fs/ocfs2/file.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * file.c
   5  *
   6  * File open, close, extend, truncate
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/capability.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/uio.h>
  33 #include <linux/sched.h>
  34 #include <linux/splice.h>
  35 #include <linux/mount.h>
  36 #include <linux/writeback.h>
  37 #include <linux/falloc.h>
  38 #include <linux/quotaops.h>
  39 #include <linux/blkdev.h>
  40
  41 #include <cluster/masklog.h>
  42
  43 #include "ocfs2.h"
  44
  45 #include "alloc.h"
  46 #include "aops.h"
  47 #include "dir.h"
  48 #include "dlmglue.h"
  49 #include "extent_map.h"
  50 #include "file.h"
  51 #include "sysfile.h"
  52 #include "inode.h"
  53 #include "ioctl.h"
  54 #include "journal.h"
  55 #include "locks.h"
  56 #include "mmap.h"
  57 #include "suballoc.h"
  58 #include "super.h"
  59 #include "xattr.h"
  60 #include "acl.h"
  61 #include "quota.h"
  62 #include "refcounttree.h"
  63 #include "ocfs2_trace.h"
  64
  65 #include "buffer_head_io.h"
  66
  67 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  68 {
  69         struct ocfs2_file_private *fp;
  70
  71         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  72         if (!fp)
  73                 return -ENOMEM;
  74
  75         fp->fp_file = file;
  76         mutex_init(&fp->fp_mutex);
  77         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  78         file->private_data = fp;
  79
  80         return 0;
  81 }
  82
  83 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  84 {
  85         struct ocfs2_file_private *fp = file->private_data;
  86         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  87
  88         if (fp) {
  89                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  90                 ocfs2_lock_res_free(&fp->fp_flock);
  91                 kfree(fp);
  92                 file->private_data = NULL;
  93         }
  94 }
  95
  96 static int ocfs2_file_open(struct inode *inode, struct file *file)
  97 {
  98         int status;
  99         int mode = file->f_flags;
 100         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 101
 102         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
 103                               (unsigned long long)OCFS2_I(inode)->ip_blkno,
 104                               file->f_path.dentry->d_name.len,
 105                               file->f_path.dentry->d_name.name, mode);
 106
 107         if (file->f_mode & FMODE_WRITE)
 108                 dquot_initialize(inode);
 109
 110         spin_lock(&oi->ip_lock);
 111
 112         /* Check that the inode hasn't been wiped from disk by another
 113          * node. If it hasn't then we're safe as long as we hold the
 114          * spin lock until our increment of open count. */
 115         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
 116                 spin_unlock(&oi->ip_lock);
 117
 118                 status = -ENOENT;
 119                 goto leave;
 120         }
 121
 122         if (mode & O_DIRECT)
 123                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 124
 125         oi->ip_open_count++;
 126         spin_unlock(&oi->ip_lock);
 127
 128         status = ocfs2_init_file_private(inode, file);
 129         if (status) {
 130                 /*
 131                  * We want to set open count back if we're failing the
 132                  * open.
 133                  */
 134                 spin_lock(&oi->ip_lock);
 135                 oi->ip_open_count--;
 136                 spin_unlock(&oi->ip_lock);
 137         }
 138
 139 leave:
 140         return status;
 141 }
 142
 143 static int ocfs2_file_release(struct inode *inode, struct file *file)
 144 {
 145         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 146
 147         spin_lock(&oi->ip_lock);
 148         if (!--oi->ip_open_count)
 149                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 150
 151         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 152                                  oi->ip_blkno,
 153                                  file->f_path.dentry->d_name.len,
 154                                  file->f_path.dentry->d_name.name,
 155                                  oi->ip_open_count);
 156         spin_unlock(&oi->ip_lock);
 157
 158         ocfs2_free_file_private(inode, file);
 159
 160         return 0;
 161 }
 162
 163 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 164 {
 165         return ocfs2_init_file_private(inode, file);
 166 }
 167
 168 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 169 {
 170         ocfs2_free_file_private(inode, file);
 171         return 0;
 172 }
 173
 174 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 175                            int datasync)
 176 {
 177         int err = 0;
 178         journal_t *journal;
 179         struct inode *inode = file->f_mapping->host;
 180         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 181
 182         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 183                               OCFS2_I(inode)->ip_blkno,
 184                               file->f_path.dentry->d_name.len,
 185                               file->f_path.dentry->d_name.name,
 186                               (unsigned long long)datasync);
 187
 188         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 189                 return -EROFS;
 190
 191         err = filemap_write_and_wait_range(inode->i_mapping, start, end);
 192         if (err)
 193                 return err;
 194
 195         /*
 196          * Probably don't need the i_mutex at all in here, just putting it here
 197          * to be consistent with how fsync used to be called, someone more
 198          * familiar with the fs could possibly remove it.
 199          */
 200         mutex_lock(&inode->i_mutex);
 201         if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
 202                 /*
 203                  * We still have to flush drive's caches to get data to the
 204                  * platter
 205                  */
 206                 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
 207                         blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 208                 goto bail;
 209         }
 210
 211         journal = osb->journal->j_journal;
 212         err = jbd2_journal_force_commit(journal);
 213
 214 bail:
 215         if (err)
 216                 mlog_errno(err);
 217         mutex_unlock(&inode->i_mutex);
 218
 219         return (err < 0) ? -EIO : 0;
 220 }
 221
 222 int ocfs2_should_update_atime(struct inode *inode,
 223                               struct vfsmount *vfsmnt)
 224 {
 225         struct timespec now;
 226         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 227
 228         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 229                 return 0;
 230
 231         if ((inode->i_flags & S_NOATIME) ||
 232             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 233                 return 0;
 234
 235         /*
 236          * We can be called with no vfsmnt structure - NFSD will
 237          * sometimes do this.
 238          *
 239          * Note that our action here is different than touch_atime() -
 240          * if we can't tell whether this is a noatime mount, then we
 241          * don't know whether to trust the value of s_atime_quantum.
 242          */
 243         if (vfsmnt == NULL)
 244                 return 0;
 245
 246         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 247             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 248                 return 0;
 249
 250         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 251                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 252                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 253                         return 1;
 254
 255                 return 0;
 256         }
 257
 258         now = CURRENT_TIME;
 259         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 260                 return 0;
 261         else
 262                 return 1;
 263 }
 264
 265 int ocfs2_update_inode_atime(struct inode *inode,
 266                              struct buffer_head *bh)
 267 {
 268         int ret;
 269         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 270         handle_t *handle;
 271         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 272
 273         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 274         if (IS_ERR(handle)) {
 275                 ret = PTR_ERR(handle);
 276                 mlog_errno(ret);
 277                 goto out;
 278         }
 279
 280         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 281                                       OCFS2_JOURNAL_ACCESS_WRITE);
 282         if (ret) {
 283                 mlog_errno(ret);
 284                 goto out_commit;
 285         }
 286
 287         /*
 288          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 289          * have i_mutex to guard against concurrent changes to other
 290          * inode fields.
 291          */
 292         inode->i_atime = CURRENT_TIME;
 293         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 294         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 295         ocfs2_journal_dirty(handle, bh);
 296
 297 out_commit:
 298         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 299 out:
 300         return ret;
 301 }
 302
 303 static int ocfs2_set_inode_size(handle_t *handle,
 304                                 struct inode *inode,
 305                                 struct buffer_head *fe_bh,
 306                                 u64 new_i_size)
 307 {
 308         int status;
 309
 310         i_size_write(inode, new_i_size);
 311         inode->i_blocks = ocfs2_inode_sector_count(inode);
 312         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 313
 314         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 315         if (status < 0) {
 316                 mlog_errno(status);
 317                 goto bail;
 318         }
 319
 320 bail:
 321         return status;
 322 }
 323
 324 int ocfs2_simple_size_update(struct inode *inode,
 325                              struct buffer_head *di_bh,
 326                              u64 new_i_size)
 327 {
 328         int ret;
 329         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 330         handle_t *handle = NULL;
 331
 332         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 333         if (IS_ERR(handle)) {
 334                 ret = PTR_ERR(handle);
 335                 mlog_errno(ret);
 336                 goto out;
 337         }
 338
 339         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 340                                    new_i_size);
 341         if (ret < 0)
 342                 mlog_errno(ret);
 343
 344         ocfs2_commit_trans(osb, handle);
 345 out:
 346         return ret;
 347 }
 348
 349 static int ocfs2_cow_file_pos(struct inode *inode,
 350                               struct buffer_head *fe_bh,
 351                               u64 offset)
 352 {
 353         int status;
 354         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 355         unsigned int num_clusters = 0;
 356         unsigned int ext_flags = 0;
 357
 358         /*
 359          * If the new offset is aligned to the range of the cluster, there is
 360          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 361          * CoW either.
 362          */
 363         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 364                 return 0;
 365
 366         status = ocfs2_get_clusters(inode, cpos, &phys,
 367                                     &num_clusters, &ext_flags);
 368         if (status) {
 369                 mlog_errno(status);
 370                 goto out;
 371         }
 372
 373         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 374                 goto out;
 375
 376         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 377
 378 out:
 379         return status;
 380 }
 381
 382 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 383                                      struct inode *inode,
 384                                      struct buffer_head *fe_bh,
 385                                      u64 new_i_size)
 386 {
 387         int status;
 388         handle_t *handle;
 389         struct ocfs2_dinode *di;
 390         u64 cluster_bytes;
 391
 392         /*
 393          * We need to CoW the cluster contains the offset if it is reflinked
 394          * since we will call ocfs2_zero_range_for_truncate later which will
 395          * write "0" from offset to the end of the cluster.
 396          */
 397         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 398         if (status) {
 399                 mlog_errno(status);
 400                 return status;
 401         }
 402
 403         /* TODO: This needs to actually orphan the inode in this
 404          * transaction. */
 405
 406         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 407         if (IS_ERR(handle)) {
 408                 status = PTR_ERR(handle);
 409                 mlog_errno(status);
 410                 goto out;
 411         }
 412
 413         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 414                                          OCFS2_JOURNAL_ACCESS_WRITE);
 415         if (status < 0) {
 416                 mlog_errno(status);
 417                 goto out_commit;
 418         }
 419
 420         /*
 421          * Do this before setting i_size.
