fs/ocfs2/file.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * file.c
   5  *
   6  * File open, close, extend, truncate
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/capability.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/uio.h>
  33 #include <linux/sched.h>
  34 #include <linux/splice.h>
  35 #include <linux/mount.h>
  36 #include <linux/writeback.h>
  37 #include <linux/falloc.h>
  38 #include <linux/quotaops.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/backing-dev.h>
  41
  42 #include <cluster/masklog.h>
  43
  44 #include "ocfs2.h"
  45
  46 #include "alloc.h"
  47 #include "aops.h"
  48 #include "dir.h"
  49 #include "dlmglue.h"
  50 #include "extent_map.h"
  51 #include "file.h"
  52 #include "sysfile.h"
  53 #include "inode.h"
  54 #include "ioctl.h"
  55 #include "journal.h"
  56 #include "locks.h"
  57 #include "mmap.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "xattr.h"
  61 #include "acl.h"
  62 #include "quota.h"
  63 #include "refcounttree.h"
  64 #include "ocfs2_trace.h"
  65
  66 #include "buffer_head_io.h"
  67
  68 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  69 {
  70         struct ocfs2_file_private *fp;
  71
  72         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  73         if (!fp)
  74                 return -ENOMEM;
  75
  76         fp->fp_file = file;
  77         mutex_init(&fp->fp_mutex);
  78         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  79         file->private_data = fp;
  80
  81         return 0;
  82 }
  83
  84 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  85 {
  86         struct ocfs2_file_private *fp = file->private_data;
  87         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  88
  89         if (fp) {
  90                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  91                 ocfs2_lock_res_free(&fp->fp_flock);
  92                 kfree(fp);
  93                 file->private_data = NULL;
  94         }
  95 }
  96
  97 static int ocfs2_file_open(struct inode *inode, struct file *file)
  98 {
  99         int status;
 100         int mode = file->f_flags;
 101         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 102
 103         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
 104                               (unsigned long long)oi->ip_blkno,
 105                               file->f_path.dentry->d_name.len,
 106                               file->f_path.dentry->d_name.name, mode);
 107
 108         if (file->f_mode & FMODE_WRITE) {
 109                 status = dquot_initialize(inode);
 110                 if (status)
 111                         goto leave;
 112         }
 113
 114         spin_lock(&oi->ip_lock);
 115
 116         /* Check that the inode hasn't been wiped from disk by another
 117          * node. If it hasn't then we're safe as long as we hold the
 118          * spin lock until our increment of open count. */
 119         if (oi->ip_flags & OCFS2_INODE_DELETED) {
 120                 spin_unlock(&oi->ip_lock);
 121
 122                 status = -ENOENT;
 123                 goto leave;
 124         }
 125
 126         if (mode & O_DIRECT)
 127                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 128
 129         oi->ip_open_count++;
 130         spin_unlock(&oi->ip_lock);
 131
 132         status = ocfs2_init_file_private(inode, file);
 133         if (status) {
 134                 /*
 135                  * We want to set open count back if we're failing the
 136                  * open.
 137                  */
 138                 spin_lock(&oi->ip_lock);
 139                 oi->ip_open_count--;
 140                 spin_unlock(&oi->ip_lock);
 141         }
 142
 143         file->f_mode |= FMODE_NOWAIT;
 144
 145 leave:
 146         return status;
 147 }
 148
 149 static int ocfs2_file_release(struct inode *inode, struct file *file)
 150 {
 151         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 152
 153         spin_lock(&oi->ip_lock);
 154         if (!--oi->ip_open_count)
 155                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 156
 157         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 158                                  oi->ip_blkno,
 159                                  file->f_path.dentry->d_name.len,
 160                                  file->f_path.dentry->d_name.name,
 161                                  oi->ip_open_count);
 162         spin_unlock(&oi->ip_lock);
 163
 164         ocfs2_free_file_private(inode, file);
 165
 166         return 0;
 167 }
 168
 169 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 170 {
 171         return ocfs2_init_file_private(inode, file);
 172 }
 173
 174 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 175 {
 176         ocfs2_free_file_private(inode, file);
 177         return 0;
 178 }
 179
 180 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 181                            int datasync)
 182 {
 183         int err = 0;
 184         struct inode *inode = file->f_mapping->host;
 185         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 186         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 187         journal_t *journal = osb->journal->j_journal;
 188         int ret;
 189         tid_t commit_tid;
 190         bool needs_barrier = false;
 191
 192         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 193                               oi->ip_blkno,
 194                               file->f_path.dentry->d_name.len,
 195                               file->f_path.dentry->d_name.name,
 196                               (unsigned long long)datasync);
 197
 198         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 199                 return -EROFS;
 200
 201         err = file_write_and_wait_range(file, start, end);
 202         if (err)
 203                 return err;
 204
 205         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 206         if (journal->j_flags & JBD2_BARRIER &&
 207             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 208                 needs_barrier = true;
 209         err = jbd2_complete_transaction(journal, commit_tid);
 210         if (needs_barrier) {
 211                 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 212                 if (!err)
 213                         err = ret;
 214         }
 215
 216         if (err)
 217                 mlog_errno(err);
 218
 219         return (err < 0) ? -EIO : 0;
 220 }
 221
 222 int ocfs2_should_update_atime(struct inode *inode,
 223                               struct vfsmount *vfsmnt)
 224 {
 225         struct timespec64 now;
 226         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 227
 228         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 229                 return 0;
 230
 231         if ((inode->i_flags & S_NOATIME) ||
 232             ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
 233                 return 0;
 234
 235         /*
 236          * We can be called with no vfsmnt structure - NFSD will
 237          * sometimes do this.
 238          *
 239          * Note that our action here is different than touch_atime() -
 240          * if we can't tell whether this is a noatime mount, then we
 241          * don't know whether to trust the value of s_atime_quantum.
 242          */
 243         if (vfsmnt == NULL)
 244                 return 0;
 245
 246         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 247             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 248                 return 0;
 249
 250         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 251                 if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 252                     (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 253                         return 1;
 254
 255                 return 0;
 256         }
 257
 258         now = current_time(inode);
 259         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 260                 return 0;
 261         else
 262                 return 1;
 263 }
 264
 265 int ocfs2_update_inode_atime(struct inode *inode,
 266                              struct buffer_head *bh)
 267 {
 268         int ret;
 269         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 270         handle_t *handle;
 271         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 272
 273         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 274         if (IS_ERR(handle)) {
 275                 ret = PTR_ERR(handle);
 276                 mlog_errno(ret);
 277                 goto out;
 278         }
 279
 280         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 281                                       OCFS2_JOURNAL_ACCESS_WRITE);
 282         if (ret) {
 283                 mlog_errno(ret);
 284                 goto out_commit;
 285         }
 286
 287         /*
 288          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 289          * have i_mutex to guard against concurrent changes to other
 290          * inode fields.
 291          */
 292         inode->i_atime = current_time(inode);
 293         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 294         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 295         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 296         ocfs2_journal_dirty(handle, bh);
 297
 298 out_commit:
 299         ocfs2_commit_trans(osb, handle);
 300 out:
 301         return ret;
 302 }
 303
 304 int ocfs2_set_inode_size(handle_t *handle,
 305                                 struct inode *inode,
 306                                 struct buffer_head *fe_bh,
 307                                 u64 new_i_size)
 308 {
 309         int status;
 310
 311         i_size_write(inode, new_i_size);
 312         inode->i_blocks = ocfs2_inode_sector_count(inode);
 313         inode->i_ctime = inode->i_mtime = current_time(inode);
 314
 315         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 316         if (status < 0) {
 317                 mlog_errno(status);
 318                 goto bail;
 319         }
 320
 321 bail:
 322         return status;
 323 }
 324
 325 int ocfs2_simple_size_update(struct inode *inode,
 326                              struct buffer_head *di_bh,
 327                              u64 new_i_size)
 328 {
 329         int ret;
 330         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 331         handle_t *handle = NULL;
 332
 333         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 334         if (IS_ERR(handle)) {
 335                 ret = PTR_ERR(handle);
 336                 mlog_errno(ret);
 337                 goto out;
 338         }
 339
 340         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 341                                    new_i_size);
 342         if (ret < 0)
 343                 mlog_errno(ret);
 344
 345         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 346         ocfs2_commit_trans(osb, handle);
 347 out:
 348         return ret;
 349 }
 350
 351 static int ocfs2_cow_file_pos(struct inode *inode,
 352                               struct buffer_head *fe_bh,
 353                               u64 offset)
 354 {
 355         int status;
 356         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 357         unsigned int num_clusters = 0;
 358         unsigned int ext_flags = 0;
 359
 360         /*
 361          * If the new offset is aligned to the range of the cluster, there is
 362          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 363          * CoW either.
