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
  40 #define MLOG_MASK_PREFIX ML_INODE
  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
  64 #include "buffer_head_io.h"
  65
  66 static int ocfs2_sync_inode(struct inode *inode)
  67 {
  68         filemap_fdatawrite(inode->i_mapping);
  69         return sync_mapping_buffers(inode->i_mapping);
  70 }
  71
  72 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  73 {
  74         struct ocfs2_file_private *fp;
  75
  76         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  77         if (!fp)
  78                 return -ENOMEM;
  79
  80         fp->fp_file = file;
  81         mutex_init(&fp->fp_mutex);
  82         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  83         file->private_data = fp;
  84
  85         return 0;
  86 }
  87
  88 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  89 {
  90         struct ocfs2_file_private *fp = file->private_data;
  91         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  92
  93         if (fp) {
  94                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  95                 ocfs2_lock_res_free(&fp->fp_flock);
  96                 kfree(fp);
  97                 file->private_data = NULL;
  98         }
  99 }
 100
 101 static int ocfs2_file_open(struct inode *inode, struct file *file)
 102 {
 103         int status;
 104         int mode = file->f_flags;
 105         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 106
 107         mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
 108                    file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
 109
 110         if (file->f_mode & FMODE_WRITE)
 111                 dquot_initialize(inode);
 112
 113         spin_lock(&oi->ip_lock);
 114
 115         /* Check that the inode hasn't been wiped from disk by another
 116          * node. If it hasn't then we're safe as long as we hold the
 117          * spin lock until our increment of open count. */
 118         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
 119                 spin_unlock(&oi->ip_lock);
 120
 121                 status = -ENOENT;
 122                 goto leave;
 123         }
 124
 125         if (mode & O_DIRECT)
 126                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 127
 128         oi->ip_open_count++;
 129         spin_unlock(&oi->ip_lock);
 130
 131         status = ocfs2_init_file_private(inode, file);
 132         if (status) {
 133                 /*
 134                  * We want to set open count back if we're failing the
 135                  * open.
 136                  */
 137                 spin_lock(&oi->ip_lock);
 138                 oi->ip_open_count--;
 139                 spin_unlock(&oi->ip_lock);
 140         }
 141
 142 leave:
 143         mlog_exit(status);
 144         return status;
 145 }
 146
 147 static int ocfs2_file_release(struct inode *inode, struct file *file)
 148 {
 149         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 150
 151         mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
 152                        file->f_path.dentry->d_name.len,
 153                        file->f_path.dentry->d_name.name);
 154
 155         spin_lock(&oi->ip_lock);
 156         if (!--oi->ip_open_count)
 157                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 158         spin_unlock(&oi->ip_lock);
 159
 160         ocfs2_free_file_private(inode, file);
 161
 162         mlog_exit(0);
 163
 164         return 0;
 165 }
 166
 167 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 168 {
 169         return ocfs2_init_file_private(inode, file);
 170 }
 171
 172 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 173 {
 174         ocfs2_free_file_private(inode, file);
 175         return 0;
 176 }
 177
 178 static int ocfs2_sync_file(struct file *file, int datasync)
 179 {
 180         int err = 0;
 181         journal_t *journal;
 182         struct dentry *dentry = file->f_path.dentry;
 183         struct inode *inode = file->f_mapping->host;
 184         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 185
 186         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
 187                    dentry->d_name.len, dentry->d_name.name);
 188
 189         err = ocfs2_sync_inode(dentry->d_inode);
 190         if (err)
 191                 goto bail;
 192
 193         if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
 194                 goto bail;
 195
 196         journal = osb->journal->j_journal;
 197         err = jbd2_journal_force_commit(journal);
 198
 199 bail:
 200         mlog_exit(err);
 201
 202         return (err < 0) ? -EIO : 0;
 203 }
 204
 205 int ocfs2_should_update_atime(struct inode *inode,
 206                               struct vfsmount *vfsmnt)
 207 {
 208         struct timespec now;
 209         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 210
 211         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 212                 return 0;
 213
 214         if ((inode->i_flags & S_NOATIME) ||
 215             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 216                 return 0;
 217
 218         /*
 219          * We can be called with no vfsmnt structure - NFSD will
 220          * sometimes do this.
 221          *
 222          * Note that our action here is different than touch_atime() -
 223          * if we can't tell whether this is a noatime mount, then we
 224          * don't know whether to trust the value of s_atime_quantum.
 225          */
 226         if (vfsmnt == NULL)
 227                 return 0;
 228
 229         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 230             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 231                 return 0;
 232
 233         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 234                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 235                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 236                         return 1;
 237
 238                 return 0;
 239         }
 240
 241         now = CURRENT_TIME;
 242         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 243                 return 0;
 244         else
 245                 return 1;
 246 }
 247
 248 int ocfs2_update_inode_atime(struct inode *inode,
 249                              struct buffer_head *bh)
 250 {
 251         int ret;
 252         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 253         handle_t *handle;
 254         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 255
 256         mlog_entry_void();
 257
 258         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 259         if (IS_ERR(handle)) {
 260                 ret = PTR_ERR(handle);
 261                 mlog_errno(ret);
 262                 goto out;
 263         }
 264
 265         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 266                                       OCFS2_JOURNAL_ACCESS_WRITE);
 267         if (ret) {
 268                 mlog_errno(ret);
 269                 goto out_commit;
 270         }
 271
 272         /*
 273          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 274          * have i_mutex to guard against concurrent changes to other
 275          * inode fields.
 276          */
 277         inode->i_atime = CURRENT_TIME;
 278         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 279         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 280         ocfs2_journal_dirty(handle, bh);
 281
 282 out_commit:
 283         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 284 out:
 285         mlog_exit(ret);
 286         return ret;
 287 }
 288
 289 static int ocfs2_set_inode_size(handle_t *handle,
 290                                 struct inode *inode,
 291                                 struct buffer_head *fe_bh,
 292                                 u64 new_i_size)
 293 {
 294         int status;
 295
 296         mlog_entry_void();
 297         i_size_write(inode, new_i_size);
 298         inode->i_blocks = ocfs2_inode_sector_count(inode);
 299         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 300
 301         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 302         if (status < 0) {
 303                 mlog_errno(status);
 304                 goto bail;
 305         }
 306
 307 bail:
 308         mlog_exit(status);
 309         return status;
 310 }
 311
 312 int ocfs2_simple_size_update(struct inode *inode,
 313                              struct buffer_head *di_bh,
 314                              u64 new_i_size)
 315 {
 316         int ret;
 317         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 318         handle_t *handle = NULL;
 319
 320         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 321         if (IS_ERR(handle)) {
 322                 ret = PTR_ERR(handle);
 323                 mlog_errno(ret);
 324                 goto out;
 325         }
 326
 327         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 328                                    new_i_size);
 329         if (ret < 0)
 330                 mlog_errno(ret);
 331
 332         ocfs2_commit_trans(osb, handle);
 333 out:
 334         return ret;
 335 }
 336
 337 static int ocfs2_cow_file_pos(struct inode *inode,
 338                               struct buffer_head *fe_bh,
 339                               u64 offset)
 340 {
 341         int status;
 342         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 343         unsigned int num_clusters = 0;
 344         unsigned int ext_flags = 0;
 345
 346         /*
 347          * If the new offset is aligned to the range of the cluster, there is
 348          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 349          * CoW either.
 350          */
 351         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 352                 return 0;
 353
 354         status = ocfs2_get_clusters(inode, cpos, &phys,
 355                                     &num_clusters, &ext_flags);
 356         if (status) {
 357                 mlog_errno(status);
 358                 goto out;
 359         }
 360
 361         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 362                 goto out;
 363
 364         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 365
 366 out:
 367         return status;
 368 }
 369
 370 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 371                                      struct inode *inode,
 372                                      struct buffer_head *fe_bh,
 373                                      u64 new_i_size)
 374 {
 375         int status;
 376         handle_t *handle;
 377         struct ocfs2_dinode *di;
 378         u64 cluster_bytes;
 379
 380         mlog_entry_void();
 381
 382         /*
 383          * We need to CoW the cluster contains the offset if it is reflinked
 384          * since we will call ocfs2_zero_range_for_truncate later which will
 385          * write "0" from offset to the end of the cluster.
 386          */
 387         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 388         if (status) {
 389                 mlog_errno(status);
 390                 return status;
 391         }
 392
 393         /* TODO: This needs to actually orphan the inode in this
 394          * transaction. */
 395
 396         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 397         if (IS_ERR(handle)) {
 398                 status = PTR_ERR(handle);
 399                 mlog_errno(status);
 400                 goto out;
 401         }
 402
 403         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 404                                          OCFS2_JOURNAL_ACCESS_WRITE);
 405         if (status < 0) {
 406                 mlog_errno(status);
 407                 goto out_commit;
 408         }
 409
 410         /*
 411          * Do this before setting i_size.
