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);
 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                                                       loff_t start_byte,
 711                                                       loff_t length)
 712 {
 713         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 714         handle_t *handle = NULL;
 715         int ret = 0;
 716
 717         if (!ocfs2_should_order_data(inode))
 718                 goto out;
 719
 720         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 721         if (IS_ERR(handle)) {
 722                 ret = -ENOMEM;
 723                 mlog_errno(ret);
 724                 goto out;
 725         }
 726
 727         ret = ocfs2_jbd2_inode_add_write(handle, inode, start_byte, length);
 728         if (ret < 0) {
 729                 mlog_errno(ret);
 730                 goto out;
 731         }
 732
 733         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 734                                       OCFS2_JOURNAL_ACCESS_WRITE);
 735         if (ret)
 736                 mlog_errno(ret);
 737         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 738
 739 out:
 740         if (ret) {
 741                 if (!IS_ERR(handle))
 742                         ocfs2_commit_trans(osb, handle);
 743                 handle = ERR_PTR(ret);
 744         }
 745         return handle;
 746 }
 747
 748 /* Some parts of this taken from generic_cont_expand, which turned out
 749  * to be too fragile to do exactly what we need without us having to
 750  * worry about recursive locking in ->write_begin() and ->write_end(). */
 751 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 752                                  u64 abs_to, struct buffer_head *di_bh)
 753 {
 754         struct address_space *mapping = inode->i_mapping;
 755         struct page *page;
 756         unsigned long index = abs_from >> PAGE_SHIFT;
 757         handle_t *handle;
 758         int ret = 0;
 759         unsigned zero_from, zero_to, block_start, block_end;
 760         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 761
 762         BUG_ON(abs_from >= abs_to);
 763         BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 764         BUG_ON(abs_from & (inode->i_blkbits - 1));
 765
 766         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh,
 767                                                       abs_from,
 768                                                       abs_to - abs_from);
 769         if (IS_ERR(handle)) {
 770                 ret = PTR_ERR(handle);
 771                 goto out;
 772         }
 773
 774         page = find_or_create_page(mapping, index, GFP_NOFS);
 775         if (!page) {
 776                 ret = -ENOMEM;
 777                 mlog_errno(ret);
 778                 goto out_commit_trans;
 779         }
 780
 781         /* Get the offsets within the page that we want to zero */
 782         zero_from = abs_from & (PAGE_SIZE - 1);
 783         zero_to = abs_to & (PAGE_SIZE - 1);
 784         if (!zero_to)
 785                 zero_to = PAGE_SIZE;
 786
 787         trace_ocfs2_write_zero_page(
 788                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 789                         (unsigned long long)abs_from,
 790                         (unsigned long long)abs_to,
 791                         index, zero_from, zero_to);
 792
 793         /* We know that zero_from is block aligned */
 794         for (block_start = zero_from; block_start < zero_to;
 795              block_start = block_end) {
 796                 block_end = block_start + i_blocksize(inode);
 797
 798                 /*
 799                  * block_start is block-aligned.  Bump it by one to force
 800                  * __block_write_begin and block_commit_write to zero the
 801                  * whole block.
 802                  */
 803                 ret = __block_write_begin(page, block_start + 1, 0,
 804                                           ocfs2_get_block);
 805                 if (ret < 0) {
 806                         mlog_errno(ret);
 807                         goto out_unlock;
 808                 }
 809
 810
 811                 /* must not update i_size! */
 812                 ret = block_commit_write(page, block_start + 1,
 813                                          block_start + 1);
 814                 if (ret < 0)
 815                         mlog_errno(ret);
 816                 else
 817                         ret = 0;
 818         }
 819
 820         /*
 821          * fs-writeback will release the dirty pages without page lock
 822          * whose offset are over inode size, the release happens at
 823          * block_write_full_page().
 824          */
 825         i_size_write(inode, abs_to);
 826         inode->i_blocks = ocfs2_inode_sector_count(inode);
 827         di->i_size = cpu_to_le64((u64)i_size_read(inode));
 828         inode->i_mtime = inode->i_ctime = current_time(inode);
 829         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 830         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 831         di->i_mtime_nsec = di->i_ctime_nsec;
 832         if (handle) {
 833                 ocfs2_journal_dirty(handle, di_bh);
 834                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 835         }
 836
 837 out_unlock:
 838         unlock_page(page);
 839         put_page(page);
 840 out_commit_trans:
 841         if (handle)
 842                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 843 out:
 844         return ret;
 845 }
 846
 847 /*
 848  * Find the next range to zero.  We do this in terms of bytes because
 849  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 850  * pagecache.  We may return multiple extents.
 851  *
 852  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 853  * needs to be zeroed.  range_start and range_end return the next zeroing
 854  * range.  A subsequent call should pass the previous range_end as its
 855  * zero_start.  If range_end is 0, there's nothing to do.
 856  *
 857  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 858  */
 859 static int ocfs2_zero_extend_get_range(struct inode *inode,
 860                                        struct buffer_head *di_bh,
 861                                        u64 zero_start, u64 zero_end,
 862                                        u64 *range_start, u64 *range_end)
 863 {
 864         int rc = 0, needs_cow = 0;
 865         u32 p_cpos, zero_clusters = 0;
 866         u32 zero_cpos =
 867                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 868         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 869         unsigned int num_clusters = 0;
 870         unsigned int ext_flags = 0;
 871
 872         while (zero_cpos < last_cpos) {
 873                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 874                                         &num_clusters, &ext_flags);
 875                 if (rc) {
 876                         mlog_errno(rc);
 877                         goto out;
 878                 }
 879
 880                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 881                         zero_clusters = num_clusters;
 882                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 883                                 needs_cow = 1;
 884                         break;
 885                 }
 886
 887                 zero_cpos += num_clusters;
 888         }
 889         if (!zero_clusters) {
 890                 *range_end = 0;
 891                 goto out;
 892         }
 893
 894         while ((zero_cpos + zero_clusters) < last_cpos) {
 895                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 896                                         &p_cpos, &num_clusters,
 897                                         &ext_flags);
 898                 if (rc) {
 899                         mlog_errno(rc);
 900                         goto out;
 901                 }
 902
 903                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 904                         break;
 905                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 906                         needs_cow = 1;
 907                 zero_clusters += num_clusters;
 908         }
 909         if ((zero_cpos + zero_clusters) > last_cpos)
 910                 zero_clusters = last_cpos - zero_cpos;
 911
 912         if (needs_cow) {
 913                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 914                                         zero_clusters, UINT_MAX);
 915                 if (rc) {
 916                         mlog_errno(rc);
 917                         goto out;
 918                 }
 919         }
 920
 921         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 922         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 923                                              zero_cpos + zero_clusters);
 924
 925 out:
 926         return rc;
 927 }
 928
 929 /*
 930  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 931  * has made sure that the entire range needs zeroing.
 932  */
 933 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 934                                    u64 range_end, struct buffer_head *di_bh)
 935 {
 936         int rc = 0;
 937         u64 next_pos;
 938         u64 zero_pos = range_start;
 939
 940         trace_ocfs2_zero_extend_range(
 941                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 942                         (unsigned long long)range_start,
 943                         (unsigned long long)range_end);
 944         BUG_ON(range_start >= range_end);
 945
 946         while (zero_pos < range_end) {
 947                 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 948                 if (next_pos > range_end)
 949                         next_pos = range_end;
 950                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 951                 if (rc < 0) {
 952                         mlog_errno(rc);
 953                         break;
 954                 }
 955                 zero_pos = next_pos;
 956
 957                 /*
 958                  * Very large extends have the potential to lock up
 959                  * the cpu for extended periods of time.
