fs/ocfs2/file.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * file.c
   5  *
   6  * File open, close, extend, truncate
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/capability.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/uio.h>
  33 #include <linux/sched.h>
  34 #include <linux/splice.h>
  35 #include <linux/mount.h>
  36 #include <linux/writeback.h>
  37 #include <linux/falloc.h>
  38 #include <linux/quotaops.h>
  39 #include <linux/blkdev.h>
  40
  41 #include <cluster/masklog.h>
  42
  43 #include "ocfs2.h"
  44
  45 #include "alloc.h"
  46 #include "aops.h"
  47 #include "dir.h"
  48 #include "dlmglue.h"
  49 #include "extent_map.h"
  50 #include "file.h"
  51 #include "sysfile.h"
  52 #include "inode.h"
  53 #include "ioctl.h"
  54 #include "journal.h"
  55 #include "locks.h"
  56 #include "mmap.h"
  57 #include "suballoc.h"
  58 #include "super.h"
  59 #include "xattr.h"
  60 #include "acl.h"
  61 #include "quota.h"
  62 #include "refcounttree.h"
  63 #include "ocfs2_trace.h"
  64
  65 #include "buffer_head_io.h"
  66
  67 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  68 {
  69         struct ocfs2_file_private *fp;
  70
  71         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  72         if (!fp)
  73                 return -ENOMEM;
  74
  75         fp->fp_file = file;
  76         mutex_init(&fp->fp_mutex);
  77         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  78         file->private_data = fp;
  79
  80         return 0;
  81 }
  82
  83 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  84 {
  85         struct ocfs2_file_private *fp = file->private_data;
  86         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  87
  88         if (fp) {
  89                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  90                 ocfs2_lock_res_free(&fp->fp_flock);
  91                 kfree(fp);
  92                 file->private_data = NULL;
  93         }
  94 }
  95
  96 static int ocfs2_file_open(struct inode *inode, struct file *file)
  97 {
  98         int status;
  99         int mode = file->f_flags;
 100         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 101
 102         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
 103                               (unsigned long long)OCFS2_I(inode)->ip_blkno,
 104                               file->f_path.dentry->d_name.len,
 105                               file->f_path.dentry->d_name.name, mode);
 106
 107         if (file->f_mode & FMODE_WRITE)
 108                 dquot_initialize(inode);
 109
 110         spin_lock(&oi->ip_lock);
 111
 112         /* Check that the inode hasn't been wiped from disk by another
 113          * node. If it hasn't then we're safe as long as we hold the
 114          * spin lock until our increment of open count. */
 115         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
 116                 spin_unlock(&oi->ip_lock);
 117
 118                 status = -ENOENT;
 119                 goto leave;
 120         }
 121
 122         if (mode & O_DIRECT)
 123                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 124
 125         oi->ip_open_count++;
 126         spin_unlock(&oi->ip_lock);
 127
 128         status = ocfs2_init_file_private(inode, file);
 129         if (status) {
 130                 /*
 131                  * We want to set open count back if we're failing the
 132                  * open.
 133                  */
 134                 spin_lock(&oi->ip_lock);
 135                 oi->ip_open_count--;
 136                 spin_unlock(&oi->ip_lock);
 137         }
 138
 139 leave:
 140         return status;
 141 }
 142
 143 static int ocfs2_file_release(struct inode *inode, struct file *file)
 144 {
 145         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 146
 147         spin_lock(&oi->ip_lock);
 148         if (!--oi->ip_open_count)
 149                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 150
 151         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 152                                  oi->ip_blkno,
 153                                  file->f_path.dentry->d_name.len,
 154                                  file->f_path.dentry->d_name.name,
 155                                  oi->ip_open_count);
 156         spin_unlock(&oi->ip_lock);
 157
 158         ocfs2_free_file_private(inode, file);
 159
 160         return 0;
 161 }
 162
 163 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 164 {
 165         return ocfs2_init_file_private(inode, file);
 166 }
 167
 168 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 169 {
 170         ocfs2_free_file_private(inode, file);
 171         return 0;
 172 }
 173
 174 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 175                            int datasync)
 176 {
 177         int err = 0;
 178         struct inode *inode = file->f_mapping->host;
 179         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 180         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 181         journal_t *journal = osb->journal->j_journal;
 182         int ret;
 183         tid_t commit_tid;
 184         bool needs_barrier = false;
 185
 186         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 187                               OCFS2_I(inode)->ip_blkno,
 188                               file->f_path.dentry->d_name.len,
 189                               file->f_path.dentry->d_name.name,
 190                               (unsigned long long)datasync);
 191
 192         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 193                 return -EROFS;
 194
 195         err = filemap_write_and_wait_range(inode->i_mapping, start, end);
 196         if (err)
 197                 return err;
 198
 199         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 200         if (journal->j_flags & JBD2_BARRIER &&
 201             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 202                 needs_barrier = true;
 203         err = jbd2_complete_transaction(journal, commit_tid);
 204         if (needs_barrier) {
 205                 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 206                 if (!err)
 207                         err = ret;
 208         }
 209
 210         if (err)
 211                 mlog_errno(err);
 212
 213         return (err < 0) ? -EIO : 0;
 214 }
 215
 216 int ocfs2_should_update_atime(struct inode *inode,
 217                               struct vfsmount *vfsmnt)
 218 {
 219         struct timespec now;
 220         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 221
 222         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 223                 return 0;
 224
 225         if ((inode->i_flags & S_NOATIME) ||
 226             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 227                 return 0;
 228
 229         /*
 230          * We can be called with no vfsmnt structure - NFSD will
 231          * sometimes do this.
 232          *
 233          * Note that our action here is different than touch_atime() -
 234          * if we can't tell whether this is a noatime mount, then we
 235          * don't know whether to trust the value of s_atime_quantum.
 236          */
 237         if (vfsmnt == NULL)
 238                 return 0;
 239
 240         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 241             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 242                 return 0;
 243
 244         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 245                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 246                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 247                         return 1;
 248
 249                 return 0;
 250         }
 251
 252         now = CURRENT_TIME;
 253         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 254                 return 0;
 255         else
 256                 return 1;
 257 }
 258
 259 int ocfs2_update_inode_atime(struct inode *inode,
 260                              struct buffer_head *bh)
 261 {
 262         int ret;
 263         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 264         handle_t *handle;
 265         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 266
 267         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 268         if (IS_ERR(handle)) {
 269                 ret = PTR_ERR(handle);
 270                 mlog_errno(ret);
 271                 goto out;
 272         }
 273
 274         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 275                                       OCFS2_JOURNAL_ACCESS_WRITE);
 276         if (ret) {
 277                 mlog_errno(ret);
 278                 goto out_commit;
 279         }
 280
 281         /*
 282          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 283          * have i_mutex to guard against concurrent changes to other
 284          * inode fields.
 285          */
 286         inode->i_atime = CURRENT_TIME;
 287         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 288         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 289         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 290         ocfs2_journal_dirty(handle, bh);
 291
 292 out_commit:
 293         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 294 out:
 295         return ret;
 296 }
 297
 298 static int ocfs2_set_inode_size(handle_t *handle,
 299                                 struct inode *inode,
 300                                 struct buffer_head *fe_bh,
 301                                 u64 new_i_size)
 302 {
 303         int status;
 304
 305         i_size_write(inode, new_i_size);
 306         inode->i_blocks = ocfs2_inode_sector_count(inode);
 307         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 308
 309         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 310         if (status < 0) {
 311                 mlog_errno(status);
 312                 goto bail;
 313         }
 314
 315 bail:
 316         return status;
 317 }
 318
 319 int ocfs2_simple_size_update(struct inode *inode,
 320                              struct buffer_head *di_bh,
 321                              u64 new_i_size)
 322 {
 323         int ret;
 324         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 325         handle_t *handle = NULL;
 326
 327         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 328         if (IS_ERR(handle)) {
 329                 ret = PTR_ERR(handle);
 330                 mlog_errno(ret);
 331                 goto out;
 332         }
 333
 334         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 335                                    new_i_size);
 336         if (ret < 0)
 337                 mlog_errno(ret);
 338
 339         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 340         ocfs2_commit_trans(osb, handle);
 341 out:
 342         return ret;
 343 }
 344
 345 static int ocfs2_cow_file_pos(struct inode *inode,
 346                               struct buffer_head *fe_bh,
 347                               u64 offset)
 348 {
 349         int status;
 350         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 351         unsigned int num_clusters = 0;
 352         unsigned int ext_flags = 0;
 353
 354         /*
 355          * If the new offset is aligned to the range of the cluster, there is
 356          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 357          * CoW either.
 358          */
 359         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 360                 return 0;
 361
 362         status = ocfs2_get_clusters(inode, cpos, &phys,
 363                                     &num_clusters, &ext_flags);
 364         if (status) {
 365                 mlog_errno(status);
 366                 goto out;
 367         }
 368
 369         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 370                 goto out;
 371
 372         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 373
 374 out:
 375         return status;
 376 }
 377
 378 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 379                                      struct inode *inode,
 380                                      struct buffer_head *fe_bh,
 381                                      u64 new_i_size)
 382 {
 383         int status;
 384         handle_t *handle;
 385         struct ocfs2_dinode *di;
 386         u64 cluster_bytes;
 387
 388         /*
 389          * We need to CoW the cluster contains the offset if it is reflinked
 390          * since we will call ocfs2_zero_range_for_truncate later which will
 391          * write "0" from offset to the end of the cluster.
