TREE
1057         ext3_init_tree_desc(base, inode);
1058         tree.private = &bp;
1059 #endif
1060         bp.blocks = blocks;
1061         bp.created = created;
1062         bp.start = block;
1063         bp.init_num = bp.num = num;
1064         bp.create = create;
1065
1066         ext3_down_truncate_sem(inode);
1067         err = fsfilt_ext3_ext_walk_space(base, block, num, ext3_ext_new_extent_cb, &bp);
1068         ext3_ext_invalidate_cache(base);
1069         ext3_up_truncate_sem(inode);
1070
1071         return err;
1072 }
1073
1074 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1075                                     int pages, unsigned long *blocks,
1076                                     int *created, int create)
1077 {
1078         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1079         int rc = 0, i = 0;
1080         struct page *fp = NULL;
1081         int clen = 0;
1082
1083         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1084                 inode->i_ino, pages, (*page)->index);
1085
1086         /* pages are sorted already. so, we just have to find
1087          * contig. space and process them properly */
1088         while (i < pages) {
1089                 if (fp == NULL) {
1090                         /* start new extent */
1091                         fp = *page++;
1092                         clen = 1;
1093                         i++;
1094                         continue;
1095                 } else if (fp->index + clen == (*page)->index) {
1096                         /* continue the extent */
1097                         page++;
1098                         clen++;
1099                         i++;
1100                         continue;
1101                 }
1102
1103                 /* process found extent */
1104                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1105                                         clen * blocks_per_page, blocks,
1106                                         created, create);
1107                 if (rc)
1108                         GOTO(cleanup, rc);
1109
1110                 /* look for next extent */
1111                 fp = NULL;
1112                 blocks += blocks_per_page * clen;
1113                 created += blocks_per_page * clen;
1114         }
1115
1116         if (fp)
1117                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1118                                         clen * blocks_per_page, blocks,
1119                                         created, create);
1120 cleanup:
1121         return rc;
1122 }
1123 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1124
1125 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1126                                unsigned long *blocks, int *created, int create);
1127 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1128                                    int pages, unsigned long *blocks,
1129                                    int *created, int create)
1130 {
1131         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1132         unsigned long *b;
1133         int rc = 0, i, *cr;
1134
1135         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1136                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1137                 if (rc) {
1138                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1139                                inode->i_ino, *b, *cr, create, rc);
1140                         break;
1141                 }
1142
1143                 b += blocks_per_page;
1144                 cr += blocks_per_page;
1145         }
1146         return rc;
1147 }
1148
1149 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1150                                 int pages, unsigned long *blocks,
1151                                 int *created, int create,
1152                                 struct semaphore *optional_sem)
1153 {
1154         int rc;
1155 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1156         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1157                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1158                                                      blocks, created, create);
1159                 return rc;
1160         }
1161 #endif
1162         if (optional_sem != NULL)
1163                 down(optional_sem);
1164         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1165                                             created, create);
1166         if (optional_sem != NULL)
1167                 up(optional_sem);
1168
1169         return rc;
1170 }
1171
1172 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1173                                    int size, loff_t *offs)
1174 {
1175         struct inode *inode = file->f_dentry->d_inode;
1176         unsigned long block;
1177         struct buffer_head *bh;
1178         int err, blocksize, csize, boffs;
1179
1180         /* prevent reading after eof */
1181         lock_kernel();
1182         if (i_size_read(inode) < *offs + size) {
1183                 size = i_size_read(inode) - *offs;
1184                 unlock_kernel();
1185                 if (size < 0) {
1186                         CERROR("size %llu is too short for read %u@%llu\n",
1187                                i_size_read(inode), size, *offs);
1188                         return -EIO;
1189                 } else if (size == 0) {
1190                         return 0;
1191                 }
1192         } else {
1193                 unlock_kernel();
1194         }
1195
1196         blocksize = 1 << inode->i_blkbits;
1197
1198         while (size > 0) {
1199                 block = *offs >> inode->i_blkbits;
