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 #include <linux/quotaio_v1.h>
  61 #include <linux/quotaio_v2.h>
  62 #if defined(HAVE_EXT3_XATTR_H)
  63 #include <ext3/xattr.h>
  64 #else
  65 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  66 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  67 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  68 #endif
  69
  70 #include <libcfs/libcfs.h>
  71 #include <lustre_fsfilt.h>
  72 #include <obd.h>
  73 #include <lustre_quota.h>
  74 #include <linux/lustre_compat25.h>
  75 #include <linux/lprocfs_status.h>
  76
  77 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
  78 #ifdef HAVE_EXT4_LDISKFS
  79 #include <ext4/ext4_extents.h>
  80 #else
  81 #include <linux/ext3_extents.h>
  82 #endif
  83 #endif
  84
  85 #include "lustre_quota_fmt.h"
  86
  87 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
  88 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
  89 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
  90 #else
  91 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
  92 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
  93 #endif
  94
  95 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
  96 /* for kernels 2.6.18 and later */
  97 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  98 #else
  99 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
 100 #endif
 101
 102 #define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks)
 103 #define fsfilt_ext3_journal_stop(handle)          ext3_journal_stop(handle)
 104
 105 static cfs_mem_cache_t *fcb_cache;
 106
 107 struct fsfilt_cb_data {
 108         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 109         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 110         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 111         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 112         void *cb_data;                  /* MDS/OST completion function data */
 113 };
 114
 115 #ifndef EXT3_XATTR_INDEX_TRUSTED        /* temporary until we hit l28 kernel */
 116 #define EXT3_XATTR_INDEX_TRUSTED        4
 117 #endif
 118
 119 #ifdef HAVE_EXT4_LDISKFS
 120 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 121 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 122 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
 123 #else
 124 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
 125 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
 126 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
 127 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
 128 #define ext3_ext_store_pblock(ex, pblock)  ((ex)->ee_start = cpu_to_le32(pblock))
 129 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
 130 #endif
 131
 132 static char *fsfilt_ext3_get_label(struct super_block *sb)
 133 {
 134         return EXT3_SB(sb)->s_es->s_volume_name;
 135 }
 136
 137 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 138 {
 139         /* see e.g. fsfilt_ext3_write_record() */
 140         journal_t *journal;
 141         handle_t *handle;
 142         int err;
 143
 144         journal = EXT3_SB(sb)->s_journal;
 145         handle = ext3_journal_start_sb(sb, 1);
 146         if (IS_ERR(handle)) {
 147                 CERROR("can't start transaction\n");
 148                 return(PTR_ERR(handle));
 149         }
 150
 151         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 152         if (err)
 153                 goto out;
 154
 155         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 156                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 157
 158         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 159
 160 out:
 161         ext3_journal_stop(handle);
 162
 163         return(err);
 164 }
 165
 166 static char *fsfilt_ext3_uuid(struct super_block *sb)
 167 {
 168         return EXT3_SB(sb)->s_es->s_uuid;
 169 }
 170
 171 #ifdef HAVE_DISK_INODE_VERSION
 172
 173 static __u64 get_i_version(struct inode *inode)
 174 {
 175 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 176         return inode->i_version;
 177 #else
 178         return EXT3_I(inode)->i_fs_version;
 179 #endif
 180 }
 181
 182 static void set_i_version(struct inode *inode, __u64 new_version)
 183 {
 184 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
 185         inode->i_version = new_version;
 186 #else
 187         (EXT3_I(inode))->i_fs_version = new_version;
 188 #endif
 189 }
 190
 191 /*
 192  * Get the 64-bit version for an inode.
 193  */
 194 static __u64 fsfilt_ext3_get_version(struct inode *inode)
 195 {
 196         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
 197                get_i_version(inode), inode->i_ino);
 198         return get_i_version(inode);
 199 }
 200
 201 /*
 202  * Set the 64-bit version and return the old version.
 203  */
 204 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
 205 {
 206         __u64 old_version = get_i_version(inode);
 207
 208         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
 209                new_version, old_version, inode->i_ino);
 210         set_i_version(inode, new_version);
 211         /* version is set after all inode operations are finished, so we should
 212          * mark it dirty here */
 213         inode->i_sb->s_op->dirty_inode(inode);
 214         return old_version;
 215 }
 216
 217 #endif
 218
 219 /*
 220  * We don't currently need any additional blocks for rmdir and
 221  * unlink transactions because we are storing the OST oa_id inside
 222  * the inode (which we will be changing anyways as part of this
 223  * transaction).
 224  */
 225 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
 226                                int logs)
 227 {
 228         /* For updates to the last received file */
 229         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 230         journal_t *journal;
 231         void *handle;
 232
 233         if (current->journal_info) {
 234                 CDEBUG(D_INODE, "increasing refcount on %p\n",
 235                        current->journal_info);
 236                 goto journal_start;
 237         }
 238
 239         switch(op) {
 240         case FSFILT_OP_RMDIR:
 241         case FSFILT_OP_UNLINK:
 242                 /* delete one file + create/update logs for each stripe */
 243                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 244                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 245                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 246                 break;
 247         case FSFILT_OP_RENAME:
 248                 /* modify additional directory */
 249                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
 250                 /* no break */
 251         case FSFILT_OP_SYMLINK:
 252                 /* additional block + block bitmap + GDT for long symlink */
 253                 nblocks += 3;
 254                 /* no break */
 255         case FSFILT_OP_CREATE: {
 256 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
 257                 static int warned;
 258                 if (!warned) {
 259                         if (!test_opt(inode->i_sb, EXTENTS)) {
 260                                 warned = 1;
 261                         } else if (((EXT3_I(inode)->i_flags &
 262                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
 263                               cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
 264                                 CWARN("extent-mapped directory found with "
 265                                       "ext3-based ldiskfs - contact "
 266                                       "http://bugzilla.lustre.org/\n");
 267                                 warned = 1;
 268                         }
 269                 }
 270 #endif
 271                 /* no break */
 272         }
 273         case FSFILT_OP_MKDIR:
 274         case FSFILT_OP_MKNOD:
 275                 /* modify one inode + block bitmap + GDT */
 276                 nblocks += 3;
 277                 /* no break */
 278         case FSFILT_OP_LINK:
 279                 /* modify parent directory */
 280                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 281                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 282                 /* create/update logs for each stripe */
 283                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 284                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 285                 break;
 286         case FSFILT_OP_SETATTR:
 287                 /* Setattr on inode */
 288                 nblocks += 1;
 289                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 290                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 291                 /* quota chown log for each stripe */
 292                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 293                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 294                 break;
 295         case FSFILT_OP_CANCEL_UNLINK:
 296                 /* blocks for log header bitmap update OR
 297                  * blocks for catalog header bitmap update + unlink of logs */
 298                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 299                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 300                 break;
 301         case FSFILT_OP_JOIN:
 302                 /* delete 2 file(file + array id) + create 1 file (array id)
 303                  * create/update logs for each stripe */
 304                 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
 305
 306                 /*create array log for head file*/
 307                 nblocks += 3;
 308                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 309                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
 310                 /*update head file array */
 311                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 312                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
 313                 break;
 314         default: CERROR("unknown transaction start op %d\n", op);
 315                 LBUG();
 316         }
 317
 318         LASSERT(current->journal_info == desc_private);
 319         journal = EXT3_SB(inode->i_sb)->s_journal;
 320         if (nblocks > journal->j_max_transaction_buffers) {
 321                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 322                        nblocks, op, logs, journal->j_max_transaction_buffers);
 323                 nblocks = journal->j_max_transaction_buffers;
 324         }
 325
 326  journal_start:
 327         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 328         handle = fsfilt_ext3_journal_start(inode, nblocks);
 329
 330         if (!IS_ERR(handle))
 331                 LASSERT(current->journal_info == handle);
 332         else
 333                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 334                        op, nblocks, PTR_ERR(handle));
 335         return handle;
 336 }
 337
 338 /*
 339  * Calculate the number of buffer credits needed to write multiple pages in
 340  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 341  * doesn't have a nice API for calculating this sort of thing in advance.
 342  *
 343  * See comment above ext3_writepage_trans_blocks for details.  We assume
 344  * no data journaling is being done, but it does allow for all of the pages
 345  * being non-contiguous.  If we are guaranteed contiguous pages we could
 346  * reduce the number of (d)indirect blocks a lot.