 364          */
 365         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 366                 return 0;
 367
 368         status = ocfs2_get_clusters(inode, cpos, &phys,
 369                                     &num_clusters, &ext_flags);
 370         if (status) {
 371                 mlog_errno(status);
 372                 goto out;
 373         }
 374
 375         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 376                 goto out;
 377
 378         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 379
 380 out:
 381         return status;
 382 }
 383
 384 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 385                                      struct inode *inode,
 386                                      struct buffer_head *fe_bh,
 387                                      u64 new_i_size)
 388 {
 389         int status;
 390         handle_t *handle;
 391         struct ocfs2_dinode *di;
 392         u64 cluster_bytes;
 393
 394         /*
 395          * We need to CoW the cluster contains the offset if it is reflinked
 396          * since we will call ocfs2_zero_range_for_truncate later which will
 397          * write "0" from offset to the end of the cluster.
 398          */
 399         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 400         if (status) {
 401                 mlog_errno(status);
 402                 return status;
 403         }
 404
 405         /* TODO: This needs to actually orphan the inode in this
 406          * transaction. */
 407
 408         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 409         if (IS_ERR(handle)) {
 410                 status = PTR_ERR(handle);
 411                 mlog_errno(status);
 412                 goto out;
 413         }
 414
 415         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 416                                          OCFS2_JOURNAL_ACCESS_WRITE);
 417         if (status < 0) {
 418                 mlog_errno(status);
 419                 goto out_commit;
 420         }
 421
 422         /*
 423          * Do this before setting i_size.
 424          */
 425         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 426         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 427                                                cluster_bytes);
 428         if (status) {
 429                 mlog_errno(status);
 430                 goto out_commit;
 431         }
 432
 433         i_size_write(inode, new_i_size);
 434         inode->i_ctime = inode->i_mtime = current_time(inode);
 435
 436         di = (struct ocfs2_dinode *) fe_bh->b_data;
 437         di->i_size = cpu_to_le64(new_i_size);
 438         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 439         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 440         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 441
 442         ocfs2_journal_dirty(handle, fe_bh);
 443
 444 out_commit:
 445         ocfs2_commit_trans(osb, handle);
 446 out:
 447         return status;
 448 }
 449
 450 int ocfs2_truncate_file(struct inode *inode,
 451                                struct buffer_head *di_bh,
 452                                u64 new_i_size)
 453 {
 454         int status = 0;
 455         struct ocfs2_dinode *fe = NULL;
 456         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 457
 458         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 459          * already validated it */
 460         fe = (struct ocfs2_dinode *) di_bh->b_data;
 461
 462         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 463                                   (unsigned long long)le64_to_cpu(fe->i_size),
 464                                   (unsigned long long)new_i_size);
 465
 466         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 467                         "Inode %llu, inode i_size = %lld != di "
 468                         "i_size = %llu, i_flags = 0x%x\n",
 469                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 470                         i_size_read(inode),
 471                         (unsigned long long)le64_to_cpu(fe->i_size),
 472                         le32_to_cpu(fe->i_flags));
 473
 474         if (new_i_size > le64_to_cpu(fe->i_size)) {
 475                 trace_ocfs2_truncate_file_error(
 476                         (unsigned long long)le64_to_cpu(fe->i_size),
 477                         (unsigned long long)new_i_size);
 478                 status = -EINVAL;
 479                 mlog_errno(status);
 480                 goto bail;
 481         }
 482
 483         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 484
 485         ocfs2_resv_discard(&osb->osb_la_resmap,
 486                            &OCFS2_I(inode)->ip_la_data_resv);
 487
 488         /*
 489          * The inode lock forced other nodes to sync and drop their
 490          * pages, which (correctly) happens even if we have a truncate
 491          * without allocation change - ocfs2 cluster sizes can be much
 492          * greater than page size, so we have to truncate them
 493          * anyway.
 494          */
 495         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 496         truncate_inode_pages(inode->i_mapping, new_i_size);
 497
 498         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 499                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 500                                                i_size_read(inode), 1);
 501                 if (status)
 502                         mlog_errno(status);
 503
 504                 goto bail_unlock_sem;
 505         }
 506
 507         /* alright, we're going to need to do a full blown alloc size
 508          * change. Orphan the inode so that recovery can complete the
 509          * truncate if necessary. This does the task of marking
 510          * i_size. */
 511         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 512         if (status < 0) {
 513                 mlog_errno(status);
 514                 goto bail_unlock_sem;
 515         }
 516
 517         status = ocfs2_commit_truncate(osb, inode, di_bh);
 518         if (status < 0) {
 519                 mlog_errno(status);
 520                 goto bail_unlock_sem;
 521         }
 522
 523         /* TODO: orphan dir cleanup here. */
 524 bail_unlock_sem:
 525         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 526
 527 bail:
 528         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 529                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 530
 531         return status;
 532 }
 533
 534 /*
 535  * extend file allocation only here.
 536  * we'll update all the disk stuff, and oip->alloc_size
 537  *
 538  * expect stuff to be locked, a transaction started and enough data /
 539  * metadata reservations in the contexts.
 540  *
 541  * Will return -EAGAIN, and a reason if a restart is needed.
 542  * If passed in, *reason will always be set, even in error.
 543  */
 544 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 545                          struct inode *inode,
 546                          u32 *logical_offset,
 547                          u32 clusters_to_add,
 548                          int mark_unwritten,
 549                          struct buffer_head *fe_bh,
 550                          handle_t *handle,
 551                          struct ocfs2_alloc_context *data_ac,
 552                          struct ocfs2_alloc_context *meta_ac,
 553                          enum ocfs2_alloc_restarted *reason_ret)
 554 {
 555         int ret;
 556         struct ocfs2_extent_tree et;
 557
 558         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 559         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 560                                           clusters_to_add, mark_unwritten,
 561                                           data_ac, meta_ac, reason_ret);
 562
 563         return ret;
 564 }
 565
 566 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 567                                    u32 clusters_to_add, int mark_unwritten)
 568 {
 569         int status = 0;
 570         int restart_func = 0;
 571         int credits;
 572         u32 prev_clusters;
 573         struct buffer_head *bh = NULL;
 574         struct ocfs2_dinode *fe = NULL;
 575         handle_t *handle = NULL;
 576         struct ocfs2_alloc_context *data_ac = NULL;
 577         struct ocfs2_alloc_context *meta_ac = NULL;
 578         enum ocfs2_alloc_restarted why = RESTART_NONE;
 579         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 580         struct ocfs2_extent_tree et;
 581         int did_quota = 0;
 582
 583         /*
 584          * Unwritten extent only exists for file systems which
 585          * support holes.
