fs/ocfs2/file.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * file.c
   5  *
   6  * File open, close, extend, truncate
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/capability.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/uio.h>
  33 #include <linux/sched.h>
  34 #include <linux/splice.h>
  35 #include <linux/mount.h>
  36 #include <linux/writeback.h>
  37 #include <linux/falloc.h>
  38 #include <linux/quotaops.h>
  39 #include <linux/blkdev.h>
  40
  41 #include <cluster/masklog.h>
  42
  43 #include "ocfs2.h"
  44
  45 #include "alloc.h"
  46 #include "aops.h"
  47 #include "dir.h"
  48 #include "dlmglue.h"
  49 #include "extent_map.h"
  50 #include "file.h"
  51 #include "sysfile.h"
  52 #include "inode.h"
  53 #include "ioctl.h"
  54 #include "journal.h"
  55 #include "locks.h"
  56 #include "mmap.h"
  57 #include "suballoc.h"
  58 #include "super.h"
  59 #include "xattr.h"
  60 #include "acl.h"
  61 #include "quota.h"
  62 #include "refcounttree.h"
  63 #include "ocfs2_trace.h"
  64
  65 #include "buffer_head_io.h"
  66
  67 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  68 {
  69         struct ocfs2_file_private *fp;
  70
  71         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  72         if (!fp)
  73                 return -ENOMEM;
  74
  75         fp->fp_file = file;
  76         mutex_init(&fp->fp_mutex);
  77         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  78         file->private_data = fp;
  79
  80         return 0;
  81 }
  82
  83 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  84 {
  85         struct ocfs2_file_private *fp = file->private_data;
  86         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  87
  88         if (fp) {
  89                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  90                 ocfs2_lock_res_free(&fp->fp_flock);
  91                 kfree(fp);
  92                 file->private_data = NULL;
  93         }
  94 }
  95
  96 static int ocfs2_file_open(struct inode *inode, struct file *file)
  97 {
  98         int status;
  99         int mode = file->f_flags;
 100         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 101
 102         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
 103                               (unsigned long long)OCFS2_I(inode)->ip_blkno,
 104                               file->f_path.dentry->d_name.len,
 105                               file->f_path.dentry->d_name.name, mode);
 106
 107         if (file->f_mode & FMODE_WRITE)
 108                 dquot_initialize(inode);
 109
 110         spin_lock(&oi->ip_lock);
 111
 112         /* Check that the inode hasn't been wiped from disk by another
 113          * node. If it hasn't then we're safe as long as we hold the
 114          * spin lock until our increment of open count. */
 115         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
 116                 spin_unlock(&oi->ip_lock);
 117
 118                 status = -ENOENT;
 119                 goto leave;
 120         }
 121
 122         if (mode & O_DIRECT)
 123                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 124
 125         oi->ip_open_count++;
 126         spin_unlock(&oi->ip_lock);
 127
 128         status = ocfs2_init_file_private(inode, file);
 129         if (status) {
 130                 /*
 131                  * We want to set open count back if we're failing the
 132                  * open.
 133                  */
 134                 spin_lock(&oi->ip_lock);
 135                 oi->ip_open_count--;
 136                 spin_unlock(&oi->ip_lock);
 137         }
 138
 139 leave:
 140         return status;
 141 }
 142
 143 static int ocfs2_file_release(struct inode *inode, struct file *file)
 144 {
 145         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 146
 147         spin_lock(&oi->ip_lock);
 148         if (!--oi->ip_open_count)
 149                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 150
 151         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 152                                  oi->ip_blkno,
 153                                  file->f_path.dentry->d_name.len,
 154                                  file->f_path.dentry->d_name.name,
 155                                  oi->ip_open_count);
 156         spin_unlock(&oi->ip_lock);
 157
 158         ocfs2_free_file_private(inode, file);
 159
 160         return 0;
 161 }
 162
 163 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 164 {
 165         return ocfs2_init_file_private(inode, file);
 166 }
 167
 168 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 169 {
 170         ocfs2_free_file_private(inode, file);
 171         return 0;
 172 }
 173
 174 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 175                            int datasync)
 176 {
 177         int err = 0;
 178         journal_t *journal;
 179         struct inode *inode = file->f_mapping->host;
 180         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 181
 182         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 183                               OCFS2_I(inode)->ip_blkno,
 184                               file->f_path.dentry->d_name.len,
 185                               file->f_path.dentry->d_name.name,
 186                               (unsigned long long)datasync);
 187
 188         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 189                 return -EROFS;
 190
 191         err = filemap_write_and_wait_range(inode->i_mapping, start, end);
 192         if (err)
 193                 return err;
 194
 195         /*
 196          * Probably don't need the i_mutex at all in here, just putting it here
 197          * to be consistent with how fsync used to be called, someone more
 198          * familiar with the fs could possibly remove it.
 199          */
 200         mutex_lock(&inode->i_mutex);
 201         if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
 202                 /*
 203                  * We still have to flush drive's caches to get data to the
 204                  * platter
 205                  */
 206                 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
 207                         blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 208                 goto bail;
 209         }
 210
 211         journal = osb->journal->j_journal;
 212         err = jbd2_journal_force_commit(journal);
 213
 214 bail:
 215         if (err)
 216                 mlog_errno(err);
 217         mutex_unlock(&inode->i_mutex);
 218
 219         return (err < 0) ? -EIO : 0;
 220 }
 221
 222 int ocfs2_should_update_atime(struct inode *inode,
 223                               struct vfsmount *vfsmnt)
 224 {
 225         struct timespec now;
 226         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 227
 228         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 229                 return 0;
 230
 231         if ((inode->i_flags & S_NOATIME) ||
 232             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 233                 return 0;
 234
 235         /*
 236          * We can be called with no vfsmnt structure - NFSD will
 237          * sometimes do this.
 238          *
 239          * Note that our action here is different than touch_atime() -
 240          * if we can't tell whether this is a noatime mount, then we
 241          * don't know whether to trust the value of s_atime_quantum.
 242          */
 243         if (vfsmnt == NULL)
 244                 return 0;
 245
 246         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 247             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 248                 return 0;
 249
 250         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 251                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 252                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 253                         return 1;
 254
 255                 return 0;
 256         }
 257
 258         now = CURRENT_TIME;
 259         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 260                 return 0;
 261         else
 262                 return 1;
 263 }
 264
 265 int ocfs2_update_inode_atime(struct inode *inode,
 266                              struct buffer_head *bh)
 267 {
 268         int ret;
 269         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 270         handle_t *handle;
 271         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 272
 273         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 274         if (IS_ERR(handle)) {
 275                 ret = PTR_ERR(handle);
 276                 mlog_errno(ret);
 277                 goto out;
 278         }
 279
 280         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 281                                       OCFS2_JOURNAL_ACCESS_WRITE);
 282         if (ret) {
 283                 mlog_errno(ret);
 284                 goto out_commit;
 285         }
 286
 287         /*
 288          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 289          * have i_mutex to guard against concurrent changes to other
 290          * inode fields.
 291          */
 292         inode->i_atime = CURRENT_TIME;
 293         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 294         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 295         ocfs2_journal_dirty(handle, bh);
 296
 297 out_commit:
 298         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 299 out:
 300         return ret;
 301 }
 302
 303 static int ocfs2_set_inode_size(handle_t *handle,
 304                                 struct inode *inode,
 305                                 struct buffer_head *fe_bh,
 306                                 u64 new_i_size)
 307 {
 308         int status;
 309
 310         i_size_write(inode, new_i_size);
 311         inode->i_blocks = ocfs2_inode_sector_count(inode);
 312         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 313
 314         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 315         if (status < 0) {
 316                 mlog_errno(status);
 317                 goto bail;
 318         }
 319
 320 bail:
 321         return status;
 322 }
 323
 324 int ocfs2_simple_size_update(struct inode *inode,
 325                              struct buffer_head *di_bh,
 326                              u64 new_i_size)
 327 {
 328         int ret;
 329         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 330         handle_t *handle = NULL;
 331
 332         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 333         if (IS_ERR(handle)) {
 334                 ret = PTR_ERR(handle);
 335                 mlog_errno(ret);
 336                 goto out;
 337         }
 338
 339         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 340                                    new_i_size);
 341         if (ret < 0)
 342                 mlog_errno(ret);
 343
 344         ocfs2_commit_trans(osb, handle);
 345 out:
 346         return ret;
 347 }
 348
 349 static int ocfs2_cow_file_pos(struct inode *inode,
 350                               struct buffer_head *fe_bh,
 351                               u64 offset)
 352 {
 353         int status;
 354         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 355         unsigned int num_clusters = 0;
 356         unsigned int ext_flags = 0;
 357
 358         /*
 359          * If the new offset is aligned to the range of the cluster, there is
 360          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 361          * CoW either.
 362          */
 363         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 364                 return 0;
 365
 366         status = ocfs2_get_clusters(inode, cpos, &phys,
 367                                     &num_clusters, &ext_flags);
 368         if (status) {
 369                 mlog_errno(status);
 370                 goto out;
 371         }
 372
 373         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 374                 goto out;
 375
 376         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 377
 378 out:
 379         return status;
 380 }
 381
 382 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 383                                      struct inode *inode,
 384                                      struct buffer_head *fe_bh,
 385                                      u64 new_i_size)
 386 {
 387         int status;
 388         handle_t *handle;
 389         struct ocfs2_dinode *di;
 390         u64 cluster_bytes;
 391
 392         /*
 393          * We need to CoW the cluster contains the offset if it is reflinked
 394          * since we will call ocfs2_zero_range_for_truncate later which will
 395          * write "0" from offset to the end of the cluster.