 347  *
 348  * With N blocks per page and P pages, for each inode we have at most:
 349  * N*P indirect
 350  * min(N*P, blocksize/4 + 1) dindirect blocks
 351  * niocount tindirect
 352  *
 353  * For the entire filesystem, we have at most:
 354  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 355  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 356  * objcount inode blocks
 357  * 1 superblock
 358  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 359  *
 360  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 361  */
 362 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 363                                       int niocount, struct niobuf_local *nb)
 364 {
 365         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 366         __u64 next_indir;
 367         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 368         int nbitmaps = 0, ngdblocks;
 369         int needed = objcount + 1; /* inodes + superblock */
 370         int i, j;
 371
 372         for (i = 0, j = 0; i < objcount; i++, fso++) {
 373                 /* two or more dindirect blocks in case we cross boundary */
 374                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 375                                     nb[j].offset) >>
 376                                    sb->s_blocksize_bits) /
 377                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 378                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 379
 380                 /* leaf, indirect, tindirect blocks for first block */
 381                 nbitmaps += blockpp + 2;
 382
 383                 j += fso->fso_bufcnt;
 384         }
 385
 386         next_indir = nb[0].offset +
 387                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 388         for (i = 1; i < niocount; i++) {
 389                 if (nb[i].offset >= next_indir) {
 390                         nbitmaps++;     /* additional indirect */
 391                         next_indir = nb[i].offset +
 392                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 393                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 394                         nbitmaps++;     /* additional indirect */
 395                 }
 396                 nbitmaps += blockpp;    /* each leaf in different group? */
 397         }
 398
 399         ngdblocks = nbitmaps;
 400         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 401                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 402         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 403                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 404
 405         needed += nbitmaps + ngdblocks;
 406
 407         /* last_rcvd update */
 408         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
 409
 410 #if defined(CONFIG_QUOTA)
 411         /* We assume that there will be 1 bit set in s_dquot.flags for each
 412          * quota file that is active.  This is at least true for now.
 413          */
 414         needed += hweight32(sb_any_quota_enabled(sb)) *
 415                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 416 #endif
 417
 418         return needed;
 419 }
 420
 421 /* We have to start a huge journal transaction here to hold all of the
 422  * metadata for the pages being written here.  This is necessitated by
 423  * the fact that we do lots of prepare_write operations before we do
 424  * any of the matching commit_write operations, so even if we split
 425  * up to use "smaller" transactions none of them could complete until
 426  * all of them were opened.  By having a single journal transaction,
 427  * we eliminate duplicate reservations for common blocks like the
 428  * superblock and group descriptors or bitmaps.
 429  *
 430  * We will start the transaction here, but each prepare_write will
 431  * add a refcount to the transaction, and each commit_write will
 432  * remove a refcount.  The transaction will be closed when all of
 433  * the pages have been written.
 434  */
 435 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 436                                    int niocount, struct niobuf_local *nb,
 437                                    void *desc_private, int logs)
 438 {
 439         journal_t *journal;
 440         handle_t *handle;
 441         int needed;
 442         ENTRY;
 443
 444         LASSERT(current->journal_info == desc_private);
 445         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 446         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 447
 448         /* The number of blocks we could _possibly_ dirty can very large.
 449          * We reduce our request if it is absurd (and we couldn't get that
 450          * many credits for a single handle anyways).
 451          *
 452          * At some point we have to limit the size of I/Os sent at one time,
 453          * increase the size of the journal, or we have to calculate the
 454          * actual journal requirements more carefully by checking all of
 455          * the blocks instead of being maximally pessimistic.  It remains to
 456          * be seen if this is a real problem or not.
 457          */
 458         if (needed > journal->j_max_transaction_buffers) {
 459                 CERROR("want too many journal credits (%d) using %d instead\n",
 460                        needed, journal->j_max_transaction_buffers);
 461                 needed = journal->j_max_transaction_buffers;
 462         }
 463
 464         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 465         handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
 466         if (IS_ERR(handle)) {
 467                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 468                        PTR_ERR(handle));
 469         } else {
 470                 LASSERT(handle->h_buffer_credits >= needed);
 471                 LASSERT(current->journal_info == handle);
 472         }
 473
 474         RETURN(handle);
 475 }
 476
 477 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 478 {
 479        handle_t *handle = h;
 480
 481        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 482        if (nblocks == 0) {
 483                handle->h_buffer_credits = 0;
 484                CWARN("setting credits of handle %p to zero by request\n", h);
 485        }
 486
 487        if (handle->h_buffer_credits > nblocks)
 488                 return 0;
 489        if (ext3_journal_extend(handle, nblocks) == 0)
 490                 return 0;
 491
 492        ext3_mark_inode_dirty(handle, inode);
 493        return ext3_journal_restart(handle, nblocks);
 494 }
 495
 496 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 497 {
 498         int rc;
 499         handle_t *handle = h;
 500
 501         LASSERT(current->journal_info == handle);
 502         if (force_sync)
 503                 handle->h_sync = 1; /* recovery likes this */
 504
 505         rc = fsfilt_ext3_journal_stop(handle);
 506
 507         return rc;
 508 }
 509
 510 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 511                                     void **wait_handle)
 512 {
 513         unsigned long tid;
 514         transaction_t *transaction;
 515         handle_t *handle = h;
 516         journal_t *journal;
 517         int rc;
 518
 519         LASSERT(current->journal_info == handle);
 520
 521         transaction = handle->h_transaction;
 522         journal = transaction->t_journal;
 523         tid = transaction->t_tid;
 524         /* we don't want to be blocked */
 525         handle->h_sync = 0;
 526         rc = fsfilt_ext3_journal_stop(handle);
 527         if (rc) {
 528                 CERROR("error while stopping transaction: %d\n", rc);
 529                 return rc;
 530         }
 531         fsfilt_log_start_commit(journal, tid);
 532
 533         *wait_handle = (void *) tid;
 534         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 535         return 0;
 536 }
 537
 538 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 539 {
 540         journal_t *journal = EXT3_JOURNAL(inode);
 541         tid_t tid = (tid_t)(long)h;
 542
 543         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 544         if (unlikely(is_journal_aborted(journal)))
 545                 return -EIO;
 546
 547         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 548
 549         if (unlikely(is_journal_aborted(journal)))
 550                 return -EIO;
 551         return 0;
 552 }
 553
 554 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 555                                struct iattr *iattr, int do_trunc)
 556 {
 557         struct inode *inode = dentry->d_inode;
 558         int rc = 0;
 559
 560         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 561          * We are already writing the inode to disk as part of this
 562          * transaction and want to avoid a lot of extra inode writeout
 563          * later on. b=9828 */
 564         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 565                 /* ATTR_SIZE would invoke truncate: clear it */
 566                 iattr->ia_valid &= ~ATTR_SIZE;
 567                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 568                 i_size_write(inode, iattr->ia_size);
 569
 570                 if (iattr->ia_valid & ATTR_UID)
 571                         inode->i_uid = iattr->ia_uid;
 572                 if (iattr->ia_valid & ATTR_GID)
 573                         inode->i_gid = iattr->ia_gid;
 574                 if (iattr->ia_valid & ATTR_ATIME)
 575                         inode->i_atime = iattr->ia_atime;
 576                 if (iattr->ia_valid & ATTR_MTIME)
 577                         inode->i_mtime = iattr->ia_mtime;
 578                 if (iattr->ia_valid & ATTR_CTIME)
 579                         inode->i_ctime = iattr->ia_ctime;
 580                 if (iattr->ia_valid & ATTR_MODE) {
 581                         inode->i_mode = iattr->ia_mode;
 582
 583                         if (!in_group_p(inode->i_gid) &&
 584                             !cfs_capable(CFS_CAP_FSETID))
 585                                 inode->i_mode &= ~S_ISGID;
 586                 }
 587
 588                 inode->i_sb->s_op->dirty_inode(inode);
 589
 590                 goto out;
 591         }
 592
 593         /* Don't allow setattr to change file type */
 594         if (iattr->ia_valid & ATTR_MODE)
 595                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 596                                  (iattr->ia_mode & ~S_IFMT);
 597
 598         /* We set these flags on the client, but have already checked perms
 599          * so don't confuse inode_change_ok. */