 422          */
 423         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 424         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 425                                                cluster_bytes);
 426         if (status) {
 427                 mlog_errno(status);
 428                 goto out_commit;
 429         }
 430
 431         i_size_write(inode, new_i_size);
 432         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 433
 434         di = (struct ocfs2_dinode *) fe_bh->b_data;
 435         di->i_size = cpu_to_le64(new_i_size);
 436         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 437         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 438
 439         ocfs2_journal_dirty(handle, fe_bh);
 440
 441 out_commit:
 442         ocfs2_commit_trans(osb, handle);
 443 out:
 444         return status;
 445 }
 446
 447 static int ocfs2_truncate_file(struct inode *inode,
 448                                struct buffer_head *di_bh,
 449                                u64 new_i_size)
 450 {
 451         int status = 0;
 452         struct ocfs2_dinode *fe = NULL;
 453         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 454
 455         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 456          * already validated it */
 457         fe = (struct ocfs2_dinode *) di_bh->b_data;
 458
 459         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 460                                   (unsigned long long)le64_to_cpu(fe->i_size),
 461                                   (unsigned long long)new_i_size);
 462
 463         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 464                         "Inode %llu, inode i_size = %lld != di "
 465                         "i_size = %llu, i_flags = 0x%x\n",
 466                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 467                         i_size_read(inode),
 468                         (unsigned long long)le64_to_cpu(fe->i_size),
 469                         le32_to_cpu(fe->i_flags));
 470
 471         if (new_i_size > le64_to_cpu(fe->i_size)) {
 472                 trace_ocfs2_truncate_file_error(
 473                         (unsigned long long)le64_to_cpu(fe->i_size),
 474                         (unsigned long long)new_i_size);
 475                 status = -EINVAL;
 476                 mlog_errno(status);
 477                 goto bail;
 478         }
 479
 480         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 481
 482         ocfs2_resv_discard(&osb->osb_la_resmap,
 483                            &OCFS2_I(inode)->ip_la_data_resv);
 484
 485         /*
 486          * The inode lock forced other nodes to sync and drop their
 487          * pages, which (correctly) happens even if we have a truncate
 488          * without allocation change - ocfs2 cluster sizes can be much
 489          * greater than page size, so we have to truncate them
 490          * anyway.
 491          */
 492         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 493         truncate_inode_pages(inode->i_mapping, new_i_size);
 494
 495         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 496                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 497                                                i_size_read(inode), 1);
 498                 if (status)
 499                         mlog_errno(status);
 500
 501                 goto bail_unlock_sem;
 502         }
 503
 504         /* alright, we're going to need to do a full blown alloc size
 505          * change. Orphan the inode so that recovery can complete the
 506          * truncate if necessary. This does the task of marking
 507          * i_size. */
 508         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 509         if (status < 0) {
 510                 mlog_errno(status);
 511                 goto bail_unlock_sem;
 512         }
 513
 514         status = ocfs2_commit_truncate(osb, inode, di_bh);
 515         if (status < 0) {
 516                 mlog_errno(status);
 517                 goto bail_unlock_sem;
 518         }
 519
 520         /* TODO: orphan dir cleanup here. */
 521 bail_unlock_sem:
 522         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 523
 524 bail:
 525         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 526                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 527
 528         return status;
 529 }
 530
 531 /*
 532  * extend file allocation only here.
 533  * we'll update all the disk stuff, and oip->alloc_size
 534  *
 535  * expect stuff to be locked, a transaction started and enough data /
 536  * metadata reservations in the contexts.
 537  *
 538  * Will return -EAGAIN, and a reason if a restart is needed.
 539  * If passed in, *reason will always be set, even in error.
 540  */
 541 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 542                          struct inode *inode,
 543                          u32 *logical_offset,
 544                          u32 clusters_to_add,
 545                          int mark_unwritten,
 546                          struct buffer_head *fe_bh,
 547                          handle_t *handle,
 548                          struct ocfs2_alloc_context *data_ac,
 549                          struct ocfs2_alloc_context *meta_ac,
 550                          enum ocfs2_alloc_restarted *reason_ret)
 551 {
 552         int ret;
 553         struct ocfs2_extent_tree et;
 554
 555         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 556         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 557                                           clusters_to_add, mark_unwritten,
 558                                           data_ac, meta_ac, reason_ret);
 559
 560         return ret;
 561 }
 562
 563 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 564                                      u32 clusters_to_add, int mark_unwritten)
 565 {
 566         int status = 0;
 567         int restart_func = 0;
 568         int credits;
 569         u32 prev_clusters;
 570         struct buffer_head *bh = NULL;
 571         struct ocfs2_dinode *fe = NULL;
 572         handle_t *handle = NULL;
 573         struct ocfs2_alloc_context *data_ac = NULL;
 574         struct ocfs2_alloc_context *meta_ac = NULL;
 575         enum ocfs2_alloc_restarted why;
 576         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 577         struct ocfs2_extent_tree et;
 578         int did_quota = 0;
 579
 580         /*
 581          * Unwritten extent only exists for file systems which
 582          * support holes.
 583          */
 584         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 585
 586         status = ocfs2_read_inode_block(inode, &bh);
 587         if (status < 0) {
 588                 mlog_errno(status);
 589                 goto leave;
 590         }
 591         fe = (struct ocfs2_dinode *) bh->b_data;
 592
 593 restart_all:
 594         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 595
 596         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 597         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 598                                        &data_ac, &meta_ac);
 599         if (status) {
 600                 mlog_errno(status);
 601                 goto leave;
 602         }
 603
 604         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 605         handle = ocfs2_start_trans(osb, credits);
 606         if (IS_ERR(handle)) {
 607                 status = PTR_ERR(handle);
 608                 handle = NULL;
 609                 mlog_errno(status);
 610                 goto leave;
 611         }
 612
 613 restarted_transaction:
 614         trace_ocfs2_extend_allocation(
 615                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 616                 (unsigned long long)i_size_read(inode),
 617                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 618                 why, restart_func);
 619
 620         status = dquot_alloc_space_nodirty(inode,
 621                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 622         if (status)
 623                 goto leave;
 624         did_quota = 1;
 625
 626         /* reserve a write to the file entry early on - that we if we
 627          * run out of credits in the allocation path, we can still
 628          * update i_size. */
 629         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 630                                          OCFS2_JOURNAL_ACCESS_WRITE);
 631         if (status < 0) {
 632                 mlog_errno(status);
 633                 goto leave;
 634         }
 635
 636         prev_clusters = OCFS2_I(inode)->ip_clusters;
 637
 638         status = ocfs2_add_inode_data(osb,
 639                                       inode,
 640                                       &logical_start,
 641                                       clusters_to_add,
 642                                       mark_unwritten,
 643                                       bh,
 644                                       handle,
 645                                       data_ac,
 646                                       meta_ac,
 647                                       &why);
 648         if ((status < 0) && (status != -EAGAIN)) {
 649                 if (status != -ENOSPC)
 650                         mlog_errno(status);
 651                 goto leave;
 652         }
 653
 654         ocfs2_journal_dirty(handle, bh);
 655
 656         spin_lock(&OCFS2_I(inode)->ip_lock);
 657         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 658         spin_unlock(&OCFS2_I(inode)->ip_lock);
 659         /* Release unused quota reservation */
 660         dquot_free_space(inode,
 661                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 662         did_quota = 0;
 663
 664         if (why != RESTART_NONE && clusters_to_add) {
 665                 if (why == RESTART_META) {
 666                         restart_func = 1;
 667                         status = 0;
 668                 } else {
 669                         BUG_ON(why != RESTART_TRANS);
 670
 671                         status = ocfs2_allocate_extend_trans(handle, 1);
 672                         if (status < 0) {
 673                                 /* handle still has to be committed at
 674                                  * this point. */
 675                                 status = -ENOMEM;
 676                                 mlog_errno(status);
 677                                 goto leave;
 678                         }
 679                         goto restarted_transaction;
 680                 }
 681         }
 682
 683         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 684              le32_to_cpu(fe->i_clusters),
 685              (unsigned long long)le64_to_cpu(fe->i_size),
 686              OCFS2_I(inode)->ip_clusters,
 687              (unsigned long long)i_size_read(inode));
 688
 689 leave:
 690         if (status < 0 && did_quota)
 691                 dquot_free_space(inode,
 692                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 693         if (handle) {
 694                 ocfs2_commit_trans(osb, handle);
 695                 handle = NULL;
 696         }
 697         if (data_ac) {
 698                 ocfs2_free_alloc_context(data_ac);
 699                 data_ac = NULL;
 700         }
 701         if (meta_ac) {
 702                 ocfs2_free_alloc_context(meta_ac);
 703                 meta_ac = NULL;
 704         }
 705         if ((!status) && restart_func) {
 706                 restart_func = 0;
 707                 goto restart_all;
 708         }
 709         brelse(bh);
 710         bh = NULL;
 711
 712         return status;
 713 }
 714
 715 /*
 716  * While a write will already be ordering the data, a truncate will not.
 717  * Thus, we need to explicitly order the zeroed pages.