 960                  */
 961                 cond_resched();
 962         }
 963
 964         return rc;
 965 }
 966
 967 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 968                       loff_t zero_to_size)
 969 {
 970         int ret = 0;
 971         u64 zero_start, range_start = 0, range_end = 0;
 972         struct super_block *sb = inode->i_sb;
 973
 974         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 975         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 976                                 (unsigned long long)zero_start,
 977                                 (unsigned long long)i_size_read(inode));
 978         while (zero_start < zero_to_size) {
 979                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 980                                                   zero_to_size,
 981                                                   &range_start,
 982                                                   &range_end);
 983                 if (ret) {
 984                         mlog_errno(ret);
 985                         break;
 986                 }
 987                 if (!range_end)
 988                         break;
 989                 /* Trim the ends */
 990                 if (range_start < zero_start)
 991                         range_start = zero_start;
 992                 if (range_end > zero_to_size)
 993                         range_end = zero_to_size;
 994
 995                 ret = ocfs2_zero_extend_range(inode, range_start,
 996                                               range_end, di_bh);
 997                 if (ret) {
 998                         mlog_errno(ret);
 999                         break;
1000                 }
1001                 zero_start = range_end;
1002         }
1003
1004         return ret;
1005 }
1006
1007 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1008                           u64 new_i_size, u64 zero_to)
1009 {
1010         int ret;
1011         u32 clusters_to_add;
1012         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1013
1014         /*
1015          * Only quota files call this without a bh, and they can't be
1016          * refcounted.
1017          */
1018         BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1019         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1020
1021         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1022         if (clusters_to_add < oi->ip_clusters)
1023                 clusters_to_add = 0;
1024         else
1025                 clusters_to_add -= oi->ip_clusters;
1026
1027         if (clusters_to_add) {
1028                 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1029                                               clusters_to_add, 0);
1030                 if (ret) {
1031                         mlog_errno(ret);
1032                         goto out;
1033                 }
1034         }
1035
1036         /*
1037          * Call this even if we don't add any clusters to the tree. We
1038          * still need to zero the area between the old i_size and the
1039          * new i_size.
1040          */
1041         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1042         if (ret < 0)
1043                 mlog_errno(ret);
1044
1045 out:
1046         return ret;
1047 }
1048
1049 static int ocfs2_extend_file(struct inode *inode,
1050                              struct buffer_head *di_bh,
1051                              u64 new_i_size)
1052 {
1053         int ret = 0;
1054         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1055
1056         BUG_ON(!di_bh);
1057
1058         /* setattr sometimes calls us like this. */
1059         if (new_i_size == 0)
1060                 goto out;
1061
1062         if (i_size_read(inode) == new_i_size)
1063                 goto out;
1064         BUG_ON(new_i_size < i_size_read(inode));
1065
1066         /*
1067          * The alloc sem blocks people in read/write from reading our
1068          * allocation until we're done changing it. We depend on
1069          * i_mutex to block other extend/truncate calls while we're
1070          * here.  We even have to hold it for sparse files because there
1071          * might be some tail zeroing.
1072          */
1073         down_write(&oi->ip_alloc_sem);
1074
1075         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1076                 /*
1077                  * We can optimize small extends by keeping the inodes
1078                  * inline data.
1079                  */
1080                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1081                         up_write(&oi->ip_alloc_sem);
1082                         goto out_update_size;
1083                 }
1084
1085                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1086                 if (ret) {
1087                         up_write(&oi->ip_alloc_sem);
1088                         mlog_errno(ret);
1089                         goto out;
1090                 }
1091         }
1092
1093         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1094                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1095         else
1096                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1097                                             new_i_size);
1098
1099         up_write(&oi->ip_alloc_sem);
1100
1101         if (ret < 0) {
1102                 mlog_errno(ret);
1103                 goto out;
1104         }
1105
1106 out_update_size:
1107         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1108         if (ret < 0)
1109                 mlog_errno(ret);
1110
1111 out:
1112         return ret;
1113 }
1114
1115 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1116 {
1117         int status = 0, size_change;
1118         int inode_locked = 0;
1119         struct inode *inode = d_inode(dentry);
1120         struct super_block *sb = inode->i_sb;
1121         struct ocfs2_super *osb = OCFS2_SB(sb);
1122         struct buffer_head *bh = NULL;
1123         handle_t *handle = NULL;
1124         struct dquot *transfer_to[MAXQUOTAS] = { };
1125         int qtype;
1126         int had_lock;
1127         struct ocfs2_lock_holder oh;
1128
1129         trace_ocfs2_setattr(inode, dentry,
1130                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1131                             dentry->d_name.len, dentry->d_name.name,
1132                             attr->ia_valid, attr->ia_mode,
1133                             from_kuid(&init_user_ns, attr->ia_uid),
1134                             from_kgid(&init_user_ns, attr->ia_gid));
1135
1136         /* ensuring we don't even attempt to truncate a symlink */
1137         if (S_ISLNK(inode->i_mode))
1138                 attr->ia_valid &= ~ATTR_SIZE;
1139
1140 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1141                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1142         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1143                 return 0;
1144
1145         status = setattr_prepare(dentry, attr);
1146         if (status)
1147                 return status;
1148
1149         if (is_quota_modification(inode, attr)) {
1150                 status = dquot_initialize(inode);
1151                 if (status)
1152                         return status;
1153         }
1154         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1155         if (size_change) {
1156                 /*
1157                  * Here we should wait dio to finish before inode lock
1158                  * to avoid a deadlock between ocfs2_setattr() and
1159                  * ocfs2_dio_end_io_write()
1160                  */
1161                 inode_dio_wait(inode);
1162
1163                 status = ocfs2_rw_lock(inode, 1);
1164                 if (status < 0) {
1165                         mlog_errno(status);
1166                         goto bail;
1167                 }
1168         }
1169
1170         had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1171         if (had_lock < 0) {
1172                 status = had_lock;
1173                 goto bail_unlock_rw;
1174         } else if (had_lock) {
1175                 /*
1176                  * As far as we know, ocfs2_setattr() could only be the first
1177                  * VFS entry point in the call chain of recursive cluster
1178                  * locking issue.
1179                  *
1180                  * For instance:
1181                  * chmod_common()
1182                  *  notify_change()
1183                  *   ocfs2_setattr()
1184                  *    posix_acl_chmod()
1185                  *     ocfs2_iop_get_acl()
1186                  *
1187                  * But, we're not 100% sure if it's always true, because the
1188                  * ordering of the VFS entry points in the call chain is out
1189                  * of our control. So, we'd better dump the stack here to
1190                  * catch the other cases of recursive locking.