1200                 boffs = *offs & (blocksize - 1);
1201                 csize = min(blocksize - boffs, size);
1202                 bh = ext3_bread(NULL, inode, block, 0, &err);
1203                 if (!bh) {
1204                         CERROR("can't read block: %d\n", err);
1205                         return err;
1206                 }
1207
1208                 memcpy(buf, bh->b_data + boffs, csize);
1209                 brelse(bh);
1210
1211                 *offs += csize;
1212                 buf += csize;
1213                 size -= csize;
1214         }
1215         return 0;
1216 }
1217
1218 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1219                                     loff_t *offs, int force_sync)
1220 {
1221         struct buffer_head *bh = NULL;
1222         unsigned long block;
1223         struct inode *inode = file->f_dentry->d_inode;
1224         loff_t old_size = i_size_read(inode), offset = *offs;
1225         loff_t new_size = i_size_read(inode);
1226         handle_t *handle;
1227         int err = 0, block_count = 0, blocksize, size, boffs;
1228
1229         /* Determine how many transaction credits are needed */
1230         blocksize = 1 << inode->i_blkbits;
1231         block_count = (*offs & (blocksize - 1)) + bufsize;
1232         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1233
1234         handle = ext3_journal_start(inode,
1235                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1236         if (IS_ERR(handle)) {
1237                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1238                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1239                 return PTR_ERR(handle);
1240         }
1241
1242         while (bufsize > 0) {
1243                 if (bh != NULL)
1244                         brelse(bh);
1245
1246                 block = offset >> inode->i_blkbits;
1247                 boffs = offset & (blocksize - 1);
1248                 size = min(blocksize - boffs, bufsize);
1249                 bh = ext3_bread(handle, inode, block, 1, &err);
1250                 if (!bh) {
1251                         CERROR("can't read/create block: %d\n", err);
1252                         goto out;
1253                 }
1254
1255                 err = ext3_journal_get_write_access(handle, bh);
1256                 if (err) {
1257                         CERROR("journal_get_write_access() returned error %d\n",
1258                                err);
1259                         goto out;
1260                 }
1261                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1262                 memcpy(bh->b_data + boffs, buf, size);
1263                 err = ext3_journal_dirty_metadata(handle, bh);
1264                 if (err) {
1265                         CERROR("journal_dirty_metadata() returned error %d\n",
1266                                err);
1267                         goto out;
1268                 }
1269                 if (offset + size > new_size)
1270                         new_size = offset + size;
1271                 offset += size;
1272                 bufsize -= size;
1273                 buf += size;
1274         }
1275
1276         if (force_sync)
1277                 handle->h_sync = 1; /* recovery likes this */
1278 out:
1279         if (bh)
1280                 brelse(bh);
1281
1282         /* correct in-core and on-disk sizes */
1283         if (new_size > i_size_read(inode)) {
1284                 lock_kernel();
1285                 if (new_size > i_size_read(inode))
1286                         i_size_write(inode, new_size);
1287                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1288                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1289                 if (i_size_read(inode) > old_size)
1290                         mark_inode_dirty(inode);
1291                 unlock_kernel();
1292         }
1293
1294         ext3_journal_stop(handle);
1295
1296         if (err == 0)
1297                 *offs = offset;
1298         return err;
1299 }
1300
1301 static int fsfilt_ext3_setup(struct super_block *sb)
1302 {
1303 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1304         struct ext3_sb_info *sbi = EXT3_SB(sb);
1305 #endif
1306 #if 0
1307         sbi->dx_lock = fsfilt_ext3_dx_lock;
1308         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1309 #endif
1310 #ifdef S_PDIROPS
1311         CWARN("Enabling PDIROPS\n");
1312         set_opt(sbi->s_mount_opt, PDIROPS);
1313         sb->s_flags |= S_PDIROPS;
1314 #endif
1315         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1316                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1317 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1318         /* enable journaled quota support */
1319         /* kfreed in ext3_put_super() */
1320         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1321         if (!sbi->s_qf_names[USRQUOTA])
1322                 return -ENOMEM;
1323         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1324         if (!sbi->s_qf_names[GRPQUOTA]) {
1325                 kfree(sbi->s_qf_names[USRQUOTA]);
1326                 sbi->s_qf_names[USRQUOTA] = NULL;
1327                 return -ENOMEM;
1328         }
1329         sbi->s_jquota_fmt = QFMT_VFS_V0;
1330 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1331         set_opt(sbi->s_mount_opt, QUOTA);
1332 #endif
1333 #endif
1334         return 0;
1335 }
1336
1337 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1338    objects. Logs is number of logfiles to update */
1339 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1340 {
1341         if ( !fso ) {
1342                 switch(op) {