 396          */
 397         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 398         if (status) {
 399                 mlog_errno(status);
 400                 return status;
 401         }
 402
 403         /* TODO: This needs to actually orphan the inode in this
 404          * transaction. */
 405
 406         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 407         if (IS_ERR(handle)) {
 408                 status = PTR_ERR(handle);
 409                 mlog_errno(status);
 410                 goto out;
 411         }
 412
 413         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 414                                          OCFS2_JOURNAL_ACCESS_WRITE);
 415         if (status < 0) {
 416                 mlog_errno(status);
 417                 goto out_commit;
 418         }
 419
 420         /*
 421          * Do this before setting i_size.
 422          */
 423         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 424         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 425                                                cluster_bytes);
 426         if (status) {
 427                 mlog_errno(status);
 428                 goto out_commit;
 429         }
 430
 431         i_size_write(inode, new_i_size);
 432         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 433
 434         di = (struct ocfs2_dinode *) fe_bh->b_data;
 435         di->i_size = cpu_to_le64(new_i_size);
 436         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 437         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 438
 439         ocfs2_journal_dirty(handle, fe_bh);
 440
 441 out_commit:
 442         ocfs2_commit_trans(osb, handle);
 443 out:
 444         return status;
 445 }
 446
 447 static int ocfs2_truncate_file(struct inode *inode,
 448                                struct buffer_head *di_bh,
 449                                u64 new_i_size)
 450 {
 451         int status = 0;
 452         struct ocfs2_dinode *fe = NULL;
 453         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 454
 455         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 456          * already validated it */
 457         fe = (struct ocfs2_dinode *) di_bh->b_data;
 458
 459         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 460                                   (unsigned long long)le64_to_cpu(fe->i_size),
 461                                   (unsigned long long)new_i_size);
 462
 463         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 464                         "Inode %llu, inode i_size = %lld != di "
 465                         "i_size = %llu, i_flags = 0x%x\n",
 466                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 467                         i_size_read(inode),
 468                         (unsigned long long)le64_to_cpu(fe->i_size),
 469                         le32_to_cpu(fe->i_flags));
 470
 471         if (new_i_size > le64_to_cpu(fe->i_size)) {
 472                 trace_ocfs2_truncate_file_error(
 473                         (unsigned long long)le64_to_cpu(fe->i_size),
 474                         (unsigned long long)new_i_size);
 475                 status = -EINVAL;
 476                 mlog_errno(status);
 477                 goto bail;
 478         }
 479
 480         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 481
 482         ocfs2_resv_discard(&osb->osb_la_resmap,
 483                            &OCFS2_I(inode)->ip_la_data_resv);
 484
 485         /*
 486          * The inode lock forced other nodes to sync and drop their
 487          * pages, which (correctly) happens even if we have a truncate
 488          * without allocation change - ocfs2 cluster sizes can be much
 489          * greater than page size, so we have to truncate them
 490          * anyway.
 491          */
 492         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 493         truncate_inode_pages(inode->i_mapping, new_i_size);
 494
 495         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 496                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 497                                                i_size_read(inode), 1);
 498                 if (status)
 499                         mlog_errno(status);
 500
 501                 goto bail_unlock_sem;
 502         }
 503
 504         /* alright, we're going to need to do a full blown alloc size
 505          * change. Orphan the inode so that recovery can complete the
 506          * truncate if necessary. This does the task of marking
 507          * i_size. */
 508         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 509         if (status < 0) {
 510                 mlog_errno(status);
 511                 goto bail_unlock_sem;
 512         }
 513
 514         status = ocfs2_commit_truncate(osb, inode, di_bh);
 515         if (status < 0) {
 516                 mlog_errno(status);
 517                 goto bail_unlock_sem;
 518         }
 519
 520         /* TODO: orphan dir cleanup here. */
 521 bail_unlock_sem:
 522         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 523
 524 bail:
 525         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 526                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 527
 528         return status;
 529 }
 530
 531 /*
 532  * extend file allocation only here.
 533  * we'll update all the disk stuff, and oip->alloc_size
 534  *
 535  * expect stuff to be locked, a transaction started and enough data /
 536  * metadata reservations in the contexts.
 537  *
 538  * Will return -EAGAIN, and a reason if a restart is needed.
 539  * If passed in, *reason will always be set, even in error.
 540  */
 541 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 542                          struct inode *inode,
 543                          u32 *logical_offset,
 544                          u32 clusters_to_add,
 545                          int mark_unwritten,
 546                          struct buffer_head *fe_bh,
 547                          handle_t *handle,
 548                          struct ocfs2_alloc_context *data_ac,
 549                          struct ocfs2_alloc_context *meta_ac,
 550                          enum ocfs2_alloc_restarted *reason_ret)
 551 {
 552         int ret;
 553         struct ocfs2_extent_tree et;
 554
 555         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 556         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 557                                           clusters_to_add, mark_unwritten,
 558                                           data_ac, meta_ac, reason_ret);
 559
 560         return ret;
 561 }
 562
 563 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 564                                      u32 clusters_to_add, int mark_unwritten)
 565 {
 566         int status = 0;
 567         int restart_func = 0;
 568         int credits;
 569         u32 prev_clusters;
 570         struct buffer_head *bh = NULL;
 571         struct ocfs2_dinode *fe = NULL;
 572         handle_t *handle = NULL;
 573         struct ocfs2_alloc_context *data_ac = NULL;
 574         struct ocfs2_alloc_context *meta_ac = NULL;
 575         enum ocfs2_alloc_restarted why;
 576         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 577         struct ocfs2_extent_tree et;
 578         int did_quota = 0;
 579
 580         /*
 581          * Unwritten extent only exists for file systems which
 582          * support holes.
 583          */
 584         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 585
 586         status = ocfs2_read_inode_block(inode, &bh);
 587         if (status < 0) {
 588                 mlog_errno(status);
 589                 goto leave;
 590         }
 591         fe = (struct ocfs2_dinode *) bh->b_data;
 592
 593 restart_all:
 594         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 595
 596         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 597         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 598                                        &data_ac, &meta_ac);
 599         if (status) {
 600                 mlog_errno(status);
 601                 goto leave;
 602         }
 603
 604         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 605         handle = ocfs2_start_trans(osb, credits);
 606         if (IS_ERR(handle)) {
 607                 status = PTR_ERR(handle);
 608                 handle = NULL;
 609                 mlog_errno(status);
 610                 goto leave;
 611         }
 612
 613 restarted_transaction:
 614         trace_ocfs2_extend_allocation(
 615                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 616                 (unsigned long long)i_size_read(inode),
 617                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 618                 why, restart_func);
 619
 620         status = dquot_alloc_space_nodirty(inode,
 621                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 622         if (status)
 623                 goto leave;
 624         did_quota = 1;
 625
 626         /* reserve a write to the file entry early on - that we if we
 627          * run out of credits in the allocation path, we can still
 628          * update i_size. */
 629         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 630                                          OCFS2_JOURNAL_ACCESS_WRITE);
 631         if (status < 0) {
 632                 mlog_errno(status);
 633                 goto leave;
 634         }
 635
 636         prev_clusters = OCFS2_I(inode)->ip_clusters;
 637
 638         status = ocfs2_add_inode_data(osb,
 639                                       inode,
 640                                       &logical_start,
 641                                       clusters_to_add,
 642                                       mark_unwritten,
 643                                       bh,
 644                                       handle,
 645                                       data_ac,
 646                                       meta_ac,
 647                                       &why);
 648         if ((status < 0) && (status != -EAGAIN)) {
 649                 if (status != -ENOSPC)
 650                         mlog_errno(status);
 651                 goto leave;
 652         }
 653
 654         ocfs2_journal_dirty(handle, bh);
 655
 656         spin_lock(&OCFS2_I(inode)->ip_lock);
 657         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 658         spin_unlock(&OCFS2_I(inode)->ip_lock);
 659         /* Release unused quota reservation */
 660         dquot_free_space(inode,
 661                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 662         did_quota = 0;
 663
 664         if (why != RESTART_NONE && clusters_to_add) {
 665                 if (why == RESTART_META) {
 666                         restart_func = 1;
 667                         status = 0;
 668                 } else {
 669                         BUG_ON(why != RESTART_TRANS);
 670
 671                         status = ocfs2_allocate_extend_trans(handle, 1);
 672                         if (status < 0) {
 673                                 /* handle still has to be committed at
 674                                  * this point. */
 675                                 status = -ENOMEM;
 676                                 mlog_errno(status);
 677                                 goto leave;
 678                         }
 679                         goto restarted_transaction;
 680                 }
 681         }
 682
 683         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 684              le32_to_cpu(fe->i_clusters),
 685              (unsigned long long)le64_to_cpu(fe->i_size),
 686              OCFS2_I(inode)->ip_clusters,
 687              (unsigned long long)i_size_read(inode));
 688
 689 leave:
 690         if (status < 0 && did_quota)
 691                 dquot_free_space(inode,
 692                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 693         if (handle) {
 694                 ocfs2_commit_trans(osb, handle);
 695                 handle = NULL;
 696         }
 697         if (data_ac) {
 698                 ocfs2_free_alloc_context(data_ac);
 699                 data_ac = NULL;
 700         }
 701         if (meta_ac) {
 702                 ocfs2_free_alloc_context(meta_ac);
 703                 meta_ac = NULL;
 704         }
 705         if ((!status) && restart_func) {
 706                 restart_func = 0;
 707                 goto restart_all;
 708         }
 709         brelse(bh);
 710         bh = NULL;
 711
 712         return status;
 713 }
 714
 715 /*
 716  * While a write will already be ordering the data, a truncate will not.