 600         iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
 601
 602         if (inode->i_op->setattr) {
 603                 rc = inode->i_op->setattr(dentry, iattr);
 604         } else {
 605                 rc = inode_change_ok(inode, iattr);
 606                 if (!rc)
 607                         rc = inode_setattr(inode, iattr);
 608         }
 609
 610  out:
 611         RETURN(rc);
 612 }
 613
 614 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 615                                  unsigned int cmd, unsigned long arg)
 616 {
 617         int rc = 0;
 618         ENTRY;
 619
 620         /* FIXME: Can't do this because of nested transaction deadlock */
 621         if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 622                 CERROR("can't set data journal flag on file\n");
 623                 RETURN(-EPERM);
 624         }
 625
 626 #ifdef HAVE_EXT4_LDISKFS
 627         /* ext4_ioctl does not have a inode argument */
 628         if (inode->i_fop->unlocked_ioctl)
 629                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 630 #else
 631         if (inode->i_fop->ioctl)
 632                 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
 633 #endif
 634         else
 635                 RETURN(-ENOTTY);
 636
 637         RETURN(rc);
 638 }
 639
 640 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 641                               void *lmm, int lmm_size, const char *name)
 642 {
 643         int rc;
 644
 645         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 646
 647         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 648                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 649
 650
 651         if (rc && rc != -EROFS)
 652                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 653                        inode->i_ino, rc);
 654         return rc;
 655 }
 656
 657 /* Must be called with i_mutex held */
 658 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 659                               const char *name)
 660 {
 661         int rc;
 662
 663         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
 664
 665         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 666                             name, lmm, lmm_size);
 667
 668         /* This gives us the MD size */
 669         if (lmm == NULL)
 670                 return (rc == -ENODATA) ? 0 : rc;
 671
 672         if (rc < 0) {
 673                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 674                        EXT3_XATTR_INDEX_TRUSTED, name,
 675                        inode->i_ino, rc);
 676                 memset(lmm, 0, lmm_size);
 677                 return (rc == -ENODATA) ? 0 : rc;
 678         }
 679
 680         return rc;
 681 }
 682
 683 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 684 {
 685         submit_bio(rw, bio);
 686         return 0;
 687 }
 688
 689 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 690                                     loff_t *off)
 691 {
 692         struct inode *inode = file->f_dentry->d_inode;
 693         int rc = 0;
 694
 695         if (S_ISREG(inode->i_mode))
 696                 rc = file->f_op->read(file, buf, count, off);
 697         else {
 698                 const int blkbits = inode->i_sb->s_blocksize_bits;
 699                 const int blksize = inode->i_sb->s_blocksize;
 700
 701                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 702                        (unsigned long)count, inode->i_ino, *off);
 703                 while (count > 0) {
 704                         struct buffer_head *bh;
 705
 706                         bh = NULL;
 707                         if (*off < i_size_read(inode)) {
 708                                 int err = 0;
 709
 710                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 711                                                 0, &err);
 712
 713                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 714
 715                                 if (bh) {
 716                                         memcpy(buf, bh->b_data, blksize);
 717                                         brelse(bh);
 718                                 } else if (err) {
 719                                         /* XXX in theory we should just fake
 720                                          * this buffer and continue like ext3,
 721                                          * especially if this is a partial read
 722                                          */
 723                                         CERROR("error read dir %lu+%llu: %d\n",
 724                                                inode->i_ino, *off, err);
 725                                         RETURN(err);
 726                                 }
 727                         }
 728                         if (!bh) {
 729                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 730
 731                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 732                                 memset(fake, 0, sizeof(*fake));
 733                                 fake->rec_len = cpu_to_le16(blksize);
 734                         }
 735                         count -= blksize;
 736                         buf += blksize;
 737                         *off += blksize;
 738                         rc += blksize;
 739                 }
 740         }
 741
 742         return rc;
 743 }
 744
 745 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 746 {
 747         struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
 748
 749         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 750
 751         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 752 }
 753
 754 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 755                                       void *handle, fsfilt_cb_t cb_func,
 756                                       void *cb_data)
 757 {
 758         struct fsfilt_cb_data *fcb;
 759
 760         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 761         if (fcb == NULL)
 762                 RETURN(-ENOMEM);
 763
 764         fcb->cb_func = cb_func;
 765         fcb->cb_obd = obd;
 766         fcb->cb_last_rcvd = last_rcvd;
 767         fcb->cb_data = cb_data;
 768
 769         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 770         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
 771                                     (struct journal_callback *)fcb);
 772
 773         return 0;
 774 }
 775
 776 /*
 777  * We need to hack the return value for the free inode counts because
 778  * the current EA code requires one filesystem block per inode with EAs,
 779  * so it is possible to run out of blocks before we run out of inodes.
 780  *
 781  * This can be removed when the ext3 EA code is fixed.
 782  */
 783 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 784 {
 785         struct kstatfs sfs;
 786         int rc;
 787
 788         memset(&sfs, 0, sizeof(sfs));
 789         rc = ll_do_statfs(sb, &sfs);
 790         if (!rc && sfs.f_bfree < sfs.f_ffree) {
 791                 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
 792                 sfs.f_ffree = sfs.f_bfree;
 793         }
 794
 795         statfs_pack(osfs, &sfs);
 796         return rc;
 797 }
 798
 799 static int fsfilt_ext3_sync(struct super_block *sb)
 800 {
 801         return ext3_force_commit(sb);
 802 }
 803
 804 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
 805 #warning "kernel code has old extents/mballoc patch, disabling"
 806 #undef EXT3_MULTIBLOCK_ALLOCATOR
 807 #endif
 808 #ifndef EXT3_EXTENTS_FL
 809 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 810 #endif
 811
 812 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
 813 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
 814 #define fsfilt_up_truncate_sem(inode)  up(&EXT3_I(inode)->truncate_sem);
 815 #define fsfilt_down_truncate_sem(inode)  down(&EXT3_I(inode)->truncate_sem);
 816 #else
 817 #ifdef HAVE_EXT4_LDISKFS
 818 #define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem));
 819 #define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem));
 820 #else
 821 #define fsfilt_up_truncate_sem(inode)  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 822 #define fsfilt_down_truncate_sem(inode)  mutex_lock(&EXT3_I(inode)->truncate_mutex);
 823 #endif
 824 #endif
 825
 826 #ifndef EXT_ASSERT
 827 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 828 #endif
 829
 830 #ifdef EXT3_EXT_HAS_NO_TREE
 831 /* for kernels 2.6.18 and later */
 832 #ifdef HAVE_EXT4_LDISKFS
 833 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 834 #else
 835 #define EXT_GENERATION(inode)           ext_generation(inode)
 836 #endif
 837 #define ext3_ext_base                   inode
 838 #define ext3_ext_base2inode(inode)      (inode)
 839 #define EXT_DEPTH(inode)                ext_depth(inode)
 840 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 841                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 842 #else
 843 #define ext3_ext_base                   ext3_extents_tree
 844 #define ext3_ext_base2inode(tree)       (tree->inode)
 845 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
 846                         ext3_ext_walk_space(tree, block, num, cb);
 847 #endif
 848
 849 #include <linux/lustre_version.h>
 850
 851 struct bpointers {
 852         unsigned long *blocks;
 853         int *created;
 854         unsigned long start;
 855         int num;
 856         int init_num;
 857         int create;
 858 };
 859
 860 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 861                                unsigned long block, int *aflags)
 862 {
 863         struct ext3_inode_info *ei = EXT3_I(inode);
 864         unsigned long bg_start;
 865         unsigned long colour;
 866         int depth;
 867
 868         if (path) {
 869                 struct ext3_extent *ex;
 870                 depth = path->p_depth;
 871
 872                 /* try to predict block placement */
 873                 if ((ex = path[depth].p_ext))
 874                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 875
 876                 /* it looks index is empty
 877                  * try to find starting from index itself */
 878                 if (path[depth].p_bh)
 879                         return path[depth].p_bh->b_blocknr;
 880         }
 881
 882         /* OK. use inode's group */
 883         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 884                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 885         colour = (current->pid % 16) *