 412          */
 413         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 414         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 415                                                cluster_bytes);
 416         if (status) {
 417                 mlog_errno(status);
 418                 goto out_commit;
 419         }
 420
 421         i_size_write(inode, new_i_size);
 422         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 423
 424         di = (struct ocfs2_dinode *) fe_bh->b_data;
 425         di->i_size = cpu_to_le64(new_i_size);
 426         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 427         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 428
 429         ocfs2_journal_dirty(handle, fe_bh);
 430
 431 out_commit:
 432         ocfs2_commit_trans(osb, handle);
 433 out:
 434
 435         mlog_exit(status);
 436         return status;
 437 }
 438
 439 static int ocfs2_truncate_file(struct inode *inode,
 440                                struct buffer_head *di_bh,
 441                                u64 new_i_size)
 442 {
 443         int status = 0;
 444         struct ocfs2_dinode *fe = NULL;
 445         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 446
 447         mlog_entry("(inode = %llu, new_i_size = %llu\n",
 448                    (unsigned long long)OCFS2_I(inode)->ip_blkno,
 449                    (unsigned long long)new_i_size);
 450
 451         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 452          * already validated it */
 453         fe = (struct ocfs2_dinode *) di_bh->b_data;
 454
 455         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 456                         "Inode %llu, inode i_size = %lld != di "
 457                         "i_size = %llu, i_flags = 0x%x\n",
 458                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 459                         i_size_read(inode),
 460                         (unsigned long long)le64_to_cpu(fe->i_size),
 461                         le32_to_cpu(fe->i_flags));
 462
 463         if (new_i_size > le64_to_cpu(fe->i_size)) {
 464                 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
 465                      (unsigned long long)le64_to_cpu(fe->i_size),
 466                      (unsigned long long)new_i_size);
 467                 status = -EINVAL;
 468                 mlog_errno(status);
 469                 goto bail;
 470         }
 471
 472         mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
 473              (unsigned long long)le64_to_cpu(fe->i_blkno),
 474              (unsigned long long)le64_to_cpu(fe->i_size),
 475              (unsigned long long)new_i_size);
 476
 477         /* lets handle the simple truncate cases before doing any more
 478          * cluster locking. */
 479         if (new_i_size == le64_to_cpu(fe->i_size))
 480                 goto bail;
 481
 482         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 483
 484         ocfs2_resv_discard(&osb->osb_la_resmap,
 485                            &OCFS2_I(inode)->ip_la_data_resv);
 486
 487         /*
 488          * The inode lock forced other nodes to sync and drop their
 489          * pages, which (correctly) happens even if we have a truncate
 490          * without allocation change - ocfs2 cluster sizes can be much
 491          * greater than page size, so we have to truncate them
 492          * anyway.
 493          */
 494         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 495         truncate_inode_pages(inode->i_mapping, new_i_size);
 496
 497         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 498                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 499                                                i_size_read(inode), 1);
 500                 if (status)
 501                         mlog_errno(status);
 502
 503                 goto bail_unlock_sem;
 504         }
 505
 506         /* alright, we're going to need to do a full blown alloc size
 507          * change. Orphan the inode so that recovery can complete the
 508          * truncate if necessary. This does the task of marking
 509          * i_size. */
 510         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 511         if (status < 0) {
 512                 mlog_errno(status);
 513                 goto bail_unlock_sem;
 514         }
 515
 516         status = ocfs2_commit_truncate(osb, inode, di_bh);
 517         if (status < 0) {
 518                 mlog_errno(status);
 519                 goto bail_unlock_sem;
 520         }
 521
 522         /* TODO: orphan dir cleanup here. */
 523 bail_unlock_sem:
 524         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 525
 526 bail:
 527         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 528                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 529
 530         mlog_exit(status);
 531         return status;
 532 }
 533
 534 /*
 535  * extend file allocation only here.
 536  * we'll update all the disk stuff, and oip->alloc_size
 537  *
 538  * expect stuff to be locked, a transaction started and enough data /
 539  * metadata reservations in the contexts.
 540  *
 541  * Will return -EAGAIN, and a reason if a restart is needed.
 542  * If passed in, *reason will always be set, even in error.
 543  */
 544 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 545                          struct inode *inode,
 546                          u32 *logical_offset,
 547                          u32 clusters_to_add,
 548                          int mark_unwritten,
 549                          struct buffer_head *fe_bh,
 550                          handle_t *handle,
 551                          struct ocfs2_alloc_context *data_ac,
 552                          struct ocfs2_alloc_context *meta_ac,
 553                          enum ocfs2_alloc_restarted *reason_ret)
 554 {
 555         int ret;
 556         struct ocfs2_extent_tree et;
 557
 558         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 559         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 560                                           clusters_to_add, mark_unwritten,
 561                                           data_ac, meta_ac, reason_ret);
 562
 563         return ret;
 564 }
 565
 566 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 567                                      u32 clusters_to_add, int mark_unwritten)
 568 {
 569         int status = 0;
 570         int restart_func = 0;
 571         int credits;
 572         u32 prev_clusters;
 573         struct buffer_head *bh = NULL;
 574         struct ocfs2_dinode *fe = NULL;
 575         handle_t *handle = NULL;
 576         struct ocfs2_alloc_context *data_ac = NULL;
 577         struct ocfs2_alloc_context *meta_ac = NULL;
 578         enum ocfs2_alloc_restarted why;
 579         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 580         struct ocfs2_extent_tree et;
 581         int did_quota = 0;
 582
 583         mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
 584
 585         /*
 586          * This function only exists for file systems which don't
 587          * support holes.
 588          */
 589         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 590
 591         status = ocfs2_read_inode_block(inode, &bh);
 592         if (status < 0) {
 593                 mlog_errno(status);
 594                 goto leave;
 595         }
 596         fe = (struct ocfs2_dinode *) bh->b_data;
 597
 598 restart_all:
 599         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 600
 601         mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
 602              "clusters_to_add = %u\n",
 603              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 604              (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
 605              clusters_to_add);
 606         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 607         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 608                                        &data_ac, &meta_ac);
 609         if (status) {
 610                 mlog_errno(status);
 611                 goto leave;
 612         }
 613
 614         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
 615                                             clusters_to_add);
 616         handle = ocfs2_start_trans(osb, credits);
 617         if (IS_ERR(handle)) {
 618                 status = PTR_ERR(handle);
 619                 handle = NULL;
 620                 mlog_errno(status);
 621                 goto leave;
 622         }
 623
 624 restarted_transaction:
 625         status = dquot_alloc_space_nodirty(inode,
 626                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 627         if (status)
 628                 goto leave;
 629         did_quota = 1;
 630
 631         /* reserve a write to the file entry early on - that we if we
 632          * run out of credits in the allocation path, we can still
 633          * update i_size. */
 634         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 635                                          OCFS2_JOURNAL_ACCESS_WRITE);
 636         if (status < 0) {
 637                 mlog_errno(status);
 638                 goto leave;
 639         }
 640
 641         prev_clusters = OCFS2_I(inode)->ip_clusters;
 642
 643         status = ocfs2_add_inode_data(osb,
 644                                       inode,
 645                                       &logical_start,
 646                                       clusters_to_add,
 647                                       mark_unwritten,
 648                                       bh,
 649                                       handle,
 650                                       data_ac,
 651                                       meta_ac,
 652                                       &why);
 653         if ((status < 0) && (status != -EAGAIN)) {
 654                 if (status != -ENOSPC)
 655                         mlog_errno(status);
 656                 goto leave;
 657         }
 658
 659         ocfs2_journal_dirty(handle, bh);
 660
 661         spin_lock(&OCFS2_I(inode)->ip_lock);
 662         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 663         spin_unlock(&OCFS2_I(inode)->ip_lock);
 664         /* Release unused quota reservation */
 665         dquot_free_space(inode,
 666                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 667         did_quota = 0;
 668
 669         if (why != RESTART_NONE && clusters_to_add) {
 670                 if (why == RESTART_META) {
 671                         mlog(0, "restarting function.\n");
 672                         restart_func = 1;
 673                         status = 0;
 674                 } else {
 675                         BUG_ON(why != RESTART_TRANS);
 676
 677                         mlog(0, "restarting transaction.\n");
 678                         /* TODO: This can be more intelligent. */
 679                         credits = ocfs2_calc_extend_credits(osb->sb,
 680                                                             &fe->id2.i_list,
 681                                                             clusters_to_add);
 682                         status = ocfs2_extend_trans(handle, credits);
 683                         if (status < 0) {
 684                                 /* handle still has to be committed at
 685                                  * this point. */
 686                                 status = -ENOMEM;
 687                                 mlog_errno(status);
 688                                 goto leave;
 689                         }
 690                         goto restarted_transaction;
 691                 }
 692         }
 693
 694         mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
 695              le32_to_cpu(fe->i_clusters),
 696              (unsigned long long)le64_to_cpu(fe->i_size));
 697         mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
 698              OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
 699
 700 leave:
 701         if (status < 0 && did_quota)
 702                 dquot_free_space(inode,
 703                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 704         if (handle) {
 705                 ocfs2_commit_trans(osb, handle);
 706                 handle = NULL;
 707         }
 708         if (data_ac) {
 709                 ocfs2_free_alloc_context(data_ac);
 710                 data_ac = NULL;
 711         }
 712         if (meta_ac) {
 713                 ocfs2_free_alloc_context(meta_ac);
 714                 meta_ac = NULL;
 715         }
 716         if ((!status) && restart_func) {
 717                 restart_func = 0;
 718                 goto restart_all;
 719         }
 720         brelse(bh);
 721         bh = NULL;
 722
 723         mlog_exit(status);
 724         return status;
 725 }
 726
 727 /*
 728  * While a write will already be ordering the data, a truncate will not.
 729  * Thus, we need to explicitly order the zeroed pages.