 586          */
 587         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 588
 589         status = ocfs2_read_inode_block(inode, &bh);
 590         if (status < 0) {
 591                 mlog_errno(status);
 592                 goto leave;
 593         }
 594         fe = (struct ocfs2_dinode *) bh->b_data;
 595
 596 restart_all:
 597         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 598
 599         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 600         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 601                                        &data_ac, &meta_ac);
 602         if (status) {
 603                 mlog_errno(status);
 604                 goto leave;
 605         }
 606
 607         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 608         handle = ocfs2_start_trans(osb, credits);
 609         if (IS_ERR(handle)) {
 610                 status = PTR_ERR(handle);
 611                 handle = NULL;
 612                 mlog_errno(status);
 613                 goto leave;
 614         }
 615
 616 restarted_transaction:
 617         trace_ocfs2_extend_allocation(
 618                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 619                 (unsigned long long)i_size_read(inode),
 620                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 621                 why, restart_func);
 622
 623         status = dquot_alloc_space_nodirty(inode,
 624                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 625         if (status)
 626                 goto leave;
 627         did_quota = 1;
 628
 629         /* reserve a write to the file entry early on - that we if we
 630          * run out of credits in the allocation path, we can still
 631          * update i_size. */
 632         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 633                                          OCFS2_JOURNAL_ACCESS_WRITE);
 634         if (status < 0) {
 635                 mlog_errno(status);
 636                 goto leave;
 637         }
 638
 639         prev_clusters = OCFS2_I(inode)->ip_clusters;
 640
 641         status = ocfs2_add_inode_data(osb,
 642                                       inode,
 643                                       &logical_start,
 644                                       clusters_to_add,
 645                                       mark_unwritten,
 646                                       bh,
 647                                       handle,
 648                                       data_ac,
 649                                       meta_ac,
 650                                       &why);
 651         if ((status < 0) && (status != -EAGAIN)) {
 652                 if (status != -ENOSPC)
 653                         mlog_errno(status);
 654                 goto leave;
 655         }
 656         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 657         ocfs2_journal_dirty(handle, bh);
 658
 659         spin_lock(&OCFS2_I(inode)->ip_lock);
 660         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 661         spin_unlock(&OCFS2_I(inode)->ip_lock);
 662         /* Release unused quota reservation */
 663         dquot_free_space(inode,
 664                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 665         did_quota = 0;
 666
 667         if (why != RESTART_NONE && clusters_to_add) {
 668                 if (why == RESTART_META) {
 669                         restart_func = 1;
 670                         status = 0;
 671                 } else {
 672                         BUG_ON(why != RESTART_TRANS);
 673
 674                         status = ocfs2_allocate_extend_trans(handle, 1);
 675                         if (status < 0) {
 676                                 /* handle still has to be committed at
 677                                  * this point. */
 678                                 status = -ENOMEM;
 679                                 mlog_errno(status);
 680                                 goto leave;
 681                         }
 682                         goto restarted_transaction;
 683                 }
 684         }
 685
 686         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 687              le32_to_cpu(fe->i_clusters),
 688              (unsigned long long)le64_to_cpu(fe->i_size),
 689              OCFS2_I(inode)->ip_clusters,
 690              (unsigned long long)i_size_read(inode));
 691
 692 leave:
 693         if (status < 0 && did_quota)
 694                 dquot_free_space(inode,
 695                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 696         if (handle) {
 697                 ocfs2_commit_trans(osb, handle);
 698                 handle = NULL;
 699         }
 700         if (data_ac) {
 701                 ocfs2_free_alloc_context(data_ac);
 702                 data_ac = NULL;
 703         }
 704         if (meta_ac) {
 705                 ocfs2_free_alloc_context(meta_ac);
 706                 meta_ac = NULL;
 707         }
 708         if ((!status) && restart_func) {
 709                 restart_func = 0;
 710                 goto restart_all;
 711         }
 712         brelse(bh);
 713         bh = NULL;
 714
 715         return status;
 716 }
 717
 718 /*
 719  * While a write will already be ordering the data, a truncate will not.
 720  * Thus, we need to explicitly order the zeroed pages.
 721  */
 722 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 723                                                 struct buffer_head *di_bh)
 724 {
 725         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 726         handle_t *handle = NULL;
 727         int ret = 0;
 728
 729         if (!ocfs2_should_order_data(inode))
 730                 goto out;
 731
 732         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 733         if (IS_ERR(handle)) {
 734                 ret = -ENOMEM;
 735                 mlog_errno(ret);
 736                 goto out;
 737         }
 738
 739         ret = ocfs2_jbd2_file_inode(handle, inode);
 740         if (ret < 0) {
 741                 mlog_errno(ret);
 742                 goto out;
 743         }
 744
 745         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 746                                       OCFS2_JOURNAL_ACCESS_WRITE);
 747         if (ret)
 748                 mlog_errno(ret);
 749         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 750
 751 out:
 752         if (ret) {
 753                 if (!IS_ERR(handle))
 754                         ocfs2_commit_trans(osb, handle);
 755                 handle = ERR_PTR(ret);
 756         }
 757         return handle;
 758 }
 759
 760 /* Some parts of this taken from generic_cont_expand, which turned out
 761  * to be too fragile to do exactly what we need without us having to
 762  * worry about recursive locking in ->write_begin() and ->write_end(). */
 763 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 764                                  u64 abs_to, struct buffer_head *di_bh)
 765 {
 766         struct address_space *mapping = inode->i_mapping;
 767         struct page *page;
 768         unsigned long index = abs_from >> PAGE_SHIFT;
 769         handle_t *handle;
 770         int ret = 0;
 771         unsigned zero_from, zero_to, block_start, block_end;
 772         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 773
 774         BUG_ON(abs_from >= abs_to);
 775         BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 776         BUG_ON(abs_from & (inode->i_blkbits - 1));
 777
 778         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
 779         if (IS_ERR(handle)) {
 780                 ret = PTR_ERR(handle);
 781                 goto out;
 782         }
 783
 784         page = find_or_create_page(mapping, index, GFP_NOFS);
 785         if (!page) {
 786                 ret = -ENOMEM;
 787                 mlog_errno(ret);
 788                 goto out_commit_trans;
 789         }
 790
 791         /* Get the offsets within the page that we want to zero */
 792         zero_from = abs_from & (PAGE_SIZE - 1);
 793         zero_to = abs_to & (PAGE_SIZE - 1);
 794         if (!zero_to)
 795                 zero_to = PAGE_SIZE;
 796
 797         trace_ocfs2_write_zero_page(
 798                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 799                         (unsigned long long)abs_from,
 800                         (unsigned long long)abs_to,
 801                         index, zero_from, zero_to);
 802
 803         /* We know that zero_from is block aligned */
 804         for (block_start = zero_from; block_start < zero_to;
 805              block_start = block_end) {
 806                 block_end = block_start + i_blocksize(inode);
 807
 808                 /*
 809                  * block_start is block-aligned.  Bump it by one to force
 810                  * __block_write_begin and block_commit_write to zero the
 811                  * whole block.
 812                  */
 813                 ret = __block_write_begin(page, block_start + 1, 0,
 814                                           ocfs2_get_block);
 815                 if (ret < 0) {
 816                         mlog_errno(ret);
 817                         goto out_unlock;
 818                 }
 819
 820
 821                 /* must not update i_size! */
 822                 ret = block_commit_write(page, block_start + 1,
 823                                          block_start + 1);
 824                 if (ret < 0)
 825                         mlog_errno(ret);
 826                 else
 827                         ret = 0;
 828         }
 829
 830         /*
 831          * fs-writeback will release the dirty pages without page lock
 832          * whose offset are over inode size, the release happens at
 833          * block_write_full_page().
 834          */
 835         i_size_write(inode, abs_to);
 836         inode->i_blocks = ocfs2_inode_sector_count(inode);
 837         di->i_size = cpu_to_le64((u64)i_size_read(inode));
 838         inode->i_mtime = inode->i_ctime = current_time(inode);
 839         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 840         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 841         di->i_mtime_nsec = di->i_ctime_nsec;
 842         if (handle) {
 843                 ocfs2_journal_dirty(handle, di_bh);
 844                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 845         }
 846
 847 out_unlock:
 848         unlock_page(page);
 849         put_page(page);
 850 out_commit_trans:
 851         if (handle)
 852                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 853 out:
 854         return ret;
 855 }
 856
 857 /*
 858  * Find the next range to zero.  We do this in terms of bytes because
 859  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 860  * pagecache.  We may return multiple extents.
 861  *
 862  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 863  * needs to be zeroed.  range_start and range_end return the next zeroing
 864  * range.  A subsequent call should pass the previous range_end as its
 865  * zero_start.  If range_end is 0, there's nothing to do.
 866  *
 867  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 868  */
 869 static int ocfs2_zero_extend_get_range(struct inode *inode,
 870                                        struct buffer_head *di_bh,
 871                                        u64 zero_start, u64 zero_end,
 872                                        u64 *range_start, u64 *range_end)
 873 {
 874         int rc = 0, needs_cow = 0;
 875         u32 p_cpos, zero_clusters = 0;
 876         u32 zero_cpos =
 877                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 878         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 879         unsigned int num_clusters = 0;
 880         unsigned int ext_flags = 0;
 881
 882         while (zero_cpos < last_cpos) {
 883                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 884                                         &num_clusters, &ext_flags);
 885                 if (rc) {
 886                         mlog_errno(rc);
 887                         goto out;
 888                 }
 889
 890                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 891                         zero_clusters = num_clusters;
 892                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 893                                 needs_cow = 1;
 894                         break;
 895                 }
 896
 897                 zero_cpos += num_clusters;
 898         }
 899         if (!zero_clusters) {
 900                 *range_end = 0;
 901                 goto out;
 902         }
 903
 904         while ((zero_cpos + zero_clusters) < last_cpos) {
 905                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 906                                         &p_cpos, &num_clusters,
 907                                         &ext_flags);
 908                 if (rc) {
 909                         mlog_errno(rc);
 910                         goto out;
 911                 }
 912
 913                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 914                         break;
 915                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 916                         needs_cow = 1;
 917                 zero_clusters += num_clusters;
 918         }
 919         if ((zero_cpos + zero_clusters) > last_cpos)
 920                 zero_clusters = last_cpos - zero_cpos;
 921
 922         if (needs_cow) {
 923                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 924                                         zero_clusters, UINT_MAX);
 925                 if (rc) {
 926                         mlog_errno(rc);
 927                         goto out;
 928                 }
 929         }
 930
 931         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 932         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 933                                              zero_cpos + zero_clusters);
 934
 935 out:
 936         return rc;
 937 }
 938
 939 /*
 940  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 941  * has made sure that the entire range needs zeroing.
