lustre/lvfs/fsfilt_ext3.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  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_LINUX_EXPORTFS_H
  50 #include <linux/exportfs.h>
  51 #endif
  52 #include <ext4/ext4.h>
  53 #include <ext4/ext4_jbd2.h>
  54 #include <linux/version.h>
  55 #include <linux/bitops.h>
  56 #include <linux/quota.h>
  57 #ifdef HAVE_QUOTAIO_H
  58 # include <linux/quotaio_v2.h>
  59 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
  60 # include <quota/quotaio_v2.h>
  61 # include <quota/quota_tree.h>
  62 # define V2_DQTREEOFF    QT_TREEOFF
  63 #elif defined(HAVE_FS_QUOTAIO_H)
  64 # include <quotaio_v2.h>
  65 # include <quota_tree.h>
  66 # define V2_DQTREEOFF    QT_TREEOFF
  67 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
  68 #endif
  69
  70 #ifdef QFMT_VFS_V1
  71 #define QFMT_LUSTRE QFMT_VFS_V1
  72 #else
  73 #define QFMT_LUSTRE QFMT_VFS_V0
  74 #endif
  75
  76 #if defined(HAVE_EXT3_XATTR_H)
  77 # include <ext3/xattr.h>
  78 #elif !defined(EXT3_XATTR_INDEX_TRUSTED)
  79 /* ext3 xattr.h not available in rh style kernel-devel rpm */
  80 /* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
  81 # define EXT3_XATTR_INDEX_TRUSTED        4
  82 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
  83 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
  84 #endif
  85
  86 #include <libcfs/libcfs.h>
  87 #include <lustre_fsfilt.h>
  88 #include <obd.h>
  89 #include <lustre_quota.h>
  90 #include <linux/lustre_compat25.h>
  91 #include <linux/lprocfs_status.h>
  92
  93 #include <ext4/ext4_extents.h>
  94
  95 #include "lustre_quota_fmt.h"
  96
  97 /* for kernels 2.6.18 and later */
  98 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
  99
 100 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
 101                ext3_ext_insert_extent(handle, inode, path, newext, flag)
 102
 103 #define ext3_mb_discard_inode_preallocations(inode) \
 104                  ext3_discard_preallocations(inode)
 105
 106 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
 107 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
 108
 109 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 110 # define journal_callback ext4_journal_cb_entry
 111 # define fsfilt_journal_callback_set(handle, func, jcb) \
 112          ext4_journal_callback_add(handle, func, jcb)
 113 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
 114 # define fsfilt_journal_callback_set(handle, func, jcb) \
 115          jbd2_journal_callback_set(handle, func, jcb)
 116 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
 117 # define fsfilt_journal_callback_set(handle, func, jcb) \
 118          journal_callback_set(handle, func, jcb)
 119 #else
 120 # error missing journal commit callback
 121 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
 122
 123 static cfs_mem_cache_t *fcb_cache;
 124
 125 struct fsfilt_cb_data {
 126         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
 127         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
 128         struct obd_device *cb_obd;      /* MDS/OBD completion device */
 129         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
 130         void *cb_data;                  /* MDS/OST completion function data */
 131 };
 132
 133 #ifndef ext3_find_next_bit
 134 #define ext3_find_next_bit           ext2_find_next_bit
 135 #endif
 136
 137 #ifndef ext2_find_next_bit
 138 #ifdef __LITTLE_ENDIAN
 139 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
 140 #else
 141 error "Need implementation of find_next_bit on big-endian systems"
 142 #endif  /* __LITTLE_ENDIAN */
 143 #endif  /* !ext2_find_next_le_bit */
 144
 145 static char *fsfilt_ext3_get_label(struct super_block *sb)
 146 {
 147         return EXT3_SB(sb)->s_es->s_volume_name;
 148 }
 149
 150 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
 151 {
 152         /* see e.g. fsfilt_ext3_write_record() */
 153         journal_t *journal;
 154         handle_t *handle;
 155         int err;
 156
 157         journal = EXT3_SB(sb)->s_journal;
 158         handle = ext3_journal_start_sb(sb, 1);
 159         if (IS_ERR(handle)) {
 160                 CERROR("can't start transaction\n");
 161                 return(PTR_ERR(handle));
 162         }
 163
 164         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 165         if (err)
 166                 goto out;
 167
 168         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
 169                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
 170
 171         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 172
 173 out:
 174         ext3_journal_stop(handle);
 175
 176         return(err);
 177 }
 178
 179 static char *fsfilt_ext3_uuid(struct super_block *sb)
 180 {
 181         return EXT3_SB(sb)->s_es->s_uuid;
 182 }
 183
 184 #ifdef HAVE_DISK_INODE_VERSION
 185
 186 static __u64 get_i_version(struct inode *inode)
 187 {
 188         return EXT3_I(inode)->i_fs_version;
 189 }
 190
 191 static void set_i_version(struct inode *inode, __u64 new_version)
 192 {
 193         (EXT3_I(inode))->i_fs_version = new_version;
 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 += EXT3_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                 /* no break */
 262         }
 263         case FSFILT_OP_MKDIR:
 264         case FSFILT_OP_MKNOD:
 265                 /* modify one inode + block bitmap + GDT */
 266                 nblocks += 3;
 267                 /* no break */
 268         case FSFILT_OP_LINK:
 269                 /* modify parent directory */
 270                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 271                            EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
 272                 /* create/update logs for each stripe */
 273                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 274                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 275                 break;
 276         case FSFILT_OP_SETATTR:
 277                 /* Setattr on inode */
 278                 nblocks += 1;
 279                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 280                            EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
 281                 /* quota chown log for each stripe */
 282                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
 283                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
 284                 break;
 285         case FSFILT_OP_CANCEL_UNLINK:
 286                 /* blocks for log header bitmap update OR
 287                  * blocks for catalog header bitmap update + unlink of logs */
 288                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
 289                         EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
 290                 break;
 291         default: CERROR("unknown transaction start op %d\n", op);
 292                 LBUG();
 293         }
 294
 295         LASSERT(current->journal_info == desc_private);
 296         journal = EXT3_SB(inode->i_sb)->s_journal;
 297         if (nblocks > journal->j_max_transaction_buffers) {
 298                 CWARN("too many credits %d for op %ux%u using %d instead\n",
 299                        nblocks, op, logs, journal->j_max_transaction_buffers);
 300                 nblocks = journal->j_max_transaction_buffers;
 301         }
 302
 303  journal_start:
 304         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
 305         handle = ext3_journal_start(inode, nblocks);
 306
 307         if (!IS_ERR(handle))
 308                 LASSERT(current->journal_info == handle);
 309         else
 310                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
 311                        op, nblocks, PTR_ERR(handle));
 312         return handle;
 313 }
 314
 315 /*
 316  * Calculate the number of buffer credits needed to write multiple pages in
 317  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
 318  * doesn't have a nice API for calculating this sort of thing in advance.
 319  *
 320  * See comment above ext3_writepage_trans_blocks for details.  We assume
 321  * no data journaling is being done, but it does allow for all of the pages
 322  * being non-contiguous.  If we are guaranteed contiguous pages we could
 323  * reduce the number of (d)indirect blocks a lot.
 324  *
 325  * With N blocks per page and P pages, for each inode we have at most:
 326  * N*P indirect
 327  * min(N*P, blocksize/4 + 1) dindirect blocks
 328  * niocount tindirect
 329  *
 330  * For the entire filesystem, we have at most:
 331  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 332  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 333  * objcount inode blocks
 334  * 1 superblock
 335  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 336  *
 337  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
 338  */
 339 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
 340                                       int niocount, struct niobuf_local *nb)
 341 {
 342         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
 343         __u64 next_indir;
 344         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
 345         int nbitmaps = 0, ngdblocks;
 346         int needed = objcount + 1; /* inodes + superblock */
 347         int i, j;
 348
 349         for (i = 0, j = 0; i < objcount; i++, fso++) {
 350                 /* two or more dindirect blocks in case we cross boundary */
 351                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
 352                                     nb[j].offset) >>
 353                                    sb->s_blocksize_bits) /
 354                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
 355                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
 356
 357                 /* leaf, indirect, tindirect blocks for first block */
 358                 nbitmaps += blockpp + 2;
 359
 360                 j += fso->fso_bufcnt;
 361         }
 362
 363         next_indir = nb[0].offset +
 364                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
 365         for (i = 1; i < niocount; i++) {
 366                 if (nb[i].offset >= next_indir) {
 367                         nbitmaps++;     /* additional indirect */
 368                         next_indir = nb[i].offset +
 369                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
 370                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
 371                         nbitmaps++;     /* additional indirect */
 372                 }
 373                 nbitmaps += blockpp;    /* each leaf in different group? */
 374         }
 375
 376         ngdblocks = nbitmaps;
 377         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 378                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 379         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 380                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 381
 382         needed += nbitmaps + ngdblocks;
 383
 384         /* last_rcvd update */
 385         needed += EXT3_DATA_TRANS_BLOCKS(sb);
 386
 387 #if defined(CONFIG_QUOTA)
 388         /* We assume that there will be 1 bit set in s_dquot.flags for each
 389          * quota file that is active.  This is at least true for now.