1191                  */
1192                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1193                 dump_stack();
1194         }
1195         inode_locked = 1;
1196
1197         if (size_change) {
1198                 status = inode_newsize_ok(inode, attr->ia_size);
1199                 if (status)
1200                         goto bail_unlock;
1201
1202                 if (i_size_read(inode) >= attr->ia_size) {
1203                         if (ocfs2_should_order_data(inode)) {
1204                                 status = ocfs2_begin_ordered_truncate(inode,
1205                                                                       attr->ia_size);
1206                                 if (status)
1207                                         goto bail_unlock;
1208                         }
1209                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1210                 } else
1211                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1212                 if (status < 0) {
1213                         if (status != -ENOSPC)
1214                                 mlog_errno(status);
1215                         status = -ENOSPC;
1216                         goto bail_unlock;
1217                 }
1218         }
1219
1220         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1221             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1222                 /*
1223                  * Gather pointers to quota structures so that allocation /
1224                  * freeing of quota structures happens here and not inside
1225                  * dquot_transfer() where we have problems with lock ordering
1226                  */
1227                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1228                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1229                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1230                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1231                         if (IS_ERR(transfer_to[USRQUOTA])) {
1232                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1233                                 transfer_to[USRQUOTA] = NULL;
1234                                 goto bail_unlock;
1235                         }
1236                 }
1237                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1238                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1239                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1240                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1241                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1242                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1243                                 transfer_to[GRPQUOTA] = NULL;
1244                                 goto bail_unlock;
1245                         }
1246                 }
1247                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1248                                            2 * ocfs2_quota_trans_credits(sb));
1249                 if (IS_ERR(handle)) {
1250                         status = PTR_ERR(handle);
1251                         mlog_errno(status);
1252                         goto bail_unlock;
1253                 }
1254                 status = __dquot_transfer(inode, transfer_to);
1255                 if (status < 0)
1256                         goto bail_commit;
1257         } else {
1258                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1259                 if (IS_ERR(handle)) {
1260                         status = PTR_ERR(handle);
1261                         mlog_errno(status);
1262                         goto bail_unlock;
1263                 }
1264         }
1265
1266         setattr_copy(inode, attr);
1267         mark_inode_dirty(inode);
1268
1269         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1270         if (status < 0)
1271                 mlog_errno(status);
1272
1273 bail_commit:
1274         ocfs2_commit_trans(osb, handle);
1275 bail_unlock:
1276         if (status && inode_locked) {
1277                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1278                 inode_locked = 0;
1279         }
1280 bail_unlock_rw:
1281         if (size_change)
1282                 ocfs2_rw_unlock(inode, 1);
1283 bail:
1284
1285         /* Release quota pointers in case we acquired them */
1286         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1287                 dqput(transfer_to[qtype]);
1288
1289         if (!status && attr->ia_valid & ATTR_MODE) {
1290                 status = ocfs2_acl_chmod(inode, bh);
1291                 if (status < 0)
1292                         mlog_errno(status);
1293         }
1294         if (inode_locked)
1295                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1296
1297         brelse(bh);
1298         return status;
1299 }
1300
1301 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1302                   u32 request_mask, unsigned int flags)
1303 {
1304         struct inode *inode = d_inode(path->dentry);
1305         struct super_block *sb = path->dentry->d_sb;
1306         struct ocfs2_super *osb = sb->s_fs_info;
1307         int err;
1308
1309         err = ocfs2_inode_revalidate(path->dentry);
1310         if (err) {
1311                 if (err != -ENOENT)
1312                         mlog_errno(err);
1313                 goto bail;
1314         }
1315
1316         generic_fillattr(inode, stat);
1317         /*
1318          * If there is inline data in the inode, the inode will normally not
1319          * have data blocks allocated (it may have an external xattr block).
1320          * Report at least one sector for such files, so tools like tar, rsync,
1321          * others don't incorrectly think the file is completely sparse.
1322          */
1323         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1324                 stat->blocks += (stat->size + 511)>>9;
1325
1326         /* We set the blksize from the cluster size for performance */
1327         stat->blksize = osb->s_clustersize;
1328
1329 bail:
1330         return err;
1331 }
1332
1333 int ocfs2_permission(struct inode *inode, int mask)
1334 {
1335         int ret, had_lock;
1336         struct ocfs2_lock_holder oh;
1337
1338         if (mask & MAY_NOT_BLOCK)
1339                 return -ECHILD;
1340
1341         had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1342         if (had_lock < 0) {
1343                 ret = had_lock;
1344                 goto out;
1345         } else if (had_lock) {
1346                 /* See comments in ocfs2_setattr() for details.
1347                  * The call chain of this case could be:
1348                  * do_sys_open()
1349                  *  may_open()
1350                  *   inode_permission()
1351                  *    ocfs2_permission()
1352                  *     ocfs2_iop_get_acl()
1353                  */
1354                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1355                 dump_stack();
1356         }
1357
1358         ret = generic_permission(&init_user_ns, inode, mask);
1359
1360         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1361 out:
1362         return ret;
1363 }
1364
1365 static int __ocfs2_write_remove_suid(struct inode *inode,
1366                                      struct buffer_head *bh)
1367 {
1368         int ret;
1369         handle_t *handle;
1370         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1371         struct ocfs2_dinode *di;
1372
1373         trace_ocfs2_write_remove_suid(
1374                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1375                         inode->i_mode);
1376
1377         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1378         if (IS_ERR(handle)) {
1379                 ret = PTR_ERR(handle);
1380                 mlog_errno(ret);
1381                 goto out;
1382         }
1383
1384         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1385                                       OCFS2_JOURNAL_ACCESS_WRITE);
1386         if (ret < 0) {
1387                 mlog_errno(ret);
1388                 goto out_trans;
1389         }
1390
1391         inode->i_mode &= ~S_ISUID;
1392         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1393                 inode->i_mode &= ~S_ISGID;
1394
1395         di = (struct ocfs2_dinode *) bh->b_data;
1396         di->i_mode = cpu_to_le16(inode->i_mode);
1397         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1398
1399         ocfs2_journal_dirty(handle, bh);
1400
1401 out_trans:
1402         ocfs2_commit_trans(osb, handle);
1403 out:
1404         return ret;
1405 }
1406
1407 static int ocfs2_write_remove_suid(struct inode *inode)
1408 {
1409         int ret;
1410         struct buffer_head *bh = NULL;
1411
1412         ret = ocfs2_read_inode_block(inode, &bh);
1413         if (ret < 0) {
1414                 mlog_errno(ret);
1415                 goto out;
1416         }
1417
1418         ret =  __ocfs2_write_remove_suid(inode, bh);
1419 out:
1420         brelse(bh);
1421         return ret;
1422 }
1423
1424 /*
1425  * Allocate enough extents to cover the region starting at byte offset
1426  * start for len bytes. Existing extents are skipped, any extents
1427  * added are marked as "unwritten".
1428  */
1429 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1430                                             u64 start, u64 len)
1431 {
1432         int ret;
1433         u32 cpos, phys_cpos, clusters, alloc_size;
1434         u64 end = start + len;
1435         struct buffer_head *di_bh = NULL;
1436
1437         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1438                 ret = ocfs2_read_inode_block(inode, &di_bh);
1439                 if (ret) {
1440                         mlog_errno(ret);
1441                         goto out;
1442                 }
1443
1444                 /*
1445                  * Nothing to do if the requested reservation range
1446                  * fits within the inode.
1447                  */
1448                 if (ocfs2_size_fits_inline_data(di_bh, end))
1449                         goto out;
1450
1451                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1452                 if (ret) {
1453                         mlog_errno(ret);
1454                         goto out;
1455                 }
1456         }
1457
1458         /*
1459          * We consider both start and len to be inclusive.
1460          */
1461         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1462         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1463         clusters -= cpos;
1464
1465         while (clusters) {
1466                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1467                                          &alloc_size, NULL);
1468                 if (ret) {
1469                         mlog_errno(ret);
1470                         goto out;
1471                 }
1472
1473                 /*
1474                  * Hole or existing extent len can be arbitrary, so
1475                  * cap it to our own allocation request.
1476                  */
1477                 if (alloc_size > clusters)
1478                         alloc_size = clusters;
1479
1480                 if (phys_cpos) {
1481                         /*
1482                          * We already have an allocation at this
1483                          * region so we can safely skip it.