 942  */
 943 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 944                                    u64 range_end, struct buffer_head *di_bh)
 945 {
 946         int rc = 0;
 947         u64 next_pos;
 948         u64 zero_pos = range_start;
 949
 950         trace_ocfs2_zero_extend_range(
 951                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 952                         (unsigned long long)range_start,
 953                         (unsigned long long)range_end);
 954         BUG_ON(range_start >= range_end);
 955
 956         while (zero_pos < range_end) {
 957                 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 958                 if (next_pos > range_end)
 959                         next_pos = range_end;
 960                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 961                 if (rc < 0) {
 962                         mlog_errno(rc);
 963                         break;
 964                 }
 965                 zero_pos = next_pos;
 966
 967                 /*
 968                  * Very large extends have the potential to lock up
 969                  * the cpu for extended periods of time.
 970                  */
 971                 cond_resched();
 972         }
 973
 974         return rc;
 975 }
 976
 977 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 978                       loff_t zero_to_size)
 979 {
 980         int ret = 0;
 981         u64 zero_start, range_start = 0, range_end = 0;
 982         struct super_block *sb = inode->i_sb;
 983
 984         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 985         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 986                                 (unsigned long long)zero_start,
 987                                 (unsigned long long)i_size_read(inode));
 988         while (zero_start < zero_to_size) {
 989                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 990                                                   zero_to_size,
 991                                                   &range_start,
 992                                                   &range_end);
 993                 if (ret) {
 994                         mlog_errno(ret);
 995                         break;
 996                 }
 997                 if (!range_end)
 998                         break;
 999                 /* Trim the ends */
1000                 if (range_start < zero_start)
1001                         range_start = zero_start;
1002                 if (range_end > zero_to_size)
1003                         range_end = zero_to_size;
1004
1005                 ret = ocfs2_zero_extend_range(inode, range_start,
1006                                               range_end, di_bh);
1007                 if (ret) {
1008                         mlog_errno(ret);
1009                         break;
1010                 }
1011                 zero_start = range_end;
1012         }
1013
1014         return ret;
1015 }
1016
1017 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1018                           u64 new_i_size, u64 zero_to)
1019 {
1020         int ret;
1021         u32 clusters_to_add;
1022         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1023
1024         /*
1025          * Only quota files call this without a bh, and they can't be
1026          * refcounted.
1027          */
1028         BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1029         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1030
1031         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1032         if (clusters_to_add < oi->ip_clusters)
1033                 clusters_to_add = 0;
1034         else
1035                 clusters_to_add -= oi->ip_clusters;
1036
1037         if (clusters_to_add) {
1038                 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1039                                               clusters_to_add, 0);
1040                 if (ret) {
1041                         mlog_errno(ret);
1042                         goto out;
1043                 }
1044         }
1045
1046         /*
1047          * Call this even if we don't add any clusters to the tree. We
1048          * still need to zero the area between the old i_size and the
1049          * new i_size.
1050          */
1051         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1052         if (ret < 0)
1053                 mlog_errno(ret);
1054
1055 out:
1056         return ret;
1057 }
1058
1059 static int ocfs2_extend_file(struct inode *inode,
1060                              struct buffer_head *di_bh,
1061                              u64 new_i_size)
1062 {
1063         int ret = 0;
1064         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1065
1066         BUG_ON(!di_bh);
1067
1068         /* setattr sometimes calls us like this. */
1069         if (new_i_size == 0)
1070                 goto out;
1071
1072         if (i_size_read(inode) == new_i_size)
1073                 goto out;
1074         BUG_ON(new_i_size < i_size_read(inode));
1075
1076         /*
1077          * The alloc sem blocks people in read/write from reading our
1078          * allocation until we're done changing it. We depend on
1079          * i_mutex to block other extend/truncate calls while we're
1080          * here.  We even have to hold it for sparse files because there
1081          * might be some tail zeroing.
1082          */
1083         down_write(&oi->ip_alloc_sem);
1084
1085         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1086                 /*
1087                  * We can optimize small extends by keeping the inodes
1088                  * inline data.
1089                  */
1090                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1091                         up_write(&oi->ip_alloc_sem);
1092                         goto out_update_size;
1093                 }
1094
1095                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1096                 if (ret) {
1097                         up_write(&oi->ip_alloc_sem);
1098                         mlog_errno(ret);
1099                         goto out;
1100                 }
1101         }
1102
1103         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1104                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1105         else
1106                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1107                                             new_i_size);
1108
1109         up_write(&oi->ip_alloc_sem);
1110
1111         if (ret < 0) {
1112                 mlog_errno(ret);
1113                 goto out;
1114         }
1115
1116 out_update_size:
1117         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1118         if (ret < 0)
1119                 mlog_errno(ret);
1120
1121 out:
1122         return ret;
1123 }
1124
1125 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1126 {
1127         int status = 0, size_change;
1128         int inode_locked = 0;
1129         struct inode *inode = d_inode(dentry);
1130         struct super_block *sb = inode->i_sb;
1131         struct ocfs2_super *osb = OCFS2_SB(sb);
1132         struct buffer_head *bh = NULL;
1133         handle_t *handle = NULL;
1134         struct dquot *transfer_to[MAXQUOTAS] = { };
1135         int qtype;
1136         int had_lock;
1137         struct ocfs2_lock_holder oh;
1138
1139         trace_ocfs2_setattr(inode, dentry,
1140                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1141                             dentry->d_name.len, dentry->d_name.name,
1142                             attr->ia_valid, attr->ia_mode,
1143                             from_kuid(&init_user_ns, attr->ia_uid),
1144                             from_kgid(&init_user_ns, attr->ia_gid));
1145
1146         /* ensuring we don't even attempt to truncate a symlink */
1147         if (S_ISLNK(inode->i_mode))
1148                 attr->ia_valid &= ~ATTR_SIZE;
1149
1150 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1151                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1152         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1153                 return 0;
1154
1155         status = setattr_prepare(dentry, attr);
1156         if (status)
1157                 return status;
1158
1159         if (is_quota_modification(inode, attr)) {
1160                 status = dquot_initialize(inode);
1161                 if (status)
1162                         return status;
1163         }
1164         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1165         if (size_change) {
1166                 /*
1167                  * Here we should wait dio to finish before inode lock
1168                  * to avoid a deadlock between ocfs2_setattr() and
1169                  * ocfs2_dio_end_io_write()
1170                  */
1171                 inode_dio_wait(inode);
1172
1173                 status = ocfs2_rw_lock(inode, 1);
1174                 if (status < 0) {
1175                         mlog_errno(status);
1176                         goto bail;
1177                 }
1178         }
1179
1180         had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1181         if (had_lock < 0) {
1182                 status = had_lock;
1183                 goto bail_unlock_rw;
1184         } else if (had_lock) {
1185                 /*
1186                  * As far as we know, ocfs2_setattr() could only be the first
1187                  * VFS entry point in the call chain of recursive cluster
1188                  * locking issue.
1189                  *
1190                  * For instance:
1191                  * chmod_common()
1192                  *  notify_change()
1193                  *   ocfs2_setattr()
1194                  *    posix_acl_chmod()
1195                  *     ocfs2_iop_get_acl()
1196                  *
1197                  * But, we're not 100% sure if it's always true, because the
1198                  * ordering of the VFS entry points in the call chain is out
1199                  * of our control. So, we'd better dump the stack here to
1200                  * catch the other cases of recursive locking.