 886                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 887         return bg_start + colour + block;
 888 }
 889
 890 #define ll_unmap_underlying_metadata(sb, blocknr) \
 891         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 892
 893 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 894 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 895                                 struct ext3_ext_path *path, unsigned long block,
 896                                 unsigned long *count, int *err)
 897 {
 898         unsigned long pblock, goal;
 899         int aflags = 0;
 900         struct inode *inode = ext3_ext_base2inode(base);
 901
 902         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 903         aflags |= 2; /* block have been already reserved */
 904         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 905         return pblock;
 906
 907 }
 908 #else
 909 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 910                                 struct ext3_ext_path *path, unsigned long block,
 911                                 unsigned long *count, int *err)
 912 {
 913         struct inode *inode = ext3_ext_base2inode(base);
 914         struct ext3_allocation_request ar;
 915         unsigned long pblock;
 916         int aflags;
 917
 918         /* find neighbour allocated blocks */
 919         ar.lleft = block;
 920         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 921         if (*err)
 922                 return 0;
 923         ar.lright = block;
 924         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 925         if (*err)
 926                 return 0;
 927
 928         /* allocate new block */
 929         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 930         ar.inode = inode;
 931         ar.logical = block;
 932         ar.len = *count;
 933         ar.flags = EXT3_MB_HINT_DATA;
 934         pblock = ext3_mb_new_blocks(handle, &ar, err);
 935         *count = ar.len;
 936         return pblock;
 937 }
 938 #endif
 939
 940 #ifdef EXT3_EXT_HAS_NO_TREE
 941 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 942                                   struct ext3_ext_path *path,
 943                                   struct ext3_ext_cache *cex,
 944 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 945                                    struct ext3_extent *ex,
 946 #endif
 947                                   void *cbdata)
 948 {
 949         struct bpointers *bp = cbdata;
 950 #else
 951 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 952                                   struct ext3_ext_path *path,
 953                                   struct ext3_ext_cache *cex
 954 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 955                                   , struct ext3_extent *ex
 956 #endif
 957                                  )
 958 {
 959         struct bpointers *bp = base->private;
 960 #endif
 961         struct inode *inode = ext3_ext_base2inode(base);
 962         struct ext3_extent nex;
 963         unsigned long pblock;
 964         unsigned long tgen;
 965         int err, i;
 966         unsigned long count;
 967         handle_t *handle;
 968
 969         i = EXT_DEPTH(base);
 970         EXT_ASSERT(i == path->p_depth);
 971         EXT_ASSERT(path[i].p_hdr);
 972
 973         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 974                 err = EXT_CONTINUE;
 975                 goto map;
 976         }
 977
 978         if (bp->create == 0) {
 979                 i = 0;
 980                 if (cex->ec_block < bp->start)
 981                         i = bp->start - cex->ec_block;
 982                 if (i >= cex->ec_len)
 983                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 984                                         i, cex->ec_len);
 985                 for (; i < cex->ec_len && bp->num; i++) {
 986                         *(bp->created) = 0;
 987                         bp->created++;
 988                         *(bp->blocks) = 0;
 989                         bp->blocks++;
 990                         bp->num--;
 991                         bp->start++;
 992                 }
 993
 994                 return EXT_CONTINUE;
 995         }
 996
 997         tgen = EXT_GENERATION(base);
 998         count = ext3_ext_calc_credits_for_insert(base, path);
 999         fsfilt_up_truncate_sem(inode);
1000
1001         handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1002         if (IS_ERR(handle)) {
1003                 fsfilt_down_truncate_sem(inode);
1004                 return PTR_ERR(handle);
1005         }
1006
1007         fsfilt_down_truncate_sem(inode);
1008         if (tgen != EXT_GENERATION(base)) {
1009                 /* the tree has changed. so path can be invalid at moment */
1010                 fsfilt_ext3_journal_stop(handle);
1011                 return EXT_REPEAT;
1012         }
1013
1014         count = cex->ec_len;
1015         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1016         if (!pblock)
1017                 goto out;
1018         EXT_ASSERT(count <= cex->ec_len);
1019
1020         /* insert new extent */
1021         nex.ee_block = cpu_to_le32(cex->ec_block);
1022         ext3_ext_store_pblock(&nex, pblock);
1023         nex.ee_len = cpu_to_le16(count);
1024         err = ext3_ext_insert_extent(handle, base, path, &nex);
1025         if (err) {
1026                 /* free data blocks we just allocated */
1027                 /* not a good idea to call discard here directly,
1028                  * but otherwise we'd need to call it every free() */
1029 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1030                 ext3_mb_discard_inode_preallocations(inode);
1031 #endif
1032                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1033                                  cpu_to_le16(nex.ee_len), 0);
1034                 goto out;
1035         }
1036
1037         /*
1038          * Putting len of the actual extent we just inserted,
1039          * we are asking ext3_ext_walk_space() to continue
1040          * scaning after that block
1041          */
1042         cex->ec_len = le16_to_cpu(nex.ee_len);
1043         cex->ec_start = ext_pblock(&nex);
1044         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1045         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1046
1047 out:
1048         fsfilt_ext3_journal_stop(handle);
1049 map:
1050         if (err >= 0) {
1051                 /* map blocks */
1052                 if (bp->num == 0) {
1053                         CERROR("hmm. why do we find this extent?\n");
1054                         CERROR("initial space: %lu:%u\n",
1055                                 bp->start, bp->init_num);
1056                         CERROR("current extent: %u/%u/%llu %d\n",
1057                                 cex->ec_block, cex->ec_len,
1058                                 (unsigned long long)cex->ec_start,
1059                                 cex->ec_type);
1060                 }
1061                 i = 0;
1062                 if (cex->ec_block < bp->start)
1063                         i = bp->start - cex->ec_block;
1064                 if (i >= cex->ec_len)
1065                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1066                                         i, cex->ec_len);
1067                 for (; i < cex->ec_len && bp->num; i++) {
1068                         *(bp->blocks) = cex->ec_start + i;
1069                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1070                                 *(bp->created) = 0;
1071                         } else {
1072                                 *(bp->created) = 1;
1073                                 /* unmap any possible underlying metadata from
1074                                  * the block device mapping.  bug 6998. */
1075                                 ll_unmap_underlying_metadata(inode->i_sb,
1076                                                              *(bp->blocks));
1077                         }
1078                         bp->created++;
1079                         bp->blocks++;
1080                         bp->num--;
1081                         bp->start++;
1082                 }
1083         }
1084         return err;
1085 }
1086
1087 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1088                        unsigned long num, unsigned long *blocks,
1089                        int *created, int create)
1090 {
1091 #ifdef EXT3_EXT_HAS_NO_TREE
1092         struct ext3_ext_base *base = inode;
1093 #else
1094         struct ext3_extents_tree tree;
1095         struct ext3_ext_base *base = &tree;
1096 #endif
1097         struct bpointers bp;
1098         int err;
1099
1100         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1101                block, block + num - 1, (unsigned) inode->i_ino);
1102
1103 #ifndef EXT3_EXT_HAS_NO_TREE
1104         ext3_init_tree_desc(base, inode);
1105         tree.private = &bp;
1106 #endif
1107         bp.blocks = blocks;
1108         bp.created = created;
1109         bp.start = block;
1110         bp.init_num = bp.num = num;
1111         bp.create = create;
1112
1113         fsfilt_down_truncate_sem(inode);
1114         err = fsfilt_ext3_ext_walk_space(base, block, num,
1115                                          ext3_ext_new_extent_cb, &bp);
1116         ext3_ext_invalidate_cache(base);
1117         fsfilt_up_truncate_sem(inode);
1118
1119         return err;
1120 }
1121
1122 int fsfilt_ext3_map_ext_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         int rc = 0, i = 0;
1128         struct page *fp = NULL;
1129         int clen = 0;
1130
1131         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1132                 inode->i_ino, pages, (*page)->index);
1133
1134         /* pages are sorted already. so, we just have to find
1135          * contig. space and process them properly */
1136         while (i < pages) {
1137                 if (fp == NULL) {
1138                         /* start new extent */
1139                         fp = *page++;
1140                         clen = 1;
1141                         i++;
1142                         continue;
1143                 } else if (fp->index + clen == (*page)->index) {
1144                         /* continue the extent */
1145                         page++;
1146                         clen++;
1147                         i++;
1148                         continue;
1149                 }
1150
1151                 /* process found extent */
1152                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1153                                         clen * blocks_per_page, blocks,