 730  */
 731 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
 732 {
 733         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 734         handle_t *handle = NULL;
 735         int ret = 0;
 736
 737         if (!ocfs2_should_order_data(inode))
 738                 goto out;
 739
 740         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 741         if (IS_ERR(handle)) {
 742                 ret = -ENOMEM;
 743                 mlog_errno(ret);
 744                 goto out;
 745         }
 746
 747         ret = ocfs2_jbd2_file_inode(handle, inode);
 748         if (ret < 0)
 749                 mlog_errno(ret);
 750
 751 out:
 752         if (ret) {
 753                 if (!IS_ERR(handle))
 754                         ocfs2_commit_trans(osb, handle);
 755                 handle = ERR_PTR(ret);
 756         }
 757         return handle;
 758 }
 759
 760 /* Some parts of this taken from generic_cont_expand, which turned out
 761  * to be too fragile to do exactly what we need without us having to
 762  * worry about recursive locking in ->write_begin() and ->write_end(). */
 763 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 764                                  u64 abs_to)
 765 {
 766         struct address_space *mapping = inode->i_mapping;
 767         struct page *page;
 768         unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
 769         handle_t *handle = NULL;
 770         int ret = 0;
 771         unsigned zero_from, zero_to, block_start, block_end;
 772
 773         BUG_ON(abs_from >= abs_to);
 774         BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
 775         BUG_ON(abs_from & (inode->i_blkbits - 1));
 776
 777         page = grab_cache_page(mapping, index);
 778         if (!page) {
 779                 ret = -ENOMEM;
 780                 mlog_errno(ret);
 781                 goto out;
 782         }
 783
 784         /* Get the offsets within the page that we want to zero */
 785         zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
 786         zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
 787         if (!zero_to)
 788                 zero_to = PAGE_CACHE_SIZE;
 789
 790         mlog(0,
 791              "abs_from = %llu, abs_to = %llu, index = %lu, zero_from = %u, zero_to = %u\n",
 792              (unsigned long long)abs_from, (unsigned long long)abs_to,
 793              index, zero_from, zero_to);
 794
 795         /* We know that zero_from is block aligned */
 796         for (block_start = zero_from; block_start < zero_to;
 797              block_start = block_end) {
 798                 block_end = block_start + (1 << inode->i_blkbits);
 799
 800                 /*
 801                  * block_start is block-aligned.  Bump it by one to
 802                  * force ocfs2_{prepare,commit}_write() to zero the
 803                  * whole block.
 804                  */
 805                 ret = ocfs2_prepare_write_nolock(inode, page,
 806                                                  block_start + 1,
 807                                                  block_start + 1);
 808                 if (ret < 0) {
 809                         mlog_errno(ret);
 810                         goto out_unlock;
 811                 }
 812
 813                 if (!handle) {
 814                         handle = ocfs2_zero_start_ordered_transaction(inode);
 815                         if (IS_ERR(handle)) {
 816                                 ret = PTR_ERR(handle);
 817                                 handle = NULL;
 818                                 break;
 819                         }
 820                 }
 821
 822                 /* must not update i_size! */
 823                 ret = block_commit_write(page, block_start + 1,
 824                                          block_start + 1);
 825                 if (ret < 0)
 826                         mlog_errno(ret);
 827                 else
 828                         ret = 0;
 829         }
 830
 831         if (handle)
 832                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 833
 834 out_unlock:
 835         unlock_page(page);
 836         page_cache_release(page);
 837 out:
 838         return ret;
 839 }
 840
 841 /*
 842  * Find the next range to zero.  We do this in terms of bytes because
 843  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 844  * pagecache.  We may return multiple extents.
 845  *
 846  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 847  * needs to be zeroed.  range_start and range_end return the next zeroing
 848  * range.  A subsequent call should pass the previous range_end as its
 849  * zero_start.  If range_end is 0, there's nothing to do.
 850  *
 851  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 852  */
 853 static int ocfs2_zero_extend_get_range(struct inode *inode,
 854                                        struct buffer_head *di_bh,
 855                                        u64 zero_start, u64 zero_end,
 856                                        u64 *range_start, u64 *range_end)
 857 {
 858         int rc = 0, needs_cow = 0;
 859         u32 p_cpos, zero_clusters = 0;
 860         u32 zero_cpos =
 861                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 862         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 863         unsigned int num_clusters = 0;
 864         unsigned int ext_flags = 0;
 865
 866         while (zero_cpos < last_cpos) {
 867                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 868                                         &num_clusters, &ext_flags);
 869                 if (rc) {
 870                         mlog_errno(rc);
 871                         goto out;
 872                 }
 873
 874                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 875                         zero_clusters = num_clusters;
 876                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 877                                 needs_cow = 1;
 878                         break;
 879                 }
 880
 881                 zero_cpos += num_clusters;
 882         }
 883         if (!zero_clusters) {
 884                 *range_end = 0;
 885                 goto out;
 886         }
 887
 888         while ((zero_cpos + zero_clusters) < last_cpos) {
 889                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 890                                         &p_cpos, &num_clusters,
 891                                         &ext_flags);
 892                 if (rc) {
 893                         mlog_errno(rc);
 894                         goto out;
 895                 }
 896
 897                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 898                         break;
 899                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 900                         needs_cow = 1;
 901                 zero_clusters += num_clusters;
 902         }
 903         if ((zero_cpos + zero_clusters) > last_cpos)
 904                 zero_clusters = last_cpos - zero_cpos;
 905
 906         if (needs_cow) {
 907                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos, zero_clusters,
 908                                         UINT_MAX);
 909                 if (rc) {
 910                         mlog_errno(rc);
 911                         goto out;
 912                 }
 913         }
 914
 915         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 916         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 917                                              zero_cpos + zero_clusters);
 918
 919 out:
 920         return rc;
 921 }
 922
 923 /*
 924  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 925  * has made sure that the entire range needs zeroing.
 926  */
 927 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 928                                    u64 range_end)
 929 {
 930         int rc = 0;
 931         u64 next_pos;
 932         u64 zero_pos = range_start;
 933
 934         mlog(0, "range_start = %llu, range_end = %llu\n",
 935              (unsigned long long)range_start,
 936              (unsigned long long)range_end);
 937         BUG_ON(range_start >= range_end);
 938
 939         while (zero_pos < range_end) {
 940                 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
 941                 if (next_pos > range_end)
 942                         next_pos = range_end;
 943                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
 944                 if (rc < 0) {
 945                         mlog_errno(rc);
 946                         break;
 947                 }
 948                 zero_pos = next_pos;
 949
 950                 /*
 951                  * Very large extends have the potential to lock up
 952                  * the cpu for extended periods of time.
 953                  */
 954                 cond_resched();
 955         }
 956
 957         return rc;
 958 }
 959
 960 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 961                       loff_t zero_to_size)
 962 {
 963         int ret = 0;
 964         u64 zero_start, range_start = 0, range_end = 0;
 965         struct super_block *sb = inode->i_sb;
 966
 967         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 968         mlog(0, "zero_start %llu for i_size %llu\n",
 969              (unsigned long long)zero_start,
 970              (unsigned long long)i_size_read(inode));
 971         while (zero_start < zero_to_size) {
 972                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 973                                                   zero_to_size,
 974                                                   &range_start,
 975                                                   &range_end);
 976                 if (ret) {
 977                         mlog_errno(ret);
 978                         break;
 979                 }
 980                 if (!range_end)
 981                         break;
 982                 /* Trim the ends */
 983                 if (range_start < zero_start)
 984                         range_start = zero_start;
 985                 if (range_end > zero_to_size)
 986                         range_end = zero_to_size;
 987
 988                 ret = ocfs2_zero_extend_range(inode, range_start,
 989                                               range_end);
 990                 if (ret) {
 991                         mlog_errno(ret);
 992                         break;
 993                 }
 994                 zero_start = range_end;
 995         }
 996
 997         return ret;
 998 }
 999
1000 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1001                           u64 new_i_size, u64 zero_to)
1002 {
1003         int ret;
1004         u32 clusters_to_add;
1005         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1006
1007         /*
1008          * Only quota files call this without a bh, and they can't be
1009          * refcounted.
1010          */
1011         BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1012         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1013
1014         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1015         if (clusters_to_add < oi->ip_clusters)
1016                 clusters_to_add = 0;
1017         else
1018                 clusters_to_add -= oi->ip_clusters;
1019
1020         if (clusters_to_add) {
1021                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1022                                                 clusters_to_add, 0);
1023                 if (ret) {
1024                         mlog_errno(ret);
1025                         goto out;
1026                 }
1027         }
1028
1029         /*
1030          * Call this even if we don't add any clusters to the tree. We
1031          * still need to zero the area between the old i_size and the
1032          * new i_size.
1033          */
1034         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1035         if (ret < 0)
1036                 mlog_errno(ret);
1037
1038 out:
1039         return ret;
1040 }
1041
1042 static int ocfs2_extend_file(struct inode *inode,
1043                              struct buffer_head *di_bh,
1044                              u64 new_i_size)
1045 {
1046         int ret = 0;
1047         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1048
1049         BUG_ON(!di_bh);
1050
1051         /* setattr sometimes calls us like this. */
1052         if (new_i_size == 0)
1053                 goto out;
1054
1055         if (i_size_read(inode) == new_i_size)
1056                 goto out;
1057         BUG_ON(new_i_size < i_size_read(inode));
1058
1059         /*
1060          * The alloc sem blocks people in read/write from reading our
1061          * allocation until we're done changing it. We depend on
1062          * i_mutex to block other extend/truncate calls while we're
1063          * here.  We even have to hold it for sparse files because there
1064          * might be some tail zeroing.
1065          */
1066         down_write(&oi->ip_alloc_sem);
1067
1068         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1069                 /*
1070                  * We can optimize small extends by keeping the inodes
1071                  * inline data.
