fs/ocfs2/file.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * file.c
   6  *
   7  * File open, close, extend, truncate
   8  *
   9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
  10  */
  11
  12 #include <linux/capability.h>
  13 #include <linux/fs.h>
  14 #include <linux/types.h>
  15 #include <linux/slab.h>
  16 #include <linux/highmem.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/uio.h>
  19 #include <linux/sched.h>
  20 #include <linux/splice.h>
  21 #include <linux/mount.h>
  22 #include <linux/writeback.h>
  23 #include <linux/falloc.h>
  24 #include <linux/quotaops.h>
  25 #include <linux/blkdev.h>
  26 #include <linux/backing-dev.h>
  27
  28 #include <cluster/masklog.h>
  29
  30 #include "ocfs2.h"
  31
  32 #include "alloc.h"
  33 #include "aops.h"
  34 #include "dir.h"
  35 #include "dlmglue.h"
  36 #include "extent_map.h"
  37 #include "file.h"
  38 #include "sysfile.h"
  39 #include "inode.h"
  40 #include "ioctl.h"
  41 #include "journal.h"
  42 #include "locks.h"
  43 #include "mmap.h"
  44 #include "suballoc.h"
  45 #include "super.h"
  46 #include "xattr.h"
  47 #include "acl.h"
  48 #include "quota.h"
  49 #include "refcounttree.h"
  50 #include "ocfs2_trace.h"
  51
  52 #include "buffer_head_io.h"
  53
  54 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  55 {
  56         struct ocfs2_file_private *fp;
  57
  58         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  59         if (!fp)
  60                 return -ENOMEM;
  61
  62         fp->fp_file = file;
  63         mutex_init(&fp->fp_mutex);
  64         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  65         file->private_data = fp;
  66
  67         return 0;
  68 }
  69
  70 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  71 {
  72         struct ocfs2_file_private *fp = file->private_data;
  73         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  74
  75         if (fp) {
  76                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  77                 ocfs2_lock_res_free(&fp->fp_flock);
  78                 kfree(fp);
  79                 file->private_data = NULL;
  80         }
  81 }
  82
  83 static int ocfs2_file_open(struct inode *inode, struct file *file)
  84 {
  85         int status;
  86         int mode = file->f_flags;
  87         struct ocfs2_inode_info *oi = OCFS2_I(inode);
  88
  89         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
  90                               (unsigned long long)oi->ip_blkno,
  91                               file->f_path.dentry->d_name.len,
  92                               file->f_path.dentry->d_name.name, mode);
  93
  94         if (file->f_mode & FMODE_WRITE) {
  95                 status = dquot_initialize(inode);
  96                 if (status)
  97                         goto leave;
  98         }
  99
 100         spin_lock(&oi->ip_lock);
 101
 102         /* Check that the inode hasn't been wiped from disk by another
 103          * node. If it hasn't then we're safe as long as we hold the
 104          * spin lock until our increment of open count. */
 105         if (oi->ip_flags & OCFS2_INODE_DELETED) {
 106                 spin_unlock(&oi->ip_lock);
 107
 108                 status = -ENOENT;
 109                 goto leave;
 110         }
 111
 112         if (mode & O_DIRECT)
 113                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 114
 115         oi->ip_open_count++;
 116         spin_unlock(&oi->ip_lock);
 117
 118         status = ocfs2_init_file_private(inode, file);
 119         if (status) {
 120                 /*
 121                  * We want to set open count back if we're failing the
 122                  * open.
 123                  */
 124                 spin_lock(&oi->ip_lock);
 125                 oi->ip_open_count--;
 126                 spin_unlock(&oi->ip_lock);
 127         }
 128
 129         file->f_mode |= FMODE_NOWAIT;
 130
 131 leave:
 132         return status;
 133 }
 134
 135 static int ocfs2_file_release(struct inode *inode, struct file *file)
 136 {
 137         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 138
 139         spin_lock(&oi->ip_lock);
 140         if (!--oi->ip_open_count)
 141                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 142
 143         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 144                                  oi->ip_blkno,
 145                                  file->f_path.dentry->d_name.len,
 146                                  file->f_path.dentry->d_name.name,
 147                                  oi->ip_open_count);
 148         spin_unlock(&oi->ip_lock);
 149
 150         ocfs2_free_file_private(inode, file);
 151
 152         return 0;
 153 }
 154
 155 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 156 {
 157         return ocfs2_init_file_private(inode, file);
 158 }
 159
 160 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 161 {
 162         ocfs2_free_file_private(inode, file);
 163         return 0;
 164 }
 165
 166 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 167                            int datasync)
 168 {
 169         int err = 0;
 170         struct inode *inode = file->f_mapping->host;
 171         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 172         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 173         journal_t *journal = osb->journal->j_journal;
 174         int ret;
 175         tid_t commit_tid;
 176         bool needs_barrier = false;
 177
 178         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 179                               oi->ip_blkno,
 180                               file->f_path.dentry->d_name.len,
 181                               file->f_path.dentry->d_name.name,
 182                               (unsigned long long)datasync);
 183
 184         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 185                 return -EROFS;
 186
 187         err = file_write_and_wait_range(file, start, end);
 188         if (err)
 189                 return err;
 190
 191         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 192         if (journal->j_flags & JBD2_BARRIER &&
 193             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 194                 needs_barrier = true;
 195         err = jbd2_complete_transaction(journal, commit_tid);
 196         if (needs_barrier) {
 197                 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 198                 if (!err)
 199                         err = ret;
 200         }
 201
 202         if (err)
 203                 mlog_errno(err);
 204
 205         return (err < 0) ? -EIO : 0;
 206 }
 207
 208 int ocfs2_should_update_atime(struct inode *inode,
 209                               struct vfsmount *vfsmnt)
 210 {
 211         struct timespec64 now;
 212         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 213
 214         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 215                 return 0;
 216
 217         if ((inode->i_flags & S_NOATIME) ||
 218             ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
 219                 return 0;
 220
 221         /*
 222          * We can be called with no vfsmnt structure - NFSD will
 223          * sometimes do this.
 224          *
 225          * Note that our action here is different than touch_atime() -
 226          * if we can't tell whether this is a noatime mount, then we
 227          * don't know whether to trust the value of s_atime_quantum.
 228          */
 229         if (vfsmnt == NULL)
 230                 return 0;
 231
 232         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 233             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 234                 return 0;
 235
 236         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 237                 if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 238                     (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 239                         return 1;
 240
 241                 return 0;
 242         }
 243
 244         now = current_time(inode);
 245         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 246                 return 0;
 247         else
 248                 return 1;
 249 }
 250
 251 int ocfs2_update_inode_atime(struct inode *inode,
 252                              struct buffer_head *bh)
 253 {
 254         int ret;
 255         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 256         handle_t *handle;
 257         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 258
 259         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 260         if (IS_ERR(handle)) {
 261                 ret = PTR_ERR(handle);
 262                 mlog_errno(ret);
 263                 goto out;
 264         }
 265
 266         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 267                                       OCFS2_JOURNAL_ACCESS_WRITE);
 268         if (ret) {
 269                 mlog_errno(ret);
 270                 goto out_commit;
 271         }
 272
 273         /*
 274          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 275          * have i_mutex to guard against concurrent changes to other
 276          * inode fields.
 277          */
 278         inode->i_atime = current_time(inode);
 279         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 280         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 281         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 282         ocfs2_journal_dirty(handle, bh);
 283
 284 out_commit:
 285         ocfs2_commit_trans(osb, handle);
 286 out:
 287         return ret;
 288 }
 289
 290 int ocfs2_set_inode_size(handle_t *handle,
 291                                 struct inode *inode,
 292                                 struct buffer_head *fe_bh,
 293                                 u64 new_i_size)
 294 {
 295         int status;
 296
 297         i_size_write(inode, new_i_size);
 298         inode->i_blocks = ocfs2_inode_sector_count(inode);
 299         inode->i_ctime = inode->i_mtime = current_time(inode);
 300
 301         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 302         if (status < 0) {
 303                 mlog_errno(status);
 304                 goto bail;
 305         }
 306
 307 bail:
 308         return status;
 309 }
 310
 311 int ocfs2_simple_size_update(struct inode *inode,
 312                              struct buffer_head *di_bh,
 313                              u64 new_i_size)
 314 {
 315         int ret;
 316         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 317         handle_t *handle = NULL;
 318
 319         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 320         if (IS_ERR(handle)) {
 321                 ret = PTR_ERR(handle);
 322                 mlog_errno(ret);
 323                 goto out;
 324         }
 325
 326         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 327                                    new_i_size);
 328         if (ret < 0)
 329                 mlog_errno(ret);
 330
 331         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 332         ocfs2_commit_trans(osb, handle);
 333 out:
 334         return ret;
 335 }
 336
 337 static int ocfs2_cow_file_pos(struct inode *inode,
 338                               struct buffer_head *fe_bh,
 339                               u64 offset)
 340 {
 341         int status;
 342         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 343         unsigned int num_clusters = 0;
 344         unsigned int ext_flags = 0;
 345
 346         /*
 347          * If the new offset is aligned to the range of the cluster, there is
 348          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 349          * CoW either.
 350          */
 351         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 352                 return 0;
 353
 354         status = ocfs2_get_clusters(inode, cpos, &phys,
 355                                     &num_clusters, &ext_flags);
 356         if (status) {
 357                 mlog_errno(status);
 358                 goto out;
 359         }
 360
 361         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 362                 goto out;
 363
 364         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 365
 366 out:
 367         return status;
 368 }
 369
 370 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 371                                      struct inode *inode,
 372                                      struct buffer_head *fe_bh,
 373                                      u64 new_i_size)
 374 {
 375         int status;
 376         handle_t *handle;
 377         struct ocfs2_dinode *di;
 378         u64 cluster_bytes;
 379
 380         /*
 381          * We need to CoW the cluster contains the offset if it is reflinked
 382          * since we will call ocfs2_zero_range_for_truncate later which will
 383          * write "0" from offset to the end of the cluster.