1484                          */
1485                         goto next;
1486                 }
1487
1488                 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1489                 if (ret) {
1490                         if (ret != -ENOSPC)
1491                                 mlog_errno(ret);
1492                         goto out;
1493                 }
1494
1495 next:
1496                 cpos += alloc_size;
1497                 clusters -= alloc_size;
1498         }
1499
1500         ret = 0;
1501 out:
1502
1503         brelse(di_bh);
1504         return ret;
1505 }
1506
1507 /*
1508  * Truncate a byte range, avoiding pages within partial clusters. This
1509  * preserves those pages for the zeroing code to write to.
1510  */
1511 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1512                                          u64 byte_len)
1513 {
1514         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1515         loff_t start, end;
1516         struct address_space *mapping = inode->i_mapping;
1517
1518         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1519         end = byte_start + byte_len;
1520         end = end & ~(osb->s_clustersize - 1);
1521
1522         if (start < end) {
1523                 unmap_mapping_range(mapping, start, end - start, 0);
1524                 truncate_inode_pages_range(mapping, start, end - 1);
1525         }
1526 }
1527
1528 static int ocfs2_zero_partial_clusters(struct inode *inode,
1529                                        u64 start, u64 len)
1530 {
1531         int ret = 0;
1532         u64 tmpend = 0;
1533         u64 end = start + len;
1534         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1535         unsigned int csize = osb->s_clustersize;
1536         handle_t *handle;
1537
1538         /*
1539          * The "start" and "end" values are NOT necessarily part of
1540          * the range whose allocation is being deleted. Rather, this
1541          * is what the user passed in with the request. We must zero
1542          * partial clusters here. There's no need to worry about
1543          * physical allocation - the zeroing code knows to skip holes.
1544          */
1545         trace_ocfs2_zero_partial_clusters(
1546                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1547                 (unsigned long long)start, (unsigned long long)end);
1548
1549         /*
1550          * If both edges are on a cluster boundary then there's no
1551          * zeroing required as the region is part of the allocation to
1552          * be truncated.
1553          */
1554         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1555                 goto out;
1556
1557         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1558         if (IS_ERR(handle)) {
1559                 ret = PTR_ERR(handle);
1560                 mlog_errno(ret);
1561                 goto out;
1562         }
1563
1564         /*
1565          * If start is on a cluster boundary and end is somewhere in another
1566          * cluster, we have not COWed the cluster starting at start, unless
1567          * end is also within the same cluster. So, in this case, we skip this
1568          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1569          * to the next one.
1570          */
1571         if ((start & (csize - 1)) != 0) {
1572                 /*
1573                  * We want to get the byte offset of the end of the 1st
1574                  * cluster.
1575                  */
1576                 tmpend = (u64)osb->s_clustersize +
1577                         (start & ~(osb->s_clustersize - 1));
1578                 if (tmpend > end)
1579                         tmpend = end;
1580
1581                 trace_ocfs2_zero_partial_clusters_range1(
1582                         (unsigned long long)start,
1583                         (unsigned long long)tmpend);
1584
1585                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1586                                                     tmpend);
1587                 if (ret)
1588                         mlog_errno(ret);
1589         }
1590
1591         if (tmpend < end) {
1592                 /*
1593                  * This may make start and end equal, but the zeroing
1594                  * code will skip any work in that case so there's no
1595                  * need to catch it up here.
1596                  */
1597                 start = end & ~(osb->s_clustersize - 1);
1598
1599                 trace_ocfs2_zero_partial_clusters_range2(
1600                         (unsigned long long)start, (unsigned long long)end);
1601
1602                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1603                 if (ret)
1604                         mlog_errno(ret);
1605         }
1606         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1607
1608         ocfs2_commit_trans(osb, handle);
1609 out:
1610         return ret;
1611 }
1612
1613 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1614 {
1615         int i;
1616         struct ocfs2_extent_rec *rec = NULL;
1617
1618         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1619
1620                 rec = &el->l_recs[i];
1621
1622                 if (le32_to_cpu(rec->e_cpos) < pos)
1623                         break;
1624         }
1625
1626         return i;
1627 }
1628
1629 /*
1630  * Helper to calculate the punching pos and length in one run, we handle the
1631  * following three cases in order:
1632  *
1633  * - remove the entire record
1634  * - remove a partial record
1635  * - no record needs to be removed (hole-punching completed)
1636 */
1637 static void ocfs2_calc_trunc_pos(struct inode *inode,
1638                                  struct ocfs2_extent_list *el,
1639                                  struct ocfs2_extent_rec *rec,
1640                                  u32 trunc_start, u32 *trunc_cpos,
1641                                  u32 *trunc_len, u32 *trunc_end,
1642                                  u64 *blkno, int *done)
1643 {
1644         int ret = 0;
1645         u32 coff, range;
1646
1647         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1648
1649         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1650                 /*
1651                  * remove an entire extent record.
1652                  */
1653                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1654                 /*
1655                  * Skip holes if any.
1656                  */
1657                 if (range < *trunc_end)
1658                         *trunc_end = range;
1659                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1660                 *blkno = le64_to_cpu(rec->e_blkno);
1661                 *trunc_end = le32_to_cpu(rec->e_cpos);
1662         } else if (range > trunc_start) {
1663                 /*
1664                  * remove a partial extent record, which means we're
1665                  * removing the last extent record.
1666                  */
1667                 *trunc_cpos = trunc_start;
1668                 /*
1669                  * skip hole if any.
1670                  */
1671                 if (range < *trunc_end)
1672                         *trunc_end = range;
1673                 *trunc_len = *trunc_end - trunc_start;
1674                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1675                 *blkno = le64_to_cpu(rec->e_blkno) +
1676                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1677                 *trunc_end = trunc_start;
1678         } else {
1679                 /*
1680                  * It may have two following possibilities:
1681                  *
1682                  * - last record has been removed
1683                  * - trunc_start was within a hole
1684                  *
1685                  * both two cases mean the completion of hole punching.
1686                  */
1687                 ret = 1;
1688         }
1689
1690         *done = ret;
1691 }
1692
1693 int ocfs2_remove_inode_range(struct inode *inode,
1694                              struct buffer_head *di_bh, u64 byte_start,
1695                              u64 byte_len)
1696 {
1697         int ret = 0, flags = 0, done = 0, i;
1698         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1699         u32 cluster_in_el;
1700         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1701         struct ocfs2_cached_dealloc_ctxt dealloc;
1702         struct address_space *mapping = inode->i_mapping;
1703         struct ocfs2_extent_tree et;
1704         struct ocfs2_path *path = NULL;
1705         struct ocfs2_extent_list *el = NULL;
1706         struct ocfs2_extent_rec *rec = NULL;
1707         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1708         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1709
1710         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1711         ocfs2_init_dealloc_ctxt(&dealloc);
1712
1713         trace_ocfs2_remove_inode_range(
1714                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1715                         (unsigned long long)byte_start,
1716                         (unsigned long long)byte_len);
1717
1718         if (byte_len == 0)
1719                 return 0;
1720
1721         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1722                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1723                                             byte_start + byte_len, 0);
1724                 if (ret) {
1725                         mlog_errno(ret);
1726                         goto out;
1727                 }
1728                 /*
1729                  * There's no need to get fancy with the page cache
1730                  * truncate of an inline-data inode. We're talking
1731                  * about less than a page here, which will be cached
1732                  * in the dinode buffer anyway.