 717  * Thus, we need to explicitly order the zeroed pages.
 718  */
 719 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
 720 {
 721         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 722         handle_t *handle = NULL;
 723         int ret = 0;
 724
 725         if (!ocfs2_should_order_data(inode))
 726                 goto out;
 727
 728         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 729         if (IS_ERR(handle)) {
 730                 ret = -ENOMEM;
 731                 mlog_errno(ret);
 732                 goto out;
 733         }
 734
 735         ret = ocfs2_jbd2_file_inode(handle, inode);
 736         if (ret < 0)
 737                 mlog_errno(ret);
 738
 739 out:
 740         if (ret) {
 741                 if (!IS_ERR(handle))
 742                         ocfs2_commit_trans(osb, handle);
 743                 handle = ERR_PTR(ret);
 744         }
 745         return handle;
 746 }
 747
 748 /* Some parts of this taken from generic_cont_expand, which turned out
 749  * to be too fragile to do exactly what we need without us having to
 750  * worry about recursive locking in ->write_begin() and ->write_end(). */
 751 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 752                                  u64 abs_to)
 753 {
 754         struct address_space *mapping = inode->i_mapping;
 755         struct page *page;
 756         unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
 757         handle_t *handle = NULL;
 758         int ret = 0;
 759         unsigned zero_from, zero_to, block_start, block_end;
 760
 761         BUG_ON(abs_from >= abs_to);
 762         BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
 763         BUG_ON(abs_from & (inode->i_blkbits - 1));
 764
 765         page = find_or_create_page(mapping, index, GFP_NOFS);
 766         if (!page) {
 767                 ret = -ENOMEM;
 768                 mlog_errno(ret);
 769                 goto out;
 770         }
 771
 772         /* Get the offsets within the page that we want to zero */
 773         zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
 774         zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
 775         if (!zero_to)
 776                 zero_to = PAGE_CACHE_SIZE;
 777
 778         trace_ocfs2_write_zero_page(
 779                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 780                         (unsigned long long)abs_from,
 781                         (unsigned long long)abs_to,
 782                         index, zero_from, zero_to);
 783
 784         /* We know that zero_from is block aligned */
 785         for (block_start = zero_from; block_start < zero_to;
 786              block_start = block_end) {
 787                 block_end = block_start + (1 << inode->i_blkbits);
 788
 789                 /*
 790                  * block_start is block-aligned.  Bump it by one to force
 791                  * __block_write_begin and block_commit_write to zero the
 792                  * whole block.
 793                  */
 794                 ret = __block_write_begin(page, block_start + 1, 0,
 795                                           ocfs2_get_block);
 796                 if (ret < 0) {
 797                         mlog_errno(ret);
 798                         goto out_unlock;
 799                 }
 800
 801                 if (!handle) {
 802                         handle = ocfs2_zero_start_ordered_transaction(inode);
 803                         if (IS_ERR(handle)) {
 804                                 ret = PTR_ERR(handle);
 805                                 handle = NULL;
 806                                 break;
 807                         }
 808                 }
 809
 810                 /* must not update i_size! */
 811                 ret = block_commit_write(page, block_start + 1,
 812                                          block_start + 1);
 813                 if (ret < 0)
 814                         mlog_errno(ret);
 815                 else
 816                         ret = 0;
 817         }
 818
 819         if (handle)
 820                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 821
 822 out_unlock:
 823         unlock_page(page);
 824         page_cache_release(page);
 825 out:
 826         return ret;
 827 }
 828
 829 /*
 830  * Find the next range to zero.  We do this in terms of bytes because
 831  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 832  * pagecache.  We may return multiple extents.
 833  *
 834  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 835  * needs to be zeroed.  range_start and range_end return the next zeroing
 836  * range.  A subsequent call should pass the previous range_end as its
 837  * zero_start.  If range_end is 0, there's nothing to do.
 838  *
 839  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 840  */
 841 static int ocfs2_zero_extend_get_range(struct inode *inode,
 842                                        struct buffer_head *di_bh,
 843                                        u64 zero_start, u64 zero_end,
 844                                        u64 *range_start, u64 *range_end)
 845 {
 846         int rc = 0, needs_cow = 0;
 847         u32 p_cpos, zero_clusters = 0;
 848         u32 zero_cpos =
 849                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 850         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 851         unsigned int num_clusters = 0;
 852         unsigned int ext_flags = 0;
 853
 854         while (zero_cpos < last_cpos) {
 855                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 856                                         &num_clusters, &ext_flags);
 857                 if (rc) {
 858                         mlog_errno(rc);
 859                         goto out;
 860                 }
 861
 862                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 863                         zero_clusters = num_clusters;
 864                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 865                                 needs_cow = 1;
 866                         break;
 867                 }
 868
 869                 zero_cpos += num_clusters;
 870         }
 871         if (!zero_clusters) {
 872                 *range_end = 0;
 873                 goto out;
 874         }
 875
 876         while ((zero_cpos + zero_clusters) < last_cpos) {
 877                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 878                                         &p_cpos, &num_clusters,
 879                                         &ext_flags);
 880                 if (rc) {
 881                         mlog_errno(rc);
 882                         goto out;
 883                 }
 884
 885                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 886                         break;
 887                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 888                         needs_cow = 1;
 889                 zero_clusters += num_clusters;
 890         }
 891         if ((zero_cpos + zero_clusters) > last_cpos)
 892                 zero_clusters = last_cpos - zero_cpos;
 893
 894         if (needs_cow) {
 895                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 896                                         zero_clusters, UINT_MAX);
 897                 if (rc) {
 898                         mlog_errno(rc);
 899                         goto out;
 900                 }
 901         }
 902
 903         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 904         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 905                                              zero_cpos + zero_clusters);
 906
 907 out:
 908         return rc;
 909 }
 910
 911 /*
 912  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 913  * has made sure that the entire range needs zeroing.
 914  */
 915 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 916                                    u64 range_end)
 917 {
 918         int rc = 0;
 919         u64 next_pos;
 920         u64 zero_pos = range_start;
 921
 922         trace_ocfs2_zero_extend_range(
 923                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 924                         (unsigned long long)range_start,
 925                         (unsigned long long)range_end);
 926         BUG_ON(range_start >= range_end);
 927
 928         while (zero_pos < range_end) {
 929                 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
 930                 if (next_pos > range_end)
 931                         next_pos = range_end;
 932                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
 933                 if (rc < 0) {
 934                         mlog_errno(rc);
 935                         break;
 936                 }
 937                 zero_pos = next_pos;
 938
 939                 /*
 940                  * Very large extends have the potential to lock up
 941                  * the cpu for extended periods of time.
 942                  */
 943                 cond_resched();
 944         }
 945
 946         return rc;
 947 }
 948
 949 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 950                       loff_t zero_to_size)
 951 {
 952         int ret = 0;
 953         u64 zero_start, range_start = 0, range_end = 0;
 954         struct super_block *sb = inode->i_sb;
 955
 956         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 957         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 958                                 (unsigned long long)zero_start,
 959                                 (unsigned long long)i_size_read(inode));
 960         while (zero_start < zero_to_size) {
 961                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 962                                                   zero_to_size,
 963                                                   &range_start,
 964                                                   &range_end);
 965                 if (ret) {
 966                         mlog_errno(ret);
 967                         break;
 968                 }
 969                 if (!range_end)
 970                         break;
 971                 /* Trim the ends */
 972                 if (range_start < zero_start)
 973                         range_start = zero_start;
 974                 if (range_end > zero_to_size)
 975                         range_end = zero_to_size;
 976
 977                 ret = ocfs2_zero_extend_range(inode, range_start,
 978                                               range_end);
 979                 if (ret) {
 980                         mlog_errno(ret);
 981                         break;
 982                 }
 983                 zero_start = range_end;
 984         }
 985
 986         return ret;
 987 }
 988
 989 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
 990                           u64 new_i_size, u64 zero_to)
 991 {
 992         int ret;
 993         u32 clusters_to_add;
 994         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 995
 996         /*
 997          * Only quota files call this without a bh, and they can't be
 998          * refcounted.
 999          */
1000         BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1001         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1002
1003         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1004         if (clusters_to_add < oi->ip_clusters)
1005                 clusters_to_add = 0;
1006         else
1007                 clusters_to_add -= oi->ip_clusters;
1008
1009         if (clusters_to_add) {
1010                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1011                                                 clusters_to_add, 0);
1012                 if (ret) {
1013                         mlog_errno(ret);
1014                         goto out;
1015                 }
1016         }
1017
1018         /*
1019          * Call this even if we don't add any clusters to the tree. We
1020          * still need to zero the area between the old i_size and the
1021          * new i_size.
