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         err = filemap_write_and_wait_range(inode->i_mapping, start, end);
 189         if (err)
 190                 return err;
 191
 192         /*
 193          * Probably don't need the i_mutex at all in here, just putting it here
 194          * to be consistent with how fsync used to be called, someone more
 195          * familiar with the fs could possibly remove it.
 196          */
 197         mutex_lock(&inode->i_mutex);
 198         if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
 199                 /*
 200                  * We still have to flush drive's caches to get data to the
 201                  * platter
 202                  */
 203                 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
 204                         blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 205                 goto bail;
 206         }
 207
 208         journal = osb->journal->j_journal;
 209         err = jbd2_journal_force_commit(journal);
 210
 211 bail:
 212         if (err)
 213                 mlog_errno(err);
 214         mutex_unlock(&inode->i_mutex);
 215
 216         return (err < 0) ? -EIO : 0;
 217 }
 218
 219 int ocfs2_should_update_atime(struct inode *inode,
 220                               struct vfsmount *vfsmnt)
 221 {
 222         struct timespec now;
 223         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 224
 225         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 226                 return 0;
 227
 228         if ((inode->i_flags & S_NOATIME) ||
 229             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 230                 return 0;
 231
 232         /*
 233          * We can be called with no vfsmnt structure - NFSD will
 234          * sometimes do this.
 235          *
 236          * Note that our action here is different than touch_atime() -
 237          * if we can't tell whether this is a noatime mount, then we
 238          * don't know whether to trust the value of s_atime_quantum.
 239          */
 240         if (vfsmnt == NULL)
 241                 return 0;
 242
 243         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 244             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 245                 return 0;
 246
 247         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 248                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 249                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 250                         return 1;
 251
 252                 return 0;
 253         }
 254
 255         now = CURRENT_TIME;
 256         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 257                 return 0;
 258         else
 259                 return 1;
 260 }
 261
 262 int ocfs2_update_inode_atime(struct inode *inode,
 263                              struct buffer_head *bh)
 264 {
 265         int ret;
 266         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 267         handle_t *handle;
 268         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 269
 270         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 271         if (IS_ERR(handle)) {
 272                 ret = PTR_ERR(handle);
 273                 mlog_errno(ret);
 274                 goto out;
 275         }
 276
 277         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 278                                       OCFS2_JOURNAL_ACCESS_WRITE);
 279         if (ret) {
 280                 mlog_errno(ret);
 281                 goto out_commit;
 282         }
 283
 284         /*
 285          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 286          * have i_mutex to guard against concurrent changes to other
 287          * inode fields.
 288          */
 289         inode->i_atime = CURRENT_TIME;
 290         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 291         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 292         ocfs2_journal_dirty(handle, bh);
 293
 294 out_commit:
 295         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 296 out:
 297         return ret;
 298 }
 299
 300 static int ocfs2_set_inode_size(handle_t *handle,
 301                                 struct inode *inode,
 302                                 struct buffer_head *fe_bh,
 303                                 u64 new_i_size)
 304 {
 305         int status;
 306
 307         i_size_write(inode, new_i_size);
 308         inode->i_blocks = ocfs2_inode_sector_count(inode);
 309         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 310
 311         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 312         if (status < 0) {
 313                 mlog_errno(status);
 314                 goto bail;
 315         }
 316
 317 bail:
 318         return status;
 319 }
 320
 321 int ocfs2_simple_size_update(struct inode *inode,
 322                              struct buffer_head *di_bh,
 323                              u64 new_i_size)
 324 {
 325         int ret;
 326         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 327         handle_t *handle = NULL;
 328
 329         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 330         if (IS_ERR(handle)) {
 331                 ret = PTR_ERR(handle);
 332                 mlog_errno(ret);
 333                 goto out;
 334         }
 335
 336         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 337                                    new_i_size);
 338         if (ret < 0)
 339                 mlog_errno(ret);
 340
 341         ocfs2_commit_trans(osb, handle);
 342 out:
 343         return ret;
 344 }
 345
 346 static int ocfs2_cow_file_pos(struct inode *inode,
 347                               struct buffer_head *fe_bh,
 348                               u64 offset)
 349 {
 350         int status;
 351         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 352         unsigned int num_clusters = 0;
 353         unsigned int ext_flags = 0;
 354
 355         /*
 356          * If the new offset is aligned to the range of the cluster, there is
 357          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 358          * CoW either.
 359          */
 360         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 361                 return 0;
 362
 363         status = ocfs2_get_clusters(inode, cpos, &phys,
 364                                     &num_clusters, &ext_flags);
 365         if (status) {
 366                 mlog_errno(status);
 367                 goto out;
 368         }
 369
 370         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 371                 goto out;
 372
 373         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 374
 375 out:
 376         return status;
 377 }
 378
 379 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 380                                      struct inode *inode,
 381                                      struct buffer_head *fe_bh,
 382                                      u64 new_i_size)
 383 {
 384         int status;
 385         handle_t *handle;
 386         struct ocfs2_dinode *di;
 387         u64 cluster_bytes;
 388
 389         /*
 390          * We need to CoW the cluster contains the offset if it is reflinked
 391          * since we will call ocfs2_zero_range_for_truncate later which will
 392          * write "0" from offset to the end of the cluster.
 393          */
 394         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 395         if (status) {
 396                 mlog_errno(status);
 397                 return status;
 398         }
 399
 400         /* TODO: This needs to actually orphan the inode in this
 401          * transaction. */
 402
 403         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 404         if (IS_ERR(handle)) {
 405                 status = PTR_ERR(handle);
 406                 mlog_errno(status);
 407                 goto out;
 408         }
 409
 410         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 411                                          OCFS2_JOURNAL_ACCESS_WRITE);
 412         if (status < 0) {
 413                 mlog_errno(status);
 414                 goto out_commit;
 415         }
 416
 417         /*
 418          * Do this before setting i_size.
 419          */
 420         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 421         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 422                                                cluster_bytes);
 423         if (status) {
 424                 mlog_errno(status);
 425                 goto out_commit;
 426         }
 427
 428         i_size_write(inode, new_i_size);
 429         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 430
 431         di = (struct ocfs2_dinode *) fe_bh->b_data;
 432         di->i_size = cpu_to_le64(new_i_size);
 433         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 434         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 435
 436         ocfs2_journal_dirty(handle, fe_bh);
 437
 438 out_commit:
 439         ocfs2_commit_trans(osb, handle);
 440 out:
 441         return status;
 442 }
 443
 444 static int ocfs2_truncate_file(struct inode *inode,
 445                                struct buffer_head *di_bh,
 446                                u64 new_i_size)
 447 {
 448         int status = 0;
 449         struct ocfs2_dinode *fe = NULL;
 450         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 451
 452         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 453          * already validated it */
 454         fe = (struct ocfs2_dinode *) di_bh->b_data;
 455
 456         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 457                                   (unsigned long long)le64_to_cpu(fe->i_size),
 458                                   (unsigned long long)new_i_size);
 459
 460         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 461                         "Inode %llu, inode i_size = %lld != di "
 462                         "i_size = %llu, i_flags = 0x%x\n",
 463                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 464                         i_size_read(inode),
 465                         (unsigned long long)le64_to_cpu(fe->i_size),
 466                         le32_to_cpu(fe->i_flags));
 467
 468         if (new_i_size > le64_to_cpu(fe->i_size)) {
 469                 trace_ocfs2_truncate_file_error(
 470                         (unsigned long long)le64_to_cpu(fe->i_size),
 471                         (unsigned long long)new_i_size);
 472                 status = -EINVAL;
 473                 mlog_errno(status);
 474                 goto bail;
 475         }
 476
 477         /* lets handle the simple truncate cases before doing any more
 478          * cluster locking. */
 479         if (new_i_size == le64_to_cpu(fe->i_size))
 480                 goto bail;
 481
 482         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 483
 484         ocfs2_resv_discard(&osb->osb_la_resmap,
 485                            &OCFS2_I(inode)->ip_la_data_resv);
 486
 487         /*
 488          * The inode lock forced other nodes to sync and drop their
 489          * pages, which (correctly) happens even if we have a truncate
 490          * without allocation change - ocfs2 cluster sizes can be much
 491          * greater than page size, so we have to truncate them
 492          * anyway.
 493          */
 494         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 495         truncate_inode_pages(inode->i_mapping, new_i_size);
 496
 497         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 498                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 499                                                i_size_read(inode), 1);
 500                 if (status)
 501                         mlog_errno(status);
 502
 503                 goto bail_unlock_sem;
 504         }
 505
 506         /* alright, we're going to need to do a full blown alloc size
 507          * change. Orphan the inode so that recovery can complete the
 508          * truncate if necessary. This does the task of marking
 509          * i_size. */
 510         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 511         if (status < 0) {
 512                 mlog_errno(status);
 513                 goto bail_unlock_sem;
 514         }
 515
 516         status = ocfs2_commit_truncate(osb, inode, di_bh);
 517         if (status < 0) {
 518                 mlog_errno(status);
 519                 goto bail_unlock_sem;
 520         }
 521
 522         /* TODO: orphan dir cleanup here. */
 523 bail_unlock_sem:
 524         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 525
 526 bail:
 527         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 528                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 529
 530         return status;
 531 }
 532
 533 /*
 534  * extend file allocation only here.
