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 #include <linux/backing-dev.h>
  41
  42 #include <cluster/masklog.h>
  43
  44 #include "ocfs2.h"
  45
  46 #include "alloc.h"
  47 #include "aops.h"
  48 #include "dir.h"
  49 #include "dlmglue.h"
  50 #include "extent_map.h"
  51 #include "file.h"
  52 #include "sysfile.h"
  53 #include "inode.h"
  54 #include "ioctl.h"
  55 #include "journal.h"
  56 #include "locks.h"
  57 #include "mmap.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "xattr.h"
  61 #include "acl.h"
  62 #include "quota.h"
  63 #include "refcounttree.h"
  64 #include "ocfs2_trace.h"
  65
  66 #include "buffer_head_io.h"
  67
  68 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  69 {
  70         struct ocfs2_file_private *fp;
  71
  72         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  73         if (!fp)
  74                 return -ENOMEM;
  75
  76         fp->fp_file = file;
  77         mutex_init(&fp->fp_mutex);
  78         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  79         file->private_data = fp;
  80
  81         return 0;
  82 }
  83
  84 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  85 {
  86         struct ocfs2_file_private *fp = file->private_data;
  87         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  88
  89         if (fp) {
  90                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  91                 ocfs2_lock_res_free(&fp->fp_flock);
  92                 kfree(fp);
  93                 file->private_data = NULL;
  94         }
  95 }
  96
  97 static int ocfs2_file_open(struct inode *inode, struct file *file)
  98 {
  99         int status;
 100         int mode = file->f_flags;
 101         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 102
 103         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
 104                               (unsigned long long)OCFS2_I(inode)->ip_blkno,
 105                               file->f_path.dentry->d_name.len,
 106                               file->f_path.dentry->d_name.name, mode);
 107
 108         if (file->f_mode & FMODE_WRITE) {
 109                 status = dquot_initialize(inode);
 110                 if (status)
 111                         goto leave;
 112         }
 113
 114         spin_lock(&oi->ip_lock);
 115
 116         /* Check that the inode hasn't been wiped from disk by another
 117          * node. If it hasn't then we're safe as long as we hold the
 118          * spin lock until our increment of open count. */
 119         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
 120                 spin_unlock(&oi->ip_lock);
 121
 122                 status = -ENOENT;
 123                 goto leave;
 124         }
 125
 126         if (mode & O_DIRECT)
 127                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 128
 129         oi->ip_open_count++;
 130         spin_unlock(&oi->ip_lock);
 131
 132         status = ocfs2_init_file_private(inode, file);
 133         if (status) {
 134                 /*
 135                  * We want to set open count back if we're failing the
 136                  * open.
 137                  */
 138                 spin_lock(&oi->ip_lock);
 139                 oi->ip_open_count--;
 140                 spin_unlock(&oi->ip_lock);
 141         }
 142
 143 leave:
 144         return status;
 145 }
 146
 147 static int ocfs2_file_release(struct inode *inode, struct file *file)
 148 {
 149         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 150
 151         spin_lock(&oi->ip_lock);
 152         if (!--oi->ip_open_count)
 153                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 154
 155         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 156                                  oi->ip_blkno,
 157                                  file->f_path.dentry->d_name.len,
 158                                  file->f_path.dentry->d_name.name,
 159                                  oi->ip_open_count);
 160         spin_unlock(&oi->ip_lock);
 161
 162         ocfs2_free_file_private(inode, file);
 163
 164         return 0;
 165 }
 166
 167 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 168 {
 169         return ocfs2_init_file_private(inode, file);
 170 }
 171
 172 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 173 {
 174         ocfs2_free_file_private(inode, file);
 175         return 0;
 176 }
 177
 178 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 179                            int datasync)
 180 {
 181         int err = 0;
 182         struct inode *inode = file->f_mapping->host;
 183         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 184         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 185         journal_t *journal = osb->journal->j_journal;
 186         int ret;
 187         tid_t commit_tid;
 188         bool needs_barrier = false;
 189
 190         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 191                               OCFS2_I(inode)->ip_blkno,
 192                               file->f_path.dentry->d_name.len,
 193                               file->f_path.dentry->d_name.name,
 194                               (unsigned long long)datasync);
 195
 196         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 197                 return -EROFS;
 198
 199         err = file_write_and_wait_range(file, start, end);
 200         if (err)
 201                 return err;
 202
 203         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 204         if (journal->j_flags & JBD2_BARRIER &&
 205             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 206                 needs_barrier = true;
 207         err = jbd2_complete_transaction(journal, commit_tid);
 208         if (needs_barrier) {
 209                 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 210                 if (!err)
 211                         err = ret;
 212         }
 213
 214         if (err)
 215                 mlog_errno(err);
 216
 217         return (err < 0) ? -EIO : 0;
 218 }
 219
 220 int ocfs2_should_update_atime(struct inode *inode,
 221                               struct vfsmount *vfsmnt)
 222 {
 223         struct timespec now;
 224         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 225
 226         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 227                 return 0;
 228
 229         if ((inode->i_flags & S_NOATIME) ||
 230             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 231                 return 0;
 232
 233         /*
 234          * We can be called with no vfsmnt structure - NFSD will
 235          * sometimes do this.
 236          *
 237          * Note that our action here is different than touch_atime() -
 238          * if we can't tell whether this is a noatime mount, then we
 239          * don't know whether to trust the value of s_atime_quantum.
 240          */
 241         if (vfsmnt == NULL)
 242                 return 0;
 243
 244         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 245             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 246                 return 0;
 247
 248         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 249                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 250                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 251                         return 1;
 252
 253                 return 0;
 254         }
 255
 256         now = current_time(inode);
 257         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 258                 return 0;
 259         else
 260                 return 1;
 261 }
 262
 263 int ocfs2_update_inode_atime(struct inode *inode,
 264                              struct buffer_head *bh)
 265 {
 266         int ret;
 267         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 268         handle_t *handle;
 269         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 270
 271         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 272         if (IS_ERR(handle)) {
 273                 ret = PTR_ERR(handle);
 274                 mlog_errno(ret);
 275                 goto out;
 276         }
 277
 278         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 279                                       OCFS2_JOURNAL_ACCESS_WRITE);
 280         if (ret) {
 281                 mlog_errno(ret);
 282                 goto out_commit;
 283         }
 284
 285         /*
 286          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 287          * have i_mutex to guard against concurrent changes to other
 288          * inode fields.
 289          */
 290         inode->i_atime = current_time(inode);
 291         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 292         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 293         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 294         ocfs2_journal_dirty(handle, bh);
 295
 296 out_commit:
 297         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 298 out:
 299         return ret;
 300 }
 301
 302 int ocfs2_set_inode_size(handle_t *handle,
 303                                 struct inode *inode,
 304                                 struct buffer_head *fe_bh,
 305                                 u64 new_i_size)
 306 {
 307         int status;
 308
 309         i_size_write(inode, new_i_size);
 310         inode->i_blocks = ocfs2_inode_sector_count(inode);
 311         inode->i_ctime = inode->i_mtime = current_time(inode);
 312
 313         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 314         if (status < 0) {
 315                 mlog_errno(status);
 316                 goto bail;
 317         }
 318
 319 bail:
 320         return status;
 321 }
 322
 323 int ocfs2_simple_size_update(struct inode *inode,
 324                              struct buffer_head *di_bh,
 325                              u64 new_i_size)
 326 {
 327         int ret;
 328         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 329         handle_t *handle = NULL;
 330
 331         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 332         if (IS_ERR(handle)) {
 333                 ret = PTR_ERR(handle);
 334                 mlog_errno(ret);
 335                 goto out;
 336         }
 337
 338         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 339                                    new_i_size);
 340         if (ret < 0)
 341                 mlog_errno(ret);
 342
 343         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 344         ocfs2_commit_trans(osb, handle);
 345 out:
 346         return ret;
 347 }
 348
 349 static int ocfs2_cow_file_pos(struct inode *inode,
 350                               struct buffer_head *fe_bh,
 351                               u64 offset)
 352 {
 353         int status;
 354         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 355         unsigned int num_clusters = 0;
 356         unsigned int ext_flags = 0;
 357
 358         /*
 359          * If the new offset is aligned to the range of the cluster, there is
 360          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 361          * CoW either.
 362          */
 363         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 364                 return 0;
 365
 366         status = ocfs2_get_clusters(inode, cpos, &phys,
 367                                     &num_clusters, &ext_flags);
 368         if (status) {
 369                 mlog_errno(status);
 370                 goto out;
 371         }
 372
 373         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 374                 goto out;
 375
 376         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 377
 378 out:
 379         return status;
 380 }
 381
 382 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 383                                      struct inode *inode,
 384                                      struct buffer_head *fe_bh,
 385                                      u64 new_i_size)
 386 {
 387         int status;
 388         handle_t *handle;
 389         struct ocfs2_dinode *di;
 390         u64 cluster_bytes;
 391
 392         /*
 393          * We need to CoW the cluster contains the offset if it is reflinked
 394          * since we will call ocfs2_zero_range_for_truncate later which will
 395          * write "0" from offset to the end of the cluster.
 396          */
 397         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 398         if (status) {
 399                 mlog_errno(status);
 400                 return status;
 401         }
 402
 403         /* TODO: This needs to actually orphan the inode in this
 404          * transaction. */
 405
 406         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 407         if (IS_ERR(handle)) {
 408                 status = PTR_ERR(handle);
 409                 mlog_errno(status);
 410                 goto out;
 411         }
 412
 413         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 414                                          OCFS2_JOURNAL_ACCESS_WRITE);
 415         if (status < 0) {
 416                 mlog_errno(status);
 417                 goto out_commit;
 418         }
 419
 420         /*
 421          * Do this before setting i_size.