 390          */
 391         needed += hweight32(ll_sb_any_quota_active(sb)) *
 392                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
 393 #endif
 394
 395         return needed;
 396 }
 397
 398 /* We have to start a huge journal transaction here to hold all of the
 399  * metadata for the pages being written here.  This is necessitated by
 400  * the fact that we do lots of prepare_write operations before we do
 401  * any of the matching commit_write operations, so even if we split
 402  * up to use "smaller" transactions none of them could complete until
 403  * all of them were opened.  By having a single journal transaction,
 404  * we eliminate duplicate reservations for common blocks like the
 405  * superblock and group descriptors or bitmaps.
 406  *
 407  * We will start the transaction here, but each prepare_write will
 408  * add a refcount to the transaction, and each commit_write will
 409  * remove a refcount.  The transaction will be closed when all of
 410  * the pages have been written.
 411  */
 412 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
 413                                    int niocount, struct niobuf_local *nb,
 414                                    void *desc_private, int logs)
 415 {
 416         journal_t *journal;
 417         handle_t *handle;
 418         int needed;
 419         ENTRY;
 420
 421         LASSERT(current->journal_info == desc_private);
 422         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
 423         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
 424
 425         /* The number of blocks we could _possibly_ dirty can very large.
 426          * We reduce our request if it is absurd (and we couldn't get that
 427          * many credits for a single handle anyways).
 428          *
 429          * At some point we have to limit the size of I/Os sent at one time,
 430          * increase the size of the journal, or we have to calculate the
 431          * actual journal requirements more carefully by checking all of
 432          * the blocks instead of being maximally pessimistic.  It remains to
 433          * be seen if this is a real problem or not.
 434          */
 435         if (needed > journal->j_max_transaction_buffers) {
 436                 CERROR("want too many journal credits (%d) using %d instead\n",
 437                        needed, journal->j_max_transaction_buffers);
 438                 needed = journal->j_max_transaction_buffers;
 439         }
 440
 441         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
 442         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
 443         if (IS_ERR(handle)) {
 444                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
 445                        PTR_ERR(handle));
 446         } else {
 447                 LASSERT(handle->h_buffer_credits >= needed);
 448                 LASSERT(current->journal_info == handle);
 449         }
 450
 451         RETURN(handle);
 452 }
 453
 454 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
 455 {
 456        handle_t *handle = h;
 457
 458        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
 459        if (nblocks == 0) {
 460                handle->h_buffer_credits = 0;
 461                CWARN("setting credits of handle %p to zero by request\n", h);
 462        }
 463
 464        if (handle->h_buffer_credits > nblocks)
 465                 return 0;
 466        if (ext3_journal_extend(handle, nblocks) == 0)
 467                 return 0;
 468
 469        ext3_mark_inode_dirty(handle, inode);
 470        return ext3_journal_restart(handle, nblocks);
 471 }
 472
 473 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
 474 {
 475         int rc;
 476         handle_t *handle = h;
 477
 478         LASSERT(current->journal_info == handle);
 479         if (force_sync)
 480                 handle->h_sync = 1; /* recovery likes this */
 481
 482         rc = ext3_journal_stop(handle);
 483
 484         return rc;
 485 }
 486
 487 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
 488                                     void **wait_handle)
 489 {
 490         unsigned long tid;
 491         transaction_t *transaction;
 492         handle_t *handle = h;
 493         journal_t *journal;
 494         int rc;
 495
 496         LASSERT(current->journal_info == handle);
 497
 498         transaction = handle->h_transaction;
 499         journal = transaction->t_journal;
 500         tid = transaction->t_tid;
 501         /* we don't want to be blocked */
 502         handle->h_sync = 0;
 503         rc = ext3_journal_stop(handle);
 504         if (rc) {
 505                 CERROR("error while stopping transaction: %d\n", rc);
 506                 return rc;
 507         }
 508         fsfilt_log_start_commit(journal, tid);
 509
 510         *wait_handle = (void *) tid;
 511         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
 512         return 0;
 513 }
 514
 515 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
 516 {
 517         journal_t *journal = EXT3_JOURNAL(inode);
 518         tid_t tid = (tid_t)(long)h;
 519
 520         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
 521         if (unlikely(is_journal_aborted(journal)))
 522                 return -EIO;
 523
 524         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
 525
 526         if (unlikely(is_journal_aborted(journal)))
 527                 return -EIO;
 528         return 0;
 529 }
 530
 531 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
 532                                struct iattr *iattr, int do_trunc)
 533 {
 534         struct inode *inode = dentry->d_inode;
 535         int rc = 0;
 536
 537 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
 538         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
 539          * timestamps to appear to be in the far future, due old timestamp
 540          * being stored on disk as an unsigned value.  This fixes up any
 541          * bad values held by the client before storing them on disk,
 542          * and ensures any timestamp updates are correct.  LU-1042 */
 543         if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
 544                      !(iattr->ia_valid & ATTR_ATIME))) {
 545                 iattr->ia_valid |= ATTR_ATIME;
 546                 LTIME_S(iattr->ia_atime) = 0;
 547         }
 548         if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
 549                      !(iattr->ia_valid & ATTR_MTIME))) {
 550                 iattr->ia_valid |= ATTR_MTIME;
 551                 LTIME_S(iattr->ia_mtime) = 0;
 552         }
 553         if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
 554                       LTIME_S(inode->i_ctime) == 0) &&
 555                      !(iattr->ia_valid & ATTR_CTIME))) {
 556                 iattr->ia_valid |= ATTR_CTIME;
 557                 LTIME_S(iattr->ia_ctime) = 0;
 558         }
 559 #else
 560 #warning "remove old LU-221/LU-1042 workaround code"
 561 #endif
 562
 563         /* When initializating timestamps for new inodes, use the filesystem
 564          * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
 565          * that this is potentially an invalid inode.  Files with an old ctime
 566          * migrated to a newly-formatted OST with a newer s_mkfs_time will not
 567          * hit this check, since it is only for ctime == 0.  LU-1010/LU-1042 */
 568         if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
 569                 LTIME_S(iattr->ia_ctime) =
 570                         EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
 571
 572         /* Avoid marking the inode dirty on the superblock list unnecessarily.