 392          */
 393         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 394         if (status) {
 395                 mlog_errno(status);
 396                 return status;
 397         }
 398
 399         /* TODO: This needs to actually orphan the inode in this
 400          * transaction. */
 401
 402         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 403         if (IS_ERR(handle)) {
 404                 status = PTR_ERR(handle);
 405                 mlog_errno(status);
 406                 goto out;
 407         }
 408
 409         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 410                                          OCFS2_JOURNAL_ACCESS_WRITE);
 411         if (status < 0) {
 412                 mlog_errno(status);
 413                 goto out_commit;
 414         }
 415
 416         /*
 417          * Do this before setting i_size.
 418          */
 419         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 420         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 421                                                cluster_bytes);
 422         if (status) {
 423                 mlog_errno(status);
 424                 goto out_commit;
 425         }
 426
 427         i_size_write(inode, new_i_size);
 428         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 429
 430         di = (struct ocfs2_dinode *) fe_bh->b_data;
 431         di->i_size = cpu_to_le64(new_i_size);
 432         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 433         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 434         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 435
 436         ocfs2_journal_dirty(handle, fe_bh);
 437
 438 out_commit:
 439         ocfs2_commit_trans(osb, handle);
 440 out:
 441         return status;
 442 }
 443
 444 static int ocfs2_truncate_file(struct inode *inode,
 445                                struct buffer_head *di_bh,
 446                                u64 new_i_size)
 447 {
 448         int status = 0;
 449         struct ocfs2_dinode *fe = NULL;
 450         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 451
 452         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 453          * already validated it */
 454         fe = (struct ocfs2_dinode *) di_bh->b_data;
 455
 456         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 457                                   (unsigned long long)le64_to_cpu(fe->i_size),
 458                                   (unsigned long long)new_i_size);
 459
 460         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 461                         "Inode %llu, inode i_size = %lld != di "
 462                         "i_size = %llu, i_flags = 0x%x\n",
 463                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 464                         i_size_read(inode),
 465                         (unsigned long long)le64_to_cpu(fe->i_size),
 466                         le32_to_cpu(fe->i_flags));
 467
 468         if (new_i_size > le64_to_cpu(fe->i_size)) {
 469                 trace_ocfs2_truncate_file_error(
 470                         (unsigned long long)le64_to_cpu(fe->i_size),
 471                         (unsigned long long)new_i_size);
 472                 status = -EINVAL;
 473                 mlog_errno(status);
 474                 goto bail;
 475         }
 476
 477         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 478
 479         ocfs2_resv_discard(&osb->osb_la_resmap,
 480                            &OCFS2_I(inode)->ip_la_data_resv);
 481
 482         /*
 483          * The inode lock forced other nodes to sync and drop their
 484          * pages, which (correctly) happens even if we have a truncate
 485          * without allocation change - ocfs2 cluster sizes can be much
 486          * greater than page size, so we have to truncate them
 487          * anyway.
 488          */
 489         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 490         truncate_inode_pages(inode->i_mapping, new_i_size);
 491
 492         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 493                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 494                                                i_size_read(inode), 1);
 495                 if (status)
 496                         mlog_errno(status);
 497
 498                 goto bail_unlock_sem;
 499         }
 500
 501         /* alright, we're going to need to do a full blown alloc size
 502          * change. Orphan the inode so that recovery can complete the
 503          * truncate if necessary. This does the task of marking
 504          * i_size. */
 505         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 506         if (status < 0) {
 507                 mlog_errno(status);
 508                 goto bail_unlock_sem;
 509         }
 510
 511         status = ocfs2_commit_truncate(osb, inode, di_bh);
 512         if (status < 0) {
 513                 mlog_errno(status);
 514                 goto bail_unlock_sem;
 515         }
 516
 517         /* TODO: orphan dir cleanup here. */
 518 bail_unlock_sem:
 519         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 520
 521 bail:
 522         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 523                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 524
 525         return status;
 526 }
 527
 528 /*
 529  * extend file allocation only here.
 530  * we'll update all the disk stuff, and oip->alloc_size
 531  *
 532  * expect stuff to be locked, a transaction started and enough data /
 533  * metadata reservations in the contexts.
 534  *
 535  * Will return -EAGAIN, and a reason if a restart is needed.
 536  * If passed in, *reason will always be set, even in error.
 537  */
 538 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 539                          struct inode *inode,
 540                          u32 *logical_offset,
 541                          u32 clusters_to_add,
 542                          int mark_unwritten,
 543                          struct buffer_head *fe_bh,
 544                          handle_t *handle,
 545                          struct ocfs2_alloc_context *data_ac,
 546                          struct ocfs2_alloc_context *meta_ac,
 547                          enum ocfs2_alloc_restarted *reason_ret)
 548 {
 549         int ret;
 550         struct ocfs2_extent_tree et;
 551
 552         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 553         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 554                                           clusters_to_add, mark_unwritten,
 555                                           data_ac, meta_ac, reason_ret);
 556
 557         return ret;
 558 }
 559
 560 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 561                                      u32 clusters_to_add, int mark_unwritten)
 562 {
 563         int status = 0;
 564         int restart_func = 0;
 565         int credits;
 566         u32 prev_clusters;
 567         struct buffer_head *bh = NULL;
 568         struct ocfs2_dinode *fe = NULL;
 569         handle_t *handle = NULL;
 570         struct ocfs2_alloc_context *data_ac = NULL;
 571         struct ocfs2_alloc_context *meta_ac = NULL;
 572         enum ocfs2_alloc_restarted why;
 573         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 574         struct ocfs2_extent_tree et;
 575         int did_quota = 0;
 576
 577         /*
 578          * Unwritten extent only exists for file systems which
 579          * support holes.
 580          */
 581         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 582
 583         status = ocfs2_read_inode_block(inode, &bh);
 584         if (status < 0) {
 585                 mlog_errno(status);
 586                 goto leave;
 587         }
 588         fe = (struct ocfs2_dinode *) bh->b_data;
 589
 590 restart_all:
 591         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 592
 593         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 594         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 595                                        &data_ac, &meta_ac);
 596         if (status) {
 597                 mlog_errno(status);
 598                 goto leave;
 599         }
 600
 601         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 602         handle = ocfs2_start_trans(osb, credits);
 603         if (IS_ERR(handle)) {
 604                 status = PTR_ERR(handle);
 605                 handle = NULL;
 606                 mlog_errno(status);
 607                 goto leave;
 608         }
 609
 610 restarted_transaction:
 611         trace_ocfs2_extend_allocation(
 612                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 613                 (unsigned long long)i_size_read(inode),
 614                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 615                 why, restart_func);
 616
 617         status = dquot_alloc_space_nodirty(inode,
 618                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 619         if (status)
 620                 goto leave;
 621         did_quota = 1;
 622
 623         /* reserve a write to the file entry early on - that we if we
 624          * run out of credits in the allocation path, we can still
 625          * update i_size. */
 626         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 627                                          OCFS2_JOURNAL_ACCESS_WRITE);
 628         if (status < 0) {
 629                 mlog_errno(status);
 630                 goto leave;
 631         }
 632
 633         prev_clusters = OCFS2_I(inode)->ip_clusters;
 634
 635         status = ocfs2_add_inode_data(osb,
 636                                       inode,
 637                                       &logical_start,
 638                                       clusters_to_add,
 639                                       mark_unwritten,
 640                                       bh,
 641                                       handle,
 642                                       data_ac,
 643                                       meta_ac,
 644                                       &why);
 645         if ((status < 0) && (status != -EAGAIN)) {
 646                 if (status != -ENOSPC)
 647                         mlog_errno(status);
 648                 goto leave;
 649         }
 650         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 651         ocfs2_journal_dirty(handle, bh);
 652
 653         spin_lock(&OCFS2_I(inode)->ip_lock);
 654         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 655         spin_unlock(&OCFS2_I(inode)->ip_lock);
 656         /* Release unused quota reservation */
 657         dquot_free_space(inode,
 658                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 659         did_quota = 0;
 660
 661         if (why != RESTART_NONE && clusters_to_add) {
 662                 if (why == RESTART_META) {
 663                         restart_func = 1;
 664                         status = 0;
 665                 } else {
 666                         BUG_ON(why != RESTART_TRANS);
 667
 668                         status = ocfs2_allocate_extend_trans(handle, 1);
 669                         if (status < 0) {
 670                                 /* handle still has to be committed at
 671                                  * this point. */
 672                                 status = -ENOMEM;
 673                                 mlog_errno(status);
 674                                 goto leave;
 675                         }
 676                         goto restarted_transaction;
 677                 }
 678         }
 679
 680         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 681              le32_to_cpu(fe->i_clusters),
 682              (unsigned long long)le64_to_cpu(fe->i_size),
 683              OCFS2_I(inode)->ip_clusters,
 684              (unsigned long long)i_size_read(inode));
 685
 686 leave:
 687         if (status < 0 && did_quota)
 688                 dquot_free_space(inode,
 689                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 690         if (handle) {
 691                 ocfs2_commit_trans(osb, handle);
 692                 handle = NULL;
 693         }
 694         if (data_ac) {
 695                 ocfs2_free_alloc_context(data_ac);
 696                 data_ac = NULL;
 697         }
 698         if (meta_ac) {
 699                 ocfs2_free_alloc_context(meta_ac);
 700                 meta_ac = NULL;
 701         }
 702         if ((!status) && restart_func) {
 703                 restart_func = 0;
 704                 goto restart_all;
 705         }
 706         brelse(bh);
 707         bh = NULL;
 708
 709         return status;
 710 }
 711
 712 /*
 713  * While a write will already be ordering the data, a truncate will not.