 535  * we'll update all the disk stuff, and oip->alloc_size
 536  *
 537  * expect stuff to be locked, a transaction started and enough data /
 538  * metadata reservations in the contexts.
 539  *
 540  * Will return -EAGAIN, and a reason if a restart is needed.
 541  * If passed in, *reason will always be set, even in error.
 542  */
 543 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 544                          struct inode *inode,
 545                          u32 *logical_offset,
 546                          u32 clusters_to_add,
 547                          int mark_unwritten,
 548                          struct buffer_head *fe_bh,
 549                          handle_t *handle,
 550                          struct ocfs2_alloc_context *data_ac,
 551                          struct ocfs2_alloc_context *meta_ac,
 552                          enum ocfs2_alloc_restarted *reason_ret)
 553 {
 554         int ret;
 555         struct ocfs2_extent_tree et;
 556
 557         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 558         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 559                                           clusters_to_add, mark_unwritten,
 560                                           data_ac, meta_ac, reason_ret);
 561
 562         return ret;
 563 }
 564
 565 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 566                                      u32 clusters_to_add, int mark_unwritten)
 567 {
 568         int status = 0;
 569         int restart_func = 0;
 570         int credits;
 571         u32 prev_clusters;
 572         struct buffer_head *bh = NULL;
 573         struct ocfs2_dinode *fe = NULL;
 574         handle_t *handle = NULL;
 575         struct ocfs2_alloc_context *data_ac = NULL;
 576         struct ocfs2_alloc_context *meta_ac = NULL;
 577         enum ocfs2_alloc_restarted why;
 578         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 579         struct ocfs2_extent_tree et;
 580         int did_quota = 0;
 581
 582         /*
 583          * This function only exists for file systems which don't
 584          * support holes.
 585          */
 586         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 587
 588         status = ocfs2_read_inode_block(inode, &bh);
 589         if (status < 0) {
 590                 mlog_errno(status);
 591                 goto leave;
 592         }
 593         fe = (struct ocfs2_dinode *) bh->b_data;
 594
 595 restart_all:
 596         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 597
 598         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 599         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 600                                        &data_ac, &meta_ac);
 601         if (status) {
 602                 mlog_errno(status);
 603                 goto leave;
 604         }
 605
 606         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
 607                                             clusters_to_add);
 608         handle = ocfs2_start_trans(osb, credits);
 609         if (IS_ERR(handle)) {
 610                 status = PTR_ERR(handle);
 611                 handle = NULL;
 612                 mlog_errno(status);
 613                 goto leave;
 614         }
 615
 616 restarted_transaction:
 617         trace_ocfs2_extend_allocation(
 618                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 619                 (unsigned long long)i_size_read(inode),
 620                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 621                 why, restart_func);
 622
 623         status = dquot_alloc_space_nodirty(inode,
 624                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 625         if (status)
 626                 goto leave;
 627         did_quota = 1;
 628
 629         /* reserve a write to the file entry early on - that we if we
 630          * run out of credits in the allocation path, we can still
 631          * update i_size. */
 632         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 633                                          OCFS2_JOURNAL_ACCESS_WRITE);
 634         if (status < 0) {
 635                 mlog_errno(status);
 636                 goto leave;
 637         }
 638
 639         prev_clusters = OCFS2_I(inode)->ip_clusters;
 640
 641         status = ocfs2_add_inode_data(osb,
 642                                       inode,
 643                                       &logical_start,
 644                                       clusters_to_add,
 645                                       mark_unwritten,
 646                                       bh,
 647                                       handle,
 648                                       data_ac,
 649                                       meta_ac,
 650                                       &why);
 651         if ((status < 0) && (status != -EAGAIN)) {
 652                 if (status != -ENOSPC)
 653                         mlog_errno(status);
 654                 goto leave;
 655         }
 656
 657         ocfs2_journal_dirty(handle, bh);
 658
 659         spin_lock(&OCFS2_I(inode)->ip_lock);
 660         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 661         spin_unlock(&OCFS2_I(inode)->ip_lock);
 662         /* Release unused quota reservation */
 663         dquot_free_space(inode,
 664                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 665         did_quota = 0;
 666
 667         if (why != RESTART_NONE && clusters_to_add) {
 668                 if (why == RESTART_META) {
 669                         restart_func = 1;
 670                         status = 0;
 671                 } else {
 672                         BUG_ON(why != RESTART_TRANS);
 673
 674                         status = ocfs2_allocate_extend_trans(handle, 1);
 675                         if (status < 0) {
 676                                 /* handle still has to be committed at
 677                                  * this point. */
 678                                 status = -ENOMEM;
 679                                 mlog_errno(status);
 680                                 goto leave;
 681                         }
 682                         goto restarted_transaction;
 683                 }
 684         }
 685
 686         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 687              le32_to_cpu(fe->i_clusters),
 688              (unsigned long long)le64_to_cpu(fe->i_size),
 689              OCFS2_I(inode)->ip_clusters,
 690              (unsigned long long)i_size_read(inode));
 691
 692 leave:
 693         if (status < 0 && did_quota)
 694                 dquot_free_space(inode,
 695                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 696         if (handle) {
 697                 ocfs2_commit_trans(osb, handle);
 698                 handle = NULL;
 699         }
 700         if (data_ac) {
 701                 ocfs2_free_alloc_context(data_ac);
 702                 data_ac = NULL;
 703         }
 704         if (meta_ac) {
 705                 ocfs2_free_alloc_context(meta_ac);
 706                 meta_ac = NULL;
 707         }
 708         if ((!status) && restart_func) {
 709                 restart_func = 0;
 710                 goto restart_all;
 711         }
 712         brelse(bh);
 713         bh = NULL;
 714
 715         return status;
 716 }
 717
 718 /*
 719  * While a write will already be ordering the data, a truncate will not.
 720  * Thus, we need to explicitly order the zeroed pages.
 721  */
 722 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
 723 {
 724         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 725         handle_t *handle = NULL;
 726         int ret = 0;
 727
 728         if (!ocfs2_should_order_data(inode))
 729                 goto out;
 730
 731         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 732         if (IS_ERR(handle)) {
 733                 ret = -ENOMEM;
 734                 mlog_errno(ret);
 735                 goto out;
 736         }
 737
 738         ret = ocfs2_jbd2_file_inode(handle, inode);
 739         if (ret < 0)
 740                 mlog_errno(ret);
 741
 742 out:
 743         if (ret) {
 744                 if (!IS_ERR(handle))
 745                         ocfs2_commit_trans(osb, handle);
 746                 handle = ERR_PTR(ret);
 747         }
 748         return handle;
 749 }
 750
 751 /* Some parts of this taken from generic_cont_expand, which turned out
 752  * to be too fragile to do exactly what we need without us having to
 753  * worry about recursive locking in ->write_begin() and ->write_end(). */
 754 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 755                                  u64 abs_to)
 756 {
 757         struct address_space *mapping = inode->i_mapping;
 758         struct page *page;
 759         unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
 760         handle_t *handle = NULL;
 761         int ret = 0;
 762         unsigned zero_from, zero_to, block_start, block_end;
 763
 764         BUG_ON(abs_from >= abs_to);
 765         BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
 766         BUG_ON(abs_from & (inode->i_blkbits - 1));
 767
 768         page = find_or_create_page(mapping, index, GFP_NOFS);
 769         if (!page) {
 770                 ret = -ENOMEM;
 771                 mlog_errno(ret);
 772                 goto out;
 773         }
 774
 775         /* Get the offsets within the page that we want to zero */
 776         zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
 777         zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
 778         if (!zero_to)
 779                 zero_to = PAGE_CACHE_SIZE;
 780
 781         trace_ocfs2_write_zero_page(
 782                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 783                         (unsigned long long)abs_from,
 784                         (unsigned long long)abs_to,
 785                         index, zero_from, zero_to);
 786
 787         /* We know that zero_from is block aligned */
 788         for (block_start = zero_from; block_start < zero_to;
 789              block_start = block_end) {
 790                 block_end = block_start + (1 << inode->i_blkbits);
 791
 792                 /*
 793                  * block_start is block-aligned.  Bump it by one to force
 794                  * __block_write_begin and block_commit_write to zero the
 795                  * whole block.
 796                  */
 797                 ret = __block_write_begin(page, block_start + 1, 0,
 798                                           ocfs2_get_block);
 799                 if (ret < 0) {
 800                         mlog_errno(ret);
 801                         goto out_unlock;
 802                 }
 803
 804                 if (!handle) {
 805                         handle = ocfs2_zero_start_ordered_transaction(inode);
 806                         if (IS_ERR(handle)) {
 807                                 ret = PTR_ERR(handle);
 808                                 handle = NULL;
 809                                 break;
 810                         }
 811                 }
 812
 813                 /* must not update i_size! */
 814                 ret = block_commit_write(page, block_start + 1,
 815                                          block_start + 1);
 816                 if (ret < 0)
 817                         mlog_errno(ret);
 818                 else
 819                         ret = 0;
 820         }
 821
 822         if (handle)
 823                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 824
 825 out_unlock:
 826         unlock_page(page);
 827         page_cache_release(page);
 828 out:
 829         return ret;
 830 }
 831
 832 /*
 833  * Find the next range to zero.  We do this in terms of bytes because
 834  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 835  * pagecache.  We may return multiple extents.