 422          */
 423         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 424         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 425                                                cluster_bytes);
 426         if (status) {
 427                 mlog_errno(status);
 428                 goto out_commit;
 429         }
 430
 431         i_size_write(inode, new_i_size);
 432         inode->i_ctime = inode->i_mtime = current_time(inode);
 433
 434         di = (struct ocfs2_dinode *) fe_bh->b_data;
 435         di->i_size = cpu_to_le64(new_i_size);
 436         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 437         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 438         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 439
 440         ocfs2_journal_dirty(handle, fe_bh);
 441
 442 out_commit:
 443         ocfs2_commit_trans(osb, handle);
 444 out:
 445         return status;
 446 }
 447
 448 int ocfs2_truncate_file(struct inode *inode,
 449                                struct buffer_head *di_bh,
 450                                u64 new_i_size)
 451 {
 452         int status = 0;
 453         struct ocfs2_dinode *fe = NULL;
 454         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 455
 456         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 457          * already validated it */
 458         fe = (struct ocfs2_dinode *) di_bh->b_data;
 459
 460         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 461                                   (unsigned long long)le64_to_cpu(fe->i_size),
 462                                   (unsigned long long)new_i_size);
 463
 464         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 465                         "Inode %llu, inode i_size = %lld != di "
 466                         "i_size = %llu, i_flags = 0x%x\n",
 467                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 468                         i_size_read(inode),
 469                         (unsigned long long)le64_to_cpu(fe->i_size),
 470                         le32_to_cpu(fe->i_flags));
 471
 472         if (new_i_size > le64_to_cpu(fe->i_size)) {
 473                 trace_ocfs2_truncate_file_error(
 474                         (unsigned long long)le64_to_cpu(fe->i_size),
 475                         (unsigned long long)new_i_size);
 476                 status = -EINVAL;
 477                 mlog_errno(status);
 478                 goto bail;
 479         }
 480
 481         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 482
 483         ocfs2_resv_discard(&osb->osb_la_resmap,
 484                            &OCFS2_I(inode)->ip_la_data_resv);
 485
 486         /*
 487          * The inode lock forced other nodes to sync and drop their
 488          * pages, which (correctly) happens even if we have a truncate
 489          * without allocation change - ocfs2 cluster sizes can be much
 490          * greater than page size, so we have to truncate them
 491          * anyway.
 492          */
 493
 494         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 495                 unmap_mapping_range(inode->i_mapping,
 496                                     new_i_size + PAGE_SIZE - 1, 0, 1);
 497                 truncate_inode_pages(inode->i_mapping, new_i_size);
 498                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 499                                                i_size_read(inode), 1);
 500                 if (status)
 501                         mlog_errno(status);
 502
 503                 goto bail_unlock_sem;
 504         }
 505
 506         /* alright, we're going to need to do a full blown alloc size
 507          * change. Orphan the inode so that recovery can complete the
 508          * truncate if necessary. This does the task of marking
 509          * i_size. */
 510         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 511         if (status < 0) {
 512                 mlog_errno(status);
 513                 goto bail_unlock_sem;
 514         }
 515
 516         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 517         truncate_inode_pages(inode->i_mapping, new_i_size);
 518
 519         status = ocfs2_commit_truncate(osb, inode, di_bh);
 520         if (status < 0) {
 521                 mlog_errno(status);
 522                 goto bail_unlock_sem;
 523         }
 524
 525         /* TODO: orphan dir cleanup here. */
 526 bail_unlock_sem:
 527         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 528
 529 bail:
 530         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 531                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 532
 533         return status;
 534 }
 535
 536 /*
 537  * extend file allocation only here.
 538  * we'll update all the disk stuff, and oip->alloc_size
 539  *
 540  * expect stuff to be locked, a transaction started and enough data /
 541  * metadata reservations in the contexts.
 542  *
 543  * Will return -EAGAIN, and a reason if a restart is needed.
 544  * If passed in, *reason will always be set, even in error.
 545  */
 546 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 547                          struct inode *inode,
 548                          u32 *logical_offset,
 549                          u32 clusters_to_add,
 550                          int mark_unwritten,
 551                          struct buffer_head *fe_bh,
 552                          handle_t *handle,
 553                          struct ocfs2_alloc_context *data_ac,
 554                          struct ocfs2_alloc_context *meta_ac,
 555                          enum ocfs2_alloc_restarted *reason_ret)
 556 {
 557         int ret;
 558         struct ocfs2_extent_tree et;
 559
 560         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 561         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 562                                           clusters_to_add, mark_unwritten,
 563                                           data_ac, meta_ac, reason_ret);
 564
 565         return ret;
 566 }
 567
 568 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 569                                      u32 clusters_to_add, int mark_unwritten)
 570 {
 571         int status = 0;
 572         int restart_func = 0;
 573         int credits;
 574         u32 prev_clusters;
 575         struct buffer_head *bh = NULL;
 576         struct ocfs2_dinode *fe = NULL;
 577         handle_t *handle = NULL;
 578         struct ocfs2_alloc_context *data_ac = NULL;
 579         struct ocfs2_alloc_context *meta_ac = NULL;
 580         enum ocfs2_alloc_restarted why = RESTART_NONE;
 581         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 582         struct ocfs2_extent_tree et;
 583         int did_quota = 0;
 584
 585         /*
 586          * Unwritten extent only exists for file systems which
 587          * support holes.
 588          */
 589         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 590
 591         status = ocfs2_read_inode_block(inode, &bh);
 592         if (status < 0) {
 593                 mlog_errno(status);
 594                 goto leave;
 595         }
 596         fe = (struct ocfs2_dinode *) bh->b_data;
 597
 598 restart_all:
 599         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 600
 601         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 602         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 603                                        &data_ac, &meta_ac);
 604         if (status) {
 605                 mlog_errno(status);
 606                 goto leave;
 607         }
 608
 609         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 610         handle = ocfs2_start_trans(osb, credits);
 611         if (IS_ERR(handle)) {
 612                 status = PTR_ERR(handle);
 613                 handle = NULL;
 614                 mlog_errno(status);
 615                 goto leave;
 616         }
 617
 618 restarted_transaction:
 619         trace_ocfs2_extend_allocation(
 620                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 621                 (unsigned long long)i_size_read(inode),
 622                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 623                 why, restart_func);
 624
 625         status = dquot_alloc_space_nodirty(inode,
 626                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 627         if (status)
 628                 goto leave;
 629         did_quota = 1;
 630
 631         /* reserve a write to the file entry early on - that we if we
 632          * run out of credits in the allocation path, we can still
 633          * update i_size. */
 634         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 635                                          OCFS2_JOURNAL_ACCESS_WRITE);
 636         if (status < 0) {
 637                 mlog_errno(status);
 638                 goto leave;
 639         }
 640
 641         prev_clusters = OCFS2_I(inode)->ip_clusters;
 642
 643         status = ocfs2_add_inode_data(osb,
 644                                       inode,
 645                                       &logical_start,
 646                                       clusters_to_add,
 647                                       mark_unwritten,
 648                                       bh,
 649                                       handle,
 650                                       data_ac,
 651                                       meta_ac,
 652                                       &why);
 653         if ((status < 0) && (status != -EAGAIN)) {
 654                 if (status != -ENOSPC)
 655                         mlog_errno(status);
 656                 goto leave;
 657         }
 658         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 659         ocfs2_journal_dirty(handle, bh);
 660
 661         spin_lock(&OCFS2_I(inode)->ip_lock);
 662         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 663         spin_unlock(&OCFS2_I(inode)->ip_lock);
 664         /* Release unused quota reservation */
 665         dquot_free_space(inode,
 666                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 667         did_quota = 0;
 668
 669         if (why != RESTART_NONE && clusters_to_add) {
 670                 if (why == RESTART_META) {
 671                         restart_func = 1;
 672                         status = 0;
 673                 } else {
 674                         BUG_ON(why != RESTART_TRANS);
 675
 676                         status = ocfs2_allocate_extend_trans(handle, 1);
 677                         if (status < 0) {
 678                                 /* handle still has to be committed at
 679                                  * this point. */
 680                                 status = -ENOMEM;
 681                                 mlog_errno(status);
 682                                 goto leave;
 683                         }
 684                         goto restarted_transaction;
 685                 }
 686         }
 687
 688         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 689              le32_to_cpu(fe->i_clusters),
 690              (unsigned long long)le64_to_cpu(fe->i_size),
 691              OCFS2_I(inode)->ip_clusters,
 692              (unsigned long long)i_size_read(inode));
 693
 694 leave:
 695         if (status < 0 && did_quota)
 696                 dquot_free_space(inode,
 697                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 698         if (handle) {
 699                 ocfs2_commit_trans(osb, handle);
 700                 handle = NULL;
 701         }
 702         if (data_ac) {
 703                 ocfs2_free_alloc_context(data_ac);
 704                 data_ac = NULL;
 705         }
 706         if (meta_ac) {
 707                 ocfs2_free_alloc_context(meta_ac);
 708                 meta_ac = NULL;
 709         }
 710         if ((!status) && restart_func) {
 711                 restart_func = 0;
 712                 goto restart_all;
 713         }
 714         brelse(bh);
 715         bh = NULL;
 716
 717         return status;
 718 }
 719
 720 /*
 721  * While a write will already be ordering the data, a truncate will not.
 722  * Thus, we need to explicitly order the zeroed pages.
