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 <ext3/xattr.h>
  58
  59 #include <libcfs/libcfs.h>
  60 #include <lustre_fsfilt.h>
  61 #include <obd.h>
  62 #include <lustre_quota.h>
  63 #include <linux/lustre_compat25.h>
  64 #include <linux/lprocfs_status.h>
  65
  66 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  67 #include <linux/ext3_extents.h>
  68 #endif
  69
  70 #include "lustre_quota_fmt.h"
  71
  72 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  73 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  74 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  75 #else
  76 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  77 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  78 #endif
  79
  80 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
  81 /* for kernels 2.6.18 and later */
  82 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  83 #else
  84 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
  85 #endif
  86
  87 #define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks)
  88 #define fsfilt_ext3_journal_stop(handle)          ext3_journal_stop(handle)
  89
  90 static cfs_mem_cache_t *fcb_cache;
  91
  92 struct fsfilt_cb_data {
  93         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
  94         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
  95         struct obd_device *cb_obd;      /* MDS/OBD completion device */
  96         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
  97         void *cb_data;                  /* MDS/OST completion function data */
  98 };
  99
 100 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 101 #define EXT3_XATTR_INDEX_TRUSTED        4
 102 #endif
 103
 104 static char *fsfilt_ext3_get_label(struct super_block *sb)
 105 {
 106         return EXT3_SB(sb)->s_es->s_volume_name;
 107 }
 108
 109 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 110 {
 111         /* see e.g. fsfilt_ext3_write_record() */
 112         journal_t *journal;
 113         handle_t *handle;
 114         int err;
 115
 116         journal = EXT3_SB(sb)->s_journal;
 117         handle = journal_start(journal, 1);
 118         if (IS_ERR(handle)) {
 119                 CERROR("can't start transaction\n");
 120                 return(PTR_ERR(handle));
 121         }
 122
 123         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 124         if (err)
 125                 goto out;
 126
 127         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 128                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 129
 130         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 131
 132 out:
 133         journal_stop(handle);
 134
 135         return(err);
 136 }
 137
 138 static char *fsfilt_ext3_uuid(struct super_block *sb)
 139 {
 140         return EXT3_SB(sb)->s_es->s_uuid;
 141 }
 142
 143 #ifdef HAVE_DISK_INODE_VERSION
 144 /*
 145  * Get the 64-bit version for an inode.
 146  */
 147 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 148 {
 149         return EXT3_I(inode)->i_fs_version;
 150 }
 151
 152 /*
 153  * Set the 64-bit version and return the old version.
 154  */
 155 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 156 {
 157         __u64 old_version = EXT3_I(inode)->i_fs_version;
 158
 159         (EXT3_I(inode))->i_fs_version = new_version;
 160         return old_version;
 161 }
 162
 163 #endif
 164
 165 /*
 166  * We don't currently need any additional blocks for rmdir and
 167  * unlink transactions because we are storing the OST oa_id inside
 168  * the inode (which we will be changing anyways as part of this
 169  * transaction).
 170  */
 171 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 172                                int logs)
 173 {
 174         /* For updates to the last received file */
 175         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 176         journal_t *journal;
 177         void *handle;
 178
 179         if (current->journal_info) {
 180                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 181                        current->journal_info);
 182                 goto journal_start;
 183         }
 184
 185         switch(op) {
 186         case FSFILT_OP_RMDIR:
 187         case FSFILT_OP_UNLINK:
 188                 /* delete one file + create/update logs for each stripe */
 189                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 190                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 191                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 192                 break;
 193         case FSFILT_OP_RENAME:
 194                 /* modify additional directory */
 195                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 196                 /* no break */
 197         case FSFILT_OP_SYMLINK:
 198                 /* additional block + block bitmap + GDT for long symlink */
 199                 nblocks += 3;
 200                 /* no break */
 201         case FSFILT_OP_CREATE: {
 202 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
 203                 static int warned;
 204                 if (!warned) {
 205                         if (!test_opt(inode->i_sb, EXTENTS)) {
 206                                 warned = 1;
 207                         } else if (((EXT3_I(inode)->i_flags &
 208                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 209                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 210                                 CWARN("extent-mapped directory found - contact "
 211                                       "http://bugzilla.lustre.org/\n");
 212                                 warned = 1;
 213                         }
 214                 }
 215 #endif
 216                 /* no break */
 217         }
 218         case FSFILT_OP_MKDIR:
 219         case FSFILT_OP_MKNOD:
 220                 /* modify one inode + block bitmap + GDT */
 221                 nblocks += 3;
 222                 /* no break */
 223         case FSFILT_OP_LINK:
 224                 /* modify parent directory */
 225                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 226                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 227                 /* create/update logs for each stripe */
 228                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 229                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 230                 break;
 231         case FSFILT_OP_SETATTR:
 232                 /* Setattr on inode */
 233                 nblocks += 1;
 234                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 235                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 236                 /* quota chown log for each stripe */
 237                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 238                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 239                 break;
 240         case FSFILT_OP_CANCEL_UNLINK:
 241                 /* blocks for log header bitmap update OR
 242                  * blocks for catalog header bitmap update + unlink of logs */
 243                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 244                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 245                 break;
 246         case FSFILT_OP_JOIN:
 247                 /* delete 2 file(file + array id) + create 1 file (array id)
 248                  * create/update logs for each stripe */
 249                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 250
 251                 /*create array log for head file*/
 252                 nblocks += 3;
 253                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 254                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
 255                 /*update head file array */
 256                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 257                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 258                 break;
 259         default: CERROR("unknown transaction start op %d\n", op);
 260                 LBUG();
 261         }
 262
 263         LASSERT(current->journal_info == desc_private);
 264         journal = EXT3_SB(inode->i_sb)->s_journal;
 265         if (nblocks > journal->j_max_transaction_buffers) {
 266                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 267                        nblocks, op, logs, journal->j_max_transaction_buffers);
 268                 nblocks = journal->j_max_transaction_buffers;
 269         }
 270
 271  journal_start:
 272         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 273         handle = fsfilt_ext3_journal_start(inode, nblocks);
 274
 275         if (!IS_ERR(handle))
 276                 LASSERT(current->journal_info == handle);
 277         else
 278                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 279                        op, nblocks, PTR_ERR(handle));
 280         return handle;
 281 }
 282
 283 /*
 284  * Calculate the number of buffer credits needed to write multiple pages in
 285  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 286  * doesn't have a nice API for calculating this sort of thing in advance.
 287  *
 288  * See comment above ext3_writepage_trans_blocks for details.  We assume
 289  * no data journaling is being done, but it does allow for all of the pages
 290  * being non-contiguous.  If we are guaranteed contiguous pages we could
 291  * reduce the number of (d)indirect blocks a lot.
 292  *
 293  * With N blocks per page and P pages, for each inode we have at most:
 294  * N*P indirect
 295  * min(N*P, blocksize/4 + 1) dindirect blocks
 296  * niocount tindirect
 297  *
 298  * For the entire filesystem, we have at most:
 299  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 300  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 301  * objcount inode blocks
 302  * 1 superblock
 303  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 304  *
 305  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 306  */
 307 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 308                                       int niocount, struct niobuf_local *nb)
 309 {
 310         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 311         __u64 next_indir;
 312         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 313         int nbitmaps = 0, ngdblocks;
 314         int needed = objcount + 1; /* inodes + superblock */
 315         int i, j;
 316
 317         for (i = 0, j = 0; i < objcount; i++, fso++) {
 318                 /* two or more dindirect blocks in case we cross boundary */
 319                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 320                                     nb[j].offset) >>
 321                                    sb->s_blocksize_bits) /
 322                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 323                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 324
 325                 /* leaf, indirect, tindirect blocks for first block */
 326                 nbitmaps += blockpp + 2;
 327
 328                 j += fso->fso_bufcnt;
 329         }
 330
 331         next_indir = nb[0].offset +
 332                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 333         for (i = 1; i < niocount; i++) {
 334                 if (nb[i].offset >= next_indir) {
 335                         nbitmaps++;     /* additional indirect */
 336                         next_indir = nb[i].offset +
 337                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 338                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 339                         nbitmaps++;     /* additional indirect */
 340                 }
 341                 nbitmaps += blockpp;    /* each leaf in different group? */
 342         }
 343
 344         ngdblocks = nbitmaps;
 345         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 346                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 347         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 348                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 349
 350         needed += nbitmaps + ngdblocks;
 351
 352         /* last_rcvd update */
 353         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 354
 355 #if defined(CONFIG_QUOTA)
 356         /* We assume that there will be 1 bit set in s_dquot.flags for each
 357          * quota file that is active.  This is at least true for now.