 714  * Thus, we need to explicitly order the zeroed pages.
 715  */
 716 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 717                                                 struct buffer_head *di_bh)
 718 {
 719         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 720         handle_t *handle = NULL;
 721         int ret = 0;
 722
 723         if (!ocfs2_should_order_data(inode))
 724                 goto out;
 725
 726         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 727         if (IS_ERR(handle)) {
 728                 ret = -ENOMEM;
 729                 mlog_errno(ret);
 730                 goto out;
 731         }
 732
 733         ret = ocfs2_jbd2_file_inode(handle, inode);
 734         if (ret < 0) {
 735                 mlog_errno(ret);
 736                 goto out;
 737         }
 738
 739         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 740                                       OCFS2_JOURNAL_ACCESS_WRITE);
 741         if (ret)
 742                 mlog_errno(ret);
 743         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 744
 745 out:
 746         if (ret) {
 747                 if (!IS_ERR(handle))
 748                         ocfs2_commit_trans(osb, handle);
 749                 handle = ERR_PTR(ret);
 750         }
 751         return handle;
 752 }
 753
 754 /* Some parts of this taken from generic_cont_expand, which turned out
 755  * to be too fragile to do exactly what we need without us having to
 756  * worry about recursive locking in ->write_begin() and ->write_end(). */
 757 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 758                                  u64 abs_to, struct buffer_head *di_bh)
 759 {
 760         struct address_space *mapping = inode->i_mapping;
 761         struct page *page;
 762         unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
 763         handle_t *handle = NULL;
 764         int ret = 0;
 765         unsigned zero_from, zero_to, block_start, block_end;
 766         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 767
 768         BUG_ON(abs_from >= abs_to);
 769         BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
 770         BUG_ON(abs_from & (inode->i_blkbits - 1));
 771
 772         page = find_or_create_page(mapping, index, GFP_NOFS);
 773         if (!page) {
 774                 ret = -ENOMEM;
 775                 mlog_errno(ret);
 776                 goto out;
 777         }
 778
 779         /* Get the offsets within the page that we want to zero */
 780         zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
 781         zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
 782         if (!zero_to)
 783                 zero_to = PAGE_CACHE_SIZE;
 784
 785         trace_ocfs2_write_zero_page(
 786                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 787                         (unsigned long long)abs_from,
 788                         (unsigned long long)abs_to,
 789                         index, zero_from, zero_to);
 790
 791         /* We know that zero_from is block aligned */
 792         for (block_start = zero_from; block_start < zero_to;
 793              block_start = block_end) {
 794                 block_end = block_start + (1 << inode->i_blkbits);
 795
 796                 /*
 797                  * block_start is block-aligned.  Bump it by one to force
 798                  * __block_write_begin and block_commit_write to zero the
 799                  * whole block.
 800                  */
 801                 ret = __block_write_begin(page, block_start + 1, 0,
 802                                           ocfs2_get_block);
 803                 if (ret < 0) {
 804                         mlog_errno(ret);
 805                         goto out_unlock;
 806                 }
 807
 808                 if (!handle) {
 809                         handle = ocfs2_zero_start_ordered_transaction(inode,
 810                                                                       di_bh);
 811                         if (IS_ERR(handle)) {
 812                                 ret = PTR_ERR(handle);
 813                                 handle = NULL;
 814                                 break;
 815                         }
 816                 }
 817
 818                 /* must not update i_size! */
 819                 ret = block_commit_write(page, block_start + 1,
 820                                          block_start + 1);
 821                 if (ret < 0)
 822                         mlog_errno(ret);
 823                 else
 824                         ret = 0;
 825         }
 826
 827         if (handle) {
 828                 /*
 829                  * fs-writeback will release the dirty pages without page lock
 830                  * whose offset are over inode size, the release happens at
 831                  * block_write_full_page().
 832                  */
 833                 i_size_write(inode, abs_to);
 834                 inode->i_blocks = ocfs2_inode_sector_count(inode);
 835                 di->i_size = cpu_to_le64((u64)i_size_read(inode));
 836                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 837                 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 838                 di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 839                 di->i_mtime_nsec = di->i_ctime_nsec;
 840                 ocfs2_journal_dirty(handle, di_bh);
 841                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 842                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 843         }
 844
 845 out_unlock:
 846         unlock_page(page);
 847         page_cache_release(page);
 848 out:
 849         return ret;
 850 }
 851
 852 /*
 853  * Find the next range to zero.  We do this in terms of bytes because
 854  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 855  * pagecache.  We may return multiple extents.
 856  *
 857  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 858  * needs to be zeroed.  range_start and range_end return the next zeroing
 859  * range.  A subsequent call should pass the previous range_end as its
 860  * zero_start.  If range_end is 0, there's nothing to do.
 861  *
 862  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 863  */
 864 static int ocfs2_zero_extend_get_range(struct inode *inode,
 865                                        struct buffer_head *di_bh,
 866                                        u64 zero_start, u64 zero_end,
 867                                        u64 *range_start, u64 *range_end)
 868 {
 869         int rc = 0, needs_cow = 0;
 870         u32 p_cpos, zero_clusters = 0;
 871         u32 zero_cpos =
 872                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 873         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 874         unsigned int num_clusters = 0;
 875         unsigned int ext_flags = 0;
 876
 877         while (zero_cpos < last_cpos) {
 878                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 879                                         &num_clusters, &ext_flags);
 880                 if (rc) {
 881                         mlog_errno(rc);
 882                         goto out;
 883                 }
 884
 885                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 886                         zero_clusters = num_clusters;
 887                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 888                                 needs_cow = 1;
 889                         break;
 890                 }
 891
 892                 zero_cpos += num_clusters;
 893         }
 894         if (!zero_clusters) {
 895                 *range_end = 0;
 896                 goto out;
 897         }
 898
 899         while ((zero_cpos + zero_clusters) < last_cpos) {
 900                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 901                                         &p_cpos, &num_clusters,
 902                                         &ext_flags);
 903                 if (rc) {
 904                         mlog_errno(rc);
 905                         goto out;
 906                 }
 907
 908                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 909                         break;
 910                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 911                         needs_cow = 1;
 912                 zero_clusters += num_clusters;
 913         }
 914         if ((zero_cpos + zero_clusters) > last_cpos)
 915                 zero_clusters = last_cpos - zero_cpos;
 916
 917         if (needs_cow) {
 918                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 919                                         zero_clusters, UINT_MAX);
 920                 if (rc) {
 921                         mlog_errno(rc);
 922                         goto out;
 923                 }
 924         }
 925
 926         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 927         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 928                                              zero_cpos + zero_clusters);
 929
 930 out:
 931         return rc;
 932 }
 933
 934 /*
 935  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 936  * has made sure that the entire range needs zeroing.
 937  */
 938 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 939                                    u64 range_end, struct buffer_head *di_bh)
 940 {
 941         int rc = 0;
 942         u64 next_pos;
 943         u64 zero_pos = range_start;
 944
 945         trace_ocfs2_zero_extend_range(
 946                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 947                         (unsigned long long)range_start,
 948                         (unsigned long long)range_end);
 949         BUG_ON(range_start >= range_end);
 950
 951         while (zero_pos < range_end) {
 952                 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
 953                 if (next_pos > range_end)
 954                         next_pos = range_end;
 955                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 956                 if (rc < 0) {
 957                         mlog_errno(rc);
 958                         break;
 959                 }
 960                 zero_pos = next_pos;
 961
 962                 /*
 963                  * Very large extends have the potential to lock up
 964                  * the cpu for extended periods of time.
 965                  */
 966                 cond_resched();
 967         }
 968
 969         return rc;
 970 }
 971
 972 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 973                       loff_t zero_to_size)
 974 {
 975         int ret = 0;
 976         u64 zero_start, range_start = 0, range_end = 0;
 977         struct super_block *sb = inode->i_sb;
 978
 979         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 980         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 981                                 (unsigned long long)zero_start,
 982                                 (unsigned long long)i_size_read(inode));
 983         while (zero_start < zero_to_size) {
 984                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 985                                                   zero_to_size,
 986                                                   &range_start,
 987                                                   &range_end);
 988                 if (ret) {
 989                         mlog_errno(ret);
 990                         break;
 991                 }
 992                 if (!range_end)
 993                         break;
 994                 /* Trim the ends */
 995                 if (range_start < zero_start)
 996                         range_start = zero_start;
 997                 if (range_end > zero_to_size)
 998                         range_end = zero_to_size;
 999
1000                 ret = ocfs2_zero_extend_range(inode, range_start,
1001                                               range_end, di_bh);
1002                 if (ret) {
1003                         mlog_errno(ret);
1004                         break;
1005                 }
1006                 zero_start = range_end;
1007         }
1008
1009         return ret;
1010 }
1011
1012 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1013                           u64 new_i_size, u64 zero_to)
1014 {
1015         int ret;
1016         u32 clusters_to_add;
1017         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1018
1019         /*
1020          * Only quota files call this without a bh, and they can't be
1021          * refcounted.
1022          */
1023         BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1024         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1025
1026         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1027         if (clusters_to_add < oi->ip_clusters)
1028                 clusters_to_add = 0;
1029         else
1030                 clusters_to_add -= oi->ip_clusters;
1031
1032         if (clusters_to_add) {
1033                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1034                                                 clusters_to_add, 0);
1035                 if (ret) {
1036                         mlog_errno(ret);
1037                         goto out;
1038                 }
1039         }
1040
1041         /*
1042          * Call this even if we don't add any clusters to the tree. We
1043          * still need to zero the area between the old i_size and the
1044          * new i_size.