 384          */
 385         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 386         if (status) {
 387                 mlog_errno(status);
 388                 return status;
 389         }
 390
 391         /* TODO: This needs to actually orphan the inode in this
 392          * transaction. */
 393
 394         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 395         if (IS_ERR(handle)) {
 396                 status = PTR_ERR(handle);
 397                 mlog_errno(status);
 398                 goto out;
 399         }
 400
 401         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 402                                          OCFS2_JOURNAL_ACCESS_WRITE);
 403         if (status < 0) {
 404                 mlog_errno(status);
 405                 goto out_commit;
 406         }
 407
 408         /*
 409          * Do this before setting i_size.
 410          */
 411         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 412         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 413                                                cluster_bytes);
 414         if (status) {
 415                 mlog_errno(status);
 416                 goto out_commit;
 417         }
 418
 419         i_size_write(inode, new_i_size);
 420         inode->i_ctime = inode->i_mtime = current_time(inode);
 421
 422         di = (struct ocfs2_dinode *) fe_bh->b_data;
 423         di->i_size = cpu_to_le64(new_i_size);
 424         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 425         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 426         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 427
 428         ocfs2_journal_dirty(handle, fe_bh);
 429
 430 out_commit:
 431         ocfs2_commit_trans(osb, handle);
 432 out:
 433         return status;
 434 }
 435
 436 int ocfs2_truncate_file(struct inode *inode,
 437                                struct buffer_head *di_bh,
 438                                u64 new_i_size)
 439 {
 440         int status = 0;
 441         struct ocfs2_dinode *fe = NULL;
 442         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 443
 444         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 445          * already validated it */
 446         fe = (struct ocfs2_dinode *) di_bh->b_data;
 447
 448         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 449                                   (unsigned long long)le64_to_cpu(fe->i_size),
 450                                   (unsigned long long)new_i_size);
 451
 452         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 453                         "Inode %llu, inode i_size = %lld != di "
 454                         "i_size = %llu, i_flags = 0x%x\n",
 455                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 456                         i_size_read(inode),
 457                         (unsigned long long)le64_to_cpu(fe->i_size),
 458                         le32_to_cpu(fe->i_flags));
 459
 460         if (new_i_size > le64_to_cpu(fe->i_size)) {
 461                 trace_ocfs2_truncate_file_error(
 462                         (unsigned long long)le64_to_cpu(fe->i_size),
 463                         (unsigned long long)new_i_size);
 464                 status = -EINVAL;
 465                 mlog_errno(status);
 466                 goto bail;
 467         }
 468
 469         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 470
 471         ocfs2_resv_discard(&osb->osb_la_resmap,
 472                            &OCFS2_I(inode)->ip_la_data_resv);
 473
 474         /*
 475          * The inode lock forced other nodes to sync and drop their
 476          * pages, which (correctly) happens even if we have a truncate
 477          * without allocation change - ocfs2 cluster sizes can be much
 478          * greater than page size, so we have to truncate them
 479          * anyway.
 480          */
 481         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 482         truncate_inode_pages(inode->i_mapping, new_i_size);
 483
 484         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 485                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 486                                                i_size_read(inode), 1);
 487                 if (status)
 488                         mlog_errno(status);
 489
 490                 goto bail_unlock_sem;
 491         }
 492
 493         /* alright, we're going to need to do a full blown alloc size
 494          * change. Orphan the inode so that recovery can complete the
 495          * truncate if necessary. This does the task of marking
 496          * i_size. */
 497         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 498         if (status < 0) {
 499                 mlog_errno(status);
 500                 goto bail_unlock_sem;
 501         }
 502
 503         status = ocfs2_commit_truncate(osb, inode, di_bh);
 504         if (status < 0) {
 505                 mlog_errno(status);
 506                 goto bail_unlock_sem;
 507         }
 508
 509         /* TODO: orphan dir cleanup here. */
 510 bail_unlock_sem:
 511         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 512
 513 bail:
 514         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 515                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 516
 517         return status;
 518 }
 519
 520 /*
 521  * extend file allocation only here.
 522  * we'll update all the disk stuff, and oip->alloc_size
 523  *
 524  * expect stuff to be locked, a transaction started and enough data /
 525  * metadata reservations in the contexts.
 526  *
 527  * Will return -EAGAIN, and a reason if a restart is needed.
 528  * If passed in, *reason will always be set, even in error.
 529  */
 530 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 531                          struct inode *inode,
 532                          u32 *logical_offset,
 533                          u32 clusters_to_add,
 534                          int mark_unwritten,
 535                          struct buffer_head *fe_bh,
 536                          handle_t *handle,
 537                          struct ocfs2_alloc_context *data_ac,
 538                          struct ocfs2_alloc_context *meta_ac,
 539                          enum ocfs2_alloc_restarted *reason_ret)
 540 {
 541         int ret;
 542         struct ocfs2_extent_tree et;
 543
 544         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 545         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 546                                           clusters_to_add, mark_unwritten,
 547                                           data_ac, meta_ac, reason_ret);
 548
 549         return ret;
 550 }
 551
 552 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 553                                    u32 clusters_to_add, int mark_unwritten)
 554 {
 555         int status = 0;
 556         int restart_func = 0;
 557         int credits;
 558         u32 prev_clusters;
 559         struct buffer_head *bh = NULL;
 560         struct ocfs2_dinode *fe = NULL;
 561         handle_t *handle = NULL;
 562         struct ocfs2_alloc_context *data_ac = NULL;
 563         struct ocfs2_alloc_context *meta_ac = NULL;
 564         enum ocfs2_alloc_restarted why = RESTART_NONE;
 565         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 566         struct ocfs2_extent_tree et;
 567         int did_quota = 0;
 568
 569         /*
 570          * Unwritten extent only exists for file systems which
 571          * support holes.
 572          */
 573         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 574
 575         status = ocfs2_read_inode_block(inode, &bh);
 576         if (status < 0) {
 577                 mlog_errno(status);
 578                 goto leave;
 579         }
 580         fe = (struct ocfs2_dinode *) bh->b_data;
 581
 582 restart_all:
 583         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 584
 585         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 586         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 587                                        &data_ac, &meta_ac);
 588         if (status) {
 589                 mlog_errno(status);
 590                 goto leave;
 591         }
 592
 593         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 594         handle = ocfs2_start_trans(osb, credits);
 595         if (IS_ERR(handle)) {
 596                 status = PTR_ERR(handle);
 597                 handle = NULL;
 598                 mlog_errno(status);
 599                 goto leave;
 600         }
 601
 602 restarted_transaction:
 603         trace_ocfs2_extend_allocation(
 604                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 605                 (unsigned long long)i_size_read(inode),
 606                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 607                 why, restart_func);
 608
 609         status = dquot_alloc_space_nodirty(inode,
 610                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 611         if (status)
 612                 goto leave;
 613         did_quota = 1;
 614
 615         /* reserve a write to the file entry early on - that we if we
 616          * run out of credits in the allocation path, we can still
 617          * update i_size. */
 618         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 619                                          OCFS2_JOURNAL_ACCESS_WRITE);
 620         if (status < 0) {
 621                 mlog_errno(status);
 622                 goto leave;
 623         }
 624
 625         prev_clusters = OCFS2_I(inode)->ip_clusters;
 626
 627         status = ocfs2_add_inode_data(osb,
 628                                       inode,
 629                                       &logical_start,
 630                                       clusters_to_add,
 631                                       mark_unwritten,
 632                                       bh,
 633                                       handle,
 634                                       data_ac,
 635                                       meta_ac,
 636                                       &why);
 637         if ((status < 0) && (status != -EAGAIN)) {
 638                 if (status != -ENOSPC)
 639                         mlog_errno(status);
 640                 goto leave;
 641         }
 642         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 643         ocfs2_journal_dirty(handle, bh);
 644
 645         spin_lock(&OCFS2_I(inode)->ip_lock);
 646         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 647         spin_unlock(&OCFS2_I(inode)->ip_lock);
 648         /* Release unused quota reservation */
 649         dquot_free_space(inode,
 650                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 651         did_quota = 0;
 652
 653         if (why != RESTART_NONE && clusters_to_add) {
 654                 if (why == RESTART_META) {
 655                         restart_func = 1;
 656                         status = 0;
 657                 } else {
 658                         BUG_ON(why != RESTART_TRANS);
 659
 660                         status = ocfs2_allocate_extend_trans(handle, 1);
 661                         if (status < 0) {
 662                                 /* handle still has to be committed at
 663                                  * this point. */
 664                                 status = -ENOMEM;
 665                                 mlog_errno(status);
 666                                 goto leave;
 667                         }
 668                         goto restarted_transaction;
 669                 }
 670         }
 671
 672         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 673              le32_to_cpu(fe->i_clusters),
 674              (unsigned long long)le64_to_cpu(fe->i_size),
 675              OCFS2_I(inode)->ip_clusters,
 676              (unsigned long long)i_size_read(inode));
 677
 678 leave:
 679         if (status < 0 && did_quota)
 680                 dquot_free_space(inode,
 681                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 682         if (handle) {
 683                 ocfs2_commit_trans(osb, handle);
 684                 handle = NULL;
 685         }
 686         if (data_ac) {
 687                 ocfs2_free_alloc_context(data_ac);
 688                 data_ac = NULL;
 689         }
 690         if (meta_ac) {
 691                 ocfs2_free_alloc_context(meta_ac);
 692                 meta_ac = NULL;
 693         }
 694         if ((!status) && restart_func) {
 695                 restart_func = 0;
 696                 goto restart_all;
 697         }
 698         brelse(bh);
 699         bh = NULL;
 700
 701         return status;
 702 }
 703
 704 /*
 705  * While a write will already be ordering the data, a truncate will not.
 706  * Thus, we need to explicitly order the zeroed pages.