 573          * We are already writing the inode to disk as part of this
 574          * transaction and want to avoid a lot of extra inode writeout
 575          * later on. b=9828 */
 576         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
 577                 /* ATTR_SIZE would invoke truncate: clear it */
 578                 iattr->ia_valid &= ~ATTR_SIZE;
 579                 EXT3_I(inode)->i_disksize = iattr->ia_size;
 580                 i_size_write(inode, iattr->ia_size);
 581
 582                 if (iattr->ia_valid & ATTR_UID)
 583                         inode->i_uid = iattr->ia_uid;
 584                 if (iattr->ia_valid & ATTR_GID)
 585                         inode->i_gid = iattr->ia_gid;
 586                 if (iattr->ia_valid & ATTR_ATIME)
 587                         inode->i_atime = iattr->ia_atime;
 588                 if (iattr->ia_valid & ATTR_MTIME)
 589                         inode->i_mtime = iattr->ia_mtime;
 590                 if (iattr->ia_valid & ATTR_CTIME)
 591                         inode->i_ctime = iattr->ia_ctime;
 592                 if (iattr->ia_valid & ATTR_MODE) {
 593                         inode->i_mode = iattr->ia_mode;
 594
 595                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
 596                             !cfs_capable(CFS_CAP_FSETID))
 597                                 inode->i_mode &= ~S_ISGID;
 598                 }
 599
 600                 inode->i_sb->s_op->dirty_inode(inode);
 601
 602                 goto out;
 603         }
 604
 605         /* Don't allow setattr to change file type */
 606         if (iattr->ia_valid & ATTR_MODE)
 607                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
 608                                  (iattr->ia_mode & ~S_IFMT);
 609
 610         /* We set these flags on the client, but have already checked perms
 611          * so don't confuse inode_change_ok. */
 612         iattr->ia_valid &= ~TIMES_SET_FLAGS;
 613
 614         if (inode->i_op->setattr) {
 615                 rc = inode->i_op->setattr(dentry, iattr);
 616         } else {
 617 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
 618                 rc = inode_change_ok(inode, iattr);
 619                 if (!rc)
 620 #endif
 621                         rc = simple_setattr(dentry, iattr);
 622         }
 623
 624  out:
 625         RETURN(rc);
 626 }
 627
 628 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
 629                                  unsigned int cmd, unsigned long arg)
 630 {
 631         int rc = 0;
 632         ENTRY;
 633
 634         /* FIXME: Can't do this because of nested transaction deadlock */
 635         if (cmd == EXT3_IOC_SETFLAGS) {
 636                 /* We can't enable data journaling on OST objects, because
 637                 * this forces the transaction to be closed in order to
 638                 * flush the journal, but the caller will already have a
 639                 * compound transaction open to update the last_rcvd file,
 640                 * and this thread would deadlock trying to set the flag. */
 641                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
 642                         CERROR("can't set data journal flag on file\n");
 643                         RETURN(-EPERM);
 644                 }
 645                 /* Because the MDS does not see the EXTENTS_FL set on the
 646                  * OST objects, mask this flag into all set flags.  It is
 647                  * not legal to clear this flag in any case, so we are not
 648                  * changing the functionality by doing this.  b=22911 */
 649                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
 650         }
 651
 652         /* ext4_ioctl does not have a inode argument */
 653         if (inode->i_fop->unlocked_ioctl)
 654                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
 655         else
 656                 RETURN(-ENOTTY);
 657
 658         RETURN(rc);
 659 }
 660
 661 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
 662                               void *lmm, int lmm_size, const char *name)
 663 {
 664         int rc;
 665
 666         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 667
 668         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
 669                                    name, lmm, lmm_size, XATTR_NO_CTIME);
 670
 671         if (rc && rc != -EROFS)
 672                 CERROR("error adding MD data to inode %lu: rc = %d\n",
 673                        inode->i_ino, rc);
 674         return rc;
 675 }
 676
 677 /* Must be called with i_mutex held */
 678 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
 679                               const char *name)
 680 {
 681         int rc;
 682
 683         LASSERT(mutex_trylock(&inode->i_mutex) == 0);
 684
 685         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
 686                             name, lmm, lmm_size);
 687
 688         /* This gives us the MD size */
 689         if (lmm == NULL)
 690                 return (rc == -ENODATA) ? 0 : rc;
 691
 692         if (rc < 0) {
 693                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
 694                        EXT3_XATTR_INDEX_TRUSTED, name,
 695                        inode->i_ino, rc);
 696                 memset(lmm, 0, lmm_size);
 697                 return (rc == -ENODATA) ? 0 : rc;
 698         }
 699
 700         return rc;
 701 }
 702
 703 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
 704 {
 705         submit_bio(rw, bio);
 706         return 0;
 707 }
 708
 709 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
 710                                     loff_t *off)
 711 {
 712         struct inode *inode = file->f_dentry->d_inode;
 713         int rc = 0;
 714
 715         if (S_ISREG(inode->i_mode))
 716                 rc = file->f_op->read(file, buf, count, off);
 717         else {
 718                 const int blkbits = inode->i_sb->s_blocksize_bits;
 719                 const int blksize = inode->i_sb->s_blocksize;
 720
 721                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
 722                        (unsigned long)count, inode->i_ino, *off);
 723                 while (count > 0) {
 724                         struct buffer_head *bh;
 725
 726                         bh = NULL;
 727                         if (*off < i_size_read(inode)) {
 728                                 int err = 0;
 729
 730                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
 731                                                 0, &err);
 732
 733                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
 734
 735                                 if (bh) {
 736                                         memcpy(buf, bh->b_data, blksize);
 737                                         brelse(bh);
 738                                 } else if (err) {
 739                                         /* XXX in theory we should just fake
 740                                          * this buffer and continue like ext3,
 741                                          * especially if this is a partial read
 742                                          */
 743                                         CERROR("error read dir %lu+%llu: %d\n",
 744                                                inode->i_ino, *off, err);
 745                                         RETURN(err);
 746                                 }
 747                         }
 748                         if (!bh) {
 749                                 struct ext3_dir_entry_2 *fake = (void *)buf;
 750
 751                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
 752                                 memset(fake, 0, sizeof(*fake));
 753                                 fake->rec_len = cpu_to_le16(blksize);
 754                         }
 755                         count -= blksize;
 756                         buf += blksize;
 757                         *off += blksize;
 758                         rc += blksize;
 759                 }
 760         }
 761
 762         return rc;
 763 }
 764
 765 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
 766 static void fsfilt_ext3_cb_func(struct super_block *sb,
 767                                 struct journal_callback *jcb, int error)
 768 #else
 769 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
 770 #endif
 771 {
 772         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
 773
 774         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
 775
 776         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
 777 }
 778
 779 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
 780                                       void *handle, fsfilt_cb_t cb_func,
 781                                       void *cb_data)
 782 {
 783         struct fsfilt_cb_data *fcb;
 784
 785         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
 786         if (fcb == NULL)
 787                 RETURN(-ENOMEM);
 788
 789         fcb->cb_func = cb_func;
 790         fcb->cb_obd = obd;
 791         fcb->cb_last_rcvd = last_rcvd;
 792         fcb->cb_data = cb_data;
 793
 794         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
 795         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
 796
 797         return 0;
 798 }
 799
 800 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
 801 {
 802         struct kstatfs sfs;
 803         int rc;
 804
 805         memset(&sfs, 0, sizeof(sfs));
 806         rc = sb->s_op->statfs(sb->s_root, &sfs);
 807         statfs_pack(osfs, &sfs);
 808         return rc;
 809 }
 810
 811 static int fsfilt_ext3_sync(struct super_block *sb)
 812 {
 813         return ext3_force_commit(sb);
 814 }
 815
 816 #ifndef EXT3_EXTENTS_FL
 817 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
 818 #endif
 819
 820 #ifndef EXT_ASSERT
 821 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
 822 #endif
 823
 824 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
 825 #define ext3_ext_base                   inode
 826 #define ext3_ext_base2inode(inode)      (inode)
 827 #define EXT_DEPTH(inode)                ext_depth(inode)
 828 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
 829                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
 830
 831 struct bpointers {
 832         unsigned long *blocks;
 833         int *created;
 834         unsigned long start;
 835         int num;
 836         int init_num;
 837         int create;
 838 };
 839
 840 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
 841                                unsigned long block, int *aflags)
 842 {
 843         struct ext3_inode_info *ei = EXT3_I(inode);
 844         unsigned long bg_start;
 845         unsigned long colour;
 846         int depth;
 847
 848         if (path) {
 849                 struct ext3_extent *ex;
 850                 depth = path->p_depth;
 851
 852                 /* try to predict block placement */
 853                 if ((ex = path[depth].p_ext))
 854                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
 855
 856                 /* it looks index is empty
 857                  * try to find starting from index itself */
 858                 if (path[depth].p_bh)
 859                         return path[depth].p_bh->b_blocknr;
 860         }
 861
 862         /* OK. use inode's group */
 863         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 864                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 865         colour = (current->pid % 16) *