 723  */
 724 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 725                                                 struct buffer_head *di_bh)
 726 {
 727         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 728         handle_t *handle = NULL;
 729         int ret = 0;
 730
 731         if (!ocfs2_should_order_data(inode))
 732                 goto out;
 733
 734         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 735         if (IS_ERR(handle)) {
 736                 ret = -ENOMEM;
 737                 mlog_errno(ret);
 738                 goto out;
 739         }
 740
 741         ret = ocfs2_jbd2_file_inode(handle, inode);
 742         if (ret < 0) {
 743                 mlog_errno(ret);
 744                 goto out;
 745         }
 746
 747         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 748                                       OCFS2_JOURNAL_ACCESS_WRITE);
 749         if (ret)
 750                 mlog_errno(ret);
 751         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 752
 753 out:
 754         if (ret) {
 755                 if (!IS_ERR(handle))
 756                         ocfs2_commit_trans(osb, handle);
 757                 handle = ERR_PTR(ret);
 758         }
 759         return handle;
 760 }
 761
 762 /* Some parts of this taken from generic_cont_expand, which turned out
 763  * to be too fragile to do exactly what we need without us having to
 764  * worry about recursive locking in ->write_begin() and ->write_end(). */
 765 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 766                                  u64 abs_to, struct buffer_head *di_bh)
 767 {
 768         struct address_space *mapping = inode->i_mapping;
 769         struct page *page;
 770         unsigned long index = abs_from >> PAGE_SHIFT;
 771         handle_t *handle;
 772         int ret = 0;
 773         unsigned zero_from, zero_to, block_start, block_end;
 774         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 775
 776         BUG_ON(abs_from >= abs_to);
 777         BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 778         BUG_ON(abs_from & (inode->i_blkbits - 1));
 779
 780         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
 781         if (IS_ERR(handle)) {
 782                 ret = PTR_ERR(handle);
 783                 goto out;
 784         }
 785
 786         page = find_or_create_page(mapping, index, GFP_NOFS);
 787         if (!page) {
 788                 ret = -ENOMEM;
 789                 mlog_errno(ret);
 790                 goto out_commit_trans;
 791         }
 792
 793         /* Get the offsets within the page that we want to zero */
 794         zero_from = abs_from & (PAGE_SIZE - 1);
 795         zero_to = abs_to & (PAGE_SIZE - 1);
 796         if (!zero_to)
 797                 zero_to = PAGE_SIZE;
 798
 799         trace_ocfs2_write_zero_page(
 800                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 801                         (unsigned long long)abs_from,
 802                         (unsigned long long)abs_to,
 803                         index, zero_from, zero_to);
 804
 805         /* We know that zero_from is block aligned */
 806         for (block_start = zero_from; block_start < zero_to;
 807              block_start = block_end) {
 808                 block_end = block_start + i_blocksize(inode);
 809
 810                 /*
 811                  * block_start is block-aligned.  Bump it by one to force
 812                  * __block_write_begin and block_commit_write to zero the
 813                  * whole block.
 814                  */
 815                 ret = __block_write_begin(page, block_start + 1, 0,
 816                                           ocfs2_get_block);
 817                 if (ret < 0) {
 818                         mlog_errno(ret);
 819                         goto out_unlock;
 820                 }
 821
 822
 823                 /* must not update i_size! */
 824                 ret = block_commit_write(page, block_start + 1,
 825                                          block_start + 1);
 826                 if (ret < 0)
 827                         mlog_errno(ret);
 828                 else
 829                         ret = 0;
 830         }
 831
 832         /*
 833          * fs-writeback will release the dirty pages without page lock
 834          * whose offset are over inode size, the release happens at
 835          * block_write_full_page().
 836          */
 837         i_size_write(inode, abs_to);
 838         inode->i_blocks = ocfs2_inode_sector_count(inode);
 839         di->i_size = cpu_to_le64((u64)i_size_read(inode));
 840         inode->i_mtime = inode->i_ctime = current_time(inode);
 841         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 842         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 843         di->i_mtime_nsec = di->i_ctime_nsec;
 844         if (handle) {
 845                 ocfs2_journal_dirty(handle, di_bh);
 846                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 847         }
 848
 849 out_unlock:
 850         unlock_page(page);
 851         put_page(page);
 852 out_commit_trans:
 853         if (handle)
 854                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 855 out:
 856         return ret;
 857 }
 858
 859 /*
 860  * Find the next range to zero.  We do this in terms of bytes because
 861  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 862  * pagecache.  We may return multiple extents.
 863  *
 864  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 865  * needs to be zeroed.  range_start and range_end return the next zeroing
 866  * range.  A subsequent call should pass the previous range_end as its
 867  * zero_start.  If range_end is 0, there's nothing to do.
 868  *
 869  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 870  */
 871 static int ocfs2_zero_extend_get_range(struct inode *inode,
 872                                        struct buffer_head *di_bh,
 873                                        u64 zero_start, u64 zero_end,
 874                                        u64 *range_start, u64 *range_end)
 875 {
 876         int rc = 0, needs_cow = 0;
 877         u32 p_cpos, zero_clusters = 0;
 878         u32 zero_cpos =
 879                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 880         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 881         unsigned int num_clusters = 0;
 882         unsigned int ext_flags = 0;
 883
 884         while (zero_cpos < last_cpos) {
 885                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 886                                         &num_clusters, &ext_flags);
 887                 if (rc) {
 888                         mlog_errno(rc);
 889                         goto out;
 890                 }
 891
 892                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 893                         zero_clusters = num_clusters;
 894                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 895                                 needs_cow = 1;
 896                         break;
 897                 }
 898
 899                 zero_cpos += num_clusters;
 900         }
 901         if (!zero_clusters) {
 902                 *range_end = 0;
 903                 goto out;
 904         }
 905
 906         while ((zero_cpos + zero_clusters) < last_cpos) {
 907                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 908                                         &p_cpos, &num_clusters,
 909                                         &ext_flags);
 910                 if (rc) {
 911                         mlog_errno(rc);
 912                         goto out;
 913                 }
 914
 915                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 916                         break;
 917                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 918                         needs_cow = 1;
 919                 zero_clusters += num_clusters;
 920         }
 921         if ((zero_cpos + zero_clusters) > last_cpos)
 922                 zero_clusters = last_cpos - zero_cpos;
 923
 924         if (needs_cow) {
 925                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 926                                         zero_clusters, UINT_MAX);
 927                 if (rc) {
 928                         mlog_errno(rc);
 929                         goto out;
 930                 }
 931         }
 932
 933         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 934         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 935                                              zero_cpos + zero_clusters);
 936
 937 out:
 938         return rc;
 939 }
 940
 941 /*
 942  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 943  * has made sure that the entire range needs zeroing.
 944  */
 945 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 946                                    u64 range_end, struct buffer_head *di_bh)
 947 {
 948         int rc = 0;
 949         u64 next_pos;
 950         u64 zero_pos = range_start;
 951
 952         trace_ocfs2_zero_extend_range(
 953                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 954                         (unsigned long long)range_start,
 955                         (unsigned long long)range_end);
 956         BUG_ON(range_start >= range_end);
 957
 958         while (zero_pos < range_end) {
 959                 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 960                 if (next_pos > range_end)
 961                         next_pos = range_end;
 962                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 963                 if (rc < 0) {
 964                         mlog_errno(rc);
 965                         break;
 966                 }
 967                 zero_pos = next_pos;
 968
 969                 /*
 970                  * Very large extends have the potential to lock up
 971                  * the cpu for extended periods of time.
 972                  */
 973                 cond_resched();
 974         }
 975
 976         return rc;
 977 }
 978
 979 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 980                       loff_t zero_to_size)
 981 {
 982         int ret = 0;
 983         u64 zero_start, range_start = 0, range_end = 0;
 984         struct super_block *sb = inode->i_sb;
 985
 986         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 987         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 988                                 (unsigned long long)zero_start,
 989                                 (unsigned long long)i_size_read(inode));
 990         while (zero_start < zero_to_size) {
 991                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 992                                                   zero_to_size,
 993                                                   &range_start,
 994                                                   &range_end);
 995                 if (ret) {
 996                         mlog_errno(ret);
 997                         break;
 998                 }
 999                 if (!range_end)
1000                         break;
1001                 /* Trim the ends */
1002                 if (range_start < zero_start)
1003                         range_start = zero_start;
1004                 if (range_end > zero_to_size)
1005                         range_end = zero_to_size;
1006
1007                 ret = ocfs2_zero_extend_range(inode, range_start,
1008                                               range_end, di_bh);
1009                 if (ret) {
1010                         mlog_errno(ret);
1011                         break;
1012                 }
1013                 zero_start = range_end;
1014         }
1015
1016         return ret;
1017 }
1018
1019 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1020                           u64 new_i_size, u64 zero_to)
1021 {
1022         int ret;
1023         u32 clusters_to_add;
1024         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1025
1026         /*
1027          * Only quota files call this without a bh, and they can't be
1028          * refcounted.
1029          */
1030         BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1031         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1032
1033         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1034         if (clusters_to_add < oi->ip_clusters)
1035                 clusters_to_add = 0;
1036         else
1037                 clusters_to_add -= oi->ip_clusters;
1038
1039         if (clusters_to_add) {
1040                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1041                                                 clusters_to_add, 0);
1042                 if (ret) {
1043                         mlog_errno(ret);
1044                         goto out;
1045                 }
1046         }
1047
1048         /*
1049          * Call this even if we don't add any clusters to the tree. We
1050          * still need to zero the area between the old i_size and the
1051          * new i_size.