 707  */
 708 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 709                                                 struct buffer_head *di_bh)
 710 {
 711         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 712         handle_t *handle = NULL;
 713         int ret = 0;
 714
 715         if (!ocfs2_should_order_data(inode))
 716                 goto out;
 717
 718         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 719         if (IS_ERR(handle)) {
 720                 ret = -ENOMEM;
 721                 mlog_errno(ret);
 722                 goto out;
 723         }
 724
 725         ret = ocfs2_jbd2_file_inode(handle, inode);
 726         if (ret < 0) {
 727                 mlog_errno(ret);
 728                 goto out;
 729         }
 730
 731         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 732                                       OCFS2_JOURNAL_ACCESS_WRITE);
 733         if (ret)
 734                 mlog_errno(ret);
 735         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 736
 737 out:
 738         if (ret) {
 739                 if (!IS_ERR(handle))
 740                         ocfs2_commit_trans(osb, handle);
 741                 handle = ERR_PTR(ret);
 742         }
 743         return handle;
 744 }
 745
 746 /* Some parts of this taken from generic_cont_expand, which turned out
 747  * to be too fragile to do exactly what we need without us having to
 748  * worry about recursive locking in ->write_begin() and ->write_end(). */
 749 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 750                                  u64 abs_to, struct buffer_head *di_bh)
 751 {
 752         struct address_space *mapping = inode->i_mapping;
 753         struct page *page;
 754         unsigned long index = abs_from >> PAGE_SHIFT;
 755         handle_t *handle;
 756         int ret = 0;
 757         unsigned zero_from, zero_to, block_start, block_end;
 758         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 759
 760         BUG_ON(abs_from >= abs_to);
 761         BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 762         BUG_ON(abs_from & (inode->i_blkbits - 1));
 763
 764         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
 765         if (IS_ERR(handle)) {
 766                 ret = PTR_ERR(handle);
 767                 goto out;
 768         }
 769
 770         page = find_or_create_page(mapping, index, GFP_NOFS);
 771         if (!page) {
 772                 ret = -ENOMEM;
 773                 mlog_errno(ret);
 774                 goto out_commit_trans;
 775         }
 776
 777         /* Get the offsets within the page that we want to zero */
 778         zero_from = abs_from & (PAGE_SIZE - 1);
 779         zero_to = abs_to & (PAGE_SIZE - 1);
 780         if (!zero_to)
 781                 zero_to = PAGE_SIZE;
 782
 783         trace_ocfs2_write_zero_page(
 784                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 785                         (unsigned long long)abs_from,
 786                         (unsigned long long)abs_to,
 787                         index, zero_from, zero_to);
 788
 789         /* We know that zero_from is block aligned */
 790         for (block_start = zero_from; block_start < zero_to;
 791              block_start = block_end) {
 792                 block_end = block_start + i_blocksize(inode);
 793
 794                 /*
 795                  * block_start is block-aligned.  Bump it by one to force
 796                  * __block_write_begin and block_commit_write to zero the
 797                  * whole block.
 798                  */
 799                 ret = __block_write_begin(page, block_start + 1, 0,
 800                                           ocfs2_get_block);
 801                 if (ret < 0) {
 802                         mlog_errno(ret);
 803                         goto out_unlock;
 804                 }
 805
 806
 807                 /* must not update i_size! */
 808                 ret = block_commit_write(page, block_start + 1,
 809                                          block_start + 1);
 810                 if (ret < 0)
 811                         mlog_errno(ret);
 812                 else
 813                         ret = 0;
 814         }
 815
 816         /*
 817          * fs-writeback will release the dirty pages without page lock
 818          * whose offset are over inode size, the release happens at
 819          * block_write_full_page().
 820          */
 821         i_size_write(inode, abs_to);
 822         inode->i_blocks = ocfs2_inode_sector_count(inode);
 823         di->i_size = cpu_to_le64((u64)i_size_read(inode));
 824         inode->i_mtime = inode->i_ctime = current_time(inode);
 825         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 826         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 827         di->i_mtime_nsec = di->i_ctime_nsec;
 828         if (handle) {
 829                 ocfs2_journal_dirty(handle, di_bh);
 830                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 831         }
 832
 833 out_unlock:
 834         unlock_page(page);
 835         put_page(page);
 836 out_commit_trans:
 837         if (handle)
 838                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 839 out:
 840         return ret;
 841 }
 842
 843 /*
 844  * Find the next range to zero.  We do this in terms of bytes because
 845  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 846  * pagecache.  We may return multiple extents.
 847  *
 848  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 849  * needs to be zeroed.  range_start and range_end return the next zeroing
 850  * range.  A subsequent call should pass the previous range_end as its
 851  * zero_start.  If range_end is 0, there's nothing to do.
 852  *
 853  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 854  */
 855 static int ocfs2_zero_extend_get_range(struct inode *inode,
 856                                        struct buffer_head *di_bh,
 857                                        u64 zero_start, u64 zero_end,
 858                                        u64 *range_start, u64 *range_end)
 859 {
 860         int rc = 0, needs_cow = 0;
 861         u32 p_cpos, zero_clusters = 0;
 862         u32 zero_cpos =
 863                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 864         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 865         unsigned int num_clusters = 0;
 866         unsigned int ext_flags = 0;
 867
 868         while (zero_cpos < last_cpos) {
 869                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 870                                         &num_clusters, &ext_flags);
 871                 if (rc) {
 872                         mlog_errno(rc);
 873                         goto out;
 874                 }
 875
 876                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 877                         zero_clusters = num_clusters;
 878                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 879                                 needs_cow = 1;
 880                         break;
 881                 }
 882
 883                 zero_cpos += num_clusters;
 884         }
 885         if (!zero_clusters) {
 886                 *range_end = 0;
 887                 goto out;
 888         }
 889
 890         while ((zero_cpos + zero_clusters) < last_cpos) {
 891                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 892                                         &p_cpos, &num_clusters,
 893                                         &ext_flags);
 894                 if (rc) {
 895                         mlog_errno(rc);
 896                         goto out;
 897                 }
 898
 899                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 900                         break;
 901                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 902                         needs_cow = 1;
 903                 zero_clusters += num_clusters;
 904         }
 905         if ((zero_cpos + zero_clusters) > last_cpos)
 906                 zero_clusters = last_cpos - zero_cpos;
 907
 908         if (needs_cow) {
 909                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 910                                         zero_clusters, UINT_MAX);
 911                 if (rc) {
 912                         mlog_errno(rc);
 913                         goto out;
 914                 }
 915         }
 916
 917         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 918         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 919                                              zero_cpos + zero_clusters);
 920
 921 out:
 922         return rc;
 923 }
 924
 925 /*
 926  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 927  * has made sure that the entire range needs zeroing.
 928  */
 929 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 930                                    u64 range_end, struct buffer_head *di_bh)
 931 {
 932         int rc = 0;
 933         u64 next_pos;
 934         u64 zero_pos = range_start;
 935
 936         trace_ocfs2_zero_extend_range(
 937                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 938                         (unsigned long long)range_start,
 939                         (unsigned long long)range_end);
 940         BUG_ON(range_start >= range_end);
 941
 942         while (zero_pos < range_end) {
 943                 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 944                 if (next_pos > range_end)
 945                         next_pos = range_end;
 946                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 947                 if (rc < 0) {
 948                         mlog_errno(rc);
 949                         break;
 950                 }
 951                 zero_pos = next_pos;
 952
 953                 /*
 954                  * Very large extends have the potential to lock up
 955                  * the cpu for extended periods of time.
 956                  */
 957                 cond_resched();
 958         }
 959
 960         return rc;
 961 }
 962
 963 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 964                       loff_t zero_to_size)
 965 {
 966         int ret = 0;
 967         u64 zero_start, range_start = 0, range_end = 0;
 968         struct super_block *sb = inode->i_sb;
 969
 970         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 971         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 972                                 (unsigned long long)zero_start,
 973                                 (unsigned long long)i_size_read(inode));
 974         while (zero_start < zero_to_size) {
 975                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 976                                                   zero_to_size,
 977                                                   &range_start,
 978                                                   &range_end);
 979                 if (ret) {
 980                         mlog_errno(ret);
 981                         break;
 982                 }
 983                 if (!range_end)
 984                         break;
 985                 /* Trim the ends */
 986                 if (range_start < zero_start)
 987                         range_start = zero_start;
 988                 if (range_end > zero_to_size)
 989                         range_end = zero_to_size;
 990
 991                 ret = ocfs2_zero_extend_range(inode, range_start,
 992                                               range_end, di_bh);
 993                 if (ret) {
 994                         mlog_errno(ret);
 995                         break;
 996                 }
 997                 zero_start = range_end;
 998         }
 999
1000         return ret;
1001 }
1002
1003 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1004                           u64 new_i_size, u64 zero_to)
1005 {
1006         int ret;
1007         u32 clusters_to_add;
1008         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1009
1010         /*
1011          * Only quota files call this without a bh, and they can't be
1012          * refcounted.
1013          */
1014         BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1015         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1016
1017         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1018         if (clusters_to_add < oi->ip_clusters)
1019                 clusters_to_add = 0;
1020         else
1021                 clusters_to_add -= oi->ip_clusters;
1022
1023         if (clusters_to_add) {
1024                 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1025                                               clusters_to_add, 0);
1026                 if (ret) {
1027                         mlog_errno(ret);
1028                         goto out;
1029                 }
1030         }
1031
1032         /*
1033          * Call this even if we don't add any clusters to the tree. We
1034          * still need to zero the area between the old i_size and the
1035          * new i_size.
1036          */
1037         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1038         if (ret < 0)
1039                 mlog_errno(ret);
1040
1041 out:
1042         return ret;
1043 }
1044
1045 static int ocfs2_extend_file(struct inode *inode,
1046                              struct buffer_head *di_bh,
1047                              u64 new_i_size)
1048 {
1049         int ret = 0;
1050         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1051
1052         BUG_ON(!di_bh);
1053
1054         /* setattr sometimes calls us like this. */
1055         if (new_i_size == 0)
1056                 goto out;
1057
1058         if (i_size_read(inode) == new_i_size)
1059                 goto out;
1060         BUG_ON(new_i_size < i_size_read(inode));
1061
1062         /*
1063          * The alloc sem blocks people in read/write from reading our
1064          * allocation until we're done changing it. We depend on
1065          * i_mutex to block other extend/truncate calls while we're
1066          * here.  We even have to hold it for sparse files because there
1067          * might be some tail zeroing.