1343                 case FSFILT_OP_CREATE:
1344                                  /* directory leaf, index & indirect & EA*/
1345                         return 4 + 3 * logs;
1346                 case FSFILT_OP_UNLINK:
1347                         return 3 * logs;
1348                 }
1349         } else {
1350                 int i;
1351                 int needed = 0;
1352                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1353                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1354                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1355                 for (i = 0; i < op; i++, fso++) {
1356                         int nblocks = fso->fso_bufcnt * blockpp;
1357                         int ndindirect = min(nblocks, addrpp + 1);
1358                         int nindir = nblocks + ndindirect + 1;
1359
1360                         needed += nindir;
1361                 }
1362                 return needed + 3 * logs;
1363         }
1364
1365         return 0;
1366 }
1367
1368 #ifdef HAVE_QUOTA_SUPPORT
1369 #define DQINFO_COPY(out, in)                    \
1370 do {                                            \
1371         Q_COPY(out, in, dqi_bgrace);            \
1372         Q_COPY(out, in, dqi_igrace);            \
1373         Q_COPY(out, in, dqi_flags);             \
1374         Q_COPY(out, in, dqi_valid);             \
1375 } while (0)
1376
1377 #define DQBLK_COPY(out, in)                     \
1378 do {                                            \
1379         Q_COPY(out, in, dqb_bhardlimit);        \
1380         Q_COPY(out, in, dqb_bsoftlimit);        \
1381         Q_COPY(out, in, dqb_curspace);          \
1382         Q_COPY(out, in, dqb_ihardlimit);        \
1383         Q_COPY(out, in, dqb_isoftlimit);        \
1384         Q_COPY(out, in, dqb_curinodes);         \
1385         Q_COPY(out, in, dqb_btime);             \
1386         Q_COPY(out, in, dqb_itime);             \
1387         Q_COPY(out, in, dqb_valid);             \
1388 } while (0)
1389
1390
1391
1392 static int fsfilt_ext3_quotactl(struct super_block *sb,
1393                                 struct obd_quotactl *oqc)
1394 {
1395         int i, rc = 0, error = 0;
1396         struct quotactl_ops *qcop;
1397         struct if_dqinfo *info;
1398         struct if_dqblk *dqblk;
1399         ENTRY;
1400
1401         if (!sb->s_qcop)
1402                 RETURN(-ENOSYS);
1403
1404         OBD_ALLOC_PTR(info);
1405         if (!info)
1406                 RETURN(-ENOMEM);
1407         OBD_ALLOC_PTR(dqblk);
1408         if (!dqblk) {
1409                 OBD_FREE_PTR(info);
1410                 RETURN(-ENOMEM);
1411         }
1412
1413         DQINFO_COPY(info, &oqc->qc_dqinfo);
1414         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1415
1416         qcop = sb->s_qcop;
1417         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1418                 for (i = 0; i < MAXQUOTAS; i++) {
1419                         if (!Q_TYPESET(oqc, i))
1420                                 continue;
1421
1422                         if (oqc->qc_cmd == Q_QUOTAON) {
1423                                 lustre_quota_version_t qfmt = oqc->qc_id;
1424                                 char *name[][MAXQUOTAS] = LUSTRE_OPQFILES_NAMES;
1425
1426                                 if (!qcop->quota_on)
1427                                         GOTO(out, rc = -ENOSYS);
1428
1429                                 rc = qcop->quota_on(sb, i, QFMT_VFS_V0,
1430                                                     name[qfmt][i]);
1431 #ifdef HAVE_QUOTA64
1432                                 if (rc == -ENOENT || rc == -EINVAL) {
1433                                         /* see bug 13904 */
1434                                         rc = lustre_slave_quota_convert(qfmt, i);
1435                                         if (!rc)
1436                                                 rc = qcop->quota_on(sb, i,
1437                                                                 QFMT_VFS_V0,
1438                                                                 name[qfmt][i]);
1439                                         else if (rc == -ESTALE)
1440                                                 rc = -ENOENT;
1441                                 }
1442 #endif
1443                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1444                                 if (!qcop->quota_off)
1445                                         GOTO(out, rc = -ENOSYS);
1446                                 rc = qcop->quota_off(sb, i);
1447                         }
1448
1449                         if (rc == -EBUSY)
1450                                 error = rc;
1451                         else if (rc)
1452                                 GOTO(out, rc);
1453                 }
1454                 GOTO(out, rc ?: error);
1455         }
1456
1457         switch (oqc->qc_cmd) {
1458         case Q_GETOINFO:
1459         case Q_GETINFO:
1460                 if (!qcop->get_info)
1461                         GOTO(out, rc = -ENOSYS);
1462                 rc = qcop->get_info(sb, oqc->qc_type, info);
1463                 break;
1464         case Q_SETQUOTA:
1465         case Q_INITQUOTA:
1466                 if (!qcop->set_dqblk)
1467                         GOTO(out, rc = -ENOSYS);
1468                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1469                 break;
1470         case Q_GETOQUOTA:
1471         case Q_GETQUOTA:
1472                 if (!qcop->get_dqblk)
1473                         GOTO(out, rc = -ENOSYS);
1474                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1475                 if (!rc)