1052          */
1053         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1054         if (ret < 0)
1055                 mlog_errno(ret);
1056
1057 out:
1058         return ret;
1059 }
1060
1061 static int ocfs2_extend_file(struct inode *inode,
1062                              struct buffer_head *di_bh,
1063                              u64 new_i_size)
1064 {
1065         int ret = 0;
1066         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1067
1068         BUG_ON(!di_bh);
1069
1070         /* setattr sometimes calls us like this. */
1071         if (new_i_size == 0)
1072                 goto out;
1073
1074         if (i_size_read(inode) == new_i_size)
1075                 goto out;
1076         BUG_ON(new_i_size < i_size_read(inode));
1077
1078         /*
1079          * The alloc sem blocks people in read/write from reading our
1080          * allocation until we're done changing it. We depend on
1081          * i_mutex to block other extend/truncate calls while we're
1082          * here.  We even have to hold it for sparse files because there
1083          * might be some tail zeroing.
1084          */
1085         down_write(&oi->ip_alloc_sem);
1086
1087         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1088                 /*
1089                  * We can optimize small extends by keeping the inodes
1090                  * inline data.
1091                  */
1092                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1093                         up_write(&oi->ip_alloc_sem);
1094                         goto out_update_size;
1095                 }
1096
1097                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1098                 if (ret) {
1099                         up_write(&oi->ip_alloc_sem);
1100                         mlog_errno(ret);
1101                         goto out;
1102                 }
1103         }
1104
1105         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1106                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1107         else
1108                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1109                                             new_i_size);
1110
1111         up_write(&oi->ip_alloc_sem);
1112
1113         if (ret < 0) {
1114                 mlog_errno(ret);
1115                 goto out;
1116         }
1117
1118 out_update_size:
1119         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1120         if (ret < 0)
1121                 mlog_errno(ret);
1122
1123 out:
1124         return ret;
1125 }
1126
1127 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1128 {
1129         int status = 0, size_change;
1130         int inode_locked = 0;
1131         struct inode *inode = d_inode(dentry);
1132         struct super_block *sb = inode->i_sb;
1133         struct ocfs2_super *osb = OCFS2_SB(sb);
1134         struct buffer_head *bh = NULL;
1135         handle_t *handle = NULL;
1136         struct dquot *transfer_to[MAXQUOTAS] = { };
1137         int qtype;
1138         int had_lock;
1139         struct ocfs2_lock_holder oh;
1140
1141         trace_ocfs2_setattr(inode, dentry,
1142                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1143                             dentry->d_name.len, dentry->d_name.name,
1144                             attr->ia_valid, attr->ia_mode,
1145                             from_kuid(&init_user_ns, attr->ia_uid),
1146                             from_kgid(&init_user_ns, attr->ia_gid));
1147
1148         /* ensuring we don't even attempt to truncate a symlink */
1149         if (S_ISLNK(inode->i_mode))
1150                 attr->ia_valid &= ~ATTR_SIZE;
1151
1152 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1153                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1154         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1155                 return 0;
1156
1157         status = setattr_prepare(dentry, attr);
1158         if (status)
1159                 return status;
1160
1161         if (is_quota_modification(inode, attr)) {
1162                 status = dquot_initialize(inode);
1163                 if (status)
1164                         return status;
1165         }
1166         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1167         if (size_change) {
1168                 /*
1169                  * Here we should wait dio to finish before inode lock
1170                  * to avoid a deadlock between ocfs2_setattr() and
1171                  * ocfs2_dio_end_io_write()
1172                  */
1173                 inode_dio_wait(inode);
1174
1175                 status = ocfs2_rw_lock(inode, 1);
1176                 if (status < 0) {
1177                         mlog_errno(status);
1178                         goto bail;
1179                 }
1180         }
1181
1182         had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1183         if (had_lock < 0) {
1184                 status = had_lock;
1185                 goto bail_unlock_rw;
1186         } else if (had_lock) {
1187                 /*
1188                  * As far as we know, ocfs2_setattr() could only be the first
1189                  * VFS entry point in the call chain of recursive cluster
1190                  * locking issue.
1191                  *
1192                  * For instance:
1193                  * chmod_common()
1194                  *  notify_change()
1195                  *   ocfs2_setattr()
1196                  *    posix_acl_chmod()
1197                  *     ocfs2_iop_get_acl()
1198                  *
1199                  * But, we're not 100% sure if it's always true, because the
1200                  * ordering of the VFS entry points in the call chain is out
1201                  * of our control. So, we'd better dump the stack here to
1202                  * catch the other cases of recursive locking.
1203                  */
1204                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1205                 dump_stack();
1206         }
1207         inode_locked = 1;
1208
1209         if (size_change) {
1210                 status = inode_newsize_ok(inode, attr->ia_size);
1211                 if (status)
1212                         goto bail_unlock;
1213
1214                 if (i_size_read(inode) >= attr->ia_size) {
1215                         if (ocfs2_should_order_data(inode)) {
1216                                 status = ocfs2_begin_ordered_truncate(inode,
1217                                                                       attr->ia_size);
1218                                 if (status)
1219                                         goto bail_unlock;
1220                         }
1221                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1222                 } else
1223                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1224                 if (status < 0) {
1225                         if (status != -ENOSPC)
1226                                 mlog_errno(status);
1227                         status = -ENOSPC;
1228                         goto bail_unlock;
1229                 }
1230         }
1231
1232         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1233             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1234                 /*
1235                  * Gather pointers to quota structures so that allocation /
1236                  * freeing of quota structures happens here and not inside
1237                  * dquot_transfer() where we have problems with lock ordering
1238                  */
1239                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1240                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1241                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1242                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1243                         if (IS_ERR(transfer_to[USRQUOTA])) {
1244                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1245                                 goto bail_unlock;
1246                         }
1247                 }
1248                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1249                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1250                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1251                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1252                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1253                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1254                                 goto bail_unlock;
1255                         }
1256                 }
1257                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1258                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1259                                            2 * ocfs2_quota_trans_credits(sb));
1260                 if (IS_ERR(handle)) {
1261                         status = PTR_ERR(handle);
1262                         mlog_errno(status);
1263                         goto bail_unlock_alloc;
1264                 }
1265                 status = __dquot_transfer(inode, transfer_to);
1266                 if (status < 0)
1267                         goto bail_commit;
1268         } else {
1269                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1270                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1271                 if (IS_ERR(handle)) {
1272                         status = PTR_ERR(handle);
1273                         mlog_errno(status);
1274                         goto bail_unlock_alloc;
1275                 }
1276         }
1277
1278         setattr_copy(inode, attr);
1279         mark_inode_dirty(inode);
1280
1281         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1282         if (status < 0)
1283                 mlog_errno(status);
1284
1285 bail_commit:
1286         ocfs2_commit_trans(osb, handle);
1287 bail_unlock_alloc:
1288         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1289 bail_unlock:
1290         if (status && inode_locked) {
1291                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1292                 inode_locked = 0;
1293         }
1294 bail_unlock_rw:
1295         if (size_change)
1296                 ocfs2_rw_unlock(inode, 1);
1297 bail:
1298
1299         /* Release quota pointers in case we acquired them */
1300         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1301                 dqput(transfer_to[qtype]);
1302
1303         if (!status && attr->ia_valid & ATTR_MODE) {
1304                 status = ocfs2_acl_chmod(inode, bh);
1305                 if (status < 0)
1306                         mlog_errno(status);
1307         }
1308         if (inode_locked)
1309                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1310
1311         brelse(bh);
1312         return status;
1313 }
1314
1315 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1316                   u32 request_mask, unsigned int flags)
1317 {
1318         struct inode *inode = d_inode(path->dentry);
1319         struct super_block *sb = path->dentry->d_sb;
1320         struct ocfs2_super *osb = sb->s_fs_info;
1321         int err;
1322
1323         err = ocfs2_inode_revalidate(path->dentry);
1324         if (err) {
1325                 if (err != -ENOENT)
1326                         mlog_errno(err);
1327                 goto bail;
1328         }
1329
1330         generic_fillattr(inode, stat);
1331         /*
1332          * If there is inline data in the inode, the inode will normally not
1333          * have data blocks allocated (it may have an external xattr block).
1334          * Report at least one sector for such files, so tools like tar, rsync,
1335          * others don't incorrectly think the file is completely sparse.
1336          */
1337         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1338                 stat->blocks += (stat->size + 511)>>9;
1339
1340         /* We set the blksize from the cluster size for performance */
1341         stat->blksize = osb->s_clustersize;
1342
1343 bail:
1344         return err;
1345 }
1346
1347 int ocfs2_permission(struct inode *inode, int mask)
1348 {
1349         int ret, had_lock;
1350         struct ocfs2_lock_holder oh;
1351
1352         if (mask & MAY_NOT_BLOCK)
1353                 return -ECHILD;
1354
1355         had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1356         if (had_lock < 0) {
1357                 ret = had_lock;
1358                 goto out;
1359         } else if (had_lock) {
1360                 /* See comments in ocfs2_setattr() for details.