1068          */
1069         down_write(&oi->ip_alloc_sem);
1070
1071         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1072                 /*
1073                  * We can optimize small extends by keeping the inodes
1074                  * inline data.
1075                  */
1076                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1077                         up_write(&oi->ip_alloc_sem);
1078                         goto out_update_size;
1079                 }
1080
1081                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1082                 if (ret) {
1083                         up_write(&oi->ip_alloc_sem);
1084                         mlog_errno(ret);
1085                         goto out;
1086                 }
1087         }
1088
1089         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1090                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1091         else
1092                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1093                                             new_i_size);
1094
1095         up_write(&oi->ip_alloc_sem);
1096
1097         if (ret < 0) {
1098                 mlog_errno(ret);
1099                 goto out;
1100         }
1101
1102 out_update_size:
1103         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1104         if (ret < 0)
1105                 mlog_errno(ret);
1106
1107 out:
1108         return ret;
1109 }
1110
1111 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1112 {
1113         int status = 0, size_change;
1114         int inode_locked = 0;
1115         struct inode *inode = d_inode(dentry);
1116         struct super_block *sb = inode->i_sb;
1117         struct ocfs2_super *osb = OCFS2_SB(sb);
1118         struct buffer_head *bh = NULL;
1119         handle_t *handle = NULL;
1120         struct dquot *transfer_to[MAXQUOTAS] = { };
1121         int qtype;
1122         int had_lock;
1123         struct ocfs2_lock_holder oh;
1124
1125         trace_ocfs2_setattr(inode, dentry,
1126                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1127                             dentry->d_name.len, dentry->d_name.name,
1128                             attr->ia_valid, attr->ia_mode,
1129                             from_kuid(&init_user_ns, attr->ia_uid),
1130                             from_kgid(&init_user_ns, attr->ia_gid));
1131
1132         /* ensuring we don't even attempt to truncate a symlink */
1133         if (S_ISLNK(inode->i_mode))
1134                 attr->ia_valid &= ~ATTR_SIZE;
1135
1136 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1137                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1138         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1139                 return 0;
1140
1141         status = setattr_prepare(dentry, attr);
1142         if (status)
1143                 return status;
1144
1145         if (is_quota_modification(inode, attr)) {
1146                 status = dquot_initialize(inode);
1147                 if (status)
1148                         return status;
1149         }
1150         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1151         if (size_change) {
1152                 /*
1153                  * Here we should wait dio to finish before inode lock
1154                  * to avoid a deadlock between ocfs2_setattr() and
1155                  * ocfs2_dio_end_io_write()
1156                  */
1157                 inode_dio_wait(inode);
1158
1159                 status = ocfs2_rw_lock(inode, 1);
1160                 if (status < 0) {
1161                         mlog_errno(status);
1162                         goto bail;
1163                 }
1164         }
1165
1166         had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1167         if (had_lock < 0) {
1168                 status = had_lock;
1169                 goto bail_unlock_rw;
1170         } else if (had_lock) {
1171                 /*
1172                  * As far as we know, ocfs2_setattr() could only be the first
1173                  * VFS entry point in the call chain of recursive cluster
1174                  * locking issue.
1175                  *
1176                  * For instance:
1177                  * chmod_common()
1178                  *  notify_change()
1179                  *   ocfs2_setattr()
1180                  *    posix_acl_chmod()
1181                  *     ocfs2_iop_get_acl()
1182                  *
1183                  * But, we're not 100% sure if it's always true, because the
1184                  * ordering of the VFS entry points in the call chain is out
1185                  * of our control. So, we'd better dump the stack here to
1186                  * catch the other cases of recursive locking.
1187                  */
1188                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1189                 dump_stack();
1190         }
1191         inode_locked = 1;
1192
1193         if (size_change) {
1194                 status = inode_newsize_ok(inode, attr->ia_size);
1195                 if (status)
1196                         goto bail_unlock;
1197
1198                 if (i_size_read(inode) >= attr->ia_size) {
1199                         if (ocfs2_should_order_data(inode)) {
1200                                 status = ocfs2_begin_ordered_truncate(inode,
1201                                                                       attr->ia_size);
1202                                 if (status)
1203                                         goto bail_unlock;
1204                         }
1205                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1206                 } else
1207                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1208                 if (status < 0) {
1209                         if (status != -ENOSPC)
1210                                 mlog_errno(status);
1211                         status = -ENOSPC;
1212                         goto bail_unlock;
1213                 }
1214         }
1215
1216         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1217             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1218                 /*
1219                  * Gather pointers to quota structures so that allocation /
1220                  * freeing of quota structures happens here and not inside
1221                  * dquot_transfer() where we have problems with lock ordering
1222                  */
1223                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1224                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1225                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1226                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1227                         if (IS_ERR(transfer_to[USRQUOTA])) {
1228                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1229                                 goto bail_unlock;
1230                         }
1231                 }
1232                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1233                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1234                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1235                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1236                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1237                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1238                                 goto bail_unlock;
1239                         }
1240                 }
1241                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1242                                            2 * ocfs2_quota_trans_credits(sb));
1243                 if (IS_ERR(handle)) {
1244                         status = PTR_ERR(handle);
1245                         mlog_errno(status);
1246                         goto bail_unlock;
1247                 }
1248                 status = __dquot_transfer(inode, transfer_to);
1249                 if (status < 0)
1250                         goto bail_commit;
1251         } else {
1252                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1253                 if (IS_ERR(handle)) {
1254                         status = PTR_ERR(handle);
1255                         mlog_errno(status);
1256                         goto bail_unlock;
1257                 }
1258         }
1259
1260         setattr_copy(inode, attr);
1261         mark_inode_dirty(inode);
1262
1263         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1264         if (status < 0)
1265                 mlog_errno(status);
1266
1267 bail_commit:
1268         ocfs2_commit_trans(osb, handle);
1269 bail_unlock:
1270         if (status && inode_locked) {
1271                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1272                 inode_locked = 0;
1273         }
1274 bail_unlock_rw:
1275         if (size_change)
1276                 ocfs2_rw_unlock(inode, 1);
1277 bail:
1278
1279         /* Release quota pointers in case we acquired them */
1280         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1281                 dqput(transfer_to[qtype]);
1282
1283         if (!status && attr->ia_valid & ATTR_MODE) {
1284                 status = ocfs2_acl_chmod(inode, bh);
1285                 if (status < 0)
1286                         mlog_errno(status);
1287         }
1288         if (inode_locked)
1289                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1290
1291         brelse(bh);
1292         return status;
1293 }
1294
1295 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1296                   u32 request_mask, unsigned int flags)
1297 {
1298         struct inode *inode = d_inode(path->dentry);
1299         struct super_block *sb = path->dentry->d_sb;
1300         struct ocfs2_super *osb = sb->s_fs_info;
1301         int err;
1302
1303         err = ocfs2_inode_revalidate(path->dentry);
1304         if (err) {
1305                 if (err != -ENOENT)
1306                         mlog_errno(err);
1307                 goto bail;
1308         }
1309
1310         generic_fillattr(inode, stat);
1311         /*
1312          * If there is inline data in the inode, the inode will normally not
1313          * have data blocks allocated (it may have an external xattr block).
1314          * Report at least one sector for such files, so tools like tar, rsync,
1315          * others don't incorrectly think the file is completely sparse.
1316          */
1317         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1318                 stat->blocks += (stat->size + 511)>>9;
1319
1320         /* We set the blksize from the cluster size for performance */
1321         stat->blksize = osb->s_clustersize;
1322
1323 bail:
1324         return err;
1325 }
1326
1327 int ocfs2_permission(struct inode *inode, int mask)
1328 {
1329         int ret, had_lock;
1330         struct ocfs2_lock_holder oh;
1331
1332         if (mask & MAY_NOT_BLOCK)
1333                 return -ECHILD;
1334
1335         had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1336         if (had_lock < 0) {
1337                 ret = had_lock;
1338                 goto out;
1339         } else if (had_lock) {
1340                 /* See comments in ocfs2_setattr() for details.
1341                  * The call chain of this case could be:
1342                  * do_sys_open()
1343                  *  may_open()
1344                  *   inode_permission()
1345                  *    ocfs2_permission()
1346                  *     ocfs2_iop_get_acl()
1347                  */
1348                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1349                 dump_stack();
1350         }
1351
1352         ret = generic_permission(inode, mask);
1353
1354         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1355 out:
1356         return ret;
1357 }
1358
1359 static int __ocfs2_write_remove_suid(struct inode *inode,
1360                                      struct buffer_head *bh)
1361 {
1362         int ret;
1363         handle_t *handle;
1364         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1365         struct ocfs2_dinode *di;
1366
1367         trace_ocfs2_write_remove_suid(
1368                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1369                         inode->i_mode);
1370
1371         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1372         if (IS_ERR(handle)) {
1373                 ret = PTR_ERR(handle);
1374                 mlog_errno(ret);
1375                 goto out;
1376         }
1377
1378         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1379                                       OCFS2_JOURNAL_ACCESS_WRITE);
1380         if (ret < 0) {
1381                 mlog_errno(ret);
1382                 goto out_trans;
1383         }
1384
1385         inode->i_mode &= ~S_ISUID;
1386         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1387                 inode->i_mode &= ~S_ISGID;
1388
1389         di = (struct ocfs2_dinode *) bh->b_data;
1390         di->i_mode = cpu_to_le16(inode->i_mode);
1391         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1392
1393         ocfs2_journal_dirty(handle, bh);
1394
1395 out_trans:
1396         ocfs2_commit_trans(osb, handle);
1397 out:
1398         return ret;
1399 }
1400
1401 static int ocfs2_write_remove_suid(struct inode *inode)
1402 {
1403         int ret;
1404         struct buffer_head *bh = NULL;
1405
1406         ret = ocfs2_read_inode_block(inode, &bh);
1407         if (ret < 0) {
1408                 mlog_errno(ret);
1409                 goto out;
1410         }
1411
1412         ret =  __ocfs2_write_remove_suid(inode, bh);
1413 out:
1414         brelse(bh);
1415         return ret;
1416 }
1417
1418 /*
1419  * Allocate enough extents to cover the region starting at byte offset
1420  * start for len bytes. Existing extents are skipped, any extents
1421  * added are marked as "unwritten".