 358          */
 359         needed += hweight32(sb_any_quota_enabled(sb)) *
 360                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 361 #endif
 362
 363         return needed;
 364 }
 365
 366 /* We have to start a huge journal transaction here to hold all of the
 367  * metadata for the pages being written here.  This is necessitated by
 368  * the fact that we do lots of prepare_write operations before we do
 369  * any of the matching commit_write operations, so even if we split
 370  * up to use "smaller" transactions none of them could complete until
 371  * all of them were opened.  By having a single journal transaction,
 372  * we eliminate duplicate reservations for common blocks like the
 373  * superblock and group descriptors or bitmaps.
 374  *
 375  * We will start the transaction here, but each prepare_write will
 376  * add a refcount to the transaction, and each commit_write will
 377  * remove a refcount.  The transaction will be closed when all of
 378  * the pages have been written.
 379  */
 380 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 381                                    int niocount, struct niobuf_local *nb,
 382                                    void *desc_private, int logs)
 383 {
 384         journal_t *journal;
 385         handle_t *handle;
 386         int needed;
 387         ENTRY;
 388
 389         LASSERT(current->journal_info == desc_private);
 390         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 391         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 392
 393         /* The number of blocks we could _possibly_ dirty can very large.
 394          * We reduce our request if it is absurd (and we couldn't get that
 395          * many credits for a single handle anyways).
 396          *
 397          * At some point we have to limit the size of I/Os sent at one time,
 398          * increase the size of the journal, or we have to calculate the
 399          * actual journal requirements more carefully by checking all of
 400          * the blocks instead of being maximally pessimistic.  It remains to
 401          * be seen if this is a real problem or not.
 402          */
 403         if (needed > journal->j_max_transaction_buffers) {
 404                 CERROR("want too many journal credits (%d) using %d instead\n",
 405                        needed, journal->j_max_transaction_buffers);
 406                 needed = journal->j_max_transaction_buffers;
 407         }
 408
 409         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 410         handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
 411         if (IS_ERR(handle)) {
 412                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 413                        PTR_ERR(handle));
 414         } else {
 415                 LASSERT(handle->h_buffer_credits >= needed);
 416                 LASSERT(current->journal_info == handle);
 417         }
 418
 419         RETURN(handle);
 420 }
 421
 422 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 423 {
 424        handle_t *handle = h;
 425
 426        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 427        if (nblocks == 0) {
 428                handle->h_buffer_credits = 0;
 429                CWARN("setting credits of handle %p to zero by request\n", h);
 430        }
 431
 432        if (handle->h_buffer_credits > nblocks)
 433                 return 0;
 434        if (journal_extend(handle, nblocks) == 0)
 435                 return 0;
 436
 437        ext3_mark_inode_dirty(handle, inode);
 438        return journal_restart(handle, nblocks);
 439 }
 440
 441 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 442 {
 443         int rc;
 444         handle_t *handle = h;
 445
 446         LASSERT(current->journal_info == handle);
 447         if (force_sync)
 448                 handle->h_sync = 1; /* recovery likes this */
 449
 450         rc = fsfilt_ext3_journal_stop(handle);
 451
 452         return rc;
 453 }
 454
 455 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 456                                     void **wait_handle)
 457 {
 458         unsigned long tid;
 459         transaction_t *transaction;
 460         handle_t *handle = h;
 461         journal_t *journal;
 462         int rc;
 463
 464         LASSERT(current->journal_info == handle);
 465
 466         transaction = handle->h_transaction;
 467         journal = transaction->t_journal;
 468         tid = transaction->t_tid;
 469         /* we don't want to be blocked */
 470         handle->h_sync = 0;
 471         rc = fsfilt_ext3_journal_stop(handle);
 472         if (rc) {
 473                 CERROR("error while stopping transaction: %d\n", rc);
 474                 return rc;
 475         }
 476         log_start_commit(journal, tid);
 477
 478         *wait_handle = (void *) tid;
 479         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 480         return 0;
 481 }
 482
 483 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 484 {
 485         journal_t *journal = EXT3_JOURNAL(inode);
 486         tid_t tid = (tid_t)(long)h;
 487
 488         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 489         if (unlikely(is_journal_aborted(journal)))
 490                 return -EIO;
 491
 492         log_wait_commit(EXT3_JOURNAL(inode), tid);
 493
 494         if (unlikely(is_journal_aborted(journal)))
 495                 return -EIO;
 496         return 0;
 497 }
 498
 499 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 500                                struct iattr *iattr, int do_trunc)
 501 {
 502         struct inode *inode = dentry->d_inode;
 503         int rc = 0;
 504
 505         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 506          * We are already writing the inode to disk as part of this
 507          * transaction and want to avoid a lot of extra inode writeout
 508          * later on. b=9828 */
 509         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 510                 /* ATTR_SIZE would invoke truncate: clear it */
 511                 iattr->ia_valid &= ~ATTR_SIZE;
 512                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 513                 i_size_write(inode, iattr->ia_size);
 514
 515                 if (iattr->ia_valid & ATTR_UID)
 516                         inode->i_uid = iattr->ia_uid;
 517                 if (iattr->ia_valid & ATTR_GID)
 518                         inode->i_gid = iattr->ia_gid;
 519                 if (iattr->ia_valid & ATTR_ATIME)
 520                         inode->i_atime = iattr->ia_atime;
 521                 if (iattr->ia_valid & ATTR_MTIME)
 522                         inode->i_mtime = iattr->ia_mtime;
 523                 if (iattr->ia_valid & ATTR_CTIME)
 524                         inode->i_ctime = iattr->ia_ctime;
 525                 if (iattr->ia_valid & ATTR_MODE) {
 526                         inode->i_mode = iattr->ia_mode;
 527
 528                         if (!in_group_p(inode->i_gid) &&
 529                             !cfs_capable(CFS_CAP_FSETID))
 530                                 inode->i_mode &= ~S_ISGID;
 531                 }
 532
 533                 inode->i_sb->s_op->dirty_inode(inode);
 534
 535                 goto out;
 536         }
 537
 538         /* Don't allow setattr to change file type */
 539         if (iattr->ia_valid & ATTR_MODE)
 540                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 541                                  (iattr->ia_mode & ~S_IFMT);
 542
 543         /* We set these flags on the client, but have already checked perms
 544          * so don't confuse inode_change_ok. */
 545         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 546
 547         if (inode->i_op->setattr) {
 548                 rc = inode->i_op->setattr(dentry, iattr);
 549         } else {
 550                 rc = inode_change_ok(inode, iattr);
 551                 if (!rc)
 552                         rc = inode_setattr(inode, iattr);
 553         }
 554
 555  out:
 556         RETURN(rc);
 557 }
 558
 559 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
 560                                  unsigned int cmd, unsigned long arg)
 561 {
 562         int rc = 0;
 563         ENTRY;
 564
 565         /* FIXME: Can't do this because of nested transaction deadlock */
 566         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 567                 CERROR("can't set data journal flag on file\n");
 568                 RETURN(-EPERM);
 569         }
 570
 571         if (inode->i_fop->ioctl)
 572                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 573         else
 574                 RETURN(-ENOTTY);
 575
 576         RETURN(rc);
 577 }
 578
 579 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 580                               void *lmm, int lmm_size, const char *name)
 581 {
 582         int rc;
 583
 584         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 585
 586         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 587                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 588
 589
 590         if (rc && rc != -EROFS)
 591                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 592                        inode->i_ino, rc);
 593         return rc;
 594 }
 595
 596 /* Must be called with i_mutex held */
 597 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 598                               const char *name)
 599 {
 600         int rc;
 601
 602         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 603
 604         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 605                             name, lmm, lmm_size);
 606
 607         /* This gives us the MD size */
 608         if (lmm == NULL)
 609                 return (rc == -ENODATA) ? 0 : rc;
 610
 611         if (rc < 0) {
 612                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 613                        EXT3_XATTR_INDEX_TRUSTED, name,
 614                        inode->i_ino, rc);