1072                  */
1073                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1074                         up_write(&oi->ip_alloc_sem);
1075                         goto out_update_size;
1076                 }
1077
1078                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1079                 if (ret) {
1080                         up_write(&oi->ip_alloc_sem);
1081                         mlog_errno(ret);
1082                         goto out;
1083                 }
1084         }
1085
1086         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1087                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1088         else
1089                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1090                                             new_i_size);
1091
1092         up_write(&oi->ip_alloc_sem);
1093
1094         if (ret < 0) {
1095                 mlog_errno(ret);
1096                 goto out;
1097         }
1098
1099 out_update_size:
1100         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1101         if (ret < 0)
1102                 mlog_errno(ret);
1103
1104 out:
1105         return ret;
1106 }
1107
1108 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1109 {
1110         int status = 0, size_change;
1111         struct inode *inode = dentry->d_inode;
1112         struct super_block *sb = inode->i_sb;
1113         struct ocfs2_super *osb = OCFS2_SB(sb);
1114         struct buffer_head *bh = NULL;
1115         handle_t *handle = NULL;
1116         struct dquot *transfer_to[MAXQUOTAS] = { };
1117         int qtype;
1118
1119         mlog_entry("(0x%p, '%.*s')\n", dentry,
1120                    dentry->d_name.len, dentry->d_name.name);
1121
1122         /* ensuring we don't even attempt to truncate a symlink */
1123         if (S_ISLNK(inode->i_mode))
1124                 attr->ia_valid &= ~ATTR_SIZE;
1125
1126         if (attr->ia_valid & ATTR_MODE)
1127                 mlog(0, "mode change: %d\n", attr->ia_mode);
1128         if (attr->ia_valid & ATTR_UID)
1129                 mlog(0, "uid change: %d\n", attr->ia_uid);
1130         if (attr->ia_valid & ATTR_GID)
1131                 mlog(0, "gid change: %d\n", attr->ia_gid);
1132         if (attr->ia_valid & ATTR_SIZE)
1133                 mlog(0, "size change...\n");
1134         if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
1135                 mlog(0, "time change...\n");
1136
1137 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1138                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1139         if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
1140                 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
1141                 return 0;
1142         }
1143
1144         status = inode_change_ok(inode, attr);
1145         if (status)
1146                 return status;
1147
1148         if (is_quota_modification(inode, attr))
1149                 dquot_initialize(inode);
1150         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1151         if (size_change) {
1152                 status = ocfs2_rw_lock(inode, 1);
1153                 if (status < 0) {
1154                         mlog_errno(status);
1155                         goto bail;
1156                 }
1157         }
1158
1159         status = ocfs2_inode_lock(inode, &bh, 1);
1160         if (status < 0) {
1161                 if (status != -ENOENT)
1162                         mlog_errno(status);
1163                 goto bail_unlock_rw;
1164         }
1165
1166         if (size_change && attr->ia_size != i_size_read(inode)) {
1167                 status = inode_newsize_ok(inode, attr->ia_size);
1168                 if (status)
1169                         goto bail_unlock;
1170
1171                 if (i_size_read(inode) > attr->ia_size) {
1172                         if (ocfs2_should_order_data(inode)) {
1173                                 status = ocfs2_begin_ordered_truncate(inode,
1174                                                                       attr->ia_size);
1175                                 if (status)
1176                                         goto bail_unlock;
1177                         }
1178                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1179                 } else
1180                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1181                 if (status < 0) {
1182                         if (status != -ENOSPC)
1183                                 mlog_errno(status);
1184                         status = -ENOSPC;
1185                         goto bail_unlock;
1186                 }
1187         }
1188
1189         if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
1190             (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
1191                 /*
1192                  * Gather pointers to quota structures so that allocation /
1193                  * freeing of quota structures happens here and not inside
1194                  * dquot_transfer() where we have problems with lock ordering
1195                  */
1196                 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
1197                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1198                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1199                         transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
1200                                                       USRQUOTA);
1201                         if (!transfer_to[USRQUOTA]) {
1202                                 status = -ESRCH;
1203                                 goto bail_unlock;
1204                         }
1205                 }
1206                 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
1207                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1208                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1209                         transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
1210                                                       GRPQUOTA);
1211                         if (!transfer_to[GRPQUOTA]) {
1212                                 status = -ESRCH;
1213                                 goto bail_unlock;
1214                         }
1215                 }
1216                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1217                                            2 * ocfs2_quota_trans_credits(sb));
1218                 if (IS_ERR(handle)) {
1219                         status = PTR_ERR(handle);
1220                         mlog_errno(status);
1221                         goto bail_unlock;
1222                 }
1223                 status = __dquot_transfer(inode, transfer_to);
1224                 if (status < 0)
1225                         goto bail_commit;
1226         } else {
1227                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1228                 if (IS_ERR(handle)) {
1229                         status = PTR_ERR(handle);
1230                         mlog_errno(status);
1231                         goto bail_unlock;
1232                 }
1233         }
1234
1235         /*
1236          * This will intentionally not wind up calling simple_setsize(),
1237          * since all the work for a size change has been done above.
1238          * Otherwise, we could get into problems with truncate as
1239          * ip_alloc_sem is used there to protect against i_size
1240          * changes.
1241          *
1242          * XXX: this means the conditional below can probably be removed.
1243          */
1244         if ((attr->ia_valid & ATTR_SIZE) &&
1245             attr->ia_size != i_size_read(inode)) {
1246                 status = vmtruncate(inode, attr->ia_size);
1247                 if (status) {
1248                         mlog_errno(status);
1249                         goto bail_commit;
1250                 }
1251         }
1252
1253         setattr_copy(inode, attr);
1254         mark_inode_dirty(inode);
1255
1256         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1257         if (status < 0)
1258                 mlog_errno(status);
1259
1260 bail_commit:
1261         ocfs2_commit_trans(osb, handle);
1262 bail_unlock:
1263         ocfs2_inode_unlock(inode, 1);
1264 bail_unlock_rw:
1265         if (size_change)
1266                 ocfs2_rw_unlock(inode, 1);
1267 bail:
1268         brelse(bh);
1269
1270         /* Release quota pointers in case we acquired them */
1271         for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1272                 dqput(transfer_to[qtype]);
1273
1274         if (!status && attr->ia_valid & ATTR_MODE) {
1275                 status = ocfs2_acl_chmod(inode);
1276                 if (status < 0)
1277                         mlog_errno(status);
1278         }
1279
1280         mlog_exit(status);
1281         return status;
1282 }
1283
1284 int ocfs2_getattr(struct vfsmount *mnt,
1285                   struct dentry *dentry,
1286                   struct kstat *stat)
1287 {
1288         struct inode *inode = dentry->d_inode;
1289         struct super_block *sb = dentry->d_inode->i_sb;
1290         struct ocfs2_super *osb = sb->s_fs_info;
1291         int err;
1292
1293         mlog_entry_void();
1294
1295         err = ocfs2_inode_revalidate(dentry);
1296         if (err) {
1297                 if (err != -ENOENT)
1298                         mlog_errno(err);
1299                 goto bail;
1300         }
1301
1302         generic_fillattr(inode, stat);
1303
1304         /* We set the blksize from the cluster size for performance */
1305         stat->blksize = osb->s_clustersize;
1306
1307 bail:
1308         mlog_exit(err);
1309
1310         return err;
1311 }
1312
1313 int ocfs2_permission(struct inode *inode, int mask)
1314 {
1315         int ret;
1316
1317         mlog_entry_void();
1318
1319         ret = ocfs2_inode_lock(inode, NULL, 0);
1320         if (ret) {
1321                 if (ret != -ENOENT)
1322                         mlog_errno(ret);
1323                 goto out;
1324         }
1325
1326         ret = generic_permission(inode, mask, ocfs2_check_acl);
1327
1328         ocfs2_inode_unlock(inode, 0);
1329 out:
1330         mlog_exit(ret);
1331         return ret;
1332 }
1333
1334 static int __ocfs2_write_remove_suid(struct inode *inode,
1335                                      struct buffer_head *bh)
1336 {
1337         int ret;
1338         handle_t *handle;
1339         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1340         struct ocfs2_dinode *di;
1341
1342         mlog_entry("(Inode %llu, mode 0%o)\n",
1343                    (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
1344
1345         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1346         if (IS_ERR(handle)) {
1347                 ret = PTR_ERR(handle);
1348                 mlog_errno(ret);
1349                 goto out;
1350         }
1351
1352         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1353                                       OCFS2_JOURNAL_ACCESS_WRITE);
1354         if (ret < 0) {
1355                 mlog_errno(ret);
1356                 goto out_trans;
1357         }
1358
1359         inode->i_mode &= ~S_ISUID;
1360         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1361                 inode->i_mode &= ~S_ISGID;
1362
1363         di = (struct ocfs2_dinode *) bh->b_data;
1364         di->i_mode = cpu_to_le16(inode->i_mode);
1365
1366         ocfs2_journal_dirty(handle, bh);
1367
1368 out_trans:
1369         ocfs2_commit_trans(osb, handle);
1370 out:
1371         mlog_exit(ret);
1372         return ret;
1373 }
1374
1375 /*
1376  * Will look for holes and unwritten extents in the range starting at
1377  * pos for count bytes (inclusive).