1422  */
1423 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1424                                             u64 start, u64 len)
1425 {
1426         int ret;
1427         u32 cpos, phys_cpos, clusters, alloc_size;
1428         u64 end = start + len;
1429         struct buffer_head *di_bh = NULL;
1430
1431         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1432                 ret = ocfs2_read_inode_block(inode, &di_bh);
1433                 if (ret) {
1434                         mlog_errno(ret);
1435                         goto out;
1436                 }
1437
1438                 /*
1439                  * Nothing to do if the requested reservation range
1440                  * fits within the inode.
1441                  */
1442                 if (ocfs2_size_fits_inline_data(di_bh, end))
1443                         goto out;
1444
1445                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1446                 if (ret) {
1447                         mlog_errno(ret);
1448                         goto out;
1449                 }
1450         }
1451
1452         /*
1453          * We consider both start and len to be inclusive.
1454          */
1455         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1456         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1457         clusters -= cpos;
1458
1459         while (clusters) {
1460                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1461                                          &alloc_size, NULL);
1462                 if (ret) {
1463                         mlog_errno(ret);
1464                         goto out;
1465                 }
1466
1467                 /*
1468                  * Hole or existing extent len can be arbitrary, so
1469                  * cap it to our own allocation request.
1470                  */
1471                 if (alloc_size > clusters)
1472                         alloc_size = clusters;
1473
1474                 if (phys_cpos) {
1475                         /*
1476                          * We already have an allocation at this
1477                          * region so we can safely skip it.
1478                          */
1479                         goto next;
1480                 }
1481
1482                 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1483                 if (ret) {
1484                         if (ret != -ENOSPC)
1485                                 mlog_errno(ret);
1486                         goto out;
1487                 }
1488
1489 next:
1490                 cpos += alloc_size;
1491                 clusters -= alloc_size;
1492         }
1493
1494         ret = 0;
1495 out:
1496
1497         brelse(di_bh);
1498         return ret;
1499 }
1500
1501 /*
1502  * Truncate a byte range, avoiding pages within partial clusters. This
1503  * preserves those pages for the zeroing code to write to.
1504  */
1505 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1506                                          u64 byte_len)
1507 {
1508         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1509         loff_t start, end;
1510         struct address_space *mapping = inode->i_mapping;
1511
1512         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1513         end = byte_start + byte_len;
1514         end = end & ~(osb->s_clustersize - 1);
1515
1516         if (start < end) {
1517                 unmap_mapping_range(mapping, start, end - start, 0);
1518                 truncate_inode_pages_range(mapping, start, end - 1);
1519         }
1520 }
1521
1522 static int ocfs2_zero_partial_clusters(struct inode *inode,
1523                                        u64 start, u64 len)
1524 {
1525         int ret = 0;
1526         u64 tmpend = 0;
1527         u64 end = start + len;
1528         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1529         unsigned int csize = osb->s_clustersize;
1530         handle_t *handle;
1531
1532         /*
1533          * The "start" and "end" values are NOT necessarily part of
1534          * the range whose allocation is being deleted. Rather, this
1535          * is what the user passed in with the request. We must zero
1536          * partial clusters here. There's no need to worry about
1537          * physical allocation - the zeroing code knows to skip holes.
1538          */
1539         trace_ocfs2_zero_partial_clusters(
1540                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1541                 (unsigned long long)start, (unsigned long long)end);
1542
1543         /*
1544          * If both edges are on a cluster boundary then there's no
1545          * zeroing required as the region is part of the allocation to
1546          * be truncated.
1547          */
1548         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1549                 goto out;
1550
1551         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1552         if (IS_ERR(handle)) {
1553                 ret = PTR_ERR(handle);
1554                 mlog_errno(ret);
1555                 goto out;
1556         }
1557
1558         /*
1559          * If start is on a cluster boundary and end is somewhere in another
1560          * cluster, we have not COWed the cluster starting at start, unless
1561          * end is also within the same cluster. So, in this case, we skip this
1562          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1563          * to the next one.
1564          */
1565         if ((start & (csize - 1)) != 0) {
1566                 /*
1567                  * We want to get the byte offset of the end of the 1st
1568                  * cluster.
1569                  */
1570                 tmpend = (u64)osb->s_clustersize +
1571                         (start & ~(osb->s_clustersize - 1));
1572                 if (tmpend > end)
1573                         tmpend = end;
1574
1575                 trace_ocfs2_zero_partial_clusters_range1(
1576                         (unsigned long long)start,
1577                         (unsigned long long)tmpend);
1578
1579                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1580                                                     tmpend);
1581                 if (ret)
1582                         mlog_errno(ret);
1583         }
1584
1585         if (tmpend < end) {
1586                 /*
1587                  * This may make start and end equal, but the zeroing
1588                  * code will skip any work in that case so there's no
1589                  * need to catch it up here.
1590                  */
1591                 start = end & ~(osb->s_clustersize - 1);
1592
1593                 trace_ocfs2_zero_partial_clusters_range2(
1594                         (unsigned long long)start, (unsigned long long)end);
1595
1596                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1597                 if (ret)
1598                         mlog_errno(ret);
1599         }
1600         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1601
1602         ocfs2_commit_trans(osb, handle);
1603 out:
1604         return ret;
1605 }
1606
1607 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1608 {
1609         int i;
1610         struct ocfs2_extent_rec *rec = NULL;
1611
1612         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1613
1614                 rec = &el->l_recs[i];
1615
1616                 if (le32_to_cpu(rec->e_cpos) < pos)
1617                         break;
1618         }
1619
1620         return i;
1621 }
1622
1623 /*
1624  * Helper to calculate the punching pos and length in one run, we handle the
1625  * following three cases in order:
1626  *
1627  * - remove the entire record
1628  * - remove a partial record
1629  * - no record needs to be removed (hole-punching completed)
1630 */
1631 static void ocfs2_calc_trunc_pos(struct inode *inode,
1632                                  struct ocfs2_extent_list *el,
1633                                  struct ocfs2_extent_rec *rec,
1634                                  u32 trunc_start, u32 *trunc_cpos,
1635                                  u32 *trunc_len, u32 *trunc_end,
1636                                  u64 *blkno, int *done)
1637 {
1638         int ret = 0;
1639         u32 coff, range;
1640
1641         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1642
1643         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1644                 /*
1645                  * remove an entire extent record.
1646                  */
1647                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1648                 /*
1649                  * Skip holes if any.
1650                  */
1651                 if (range < *trunc_end)
1652                         *trunc_end = range;
1653                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1654                 *blkno = le64_to_cpu(rec->e_blkno);
1655                 *trunc_end = le32_to_cpu(rec->e_cpos);
1656         } else if (range > trunc_start) {
1657                 /*
1658                  * remove a partial extent record, which means we're
1659                  * removing the last extent record.
1660                  */
1661                 *trunc_cpos = trunc_start;
1662                 /*
1663                  * skip hole if any.
1664                  */
1665                 if (range < *trunc_end)
1666                         *trunc_end = range;
1667                 *trunc_len = *trunc_end - trunc_start;
1668                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1669                 *blkno = le64_to_cpu(rec->e_blkno) +
1670                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1671                 *trunc_end = trunc_start;
1672         } else {
1673                 /*
1674                  * It may have two following possibilities:
1675                  *
1676                  * - last record has been removed
1677                  * - trunc_start was within a hole
1678                  *
1679                  * both two cases mean the completion of hole punching.
1680                  */
1681                 ret = 1;
1682         }
1683
1684         *done = ret;
1685 }
1686
1687 int ocfs2_remove_inode_range(struct inode *inode,
1688                              struct buffer_head *di_bh, u64 byte_start,
1689                              u64 byte_len)
1690 {
1691         int ret = 0, flags = 0, done = 0, i;
1692         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1693         u32 cluster_in_el;
1694         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1695         struct ocfs2_cached_dealloc_ctxt dealloc;
1696         struct address_space *mapping = inode->i_mapping;
1697         struct ocfs2_extent_tree et;
1698         struct ocfs2_path *path = NULL;
1699         struct ocfs2_extent_list *el = NULL;
1700         struct ocfs2_extent_rec *rec = NULL;
1701         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1702         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1703
1704         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1705         ocfs2_init_dealloc_ctxt(&dealloc);
1706
1707         trace_ocfs2_remove_inode_range(
1708                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1709                         (unsigned long long)byte_start,
1710                         (unsigned long long)byte_len);
1711
1712         if (byte_len == 0)
1713                 return 0;
1714
1715         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1716                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1717                                             byte_start + byte_len, 0);
1718                 if (ret) {
1719                         mlog_errno(ret);
1720                         goto out;
1721                 }
1722                 /*
1723                  * There's no need to get fancy with the page cache
1724                  * truncate of an inline-data inode. We're talking
1725                  * about less than a page here, which will be cached
1726                  * in the dinode buffer anyway.
1727                  */
1728                 unmap_mapping_range(mapping, 0, 0, 0);
1729                 truncate_inode_pages(mapping, 0);
1730                 goto out;
1731         }
1732
1733         /*
1734          * For reflinks, we may need to CoW 2 clusters which might be
1735          * partially zero'd later, if hole's start and end offset were
1736          * within one cluster(means is not exactly aligned to clustersize).