1154                                         created, create);
1155                 if (rc)
1156                         GOTO(cleanup, rc);
1157
1158                 /* look for next extent */
1159                 fp = NULL;
1160                 blocks += blocks_per_page * clen;
1161                 created += blocks_per_page * clen;
1162         }
1163
1164         if (fp)
1165                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1166                                         clen * blocks_per_page, blocks,
1167                                         created, create);
1168 cleanup:
1169         return rc;
1170 }
1171 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1172
1173 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1174                                unsigned long *blocks, int *created, int create);
1175 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1176                                    int pages, unsigned long *blocks,
1177                                    int *created, int create)
1178 {
1179         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1180         unsigned long *b;
1181         int rc = 0, i, *cr;
1182
1183         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1184                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1185                 if (rc) {
1186                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1187                                inode->i_ino, *b, *cr, create, rc);
1188                         break;
1189                 }
1190
1191                 b += blocks_per_page;
1192                 cr += blocks_per_page;
1193         }
1194         return rc;
1195 }
1196
1197 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1198                                 int pages, unsigned long *blocks,
1199                                 int *created, int create,
1200                                 struct semaphore *optional_sem)
1201 {
1202         int rc;
1203 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1204         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1205                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1206                                                      blocks, created, create);
1207                 return rc;
1208         }
1209 #endif
1210         if (optional_sem != NULL)
1211                 down(optional_sem);
1212         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1213                                             created, create);
1214         if (optional_sem != NULL)
1215                 up(optional_sem);
1216
1217         return rc;
1218 }
1219
1220 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1221 {
1222         unsigned long block;
1223         struct buffer_head *bh;
1224         int err, blocksize, csize, boffs, osize = size;
1225
1226         /* prevent reading after eof */
1227         lock_kernel();
1228         if (i_size_read(inode) < *offs + size) {
1229                 size = i_size_read(inode) - *offs;
1230                 unlock_kernel();
1231                 if (size < 0) {
1232                         CERROR("size %llu is too short for read %u@%llu\n",
1233                                i_size_read(inode), size, *offs);
1234                         return -EIO;
1235                 } else if (size == 0) {
1236                         return 0;
1237                 }
1238         } else {
1239                 unlock_kernel();
1240         }
1241
1242         blocksize = 1 << inode->i_blkbits;
1243
1244         while (size > 0) {
1245                 block = *offs >> inode->i_blkbits;
1246                 boffs = *offs & (blocksize - 1);
1247                 csize = min(blocksize - boffs, size);
1248                 bh = ext3_bread(NULL, inode, block, 0, &err);
1249                 if (!bh) {
1250                         CERROR("can't read block: %d\n", err);
1251                         return err;
1252                 }
1253
1254                 memcpy(buf, bh->b_data + boffs, csize);
1255                 brelse(bh);
1256
1257                 *offs += csize;
1258                 buf += csize;
1259                 size -= csize;
1260         }
1261         return osize;
1262 }
1263 EXPORT_SYMBOL(fsfilt_ext3_read);
1264
1265 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1266                                    int size, loff_t *offs)
1267 {
1268         int rc;
1269         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1270         if (rc > 0)
1271                 rc = 0;
1272         return rc;
1273 }
1274
1275 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1276                                 loff_t *offs, handle_t *handle)
1277 {
1278         struct buffer_head *bh = NULL;
1279         loff_t old_size = i_size_read(inode), offset = *offs;
1280         loff_t new_size = i_size_read(inode);
1281         unsigned long block;
1282         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1283
1284         while (bufsize > 0) {
1285                 if (bh != NULL)
1286                         brelse(bh);
1287
1288                 block = offset >> inode->i_blkbits;
1289                 boffs = offset & (blocksize - 1);
1290                 size = min(blocksize - boffs, bufsize);
1291                 bh = ext3_bread(handle, inode, block, 1, &err);
1292                 if (!bh) {
1293                         CERROR("can't read/create block: %d\n", err);
1294                         break;
1295                 }
1296
1297                 err = ext3_journal_get_write_access(handle, bh);
1298                 if (err) {
1299                         CERROR("journal_get_write_access() returned error %d\n",
1300                                err);
1301                         break;
1302                 }
1303                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1304                 memcpy(bh->b_data + boffs, buf, size);
1305                 err = ext3_journal_dirty_metadata(handle, bh);
1306                 if (err) {
1307                         CERROR("journal_dirty_metadata() returned error %d\n",
1308                                err);
1309                         break;
1310                 }
1311                 if (offset + size > new_size)
1312                         new_size = offset + size;
1313                 offset += size;
1314                 bufsize -= size;
1315                 buf += size;
1316         }
1317         if (bh)
1318                 brelse(bh);
1319
1320         /* correct in-core and on-disk sizes */
1321         if (new_size > i_size_read(inode)) {
1322                 lock_kernel();
1323                 if (new_size > i_size_read(inode))
1324                         i_size_write(inode, new_size);
1325                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1326                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1327                 if (i_size_read(inode) > old_size)
1328                         mark_inode_dirty(inode);
1329                 unlock_kernel();
1330         }
1331
1332         if (err == 0)
1333                 *offs = offset;
1334         return err;
1335 }
1336 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1337
1338 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1339                                     loff_t *offs, int force_sync)
1340 {
1341         struct inode *inode = file->f_dentry->d_inode;
1342         handle_t *handle;
1343         int err, block_count = 0, blocksize;
1344
1345         /* Determine how many transaction credits are needed */
1346         blocksize = 1 << inode->i_blkbits;
1347         block_count = (*offs & (blocksize - 1)) + bufsize;
1348         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1349
1350         handle = fsfilt_ext3_journal_start(inode,
1351                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1352         if (IS_ERR(handle)) {
1353                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1354                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1355                 return PTR_ERR(handle);
1356         }
1357
1358         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1359
1360         if (!err && force_sync)
1361                 handle->h_sync = 1; /* recovery likes this */
1362
1363         fsfilt_ext3_journal_stop(handle);
1364
1365         return err;
1366 }
1367
1368 static int fsfilt_ext3_setup(struct super_block *sb)
1369 {
1370 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1371      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1372         struct ext3_sb_info *sbi = EXT3_SB(sb);
1373 #if 0
1374         sbi->dx_lock = fsfilt_ext3_dx_lock;
1375         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1376 #endif
1377 #endif
1378 #ifdef S_PDIROPS
1379         CWARN("Enabling PDIROPS\n");
1380         set_opt(sbi->s_mount_opt, PDIROPS);
1381         sb->s_flags |= S_PDIROPS;
1382 #endif
1383         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1384                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1385 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1386         /* enable journaled quota support */
1387         /* kfreed in ext3_put_super() */
1388         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1389         if (!sbi->s_qf_names[USRQUOTA])
1390                 return -ENOMEM;
1391         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1392         if (!sbi->s_qf_names[GRPQUOTA]) {
1393                 kfree(sbi->s_qf_names[USRQUOTA]);
1394                 sbi->s_qf_names[USRQUOTA] = NULL;
1395                 return -ENOMEM;
1396         }
1397         sbi->s_jquota_fmt = QFMT_VFS_V0;
1398 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1399         set_opt(sbi->s_mount_opt, QUOTA);
1400 #endif
1401 #endif
1402         return 0;
1403 }
1404
1405 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1406    objects. Logs is number of logfiles to update */
1407 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1408 {
1409         if ( !fso ) {
1410                 switch(op) {
1411                 case FSFILT_OP_CREATE:
1412                                  /* directory leaf, index & indirect & EA*/
1413                         return 4 + 3 * logs;
1414                 case FSFILT_OP_UNLINK:
1415                         return 3 * logs;
1416                 }
1417         } else {
1418                 int i;
1419                 int needed = 0;
1420                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1421                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1422                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1423                 for (i = 0; i < op; i++, fso++) {
