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/slab.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/quotaops.h>
  49 #ifdef HAVE_EXT4_LDISKFS
  50 #include <ext4/ext4.h>
  51 #include <ext4/ext4_jbd2.h>
  52 #else
  53 #include <linux/jbd.h>
  54 #include <linux/ext3_fs.h>
  55 #include <linux/ext3_jbd.h>
  56 #endif
  57 #include <linux/version.h>
  58 #include <linux/bitops.h>
  59 #include <linux/quota.h>
  60 #ifdef HAVE_QUOTAIO_V1_H
  61 # include <linux/quotaio_v1.h>
  62 # include <linux/quotaio_v2.h>
  63 #else
  64 # include <quotaio_v1.h>
  65 # include <quotaio_v2.h>
  66 # include <quota_tree.h>
  67 # define V2_DQTREEOFF    QT_TREEOFF
  68 #endif
  69
  70 #if defined(HAVE_EXT3_XATTR_H)
  71 #include <ext3/xattr.h>
  72 #else
  73 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  74 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  75 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  76 #endif
  77
  78 #include <libcfs/libcfs.h>
  79 #include <lustre_fsfilt.h>
  80 #include <obd.h>
  81 #include <lustre_quota.h>
  82 #include <linux/lustre_compat25.h>
  83 #include <linux/lprocfs_status.h>
  84
  85 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  86 #ifdef HAVE_EXT4_LDISKFS
  87 #include <ext4/ext4_extents.h>
  88 #else
  89 #include <linux/ext3_extents.h>
  90 #endif
  91 #endif
  92
  93 #include "lustre_quota_fmt.h"
  94
  95 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  96 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  97 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  98 #else
  99 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
 100 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
 101 #endif
 102
 103 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
 104 /* for kernels 2.6.18 and later */
 105 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
 106 #else
 107 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 108 #endif
 109
 110 static cfs_mem_cache_t *fcb_cache;
 111
 112 struct fsfilt_cb_data {
 113         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 114         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 115         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 116         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 117         void *cb_data;                  /* MDS/OST completion function data */
 118 };
 119
 120 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 121 #define EXT3_XATTR_INDEX_TRUSTED        4
 122 #endif
 123
 124 #ifdef HAVE_EXT4_LDISKFS
 125 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 126 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 127 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 128 #else
 129 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 130 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 131 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 132 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 133 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 134 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 135 #endif
 136
 137 static char *fsfilt_ext3_get_label(struct super_block *sb)
 138 {
 139         return EXT3_SB(sb)->s_es->s_volume_name;
 140 }
 141
 142 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 143 {
 144         /* see e.g. fsfilt_ext3_write_record() */
 145         journal_t *journal;
 146         handle_t *handle;
 147         int err;
 148
 149         journal = EXT3_SB(sb)->s_journal;
 150         handle = ext3_journal_start_sb(sb, 1);
 151         if (IS_ERR(handle)) {
 152                 CERROR("can't start transaction\n");
 153                 return(PTR_ERR(handle));
 154         }
 155
 156         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 157         if (err)
 158                 goto out;
 159
 160         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 161                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 162
 163         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 164
 165 out:
 166         ext3_journal_stop(handle);
 167
 168         return(err);
 169 }
 170
 171 static char *fsfilt_ext3_uuid(struct super_block *sb)
 172 {
 173         return EXT3_SB(sb)->s_es->s_uuid;
 174 }
 175
 176 #ifdef HAVE_DISK_INODE_VERSION
 177
 178 static __u64 get_i_version(struct inode *inode)
 179 {
 180 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 181         return inode->i_version;
 182 #else
 183         return EXT3_I(inode)->i_fs_version;
 184 #endif
 185 }
 186
 187 static void set_i_version(struct inode *inode, __u64 new_version)
 188 {
 189 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 190         inode->i_version = new_version;
 191 #else
 192         (EXT3_I(inode))->i_fs_version = new_version;
 193 #endif
 194 }
 195
 196 /*
 197  * Get the 64-bit version for an inode.
 198  */
 199 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 200 {
 201         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 202                get_i_version(inode), inode->i_ino);
 203         return get_i_version(inode);
 204 }
 205
 206 /*
 207  * Set the 64-bit version and return the old version.
 208  */
 209 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 210 {
 211         __u64 old_version = get_i_version(inode);
 212
 213         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 214                new_version, old_version, inode->i_ino);
 215         set_i_version(inode, new_version);
 216         /* version is set after all inode operations are finished, so we should
 217          * mark it dirty here */
 218         inode->i_sb->s_op->dirty_inode(inode);
 219         return old_version;
 220 }
 221
 222 #endif
 223
 224 /*
 225  * We don't currently need any additional blocks for rmdir and
 226  * unlink transactions because we are storing the OST oa_id inside
 227  * the inode (which we will be changing anyways as part of this
 228  * transaction).
 229  */
 230 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 231                                int logs)
 232 {
 233         /* For updates to the last received file */
 234         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 235         journal_t *journal;
 236         void *handle;
 237
 238         if (current->journal_info) {
 239                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 240                        current->journal_info);
 241                 goto journal_start;
 242         }
 243
 244         switch(op) {
 245         case FSFILT_OP_RMDIR:
 246         case FSFILT_OP_UNLINK:
 247                 /* delete one file + create/update logs for each stripe */
 248                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 249                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 250                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 251                 break;
 252         case FSFILT_OP_RENAME:
 253                 /* modify additional directory */
 254                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 255                 /* no break */
 256         case FSFILT_OP_SYMLINK:
 257                 /* additional block + block bitmap + GDT for long symlink */
 258                 nblocks += 3;
 259                 /* no break */
 260         case FSFILT_OP_CREATE: {
 261 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 262                 static int warned;
 263                 if (!warned) {
 264                         if (!test_opt(inode->i_sb, EXTENTS)) {
 265                                 warned = 1;
 266                         } else if (((EXT3_I(inode)->i_flags &
 267                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 268                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 269                                 CWARN("extent-mapped directory found with "
 270                                       "ext3-based ldiskfs - contact "
 271                                       "http://bugzilla.lustre.org/\n");
 272                                 warned = 1;
 273                         }
 274                 }
 275 #endif
 276                 /* no break */
 277         }
 278         case FSFILT_OP_MKDIR:
 279         case FSFILT_OP_MKNOD:
 280                 /* modify one inode + block bitmap + GDT */
 281                 nblocks += 3;
 282                 /* no break */
 283         case FSFILT_OP_LINK:
 284                 /* modify parent directory */
 285                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 286                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 287                 /* create/update logs for each stripe */
 288                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 289                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 290                 break;
 291         case FSFILT_OP_SETATTR:
 292                 /* Setattr on inode */
 293                 nblocks += 1;
 294                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 295                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 296                 /* quota chown log for each stripe */
 297                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 298                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 299                 break;
 300         case FSFILT_OP_CANCEL_UNLINK:
 301                 /* blocks for log header bitmap update OR
 302                  * blocks for catalog header bitmap update + unlink of logs */
 303                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 304                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 305                 break;
 306         default: CERROR("unknown transaction start op %d\n", op);
 307                 LBUG();
 308         }
 309
 310         LASSERT(current->journal_info == desc_private);
 311         journal = EXT3_SB(inode->i_sb)->s_journal;
 312         if (nblocks > journal->j_max_transaction_buffers) {
 313                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 314                        nblocks, op, logs, journal->j_max_transaction_buffers);
 315                 nblocks = journal->j_max_transaction_buffers;
 316         }
 317
 318  journal_start:
 319         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 320         handle = ext3_journal_start(inode, nblocks);
 321
 322         if (!IS_ERR(handle))
 323                 LASSERT(current->journal_info == handle);
 324         else
 325                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 326                        op, nblocks, PTR_ERR(handle));
 327         return handle;
 328 }
 329
 330 /*
 331  * Calculate the number of buffer credits needed to write multiple pages in
 332  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 333  * doesn't have a nice API for calculating this sort of thing in advance.
 334  *
 335  * See comment above ext3_writepage_trans_blocks for details.  We assume
 336  * no data journaling is being done, but it does allow for all of the pages
 337  * being non-contiguous.  If we are guaranteed contiguous pages we could
 338  * reduce the number of (d)indirect blocks a lot.