1378  */
1379 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1380                                        size_t count)
1381 {
1382         int ret = 0;
1383         unsigned int extent_flags;
1384         u32 cpos, clusters, extent_len, phys_cpos;
1385         struct super_block *sb = inode->i_sb;
1386
1387         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1388         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1389
1390         while (clusters) {
1391                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1392                                          &extent_flags);
1393                 if (ret < 0) {
1394                         mlog_errno(ret);
1395                         goto out;
1396                 }
1397
1398                 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1399                         ret = 1;
1400                         break;
1401                 }
1402
1403                 if (extent_len > clusters)
1404                         extent_len = clusters;
1405
1406                 clusters -= extent_len;
1407                 cpos += extent_len;
1408         }
1409 out:
1410         return ret;
1411 }
1412
1413 static int ocfs2_write_remove_suid(struct inode *inode)
1414 {
1415         int ret;
1416         struct buffer_head *bh = NULL;
1417
1418         ret = ocfs2_read_inode_block(inode, &bh);
1419         if (ret < 0) {
1420                 mlog_errno(ret);
1421                 goto out;
1422         }
1423
1424         ret =  __ocfs2_write_remove_suid(inode, bh);
1425 out:
1426         brelse(bh);
1427         return ret;
1428 }
1429
1430 /*
1431  * Allocate enough extents to cover the region starting at byte offset
1432  * start for len bytes. Existing extents are skipped, any extents
1433  * added are marked as "unwritten".
1434  */
1435 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1436                                             u64 start, u64 len)
1437 {
1438         int ret;
1439         u32 cpos, phys_cpos, clusters, alloc_size;
1440         u64 end = start + len;
1441         struct buffer_head *di_bh = NULL;
1442
1443         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1444                 ret = ocfs2_read_inode_block(inode, &di_bh);
1445                 if (ret) {
1446                         mlog_errno(ret);
1447                         goto out;
1448                 }
1449
1450                 /*
1451                  * Nothing to do if the requested reservation range
1452                  * fits within the inode.
1453                  */
1454                 if (ocfs2_size_fits_inline_data(di_bh, end))
1455                         goto out;
1456
1457                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1458                 if (ret) {
1459                         mlog_errno(ret);
1460                         goto out;
1461                 }
1462         }
1463
1464         /*
1465          * We consider both start and len to be inclusive.
1466          */
1467         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1468         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1469         clusters -= cpos;
1470
1471         while (clusters) {
1472                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1473                                          &alloc_size, NULL);
1474                 if (ret) {
1475                         mlog_errno(ret);
1476                         goto out;
1477                 }
1478
1479                 /*
1480                  * Hole or existing extent len can be arbitrary, so
1481                  * cap it to our own allocation request.
1482                  */
1483                 if (alloc_size > clusters)
1484                         alloc_size = clusters;
1485
1486                 if (phys_cpos) {
1487                         /*
1488                          * We already have an allocation at this
1489                          * region so we can safely skip it.
1490                          */
1491                         goto next;
1492                 }
1493
1494                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1495                 if (ret) {
1496                         if (ret != -ENOSPC)
1497                                 mlog_errno(ret);
1498                         goto out;
1499                 }
1500
1501 next:
1502                 cpos += alloc_size;
1503                 clusters -= alloc_size;
1504         }
1505
1506         ret = 0;
1507 out:
1508
1509         brelse(di_bh);
1510         return ret;
1511 }
1512
1513 /*
1514  * Truncate a byte range, avoiding pages within partial clusters. This
1515  * preserves those pages for the zeroing code to write to.
1516  */
1517 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1518                                          u64 byte_len)
1519 {
1520         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1521         loff_t start, end;
1522         struct address_space *mapping = inode->i_mapping;
1523
1524         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1525         end = byte_start + byte_len;
1526         end = end & ~(osb->s_clustersize - 1);
1527
1528         if (start < end) {
1529                 unmap_mapping_range(mapping, start, end - start, 0);
1530                 truncate_inode_pages_range(mapping, start, end - 1);
1531         }
1532 }
1533
1534 static int ocfs2_zero_partial_clusters(struct inode *inode,
1535                                        u64 start, u64 len)
1536 {
1537         int ret = 0;
1538         u64 tmpend, end = start + len;
1539         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1540         unsigned int csize = osb->s_clustersize;
1541         handle_t *handle;
1542
1543         /*
1544          * The "start" and "end" values are NOT necessarily part of
1545          * the range whose allocation is being deleted. Rather, this
1546          * is what the user passed in with the request. We must zero
1547          * partial clusters here. There's no need to worry about
1548          * physical allocation - the zeroing code knows to skip holes.
1549          */
1550         mlog(0, "byte start: %llu, end: %llu\n",
1551              (unsigned long long)start, (unsigned long long)end);
1552
1553         /*
1554          * If both edges are on a cluster boundary then there's no
1555          * zeroing required as the region is part of the allocation to
1556          * be truncated.
1557          */
1558         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1559                 goto out;
1560
1561         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1562         if (IS_ERR(handle)) {
1563                 ret = PTR_ERR(handle);
1564                 mlog_errno(ret);
1565                 goto out;
1566         }
1567
1568         /*
1569          * We want to get the byte offset of the end of the 1st cluster.
1570          */
1571         tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1572         if (tmpend > end)
1573                 tmpend = end;
1574
1575         mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1576              (unsigned long long)start, (unsigned long long)tmpend);
1577
1578         ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1579         if (ret)
1580                 mlog_errno(ret);
1581
1582         if (tmpend < end) {
1583                 /*
1584                  * This may make start and end equal, but the zeroing
1585                  * code will skip any work in that case so there's no
1586                  * need to catch it up here.
1587                  */
1588                 start = end & ~(osb->s_clustersize - 1);
1589
1590                 mlog(0, "2nd range: start: %llu, end: %llu\n",
1591                      (unsigned long long)start, (unsigned long long)end);
1592
1593                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1594                 if (ret)
1595                         mlog_errno(ret);
1596         }
1597
1598         ocfs2_commit_trans(osb, handle);
1599 out:
1600         return ret;
1601 }
1602
1603 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1604 {
1605         int i;
1606         struct ocfs2_extent_rec *rec = NULL;
1607
1608         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1609
1610                 rec = &el->l_recs[i];
1611
1612                 if (le32_to_cpu(rec->e_cpos) < pos)
1613                         break;
1614         }
1615
1616         return i;
1617 }
1618
1619 /*
1620  * Helper to calculate the punching pos and length in one run, we handle the
1621  * following three cases in order:
1622  *
1623  * - remove the entire record
1624  * - remove a partial record
1625  * - no record needs to be removed (hole-punching completed)
1626 */
1627 static void ocfs2_calc_trunc_pos(struct inode *inode,
1628                                  struct ocfs2_extent_list *el,
1629                                  struct ocfs2_extent_rec *rec,
1630                                  u32 trunc_start, u32 *trunc_cpos,
1631                                  u32 *trunc_len, u32 *trunc_end,
1632                                  u64 *blkno, int *done)
1633 {
1634         int ret = 0;
1635         u32 coff, range;
1636
1637         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1638
1639         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1640                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1641                 /*
1642                  * Skip holes if any.
1643                  */
1644                 if (range < *trunc_end)
1645                         *trunc_end = range;
1646                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1647                 *blkno = le64_to_cpu(rec->e_blkno);
1648                 *trunc_end = le32_to_cpu(rec->e_cpos);
1649         } else if (range > trunc_start) {
1650                 *trunc_cpos = trunc_start;
1651                 *trunc_len = *trunc_end - trunc_start;
1652                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1653                 *blkno = le64_to_cpu(rec->e_blkno) +
1654                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1655                 *trunc_end = trunc_start;
1656         } else {
1657                 /*
1658                  * It may have two following possibilities:
1659                  *
1660                  * - last record has been removed
1661                  * - trunc_start was within a hole
1662                  *
1663                  * both two cases mean the completion of hole punching.
1664                  */
1665                 ret = 1;
1666         }
1667
1668         *done = ret;
1669 }
1670
1671 static int ocfs2_remove_inode_range(struct inode *inode,
1672                                     struct buffer_head *di_bh, u64 byte_start,
1673                                     u64 byte_len)
1674 {
1675         int ret = 0, flags = 0, done = 0, i;
1676         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1677         u32 cluster_in_el;
1678         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1679         struct ocfs2_cached_dealloc_ctxt dealloc;
1680         struct address_space *mapping = inode->i_mapping;
1681         struct ocfs2_extent_tree et;
1682         struct ocfs2_path *path = NULL;
1683         struct ocfs2_extent_list *el = NULL;
1684         struct ocfs2_extent_rec *rec = NULL;
1685         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1686         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1687
1688         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1689         ocfs2_init_dealloc_ctxt(&dealloc);
1690
1691         if (byte_len == 0)
1692                 return 0;
1693
1694         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1695                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1696                                             byte_start + byte_len, 0);
1697                 if (ret) {
1698                         mlog_errno(ret);
1699                         goto out;
1700                 }
1701                 /*
1702                  * There's no need to get fancy with the page cache
1703                  * truncate of an inline-data inode. We're talking
1704                  * about less than a page here, which will be cached
1705                  * in the dinode buffer anyway.
1706                  */
1707                 unmap_mapping_range(mapping, 0, 0, 0);
1708                 truncate_inode_pages(mapping, 0);
1709                 goto out;
1710         }
1711
1712         /*
1713          * For reflinks, we may need to CoW 2 clusters which might be
1714          * partially zero'd later, if hole's start and end offset were
1715          * within one cluster(means is not exactly aligned to clustersize).