1733                  */
1734                 unmap_mapping_range(mapping, 0, 0, 0);
1735                 truncate_inode_pages(mapping, 0);
1736                 goto out;
1737         }
1738
1739         /*
1740          * For reflinks, we may need to CoW 2 clusters which might be
1741          * partially zero'd later, if hole's start and end offset were
1742          * within one cluster(means is not exactly aligned to clustersize).
1743          */
1744
1745         if (ocfs2_is_refcount_inode(inode)) {
1746                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1747                 if (ret) {
1748                         mlog_errno(ret);
1749                         goto out;
1750                 }
1751
1752                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1753                 if (ret) {
1754                         mlog_errno(ret);
1755                         goto out;
1756                 }
1757         }
1758
1759         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1760         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1761         cluster_in_el = trunc_end;
1762
1763         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1764         if (ret) {
1765                 mlog_errno(ret);
1766                 goto out;
1767         }
1768
1769         path = ocfs2_new_path_from_et(&et);
1770         if (!path) {
1771                 ret = -ENOMEM;
1772                 mlog_errno(ret);
1773                 goto out;
1774         }
1775
1776         while (trunc_end > trunc_start) {
1777
1778                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1779                                       cluster_in_el);
1780                 if (ret) {
1781                         mlog_errno(ret);
1782                         goto out;
1783                 }
1784
1785                 el = path_leaf_el(path);
1786
1787                 i = ocfs2_find_rec(el, trunc_end);
1788                 /*
1789                  * Need to go to previous extent block.
1790                  */
1791                 if (i < 0) {
1792                         if (path->p_tree_depth == 0)
1793                                 break;
1794
1795                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1796                                                             path,
1797                                                             &cluster_in_el);
1798                         if (ret) {
1799                                 mlog_errno(ret);
1800                                 goto out;
1801                         }
1802
1803                         /*
1804                          * We've reached the leftmost extent block,
1805                          * it's safe to leave.
1806                          */
1807                         if (cluster_in_el == 0)
1808                                 break;
1809
1810                         /*
1811                          * The 'pos' searched for previous extent block is
1812                          * always one cluster less than actual trunc_end.
1813                          */
1814                         trunc_end = cluster_in_el + 1;
1815
1816                         ocfs2_reinit_path(path, 1);
1817
1818                         continue;
1819
1820                 } else
1821                         rec = &el->l_recs[i];
1822
1823                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1824                                      &trunc_len, &trunc_end, &blkno, &done);
1825                 if (done)
1826                         break;
1827
1828                 flags = rec->e_flags;
1829                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1830
1831                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1832                                                phys_cpos, trunc_len, flags,
1833                                                &dealloc, refcount_loc, false);
1834                 if (ret < 0) {
1835                         mlog_errno(ret);
1836                         goto out;
1837                 }
1838
1839                 cluster_in_el = trunc_end;
1840
1841                 ocfs2_reinit_path(path, 1);
1842         }
1843
1844         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1845
1846 out:
1847         ocfs2_free_path(path);
1848         ocfs2_schedule_truncate_log_flush(osb, 1);
1849         ocfs2_run_deallocs(osb, &dealloc);
1850
1851         return ret;
1852 }
1853
1854 /*
1855  * Parts of this function taken from xfs_change_file_space()
1856  */
1857 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1858                                      loff_t f_pos, unsigned int cmd,
1859                                      struct ocfs2_space_resv *sr,
1860                                      int change_size)
1861 {
1862         int ret;
1863         s64 llen;
1864         loff_t size;
1865         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1866         struct buffer_head *di_bh = NULL;
1867         handle_t *handle;
1868         unsigned long long max_off = inode->i_sb->s_maxbytes;
1869
1870         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1871                 return -EROFS;
1872
1873         inode_lock(inode);
1874
1875         /*
1876          * This prevents concurrent writes on other nodes
1877          */
1878         ret = ocfs2_rw_lock(inode, 1);
1879         if (ret) {
1880                 mlog_errno(ret);
1881                 goto out;
1882         }
1883
1884         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1885         if (ret) {
1886                 mlog_errno(ret);
1887                 goto out_rw_unlock;
1888         }
1889
1890         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1891                 ret = -EPERM;
1892                 goto out_inode_unlock;
1893         }
1894
1895         switch (sr->l_whence) {
1896         case 0: /*SEEK_SET*/
1897                 break;
1898         case 1: /*SEEK_CUR*/
1899                 sr->l_start += f_pos;
1900                 break;
1901         case 2: /*SEEK_END*/
1902                 sr->l_start += i_size_read(inode);
1903                 break;
1904         default:
1905                 ret = -EINVAL;
1906                 goto out_inode_unlock;
1907         }
1908         sr->l_whence = 0;
1909
1910         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1911
1912         if (sr->l_start < 0
1913             || sr->l_start > max_off
1914             || (sr->l_start + llen) < 0
1915             || (sr->l_start + llen) > max_off) {
1916                 ret = -EINVAL;
1917                 goto out_inode_unlock;
1918         }
1919         size = sr->l_start + sr->l_len;
1920
1921         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1922             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1923                 if (sr->l_len <= 0) {
1924                         ret = -EINVAL;
1925                         goto out_inode_unlock;
1926                 }
1927         }
1928
1929         if (file && should_remove_suid(file->f_path.dentry)) {
1930                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1931                 if (ret) {
1932                         mlog_errno(ret);
1933                         goto out_inode_unlock;
1934                 }
1935         }
1936
1937         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1938         switch (cmd) {
1939         case OCFS2_IOC_RESVSP:
1940         case OCFS2_IOC_RESVSP64:
1941                 /*
1942                  * This takes unsigned offsets, but the signed ones we
1943                  * pass have been checked against overflow above.