 339  *
 340  * With N blocks per page and P pages, for each inode we have at most:
 341  * N*P indirect
 342  * min(N*P, blocksize/4 + 1) dindirect blocks
 343  * niocount tindirect
 344  *
 345  * For the entire filesystem, we have at most:
 346  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 347  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 348  * objcount inode blocks
 349  * 1 superblock
 350  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 351  *
 352  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 353  */
 354 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 355                                       int niocount, struct niobuf_local *nb)
 356 {
 357         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 358         __u64 next_indir;
 359         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 360         int nbitmaps = 0, ngdblocks;
 361         int needed = objcount + 1; /* inodes + superblock */
 362         int i, j;
 363
 364         for (i = 0, j = 0; i < objcount; i++, fso++) {
 365                 /* two or more dindirect blocks in case we cross boundary */
 366                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 367                                     nb[j].offset) >>
 368                                    sb->s_blocksize_bits) /
 369                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 370                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 371
 372                 /* leaf, indirect, tindirect blocks for first block */
 373                 nbitmaps += blockpp + 2;
 374
 375                 j += fso->fso_bufcnt;
 376         }
 377
 378         next_indir = nb[0].offset +
 379                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 380         for (i = 1; i < niocount; i++) {
 381                 if (nb[i].offset >= next_indir) {
 382                         nbitmaps++;     /* additional indirect */
 383                         next_indir = nb[i].offset +
 384                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 385                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 386                         nbitmaps++;     /* additional indirect */
 387                 }
 388                 nbitmaps += blockpp;    /* each leaf in different group? */
 389         }
 390
 391         ngdblocks = nbitmaps;
 392         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 393                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 394         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 395                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 396
 397         needed += nbitmaps + ngdblocks;
 398
 399         /* last_rcvd update */
 400         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 401
 402 #if defined(CONFIG_QUOTA)
 403         /* We assume that there will be 1 bit set in s_dquot.flags for each
 404          * quota file that is active.  This is at least true for now.
 405          */
 406         needed += hweight32(ll_sb_any_quota_active(sb)) *
 407                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 408 #endif
 409
 410         return needed;
 411 }
 412
 413 /* We have to start a huge journal transaction here to hold all of the
 414  * metadata for the pages being written here.  This is necessitated by
 415  * the fact that we do lots of prepare_write operations before we do
 416  * any of the matching commit_write operations, so even if we split
 417  * up to use "smaller" transactions none of them could complete until
 418  * all of them were opened.  By having a single journal transaction,
 419  * we eliminate duplicate reservations for common blocks like the
 420  * superblock and group descriptors or bitmaps.
 421  *
 422  * We will start the transaction here, but each prepare_write will
 423  * add a refcount to the transaction, and each commit_write will
 424  * remove a refcount.  The transaction will be closed when all of
 425  * the pages have been written.
 426  */
 427 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 428                                    int niocount, struct niobuf_local *nb,
 429                                    void *desc_private, int logs)
 430 {
 431         journal_t *journal;
 432         handle_t *handle;
 433         int needed;
 434         ENTRY;
 435
 436         LASSERT(current->journal_info == desc_private);
 437         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 438         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 439
 440         /* The number of blocks we could _possibly_ dirty can very large.
 441          * We reduce our request if it is absurd (and we couldn't get that
 442          * many credits for a single handle anyways).
 443          *
 444          * At some point we have to limit the size of I/Os sent at one time,
 445          * increase the size of the journal, or we have to calculate the
 446          * actual journal requirements more carefully by checking all of
 447          * the blocks instead of being maximally pessimistic.  It remains to
 448          * be seen if this is a real problem or not.
 449          */
 450         if (needed > journal->j_max_transaction_buffers) {
 451                 CERROR("want too many journal credits (%d) using %d instead\n",
 452                        needed, journal->j_max_transaction_buffers);
 453                 needed = journal->j_max_transaction_buffers;
 454         }
 455
 456         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 457         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 458         if (IS_ERR(handle)) {
 459                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 460                        PTR_ERR(handle));
 461         } else {
 462                 LASSERT(handle->h_buffer_credits >= needed);
 463                 LASSERT(current->journal_info == handle);
 464         }
 465
 466         RETURN(handle);
 467 }
 468
 469 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 470 {
 471        handle_t *handle = h;
 472
 473        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 474        if (nblocks == 0) {
 475                handle->h_buffer_credits = 0;
 476                CWARN("setting credits of handle %p to zero by request\n", h);
 477        }
 478
 479        if (handle->h_buffer_credits > nblocks)
 480                 return 0;
 481        if (ext3_journal_extend(handle, nblocks) == 0)
 482                 return 0;
 483
 484        ext3_mark_inode_dirty(handle, inode);
 485        return ext3_journal_restart(handle, nblocks);
 486 }
 487
 488 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 489 {
 490         int rc;
 491         handle_t *handle = h;
 492
 493         LASSERT(current->journal_info == handle);
 494         if (force_sync)
 495                 handle->h_sync = 1; /* recovery likes this */
 496
 497         rc = ext3_journal_stop(handle);
 498
 499         return rc;
 500 }
 501
 502 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 503                                     void **wait_handle)
 504 {
 505         unsigned long tid;
 506         transaction_t *transaction;
 507         handle_t *handle = h;
 508         journal_t *journal;
 509         int rc;
 510
 511         LASSERT(current->journal_info == handle);
 512
 513         transaction = handle->h_transaction;
 514         journal = transaction->t_journal;
 515         tid = transaction->t_tid;
 516         /* we don't want to be blocked */
 517         handle->h_sync = 0;
 518         rc = ext3_journal_stop(handle);
 519         if (rc) {
 520                 CERROR("error while stopping transaction: %d\n", rc);
 521                 return rc;
 522         }
 523         fsfilt_log_start_commit(journal, tid);
 524
 525         *wait_handle = (void *) tid;
 526         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 527         return 0;
 528 }
 529
 530 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 531 {
 532         journal_t *journal = EXT3_JOURNAL(inode);
 533         tid_t tid = (tid_t)(long)h;
 534
 535         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 536         if (unlikely(is_journal_aborted(journal)))
 537                 return -EIO;
 538
 539         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 540
 541         if (unlikely(is_journal_aborted(journal)))
 542                 return -EIO;
 543         return 0;
 544 }
 545
 546 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 547                                struct iattr *iattr, int do_trunc)
 548 {
 549         struct inode *inode = dentry->d_inode;
 550         int rc = 0;
 551
 552         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 553          * We are already writing the inode to disk as part of this
 554          * transaction and want to avoid a lot of extra inode writeout
 555          * later on. b=9828 */
 556         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 557                 /* ATTR_SIZE would invoke truncate: clear it */
 558                 iattr->ia_valid &= ~ATTR_SIZE;
 559                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 560                 i_size_write(inode, iattr->ia_size);
 561
 562                 if (iattr->ia_valid & ATTR_UID)
 563                         inode->i_uid = iattr->ia_uid;
 564                 if (iattr->ia_valid & ATTR_GID)
 565                         inode->i_gid = iattr->ia_gid;
 566                 if (iattr->ia_valid & ATTR_ATIME)
 567                         inode->i_atime = iattr->ia_atime;
 568                 if (iattr->ia_valid & ATTR_MTIME)
 569                         inode->i_mtime = iattr->ia_mtime;
 570                 if (iattr->ia_valid & ATTR_CTIME)
 571                         inode->i_ctime = iattr->ia_ctime;
 572                 if (iattr->ia_valid & ATTR_MODE) {
 573                         inode->i_mode = iattr->ia_mode;
 574
 575                         if (!in_group_p(inode->i_gid) &&
 576                             !cfs_capable(CFS_CAP_FSETID))
 577                                 inode->i_mode &= ~S_ISGID;
 578                 }
 579
 580                 inode->i_sb->s_op->dirty_inode(inode);
 581
 582                 goto out;
 583         }
 584
 585         /* Don't allow setattr to change file type */
 586         if (iattr->ia_valid & ATTR_MODE)
 587                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 588                                  (iattr->ia_mode & ~S_IFMT);
 589
 590         /* We set these flags on the client, but have already checked perms