1737          */
1738
1739         if (ocfs2_is_refcount_inode(inode)) {
1740                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1741                 if (ret) {
1742                         mlog_errno(ret);
1743                         goto out;
1744                 }
1745
1746                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1747                 if (ret) {
1748                         mlog_errno(ret);
1749                         goto out;
1750                 }
1751         }
1752
1753         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1754         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1755         cluster_in_el = trunc_end;
1756
1757         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1758         if (ret) {
1759                 mlog_errno(ret);
1760                 goto out;
1761         }
1762
1763         path = ocfs2_new_path_from_et(&et);
1764         if (!path) {
1765                 ret = -ENOMEM;
1766                 mlog_errno(ret);
1767                 goto out;
1768         }
1769
1770         while (trunc_end > trunc_start) {
1771
1772                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1773                                       cluster_in_el);
1774                 if (ret) {
1775                         mlog_errno(ret);
1776                         goto out;
1777                 }
1778
1779                 el = path_leaf_el(path);
1780
1781                 i = ocfs2_find_rec(el, trunc_end);
1782                 /*
1783                  * Need to go to previous extent block.
1784                  */
1785                 if (i < 0) {
1786                         if (path->p_tree_depth == 0)
1787                                 break;
1788
1789                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1790                                                             path,
1791                                                             &cluster_in_el);
1792                         if (ret) {
1793                                 mlog_errno(ret);
1794                                 goto out;
1795                         }
1796
1797                         /*
1798                          * We've reached the leftmost extent block,
1799                          * it's safe to leave.
1800                          */
1801                         if (cluster_in_el == 0)
1802                                 break;
1803
1804                         /*
1805                          * The 'pos' searched for previous extent block is
1806                          * always one cluster less than actual trunc_end.
1807                          */
1808                         trunc_end = cluster_in_el + 1;
1809
1810                         ocfs2_reinit_path(path, 1);
1811
1812                         continue;
1813
1814                 } else
1815                         rec = &el->l_recs[i];
1816
1817                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1818                                      &trunc_len, &trunc_end, &blkno, &done);
1819                 if (done)
1820                         break;
1821
1822                 flags = rec->e_flags;
1823                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1824
1825                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1826                                                phys_cpos, trunc_len, flags,
1827                                                &dealloc, refcount_loc, false);
1828                 if (ret < 0) {
1829                         mlog_errno(ret);
1830                         goto out;
1831                 }
1832
1833                 cluster_in_el = trunc_end;
1834
1835                 ocfs2_reinit_path(path, 1);
1836         }
1837
1838         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1839
1840 out:
1841         ocfs2_free_path(path);
1842         ocfs2_schedule_truncate_log_flush(osb, 1);
1843         ocfs2_run_deallocs(osb, &dealloc);
1844
1845         return ret;
1846 }
1847
1848 /*
1849  * Parts of this function taken from xfs_change_file_space()
1850  */
1851 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1852                                      loff_t f_pos, unsigned int cmd,
1853                                      struct ocfs2_space_resv *sr,
1854                                      int change_size)
1855 {
1856         int ret;
1857         s64 llen;
1858         loff_t size;
1859         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1860         struct buffer_head *di_bh = NULL;
1861         handle_t *handle;
1862         unsigned long long max_off = inode->i_sb->s_maxbytes;
1863
1864         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1865                 return -EROFS;
1866
1867         inode_lock(inode);
1868
1869         /*
1870          * This prevents concurrent writes on other nodes
1871          */
1872         ret = ocfs2_rw_lock(inode, 1);
1873         if (ret) {
1874                 mlog_errno(ret);
1875                 goto out;
1876         }
1877
1878         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1879         if (ret) {
1880                 mlog_errno(ret);
1881                 goto out_rw_unlock;
1882         }
1883
1884         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1885                 ret = -EPERM;
1886                 goto out_inode_unlock;
1887         }
1888
1889         switch (sr->l_whence) {
1890         case 0: /*SEEK_SET*/
1891                 break;
1892         case 1: /*SEEK_CUR*/
1893                 sr->l_start += f_pos;
1894                 break;
1895         case 2: /*SEEK_END*/
1896                 sr->l_start += i_size_read(inode);
1897                 break;
1898         default:
1899                 ret = -EINVAL;
1900                 goto out_inode_unlock;
1901         }
1902         sr->l_whence = 0;
1903
1904         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1905
1906         if (sr->l_start < 0
1907             || sr->l_start > max_off
1908             || (sr->l_start + llen) < 0
1909             || (sr->l_start + llen) > max_off) {
1910                 ret = -EINVAL;
1911                 goto out_inode_unlock;
1912         }
1913         size = sr->l_start + sr->l_len;
1914
1915         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1916             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1917                 if (sr->l_len <= 0) {
1918                         ret = -EINVAL;
1919                         goto out_inode_unlock;
1920                 }
1921         }
1922
1923         if (file && should_remove_suid(file->f_path.dentry)) {
1924                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1925                 if (ret) {
1926                         mlog_errno(ret);
1927                         goto out_inode_unlock;
1928                 }
1929         }
1930
1931         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1932         switch (cmd) {
1933         case OCFS2_IOC_RESVSP:
1934         case OCFS2_IOC_RESVSP64:
1935                 /*
1936                  * This takes unsigned offsets, but the signed ones we
1937                  * pass have been checked against overflow above.
1938                  */
1939                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1940                                                        sr->l_len);
1941                 break;
1942         case OCFS2_IOC_UNRESVSP:
1943         case OCFS2_IOC_UNRESVSP64:
1944                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1945                                                sr->l_len);
1946                 break;
1947         default:
1948                 ret = -EINVAL;
1949         }
1950         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1951         if (ret) {
1952                 mlog_errno(ret);
1953                 goto out_inode_unlock;
1954         }
1955
1956         /*
1957          * We update c/mtime for these changes
1958          */
1959         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1960         if (IS_ERR(handle)) {
1961                 ret = PTR_ERR(handle);
1962                 mlog_errno(ret);
1963                 goto out_inode_unlock;
1964         }
1965
1966         if (change_size && i_size_read(inode) < size)
1967                 i_size_write(inode, size);
1968
1969         inode->i_ctime = inode->i_mtime = current_time(inode);
1970         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1971         if (ret < 0)
1972                 mlog_errno(ret);
1973
1974         if (file && (file->f_flags & O_SYNC))
1975                 handle->h_sync = 1;
1976
1977         ocfs2_commit_trans(osb, handle);
1978
1979 out_inode_unlock:
1980         brelse(di_bh);
1981         ocfs2_inode_unlock(inode, 1);
1982 out_rw_unlock:
1983         ocfs2_rw_unlock(inode, 1);
1984
1985 out:
1986         inode_unlock(inode);
1987         return ret;
1988 }
1989
1990 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1991                             struct ocfs2_space_resv *sr)
1992 {
1993         struct inode *inode = file_inode(file);
1994         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1995         int ret;
1996
1997         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1998             !ocfs2_writes_unwritten_extents(osb))
1999                 return -ENOTTY;
2000         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2001                  !ocfs2_sparse_alloc(osb))
2002                 return -ENOTTY;
2003
2004         if (!S_ISREG(inode->i_mode))
2005                 return -EINVAL;
2006
2007         if (!(file->f_mode & FMODE_WRITE))
2008                 return -EBADF;
2009
2010         ret = mnt_want_write_file(file);
2011         if (ret)
2012                 return ret;
2013         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2014         mnt_drop_write_file(file);
2015         return ret;
2016 }
2017
2018 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2019                             loff_t len)
2020 {
2021         struct inode *inode = file_inode(file);
2022         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2023         struct ocfs2_space_resv sr;
2024         int change_size = 1;
2025         int cmd = OCFS2_IOC_RESVSP64;
2026
2027         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2028                 return -EOPNOTSUPP;
2029         if (!ocfs2_writes_unwritten_extents(osb))
2030                 return -EOPNOTSUPP;
2031
2032         if (mode & FALLOC_FL_KEEP_SIZE)
2033                 change_size = 0;
2034
2035         if (mode & FALLOC_FL_PUNCH_HOLE)
2036                 cmd = OCFS2_IOC_UNRESVSP64;
2037
2038         sr.l_whence = 0;
2039         sr.l_start = (s64)offset;
2040         sr.l_len = (s64)len;
2041
2042         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2043                                          change_size);
2044 }
2045
2046 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2047                                    size_t count)
2048 {
2049         int ret = 0;
2050         unsigned int extent_flags;
2051         u32 cpos, clusters, extent_len, phys_cpos;
2052         struct super_block *sb = inode->i_sb;
2053
2054         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2055             !ocfs2_is_refcount_inode(inode) ||
2056             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2057                 return 0;
2058
2059         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2060         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2061
2062         while (clusters) {
2063                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2064                                          &extent_flags);
2065                 if (ret < 0) {
2066                         mlog_errno(ret);
2067                         goto out;
2068                 }
2069
2070                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2071                         ret = 1;
2072                         break;
2073                 }
2074
2075                 if (extent_len > clusters)
2076                         extent_len = clusters;
2077
2078                 clusters -= extent_len;
2079                 cpos += extent_len;
2080         }
2081 out:
2082         return ret;
2083 }
2084
2085 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2086 {
2087         int blockmask = inode->i_sb->s_blocksize - 1;
2088         loff_t final_size = pos + count;
2089
2090         if ((pos & blockmask) || (final_size & blockmask))
2091                 return 1;
2092         return 0;
2093 }
2094
2095 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2096                                             struct buffer_head **di_bh,
2097                                             int meta_level,
2098                                             int overwrite_io,
2099                                             int write_sem,
2100                                             int wait)
2101 {
2102         int ret = 0;
2103
2104         if (wait)
2105                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2106         else
2107                 ret = ocfs2_try_inode_lock(inode,
2108                         overwrite_io ? NULL : di_bh, meta_level);
2109         if (ret < 0)
2110                 goto out;
2111
2112         if (wait) {
2113                 if (write_sem)
2114                         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2115                 else
2116                         down_read(&OCFS2_I(inode)->ip_alloc_sem);
2117         } else {
2118                 if (write_sem)
2119                         ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2120                 else
2121                         ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2122
2123                 if (!ret) {
2124                         ret = -EAGAIN;
2125                         goto out_unlock;
2126                 }
2127         }
2128
2129         return ret;
2130
2131 out_unlock:
2132         brelse(*di_bh);
2133         ocfs2_inode_unlock(inode, meta_level);
2134 out:
2135         return ret;
2136 }
2137
2138 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2139                                                struct buffer_head **di_bh,
2140                                                int meta_level,
2141                                                int write_sem)
2142 {
2143         if (write_sem)
2144                 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2145         else
2146                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2147
2148         brelse(*di_bh);
2149         *di_bh = NULL;
2150
2151         if (meta_level >= 0)
2152                 ocfs2_inode_unlock(inode, meta_level);
2153 }
2154
2155 static int ocfs2_prepare_inode_for_write(struct file *file,
2156                                          loff_t pos, size_t count, int wait)
2157 {
2158         int ret = 0, meta_level = 0, overwrite_io = 0;
2159         int write_sem = 0;
2160         struct dentry *dentry = file->f_path.dentry;
2161         struct inode *inode = d_inode(dentry);
2162         struct buffer_head *di_bh = NULL;
2163         loff_t end;
2164         u32 cpos;
2165         u32 clusters;
2166
2167         /*
2168          * We start with a read level meta lock and only jump to an ex
2169          * if we need to make modifications here.