1022          */
1023         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1024         if (ret < 0)
1025                 mlog_errno(ret);
1026
1027 out:
1028         return ret;
1029 }
1030
1031 static int ocfs2_extend_file(struct inode *inode,
1032                              struct buffer_head *di_bh,
1033                              u64 new_i_size)
1034 {
1035         int ret = 0;
1036         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1037
1038         BUG_ON(!di_bh);
1039
1040         /* setattr sometimes calls us like this. */
1041         if (new_i_size == 0)
1042                 goto out;
1043
1044         if (i_size_read(inode) == new_i_size)
1045                 goto out;
1046         BUG_ON(new_i_size < i_size_read(inode));
1047
1048         /*
1049          * The alloc sem blocks people in read/write from reading our
1050          * allocation until we're done changing it. We depend on
1051          * i_mutex to block other extend/truncate calls while we're
1052          * here.  We even have to hold it for sparse files because there
1053          * might be some tail zeroing.
1054          */
1055         down_write(&oi->ip_alloc_sem);
1056
1057         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1058                 /*
1059                  * We can optimize small extends by keeping the inodes
1060                  * inline data.
1061                  */
1062                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1063                         up_write(&oi->ip_alloc_sem);
1064                         goto out_update_size;
1065                 }
1066
1067                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1068                 if (ret) {
1069                         up_write(&oi->ip_alloc_sem);
1070                         mlog_errno(ret);
1071                         goto out;
1072                 }
1073         }
1074
1075         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1076                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1077         else
1078                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1079                                             new_i_size);
1080
1081         up_write(&oi->ip_alloc_sem);
1082
1083         if (ret < 0) {
1084                 mlog_errno(ret);
1085                 goto out;
1086         }
1087
1088 out_update_size:
1089         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1090         if (ret < 0)
1091                 mlog_errno(ret);
1092
1093 out:
1094         return ret;
1095 }
1096
1097 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1098 {
1099         int status = 0, size_change;
1100         struct inode *inode = dentry->d_inode;
1101         struct super_block *sb = inode->i_sb;
1102         struct ocfs2_super *osb = OCFS2_SB(sb);
1103         struct buffer_head *bh = NULL;
1104         handle_t *handle = NULL;
1105         struct dquot *transfer_to[MAXQUOTAS] = { };
1106         int qtype;
1107
1108         trace_ocfs2_setattr(inode, dentry,
1109                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1110                             dentry->d_name.len, dentry->d_name.name,
1111                             attr->ia_valid, attr->ia_mode,
1112                             from_kuid(&init_user_ns, attr->ia_uid),
1113                             from_kgid(&init_user_ns, attr->ia_gid));
1114
1115         /* ensuring we don't even attempt to truncate a symlink */
1116         if (S_ISLNK(inode->i_mode))
1117                 attr->ia_valid &= ~ATTR_SIZE;
1118
1119 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1120                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1121         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1122                 return 0;
1123
1124         status = inode_change_ok(inode, attr);
1125         if (status)
1126                 return status;
1127
1128         if (is_quota_modification(inode, attr))
1129                 dquot_initialize(inode);
1130         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1131         if (size_change) {
1132                 status = ocfs2_rw_lock(inode, 1);
1133                 if (status < 0) {
1134                         mlog_errno(status);
1135                         goto bail;
1136                 }
1137         }
1138
1139         status = ocfs2_inode_lock(inode, &bh, 1);
1140         if (status < 0) {
1141                 if (status != -ENOENT)
1142                         mlog_errno(status);
1143                 goto bail_unlock_rw;
1144         }
1145
1146         if (size_change) {
1147                 status = inode_newsize_ok(inode, attr->ia_size);
1148                 if (status)
1149                         goto bail_unlock;
1150
1151                 inode_dio_wait(inode);
1152
1153                 if (i_size_read(inode) >= attr->ia_size) {
1154                         if (ocfs2_should_order_data(inode)) {
1155                                 status = ocfs2_begin_ordered_truncate(inode,
1156                                                                       attr->ia_size);
1157                                 if (status)
1158                                         goto bail_unlock;
1159                         }
1160                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1161                 } else
1162                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1163                 if (status < 0) {
1164                         if (status != -ENOSPC)
1165                                 mlog_errno(status);
1166                         status = -ENOSPC;
1167                         goto bail_unlock;
1168                 }
1169         }
1170
1171         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1172             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1173                 /*
1174                  * Gather pointers to quota structures so that allocation /
1175                  * freeing of quota structures happens here and not inside
1176                  * dquot_transfer() where we have problems with lock ordering
1177                  */
1178                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1179                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1180                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1181                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1182                         if (!transfer_to[USRQUOTA]) {
1183                                 status = -ESRCH;
1184                                 goto bail_unlock;
1185                         }
1186                 }
1187                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1188                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1189                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1190                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1191                         if (!transfer_to[GRPQUOTA]) {
1192                                 status = -ESRCH;
1193                                 goto bail_unlock;
1194                         }
1195                 }
1196                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1197                                            2 * ocfs2_quota_trans_credits(sb));
1198                 if (IS_ERR(handle)) {
1199                         status = PTR_ERR(handle);
1200                         mlog_errno(status);
1201                         goto bail_unlock;
1202                 }
1203                 status = __dquot_transfer(inode, transfer_to);
1204                 if (status < 0)
1205                         goto bail_commit;
1206         } else {
1207                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1208                 if (IS_ERR(handle)) {
1209                         status = PTR_ERR(handle);
1210                         mlog_errno(status);
1211                         goto bail_unlock;
1212                 }
1213         }
1214
1215         setattr_copy(inode, attr);
1216         mark_inode_dirty(inode);
1217
1218         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1219         if (status < 0)
1220                 mlog_errno(status);
1221
1222 bail_commit:
1223         ocfs2_commit_trans(osb, handle);
1224 bail_unlock:
1225         ocfs2_inode_unlock(inode, 1);
1226 bail_unlock_rw:
1227         if (size_change)
1228                 ocfs2_rw_unlock(inode, 1);
1229 bail:
1230         brelse(bh);
1231
1232         /* Release quota pointers in case we acquired them */
1233         for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1234                 dqput(transfer_to[qtype]);
1235
1236         if (!status && attr->ia_valid & ATTR_MODE) {
1237                 status = posix_acl_chmod(inode, inode->i_mode);
1238                 if (status < 0)
1239                         mlog_errno(status);
1240         }
1241
1242         return status;
1243 }
1244
1245 int ocfs2_getattr(struct vfsmount *mnt,
1246                   struct dentry *dentry,
1247                   struct kstat *stat)
1248 {
1249         struct inode *inode = dentry->d_inode;
1250         struct super_block *sb = dentry->d_inode->i_sb;
1251         struct ocfs2_super *osb = sb->s_fs_info;
1252         int err;
1253
1254         err = ocfs2_inode_revalidate(dentry);
1255         if (err) {
1256                 if (err != -ENOENT)
1257                         mlog_errno(err);
1258                 goto bail;
1259         }
1260
1261         generic_fillattr(inode, stat);
1262
1263         /* We set the blksize from the cluster size for performance */
1264         stat->blksize = osb->s_clustersize;
1265
1266 bail:
1267         return err;
1268 }
1269
1270 int ocfs2_permission(struct inode *inode, int mask)
1271 {
1272         int ret;
1273
1274         if (mask & MAY_NOT_BLOCK)
1275                 return -ECHILD;
1276
1277         ret = ocfs2_inode_lock(inode, NULL, 0);
1278         if (ret) {
1279                 if (ret != -ENOENT)
1280                         mlog_errno(ret);
1281                 goto out;
1282         }
1283
1284         ret = generic_permission(inode, mask);
1285
1286         ocfs2_inode_unlock(inode, 0);
1287 out:
1288         return ret;
1289 }
1290
1291 static int __ocfs2_write_remove_suid(struct inode *inode,
1292                                      struct buffer_head *bh)
1293 {
1294         int ret;
1295         handle_t *handle;
1296         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1297         struct ocfs2_dinode *di;
1298
1299         trace_ocfs2_write_remove_suid(
1300                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1301                         inode->i_mode);
1302
1303         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1304         if (IS_ERR(handle)) {
1305                 ret = PTR_ERR(handle);
1306                 mlog_errno(ret);
1307                 goto out;
1308         }
1309
1310         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1311                                       OCFS2_JOURNAL_ACCESS_WRITE);
1312         if (ret < 0) {
1313                 mlog_errno(ret);
1314                 goto out_trans;
1315         }
1316
1317         inode->i_mode &= ~S_ISUID;
1318         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1319                 inode->i_mode &= ~S_ISGID;
1320
1321         di = (struct ocfs2_dinode *) bh->b_data;
1322         di->i_mode = cpu_to_le16(inode->i_mode);
1323
1324         ocfs2_journal_dirty(handle, bh);
1325
1326 out_trans:
1327         ocfs2_commit_trans(osb, handle);
1328 out:
1329         return ret;
1330 }
1331
1332 /*
1333  * Will look for holes and unwritten extents in the range starting at
1334  * pos for count bytes (inclusive).