1716          */
1717
1718         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1719
1720                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1721                 if (ret) {
1722                         mlog_errno(ret);
1723                         goto out;
1724                 }
1725
1726                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1727                 if (ret) {
1728                         mlog_errno(ret);
1729                         goto out;
1730                 }
1731         }
1732
1733         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1734         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1735         cluster_in_el = trunc_end;
1736
1737         mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, cend: %u\n",
1738              (unsigned long long)OCFS2_I(inode)->ip_blkno,
1739              (unsigned long long)byte_start,
1740              (unsigned long long)byte_len, trunc_start, trunc_end);
1741
1742         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1743         if (ret) {
1744                 mlog_errno(ret);
1745                 goto out;
1746         }
1747
1748         path = ocfs2_new_path_from_et(&et);
1749         if (!path) {
1750                 ret = -ENOMEM;
1751                 mlog_errno(ret);
1752                 goto out;
1753         }
1754
1755         while (trunc_end > trunc_start) {
1756
1757                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1758                                       cluster_in_el);
1759                 if (ret) {
1760                         mlog_errno(ret);
1761                         goto out;
1762                 }
1763
1764                 el = path_leaf_el(path);
1765
1766                 i = ocfs2_find_rec(el, trunc_end);
1767                 /*
1768                  * Need to go to previous extent block.
1769                  */
1770                 if (i < 0) {
1771                         if (path->p_tree_depth == 0)
1772                                 break;
1773
1774                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1775                                                             path,
1776                                                             &cluster_in_el);
1777                         if (ret) {
1778                                 mlog_errno(ret);
1779                                 goto out;
1780                         }
1781
1782                         /*
1783                          * We've reached the leftmost extent block,
1784                          * it's safe to leave.
1785                          */
1786                         if (cluster_in_el == 0)
1787                                 break;
1788
1789                         /*
1790                          * The 'pos' searched for previous extent block is
1791                          * always one cluster less than actual trunc_end.
1792                          */
1793                         trunc_end = cluster_in_el + 1;
1794
1795                         ocfs2_reinit_path(path, 1);
1796
1797                         continue;
1798
1799                 } else
1800                         rec = &el->l_recs[i];
1801
1802                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1803                                      &trunc_len, &trunc_end, &blkno, &done);
1804                 if (done)
1805                         break;
1806
1807                 flags = rec->e_flags;
1808                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1809
1810                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1811                                                phys_cpos, trunc_len, flags,
1812                                                &dealloc, refcount_loc);
1813                 if (ret < 0) {
1814                         mlog_errno(ret);
1815                         goto out;
1816                 }
1817
1818                 cluster_in_el = trunc_end;
1819
1820                 ocfs2_reinit_path(path, 1);
1821         }
1822
1823         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1824
1825 out:
1826         ocfs2_schedule_truncate_log_flush(osb, 1);
1827         ocfs2_run_deallocs(osb, &dealloc);
1828
1829         return ret;
1830 }
1831
1832 /*
1833  * Parts of this function taken from xfs_change_file_space()
1834  */
1835 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1836                                      loff_t f_pos, unsigned int cmd,
1837                                      struct ocfs2_space_resv *sr,
1838                                      int change_size)
1839 {
1840         int ret;
1841         s64 llen;
1842         loff_t size;
1843         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1844         struct buffer_head *di_bh = NULL;
1845         handle_t *handle;
1846         unsigned long long max_off = inode->i_sb->s_maxbytes;
1847
1848         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1849                 return -EROFS;
1850
1851         mutex_lock(&inode->i_mutex);
1852
1853         /*
1854          * This prevents concurrent writes on other nodes
1855          */
1856         ret = ocfs2_rw_lock(inode, 1);
1857         if (ret) {
1858                 mlog_errno(ret);
1859                 goto out;
1860         }
1861
1862         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1863         if (ret) {
1864                 mlog_errno(ret);
1865                 goto out_rw_unlock;
1866         }
1867
1868         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1869                 ret = -EPERM;
1870                 goto out_inode_unlock;
1871         }
1872
1873         switch (sr->l_whence) {
1874         case 0: /*SEEK_SET*/
1875                 break;
1876         case 1: /*SEEK_CUR*/
1877                 sr->l_start += f_pos;
1878                 break;
1879         case 2: /*SEEK_END*/
1880                 sr->l_start += i_size_read(inode);
1881                 break;
1882         default:
1883                 ret = -EINVAL;
1884                 goto out_inode_unlock;
1885         }
1886         sr->l_whence = 0;
1887
1888         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1889
1890         if (sr->l_start < 0
1891             || sr->l_start > max_off
1892             || (sr->l_start + llen) < 0
1893             || (sr->l_start + llen) > max_off) {
1894                 ret = -EINVAL;
1895                 goto out_inode_unlock;
1896         }
1897         size = sr->l_start + sr->l_len;
1898
1899         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1900                 if (sr->l_len <= 0) {
1901                         ret = -EINVAL;
1902                         goto out_inode_unlock;
1903                 }
1904         }
1905
1906         if (file && should_remove_suid(file->f_path.dentry)) {
1907                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1908                 if (ret) {
1909                         mlog_errno(ret);
1910                         goto out_inode_unlock;
1911                 }
1912         }
1913
1914         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1915         switch (cmd) {
1916         case OCFS2_IOC_RESVSP:
1917         case OCFS2_IOC_RESVSP64:
1918                 /*
1919                  * This takes unsigned offsets, but the signed ones we
1920                  * pass have been checked against overflow above.
1921                  */
1922                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1923                                                        sr->l_len);
1924                 break;
1925         case OCFS2_IOC_UNRESVSP:
1926         case OCFS2_IOC_UNRESVSP64:
1927                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1928                                                sr->l_len);
1929                 break;
1930         default:
1931                 ret = -EINVAL;
1932         }
1933         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1934         if (ret) {
1935                 mlog_errno(ret);
1936                 goto out_inode_unlock;
1937         }
1938
1939         /*
1940          * We update c/mtime for these changes
1941          */
1942         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1943         if (IS_ERR(handle)) {
1944                 ret = PTR_ERR(handle);
1945                 mlog_errno(ret);
1946                 goto out_inode_unlock;
1947         }
1948
1949         if (change_size && i_size_read(inode) < size)
1950                 i_size_write(inode, size);
1951
1952         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1953         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1954         if (ret < 0)
1955                 mlog_errno(ret);
1956
1957         ocfs2_commit_trans(osb, handle);
1958
1959 out_inode_unlock:
1960         brelse(di_bh);
1961         ocfs2_inode_unlock(inode, 1);
1962 out_rw_unlock:
1963         ocfs2_rw_unlock(inode, 1);
1964
1965 out:
1966         mutex_unlock(&inode->i_mutex);
1967         return ret;
1968 }
1969
1970 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1971                             struct ocfs2_space_resv *sr)
1972 {
1973         struct inode *inode = file->f_path.dentry->d_inode;
1974         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1975
1976         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1977             !ocfs2_writes_unwritten_extents(osb))
1978                 return -ENOTTY;
1979         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1980                  !ocfs2_sparse_alloc(osb))
1981                 return -ENOTTY;
1982
1983         if (!S_ISREG(inode->i_mode))
1984                 return -EINVAL;
1985
1986         if (!(file->f_mode & FMODE_WRITE))
1987                 return -EBADF;
1988
1989         return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1990 }
1991
1992 static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1993                             loff_t len)
1994 {
1995         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1996         struct ocfs2_space_resv sr;
1997         int change_size = 1;
1998
1999         if (!ocfs2_writes_unwritten_extents(osb))
2000                 return -EOPNOTSUPP;
2001
2002         if (S_ISDIR(inode->i_mode))
2003                 return -ENODEV;
2004
2005         if (mode & FALLOC_FL_KEEP_SIZE)
2006                 change_size = 0;
2007
2008         sr.l_whence = 0;
2009         sr.l_start = (s64)offset;
2010         sr.l_len = (s64)len;
2011
2012         return __ocfs2_change_file_space(NULL, inode, offset,
2013                                          OCFS2_IOC_RESVSP64, &sr, change_size);
2014 }
2015
2016 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2017                                    size_t count)
2018 {
2019         int ret = 0;
2020         unsigned int extent_flags;
2021         u32 cpos, clusters, extent_len, phys_cpos;
2022         struct super_block *sb = inode->i_sb;
2023
2024         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2025             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2026             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2027                 return 0;
2028
2029         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2030         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2031
2032         while (clusters) {
2033                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2034                                          &extent_flags);
2035                 if (ret < 0) {
2036                         mlog_errno(ret);
2037                         goto out;
2038                 }
2039
2040                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2041                         ret = 1;
2042                         break;
2043                 }
2044
2045                 if (extent_len > clusters)
2046                         extent_len = clusters;
2047
2048                 clusters -= extent_len;
2049                 cpos += extent_len;
2050         }
2051 out:
2052         return ret;
2053 }
2054
2055 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2056                                             loff_t pos, size_t count,
2057                                             int *meta_level)
2058 {
2059         int ret;
2060         struct buffer_head *di_bh = NULL;
2061         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2062         u32 clusters =
2063                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2064
2065         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2066         if (ret) {
2067                 mlog_errno(ret);
2068                 goto out;
2069         }
2070
2071         *meta_level = 1;
2072
2073         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2074         if (ret)
2075                 mlog_errno(ret);
2076 out:
2077         brelse(di_bh);
2078         return ret;
2079 }
2080
2081 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
2082                                          loff_t *ppos,
2083                                          size_t count,
2084                                          int appending,
2085                                          int *direct_io,
2086                                          int *has_refcount)
2087 {
2088         int ret = 0, meta_level = 0;
2089         struct inode *inode = dentry->d_inode;
2090         loff_t saved_pos, end;
2091
2092         /*
2093          * We start with a read level meta lock and only jump to an ex
2094          * if we need to make modifications here.