 591          * so don't confuse inode_change_ok. */
 592         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 593
 594         if (inode->i_op->setattr) {
 595                 rc = inode->i_op->setattr(dentry, iattr);
 596         } else {
 597                 rc = inode_change_ok(inode, iattr);
 598                 if (!rc)
 599                         rc = inode_setattr(inode, iattr);
 600         }
 601
 602  out:
 603         RETURN(rc);
 604 }
 605
 606 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 607                                  unsigned int cmd, unsigned long arg)
 608 {
 609         int rc = 0;
 610         ENTRY;
 611
 612         /* FIXME: Can't do this because of nested transaction deadlock */
 613         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 614                 CERROR("can't set data journal flag on file\n");
 615                 RETURN(-EPERM);
 616         }
 617
 618 #ifdef HAVE_EXT4_LDISKFS
 619         /* ext4_ioctl does not have a inode argument */
 620         if (inode->i_fop->unlocked_ioctl)
 621                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 622 #else
 623         if (inode->i_fop->ioctl)
 624                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 625 #endif
 626         else
 627                 RETURN(-ENOTTY);
 628
 629         RETURN(rc);
 630 }
 631
 632 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 633                               void *lmm, int lmm_size, const char *name)
 634 {
 635         int rc;
 636
 637         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 638
 639         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 640                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 641
 642
 643         if (rc && rc != -EROFS)
 644                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 645                        inode->i_ino, rc);
 646         return rc;
 647 }
 648
 649 /* Must be called with i_mutex held */
 650 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 651                               const char *name)
 652 {
 653         int rc;
 654
 655         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 656
 657         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 658                             name, lmm, lmm_size);
 659
 660         /* This gives us the MD size */
 661         if (lmm == NULL)
 662                 return (rc == -ENODATA) ? 0 : rc;
 663
 664         if (rc < 0) {
 665                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 666                        EXT3_XATTR_INDEX_TRUSTED, name,
 667                        inode->i_ino, rc);
 668                 memset(lmm, 0, lmm_size);
 669                 return (rc == -ENODATA) ? 0 : rc;
 670         }
 671
 672         return rc;
 673 }
 674
 675 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 676 {
 677         submit_bio(rw, bio);
 678         return 0;
 679 }
 680
 681 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 682                                     loff_t *off)
 683 {
 684         struct inode *inode = file->f_dentry->d_inode;
 685         int rc = 0;
 686
 687         if (S_ISREG(inode->i_mode))
 688                 rc = file->f_op->read(file, buf, count, off);
 689         else {
 690                 const int blkbits = inode->i_sb->s_blocksize_bits;
 691                 const int blksize = inode->i_sb->s_blocksize;
 692
 693                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 694                        (unsigned long)count, inode->i_ino, *off);
 695                 while (count > 0) {
 696                         struct buffer_head *bh;
 697
 698                         bh = NULL;
 699                         if (*off < i_size_read(inode)) {
 700                                 int err = 0;
 701
 702                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 703                                                 0, &err);
 704
 705                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 706
 707                                 if (bh) {
 708                                         memcpy(buf, bh->b_data, blksize);
 709                                         brelse(bh);
 710                                 } else if (err) {
 711                                         /* XXX in theory we should just fake
 712                                          * this buffer and continue like ext3,
 713                                          * especially if this is a partial read
 714                                          */
 715                                         CERROR("error read dir %lu+%llu: %d\n",
 716                                                inode->i_ino, *off, err);
 717                                         RETURN(err);
 718                                 }
 719                         }
 720                         if (!bh) {
 721                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 722
 723                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 724                                 memset(fake, 0, sizeof(*fake));
 725                                 fake->rec_len = cpu_to_le16(blksize);
 726                         }
 727                         count -= blksize;
 728                         buf += blksize;
 729                         *off += blksize;
 730                         rc += blksize;
 731                 }
 732         }
 733
 734         return rc;
 735 }
 736
 737 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 738 {
 739         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 740
 741         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 742
 743         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 744 }
 745
 746 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 747                                       void *handle, fsfilt_cb_t cb_func,
 748                                       void *cb_data)
 749 {
 750         struct fsfilt_cb_data *fcb;
 751
 752         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 753         if (fcb == NULL)
 754                 RETURN(-ENOMEM);
 755
 756         fcb->cb_func = cb_func;
 757         fcb->cb_obd = obd;
 758         fcb->cb_last_rcvd = last_rcvd;
 759         fcb->cb_data = cb_data;
 760
 761         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 762         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 763                                     (struct journal_callback *)fcb);
 764
 765         return 0;
 766 }
 767
 768 /*
 769  * We need to hack the return value for the free inode counts because
 770  * the current EA code requires one filesystem block per inode with EAs,
 771  * so it is possible to run out of blocks before we run out of inodes.
 772  *
 773  * This can be removed when the ext3 EA code is fixed.
 774  */
 775 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 776 {
 777         struct kstatfs sfs;
 778         int rc;
 779
 780         memset(&sfs, 0, sizeof(sfs));
 781         rc = ll_do_statfs(sb, &sfs);
 782         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 783                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 784                 sfs.f_ffree = sfs.f_bfree;
 785         }
 786
 787         statfs_pack(osfs, &sfs);
 788         return rc;
 789 }
 790
 791 static int fsfilt_ext3_sync(struct super_block *sb)
 792 {
 793         return ext3_force_commit(sb);
 794 }
 795
 796 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 797 #warning "kernel code has old extents/mballoc patch, disabling"
 798 #undef EXT3_MULTIBLOCK_ALLOCATOR
 799 #endif
 800 #ifndef EXT3_EXTENTS_FL
 801 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 802 #endif
 803
 804 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 805 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 806 #define fsfilt_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 807 #define fsfilt_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 808 #else
 809 #ifdef HAVE_EXT4_LDISKFS
 810 #define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem));
 811 #define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem));
 812 #else
 813 #define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 814 #define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 815 #endif
 816 #endif
 817
 818 #ifndef EXT_ASSERT
 819 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 820 #endif
 821
 822 #ifdef EXT3_EXT_HAS_NO_TREE
 823 /* for kernels 2.6.18 and later */
 824 #ifdef HAVE_EXT4_LDISKFS
 825 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 826 #else
 827 #define EXT_GENERATION(inode)           ext_generation(inode)
 828 #endif
 829 #define ext3_ext_base                   inode
 830 #define ext3_ext_base2inode(inode)      (inode)
 831 #define EXT_DEPTH(inode)                ext_depth(inode)
 832 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 833                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 834 #else
 835 #define ext3_ext_base                   ext3_extents_tree
 836 #define ext3_ext_base2inode(tree)       (tree->inode)
 837 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 838                         ext3_ext_walk_space(tree, block, num, cb);
 839 #endif
 840
 841 #include <linux/lustre_version.h>
 842
 843 struct bpointers {
 844         unsigned long *blocks;
 845         int *created;
 846         unsigned long start;
 847         int num;
 848         int init_num;
 849         int create;
 850 };
 851
 852 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 853                                unsigned long block, int *aflags)
 854 {
 855         struct ext3_inode_info *ei = EXT3_I(inode);
 856         unsigned long bg_start;
 857         unsigned long colour;
 858         int depth;
 859
 860         if (path) {
 861                 struct ext3_extent *ex;
 862                 depth = path->p_depth;
 863
 864                 /* try to predict block placement */
 865                 if ((ex = path[depth].p_ext))
 866                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 867
 868                 /* it looks index is empty
 869                  * try to find starting from index itself */
 870                 if (path[depth].p_bh)
 871                         return path[depth].p_bh->b_blocknr;
 872         }
 873
 874         /* OK. use inode's group */
 875         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 876                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 877         colour = (current->pid % 16) *
 878                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 879         return bg_start + colour + block;