 866                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 867         return bg_start + colour + block;
 868 }
 869
 870 #define ll_unmap_underlying_metadata(sb, blocknr) \
 871         unmap_underlying_metadata((sb)->s_bdev, blocknr)
 872
 873 #ifndef EXT3_MB_HINT_GROUP_ALLOC
 874 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 875                                 struct ext3_ext_path *path, unsigned long block,
 876                                 unsigned long *count, int *err)
 877 {
 878         unsigned long pblock, goal;
 879         int aflags = 0;
 880         struct inode *inode = ext3_ext_base2inode(base);
 881
 882         goal = ext3_ext_find_goal(inode, path, block, &aflags);
 883         aflags |= 2; /* block have been already reserved */
 884         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
 885         return pblock;
 886
 887 }
 888 #else
 889 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
 890                                 struct ext3_ext_path *path, unsigned long block,
 891                                 unsigned long *count, int *err)
 892 {
 893         struct inode *inode = ext3_ext_base2inode(base);
 894         struct ext3_allocation_request ar;
 895         unsigned long pblock;
 896         int aflags;
 897
 898         /* find neighbour allocated blocks */
 899         ar.lleft = block;
 900         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
 901         if (*err)
 902                 return 0;
 903         ar.lright = block;
 904         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
 905         if (*err)
 906                 return 0;
 907
 908         /* allocate new block */
 909         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
 910         ar.inode = inode;
 911         ar.logical = block;
 912         ar.len = *count;
 913         ar.flags = EXT3_MB_HINT_DATA;
 914         pblock = ext3_mb_new_blocks(handle, &ar, err);
 915         *count = ar.len;
 916         return pblock;
 917 }
 918 #endif
 919
 920 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
 921                                   struct ext3_ext_path *path,
 922                                   struct ext3_ext_cache *cex,
 923 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
 924                                    struct ext3_extent *ex,
 925 #endif
 926                                   void *cbdata)
 927 {
 928         struct bpointers *bp = cbdata;
 929         struct inode *inode = ext3_ext_base2inode(base);
 930         struct ext3_extent nex;
 931         unsigned long pblock;
 932         unsigned long tgen;
 933         int err, i;
 934         unsigned long count;
 935         handle_t *handle;
 936
 937         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
 938                 err = EXT_CONTINUE;
 939                 goto map;
 940         }
 941
 942         if (bp->create == 0) {
 943                 i = 0;
 944                 if (cex->ec_block < bp->start)
 945                         i = bp->start - cex->ec_block;
 946                 if (i >= cex->ec_len)
 947                         CERROR("nothing to do?! i = %d, e_num = %u\n",
 948                                         i, cex->ec_len);
 949                 for (; i < cex->ec_len && bp->num; i++) {
 950                         *(bp->created) = 0;
 951                         bp->created++;
 952                         *(bp->blocks) = 0;
 953                         bp->blocks++;
 954                         bp->num--;
 955                         bp->start++;
 956                 }
 957
 958                 return EXT_CONTINUE;
 959         }
 960
 961         tgen = EXT_GENERATION(base);
 962         count = ext3_ext_calc_credits_for_insert(base, path);
 963
 964         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
 965         if (IS_ERR(handle)) {
 966                 return PTR_ERR(handle);
 967         }
 968
 969         if (tgen != EXT_GENERATION(base)) {
 970                 /* the tree has changed. so path can be invalid at moment */
 971                 ext3_journal_stop(handle);
 972                 return EXT_REPEAT;
 973         }
 974
 975         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
 976          * protected by i_data_sem as whole. so we patch it to store
 977          * generation to path and now verify the tree hasn't changed */
 978         down_write((&EXT4_I(inode)->i_data_sem));
 979
 980         /* validate extent, make sure the extent tree does not changed */
 981         if (EXT_GENERATION(base) != path[0].p_generation) {
 982                 /* cex is invalid, try again */
 983                 up_write(&EXT4_I(inode)->i_data_sem);
 984                 ext3_journal_stop(handle);
 985                 return EXT_REPEAT;
 986         }
 987
 988         count = cex->ec_len;
 989         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
 990         if (!pblock)
 991                 goto out;
 992         EXT_ASSERT(count <= cex->ec_len);
 993
 994         /* insert new extent */
 995         nex.ee_block = cpu_to_le32(cex->ec_block);
 996         ext3_ext_store_pblock(&nex, pblock);
 997         nex.ee_len = cpu_to_le16(count);
 998         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
 999         if (err) {
1000                 /* free data blocks we just allocated */
1001                 /* not a good idea to call discard here directly,
1002                  * but otherwise we'd need to call it every free() */
1003 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1004                 ext3_mb_discard_inode_preallocations(inode);
1005 #endif
1006                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1007                                  cpu_to_le16(nex.ee_len), 0);
1008                 goto out;
1009         }
1010
1011         /*
1012          * Putting len of the actual extent we just inserted,
1013          * we are asking ext3_ext_walk_space() to continue
1014          * scaning after that block
1015          */
1016         cex->ec_len = le16_to_cpu(nex.ee_len);
1017         cex->ec_start = ext_pblock(&nex);
1018         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1019         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1020
1021 out:
1022         up_write((&EXT4_I(inode)->i_data_sem));
1023         ext3_journal_stop(handle);
1024 map:
1025         if (err >= 0) {
1026                 /* map blocks */
1027                 if (bp->num == 0) {
1028                         CERROR("hmm. why do we find this extent?\n");
1029                         CERROR("initial space: %lu:%u\n",
1030                                 bp->start, bp->init_num);
1031                         CERROR("current extent: %u/%u/%llu %d\n",
1032                                 cex->ec_block, cex->ec_len,
1033                                 (unsigned long long)cex->ec_start,
1034                                 cex->ec_type);
1035                 }
1036                 i = 0;
1037                 if (cex->ec_block < bp->start)
1038                         i = bp->start - cex->ec_block;
1039                 if (i >= cex->ec_len)
1040                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1041                                         i, cex->ec_len);
1042                 for (; i < cex->ec_len && bp->num; i++) {
1043                         *(bp->blocks) = cex->ec_start + i;
1044                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1045                                 *(bp->created) = 0;
1046                         } else {
1047                                 *(bp->created) = 1;
1048                                 /* unmap any possible underlying metadata from
1049                                  * the block device mapping.  bug 6998. */
1050                                 ll_unmap_underlying_metadata(inode->i_sb,
1051                                                              *(bp->blocks));
1052                         }
1053                         bp->created++;
1054                         bp->blocks++;
1055                         bp->num--;
1056                         bp->start++;
1057                 }
1058         }
1059         return err;
1060 }
1061
1062 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1063                        unsigned long num, unsigned long *blocks,
1064                        int *created, int create)
1065 {
1066         struct ext3_ext_base *base = inode;
1067         struct bpointers bp;
1068         int err;
1069
1070         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1071                block, block + num - 1, (unsigned) inode->i_ino);
1072
1073         bp.blocks = blocks;
1074         bp.created = created;
1075         bp.start = block;
1076         bp.init_num = bp.num = num;
1077         bp.create = create;
1078
1079         err = fsfilt_ext3_ext_walk_space(base, block, num,
1080                                          ext3_ext_new_extent_cb, &bp);
1081         ext3_ext_invalidate_cache(base);
1082
1083         return err;
1084 }
1085
1086 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1087                                     int pages, unsigned long *blocks,
1088                                     int *created, int create)
1089 {
1090         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1091         int rc = 0, i = 0;
1092         struct page *fp = NULL;
1093         int clen = 0;
1094
1095         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1096                 inode->i_ino, pages, (*page)->index);
1097
1098         /* pages are sorted already. so, we just have to find
1099          * contig. space and process them properly */
1100         while (i < pages) {
1101                 if (fp == NULL) {
1102                         /* start new extent */
1103                         fp = *page++;
1104                         clen = 1;
1105                         i++;
1106                         continue;
1107                 } else if (fp->index + clen == (*page)->index) {
1108                         /* continue the extent */
1109                         page++;
1110                         clen++;
1111                         i++;
1112                         continue;
1113                 }
1114
1115                 /* process found extent */
1116                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1117                                         clen * blocks_per_page, blocks,
1118                                         created, create);
1119                 if (rc)
1120                         GOTO(cleanup, rc);
1121
1122                 /* look for next extent */
1123                 fp = NULL;
1124                 blocks += blocks_per_page * clen;
1125                 created += blocks_per_page * clen;
1126         }
1127
1128         if (fp)
1129                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1130                                         clen * blocks_per_page, blocks,
1131                                         created, create);
1132 cleanup:
1133         return rc;
1134 }
1135
1136 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1137                                unsigned long *blocks, int *created, int create);
1138 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1139                                    int pages, unsigned long *blocks,
1140                                    int *created, int create)
1141 {
1142         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1143         unsigned long *b;
1144         int rc = 0, i, *cr;
1145