 718  */
 719 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 720                                                 struct buffer_head *di_bh)
 721 {
 722         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 723         handle_t *handle = NULL;
 724         int ret = 0;
 725
 726         if (!ocfs2_should_order_data(inode))
 727                 goto out;
 728
 729         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 730         if (IS_ERR(handle)) {
 731                 ret = -ENOMEM;
 732                 mlog_errno(ret);
 733                 goto out;
 734         }
 735
 736         ret = ocfs2_jbd2_file_inode(handle, inode);
 737         if (ret < 0) {
 738                 mlog_errno(ret);
 739                 goto out;
 740         }
 741
 742         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 743                                       OCFS2_JOURNAL_ACCESS_WRITE);
 744         if (ret)
 745                 mlog_errno(ret);
 746
 747 out:
 748         if (ret) {
 749                 if (!IS_ERR(handle))
 750                         ocfs2_commit_trans(osb, handle);
 751                 handle = ERR_PTR(ret);
 752         }
 753         return handle;
 754 }
 755
 756 /* Some parts of this taken from generic_cont_expand, which turned out
 757  * to be too fragile to do exactly what we need without us having to
 758  * worry about recursive locking in ->write_begin() and ->write_end(). */
 759 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 760                                  u64 abs_to, struct buffer_head *di_bh)
 761 {
 762         struct address_space *mapping = inode->i_mapping;
 763         struct page *page;
 764         unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
 765         handle_t *handle = NULL;
 766         int ret = 0;
 767         unsigned zero_from, zero_to, block_start, block_end;
 768         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 769
 770         BUG_ON(abs_from >= abs_to);
 771         BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
 772         BUG_ON(abs_from & (inode->i_blkbits - 1));
 773
 774         page = find_or_create_page(mapping, index, GFP_NOFS);
 775         if (!page) {
 776                 ret = -ENOMEM;
 777                 mlog_errno(ret);
 778                 goto out;
 779         }
 780
 781         /* Get the offsets within the page that we want to zero */
 782         zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
 783         zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
 784         if (!zero_to)
 785                 zero_to = PAGE_CACHE_SIZE;
 786
 787         trace_ocfs2_write_zero_page(
 788                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 789                         (unsigned long long)abs_from,
 790                         (unsigned long long)abs_to,
 791                         index, zero_from, zero_to);
 792
 793         /* We know that zero_from is block aligned */
 794         for (block_start = zero_from; block_start < zero_to;
 795              block_start = block_end) {
 796                 block_end = block_start + (1 << inode->i_blkbits);
 797
 798                 /*
 799                  * block_start is block-aligned.  Bump it by one to force
 800                  * __block_write_begin and block_commit_write to zero the
 801                  * whole block.
 802                  */
 803                 ret = __block_write_begin(page, block_start + 1, 0,
 804                                           ocfs2_get_block);
 805                 if (ret < 0) {
 806                         mlog_errno(ret);
 807                         goto out_unlock;
 808                 }
 809
 810                 if (!handle) {
 811                         handle = ocfs2_zero_start_ordered_transaction(inode,
 812                                                                       di_bh);
 813                         if (IS_ERR(handle)) {
 814                                 ret = PTR_ERR(handle);
 815                                 handle = NULL;
 816                                 break;
 817                         }
 818                 }
 819
 820                 /* must not update i_size! */
 821                 ret = block_commit_write(page, block_start + 1,
 822                                          block_start + 1);
 823                 if (ret < 0)
 824                         mlog_errno(ret);
 825                 else
 826                         ret = 0;
 827         }
 828
 829         if (handle) {
 830                 /*
 831                  * fs-writeback will release the dirty pages without page lock
 832                  * whose offset are over inode size, the release happens at
 833                  * block_write_full_page_endio().
 834                  */
 835                 i_size_write(inode, abs_to);
 836                 inode->i_blocks = ocfs2_inode_sector_count(inode);
 837                 di->i_size = cpu_to_le64((u64)i_size_read(inode));
 838                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 839                 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 840                 di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 841                 di->i_mtime_nsec = di->i_ctime_nsec;
 842                 ocfs2_journal_dirty(handle, di_bh);
 843                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 844         }
 845
 846 out_unlock:
 847         unlock_page(page);
 848         page_cache_release(page);
 849 out:
 850         return ret;
 851 }
 852
 853 /*
 854  * Find the next range to zero.  We do this in terms of bytes because
 855  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 856  * pagecache.  We may return multiple extents.
 857  *
 858  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 859  * needs to be zeroed.  range_start and range_end return the next zeroing
 860  * range.  A subsequent call should pass the previous range_end as its
 861  * zero_start.  If range_end is 0, there's nothing to do.
 862  *
 863  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 864  */
 865 static int ocfs2_zero_extend_get_range(struct inode *inode,
 866                                        struct buffer_head *di_bh,
 867                                        u64 zero_start, u64 zero_end,
 868                                        u64 *range_start, u64 *range_end)
 869 {
 870         int rc = 0, needs_cow = 0;
 871         u32 p_cpos, zero_clusters = 0;
 872         u32 zero_cpos =
 873                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 874         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 875         unsigned int num_clusters = 0;
 876         unsigned int ext_flags = 0;
 877
 878         while (zero_cpos < last_cpos) {
 879                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 880                                         &num_clusters, &ext_flags);
 881                 if (rc) {
 882                         mlog_errno(rc);
 883                         goto out;
 884                 }
 885
 886                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 887                         zero_clusters = num_clusters;
 888                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 889                                 needs_cow = 1;
 890                         break;
 891                 }
 892
 893                 zero_cpos += num_clusters;
 894         }
 895         if (!zero_clusters) {
 896                 *range_end = 0;
 897                 goto out;
 898         }
 899
 900         while ((zero_cpos + zero_clusters) < last_cpos) {
 901                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 902                                         &p_cpos, &num_clusters,
 903                                         &ext_flags);
 904                 if (rc) {
 905                         mlog_errno(rc);
 906                         goto out;
 907                 }
 908
 909                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 910                         break;
 911                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 912                         needs_cow = 1;
 913                 zero_clusters += num_clusters;
 914         }
 915         if ((zero_cpos + zero_clusters) > last_cpos)
 916                 zero_clusters = last_cpos - zero_cpos;
 917
 918         if (needs_cow) {
 919                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 920                                         zero_clusters, UINT_MAX);
 921                 if (rc) {
 922                         mlog_errno(rc);
 923                         goto out;
 924                 }
 925         }
 926
 927         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 928         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 929                                              zero_cpos + zero_clusters);
 930
 931 out:
 932         return rc;
 933 }
 934
 935 /*
 936  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 937  * has made sure that the entire range needs zeroing.
 938  */
 939 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 940                                    u64 range_end, struct buffer_head *di_bh)
 941 {
 942         int rc = 0;
 943         u64 next_pos;
 944         u64 zero_pos = range_start;
 945
 946         trace_ocfs2_zero_extend_range(
 947                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 948                         (unsigned long long)range_start,
 949                         (unsigned long long)range_end);
 950         BUG_ON(range_start >= range_end);
 951
 952         while (zero_pos < range_end) {
 953                 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
 954                 if (next_pos > range_end)
 955                         next_pos = range_end;
 956                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 957                 if (rc < 0) {
 958                         mlog_errno(rc);
 959                         break;
 960                 }
 961                 zero_pos = next_pos;
 962
 963                 /*
 964                  * Very large extends have the potential to lock up
 965                  * the cpu for extended periods of time.
 966                  */
 967                 cond_resched();
 968         }
 969
 970         return rc;
 971 }
 972
 973 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 974                       loff_t zero_to_size)
 975 {
 976         int ret = 0;
 977         u64 zero_start, range_start = 0, range_end = 0;
 978         struct super_block *sb = inode->i_sb;
 979
 980         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 981         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 982                                 (unsigned long long)zero_start,
 983                                 (unsigned long long)i_size_read(inode));
 984         while (zero_start < zero_to_size) {
 985                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 986                                                   zero_to_size,
 987                                                   &range_start,
 988                                                   &range_end);
 989                 if (ret) {
 990                         mlog_errno(ret);
 991                         break;
 992                 }
 993                 if (!range_end)
 994                         break;
 995                 /* Trim the ends */
 996                 if (range_start < zero_start)
 997                         range_start = zero_start;
 998                 if (range_end > zero_to_size)
 999                         range_end = zero_to_size;
1000
1001                 ret = ocfs2_zero_extend_range(inode, range_start,
1002                                               range_end, di_bh);
1003                 if (ret) {
1004                         mlog_errno(ret);
1005                         break;
1006                 }
1007                 zero_start = range_end;
1008         }
1009
1010         return ret;
1011 }
1012
1013 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1014                           u64 new_i_size, u64 zero_to)
1015 {
1016         int ret;
1017         u32 clusters_to_add;
1018         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1019
1020         /*
1021          * Only quota files call this without a bh, and they can't be
1022          * refcounted.
1023          */
1024         BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1025         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1026
1027         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1028         if (clusters_to_add < oi->ip_clusters)
1029                 clusters_to_add = 0;
1030         else
1031                 clusters_to_add -= oi->ip_clusters;
1032
1033         if (clusters_to_add) {
1034                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1035                                                 clusters_to_add, 0);
1036                 if (ret) {
1037                         mlog_errno(ret);
1038                         goto out;
1039                 }
1040         }
1041
1042         /*
1043          * Call this even if we don't add any clusters to the tree. We
1044          * still need to zero the area between the old i_size and the
1045          * new i_size.
1046          */
1047         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1048         if (ret < 0)
1049                 mlog_errno(ret);
1050
1051 out:
1052         return ret;
1053 }
1054
1055 static int ocfs2_extend_file(struct inode *inode,
1056                              struct buffer_head *di_bh,
1057                              u64 new_i_size)
1058 {
1059         int ret = 0;
1060         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1061
1062         BUG_ON(!di_bh);
1063
1064         /* setattr sometimes calls us like this. */
1065         if (new_i_size == 0)
1066                 goto out;
1067
1068         if (i_size_read(inode) == new_i_size)
1069                 goto out;
1070         BUG_ON(new_i_size < i_size_read(inode));
1071
1072         /*
1073          * The alloc sem blocks people in read/write from reading our
1074          * allocation until we're done changing it. We depend on
1075          * i_mutex to block other extend/truncate calls while we're
1076          * here.  We even have to hold it for sparse files because there
1077          * might be some tail zeroing.
1078          */
1079         down_write(&oi->ip_alloc_sem);
1080
1081         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1082                 /*
1083                  * We can optimize small extends by keeping the inodes
1084                  * inline data.