 880 }
 881
 882 #define ll_unmap_underlying_metadata(sb, blocknr) \
 883         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 884
 885 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 886 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 887                                 struct ext3_ext_path *path, unsigned long block,
 888                                 unsigned long *count, int *err)
 889 {
 890         unsigned long pblock, goal;
 891         int aflags = 0;
 892         struct inode *inode = ext3_ext_base2inode(base);
 893
 894         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 895         aflags |= 2; /* block have been already reserved */
 896         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 897         return pblock;
 898
 899 }
 900 #else
 901 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 902                                 struct ext3_ext_path *path, unsigned long block,
 903                                 unsigned long *count, int *err)
 904 {
 905         struct inode *inode = ext3_ext_base2inode(base);
 906         struct ext3_allocation_request ar;
 907         unsigned long pblock;
 908         int aflags;
 909
 910         /* find neighbour allocated blocks */
 911         ar.lleft = block;
 912         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 913         if (*err)
 914                 return 0;
 915         ar.lright = block;
 916         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 917         if (*err)
 918                 return 0;
 919
 920         /* allocate new block */
 921         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 922         ar.inode = inode;
 923         ar.logical = block;
 924         ar.len = *count;
 925         ar.flags = EXT3_MB_HINT_DATA;
 926         pblock = ext3_mb_new_blocks(handle, &ar, err);
 927         *count = ar.len;
 928         return pblock;
 929 }
 930 #endif
 931
 932 #ifdef EXT3_EXT_HAS_NO_TREE
 933 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 934                                   struct ext3_ext_path *path,
 935                                   struct ext3_ext_cache *cex,
 936 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 937                                    struct ext3_extent *ex,
 938 #endif
 939                                   void *cbdata)
 940 {
 941         struct bpointers *bp = cbdata;
 942 #else
 943 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 944                                   struct ext3_ext_path *path,
 945                                   struct ext3_ext_cache *cex
 946 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 947                                   , struct ext3_extent *ex
 948 #endif
 949                                  )
 950 {
 951         struct bpointers *bp = base->private;
 952 #endif
 953         struct inode *inode = ext3_ext_base2inode(base);
 954         struct ext3_extent nex;
 955         unsigned long pblock;
 956         unsigned long tgen;
 957         int err, i;
 958         unsigned long count;
 959         handle_t *handle;
 960
 961         i = EXT_DEPTH(base);
 962         EXT_ASSERT(i == path->p_depth);
 963         EXT_ASSERT(path[i].p_hdr);
 964
 965         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 966                 err = EXT_CONTINUE;
 967                 goto map;
 968         }
 969
 970         if (bp->create == 0) {
 971                 i = 0;
 972                 if (cex->ec_block < bp->start)
 973                         i = bp->start - cex->ec_block;
 974                 if (i >= cex->ec_len)
 975                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 976                                         i, cex->ec_len);
 977                 for (; i < cex->ec_len && bp->num; i++) {
 978                         *(bp->created) = 0;
 979                         bp->created++;
 980                         *(bp->blocks) = 0;
 981                         bp->blocks++;
 982                         bp->num--;
 983                         bp->start++;
 984                 }
 985
 986                 return EXT_CONTINUE;
 987         }
 988
 989         tgen = EXT_GENERATION(base);
 990         count = ext3_ext_calc_credits_for_insert(base, path);
 991         fsfilt_up_truncate_sem(inode);
 992
 993         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 994         if (IS_ERR(handle)) {
 995                 fsfilt_down_truncate_sem(inode);
 996                 return PTR_ERR(handle);
 997         }
 998
 999         fsfilt_down_truncate_sem(inode);
1000         if (tgen != EXT_GENERATION(base)) {
1001                 /* the tree has changed. so path can be invalid at moment */
1002                 ext3_journal_stop(handle);
1003                 return EXT_REPEAT;
1004         }
1005
1006         count = cex->ec_len;
1007         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1008         if (!pblock)
1009                 goto out;
1010         EXT_ASSERT(count <= cex->ec_len);
1011
1012         /* insert new extent */
1013         nex.ee_block = cpu_to_le32(cex->ec_block);
1014         ext3_ext_store_pblock(&nex, pblock);
1015         nex.ee_len = cpu_to_le16(count);
1016         err = ext3_ext_insert_extent(handle, base, path, &nex);
1017         if (err) {
1018                 /* free data blocks we just allocated */
1019                 /* not a good idea to call discard here directly,
1020                  * but otherwise we'd need to call it every free() */
1021 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1022                 ext3_mb_discard_inode_preallocations(inode);
1023 #endif
1024                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1025                                  cpu_to_le16(nex.ee_len), 0);
1026                 goto out;
1027         }
1028
1029         /*
1030          * Putting len of the actual extent we just inserted,
1031          * we are asking ext3_ext_walk_space() to continue
1032          * scaning after that block
1033          */
1034         cex->ec_len = le16_to_cpu(nex.ee_len);
1035         cex->ec_start = ext_pblock(&nex);
1036         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1037         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1038
1039 out:
1040         ext3_journal_stop(handle);
1041 map:
1042         if (err >= 0) {
1043                 /* map blocks */
1044                 if (bp->num == 0) {
1045                         CERROR("hmm. why do we find this extent?\n");
1046                         CERROR("initial space: %lu:%u\n",
1047                                 bp->start, bp->init_num);
1048                         CERROR("current extent: %u/%u/%llu %d\n",
1049                                 cex->ec_block, cex->ec_len,
1050                                 (unsigned long long)cex->ec_start,
1051                                 cex->ec_type);
1052                 }
1053                 i = 0;
1054                 if (cex->ec_block < bp->start)
1055                         i = bp->start - cex->ec_block;
1056                 if (i >= cex->ec_len)
1057                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1058                                         i, cex->ec_len);
1059                 for (; i < cex->ec_len && bp->num; i++) {
1060                         *(bp->blocks) = cex->ec_start + i;
1061                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1062                                 *(bp->created) = 0;
1063                         } else {
1064                                 *(bp->created) = 1;
1065                                 /* unmap any possible underlying metadata from
1066                                  * the block device mapping.  bug 6998. */
1067                                 ll_unmap_underlying_metadata(inode->i_sb,
1068                                                              *(bp->blocks));
1069                         }
1070                         bp->created++;
1071                         bp->blocks++;
1072                         bp->num--;
1073                         bp->start++;
1074                 }
1075         }
1076         return err;
1077 }
1078
1079 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1080                        unsigned long num, unsigned long *blocks,
1081                        int *created, int create)
1082 {
1083 #ifdef EXT3_EXT_HAS_NO_TREE
1084         struct ext3_ext_base *base = inode;
1085 #else
1086         struct ext3_extents_tree tree;
1087         struct ext3_ext_base *base = &tree;
1088 #endif
1089         struct bpointers bp;
1090         int err;
1091
1092         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1093                block, block + num - 1, (unsigned) inode->i_ino);
1094
1095 #ifndef EXT3_EXT_HAS_NO_TREE
1096         ext3_init_tree_desc(base, inode);
1097         tree.private = &bp;
1098 #endif
1099         bp.blocks = blocks;
1100         bp.created = created;
1101         bp.start = block;
1102         bp.init_num = bp.num = num;
1103         bp.create = create;
1104
1105         fsfilt_down_truncate_sem(inode);
1106         err = fsfilt_ext3_ext_walk_space(base, block, num,
1107                                          ext3_ext_new_extent_cb, &bp);
1108         ext3_ext_invalidate_cache(base);
1109         fsfilt_up_truncate_sem(inode);
1110
1111         return err;
1112 }
1113
1114 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1115                                     int pages, unsigned long *blocks,
1116                                     int *created, int create)
1117 {
1118         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1119         int rc = 0, i = 0;
1120         struct page *fp = NULL;
1121         int clen = 0;
1122
1123         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1124                 inode->i_ino, pages, (*page)->index);
1125
1126         /* pages are sorted already. so, we just have to find
1127          * contig. space and process them properly */
1128         while (i < pages) {
1129                 if (fp == NULL) {
1130                         /* start new extent */
1131                         fp = *page++;
1132                         clen = 1;
1133                         i++;
1134                         continue;
1135                 } else if (fp->index + clen == (*page)->index) {
1136                         /* continue the extent */
1137                         page++;
1138                         clen++;
1139                         i++;
1140                         continue;
1141                 }
1142