 615                 memset(lmm, 0, lmm_size);
 616                 return (rc == -ENODATA) ? 0 : rc;
 617         }
 618
 619         return rc;
 620 }
 621
 622 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 623 {
 624         submit_bio(rw, bio);
 625         return 0;
 626 }
 627
 628 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 629                                     loff_t *off)
 630 {
 631         struct inode *inode = file->f_dentry->d_inode;
 632         int rc = 0;
 633
 634         if (S_ISREG(inode->i_mode))
 635                 rc = file->f_op->read(file, buf, count, off);
 636         else {
 637                 const int blkbits = inode->i_sb->s_blocksize_bits;
 638                 const int blksize = inode->i_sb->s_blocksize;
 639
 640                 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
 641                        count, inode->i_ino, *off);
 642                 while (count > 0) {
 643                         struct buffer_head *bh;
 644
 645                         bh = NULL;
 646                         if (*off < i_size_read(inode)) {
 647                                 int err = 0;
 648
 649                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 650                                                 0, &err);
 651
 652                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 653
 654                                 if (bh) {
 655                                         memcpy(buf, bh->b_data, blksize);
 656                                         brelse(bh);
 657                                 } else if (err) {
 658                                         /* XXX in theory we should just fake
 659                                          * this buffer and continue like ext3,
 660                                          * especially if this is a partial read
 661                                          */
 662                                         CERROR("error read dir %lu+%llu: %d\n",
 663                                                inode->i_ino, *off, err);
 664                                         RETURN(err);
 665                                 }
 666                         }
 667                         if (!bh) {
 668                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 669
 670                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 671                                 memset(fake, 0, sizeof(*fake));
 672                                 fake->rec_len = cpu_to_le16(blksize);
 673                         }
 674                         count -= blksize;
 675                         buf += blksize;
 676                         *off += blksize;
 677                         rc += blksize;
 678                 }
 679         }
 680
 681         return rc;
 682 }
 683
 684 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 685 {
 686         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 687
 688         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 689
 690         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 691 }
 692
 693 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 694                                       void *handle, fsfilt_cb_t cb_func,
 695                                       void *cb_data)
 696 {
 697         struct fsfilt_cb_data *fcb;
 698
 699         OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
 700         if (fcb == NULL)
 701                 RETURN(-ENOMEM);
 702
 703         fcb->cb_func = cb_func;
 704         fcb->cb_obd = obd;
 705         fcb->cb_last_rcvd = last_rcvd;
 706         fcb->cb_data = cb_data;
 707
 708         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 709         journal_callback_set(handle, fsfilt_ext3_cb_func,
 710                              (struct journal_callback *)fcb);
 711
 712         return 0;
 713 }
 714
 715 /*
 716  * We need to hack the return value for the free inode counts because
 717  * the current EA code requires one filesystem block per inode with EAs,
 718  * so it is possible to run out of blocks before we run out of inodes.
 719  *
 720  * This can be removed when the ext3 EA code is fixed.
 721  */
 722 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 723 {
 724         struct kstatfs sfs;
 725         int rc;
 726
 727         memset(&sfs, 0, sizeof(sfs));
 728         rc = ll_do_statfs(sb, &sfs);
 729         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 730                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 731                 sfs.f_ffree = sfs.f_bfree;
 732         }
 733
 734         statfs_pack(osfs, &sfs);
 735         return rc;
 736 }
 737
 738 static int fsfilt_ext3_sync(struct super_block *sb)
 739 {
 740         return ext3_force_commit(sb);
 741 }
 742
 743 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 744 #warning "kernel code has old extents/mballoc patch, disabling"
 745 #undef EXT3_MULTIBLOCK_ALLOCATOR
 746 #endif
 747 #ifndef EXT3_EXTENTS_FL
 748 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 749 #endif
 750
 751 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 752 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 753 #define ext3_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 754 #define ext3_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 755 #else
 756 #define ext3_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 757 #define ext3_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 758 #endif
 759
 760 #ifndef EXT_ASSERT
 761 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 762 #endif
 763
 764 #ifdef EXT3_EXT_HAS_NO_TREE
 765 /* for kernels 2.6.18 and later */
 766 #define ext3_ext_base                   inode
 767 #define ext3_ext_base2inode(inode)      (inode)
 768 #define EXT_DEPTH(inode)                ext_depth(inode)
 769 #define EXT_GENERATION(inode)           ext_generation(inode)
 770 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 771                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 772 #else
 773 #define ext3_ext_base                   ext3_extents_tree
 774 #define ext3_ext_base2inode(tree)       (tree->inode)
 775 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 776                         ext3_ext_walk_space(tree, block, num, cb);
 777 #endif
 778
 779 #include <linux/lustre_version.h>
 780 #if EXT3_EXT_MAGIC == 0xf301
 781 #define ee_start e_start
 782 #define ee_block e_block
 783 #define ee_len   e_num
 784 #endif
 785 #ifndef EXT3_BB_MAX_BLOCKS
 786 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
 787         ext3_new_blocks(handle, inode, count, goal, err)
 788 #endif
 789
 790 struct bpointers {
 791         unsigned long *blocks;
 792         int *created;
 793         unsigned long start;
 794         int num;
 795         int init_num;
 796         int create;
 797 };
 798
 799 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 800                               unsigned long block, int *aflags)
 801 {
 802         struct ext3_inode_info *ei = EXT3_I(inode);
 803         unsigned long bg_start;
 804         unsigned long colour;
 805         int depth;
 806
 807         if (path) {
 808                 struct ext3_extent *ex;
 809                 depth = path->p_depth;
 810
 811                 /* try to predict block placement */
 812                 if ((ex = path[depth].p_ext)) {
 813 #if 0
 814                         /* This prefers to eat into a contiguous extent
 815                          * rather than find an extent that the whole
 816                          * request will fit into.  This can fragment data
 817                          * block allocation and prevents our lovely 1M I/Os
 818                          * from reaching the disk intact. */
 819                         if (ex->ee_block + ex->ee_len == block)
 820                                 *aflags |= 1;
 821 #endif
 822                         return ex->ee_start + (block - ex->ee_block);
 823                 }
 824
 825                 /* it looks index is empty
 826                  * try to find starting from index itself */
 827                 if (path[depth].p_bh)
 828                         return path[depth].p_bh->b_blocknr;
 829         }
 830
 831         /* OK. use inode's group */
 832         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 833                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 834         colour = (current->pid % 16) *
 835                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 836         return bg_start + colour + block;
 837 }
 838
 839 #define ll_unmap_underlying_metadata(sb, blocknr) \
 840         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 841
 842 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 843 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 844                                 struct ext3_ext_path *path, unsigned long block,
 845                                 unsigned long *count, int *err)
 846 {
 847         unsigned long pblock, goal;
 848         int aflags = 0;
 849         struct inode *inode = ext3_ext_base2inode(base);
 850
 851         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 852         aflags |= 2; /* block have been already reserved */
 853         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 854         return pblock;
 855
 856 }
 857 #else
 858 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 859                                 struct ext3_ext_path *path, unsigned long block,
 860                                 unsigned long *count, int *err)
 861 {
 862         struct inode *inode = ext3_ext_base2inode(base);
 863         struct ext3_allocation_request ar;
 864         unsigned long pblock;
 865         int aflags;
 866
 867         /* find neighbour allocated blocks */
 868         ar.lleft = block;
 869         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 870         if (*err)
 871                 return 0;
 872         ar.lright = block;