1085                  */
1086                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1087                         up_write(&oi->ip_alloc_sem);
1088                         goto out_update_size;
1089                 }
1090
1091                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1092                 if (ret) {
1093                         up_write(&oi->ip_alloc_sem);
1094                         mlog_errno(ret);
1095                         goto out;
1096                 }
1097         }
1098
1099         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1100                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1101         else
1102                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1103                                             new_i_size);
1104
1105         up_write(&oi->ip_alloc_sem);
1106
1107         if (ret < 0) {
1108                 mlog_errno(ret);
1109                 goto out;
1110         }
1111
1112 out_update_size:
1113         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1114         if (ret < 0)
1115                 mlog_errno(ret);
1116
1117 out:
1118         return ret;
1119 }
1120
1121 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1122 {
1123         int status = 0, size_change;
1124         struct inode *inode = dentry->d_inode;
1125         struct super_block *sb = inode->i_sb;
1126         struct ocfs2_super *osb = OCFS2_SB(sb);
1127         struct buffer_head *bh = NULL;
1128         handle_t *handle = NULL;
1129         struct dquot *transfer_to[MAXQUOTAS] = { };
1130         int qtype;
1131
1132         trace_ocfs2_setattr(inode, dentry,
1133                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1134                             dentry->d_name.len, dentry->d_name.name,
1135                             attr->ia_valid, attr->ia_mode,
1136                             from_kuid(&init_user_ns, attr->ia_uid),
1137                             from_kgid(&init_user_ns, attr->ia_gid));
1138
1139         /* ensuring we don't even attempt to truncate a symlink */
1140         if (S_ISLNK(inode->i_mode))
1141                 attr->ia_valid &= ~ATTR_SIZE;
1142
1143 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1144                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1145         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1146                 return 0;
1147
1148         status = inode_change_ok(inode, attr);
1149         if (status)
1150                 return status;
1151
1152         if (is_quota_modification(inode, attr))
1153                 dquot_initialize(inode);
1154         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1155         if (size_change) {
1156                 status = ocfs2_rw_lock(inode, 1);
1157                 if (status < 0) {
1158                         mlog_errno(status);
1159                         goto bail;
1160                 }
1161         }
1162
1163         status = ocfs2_inode_lock(inode, &bh, 1);
1164         if (status < 0) {
1165                 if (status != -ENOENT)
1166                         mlog_errno(status);
1167                 goto bail_unlock_rw;
1168         }
1169
1170         if (size_change) {
1171                 status = inode_newsize_ok(inode, attr->ia_size);
1172                 if (status)
1173                         goto bail_unlock;
1174
1175                 inode_dio_wait(inode);
1176
1177                 if (i_size_read(inode) >= attr->ia_size) {
1178                         if (ocfs2_should_order_data(inode)) {
1179                                 status = ocfs2_begin_ordered_truncate(inode,
1180                                                                       attr->ia_size);
1181                                 if (status)
1182                                         goto bail_unlock;
1183                         }
1184                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1185                 } else
1186                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1187                 if (status < 0) {
1188                         if (status != -ENOSPC)
1189                                 mlog_errno(status);
1190                         status = -ENOSPC;
1191                         goto bail_unlock;
1192                 }
1193         }
1194
1195         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1196             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1197                 /*
1198                  * Gather pointers to quota structures so that allocation /
1199                  * freeing of quota structures happens here and not inside
1200                  * dquot_transfer() where we have problems with lock ordering
1201                  */
1202                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1203                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1204                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1205                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1206                         if (!transfer_to[USRQUOTA]) {
1207                                 status = -ESRCH;
1208                                 goto bail_unlock;
1209                         }
1210                 }
1211                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1212                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1213                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1214                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1215                         if (!transfer_to[GRPQUOTA]) {
1216                                 status = -ESRCH;
1217                                 goto bail_unlock;
1218                         }
1219                 }
1220                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1221                                            2 * ocfs2_quota_trans_credits(sb));
1222                 if (IS_ERR(handle)) {
1223                         status = PTR_ERR(handle);
1224                         mlog_errno(status);
1225                         goto bail_unlock;
1226                 }
1227                 status = __dquot_transfer(inode, transfer_to);
1228                 if (status < 0)
1229                         goto bail_commit;
1230         } else {
1231                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1232                 if (IS_ERR(handle)) {
1233                         status = PTR_ERR(handle);
1234                         mlog_errno(status);
1235                         goto bail_unlock;
1236                 }
1237         }
1238
1239         setattr_copy(inode, attr);
1240         mark_inode_dirty(inode);
1241
1242         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1243         if (status < 0)
1244                 mlog_errno(status);
1245
1246 bail_commit:
1247         ocfs2_commit_trans(osb, handle);
1248 bail_unlock:
1249         ocfs2_inode_unlock(inode, 1);
1250 bail_unlock_rw:
1251         if (size_change)
1252                 ocfs2_rw_unlock(inode, 1);
1253 bail:
1254         brelse(bh);
1255
1256         /* Release quota pointers in case we acquired them */
1257         for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1258                 dqput(transfer_to[qtype]);
1259
1260         if (!status && attr->ia_valid & ATTR_MODE) {
1261                 status = posix_acl_chmod(inode, inode->i_mode);
1262                 if (status < 0)
1263                         mlog_errno(status);
1264         }
1265
1266         return status;
1267 }
1268
1269 int ocfs2_getattr(struct vfsmount *mnt,
1270                   struct dentry *dentry,
1271                   struct kstat *stat)
1272 {
1273         struct inode *inode = dentry->d_inode;
1274         struct super_block *sb = dentry->d_inode->i_sb;
1275         struct ocfs2_super *osb = sb->s_fs_info;
1276         int err;
1277
1278         err = ocfs2_inode_revalidate(dentry);
1279         if (err) {
1280                 if (err != -ENOENT)
1281                         mlog_errno(err);
1282                 goto bail;
1283         }
1284
1285         generic_fillattr(inode, stat);
1286
1287         /* We set the blksize from the cluster size for performance */
1288         stat->blksize = osb->s_clustersize;
1289
1290 bail:
1291         return err;
1292 }
1293
1294 int ocfs2_permission(struct inode *inode, int mask)
1295 {
1296         int ret;
1297
1298         if (mask & MAY_NOT_BLOCK)
1299                 return -ECHILD;
1300
1301         ret = ocfs2_inode_lock(inode, NULL, 0);
1302         if (ret) {
1303                 if (ret != -ENOENT)
1304                         mlog_errno(ret);
1305                 goto out;
1306         }
1307
1308         ret = generic_permission(inode, mask);
1309
1310         ocfs2_inode_unlock(inode, 0);
1311 out:
1312         return ret;
1313 }
1314
1315 static int __ocfs2_write_remove_suid(struct inode *inode,
1316                                      struct buffer_head *bh)
1317 {
1318         int ret;
1319         handle_t *handle;
1320         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1321         struct ocfs2_dinode *di;
1322
1323         trace_ocfs2_write_remove_suid(
1324                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1325                         inode->i_mode);
1326
1327         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1328         if (IS_ERR(handle)) {
1329                 ret = PTR_ERR(handle);
1330                 mlog_errno(ret);
1331                 goto out;
1332         }
1333
1334         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1335                                       OCFS2_JOURNAL_ACCESS_WRITE);
1336         if (ret < 0) {
1337                 mlog_errno(ret);
1338                 goto out_trans;
1339         }
1340
1341         inode->i_mode &= ~S_ISUID;
1342         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1343                 inode->i_mode &= ~S_ISGID;
1344
1345         di = (struct ocfs2_dinode *) bh->b_data;
1346         di->i_mode = cpu_to_le16(inode->i_mode);
1347
1348         ocfs2_journal_dirty(handle, bh);
1349
1350 out_trans:
1351         ocfs2_commit_trans(osb, handle);
1352 out:
1353         return ret;
1354 }
1355
1356 /*
1357  * Will look for holes and unwritten extents in the range starting at
1358  * pos for count bytes (inclusive).
1359  */
1360 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1361                                        size_t count)
1362 {
1363         int ret = 0;
1364         unsigned int extent_flags;
1365         u32 cpos, clusters, extent_len, phys_cpos;
1366         struct super_block *sb = inode->i_sb;
1367
1368         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1369         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1370
1371         while (clusters) {
1372                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1373                                          &extent_flags);
1374                 if (ret < 0) {
1375                         mlog_errno(ret);
1376                         goto out;
1377                 }
1378
1379                 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1380                         ret = 1;
1381                         break;
1382                 }
1383
1384                 if (extent_len > clusters)
1385                         extent_len = clusters;
1386
1387                 clusters -= extent_len;
1388                 cpos += extent_len;
1389         }
1390 out:
1391         return ret;
1392 }
1393
1394 static int ocfs2_write_remove_suid(struct inode *inode)
1395 {
1396         int ret;
1397         struct buffer_head *bh = NULL;
1398
1399         ret = ocfs2_read_inode_block(inode, &bh);
1400         if (ret < 0) {
1401                 mlog_errno(ret);
1402                 goto out;
1403         }
1404
1405         ret =  __ocfs2_write_remove_suid(inode, bh);
1406 out:
1407         brelse(bh);
1408         return ret;
1409 }
1410
1411 /*
1412  * Allocate enough extents to cover the region starting at byte offset
1413  * start for len bytes. Existing extents are skipped, any extents
1414  * added are marked as "unwritten".
1415  */
1416 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1417                                             u64 start, u64 len)
1418 {
1419         int ret;
1420         u32 cpos, phys_cpos, clusters, alloc_size;
1421         u64 end = start + len;
1422         struct buffer_head *di_bh = NULL;
1423
1424         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1425                 ret = ocfs2_read_inode_block(inode, &di_bh);
1426                 if (ret) {
1427                         mlog_errno(ret);
1428                         goto out;
1429                 }
1430
1431                 /*
1432                  * Nothing to do if the requested reservation range
1433                  * fits within the inode.