2095          */
2096         for(;;) {
2097                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2098                 if (ret < 0) {
2099                         meta_level = -1;
2100                         mlog_errno(ret);
2101                         goto out;
2102                 }
2103
2104                 /* Clear suid / sgid if necessary. We do this here
2105                  * instead of later in the write path because
2106                  * remove_suid() calls ->setattr without any hint that
2107                  * we may have already done our cluster locking. Since
2108                  * ocfs2_setattr() *must* take cluster locks to
2109                  * proceeed, this will lead us to recursively lock the
2110                  * inode. There's also the dinode i_size state which
2111                  * can be lost via setattr during extending writes (we
2112                  * set inode->i_size at the end of a write. */
2113                 if (should_remove_suid(dentry)) {
2114                         if (meta_level == 0) {
2115                                 ocfs2_inode_unlock(inode, meta_level);
2116                                 meta_level = 1;
2117                                 continue;
2118                         }
2119
2120                         ret = ocfs2_write_remove_suid(inode);
2121                         if (ret < 0) {
2122                                 mlog_errno(ret);
2123                                 goto out_unlock;
2124                         }
2125                 }
2126
2127                 /* work on a copy of ppos until we're sure that we won't have
2128                  * to recalculate it due to relocking. */
2129                 if (appending) {
2130                         saved_pos = i_size_read(inode);
2131                         mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
2132                 } else {
2133                         saved_pos = *ppos;
2134                 }
2135
2136                 end = saved_pos + count;
2137
2138                 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2139                 if (ret == 1) {
2140                         ocfs2_inode_unlock(inode, meta_level);
2141                         meta_level = -1;
2142
2143                         ret = ocfs2_prepare_inode_for_refcount(inode,
2144                                                                saved_pos,
2145                                                                count,
2146                                                                &meta_level);
2147                         if (has_refcount)
2148                                 *has_refcount = 1;
2149                         if (direct_io)
2150                                 *direct_io = 0;
2151                 }
2152
2153                 if (ret < 0) {
2154                         mlog_errno(ret);
2155                         goto out_unlock;
2156                 }
2157
2158                 /*
2159                  * Skip the O_DIRECT checks if we don't need
2160                  * them.
2161                  */
2162                 if (!direct_io || !(*direct_io))
2163                         break;
2164
2165                 /*
2166                  * There's no sane way to do direct writes to an inode
2167                  * with inline data.
2168                  */
2169                 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2170                         *direct_io = 0;
2171                         break;
2172                 }
2173
2174                 /*
2175                  * Allowing concurrent direct writes means
2176                  * i_size changes wouldn't be synchronized, so
2177                  * one node could wind up truncating another
2178                  * nodes writes.
2179                  */
2180                 if (end > i_size_read(inode)) {
2181                         *direct_io = 0;
2182                         break;
2183                 }
2184
2185                 /*
2186                  * We don't fill holes during direct io, so
2187                  * check for them here. If any are found, the
2188                  * caller will have to retake some cluster
2189                  * locks and initiate the io as buffered.
2190                  */
2191                 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2192                 if (ret == 1) {
2193                         *direct_io = 0;
2194                         ret = 0;
2195                 } else if (ret < 0)
2196                         mlog_errno(ret);
2197                 break;
2198         }
2199
2200         if (appending)
2201                 *ppos = saved_pos;
2202
2203 out_unlock:
2204         if (meta_level >= 0)
2205                 ocfs2_inode_unlock(inode, meta_level);
2206
2207 out:
2208         return ret;
2209 }
2210
2211 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2212                                     const struct iovec *iov,
2213                                     unsigned long nr_segs,
2214                                     loff_t pos)
2215 {
2216         int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
2217         int can_do_direct, has_refcount = 0;
2218         ssize_t written = 0;
2219         size_t ocount;          /* original count */
2220         size_t count;           /* after file limit checks */
2221         loff_t old_size, *ppos = &iocb->ki_pos;
2222         u32 old_clusters;
2223         struct file *file = iocb->ki_filp;
2224         struct inode *inode = file->f_path.dentry->d_inode;
2225         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2226
2227         mlog_entry("(0x%p, %u, '%.*s')\n", file,
2228                    (unsigned int)nr_segs,
2229                    file->f_path.dentry->d_name.len,
2230                    file->f_path.dentry->d_name.name);
2231
2232         if (iocb->ki_left == 0)
2233                 return 0;
2234
2235         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2236
2237         appending = file->f_flags & O_APPEND ? 1 : 0;
2238         direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2239
2240         mutex_lock(&inode->i_mutex);
2241
2242 relock:
2243         /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
2244         if (direct_io) {
2245                 down_read(&inode->i_alloc_sem);
2246                 have_alloc_sem = 1;
2247         }
2248
2249         /* concurrent O_DIRECT writes are allowed */
2250         rw_level = !direct_io;
2251         ret = ocfs2_rw_lock(inode, rw_level);
2252         if (ret < 0) {
2253                 mlog_errno(ret);
2254                 goto out_sems;
2255         }
2256
2257         can_do_direct = direct_io;
2258         ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
2259                                             iocb->ki_left, appending,
2260                                             &can_do_direct, &has_refcount);
2261         if (ret < 0) {
2262                 mlog_errno(ret);
2263                 goto out;
2264         }
2265
2266         /*
2267          * We can't complete the direct I/O as requested, fall back to
2268          * buffered I/O.
2269          */
2270         if (direct_io && !can_do_direct) {
2271                 ocfs2_rw_unlock(inode, rw_level);
2272                 up_read(&inode->i_alloc_sem);
2273
2274                 have_alloc_sem = 0;
2275                 rw_level = -1;
2276
2277                 direct_io = 0;
2278                 goto relock;
2279         }
2280
2281         /*
2282          * To later detect whether a journal commit for sync writes is
2283          * necessary, we sample i_size, and cluster count here.
2284          */
2285         old_size = i_size_read(inode);
2286         old_clusters = OCFS2_I(inode)->ip_clusters;
2287
2288         /* communicate with ocfs2_dio_end_io */
2289         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2290
2291         ret = generic_segment_checks(iov, &nr_segs, &ocount,
2292                                      VERIFY_READ);
2293         if (ret)
2294                 goto out_dio;
2295
2296         count = ocount;
2297         ret = generic_write_checks(file, ppos, &count,
2298                                    S_ISBLK(inode->i_mode));
2299         if (ret)
2300                 goto out_dio;
2301
2302         if (direct_io) {
2303                 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2304                                                     ppos, count, ocount);
2305                 if (written < 0) {
2306                         /*
2307                          * direct write may have instantiated a few
2308                          * blocks outside i_size. Trim these off again.
2309                          * Don't need i_size_read because we hold i_mutex.
2310                          *
2311                          * XXX(hch): this looks buggy because ocfs2 did not
2312                          * actually implement ->truncate.  Take a look at
2313                          * the new truncate sequence and update this accordingly
2314                          */
2315                         if (*ppos + count > inode->i_size)
2316                                 simple_setsize(inode, inode->i_size);
2317                         ret = written;
2318                         goto out_dio;
2319                 }
2320         } else {
2321                 current->backing_dev_info = file->f_mapping->backing_dev_info;
2322                 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2323                                                       ppos, count, 0);
2324                 current->backing_dev_info = NULL;
2325         }
2326
2327 out_dio:
2328         /* buffered aio wouldn't have proper lock coverage today */
2329         BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2330
2331         if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2332             ((file->f_flags & O_DIRECT) && has_refcount)) {
2333                 ret = filemap_fdatawrite_range(file->f_mapping, pos,
2334                                                pos + count - 1);
2335                 if (ret < 0)
2336                         written = ret;
2337
2338                 if (!ret && ((old_size != i_size_read(inode)) ||
2339                              (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2340                              has_refcount)) {
2341                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2342                         if (ret < 0)
2343                                 written = ret;
2344                 }
2345
2346                 if (!ret)
2347                         ret = filemap_fdatawait_range(file->f_mapping, pos,
2348                                                       pos + count - 1);
2349         }
2350
2351         /*
2352          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2353          * function pointer which is called when o_direct io completes so that
2354          * it can unlock our rw lock.  (it's the clustered equivalent of
2355          * i_alloc_sem; protects truncate from racing with pending ios).
2356          * Unfortunately there are error cases which call end_io and others
2357          * that don't.  so we don't have to unlock the rw_lock if either an
2358          * async dio is going to do it in the future or an end_io after an
2359          * error has already done it.