1143                 /* process found extent */
1144                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1145                                         clen * blocks_per_page, blocks,
1146                                         created, create);
1147                 if (rc)
1148                         GOTO(cleanup, rc);
1149
1150                 /* look for next extent */
1151                 fp = NULL;
1152                 blocks += blocks_per_page * clen;
1153                 created += blocks_per_page * clen;
1154         }
1155
1156         if (fp)
1157                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1158                                         clen * blocks_per_page, blocks,
1159                                         created, create);
1160 cleanup:
1161         return rc;
1162 }
1163 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1164
1165 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1166                                unsigned long *blocks, int *created, int create);
1167 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1168                                    int pages, unsigned long *blocks,
1169                                    int *created, int create)
1170 {
1171         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1172         unsigned long *b;
1173         int rc = 0, i, *cr;
1174
1175         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1176                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1177                 if (rc) {
1178                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1179                                inode->i_ino, *b, *cr, create, rc);
1180                         break;
1181                 }
1182
1183                 b += blocks_per_page;
1184                 cr += blocks_per_page;
1185         }
1186         return rc;
1187 }
1188
1189 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1190                                 int pages, unsigned long *blocks,
1191                                 int *created, int create,
1192                                 struct semaphore *optional_sem)
1193 {
1194         int rc;
1195 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1196         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1197                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1198                                                      blocks, created, create);
1199                 return rc;
1200         }
1201 #endif
1202         if (optional_sem != NULL)
1203                 down(optional_sem);
1204         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1205                                             created, create);
1206         if (optional_sem != NULL)
1207                 up(optional_sem);
1208
1209         return rc;
1210 }
1211
1212 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1213 {
1214         unsigned long block;
1215         struct buffer_head *bh;
1216         int err, blocksize, csize, boffs, osize = size;
1217
1218         /* prevent reading after eof */
1219         lock_kernel();
1220         if (i_size_read(inode) < *offs + size) {
1221                 size = i_size_read(inode) - *offs;
1222                 unlock_kernel();
1223                 if (size < 0) {
1224                         CERROR("size %llu is too short for read %u@%llu\n",
1225                                i_size_read(inode), size, *offs);
1226                         return -EIO;
1227                 } else if (size == 0) {
1228                         return 0;
1229                 }
1230         } else {
1231                 unlock_kernel();
1232         }
1233
1234         blocksize = 1 << inode->i_blkbits;
1235
1236         while (size > 0) {
1237                 block = *offs >> inode->i_blkbits;
1238                 boffs = *offs & (blocksize - 1);
1239                 csize = min(blocksize - boffs, size);
1240                 bh = ext3_bread(NULL, inode, block, 0, &err);
1241                 if (!bh) {
1242                         CERROR("can't read block: %d\n", err);
1243                         return err;
1244                 }
1245
1246                 memcpy(buf, bh->b_data + boffs, csize);
1247                 brelse(bh);
1248
1249                 *offs += csize;
1250                 buf += csize;
1251                 size -= csize;
1252         }
1253         return osize;
1254 }
1255 EXPORT_SYMBOL(fsfilt_ext3_read);
1256
1257 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1258                                    int size, loff_t *offs)
1259 {
1260         int rc;
1261         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1262         if (rc > 0)
1263                 rc = 0;
1264         return rc;
1265 }
1266
1267 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1268                                 loff_t *offs, handle_t *handle)
1269 {
1270         struct buffer_head *bh = NULL;
1271         loff_t old_size = i_size_read(inode), offset = *offs;
1272         loff_t new_size = i_size_read(inode);
1273         unsigned long block;
1274         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1275
1276         while (bufsize > 0) {
1277                 if (bh != NULL)
1278                         brelse(bh);
1279
1280                 block = offset >> inode->i_blkbits;
1281                 boffs = offset & (blocksize - 1);
1282                 size = min(blocksize - boffs, bufsize);
1283                 bh = ext3_bread(handle, inode, block, 1, &err);
1284                 if (!bh) {
1285                         CERROR("can't read/create block: %d\n", err);
1286                         break;
1287                 }
1288
1289                 err = ext3_journal_get_write_access(handle, bh);
1290                 if (err) {
1291                         CERROR("journal_get_write_access() returned error %d\n",
1292                                err);
1293                         break;
1294                 }
1295                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1296                 memcpy(bh->b_data + boffs, buf, size);
1297                 err = ext3_journal_dirty_metadata(handle, bh);
1298                 if (err) {
1299                         CERROR("journal_dirty_metadata() returned error %d\n",
1300                                err);
1301                         break;
1302                 }
1303                 if (offset + size > new_size)
1304                         new_size = offset + size;
1305                 offset += size;
1306                 bufsize -= size;
1307                 buf += size;
1308         }
1309         if (bh)
1310                 brelse(bh);
1311
1312         /* correct in-core and on-disk sizes */
1313         if (new_size > i_size_read(inode)) {
1314                 lock_kernel();
1315                 if (new_size > i_size_read(inode))
1316                         i_size_write(inode, new_size);
1317                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1318                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1319                 if (i_size_read(inode) > old_size)
1320                         mark_inode_dirty(inode);
1321                 unlock_kernel();
1322         }
1323
1324         if (err == 0)
1325                 *offs = offset;
1326         return err;
1327 }
1328 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1329
1330 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1331                                     loff_t *offs, int force_sync)
1332 {
1333         struct inode *inode = file->f_dentry->d_inode;
1334         handle_t *handle;
1335         int err, block_count = 0, blocksize;
1336
1337         /* Determine how many transaction credits are needed */
1338         blocksize = 1 << inode->i_blkbits;
1339         block_count = (*offs & (blocksize - 1)) + bufsize;
1340         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1341
1342         handle = ext3_journal_start(inode,
1343                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1344         if (IS_ERR(handle)) {
1345                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1346                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1347                 return PTR_ERR(handle);
1348         }
1349
1350         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1351
1352         if (!err && force_sync)
1353                 handle->h_sync = 1; /* recovery likes this */
1354
1355         ext3_journal_stop(handle);
1356
1357         return err;
1358 }
1359
1360 static int fsfilt_ext3_setup(struct super_block *sb)
1361 {
1362 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1363      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1364         struct ext3_sb_info *sbi = EXT3_SB(sb);
1365 #if 0
1366         sbi->dx_lock = fsfilt_ext3_dx_lock;
1367         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1368 #endif
1369 #endif
1370 #ifdef S_PDIROPS
1371         CWARN("Enabling PDIROPS\n");
1372         set_opt(sbi->s_mount_opt, PDIROPS);
1373         sb->s_flags |= S_PDIROPS;
1374 #endif
1375         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1376                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1377 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1378         /* enable journaled quota support */
1379         /* kfreed in ext3_put_super() */
1380         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1381         if (!sbi->s_qf_names[USRQUOTA])
1382                 return -ENOMEM;
1383         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1384         if (!sbi->s_qf_names[GRPQUOTA]) {
1385                 kfree(sbi->s_qf_names[USRQUOTA]);
1386                 sbi->s_qf_names[USRQUOTA] = NULL;
1387                 return -ENOMEM;
1388         }
1389         sbi->s_jquota_fmt = QFMT_VFS_V0;
1390 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1391         set_opt(sbi->s_mount_opt, QUOTA);
1392 #endif
1393 #endif
1394         return 0;
1395 }
1396
1397 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1398    objects. Logs is number of logfiles to update */
1399 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1400 {
1401         if ( !fso ) {
1402                 switch(op) {
1403                 case FSFILT_OP_CREATE:
1404                                  /* directory leaf, index & indirect & EA*/
1405                         return 4 + 3 * logs;
1406                 case FSFILT_OP_UNLINK:
1407                         return 3 * logs;
1408                 }
1409         } else {
1410                 int i;
1411                 int needed = 0;