 873         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 874         if (*err)
 875                 return 0;
 876
 877         /* allocate new block */
 878         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 879         ar.inode = inode;
 880         ar.logical = block;
 881         ar.len = *count;
 882         ar.flags = EXT3_MB_HINT_DATA;
 883         pblock = ext3_mb_new_blocks(handle, &ar, err);
 884         *count = ar.len;
 885         return pblock;
 886 }
 887 #endif
 888
 889 #ifdef EXT3_EXT_HAS_NO_TREE
 890 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 891                                   struct ext3_ext_path *path,
 892                                   struct ext3_ext_cache *cex,
 893 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 894                                    struct ext3_extent *ex,
 895 #endif
 896                                   void *cbdata)
 897 {
 898         struct bpointers *bp = cbdata;
 899 #else
 900 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 901                                   struct ext3_ext_path *path,
 902                                   struct ext3_ext_cache *cex
 903 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 904                                   , struct ext3_extent *ex
 905 #endif
 906                                  )
 907 {
 908         struct bpointers *bp = base->private;
 909 #endif
 910         struct inode *inode = ext3_ext_base2inode(base);
 911         struct ext3_extent nex;
 912         unsigned long pblock;
 913         unsigned long tgen;
 914         int err, i;
 915         unsigned long count;
 916         handle_t *handle;
 917
 918         i = EXT_DEPTH(base);
 919         EXT_ASSERT(i == path->p_depth);
 920         EXT_ASSERT(path[i].p_hdr);
 921
 922         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 923                 err = EXT_CONTINUE;
 924                 goto map;
 925         }
 926
 927         if (bp->create == 0) {
 928                 i = 0;
 929                 if (cex->ec_block < bp->start)
 930                         i = bp->start - cex->ec_block;
 931                 if (i >= cex->ec_len)
 932                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 933                                         i, cex->ec_len);
 934                 for (; i < cex->ec_len && bp->num; i++) {
 935                         *(bp->created) = 0;
 936                         bp->created++;
 937                         *(bp->blocks) = 0;
 938                         bp->blocks++;
 939                         bp->num--;
 940                         bp->start++;
 941                 }
 942
 943                 return EXT_CONTINUE;
 944         }
 945
 946         tgen = EXT_GENERATION(base);
 947         count = ext3_ext_calc_credits_for_insert(base, path);
 948         ext3_up_truncate_sem(inode);
 949
 950         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 951         if (IS_ERR(handle)) {
 952                 ext3_down_truncate_sem(inode);
 953                 return PTR_ERR(handle);
 954         }
 955
 956         ext3_down_truncate_sem(inode);
 957         if (tgen != EXT_GENERATION(base)) {
 958                 /* the tree has changed. so path can be invalid at moment */
 959                 fsfilt_ext3_journal_stop(handle);
 960                 return EXT_REPEAT;
 961         }
 962
 963         count = cex->ec_len;
 964         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 965         if (!pblock)
 966                 goto out;
 967         EXT_ASSERT(count <= cex->ec_len);
 968
 969         /* insert new extent */
 970         nex.ee_block = cex->ec_block;
 971         nex.ee_start = pblock;
 972         nex.ee_len = count;
 973         err = ext3_ext_insert_extent(handle, base, path, &nex);
 974         if (err) {
 975                 /* free data blocks we just allocated */
 976                 /* not a good idea to call discard here directly,
 977                  * but otherwise we'd need to call it every free() */
 978 #ifdef EXT3_MB_HINT_GROUP_ALLOC
 979                 ext3_mb_discard_inode_preallocations(inode);
 980 #endif
 981                 ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);
 982                 goto out;
 983         }
 984
 985         /*
 986          * Putting len of the actual extent we just inserted,
 987          * we are asking ext3_ext_walk_space() to continue
 988          * scaning after that block
 989          */
 990         cex->ec_len = nex.ee_len;
 991         cex->ec_start = nex.ee_start;
 992         BUG_ON(nex.ee_len == 0);
 993         BUG_ON(nex.ee_block != cex->ec_block);
 994
 995 out:
 996         fsfilt_ext3_journal_stop(handle);
 997 map:
 998         if (err >= 0) {
 999                 /* map blocks */
1000                 if (bp->num == 0) {
1001                         CERROR("hmm. why do we find this extent?\n");
1002                         CERROR("initial space: %lu:%u\n",
1003                                 bp->start, bp->init_num);
1004                         CERROR("current extent: %u/%u/%u %d\n",
1005                                 cex->ec_block, cex->ec_len,
1006                                 cex->ec_start, cex->ec_type);
1007                 }
1008                 i = 0;
1009                 if (cex->ec_block < bp->start)
1010                         i = bp->start - cex->ec_block;
1011                 if (i >= cex->ec_len)
1012                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1013                                         i, cex->ec_len);
1014                 for (; i < cex->ec_len && bp->num; i++) {
1015                         *(bp->blocks) = cex->ec_start + i;
1016                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1017                                 *(bp->created) = 0;
1018                         } else {
1019                                 *(bp->created) = 1;
1020                                 /* unmap any possible underlying metadata from
1021                                  * the block device mapping.  bug 6998. */
1022                                 ll_unmap_underlying_metadata(inode->i_sb,
1023                                                              *(bp->blocks));
1024                         }
1025                         bp->created++;
1026                         bp->blocks++;
1027                         bp->num--;
1028                         bp->start++;
1029                 }
1030         }
1031         return err;
1032 }
1033
1034 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1035                        unsigned long num, unsigned long *blocks,
1036                        int *created, int create)
1037 {
1038 #ifdef EXT3_EXT_HAS_NO_TREE
1039         struct ext3_ext_base *base = inode;
1040 #else
1041         struct ext3_extents_tree tree;
1042         struct ext3_ext_base *base = &tree;
1043 #endif
1044         struct bpointers bp;
1045         int err;
1046
1047         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1048                block, block + num - 1, (unsigned) inode->i_ino);
1049
1050 #ifndef EXT3_EXT_HAS_NO_TREE
1051         ext3_init_tree_desc(base, inode);
1052         tree.private = &bp;
1053 #endif
1054         bp.blocks = blocks;
1055         bp.created = created;
1056         bp.start = block;
1057         bp.init_num = bp.num = num;
1058         bp.create = create;
1059
1060         ext3_down_truncate_sem(inode);
1061         err = fsfilt_ext3_ext_walk_space(base, block, num,
1062                                          ext3_ext_new_extent_cb, &bp);
1063         ext3_ext_invalidate_cache(base);
1064         ext3_up_truncate_sem(inode);
1065
1066         return err;
1067 }
1068
1069 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1070                                     int pages, unsigned long *blocks,
1071                                     int *created, int create)
1072 {
1073         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1074         int rc = 0, i = 0;
1075         struct page *fp = NULL;
1076         int clen = 0;
1077
1078         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1079                 inode->i_ino, pages, (*page)->index);
1080
1081         /* pages are sorted already. so, we just have to find
1082          * contig. space and process them properly */
1083         while (i < pages) {
1084                 if (fp == NULL) {
1085                         /* start new extent */
1086                         fp = *page++;
1087                         clen = 1;
1088                         i++;
1089                         continue;
1090                 } else if (fp->index + clen == (*page)->index) {
1091                         /* continue the extent */
1092                         page++;
1093                         clen++;
1094                         i++;
1095                         continue;
1096                 }
1097
1098                 /* process found extent */
1099                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1100                                         clen * blocks_per_page, blocks,
1101                                         created, create);
1102                 if (rc)
1103                         GOTO(cleanup, rc);
1104
1105                 /* look for next extent */
1106                 fp = NULL;
1107                 blocks += blocks_per_page * clen;
1108                 created += blocks_per_page * clen;
1109         }
1110
1111         if (fp)
1112                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1113                                         clen * blocks_per_page, blocks,
1114                                         created, create);
1115 cleanup:
1116         return rc;
1117 }
1118 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1119
1120 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1121                                unsigned long *blocks, int *created, int create);
1122 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1123                                    int pages, unsigned long *blocks,