1335  */
1336 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1337                                        size_t count)
1338 {
1339         int ret = 0;
1340         unsigned int extent_flags;
1341         u32 cpos, clusters, extent_len, phys_cpos;
1342         struct super_block *sb = inode->i_sb;
1343
1344         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1345         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1346
1347         while (clusters) {
1348                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1349                                          &extent_flags);
1350                 if (ret < 0) {
1351                         mlog_errno(ret);
1352                         goto out;
1353                 }
1354
1355                 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1356                         ret = 1;
1357                         break;
1358                 }
1359
1360                 if (extent_len > clusters)
1361                         extent_len = clusters;
1362
1363                 clusters -= extent_len;
1364                 cpos += extent_len;
1365         }
1366 out:
1367         return ret;
1368 }
1369
1370 static int ocfs2_write_remove_suid(struct inode *inode)
1371 {
1372         int ret;
1373         struct buffer_head *bh = NULL;
1374
1375         ret = ocfs2_read_inode_block(inode, &bh);
1376         if (ret < 0) {
1377                 mlog_errno(ret);
1378                 goto out;
1379         }
1380
1381         ret =  __ocfs2_write_remove_suid(inode, bh);
1382 out:
1383         brelse(bh);
1384         return ret;
1385 }
1386
1387 /*
1388  * Allocate enough extents to cover the region starting at byte offset
1389  * start for len bytes. Existing extents are skipped, any extents
1390  * added are marked as "unwritten".
1391  */
1392 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1393                                             u64 start, u64 len)
1394 {
1395         int ret;
1396         u32 cpos, phys_cpos, clusters, alloc_size;
1397         u64 end = start + len;
1398         struct buffer_head *di_bh = NULL;
1399
1400         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1401                 ret = ocfs2_read_inode_block(inode, &di_bh);
1402                 if (ret) {
1403                         mlog_errno(ret);
1404                         goto out;
1405                 }
1406
1407                 /*
1408                  * Nothing to do if the requested reservation range
1409                  * fits within the inode.
1410                  */
1411                 if (ocfs2_size_fits_inline_data(di_bh, end))
1412                         goto out;
1413
1414                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1415                 if (ret) {
1416                         mlog_errno(ret);
1417                         goto out;
1418                 }
1419         }
1420
1421         /*
1422          * We consider both start and len to be inclusive.
1423          */
1424         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1425         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1426         clusters -= cpos;
1427
1428         while (clusters) {
1429                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1430                                          &alloc_size, NULL);
1431                 if (ret) {
1432                         mlog_errno(ret);
1433                         goto out;
1434                 }
1435
1436                 /*
1437                  * Hole or existing extent len can be arbitrary, so
1438                  * cap it to our own allocation request.
1439                  */
1440                 if (alloc_size > clusters)
1441                         alloc_size = clusters;
1442
1443                 if (phys_cpos) {
1444                         /*
1445                          * We already have an allocation at this
1446                          * region so we can safely skip it.
1447                          */
1448                         goto next;
1449                 }
1450
1451                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1452                 if (ret) {
1453                         if (ret != -ENOSPC)
1454                                 mlog_errno(ret);
1455                         goto out;
1456                 }
1457
1458 next:
1459                 cpos += alloc_size;
1460                 clusters -= alloc_size;
1461         }
1462
1463         ret = 0;
1464 out:
1465
1466         brelse(di_bh);
1467         return ret;
1468 }
1469
1470 /*
1471  * Truncate a byte range, avoiding pages within partial clusters. This
1472  * preserves those pages for the zeroing code to write to.
1473  */
1474 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1475                                          u64 byte_len)
1476 {
1477         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1478         loff_t start, end;
1479         struct address_space *mapping = inode->i_mapping;
1480
1481         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1482         end = byte_start + byte_len;
1483         end = end & ~(osb->s_clustersize - 1);
1484
1485         if (start < end) {
1486                 unmap_mapping_range(mapping, start, end - start, 0);
1487                 truncate_inode_pages_range(mapping, start, end - 1);
1488         }
1489 }
1490
1491 static int ocfs2_zero_partial_clusters(struct inode *inode,
1492                                        u64 start, u64 len)
1493 {
1494         int ret = 0;
1495         u64 tmpend, end = start + len;
1496         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1497         unsigned int csize = osb->s_clustersize;
1498         handle_t *handle;
1499
1500         /*
1501          * The "start" and "end" values are NOT necessarily part of
1502          * the range whose allocation is being deleted. Rather, this
1503          * is what the user passed in with the request. We must zero
1504          * partial clusters here. There's no need to worry about
1505          * physical allocation - the zeroing code knows to skip holes.
1506          */
1507         trace_ocfs2_zero_partial_clusters(
1508                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1509                 (unsigned long long)start, (unsigned long long)end);
1510
1511         /*
1512          * If both edges are on a cluster boundary then there's no
1513          * zeroing required as the region is part of the allocation to
1514          * be truncated.
1515          */
1516         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1517                 goto out;
1518
1519         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1520         if (IS_ERR(handle)) {
1521                 ret = PTR_ERR(handle);
1522                 mlog_errno(ret);
1523                 goto out;
1524         }
1525
1526         /*
1527          * We want to get the byte offset of the end of the 1st cluster.
1528          */
1529         tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1530         if (tmpend > end)
1531                 tmpend = end;
1532
1533         trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1534                                                  (unsigned long long)tmpend);
1535
1536         ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1537         if (ret)
1538                 mlog_errno(ret);
1539
1540         if (tmpend < end) {
1541                 /*
1542                  * This may make start and end equal, but the zeroing
1543                  * code will skip any work in that case so there's no
1544                  * need to catch it up here.
1545                  */
1546                 start = end & ~(osb->s_clustersize - 1);
1547
1548                 trace_ocfs2_zero_partial_clusters_range2(
1549                         (unsigned long long)start, (unsigned long long)end);
1550
1551                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1552                 if (ret)
1553                         mlog_errno(ret);
1554         }
1555
1556         ocfs2_commit_trans(osb, handle);
1557 out:
1558         return ret;
1559 }
1560
1561 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1562 {
1563         int i;
1564         struct ocfs2_extent_rec *rec = NULL;
1565
1566         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1567
1568                 rec = &el->l_recs[i];
1569
1570                 if (le32_to_cpu(rec->e_cpos) < pos)
1571                         break;
1572         }
1573
1574         return i;
1575 }
1576
1577 /*
1578  * Helper to calculate the punching pos and length in one run, we handle the
1579  * following three cases in order:
1580  *
1581  * - remove the entire record
1582  * - remove a partial record
1583  * - no record needs to be removed (hole-punching completed)
1584 */
1585 static void ocfs2_calc_trunc_pos(struct inode *inode,
1586                                  struct ocfs2_extent_list *el,
1587                                  struct ocfs2_extent_rec *rec,
1588                                  u32 trunc_start, u32 *trunc_cpos,
1589                                  u32 *trunc_len, u32 *trunc_end,
1590                                  u64 *blkno, int *done)
1591 {
1592         int ret = 0;
1593         u32 coff, range;
1594
1595         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1596
1597         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1598                 /*
1599                  * remove an entire extent record.
1600                  */
1601                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1602                 /*
1603                  * Skip holes if any.
1604                  */
1605                 if (range < *trunc_end)
1606                         *trunc_end = range;
1607                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1608                 *blkno = le64_to_cpu(rec->e_blkno);
1609                 *trunc_end = le32_to_cpu(rec->e_cpos);
1610         } else if (range > trunc_start) {
1611                 /*
1612                  * remove a partial extent record, which means we're
1613                  * removing the last extent record.
1614                  */
1615                 *trunc_cpos = trunc_start;
1616                 /*
1617                  * skip hole if any.
1618                  */
1619                 if (range < *trunc_end)
1620                         *trunc_end = range;
1621                 *trunc_len = *trunc_end - trunc_start;
1622                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1623                 *blkno = le64_to_cpu(rec->e_blkno) +
1624                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1625                 *trunc_end = trunc_start;
1626         } else {
1627                 /*
1628                  * It may have two following possibilities:
1629                  *
1630                  * - last record has been removed
1631                  * - trunc_start was within a hole
1632                  *
1633                  * both two cases mean the completion of hole punching.
1634                  */
1635                 ret = 1;
1636         }
1637
1638         *done = ret;
1639 }
1640
1641 static int ocfs2_remove_inode_range(struct inode *inode,
1642                                     struct buffer_head *di_bh, u64 byte_start,
1643                                     u64 byte_len)
1644 {
1645         int ret = 0, flags = 0, done = 0, i;
1646         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1647         u32 cluster_in_el;
1648         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1649         struct ocfs2_cached_dealloc_ctxt dealloc;
1650         struct address_space *mapping = inode->i_mapping;
1651         struct ocfs2_extent_tree et;
1652         struct ocfs2_path *path = NULL;
1653         struct ocfs2_extent_list *el = NULL;
1654         struct ocfs2_extent_rec *rec = NULL;
1655         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1656         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1657
1658         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1659         ocfs2_init_dealloc_ctxt(&dealloc);
1660
1661         trace_ocfs2_remove_inode_range(
1662                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1663                         (unsigned long long)byte_start,
1664                         (unsigned long long)byte_len);
1665
1666         if (byte_len == 0)
1667                 return 0;
1668
1669         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1670                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1671                                             byte_start + byte_len, 0);
1672                 if (ret) {
1673                         mlog_errno(ret);
1674                         goto out;
1675                 }
1676                 /*
1677                  * There's no need to get fancy with the page cache
1678                  * truncate of an inline-data inode. We're talking
1679                  * about less than a page here, which will be cached
1680                  * in the dinode buffer anyway.
1681                  */
1682                 unmap_mapping_range(mapping, 0, 0, 0);
1683                 truncate_inode_pages(mapping, 0);
1684                 goto out;
1685         }
1686
1687         /*
1688          * For reflinks, we may need to CoW 2 clusters which might be
1689          * partially zero'd later, if hole's start and end offset were
1690          * within one cluster(means is not exactly aligned to clustersize).