1424                         int nblocks = fso->fso_bufcnt * blockpp;
1425                         int ndindirect = min(nblocks, addrpp + 1);
1426                         int nindir = nblocks + ndindirect + 1;
1427
1428                         needed += nindir;
1429                 }
1430                 return needed + 3 * logs;
1431         }
1432
1433         return 0;
1434 }
1435
1436 #ifdef HAVE_QUOTA_SUPPORT
1437 #define DQINFO_COPY(out, in)                    \
1438 do {                                            \
1439         Q_COPY(out, in, dqi_bgrace);            \
1440         Q_COPY(out, in, dqi_igrace);            \
1441         Q_COPY(out, in, dqi_flags);             \
1442         Q_COPY(out, in, dqi_valid);             \
1443 } while (0)
1444
1445 #define DQBLK_COPY(out, in)                     \
1446 do {                                            \
1447         Q_COPY(out, in, dqb_bhardlimit);        \
1448         Q_COPY(out, in, dqb_bsoftlimit);        \
1449         Q_COPY(out, in, dqb_curspace);          \
1450         Q_COPY(out, in, dqb_ihardlimit);        \
1451         Q_COPY(out, in, dqb_isoftlimit);        \
1452         Q_COPY(out, in, dqb_curinodes);         \
1453         Q_COPY(out, in, dqb_btime);             \
1454         Q_COPY(out, in, dqb_itime);             \
1455         Q_COPY(out, in, dqb_valid);             \
1456 } while (0)
1457
1458 static int fsfilt_ext3_quotactl(struct super_block *sb,
1459                                 struct obd_quotactl *oqc)
1460 {
1461         int i, rc = 0, error = 0;
1462         struct quotactl_ops *qcop;
1463         struct if_dqinfo *info;
1464         struct if_dqblk *dqblk;
1465         ENTRY;
1466
1467         if (!sb->s_qcop)
1468                 RETURN(-ENOSYS);
1469
1470         OBD_ALLOC_PTR(info);
1471         if (!info)
1472                 RETURN(-ENOMEM);
1473         OBD_ALLOC_PTR(dqblk);
1474         if (!dqblk) {
1475                 OBD_FREE_PTR(info);
1476                 RETURN(-ENOMEM);
1477         }
1478
1479         DQINFO_COPY(info, &oqc->qc_dqinfo);
1480         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1481
1482         qcop = sb->s_qcop;
1483         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1484                 for (i = 0; i < MAXQUOTAS; i++) {
1485                         if (!Q_TYPESET(oqc, i))
1486                                 continue;
1487
1488                         if (oqc->qc_cmd == Q_QUOTAON) {
1489                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1490
1491                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1492
1493                                 if (!qcop->quota_on)
1494                                         GOTO(out, rc = -ENOSYS);
1495
1496                                 rc = qcop->quota_on(sb, i, QFMT_VFS_V0,
1497                                                     name[i]);
1498                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1499                                 if (!qcop->quota_off)
1500                                         GOTO(out, rc = -ENOSYS);
1501                                 rc = qcop->quota_off(sb, i);
1502                         }
1503
1504                         if (rc == -EBUSY)
1505                                 error = rc;
1506                         else if (rc)
1507                                 GOTO(out, rc);
1508                 }
1509                 GOTO(out, rc ?: error);
1510         }
1511
1512         switch (oqc->qc_cmd) {
1513         case Q_GETOINFO:
1514         case Q_GETINFO:
1515                 if (!qcop->get_info)
1516                         GOTO(out, rc = -ENOSYS);
1517                 rc = qcop->get_info(sb, oqc->qc_type, info);
1518                 break;
1519         case Q_SETQUOTA:
1520         case Q_INITQUOTA:
1521                 if (!qcop->set_dqblk)
1522                         GOTO(out, rc = -ENOSYS);
1523                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1524                 break;
1525         case Q_GETOQUOTA:
1526         case Q_GETQUOTA:
1527                 if (!qcop->get_dqblk)
1528                         GOTO(out, rc = -ENOSYS);
1529                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1530                 if (!rc)
1531                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1532                 break;
1533         case Q_SYNC:
1534                 if (!sb->s_qcop->quota_sync)
1535                         GOTO(out, rc = -ENOSYS);
1536                 qcop->quota_sync(sb, oqc->qc_type);
1537                 break;
1538         case Q_FINVALIDATE:
1539                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1540                 for (i = 0; i < MAXQUOTAS; i++) {
1541                         struct file *fp;
1542                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1543
1544                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1545
1546                         if (!Q_TYPESET(oqc, i))
1547                                 continue;
1548
1549                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1550                         if (IS_ERR(fp)) {
1551                                 rc = PTR_ERR(fp);
1552                                 CERROR("error invalidating operational quota file"
1553                                        " %s (rc:%d)\n", name[i], rc);
1554                         } else {
1555                                 filp_close(fp, 0);
1556                         }
1557
1558                 }
1559                 break;
1560         default:
1561                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1562                 LBUG();
1563         }
1564 out:
1565         DQINFO_COPY(&oqc->qc_dqinfo, info);
1566         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1567
1568         OBD_FREE_PTR(info);
1569         OBD_FREE_PTR(dqblk);
1570
1571         if (rc)
1572                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1573                                 "failed: %d\n",
1574                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1575         RETURN(rc);
1576 }
1577
1578 struct chk_dqblk{
1579         struct hlist_node       dqb_hash;        /** quotacheck hash */
1580         struct list_head        dqb_list;        /** in list also */
1581         qid_t                   dqb_id;          /** uid/gid */
1582         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1583         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1584         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1585         qsize_t                 dqb_curspace;    /** current space */
1586         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1587         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1588         qsize_t                 dqb_curinodes;   /** current inodes */
1589         __u64                   dqb_btime;       /** block grace time */
1590         __u64                   dqb_itime;       /** inode grace time */
1591         __u32                   dqb_valid;       /** flag for above fields */
1592 };
1593
1594 static inline unsigned int chkquot_hash(qid_t id, int type)
1595                                         __attribute__((__const__));
1596
1597 static inline unsigned int chkquot_hash(qid_t id, int type)
1598 {
1599         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1600 }
1601
1602 static inline struct chk_dqblk *
1603 find_chkquot(struct hlist_head *head, qid_t id, int type)
1604 {
1605         struct hlist_node *node;
1606         struct chk_dqblk *cdqb;
1607
1608         hlist_for_each(node, head) {
1609                 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1610                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1611                         return cdqb;
1612         }
1613
1614         return NULL;
1615 }
1616
1617 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1618 {
1619         struct chk_dqblk *cdqb;
1620
1621         OBD_ALLOC_PTR(cdqb);
1622         if (cdqb) {
1623                 INIT_HLIST_NODE(&cdqb->dqb_hash);
1624                 INIT_LIST_HEAD(&cdqb->dqb_list);
1625                 cdqb->dqb_id = id;
1626                 cdqb->dqb_type = type;
1627         }
1628
1629         return cdqb;
1630 }
1631
1632 static struct chk_dqblk *
1633 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1634       qid_t id, int type, int first_check)
1635 {
1636         struct hlist_head *head = hash + chkquot_hash(id, type);
1637         struct if_dqblk dqb;
1638         struct chk_dqblk *cdqb;
1639         int rc;
1640
1641         cdqb = find_chkquot(head, id, type);
1642         if (cdqb)
1643                 return cdqb;
1644
1645         cdqb = alloc_chkquot(id, type);
1646         if (!cdqb)
1647                 return NULL;
1648
1649         if (!first_check) {
1650                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1651                 if (rc) {
1652                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1653                                id, type, rc);
1654                 } else {
1655                         DQBLK_COPY(cdqb, &dqb);
1656                         cdqb->dqb_curspace = 0;
1657                         cdqb->dqb_curinodes = 0;
1658                 }
1659         }
1660
1661         hlist_add_head(&cdqb->dqb_hash, head);
1662         list_add_tail(&cdqb->dqb_list, list);
1663
1664         return cdqb;
1665 }
1666
1667 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1668 {
1669         struct obd_quotactl *oqctl;
1670         int rc;
1671
1672         OBD_ALLOC_PTR(oqctl);
1673         if (!oqctl)
1674                 RETURN(-ENOMEM);
1675
1676         oqctl->qc_cmd = cmd;
1677         oqctl->qc_id = qfmt;
1678         oqctl->qc_type = type;
1679         rc = fsfilt_ext3_quotactl(sb, oqctl);
1680
1681         OBD_FREE_PTR(oqctl);
1682         return rc;
1683 }
1684
1685 static inline int read_old_dqinfo(struct super_block *sb, int type,
1686                                   struct if_dqinfo *dqinfo)
1687 {
1688         struct obd_quotactl *oqctl;
1689         int rc;
1690         ENTRY;
1691
1692         OBD_ALLOC_PTR(oqctl);
1693         if (!oqctl)
1694                 RETURN(-ENOMEM);
1695
1696         oqctl->qc_cmd = Q_GETINFO;
1697         oqctl->qc_type = type;
1698         rc = fsfilt_ext3_quotactl(sb, oqctl);
1699         if (!rc)
1700                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1701
1702         OBD_FREE_PTR(oqctl);
1703         RETURN(rc);
1704 }
1705
1706 static inline struct ext3_group_desc *