1361                  * The call chain of this case could be:
1362                  * do_sys_open()
1363                  *  may_open()
1364                  *   inode_permission()
1365                  *    ocfs2_permission()
1366                  *     ocfs2_iop_get_acl()
1367                  */
1368                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1369                 dump_stack();
1370         }
1371
1372         ret = generic_permission(inode, mask);
1373
1374         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1375 out:
1376         return ret;
1377 }
1378
1379 static int __ocfs2_write_remove_suid(struct inode *inode,
1380                                      struct buffer_head *bh)
1381 {
1382         int ret;
1383         handle_t *handle;
1384         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1385         struct ocfs2_dinode *di;
1386
1387         trace_ocfs2_write_remove_suid(
1388                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1389                         inode->i_mode);
1390
1391         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1392         if (IS_ERR(handle)) {
1393                 ret = PTR_ERR(handle);
1394                 mlog_errno(ret);
1395                 goto out;
1396         }
1397
1398         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1399                                       OCFS2_JOURNAL_ACCESS_WRITE);
1400         if (ret < 0) {
1401                 mlog_errno(ret);
1402                 goto out_trans;
1403         }
1404
1405         inode->i_mode &= ~S_ISUID;
1406         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1407                 inode->i_mode &= ~S_ISGID;
1408
1409         di = (struct ocfs2_dinode *) bh->b_data;
1410         di->i_mode = cpu_to_le16(inode->i_mode);
1411         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1412
1413         ocfs2_journal_dirty(handle, bh);
1414
1415 out_trans:
1416         ocfs2_commit_trans(osb, handle);
1417 out:
1418         return ret;
1419 }
1420
1421 static int ocfs2_write_remove_suid(struct inode *inode)
1422 {
1423         int ret;
1424         struct buffer_head *bh = NULL;
1425
1426         ret = ocfs2_read_inode_block(inode, &bh);
1427         if (ret < 0) {
1428                 mlog_errno(ret);
1429                 goto out;
1430         }
1431
1432         ret =  __ocfs2_write_remove_suid(inode, bh);
1433 out:
1434         brelse(bh);
1435         return ret;
1436 }
1437
1438 /*
1439  * Allocate enough extents to cover the region starting at byte offset
1440  * start for len bytes. Existing extents are skipped, any extents
1441  * added are marked as "unwritten".
1442  */
1443 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1444                                             u64 start, u64 len)
1445 {
1446         int ret;
1447         u32 cpos, phys_cpos, clusters, alloc_size;
1448         u64 end = start + len;
1449         struct buffer_head *di_bh = NULL;
1450
1451         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1452                 ret = ocfs2_read_inode_block(inode, &di_bh);
1453                 if (ret) {
1454                         mlog_errno(ret);
1455                         goto out;
1456                 }
1457
1458                 /*
1459                  * Nothing to do if the requested reservation range
1460                  * fits within the inode.
1461                  */
1462                 if (ocfs2_size_fits_inline_data(di_bh, end))
1463                         goto out;
1464
1465                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1466                 if (ret) {
1467                         mlog_errno(ret);
1468                         goto out;
1469                 }
1470         }
1471
1472         /*
1473          * We consider both start and len to be inclusive.
1474          */
1475         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1476         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1477         clusters -= cpos;
1478
1479         while (clusters) {
1480                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1481                                          &alloc_size, NULL);
1482                 if (ret) {
1483                         mlog_errno(ret);
1484                         goto out;
1485                 }
1486
1487                 /*
1488                  * Hole or existing extent len can be arbitrary, so
1489                  * cap it to our own allocation request.
1490                  */
1491                 if (alloc_size > clusters)
1492                         alloc_size = clusters;
1493
1494                 if (phys_cpos) {
1495                         /*
1496                          * We already have an allocation at this
1497                          * region so we can safely skip it.
1498                          */
1499                         goto next;
1500                 }
1501
1502                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1503                 if (ret) {
1504                         if (ret != -ENOSPC)
1505                                 mlog_errno(ret);
1506                         goto out;
1507                 }
1508
1509 next:
1510                 cpos += alloc_size;
1511                 clusters -= alloc_size;
1512         }
1513
1514         ret = 0;
1515 out:
1516
1517         brelse(di_bh);
1518         return ret;
1519 }
1520
1521 /*
1522  * Truncate a byte range, avoiding pages within partial clusters. This
1523  * preserves those pages for the zeroing code to write to.
1524  */
1525 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1526                                          u64 byte_len)
1527 {
1528         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1529         loff_t start, end;
1530         struct address_space *mapping = inode->i_mapping;
1531
1532         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1533         end = byte_start + byte_len;
1534         end = end & ~(osb->s_clustersize - 1);
1535
1536         if (start < end) {
1537                 unmap_mapping_range(mapping, start, end - start, 0);
1538                 truncate_inode_pages_range(mapping, start, end - 1);
1539         }
1540 }
1541
1542 /*
1543  * zero out partial blocks of one cluster.
1544  *
1545  * start: file offset where zero starts, will be made upper block aligned.
1546  * len: it will be trimmed to the end of current cluster if "start + len"
1547  *      is bigger than it.
1548  */
1549 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1550                                         u64 start, u64 len)
1551 {
1552         int ret;
1553         u64 start_block, end_block, nr_blocks;
1554         u64 p_block, offset;
1555         u32 cluster, p_cluster, nr_clusters;
1556         struct super_block *sb = inode->i_sb;
1557         u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1558
1559         if (start + len < end)
1560                 end = start + len;
1561
1562         start_block = ocfs2_blocks_for_bytes(sb, start);
1563         end_block = ocfs2_blocks_for_bytes(sb, end);
1564         nr_blocks = end_block - start_block;
1565         if (!nr_blocks)
1566                 return 0;
1567
1568         cluster = ocfs2_bytes_to_clusters(sb, start);
1569         ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1570                                 &nr_clusters, NULL);
1571         if (ret)
1572                 return ret;
1573         if (!p_cluster)
1574                 return 0;
1575
1576         offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1577         p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1578         return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1579 }
1580
1581 static int ocfs2_zero_partial_clusters(struct inode *inode,
1582                                        u64 start, u64 len)
1583 {
1584         int ret = 0;
1585         u64 tmpend = 0;
1586         u64 end = start + len;
1587         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1588         unsigned int csize = osb->s_clustersize;
1589         handle_t *handle;
1590         loff_t isize = i_size_read(inode);
1591
1592         /*
1593          * The "start" and "end" values are NOT necessarily part of
1594          * the range whose allocation is being deleted. Rather, this
1595          * is what the user passed in with the request. We must zero
1596          * partial clusters here. There's no need to worry about
1597          * physical allocation - the zeroing code knows to skip holes.
1598          */
1599         trace_ocfs2_zero_partial_clusters(
1600                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1601                 (unsigned long long)start, (unsigned long long)end);
1602
1603         /*
1604          * If both edges are on a cluster boundary then there's no
1605          * zeroing required as the region is part of the allocation to
1606          * be truncated.
1607          */
1608         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1609                 goto out;
1610
1611         /* No page cache for EOF blocks, issue zero out to disk. */
1612         if (end > isize) {
1613                 /*
1614                  * zeroout eof blocks in last cluster starting from
1615                  * "isize" even "start" > "isize" because it is
1616                  * complicated to zeroout just at "start" as "start"
1617                  * may be not aligned with block size, buffer write
1618                  * would be required to do that, but out of eof buffer
1619                  * write is not supported.
1620                  */
1621                 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1622                                         end - isize);
1623                 if (ret) {
1624                         mlog_errno(ret);
1625                         goto out;
1626                 }
1627                 if (start >= isize)
1628                         goto out;
1629                 end = isize;
1630         }
1631         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1632         if (IS_ERR(handle)) {
1633                 ret = PTR_ERR(handle);
1634                 mlog_errno(ret);
1635                 goto out;
1636         }
1637
1638         /*
1639          * If start is on a cluster boundary and end is somewhere in another
1640          * cluster, we have not COWed the cluster starting at start, unless
1641          * end is also within the same cluster. So, in this case, we skip this
1642          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1643          * to the next one.
1644          */
1645         if ((start & (csize - 1)) != 0) {
1646                 /*
1647                  * We want to get the byte offset of the end of the 1st
1648                  * cluster.
1649                  */
1650                 tmpend = (u64)osb->s_clustersize +
1651                         (start & ~(osb->s_clustersize - 1));
1652                 if (tmpend > end)
1653                         tmpend = end;
1654
1655                 trace_ocfs2_zero_partial_clusters_range1(
1656                         (unsigned long long)start,
1657                         (unsigned long long)tmpend);
1658
1659                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1660                                                     tmpend);
1661                 if (ret)
1662                         mlog_errno(ret);
1663         }
1664
1665         if (tmpend < end) {
1666                 /*
1667                  * This may make start and end equal, but the zeroing
1668                  * code will skip any work in that case so there's no
1669                  * need to catch it up here.
1670                  */
1671                 start = end & ~(osb->s_clustersize - 1);
1672
1673                 trace_ocfs2_zero_partial_clusters_range2(
1674                         (unsigned long long)start, (unsigned long long)end);
1675
1676                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1677                 if (ret)
1678                         mlog_errno(ret);
1679         }
1680         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1681
1682         ocfs2_commit_trans(osb, handle);
1683 out:
1684         return ret;
1685 }
1686
1687 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1688 {
1689         int i;
1690         struct ocfs2_extent_rec *rec = NULL;
1691
1692         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1693
1694                 rec = &el->l_recs[i];
1695
1696                 if (le32_to_cpu(rec->e_cpos) < pos)
1697                         break;
1698         }
1699
1700         return i;
1701 }
1702
1703 /*
1704  * Helper to calculate the punching pos and length in one run, we handle the
1705  * following three cases in order:
1706  *
1707  * - remove the entire record
1708  * - remove a partial record
1709  * - no record needs to be removed (hole-punching completed)
1710 */
1711 static void ocfs2_calc_trunc_pos(struct inode *inode,
1712                                  struct ocfs2_extent_list *el,
1713                                  struct ocfs2_extent_rec *rec,
1714                                  u32 trunc_start, u32 *trunc_cpos,
1715                                  u32 *trunc_len, u32 *trunc_end,
1716                                  u64 *blkno, int *done)
1717 {
1718         int ret = 0;
1719         u32 coff, range;
1720
1721         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1722
1723         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1724                 /*
1725                  * remove an entire extent record.