1124                                    int *created, int create)
1125 {
1126         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1127         unsigned long *b;
1128         int rc = 0, i, *cr;
1129
1130         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1131                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1132                 if (rc) {
1133                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1134                                inode->i_ino, *b, *cr, create, rc);
1135                         break;
1136                 }
1137
1138                 b += blocks_per_page;
1139                 cr += blocks_per_page;
1140         }
1141         return rc;
1142 }
1143
1144 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1145                                 int pages, unsigned long *blocks,
1146                                 int *created, int create,
1147                                 struct semaphore *optional_sem)
1148 {
1149         int rc;
1150 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1151         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1152                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1153                                                      blocks, created, create);
1154                 return rc;
1155         }
1156 #endif
1157         if (optional_sem != NULL)
1158                 down(optional_sem);
1159         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1160                                             created, create);
1161         if (optional_sem != NULL)
1162                 up(optional_sem);
1163
1164         return rc;
1165 }
1166
1167 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1168 {
1169         unsigned long block;
1170         struct buffer_head *bh;
1171         int err, blocksize, csize, boffs, osize = size;
1172
1173         /* prevent reading after eof */
1174         lock_kernel();
1175         if (i_size_read(inode) < *offs + size) {
1176                 size = i_size_read(inode) - *offs;
1177                 unlock_kernel();
1178                 if (size < 0) {
1179                         CERROR("size %llu is too short for read %u@%llu\n",
1180                                i_size_read(inode), size, *offs);
1181                         return -EIO;
1182                 } else if (size == 0) {
1183                         return 0;
1184                 }
1185         } else {
1186                 unlock_kernel();
1187         }
1188
1189         blocksize = 1 << inode->i_blkbits;
1190
1191         while (size > 0) {
1192                 block = *offs >> inode->i_blkbits;
1193                 boffs = *offs & (blocksize - 1);
1194                 csize = min(blocksize - boffs, size);
1195                 bh = ext3_bread(NULL, inode, block, 0, &err);
1196                 if (!bh) {
1197                         CERROR("can't read block: %d\n", err);
1198                         return err;
1199                 }
1200
1201                 memcpy(buf, bh->b_data + boffs, csize);
1202                 brelse(bh);
1203
1204                 *offs += csize;
1205                 buf += csize;
1206                 size -= csize;
1207         }
1208         return osize;
1209 }
1210 EXPORT_SYMBOL(fsfilt_ext3_read);
1211
1212 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1213                                    int size, loff_t *offs)
1214 {
1215         int rc;
1216         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1217         if (rc > 0)
1218                 rc = 0;
1219         return rc;
1220 }
1221
1222 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1223                                 loff_t *offs, handle_t *handle)
1224 {
1225         struct buffer_head *bh = NULL;
1226         loff_t old_size = i_size_read(inode), offset = *offs;
1227         loff_t new_size = i_size_read(inode);
1228         unsigned long block;
1229         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1230
1231         while (bufsize > 0) {
1232                 if (bh != NULL)
1233                         brelse(bh);
1234
1235                 block = offset >> inode->i_blkbits;
1236                 boffs = offset & (blocksize - 1);
1237                 size = min(blocksize - boffs, bufsize);
1238                 bh = ext3_bread(handle, inode, block, 1, &err);
1239                 if (!bh) {
1240                         CERROR("can't read/create block: %d\n", err);
1241                         break;
1242                 }
1243
1244                 err = ext3_journal_get_write_access(handle, bh);
1245                 if (err) {
1246                         CERROR("journal_get_write_access() returned error %d\n",
1247                                err);
1248                         break;
1249                 }
1250                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1251                 memcpy(bh->b_data + boffs, buf, size);
1252                 err = ext3_journal_dirty_metadata(handle, bh);
1253                 if (err) {
1254                         CERROR("journal_dirty_metadata() returned error %d\n",
1255                                err);
1256                         break;
1257                 }
1258                 if (offset + size > new_size)
1259                         new_size = offset + size;
1260                 offset += size;
1261                 bufsize -= size;
1262                 buf += size;
1263         }
1264         if (bh)
1265                 brelse(bh);
1266
1267         /* correct in-core and on-disk sizes */
1268         if (new_size > i_size_read(inode)) {
1269                 lock_kernel();
1270                 if (new_size > i_size_read(inode))
1271                         i_size_write(inode, new_size);
1272                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1273                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1274                 if (i_size_read(inode) > old_size)
1275                         mark_inode_dirty(inode);
1276                 unlock_kernel();
1277         }
1278
1279         if (err == 0)
1280                 *offs = offset;
1281         return err;
1282 }
1283 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1284
1285 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1286                                     loff_t *offs, int force_sync)
1287 {
1288         struct inode *inode = file->f_dentry->d_inode;
1289         handle_t *handle;
1290         int err, block_count = 0, blocksize;
1291
1292         /* Determine how many transaction credits are needed */
1293         blocksize = 1 << inode->i_blkbits;
1294         block_count = (*offs & (blocksize - 1)) + bufsize;
1295         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1296
1297         handle = fsfilt_ext3_journal_start(inode,
1298                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1299         if (IS_ERR(handle)) {
1300                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1301                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1302                 return PTR_ERR(handle);
1303         }
1304
1305         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1306
1307         if (!err && force_sync)
1308                 handle->h_sync = 1; /* recovery likes this */
1309
1310         fsfilt_ext3_journal_stop(handle);
1311
1312         return err;
1313 }
1314
1315 static int fsfilt_ext3_setup(struct super_block *sb)
1316 {
1317 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1318      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1319         struct ext3_sb_info *sbi = EXT3_SB(sb);
1320 #if 0
1321         sbi->dx_lock = fsfilt_ext3_dx_lock;
1322         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1323 #endif
1324 #endif
1325 #ifdef S_PDIROPS
1326         CWARN("Enabling PDIROPS\n");
1327         set_opt(sbi->s_mount_opt, PDIROPS);
1328         sb->s_flags |= S_PDIROPS;
1329 #endif
1330         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1331                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1332 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1333         /* enable journaled quota support */
1334         /* kfreed in ext3_put_super() */
1335         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1336         if (!sbi->s_qf_names[USRQUOTA])
1337                 return -ENOMEM;
1338         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1339         if (!sbi->s_qf_names[GRPQUOTA]) {
1340                 kfree(sbi->s_qf_names[USRQUOTA]);
1341                 sbi->s_qf_names[USRQUOTA] = NULL;
1342                 return -ENOMEM;
1343         }
1344         sbi->s_jquota_fmt = QFMT_VFS_V0;
1345 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1346         set_opt(sbi->s_mount_opt, QUOTA);
1347 #endif
1348 #endif
1349         return 0;
1350 }
1351
1352 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1353    objects. Logs is number of logfiles to update */
1354 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1355 {
1356         if ( !fso ) {
1357                 switch(op) {
1358                 case FSFILT_OP_CREATE:
1359                                  /* directory leaf, index & indirect & EA*/
1360                         return 4 + 3 * logs;
1361                 case FSFILT_OP_UNLINK:
1362                         return 3 * logs;
1363                 }
1364         } else {
1365                 int i;
1366                 int needed = 0;
1367                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1368                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1369                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1370                 for (i = 0; i < op; i++, fso++) {
1371                         int nblocks = fso->fso_bufcnt * blockpp;
1372                         int ndindirect = min(nblocks, addrpp + 1);
1373                         int nindir = nblocks + ndindirect + 1;
1374
1375                         needed += nindir;