1707 get_group_desc(struct super_block *sb, int group)
1708 {
1709         unsigned long desc_block, desc;
1710         struct ext3_group_desc *gdp;
1711
1712         desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1713         desc = group % EXT3_DESC_PER_BLOCK(sb);
1714         gdp = (struct ext3_group_desc *)
1715               EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1716
1717         return gdp + desc;
1718 }
1719
1720
1721 #ifndef HAVE_EXT4_LDISKFS
1722 static inline struct buffer_head *
1723 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1724 {
1725         struct ext3_group_desc *desc;
1726         struct buffer_head *bh;
1727
1728         desc = get_group_desc(sb, group);
1729         bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1730         return bh;
1731 }
1732 #endif
1733
1734 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1735                                      struct buffer_head *bitmap_bh,
1736                                      int index, unsigned long ino)
1737 {
1738         struct inode *inode = NULL;
1739
1740
1741         if (ext3_test_bit(index, bitmap_bh->b_data))
1742 #ifdef HAVE_EXT4_LDISKFS
1743                 inode = ext4_iget(sb, ino);
1744                 if (IS_ERR(inode))
1745                 /* Newer kernels return an error instead of a NULL pointer */
1746                         inode = NULL;
1747 #else
1748                 inode = iget(sb, ino);
1749 #endif
1750         return inode;
1751 }
1752
1753 struct qchk_ctxt {
1754         struct hlist_head       qckt_hash[NR_DQHASH];        /* quotacheck hash */
1755         struct list_head        qckt_list;                   /* quotacheck list */
1756         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1757         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];      /* old dqinfo */
1758 };
1759
1760 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1761                            struct obd_quotactl *oqc)
1762 {
1763         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1764         loff_t size = 0;
1765         qid_t qid[MAXQUOTAS];
1766         int cnt, i, rc = 0;
1767
1768         if (!inode)
1769                 return 0;
1770
1771         qid[USRQUOTA] = inode->i_uid;
1772         qid[GRPQUOTA] = inode->i_gid;
1773
1774         if (S_ISDIR(inode->i_mode) ||
1775             S_ISREG(inode->i_mode) ||
1776             S_ISLNK(inode->i_mode))
1777                 size = inode_get_bytes(inode);
1778
1779         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1780                 if (!Q_TYPESET(oqc, cnt))
1781                         continue;
1782
1783                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1784                                 &qctxt->qckt_list, qid[cnt], cnt,
1785                                 qctxt->qckt_first_check[cnt]);
1786                 if (!cdqb[cnt]) {
1787                         rc = -ENOMEM;
1788                         break;
1789                 }
1790
1791                 cdqb[cnt]->dqb_curspace += size;
1792                 cdqb[cnt]->dqb_curinodes++;
1793         }
1794
1795         if (rc) {
1796                 for (i = 0; i < cnt; i++) {
1797                         if (!Q_TYPESET(oqc, i))
1798                                 continue;
1799                         LASSERT(cdqb[i]);
1800                         cdqb[i]->dqb_curspace -= size;
1801                         cdqb[i]->dqb_curinodes--;
1802                 }
1803         }
1804
1805         return rc;
1806 }
1807
1808 /* write dqinfo struct in a new quota file */
1809 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1810 {
1811         struct v2_disk_dqinfo dqinfo;
1812         __u32 blocks = V2_DQTREEOFF + 1;
1813         loff_t offset = V2_DQINFOOFF;
1814
1815         if (info) {
1816                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1817                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1818                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1819                                                ~DQF_INFO_DIRTY);
1820         } else {
1821                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1822                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1823                 dqinfo.dqi_flags = 0;
1824         }
1825
1826         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1827         dqinfo.dqi_free_blk = 0;
1828         dqinfo.dqi_free_entry = 0;
1829
1830         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1831 }
1832
1833 static int v3_write_dqheader(struct file *f, int type)
1834 {
1835         static const __u32 quota_magics[] = V2_INITQMAGICS;
1836         static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1837         struct v2_disk_dqheader dqhead;
1838         loff_t offset = 0;
1839
1840         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1841         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1842
1843         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1844         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1845
1846         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1847 }
1848
1849 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1850                                   struct obd_quotactl *oqc)
1851 {
1852         int i, rc = 0;
1853         ENTRY;
1854
1855         for (i = 0; i < MAXQUOTAS; i++) {
1856                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1857                                          NULL : &qctxt->qckt_dqinfo[i];
1858                 struct file *file;
1859                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1860
1861                 if (!Q_TYPESET(oqc, i))
1862                         continue;
1863
1864                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1865
1866                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1867                 if (IS_ERR(file)) {
1868                         rc = PTR_ERR(file);
1869                         CERROR("can't create %s file: rc = %d\n",
1870                                name[i], rc);
1871                         GOTO(out, rc);
1872                 }
1873
1874                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1875                         CERROR("file %s is not regular", name[i]);
1876                         filp_close(file, 0);
1877                         GOTO(out, rc = -EINVAL);
1878                 }
1879
1880                 DQUOT_DROP(file->f_dentry->d_inode);
1881
1882                 rc = v3_write_dqheader(file, i);
1883                 if (rc) {
1884                         filp_close(file, 0);
1885                         GOTO(out, rc);
1886                 }
1887
1888                 rc = v3_write_dqinfo(file, i, info);
1889                 filp_close(file, 0);
1890                 if (rc)
1891                         GOTO(out, rc);
1892         }
1893
1894 out:
1895         RETURN(rc);
1896 }
1897
1898
1899 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1900                           struct chk_dqblk *cdqb)
1901 {
1902         struct obd_quotactl *oqc;
1903         long now;
1904         int rc;
1905         ENTRY;
1906
1907         OBD_ALLOC_PTR(oqc);
1908         if (!oqc)
1909                 RETURN(-ENOMEM);
1910
1911         now = cfs_time_current_sec();
1912
1913         if (cdqb->dqb_bsoftlimit &&
1914             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1915             !cdqb->dqb_btime)
1916                 cdqb->dqb_btime =
1917                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1918
1919         if (cdqb->dqb_isoftlimit &&
1920             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1921             !cdqb->dqb_itime)
1922                 cdqb->dqb_itime =
1923                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1924
1925         cdqb->dqb_valid = QIF_ALL;
1926
1927         oqc->qc_cmd = Q_SETQUOTA;
1928         oqc->qc_type = cdqb->dqb_type;
1929         oqc->qc_id = cdqb->dqb_id;
1930         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1931
1932         rc = fsfilt_ext3_quotactl(sb, oqc);
1933         OBD_FREE_PTR(oqc);
1934         RETURN(rc);
1935 }
1936
1937 static int prune_chkquots(struct super_block *sb,
1938                           struct qchk_ctxt *qctxt, int error)
1939 {
1940         struct chk_dqblk *cdqb, *tmp;
1941         int rc;
1942
1943         list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1944                 if (!error) {
1945                         rc = commit_chkquot(sb, qctxt, cdqb);
1946                         if (rc)
1947                                 error = rc;
1948                 }
1949                 hlist_del_init(&cdqb->dqb_hash);
1950                 list_del(&cdqb->dqb_list);
1951                 OBD_FREE_PTR(cdqb);
1952         }
1953
1954         return error;
1955 }
1956
1957 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1958                                   struct obd_quotactl *oqc)
1959 {
1960         struct ext3_sb_info *sbi = EXT3_SB(sb);
1961         int i, group;
1962         struct qchk_ctxt *qctxt;
1963         struct buffer_head *bitmap_bh = NULL;
1964         unsigned long ino;
1965         struct inode *inode;
1966         int rc = 0;
1967         ENTRY;
1968
1969         /* turn on quota and read dqinfo if existed */
1970         OBD_ALLOC_PTR(qctxt);
1971         if (!qctxt) {
1972                 oqc->qc_stat = -ENOMEM;
1973                 RETURN(-ENOMEM);
1974         }
1975
1976         for (i = 0; i < NR_DQHASH; i++)
1977                 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1978         INIT_LIST_HEAD(&qctxt->qckt_list);
1979
1980         for (i = 0; i < MAXQUOTAS; i++) {
1981                 if (!Q_TYPESET(oqc, i))
1982                         continue;
1983
1984                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1985                 if (!rc || rc == -EBUSY) {
1986                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1987                         if (rc)
1988                                 GOTO(out, rc);
1989                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1990                         qctxt->qckt_first_check[i] = 1;
1991                 } else if (rc) {
1992                         GOTO(out, rc);
1993                 }
1994         }
1995
1996         /* check quota and update in hash */
1997         for (group = 0; group < sbi->s_groups_count; group++) {
1998                 struct ext3_group_desc *desc;
1999                 desc = get_group_desc(sb, group);
2000                 if (!desc)
2001                         GOTO(out, -EIO);