1726                  */
1727                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1728                 /*
1729                  * Skip holes if any.
1730                  */
1731                 if (range < *trunc_end)
1732                         *trunc_end = range;
1733                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1734                 *blkno = le64_to_cpu(rec->e_blkno);
1735                 *trunc_end = le32_to_cpu(rec->e_cpos);
1736         } else if (range > trunc_start) {
1737                 /*
1738                  * remove a partial extent record, which means we're
1739                  * removing the last extent record.
1740                  */
1741                 *trunc_cpos = trunc_start;
1742                 /*
1743                  * skip hole if any.
1744                  */
1745                 if (range < *trunc_end)
1746                         *trunc_end = range;
1747                 *trunc_len = *trunc_end - trunc_start;
1748                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1749                 *blkno = le64_to_cpu(rec->e_blkno) +
1750                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1751                 *trunc_end = trunc_start;
1752         } else {
1753                 /*
1754                  * It may have two following possibilities:
1755                  *
1756                  * - last record has been removed
1757                  * - trunc_start was within a hole
1758                  *
1759                  * both two cases mean the completion of hole punching.
1760                  */
1761                 ret = 1;
1762         }
1763
1764         *done = ret;
1765 }
1766
1767 int ocfs2_remove_inode_range(struct inode *inode,
1768                              struct buffer_head *di_bh, u64 byte_start,
1769                              u64 byte_len)
1770 {
1771         int ret = 0, flags = 0, done = 0, i;
1772         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1773         u32 cluster_in_el;
1774         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1775         struct ocfs2_cached_dealloc_ctxt dealloc;
1776         struct address_space *mapping = inode->i_mapping;
1777         struct ocfs2_extent_tree et;
1778         struct ocfs2_path *path = NULL;
1779         struct ocfs2_extent_list *el = NULL;
1780         struct ocfs2_extent_rec *rec = NULL;
1781         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1782         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1783
1784         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1785         ocfs2_init_dealloc_ctxt(&dealloc);
1786
1787         trace_ocfs2_remove_inode_range(
1788                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1789                         (unsigned long long)byte_start,
1790                         (unsigned long long)byte_len);
1791
1792         if (byte_len == 0)
1793                 return 0;
1794
1795         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1796                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1797                                             byte_start + byte_len, 0);
1798                 if (ret) {
1799                         mlog_errno(ret);
1800                         goto out;
1801                 }
1802                 /*
1803                  * There's no need to get fancy with the page cache
1804                  * truncate of an inline-data inode. We're talking
1805                  * about less than a page here, which will be cached
1806                  * in the dinode buffer anyway.
1807                  */
1808                 unmap_mapping_range(mapping, 0, 0, 0);
1809                 truncate_inode_pages(mapping, 0);
1810                 goto out;
1811         }
1812
1813         /*
1814          * For reflinks, we may need to CoW 2 clusters which might be
1815          * partially zero'd later, if hole's start and end offset were
1816          * within one cluster(means is not exactly aligned to clustersize).
1817          */
1818
1819         if (ocfs2_is_refcount_inode(inode)) {
1820                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1821                 if (ret) {
1822                         mlog_errno(ret);
1823                         goto out;
1824                 }
1825
1826                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1827                 if (ret) {
1828                         mlog_errno(ret);
1829                         goto out;
1830                 }
1831         }
1832
1833         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1834         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1835         cluster_in_el = trunc_end;
1836
1837         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1838         if (ret) {
1839                 mlog_errno(ret);
1840                 goto out;
1841         }
1842
1843         path = ocfs2_new_path_from_et(&et);
1844         if (!path) {
1845                 ret = -ENOMEM;
1846                 mlog_errno(ret);
1847                 goto out;
1848         }
1849
1850         while (trunc_end > trunc_start) {
1851
1852                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1853                                       cluster_in_el);
1854                 if (ret) {
1855                         mlog_errno(ret);
1856                         goto out;
1857                 }
1858
1859                 el = path_leaf_el(path);
1860
1861                 i = ocfs2_find_rec(el, trunc_end);
1862                 /*
1863                  * Need to go to previous extent block.
1864                  */
1865                 if (i < 0) {
1866                         if (path->p_tree_depth == 0)
1867                                 break;
1868
1869                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1870                                                             path,
1871                                                             &cluster_in_el);
1872                         if (ret) {
1873                                 mlog_errno(ret);
1874                                 goto out;
1875                         }
1876
1877                         /*
1878                          * We've reached the leftmost extent block,
1879                          * it's safe to leave.
1880                          */
1881                         if (cluster_in_el == 0)
1882                                 break;
1883
1884                         /*
1885                          * The 'pos' searched for previous extent block is
1886                          * always one cluster less than actual trunc_end.
1887                          */
1888                         trunc_end = cluster_in_el + 1;
1889
1890                         ocfs2_reinit_path(path, 1);
1891
1892                         continue;
1893
1894                 } else
1895                         rec = &el->l_recs[i];
1896
1897                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1898                                      &trunc_len, &trunc_end, &blkno, &done);
1899                 if (done)
1900                         break;
1901
1902                 flags = rec->e_flags;
1903                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1904
1905                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1906                                                phys_cpos, trunc_len, flags,
1907                                                &dealloc, refcount_loc, false);
1908                 if (ret < 0) {
1909                         mlog_errno(ret);
1910                         goto out;
1911                 }
1912
1913                 cluster_in_el = trunc_end;
1914
1915                 ocfs2_reinit_path(path, 1);
1916         }
1917
1918         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1919
1920 out:
1921         ocfs2_free_path(path);
1922         ocfs2_schedule_truncate_log_flush(osb, 1);
1923         ocfs2_run_deallocs(osb, &dealloc);
1924
1925         return ret;
1926 }
1927
1928 /*
1929  * Parts of this function taken from xfs_change_file_space()
1930  */
1931 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1932                                      loff_t f_pos, unsigned int cmd,
1933                                      struct ocfs2_space_resv *sr,
1934                                      int change_size)
1935 {
1936         int ret;
1937         s64 llen;
1938         loff_t size, orig_isize;
1939         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1940         struct buffer_head *di_bh = NULL;
1941         handle_t *handle;
1942         unsigned long long max_off = inode->i_sb->s_maxbytes;
1943
1944         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1945                 return -EROFS;
1946
1947         inode_lock(inode);
1948
1949         /*
1950          * This prevents concurrent writes on other nodes
1951          */
1952         ret = ocfs2_rw_lock(inode, 1);
1953         if (ret) {
1954                 mlog_errno(ret);
1955                 goto out;
1956         }
1957
1958         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1959         if (ret) {
1960                 mlog_errno(ret);
1961                 goto out_rw_unlock;
1962         }
1963
1964         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1965                 ret = -EPERM;
1966                 goto out_inode_unlock;
1967         }
1968
1969         switch (sr->l_whence) {
1970         case 0: /*SEEK_SET*/
1971                 break;
1972         case 1: /*SEEK_CUR*/
1973                 sr->l_start += f_pos;
1974                 break;
1975         case 2: /*SEEK_END*/
1976                 sr->l_start += i_size_read(inode);
1977                 break;
1978         default:
1979                 ret = -EINVAL;
1980                 goto out_inode_unlock;
1981         }
1982         sr->l_whence = 0;
1983
1984         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1985
1986         if (sr->l_start < 0
1987             || sr->l_start > max_off
1988             || (sr->l_start + llen) < 0
1989             || (sr->l_start + llen) > max_off) {
1990                 ret = -EINVAL;
1991                 goto out_inode_unlock;
1992         }
1993         size = sr->l_start + sr->l_len;
1994
1995         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1996             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1997                 if (sr->l_len <= 0) {
1998                         ret = -EINVAL;
1999                         goto out_inode_unlock;
2000                 }
2001         }
2002
2003         if (file && should_remove_suid(file->f_path.dentry)) {
2004                 ret = __ocfs2_write_remove_suid(inode, di_bh);
2005                 if (ret) {
2006                         mlog_errno(ret);
2007                         goto out_inode_unlock;
2008                 }
2009         }
2010
2011         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2012         switch (cmd) {
2013         case OCFS2_IOC_RESVSP:
2014         case OCFS2_IOC_RESVSP64:
2015                 /*
2016                  * This takes unsigned offsets, but the signed ones we
2017                  * pass have been checked against overflow above.