1045          */
1046         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1047         if (ret < 0)
1048                 mlog_errno(ret);
1049
1050 out:
1051         return ret;
1052 }
1053
1054 static int ocfs2_extend_file(struct inode *inode,
1055                              struct buffer_head *di_bh,
1056                              u64 new_i_size)
1057 {
1058         int ret = 0;
1059         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1060
1061         BUG_ON(!di_bh);
1062
1063         /* setattr sometimes calls us like this. */
1064         if (new_i_size == 0)
1065                 goto out;
1066
1067         if (i_size_read(inode) == new_i_size)
1068                 goto out;
1069         BUG_ON(new_i_size < i_size_read(inode));
1070
1071         /*
1072          * The alloc sem blocks people in read/write from reading our
1073          * allocation until we're done changing it. We depend on
1074          * i_mutex to block other extend/truncate calls while we're
1075          * here.  We even have to hold it for sparse files because there
1076          * might be some tail zeroing.
1077          */
1078         down_write(&oi->ip_alloc_sem);
1079
1080         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1081                 /*
1082                  * We can optimize small extends by keeping the inodes
1083                  * inline data.
1084                  */
1085                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1086                         up_write(&oi->ip_alloc_sem);
1087                         goto out_update_size;
1088                 }
1089
1090                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1091                 if (ret) {
1092                         up_write(&oi->ip_alloc_sem);
1093                         mlog_errno(ret);
1094                         goto out;
1095                 }
1096         }
1097
1098         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1099                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1100         else
1101                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1102                                             new_i_size);
1103
1104         up_write(&oi->ip_alloc_sem);
1105
1106         if (ret < 0) {
1107                 mlog_errno(ret);
1108                 goto out;
1109         }
1110
1111 out_update_size:
1112         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1113         if (ret < 0)
1114                 mlog_errno(ret);
1115
1116 out:
1117         return ret;
1118 }
1119
1120 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1121 {
1122         int status = 0, size_change;
1123         struct inode *inode = dentry->d_inode;
1124         struct super_block *sb = inode->i_sb;
1125         struct ocfs2_super *osb = OCFS2_SB(sb);
1126         struct buffer_head *bh = NULL;
1127         handle_t *handle = NULL;
1128         struct dquot *transfer_to[MAXQUOTAS] = { };
1129         int qtype;
1130
1131         trace_ocfs2_setattr(inode, dentry,
1132                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1133                             dentry->d_name.len, dentry->d_name.name,
1134                             attr->ia_valid, attr->ia_mode,
1135                             from_kuid(&init_user_ns, attr->ia_uid),
1136                             from_kgid(&init_user_ns, attr->ia_gid));
1137
1138         /* ensuring we don't even attempt to truncate a symlink */
1139         if (S_ISLNK(inode->i_mode))
1140                 attr->ia_valid &= ~ATTR_SIZE;
1141
1142 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1143                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1144         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1145                 return 0;
1146
1147         status = inode_change_ok(inode, attr);
1148         if (status)
1149                 return status;
1150
1151         if (is_quota_modification(inode, attr))
1152                 dquot_initialize(inode);
1153         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1154         if (size_change) {
1155                 status = ocfs2_rw_lock(inode, 1);
1156                 if (status < 0) {
1157                         mlog_errno(status);
1158                         goto bail;
1159                 }
1160         }
1161
1162         status = ocfs2_inode_lock(inode, &bh, 1);
1163         if (status < 0) {
1164                 if (status != -ENOENT)
1165                         mlog_errno(status);
1166                 goto bail_unlock_rw;
1167         }
1168
1169         if (size_change) {
1170                 status = inode_newsize_ok(inode, attr->ia_size);
1171                 if (status)
1172                         goto bail_unlock;
1173
1174                 inode_dio_wait(inode);
1175
1176                 if (i_size_read(inode) >= attr->ia_size) {
1177                         if (ocfs2_should_order_data(inode)) {
1178                                 status = ocfs2_begin_ordered_truncate(inode,
1179                                                                       attr->ia_size);
1180                                 if (status)
1181                                         goto bail_unlock;
1182                         }
1183                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1184                 } else
1185                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1186                 if (status < 0) {
1187                         if (status != -ENOSPC)
1188                                 mlog_errno(status);
1189                         status = -ENOSPC;
1190                         goto bail_unlock;
1191                 }
1192         }
1193
1194         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1195             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1196                 /*
1197                  * Gather pointers to quota structures so that allocation /
1198                  * freeing of quota structures happens here and not inside
1199                  * dquot_transfer() where we have problems with lock ordering
1200                  */
1201                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1202                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1203                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1204                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1205                         if (!transfer_to[USRQUOTA]) {
1206                                 status = -ESRCH;
1207                                 goto bail_unlock;
1208                         }
1209                 }
1210                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1211                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1212                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1213                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1214                         if (!transfer_to[GRPQUOTA]) {
1215                                 status = -ESRCH;
1216                                 goto bail_unlock;
1217                         }
1218                 }
1219                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1220                                            2 * ocfs2_quota_trans_credits(sb));
1221                 if (IS_ERR(handle)) {
1222                         status = PTR_ERR(handle);
1223                         mlog_errno(status);
1224                         goto bail_unlock;
1225                 }
1226                 status = __dquot_transfer(inode, transfer_to);
1227                 if (status < 0)
1228                         goto bail_commit;
1229         } else {
1230                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1231                 if (IS_ERR(handle)) {
1232                         status = PTR_ERR(handle);
1233                         mlog_errno(status);
1234                         goto bail_unlock;
1235                 }
1236         }
1237
1238         setattr_copy(inode, attr);
1239         mark_inode_dirty(inode);
1240
1241         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1242         if (status < 0)
1243                 mlog_errno(status);
1244
1245 bail_commit:
1246         ocfs2_commit_trans(osb, handle);
1247 bail_unlock:
1248         ocfs2_inode_unlock(inode, 1);
1249 bail_unlock_rw:
1250         if (size_change)
1251                 ocfs2_rw_unlock(inode, 1);
1252 bail:
1253         brelse(bh);
1254
1255         /* Release quota pointers in case we acquired them */
1256         for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1257                 dqput(transfer_to[qtype]);
1258
1259         if (!status && attr->ia_valid & ATTR_MODE) {
1260                 status = posix_acl_chmod(inode, inode->i_mode);
1261                 if (status < 0)
1262                         mlog_errno(status);
1263         }
1264
1265         return status;
1266 }
1267
1268 int ocfs2_getattr(struct vfsmount *mnt,
1269                   struct dentry *dentry,
1270                   struct kstat *stat)
1271 {
1272         struct inode *inode = dentry->d_inode;
1273         struct super_block *sb = dentry->d_inode->i_sb;
1274         struct ocfs2_super *osb = sb->s_fs_info;
1275         int err;
1276
1277         err = ocfs2_inode_revalidate(dentry);
1278         if (err) {
1279                 if (err != -ENOENT)
1280                         mlog_errno(err);
1281                 goto bail;
1282         }
1283
1284         generic_fillattr(inode, stat);
1285
1286         /* We set the blksize from the cluster size for performance */
1287         stat->blksize = osb->s_clustersize;
1288
1289 bail:
1290         return err;
1291 }
1292
1293 int ocfs2_permission(struct inode *inode, int mask)
1294 {
1295         int ret;
1296
1297         if (mask & MAY_NOT_BLOCK)
1298                 return -ECHILD;
1299
1300         ret = ocfs2_inode_lock(inode, NULL, 0);
1301         if (ret) {
1302                 if (ret != -ENOENT)
1303                         mlog_errno(ret);
1304                 goto out;
1305         }
1306
1307         ret = generic_permission(inode, mask);
1308
1309         ocfs2_inode_unlock(inode, 0);
1310 out:
1311         return ret;
1312 }
1313
1314 static int __ocfs2_write_remove_suid(struct inode *inode,
1315                                      struct buffer_head *bh)
1316 {
1317         int ret;
1318         handle_t *handle;
1319         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1320         struct ocfs2_dinode *di;
1321
1322         trace_ocfs2_write_remove_suid(
1323                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1324                         inode->i_mode);
1325
1326         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1327         if (IS_ERR(handle)) {
1328                 ret = PTR_ERR(handle);
1329                 mlog_errno(ret);
1330                 goto out;
1331         }
1332
1333         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1334                                       OCFS2_JOURNAL_ACCESS_WRITE);
1335         if (ret < 0) {
1336                 mlog_errno(ret);
1337                 goto out_trans;
1338         }
1339
1340         inode->i_mode &= ~S_ISUID;
1341         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1342                 inode->i_mode &= ~S_ISGID;
1343
1344         di = (struct ocfs2_dinode *) bh->b_data;
1345         di->i_mode = cpu_to_le16(inode->i_mode);
1346         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1347
1348         ocfs2_journal_dirty(handle, bh);
1349
1350 out_trans:
1351         ocfs2_commit_trans(osb, handle);
1352 out:
1353         return ret;
1354 }
1355
1356 /*
1357  * Will look for holes and unwritten extents in the range starting at
1358  * pos for count bytes (inclusive).