1944                  */
1945                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1946                                                        sr->l_len);
1947                 break;
1948         case OCFS2_IOC_UNRESVSP:
1949         case OCFS2_IOC_UNRESVSP64:
1950                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1951                                                sr->l_len);
1952                 break;
1953         default:
1954                 ret = -EINVAL;
1955         }
1956         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1957         if (ret) {
1958                 mlog_errno(ret);
1959                 goto out_inode_unlock;
1960         }
1961
1962         /*
1963          * We update c/mtime for these changes
1964          */
1965         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1966         if (IS_ERR(handle)) {
1967                 ret = PTR_ERR(handle);
1968                 mlog_errno(ret);
1969                 goto out_inode_unlock;
1970         }
1971
1972         if (change_size && i_size_read(inode) < size)
1973                 i_size_write(inode, size);
1974
1975         inode->i_ctime = inode->i_mtime = current_time(inode);
1976         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1977         if (ret < 0)
1978                 mlog_errno(ret);
1979
1980         if (file && (file->f_flags & O_SYNC))
1981                 handle->h_sync = 1;
1982
1983         ocfs2_commit_trans(osb, handle);
1984
1985 out_inode_unlock:
1986         brelse(di_bh);
1987         ocfs2_inode_unlock(inode, 1);
1988 out_rw_unlock:
1989         ocfs2_rw_unlock(inode, 1);
1990
1991 out:
1992         inode_unlock(inode);
1993         return ret;
1994 }
1995
1996 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1997                             struct ocfs2_space_resv *sr)
1998 {
1999         struct inode *inode = file_inode(file);
2000         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2001         int ret;
2002
2003         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2004             !ocfs2_writes_unwritten_extents(osb))
2005                 return -ENOTTY;
2006         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2007                  !ocfs2_sparse_alloc(osb))
2008                 return -ENOTTY;
2009
2010         if (!S_ISREG(inode->i_mode))
2011                 return -EINVAL;
2012
2013         if (!(file->f_mode & FMODE_WRITE))
2014                 return -EBADF;
2015
2016         ret = mnt_want_write_file(file);
2017         if (ret)
2018                 return ret;
2019         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2020         mnt_drop_write_file(file);
2021         return ret;
2022 }
2023
2024 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2025                             loff_t len)
2026 {
2027         struct inode *inode = file_inode(file);
2028         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2029         struct ocfs2_space_resv sr;
2030         int change_size = 1;
2031         int cmd = OCFS2_IOC_RESVSP64;
2032
2033         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2034                 return -EOPNOTSUPP;
2035         if (!ocfs2_writes_unwritten_extents(osb))
2036                 return -EOPNOTSUPP;
2037
2038         if (mode & FALLOC_FL_KEEP_SIZE)
2039                 change_size = 0;
2040
2041         if (mode & FALLOC_FL_PUNCH_HOLE)
2042                 cmd = OCFS2_IOC_UNRESVSP64;
2043
2044         sr.l_whence = 0;
2045         sr.l_start = (s64)offset;
2046         sr.l_len = (s64)len;
2047
2048         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2049                                          change_size);
2050 }
2051
2052 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2053                                    size_t count)
2054 {
2055         int ret = 0;
2056         unsigned int extent_flags;
2057         u32 cpos, clusters, extent_len, phys_cpos;
2058         struct super_block *sb = inode->i_sb;
2059
2060         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2061             !ocfs2_is_refcount_inode(inode) ||
2062             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2063                 return 0;
2064
2065         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2066         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2067
2068         while (clusters) {
2069                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2070                                          &extent_flags);
2071                 if (ret < 0) {
2072                         mlog_errno(ret);
2073                         goto out;
2074                 }
2075
2076                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2077                         ret = 1;
2078                         break;
2079                 }
2080
2081                 if (extent_len > clusters)
2082                         extent_len = clusters;
2083
2084                 clusters -= extent_len;
2085                 cpos += extent_len;
2086         }
2087 out:
2088         return ret;
2089 }
2090
2091 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2092 {
2093         int blockmask = inode->i_sb->s_blocksize - 1;
2094         loff_t final_size = pos + count;
2095
2096         if ((pos & blockmask) || (final_size & blockmask))
2097                 return 1;
2098         return 0;
2099 }
2100
2101 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2102                                             struct buffer_head **di_bh,
2103                                             int meta_level,
2104                                             int write_sem,
2105                                             int wait)
2106 {
2107         int ret = 0;
2108
2109         if (wait)
2110                 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2111         else
2112                 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2113         if (ret < 0)
2114                 goto out;
2115
2116         if (wait) {
2117                 if (write_sem)
2118                         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2119                 else
2120                         down_read(&OCFS2_I(inode)->ip_alloc_sem);
2121         } else {
2122                 if (write_sem)
2123                         ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2124                 else
2125                         ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2126
2127                 if (!ret) {
2128                         ret = -EAGAIN;
2129                         goto out_unlock;
2130                 }
2131         }
2132
2133         return ret;
2134
2135 out_unlock:
2136         brelse(*di_bh);
2137         *di_bh = NULL;
2138         ocfs2_inode_unlock(inode, meta_level);
2139 out:
2140         return ret;
2141 }
2142
2143 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2144                                                struct buffer_head **di_bh,
2145                                                int meta_level,
2146                                                int write_sem)
2147 {
2148         if (write_sem)
2149                 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2150         else
2151                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2152
2153         brelse(*di_bh);
2154         *di_bh = NULL;
2155
2156         if (meta_level >= 0)
2157                 ocfs2_inode_unlock(inode, meta_level);
2158 }
2159
2160 static int ocfs2_prepare_inode_for_write(struct file *file,
2161                                          loff_t pos, size_t count, int wait)
2162 {
2163         int ret = 0, meta_level = 0, overwrite_io = 0;
2164         int write_sem = 0;
2165         struct dentry *dentry = file->f_path.dentry;
2166         struct inode *inode = d_inode(dentry);
2167         struct buffer_head *di_bh = NULL;
2168         u32 cpos;
2169         u32 clusters;
2170
2171         /*
2172          * We start with a read level meta lock and only jump to an ex
2173          * if we need to make modifications here.
2174          */
2175         for(;;) {
2176                 ret = ocfs2_inode_lock_for_extent_tree(inode,
2177                                                        &di_bh,
2178                                                        meta_level,
2179                                                        write_sem,
2180                                                        wait);
2181                 if (ret < 0) {
2182                         if (ret != -EAGAIN)
2183                                 mlog_errno(ret);
2184                         goto out;
2185                 }
2186
2187                 /*
2188                  * Check if IO will overwrite allocated blocks in case
2189                  * IOCB_NOWAIT flag is set.
2190                  */
2191                 if (!wait && !overwrite_io) {
2192                         overwrite_io = 1;
2193
2194                         ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2195                         if (ret < 0) {
2196                                 if (ret != -EAGAIN)
2197                                         mlog_errno(ret);
2198                                 goto out_unlock;
2199                         }
2200                 }
2201
2202                 /* Clear suid / sgid if necessary. We do this here
2203                  * instead of later in the write path because
2204                  * remove_suid() calls ->setattr without any hint that
2205                  * we may have already done our cluster locking. Since
2206                  * ocfs2_setattr() *must* take cluster locks to
2207                  * proceed, this will lead us to recursively lock the
2208                  * inode. There's also the dinode i_size state which
2209                  * can be lost via setattr during extending writes (we
2210                  * set inode->i_size at the end of a write. */
2211                 if (should_remove_suid(dentry)) {
2212                         if (meta_level == 0) {
2213                                 ocfs2_inode_unlock_for_extent_tree(inode,
2214                                                                    &di_bh,
2215                                                                    meta_level,
2216                                                                    write_sem);
2217                                 meta_level = 1;
2218                                 continue;
2219                         }
2220
2221                         ret = ocfs2_write_remove_suid(inode);
2222                         if (ret < 0) {
2223                                 mlog_errno(ret);
2224                                 goto out_unlock;
2225                         }
2226                 }
2227
2228                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2229                 if (ret == 1) {
2230                         ocfs2_inode_unlock_for_extent_tree(inode,
2231                                                            &di_bh,
2232                                                            meta_level,
2233                                                            write_sem);
2234                         meta_level = 1;
2235                         write_sem = 1;
2236                         ret = ocfs2_inode_lock_for_extent_tree(inode,
2237                                                                &di_bh,
2238                                                                meta_level,
2239                                                                write_sem,
2240                                                                wait);
2241                         if (ret < 0) {
2242                                 if (ret != -EAGAIN)
2243                                         mlog_errno(ret);
2244                                 goto out;
2245                         }
2246
2247                         cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2248                         clusters =
2249                                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2250                         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2251                 }
2252
2253                 if (ret < 0) {
2254                         if (ret != -EAGAIN)
2255                                 mlog_errno(ret);
2256                         goto out_unlock;
2257                 }
2258
2259                 break;
2260         }
2261
2262 out_unlock:
2263         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2264                                             pos, count, wait);
2265
2266         ocfs2_inode_unlock_for_extent_tree(inode,
2267                                            &di_bh,
2268                                            meta_level,
2269                                            write_sem);
2270
2271 out:
2272         return ret;
2273 }
2274
2275 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2276                                     struct iov_iter *from)
2277 {
2278         int rw_level;
2279         ssize_t written = 0;
2280         ssize_t ret;
2281         size_t count = iov_iter_count(from);
2282         struct file *file = iocb->ki_filp;
2283         struct inode *inode = file_inode(file);
2284         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2285         int full_coherency = !(osb->s_mount_opt &
2286                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2287         void *saved_ki_complete = NULL;
2288         int append_write = ((iocb->ki_pos + count) >=
2289                         i_size_read(inode) ? 1 : 0);
2290         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2291         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2292
2293         trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2294                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2295                 file->f_path.dentry->d_name.len,
2296                 file->f_path.dentry->d_name.name,
2297                 (unsigned int)from->nr_segs);   /* GRRRRR */
2298
2299         if (!direct_io && nowait)
2300                 return -EOPNOTSUPP;
2301
2302         if (count == 0)
2303                 return 0;
2304
2305         if (nowait) {
2306                 if (!inode_trylock(inode))
2307                         return -EAGAIN;
2308         } else
2309                 inode_lock(inode);
2310
2311         /*
2312          * Concurrent O_DIRECT writes are allowed with
2313          * mount_option "coherency=buffered".