1376                 }
1377                 return needed + 3 * logs;
1378         }
1379
1380         return 0;
1381 }
1382
1383 #ifdef HAVE_QUOTA_SUPPORT
1384 #define DQINFO_COPY(out, in)                    \
1385 do {                                            \
1386         Q_COPY(out, in, dqi_bgrace);            \
1387         Q_COPY(out, in, dqi_igrace);            \
1388         Q_COPY(out, in, dqi_flags);             \
1389         Q_COPY(out, in, dqi_valid);             \
1390 } while (0)
1391
1392 #define DQBLK_COPY(out, in)                     \
1393 do {                                            \
1394         Q_COPY(out, in, dqb_bhardlimit);        \
1395         Q_COPY(out, in, dqb_bsoftlimit);        \
1396         Q_COPY(out, in, dqb_curspace);          \
1397         Q_COPY(out, in, dqb_ihardlimit);        \
1398         Q_COPY(out, in, dqb_isoftlimit);        \
1399         Q_COPY(out, in, dqb_curinodes);         \
1400         Q_COPY(out, in, dqb_btime);             \
1401         Q_COPY(out, in, dqb_itime);             \
1402         Q_COPY(out, in, dqb_valid);             \
1403 } while (0)
1404
1405 static int fsfilt_ext3_quotactl(struct super_block *sb,
1406                                 struct obd_quotactl *oqc)
1407 {
1408         int i, rc = 0, error = 0;
1409         struct quotactl_ops *qcop;
1410         struct if_dqinfo *info;
1411         struct if_dqblk *dqblk;
1412         ENTRY;
1413
1414         if (!sb->s_qcop)
1415                 RETURN(-ENOSYS);
1416
1417         OBD_ALLOC_PTR(info);
1418         if (!info)
1419                 RETURN(-ENOMEM);
1420         OBD_ALLOC_PTR(dqblk);
1421         if (!dqblk) {
1422                 OBD_FREE_PTR(info);
1423                 RETURN(-ENOMEM);
1424         }
1425
1426         DQINFO_COPY(info, &oqc->qc_dqinfo);
1427         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1428
1429         qcop = sb->s_qcop;
1430         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1431                 for (i = 0; i < MAXQUOTAS; i++) {
1432                         if (!Q_TYPESET(oqc, i))
1433                                 continue;
1434
1435                         if (oqc->qc_cmd == Q_QUOTAON) {
1436                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1437
1438                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1439
1440                                 if (!qcop->quota_on)
1441                                         GOTO(out, rc = -ENOSYS);
1442
1443                                 rc = qcop->quota_on(sb, i, QFMT_VFS_V0,
1444                                                     name[i]);
1445                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1446                                 if (!qcop->quota_off)
1447                                         GOTO(out, rc = -ENOSYS);
1448                                 rc = qcop->quota_off(sb, i);
1449                         }
1450
1451                         if (rc == -EBUSY)
1452                                 error = rc;
1453                         else if (rc)
1454                                 GOTO(out, rc);
1455                 }
1456                 GOTO(out, rc ?: error);
1457         }
1458
1459         switch (oqc->qc_cmd) {
1460         case Q_GETOINFO:
1461         case Q_GETINFO:
1462                 if (!qcop->get_info)
1463                         GOTO(out, rc = -ENOSYS);
1464                 rc = qcop->get_info(sb, oqc->qc_type, info);
1465                 break;
1466         case Q_SETQUOTA:
1467         case Q_INITQUOTA:
1468                 if (!qcop->set_dqblk)
1469                         GOTO(out, rc = -ENOSYS);
1470                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1471                 break;
1472         case Q_GETOQUOTA:
1473         case Q_GETQUOTA:
1474                 if (!qcop->get_dqblk)
1475                         GOTO(out, rc = -ENOSYS);
1476                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1477                 if (!rc)
1478                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1479                 break;
1480         case Q_SYNC:
1481                 if (!sb->s_qcop->quota_sync)
1482                         GOTO(out, rc = -ENOSYS);
1483                 qcop->quota_sync(sb, oqc->qc_type);
1484                 break;
1485         case Q_FINVALIDATE:
1486                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1487                 for (i = 0; i < MAXQUOTAS; i++) {
1488                         struct file *fp;
1489                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1490
1491                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1492
1493                         if (!Q_TYPESET(oqc, i))
1494                                 continue;
1495
1496                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1497                         if (IS_ERR(fp)) {
1498                                 rc = PTR_ERR(fp);
1499                                 CERROR("error invalidating operational quota file"
1500                                        " %s (rc:%d)\n", name[i], rc);
1501                         } else {
1502                                 filp_close(fp, 0);
1503                         }
1504
1505                 }
1506                 break;
1507         default:
1508                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1509                 LBUG();
1510         }
1511 out:
1512         DQINFO_COPY(&oqc->qc_dqinfo, info);
1513         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1514
1515         OBD_FREE_PTR(info);
1516         OBD_FREE_PTR(dqblk);
1517
1518         if (rc)
1519                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1520                                 "failed: %d\n",
1521                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1522         RETURN(rc);
1523 }
1524
1525 struct chk_dqblk{
1526         struct hlist_node       dqb_hash;        /** quotacheck hash */
1527         struct list_head        dqb_list;        /** in list also */
1528         qid_t                   dqb_id;          /** uid/gid */
1529         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1530         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1531         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1532         qsize_t                 dqb_curspace;    /** current space */
1533         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1534         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1535         qsize_t                 dqb_curinodes;   /** current inodes */
1536         __u64                   dqb_btime;       /** block grace time */
1537         __u64                   dqb_itime;       /** inode grace time */
1538         __u32                   dqb_valid;       /** flag for above fields */
1539 };
1540
1541 static inline unsigned int chkquot_hash(qid_t id, int type)
1542                                         __attribute__((__const__));
1543
1544 static inline unsigned int chkquot_hash(qid_t id, int type)
1545 {
1546         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1547 }
1548
1549 static inline struct chk_dqblk *
1550 find_chkquot(struct hlist_head *head, qid_t id, int type)
1551 {
1552         struct hlist_node *node;
1553         struct chk_dqblk *cdqb;
1554
1555         hlist_for_each(node, head) {
1556                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1557                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1558                         return cdqb;
1559         }
1560
1561         return NULL;
1562 }
1563
1564 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1565 {
1566         struct chk_dqblk *cdqb;
1567
1568         OBD_ALLOC_PTR(cdqb);
1569         if (cdqb) {
1570                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1571                 INIT_LIST_HEAD(&cdqb->dqb_list);
1572                 cdqb->dqb_id = id;
1573                 cdqb->dqb_type = type;
1574         }
1575
1576         return cdqb;
1577 }
1578
1579 static struct chk_dqblk *
1580 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1581       qid_t id, int type, int first_check)
1582 {
1583         struct hlist_head *head = hash + chkquot_hash(id, type);
1584         struct if_dqblk dqb;
1585         struct chk_dqblk *cdqb;
1586         int rc;
1587
1588         cdqb = find_chkquot(head, id, type);
1589         if (cdqb)
1590                 return cdqb;
1591
1592         cdqb = alloc_chkquot(id, type);
1593         if (!cdqb)
1594                 return NULL;
1595
1596         if (!first_check) {
1597                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1598                 if (rc) {
1599                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1600                                id, type, rc);
1601                 } else {
1602                         DQBLK_COPY(cdqb, &dqb);
1603                         cdqb->dqb_curspace = 0;
1604                         cdqb->dqb_curinodes = 0;
1605                 }
1606         }
1607
1608         hlist_add_head(&cdqb->dqb_hash, head);
1609         list_add_tail(&cdqb->dqb_list, list);
1610
1611         return cdqb;
1612 }
1613
1614 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1615 {
1616         struct obd_quotactl *oqctl;
1617         int rc;
1618
1619         OBD_ALLOC_PTR(oqctl);
1620         if (!oqctl)
1621                 RETURN(-ENOMEM);
1622
1623         oqctl->qc_cmd = cmd;
1624         oqctl->qc_id = qfmt;
1625         oqctl->qc_type = type;
1626         rc = fsfilt_ext3_quotactl(sb, oqctl);
1627
1628         OBD_FREE_PTR(oqctl);
1629         return rc;
1630 }
1631
1632 static inline int read_old_dqinfo(struct super_block *sb, int type,
1633                                   struct if_dqinfo *dqinfo)
1634 {
1635         struct obd_quotactl *oqctl;
1636         int rc;
1637         ENTRY;
1638
1639         OBD_ALLOC_PTR(oqctl);
1640         if (!oqctl)
1641                 RETURN(-ENOMEM);