1359  */
1360 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1361                                        size_t count)
1362 {
1363         int ret = 0;
1364         unsigned int extent_flags;
1365         u32 cpos, clusters, extent_len, phys_cpos;
1366         struct super_block *sb = inode->i_sb;
1367
1368         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1369         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1370
1371         while (clusters) {
1372                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1373                                          &extent_flags);
1374                 if (ret < 0) {
1375                         mlog_errno(ret);
1376                         goto out;
1377                 }
1378
1379                 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1380                         ret = 1;
1381                         break;
1382                 }
1383
1384                 if (extent_len > clusters)
1385                         extent_len = clusters;
1386
1387                 clusters -= extent_len;
1388                 cpos += extent_len;
1389         }
1390 out:
1391         return ret;
1392 }
1393
1394 static int ocfs2_write_remove_suid(struct inode *inode)
1395 {
1396         int ret;
1397         struct buffer_head *bh = NULL;
1398
1399         ret = ocfs2_read_inode_block(inode, &bh);
1400         if (ret < 0) {
1401                 mlog_errno(ret);
1402                 goto out;
1403         }
1404
1405         ret =  __ocfs2_write_remove_suid(inode, bh);
1406 out:
1407         brelse(bh);
1408         return ret;
1409 }
1410
1411 /*
1412  * Allocate enough extents to cover the region starting at byte offset
1413  * start for len bytes. Existing extents are skipped, any extents
1414  * added are marked as "unwritten".
1415  */
1416 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1417                                             u64 start, u64 len)
1418 {
1419         int ret;
1420         u32 cpos, phys_cpos, clusters, alloc_size;
1421         u64 end = start + len;
1422         struct buffer_head *di_bh = NULL;
1423
1424         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1425                 ret = ocfs2_read_inode_block(inode, &di_bh);
1426                 if (ret) {
1427                         mlog_errno(ret);
1428                         goto out;
1429                 }
1430
1431                 /*
1432                  * Nothing to do if the requested reservation range
1433                  * fits within the inode.
1434                  */
1435                 if (ocfs2_size_fits_inline_data(di_bh, end))
1436                         goto out;
1437
1438                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1439                 if (ret) {
1440                         mlog_errno(ret);
1441                         goto out;
1442                 }
1443         }
1444
1445         /*
1446          * We consider both start and len to be inclusive.
1447          */
1448         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1449         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1450         clusters -= cpos;
1451
1452         while (clusters) {
1453                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1454                                          &alloc_size, NULL);
1455                 if (ret) {
1456                         mlog_errno(ret);
1457                         goto out;
1458                 }
1459
1460                 /*
1461                  * Hole or existing extent len can be arbitrary, so
1462                  * cap it to our own allocation request.
1463                  */
1464                 if (alloc_size > clusters)
1465                         alloc_size = clusters;
1466
1467                 if (phys_cpos) {
1468                         /*
1469                          * We already have an allocation at this
1470                          * region so we can safely skip it.
1471                          */
1472                         goto next;
1473                 }
1474
1475                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1476                 if (ret) {
1477                         if (ret != -ENOSPC)
1478                                 mlog_errno(ret);
1479                         goto out;
1480                 }
1481
1482 next:
1483                 cpos += alloc_size;
1484                 clusters -= alloc_size;
1485         }
1486
1487         ret = 0;
1488 out:
1489
1490         brelse(di_bh);
1491         return ret;
1492 }
1493
1494 /*
1495  * Truncate a byte range, avoiding pages within partial clusters. This
1496  * preserves those pages for the zeroing code to write to.
1497  */
1498 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1499                                          u64 byte_len)
1500 {
1501         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1502         loff_t start, end;
1503         struct address_space *mapping = inode->i_mapping;
1504
1505         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1506         end = byte_start + byte_len;
1507         end = end & ~(osb->s_clustersize - 1);
1508
1509         if (start < end) {
1510                 unmap_mapping_range(mapping, start, end - start, 0);
1511                 truncate_inode_pages_range(mapping, start, end - 1);
1512         }
1513 }
1514
1515 static int ocfs2_zero_partial_clusters(struct inode *inode,
1516                                        u64 start, u64 len)
1517 {
1518         int ret = 0;
1519         u64 tmpend, end = start + len;
1520         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1521         unsigned int csize = osb->s_clustersize;
1522         handle_t *handle;
1523
1524         /*
1525          * The "start" and "end" values are NOT necessarily part of
1526          * the range whose allocation is being deleted. Rather, this
1527          * is what the user passed in with the request. We must zero
1528          * partial clusters here. There's no need to worry about
1529          * physical allocation - the zeroing code knows to skip holes.
1530          */
1531         trace_ocfs2_zero_partial_clusters(
1532                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1533                 (unsigned long long)start, (unsigned long long)end);
1534
1535         /*
1536          * If both edges are on a cluster boundary then there's no
1537          * zeroing required as the region is part of the allocation to
1538          * be truncated.
1539          */
1540         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1541                 goto out;
1542
1543         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1544         if (IS_ERR(handle)) {
1545                 ret = PTR_ERR(handle);
1546                 mlog_errno(ret);
1547                 goto out;
1548         }
1549
1550         /*
1551          * We want to get the byte offset of the end of the 1st cluster.
1552          */
1553         tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1554         if (tmpend > end)
1555                 tmpend = end;
1556
1557         trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1558                                                  (unsigned long long)tmpend);
1559
1560         ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1561         if (ret)
1562                 mlog_errno(ret);
1563
1564         if (tmpend < end) {
1565                 /*
1566                  * This may make start and end equal, but the zeroing
1567                  * code will skip any work in that case so there's no
1568                  * need to catch it up here.
1569                  */
1570                 start = end & ~(osb->s_clustersize - 1);
1571
1572                 trace_ocfs2_zero_partial_clusters_range2(
1573                         (unsigned long long)start, (unsigned long long)end);
1574
1575                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1576                 if (ret)
1577                         mlog_errno(ret);
1578         }
1579         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1580
1581         ocfs2_commit_trans(osb, handle);
1582 out:
1583         return ret;
1584 }
1585
1586 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1587 {
1588         int i;
1589         struct ocfs2_extent_rec *rec = NULL;
1590
1591         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1592
1593                 rec = &el->l_recs[i];
1594
1595                 if (le32_to_cpu(rec->e_cpos) < pos)
1596                         break;
1597         }
1598
1599         return i;
1600 }
1601
1602 /*
1603  * Helper to calculate the punching pos and length in one run, we handle the
1604  * following three cases in order:
1605  *
1606  * - remove the entire record
1607  * - remove a partial record
1608  * - no record needs to be removed (hole-punching completed)
1609 */
1610 static void ocfs2_calc_trunc_pos(struct inode *inode,
1611                                  struct ocfs2_extent_list *el,
1612                                  struct ocfs2_extent_rec *rec,
1613                                  u32 trunc_start, u32 *trunc_cpos,
1614                                  u32 *trunc_len, u32 *trunc_end,
1615                                  u64 *blkno, int *done)
1616 {
1617         int ret = 0;
1618         u32 coff, range;
1619
1620         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1621
1622         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1623                 /*
1624                  * remove an entire extent record.
1625                  */
1626                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1627                 /*
1628                  * Skip holes if any.
1629                  */
1630                 if (range < *trunc_end)
1631                         *trunc_end = range;
1632                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1633                 *blkno = le64_to_cpu(rec->e_blkno);
1634                 *trunc_end = le32_to_cpu(rec->e_cpos);
1635         } else if (range > trunc_start) {
1636                 /*
1637                  * remove a partial extent record, which means we're
1638                  * removing the last extent record.
1639                  */
1640                 *trunc_cpos = trunc_start;
1641                 /*
1642                  * skip hole if any.
1643                  */
1644                 if (range < *trunc_end)
1645                         *trunc_end = range;
1646                 *trunc_len = *trunc_end - trunc_start;
1647                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1648                 *blkno = le64_to_cpu(rec->e_blkno) +
1649                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1650                 *trunc_end = trunc_start;
1651         } else {
1652                 /*
1653                  * It may have two following possibilities:
1654                  *
1655                  * - last record has been removed
1656                  * - trunc_start was within a hole
1657                  *
1658                  * both two cases mean the completion of hole punching.
1659                  */
1660                 ret = 1;
1661         }
1662
1663         *done = ret;
1664 }
1665
1666 static int ocfs2_remove_inode_range(struct inode *inode,
1667                                     struct buffer_head *di_bh, u64 byte_start,
1668                                     u64 byte_len)
1669 {
1670         int ret = 0, flags = 0, done = 0, i;
1671         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1672         u32 cluster_in_el;
1673         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1674         struct ocfs2_cached_dealloc_ctxt dealloc;
1675         struct address_space *mapping = inode->i_mapping;
1676         struct ocfs2_extent_tree et;
1677         struct ocfs2_path *path = NULL;
1678         struct ocfs2_extent_list *el = NULL;
1679         struct ocfs2_extent_rec *rec = NULL;
1680         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1681         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1682
1683         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1684         ocfs2_init_dealloc_ctxt(&dealloc);
1685
1686         trace_ocfs2_remove_inode_range(
1687                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1688                         (unsigned long long)byte_start,
1689                         (unsigned long long)byte_len);
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         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1738         if (ret) {
1739                 mlog_errno(ret);
1740                 goto out;
1741         }
1742
1743         path = ocfs2_new_path_from_et(&et);
1744         if (!path) {
1745                 ret = -ENOMEM;
1746                 mlog_errno(ret);
1747                 goto out;
1748         }
1749
1750         while (trunc_end > trunc_start) {
1751
1752                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1753                                       cluster_in_el);
1754                 if (ret) {
1755                         mlog_errno(ret);
1756                         goto out;
1757                 }
1758
1759                 el = path_leaf_el(path);
1760
1761                 i = ocfs2_find_rec(el, trunc_end);
1762                 /*
1763                  * Need to go to previous extent block.