1434                  */
1435                 if (ocfs2_size_fits_inline_data(di_bh, end))
1436                         goto out;
1437
1438                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1439                 if (ret) {
1440                         mlog_errno(ret);
1441                         goto out;
1442                 }
1443         }
1444
1445         /*
1446          * We consider both start and len to be inclusive.
1447          */
1448         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1449         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1450         clusters -= cpos;
1451
1452         while (clusters) {
1453                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1454                                          &alloc_size, NULL);
1455                 if (ret) {
1456                         mlog_errno(ret);
1457                         goto out;
1458                 }
1459
1460                 /*
1461                  * Hole or existing extent len can be arbitrary, so
1462                  * cap it to our own allocation request.
1463                  */
1464                 if (alloc_size > clusters)
1465                         alloc_size = clusters;
1466
1467                 if (phys_cpos) {
1468                         /*
1469                          * We already have an allocation at this
1470                          * region so we can safely skip it.
1471                          */
1472                         goto next;
1473                 }
1474
1475                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1476                 if (ret) {
1477                         if (ret != -ENOSPC)
1478                                 mlog_errno(ret);
1479                         goto out;
1480                 }
1481
1482 next:
1483                 cpos += alloc_size;
1484                 clusters -= alloc_size;
1485         }
1486
1487         ret = 0;
1488 out:
1489
1490         brelse(di_bh);
1491         return ret;
1492 }
1493
1494 /*
1495  * Truncate a byte range, avoiding pages within partial clusters. This
1496  * preserves those pages for the zeroing code to write to.
1497  */
1498 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1499                                          u64 byte_len)
1500 {
1501         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1502         loff_t start, end;
1503         struct address_space *mapping = inode->i_mapping;
1504
1505         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1506         end = byte_start + byte_len;
1507         end = end & ~(osb->s_clustersize - 1);
1508
1509         if (start < end) {
1510                 unmap_mapping_range(mapping, start, end - start, 0);
1511                 truncate_inode_pages_range(mapping, start, end - 1);
1512         }
1513 }
1514
1515 static int ocfs2_zero_partial_clusters(struct inode *inode,
1516                                        u64 start, u64 len)
1517 {
1518         int ret = 0;
1519         u64 tmpend, end = start + len;
1520         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1521         unsigned int csize = osb->s_clustersize;
1522         handle_t *handle;
1523
1524         /*
1525          * The "start" and "end" values are NOT necessarily part of
1526          * the range whose allocation is being deleted. Rather, this
1527          * is what the user passed in with the request. We must zero
1528          * partial clusters here. There's no need to worry about
1529          * physical allocation - the zeroing code knows to skip holes.
1530          */
1531         trace_ocfs2_zero_partial_clusters(
1532                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1533                 (unsigned long long)start, (unsigned long long)end);
1534
1535         /*
1536          * If both edges are on a cluster boundary then there's no
1537          * zeroing required as the region is part of the allocation to
1538          * be truncated.
1539          */
1540         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1541                 goto out;
1542
1543         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1544         if (IS_ERR(handle)) {
1545                 ret = PTR_ERR(handle);
1546                 mlog_errno(ret);
1547                 goto out;
1548         }
1549
1550         /*
1551          * We want to get the byte offset of the end of the 1st cluster.
1552          */
1553         tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1554         if (tmpend > end)
1555                 tmpend = end;
1556
1557         trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1558                                                  (unsigned long long)tmpend);
1559
1560         ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1561         if (ret)
1562                 mlog_errno(ret);
1563
1564         if (tmpend < end) {
1565                 /*
1566                  * This may make start and end equal, but the zeroing
1567                  * code will skip any work in that case so there's no
1568                  * need to catch it up here.
1569                  */
1570                 start = end & ~(osb->s_clustersize - 1);
1571
1572                 trace_ocfs2_zero_partial_clusters_range2(
1573                         (unsigned long long)start, (unsigned long long)end);
1574
1575                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1576                 if (ret)
1577                         mlog_errno(ret);
1578         }
1579
1580         ocfs2_commit_trans(osb, handle);
1581 out:
1582         return ret;
1583 }
1584
1585 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1586 {
1587         int i;
1588         struct ocfs2_extent_rec *rec = NULL;
1589
1590         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1591
1592                 rec = &el->l_recs[i];
1593
1594                 if (le32_to_cpu(rec->e_cpos) < pos)
1595                         break;
1596         }
1597
1598         return i;
1599 }
1600
1601 /*
1602  * Helper to calculate the punching pos and length in one run, we handle the
1603  * following three cases in order:
1604  *
1605  * - remove the entire record
1606  * - remove a partial record
1607  * - no record needs to be removed (hole-punching completed)
1608 */
1609 static void ocfs2_calc_trunc_pos(struct inode *inode,
1610                                  struct ocfs2_extent_list *el,
1611                                  struct ocfs2_extent_rec *rec,
1612                                  u32 trunc_start, u32 *trunc_cpos,
1613                                  u32 *trunc_len, u32 *trunc_end,
1614                                  u64 *blkno, int *done)
1615 {
1616         int ret = 0;
1617         u32 coff, range;
1618
1619         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1620
1621         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1622                 /*
1623                  * remove an entire extent record.
1624                  */
1625                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1626                 /*
1627                  * Skip holes if any.
1628                  */
1629                 if (range < *trunc_end)
1630                         *trunc_end = range;
1631                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1632                 *blkno = le64_to_cpu(rec->e_blkno);
1633                 *trunc_end = le32_to_cpu(rec->e_cpos);
1634         } else if (range > trunc_start) {
1635                 /*
1636                  * remove a partial extent record, which means we're
1637                  * removing the last extent record.
1638                  */
1639                 *trunc_cpos = trunc_start;
1640                 /*
1641                  * skip hole if any.
1642                  */
1643                 if (range < *trunc_end)
1644                         *trunc_end = range;
1645                 *trunc_len = *trunc_end - trunc_start;
1646                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1647                 *blkno = le64_to_cpu(rec->e_blkno) +
1648                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1649                 *trunc_end = trunc_start;
1650         } else {
1651                 /*
1652                  * It may have two following possibilities:
1653                  *
1654                  * - last record has been removed
1655                  * - trunc_start was within a hole
1656                  *
1657                  * both two cases mean the completion of hole punching.
1658                  */
1659                 ret = 1;
1660         }
1661
1662         *done = ret;
1663 }
1664
1665 static int ocfs2_remove_inode_range(struct inode *inode,
1666                                     struct buffer_head *di_bh, u64 byte_start,
1667                                     u64 byte_len)
1668 {
1669         int ret = 0, flags = 0, done = 0, i;
1670         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1671         u32 cluster_in_el;
1672         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1673         struct ocfs2_cached_dealloc_ctxt dealloc;
1674         struct address_space *mapping = inode->i_mapping;
1675         struct ocfs2_extent_tree et;
1676         struct ocfs2_path *path = NULL;
1677         struct ocfs2_extent_list *el = NULL;
1678         struct ocfs2_extent_rec *rec = NULL;
1679         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1680         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1681
1682         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1683         ocfs2_init_dealloc_ctxt(&dealloc);
1684
1685         trace_ocfs2_remove_inode_range(
1686                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1687                         (unsigned long long)byte_start,
1688                         (unsigned long long)byte_len);
1689
1690         if (byte_len == 0)
1691                 return 0;
1692
1693         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1694                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1695                                             byte_start + byte_len, 0);
1696                 if (ret) {
1697                         mlog_errno(ret);
1698                         goto out;
1699                 }
1700                 /*
1701                  * There's no need to get fancy with the page cache
1702                  * truncate of an inline-data inode. We're talking
1703                  * about less than a page here, which will be cached
1704                  * in the dinode buffer anyway.
1705                  */
1706                 unmap_mapping_range(mapping, 0, 0, 0);
1707                 truncate_inode_pages(mapping, 0);
1708                 goto out;
1709         }
1710
1711         /*
1712          * For reflinks, we may need to CoW 2 clusters which might be
1713          * partially zero'd later, if hole's start and end offset were
1714          * within one cluster(means is not exactly aligned to clustersize).
1715          */
1716
1717         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1718
1719                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1720                 if (ret) {
1721                         mlog_errno(ret);
1722                         goto out;
1723                 }
1724
1725                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1726                 if (ret) {
1727                         mlog_errno(ret);
1728                         goto out;
1729                 }
1730         }
1731
1732         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1733         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1734         cluster_in_el = trunc_end;
1735
1736         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1737         if (ret) {
1738                 mlog_errno(ret);
1739                 goto out;
1740         }
1741
1742         path = ocfs2_new_path_from_et(&et);
1743         if (!path) {
1744                 ret = -ENOMEM;
1745                 mlog_errno(ret);
1746                 goto out;
1747         }
1748
1749         while (trunc_end > trunc_start) {
1750
1751                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1752                                       cluster_in_el);
1753                 if (ret) {
1754                         mlog_errno(ret);
1755                         goto out;
1756                 }
1757
1758                 el = path_leaf_el(path);
1759
1760                 i = ocfs2_find_rec(el, trunc_end);
1761                 /*
1762                  * Need to go to previous extent block.
1763                  */
1764                 if (i < 0) {
1765                         if (path->p_tree_depth == 0)
1766                                 break;
1767
1768                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1769                                                             path,
1770                                                             &cluster_in_el);
1771                         if (ret) {
1772                                 mlog_errno(ret);
1773                                 goto out;
1774                         }
1775
1776                         /*
1777                          * We've reached the leftmost extent block,
1778                          * it's safe to leave.
1779                          */
1780                         if (cluster_in_el == 0)
1781                                 break;
1782
1783                         /*
1784                          * The 'pos' searched for previous extent block is
1785                          * always one cluster less than actual trunc_end.