1476                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1477                 break;
1478         case Q_SYNC:
1479                 if (!sb->s_qcop->quota_sync)
1480                         GOTO(out, rc = -ENOSYS);
1481                 qcop->quota_sync(sb, oqc->qc_type);
1482                 break;
1483         case Q_FINVALIDATE:
1484                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1485                 for (i = 0; i < MAXQUOTAS; i++) {
1486                         struct file *fp;
1487                         lustre_quota_version_t qfmt = oqc->qc_id;
1488                         char *name[][MAXQUOTAS] = LUSTRE_OPQFILES_NAMES;
1489
1490                         if (!Q_TYPESET(oqc, i))
1491                                 continue;
1492
1493                         fp = filp_open(name[qfmt][i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1494                         if (IS_ERR(fp)) {
1495                                 rc = PTR_ERR(fp);
1496                                 CERROR("error invalidating operational quota file"
1497                                        " %s (rc:%d)\n", name[qfmt][i], rc);
1498                         } else {
1499                                 filp_close(fp, 0);
1500                         }
1501
1502                 }
1503                 break;
1504         default:
1505                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1506                 LBUG();
1507         }
1508 out:
1509         DQINFO_COPY(&oqc->qc_dqinfo, info);
1510         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1511
1512         OBD_FREE_PTR(info);
1513         OBD_FREE_PTR(dqblk);
1514
1515         if (rc)
1516                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1517                                 "failed: %d\n",
1518                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1519         RETURN(rc);
1520 }
1521
1522 struct chk_dqblk{
1523         struct hlist_node       dqb_hash;        /* quotacheck hash */
1524         struct list_head        dqb_list;        /* in list also */
1525         qid_t                   dqb_id;          /* uid/gid */
1526         short                   dqb_type;        /* USRQUOTA/GRPQUOTA */
1527         qsize_t                 dqb_bhardlimit;  /* block hard limit */
1528         qsize_t                 dqb_bsoftlimit;  /* block soft limit */
1529         qsize_t                 dqb_curspace;    /* current space */
1530         qsize_t                 dqb_ihardlimit;  /* inode hard limit */
1531         qsize_t                 dqb_isoftlimit;  /* inode soft limit */
1532         qsize_t                 dqb_curinodes;   /* current inodes */
1533         __u64                   dqb_btime;       /* block grace time */
1534         __u64                   dqb_itime;       /* inode grace time */
1535         __u32                   dqb_valid;       /* flag for above fields */
1536 };
1537
1538 static inline unsigned int chkquot_hash(qid_t id, int type)
1539                                         __attribute__((__const__));
1540
1541 static inline unsigned int chkquot_hash(qid_t id, int type)
1542 {
1543         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1544 }
1545
1546 static inline struct chk_dqblk *
1547 find_chkquot(struct hlist_head *head, qid_t id, int type)
1548 {
1549         struct hlist_node *node;
1550         struct chk_dqblk *cdqb;
1551
1552         hlist_for_each(node, head) {
1553                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1554                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1555                         return cdqb;
1556         }
1557
1558         return NULL;
1559 }
1560
1561 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1562 {
1563         struct chk_dqblk *cdqb;
1564
1565         OBD_ALLOC_PTR(cdqb);
1566         if (cdqb) {
1567                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1568                 INIT_LIST_HEAD(&cdqb->dqb_list);
1569                 cdqb->dqb_id = id;
1570                 cdqb->dqb_type = type;
1571         }
1572
1573         return cdqb;
1574 }
1575
1576 static struct chk_dqblk *
1577 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1578       qid_t id, int type, int first_check)
1579 {
1580         struct hlist_head *head = hash + chkquot_hash(id, type);
1581         struct if_dqblk dqb;
1582         struct chk_dqblk *cdqb;
1583         int rc;
1584
1585         cdqb = find_chkquot(head, id, type);
1586         if (cdqb)
1587                 return cdqb;
1588
1589         cdqb = alloc_chkquot(id, type);
1590         if (!cdqb)
1591                 return NULL;
1592
1593         if (!first_check) {
1594                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1595                 if (rc) {
1596                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1597                                id, type, rc);
1598                 } else {
1599                         DQBLK_COPY(cdqb, &dqb);
1600                         cdqb->dqb_curspace = 0;
1601                         cdqb->dqb_curinodes = 0;
1602                 }
1603         }
1604
1605         hlist_add_head(&cdqb->dqb_hash, head);
1606         list_add_tail(&cdqb->dqb_list, list);
1607
1608         return cdqb;
1609 }
1610
1611 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1612 {
1613         struct obd_quotactl *oqctl;
1614         int rc;
1615
1616         OBD_ALLOC_PTR(oqctl);
1617         if (!oqctl)
1618                 RETURN(-ENOMEM);
1619
1620         oqctl->qc_cmd = cmd;
1621         oqctl->qc_id = qfmt;
1622         oqctl->qc_type = type;
1623         rc = fsfilt_ext3_quotactl(sb, oqctl);
1624