1764                  */
1765                 if (i < 0) {
1766                         if (path->p_tree_depth == 0)
1767                                 break;
1768
1769                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1770                                                             path,
1771                                                             &cluster_in_el);
1772                         if (ret) {
1773                                 mlog_errno(ret);
1774                                 goto out;
1775                         }
1776
1777                         /*
1778                          * We've reached the leftmost extent block,
1779                          * it's safe to leave.
1780                          */
1781                         if (cluster_in_el == 0)
1782                                 break;
1783
1784                         /*
1785                          * The 'pos' searched for previous extent block is
1786                          * always one cluster less than actual trunc_end.
1787                          */
1788                         trunc_end = cluster_in_el + 1;
1789
1790                         ocfs2_reinit_path(path, 1);
1791
1792                         continue;
1793
1794                 } else
1795                         rec = &el->l_recs[i];
1796
1797                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1798                                      &trunc_len, &trunc_end, &blkno, &done);
1799                 if (done)
1800                         break;
1801
1802                 flags = rec->e_flags;
1803                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1804
1805                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1806                                                phys_cpos, trunc_len, flags,
1807                                                &dealloc, refcount_loc);
1808                 if (ret < 0) {
1809                         mlog_errno(ret);
1810                         goto out;
1811                 }
1812
1813                 cluster_in_el = trunc_end;
1814
1815                 ocfs2_reinit_path(path, 1);
1816         }
1817
1818         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1819
1820 out:
1821         ocfs2_free_path(path);
1822         ocfs2_schedule_truncate_log_flush(osb, 1);
1823         ocfs2_run_deallocs(osb, &dealloc);
1824
1825         return ret;
1826 }
1827
1828 /*
1829  * Parts of this function taken from xfs_change_file_space()
1830  */
1831 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1832                                      loff_t f_pos, unsigned int cmd,
1833                                      struct ocfs2_space_resv *sr,
1834                                      int change_size)
1835 {
1836         int ret;
1837         s64 llen;
1838         loff_t size;
1839         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1840         struct buffer_head *di_bh = NULL;
1841         handle_t *handle;
1842         unsigned long long max_off = inode->i_sb->s_maxbytes;
1843
1844         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1845                 return -EROFS;
1846
1847         mutex_lock(&inode->i_mutex);
1848
1849         /*
1850          * This prevents concurrent writes on other nodes
1851          */
1852         ret = ocfs2_rw_lock(inode, 1);
1853         if (ret) {
1854                 mlog_errno(ret);
1855                 goto out;
1856         }
1857
1858         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1859         if (ret) {
1860                 mlog_errno(ret);
1861                 goto out_rw_unlock;
1862         }
1863
1864         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1865                 ret = -EPERM;
1866                 goto out_inode_unlock;
1867         }
1868
1869         switch (sr->l_whence) {
1870         case 0: /*SEEK_SET*/
1871                 break;
1872         case 1: /*SEEK_CUR*/
1873                 sr->l_start += f_pos;
1874                 break;
1875         case 2: /*SEEK_END*/
1876                 sr->l_start += i_size_read(inode);
1877                 break;
1878         default:
1879                 ret = -EINVAL;
1880                 goto out_inode_unlock;
1881         }
1882         sr->l_whence = 0;
1883
1884         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1885
1886         if (sr->l_start < 0
1887             || sr->l_start > max_off
1888             || (sr->l_start + llen) < 0
1889             || (sr->l_start + llen) > max_off) {
1890                 ret = -EINVAL;
1891                 goto out_inode_unlock;
1892         }
1893         size = sr->l_start + sr->l_len;
1894
1895         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1896             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1897                 if (sr->l_len <= 0) {
1898                         ret = -EINVAL;
1899                         goto out_inode_unlock;
1900                 }
1901         }
1902
1903         if (file && should_remove_suid(file->f_path.dentry)) {
1904                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1905                 if (ret) {
1906                         mlog_errno(ret);
1907                         goto out_inode_unlock;
1908                 }
1909         }
1910
1911         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1912         switch (cmd) {
1913         case OCFS2_IOC_RESVSP:
1914         case OCFS2_IOC_RESVSP64:
1915                 /*
1916                  * This takes unsigned offsets, but the signed ones we
1917                  * pass have been checked against overflow above.
1918                  */
1919                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1920                                                        sr->l_len);
1921                 break;
1922         case OCFS2_IOC_UNRESVSP:
1923         case OCFS2_IOC_UNRESVSP64:
1924                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1925                                                sr->l_len);
1926                 break;
1927         default:
1928                 ret = -EINVAL;
1929         }
1930         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1931         if (ret) {
1932                 mlog_errno(ret);
1933                 goto out_inode_unlock;
1934         }
1935
1936         /*
1937          * We update c/mtime for these changes
1938          */
1939         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1940         if (IS_ERR(handle)) {
1941                 ret = PTR_ERR(handle);
1942                 mlog_errno(ret);
1943                 goto out_inode_unlock;
1944         }
1945
1946         if (change_size && i_size_read(inode) < size)
1947                 i_size_write(inode, size);
1948
1949         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1950         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1951         if (ret < 0)
1952                 mlog_errno(ret);
1953
1954         if (file && (file->f_flags & O_SYNC))
1955                 handle->h_sync = 1;
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_inode(file);
1974         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1975         int ret;
1976
1977         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1978             !ocfs2_writes_unwritten_extents(osb))
1979                 return -ENOTTY;
1980         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1981                  !ocfs2_sparse_alloc(osb))
1982                 return -ENOTTY;
1983
1984         if (!S_ISREG(inode->i_mode))
1985                 return -EINVAL;
1986
1987         if (!(file->f_mode & FMODE_WRITE))
1988                 return -EBADF;
1989
1990         ret = mnt_want_write_file(file);
1991         if (ret)
1992                 return ret;
1993         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1994         mnt_drop_write_file(file);
1995         return ret;
1996 }
1997
1998 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1999                             loff_t len)
2000 {
2001         struct inode *inode = file_inode(file);
2002         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2003         struct ocfs2_space_resv sr;
2004         int change_size = 1;
2005         int cmd = OCFS2_IOC_RESVSP64;
2006
2007         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2008                 return -EOPNOTSUPP;
2009         if (!ocfs2_writes_unwritten_extents(osb))
2010                 return -EOPNOTSUPP;
2011
2012         if (mode & FALLOC_FL_KEEP_SIZE)
2013                 change_size = 0;
2014
2015         if (mode & FALLOC_FL_PUNCH_HOLE)
2016                 cmd = OCFS2_IOC_UNRESVSP64;
2017
2018         sr.l_whence = 0;
2019         sr.l_start = (s64)offset;
2020         sr.l_len = (s64)len;
2021
2022         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2023                                          change_size);
2024 }
2025
2026 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2027                                    size_t count)
2028 {
2029         int ret = 0;
2030         unsigned int extent_flags;
2031         u32 cpos, clusters, extent_len, phys_cpos;
2032         struct super_block *sb = inode->i_sb;
2033
2034         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2035             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2036             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2037                 return 0;
2038
2039         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2040         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2041
2042         while (clusters) {
2043                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2044                                          &extent_flags);
2045                 if (ret < 0) {
2046                         mlog_errno(ret);
2047                         goto out;
2048                 }
2049
2050                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2051                         ret = 1;
2052                         break;
2053                 }
2054
2055                 if (extent_len > clusters)
2056                         extent_len = clusters;
2057
2058                 clusters -= extent_len;
2059                 cpos += extent_len;
2060         }
2061 out:
2062         return ret;
2063 }
2064
2065 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2066 {
2067         int blockmask = inode->i_sb->s_blocksize - 1;
2068         loff_t final_size = pos + count;
2069
2070         if ((pos & blockmask) || (final_size & blockmask))
2071                 return 1;
2072         return 0;
2073 }
2074
2075 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2076                                             struct file *file,
2077                                             loff_t pos, size_t count,
2078                                             int *meta_level)
2079 {
2080         int ret;
2081         struct buffer_head *di_bh = NULL;
2082         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2083         u32 clusters =
2084                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2085
2086         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2087         if (ret) {
2088                 mlog_errno(ret);
2089                 goto out;
2090         }
2091
2092         *meta_level = 1;
2093
2094         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2095         if (ret)
2096                 mlog_errno(ret);
2097 out:
2098         brelse(di_bh);
2099         return ret;
2100 }
2101
2102 static int ocfs2_prepare_inode_for_write(struct file *file,
2103                                          loff_t *ppos,
2104                                          size_t count,
2105                                          int appending,
2106                                          int *direct_io,
2107                                          int *has_refcount)
2108 {
2109         int ret = 0, meta_level = 0;
2110         struct dentry *dentry = file->f_path.dentry;
2111         struct inode *inode = dentry->d_inode;
2112         loff_t saved_pos = 0, end;
2113
2114         /*
2115          * We start with a read level meta lock and only jump to an ex
2116          * if we need to make modifications here.