 836  *
 837  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 838  * needs to be zeroed.  range_start and range_end return the next zeroing
 839  * range.  A subsequent call should pass the previous range_end as its
 840  * zero_start.  If range_end is 0, there's nothing to do.
 841  *
 842  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 843  */
 844 static int ocfs2_zero_extend_get_range(struct inode *inode,
 845                                        struct buffer_head *di_bh,
 846                                        u64 zero_start, u64 zero_end,
 847                                        u64 *range_start, u64 *range_end)
 848 {
 849         int rc = 0, needs_cow = 0;
 850         u32 p_cpos, zero_clusters = 0;
 851         u32 zero_cpos =
 852                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 853         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 854         unsigned int num_clusters = 0;
 855         unsigned int ext_flags = 0;
 856
 857         while (zero_cpos < last_cpos) {
 858                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 859                                         &num_clusters, &ext_flags);
 860                 if (rc) {
 861                         mlog_errno(rc);
 862                         goto out;
 863                 }
 864
 865                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 866                         zero_clusters = num_clusters;
 867                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 868                                 needs_cow = 1;
 869                         break;
 870                 }
 871
 872                 zero_cpos += num_clusters;
 873         }
 874         if (!zero_clusters) {
 875                 *range_end = 0;
 876                 goto out;
 877         }
 878
 879         while ((zero_cpos + zero_clusters) < last_cpos) {
 880                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 881                                         &p_cpos, &num_clusters,
 882                                         &ext_flags);
 883                 if (rc) {
 884                         mlog_errno(rc);
 885                         goto out;
 886                 }
 887
 888                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 889                         break;
 890                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 891                         needs_cow = 1;
 892                 zero_clusters += num_clusters;
 893         }
 894         if ((zero_cpos + zero_clusters) > last_cpos)
 895                 zero_clusters = last_cpos - zero_cpos;
 896
 897         if (needs_cow) {
 898                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 899                                         zero_clusters, UINT_MAX);
 900                 if (rc) {
 901                         mlog_errno(rc);
 902                         goto out;
 903                 }
 904         }
 905
 906         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 907         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 908                                              zero_cpos + zero_clusters);
 909
 910 out:
 911         return rc;
 912 }
 913
 914 /*
 915  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 916  * has made sure that the entire range needs zeroing.
 917  */
 918 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 919                                    u64 range_end)
 920 {
 921         int rc = 0;
 922         u64 next_pos;
 923         u64 zero_pos = range_start;
 924
 925         trace_ocfs2_zero_extend_range(
 926                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 927                         (unsigned long long)range_start,
 928                         (unsigned long long)range_end);
 929         BUG_ON(range_start >= range_end);
 930
 931         while (zero_pos < range_end) {
 932                 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
 933                 if (next_pos > range_end)
 934                         next_pos = range_end;
 935                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
 936                 if (rc < 0) {
 937                         mlog_errno(rc);
 938                         break;
 939                 }
 940                 zero_pos = next_pos;
 941
 942                 /*
 943                  * Very large extends have the potential to lock up
 944                  * the cpu for extended periods of time.
 945                  */
 946                 cond_resched();
 947         }
 948
 949         return rc;
 950 }
 951
 952 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 953                       loff_t zero_to_size)
 954 {
 955         int ret = 0;
 956         u64 zero_start, range_start = 0, range_end = 0;
 957         struct super_block *sb = inode->i_sb;
 958
 959         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 960         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 961                                 (unsigned long long)zero_start,
 962                                 (unsigned long long)i_size_read(inode));
 963         while (zero_start < zero_to_size) {
 964                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 965                                                   zero_to_size,
 966                                                   &range_start,
 967                                                   &range_end);
 968                 if (ret) {
 969                         mlog_errno(ret);
 970                         break;
 971                 }
 972                 if (!range_end)
 973                         break;
 974                 /* Trim the ends */
 975                 if (range_start < zero_start)
 976                         range_start = zero_start;
 977                 if (range_end > zero_to_size)
 978                         range_end = zero_to_size;
 979
 980                 ret = ocfs2_zero_extend_range(inode, range_start,
 981                                               range_end);
 982                 if (ret) {
 983                         mlog_errno(ret);
 984                         break;
 985                 }
 986                 zero_start = range_end;
 987         }
 988
 989         return ret;
 990 }
 991
 992 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
 993                           u64 new_i_size, u64 zero_to)
 994 {
 995         int ret;
 996         u32 clusters_to_add;
 997         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 998
 999         /*
1000          * Only quota files call this without a bh, and they can't be
1001          * refcounted.
1002          */
1003         BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1004         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1005
1006         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1007         if (clusters_to_add < oi->ip_clusters)
1008                 clusters_to_add = 0;
1009         else
1010                 clusters_to_add -= oi->ip_clusters;
1011
1012         if (clusters_to_add) {
1013                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1014                                                 clusters_to_add, 0);
1015                 if (ret) {
1016                         mlog_errno(ret);
1017                         goto out;
1018                 }
1019         }
1020
1021         /*
1022          * Call this even if we don't add any clusters to the tree. We
1023          * still need to zero the area between the old i_size and the
1024          * new i_size.
1025          */
1026         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1027         if (ret < 0)
1028                 mlog_errno(ret);
1029
1030 out:
1031         return ret;
1032 }
1033
1034 static int ocfs2_extend_file(struct inode *inode,
1035                              struct buffer_head *di_bh,
1036                              u64 new_i_size)
1037 {
1038         int ret = 0;
1039         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1040
1041         BUG_ON(!di_bh);
1042
1043         /* setattr sometimes calls us like this. */
1044         if (new_i_size == 0)
1045                 goto out;
1046
1047         if (i_size_read(inode) == new_i_size)
1048                 goto out;
1049         BUG_ON(new_i_size < i_size_read(inode));
1050
1051         /*
1052          * The alloc sem blocks people in read/write from reading our
1053          * allocation until we're done changing it. We depend on
1054          * i_mutex to block other extend/truncate calls while we're
1055          * here.  We even have to hold it for sparse files because there
1056          * might be some tail zeroing.
1057          */
1058         down_write(&oi->ip_alloc_sem);
1059
1060         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1061                 /*
1062                  * We can optimize small extends by keeping the inodes
1063                  * inline data.
1064                  */
1065                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1066                         up_write(&oi->ip_alloc_sem);
1067                         goto out_update_size;
1068                 }
1069
1070                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1071                 if (ret) {
1072                         up_write(&oi->ip_alloc_sem);
1073                         mlog_errno(ret);
1074                         goto out;
1075                 }
1076         }
1077
1078         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1079                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1080         else
1081                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1082                                             new_i_size);
1083
1084         up_write(&oi->ip_alloc_sem);
1085
1086         if (ret < 0) {
1087                 mlog_errno(ret);
1088                 goto out;
1089         }
1090
1091 out_update_size:
1092         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1093         if (ret < 0)
1094                 mlog_errno(ret);
1095
1096 out:
1097         return ret;
1098 }
1099
1100 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1101 {
1102         int status = 0, size_change;
1103         struct inode *inode = dentry->d_inode;
1104         struct super_block *sb = inode->i_sb;
1105         struct ocfs2_super *osb = OCFS2_SB(sb);
1106         struct buffer_head *bh = NULL;
1107         handle_t *handle = NULL;
1108         struct dquot *transfer_to[MAXQUOTAS] = { };
1109         int qtype;
1110
1111         trace_ocfs2_setattr(inode, dentry,
1112                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1113                             dentry->d_name.len, dentry->d_name.name,
1114                             attr->ia_valid, attr->ia_mode,
1115                             from_kuid(&init_user_ns, attr->ia_uid),
1116                             from_kgid(&init_user_ns, attr->ia_gid));
1117
1118         /* ensuring we don't even attempt to truncate a symlink */
1119         if (S_ISLNK(inode->i_mode))
1120                 attr->ia_valid &= ~ATTR_SIZE;
1121
1122 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1123                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1124         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1125                 return 0;
1126
1127         status = inode_change_ok(inode, attr);
1128         if (status)
1129                 return status;
1130
1131         if (is_quota_modification(inode, attr))
1132                 dquot_initialize(inode);
1133         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1134         if (size_change) {
1135                 status = ocfs2_rw_lock(inode, 1);
1136                 if (status < 0) {
1137                         mlog_errno(status);
1138                         goto bail;
1139                 }
1140         }
1141
1142         status = ocfs2_inode_lock(inode, &bh, 1);
1143         if (status < 0) {
1144                 if (status != -ENOENT)
1145                         mlog_errno(status);
1146                 goto bail_unlock_rw;
1147         }
1148
1149         if (size_change && attr->ia_size != i_size_read(inode)) {
1150                 status = inode_newsize_ok(inode, attr->ia_size);
1151                 if (status)
1152                         goto bail_unlock;
1153
1154                 inode_dio_wait(inode);
1155
1156                 if (i_size_read(inode) > attr->ia_size) {
1157                         if (ocfs2_should_order_data(inode)) {
1158                                 status = ocfs2_begin_ordered_truncate(inode,
1159                                                                       attr->ia_size);