1146         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1147                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1148                 if (rc) {
1149                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1150                                inode->i_ino, *b, *cr, create, rc);
1151                         break;
1152                 }
1153
1154                 b += blocks_per_page;
1155                 cr += blocks_per_page;
1156         }
1157         return rc;
1158 }
1159
1160 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1161                                 int pages, unsigned long *blocks,
1162                                 int *created, int create,
1163                                 cfs_mutex_t *optional_mutex)
1164 {
1165         int rc;
1166
1167         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1168                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1169                                                      blocks, created, create);
1170                 return rc;
1171         }
1172         if (optional_mutex != NULL)
1173                 cfs_mutex_lock(optional_mutex);
1174         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1175                                             created, create);
1176         if (optional_mutex != NULL)
1177                 cfs_mutex_unlock(optional_mutex);
1178
1179         return rc;
1180 }
1181
1182 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1183 {
1184         unsigned long block;
1185         struct buffer_head *bh;
1186         int err, blocksize, csize, boffs, osize = size;
1187
1188         /* prevent reading after eof */
1189         spin_lock(&inode->i_lock);
1190         if (i_size_read(inode) < *offs + size) {
1191                 size = i_size_read(inode) - *offs;
1192                 spin_unlock(&inode->i_lock);
1193                 if (size < 0) {
1194                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1195                                i_size_read(inode), *offs);
1196                         return -EBADR;
1197                 } else if (size == 0) {
1198                         return 0;
1199                 }
1200         } else {
1201                 spin_unlock(&inode->i_lock);
1202         }
1203
1204         blocksize = 1 << inode->i_blkbits;
1205
1206         while (size > 0) {
1207                 block = *offs >> inode->i_blkbits;
1208                 boffs = *offs & (blocksize - 1);
1209                 csize = min(blocksize - boffs, size);
1210                 bh = ext3_bread(NULL, inode, block, 0, &err);
1211                 if (!bh) {
1212                         CERROR("can't read block: %d\n", err);
1213                         return err;
1214                 }
1215
1216                 memcpy(buf, bh->b_data + boffs, csize);
1217                 brelse(bh);
1218
1219                 *offs += csize;
1220                 buf += csize;
1221                 size -= csize;
1222         }
1223         return osize;
1224 }
1225 EXPORT_SYMBOL(fsfilt_ext3_read);
1226
1227 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1228                                    int size, loff_t *offs)
1229 {
1230         int rc;
1231         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1232         if (rc > 0)
1233                 rc = 0;
1234         return rc;
1235 }
1236
1237 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1238                                 loff_t *offs, handle_t *handle)
1239 {
1240         struct buffer_head *bh = NULL;
1241         loff_t old_size = i_size_read(inode), offset = *offs;
1242         loff_t new_size = i_size_read(inode);
1243         unsigned long block;
1244         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1245
1246         while (bufsize > 0) {
1247                 if (bh != NULL)
1248                         brelse(bh);
1249
1250                 block = offset >> inode->i_blkbits;
1251                 boffs = offset & (blocksize - 1);
1252                 size = min(blocksize - boffs, bufsize);
1253                 bh = ext3_bread(handle, inode, block, 1, &err);
1254                 if (!bh) {
1255                         CERROR("can't read/create block: %d\n", err);
1256                         break;
1257                 }
1258
1259                 err = ext3_journal_get_write_access(handle, bh);
1260                 if (err) {
1261                         CERROR("journal_get_write_access() returned error %d\n",
1262                                err);
1263                         break;
1264                 }
1265                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1266                 memcpy(bh->b_data + boffs, buf, size);
1267                 err = ext3_journal_dirty_metadata(handle, bh);
1268                 if (err) {
1269                         CERROR("journal_dirty_metadata() returned error %d\n",
1270                                err);
1271                         break;
1272                 }
1273                 if (offset + size > new_size)
1274                         new_size = offset + size;
1275                 offset += size;
1276                 bufsize -= size;
1277                 buf += size;
1278         }
1279         if (bh)
1280                 brelse(bh);
1281
1282         /* correct in-core and on-disk sizes */
1283         if (new_size > i_size_read(inode)) {
1284                 spin_lock(&inode->i_lock);
1285                 if (new_size > i_size_read(inode))
1286                         i_size_write(inode, new_size);
1287                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1288                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1289                 if (i_size_read(inode) > old_size) {
1290                         spin_unlock(&inode->i_lock);
1291                         mark_inode_dirty(inode);
1292                 } else {
1293                         spin_unlock(&inode->i_lock);
1294                 }
1295         }
1296
1297         if (err == 0)
1298                 *offs = offset;
1299         return err;
1300 }
1301 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1302
1303 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1304                                     loff_t *offs, int force_sync)
1305 {
1306         struct inode *inode = file->f_dentry->d_inode;
1307         handle_t *handle;
1308         int err, block_count = 0, blocksize;
1309
1310         /* Determine how many transaction credits are needed */
1311         blocksize = 1 << inode->i_blkbits;
1312         block_count = (*offs & (blocksize - 1)) + bufsize;
1313         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1314
1315         handle = ext3_journal_start(inode,
1316                         block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1317         if (IS_ERR(handle)) {
1318                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1319                        block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
1320                        bufsize);
1321                 return PTR_ERR(handle);
1322         }
1323
1324         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1325
1326         if (!err && force_sync)
1327                 handle->h_sync = 1; /* recovery likes this */
1328
1329         ext3_journal_stop(handle);
1330
1331         return err;
1332 }
1333
1334 static int fsfilt_ext3_setup(struct super_block *sb)
1335 {
1336 #if defined(HAVE_QUOTA_SUPPORT) || defined(S_PDIROPS)
1337         struct ext3_sb_info *sbi = EXT3_SB(sb);
1338 #if 0
1339         sbi->dx_lock = fsfilt_ext3_dx_lock;
1340         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1341 #endif
1342 #endif
1343         if (!EXT3_HAS_COMPAT_FEATURE(sb,
1344                                 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1345                 CERROR("ext3 mounted without journal\n");
1346                 return -EINVAL;
1347         }
1348
1349 #ifdef S_PDIROPS
1350         CWARN("Enabling PDIROPS\n");
1351         set_opt(sbi->s_mount_opt, PDIROPS);
1352         sb->s_flags |= S_PDIROPS;
1353 #endif
1354         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1355                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1356 #ifdef HAVE_QUOTA_SUPPORT
1357         /* enable journaled quota support */
1358         /* kfreed in ext3_put_super() */
1359         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1360         if (!sbi->s_qf_names[USRQUOTA])
1361                 return -ENOMEM;
1362         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1363         if (!sbi->s_qf_names[GRPQUOTA]) {
1364                 kfree(sbi->s_qf_names[USRQUOTA]);
1365                 sbi->s_qf_names[USRQUOTA] = NULL;
1366                 return -ENOMEM;
1367         }
1368         sbi->s_jquota_fmt = QFMT_LUSTRE;
1369         set_opt(sbi->s_mount_opt, QUOTA);
1370 #endif
1371         return 0;
1372 }
1373
1374 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1375    objects. Logs is number of logfiles to update */
1376 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1377 {
1378         if ( !fso ) {
1379                 switch(op) {
1380                 case FSFILT_OP_CREATE:
1381                                  /* directory leaf, index & indirect & EA*/
1382                         return 4 + 3 * logs;
1383                 case FSFILT_OP_UNLINK:
1384                         return 3 * logs;
1385                 }
1386         } else {
1387                 int i;
1388                 int needed = 0;
1389                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1390                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1391                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1392                 for (i = 0; i < op; i++, fso++) {
1393                         int nblocks = fso->fso_bufcnt * blockpp;
1394                         int ndindirect = min(nblocks, addrpp + 1);
1395                         int nindir = nblocks + ndindirect + 1;
1396
1397                         needed += nindir;
1398                 }
1399                 return needed + 3 * logs;
1400         }
1401
1402         return 0;
1403 }
1404
1405 #ifdef HAVE_QUOTA_SUPPORT
1406 #define DQINFO_COPY(out, in)                    \
1407 do {                                            \
1408         Q_COPY(out, in, dqi_bgrace);            \
1409         Q_COPY(out, in, dqi_igrace);            \
1410         Q_COPY(out, in, dqi_flags);             \
1411         Q_COPY(out, in, dqi_valid);             \
1412 } while (0)
1413
1414 #define DQBLK_COPY(out, in)                     \
1415 do {                                            \
1416         Q_COPY(out, in, dqb_bhardlimit);        \
1417         Q_COPY(out, in, dqb_bsoftlimit);        \
1418         Q_COPY(out, in, dqb_curspace);          \
1419         Q_COPY(out, in, dqb_ihardlimit);        \
1420         Q_COPY(out, in, dqb_isoftlimit);        \
1421         Q_COPY(out, in, dqb_curinodes);         \
1422         Q_COPY(out, in, dqb_btime);             \
1423         Q_COPY(out, in, dqb_itime);             \
1424         Q_COPY(out, in, dqb_valid);             \
1425 } while (0)
1426
1427 static int fsfilt_ext3_quotactl(struct super_block *sb,
1428                                 struct obd_quotactl *oqc)
1429 {
1430         int i, rc = 0, error = 0;
1431         const struct quotactl_ops *qcop;
1432         struct if_dqinfo *info;
1433         struct if_dqblk *dqblk;
1434         ENTRY;
1435
1436         if (!sb->s_qcop)
1437                 RETURN(-ENOSYS);
1438