2360          */
2361         if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2362                 rw_level = -1;
2363                 have_alloc_sem = 0;
2364         }
2365
2366 out:
2367         if (rw_level != -1)
2368                 ocfs2_rw_unlock(inode, rw_level);
2369
2370 out_sems:
2371         if (have_alloc_sem)
2372                 up_read(&inode->i_alloc_sem);
2373
2374         mutex_unlock(&inode->i_mutex);
2375
2376         if (written)
2377                 ret = written;
2378         mlog_exit(ret);
2379         return ret;
2380 }
2381
2382 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2383                                 struct file *out,
2384                                 struct splice_desc *sd)
2385 {
2386         int ret;
2387
2388         ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, &sd->pos,
2389                                             sd->total_len, 0, NULL, NULL);
2390         if (ret < 0) {
2391                 mlog_errno(ret);
2392                 return ret;
2393         }
2394
2395         return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2396 }
2397
2398 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2399                                        struct file *out,
2400                                        loff_t *ppos,
2401                                        size_t len,
2402                                        unsigned int flags)
2403 {
2404         int ret;
2405         struct address_space *mapping = out->f_mapping;
2406         struct inode *inode = mapping->host;
2407         struct splice_desc sd = {
2408                 .total_len = len,
2409                 .flags = flags,
2410                 .pos = *ppos,
2411                 .u.file = out,
2412         };
2413
2414         mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2415                    (unsigned int)len,
2416                    out->f_path.dentry->d_name.len,
2417                    out->f_path.dentry->d_name.name);
2418
2419         if (pipe->inode)
2420                 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
2421
2422         splice_from_pipe_begin(&sd);
2423         do {
2424                 ret = splice_from_pipe_next(pipe, &sd);
2425                 if (ret <= 0)
2426                         break;
2427
2428                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2429                 ret = ocfs2_rw_lock(inode, 1);
2430                 if (ret < 0)
2431                         mlog_errno(ret);
2432                 else {
2433                         ret = ocfs2_splice_to_file(pipe, out, &sd);
2434                         ocfs2_rw_unlock(inode, 1);
2435                 }
2436                 mutex_unlock(&inode->i_mutex);
2437         } while (ret > 0);
2438         splice_from_pipe_end(pipe, &sd);
2439
2440         if (pipe->inode)
2441                 mutex_unlock(&pipe->inode->i_mutex);
2442
2443         if (sd.num_spliced)
2444                 ret = sd.num_spliced;
2445
2446         if (ret > 0) {
2447                 unsigned long nr_pages;
2448                 int err;
2449
2450                 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2451
2452                 err = generic_write_sync(out, *ppos, ret);
2453                 if (err)
2454                         ret = err;
2455                 else
2456                         *ppos += ret;
2457
2458                 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2459         }
2460
2461         mlog_exit(ret);
2462         return ret;
2463 }
2464
2465 static ssize_t ocfs2_file_splice_read(struct file *in,
2466                                       loff_t *ppos,
2467                                       struct pipe_inode_info *pipe,
2468                                       size_t len,
2469                                       unsigned int flags)
2470 {
2471         int ret = 0, lock_level = 0;
2472         struct inode *inode = in->f_path.dentry->d_inode;
2473
2474         mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2475                    (unsigned int)len,
2476                    in->f_path.dentry->d_name.len,
2477                    in->f_path.dentry->d_name.name);
2478
2479         /*
2480          * See the comment in ocfs2_file_aio_read()
2481          */
2482         ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
2483         if (ret < 0) {
2484                 mlog_errno(ret);
2485                 goto bail;
2486         }
2487         ocfs2_inode_unlock(inode, lock_level);
2488
2489         ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2490
2491 bail:
2492         mlog_exit(ret);
2493         return ret;
2494 }
2495
2496 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2497                                    const struct iovec *iov,
2498                                    unsigned long nr_segs,
2499                                    loff_t pos)
2500 {
2501         int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2502         struct file *filp = iocb->ki_filp;
2503         struct inode *inode = filp->f_path.dentry->d_inode;
2504
2505         mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2506                    (unsigned int)nr_segs,
2507                    filp->f_path.dentry->d_name.len,
2508                    filp->f_path.dentry->d_name.name);
2509
2510         if (!inode) {
2511                 ret = -EINVAL;
2512                 mlog_errno(ret);
2513                 goto bail;
2514         }
2515
2516         /*
2517          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2518          * need locks to protect pending reads from racing with truncate.
2519          */
2520         if (filp->f_flags & O_DIRECT) {
2521                 down_read(&inode->i_alloc_sem);
2522                 have_alloc_sem = 1;
2523
2524                 ret = ocfs2_rw_lock(inode, 0);
2525                 if (ret < 0) {
2526                         mlog_errno(ret);
2527                         goto bail;
2528                 }
2529                 rw_level = 0;
2530                 /* communicate with ocfs2_dio_end_io */
2531                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2532         }
2533
2534         /*
2535          * We're fine letting folks race truncates and extending
2536          * writes with read across the cluster, just like they can
2537          * locally. Hence no rw_lock during read.
2538          *
2539          * Take and drop the meta data lock to update inode fields
2540          * like i_size. This allows the checks down below
2541          * generic_file_aio_read() a chance of actually working.
2542          */
2543         ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2544         if (ret < 0) {
2545                 mlog_errno(ret);
2546                 goto bail;
2547         }
2548         ocfs2_inode_unlock(inode, lock_level);
2549
2550         ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2551         if (ret == -EINVAL)
2552                 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2553
2554         /* buffered aio wouldn't have proper lock coverage today */
2555         BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2556
2557         /* see ocfs2_file_aio_write */
2558         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2559                 rw_level = -1;
2560                 have_alloc_sem = 0;
2561         }
2562
2563 bail:
2564         if (have_alloc_sem)
2565                 up_read(&inode->i_alloc_sem);
2566         if (rw_level != -1)
2567                 ocfs2_rw_unlock(inode, rw_level);
2568         mlog_exit(ret);
2569
2570         return ret;
2571 }
2572
2573 const struct inode_operations ocfs2_file_iops = {
2574         .setattr        = ocfs2_setattr,
2575         .getattr        = ocfs2_getattr,
2576         .permission     = ocfs2_permission,
2577         .setxattr       = generic_setxattr,
2578         .getxattr       = generic_getxattr,
2579         .listxattr      = ocfs2_listxattr,
2580         .removexattr    = generic_removexattr,
2581         .fallocate      = ocfs2_fallocate,
2582         .fiemap         = ocfs2_fiemap,
2583 };
2584
2585 const struct inode_operations ocfs2_special_file_iops = {
2586         .setattr        = ocfs2_setattr,
2587         .getattr        = ocfs2_getattr,
2588         .permission     = ocfs2_permission,
2589 };
2590
2591 /*
2592  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2593  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2594  */
2595 const struct file_operations ocfs2_fops = {
2596         .llseek         = generic_file_llseek,
2597         .read           = do_sync_read,
2598         .write          = do_sync_write,
2599         .mmap           = ocfs2_mmap,
2600         .fsync          = ocfs2_sync_file,
2601         .release        = ocfs2_file_release,
2602         .open           = ocfs2_file_open,
2603         .aio_read       = ocfs2_file_aio_read,
2604         .aio_write      = ocfs2_file_aio_write,
2605         .unlocked_ioctl = ocfs2_ioctl,
2606 #ifdef CONFIG_COMPAT
2607         .compat_ioctl   = ocfs2_compat_ioctl,
2608 #endif
2609         .lock           = ocfs2_lock,
2610         .flock          = ocfs2_flock,
2611         .splice_read    = ocfs2_file_splice_read,
2612         .splice_write   = ocfs2_file_splice_write,
2613 };
2614
2615 const struct file_operations ocfs2_dops = {
2616         .llseek         = generic_file_llseek,
2617         .read           = generic_read_dir,
2618         .readdir        = ocfs2_readdir,
2619         .fsync          = ocfs2_sync_file,
2620         .release        = ocfs2_dir_release,
2621         .open           = ocfs2_dir_open,
2622         .unlocked_ioctl = ocfs2_ioctl,
2623 #ifdef CONFIG_COMPAT
2624         .compat_ioctl   = ocfs2_compat_ioctl,
2625 #endif
2626         .lock           = ocfs2_lock,
2627         .flock          = ocfs2_flock,
2628 };
2629
2630 /*
2631  * POSIX-lockless variants of our file_operations.
2632  *
2633  * These will be used if the underlying cluster stack does not support
2634  * posix file locking, if the user passes the "localflocks" mount
2635  * option, or if we have a local-only fs.
2636  *
2637  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2638  * so we still want it in the case of no stack support for
2639  * plocks. Internally, it will do the right thing when asked to ignore
2640  * the cluster.
2641  */
2642 const struct file_operations ocfs2_fops_no_plocks = {
2643         .llseek         = generic_file_llseek,
2644         .read           = do_sync_read,
2645         .write          = do_sync_write,
2646         .mmap           = ocfs2_mmap,
2647         .fsync          = ocfs2_sync_file,
2648         .release        = ocfs2_file_release,
2649         .open           = ocfs2_file_open,
2650         .aio_read       = ocfs2_file_aio_read,
2651         .aio_write      = ocfs2_file_aio_write,
2652         .unlocked_ioctl = ocfs2_ioctl,
2653 #ifdef CONFIG_COMPAT
2654         .compat_ioctl   = ocfs2_compat_ioctl,
2655 #endif
2656         .flock          = ocfs2_flock,
2657         .splice_read    = ocfs2_file_splice_read,
2658         .splice_write   = ocfs2_file_splice_write,
2659 };
2660
2661 const struct file_operations ocfs2_dops_no_plocks = {
2662         .llseek         = generic_file_llseek,
2663         .read           = generic_read_dir,
2664         .readdir        = ocfs2_readdir,
2665         .fsync          = ocfs2_sync_file,
2666         .release        = ocfs2_dir_release,
2667         .open           = ocfs2_dir_open,
2668         .unlocked_ioctl = ocfs2_ioctl,
2669 #ifdef CONFIG_COMPAT
2670         .compat_ioctl   = ocfs2_compat_ioctl,
2671 #endif
2672         .flock          = ocfs2_flock,
2673 };