1201                  */
1202                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1203                 dump_stack();
1204         }
1205         inode_locked = 1;
1206
1207         if (size_change) {
1208                 status = inode_newsize_ok(inode, attr->ia_size);
1209                 if (status)
1210                         goto bail_unlock;
1211
1212                 if (i_size_read(inode) >= attr->ia_size) {
1213                         if (ocfs2_should_order_data(inode)) {
1214                                 status = ocfs2_begin_ordered_truncate(inode,
1215                                                                       attr->ia_size);
1216                                 if (status)
1217                                         goto bail_unlock;
1218                         }
1219                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1220                 } else
1221                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1222                 if (status < 0) {
1223                         if (status != -ENOSPC)
1224                                 mlog_errno(status);
1225                         status = -ENOSPC;
1226                         goto bail_unlock;
1227                 }
1228         }
1229
1230         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1231             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1232                 /*
1233                  * Gather pointers to quota structures so that allocation /
1234                  * freeing of quota structures happens here and not inside
1235                  * dquot_transfer() where we have problems with lock ordering
1236                  */
1237                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1238                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1239                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1240                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1241                         if (IS_ERR(transfer_to[USRQUOTA])) {
1242                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1243                                 goto bail_unlock;
1244                         }
1245                 }
1246                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1247                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1248                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1249                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1250                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1251                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1252                                 goto bail_unlock;
1253                         }
1254                 }
1255                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1256                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1257                                            2 * ocfs2_quota_trans_credits(sb));
1258                 if (IS_ERR(handle)) {
1259                         status = PTR_ERR(handle);
1260                         mlog_errno(status);
1261                         goto bail_unlock_alloc;
1262                 }
1263                 status = __dquot_transfer(inode, transfer_to);
1264                 if (status < 0)
1265                         goto bail_commit;
1266         } else {
1267                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1268                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1269                 if (IS_ERR(handle)) {
1270                         status = PTR_ERR(handle);
1271                         mlog_errno(status);
1272                         goto bail_unlock_alloc;
1273                 }
1274         }
1275
1276         setattr_copy(inode, attr);
1277         mark_inode_dirty(inode);
1278
1279         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1280         if (status < 0)
1281                 mlog_errno(status);
1282
1283 bail_commit:
1284         ocfs2_commit_trans(osb, handle);
1285 bail_unlock_alloc:
1286         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1287 bail_unlock:
1288         if (status && inode_locked) {
1289                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1290                 inode_locked = 0;
1291         }
1292 bail_unlock_rw:
1293         if (size_change)
1294                 ocfs2_rw_unlock(inode, 1);
1295 bail:
1296
1297         /* Release quota pointers in case we acquired them */
1298         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1299                 dqput(transfer_to[qtype]);
1300
1301         if (!status && attr->ia_valid & ATTR_MODE) {
1302                 status = ocfs2_acl_chmod(inode, bh);
1303                 if (status < 0)
1304                         mlog_errno(status);
1305         }
1306         if (inode_locked)
1307                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1308
1309         brelse(bh);
1310         return status;
1311 }
1312
1313 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1314                   u32 request_mask, unsigned int flags)
1315 {
1316         struct inode *inode = d_inode(path->dentry);
1317         struct super_block *sb = path->dentry->d_sb;
1318         struct ocfs2_super *osb = sb->s_fs_info;
1319         int err;
1320
1321         err = ocfs2_inode_revalidate(path->dentry);
1322         if (err) {
1323                 if (err != -ENOENT)
1324                         mlog_errno(err);
1325                 goto bail;
1326         }
1327
1328         generic_fillattr(inode, stat);
1329         /*
1330          * If there is inline data in the inode, the inode will normally not
1331          * have data blocks allocated (it may have an external xattr block).
1332          * Report at least one sector for such files, so tools like tar, rsync,
1333          * others don't incorrectly think the file is completely sparse.
1334          */
1335         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1336                 stat->blocks += (stat->size + 511)>>9;
1337
1338         /* We set the blksize from the cluster size for performance */
1339         stat->blksize = osb->s_clustersize;
1340
1341 bail:
1342         return err;
1343 }
1344
1345 int ocfs2_permission(struct inode *inode, int mask)
1346 {
1347         int ret, had_lock;
1348         struct ocfs2_lock_holder oh;
1349
1350         if (mask & MAY_NOT_BLOCK)
1351                 return -ECHILD;
1352
1353         had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1354         if (had_lock < 0) {
1355                 ret = had_lock;
1356                 goto out;
1357         } else if (had_lock) {
1358                 /* See comments in ocfs2_setattr() for details.
1359                  * The call chain of this case could be:
1360                  * do_sys_open()
1361                  *  may_open()
1362                  *   inode_permission()
1363                  *    ocfs2_permission()
1364                  *     ocfs2_iop_get_acl()
1365                  */
1366                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1367                 dump_stack();
1368         }
1369
1370         ret = generic_permission(inode, mask);
1371
1372         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1373 out:
1374         return ret;
1375 }
1376
1377 static int __ocfs2_write_remove_suid(struct inode *inode,
1378                                      struct buffer_head *bh)
1379 {
1380         int ret;
1381         handle_t *handle;
1382         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1383         struct ocfs2_dinode *di;
1384
1385         trace_ocfs2_write_remove_suid(
1386                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1387                         inode->i_mode);
1388
1389         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1390         if (IS_ERR(handle)) {
1391                 ret = PTR_ERR(handle);
1392                 mlog_errno(ret);
1393                 goto out;
1394         }
1395
1396         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1397                                       OCFS2_JOURNAL_ACCESS_WRITE);
1398         if (ret < 0) {
1399                 mlog_errno(ret);
1400                 goto out_trans;
1401         }
1402
1403         inode->i_mode &= ~S_ISUID;
1404         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1405                 inode->i_mode &= ~S_ISGID;
1406
1407         di = (struct ocfs2_dinode *) bh->b_data;
1408         di->i_mode = cpu_to_le16(inode->i_mode);
1409         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1410
1411         ocfs2_journal_dirty(handle, bh);
1412
1413 out_trans:
1414         ocfs2_commit_trans(osb, handle);
1415 out:
1416         return ret;
1417 }
1418
1419 static int ocfs2_write_remove_suid(struct inode *inode)
1420 {
1421         int ret;
1422         struct buffer_head *bh = NULL;
1423
1424         ret = ocfs2_read_inode_block(inode, &bh);
1425         if (ret < 0) {
1426                 mlog_errno(ret);
1427                 goto out;
1428         }
1429
1430         ret =  __ocfs2_write_remove_suid(inode, bh);
1431 out:
1432         brelse(bh);
1433         return ret;
1434 }
1435
1436 /*
1437  * Allocate enough extents to cover the region starting at byte offset
1438  * start for len bytes. Existing extents are skipped, any extents
1439  * added are marked as "unwritten".
1440  */
1441 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1442                                             u64 start, u64 len)
1443 {
1444         int ret;
1445         u32 cpos, phys_cpos, clusters, alloc_size;
1446         u64 end = start + len;
1447         struct buffer_head *di_bh = NULL;
1448
1449         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1450                 ret = ocfs2_read_inode_block(inode, &di_bh);
1451                 if (ret) {
1452                         mlog_errno(ret);
1453                         goto out;
1454                 }
1455
1456                 /*
1457                  * Nothing to do if the requested reservation range
1458                  * fits within the inode.
1459                  */
1460                 if (ocfs2_size_fits_inline_data(di_bh, end))
1461                         goto out;
1462
1463                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1464                 if (ret) {
1465                         mlog_errno(ret);
1466                         goto out;
1467                 }
1468         }
1469
1470         /*
1471          * We consider both start and len to be inclusive.
1472          */
1473         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1474         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1475         clusters -= cpos;
1476
1477         while (clusters) {
1478                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1479                                          &alloc_size, NULL);
1480                 if (ret) {
1481                         mlog_errno(ret);
1482                         goto out;
1483                 }
1484
1485                 /*
1486                  * Hole or existing extent len can be arbitrary, so
1487                  * cap it to our own allocation request.
1488                  */
1489                 if (alloc_size > clusters)
1490                         alloc_size = clusters;
1491
1492                 if (phys_cpos) {
1493                         /*
1494                          * We already have an allocation at this
1495                          * region so we can safely skip it.