2170          */
2171         for(;;) {
2172                 ret = ocfs2_inode_lock_for_extent_tree(inode,
2173                                                        &di_bh,
2174                                                        meta_level,
2175                                                        overwrite_io,
2176                                                        write_sem,
2177                                                        wait);
2178                 if (ret < 0) {
2179                         if (ret != -EAGAIN)
2180                                 mlog_errno(ret);
2181                         goto out;
2182                 }
2183
2184                 /*
2185                  * Check if IO will overwrite allocated blocks in case
2186                  * IOCB_NOWAIT flag is set.
2187                  */
2188                 if (!wait && !overwrite_io) {
2189                         overwrite_io = 1;
2190
2191                         ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2192                         if (ret < 0) {
2193                                 if (ret != -EAGAIN)
2194                                         mlog_errno(ret);
2195                                 goto out_unlock;
2196                         }
2197                 }
2198
2199                 /* Clear suid / sgid if necessary. We do this here
2200                  * instead of later in the write path because
2201                  * remove_suid() calls ->setattr without any hint that
2202                  * we may have already done our cluster locking. Since
2203                  * ocfs2_setattr() *must* take cluster locks to
2204                  * proceed, this will lead us to recursively lock the
2205                  * inode. There's also the dinode i_size state which
2206                  * can be lost via setattr during extending writes (we
2207                  * set inode->i_size at the end of a write. */
2208                 if (should_remove_suid(dentry)) {
2209                         if (meta_level == 0) {
2210                                 ocfs2_inode_unlock_for_extent_tree(inode,
2211                                                                    &di_bh,
2212                                                                    meta_level,
2213                                                                    write_sem);
2214                                 meta_level = 1;
2215                                 continue;
2216                         }
2217
2218                         ret = ocfs2_write_remove_suid(inode);
2219                         if (ret < 0) {
2220                                 mlog_errno(ret);
2221                                 goto out_unlock;
2222                         }
2223                 }
2224
2225                 end = pos + count;
2226
2227                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2228                 if (ret == 1) {
2229                         ocfs2_inode_unlock_for_extent_tree(inode,
2230                                                            &di_bh,
2231                                                            meta_level,
2232                                                            write_sem);
2233                         ret = ocfs2_inode_lock_for_extent_tree(inode,
2234                                                                &di_bh,
2235                                                                meta_level,
2236                                                                overwrite_io,
2237                                                                1,
2238                                                                wait);
2239                         write_sem = 1;
2240                         if (ret < 0) {
2241                                 if (ret != -EAGAIN)
2242                                         mlog_errno(ret);
2243                                 goto out;
2244                         }
2245
2246                         cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2247                         clusters =
2248                                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2249                         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2250                 }
2251
2252                 if (ret < 0) {
2253                         if (ret != -EAGAIN)
2254                                 mlog_errno(ret);
2255                         goto out_unlock;
2256                 }
2257
2258                 break;
2259         }
2260
2261 out_unlock:
2262         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2263                                             pos, count, wait);
2264
2265         ocfs2_inode_unlock_for_extent_tree(inode,
2266                                            &di_bh,
2267                                            meta_level,
2268                                            write_sem);
2269
2270 out:
2271         return ret;
2272 }
2273
2274 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2275                                     struct iov_iter *from)
2276 {
2277         int rw_level;
2278         ssize_t written = 0;
2279         ssize_t ret;
2280         size_t count = iov_iter_count(from);
2281         struct file *file = iocb->ki_filp;
2282         struct inode *inode = file_inode(file);
2283         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2284         int full_coherency = !(osb->s_mount_opt &
2285                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2286         void *saved_ki_complete = NULL;
2287         int append_write = ((iocb->ki_pos + count) >=
2288                         i_size_read(inode) ? 1 : 0);
2289         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2290         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2291
2292         trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2293                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2294                 file->f_path.dentry->d_name.len,
2295                 file->f_path.dentry->d_name.name,
2296                 (unsigned int)from->nr_segs);   /* GRRRRR */
2297
2298         if (!direct_io && nowait)
2299                 return -EOPNOTSUPP;
2300
2301         if (count == 0)
2302                 return 0;
2303
2304         if (nowait) {
2305                 if (!inode_trylock(inode))
2306                         return -EAGAIN;
2307         } else
2308                 inode_lock(inode);
2309
2310         /*
2311          * Concurrent O_DIRECT writes are allowed with
2312          * mount_option "coherency=buffered".
2313          * For append write, we must take rw EX.
2314          */
2315         rw_level = (!direct_io || full_coherency || append_write);
2316
2317         if (nowait)
2318                 ret = ocfs2_try_rw_lock(inode, rw_level);
2319         else
2320                 ret = ocfs2_rw_lock(inode, rw_level);
2321         if (ret < 0) {
2322                 if (ret != -EAGAIN)
2323                         mlog_errno(ret);
2324                 goto out_mutex;
2325         }
2326
2327         /*
2328          * O_DIRECT writes with "coherency=full" need to take EX cluster
2329          * inode_lock to guarantee coherency.
2330          */
2331         if (direct_io && full_coherency) {
2332                 /*
2333                  * We need to take and drop the inode lock to force
2334                  * other nodes to drop their caches.  Buffered I/O
2335                  * already does this in write_begin().
2336                  */
2337                 if (nowait)
2338                         ret = ocfs2_try_inode_lock(inode, NULL, 1);
2339                 else
2340                         ret = ocfs2_inode_lock(inode, NULL, 1);
2341                 if (ret < 0) {
2342                         if (ret != -EAGAIN)
2343                                 mlog_errno(ret);
2344                         goto out;
2345                 }
2346
2347                 ocfs2_inode_unlock(inode, 1);
2348         }
2349
2350         ret = generic_write_checks(iocb, from);
2351         if (ret <= 0) {
2352                 if (ret)
2353                         mlog_errno(ret);
2354                 goto out;
2355         }
2356         count = ret;
2357
2358         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2359         if (ret < 0) {
2360                 if (ret != -EAGAIN)
2361                         mlog_errno(ret);
2362                 goto out;
2363         }
2364
2365         if (direct_io && !is_sync_kiocb(iocb) &&
2366             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2367                 /*
2368                  * Make it a sync io if it's an unaligned aio.
2369                  */
2370                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2371         }
2372
2373         /* communicate with ocfs2_dio_end_io */
2374         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2375
2376         written = __generic_file_write_iter(iocb, from);
2377         /* buffered aio wouldn't have proper lock coverage today */
2378         BUG_ON(written == -EIOCBQUEUED && !direct_io);
2379
2380         /*
2381          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2382          * function pointer which is called when o_direct io completes so that
2383          * it can unlock our rw lock.
2384          * Unfortunately there are error cases which call end_io and others
2385          * that don't.  so we don't have to unlock the rw_lock if either an
2386          * async dio is going to do it in the future or an end_io after an
2387          * error has already done it.
2388          */
2389         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2390                 rw_level = -1;
2391         }
2392
2393         if (unlikely(written <= 0))
2394                 goto out;
2395
2396         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2397             IS_SYNC(inode)) {
2398                 ret = filemap_fdatawrite_range(file->f_mapping,
2399                                                iocb->ki_pos - written,
2400                                                iocb->ki_pos - 1);
2401                 if (ret < 0)
2402                         written = ret;
2403
2404                 if (!ret) {
2405                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2406                         if (ret < 0)
2407                                 written = ret;
2408                 }
2409
2410                 if (!ret)
2411                         ret = filemap_fdatawait_range(file->f_mapping,
2412                                                       iocb->ki_pos - written,
2413                                                       iocb->ki_pos - 1);
2414         }
2415
2416 out:
2417         if (saved_ki_complete)
2418                 xchg(&iocb->ki_complete, saved_ki_complete);
2419
2420         if (rw_level != -1)
2421                 ocfs2_rw_unlock(inode, rw_level);
2422
2423 out_mutex:
2424         inode_unlock(inode);
2425
2426         if (written)
2427                 ret = written;
2428         return ret;
2429 }
2430
2431 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2432                                    struct iov_iter *to)
2433 {
2434         int ret = 0, rw_level = -1, lock_level = 0;
2435         struct file *filp = iocb->ki_filp;
2436         struct inode *inode = file_inode(filp);
2437         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2438         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2439
2440         trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2441                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2442                         filp->f_path.dentry->d_name.len,
2443                         filp->f_path.dentry->d_name.name,
2444                         to->nr_segs);   /* GRRRRR */
2445
2446
2447         if (!inode) {
2448                 ret = -EINVAL;
2449                 mlog_errno(ret);
2450                 goto bail;
2451         }
2452
2453         if (!direct_io && nowait)
2454                 return -EOPNOTSUPP;
2455
2456         /*
2457          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2458          * need locks to protect pending reads from racing with truncate.