1412                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1413                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1414                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1415                 for (i = 0; i < op; i++, fso++) {
1416                         int nblocks = fso->fso_bufcnt * blockpp;
1417                         int ndindirect = min(nblocks, addrpp + 1);
1418                         int nindir = nblocks + ndindirect + 1;
1419
1420                         needed += nindir;
1421                 }
1422                 return needed + 3 * logs;
1423         }
1424
1425         return 0;
1426 }
1427
1428 #ifdef HAVE_QUOTA_SUPPORT
1429 #define DQINFO_COPY(out, in)                    \
1430 do {                                            \
1431         Q_COPY(out, in, dqi_bgrace);            \
1432         Q_COPY(out, in, dqi_igrace);            \
1433         Q_COPY(out, in, dqi_flags);             \
1434         Q_COPY(out, in, dqi_valid);             \
1435 } while (0)
1436
1437 #define DQBLK_COPY(out, in)                     \
1438 do {                                            \
1439         Q_COPY(out, in, dqb_bhardlimit);        \
1440         Q_COPY(out, in, dqb_bsoftlimit);        \
1441         Q_COPY(out, in, dqb_curspace);          \
1442         Q_COPY(out, in, dqb_ihardlimit);        \
1443         Q_COPY(out, in, dqb_isoftlimit);        \
1444         Q_COPY(out, in, dqb_curinodes);         \
1445         Q_COPY(out, in, dqb_btime);             \
1446         Q_COPY(out, in, dqb_itime);             \
1447         Q_COPY(out, in, dqb_valid);             \
1448 } while (0)
1449
1450 static int fsfilt_ext3_quotactl(struct super_block *sb,
1451                                 struct obd_quotactl *oqc)
1452 {
1453         int i, rc = 0, error = 0;
1454         struct quotactl_ops *qcop;
1455         struct if_dqinfo *info;
1456         struct if_dqblk *dqblk;
1457         ENTRY;
1458
1459         if (!sb->s_qcop)
1460                 RETURN(-ENOSYS);
1461
1462         OBD_ALLOC_PTR(info);
1463         if (!info)
1464                 RETURN(-ENOMEM);
1465         OBD_ALLOC_PTR(dqblk);
1466         if (!dqblk) {
1467                 OBD_FREE_PTR(info);
1468                 RETURN(-ENOMEM);
1469         }
1470
1471         DQINFO_COPY(info, &oqc->qc_dqinfo);
1472         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1473
1474         qcop = sb->s_qcop;
1475         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1476                 for (i = 0; i < MAXQUOTAS; i++) {
1477                         if (!Q_TYPESET(oqc, i))
1478                                 continue;
1479
1480                         if (oqc->qc_cmd == Q_QUOTAON) {
1481                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1482
1483                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1484
1485                                 rc = ll_quota_on(sb, i, QFMT_VFS_V0,
1486                                                  name[i], 0);
1487                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1488                                 rc = ll_quota_off(sb, i, 0);
1489                         }
1490
1491                         if (rc == -EBUSY)
1492                                 error = rc;
1493                         else if (rc)
1494                                 GOTO(out, rc);
1495                 }
1496                 GOTO(out, rc ?: error);
1497         }
1498
1499         switch (oqc->qc_cmd) {
1500         case Q_GETOINFO:
1501         case Q_GETINFO:
1502                 if (!qcop->get_info)
1503                         GOTO(out, rc = -ENOSYS);
1504                 rc = qcop->get_info(sb, oqc->qc_type, info);
1505                 break;
1506         case Q_SETQUOTA:
1507         case Q_INITQUOTA:
1508                 if (!qcop->set_dqblk)
1509                         GOTO(out, rc = -ENOSYS);
1510                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1511                 break;
1512         case Q_GETOQUOTA:
1513         case Q_GETQUOTA:
1514                 if (!qcop->get_dqblk)
1515                         GOTO(out, rc = -ENOSYS);
1516                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1517                 if (!rc)
1518                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1519                 break;
1520         case Q_SYNC:
1521                 if (!sb->s_qcop->quota_sync)
1522                         GOTO(out, rc = -ENOSYS);
1523                 qcop->quota_sync(sb, oqc->qc_type);
1524                 break;
1525         case Q_FINVALIDATE:
1526                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1527                 for (i = 0; i < MAXQUOTAS; i++) {
1528                         struct file *fp;
1529                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1530
1531                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1532
1533                         if (!Q_TYPESET(oqc, i))
1534                                 continue;
1535
1536                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1537                         if (IS_ERR(fp)) {
1538                                 rc = PTR_ERR(fp);
1539                                 CERROR("error invalidating operational quota file"
1540                                        " %s (rc:%d)\n", name[i], rc);
1541                         } else {
1542                                 filp_close(fp, 0);
1543                         }
1544
1545                 }
1546                 break;
1547         default:
1548                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1549                 LBUG();
1550         }
1551 out:
1552         DQINFO_COPY(&oqc->qc_dqinfo, info);
1553         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1554
1555         OBD_FREE_PTR(info);
1556         OBD_FREE_PTR(dqblk);
1557
1558         if (rc)
1559                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1560                                 "failed: %d\n",
1561                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1562         RETURN(rc);
1563 }
1564
1565 struct chk_dqblk{
1566         struct hlist_node       dqb_hash;        /** quotacheck hash */
1567         struct list_head        dqb_list;        /** in list also */
1568         qid_t                   dqb_id;          /** uid/gid */
1569         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1570         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1571         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1572         qsize_t                 dqb_curspace;    /** current space */
1573         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1574         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1575         qsize_t                 dqb_curinodes;   /** current inodes */
1576         __u64                   dqb_btime;       /** block grace time */
1577         __u64                   dqb_itime;       /** inode grace time */
1578         __u32                   dqb_valid;       /** flag for above fields */
1579 };
1580
1581 static inline unsigned int chkquot_hash(qid_t id, int type)
1582                                         __attribute__((__const__));
1583
1584 static inline unsigned int chkquot_hash(qid_t id, int type)
1585 {
1586         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1587 }
1588
1589 static inline struct chk_dqblk *
1590 find_chkquot(struct hlist_head *head, qid_t id, int type)
1591 {
1592         struct hlist_node *node;
1593         struct chk_dqblk *cdqb;
1594
1595         hlist_for_each(node, head) {
1596                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1597                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1598                         return cdqb;
1599         }
1600
1601         return NULL;
1602 }
1603
1604 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1605 {
1606         struct chk_dqblk *cdqb;
1607
1608         OBD_ALLOC_PTR(cdqb);
1609         if (cdqb) {
1610                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1611                 INIT_LIST_HEAD(&cdqb->dqb_list);
1612                 cdqb->dqb_id = id;
1613                 cdqb->dqb_type = type;
1614         }
1615
1616         return cdqb;
1617 }
1618
1619 static struct chk_dqblk *
1620 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1621       qid_t id, int type, int first_check)
1622 {
1623         struct hlist_head *head = hash + chkquot_hash(id, type);
1624         struct if_dqblk dqb;
1625         struct chk_dqblk *cdqb;
1626         int rc;
1627
1628         cdqb = find_chkquot(head, id, type);
1629         if (cdqb)
1630                 return cdqb;
1631
1632         cdqb = alloc_chkquot(id, type);
1633         if (!cdqb)
1634                 return NULL;
1635
1636         if (!first_check) {
1637                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1638                 if (rc) {
1639                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1640                                id, type, rc);
1641                 } else {
1642                         DQBLK_COPY(cdqb, &dqb);
1643                         cdqb->dqb_curspace = 0;
1644                         cdqb->dqb_curinodes = 0;
1645                 }
1646         }
1647
1648         hlist_add_head(&cdqb->dqb_hash, head);
1649         list_add_tail(&cdqb->dqb_list, list);
1650
1651         return cdqb;
1652 }
1653
1654 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1655 {
1656         struct obd_quotactl *oqctl;
1657         int rc;
1658
1659         OBD_ALLOC_PTR(oqctl);
1660         if (!oqctl)
1661                 RETURN(-ENOMEM);
1662
1663         oqctl->qc_cmd = cmd;
1664         oqctl->qc_id = qfmt;
1665         oqctl->qc_type = type;
1666         rc = fsfilt_ext3_quotactl(sb, oqctl);
1667
1668         OBD_FREE_PTR(oqctl);
1669         return rc;
1670 }
1671
1672 static inline int read_old_dqinfo(struct super_block *sb, int type,
1673                                   struct if_dqinfo *dqinfo)
1674 {
1675         struct obd_quotactl *oqctl;
1676         int rc;
1677         ENTRY;
1678
1679         OBD_ALLOC_PTR(oqctl);
1680         if (!oqctl)
1681                 RETURN(-ENOMEM);
1682
1683         oqctl->qc_cmd = Q_GETINFO;
1684         oqctl->qc_type = type;
1685         rc = fsfilt_ext3_quotactl(sb, oqctl);
1686         if (!rc)
1687                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1688
1689         OBD_FREE_PTR(oqctl);
1690         RETURN(rc);
1691 }
1692
1693 static inline struct ext3_group_desc *