1642
1643         oqctl->qc_cmd = Q_GETINFO;
1644         oqctl->qc_type = type;
1645         rc = fsfilt_ext3_quotactl(sb, oqctl);
1646         if (!rc)
1647                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1648
1649         OBD_FREE_PTR(oqctl);
1650         RETURN(rc);
1651 }
1652
1653 static inline struct ext3_group_desc *
1654 get_group_desc(struct super_block *sb, int group)
1655 {
1656         unsigned long desc_block, desc;
1657         struct ext3_group_desc *gdp;
1658
1659         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1660         desc = group % EXT3_DESC_PER_BLOCK(sb);
1661         gdp = (struct ext3_group_desc *)
1662               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1663
1664         return gdp + desc;
1665 }
1666
1667 static inline struct buffer_head *
1668 read_inode_bitmap(struct super_block *sb, unsigned long group)
1669 {
1670         struct ext3_group_desc *desc;
1671         struct buffer_head *bh;
1672
1673         desc = get_group_desc(sb, group);
1674         bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1675
1676         return bh;
1677 }
1678
1679 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1680                                      struct buffer_head *bitmap_bh,
1681                                      int index, unsigned long ino)
1682 {
1683         struct inode *inode = NULL;
1684
1685         if (ext3_test_bit(index, bitmap_bh->b_data))
1686                 inode = iget(sb, ino);
1687
1688         return inode;
1689 }
1690
1691 struct qchk_ctxt {
1692         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1693         struct list_head        qckt_list;                   /* quotacheck list */
1694         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1695         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1696 };
1697
1698 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1699                            struct obd_quotactl *oqc)
1700 {
1701         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1702         loff_t size = 0;
1703         qid_t qid[MAXQUOTAS];
1704         int cnt, i, rc = 0;
1705
1706         if (!inode)
1707                 return 0;
1708
1709         qid[USRQUOTA] = inode->i_uid;
1710         qid[GRPQUOTA] = inode->i_gid;
1711
1712         if (S_ISDIR(inode->i_mode) ||
1713             S_ISREG(inode->i_mode) ||
1714             S_ISLNK(inode->i_mode))
1715                 size = inode_get_bytes(inode);
1716
1717         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1718                 if (!Q_TYPESET(oqc, cnt))
1719                         continue;
1720
1721                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1722                                 &qctxt->qckt_list, qid[cnt], cnt,
1723                                 qctxt->qckt_first_check[cnt]);
1724                 if (!cdqb[cnt]) {
1725                         rc = -ENOMEM;
1726                         break;
1727                 }
1728
1729                 cdqb[cnt]->dqb_curspace += size;
1730                 cdqb[cnt]->dqb_curinodes++;
1731         }
1732
1733         if (rc) {
1734                 for (i = 0; i < cnt; i++) {
1735                         if (!Q_TYPESET(oqc, i))
1736                                 continue;
1737                         LASSERT(cdqb[i]);
1738                         cdqb[i]->dqb_curspace -= size;
1739                         cdqb[i]->dqb_curinodes--;
1740                 }
1741         }
1742
1743         return rc;
1744 }
1745
1746 /* write dqinfo struct in a new quota file */
1747 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1748 {
1749         struct v2_disk_dqinfo dqinfo;
1750         __u32 blocks = V2_DQTREEOFF + 1;
1751         loff_t offset = V2_DQINFOOFF;
1752
1753         if (info) {
1754                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1755                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1756                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1757                                                ~DQF_INFO_DIRTY);
1758         } else {
1759                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1760                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1761                 dqinfo.dqi_flags = 0;
1762         }
1763
1764         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1765         dqinfo.dqi_free_blk = 0;
1766         dqinfo.dqi_free_entry = 0;
1767
1768         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1769 }
1770
1771 static int v3_write_dqheader(struct file *f, int type)
1772 {
1773         static const __u32 quota_magics[] = V2_INITQMAGICS;
1774         static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1775         struct v2_disk_dqheader dqhead;
1776         loff_t offset = 0;
1777
1778         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1779         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1780
1781         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1782         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1783
1784         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1785 }
1786
1787 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1788                                   struct obd_quotactl *oqc)
1789 {
1790         int i, rc = 0;
1791         ENTRY;
1792
1793         for (i = 0; i < MAXQUOTAS; i++) {
1794                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1795                                          NULL : &qctxt->qckt_dqinfo[i];
1796                 struct file *file;
1797                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1798
1799                 if (!Q_TYPESET(oqc, i))
1800                         continue;
1801
1802                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1803
1804                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1805                 if (IS_ERR(file)) {
1806                         rc = PTR_ERR(file);
1807                         CERROR("can't create %s file: rc = %d\n",
1808                                name[i], rc);
1809                         GOTO(out, rc);
1810                 }
1811
1812                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1813                         CERROR("file %s is not regular", name[i]);
1814                         filp_close(file, 0);
1815                         GOTO(out, rc = -EINVAL);
1816                 }
1817
1818                 DQUOT_DROP(file->f_dentry->d_inode);
1819
1820                 rc = v3_write_dqheader(file, i);
1821                 if (rc) {
1822                         filp_close(file, 0);
1823                         GOTO(out, rc);
1824                 }
1825
1826                 rc = v3_write_dqinfo(file, i, info);
1827                 filp_close(file, 0);
1828                 if (rc)
1829                         GOTO(out, rc);
1830         }
1831
1832 out:
1833         RETURN(rc);
1834 }
1835
1836
1837 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1838                           struct chk_dqblk *cdqb)
1839 {
1840         struct obd_quotactl *oqc;
1841         long now;
1842         int rc;
1843         ENTRY;
1844
1845         OBD_ALLOC_PTR(oqc);
1846         if (!oqc)
1847                 RETURN(-ENOMEM);
1848
1849         now = cfs_time_current_sec();
1850
1851         if (cdqb->dqb_bsoftlimit &&
1852             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1853             !cdqb->dqb_btime)
1854                 cdqb->dqb_btime =
1855                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1856
1857         if (cdqb->dqb_isoftlimit &&
1858             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1859             !cdqb->dqb_itime)
1860                 cdqb->dqb_itime =
1861                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1862
1863         cdqb->dqb_valid = QIF_ALL;
1864
1865         oqc->qc_cmd = Q_SETQUOTA;
1866         oqc->qc_type = cdqb->dqb_type;
1867         oqc->qc_id = cdqb->dqb_id;
1868         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1869
1870         rc = fsfilt_ext3_quotactl(sb, oqc);
1871         OBD_FREE_PTR(oqc);
1872         RETURN(rc);
1873 }
1874
1875 static int prune_chkquots(struct super_block *sb,
1876                           struct qchk_ctxt *qctxt, int error)
1877 {
1878         struct chk_dqblk *cdqb, *tmp;
1879         int rc;
1880
1881         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1882                 if (!error) {
1883                         rc = commit_chkquot(sb, qctxt, cdqb);
1884                         if (rc)
1885                                 error = rc;
1886                 }
1887                 hlist_del_init(&cdqb->dqb_hash);
1888                 list_del(&cdqb->dqb_list);
1889                 OBD_FREE_PTR(cdqb);
1890         }
1891
1892         return error;
1893 }
1894
1895 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1896                                   struct obd_quotactl *oqc)
1897 {
1898         struct ext3_sb_info *sbi = EXT3_SB(sb);
1899         int i, group;
1900         struct qchk_ctxt *qctxt;
1901         struct buffer_head *bitmap_bh = NULL;
1902         unsigned long ino;
1903         struct inode *inode;
1904         int rc = 0;
1905         ENTRY;
1906
1907         /* turn on quota and read dqinfo if existed */
1908         OBD_ALLOC_PTR(qctxt);
1909         if (!qctxt) {
1910                 oqc->qc_stat = -ENOMEM;
1911                 RETURN(-ENOMEM);
1912         }
1913
1914         for (i = 0; i < NR_DQHASH; i++)
1915                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1916         INIT_LIST_HEAD(&qctxt->qckt_list);
1917
1918         for (i = 0; i < MAXQUOTAS; i++) {
1919                 if (!Q_TYPESET(oqc, i))
1920                         continue;
1921
1922                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1923                 if (!rc || rc == -EBUSY) {
1924                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1925                         if (rc)