1625         OBD_FREE_PTR(oqctl);
1626         return rc;
1627 }
1628
1629 static inline int read_old_dqinfo(struct super_block *sb, int type,
1630                                   struct if_dqinfo *dqinfo)
1631 {
1632         struct obd_quotactl *oqctl;
1633         int rc;
1634         ENTRY;
1635
1636         OBD_ALLOC_PTR(oqctl);
1637         if (!oqctl)
1638                 RETURN(-ENOMEM);
1639
1640         oqctl->qc_cmd = Q_GETINFO;
1641         oqctl->qc_type = type;
1642         rc = fsfilt_ext3_quotactl(sb, oqctl);
1643         if (!rc)
1644                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1645
1646         OBD_FREE_PTR(oqctl);
1647         RETURN(rc);
1648 }
1649
1650 static inline struct ext3_group_desc *
1651 get_group_desc(struct super_block *sb, int group)
1652 {
1653         unsigned long desc_block, desc;
1654         struct ext3_group_desc *gdp;
1655
1656         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1657         desc = group % EXT3_DESC_PER_BLOCK(sb);
1658         gdp = (struct ext3_group_desc *)
1659               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1660
1661         return gdp + desc;
1662 }
1663
1664 static inline struct buffer_head *
1665 read_inode_bitmap(struct super_block *sb, unsigned long group)
1666 {
1667         struct ext3_group_desc *desc;
1668         struct buffer_head *bh;
1669
1670         desc = get_group_desc(sb, group);
1671         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1672
1673         return bh;
1674 }
1675
1676 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1677                                      struct buffer_head *bitmap_bh,
1678                                      int index, unsigned long ino)
1679 {
1680         struct inode *inode = NULL;
1681
1682         if (ext3_test_bit(index, bitmap_bh->b_data))
1683                 inode = iget(sb, ino);
1684
1685         return inode;
1686 }
1687
1688 struct qchk_ctxt {
1689         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1690         struct list_head        qckt_list;                   /* quotacheck list */
1691         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1692         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1693 };
1694
1695 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1696                            struct obd_quotactl *oqc)
1697 {
1698         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1699         loff_t size = 0;
1700         qid_t qid[MAXQUOTAS];
1701         int cnt, i, rc = 0;
1702
1703         if (!inode)
1704                 return 0;
1705
1706         qid[USRQUOTA] = inode->i_uid;
1707         qid[GRPQUOTA] = inode->i_gid;
1708
1709         if (S_ISDIR(inode->i_mode) ||
1710             S_ISREG(inode->i_mode) ||
1711             S_ISLNK(inode->i_mode))
1712                 size = inode_get_bytes(inode);
1713
1714         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1715                 if (!Q_TYPESET(oqc, cnt))
1716                         continue;
1717
1718                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1719                                 &qctxt->qckt_list, qid[cnt], cnt,
1720                                 qctxt->qckt_first_check[cnt]);
1721                 if (!cdqb[cnt]) {
1722                         rc = -ENOMEM;
1723                         break;
1724                 }
1725
1726                 cdqb[cnt]->dqb_curspace += size;
1727                 cdqb[cnt]->dqb_curinodes++;
1728         }
1729
1730         if (rc) {
1731                 for (i = 0; i < cnt; i++) {
1732                         if (!Q_TYPESET(oqc, i))
1733                                 continue;
1734                         LASSERT(cdqb[i]);
1735                         cdqb[i]->dqb_curspace -= size;
1736                         cdqb[i]->dqb_curinodes--;
1737                 }
1738         }
1739
1740         return rc;
1741 }
1742
1743 static int v2_write_dqheader(struct file *f, int type)
1744 {
1745         static const __u32 quota_magics[] = V2_INITQMAGICS;
1746 #ifdef HAVE_QUOTA64
1747         static const __u32 quota_versions[] = V2_INITQVERSIONS_R0;
1748 #else
1749         static const __u32 quota_versions[] = V2_INITQVERSIONS;
1750 #endif
1751         struct v2_disk_dqheader dqhead;
1752         loff_t offset = 0;
1753
1754         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1755         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1756
1757         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1758         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1759
1760         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1761 }
1762
1763 /* write dqinfo struct in a new quota file */
1764 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1765 {
1766         struct v2_disk_dqinfo dqinfo;
1767         __u32 blocks = V2_DQTREEOFF + 1;
1768         loff_t offset = V2_DQINFOOFF;
1769
1770         if (info) {
1771                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1772                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1773                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1774                                                ~DQF_INFO_DIRTY);
1775         } else {
1776                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1777                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1778                 dqinfo.dqi_flags = 0;
1779         }
1780
1781         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1782         dqinfo.dqi_free_blk = 0;
1783         dqinfo.dqi_free_entry = 0;
1784
1785         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1786 }
1787
1788 #ifdef HAVE_QUOTA64