1786                          */
1787                         trunc_end = cluster_in_el + 1;
1788
1789                         ocfs2_reinit_path(path, 1);
1790
1791                         continue;
1792
1793                 } else
1794                         rec = &el->l_recs[i];
1795
1796                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1797                                      &trunc_len, &trunc_end, &blkno, &done);
1798                 if (done)
1799                         break;
1800
1801                 flags = rec->e_flags;
1802                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1803
1804                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1805                                                phys_cpos, trunc_len, flags,
1806                                                &dealloc, refcount_loc);
1807                 if (ret < 0) {
1808                         mlog_errno(ret);
1809                         goto out;
1810                 }
1811
1812                 cluster_in_el = trunc_end;
1813
1814                 ocfs2_reinit_path(path, 1);
1815         }
1816
1817         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1818
1819 out:
1820         ocfs2_free_path(path);
1821         ocfs2_schedule_truncate_log_flush(osb, 1);
1822         ocfs2_run_deallocs(osb, &dealloc);
1823
1824         return ret;
1825 }
1826
1827 /*
1828  * Parts of this function taken from xfs_change_file_space()
1829  */
1830 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1831                                      loff_t f_pos, unsigned int cmd,
1832                                      struct ocfs2_space_resv *sr,
1833                                      int change_size)
1834 {
1835         int ret;
1836         s64 llen;
1837         loff_t size;
1838         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1839         struct buffer_head *di_bh = NULL;
1840         handle_t *handle;
1841         unsigned long long max_off = inode->i_sb->s_maxbytes;
1842
1843         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1844                 return -EROFS;
1845
1846         mutex_lock(&inode->i_mutex);
1847
1848         /*
1849          * This prevents concurrent writes on other nodes
1850          */
1851         ret = ocfs2_rw_lock(inode, 1);
1852         if (ret) {
1853                 mlog_errno(ret);
1854                 goto out;
1855         }
1856
1857         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1858         if (ret) {
1859                 mlog_errno(ret);
1860                 goto out_rw_unlock;
1861         }
1862
1863         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1864                 ret = -EPERM;
1865                 goto out_inode_unlock;
1866         }
1867
1868         switch (sr->l_whence) {
1869         case 0: /*SEEK_SET*/
1870                 break;
1871         case 1: /*SEEK_CUR*/
1872                 sr->l_start += f_pos;
1873                 break;
1874         case 2: /*SEEK_END*/
1875                 sr->l_start += i_size_read(inode);
1876                 break;
1877         default:
1878                 ret = -EINVAL;
1879                 goto out_inode_unlock;
1880         }
1881         sr->l_whence = 0;
1882
1883         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1884
1885         if (sr->l_start < 0
1886             || sr->l_start > max_off
1887             || (sr->l_start + llen) < 0
1888             || (sr->l_start + llen) > max_off) {
1889                 ret = -EINVAL;
1890                 goto out_inode_unlock;
1891         }
1892         size = sr->l_start + sr->l_len;
1893
1894         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1895             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1896                 if (sr->l_len <= 0) {
1897                         ret = -EINVAL;
1898                         goto out_inode_unlock;
1899                 }
1900         }
1901
1902         if (file && should_remove_suid(file->f_path.dentry)) {
1903                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1904                 if (ret) {
1905                         mlog_errno(ret);
1906                         goto out_inode_unlock;
1907                 }
1908         }
1909
1910         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1911         switch (cmd) {
1912         case OCFS2_IOC_RESVSP:
1913         case OCFS2_IOC_RESVSP64:
1914                 /*
1915                  * This takes unsigned offsets, but the signed ones we
1916                  * pass have been checked against overflow above.
1917                  */
1918                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1919                                                        sr->l_len);
1920                 break;
1921         case OCFS2_IOC_UNRESVSP:
1922         case OCFS2_IOC_UNRESVSP64:
1923                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1924                                                sr->l_len);
1925                 break;
1926         default:
1927                 ret = -EINVAL;
1928         }
1929         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1930         if (ret) {
1931                 mlog_errno(ret);
1932                 goto out_inode_unlock;
1933         }
1934
1935         /*
1936          * We update c/mtime for these changes
1937          */
1938         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1939         if (IS_ERR(handle)) {
1940                 ret = PTR_ERR(handle);
1941                 mlog_errno(ret);
1942                 goto out_inode_unlock;
1943         }
1944
1945         if (change_size && i_size_read(inode) < size)
1946                 i_size_write(inode, size);
1947
1948         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1949         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1950         if (ret < 0)
1951                 mlog_errno(ret);
1952
1953         if (file && (file->f_flags & O_SYNC))
1954                 handle->h_sync = 1;
1955
1956         ocfs2_commit_trans(osb, handle);
1957
1958 out_inode_unlock:
1959         brelse(di_bh);
1960         ocfs2_inode_unlock(inode, 1);
1961 out_rw_unlock:
1962         ocfs2_rw_unlock(inode, 1);
1963
1964 out:
1965         mutex_unlock(&inode->i_mutex);
1966         return ret;
1967 }
1968
1969 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1970                             struct ocfs2_space_resv *sr)
1971 {
1972         struct inode *inode = file_inode(file);
1973         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1974         int ret;
1975
1976         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1977             !ocfs2_writes_unwritten_extents(osb))
1978                 return -ENOTTY;
1979         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1980                  !ocfs2_sparse_alloc(osb))
1981                 return -ENOTTY;
1982
1983         if (!S_ISREG(inode->i_mode))
1984                 return -EINVAL;
1985
1986         if (!(file->f_mode & FMODE_WRITE))
1987                 return -EBADF;
1988
1989         ret = mnt_want_write_file(file);
1990         if (ret)
1991                 return ret;
1992         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1993         mnt_drop_write_file(file);
1994         return ret;
1995 }
1996
1997 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1998                             loff_t len)
1999 {
2000         struct inode *inode = file_inode(file);
2001         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2002         struct ocfs2_space_resv sr;
2003         int change_size = 1;
2004         int cmd = OCFS2_IOC_RESVSP64;
2005
2006         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2007                 return -EOPNOTSUPP;
2008         if (!ocfs2_writes_unwritten_extents(osb))
2009                 return -EOPNOTSUPP;
2010
2011         if (mode & FALLOC_FL_KEEP_SIZE)
2012                 change_size = 0;
2013
2014         if (mode & FALLOC_FL_PUNCH_HOLE)
2015                 cmd = OCFS2_IOC_UNRESVSP64;
2016
2017         sr.l_whence = 0;
2018         sr.l_start = (s64)offset;
2019         sr.l_len = (s64)len;
2020
2021         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2022                                          change_size);
2023 }
2024
2025 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2026                                    size_t count)
2027 {
2028         int ret = 0;
2029         unsigned int extent_flags;
2030         u32 cpos, clusters, extent_len, phys_cpos;
2031         struct super_block *sb = inode->i_sb;
2032
2033         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2034             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2035             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2036                 return 0;
2037
2038         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2039         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2040
2041         while (clusters) {
2042                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2043                                          &extent_flags);
2044                 if (ret < 0) {
2045                         mlog_errno(ret);
2046                         goto out;
2047                 }
2048
2049                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2050                         ret = 1;
2051                         break;
2052                 }
2053
2054                 if (extent_len > clusters)
2055                         extent_len = clusters;
2056
2057                 clusters -= extent_len;
2058                 cpos += extent_len;
2059         }
2060 out:
2061         return ret;
2062 }
2063
2064 static void ocfs2_aiodio_wait(struct inode *inode)
2065 {
2066         wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
2067
2068         wait_event(*wq, (atomic_read(&OCFS2_I(inode)->ip_unaligned_aio) == 0));
2069 }
2070
2071 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2072 {
2073         int blockmask = inode->i_sb->s_blocksize - 1;
2074         loff_t final_size = pos + count;
2075
2076         if ((pos & blockmask) || (final_size & blockmask))
2077                 return 1;
2078         return 0;
2079 }
2080
2081 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2082                                             struct file *file,
2083                                             loff_t pos, size_t count,
2084                                             int *meta_level)
2085 {
2086         int ret;
2087         struct buffer_head *di_bh = NULL;
2088         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2089         u32 clusters =
2090                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2091
2092         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2093         if (ret) {
2094                 mlog_errno(ret);
2095                 goto out;
2096         }
2097
2098         *meta_level = 1;
2099
2100         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2101         if (ret)
2102                 mlog_errno(ret);
2103 out:
2104         brelse(di_bh);
2105         return ret;
2106 }
2107
2108 static int ocfs2_prepare_inode_for_write(struct file *file,
2109                                          loff_t *ppos,
2110                                          size_t count,
2111                                          int appending,
2112                                          int *direct_io,
2113                                          int *has_refcount)
2114 {
2115         int ret = 0, meta_level = 0;
2116         struct dentry *dentry = file->f_path.dentry;
2117         struct inode *inode = dentry->d_inode;
2118         loff_t saved_pos = 0, end;
2119
2120         /*
2121          * We start with a read level meta lock and only jump to an ex
2122          * if we need to make modifications here.
2123          */
2124         for(;;) {
2125                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2126                 if (ret < 0) {
2127                         meta_level = -1;
2128                         mlog_errno(ret);
2129                         goto out;
2130                 }
2131
2132                 /* Clear suid / sgid if necessary. We do this here
2133                  * instead of later in the write path because
2134                  * remove_suid() calls ->setattr without any hint that
2135                  * we may have already done our cluster locking. Since
2136                  * ocfs2_setattr() *must* take cluster locks to
2137                  * proceed, this will lead us to recursively lock the
2138                  * inode. There's also the dinode i_size state which
2139                  * can be lost via setattr during extending writes (we
2140                  * set inode->i_size at the end of a write. */
2141                 if (should_remove_suid(dentry)) {
2142                         if (meta_level == 0) {
2143                                 ocfs2_inode_unlock(inode, meta_level);
2144                                 meta_level = 1;
2145                                 continue;
2146                         }
2147
2148                         ret = ocfs2_write_remove_suid(inode);
2149                         if (ret < 0) {
2150                                 mlog_errno(ret);
2151                                 goto out_unlock;
2152                         }
2153                 }
2154
2155                 /* work on a copy of ppos until we're sure that we won't have
2156                  * to recalculate it due to relocking. */
2157                 if (appending)
2158                         saved_pos = i_size_read(inode);
2159                 else
2160                         saved_pos = *ppos;
2161
2162                 end = saved_pos + count;
2163
2164                 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2165                 if (ret == 1) {
2166                         ocfs2_inode_unlock(inode, meta_level);
2167                         meta_level = -1;
2168
2169                         ret = ocfs2_prepare_inode_for_refcount(inode,
2170                                                                file,
2171                                                                saved_pos,
2172                                                                count,
2173                                                                &meta_level);
2174                         if (has_refcount)
2175                                 *has_refcount = 1;
2176                         if (direct_io)
2177                                 *direct_io = 0;
2178                 }
2179
2180                 if (ret < 0) {
2181                         mlog_errno(ret);
2182                         goto out_unlock;
2183                 }
2184
2185                 /*
2186                  * Skip the O_DIRECT checks if we don't need
2187                  * them.