2002
2003                 spin_lock(sb_bgl_lock(sbi, group));
2004                 if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2005                         /* no inode in use in this group, just skip it */
2006                         spin_unlock(sb_bgl_lock(sbi, group));
2007                         continue;
2008                 }
2009                 spin_unlock(sb_bgl_lock(sbi, group));
2010
2011                 ino = group * sbi->s_inodes_per_group + 1;
2012                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2013                 if (!bitmap_bh) {
2014                         CERROR("ext3_read_inode_bitmap group %d failed", group);
2015                         GOTO(out, rc = -EIO);
2016                 }
2017
2018                 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
2019                         if (ino < sbi->s_first_ino)
2020                                 continue;
2021
2022                         inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
2023                         rc = add_inode_quota(inode, qctxt, oqc);
2024                         iput(inode);
2025                         if (rc) {
2026                                 brelse(bitmap_bh);
2027                                 GOTO(out, rc);
2028                         }
2029                 }
2030
2031                 brelse(bitmap_bh);
2032         }
2033
2034         /* read old quota limits from old quota file. (only for the user
2035          * has limits but hasn't file) */
2036 #ifdef HAVE_QUOTA_SUPPORT
2037         for (i = 0; i < MAXQUOTAS; i++) {
2038                 struct list_head id_list;
2039                 struct dquot_id *dqid, *tmp;
2040
2041                 if (!Q_TYPESET(oqc, i))
2042                         continue;
2043
2044                 if (qctxt->qckt_first_check[i])
2045                         continue;
2046
2047
2048                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2049                 INIT_LIST_HEAD(&id_list);
2050 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2051                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2052 #else
2053                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2054 #endif
2055                 if (rc)
2056                         CERROR("read old limits failed. (rc:%d)\n", rc);
2057
2058                 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2059                         list_del_init(&dqid->di_link);
2060
2061                         if (!rc)
2062                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2063                                       dqid->di_id, i,
2064                                       qctxt->qckt_first_check[i]);
2065                         OBD_FREE_PTR(dqid);
2066                 }
2067         }
2068 #endif
2069         /* turn off quota cause we are to dump chk_dqblk to files */
2070         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2071
2072         rc = create_new_quota_files(qctxt, oqc);
2073         if (rc)
2074                 GOTO(out, rc);
2075
2076         /* we use vfs functions to set dqblk, so turn quota on */
2077         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2078 out:
2079         /* dump and free chk_dqblk */
2080         rc = prune_chkquots(sb, qctxt, rc);
2081         OBD_FREE_PTR(qctxt);
2082
2083         /* turn off quota, `lfs quotacheck` will turn on when all
2084          * nodes quotacheck finish. */
2085         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2086
2087         oqc->qc_stat = rc;
2088         if (rc)
2089                 CERROR("quotacheck failed: rc = %d\n", rc);
2090
2091         RETURN(rc);
2092 }
2093
2094 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2095                                  int cmd)
2096 {
2097         int rc = 0;
2098         ENTRY;
2099
2100         if (lqi->qi_files[type] == NULL) {
2101                 CERROR("operate qinfo before it's enabled!\n");
2102                 RETURN(-EIO);
2103         }
2104
2105         switch (cmd) {
2106         case QFILE_CHK:
2107                 rc = lustre_check_quota_file(lqi, type);
2108                 break;
2109         case QFILE_RD_INFO:
2110                 rc = lustre_read_quota_info(lqi, type);
2111                 break;
2112         case QFILE_WR_INFO:
2113                 rc = lustre_write_quota_info(lqi, type);
2114                 break;
2115         case QFILE_INIT_INFO:
2116                 rc = lustre_init_quota_info(lqi, type);
2117                 break;
2118         case QFILE_CONVERT:
2119                 rc = -ENOTSUPP;
2120                 CERROR("quota CONVERT command is not supported\n");
2121                 break;
2122         default:
2123                 rc = -ENOTSUPP;
2124                 CERROR("Unsupported admin quota file cmd %d\n"
2125                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2126                        cmd);
2127                 break;
2128         }
2129         RETURN(rc);
2130 }
2131
2132 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2133                             struct list_head *list)
2134 {
2135         return lustre_get_qids(file, inode, type, list);
2136 }
2137
2138 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2139 {
2140         int rc = 0;
2141         ENTRY;
2142
2143         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2144                 CERROR("operate dquot before it's enabled!\n");
2145                 RETURN(-EIO);
2146         }
2147
2148         switch (cmd) {
2149         case QFILE_RD_DQUOT:
2150                 rc = lustre_read_dquot(dquot);
2151                 break;
2152         case QFILE_WR_DQUOT:
2153                 if (dquot->dq_dqb.dqb_ihardlimit ||
2154                     dquot->dq_dqb.dqb_isoftlimit ||
2155                     dquot->dq_dqb.dqb_bhardlimit ||
2156                     dquot->dq_dqb.dqb_bsoftlimit)
2157                         clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2158                 else
2159                         set_bit(DQ_FAKE_B, &dquot->dq_flags);
2160
2161                 rc = lustre_commit_dquot(dquot);
2162                 if (rc >= 0)
2163                         rc = 0;
2164                 break;
2165         default:
2166                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2167                 LBUG();
2168                 break;
2169         }
2170         RETURN(rc);
2171 }
2172
2173 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2174                                 struct inode *inode, int frags)
2175 {
2176 #ifdef EXT3_EXT_HAS_NO_TREE
2177         struct ext3_ext_base *base = inode;
2178 #else
2179         struct ext3_extents_tree tree;
2180         struct ext3_ext_base *base = &tree;
2181
2182         ext3_init_tree_desc(base, inode);
2183 #endif
2184         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2185          * metablocks at maxium b=16542 */
2186         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2187         return 0;
2188 }
2189
2190 #endif
2191
2192 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2193 {
2194         return (EXT3_SB(sb)->journal_bdev);
2195 }
2196 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2197
2198 static struct fsfilt_operations fsfilt_ext3_ops = {
2199         .fs_type                = "ext3",
2200         .fs_owner               = THIS_MODULE,
2201         .fs_getlabel            = fsfilt_ext3_get_label,
2202         .fs_setlabel            = fsfilt_ext3_set_label,
2203         .fs_uuid                = fsfilt_ext3_uuid,
2204         .fs_start               = fsfilt_ext3_start,
2205         .fs_brw_start           = fsfilt_ext3_brw_start,
2206         .fs_extend              = fsfilt_ext3_extend,
2207         .fs_commit              = fsfilt_ext3_commit,
2208         .fs_commit_async        = fsfilt_ext3_commit_async,
2209         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2210         .fs_setattr             = fsfilt_ext3_setattr,
2211         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2212         .fs_set_md              = fsfilt_ext3_set_md,
2213         .fs_get_md              = fsfilt_ext3_get_md,
2214         .fs_readpage            = fsfilt_ext3_readpage,
2215         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2216         .fs_statfs              = fsfilt_ext3_statfs,
2217         .fs_sync                = fsfilt_ext3_sync,
2218         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2219         .fs_write_record        = fsfilt_ext3_write_record,
2220         .fs_read_record         = fsfilt_ext3_read_record,
2221         .fs_setup               = fsfilt_ext3_setup,
2222         .fs_send_bio            = fsfilt_ext3_send_bio,
2223         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2224 #ifdef HAVE_DISK_INODE_VERSION
2225         .fs_get_version         = fsfilt_ext3_get_version,
2226         .fs_set_version         = fsfilt_ext3_set_version,
2227 #endif
2228 #ifdef HAVE_QUOTA_SUPPORT
2229         .fs_quotactl            = fsfilt_ext3_quotactl,
2230         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2231         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2232         .fs_qids                = fsfilt_ext3_qids,
2233         .fs_dquot               = fsfilt_ext3_dquot,
2234         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2235 #endif
2236         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2237 };
2238
2239 static int __init fsfilt_ext3_init(void)
2240 {
2241         int rc;
2242
2243         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2244                                          sizeof(struct fsfilt_cb_data), 0, 0);
2245         if (!fcb_cache) {
2246                 CERROR("error allocating fsfilt journal callback cache\n");
2247                 GOTO(out, rc = -ENOMEM);
2248         }
2249
2250         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2251
2252         if (rc) {
2253                 int err = cfs_mem_cache_destroy(fcb_cache);
2254                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2255         }
2256 out:
2257         return rc;
2258 }
2259
2260 static void __exit fsfilt_ext3_exit(void)
2261 {
2262         int rc;
2263
2264         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2265         rc = cfs_mem_cache_destroy(fcb_cache);
2266         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2267 }
2268
2269 module_init(fsfilt_ext3_init);
2270 module_exit(fsfilt_ext3_exit);
2271
2272 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2273 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2274 MODULE_LICENSE("GPL");