2117          */
2118         for(;;) {
2119                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2120                 if (ret < 0) {
2121                         meta_level = -1;
2122                         mlog_errno(ret);
2123                         goto out;
2124                 }
2125
2126                 /* Clear suid / sgid if necessary. We do this here
2127                  * instead of later in the write path because
2128                  * remove_suid() calls ->setattr without any hint that
2129                  * we may have already done our cluster locking. Since
2130                  * ocfs2_setattr() *must* take cluster locks to
2131                  * proceed, this will lead us to recursively lock the
2132                  * inode. There's also the dinode i_size state which
2133                  * can be lost via setattr during extending writes (we
2134                  * set inode->i_size at the end of a write. */
2135                 if (should_remove_suid(dentry)) {
2136                         if (meta_level == 0) {
2137                                 ocfs2_inode_unlock(inode, meta_level);
2138                                 meta_level = 1;
2139                                 continue;
2140                         }
2141
2142                         ret = ocfs2_write_remove_suid(inode);
2143                         if (ret < 0) {
2144                                 mlog_errno(ret);
2145                                 goto out_unlock;
2146                         }
2147                 }
2148
2149                 /* work on a copy of ppos until we're sure that we won't have
2150                  * to recalculate it due to relocking. */
2151                 if (appending)
2152                         saved_pos = i_size_read(inode);
2153                 else
2154                         saved_pos = *ppos;
2155
2156                 end = saved_pos + count;
2157
2158                 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2159                 if (ret == 1) {
2160                         ocfs2_inode_unlock(inode, meta_level);
2161                         meta_level = -1;
2162
2163                         ret = ocfs2_prepare_inode_for_refcount(inode,
2164                                                                file,
2165                                                                saved_pos,
2166                                                                count,
2167                                                                &meta_level);
2168                         if (has_refcount)
2169                                 *has_refcount = 1;
2170                         if (direct_io)
2171                                 *direct_io = 0;
2172                 }
2173
2174                 if (ret < 0) {
2175                         mlog_errno(ret);
2176                         goto out_unlock;
2177                 }
2178
2179                 /*
2180                  * Skip the O_DIRECT checks if we don't need
2181                  * them.
2182                  */
2183                 if (!direct_io || !(*direct_io))
2184                         break;
2185
2186                 /*
2187                  * There's no sane way to do direct writes to an inode
2188                  * with inline data.
2189                  */
2190                 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2191                         *direct_io = 0;
2192                         break;
2193                 }
2194
2195                 /*
2196                  * Allowing concurrent direct writes means
2197                  * i_size changes wouldn't be synchronized, so
2198                  * one node could wind up truncating another
2199                  * nodes writes.
2200                  */
2201                 if (end > i_size_read(inode)) {
2202                         *direct_io = 0;
2203                         break;
2204                 }
2205
2206                 /*
2207                  * We don't fill holes during direct io, so
2208                  * check for them here. If any are found, the
2209                  * caller will have to retake some cluster
2210                  * locks and initiate the io as buffered.
2211                  */
2212                 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2213                 if (ret == 1) {
2214                         *direct_io = 0;
2215                         ret = 0;
2216                 } else if (ret < 0)
2217                         mlog_errno(ret);
2218                 break;
2219         }
2220
2221         if (appending)
2222                 *ppos = saved_pos;
2223
2224 out_unlock:
2225         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2226                                             saved_pos, appending, count,
2227                                             direct_io, has_refcount);
2228
2229         if (meta_level >= 0)
2230                 ocfs2_inode_unlock(inode, meta_level);
2231
2232 out:
2233         return ret;
2234 }
2235
2236 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2237                                     struct iov_iter *from)
2238 {
2239         int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
2240         int can_do_direct, has_refcount = 0;
2241         ssize_t written = 0;
2242         size_t count = iov_iter_count(from);
2243         loff_t old_size, *ppos = &iocb->ki_pos;
2244         u32 old_clusters;
2245         struct file *file = iocb->ki_filp;
2246         struct inode *inode = file_inode(file);
2247         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2248         int full_coherency = !(osb->s_mount_opt &
2249                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2250         int unaligned_dio = 0;
2251
2252         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2253                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2254                 file->f_path.dentry->d_name.len,
2255                 file->f_path.dentry->d_name.name,
2256                 (unsigned int)from->nr_segs);   /* GRRRRR */
2257
2258         if (iocb->ki_nbytes == 0)
2259                 return 0;
2260
2261         appending = file->f_flags & O_APPEND ? 1 : 0;
2262         direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2263
2264         mutex_lock(&inode->i_mutex);
2265
2266         ocfs2_iocb_clear_sem_locked(iocb);
2267
2268 relock:
2269         /* to match setattr's i_mutex -> rw_lock ordering */
2270         if (direct_io) {
2271                 have_alloc_sem = 1;
2272                 /* communicate with ocfs2_dio_end_io */
2273                 ocfs2_iocb_set_sem_locked(iocb);
2274         }
2275
2276         /*
2277          * Concurrent O_DIRECT writes are allowed with
2278          * mount_option "coherency=buffered".
2279          */
2280         rw_level = (!direct_io || full_coherency);
2281
2282         ret = ocfs2_rw_lock(inode, rw_level);
2283         if (ret < 0) {
2284                 mlog_errno(ret);
2285                 goto out_sems;
2286         }
2287
2288         /*
2289          * O_DIRECT writes with "coherency=full" need to take EX cluster
2290          * inode_lock to guarantee coherency.
2291          */
2292         if (direct_io && full_coherency) {
2293                 /*
2294                  * We need to take and drop the inode lock to force
2295                  * other nodes to drop their caches.  Buffered I/O
2296                  * already does this in write_begin().
2297                  */
2298                 ret = ocfs2_inode_lock(inode, NULL, 1);
2299                 if (ret < 0) {
2300                         mlog_errno(ret);
2301                         goto out;
2302                 }
2303
2304                 ocfs2_inode_unlock(inode, 1);
2305         }
2306
2307         can_do_direct = direct_io;
2308         ret = ocfs2_prepare_inode_for_write(file, ppos,
2309                                             iocb->ki_nbytes, appending,
2310                                             &can_do_direct, &has_refcount);
2311         if (ret < 0) {
2312                 mlog_errno(ret);
2313                 goto out;
2314         }
2315
2316         if (direct_io && !is_sync_kiocb(iocb))
2317                 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_nbytes,
2318                                                       *ppos);
2319
2320         /*
2321          * We can't complete the direct I/O as requested, fall back to
2322          * buffered I/O.
2323          */
2324         if (direct_io && !can_do_direct) {
2325                 ocfs2_rw_unlock(inode, rw_level);
2326
2327                 have_alloc_sem = 0;
2328                 rw_level = -1;
2329
2330                 direct_io = 0;
2331                 goto relock;
2332         }
2333
2334         if (unaligned_dio) {
2335                 /*
2336                  * Wait on previous unaligned aio to complete before
2337                  * proceeding.
2338                  */
2339                 mutex_lock(&OCFS2_I(inode)->ip_unaligned_aio);
2340                 /* Mark the iocb as needing an unlock in ocfs2_dio_end_io */
2341                 ocfs2_iocb_set_unaligned_aio(iocb);
2342         }
2343
2344         /*
2345          * To later detect whether a journal commit for sync writes is
2346          * necessary, we sample i_size, and cluster count here.
2347          */
2348         old_size = i_size_read(inode);
2349         old_clusters = OCFS2_I(inode)->ip_clusters;
2350
2351         /* communicate with ocfs2_dio_end_io */
2352         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2353
2354         ret = generic_write_checks(file, ppos, &count,
2355                                    S_ISBLK(inode->i_mode));
2356         if (ret)
2357                 goto out_dio;
2358
2359         iov_iter_truncate(from, count);
2360         if (direct_io) {
2361                 written = generic_file_direct_write(iocb, from, *ppos);
2362                 if (written < 0) {
2363                         ret = written;
2364                         goto out_dio;
2365                 }
2366         } else {
2367                 current->backing_dev_info = file->f_mapping->backing_dev_info;
2368                 written = generic_perform_write(file, from, *ppos);
2369                 if (likely(written >= 0))
2370                         iocb->ki_pos = *ppos + written;
2371                 current->backing_dev_info = NULL;
2372         }
2373
2374 out_dio:
2375         /* buffered aio wouldn't have proper lock coverage today */
2376         BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2377
2378         if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2379             ((file->f_flags & O_DIRECT) && !direct_io)) {
2380                 ret = filemap_fdatawrite_range(file->f_mapping, *ppos,
2381                                                *ppos + count - 1);
2382                 if (ret < 0)
2383                         written = ret;
2384
2385                 if (!ret && ((old_size != i_size_read(inode)) ||
2386                              (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2387                              has_refcount)) {
2388                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2389                         if (ret < 0)
2390                                 written = ret;
2391                 }
2392
2393                 if (!ret)
2394                         ret = filemap_fdatawait_range(file->f_mapping, *ppos,
2395                                                       *ppos + count - 1);
2396         }
2397
2398         /*
2399          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2400          * function pointer which is called when o_direct io completes so that
2401          * it can unlock our rw lock.
2402          * Unfortunately there are error cases which call end_io and others
2403          * that don't.  so we don't have to unlock the rw_lock if either an
2404          * async dio is going to do it in the future or an end_io after an
2405          * error has already done it.