1439         OBD_ALLOC_PTR(info);
1440         if (!info)
1441                 RETURN(-ENOMEM);
1442         OBD_ALLOC_PTR(dqblk);
1443         if (!dqblk) {
1444                 OBD_FREE_PTR(info);
1445                 RETURN(-ENOMEM);
1446         }
1447
1448         DQINFO_COPY(info, &oqc->qc_dqinfo);
1449         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1450
1451         qcop = sb->s_qcop;
1452         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1453                 for (i = 0; i < MAXQUOTAS; i++) {
1454                         if (!Q_TYPESET(oqc, i))
1455                                 continue;
1456
1457                         if (oqc->qc_cmd == Q_QUOTAON) {
1458                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1459
1460                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1461
1462                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1463                                                  name[i], 0);
1464                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1465                                 rc = ll_quota_off(sb, i, 0);
1466                         }
1467
1468                         if (rc == -EBUSY)
1469                                 error = rc;
1470                         else if (rc)
1471                                 GOTO(out, rc);
1472                 }
1473                 GOTO(out, rc ?: error);
1474         }
1475
1476         switch (oqc->qc_cmd) {
1477         case Q_GETOINFO:
1478         case Q_GETINFO:
1479                 if (!qcop->get_info)
1480                         GOTO(out, rc = -ENOSYS);
1481                 rc = qcop->get_info(sb, oqc->qc_type, info);
1482                 break;
1483         case Q_SETQUOTA:
1484         case Q_INITQUOTA:
1485                 if (!qcop->set_dqblk)
1486                         GOTO(out, rc = -ENOSYS);
1487                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1488                 break;
1489         case Q_GETOQUOTA:
1490         case Q_GETQUOTA:
1491                 if (!qcop->get_dqblk)
1492                         GOTO(out, rc = -ENOSYS);
1493                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1494                 if (!rc)
1495                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1496                 break;
1497         case Q_SYNC:
1498                 if (!sb->s_qcop->quota_sync)
1499                         GOTO(out, rc = -ENOSYS);
1500                 qcop->quota_sync(sb, oqc->qc_type);
1501                 break;
1502         case Q_FINVALIDATE:
1503                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1504                 for (i = 0; i < MAXQUOTAS; i++) {
1505                         struct file *fp;
1506                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1507
1508                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1509
1510                         if (!Q_TYPESET(oqc, i))
1511                                 continue;
1512
1513                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1514                         if (IS_ERR(fp)) {
1515                                 rc = PTR_ERR(fp);
1516                                 CERROR("error invalidating operational quota file"
1517                                        " %s (rc:%d)\n", name[i], rc);
1518                         } else {
1519                                 filp_close(fp, 0);
1520                         }
1521
1522                 }
1523                 break;
1524         default:
1525                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1526                 LBUG();
1527         }
1528 out:
1529         DQINFO_COPY(&oqc->qc_dqinfo, info);
1530         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1531
1532         OBD_FREE_PTR(info);
1533         OBD_FREE_PTR(dqblk);
1534
1535         if (rc)
1536                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1537                                 "failed: %d\n",
1538                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1539         RETURN(rc);
1540 }
1541
1542 struct chk_dqblk{
1543         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1544         cfs_list_t              dqb_list;        /** in list also */
1545         qid_t                   dqb_id;          /** uid/gid */
1546         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1547         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1548         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1549         qsize_t                 dqb_curspace;    /** current space */
1550         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1551         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1552         qsize_t                 dqb_curinodes;   /** current inodes */
1553         obd_time                dqb_btime;       /** block grace time */
1554         obd_time                dqb_itime;       /** inode grace time */
1555         __u32                   dqb_valid;       /** flag for above fields */
1556 };
1557
1558 static inline unsigned int chkquot_hash(qid_t id, int type)
1559                                         __attribute__((__const__));
1560
1561 static inline unsigned int chkquot_hash(qid_t id, int type)
1562 {
1563         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1564 }
1565
1566 static inline struct chk_dqblk *
1567 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1568 {
1569         cfs_hlist_node_t *node;
1570         struct chk_dqblk *cdqb;
1571
1572         cfs_hlist_for_each(node, head) {
1573                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1574                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1575                         return cdqb;
1576         }
1577
1578         return NULL;
1579 }
1580
1581 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1582 {
1583         struct chk_dqblk *cdqb;
1584
1585         OBD_ALLOC_PTR(cdqb);
1586         if (cdqb) {
1587                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1588                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1589                 cdqb->dqb_id = id;
1590                 cdqb->dqb_type = type;
1591         }
1592
1593         return cdqb;
1594 }
1595
1596 static struct chk_dqblk *
1597 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1598       cfs_list_t *list, qid_t id, int type, int first_check)
1599 {
1600         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1601         struct if_dqblk dqb;
1602         struct chk_dqblk *cdqb;
1603         int rc;
1604
1605         cdqb = find_chkquot(head, id, type);
1606         if (cdqb)
1607                 return cdqb;
1608
1609         cdqb = alloc_chkquot(id, type);
1610         if (!cdqb)
1611                 return NULL;
1612
1613         if (!first_check) {
1614                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1615                 if (rc) {
1616                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1617                                id, type, rc);
1618                 } else {
1619                         DQBLK_COPY(cdqb, &dqb);
1620                         cdqb->dqb_curspace = 0;
1621                         cdqb->dqb_curinodes = 0;
1622                 }
1623         }
1624
1625         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1626         cfs_list_add_tail(&cdqb->dqb_list, list);
1627
1628         return cdqb;
1629 }
1630
1631 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1632 {
1633         struct obd_quotactl *oqctl;
1634         int rc;
1635
1636         OBD_ALLOC_PTR(oqctl);
1637         if (!oqctl)
1638                 RETURN(-ENOMEM);
1639
1640         oqctl->qc_cmd = cmd;
1641         oqctl->qc_id = qfmt;
1642         oqctl->qc_type = type;
1643         rc = fsfilt_ext3_quotactl(sb, oqctl);
1644
1645         OBD_FREE_PTR(oqctl);
1646         return rc;
1647 }
1648
1649 static inline int read_old_dqinfo(struct super_block *sb, int type,
1650                                   struct if_dqinfo *dqinfo)
1651 {
1652         struct obd_quotactl *oqctl;
1653         int rc;
1654         ENTRY;
1655
1656         OBD_ALLOC_PTR(oqctl);
1657         if (!oqctl)
1658                 RETURN(-ENOMEM);
1659
1660         oqctl->qc_cmd = Q_GETINFO;
1661         oqctl->qc_type = type;
1662         rc = fsfilt_ext3_quotactl(sb, oqctl);
1663         if (!rc)
1664                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1665
1666         OBD_FREE_PTR(oqctl);
1667         RETURN(rc);
1668 }
1669
1670 struct qchk_ctxt {
1671         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1672         cfs_list_t              qckt_list;                 /* quotacheck list */
1673         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1674         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1675 };
1676
1677 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1678                            struct obd_quotactl *oqc)
1679 {
1680         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1681         loff_t size = 0;
1682         qid_t qid[MAXQUOTAS];
1683         int cnt, i, rc = 0;
1684
1685         if (!inode)
1686                 return 0;
1687
1688         qid[USRQUOTA] = inode->i_uid;
1689         qid[GRPQUOTA] = inode->i_gid;
1690
1691         if (S_ISDIR(inode->i_mode) ||
1692             S_ISREG(inode->i_mode) ||
1693             S_ISLNK(inode->i_mode))
1694                 size = inode_get_bytes(inode);
1695
1696         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1697                 if (!Q_TYPESET(oqc, cnt))
1698                         continue;
1699
1700                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1701                                 &qctxt->qckt_list, qid[cnt], cnt,
1702                                 qctxt->qckt_first_check[cnt]);
1703                 if (!cdqb[cnt]) {
1704                         rc = -ENOMEM;
1705                         break;
1706                 }
1707
1708                 cdqb[cnt]->dqb_curspace += size;
1709                 cdqb[cnt]->dqb_curinodes++;
1710         }
1711
1712         if (rc) {
1713                 for (i = 0; i < cnt; i++) {
1714                         if (!Q_TYPESET(oqc, i))
1715                                 continue;
1716                         LASSERT(cdqb[i]);
1717                         cdqb[i]->dqb_curspace -= size;
1718                         cdqb[i]->dqb_curinodes--;
1719                 }
1720         }
1721
1722         return rc;
1723 }
1724
1725 /* write dqinfo struct in a new quota file */
1726 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1727 {
1728         struct v2_disk_dqinfo dqinfo;
1729         __u32 blocks = V2_DQTREEOFF + 1;
1730         loff_t offset = V2_DQINFOOFF;
1731
1732         if (info) {
1733                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1734                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1735                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1736                                                ~DQF_INFO_DIRTY);
1737         } else {
1738                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1739                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1740                 dqinfo.dqi_flags = 0;
1741         }
1742
1743         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1744         dqinfo.dqi_free_blk = 0;
1745         dqinfo.dqi_free_entry = 0;
1746
1747         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1748 }
1749