1160                                 if (status)
1161                                         goto bail_unlock;
1162                         }
1163                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1164                 } else
1165                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1166                 if (status < 0) {
1167                         if (status != -ENOSPC)
1168                                 mlog_errno(status);
1169                         status = -ENOSPC;
1170                         goto bail_unlock;
1171                 }
1172         }
1173
1174         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1175             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1176                 /*
1177                  * Gather pointers to quota structures so that allocation /
1178                  * freeing of quota structures happens here and not inside
1179                  * dquot_transfer() where we have problems with lock ordering
1180                  */
1181                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1182                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1183                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1184                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1185                         if (!transfer_to[USRQUOTA]) {
1186                                 status = -ESRCH;
1187                                 goto bail_unlock;
1188                         }
1189                 }
1190                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1191                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1192                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1193                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1194                         if (!transfer_to[GRPQUOTA]) {
1195                                 status = -ESRCH;
1196                                 goto bail_unlock;
1197                         }
1198                 }
1199                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1200                                            2 * ocfs2_quota_trans_credits(sb));
1201                 if (IS_ERR(handle)) {
1202                         status = PTR_ERR(handle);
1203                         mlog_errno(status);
1204                         goto bail_unlock;
1205                 }
1206                 status = __dquot_transfer(inode, transfer_to);
1207                 if (status < 0)
1208                         goto bail_commit;
1209         } else {
1210                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1211                 if (IS_ERR(handle)) {
1212                         status = PTR_ERR(handle);
1213                         mlog_errno(status);
1214                         goto bail_unlock;
1215                 }
1216         }
1217
1218         setattr_copy(inode, attr);
1219         mark_inode_dirty(inode);
1220
1221         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1222         if (status < 0)
1223                 mlog_errno(status);
1224
1225 bail_commit:
1226         ocfs2_commit_trans(osb, handle);
1227 bail_unlock:
1228         ocfs2_inode_unlock(inode, 1);
1229 bail_unlock_rw:
1230         if (size_change)
1231                 ocfs2_rw_unlock(inode, 1);
1232 bail:
1233         brelse(bh);
1234
1235         /* Release quota pointers in case we acquired them */
1236         for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1237                 dqput(transfer_to[qtype]);
1238
1239         if (!status && attr->ia_valid & ATTR_MODE) {
1240                 status = ocfs2_acl_chmod(inode);
1241                 if (status < 0)
1242                         mlog_errno(status);
1243         }
1244
1245         return status;
1246 }
1247
1248 int ocfs2_getattr(struct vfsmount *mnt,
1249                   struct dentry *dentry,
1250                   struct kstat *stat)
1251 {
1252         struct inode *inode = dentry->d_inode;
1253         struct super_block *sb = dentry->d_inode->i_sb;
1254         struct ocfs2_super *osb = sb->s_fs_info;
1255         int err;
1256
1257         err = ocfs2_inode_revalidate(dentry);
1258         if (err) {
1259                 if (err != -ENOENT)
1260                         mlog_errno(err);
1261                 goto bail;
1262         }
1263
1264         generic_fillattr(inode, stat);
1265
1266         /* We set the blksize from the cluster size for performance */
1267         stat->blksize = osb->s_clustersize;
1268
1269 bail:
1270         return err;
1271 }
1272
1273 int ocfs2_permission(struct inode *inode, int mask)
1274 {
1275         int ret;
1276
1277         if (mask & MAY_NOT_BLOCK)
1278                 return -ECHILD;
1279
1280         ret = ocfs2_inode_lock(inode, NULL, 0);
1281         if (ret) {
1282                 if (ret != -ENOENT)
1283                         mlog_errno(ret);
1284                 goto out;
1285         }
1286
1287         ret = generic_permission(inode, mask);
1288
1289         ocfs2_inode_unlock(inode, 0);
1290 out:
1291         return ret;
1292 }
1293
1294 static int __ocfs2_write_remove_suid(struct inode *inode,
1295                                      struct buffer_head *bh)
1296 {
1297         int ret;
1298         handle_t *handle;
1299         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1300         struct ocfs2_dinode *di;
1301
1302         trace_ocfs2_write_remove_suid(
1303                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1304                         inode->i_mode);
1305
1306         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1307         if (IS_ERR(handle)) {
1308                 ret = PTR_ERR(handle);
1309                 mlog_errno(ret);
1310                 goto out;
1311         }
1312
1313         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1314                                       OCFS2_JOURNAL_ACCESS_WRITE);
1315         if (ret < 0) {
1316                 mlog_errno(ret);
1317                 goto out_trans;
1318         }
1319
1320         inode->i_mode &= ~S_ISUID;
1321         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1322                 inode->i_mode &= ~S_ISGID;
1323
1324         di = (struct ocfs2_dinode *) bh->b_data;
1325         di->i_mode = cpu_to_le16(inode->i_mode);
1326
1327         ocfs2_journal_dirty(handle, bh);
1328
1329 out_trans:
1330         ocfs2_commit_trans(osb, handle);
1331 out:
1332         return ret;
1333 }
1334
1335 /*
1336  * Will look for holes and unwritten extents in the range starting at
1337  * pos for count bytes (inclusive).
1338  */
1339 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1340                                        size_t count)
1341 {
1342         int ret = 0;
1343         unsigned int extent_flags;
1344         u32 cpos, clusters, extent_len, phys_cpos;
1345         struct super_block *sb = inode->i_sb;
1346
1347         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1348         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1349
1350         while (clusters) {
1351                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1352                                          &extent_flags);
1353                 if (ret < 0) {
1354                         mlog_errno(ret);
1355                         goto out;
1356                 }
1357
1358                 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1359                         ret = 1;
1360                         break;
1361                 }
1362
1363                 if (extent_len > clusters)
1364                         extent_len = clusters;
1365
1366                 clusters -= extent_len;
1367                 cpos += extent_len;
1368         }
1369 out:
1370         return ret;
1371 }
1372
1373 static int ocfs2_write_remove_suid(struct inode *inode)
1374 {
1375         int ret;
1376         struct buffer_head *bh = NULL;
1377
1378         ret = ocfs2_read_inode_block(inode, &bh);
1379         if (ret < 0) {
1380                 mlog_errno(ret);
1381                 goto out;
1382         }
1383
1384         ret =  __ocfs2_write_remove_suid(inode, bh);
1385 out:
1386         brelse(bh);
1387         return ret;
1388 }
1389
1390 /*
1391  * Allocate enough extents to cover the region starting at byte offset
1392  * start for len bytes. Existing extents are skipped, any extents
1393  * added are marked as "unwritten".
1394  */
1395 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1396                                             u64 start, u64 len)
1397 {
1398         int ret;
1399         u32 cpos, phys_cpos, clusters, alloc_size;
1400         u64 end = start + len;
1401         struct buffer_head *di_bh = NULL;
1402
1403         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1404                 ret = ocfs2_read_inode_block(inode, &di_bh);
1405                 if (ret) {
1406                         mlog_errno(ret);
1407                         goto out;
1408                 }
1409
1410                 /*
1411                  * Nothing to do if the requested reservation range
1412                  * fits within the inode.
1413                  */
1414                 if (ocfs2_size_fits_inline_data(di_bh, end))
1415                         goto out;
1416
1417                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1418                 if (ret) {
1419                         mlog_errno(ret);
1420                         goto out;
1421                 }
1422         }
1423
1424         /*
1425          * We consider both start and len to be inclusive.
1426          */
1427         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1428         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1429         clusters -= cpos;
1430
1431         while (clusters) {
1432                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1433                                          &alloc_size, NULL);
1434                 if (ret) {
1435                         mlog_errno(ret);
1436                         goto out;
1437                 }
1438
1439                 /*
1440                  * Hole or existing extent len can be arbitrary, so
1441                  * cap it to our own allocation request.
1442                  */
1443                 if (alloc_size > clusters)
1444                         alloc_size = clusters;
1445
1446                 if (phys_cpos) {
1447                         /*
1448                          * We already have an allocation at this
1449                          * region so we can safely skip it.
1450                          */
1451                         goto next;
1452                 }
1453
1454                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1455                 if (ret) {
1456                         if (ret != -ENOSPC)
1457                                 mlog_errno(ret);
1458                         goto out;
1459                 }
1460
1461 next:
1462                 cpos += alloc_size;
1463                 clusters -= alloc_size;
1464         }
1465
1466         ret = 0;
1467 out:
1468
1469         brelse(di_bh);
1470         return ret;
1471 }
1472
1473 /*
1474  * Truncate a byte range, avoiding pages within partial clusters. This
1475  * preserves those pages for the zeroing code to write to.
1476  */
1477 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1478                                          u64 byte_len)
1479 {
1480         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1481         loff_t start, end;
1482         struct address_space *mapping = inode->i_mapping;
1483
1484         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1485         end = byte_start + byte_len;
1486         end = end & ~(osb->s_clustersize - 1);
1487
1488         if (start < end) {
1489                 unmap_mapping_range(mapping, start, end - start, 0);
1490                 truncate_inode_pages_range(mapping, start, end - 1);
1491         }
1492 }
1493
1494 static int ocfs2_zero_partial_clusters(struct inode *inode,
1495                                        u64 start, u64 len)
1496 {
1497         int ret = 0;
1498         u64 tmpend, end = start + len;
1499         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1500         unsigned int csize = osb->s_clustersize;
1501         handle_t *handle;
1502
1503         /*
1504          * The "start" and "end" values are NOT necessarily part of
1505          * the range whose allocation is being deleted. Rather, this
1506          * is what the user passed in with the request. We must zero
1507          * partial clusters here. There's no need to worry about
1508          * physical allocation - the zeroing code knows to skip holes.