1789 static int v3_write_dqheader(struct file *f, int type)
1790 {
1791         static const __u32 quota_magics[] = V2_INITQMAGICS;
1792         static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1793         struct v2_disk_dqheader dqhead;
1794         loff_t offset = 0;
1795
1796         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1797         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1798
1799         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1800         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1801
1802         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1803 }
1804
1805 /* write dqinfo struct in a new quota file */
1806 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1807 {
1808         return v2_write_dqinfo(f, type, info);
1809 }
1810 #endif
1811
1812 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1813                                   struct obd_quotactl *oqc)
1814 {
1815         int i, rc = 0;
1816         ENTRY;
1817
1818         for (i = 0; i < MAXQUOTAS; i++) {
1819                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1820                                          NULL : &qctxt->qckt_dqinfo[i];
1821                 struct file *file;
1822                 const char *name[][MAXQUOTAS] = LUSTRE_OPQFILES_NAMES;
1823                 int (*write_dqheader)(struct file *, int);
1824                 int (*write_dqinfo)(struct file *, int, struct if_dqinfo *);
1825
1826                 if (!Q_TYPESET(oqc, i))
1827                         continue;
1828
1829                 file = filp_open(name[oqc->qc_id][i],
1830                                  O_RDWR | O_CREAT | O_TRUNC, 0644);
1831                 if (IS_ERR(file)) {
1832                         rc = PTR_ERR(file);
1833                         CERROR("can't create %s file: rc = %d\n",
1834                                name[oqc->qc_id][i], rc);
1835                         GOTO(out, rc);
1836                 }
1837
1838                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1839                         CERROR("file %s is not regular", name[oqc->qc_id][i]);
1840                         filp_close(file, 0);
1841                         GOTO(out, rc = -EINVAL);
1842                 }
1843
1844                 DQUOT_DROP(file->f_dentry->d_inode);
1845
1846                 switch (oqc->qc_id) {
1847                 case LUSTRE_QUOTA_V1 : write_dqheader = v2_write_dqheader;
1848                                        write_dqinfo   = v2_write_dqinfo;
1849                                        break;
1850 #ifdef HAVE_QUOTA64
1851                 case LUSTRE_QUOTA_V2 : write_dqheader = v3_write_dqheader;
1852                                        write_dqinfo   = v3_write_dqinfo;
1853                                        break;
1854 #endif
1855                 default              : CERROR("unknown quota format!\n");
1856                                        LBUG();
1857                 }
1858
1859                 rc = (*write_dqheader)(file, i);
1860                 if (rc) {
1861                         filp_close(file, 0);
1862                         GOTO(out, rc);
1863                 }
1864
1865                 rc = (*write_dqinfo)(file, i, info);
1866                 filp_close(file, 0);
1867                 if (rc)
1868                         GOTO(out, rc);
1869         }
1870
1871 out:
1872         RETURN(rc);
1873 }
1874
1875
1876 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1877                           struct chk_dqblk *cdqb)
1878 {
1879         struct obd_quotactl *oqc;
1880         long now;
1881         int rc;
1882         ENTRY;
1883
1884         OBD_ALLOC_PTR(oqc);
1885         if (!oqc)
1886                 RETURN(-ENOMEM);
1887
1888         now = CURRENT_SECONDS;
1889
1890         if (cdqb->dqb_bsoftlimit &&
1891             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1892             !cdqb->dqb_btime)
1893                 cdqb->dqb_btime =
1894                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1895
1896         if (cdqb->dqb_isoftlimit &&
1897             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1898             !cdqb->dqb_itime)
1899                 cdqb->dqb_itime =
1900                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1901
1902         cdqb->dqb_valid = QIF_ALL;
1903
1904         oqc->qc_cmd = Q_SETQUOTA;
1905         oqc->qc_type = cdqb->dqb_type;
1906         oqc->qc_id = cdqb->dqb_id;
1907         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1908
1909         rc = fsfilt_ext3_quotactl(sb, oqc);
1910         OBD_FREE_PTR(oqc);
1911         RETURN(rc);
1912 }
1913
1914 static int prune_chkquots(struct super_block *sb,
1915                           struct qchk_ctxt *qctxt, int error)
1916 {
1917         struct chk_dqblk *cdqb, *tmp;
1918         int rc;
1919
1920         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1921                 if (!error) {
1922                         rc = commit_chkquot(sb, qctxt, cdqb);
1923                         if (rc)
1924                                 error = rc;
1925                 }
1926                 hlist_del_init(&cdqb->dqb_hash);
1927                 list_del(&cdqb->dqb_list);
1928                 OBD_FREE_PTR(cdqb);
1929         }
1930
1931         return error;
1932 }
1933
1934 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1935                                   struct obd_quotactl *oqc)
1936 {
1937         struct ext3_sb_info *sbi = EXT3_SB(sb);
1938         int i, group;
1939         struct qchk_ctxt *qctxt;
1940         struct buffer_head *bitmap_bh = NULL;
1941         unsigned long ino;