1750 static int v3_write_dqheader(struct file *f, int type)
1751 {
1752         static const __u32 quota_magics[] = V2_INITQMAGICS;
1753         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1754         struct v2_disk_dqheader dqhead;
1755         loff_t offset = 0;
1756
1757         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1758         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1759
1760         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1761         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1762
1763         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1764 }
1765
1766 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1767                                   struct obd_quotactl *oqc)
1768 {
1769         int i, rc = 0;
1770         ENTRY;
1771
1772         for (i = 0; i < MAXQUOTAS; i++) {
1773                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1774                                          NULL : &qctxt->qckt_dqinfo[i];
1775                 struct file *file;
1776                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1777
1778                 if (!Q_TYPESET(oqc, i))
1779                         continue;
1780
1781                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1782
1783                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1784                 if (IS_ERR(file)) {
1785                         rc = PTR_ERR(file);
1786                         CERROR("can't create %s file: rc = %d\n",
1787                                name[i], rc);
1788                         GOTO(out, rc);
1789                 }
1790
1791                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1792                         CERROR("file %s is not regular", name[i]);
1793                         filp_close(file, 0);
1794                         GOTO(out, rc = -EINVAL);
1795                 }
1796
1797                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1798
1799                 rc = v3_write_dqheader(file, i);
1800                 if (rc) {
1801                         filp_close(file, 0);
1802                         GOTO(out, rc);
1803                 }
1804
1805                 rc = v3_write_dqinfo(file, i, info);
1806                 filp_close(file, 0);
1807                 if (rc)
1808                         GOTO(out, rc);
1809         }
1810
1811 out:
1812         RETURN(rc);
1813 }
1814
1815
1816 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1817                           struct chk_dqblk *cdqb)
1818 {
1819         struct obd_quotactl *oqc;
1820         long now;
1821         int rc;
1822         ENTRY;
1823
1824         OBD_ALLOC_PTR(oqc);
1825         if (!oqc)
1826                 RETURN(-ENOMEM);
1827
1828         now = cfs_time_current_sec();
1829
1830         if (cdqb->dqb_bsoftlimit &&
1831             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1832             !cdqb->dqb_btime)
1833                 cdqb->dqb_btime =
1834                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1835
1836         if (cdqb->dqb_isoftlimit &&
1837             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1838             !cdqb->dqb_itime)
1839                 cdqb->dqb_itime =
1840                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1841
1842         cdqb->dqb_valid = QIF_ALL;
1843
1844         oqc->qc_cmd = Q_SETQUOTA;
1845         oqc->qc_type = cdqb->dqb_type;
1846         oqc->qc_id = cdqb->dqb_id;
1847         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1848
1849         rc = fsfilt_ext3_quotactl(sb, oqc);
1850         OBD_FREE_PTR(oqc);
1851         RETURN(rc);
1852 }
1853
1854 static int prune_chkquots(struct super_block *sb,
1855                           struct qchk_ctxt *qctxt, int error)
1856 {
1857         struct chk_dqblk *cdqb, *tmp;
1858         int rc;
1859
1860         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1861                 if (!error) {
1862                         rc = commit_chkquot(sb, qctxt, cdqb);
1863                         if (rc)
1864                                 error = rc;
1865                 }
1866                 cfs_hlist_del_init(&cdqb->dqb_hash);
1867                 cfs_list_del(&cdqb->dqb_list);
1868                 OBD_FREE_PTR(cdqb);
1869         }
1870
1871         return error;
1872 }
1873
1874 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1875 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1876 #endif
1877
1878 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1879                                   struct obd_quotactl *oqc)
1880 {
1881         struct ext3_sb_info *sbi = EXT3_SB(sb);
1882         int i, group, uninit_feat = 0;
1883         struct qchk_ctxt *qctxt;
1884         struct buffer_head *bitmap_bh = NULL;
1885         unsigned long ino, inode_inuse;
1886         struct inode *inode;
1887         int rc = 0;
1888         ENTRY;
1889
1890         /* turn on quota and read dqinfo if existed */
1891         OBD_ALLOC_PTR(qctxt);
1892         if (!qctxt) {
1893                 oqc->qc_stat = -ENOMEM;
1894                 RETURN(-ENOMEM);
1895         }
1896
1897         for (i = 0; i < NR_DQHASH; i++)
1898                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1899         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
1900
1901         for (i = 0; i < MAXQUOTAS; i++) {
1902                 if (!Q_TYPESET(oqc, i))
1903                         continue;
1904
1905                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1906                 if (!rc || rc == -EBUSY) {
1907                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1908                         if (rc)
1909                                 GOTO(out, rc);
1910                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1911                         qctxt->qckt_first_check[i] = 1;
1912                 } else if (rc) {
1913                         GOTO(out, rc);
1914                 }
1915         }
1916         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
1917                 /* This filesystem supports the uninit group feature */
1918                 uninit_feat = 1;
1919
1920         /* number of inodes that have been allocated */
1921         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
1922                       percpu_counter_sum(&sbi->s_freeinodes_counter);
1923
1924         /* check quota and update in hash */
1925         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
1926              group++) {
1927                 unsigned long used_count = sbi->s_inodes_per_group;
1928
1929                 if (uninit_feat) {
1930                         struct ext3_group_desc *desc;
1931                         desc = ext3_get_group_desc(sb, group, NULL);
1932                         if (!desc)
1933                                 GOTO(out, -EIO);
1934
1935                         /* we don't really need to take the group lock here,
1936                          * but it may be useful if one day we support online
1937                          * quotacheck */
1938                         ext4_lock_group(sb, group);
1939                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
1940                                 /* no inode in use in this group, just skip it */
1941                                 ext3_unlock_group(sb, group);
1942                                 continue;
1943                         }
1944
1945                         used_count -= ext3_itable_unused_count(sb, desc);
1946                         ext3_unlock_group(sb, group);
1947                 }
1948
1949                 ino = group * sbi->s_inodes_per_group + 1;
1950                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
1951                 if (!bitmap_bh) {
1952                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
1953                                sb->s_id, group);
1954                         GOTO(out, -EIO);
1955                 }
1956
1957                 i = 0;
1958                 while (i < used_count &&
1959                        (i = ext3_find_next_bit(bitmap_bh->b_data,
1960                                                used_count, i)) < used_count) {
1961                         inode_inuse--;
1962                         i++;
1963                         ino = i + group * sbi->s_inodes_per_group;
1964                         if (ino < sbi->s_first_ino)
1965                                 continue;
1966                         inode = ext3_iget(sb, ino);
1967                         if (!inode || IS_ERR(inode))
1968                                 continue;
1969
1970                         rc = add_inode_quota(inode, qctxt, oqc);
1971                         iput(inode);
1972                         if (rc) {
1973                                 brelse(bitmap_bh);
1974                                 GOTO(out, rc);
1975                         }
1976                 }
1977                 brelse(bitmap_bh);
1978         }
1979
1980         /* read old quota limits from old quota file. (only for the user
1981          * has limits but hasn't file) */
1982 #ifdef HAVE_QUOTA_SUPPORT
1983         for (i = 0; i < MAXQUOTAS; i++) {
1984                 cfs_list_t id_list;
1985                 struct dquot_id *dqid, *tmp;
1986
1987                 if (!Q_TYPESET(oqc, i))
1988                         continue;
1989
1990                 if (qctxt->qckt_first_check[i])
1991                         continue;
1992
1993
1994                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1995                 CFS_INIT_LIST_HEAD(&id_list);
1996                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1997                 if (rc)
1998                         CERROR("read old limits failed. (rc:%d)\n", rc);
1999
2000                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2001                         cfs_list_del_init(&dqid->di_link);
2002
2003                         if (!rc)
2004                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2005                                       dqid->di_id, i,
2006                                       qctxt->qckt_first_check[i]);
2007                         OBD_FREE_PTR(dqid);
2008                 }
2009         }
2010 #endif
2011         /* turn off quota cause we are to dump chk_dqblk to files */
2012         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2013
2014         rc = create_new_quota_files(qctxt, oqc);
2015         if (rc)
2016                 GOTO(out, rc);
2017
2018         /* we use vfs functions to set dqblk, so turn quota on */
2019         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2020 out:
2021         /* dump and free chk_dqblk */
2022         rc = prune_chkquots(sb, qctxt, rc);
2023         OBD_FREE_PTR(qctxt);
2024
2025         /* turn off quota, `lfs quotacheck` will turn on when all
2026          * nodes quotacheck finish. */
2027         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2028
2029         oqc->qc_stat = rc;
2030         if (rc)
2031                 CERROR("quotacheck failed: rc = %d\n", rc);
2032
2033         RETURN(rc);
2034 }
2035
2036 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2037                                  int cmd)
2038 {
2039         int rc = 0;
2040         ENTRY;
2041
2042         if (lqi->qi_files[type] == NULL) {
2043                 CERROR("operate qinfo before it's enabled!\n");
2044                 RETURN(-ESRCH);
2045         }
2046
2047         switch (cmd) {
2048         case QFILE_CHK:
2049                 rc = lustre_check_quota_file(lqi, type);
2050                 break;
2051         case QFILE_RD_INFO:
2052                 rc = lustre_read_quota_info(lqi, type);
2053                 break;
2054         case QFILE_WR_INFO:
2055                 rc = lustre_write_quota_info(lqi, type);
2056                 break;
2057         case QFILE_INIT_INFO:
2058                 rc = lustre_init_quota_info(lqi, type);
2059                 break;
2060         case QFILE_CONVERT:
2061                 rc = -ENOTSUPP;
2062                 CERROR("quota CONVERT command is not supported\n");
2063                 break;
2064         default:
2065                 rc = -ENOTSUPP;
2066                 CERROR("Unsupported admin quota file cmd %d\n"
2067                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2068                        cmd);
2069                 break;
2070         }
2071         RETURN(rc);
2072 }
2073
2074 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2075                             cfs_list_t *list)
2076 {
2077         return lustre_get_qids(file, inode, type, list);
2078 }
2079
2080 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2081 {
2082         int rc = 0;
2083         ENTRY;
2084
2085         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2086                 CERROR("operate dquot before it's enabled!\n");
2087                 RETURN(-ESRCH);
2088         }
2089
2090         switch (cmd) {
2091         case QFILE_RD_DQUOT:
2092                 rc = lustre_read_dquot(dquot);
2093                 break;
2094         case QFILE_WR_DQUOT:
2095                 if (dquot->dq_dqb.dqb_ihardlimit ||
2096                     dquot->dq_dqb.dqb_isoftlimit ||
2097                     dquot->dq_dqb.dqb_bhardlimit ||
2098                     dquot->dq_dqb.dqb_bsoftlimit)
2099                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2100                 else
2101                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2102
2103                 rc = lustre_commit_dquot(dquot);
2104                 if (rc >= 0)
2105                         rc = 0;
2106                 break;
2107         default:
2108                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2109                 LBUG();
2110                 break;
2111         }
2112         RETURN(rc);
2113 }
2114
2115 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2116                                 struct inode *inode, int frags)
2117 {
2118         struct ext3_ext_base *base = inode;
2119         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2120          * metablocks at maxium b=16542 */
2121         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2122         return 0;
2123 }
2124
2125 #endif
2126
2127 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2128 {
2129         return (EXT3_SB(sb)->journal_bdev);
2130 }
2131 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2132
2133 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2134                           char *buf, int count, loff_t pos)
2135 {
2136         loff_t p = pos;
2137         int rc;
2138
2139         if (!f && !inode) {
2140                 CERROR("lustre_read_quota failed for no quota file!\n");
2141                 libcfs_debug_dumpstack(NULL);
2142                 return -EINVAL;
2143         }
2144
2145         /* Support for both adm and op quota files must be provided */
2146         if (f) {
2147                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2148                 rc = rc < 0 ? rc : p - pos;
2149         } else {
2150                 struct super_block *sb = inode->i_sb;
2151                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2152         }
2153         return rc;
2154 }
2155
2156 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2157 {
2158         loff_t p = pos;
2159         int rc;
2160
2161         /* Only adm quota files are supported, op updates are handled by vfs */
2162         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2163         rc = rc < 0 ? rc : p - pos;
2164
2165         return rc;
2166 }
2167
2168 void *lustre_quota_journal_start(struct inode *inode, int delete)
2169 {
2170         handle_t *handle;
2171         unsigned block_count;
2172
2173         if (delete) {
2174                 /* each indirect block (+4) may become free, attaching to the
2175                  * header list of free blocks (+1); the data block (+1) may
2176                  * become a free block (+0) or a block with free dqentries (+0) */
2177                 block_count = (4 + 1) + 1;
2178                 handle = ext3_journal_start(inode,
2179                              block_count*EXT3_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2180         } else {
2181                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2182                  * freeblk reusage with a header update (+1); dqentry is either reused
2183                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2184                  * a new block allocation (+1) with a header update (+1)              */
2185                 block_count = (4 + 1) + 3;
2186                 handle = ext3_journal_start(inode,
2187                              block_count*EXT3_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2188
2189         }
2190
2191         return handle;
2192 }
2193
2194 void lustre_quota_journal_stop(void *handle)
2195 {
2196         ext3_journal_stop((handle_t *)handle);
2197 }
2198
2199 static int ll_decode_fh_accept(void *context, struct dentry *de)
2200 {
2201         return 1;
2202 }
2203
2204 #ifdef HAVE_EXPORTFS_DECODE_FH
2205 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2206          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2207                             acceptable, context)
2208 #else
2209 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2210          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2211                                      type, acceptable, context)
2212 # define FILEID_INO32_GEN 1
2213 extern struct export_operations export_op_default;
2214 #endif
2215
2216 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2217                                       struct fsfilt_fid *fid, int ignore_gen)
2218 {
2219         struct inode  *inode;
2220         struct dentry *result;
2221
2222         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2223                                        ll_decode_fh_accept, NULL);
2224         if (IS_ERR(result)) {
2225                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2226                        fid->ino, fid->gen, PTR_ERR(result));
2227                 return result;
2228         }
2229
2230         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2231                fid->ino, fid->gen);
2232         inode = result->d_inode;
2233         if (inode == NULL)
2234                 goto err_out;
2235
2236         if (inode->i_nlink == 0 &&
2237             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2238                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2239                               " ctime -- this may indicate disk "
2240                               "corruption (inode: %lu, link: %lu, "
2241                               "count: %d)\n", inode->i_ino,
2242                               (unsigned long)inode->i_nlink,
2243                               atomic_read(&inode->i_count));
2244                 goto err_out;
2245         }
2246         if (fid->gen && inode->i_generation != fid->gen) {
2247                 /* we didn't find the right inode.. */
2248                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2249                        "count: %d, generation %u/%u\n",
2250                        inode->i_ino, (unsigned long)inode->i_nlink,
2251                        atomic_read(&inode->i_count), inode->i_generation,
2252                        fid->gen);
2253                 goto err_out;
2254         }
2255
2256         return result;
2257 err_out:
2258         l_dput(result);
2259         return ERR_PTR(-ENOENT);
2260 }
2261
2262 static struct fsfilt_operations fsfilt_ext3_ops = {
2263         .fs_type                = "ext3",
2264         .fs_owner               = THIS_MODULE,
2265         .fs_getlabel            = fsfilt_ext3_get_label,
2266         .fs_setlabel            = fsfilt_ext3_set_label,
2267         .fs_uuid                = fsfilt_ext3_uuid,
2268         .fs_start               = fsfilt_ext3_start,
2269         .fs_brw_start           = fsfilt_ext3_brw_start,
2270         .fs_extend              = fsfilt_ext3_extend,
2271         .fs_commit              = fsfilt_ext3_commit,
2272         .fs_commit_async        = fsfilt_ext3_commit_async,
2273         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2274         .fs_setattr             = fsfilt_ext3_setattr,
2275         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2276         .fs_set_md              = fsfilt_ext3_set_md,
2277         .fs_get_md              = fsfilt_ext3_get_md,
2278         .fs_readpage            = fsfilt_ext3_readpage,
2279         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2280         .fs_statfs              = fsfilt_ext3_statfs,
2281         .fs_sync                = fsfilt_ext3_sync,
2282         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2283         .fs_write_record        = fsfilt_ext3_write_record,
2284         .fs_read_record         = fsfilt_ext3_read_record,
2285         .fs_setup               = fsfilt_ext3_setup,
2286         .fs_send_bio            = fsfilt_ext3_send_bio,
2287         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2288 #ifdef HAVE_DISK_INODE_VERSION
2289         .fs_get_version         = fsfilt_ext3_get_version,
2290         .fs_set_version         = fsfilt_ext3_set_version,
2291 #endif
2292 #ifdef HAVE_QUOTA_SUPPORT
2293         .fs_quotactl            = fsfilt_ext3_quotactl,
2294         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2295         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2296         .fs_qids                = fsfilt_ext3_qids,
2297         .fs_dquot               = fsfilt_ext3_dquot,
2298         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2299 #endif
2300         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2301         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2302 };
2303
2304 static int __init fsfilt_ext3_init(void)
2305 {
2306         int rc;
2307
2308         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2309                                          sizeof(struct fsfilt_cb_data), 0, 0);
2310         if (!fcb_cache) {
2311                 CERROR("error allocating fsfilt journal callback cache\n");
2312                 GOTO(out, rc = -ENOMEM);
2313         }
2314
2315         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2316
2317         if (rc) {
2318                 int err = cfs_mem_cache_destroy(fcb_cache);
2319                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2320         }
2321 out:
2322         return rc;
2323 }
2324
2325 static void __exit fsfilt_ext3_exit(void)
2326 {
2327         int rc;
2328
2329         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2330         rc = cfs_mem_cache_destroy(fcb_cache);
2331         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2332 }
2333
2334 module_init(fsfilt_ext3_init);
2335 module_exit(fsfilt_ext3_exit);
2336
2337 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2338 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2339 MODULE_LICENSE("GPL");