1509          */
1510         trace_ocfs2_zero_partial_clusters(
1511                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1512                 (unsigned long long)start, (unsigned long long)end);
1513
1514         /*
1515          * If both edges are on a cluster boundary then there's no
1516          * zeroing required as the region is part of the allocation to
1517          * be truncated.
1518          */
1519         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1520                 goto out;
1521
1522         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1523         if (IS_ERR(handle)) {
1524                 ret = PTR_ERR(handle);
1525                 mlog_errno(ret);
1526                 goto out;
1527         }
1528
1529         /*
1530          * We want to get the byte offset of the end of the 1st cluster.
1531          */
1532         tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1533         if (tmpend > end)
1534                 tmpend = end;
1535
1536         trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1537                                                  (unsigned long long)tmpend);
1538
1539         ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1540         if (ret)
1541                 mlog_errno(ret);
1542
1543         if (tmpend < end) {
1544                 /*
1545                  * This may make start and end equal, but the zeroing
1546                  * code will skip any work in that case so there's no
1547                  * need to catch it up here.
1548                  */
1549                 start = end & ~(osb->s_clustersize - 1);
1550
1551                 trace_ocfs2_zero_partial_clusters_range2(
1552                         (unsigned long long)start, (unsigned long long)end);
1553
1554                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1555                 if (ret)
1556                         mlog_errno(ret);
1557         }
1558
1559         ocfs2_commit_trans(osb, handle);
1560 out:
1561         return ret;
1562 }
1563
1564 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1565 {
1566         int i;
1567         struct ocfs2_extent_rec *rec = NULL;
1568
1569         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1570
1571                 rec = &el->l_recs[i];
1572
1573                 if (le32_to_cpu(rec->e_cpos) < pos)
1574                         break;
1575         }
1576
1577         return i;
1578 }
1579
1580 /*
1581  * Helper to calculate the punching pos and length in one run, we handle the
1582  * following three cases in order:
1583  *
1584  * - remove the entire record
1585  * - remove a partial record
1586  * - no record needs to be removed (hole-punching completed)
1587 */
1588 static void ocfs2_calc_trunc_pos(struct inode *inode,
1589                                  struct ocfs2_extent_list *el,
1590                                  struct ocfs2_extent_rec *rec,
1591                                  u32 trunc_start, u32 *trunc_cpos,
1592                                  u32 *trunc_len, u32 *trunc_end,
1593                                  u64 *blkno, int *done)
1594 {
1595         int ret = 0;
1596         u32 coff, range;
1597
1598         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1599
1600         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1601                 /*
1602                  * remove an entire extent record.
1603                  */
1604                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1605                 /*
1606                  * Skip holes if any.
1607                  */
1608                 if (range < *trunc_end)
1609                         *trunc_end = range;
1610                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1611                 *blkno = le64_to_cpu(rec->e_blkno);
1612                 *trunc_end = le32_to_cpu(rec->e_cpos);
1613         } else if (range > trunc_start) {
1614                 /*
1615                  * remove a partial extent record, which means we're
1616                  * removing the last extent record.
1617                  */
1618                 *trunc_cpos = trunc_start;
1619                 /*
1620                  * skip hole if any.
1621                  */
1622                 if (range < *trunc_end)
1623                         *trunc_end = range;
1624                 *trunc_len = *trunc_end - trunc_start;
1625                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1626                 *blkno = le64_to_cpu(rec->e_blkno) +
1627                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1628                 *trunc_end = trunc_start;
1629         } else {
1630                 /*
1631                  * It may have two following possibilities:
1632                  *
1633                  * - last record has been removed
1634                  * - trunc_start was within a hole
1635                  *
1636                  * both two cases mean the completion of hole punching.
1637                  */
1638                 ret = 1;
1639         }
1640
1641         *done = ret;
1642 }
1643
1644 static int ocfs2_remove_inode_range(struct inode *inode,
1645                                     struct buffer_head *di_bh, u64 byte_start,
1646                                     u64 byte_len)
1647 {
1648         int ret = 0, flags = 0, done = 0, i;
1649         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1650         u32 cluster_in_el;
1651         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1652         struct ocfs2_cached_dealloc_ctxt dealloc;
1653         struct address_space *mapping = inode->i_mapping;
1654         struct ocfs2_extent_tree et;
1655         struct ocfs2_path *path = NULL;
1656         struct ocfs2_extent_list *el = NULL;
1657         struct ocfs2_extent_rec *rec = NULL;
1658         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1659         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1660
1661         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1662         ocfs2_init_dealloc_ctxt(&dealloc);
1663
1664         trace_ocfs2_remove_inode_range(
1665                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1666                         (unsigned long long)byte_start,
1667                         (unsigned long long)byte_len);
1668
1669         if (byte_len == 0)
1670                 return 0;
1671
1672         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1673                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1674                                             byte_start + byte_len, 0);
1675                 if (ret) {
1676                         mlog_errno(ret);
1677                         goto out;
1678                 }
1679                 /*
1680                  * There's no need to get fancy with the page cache
1681                  * truncate of an inline-data inode. We're talking
1682                  * about less than a page here, which will be cached
1683                  * in the dinode buffer anyway.
1684                  */
1685                 unmap_mapping_range(mapping, 0, 0, 0);
1686                 truncate_inode_pages(mapping, 0);
1687                 goto out;
1688         }
1689
1690         /*
1691          * For reflinks, we may need to CoW 2 clusters which might be
1692          * partially zero'd later, if hole's start and end offset were
1693          * within one cluster(means is not exactly aligned to clustersize).
1694          */
1695
1696         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1697
1698                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1699                 if (ret) {
1700                         mlog_errno(ret);
1701                         goto out;
1702                 }
1703
1704                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1705                 if (ret) {
1706                         mlog_errno(ret);
1707                         goto out;
1708                 }
1709         }
1710
1711         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1712         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1713         cluster_in_el = trunc_end;
1714
1715         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1716         if (ret) {
1717                 mlog_errno(ret);
1718                 goto out;
1719         }
1720
1721         path = ocfs2_new_path_from_et(&et);
1722         if (!path) {
1723                 ret = -ENOMEM;
1724                 mlog_errno(ret);
1725                 goto out;
1726         }
1727
1728         while (trunc_end > trunc_start) {
1729
1730                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1731                                       cluster_in_el);
1732                 if (ret) {
1733                         mlog_errno(ret);
1734                         goto out;
1735                 }
1736
1737                 el = path_leaf_el(path);
1738
1739                 i = ocfs2_find_rec(el, trunc_end);
1740                 /*
1741                  * Need to go to previous extent block.
1742                  */
1743                 if (i < 0) {
1744                         if (path->p_tree_depth == 0)
1745                                 break;
1746
1747                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1748                                                             path,
1749                                                             &cluster_in_el);
1750                         if (ret) {
1751                                 mlog_errno(ret);
1752                                 goto out;
1753                         }
1754
1755                         /*
1756                          * We've reached the leftmost extent block,
1757                          * it's safe to leave.
1758                          */
1759                         if (cluster_in_el == 0)
1760                                 break;
1761
1762                         /*
1763                          * The 'pos' searched for previous extent block is
1764                          * always one cluster less than actual trunc_end.
1765                          */
1766                         trunc_end = cluster_in_el + 1;
1767
1768                         ocfs2_reinit_path(path, 1);
1769
1770                         continue;
1771
1772                 } else
1773                         rec = &el->l_recs[i];
1774
1775                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1776                                      &trunc_len, &trunc_end, &blkno, &done);
1777                 if (done)
1778                         break;
1779
1780                 flags = rec->e_flags;
1781                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1782
1783                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1784                                                phys_cpos, trunc_len, flags,
1785                                                &dealloc, refcount_loc);
1786                 if (ret < 0) {
1787                         mlog_errno(ret);
1788                         goto out;
1789                 }
1790
1791                 cluster_in_el = trunc_end;
1792
1793                 ocfs2_reinit_path(path, 1);
1794         }
1795
1796         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1797
1798 out:
1799         ocfs2_free_path(path);
1800         ocfs2_schedule_truncate_log_flush(osb, 1);
1801         ocfs2_run_deallocs(osb, &dealloc);
1802
1803         return ret;
1804 }
1805
1806 /*
1807  * Parts of this function taken from xfs_change_file_space()
1808  */
1809 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1810                                      loff_t f_pos, unsigned int cmd,
1811                                      struct ocfs2_space_resv *sr,
1812                                      int change_size)
1813 {
1814         int ret;
1815         s64 llen;
1816         loff_t size;
1817         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1818         struct buffer_head *di_bh = NULL;
1819         handle_t *handle;
1820         unsigned long long max_off = inode->i_sb->s_maxbytes;
1821
1822         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1823                 return -EROFS;
1824
1825         mutex_lock(&inode->i_mutex);
1826
1827         /*
1828          * This prevents concurrent writes on other nodes
1829          */
1830         ret = ocfs2_rw_lock(inode, 1);
1831         if (ret) {
1832                 mlog_errno(ret);
1833                 goto out;
1834         }
1835
1836         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1837         if (ret) {
1838                 mlog_errno(ret);
1839                 goto out_rw_unlock;
1840         }
1841
1842         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1843                 ret = -EPERM;
1844                 goto out_inode_unlock;
1845         }
1846
1847         switch (sr->l_whence) {
1848         case 0: /*SEEK_SET*/
1849                 break;
1850         case 1: /*SEEK_CUR*/
1851                 sr->l_start += f_pos;
1852                 break;
1853         case 2: /*SEEK_END*/
1854                 sr->l_start += i_size_read(inode);
1855                 break;
1856         default:
1857                 ret = -EINVAL;
1858                 goto out_inode_unlock;
1859         }
1860         sr->l_whence = 0;
1861
1862         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1863
1864         if (sr->l_start < 0
1865             || sr->l_start > max_off
1866             || (sr->l_start + llen) < 0
1867             || (sr->l_start + llen) > max_off) {
1868                 ret = -EINVAL;
1869                 goto out_inode_unlock;
1870         }
1871         size = sr->l_start + sr->l_len;
1872
1873         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1874                 if (sr->l_len <= 0) {
1875                         ret = -EINVAL;
1876                         goto out_inode_unlock;
1877                 }
1878         }
1879
1880         if (file && should_remove_suid(file->f_path.dentry)) {
1881                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1882                 if (ret) {
1883                         mlog_errno(ret);
1884                         goto out_inode_unlock;
1885                 }
1886         }
1887
1888         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1889         switch (cmd) {
1890         case OCFS2_IOC_RESVSP:
1891         case OCFS2_IOC_RESVSP64:
1892                 /*
1893                  * This takes unsigned offsets, but the signed ones we
1894                  * pass have been checked against overflow above.