1694 get_group_desc(struct super_block *sb, int group)
1695 {
1696         unsigned long desc_block, desc;
1697         struct ext3_group_desc *gdp;
1698
1699         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1700         desc = group % EXT3_DESC_PER_BLOCK(sb);
1701         gdp = (struct ext3_group_desc *)
1702               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1703
1704         return gdp + desc;
1705 }
1706
1707
1708 #ifndef HAVE_EXT4_LDISKFS
1709 static inline struct buffer_head *
1710 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1711 {
1712         struct ext3_group_desc *desc;
1713         struct buffer_head *bh;
1714
1715         desc = get_group_desc(sb, group);
1716         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1717         return bh;
1718 }
1719 #endif
1720
1721 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1722                                      struct buffer_head *bitmap_bh,
1723                                      int index, unsigned long ino)
1724 {
1725         struct inode *inode = NULL;
1726
1727
1728         if (ext3_test_bit(index, bitmap_bh->b_data))
1729 #ifdef HAVE_EXT4_LDISKFS
1730                 inode = ext4_iget(sb, ino);
1731                 if (IS_ERR(inode))
1732                 /* Newer kernels return an error instead of a NULL pointer */
1733                         inode = NULL;
1734 #else
1735                 inode = iget(sb, ino);
1736 #endif
1737         return inode;
1738 }
1739
1740 struct qchk_ctxt {
1741         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1742         struct list_head        qckt_list;                   /* quotacheck list */
1743         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1744         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1745 };
1746
1747 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1748                            struct obd_quotactl *oqc)
1749 {
1750         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1751         loff_t size = 0;
1752         qid_t qid[MAXQUOTAS];
1753         int cnt, i, rc = 0;
1754
1755         if (!inode)
1756                 return 0;
1757
1758         qid[USRQUOTA] = inode->i_uid;
1759         qid[GRPQUOTA] = inode->i_gid;
1760
1761         if (S_ISDIR(inode->i_mode) ||
1762             S_ISREG(inode->i_mode) ||
1763             S_ISLNK(inode->i_mode))
1764                 size = inode_get_bytes(inode);
1765
1766         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1767                 if (!Q_TYPESET(oqc, cnt))
1768                         continue;
1769
1770                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1771                                 &qctxt->qckt_list, qid[cnt], cnt,
1772                                 qctxt->qckt_first_check[cnt]);
1773                 if (!cdqb[cnt]) {
1774                         rc = -ENOMEM;
1775                         break;
1776                 }
1777
1778                 cdqb[cnt]->dqb_curspace += size;
1779                 cdqb[cnt]->dqb_curinodes++;
1780         }
1781
1782         if (rc) {
1783                 for (i = 0; i < cnt; i++) {
1784                         if (!Q_TYPESET(oqc, i))
1785                                 continue;
1786                         LASSERT(cdqb[i]);
1787                         cdqb[i]->dqb_curspace -= size;
1788                         cdqb[i]->dqb_curinodes--;
1789                 }
1790         }
1791
1792         return rc;
1793 }
1794
1795 /* write dqinfo struct in a new quota file */
1796 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1797 {
1798         struct v2_disk_dqinfo dqinfo;
1799         __u32 blocks = V2_DQTREEOFF + 1;
1800         loff_t offset = V2_DQINFOOFF;
1801
1802         if (info) {
1803                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1804                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1805                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1806                                                ~DQF_INFO_DIRTY);
1807         } else {
1808                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1809                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1810                 dqinfo.dqi_flags = 0;
1811         }
1812
1813         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1814         dqinfo.dqi_free_blk = 0;
1815         dqinfo.dqi_free_entry = 0;
1816
1817         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1818 }
1819
1820 static int v3_write_dqheader(struct file *f, int type)
1821 {
1822         static const __u32 quota_magics[] = V2_INITQMAGICS;
1823         static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1824         struct v2_disk_dqheader dqhead;
1825         loff_t offset = 0;
1826
1827         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1828         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1829
1830         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1831         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1832
1833         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1834 }
1835
1836 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1837                                   struct obd_quotactl *oqc)
1838 {
1839         int i, rc = 0;
1840         ENTRY;
1841
1842         for (i = 0; i < MAXQUOTAS; i++) {
1843                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1844                                          NULL : &qctxt->qckt_dqinfo[i];
1845                 struct file *file;
1846                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1847
1848                 if (!Q_TYPESET(oqc, i))
1849                         continue;
1850
1851                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1852
1853                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1854                 if (IS_ERR(file)) {
1855                         rc = PTR_ERR(file);
1856                         CERROR("can't create %s file: rc = %d\n",
1857                                name[i], rc);
1858                         GOTO(out, rc);
1859                 }
1860
1861                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1862                         CERROR("file %s is not regular", name[i]);
1863                         filp_close(file, 0);
1864                         GOTO(out, rc = -EINVAL);
1865                 }
1866
1867                 DQUOT_DROP(file->f_dentry->d_inode);
1868
1869                 rc = v3_write_dqheader(file, i);
1870                 if (rc) {
1871                         filp_close(file, 0);
1872                         GOTO(out, rc);
1873                 }
1874
1875                 rc = v3_write_dqinfo(file, i, info);
1876                 filp_close(file, 0);
1877                 if (rc)
1878                         GOTO(out, rc);
1879         }
1880
1881 out:
1882         RETURN(rc);
1883 }
1884
1885
1886 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1887                           struct chk_dqblk *cdqb)
1888 {
1889         struct obd_quotactl *oqc;
1890         long now;
1891         int rc;
1892         ENTRY;
1893
1894         OBD_ALLOC_PTR(oqc);
1895         if (!oqc)
1896                 RETURN(-ENOMEM);
1897
1898         now = cfs_time_current_sec();
1899
1900         if (cdqb->dqb_bsoftlimit &&
1901             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1902             !cdqb->dqb_btime)
1903                 cdqb->dqb_btime =
1904                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1905
1906         if (cdqb->dqb_isoftlimit &&
1907             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1908             !cdqb->dqb_itime)
1909                 cdqb->dqb_itime =
1910                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1911
1912         cdqb->dqb_valid = QIF_ALL;
1913
1914         oqc->qc_cmd = Q_SETQUOTA;
1915         oqc->qc_type = cdqb->dqb_type;
1916         oqc->qc_id = cdqb->dqb_id;
1917         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1918
1919         rc = fsfilt_ext3_quotactl(sb, oqc);
1920         OBD_FREE_PTR(oqc);
1921         RETURN(rc);
1922 }
1923
1924 static int prune_chkquots(struct super_block *sb,
1925                           struct qchk_ctxt *qctxt, int error)
1926 {
1927         struct chk_dqblk *cdqb, *tmp;
1928         int rc;
1929
1930         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1931                 if (!error) {
1932                         rc = commit_chkquot(sb, qctxt, cdqb);
1933                         if (rc)
1934                                 error = rc;
1935                 }
1936                 hlist_del_init(&cdqb->dqb_hash);
1937                 list_del(&cdqb->dqb_list);
1938                 OBD_FREE_PTR(cdqb);
1939         }
1940
1941         return error;
1942 }
1943
1944 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1945                                   struct obd_quotactl *oqc)
1946 {
1947         struct ext3_sb_info *sbi = EXT3_SB(sb);
1948         int i, group;
1949         struct qchk_ctxt *qctxt;
1950         struct buffer_head *bitmap_bh = NULL;
1951         unsigned long ino;
1952         struct inode *inode;
1953         int rc = 0;
1954         ENTRY;
1955
1956         /* turn on quota and read dqinfo if existed */
1957         OBD_ALLOC_PTR(qctxt);
1958         if (!qctxt) {
1959                 oqc->qc_stat = -ENOMEM;
1960                 RETURN(-ENOMEM);
1961         }
1962
1963         for (i = 0; i < NR_DQHASH; i++)
1964                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1965         INIT_LIST_HEAD(&qctxt->qckt_list);
1966
1967         for (i = 0; i < MAXQUOTAS; i++) {
1968                 if (!Q_TYPESET(oqc, i))
1969                         continue;
1970
1971                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1972                 if (!rc || rc == -EBUSY) {
1973                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1974                         if (rc)
1975                                 GOTO(out, rc);
1976                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1977                         qctxt->qckt_first_check[i] = 1;
1978                 } else if (rc) {
1979                         GOTO(out, rc);
1980                 }
1981         }
1982
1983         /* check quota and update in hash */
1984         for (group = 0; group < sbi->s_groups_count; group++) {
1985                 struct ext3_group_desc *desc;
1986                 desc = get_group_desc(sb, group);
1987                 if (!desc)
1988                         GOTO(out, -EIO);
1989