1926                                 GOTO(out, rc);
1927                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1928                         qctxt->qckt_first_check[i] = 1;
1929                 } else if (rc) {
1930                         GOTO(out, rc);
1931                 }
1932         }
1933
1934         /* check quota and update in hash */
1935         for (group = 0; group < sbi->s_groups_count; group++) {
1936                 ino = group * sbi->s_inodes_per_group + 1;
1937                 bitmap_bh = read_inode_bitmap(sb, group);
1938                 if (!bitmap_bh) {
1939                         CERROR("read_inode_bitmap group %d failed", group);
1940                         GOTO(out, rc = -EIO);
1941                 }
1942
1943                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1944                         if (ino < sbi->s_first_ino)
1945                                 continue;
1946
1947                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1948                         rc = add_inode_quota(inode, qctxt, oqc);
1949                         iput(inode);
1950                         if (rc) {
1951                                 brelse(bitmap_bh);
1952                                 GOTO(out, rc);
1953                         }
1954                 }
1955
1956                 brelse(bitmap_bh);
1957         }
1958
1959         /* read old quota limits from old quota file. (only for the user
1960          * has limits but hasn't file) */
1961 #ifdef HAVE_QUOTA_SUPPORT
1962         for (i = 0; i < MAXQUOTAS; i++) {
1963                 struct list_head id_list;
1964                 struct dquot_id *dqid, *tmp;
1965
1966                 if (!Q_TYPESET(oqc, i))
1967                         continue;
1968
1969                 if (qctxt->qckt_first_check[i])
1970                         continue;
1971
1972
1973                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1974                 INIT_LIST_HEAD(&id_list);
1975 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1976                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1977 #else
1978                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1979 #endif
1980                 if (rc)
1981                         CERROR("read old limits failed. (rc:%d)\n", rc);
1982
1983                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1984                         list_del_init(&dqid->di_link);
1985
1986                         if (!rc)
1987                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1988                                       dqid->di_id, i,
1989                                       qctxt->qckt_first_check[i]);
1990                         kfree(dqid);
1991                 }
1992         }
1993 #endif
1994         /* turn off quota cause we are to dump chk_dqblk to files */
1995         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
1996
1997         rc = create_new_quota_files(qctxt, oqc);
1998         if (rc)
1999                 GOTO(out, rc);
2000
2001         /* we use vfs functions to set dqblk, so turn quota on */
2002         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2003 out:
2004         /* dump and free chk_dqblk */
2005         rc = prune_chkquots(sb, qctxt, rc);
2006         OBD_FREE_PTR(qctxt);
2007
2008         /* turn off quota, `lfs quotacheck` will turn on when all
2009          * nodes quotacheck finish. */
2010         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2011
2012         oqc->qc_stat = rc;
2013         if (rc)
2014                 CERROR("quotacheck failed: rc = %d\n", rc);
2015
2016         RETURN(rc);
2017 }
2018
2019 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2020                                  int cmd)
2021 {
2022         int rc = 0;
2023         ENTRY;
2024
2025         if (lqi->qi_files[type] == NULL) {
2026                 CERROR("operate qinfo before it's enabled!\n");
2027                 RETURN(-EIO);
2028         }
2029
2030         switch (cmd) {
2031         case QFILE_CHK:
2032                 rc = lustre_check_quota_file(lqi, type);
2033                 break;
2034         case QFILE_RD_INFO:
2035                 rc = lustre_read_quota_info(lqi, type);
2036                 break;
2037         case QFILE_WR_INFO:
2038                 rc = lustre_write_quota_info(lqi, type);
2039                 break;
2040         case QFILE_INIT_INFO:
2041                 rc = lustre_init_quota_info(lqi, type);
2042                 break;
2043         case QFILE_CONVERT:
2044                 rc = -ENOTSUPP;
2045                 CERROR("quota CONVERT command is not supported\n");
2046                 break;
2047         default:
2048                 rc = -ENOTSUPP;
2049                 CERROR("Unsupported admin quota file cmd %d\n"
2050                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2051                        cmd);
2052                 break;
2053         }
2054         RETURN(rc);
2055 }
2056
2057 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2058                             struct list_head *list)
2059 {
2060         return lustre_get_qids(file, inode, type, list);
2061 }
2062
2063 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2064 {
2065         int rc = 0;
2066         ENTRY;
2067
2068         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2069                 CERROR("operate dquot before it's enabled!\n");
2070                 RETURN(-EIO);
2071         }
2072
2073         switch (cmd) {
2074         case QFILE_RD_DQUOT:
2075                 rc = lustre_read_dquot(dquot);
2076                 break;
2077         case QFILE_WR_DQUOT:
2078                 if (dquot->dq_dqb.dqb_ihardlimit ||
2079                     dquot->dq_dqb.dqb_isoftlimit ||
2080                     dquot->dq_dqb.dqb_bhardlimit ||
2081                     dquot->dq_dqb.dqb_bsoftlimit)
2082                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2083                 else
2084                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2085
2086                 rc = lustre_commit_dquot(dquot);
2087                 if (rc >= 0)
2088                         rc = 0;
2089                 break;
2090         default:
2091                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2092                 LBUG();
2093                 break;
2094         }
2095         RETURN(rc);
2096 }
2097 #endif
2098
2099 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2100 {
2101         return (EXT3_SB(sb)->journal_bdev);
2102 }
2103 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2104
2105 static struct fsfilt_operations fsfilt_ext3_ops = {
2106         .fs_type                = "ext3",
2107         .fs_owner               = THIS_MODULE,
2108         .fs_getlabel            = fsfilt_ext3_get_label,
2109         .fs_setlabel            = fsfilt_ext3_set_label,
2110         .fs_uuid                = fsfilt_ext3_uuid,
2111         .fs_start               = fsfilt_ext3_start,
2112         .fs_brw_start           = fsfilt_ext3_brw_start,
2113         .fs_extend              = fsfilt_ext3_extend,
2114         .fs_commit              = fsfilt_ext3_commit,
2115         .fs_commit_async        = fsfilt_ext3_commit_async,
2116         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2117         .fs_setattr             = fsfilt_ext3_setattr,
2118         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2119         .fs_set_md              = fsfilt_ext3_set_md,
2120         .fs_get_md              = fsfilt_ext3_get_md,
2121         .fs_readpage            = fsfilt_ext3_readpage,
2122         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2123         .fs_statfs              = fsfilt_ext3_statfs,
2124         .fs_sync                = fsfilt_ext3_sync,
2125         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2126         .fs_write_record        = fsfilt_ext3_write_record,
2127         .fs_read_record         = fsfilt_ext3_read_record,
2128         .fs_setup               = fsfilt_ext3_setup,
2129         .fs_send_bio            = fsfilt_ext3_send_bio,
2130         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2131 #ifdef HAVE_DISK_INODE_VERSION
2132         .fs_get_version         = fsfilt_ext3_get_version,
2133         .fs_set_version         = fsfilt_ext3_set_version,
2134 #endif
2135 #ifdef HAVE_QUOTA_SUPPORT
2136         .fs_quotactl            = fsfilt_ext3_quotactl,
2137         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2138         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2139         .fs_qids                = fsfilt_ext3_qids,
2140         .fs_dquot               = fsfilt_ext3_dquot,
2141 #endif
2142         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2143 };
2144
2145 static int __init fsfilt_ext3_init(void)
2146 {
2147         int rc;
2148
2149         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2150                                          sizeof(struct fsfilt_cb_data), 0, 0);
2151         if (!fcb_cache) {
2152                 CERROR("error allocating fsfilt journal callback cache\n");
2153                 GOTO(out, rc = -ENOMEM);
2154         }
2155
2156         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2157
2158         if (rc) {
2159                 int err = cfs_mem_cache_destroy(fcb_cache);
2160                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2161         }
2162 out:
2163         return rc;
2164 }
2165
2166 static void __exit fsfilt_ext3_exit(void)
2167 {
2168         int rc;
2169
2170         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2171         rc = cfs_mem_cache_destroy(fcb_cache);
2172         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2173 }
2174
2175 module_init(fsfilt_ext3_init);
2176 module_exit(fsfilt_ext3_exit);
2177
2178 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2179 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2180 MODULE_LICENSE("GPL");