1895                  */
1896                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1897                                                        sr->l_len);
1898                 break;
1899         case OCFS2_IOC_UNRESVSP:
1900         case OCFS2_IOC_UNRESVSP64:
1901                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1902                                                sr->l_len);
1903                 break;
1904         default:
1905                 ret = -EINVAL;
1906         }
1907         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1908         if (ret) {
1909                 mlog_errno(ret);
1910                 goto out_inode_unlock;
1911         }
1912
1913         /*
1914          * We update c/mtime for these changes
1915          */
1916         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1917         if (IS_ERR(handle)) {
1918                 ret = PTR_ERR(handle);
1919                 mlog_errno(ret);
1920                 goto out_inode_unlock;
1921         }
1922
1923         if (change_size && i_size_read(inode) < size)
1924                 i_size_write(inode, size);
1925
1926         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1927         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1928         if (ret < 0)
1929                 mlog_errno(ret);
1930
1931         if (file && (file->f_flags & O_SYNC))
1932                 handle->h_sync = 1;
1933
1934         ocfs2_commit_trans(osb, handle);
1935
1936 out_inode_unlock:
1937         brelse(di_bh);
1938         ocfs2_inode_unlock(inode, 1);
1939 out_rw_unlock:
1940         ocfs2_rw_unlock(inode, 1);
1941
1942 out:
1943         mutex_unlock(&inode->i_mutex);
1944         return ret;
1945 }
1946
1947 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1948                             struct ocfs2_space_resv *sr)
1949 {
1950         struct inode *inode = file_inode(file);
1951         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1952         int ret;
1953
1954         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1955             !ocfs2_writes_unwritten_extents(osb))
1956                 return -ENOTTY;
1957         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1958                  !ocfs2_sparse_alloc(osb))
1959                 return -ENOTTY;
1960
1961         if (!S_ISREG(inode->i_mode))
1962                 return -EINVAL;
1963
1964         if (!(file->f_mode & FMODE_WRITE))
1965                 return -EBADF;
1966
1967         ret = mnt_want_write_file(file);
1968         if (ret)
1969                 return ret;
1970         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1971         mnt_drop_write_file(file);
1972         return ret;
1973 }
1974
1975 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1976                             loff_t len)
1977 {
1978         struct inode *inode = file_inode(file);
1979         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1980         struct ocfs2_space_resv sr;
1981         int change_size = 1;
1982         int cmd = OCFS2_IOC_RESVSP64;
1983
1984         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1985                 return -EOPNOTSUPP;
1986         if (!ocfs2_writes_unwritten_extents(osb))
1987                 return -EOPNOTSUPP;
1988
1989         if (mode & FALLOC_FL_KEEP_SIZE)
1990                 change_size = 0;
1991
1992         if (mode & FALLOC_FL_PUNCH_HOLE)
1993                 cmd = OCFS2_IOC_UNRESVSP64;
1994
1995         sr.l_whence = 0;
1996         sr.l_start = (s64)offset;
1997         sr.l_len = (s64)len;
1998
1999         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2000                                          change_size);
2001 }
2002
2003 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2004                                    size_t count)
2005 {
2006         int ret = 0;
2007         unsigned int extent_flags;
2008         u32 cpos, clusters, extent_len, phys_cpos;
2009         struct super_block *sb = inode->i_sb;
2010
2011         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2012             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2013             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2014                 return 0;
2015
2016         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2017         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2018
2019         while (clusters) {
2020                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2021                                          &extent_flags);
2022                 if (ret < 0) {
2023                         mlog_errno(ret);
2024                         goto out;
2025                 }
2026
2027                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2028                         ret = 1;
2029                         break;
2030                 }
2031
2032                 if (extent_len > clusters)
2033                         extent_len = clusters;
2034
2035                 clusters -= extent_len;
2036                 cpos += extent_len;
2037         }
2038 out:
2039         return ret;
2040 }
2041
2042 static void ocfs2_aiodio_wait(struct inode *inode)
2043 {
2044         wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
2045
2046         wait_event(*wq, (atomic_read(&OCFS2_I(inode)->ip_unaligned_aio) == 0));
2047 }
2048
2049 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2050 {
2051         int blockmask = inode->i_sb->s_blocksize - 1;
2052         loff_t final_size = pos + count;
2053
2054         if ((pos & blockmask) || (final_size & blockmask))
2055                 return 1;
2056         return 0;
2057 }
2058
2059 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2060                                             struct file *file,
2061                                             loff_t pos, size_t count,
2062                                             int *meta_level)
2063 {
2064         int ret;
2065         struct buffer_head *di_bh = NULL;
2066         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2067         u32 clusters =
2068                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2069
2070         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2071         if (ret) {
2072                 mlog_errno(ret);
2073                 goto out;
2074         }
2075
2076         *meta_level = 1;
2077
2078         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2079         if (ret)
2080                 mlog_errno(ret);
2081 out:
2082         brelse(di_bh);
2083         return ret;
2084 }
2085
2086 static int ocfs2_prepare_inode_for_write(struct file *file,
2087                                          loff_t *ppos,
2088                                          size_t count,
2089                                          int appending,
2090                                          int *direct_io,
2091                                          int *has_refcount)
2092 {
2093         int ret = 0, meta_level = 0;
2094         struct dentry *dentry = file->f_path.dentry;
2095         struct inode *inode = dentry->d_inode;
2096         loff_t saved_pos = 0, end;
2097
2098         /*
2099          * We start with a read level meta lock and only jump to an ex
2100          * if we need to make modifications here.
2101          */
2102         for(;;) {
2103                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2104                 if (ret < 0) {
2105                         meta_level = -1;
2106                         mlog_errno(ret);
2107                         goto out;
2108                 }
2109
2110                 /* Clear suid / sgid if necessary. We do this here
2111                  * instead of later in the write path because
2112                  * remove_suid() calls ->setattr without any hint that
2113                  * we may have already done our cluster locking. Since
2114                  * ocfs2_setattr() *must* take cluster locks to
2115                  * proceed, this will lead us to recursively lock the
2116                  * inode. There's also the dinode i_size state which
2117                  * can be lost via setattr during extending writes (we
2118                  * set inode->i_size at the end of a write. */
2119                 if (should_remove_suid(dentry)) {
2120                         if (meta_level == 0) {
2121                                 ocfs2_inode_unlock(inode, meta_level);
2122                                 meta_level = 1;
2123                                 continue;
2124                         }
2125
2126                         ret = ocfs2_write_remove_suid(inode);
2127                         if (ret < 0) {
2128                                 mlog_errno(ret);
2129                                 goto out_unlock;
2130                         }
2131                 }
2132
2133                 /* work on a copy of ppos until we're sure that we won't have
2134                  * to recalculate it due to relocking. */
2135                 if (appending)
2136                         saved_pos = i_size_read(inode);
2137                 else
2138                         saved_pos = *ppos;
2139
2140                 end = saved_pos + count;
2141
2142                 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2143                 if (ret == 1) {
2144                         ocfs2_inode_unlock(inode, meta_level);
2145                         meta_level = -1;
2146
2147                         ret = ocfs2_prepare_inode_for_refcount(inode,
2148                                                                file,
2149                                                                saved_pos,
2150                                                                count,
2151                                                                &meta_level);
2152                         if (has_refcount)
2153                                 *has_refcount = 1;
2154                         if (direct_io)
2155                                 *direct_io = 0;
2156                 }
2157
2158                 if (ret < 0) {
2159                         mlog_errno(ret);
2160                         goto out_unlock;
2161                 }
2162
2163                 /*
2164                  * Skip the O_DIRECT checks if we don't need
2165                  * them.
2166                  */
2167                 if (!direct_io || !(*direct_io))
2168                         break;
2169
2170                 /*
2171                  * There's no sane way to do direct writes to an inode
2172                  * with inline data.