2018                  */
2019                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2020                                                        sr->l_len);
2021                 break;
2022         case OCFS2_IOC_UNRESVSP:
2023         case OCFS2_IOC_UNRESVSP64:
2024                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2025                                                sr->l_len);
2026                 break;
2027         default:
2028                 ret = -EINVAL;
2029         }
2030
2031         orig_isize = i_size_read(inode);
2032         /* zeroout eof blocks in the cluster. */
2033         if (!ret && change_size && orig_isize < size) {
2034                 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2035                                         size - orig_isize);
2036                 if (!ret)
2037                         i_size_write(inode, size);
2038         }
2039         up_write(&OCFS2_I(inode)->ip_alloc_sem);
2040         if (ret) {
2041                 mlog_errno(ret);
2042                 goto out_inode_unlock;
2043         }
2044
2045         /*
2046          * We update c/mtime for these changes
2047          */
2048         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2049         if (IS_ERR(handle)) {
2050                 ret = PTR_ERR(handle);
2051                 mlog_errno(ret);
2052                 goto out_inode_unlock;
2053         }
2054
2055         inode->i_ctime = inode->i_mtime = current_time(inode);
2056         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2057         if (ret < 0)
2058                 mlog_errno(ret);
2059
2060         if (file && (file->f_flags & O_SYNC))
2061                 handle->h_sync = 1;
2062
2063         ocfs2_commit_trans(osb, handle);
2064
2065 out_inode_unlock:
2066         brelse(di_bh);
2067         ocfs2_inode_unlock(inode, 1);
2068 out_rw_unlock:
2069         ocfs2_rw_unlock(inode, 1);
2070
2071 out:
2072         inode_unlock(inode);
2073         return ret;
2074 }
2075
2076 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2077                             struct ocfs2_space_resv *sr)
2078 {
2079         struct inode *inode = file_inode(file);
2080         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2081         int ret;
2082
2083         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2084             !ocfs2_writes_unwritten_extents(osb))
2085                 return -ENOTTY;
2086         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2087                  !ocfs2_sparse_alloc(osb))
2088                 return -ENOTTY;
2089
2090         if (!S_ISREG(inode->i_mode))
2091                 return -EINVAL;
2092
2093         if (!(file->f_mode & FMODE_WRITE))
2094                 return -EBADF;
2095
2096         ret = mnt_want_write_file(file);
2097         if (ret)
2098                 return ret;
2099         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2100         mnt_drop_write_file(file);
2101         return ret;
2102 }
2103
2104 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2105                             loff_t len)
2106 {
2107         struct inode *inode = file_inode(file);
2108         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2109         struct ocfs2_space_resv sr;
2110         int change_size = 1;
2111         int cmd = OCFS2_IOC_RESVSP64;
2112         int ret = 0;
2113
2114         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2115                 return -EOPNOTSUPP;
2116         if (!ocfs2_writes_unwritten_extents(osb))
2117                 return -EOPNOTSUPP;
2118
2119         if (mode & FALLOC_FL_KEEP_SIZE) {
2120                 change_size = 0;
2121         } else {
2122                 ret = inode_newsize_ok(inode, offset + len);
2123                 if (ret)
2124                         return ret;
2125         }
2126
2127         if (mode & FALLOC_FL_PUNCH_HOLE)
2128                 cmd = OCFS2_IOC_UNRESVSP64;
2129
2130         sr.l_whence = 0;
2131         sr.l_start = (s64)offset;
2132         sr.l_len = (s64)len;
2133
2134         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2135                                          change_size);
2136 }
2137
2138 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2139                                    size_t count)
2140 {
2141         int ret = 0;
2142         unsigned int extent_flags;
2143         u32 cpos, clusters, extent_len, phys_cpos;
2144         struct super_block *sb = inode->i_sb;
2145
2146         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2147             !ocfs2_is_refcount_inode(inode) ||
2148             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2149                 return 0;
2150
2151         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2152         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2153
2154         while (clusters) {
2155                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2156                                          &extent_flags);
2157                 if (ret < 0) {
2158                         mlog_errno(ret);
2159                         goto out;
2160                 }
2161
2162                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2163                         ret = 1;
2164                         break;
2165                 }
2166
2167                 if (extent_len > clusters)
2168                         extent_len = clusters;
2169
2170                 clusters -= extent_len;
2171                 cpos += extent_len;
2172         }
2173 out:
2174         return ret;
2175 }
2176
2177 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2178 {
2179         int blockmask = inode->i_sb->s_blocksize - 1;
2180         loff_t final_size = pos + count;
2181
2182         if ((pos & blockmask) || (final_size & blockmask))
2183                 return 1;
2184         return 0;
2185 }
2186
2187 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2188                                             struct file *file,
2189                                             loff_t pos, size_t count,
2190                                             int *meta_level)
2191 {
2192         int ret;
2193         struct buffer_head *di_bh = NULL;
2194         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2195         u32 clusters =
2196                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2197
2198         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2199         if (ret) {
2200                 mlog_errno(ret);
2201                 goto out;
2202         }
2203
2204         *meta_level = 1;
2205
2206         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2207         if (ret)
2208                 mlog_errno(ret);
2209 out:
2210         brelse(di_bh);
2211         return ret;
2212 }
2213
2214 static int ocfs2_prepare_inode_for_write(struct file *file,
2215                                          loff_t pos,
2216                                          size_t count)
2217 {
2218         int ret = 0, meta_level = 0;
2219         struct dentry *dentry = file->f_path.dentry;
2220         struct inode *inode = d_inode(dentry);
2221         loff_t end;
2222
2223         /*
2224          * We start with a read level meta lock and only jump to an ex
2225          * if we need to make modifications here.
2226          */
2227         for(;;) {
2228                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2229                 if (ret < 0) {
2230                         meta_level = -1;
2231                         mlog_errno(ret);
2232                         goto out;
2233                 }
2234
2235                 /* Clear suid / sgid if necessary. We do this here
2236                  * instead of later in the write path because
2237                  * remove_suid() calls ->setattr without any hint that
2238                  * we may have already done our cluster locking. Since
2239                  * ocfs2_setattr() *must* take cluster locks to
2240                  * proceed, this will lead us to recursively lock the
2241                  * inode. There's also the dinode i_size state which
2242                  * can be lost via setattr during extending writes (we
2243                  * set inode->i_size at the end of a write. */
2244                 if (should_remove_suid(dentry)) {
2245                         if (meta_level == 0) {
2246                                 ocfs2_inode_unlock(inode, meta_level);
2247                                 meta_level = 1;
2248                                 continue;
2249                         }
2250
2251                         ret = ocfs2_write_remove_suid(inode);
2252                         if (ret < 0) {
2253                                 mlog_errno(ret);
2254                                 goto out_unlock;
2255                         }
2256                 }
2257
2258                 end = pos + count;
2259
2260                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2261                 if (ret == 1) {
2262                         ocfs2_inode_unlock(inode, meta_level);
2263                         meta_level = -1;
2264
2265                         ret = ocfs2_prepare_inode_for_refcount(inode,
2266                                                                file,
2267                                                                pos,
2268                                                                count,
2269                                                                &meta_level);
2270                 }
2271
2272                 if (ret < 0) {
2273                         mlog_errno(ret);
2274                         goto out_unlock;
2275                 }
2276
2277                 break;
2278         }
2279
2280 out_unlock:
2281         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2282                                             pos, count);
2283
2284         if (meta_level >= 0)
2285                 ocfs2_inode_unlock(inode, meta_level);
2286
2287 out:
2288         return ret;
2289 }
2290
2291 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2292                                     struct iov_iter *from)
2293 {
2294         int direct_io, rw_level;
2295         ssize_t written = 0;
2296         ssize_t ret;
2297         size_t count = iov_iter_count(from);
2298         struct file *file = iocb->ki_filp;
2299         struct inode *inode = file_inode(file);
2300         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2301         int full_coherency = !(osb->s_mount_opt &
2302                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2303         void *saved_ki_complete = NULL;
2304         int append_write = ((iocb->ki_pos + count) >=
2305                         i_size_read(inode) ? 1 : 0);
2306
2307         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2308                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2309                 file->f_path.dentry->d_name.len,
2310                 file->f_path.dentry->d_name.name,
2311                 (unsigned int)from->nr_segs);   /* GRRRRR */
2312
2313         if (count == 0)
2314                 return 0;
2315
2316         direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2317
2318         inode_lock(inode);
2319
2320         /*
2321          * Concurrent O_DIRECT writes are allowed with
2322          * mount_option "coherency=buffered".
2323          * For append write, we must take rw EX.
2324          */
2325         rw_level = (!direct_io || full_coherency || append_write);
2326
2327         ret = ocfs2_rw_lock(inode, rw_level);
2328         if (ret < 0) {
2329                 mlog_errno(ret);
2330                 goto out_mutex;
2331         }
2332
2333         /*
2334          * O_DIRECT writes with "coherency=full" need to take EX cluster
2335          * inode_lock to guarantee coherency.
2336          */
2337         if (direct_io && full_coherency) {
2338                 /*
2339                  * We need to take and drop the inode lock to force
2340                  * other nodes to drop their caches.  Buffered I/O
2341                  * already does this in write_begin().
2342                  */
2343                 ret = ocfs2_inode_lock(inode, NULL, 1);
2344                 if (ret < 0) {
2345                         mlog_errno(ret);
2346                         goto out;
2347                 }
2348
2349                 ocfs2_inode_unlock(inode, 1);
2350         }
2351
2352         ret = generic_write_checks(iocb, from);
2353         if (ret <= 0) {
2354                 if (ret)
2355                         mlog_errno(ret);
2356                 goto out;
2357         }
2358         count = ret;
2359
2360         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count);
2361         if (ret < 0) {
2362                 mlog_errno(ret);
2363                 goto out;
2364         }
2365
2366         if (direct_io && !is_sync_kiocb(iocb) &&
2367             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2368                 /*
2369                  * Make it a sync io if it's an unaligned aio.
2370                  */
2371                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2372         }
2373
2374         /* communicate with ocfs2_dio_end_io */
2375         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2376
2377         written = __generic_file_write_iter(iocb, from);
2378         /* buffered aio wouldn't have proper lock coverage today */
2379         BUG_ON(written == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2380
2381         /*
2382          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2383          * function pointer which is called when o_direct io completes so that
2384          * it can unlock our rw lock.
2385          * Unfortunately there are error cases which call end_io and others
2386          * that don't.  so we don't have to unlock the rw_lock if either an
2387          * async dio is going to do it in the future or an end_io after an
2388          * error has already done it.