1496                          */
1497                         goto next;
1498                 }
1499
1500                 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1501                 if (ret) {
1502                         if (ret != -ENOSPC)
1503                                 mlog_errno(ret);
1504                         goto out;
1505                 }
1506
1507 next:
1508                 cpos += alloc_size;
1509                 clusters -= alloc_size;
1510         }
1511
1512         ret = 0;
1513 out:
1514
1515         brelse(di_bh);
1516         return ret;
1517 }
1518
1519 /*
1520  * Truncate a byte range, avoiding pages within partial clusters. This
1521  * preserves those pages for the zeroing code to write to.
1522  */
1523 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1524                                          u64 byte_len)
1525 {
1526         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1527         loff_t start, end;
1528         struct address_space *mapping = inode->i_mapping;
1529
1530         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1531         end = byte_start + byte_len;
1532         end = end & ~(osb->s_clustersize - 1);
1533
1534         if (start < end) {
1535                 unmap_mapping_range(mapping, start, end - start, 0);
1536                 truncate_inode_pages_range(mapping, start, end - 1);
1537         }
1538 }
1539
1540 /*
1541  * zero out partial blocks of one cluster.
1542  *
1543  * start: file offset where zero starts, will be made upper block aligned.
1544  * len: it will be trimmed to the end of current cluster if "start + len"
1545  *      is bigger than it.
1546  */
1547 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1548                                         u64 start, u64 len)
1549 {
1550         int ret;
1551         u64 start_block, end_block, nr_blocks;
1552         u64 p_block, offset;
1553         u32 cluster, p_cluster, nr_clusters;
1554         struct super_block *sb = inode->i_sb;
1555         u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1556
1557         if (start + len < end)
1558                 end = start + len;
1559
1560         start_block = ocfs2_blocks_for_bytes(sb, start);
1561         end_block = ocfs2_blocks_for_bytes(sb, end);
1562         nr_blocks = end_block - start_block;
1563         if (!nr_blocks)
1564                 return 0;
1565
1566         cluster = ocfs2_bytes_to_clusters(sb, start);
1567         ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1568                                 &nr_clusters, NULL);
1569         if (ret)
1570                 return ret;
1571         if (!p_cluster)
1572                 return 0;
1573
1574         offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1575         p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1576         return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1577 }
1578
1579 static int ocfs2_zero_partial_clusters(struct inode *inode,
1580                                        u64 start, u64 len)
1581 {
1582         int ret = 0;
1583         u64 tmpend = 0;
1584         u64 end = start + len;
1585         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1586         unsigned int csize = osb->s_clustersize;
1587         handle_t *handle;
1588         loff_t isize = i_size_read(inode);
1589
1590         /*
1591          * The "start" and "end" values are NOT necessarily part of
1592          * the range whose allocation is being deleted. Rather, this
1593          * is what the user passed in with the request. We must zero
1594          * partial clusters here. There's no need to worry about
1595          * physical allocation - the zeroing code knows to skip holes.
1596          */
1597         trace_ocfs2_zero_partial_clusters(
1598                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1599                 (unsigned long long)start, (unsigned long long)end);
1600
1601         /*
1602          * If both edges are on a cluster boundary then there's no
1603          * zeroing required as the region is part of the allocation to
1604          * be truncated.
1605          */
1606         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1607                 goto out;
1608
1609         /* No page cache for EOF blocks, issue zero out to disk. */
1610         if (end > isize) {
1611                 /*
1612                  * zeroout eof blocks in last cluster starting from
1613                  * "isize" even "start" > "isize" because it is
1614                  * complicated to zeroout just at "start" as "start"
1615                  * may be not aligned with block size, buffer write
1616                  * would be required to do that, but out of eof buffer
1617                  * write is not supported.
1618                  */
1619                 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1620                                         end - isize);
1621                 if (ret) {
1622                         mlog_errno(ret);
1623                         goto out;
1624                 }
1625                 if (start >= isize)
1626                         goto out;
1627                 end = isize;
1628         }
1629         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1630         if (IS_ERR(handle)) {
1631                 ret = PTR_ERR(handle);
1632                 mlog_errno(ret);
1633                 goto out;
1634         }
1635
1636         /*
1637          * If start is on a cluster boundary and end is somewhere in another
1638          * cluster, we have not COWed the cluster starting at start, unless
1639          * end is also within the same cluster. So, in this case, we skip this
1640          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1641          * to the next one.
1642          */
1643         if ((start & (csize - 1)) != 0) {
1644                 /*
1645                  * We want to get the byte offset of the end of the 1st
1646                  * cluster.
1647                  */
1648                 tmpend = (u64)osb->s_clustersize +
1649                         (start & ~(osb->s_clustersize - 1));
1650                 if (tmpend > end)
1651                         tmpend = end;
1652
1653                 trace_ocfs2_zero_partial_clusters_range1(
1654                         (unsigned long long)start,
1655                         (unsigned long long)tmpend);
1656
1657                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1658                                                     tmpend);
1659                 if (ret)
1660                         mlog_errno(ret);
1661         }
1662
1663         if (tmpend < end) {
1664                 /*
1665                  * This may make start and end equal, but the zeroing
1666                  * code will skip any work in that case so there's no
1667                  * need to catch it up here.
1668                  */
1669                 start = end & ~(osb->s_clustersize - 1);
1670
1671                 trace_ocfs2_zero_partial_clusters_range2(
1672                         (unsigned long long)start, (unsigned long long)end);
1673
1674                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1675                 if (ret)
1676                         mlog_errno(ret);
1677         }
1678         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1679
1680         ocfs2_commit_trans(osb, handle);
1681 out:
1682         return ret;
1683 }
1684
1685 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1686 {
1687         int i;
1688         struct ocfs2_extent_rec *rec = NULL;
1689
1690         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1691
1692                 rec = &el->l_recs[i];
1693
1694                 if (le32_to_cpu(rec->e_cpos) < pos)
1695                         break;
1696         }
1697
1698         return i;
1699 }
1700
1701 /*
1702  * Helper to calculate the punching pos and length in one run, we handle the
1703  * following three cases in order:
1704  *
1705  * - remove the entire record
1706  * - remove a partial record
1707  * - no record needs to be removed (hole-punching completed)
1708 */
1709 static void ocfs2_calc_trunc_pos(struct inode *inode,
1710                                  struct ocfs2_extent_list *el,
1711                                  struct ocfs2_extent_rec *rec,
1712                                  u32 trunc_start, u32 *trunc_cpos,
1713                                  u32 *trunc_len, u32 *trunc_end,
1714                                  u64 *blkno, int *done)
1715 {
1716         int ret = 0;
1717         u32 coff, range;
1718
1719         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1720
1721         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1722                 /*
1723                  * remove an entire extent record.
1724                  */
1725                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1726                 /*
1727                  * Skip holes if any.
1728                  */
1729                 if (range < *trunc_end)
1730                         *trunc_end = range;
1731                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1732                 *blkno = le64_to_cpu(rec->e_blkno);
1733                 *trunc_end = le32_to_cpu(rec->e_cpos);
1734         } else if (range > trunc_start) {
1735                 /*
1736                  * remove a partial extent record, which means we're
1737                  * removing the last extent record.
1738                  */
1739                 *trunc_cpos = trunc_start;
1740                 /*
1741                  * skip hole if any.
1742                  */
1743                 if (range < *trunc_end)
1744                         *trunc_end = range;
1745                 *trunc_len = *trunc_end - trunc_start;
1746                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1747                 *blkno = le64_to_cpu(rec->e_blkno) +
1748                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1749                 *trunc_end = trunc_start;
1750         } else {
1751                 /*
1752                  * It may have two following possibilities:
1753                  *
1754                  * - last record has been removed
1755                  * - trunc_start was within a hole
1756                  *
1757                  * both two cases mean the completion of hole punching.
1758                  */
1759                 ret = 1;
1760         }
1761
1762         *done = ret;
1763 }
1764
1765 int ocfs2_remove_inode_range(struct inode *inode,
1766                              struct buffer_head *di_bh, u64 byte_start,
1767                              u64 byte_len)
1768 {
1769         int ret = 0, flags = 0, done = 0, i;
1770         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1771         u32 cluster_in_el;
1772         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1773         struct ocfs2_cached_dealloc_ctxt dealloc;
1774         struct address_space *mapping = inode->i_mapping;
1775         struct ocfs2_extent_tree et;
1776         struct ocfs2_path *path = NULL;
1777         struct ocfs2_extent_list *el = NULL;
1778         struct ocfs2_extent_rec *rec = NULL;
1779         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1780         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1781
1782         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1783         ocfs2_init_dealloc_ctxt(&dealloc);
1784
1785         trace_ocfs2_remove_inode_range(
1786                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1787                         (unsigned long long)byte_start,
1788                         (unsigned long long)byte_len);
1789
1790         if (byte_len == 0)
1791                 return 0;
1792
1793         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1794                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1795                                             byte_start + byte_len, 0);
1796                 if (ret) {
1797                         mlog_errno(ret);
1798                         goto out;
1799                 }
1800                 /*
1801                  * There's no need to get fancy with the page cache
1802                  * truncate of an inline-data inode. We're talking
1803                  * about less than a page here, which will be cached
1804                  * in the dinode buffer anyway.