2173                  */
2174                 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2175                         *direct_io = 0;
2176                         break;
2177                 }
2178
2179                 /*
2180                  * Allowing concurrent direct writes means
2181                  * i_size changes wouldn't be synchronized, so
2182                  * one node could wind up truncating another
2183                  * nodes writes.
2184                  */
2185                 if (end > i_size_read(inode)) {
2186                         *direct_io = 0;
2187                         break;
2188                 }
2189
2190                 /*
2191                  * We don't fill holes during direct io, so
2192                  * check for them here. If any are found, the
2193                  * caller will have to retake some cluster
2194                  * locks and initiate the io as buffered.
2195                  */
2196                 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2197                 if (ret == 1) {
2198                         *direct_io = 0;
2199                         ret = 0;
2200                 } else if (ret < 0)
2201                         mlog_errno(ret);
2202                 break;
2203         }
2204
2205         if (appending)
2206                 *ppos = saved_pos;
2207
2208 out_unlock:
2209         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2210                                             saved_pos, appending, count,
2211                                             direct_io, has_refcount);
2212
2213         if (meta_level >= 0)
2214                 ocfs2_inode_unlock(inode, meta_level);
2215
2216 out:
2217         return ret;
2218 }
2219
2220 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2221                                     const struct iovec *iov,
2222                                     unsigned long nr_segs,
2223                                     loff_t pos)
2224 {
2225         int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
2226         int can_do_direct, has_refcount = 0;
2227         ssize_t written = 0;
2228         size_t ocount;          /* original count */
2229         size_t count;           /* after file limit checks */
2230         loff_t old_size, *ppos = &iocb->ki_pos;
2231         u32 old_clusters;
2232         struct file *file = iocb->ki_filp;
2233         struct inode *inode = file_inode(file);
2234         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2235         int full_coherency = !(osb->s_mount_opt &
2236                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2237         int unaligned_dio = 0;
2238
2239         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2240                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2241                 file->f_path.dentry->d_name.len,
2242                 file->f_path.dentry->d_name.name,
2243                 (unsigned int)nr_segs);
2244
2245         if (iocb->ki_nbytes == 0)
2246                 return 0;
2247
2248         appending = file->f_flags & O_APPEND ? 1 : 0;
2249         direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2250
2251         mutex_lock(&inode->i_mutex);
2252
2253         ocfs2_iocb_clear_sem_locked(iocb);
2254
2255 relock:
2256         /* to match setattr's i_mutex -> rw_lock ordering */
2257         if (direct_io) {
2258                 have_alloc_sem = 1;
2259                 /* communicate with ocfs2_dio_end_io */
2260                 ocfs2_iocb_set_sem_locked(iocb);
2261         }
2262
2263         /*
2264          * Concurrent O_DIRECT writes are allowed with
2265          * mount_option "coherency=buffered".
2266          */
2267         rw_level = (!direct_io || full_coherency);
2268
2269         ret = ocfs2_rw_lock(inode, rw_level);
2270         if (ret < 0) {
2271                 mlog_errno(ret);
2272                 goto out_sems;
2273         }
2274
2275         /*
2276          * O_DIRECT writes with "coherency=full" need to take EX cluster
2277          * inode_lock to guarantee coherency.
2278          */
2279         if (direct_io && full_coherency) {
2280                 /*
2281                  * We need to take and drop the inode lock to force
2282                  * other nodes to drop their caches.  Buffered I/O
2283                  * already does this in write_begin().
2284                  */
2285                 ret = ocfs2_inode_lock(inode, NULL, 1);
2286                 if (ret < 0) {
2287                         mlog_errno(ret);
2288                         goto out;
2289                 }
2290
2291                 ocfs2_inode_unlock(inode, 1);
2292         }
2293
2294         can_do_direct = direct_io;
2295         ret = ocfs2_prepare_inode_for_write(file, ppos,
2296                                             iocb->ki_nbytes, appending,
2297                                             &can_do_direct, &has_refcount);
2298         if (ret < 0) {
2299                 mlog_errno(ret);
2300                 goto out;
2301         }
2302
2303         if (direct_io && !is_sync_kiocb(iocb))
2304                 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_nbytes,
2305                                                       *ppos);
2306
2307         /*
2308          * We can't complete the direct I/O as requested, fall back to
2309          * buffered I/O.
2310          */
2311         if (direct_io && !can_do_direct) {
2312                 ocfs2_rw_unlock(inode, rw_level);
2313
2314                 have_alloc_sem = 0;
2315                 rw_level = -1;
2316
2317                 direct_io = 0;
2318                 goto relock;
2319         }
2320
2321         if (unaligned_dio) {
2322                 /*
2323                  * Wait on previous unaligned aio to complete before
2324                  * proceeding.
2325                  */
2326                 ocfs2_aiodio_wait(inode);
2327
2328                 /* Mark the iocb as needing a decrement in ocfs2_dio_end_io */
2329                 atomic_inc(&OCFS2_I(inode)->ip_unaligned_aio);
2330                 ocfs2_iocb_set_unaligned_aio(iocb);
2331         }
2332
2333         /*
2334          * To later detect whether a journal commit for sync writes is
2335          * necessary, we sample i_size, and cluster count here.
2336          */
2337         old_size = i_size_read(inode);
2338         old_clusters = OCFS2_I(inode)->ip_clusters;
2339
2340         /* communicate with ocfs2_dio_end_io */
2341         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2342
2343         ret = generic_segment_checks(iov, &nr_segs, &ocount,
2344                                      VERIFY_READ);
2345         if (ret)
2346                 goto out_dio;
2347
2348         count = ocount;
2349         ret = generic_write_checks(file, ppos, &count,
2350                                    S_ISBLK(inode->i_mode));
2351         if (ret)
2352                 goto out_dio;
2353
2354         if (direct_io) {
2355                 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2356                                                     ppos, count, ocount);
2357                 if (written < 0) {
2358                         ret = written;
2359                         goto out_dio;
2360                 }
2361         } else {
2362                 current->backing_dev_info = file->f_mapping->backing_dev_info;
2363                 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2364                                                       ppos, count, 0);
2365                 current->backing_dev_info = NULL;
2366         }
2367
2368 out_dio:
2369         /* buffered aio wouldn't have proper lock coverage today */
2370         BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2371
2372         if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2373             ((file->f_flags & O_DIRECT) && !direct_io)) {
2374                 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2375                                                pos + count - 1);
2376                 if (ret < 0)
2377                         written = ret;
2378
2379                 if (!ret && ((old_size != i_size_read(inode)) ||
2380                              (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2381                              has_refcount)) {
2382                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2383                         if (ret < 0)
2384                                 written = ret;
2385                 }
2386
2387                 if (!ret)
2388                         ret = filemap_fdatawait_range(file->f_mapping, pos,
2389                                                       pos + count - 1);
2390         }
2391
2392         /*
2393          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2394          * function pointer which is called when o_direct io completes so that
2395          * it can unlock our rw lock.
2396          * Unfortunately there are error cases which call end_io and others
2397          * that don't.  so we don't have to unlock the rw_lock if either an
2398          * async dio is going to do it in the future or an end_io after an
2399          * error has already done it.