2188                  */
2189                 if (!direct_io || !(*direct_io))
2190                         break;
2191
2192                 /*
2193                  * There's no sane way to do direct writes to an inode
2194                  * with inline data.
2195                  */
2196                 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2197                         *direct_io = 0;
2198                         break;
2199                 }
2200
2201                 /*
2202                  * Allowing concurrent direct writes means
2203                  * i_size changes wouldn't be synchronized, so
2204                  * one node could wind up truncating another
2205                  * nodes writes.
2206                  */
2207                 if (end > i_size_read(inode)) {
2208                         *direct_io = 0;
2209                         break;
2210                 }
2211
2212                 /*
2213                  * We don't fill holes during direct io, so
2214                  * check for them here. If any are found, the
2215                  * caller will have to retake some cluster
2216                  * locks and initiate the io as buffered.
2217                  */
2218                 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2219                 if (ret == 1) {
2220                         *direct_io = 0;
2221                         ret = 0;
2222                 } else if (ret < 0)
2223                         mlog_errno(ret);
2224                 break;
2225         }
2226
2227         if (appending)
2228                 *ppos = saved_pos;
2229
2230 out_unlock:
2231         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2232                                             saved_pos, appending, count,
2233                                             direct_io, has_refcount);
2234
2235         if (meta_level >= 0)
2236                 ocfs2_inode_unlock(inode, meta_level);
2237
2238 out:
2239         return ret;
2240 }
2241
2242 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2243                                     const struct iovec *iov,
2244                                     unsigned long nr_segs,
2245                                     loff_t pos)
2246 {
2247         int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
2248         int can_do_direct, has_refcount = 0;
2249         ssize_t written = 0;
2250         size_t ocount;          /* original count */
2251         size_t count;           /* after file limit checks */
2252         loff_t old_size, *ppos = &iocb->ki_pos;
2253         u32 old_clusters;
2254         struct file *file = iocb->ki_filp;
2255         struct inode *inode = file_inode(file);
2256         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2257         int full_coherency = !(osb->s_mount_opt &
2258                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2259         int unaligned_dio = 0;
2260
2261         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2262                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2263                 file->f_path.dentry->d_name.len,
2264                 file->f_path.dentry->d_name.name,
2265                 (unsigned int)nr_segs);
2266
2267         if (iocb->ki_nbytes == 0)
2268                 return 0;
2269
2270         appending = file->f_flags & O_APPEND ? 1 : 0;
2271         direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2272
2273         mutex_lock(&inode->i_mutex);
2274
2275         ocfs2_iocb_clear_sem_locked(iocb);
2276
2277 relock:
2278         /* to match setattr's i_mutex -> rw_lock ordering */
2279         if (direct_io) {
2280                 have_alloc_sem = 1;
2281                 /* communicate with ocfs2_dio_end_io */
2282                 ocfs2_iocb_set_sem_locked(iocb);
2283         }
2284
2285         /*
2286          * Concurrent O_DIRECT writes are allowed with
2287          * mount_option "coherency=buffered".
2288          */
2289         rw_level = (!direct_io || full_coherency);
2290
2291         ret = ocfs2_rw_lock(inode, rw_level);
2292         if (ret < 0) {
2293                 mlog_errno(ret);
2294                 goto out_sems;
2295         }
2296
2297         /*
2298          * O_DIRECT writes with "coherency=full" need to take EX cluster
2299          * inode_lock to guarantee coherency.
2300          */
2301         if (direct_io && full_coherency) {
2302                 /*
2303                  * We need to take and drop the inode lock to force
2304                  * other nodes to drop their caches.  Buffered I/O
2305                  * already does this in write_begin().
2306                  */
2307                 ret = ocfs2_inode_lock(inode, NULL, 1);
2308                 if (ret < 0) {
2309                         mlog_errno(ret);
2310                         goto out;
2311                 }
2312
2313                 ocfs2_inode_unlock(inode, 1);
2314         }
2315
2316         can_do_direct = direct_io;
2317         ret = ocfs2_prepare_inode_for_write(file, ppos,
2318                                             iocb->ki_nbytes, appending,
2319                                             &can_do_direct, &has_refcount);
2320         if (ret < 0) {
2321                 mlog_errno(ret);
2322                 goto out;
2323         }
2324
2325         if (direct_io && !is_sync_kiocb(iocb))
2326                 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_nbytes,
2327                                                       *ppos);
2328
2329         /*
2330          * We can't complete the direct I/O as requested, fall back to
2331          * buffered I/O.
2332          */
2333         if (direct_io && !can_do_direct) {
2334                 ocfs2_rw_unlock(inode, rw_level);
2335
2336                 have_alloc_sem = 0;
2337                 rw_level = -1;
2338
2339                 direct_io = 0;
2340                 goto relock;
2341         }
2342
2343         if (unaligned_dio) {
2344                 /*
2345                  * Wait on previous unaligned aio to complete before
2346                  * proceeding.
2347                  */
2348                 ocfs2_aiodio_wait(inode);
2349
2350                 /* Mark the iocb as needing a decrement in ocfs2_dio_end_io */
2351                 atomic_inc(&OCFS2_I(inode)->ip_unaligned_aio);
2352                 ocfs2_iocb_set_unaligned_aio(iocb);
2353         }
2354
2355         /*
2356          * To later detect whether a journal commit for sync writes is
2357          * necessary, we sample i_size, and cluster count here.
2358          */
2359         old_size = i_size_read(inode);
2360         old_clusters = OCFS2_I(inode)->ip_clusters;
2361
2362         /* communicate with ocfs2_dio_end_io */
2363         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2364
2365         ret = generic_segment_checks(iov, &nr_segs, &ocount,
2366                                      VERIFY_READ);
2367         if (ret)
2368                 goto out_dio;
2369
2370         count = ocount;
2371         ret = generic_write_checks(file, ppos, &count,
2372                                    S_ISBLK(inode->i_mode));
2373         if (ret)
2374                 goto out_dio;
2375
2376         if (direct_io) {
2377                 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2378                                                     ppos, count, ocount);
2379                 if (written < 0) {
2380                         ret = written;
2381                         goto out_dio;
2382                 }
2383         } else {
2384                 current->backing_dev_info = file->f_mapping->backing_dev_info;
2385                 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2386                                                       ppos, count, 0);
2387                 current->backing_dev_info = NULL;
2388         }
2389
2390 out_dio:
2391         /* buffered aio wouldn't have proper lock coverage today */
2392         BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2393
2394         if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2395             ((file->f_flags & O_DIRECT) && !direct_io)) {
2396                 ret = filemap_fdatawrite_range(file->f_mapping, *ppos,
2397                                                *ppos + count - 1);
2398                 if (ret < 0)
2399                         written = ret;
2400
2401                 if (!ret && ((old_size != i_size_read(inode)) ||
2402                              (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2403                              has_refcount)) {
2404                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2405                         if (ret < 0)
2406                                 written = ret;
2407                 }
2408
2409                 if (!ret)
2410                         ret = filemap_fdatawait_range(file->f_mapping, *ppos,
2411                                                       *ppos + count - 1);
2412         }
2413
2414         /*
2415          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2416          * function pointer which is called when o_direct io completes so that
2417          * it can unlock our rw lock.
2418          * Unfortunately there are error cases which call end_io and others
2419          * that don't.  so we don't have to unlock the rw_lock if either an
2420          * async dio is going to do it in the future or an end_io after an
2421          * error has already done it.