2314          * For append write, we must take rw EX.
2315          */
2316         rw_level = (!direct_io || full_coherency || append_write);
2317
2318         if (nowait)
2319                 ret = ocfs2_try_rw_lock(inode, rw_level);
2320         else
2321                 ret = ocfs2_rw_lock(inode, rw_level);
2322         if (ret < 0) {
2323                 if (ret != -EAGAIN)
2324                         mlog_errno(ret);
2325                 goto out_mutex;
2326         }
2327
2328         /*
2329          * O_DIRECT writes with "coherency=full" need to take EX cluster
2330          * inode_lock to guarantee coherency.
2331          */
2332         if (direct_io && full_coherency) {
2333                 /*
2334                  * We need to take and drop the inode lock to force
2335                  * other nodes to drop their caches.  Buffered I/O
2336                  * already does this in write_begin().
2337                  */
2338                 if (nowait)
2339                         ret = ocfs2_try_inode_lock(inode, NULL, 1);
2340                 else
2341                         ret = ocfs2_inode_lock(inode, NULL, 1);
2342                 if (ret < 0) {
2343                         if (ret != -EAGAIN)
2344                                 mlog_errno(ret);
2345                         goto out;
2346                 }
2347
2348                 ocfs2_inode_unlock(inode, 1);
2349         }
2350
2351         ret = generic_write_checks(iocb, from);
2352         if (ret <= 0) {
2353                 if (ret)
2354                         mlog_errno(ret);
2355                 goto out;
2356         }
2357         count = ret;
2358
2359         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2360         if (ret < 0) {
2361                 if (ret != -EAGAIN)
2362                         mlog_errno(ret);
2363                 goto out;
2364         }
2365
2366         if (direct_io && !is_sync_kiocb(iocb) &&
2367             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2368                 /*
2369                  * Make it a sync io if it's an unaligned aio.
2370                  */
2371                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2372         }
2373
2374         /* communicate with ocfs2_dio_end_io */
2375         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2376
2377         written = __generic_file_write_iter(iocb, from);
2378         /* buffered aio wouldn't have proper lock coverage today */
2379         BUG_ON(written == -EIOCBQUEUED && !direct_io);
2380
2381         /*
2382          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2383          * function pointer which is called when o_direct io completes so that
2384          * it can unlock our rw lock.
2385          * Unfortunately there are error cases which call end_io and others
2386          * that don't.  so we don't have to unlock the rw_lock if either an
2387          * async dio is going to do it in the future or an end_io after an
2388          * error has already done it.
2389          */
2390         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2391                 rw_level = -1;
2392         }
2393
2394         if (unlikely(written <= 0))
2395                 goto out;
2396
2397         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2398             IS_SYNC(inode)) {
2399                 ret = filemap_fdatawrite_range(file->f_mapping,
2400                                                iocb->ki_pos - written,
2401                                                iocb->ki_pos - 1);
2402                 if (ret < 0)
2403                         written = ret;
2404
2405                 if (!ret) {
2406                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2407                         if (ret < 0)
2408                                 written = ret;
2409                 }
2410
2411                 if (!ret)
2412                         ret = filemap_fdatawait_range(file->f_mapping,
2413                                                       iocb->ki_pos - written,
2414                                                       iocb->ki_pos - 1);
2415         }
2416
2417 out:
2418         if (saved_ki_complete)
2419                 xchg(&iocb->ki_complete, saved_ki_complete);
2420
2421         if (rw_level != -1)
2422                 ocfs2_rw_unlock(inode, rw_level);
2423
2424 out_mutex:
2425         inode_unlock(inode);
2426
2427         if (written)
2428                 ret = written;
2429         return ret;
2430 }
2431
2432 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2433                                    struct iov_iter *to)
2434 {
2435         int ret = 0, rw_level = -1, lock_level = 0;
2436         struct file *filp = iocb->ki_filp;
2437         struct inode *inode = file_inode(filp);
2438         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2439         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2440
2441         trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2442                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2443                         filp->f_path.dentry->d_name.len,
2444                         filp->f_path.dentry->d_name.name,
2445                         to->nr_segs);   /* GRRRRR */
2446
2447
2448         if (!inode) {
2449                 ret = -EINVAL;
2450                 mlog_errno(ret);
2451                 goto bail;
2452         }
2453
2454         if (!direct_io && nowait)
2455                 return -EOPNOTSUPP;
2456
2457         /*
2458          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2459          * need locks to protect pending reads from racing with truncate.
2460          */
2461         if (direct_io) {
2462                 if (nowait)
2463                         ret = ocfs2_try_rw_lock(inode, 0);
2464                 else
2465                         ret = ocfs2_rw_lock(inode, 0);
2466
2467                 if (ret < 0) {
2468                         if (ret != -EAGAIN)
2469                                 mlog_errno(ret);
2470                         goto bail;
2471                 }
2472                 rw_level = 0;
2473                 /* communicate with ocfs2_dio_end_io */
2474                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2475         }
2476
2477         /*
2478          * We're fine letting folks race truncates and extending
2479          * writes with read across the cluster, just like they can
2480          * locally. Hence no rw_lock during read.
2481          *
2482          * Take and drop the meta data lock to update inode fields
2483          * like i_size. This allows the checks down below
2484          * generic_file_read_iter() a chance of actually working.
2485          */
2486         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2487                                      !nowait);
2488         if (ret < 0) {
2489                 if (ret != -EAGAIN)
2490                         mlog_errno(ret);
2491                 goto bail;
2492         }
2493         ocfs2_inode_unlock(inode, lock_level);
2494
2495         ret = generic_file_read_iter(iocb, to);
2496         trace_generic_file_read_iter_ret(ret);
2497
2498         /* buffered aio wouldn't have proper lock coverage today */
2499         BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2500
2501         /* see ocfs2_file_write_iter */
2502         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2503                 rw_level = -1;
2504         }
2505
2506 bail:
2507         if (rw_level != -1)
2508                 ocfs2_rw_unlock(inode, rw_level);
2509
2510         return ret;
2511 }
2512
2513 /* Refer generic_file_llseek_unlocked() */
2514 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2515 {
2516         struct inode *inode = file->f_mapping->host;
2517         int ret = 0;
2518
2519         inode_lock(inode);
2520
2521         switch (whence) {
2522         case SEEK_SET:
2523                 break;
2524         case SEEK_END:
2525                 /* SEEK_END requires the OCFS2 inode lock for the file
2526                  * because it references the file's size.