1805                  */
1806                 unmap_mapping_range(mapping, 0, 0, 0);
1807                 truncate_inode_pages(mapping, 0);
1808                 goto out;
1809         }
1810
1811         /*
1812          * For reflinks, we may need to CoW 2 clusters which might be
1813          * partially zero'd later, if hole's start and end offset were
1814          * within one cluster(means is not exactly aligned to clustersize).
1815          */
1816
1817         if (ocfs2_is_refcount_inode(inode)) {
1818                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1819                 if (ret) {
1820                         mlog_errno(ret);
1821                         goto out;
1822                 }
1823
1824                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1825                 if (ret) {
1826                         mlog_errno(ret);
1827                         goto out;
1828                 }
1829         }
1830
1831         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1832         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1833         cluster_in_el = trunc_end;
1834
1835         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1836         if (ret) {
1837                 mlog_errno(ret);
1838                 goto out;
1839         }
1840
1841         path = ocfs2_new_path_from_et(&et);
1842         if (!path) {
1843                 ret = -ENOMEM;
1844                 mlog_errno(ret);
1845                 goto out;
1846         }
1847
1848         while (trunc_end > trunc_start) {
1849
1850                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1851                                       cluster_in_el);
1852                 if (ret) {
1853                         mlog_errno(ret);
1854                         goto out;
1855                 }
1856
1857                 el = path_leaf_el(path);
1858
1859                 i = ocfs2_find_rec(el, trunc_end);
1860                 /*
1861                  * Need to go to previous extent block.
1862                  */
1863                 if (i < 0) {
1864                         if (path->p_tree_depth == 0)
1865                                 break;
1866
1867                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1868                                                             path,
1869                                                             &cluster_in_el);
1870                         if (ret) {
1871                                 mlog_errno(ret);
1872                                 goto out;
1873                         }
1874
1875                         /*
1876                          * We've reached the leftmost extent block,
1877                          * it's safe to leave.
1878                          */
1879                         if (cluster_in_el == 0)
1880                                 break;
1881
1882                         /*
1883                          * The 'pos' searched for previous extent block is
1884                          * always one cluster less than actual trunc_end.
1885                          */
1886                         trunc_end = cluster_in_el + 1;
1887
1888                         ocfs2_reinit_path(path, 1);
1889
1890                         continue;
1891
1892                 } else
1893                         rec = &el->l_recs[i];
1894
1895                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1896                                      &trunc_len, &trunc_end, &blkno, &done);
1897                 if (done)
1898                         break;
1899
1900                 flags = rec->e_flags;
1901                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1902
1903                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1904                                                phys_cpos, trunc_len, flags,
1905                                                &dealloc, refcount_loc, false);
1906                 if (ret < 0) {
1907                         mlog_errno(ret);
1908                         goto out;
1909                 }
1910
1911                 cluster_in_el = trunc_end;
1912
1913                 ocfs2_reinit_path(path, 1);
1914         }
1915
1916         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1917
1918 out:
1919         ocfs2_free_path(path);
1920         ocfs2_schedule_truncate_log_flush(osb, 1);
1921         ocfs2_run_deallocs(osb, &dealloc);
1922
1923         return ret;
1924 }
1925
1926 /*
1927  * Parts of this function taken from xfs_change_file_space()
1928  */
1929 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1930                                      loff_t f_pos, unsigned int cmd,
1931                                      struct ocfs2_space_resv *sr,
1932                                      int change_size)
1933 {
1934         int ret;
1935         s64 llen;
1936         loff_t size, orig_isize;
1937         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1938         struct buffer_head *di_bh = NULL;
1939         handle_t *handle;
1940         unsigned long long max_off = inode->i_sb->s_maxbytes;
1941
1942         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1943                 return -EROFS;
1944
1945         inode_lock(inode);
1946
1947         /*
1948          * This prevents concurrent writes on other nodes
1949          */
1950         ret = ocfs2_rw_lock(inode, 1);
1951         if (ret) {
1952                 mlog_errno(ret);
1953                 goto out;
1954         }
1955
1956         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1957         if (ret) {
1958                 mlog_errno(ret);
1959                 goto out_rw_unlock;
1960         }
1961
1962         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1963                 ret = -EPERM;
1964                 goto out_inode_unlock;
1965         }
1966
1967         switch (sr->l_whence) {
1968         case 0: /*SEEK_SET*/
1969                 break;
1970         case 1: /*SEEK_CUR*/
1971                 sr->l_start += f_pos;
1972                 break;
1973         case 2: /*SEEK_END*/
1974                 sr->l_start += i_size_read(inode);
1975                 break;
1976         default:
1977                 ret = -EINVAL;
1978                 goto out_inode_unlock;
1979         }
1980         sr->l_whence = 0;
1981
1982         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1983
1984         if (sr->l_start < 0
1985             || sr->l_start > max_off
1986             || (sr->l_start + llen) < 0
1987             || (sr->l_start + llen) > max_off) {
1988                 ret = -EINVAL;
1989                 goto out_inode_unlock;
1990         }
1991         size = sr->l_start + sr->l_len;
1992
1993         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1994             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1995                 if (sr->l_len <= 0) {
1996                         ret = -EINVAL;
1997                         goto out_inode_unlock;
1998                 }
1999         }
2000
2001         if (file && should_remove_suid(file->f_path.dentry)) {
2002                 ret = __ocfs2_write_remove_suid(inode, di_bh);
2003                 if (ret) {
2004                         mlog_errno(ret);
2005                         goto out_inode_unlock;
2006                 }
2007         }
2008
2009         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2010         switch (cmd) {
2011         case OCFS2_IOC_RESVSP:
2012         case OCFS2_IOC_RESVSP64:
2013                 /*
2014                  * This takes unsigned offsets, but the signed ones we
2015                  * pass have been checked against overflow above.