1942         struct inode *inode;
1943         int rc = 0;
1944         ENTRY;
1945
1946         /* turn on quota and read dqinfo if existed */
1947         OBD_ALLOC_PTR(qctxt);
1948         if (!qctxt) {
1949                 oqc->qc_stat = -ENOMEM;
1950                 RETURN(-ENOMEM);
1951         }
1952
1953         for (i = 0; i < NR_DQHASH; i++)
1954                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1955         INIT_LIST_HEAD(&qctxt->qckt_list);
1956
1957         for (i = 0; i < MAXQUOTAS; i++) {
1958                 if (!Q_TYPESET(oqc, i))
1959                         continue;
1960
1961                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1962                 if (!rc || rc == -EBUSY) {
1963                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1964                         if (rc)
1965                                 GOTO(out, rc);
1966                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1967                         qctxt->qckt_first_check[i] = 1;
1968                 } else if (rc) {
1969                         GOTO(out, rc);
1970                 }
1971         }
1972
1973         /* check quota and update in hash */
1974         for (group = 0; group < sbi->s_groups_count; group++) {
1975                 ino = group * sbi->s_inodes_per_group + 1;
1976                 bitmap_bh = read_inode_bitmap(sb, group);
1977                 if (!bitmap_bh) {
1978                         CERROR("read_inode_bitmap group %d failed", group);
1979                         GOTO(out, -EIO);
1980                 }
1981
1982                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1983                         if (ino < sbi->s_first_ino)
1984                                 continue;
1985
1986                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1987                         rc = add_inode_quota(inode, qctxt, oqc);
1988                         iput(inode);
1989                         if (rc) {
1990                                 brelse(bitmap_bh);
1991                                 GOTO(out, rc);
1992                         }
1993                 }
1994
1995                 brelse(bitmap_bh);
1996         }
1997
1998         /* read old quota limits from old quota file. (only for the user
1999          * has limits but hasn't file) */
2000 #ifdef HAVE_QUOTA_SUPPORT
2001         for (i = 0; i < MAXQUOTAS; i++) {
2002                 struct list_head id_list;
2003                 struct dquot_id *dqid, *tmp;
2004
2005                 if (!Q_TYPESET(oqc, i))
2006                         continue;
2007
2008                 if (qctxt->qckt_first_check[i])
2009                         continue;
2010
2011
2012                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2013                 INIT_LIST_HEAD(&id_list);
2014 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2015                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2016 #else
2017                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2018 #endif
2019                 if (rc)
2020                         CERROR("read old limits failed. (rc:%d)\n", rc);
2021
2022                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2023                         list_del_init(&dqid->di_link);
2024
2025                         if (!rc)
2026                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2027                                       dqid->di_id, i,
2028                                       qctxt->qckt_first_check[i]);
2029                         kfree(dqid);
2030                 }
2031         }
2032 #endif
2033         /* turn off quota cause we are to dump chk_dqblk to files */
2034         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2035
2036         rc = create_new_quota_files(qctxt, oqc);
2037         if (rc)
2038                 GOTO(out, rc);
2039
2040         /* we use vfs functions to set dqblk, so turn quota on */
2041         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2042 out:
2043         /* dump and free chk_dqblk */
2044         rc = prune_chkquots(sb, qctxt, rc);
2045         OBD_FREE_PTR(qctxt);
2046
2047         /* turn off quota, `lfs quotacheck` will turn on when all
2048          * nodes quotacheck finish. */
2049         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2050
2051         oqc->qc_stat = rc;
2052         if (rc)
2053                 CERROR("quotacheck failed: rc = %d\n", rc);
2054
2055         RETURN(rc);
2056 }
2057
2058 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2059                                  int cmd)
2060 {
2061         int rc = 0;
2062         ENTRY;
2063
2064         if (lqi->qi_files[type] == NULL) {
2065                 CERROR("operate qinfo before it's enabled!\n");
2066                 RETURN(-EIO);
2067         }
2068
2069         switch (cmd) {
2070         case QFILE_CHK:
2071                 rc = lustre_check_quota_file(lqi, type);
2072                 break;
2073         case QFILE_RD_INFO:
2074                 rc = lustre_read_quota_info(lqi, type);
2075                 break;
2076         case QFILE_WR_INFO:
2077                 rc = lustre_write_quota_info(lqi, type);
2078                 break;
2079         case QFILE_INIT_INFO:
2080                 rc = lustre_init_quota_info(lqi, type);
2081                 break;
2082         case QFILE_CONVERT:
2083                 rc = lustre_quota_convert(lqi, type);
2084                 break;
2085         default:
2086                 rc = -ENOTSUPP;
2087                 CERROR("Unsupported admin quota file cmd %d\n"
2088                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2089                        cmd);