1990                 spin_lock(sb_bgl_lock(sbi, group));
1991                 if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
1992                         /* no inode in use in this group, just skip it */
1993                         spin_unlock(sb_bgl_lock(sbi, group));
1994                         continue;
1995                 }
1996                 spin_unlock(sb_bgl_lock(sbi, group));
1997
1998                 ino = group * sbi->s_inodes_per_group + 1;
1999                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2000                 if (!bitmap_bh) {
2001                         CERROR("ext3_read_inode_bitmap group %d failed", group);
2002                         GOTO(out, rc = -EIO);
2003                 }
2004
2005                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
2006                         if (ino < sbi->s_first_ino)
2007                                 continue;
2008
2009                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
2010                         rc = add_inode_quota(inode, qctxt, oqc);
2011                         iput(inode);
2012                         if (rc) {
2013                                 brelse(bitmap_bh);
2014                                 GOTO(out, rc);
2015                         }
2016                 }
2017
2018                 brelse(bitmap_bh);
2019         }
2020
2021         /* read old quota limits from old quota file. (only for the user
2022          * has limits but hasn't file) */
2023 #ifdef HAVE_QUOTA_SUPPORT
2024         for (i = 0; i < MAXQUOTAS; i++) {
2025                 struct list_head id_list;
2026                 struct dquot_id *dqid, *tmp;
2027
2028                 if (!Q_TYPESET(oqc, i))
2029                         continue;
2030
2031                 if (qctxt->qckt_first_check[i])
2032                         continue;
2033
2034
2035                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2036                 INIT_LIST_HEAD(&id_list);
2037 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2038                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2039 #else
2040                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2041 #endif
2042                 if (rc)
2043                         CERROR("read old limits failed. (rc:%d)\n", rc);
2044
2045                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2046                         list_del_init(&dqid->di_link);
2047
2048                         if (!rc)
2049                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2050                                       dqid->di_id, i,
2051                                       qctxt->qckt_first_check[i]);
2052                         OBD_FREE_PTR(dqid);
2053                 }
2054         }
2055 #endif
2056         /* turn off quota cause we are to dump chk_dqblk to files */
2057         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2058
2059         rc = create_new_quota_files(qctxt, oqc);
2060         if (rc)
2061                 GOTO(out, rc);
2062
2063         /* we use vfs functions to set dqblk, so turn quota on */
2064         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2065 out:
2066         /* dump and free chk_dqblk */
2067         rc = prune_chkquots(sb, qctxt, rc);
2068         OBD_FREE_PTR(qctxt);
2069
2070         /* turn off quota, `lfs quotacheck` will turn on when all
2071          * nodes quotacheck finish. */
2072         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2073
2074         oqc->qc_stat = rc;
2075         if (rc)
2076                 CERROR("quotacheck failed: rc = %d\n", rc);
2077
2078         RETURN(rc);
2079 }
2080
2081 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2082                                  int cmd)
2083 {
2084         int rc = 0;
2085         ENTRY;
2086
2087         if (lqi->qi_files[type] == NULL) {
2088                 CERROR("operate qinfo before it's enabled!\n");
2089                 RETURN(-ESRCH);
2090         }
2091
2092         switch (cmd) {
2093         case QFILE_CHK:
2094                 rc = lustre_check_quota_file(lqi, type);
2095                 break;
2096         case QFILE_RD_INFO:
2097                 rc = lustre_read_quota_info(lqi, type);
2098                 break;
2099         case QFILE_WR_INFO:
2100                 rc = lustre_write_quota_info(lqi, type);
2101                 break;
2102         case QFILE_INIT_INFO:
2103                 rc = lustre_init_quota_info(lqi, type);
2104                 break;
2105         case QFILE_CONVERT:
2106                 rc = -ENOTSUPP;
2107                 CERROR("quota CONVERT command is not supported\n");
2108                 break;
2109         default:
2110                 rc = -ENOTSUPP;
2111                 CERROR("Unsupported admin quota file cmd %d\n"
2112                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2113                        cmd);
2114                 break;
2115         }
2116         RETURN(rc);
2117 }
2118
2119 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2120                             struct list_head *list)
2121 {
2122         return lustre_get_qids(file, inode, type, list);
2123 }
2124
2125 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2126 {
2127         int rc = 0;
2128         ENTRY;
2129
2130         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2131                 CERROR("operate dquot before it's enabled!\n");
2132                 RETURN(-ESRCH);
2133         }
2134
2135         switch (cmd) {
2136         case QFILE_RD_DQUOT:
2137                 rc = lustre_read_dquot(dquot);
2138                 break;
2139         case QFILE_WR_DQUOT:
2140                 if (dquot->dq_dqb.dqb_ihardlimit ||
2141                     dquot->dq_dqb.dqb_isoftlimit ||
2142                     dquot->dq_dqb.dqb_bhardlimit ||
2143                     dquot->dq_dqb.dqb_bsoftlimit)
2144                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2145                 else
2146                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2147
2148                 rc = lustre_commit_dquot(dquot);
2149                 if (rc >= 0)
2150                         rc = 0;
2151                 break;
2152         default:
2153                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2154                 LBUG();
2155                 break;
2156         }
2157         RETURN(rc);
2158 }
2159
2160 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2161                                 struct inode *inode, int frags)
2162 {
2163 #ifdef EXT3_EXT_HAS_NO_TREE
2164         struct ext3_ext_base *base = inode;
2165 #else
2166         struct ext3_extents_tree tree;
2167         struct ext3_ext_base *base = &tree;
2168
2169         ext3_init_tree_desc(base, inode);
2170 #endif
2171         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2172          * metablocks at maxium b=16542 */
2173         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2174         return 0;
2175 }
2176
2177 #endif
2178
2179 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2180 {
2181         return (EXT3_SB(sb)->journal_bdev);
2182 }
2183 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2184
2185 static struct fsfilt_operations fsfilt_ext3_ops = {
2186         .fs_type                = "ext3",
2187         .fs_owner               = THIS_MODULE,
2188         .fs_getlabel            = fsfilt_ext3_get_label,
2189         .fs_setlabel            = fsfilt_ext3_set_label,
2190         .fs_uuid                = fsfilt_ext3_uuid,
2191         .fs_start               = fsfilt_ext3_start,
2192         .fs_brw_start           = fsfilt_ext3_brw_start,
2193         .fs_extend              = fsfilt_ext3_extend,
2194         .fs_commit              = fsfilt_ext3_commit,
2195         .fs_commit_async        = fsfilt_ext3_commit_async,
2196         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2197         .fs_setattr             = fsfilt_ext3_setattr,
2198         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2199         .fs_set_md              = fsfilt_ext3_set_md,
2200         .fs_get_md              = fsfilt_ext3_get_md,
2201         .fs_readpage            = fsfilt_ext3_readpage,
2202         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2203         .fs_statfs              = fsfilt_ext3_statfs,
2204         .fs_sync                = fsfilt_ext3_sync,
2205         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2206         .fs_write_record        = fsfilt_ext3_write_record,
2207         .fs_read_record         = fsfilt_ext3_read_record,
2208         .fs_setup               = fsfilt_ext3_setup,
2209         .fs_send_bio            = fsfilt_ext3_send_bio,
2210         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2211 #ifdef HAVE_DISK_INODE_VERSION
2212         .fs_get_version         = fsfilt_ext3_get_version,
2213         .fs_set_version         = fsfilt_ext3_set_version,
2214 #endif
2215 #ifdef HAVE_QUOTA_SUPPORT
2216         .fs_quotactl            = fsfilt_ext3_quotactl,
2217         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2218         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2219         .fs_qids                = fsfilt_ext3_qids,
2220         .fs_dquot               = fsfilt_ext3_dquot,
2221         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2222 #endif
2223         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2224 };
2225
2226 static int __init fsfilt_ext3_init(void)
2227 {
2228         int rc;
2229
2230         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2231                                          sizeof(struct fsfilt_cb_data), 0, 0);
2232         if (!fcb_cache) {
2233                 CERROR("error allocating fsfilt journal callback cache\n");
2234                 GOTO(out, rc = -ENOMEM);
2235         }
2236
2237         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2238
2239         if (rc) {
2240                 int err = cfs_mem_cache_destroy(fcb_cache);
2241                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2242         }
2243 out:
2244         return rc;
2245 }
2246
2247 static void __exit fsfilt_ext3_exit(void)
2248 {
2249         int rc;
2250
2251         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2252         rc = cfs_mem_cache_destroy(fcb_cache);
2253         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2254 }
2255
2256 module_init(fsfilt_ext3_init);
2257 module_exit(fsfilt_ext3_exit);
2258
2259 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2260 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2261 MODULE_LICENSE("GPL");