2527                  */
2528                 ret = ocfs2_inode_lock(inode, NULL, 0);
2529                 if (ret < 0) {
2530                         mlog_errno(ret);
2531                         goto out;
2532                 }
2533                 offset += i_size_read(inode);
2534                 ocfs2_inode_unlock(inode, 0);
2535                 break;
2536         case SEEK_CUR:
2537                 if (offset == 0) {
2538                         offset = file->f_pos;
2539                         goto out;
2540                 }
2541                 offset += file->f_pos;
2542                 break;
2543         case SEEK_DATA:
2544         case SEEK_HOLE:
2545                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2546                 if (ret)
2547                         goto out;
2548                 break;
2549         default:
2550                 ret = -EINVAL;
2551                 goto out;
2552         }
2553
2554         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2555
2556 out:
2557         inode_unlock(inode);
2558         if (ret)
2559                 return ret;
2560         return offset;
2561 }
2562
2563 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2564                                      struct file *file_out, loff_t pos_out,
2565                                      loff_t len, unsigned int remap_flags)
2566 {
2567         struct inode *inode_in = file_inode(file_in);
2568         struct inode *inode_out = file_inode(file_out);
2569         struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2570         struct buffer_head *in_bh = NULL, *out_bh = NULL;
2571         bool same_inode = (inode_in == inode_out);
2572         loff_t remapped = 0;
2573         ssize_t ret;
2574
2575         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2576                 return -EINVAL;
2577         if (!ocfs2_refcount_tree(osb))
2578                 return -EOPNOTSUPP;
2579         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2580                 return -EROFS;
2581
2582         /* Lock both files against IO */
2583         ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2584         if (ret)
2585                 return ret;
2586
2587         /* Check file eligibility and prepare for block sharing. */
2588         ret = -EINVAL;
2589         if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2590             (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2591                 goto out_unlock;
2592
2593         ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2594                         &len, remap_flags);
2595         if (ret < 0 || len == 0)
2596                 goto out_unlock;
2597
2598         /* Lock out changes to the allocation maps and remap. */
2599         down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2600         if (!same_inode)
2601                 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2602                                   SINGLE_DEPTH_NESTING);
2603
2604         /* Zap any page cache for the destination file's range. */
2605         truncate_inode_pages_range(&inode_out->i_data,
2606                                    round_down(pos_out, PAGE_SIZE),
2607                                    round_up(pos_out + len, PAGE_SIZE) - 1);
2608
2609         remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2610                         inode_out, out_bh, pos_out, len);
2611         up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2612         if (!same_inode)
2613                 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2614         if (remapped < 0) {
2615                 ret = remapped;
2616                 mlog_errno(ret);
2617                 goto out_unlock;
2618         }
2619
2620         /*
2621          * Empty the extent map so that we may get the right extent
2622          * record from the disk.
2623          */
2624         ocfs2_extent_map_trunc(inode_in, 0);
2625         ocfs2_extent_map_trunc(inode_out, 0);
2626
2627         ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2628         if (ret) {
2629                 mlog_errno(ret);
2630                 goto out_unlock;
2631         }
2632
2633 out_unlock:
2634         ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2635         return remapped > 0 ? remapped : ret;
2636 }
2637
2638 const struct inode_operations ocfs2_file_iops = {
2639         .setattr        = ocfs2_setattr,
2640         .getattr        = ocfs2_getattr,
2641         .permission     = ocfs2_permission,
2642         .listxattr      = ocfs2_listxattr,
2643         .fiemap         = ocfs2_fiemap,
2644         .get_acl        = ocfs2_iop_get_acl,
2645         .set_acl        = ocfs2_iop_set_acl,
2646 };
2647
2648 const struct inode_operations ocfs2_special_file_iops = {
2649         .setattr        = ocfs2_setattr,
2650         .getattr        = ocfs2_getattr,
2651         .permission     = ocfs2_permission,
2652         .get_acl        = ocfs2_iop_get_acl,
2653         .set_acl        = ocfs2_iop_set_acl,
2654 };
2655
2656 /*
2657  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2658  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2659  */
2660 const struct file_operations ocfs2_fops = {
2661         .llseek         = ocfs2_file_llseek,
2662         .mmap           = ocfs2_mmap,
2663         .fsync          = ocfs2_sync_file,
2664         .release        = ocfs2_file_release,
2665         .open           = ocfs2_file_open,
2666         .read_iter      = ocfs2_file_read_iter,
2667         .write_iter     = ocfs2_file_write_iter,
2668         .unlocked_ioctl = ocfs2_ioctl,
2669 #ifdef CONFIG_COMPAT
2670         .compat_ioctl   = ocfs2_compat_ioctl,
2671 #endif
2672         .lock           = ocfs2_lock,
2673         .flock          = ocfs2_flock,
2674         .splice_read    = generic_file_splice_read,
2675         .splice_write   = iter_file_splice_write,
2676         .fallocate      = ocfs2_fallocate,
2677         .remap_file_range = ocfs2_remap_file_range,
2678 };
2679
2680 const struct file_operations ocfs2_dops = {
2681         .llseek         = generic_file_llseek,
2682         .read           = generic_read_dir,
2683         .iterate        = ocfs2_readdir,
2684         .fsync          = ocfs2_sync_file,
2685         .release        = ocfs2_dir_release,
2686         .open           = ocfs2_dir_open,
2687         .unlocked_ioctl = ocfs2_ioctl,
2688 #ifdef CONFIG_COMPAT
2689         .compat_ioctl   = ocfs2_compat_ioctl,
2690 #endif
2691         .lock           = ocfs2_lock,
2692         .flock          = ocfs2_flock,
2693 };
2694
2695 /*
2696  * POSIX-lockless variants of our file_operations.
2697  *
2698  * These will be used if the underlying cluster stack does not support
2699  * posix file locking, if the user passes the "localflocks" mount
2700  * option, or if we have a local-only fs.
2701  *
2702  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2703  * so we still want it in the case of no stack support for
2704  * plocks. Internally, it will do the right thing when asked to ignore
2705  * the cluster.
2706  */
2707 const struct file_operations ocfs2_fops_no_plocks = {
2708         .llseek         = ocfs2_file_llseek,
2709         .mmap           = ocfs2_mmap,
2710         .fsync          = ocfs2_sync_file,
2711         .release        = ocfs2_file_release,
2712         .open           = ocfs2_file_open,
2713         .read_iter      = ocfs2_file_read_iter,
2714         .write_iter     = ocfs2_file_write_iter,
2715         .unlocked_ioctl = ocfs2_ioctl,
2716 #ifdef CONFIG_COMPAT
2717         .compat_ioctl   = ocfs2_compat_ioctl,
2718 #endif
2719         .flock          = ocfs2_flock,
2720         .splice_read    = generic_file_splice_read,
2721         .splice_write   = iter_file_splice_write,
2722         .fallocate      = ocfs2_fallocate,
2723         .remap_file_range = ocfs2_remap_file_range,
2724 };
2725
2726 const struct file_operations ocfs2_dops_no_plocks = {
2727         .llseek         = generic_file_llseek,
2728         .read           = generic_read_dir,
2729         .iterate        = ocfs2_readdir,
2730         .fsync          = ocfs2_sync_file,
2731         .release        = ocfs2_dir_release,
2732         .open           = ocfs2_dir_open,
2733         .unlocked_ioctl = ocfs2_ioctl,
2734 #ifdef CONFIG_COMPAT
2735         .compat_ioctl   = ocfs2_compat_ioctl,
2736 #endif
2737         .flock          = ocfs2_flock,
2738 };