2389          */
2390         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2391                 rw_level = -1;
2392         }
2393
2394         if (unlikely(written <= 0))
2395                 goto out;
2396
2397         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2398             IS_SYNC(inode)) {
2399                 ret = filemap_fdatawrite_range(file->f_mapping,
2400                                                iocb->ki_pos - written,
2401                                                iocb->ki_pos - 1);
2402                 if (ret < 0)
2403                         written = ret;
2404
2405                 if (!ret) {
2406                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2407                         if (ret < 0)
2408                                 written = ret;
2409                 }
2410
2411                 if (!ret)
2412                         ret = filemap_fdatawait_range(file->f_mapping,
2413                                                       iocb->ki_pos - written,
2414                                                       iocb->ki_pos - 1);
2415         }
2416
2417 out:
2418         if (saved_ki_complete)
2419                 xchg(&iocb->ki_complete, saved_ki_complete);
2420
2421         if (rw_level != -1)
2422                 ocfs2_rw_unlock(inode, rw_level);
2423
2424 out_mutex:
2425         inode_unlock(inode);
2426
2427         if (written)
2428                 ret = written;
2429         return ret;
2430 }
2431
2432 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2433                                    struct iov_iter *to)
2434 {
2435         int ret = 0, rw_level = -1, lock_level = 0;
2436         struct file *filp = iocb->ki_filp;
2437         struct inode *inode = file_inode(filp);
2438
2439         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2440                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2441                         filp->f_path.dentry->d_name.len,
2442                         filp->f_path.dentry->d_name.name,
2443                         to->nr_segs);   /* GRRRRR */
2444
2445
2446         if (!inode) {
2447                 ret = -EINVAL;
2448                 mlog_errno(ret);
2449                 goto bail;
2450         }
2451
2452         /*
2453          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2454          * need locks to protect pending reads from racing with truncate.
2455          */
2456         if (iocb->ki_flags & IOCB_DIRECT) {
2457                 ret = ocfs2_rw_lock(inode, 0);
2458                 if (ret < 0) {
2459                         mlog_errno(ret);
2460                         goto bail;
2461                 }
2462                 rw_level = 0;
2463                 /* communicate with ocfs2_dio_end_io */
2464                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2465         }
2466
2467         /*
2468          * We're fine letting folks race truncates and extending
2469          * writes with read across the cluster, just like they can
2470          * locally. Hence no rw_lock during read.
2471          *
2472          * Take and drop the meta data lock to update inode fields
2473          * like i_size. This allows the checks down below
2474          * generic_file_aio_read() a chance of actually working.
2475          */
2476         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2477         if (ret < 0) {
2478                 mlog_errno(ret);
2479                 goto bail;
2480         }
2481         ocfs2_inode_unlock(inode, lock_level);
2482
2483         ret = generic_file_read_iter(iocb, to);
2484         trace_generic_file_aio_read_ret(ret);
2485
2486         /* buffered aio wouldn't have proper lock coverage today */
2487         BUG_ON(ret == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2488
2489         /* see ocfs2_file_write_iter */
2490         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2491                 rw_level = -1;
2492         }
2493
2494 bail:
2495         if (rw_level != -1)
2496                 ocfs2_rw_unlock(inode, rw_level);
2497
2498         return ret;
2499 }
2500
2501 /* Refer generic_file_llseek_unlocked() */
2502 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2503 {
2504         struct inode *inode = file->f_mapping->host;
2505         int ret = 0;
2506
2507         inode_lock(inode);
2508
2509         switch (whence) {
2510         case SEEK_SET:
2511                 break;
2512         case SEEK_END:
2513                 /* SEEK_END requires the OCFS2 inode lock for the file
2514                  * because it references the file's size.
2515                  */
2516                 ret = ocfs2_inode_lock(inode, NULL, 0);
2517                 if (ret < 0) {
2518                         mlog_errno(ret);
2519                         goto out;
2520                 }
2521                 offset += i_size_read(inode);
2522                 ocfs2_inode_unlock(inode, 0);
2523                 break;
2524         case SEEK_CUR:
2525                 if (offset == 0) {
2526                         offset = file->f_pos;
2527                         goto out;
2528                 }
2529                 offset += file->f_pos;
2530                 break;
2531         case SEEK_DATA:
2532         case SEEK_HOLE:
2533                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2534                 if (ret)
2535                         goto out;
2536                 break;
2537         default:
2538                 ret = -EINVAL;
2539                 goto out;
2540         }
2541
2542         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2543
2544 out:
2545         inode_unlock(inode);
2546         if (ret)
2547                 return ret;
2548         return offset;
2549 }
2550
2551 static int ocfs2_file_clone_range(struct file *file_in,
2552                                   loff_t pos_in,
2553                                   struct file *file_out,
2554                                   loff_t pos_out,
2555                                   u64 len)
2556 {
2557         return ocfs2_reflink_remap_range(file_in, pos_in, file_out, pos_out,
2558                                          len, false);
2559 }
2560
2561 static ssize_t ocfs2_file_dedupe_range(struct file *src_file,
2562                                        u64 loff,
2563                                        u64 len,
2564                                        struct file *dst_file,
2565                                        u64 dst_loff)
2566 {
2567         int error;
2568
2569         error = ocfs2_reflink_remap_range(src_file, loff, dst_file, dst_loff,
2570                                           len, true);
2571         if (error)
2572                 return error;
2573         return len;
2574 }
2575
2576 const struct inode_operations ocfs2_file_iops = {
2577         .setattr        = ocfs2_setattr,
2578         .getattr        = ocfs2_getattr,
2579         .permission     = ocfs2_permission,
2580         .listxattr      = ocfs2_listxattr,
2581         .fiemap         = ocfs2_fiemap,
2582         .get_acl        = ocfs2_iop_get_acl,
2583         .set_acl        = ocfs2_iop_set_acl,
2584 };
2585
2586 const struct inode_operations ocfs2_special_file_iops = {
2587         .setattr        = ocfs2_setattr,
2588         .getattr        = ocfs2_getattr,
2589         .permission     = ocfs2_permission,
2590         .get_acl        = ocfs2_iop_get_acl,
2591         .set_acl        = ocfs2_iop_set_acl,
2592 };
2593
2594 /*
2595  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2596  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2597  */
2598 const struct file_operations ocfs2_fops = {
2599         .llseek         = ocfs2_file_llseek,
2600         .mmap           = ocfs2_mmap,
2601         .fsync          = ocfs2_sync_file,
2602         .release        = ocfs2_file_release,
2603         .open           = ocfs2_file_open,
2604         .read_iter      = ocfs2_file_read_iter,
2605         .write_iter     = ocfs2_file_write_iter,
2606         .unlocked_ioctl = ocfs2_ioctl,
2607 #ifdef CONFIG_COMPAT
2608         .compat_ioctl   = ocfs2_compat_ioctl,
2609 #endif
2610         .lock           = ocfs2_lock,
2611         .flock          = ocfs2_flock,
2612         .splice_read    = generic_file_splice_read,
2613         .splice_write   = iter_file_splice_write,
2614         .fallocate      = ocfs2_fallocate,
2615         .clone_file_range = ocfs2_file_clone_range,
2616         .dedupe_file_range = ocfs2_file_dedupe_range,
2617 };
2618
2619 const struct file_operations ocfs2_dops = {
2620         .llseek         = generic_file_llseek,
2621         .read           = generic_read_dir,
2622         .iterate        = ocfs2_readdir,
2623         .fsync          = ocfs2_sync_file,
2624         .release        = ocfs2_dir_release,
2625         .open           = ocfs2_dir_open,
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 };
2633
2634 /*
2635  * POSIX-lockless variants of our file_operations.
2636  *
2637  * These will be used if the underlying cluster stack does not support
2638  * posix file locking, if the user passes the "localflocks" mount
2639  * option, or if we have a local-only fs.
2640  *
2641  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2642  * so we still want it in the case of no stack support for
2643  * plocks. Internally, it will do the right thing when asked to ignore
2644  * the cluster.
2645  */
2646 const struct file_operations ocfs2_fops_no_plocks = {
2647         .llseek         = ocfs2_file_llseek,
2648         .mmap           = ocfs2_mmap,
2649         .fsync          = ocfs2_sync_file,
2650         .release        = ocfs2_file_release,
2651         .open           = ocfs2_file_open,
2652         .read_iter      = ocfs2_file_read_iter,
2653         .write_iter     = ocfs2_file_write_iter,
2654         .unlocked_ioctl = ocfs2_ioctl,
2655 #ifdef CONFIG_COMPAT
2656         .compat_ioctl   = ocfs2_compat_ioctl,
2657 #endif
2658         .flock          = ocfs2_flock,
2659         .splice_read    = generic_file_splice_read,
2660         .splice_write   = iter_file_splice_write,
2661         .fallocate      = ocfs2_fallocate,
2662         .clone_file_range = ocfs2_file_clone_range,
2663         .dedupe_file_range = ocfs2_file_dedupe_range,
2664 };
2665
2666 const struct file_operations ocfs2_dops_no_plocks = {
2667         .llseek         = generic_file_llseek,
2668         .read           = generic_read_dir,
2669         .iterate        = ocfs2_readdir,
2670         .fsync          = ocfs2_sync_file,
2671         .release        = ocfs2_dir_release,
2672         .open           = ocfs2_dir_open,
2673         .unlocked_ioctl = ocfs2_ioctl,
2674 #ifdef CONFIG_COMPAT
2675         .compat_ioctl   = ocfs2_compat_ioctl,
2676 #endif
2677         .flock          = ocfs2_flock,
2678 };