2090                 break;
2091         }
2092         RETURN(rc);
2093 }
2094
2095 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2096                             struct list_head *list)
2097 {
2098         return lustre_get_qids(file, inode, type, list);
2099 }
2100
2101 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2102 {
2103         int rc = 0;
2104         ENTRY;
2105
2106         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2107                 CERROR("operate dquot before it's enabled!\n");
2108                 RETURN(-EIO);
2109         }
2110
2111         switch (cmd) {
2112         case QFILE_RD_DQUOT:
2113                 rc = lustre_read_dquot(dquot);
2114                 break;
2115         case QFILE_WR_DQUOT:
2116                 if (dquot->dq_dqb.dqb_ihardlimit ||
2117                     dquot->dq_dqb.dqb_isoftlimit ||
2118                     dquot->dq_dqb.dqb_bhardlimit ||
2119                     dquot->dq_dqb.dqb_bsoftlimit)
2120                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2121                 else
2122                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2123
2124                 rc = lustre_commit_dquot(dquot);
2125                 if (rc >= 0)
2126                         rc = 0;
2127                 break;
2128         default:
2129                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2130                 LBUG();
2131                 break;
2132         }
2133         RETURN(rc);
2134 }
2135
2136 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2137                                 struct inode *inode, int frags)
2138 {
2139 #ifdef EXT3_EXT_HAS_NO_TREE
2140         struct ext3_ext_base *base = inode;
2141 #else
2142         struct ext3_extents_tree tree;
2143         struct ext3_ext_base *base = &tree;
2144
2145         ext3_init_tree_desc(base, inode);
2146 #endif
2147         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2148          * metablocks at maxium b=16542 */
2149         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2150         return 0;
2151 }
2152
2153 #endif
2154
2155 static lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2156 {
2157         return (EXT3_SB(sb)->journal_bdev);
2158 }
2159
2160 static struct fsfilt_operations fsfilt_ext3_ops = {
2161         .fs_type                = "ext3",
2162         .fs_owner               = THIS_MODULE,
2163         .fs_getlabel            = fsfilt_ext3_get_label,
2164         .fs_setlabel            = fsfilt_ext3_set_label,
2165         .fs_uuid                = fsfilt_ext3_uuid,
2166         .fs_start               = fsfilt_ext3_start,
2167         .fs_brw_start           = fsfilt_ext3_brw_start,
2168         .fs_extend              = fsfilt_ext3_extend,
2169         .fs_commit              = fsfilt_ext3_commit,
2170         .fs_commit_async        = fsfilt_ext3_commit_async,
2171         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2172         .fs_setattr             = fsfilt_ext3_setattr,
2173         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2174         .fs_set_md              = fsfilt_ext3_set_md,
2175         .fs_get_md              = fsfilt_ext3_get_md,
2176         .fs_readpage            = fsfilt_ext3_readpage,
2177         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2178         .fs_statfs              = fsfilt_ext3_statfs,
2179         .fs_sync                = fsfilt_ext3_sync,
2180         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2181         .fs_write_record        = fsfilt_ext3_write_record,
2182         .fs_read_record         = fsfilt_ext3_read_record,
2183         .fs_setup               = fsfilt_ext3_setup,
2184         .fs_send_bio            = fsfilt_ext3_send_bio,
2185         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2186 #ifdef HAVE_DISK_INODE_VERSION
2187         .fs_get_version         = fsfilt_ext3_get_version,
2188         .fs_set_version         = fsfilt_ext3_set_version,
2189 #endif
2190 #ifdef HAVE_QUOTA_SUPPORT
2191         .fs_quotactl            = fsfilt_ext3_quotactl,
2192         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2193         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2194         .fs_qids                = fsfilt_ext3_qids,
2195         .fs_dquot               = fsfilt_ext3_dquot,
2196         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2197 #endif
2198         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2199 };
2200
2201 static int __init fsfilt_ext3_init(void)
2202 {
2203         int rc;
2204
2205         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2206                                          sizeof(struct fsfilt_cb_data), 0, 0);
2207         if (!fcb_cache) {
2208                 CERROR("error allocating fsfilt journal callback cache\n");
2209                 GOTO(out, rc = -ENOMEM);
2210         }
2211
2212         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2213
2214         if (rc) {
2215                 int err = cfs_mem_cache_destroy(fcb_cache);
2216                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2217         }
2218 out:
2219         return rc;
2220 }
2221
2222 static void __exit fsfilt_ext3_exit(void)
2223 {
2224         int rc;
2225
2226         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2227         rc = cfs_mem_cache_destroy(fcb_cache);
2228         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2229 }
2230
2231 module_init(fsfilt_ext3_init);
2232 module_exit(fsfilt_ext3_exit);
2233
2234 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2235 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2236 MODULE_LICENSE("GPL");