2400          */
2401         if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2402                 rw_level = -1;
2403                 have_alloc_sem = 0;
2404                 unaligned_dio = 0;
2405         }
2406
2407         if (unaligned_dio) {
2408                 ocfs2_iocb_clear_unaligned_aio(iocb);
2409                 atomic_dec(&OCFS2_I(inode)->ip_unaligned_aio);
2410         }
2411
2412 out:
2413         if (rw_level != -1)
2414                 ocfs2_rw_unlock(inode, rw_level);
2415
2416 out_sems:
2417         if (have_alloc_sem)
2418                 ocfs2_iocb_clear_sem_locked(iocb);
2419
2420         mutex_unlock(&inode->i_mutex);
2421
2422         if (written)
2423                 ret = written;
2424         return ret;
2425 }
2426
2427 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2428                                 struct file *out,
2429                                 struct splice_desc *sd)
2430 {
2431         int ret;
2432
2433         ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2434                                             sd->total_len, 0, NULL, NULL);
2435         if (ret < 0) {
2436                 mlog_errno(ret);
2437                 return ret;
2438         }
2439
2440         return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2441 }
2442
2443 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2444                                        struct file *out,
2445                                        loff_t *ppos,
2446                                        size_t len,
2447                                        unsigned int flags)
2448 {
2449         int ret;
2450         struct address_space *mapping = out->f_mapping;
2451         struct inode *inode = mapping->host;
2452         struct splice_desc sd = {
2453                 .total_len = len,
2454                 .flags = flags,
2455                 .pos = *ppos,
2456                 .u.file = out,
2457         };
2458
2459
2460         trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2461                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2462                         out->f_path.dentry->d_name.len,
2463                         out->f_path.dentry->d_name.name, len);
2464
2465         pipe_lock(pipe);
2466
2467         splice_from_pipe_begin(&sd);
2468         do {
2469                 ret = splice_from_pipe_next(pipe, &sd);
2470                 if (ret <= 0)
2471                         break;
2472
2473                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2474                 ret = ocfs2_rw_lock(inode, 1);
2475                 if (ret < 0)
2476                         mlog_errno(ret);
2477                 else {
2478                         ret = ocfs2_splice_to_file(pipe, out, &sd);
2479                         ocfs2_rw_unlock(inode, 1);
2480                 }
2481                 mutex_unlock(&inode->i_mutex);
2482         } while (ret > 0);
2483         splice_from_pipe_end(pipe, &sd);
2484
2485         pipe_unlock(pipe);
2486
2487         if (sd.num_spliced)
2488                 ret = sd.num_spliced;
2489
2490         if (ret > 0) {
2491                 int err;
2492
2493                 err = generic_write_sync(out, *ppos, ret);
2494                 if (err)
2495                         ret = err;
2496                 else
2497                         *ppos += ret;
2498
2499                 balance_dirty_pages_ratelimited(mapping);
2500         }
2501
2502         return ret;
2503 }
2504
2505 static ssize_t ocfs2_file_splice_read(struct file *in,
2506                                       loff_t *ppos,
2507                                       struct pipe_inode_info *pipe,
2508                                       size_t len,
2509                                       unsigned int flags)
2510 {
2511         int ret = 0, lock_level = 0;
2512         struct inode *inode = file_inode(in);
2513
2514         trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2515                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2516                         in->f_path.dentry->d_name.len,
2517                         in->f_path.dentry->d_name.name, len);
2518
2519         /*
2520          * See the comment in ocfs2_file_aio_read()
2521          */
2522         ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2523         if (ret < 0) {
2524                 mlog_errno(ret);
2525                 goto bail;
2526         }
2527         ocfs2_inode_unlock(inode, lock_level);
2528
2529         ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2530
2531 bail:
2532         return ret;
2533 }
2534
2535 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2536                                    const struct iovec *iov,
2537                                    unsigned long nr_segs,
2538                                    loff_t pos)
2539 {
2540         int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2541         struct file *filp = iocb->ki_filp;
2542         struct inode *inode = file_inode(filp);
2543
2544         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2545                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2546                         filp->f_path.dentry->d_name.len,
2547                         filp->f_path.dentry->d_name.name, nr_segs);
2548
2549
2550         if (!inode) {
2551                 ret = -EINVAL;
2552                 mlog_errno(ret);
2553                 goto bail;
2554         }
2555
2556         ocfs2_iocb_clear_sem_locked(iocb);
2557
2558         /*
2559          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2560          * need locks to protect pending reads from racing with truncate.
2561          */
2562         if (filp->f_flags & O_DIRECT) {
2563                 have_alloc_sem = 1;
2564                 ocfs2_iocb_set_sem_locked(iocb);
2565
2566                 ret = ocfs2_rw_lock(inode, 0);
2567                 if (ret < 0) {
2568                         mlog_errno(ret);
2569                         goto bail;
2570                 }
2571                 rw_level = 0;
2572                 /* communicate with ocfs2_dio_end_io */
2573                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2574         }
2575
2576         /*
2577          * We're fine letting folks race truncates and extending
2578          * writes with read across the cluster, just like they can
2579          * locally. Hence no rw_lock during read.
2580          *
2581          * Take and drop the meta data lock to update inode fields
2582          * like i_size. This allows the checks down below
2583          * generic_file_aio_read() a chance of actually working.
2584          */
2585         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2586         if (ret < 0) {
2587                 mlog_errno(ret);
2588                 goto bail;
2589         }
2590         ocfs2_inode_unlock(inode, lock_level);
2591
2592         ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2593         trace_generic_file_aio_read_ret(ret);
2594
2595         /* buffered aio wouldn't have proper lock coverage today */
2596         BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2597
2598         /* see ocfs2_file_aio_write */
2599         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2600                 rw_level = -1;
2601                 have_alloc_sem = 0;
2602         }
2603
2604 bail:
2605         if (have_alloc_sem)
2606                 ocfs2_iocb_clear_sem_locked(iocb);
2607
2608         if (rw_level != -1)
2609                 ocfs2_rw_unlock(inode, rw_level);
2610
2611         return ret;
2612 }
2613
2614 /* Refer generic_file_llseek_unlocked() */
2615 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2616 {
2617         struct inode *inode = file->f_mapping->host;
2618         int ret = 0;
2619
2620         mutex_lock(&inode->i_mutex);
2621
2622         switch (whence) {
2623         case SEEK_SET:
2624                 break;
2625         case SEEK_END:
2626                 offset += inode->i_size;
2627                 break;
2628         case SEEK_CUR:
2629                 if (offset == 0) {
2630                         offset = file->f_pos;
2631                         goto out;
2632                 }
2633                 offset += file->f_pos;
2634                 break;
2635         case SEEK_DATA:
2636         case SEEK_HOLE:
2637                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2638                 if (ret)
2639                         goto out;
2640                 break;
2641         default:
2642                 ret = -EINVAL;
2643                 goto out;
2644         }
2645
2646         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2647
2648 out:
2649         mutex_unlock(&inode->i_mutex);
2650         if (ret)
2651                 return ret;
2652         return offset;
2653 }
2654
2655 const struct inode_operations ocfs2_file_iops = {
2656         .setattr        = ocfs2_setattr,
2657         .getattr        = ocfs2_getattr,
2658         .permission     = ocfs2_permission,
2659         .setxattr       = generic_setxattr,
2660         .getxattr       = generic_getxattr,
2661         .listxattr      = ocfs2_listxattr,
2662         .removexattr    = generic_removexattr,
2663         .fiemap         = ocfs2_fiemap,
2664         .get_acl        = ocfs2_iop_get_acl,
2665 };
2666
2667 const struct inode_operations ocfs2_special_file_iops = {
2668         .setattr        = ocfs2_setattr,
2669         .getattr        = ocfs2_getattr,
2670         .permission     = ocfs2_permission,
2671         .get_acl        = ocfs2_iop_get_acl,
2672 };
2673
2674 /*
2675  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2676  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2677  */
2678 const struct file_operations ocfs2_fops = {
2679         .llseek         = ocfs2_file_llseek,
2680         .read           = do_sync_read,
2681         .write          = do_sync_write,
2682         .mmap           = ocfs2_mmap,
2683         .fsync          = ocfs2_sync_file,
2684         .release        = ocfs2_file_release,
2685         .open           = ocfs2_file_open,
2686         .aio_read       = ocfs2_file_aio_read,
2687         .aio_write      = ocfs2_file_aio_write,
2688         .unlocked_ioctl = ocfs2_ioctl,
2689 #ifdef CONFIG_COMPAT
2690         .compat_ioctl   = ocfs2_compat_ioctl,
2691 #endif
2692         .lock           = ocfs2_lock,
2693         .flock          = ocfs2_flock,
2694         .splice_read    = ocfs2_file_splice_read,
2695         .splice_write   = ocfs2_file_splice_write,
2696         .fallocate      = ocfs2_fallocate,
2697 };
2698
2699 const struct file_operations ocfs2_dops = {
2700         .llseek         = generic_file_llseek,
2701         .read           = generic_read_dir,
2702         .iterate        = ocfs2_readdir,
2703         .fsync          = ocfs2_sync_file,
2704         .release        = ocfs2_dir_release,
2705         .open           = ocfs2_dir_open,
2706         .unlocked_ioctl = ocfs2_ioctl,
2707 #ifdef CONFIG_COMPAT
2708         .compat_ioctl   = ocfs2_compat_ioctl,
2709 #endif
2710         .lock           = ocfs2_lock,
2711         .flock          = ocfs2_flock,
2712 };
2713
2714 /*
2715  * POSIX-lockless variants of our file_operations.
2716  *
2717  * These will be used if the underlying cluster stack does not support
2718  * posix file locking, if the user passes the "localflocks" mount
2719  * option, or if we have a local-only fs.
2720  *
2721  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2722  * so we still want it in the case of no stack support for
2723  * plocks. Internally, it will do the right thing when asked to ignore
2724  * the cluster.
2725  */
2726 const struct file_operations ocfs2_fops_no_plocks = {
2727         .llseek         = ocfs2_file_llseek,
2728         .read           = do_sync_read,
2729         .write          = do_sync_write,
2730         .mmap           = ocfs2_mmap,
2731         .fsync          = ocfs2_sync_file,
2732         .release        = ocfs2_file_release,
2733         .open           = ocfs2_file_open,
2734         .aio_read       = ocfs2_file_aio_read,
2735         .aio_write      = ocfs2_file_aio_write,
2736         .unlocked_ioctl = ocfs2_ioctl,
2737 #ifdef CONFIG_COMPAT
2738         .compat_ioctl   = ocfs2_compat_ioctl,
2739 #endif
2740         .flock          = ocfs2_flock,
2741         .splice_read    = ocfs2_file_splice_read,
2742         .splice_write   = ocfs2_file_splice_write,
2743         .fallocate      = ocfs2_fallocate,
2744 };
2745
2746 const struct file_operations ocfs2_dops_no_plocks = {
2747         .llseek         = generic_file_llseek,
2748         .read           = generic_read_dir,
2749         .iterate        = ocfs2_readdir,
2750         .fsync          = ocfs2_sync_file,
2751         .release        = ocfs2_dir_release,
2752         .open           = ocfs2_dir_open,
2753         .unlocked_ioctl = ocfs2_ioctl,
2754 #ifdef CONFIG_COMPAT
2755         .compat_ioctl   = ocfs2_compat_ioctl,
2756 #endif
2757         .flock          = ocfs2_flock,
2758 };