1691          */
1692
1693         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1694
1695                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1696                 if (ret) {
1697                         mlog_errno(ret);
1698                         goto out;
1699                 }
1700
1701                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1702                 if (ret) {
1703                         mlog_errno(ret);
1704                         goto out;
1705                 }
1706         }
1707
1708         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1709         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1710         cluster_in_el = trunc_end;
1711
1712         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1713         if (ret) {
1714                 mlog_errno(ret);
1715                 goto out;
1716         }
1717
1718         path = ocfs2_new_path_from_et(&et);
1719         if (!path) {
1720                 ret = -ENOMEM;
1721                 mlog_errno(ret);
1722                 goto out;
1723         }
1724
1725         while (trunc_end > trunc_start) {
1726
1727                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1728                                       cluster_in_el);
1729                 if (ret) {
1730                         mlog_errno(ret);
1731                         goto out;
1732                 }
1733
1734                 el = path_leaf_el(path);
1735
1736                 i = ocfs2_find_rec(el, trunc_end);
1737                 /*
1738                  * Need to go to previous extent block.
1739                  */
1740                 if (i < 0) {
1741                         if (path->p_tree_depth == 0)
1742                                 break;
1743
1744                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1745                                                             path,
1746                                                             &cluster_in_el);
1747                         if (ret) {
1748                                 mlog_errno(ret);
1749                                 goto out;
1750                         }
1751
1752                         /*
1753                          * We've reached the leftmost extent block,
1754                          * it's safe to leave.
1755                          */
1756                         if (cluster_in_el == 0)
1757                                 break;
1758
1759                         /*
1760                          * The 'pos' searched for previous extent block is
1761                          * always one cluster less than actual trunc_end.
1762                          */
1763                         trunc_end = cluster_in_el + 1;
1764
1765                         ocfs2_reinit_path(path, 1);
1766
1767                         continue;
1768
1769                 } else
1770                         rec = &el->l_recs[i];
1771
1772                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1773                                      &trunc_len, &trunc_end, &blkno, &done);
1774                 if (done)
1775                         break;
1776
1777                 flags = rec->e_flags;
1778                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1779
1780                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1781                                                phys_cpos, trunc_len, flags,
1782                                                &dealloc, refcount_loc);
1783                 if (ret < 0) {
1784                         mlog_errno(ret);
1785                         goto out;
1786                 }
1787
1788                 cluster_in_el = trunc_end;
1789
1790                 ocfs2_reinit_path(path, 1);
1791         }
1792
1793         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1794
1795 out:
1796         ocfs2_free_path(path);
1797         ocfs2_schedule_truncate_log_flush(osb, 1);
1798         ocfs2_run_deallocs(osb, &dealloc);
1799
1800         return ret;
1801 }
1802
1803 /*
1804  * Parts of this function taken from xfs_change_file_space()
1805  */
1806 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1807                                      loff_t f_pos, unsigned int cmd,
1808                                      struct ocfs2_space_resv *sr,
1809                                      int change_size)
1810 {
1811         int ret;
1812         s64 llen;
1813         loff_t size;
1814         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1815         struct buffer_head *di_bh = NULL;
1816         handle_t *handle;
1817         unsigned long long max_off = inode->i_sb->s_maxbytes;
1818
1819         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1820                 return -EROFS;
1821
1822         mutex_lock(&inode->i_mutex);
1823
1824         /*
1825          * This prevents concurrent writes on other nodes
1826          */
1827         ret = ocfs2_rw_lock(inode, 1);
1828         if (ret) {
1829                 mlog_errno(ret);
1830                 goto out;
1831         }
1832
1833         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1834         if (ret) {
1835                 mlog_errno(ret);
1836                 goto out_rw_unlock;
1837         }
1838
1839         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1840                 ret = -EPERM;
1841                 goto out_inode_unlock;
1842         }
1843
1844         switch (sr->l_whence) {
1845         case 0: /*SEEK_SET*/
1846                 break;
1847         case 1: /*SEEK_CUR*/
1848                 sr->l_start += f_pos;
1849                 break;
1850         case 2: /*SEEK_END*/
1851                 sr->l_start += i_size_read(inode);
1852                 break;
1853         default:
1854                 ret = -EINVAL;
1855                 goto out_inode_unlock;
1856         }
1857         sr->l_whence = 0;
1858
1859         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1860
1861         if (sr->l_start < 0
1862             || sr->l_start > max_off
1863             || (sr->l_start + llen) < 0
1864             || (sr->l_start + llen) > max_off) {
1865                 ret = -EINVAL;
1866                 goto out_inode_unlock;
1867         }
1868         size = sr->l_start + sr->l_len;
1869
1870         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1871             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1872                 if (sr->l_len <= 0) {
1873                         ret = -EINVAL;
1874                         goto out_inode_unlock;
1875                 }
1876         }
1877
1878         if (file && should_remove_suid(file->f_path.dentry)) {
1879                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1880                 if (ret) {
1881                         mlog_errno(ret);
1882                         goto out_inode_unlock;
1883                 }
1884         }
1885
1886         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1887         switch (cmd) {
1888         case OCFS2_IOC_RESVSP:
1889         case OCFS2_IOC_RESVSP64:
1890                 /*
1891                  * This takes unsigned offsets, but the signed ones we
1892                  * pass have been checked against overflow above.
1893                  */
1894                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1895                                                        sr->l_len);
1896                 break;
1897         case OCFS2_IOC_UNRESVSP:
1898         case OCFS2_IOC_UNRESVSP64:
1899                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1900                                                sr->l_len);
1901                 break;
1902         default:
1903                 ret = -EINVAL;
1904         }
1905         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1906         if (ret) {
1907                 mlog_errno(ret);
1908                 goto out_inode_unlock;
1909         }
1910
1911         /*
1912          * We update c/mtime for these changes
1913          */
1914         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1915         if (IS_ERR(handle)) {
1916                 ret = PTR_ERR(handle);
1917                 mlog_errno(ret);
1918                 goto out_inode_unlock;
1919         }
1920
1921         if (change_size && i_size_read(inode) < size)
1922                 i_size_write(inode, size);
1923
1924         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1925         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1926         if (ret < 0)
1927                 mlog_errno(ret);
1928
1929         if (file && (file->f_flags & O_SYNC))
1930                 handle->h_sync = 1;
1931
1932         ocfs2_commit_trans(osb, handle);
1933
1934 out_inode_unlock:
1935         brelse(di_bh);
1936         ocfs2_inode_unlock(inode, 1);
1937 out_rw_unlock:
1938         ocfs2_rw_unlock(inode, 1);
1939
1940 out:
1941         mutex_unlock(&inode->i_mutex);
1942         return ret;
1943 }
1944
1945 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1946                             struct ocfs2_space_resv *sr)
1947 {
1948         struct inode *inode = file_inode(file);
1949         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1950         int ret;
1951
1952         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1953             !ocfs2_writes_unwritten_extents(osb))
1954                 return -ENOTTY;
1955         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1956                  !ocfs2_sparse_alloc(osb))
1957                 return -ENOTTY;
1958
1959         if (!S_ISREG(inode->i_mode))
1960                 return -EINVAL;
1961
1962         if (!(file->f_mode & FMODE_WRITE))
1963                 return -EBADF;
1964
1965         ret = mnt_want_write_file(file);
1966         if (ret)
1967                 return ret;
1968         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1969         mnt_drop_write_file(file);
1970         return ret;
1971 }
1972
1973 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1974                             loff_t len)
1975 {
1976         struct inode *inode = file_inode(file);
1977         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1978         struct ocfs2_space_resv sr;
1979         int change_size = 1;
1980         int cmd = OCFS2_IOC_RESVSP64;
1981
1982         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1983                 return -EOPNOTSUPP;
1984         if (!ocfs2_writes_unwritten_extents(osb))
1985                 return -EOPNOTSUPP;
1986
1987         if (mode & FALLOC_FL_KEEP_SIZE)
1988                 change_size = 0;
1989
1990         if (mode & FALLOC_FL_PUNCH_HOLE)
1991                 cmd = OCFS2_IOC_UNRESVSP64;
1992
1993         sr.l_whence = 0;
1994         sr.l_start = (s64)offset;
1995         sr.l_len = (s64)len;
1996
1997         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
1998                                          change_size);
1999 }
2000
2001 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2002                                    size_t count)
2003 {
2004         int ret = 0;
2005         unsigned int extent_flags;
2006         u32 cpos, clusters, extent_len, phys_cpos;
2007         struct super_block *sb = inode->i_sb;
2008
2009         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2010             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2011             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2012                 return 0;
2013
2014         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2015         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2016
2017         while (clusters) {
2018                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2019                                          &extent_flags);
2020                 if (ret < 0) {
2021                         mlog_errno(ret);
2022                         goto out;
2023                 }
2024
2025                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2026                         ret = 1;
2027                         break;
2028                 }
2029
2030                 if (extent_len > clusters)
2031                         extent_len = clusters;
2032
2033                 clusters -= extent_len;
2034                 cpos += extent_len;
2035         }
2036 out:
2037         return ret;
2038 }
2039
2040 static void ocfs2_aiodio_wait(struct inode *inode)
2041 {
2042         wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
2043
2044         wait_event(*wq, (atomic_read(&OCFS2_I(inode)->ip_unaligned_aio) == 0));
2045 }
2046
2047 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2048 {
2049         int blockmask = inode->i_sb->s_blocksize - 1;
2050         loff_t final_size = pos + count;
2051
2052         if ((pos & blockmask) || (final_size & blockmask))
2053                 return 1;
2054         return 0;
2055 }
2056
2057 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2058                                             struct file *file,
2059                                             loff_t pos, size_t count,
2060                                             int *meta_level)
2061 {
2062         int ret;
2063         struct buffer_head *di_bh = NULL;
2064         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2065         u32 clusters =
2066                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2067
2068         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2069         if (ret) {
2070                 mlog_errno(ret);
2071                 goto out;
2072         }
2073
2074         *meta_level = 1;
2075
2076         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2077         if (ret)
2078                 mlog_errno(ret);
2079 out:
2080         brelse(di_bh);
2081         return ret;
2082 }
2083
2084 static int ocfs2_prepare_inode_for_write(struct file *file,
2085                                          loff_t *ppos,
2086                                          size_t count,
2087                                          int appending,
2088                                          int *direct_io,
2089                                          int *has_refcount)
2090 {
2091         int ret = 0, meta_level = 0;
2092         struct dentry *dentry = file->f_path.dentry;
2093         struct inode *inode = dentry->d_inode;
2094         loff_t saved_pos = 0, end;
2095
2096         /*
2097          * We start with a read level meta lock and only jump to an ex
2098          * if we need to make modifications here.