2422          */
2423         if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2424                 rw_level = -1;
2425                 have_alloc_sem = 0;
2426                 unaligned_dio = 0;
2427         }
2428
2429         if (unaligned_dio) {
2430                 ocfs2_iocb_clear_unaligned_aio(iocb);
2431                 atomic_dec(&OCFS2_I(inode)->ip_unaligned_aio);
2432         }
2433
2434 out:
2435         if (rw_level != -1)
2436                 ocfs2_rw_unlock(inode, rw_level);
2437
2438 out_sems:
2439         if (have_alloc_sem)
2440                 ocfs2_iocb_clear_sem_locked(iocb);
2441
2442         mutex_unlock(&inode->i_mutex);
2443
2444         if (written)
2445                 ret = written;
2446         return ret;
2447 }
2448
2449 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2450                                 struct file *out,
2451                                 struct splice_desc *sd)
2452 {
2453         int ret;
2454
2455         ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2456                                             sd->total_len, 0, NULL, NULL);
2457         if (ret < 0) {
2458                 mlog_errno(ret);
2459                 return ret;
2460         }
2461
2462         return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2463 }
2464
2465 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2466                                        struct file *out,
2467                                        loff_t *ppos,
2468                                        size_t len,
2469                                        unsigned int flags)
2470 {
2471         int ret;
2472         struct address_space *mapping = out->f_mapping;
2473         struct inode *inode = mapping->host;
2474         struct splice_desc sd = {
2475                 .total_len = len,
2476                 .flags = flags,
2477                 .pos = *ppos,
2478                 .u.file = out,
2479         };
2480
2481
2482         trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2483                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2484                         out->f_path.dentry->d_name.len,
2485                         out->f_path.dentry->d_name.name, len);
2486
2487         pipe_lock(pipe);
2488
2489         splice_from_pipe_begin(&sd);
2490         do {
2491                 ret = splice_from_pipe_next(pipe, &sd);
2492                 if (ret <= 0)
2493                         break;
2494
2495                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2496                 ret = ocfs2_rw_lock(inode, 1);
2497                 if (ret < 0)
2498                         mlog_errno(ret);
2499                 else {
2500                         ret = ocfs2_splice_to_file(pipe, out, &sd);
2501                         ocfs2_rw_unlock(inode, 1);
2502                 }
2503                 mutex_unlock(&inode->i_mutex);
2504         } while (ret > 0);
2505         splice_from_pipe_end(pipe, &sd);
2506
2507         pipe_unlock(pipe);
2508
2509         if (sd.num_spliced)
2510                 ret = sd.num_spliced;
2511
2512         if (ret > 0) {
2513                 int err;
2514
2515                 err = generic_write_sync(out, *ppos, ret);
2516                 if (err)
2517                         ret = err;
2518                 else
2519                         *ppos += ret;
2520
2521                 balance_dirty_pages_ratelimited(mapping);
2522         }
2523
2524         return ret;
2525 }
2526
2527 static ssize_t ocfs2_file_splice_read(struct file *in,
2528                                       loff_t *ppos,
2529                                       struct pipe_inode_info *pipe,
2530                                       size_t len,
2531                                       unsigned int flags)
2532 {
2533         int ret = 0, lock_level = 0;
2534         struct inode *inode = file_inode(in);
2535
2536         trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2537                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2538                         in->f_path.dentry->d_name.len,
2539                         in->f_path.dentry->d_name.name, len);
2540
2541         /*
2542          * See the comment in ocfs2_file_aio_read()
2543          */
2544         ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2545         if (ret < 0) {
2546                 mlog_errno(ret);
2547                 goto bail;
2548         }
2549         ocfs2_inode_unlock(inode, lock_level);
2550
2551         ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2552
2553 bail:
2554         return ret;
2555 }
2556
2557 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2558                                    const struct iovec *iov,
2559                                    unsigned long nr_segs,
2560                                    loff_t pos)
2561 {
2562         int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2563         struct file *filp = iocb->ki_filp;
2564         struct inode *inode = file_inode(filp);
2565
2566         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2567                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2568                         filp->f_path.dentry->d_name.len,
2569                         filp->f_path.dentry->d_name.name, nr_segs);
2570
2571
2572         if (!inode) {
2573                 ret = -EINVAL;
2574                 mlog_errno(ret);
2575                 goto bail;
2576         }
2577
2578         ocfs2_iocb_clear_sem_locked(iocb);
2579
2580         /*
2581          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2582          * need locks to protect pending reads from racing with truncate.
2583          */
2584         if (filp->f_flags & O_DIRECT) {
2585                 have_alloc_sem = 1;
2586                 ocfs2_iocb_set_sem_locked(iocb);
2587
2588                 ret = ocfs2_rw_lock(inode, 0);
2589                 if (ret < 0) {
2590                         mlog_errno(ret);
2591                         goto bail;
2592                 }
2593                 rw_level = 0;
2594                 /* communicate with ocfs2_dio_end_io */
2595                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2596         }
2597
2598         /*
2599          * We're fine letting folks race truncates and extending
2600          * writes with read across the cluster, just like they can
2601          * locally. Hence no rw_lock during read.
2602          *
2603          * Take and drop the meta data lock to update inode fields
2604          * like i_size. This allows the checks down below
2605          * generic_file_aio_read() a chance of actually working.
2606          */
2607         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2608         if (ret < 0) {
2609                 mlog_errno(ret);
2610                 goto bail;
2611         }
2612         ocfs2_inode_unlock(inode, lock_level);
2613
2614         ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2615         trace_generic_file_aio_read_ret(ret);
2616
2617         /* buffered aio wouldn't have proper lock coverage today */
2618         BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2619
2620         /* see ocfs2_file_aio_write */
2621         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2622                 rw_level = -1;
2623                 have_alloc_sem = 0;
2624         }
2625
2626 bail:
2627         if (have_alloc_sem)
2628                 ocfs2_iocb_clear_sem_locked(iocb);
2629
2630         if (rw_level != -1)
2631                 ocfs2_rw_unlock(inode, rw_level);
2632
2633         return ret;
2634 }
2635
2636 /* Refer generic_file_llseek_unlocked() */
2637 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2638 {
2639         struct inode *inode = file->f_mapping->host;
2640         int ret = 0;
2641
2642         mutex_lock(&inode->i_mutex);
2643
2644         switch (whence) {
2645         case SEEK_SET:
2646                 break;
2647         case SEEK_END:
2648                 offset += inode->i_size;
2649                 break;
2650         case SEEK_CUR:
2651                 if (offset == 0) {
2652                         offset = file->f_pos;
2653                         goto out;
2654                 }
2655                 offset += file->f_pos;
2656                 break;
2657         case SEEK_DATA:
2658         case SEEK_HOLE:
2659                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2660                 if (ret)
2661                         goto out;
2662                 break;
2663         default:
2664                 ret = -EINVAL;
2665                 goto out;
2666         }
2667
2668         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2669
2670 out:
2671         mutex_unlock(&inode->i_mutex);
2672         if (ret)
2673                 return ret;
2674         return offset;
2675 }
2676
2677 const struct inode_operations ocfs2_file_iops = {
2678         .setattr        = ocfs2_setattr,
2679         .getattr        = ocfs2_getattr,
2680         .permission     = ocfs2_permission,
2681         .setxattr       = generic_setxattr,
2682         .getxattr       = generic_getxattr,
2683         .listxattr      = ocfs2_listxattr,
2684         .removexattr    = generic_removexattr,
2685         .fiemap         = ocfs2_fiemap,
2686         .get_acl        = ocfs2_iop_get_acl,
2687         .set_acl        = ocfs2_iop_set_acl,
2688 };
2689
2690 const struct inode_operations ocfs2_special_file_iops = {
2691         .setattr        = ocfs2_setattr,
2692         .getattr        = ocfs2_getattr,
2693         .permission     = ocfs2_permission,
2694         .get_acl        = ocfs2_iop_get_acl,
2695         .set_acl        = ocfs2_iop_set_acl,
2696 };
2697
2698 /*
2699  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2700  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2701  */
2702 const struct file_operations ocfs2_fops = {
2703         .llseek         = ocfs2_file_llseek,
2704         .read           = do_sync_read,
2705         .write          = do_sync_write,
2706         .mmap           = ocfs2_mmap,
2707         .fsync          = ocfs2_sync_file,
2708         .release        = ocfs2_file_release,
2709         .open           = ocfs2_file_open,
2710         .aio_read       = ocfs2_file_aio_read,
2711         .aio_write      = ocfs2_file_aio_write,
2712         .unlocked_ioctl = ocfs2_ioctl,
2713 #ifdef CONFIG_COMPAT
2714         .compat_ioctl   = ocfs2_compat_ioctl,
2715 #endif
2716         .lock           = ocfs2_lock,
2717         .flock          = ocfs2_flock,
2718         .splice_read    = ocfs2_file_splice_read,
2719         .splice_write   = ocfs2_file_splice_write,
2720         .fallocate      = ocfs2_fallocate,
2721 };
2722
2723 const struct file_operations ocfs2_dops = {
2724         .llseek         = generic_file_llseek,
2725         .read           = generic_read_dir,
2726         .iterate        = ocfs2_readdir,
2727         .fsync          = ocfs2_sync_file,
2728         .release        = ocfs2_dir_release,
2729         .open           = ocfs2_dir_open,
2730         .unlocked_ioctl = ocfs2_ioctl,
2731 #ifdef CONFIG_COMPAT
2732         .compat_ioctl   = ocfs2_compat_ioctl,
2733 #endif
2734         .lock           = ocfs2_lock,
2735         .flock          = ocfs2_flock,
2736 };
2737
2738 /*
2739  * POSIX-lockless variants of our file_operations.
2740  *
2741  * These will be used if the underlying cluster stack does not support
2742  * posix file locking, if the user passes the "localflocks" mount
2743  * option, or if we have a local-only fs.
2744  *
2745  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2746  * so we still want it in the case of no stack support for
2747  * plocks. Internally, it will do the right thing when asked to ignore
2748  * the cluster.
2749  */
2750 const struct file_operations ocfs2_fops_no_plocks = {
2751         .llseek         = ocfs2_file_llseek,
2752         .read           = do_sync_read,
2753         .write          = do_sync_write,
2754         .mmap           = ocfs2_mmap,
2755         .fsync          = ocfs2_sync_file,
2756         .release        = ocfs2_file_release,
2757         .open           = ocfs2_file_open,
2758         .aio_read       = ocfs2_file_aio_read,
2759         .aio_write      = ocfs2_file_aio_write,
2760         .unlocked_ioctl = ocfs2_ioctl,
2761 #ifdef CONFIG_COMPAT
2762         .compat_ioctl   = ocfs2_compat_ioctl,
2763 #endif
2764         .flock          = ocfs2_flock,
2765         .splice_read    = ocfs2_file_splice_read,
2766         .splice_write   = ocfs2_file_splice_write,
2767         .fallocate      = ocfs2_fallocate,
2768 };
2769
2770 const struct file_operations ocfs2_dops_no_plocks = {
2771         .llseek         = generic_file_llseek,
2772         .read           = generic_read_dir,
2773         .iterate        = ocfs2_readdir,
2774         .fsync          = ocfs2_sync_file,
2775         .release        = ocfs2_dir_release,
2776         .open           = ocfs2_dir_open,
2777         .unlocked_ioctl = ocfs2_ioctl,
2778 #ifdef CONFIG_COMPAT
2779         .compat_ioctl   = ocfs2_compat_ioctl,
2780 #endif
2781         .flock          = ocfs2_flock,
2782 };