2459          */
2460         if (direct_io) {
2461                 if (nowait)
2462                         ret = ocfs2_try_rw_lock(inode, 0);
2463                 else
2464                         ret = ocfs2_rw_lock(inode, 0);
2465
2466                 if (ret < 0) {
2467                         if (ret != -EAGAIN)
2468                                 mlog_errno(ret);
2469                         goto bail;
2470                 }
2471                 rw_level = 0;
2472                 /* communicate with ocfs2_dio_end_io */
2473                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2474         }
2475
2476         /*
2477          * We're fine letting folks race truncates and extending
2478          * writes with read across the cluster, just like they can
2479          * locally. Hence no rw_lock during read.
2480          *
2481          * Take and drop the meta data lock to update inode fields
2482          * like i_size. This allows the checks down below
2483          * generic_file_read_iter() a chance of actually working.
2484          */
2485         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2486                                      !nowait);
2487         if (ret < 0) {
2488                 if (ret != -EAGAIN)
2489                         mlog_errno(ret);
2490                 goto bail;
2491         }
2492         ocfs2_inode_unlock(inode, lock_level);
2493
2494         ret = generic_file_read_iter(iocb, to);
2495         trace_generic_file_read_iter_ret(ret);
2496
2497         /* buffered aio wouldn't have proper lock coverage today */
2498         BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2499
2500         /* see ocfs2_file_write_iter */
2501         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2502                 rw_level = -1;
2503         }
2504
2505 bail:
2506         if (rw_level != -1)
2507                 ocfs2_rw_unlock(inode, rw_level);
2508
2509         return ret;
2510 }
2511
2512 /* Refer generic_file_llseek_unlocked() */
2513 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2514 {
2515         struct inode *inode = file->f_mapping->host;
2516         int ret = 0;
2517
2518         inode_lock(inode);
2519
2520         switch (whence) {
2521         case SEEK_SET:
2522                 break;
2523         case SEEK_END:
2524                 /* SEEK_END requires the OCFS2 inode lock for the file
2525                  * because it references the file's size.
2526                  */
2527                 ret = ocfs2_inode_lock(inode, NULL, 0);
2528                 if (ret < 0) {
2529                         mlog_errno(ret);
2530                         goto out;
2531                 }
2532                 offset += i_size_read(inode);
2533                 ocfs2_inode_unlock(inode, 0);
2534                 break;
2535         case SEEK_CUR:
2536                 if (offset == 0) {
2537                         offset = file->f_pos;
2538                         goto out;
2539                 }
2540                 offset += file->f_pos;
2541                 break;
2542         case SEEK_DATA:
2543         case SEEK_HOLE:
2544                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2545                 if (ret)
2546                         goto out;
2547                 break;
2548         default:
2549                 ret = -EINVAL;
2550                 goto out;
2551         }
2552
2553         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2554
2555 out:
2556         inode_unlock(inode);
2557         if (ret)
2558                 return ret;
2559         return offset;
2560 }
2561
2562 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2563                                      struct file *file_out, loff_t pos_out,
2564                                      loff_t len, unsigned int remap_flags)
2565 {
2566         struct inode *inode_in = file_inode(file_in);
2567         struct inode *inode_out = file_inode(file_out);
2568         struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2569         struct buffer_head *in_bh = NULL, *out_bh = NULL;
2570         bool same_inode = (inode_in == inode_out);
2571         loff_t remapped = 0;
2572         ssize_t ret;
2573
2574         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2575                 return -EINVAL;
2576         if (!ocfs2_refcount_tree(osb))
2577                 return -EOPNOTSUPP;
2578         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2579                 return -EROFS;
2580
2581         /* Lock both files against IO */
2582         ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2583         if (ret)
2584                 return ret;
2585
2586         /* Check file eligibility and prepare for block sharing. */
2587         ret = -EINVAL;
2588         if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2589             (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2590                 goto out_unlock;
2591
2592         ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2593                         &len, remap_flags);
2594         if (ret < 0 || len == 0)
2595                 goto out_unlock;
2596
2597         /* Lock out changes to the allocation maps and remap. */
2598         down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2599         if (!same_inode)
2600                 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2601                                   SINGLE_DEPTH_NESTING);
2602
2603         /* Zap any page cache for the destination file's range. */
2604         truncate_inode_pages_range(&inode_out->i_data,
2605                                    round_down(pos_out, PAGE_SIZE),
2606                                    round_up(pos_out + len, PAGE_SIZE) - 1);
2607
2608         remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2609                         inode_out, out_bh, pos_out, len);
2610         up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2611         if (!same_inode)
2612                 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2613         if (remapped < 0) {
2614                 ret = remapped;
2615                 mlog_errno(ret);
2616                 goto out_unlock;
2617         }
2618
2619         /*
2620          * Empty the extent map so that we may get the right extent
2621          * record from the disk.
2622          */
2623         ocfs2_extent_map_trunc(inode_in, 0);
2624         ocfs2_extent_map_trunc(inode_out, 0);
2625
2626         ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2627         if (ret) {
2628                 mlog_errno(ret);
2629                 goto out_unlock;
2630         }
2631
2632 out_unlock:
2633         ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2634         return remapped > 0 ? remapped : ret;
2635 }
2636
2637 const struct inode_operations ocfs2_file_iops = {
2638         .setattr        = ocfs2_setattr,
2639         .getattr        = ocfs2_getattr,
2640         .permission     = ocfs2_permission,
2641         .listxattr      = ocfs2_listxattr,
2642         .fiemap         = ocfs2_fiemap,
2643         .get_acl        = ocfs2_iop_get_acl,
2644         .set_acl        = ocfs2_iop_set_acl,
2645 };
2646
2647 const struct inode_operations ocfs2_special_file_iops = {
2648         .setattr        = ocfs2_setattr,
2649         .getattr        = ocfs2_getattr,
2650         .permission     = ocfs2_permission,
2651         .get_acl        = ocfs2_iop_get_acl,
2652         .set_acl        = ocfs2_iop_set_acl,
2653 };
2654
2655 /*
2656  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2657  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2658  */
2659 const struct file_operations ocfs2_fops = {
2660         .llseek         = ocfs2_file_llseek,
2661         .mmap           = ocfs2_mmap,
2662         .fsync          = ocfs2_sync_file,
2663         .release        = ocfs2_file_release,
2664         .open           = ocfs2_file_open,
2665         .read_iter      = ocfs2_file_read_iter,
2666         .write_iter     = ocfs2_file_write_iter,
2667         .unlocked_ioctl = ocfs2_ioctl,
2668 #ifdef CONFIG_COMPAT
2669         .compat_ioctl   = ocfs2_compat_ioctl,
2670 #endif
2671         .lock           = ocfs2_lock,
2672         .flock          = ocfs2_flock,
2673         .splice_read    = generic_file_splice_read,
2674         .splice_write   = iter_file_splice_write,
2675         .fallocate      = ocfs2_fallocate,
2676         .remap_file_range = ocfs2_remap_file_range,
2677 };
2678
2679 const struct file_operations ocfs2_dops = {
2680         .llseek         = generic_file_llseek,
2681         .read           = generic_read_dir,
2682         .iterate        = ocfs2_readdir,
2683         .fsync          = ocfs2_sync_file,
2684         .release        = ocfs2_dir_release,
2685         .open           = ocfs2_dir_open,
2686         .unlocked_ioctl = ocfs2_ioctl,
2687 #ifdef CONFIG_COMPAT
2688         .compat_ioctl   = ocfs2_compat_ioctl,
2689 #endif
2690         .lock           = ocfs2_lock,
2691         .flock          = ocfs2_flock,
2692 };
2693
2694 /*
2695  * POSIX-lockless variants of our file_operations.
2696  *
2697  * These will be used if the underlying cluster stack does not support
2698  * posix file locking, if the user passes the "localflocks" mount
2699  * option, or if we have a local-only fs.
2700  *
2701  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2702  * so we still want it in the case of no stack support for
2703  * plocks. Internally, it will do the right thing when asked to ignore
2704  * the cluster.
2705  */
2706 const struct file_operations ocfs2_fops_no_plocks = {
2707         .llseek         = ocfs2_file_llseek,
2708         .mmap           = ocfs2_mmap,
2709         .fsync          = ocfs2_sync_file,
2710         .release        = ocfs2_file_release,
2711         .open           = ocfs2_file_open,
2712         .read_iter      = ocfs2_file_read_iter,
2713         .write_iter     = ocfs2_file_write_iter,
2714         .unlocked_ioctl = ocfs2_ioctl,
2715 #ifdef CONFIG_COMPAT
2716         .compat_ioctl   = ocfs2_compat_ioctl,
2717 #endif
2718         .flock          = ocfs2_flock,
2719         .splice_read    = generic_file_splice_read,
2720         .splice_write   = iter_file_splice_write,
2721         .fallocate      = ocfs2_fallocate,
2722         .remap_file_range = ocfs2_remap_file_range,
2723 };
2724
2725 const struct file_operations ocfs2_dops_no_plocks = {
2726         .llseek         = generic_file_llseek,
2727         .read           = generic_read_dir,
2728         .iterate        = ocfs2_readdir,
2729         .fsync          = ocfs2_sync_file,
2730         .release        = ocfs2_dir_release,
2731         .open           = ocfs2_dir_open,
2732         .unlocked_ioctl = ocfs2_ioctl,
2733 #ifdef CONFIG_COMPAT
2734         .compat_ioctl   = ocfs2_compat_ioctl,
2735 #endif
2736         .flock          = ocfs2_flock,
2737 };