2099          */
2100         for(;;) {
2101                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2102                 if (ret < 0) {
2103                         meta_level = -1;
2104                         mlog_errno(ret);
2105                         goto out;
2106                 }
2107
2108                 /* Clear suid / sgid if necessary. We do this here
2109                  * instead of later in the write path because
2110                  * remove_suid() calls ->setattr without any hint that
2111                  * we may have already done our cluster locking. Since
2112                  * ocfs2_setattr() *must* take cluster locks to
2113                  * proceed, this will lead us to recursively lock the
2114                  * inode. There's also the dinode i_size state which
2115                  * can be lost via setattr during extending writes (we
2116                  * set inode->i_size at the end of a write. */
2117                 if (should_remove_suid(dentry)) {
2118                         if (meta_level == 0) {
2119                                 ocfs2_inode_unlock(inode, meta_level);
2120                                 meta_level = 1;
2121                                 continue;
2122                         }
2123
2124                         ret = ocfs2_write_remove_suid(inode);
2125                         if (ret < 0) {
2126                                 mlog_errno(ret);
2127                                 goto out_unlock;
2128                         }
2129                 }
2130
2131                 /* work on a copy of ppos until we're sure that we won't have
2132                  * to recalculate it due to relocking. */
2133                 if (appending)
2134                         saved_pos = i_size_read(inode);
2135                 else
2136                         saved_pos = *ppos;
2137
2138                 end = saved_pos + count;
2139
2140                 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2141                 if (ret == 1) {
2142                         ocfs2_inode_unlock(inode, meta_level);
2143                         meta_level = -1;
2144
2145                         ret = ocfs2_prepare_inode_for_refcount(inode,
2146                                                                file,
2147                                                                saved_pos,
2148                                                                count,
2149                                                                &meta_level);
2150                         if (has_refcount)
2151                                 *has_refcount = 1;
2152                         if (direct_io)
2153                                 *direct_io = 0;
2154                 }
2155
2156                 if (ret < 0) {
2157                         mlog_errno(ret);
2158                         goto out_unlock;
2159                 }
2160
2161                 /*
2162                  * Skip the O_DIRECT checks if we don't need
2163                  * them.
2164                  */
2165                 if (!direct_io || !(*direct_io))
2166                         break;
2167
2168                 /*
2169                  * There's no sane way to do direct writes to an inode
2170                  * with inline data.
2171                  */
2172                 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2173                         *direct_io = 0;
2174                         break;
2175                 }
2176
2177                 /*
2178                  * Allowing concurrent direct writes means
2179                  * i_size changes wouldn't be synchronized, so
2180                  * one node could wind up truncating another
2181                  * nodes writes.
2182                  */
2183                 if (end > i_size_read(inode)) {
2184                         *direct_io = 0;
2185                         break;
2186                 }
2187
2188                 /*
2189                  * We don't fill holes during direct io, so
2190                  * check for them here. If any are found, the
2191                  * caller will have to retake some cluster
2192                  * locks and initiate the io as buffered.
2193                  */
2194                 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2195                 if (ret == 1) {
2196                         *direct_io = 0;
2197                         ret = 0;
2198                 } else if (ret < 0)
2199                         mlog_errno(ret);
2200                 break;
2201         }
2202
2203         if (appending)
2204                 *ppos = saved_pos;
2205
2206 out_unlock:
2207         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2208                                             saved_pos, appending, count,
2209                                             direct_io, has_refcount);
2210
2211         if (meta_level >= 0)
2212                 ocfs2_inode_unlock(inode, meta_level);
2213
2214 out:
2215         return ret;
2216 }
2217
2218 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2219                                     const struct iovec *iov,
2220                                     unsigned long nr_segs,
2221                                     loff_t pos)
2222 {
2223         int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
2224         int can_do_direct, has_refcount = 0;
2225         ssize_t written = 0;
2226         size_t ocount;          /* original count */
2227         size_t count;           /* after file limit checks */
2228         loff_t old_size, *ppos = &iocb->ki_pos;
2229         u32 old_clusters;
2230         struct file *file = iocb->ki_filp;
2231         struct inode *inode = file_inode(file);
2232         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2233         int full_coherency = !(osb->s_mount_opt &
2234                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2235         int unaligned_dio = 0;
2236
2237         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2238                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2239                 file->f_path.dentry->d_name.len,
2240                 file->f_path.dentry->d_name.name,
2241                 (unsigned int)nr_segs);
2242
2243         if (iocb->ki_nbytes == 0)
2244                 return 0;
2245
2246         appending = file->f_flags & O_APPEND ? 1 : 0;
2247         direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2248
2249         mutex_lock(&inode->i_mutex);
2250
2251         ocfs2_iocb_clear_sem_locked(iocb);
2252
2253 relock:
2254         /* to match setattr's i_mutex -> rw_lock ordering */
2255         if (direct_io) {
2256                 have_alloc_sem = 1;
2257                 /* communicate with ocfs2_dio_end_io */
2258                 ocfs2_iocb_set_sem_locked(iocb);
2259         }
2260
2261         /*
2262          * Concurrent O_DIRECT writes are allowed with
2263          * mount_option "coherency=buffered".
2264          */
2265         rw_level = (!direct_io || full_coherency);
2266
2267         ret = ocfs2_rw_lock(inode, rw_level);
2268         if (ret < 0) {
2269                 mlog_errno(ret);
2270                 goto out_sems;
2271         }
2272
2273         /*
2274          * O_DIRECT writes with "coherency=full" need to take EX cluster
2275          * inode_lock to guarantee coherency.
2276          */
2277         if (direct_io && full_coherency) {
2278                 /*
2279                  * We need to take and drop the inode lock to force
2280                  * other nodes to drop their caches.  Buffered I/O
2281                  * already does this in write_begin().
2282                  */
2283                 ret = ocfs2_inode_lock(inode, NULL, 1);
2284                 if (ret < 0) {
2285                         mlog_errno(ret);
2286                         goto out;
2287                 }
2288
2289                 ocfs2_inode_unlock(inode, 1);
2290         }
2291
2292         can_do_direct = direct_io;
2293         ret = ocfs2_prepare_inode_for_write(file, ppos,
2294                                             iocb->ki_nbytes, appending,
2295                                             &can_do_direct, &has_refcount);
2296         if (ret < 0) {
2297                 mlog_errno(ret);
2298                 goto out;
2299         }
2300
2301         if (direct_io && !is_sync_kiocb(iocb))
2302                 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_nbytes,
2303                                                       *ppos);
2304
2305         /*
2306          * We can't complete the direct I/O as requested, fall back to
2307          * buffered I/O.
2308          */
2309         if (direct_io && !can_do_direct) {
2310                 ocfs2_rw_unlock(inode, rw_level);
2311
2312                 have_alloc_sem = 0;
2313                 rw_level = -1;
2314
2315                 direct_io = 0;
2316                 goto relock;
2317         }
2318
2319         if (unaligned_dio) {
2320                 /*
2321                  * Wait on previous unaligned aio to complete before
2322                  * proceeding.
2323                  */
2324                 ocfs2_aiodio_wait(inode);
2325
2326                 /* Mark the iocb as needing a decrement in ocfs2_dio_end_io */
2327                 atomic_inc(&OCFS2_I(inode)->ip_unaligned_aio);
2328                 ocfs2_iocb_set_unaligned_aio(iocb);
2329         }
2330
2331         /*
2332          * To later detect whether a journal commit for sync writes is
2333          * necessary, we sample i_size, and cluster count here.