2406          */
2407         if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2408                 rw_level = -1;
2409                 have_alloc_sem = 0;
2410                 unaligned_dio = 0;
2411         }
2412
2413         if (unaligned_dio) {
2414                 ocfs2_iocb_clear_unaligned_aio(iocb);
2415                 mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio);
2416         }
2417
2418 out:
2419         if (rw_level != -1)
2420                 ocfs2_rw_unlock(inode, rw_level);
2421
2422 out_sems:
2423         if (have_alloc_sem)
2424                 ocfs2_iocb_clear_sem_locked(iocb);
2425
2426         mutex_unlock(&inode->i_mutex);
2427
2428         if (written)
2429                 ret = written;
2430         return ret;
2431 }
2432
2433 static ssize_t ocfs2_file_splice_read(struct file *in,
2434                                       loff_t *ppos,
2435                                       struct pipe_inode_info *pipe,
2436                                       size_t len,
2437                                       unsigned int flags)
2438 {
2439         int ret = 0, lock_level = 0;
2440         struct inode *inode = file_inode(in);
2441
2442         trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2443                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2444                         in->f_path.dentry->d_name.len,
2445                         in->f_path.dentry->d_name.name, len);
2446
2447         /*
2448          * See the comment in ocfs2_file_read_iter()
2449          */
2450         ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2451         if (ret < 0) {
2452                 mlog_errno(ret);
2453                 goto bail;
2454         }
2455         ocfs2_inode_unlock(inode, lock_level);
2456
2457         ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2458
2459 bail:
2460         return ret;
2461 }
2462
2463 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2464                                    struct iov_iter *to)
2465 {
2466         int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2467         struct file *filp = iocb->ki_filp;
2468         struct inode *inode = file_inode(filp);
2469
2470         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2471                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2472                         filp->f_path.dentry->d_name.len,
2473                         filp->f_path.dentry->d_name.name,
2474                         to->nr_segs);   /* GRRRRR */
2475
2476
2477         if (!inode) {
2478                 ret = -EINVAL;
2479                 mlog_errno(ret);
2480                 goto bail;
2481         }
2482
2483         ocfs2_iocb_clear_sem_locked(iocb);
2484
2485         /*
2486          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2487          * need locks to protect pending reads from racing with truncate.
2488          */
2489         if (filp->f_flags & O_DIRECT) {
2490                 have_alloc_sem = 1;
2491                 ocfs2_iocb_set_sem_locked(iocb);
2492
2493                 ret = ocfs2_rw_lock(inode, 0);
2494                 if (ret < 0) {
2495                         mlog_errno(ret);
2496                         goto bail;
2497                 }
2498                 rw_level = 0;
2499                 /* communicate with ocfs2_dio_end_io */
2500                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2501         }
2502
2503         /*
2504          * We're fine letting folks race truncates and extending
2505          * writes with read across the cluster, just like they can
2506          * locally. Hence no rw_lock during read.
2507          *
2508          * Take and drop the meta data lock to update inode fields
2509          * like i_size. This allows the checks down below
2510          * generic_file_aio_read() a chance of actually working.
2511          */
2512         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2513         if (ret < 0) {
2514                 mlog_errno(ret);
2515                 goto bail;
2516         }
2517         ocfs2_inode_unlock(inode, lock_level);
2518
2519         ret = generic_file_read_iter(iocb, to);
2520         trace_generic_file_aio_read_ret(ret);
2521
2522         /* buffered aio wouldn't have proper lock coverage today */
2523         BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2524
2525         /* see ocfs2_file_write_iter */
2526         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2527                 rw_level = -1;
2528                 have_alloc_sem = 0;
2529         }
2530
2531 bail:
2532         if (have_alloc_sem)
2533                 ocfs2_iocb_clear_sem_locked(iocb);
2534
2535         if (rw_level != -1)
2536                 ocfs2_rw_unlock(inode, rw_level);
2537
2538         return ret;
2539 }
2540
2541 /* Refer generic_file_llseek_unlocked() */
2542 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2543 {
2544         struct inode *inode = file->f_mapping->host;
2545         int ret = 0;
2546
2547         mutex_lock(&inode->i_mutex);
2548
2549         switch (whence) {
2550         case SEEK_SET:
2551                 break;
2552         case SEEK_END:
2553                 /* SEEK_END requires the OCFS2 inode lock for the file
2554                  * because it references the file's size.
2555                  */
2556                 ret = ocfs2_inode_lock(inode, NULL, 0);
2557                 if (ret < 0) {
2558                         mlog_errno(ret);
2559                         goto out;
2560                 }
2561                 offset += i_size_read(inode);
2562                 ocfs2_inode_unlock(inode, 0);
2563                 break;
2564         case SEEK_CUR:
2565                 if (offset == 0) {
2566                         offset = file->f_pos;
2567                         goto out;
2568                 }
2569                 offset += file->f_pos;
2570                 break;
2571         case SEEK_DATA:
2572         case SEEK_HOLE:
2573                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2574                 if (ret)
2575                         goto out;
2576                 break;
2577         default:
2578                 ret = -EINVAL;
2579                 goto out;
2580         }
2581
2582         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2583
2584 out:
2585         mutex_unlock(&inode->i_mutex);
2586         if (ret)
2587                 return ret;
2588         return offset;
2589 }
2590
2591 const struct inode_operations ocfs2_file_iops = {
2592         .setattr        = ocfs2_setattr,
2593         .getattr        = ocfs2_getattr,
2594         .permission     = ocfs2_permission,
2595         .setxattr       = generic_setxattr,
2596         .getxattr       = generic_getxattr,
2597         .listxattr      = ocfs2_listxattr,
2598         .removexattr    = generic_removexattr,
2599         .fiemap         = ocfs2_fiemap,
2600         .get_acl        = ocfs2_iop_get_acl,
2601         .set_acl        = ocfs2_iop_set_acl,
2602 };
2603
2604 const struct inode_operations ocfs2_special_file_iops = {
2605         .setattr        = ocfs2_setattr,
2606         .getattr        = ocfs2_getattr,
2607         .permission     = ocfs2_permission,
2608         .get_acl        = ocfs2_iop_get_acl,
2609         .set_acl        = ocfs2_iop_set_acl,
2610 };
2611
2612 /*
2613  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2614  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2615  */
2616 const struct file_operations ocfs2_fops = {
2617         .llseek         = ocfs2_file_llseek,
2618         .read           = new_sync_read,
2619         .write          = new_sync_write,
2620         .mmap           = ocfs2_mmap,
2621         .fsync          = ocfs2_sync_file,
2622         .release        = ocfs2_file_release,
2623         .open           = ocfs2_file_open,
2624         .read_iter      = ocfs2_file_read_iter,
2625         .write_iter     = ocfs2_file_write_iter,
2626         .unlocked_ioctl = ocfs2_ioctl,
2627 #ifdef CONFIG_COMPAT
2628         .compat_ioctl   = ocfs2_compat_ioctl,
2629 #endif
2630         .lock           = ocfs2_lock,
2631         .flock          = ocfs2_flock,
2632         .splice_read    = ocfs2_file_splice_read,
2633         .splice_write   = iter_file_splice_write,
2634         .fallocate      = ocfs2_fallocate,
2635 };
2636
2637 const struct file_operations ocfs2_dops = {
2638         .llseek         = generic_file_llseek,
2639         .read           = generic_read_dir,
2640         .iterate        = ocfs2_readdir,
2641         .fsync          = ocfs2_sync_file,
2642         .release        = ocfs2_dir_release,
2643         .open           = ocfs2_dir_open,
2644         .unlocked_ioctl = ocfs2_ioctl,
2645 #ifdef CONFIG_COMPAT
2646         .compat_ioctl   = ocfs2_compat_ioctl,
2647 #endif
2648         .lock           = ocfs2_lock,
2649         .flock          = ocfs2_flock,
2650 };
2651
2652 /*
2653  * POSIX-lockless variants of our file_operations.
2654  *
2655  * These will be used if the underlying cluster stack does not support
2656  * posix file locking, if the user passes the "localflocks" mount
2657  * option, or if we have a local-only fs.
2658  *
2659  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2660  * so we still want it in the case of no stack support for
2661  * plocks. Internally, it will do the right thing when asked to ignore
2662  * the cluster.
2663  */
2664 const struct file_operations ocfs2_fops_no_plocks = {
2665         .llseek         = ocfs2_file_llseek,
2666         .read           = new_sync_read,
2667         .write          = new_sync_write,
2668         .mmap           = ocfs2_mmap,
2669         .fsync          = ocfs2_sync_file,
2670         .release        = ocfs2_file_release,
2671         .open           = ocfs2_file_open,
2672         .read_iter      = ocfs2_file_read_iter,
2673         .write_iter     = ocfs2_file_write_iter,
2674         .unlocked_ioctl = ocfs2_ioctl,
2675 #ifdef CONFIG_COMPAT
2676         .compat_ioctl   = ocfs2_compat_ioctl,
2677 #endif
2678         .flock          = ocfs2_flock,
2679         .splice_read    = ocfs2_file_splice_read,
2680         .splice_write   = iter_file_splice_write,
2681         .fallocate      = ocfs2_fallocate,
2682 };
2683
2684 const struct file_operations ocfs2_dops_no_plocks = {
2685         .llseek         = generic_file_llseek,
2686         .read           = generic_read_dir,
2687         .iterate        = ocfs2_readdir,
2688         .fsync          = ocfs2_sync_file,
2689         .release        = ocfs2_dir_release,
2690         .open           = ocfs2_dir_open,
2691         .unlocked_ioctl = ocfs2_ioctl,
2692 #ifdef CONFIG_COMPAT
2693         .compat_ioctl   = ocfs2_compat_ioctl,
2694 #endif
2695         .flock          = ocfs2_flock,
2696 };