2334          */
2335         old_size = i_size_read(inode);
2336         old_clusters = OCFS2_I(inode)->ip_clusters;
2337
2338         /* communicate with ocfs2_dio_end_io */
2339         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2340
2341         ret = generic_segment_checks(iov, &nr_segs, &ocount,
2342                                      VERIFY_READ);
2343         if (ret)
2344                 goto out_dio;
2345
2346         count = ocount;
2347         ret = generic_write_checks(file, ppos, &count,
2348                                    S_ISBLK(inode->i_mode));
2349         if (ret)
2350                 goto out_dio;
2351
2352         if (direct_io) {
2353                 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2354                                                     ppos, count, ocount);
2355                 if (written < 0) {
2356                         ret = written;
2357                         goto out_dio;
2358                 }
2359         } else {
2360                 current->backing_dev_info = file->f_mapping->backing_dev_info;
2361                 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2362                                                       ppos, count, 0);
2363                 current->backing_dev_info = NULL;
2364         }
2365
2366 out_dio:
2367         /* buffered aio wouldn't have proper lock coverage today */
2368         BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2369
2370         if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2371             ((file->f_flags & O_DIRECT) && !direct_io)) {
2372                 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2373                                                pos + count - 1);
2374                 if (ret < 0)
2375                         written = ret;
2376
2377                 if (!ret && ((old_size != i_size_read(inode)) ||
2378                              (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2379                              has_refcount)) {
2380                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2381                         if (ret < 0)
2382                                 written = ret;
2383                 }
2384
2385                 if (!ret)
2386                         ret = filemap_fdatawait_range(file->f_mapping, pos,
2387                                                       pos + count - 1);
2388         }
2389
2390         /*
2391          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2392          * function pointer which is called when o_direct io completes so that
2393          * it can unlock our rw lock.
2394          * Unfortunately there are error cases which call end_io and others
2395          * that don't.  so we don't have to unlock the rw_lock if either an
2396          * async dio is going to do it in the future or an end_io after an
2397          * error has already done it.
2398          */
2399         if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2400                 rw_level = -1;
2401                 have_alloc_sem = 0;
2402                 unaligned_dio = 0;
2403         }
2404
2405         if (unaligned_dio) {
2406                 ocfs2_iocb_clear_unaligned_aio(iocb);
2407                 atomic_dec(&OCFS2_I(inode)->ip_unaligned_aio);
2408         }
2409
2410 out:
2411         if (rw_level != -1)
2412                 ocfs2_rw_unlock(inode, rw_level);
2413
2414 out_sems:
2415         if (have_alloc_sem)
2416                 ocfs2_iocb_clear_sem_locked(iocb);
2417
2418         mutex_unlock(&inode->i_mutex);
2419
2420         if (written)
2421                 ret = written;
2422         return ret;
2423 }
2424
2425 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2426                                 struct file *out,
2427                                 struct splice_desc *sd)
2428 {
2429         int ret;
2430
2431         ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2432                                             sd->total_len, 0, NULL, NULL);
2433         if (ret < 0) {
2434                 mlog_errno(ret);
2435                 return ret;
2436         }
2437
2438         return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2439 }
2440
2441 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2442                                        struct file *out,
2443                                        loff_t *ppos,
2444                                        size_t len,
2445                                        unsigned int flags)
2446 {
2447         int ret;
2448         struct address_space *mapping = out->f_mapping;
2449         struct inode *inode = mapping->host;
2450         struct splice_desc sd = {
2451                 .total_len = len,
2452                 .flags = flags,
2453                 .pos = *ppos,
2454                 .u.file = out,
2455         };
2456
2457
2458         trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2459                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2460                         out->f_path.dentry->d_name.len,
2461                         out->f_path.dentry->d_name.name, len);
2462
2463         pipe_lock(pipe);
2464
2465         splice_from_pipe_begin(&sd);
2466         do {
2467                 ret = splice_from_pipe_next(pipe, &sd);
2468                 if (ret <= 0)
2469                         break;
2470
2471                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2472                 ret = ocfs2_rw_lock(inode, 1);
2473                 if (ret < 0)
2474                         mlog_errno(ret);
2475                 else {
2476                         ret = ocfs2_splice_to_file(pipe, out, &sd);
2477                         ocfs2_rw_unlock(inode, 1);
2478                 }
2479                 mutex_unlock(&inode->i_mutex);
2480         } while (ret > 0);
2481         splice_from_pipe_end(pipe, &sd);
2482
2483         pipe_unlock(pipe);
2484
2485         if (sd.num_spliced)
2486                 ret = sd.num_spliced;
2487
2488         if (ret > 0) {
2489                 int err;
2490
2491                 err = generic_write_sync(out, *ppos, ret);
2492                 if (err)
2493                         ret = err;
2494                 else
2495                         *ppos += ret;
2496
2497                 balance_dirty_pages_ratelimited(mapping);
2498         }
2499
2500         return ret;
2501 }
2502
2503 static ssize_t ocfs2_file_splice_read(struct file *in,
2504                                       loff_t *ppos,
2505                                       struct pipe_inode_info *pipe,
2506                                       size_t len,
2507                                       unsigned int flags)
2508 {
2509         int ret = 0, lock_level = 0;
2510         struct inode *inode = file_inode(in);
2511
2512         trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2513                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2514                         in->f_path.dentry->d_name.len,
2515                         in->f_path.dentry->d_name.name, len);
2516
2517         /*
2518          * See the comment in ocfs2_file_aio_read()
2519          */
2520         ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2521         if (ret < 0) {
2522                 mlog_errno(ret);
2523                 goto bail;
2524         }
2525         ocfs2_inode_unlock(inode, lock_level);
2526
2527         ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2528
2529 bail:
2530         return ret;
2531 }
2532
2533 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2534                                    const struct iovec *iov,
2535                                    unsigned long nr_segs,
2536                                    loff_t pos)
2537 {
2538         int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2539         struct file *filp = iocb->ki_filp;
2540         struct inode *inode = file_inode(filp);
2541
2542         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2543                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2544                         filp->f_path.dentry->d_name.len,
2545                         filp->f_path.dentry->d_name.name, nr_segs);
2546
2547
2548         if (!inode) {
2549                 ret = -EINVAL;
2550                 mlog_errno(ret);
2551                 goto bail;
2552         }
2553
2554         ocfs2_iocb_clear_sem_locked(iocb);
2555
2556         /*
2557          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2558          * need locks to protect pending reads from racing with truncate.
2559          */
2560         if (filp->f_flags & O_DIRECT) {
2561                 have_alloc_sem = 1;
2562                 ocfs2_iocb_set_sem_locked(iocb);
2563
2564                 ret = ocfs2_rw_lock(inode, 0);
2565                 if (ret < 0) {
2566                         mlog_errno(ret);
2567                         goto bail;
2568                 }
2569                 rw_level = 0;
2570                 /* communicate with ocfs2_dio_end_io */
2571                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2572         }
2573
2574         /*
2575          * We're fine letting folks race truncates and extending
2576          * writes with read across the cluster, just like they can
2577          * locally. Hence no rw_lock during read.
2578          *
2579          * Take and drop the meta data lock to update inode fields
2580          * like i_size. This allows the checks down below
2581          * generic_file_aio_read() a chance of actually working.
2582          */
2583         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2584         if (ret < 0) {
2585                 mlog_errno(ret);
2586                 goto bail;
2587         }
2588         ocfs2_inode_unlock(inode, lock_level);
2589
2590         ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2591         trace_generic_file_aio_read_ret(ret);
2592
2593         /* buffered aio wouldn't have proper lock coverage today */
2594         BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2595
2596         /* see ocfs2_file_aio_write */
2597         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2598                 rw_level = -1;
2599                 have_alloc_sem = 0;
2600         }
2601
2602 bail:
2603         if (have_alloc_sem)
2604                 ocfs2_iocb_clear_sem_locked(iocb);
2605
2606         if (rw_level != -1)
2607                 ocfs2_rw_unlock(inode, rw_level);
2608
2609         return ret;
2610 }
2611
2612 /* Refer generic_file_llseek_unlocked() */
2613 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2614 {
2615         struct inode *inode = file->f_mapping->host;
2616         int ret = 0;
2617
2618         mutex_lock(&inode->i_mutex);
2619
2620         switch (whence) {
2621         case SEEK_SET:
2622                 break;
2623         case SEEK_END:
2624                 offset += inode->i_size;
2625                 break;
2626         case SEEK_CUR:
2627                 if (offset == 0) {
2628                         offset = file->f_pos;
2629                         goto out;
2630                 }
2631                 offset += file->f_pos;
2632                 break;
2633         case SEEK_DATA:
2634         case SEEK_HOLE:
2635                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2636                 if (ret)
2637                         goto out;
2638                 break;
2639         default:
2640                 ret = -EINVAL;
2641                 goto out;
2642         }
2643
2644         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2645
2646 out:
2647         mutex_unlock(&inode->i_mutex);
2648         if (ret)
2649                 return ret;
2650         return offset;
2651 }
2652
2653 const struct inode_operations ocfs2_file_iops = {
2654         .setattr        = ocfs2_setattr,
2655         .getattr        = ocfs2_getattr,
2656         .permission     = ocfs2_permission,
2657         .setxattr       = generic_setxattr,
2658         .getxattr       = generic_getxattr,
2659         .listxattr      = ocfs2_listxattr,
2660         .removexattr    = generic_removexattr,
2661         .fiemap         = ocfs2_fiemap,
2662         .get_acl        = ocfs2_iop_get_acl,
2663         .set_acl        = ocfs2_iop_set_acl,
2664 };
2665
2666 const struct inode_operations ocfs2_special_file_iops = {
2667         .setattr        = ocfs2_setattr,
2668         .getattr        = ocfs2_getattr,
2669         .permission     = ocfs2_permission,
2670         .get_acl        = ocfs2_iop_get_acl,
2671         .set_acl        = ocfs2_iop_set_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 };