2016                  */
2017                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2018                                                        sr->l_len);
2019                 break;
2020         case OCFS2_IOC_UNRESVSP:
2021         case OCFS2_IOC_UNRESVSP64:
2022                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2023                                                sr->l_len);
2024                 break;
2025         default:
2026                 ret = -EINVAL;
2027         }
2028
2029         orig_isize = i_size_read(inode);
2030         /* zeroout eof blocks in the cluster. */
2031         if (!ret && change_size && orig_isize < size) {
2032                 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2033                                         size - orig_isize);
2034                 if (!ret)
2035                         i_size_write(inode, size);
2036         }
2037         up_write(&OCFS2_I(inode)->ip_alloc_sem);
2038         if (ret) {
2039                 mlog_errno(ret);
2040                 goto out_inode_unlock;
2041         }
2042
2043         /*
2044          * We update c/mtime for these changes
2045          */
2046         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2047         if (IS_ERR(handle)) {
2048                 ret = PTR_ERR(handle);
2049                 mlog_errno(ret);
2050                 goto out_inode_unlock;
2051         }
2052
2053         inode->i_ctime = inode->i_mtime = current_time(inode);
2054         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2055         if (ret < 0)
2056                 mlog_errno(ret);
2057
2058         if (file && (file->f_flags & O_SYNC))
2059                 handle->h_sync = 1;
2060
2061         ocfs2_commit_trans(osb, handle);
2062
2063 out_inode_unlock:
2064         brelse(di_bh);
2065         ocfs2_inode_unlock(inode, 1);
2066 out_rw_unlock:
2067         ocfs2_rw_unlock(inode, 1);
2068
2069 out:
2070         inode_unlock(inode);
2071         return ret;
2072 }
2073
2074 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2075                             struct ocfs2_space_resv *sr)
2076 {
2077         struct inode *inode = file_inode(file);
2078         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2079         int ret;
2080
2081         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2082             !ocfs2_writes_unwritten_extents(osb))
2083                 return -ENOTTY;
2084         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2085                  !ocfs2_sparse_alloc(osb))
2086                 return -ENOTTY;
2087
2088         if (!S_ISREG(inode->i_mode))
2089                 return -EINVAL;
2090
2091         if (!(file->f_mode & FMODE_WRITE))
2092                 return -EBADF;
2093
2094         ret = mnt_want_write_file(file);
2095         if (ret)
2096                 return ret;
2097         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2098         mnt_drop_write_file(file);
2099         return ret;
2100 }
2101
2102 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2103                             loff_t len)
2104 {
2105         struct inode *inode = file_inode(file);
2106         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2107         struct ocfs2_space_resv sr;
2108         int change_size = 1;
2109         int cmd = OCFS2_IOC_RESVSP64;
2110
2111         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2112                 return -EOPNOTSUPP;
2113         if (!ocfs2_writes_unwritten_extents(osb))
2114                 return -EOPNOTSUPP;
2115
2116         if (mode & FALLOC_FL_KEEP_SIZE)
2117                 change_size = 0;
2118
2119         if (mode & FALLOC_FL_PUNCH_HOLE)
2120                 cmd = OCFS2_IOC_UNRESVSP64;
2121
2122         sr.l_whence = 0;
2123         sr.l_start = (s64)offset;
2124         sr.l_len = (s64)len;
2125
2126         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2127                                          change_size);
2128 }
2129
2130 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2131                                    size_t count)
2132 {
2133         int ret = 0;
2134         unsigned int extent_flags;
2135         u32 cpos, clusters, extent_len, phys_cpos;
2136         struct super_block *sb = inode->i_sb;
2137
2138         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2139             !ocfs2_is_refcount_inode(inode) ||
2140             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2141                 return 0;
2142
2143         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2144         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2145
2146         while (clusters) {
2147                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2148                                          &extent_flags);
2149                 if (ret < 0) {
2150                         mlog_errno(ret);
2151                         goto out;
2152                 }
2153
2154                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2155                         ret = 1;
2156                         break;
2157                 }
2158
2159                 if (extent_len > clusters)
2160                         extent_len = clusters;
2161
2162                 clusters -= extent_len;
2163                 cpos += extent_len;
2164         }
2165 out:
2166         return ret;
2167 }
2168
2169 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2170 {
2171         int blockmask = inode->i_sb->s_blocksize - 1;
2172         loff_t final_size = pos + count;
2173
2174         if ((pos & blockmask) || (final_size & blockmask))
2175                 return 1;
2176         return 0;
2177 }
2178
2179 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2180                                             struct buffer_head **di_bh,
2181                                             int meta_level,
2182                                             int write_sem,
2183                                             int wait)
2184 {
2185         int ret = 0;
2186
2187         if (wait)
2188                 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2189         else
2190                 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2191         if (ret < 0)
2192                 goto out;
2193
2194         if (wait) {
2195                 if (write_sem)
2196                         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2197                 else
2198                         down_read(&OCFS2_I(inode)->ip_alloc_sem);
2199         } else {
2200                 if (write_sem)
2201                         ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2202                 else
2203                         ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2204
2205                 if (!ret) {
2206                         ret = -EAGAIN;
2207                         goto out_unlock;
2208                 }
2209         }
2210
2211         return ret;
2212
2213 out_unlock:
2214         brelse(*di_bh);
2215         *di_bh = NULL;
2216         ocfs2_inode_unlock(inode, meta_level);
2217 out:
2218         return ret;
2219 }
2220
2221 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2222                                                struct buffer_head **di_bh,
2223                                                int meta_level,
2224                                                int write_sem)
2225 {
2226         if (write_sem)
2227                 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2228         else
2229                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2230
2231         brelse(*di_bh);
2232         *di_bh = NULL;
2233
2234         if (meta_level >= 0)
2235                 ocfs2_inode_unlock(inode, meta_level);
2236 }
2237
2238 static int ocfs2_prepare_inode_for_write(struct file *file,
2239                                          loff_t pos, size_t count, int wait)
2240 {
2241         int ret = 0, meta_level = 0, overwrite_io = 0;
2242         int write_sem = 0;
2243         struct dentry *dentry = file->f_path.dentry;
2244         struct inode *inode = d_inode(dentry);
2245         struct buffer_head *di_bh = NULL;
2246         loff_t end;
2247         u32 cpos;
2248         u32 clusters;
2249
2250         /*
2251          * We start with a read level meta lock and only jump to an ex
2252          * if we need to make modifications here.
2253          */
2254         for(;;) {
2255                 ret = ocfs2_inode_lock_for_extent_tree(inode,
2256                                                        &di_bh,
2257                                                        meta_level,
2258                                                        write_sem,
2259                                                        wait);
2260                 if (ret < 0) {
2261                         if (ret != -EAGAIN)
2262                                 mlog_errno(ret);
2263                         goto out;
2264                 }
2265
2266                 /*
2267                  * Check if IO will overwrite allocated blocks in case
2268                  * IOCB_NOWAIT flag is set.
2269                  */
2270                 if (!wait && !overwrite_io) {
2271                         overwrite_io = 1;
2272
2273                         ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2274                         if (ret < 0) {
2275                                 if (ret != -EAGAIN)
2276                                         mlog_errno(ret);
2277                                 goto out_unlock;
2278                         }
2279                 }
2280
2281                 /* Clear suid / sgid if necessary. We do this here
2282                  * instead of later in the write path because
2283                  * remove_suid() calls ->setattr without any hint that
2284                  * we may have already done our cluster locking. Since
2285                  * ocfs2_setattr() *must* take cluster locks to
2286                  * proceed, this will lead us to recursively lock the
2287                  * inode. There's also the dinode i_size state which
2288                  * can be lost via setattr during extending writes (we
2289                  * set inode->i_size at the end of a write. */
2290                 if (should_remove_suid(dentry)) {
2291                         if (meta_level == 0) {
2292                                 ocfs2_inode_unlock_for_extent_tree(inode,
2293                                                                    &di_bh,
2294                                                                    meta_level,
2295                                                                    write_sem);
2296                                 meta_level = 1;
2297                                 continue;
2298                         }
2299
2300                         ret = ocfs2_write_remove_suid(inode);
2301                         if (ret < 0) {
2302                                 mlog_errno(ret);
2303                                 goto out_unlock;
2304                         }
2305                 }
2306
2307                 end = pos + count;
2308
2309                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2310                 if (ret == 1) {
2311                         ocfs2_inode_unlock_for_extent_tree(inode,
2312                                                            &di_bh,
2313                                                            meta_level,
2314                                                            write_sem);
2315                         meta_level = 1;
2316                         write_sem = 1;
2317                         ret = ocfs2_inode_lock_for_extent_tree(inode,
2318                                                                &di_bh,
2319                                                                meta_level,
2320                                                                write_sem,
2321                                                                wait);
2322                         if (ret < 0) {
2323                                 if (ret != -EAGAIN)
2324                                         mlog_errno(ret);
2325                                 goto out;
2326                         }
2327
2328                         cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2329                         clusters =
2330                                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2331                         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2332                 }
2333
2334                 if (ret < 0) {
2335                         if (ret != -EAGAIN)
2336                                 mlog_errno(ret);
2337                         goto out_unlock;
2338                 }
2339
2340                 break;
2341         }
2342
2343 out_unlock:
2344         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2345                                             pos, count, wait);
2346
2347         ocfs2_inode_unlock_for_extent_tree(inode,
2348                                            &di_bh,
2349                                            meta_level,
2350                                            write_sem);
2351
2352 out:
2353         return ret;
2354 }
2355
2356 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2357                                     struct iov_iter *from)
2358 {
2359         int rw_level;
2360         ssize_t written = 0;
2361         ssize_t ret;
2362         size_t count = iov_iter_count(from);
2363         struct file *file = iocb->ki_filp;
2364         struct inode *inode = file_inode(file);
2365         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2366         int full_coherency = !(osb->s_mount_opt &
2367                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2368         void *saved_ki_complete = NULL;
2369         int append_write = ((iocb->ki_pos + count) >=
2370                         i_size_read(inode) ? 1 : 0);
2371         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2372         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2373
2374         trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2375                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2376                 file->f_path.dentry->d_name.len,
2377                 file->f_path.dentry->d_name.name,
2378                 (unsigned int)from->nr_segs);   /* GRRRRR */
2379
2380         if (!direct_io && nowait)
2381                 return -EOPNOTSUPP;
2382
2383         if (count == 0)
2384                 return 0;
2385
2386         if (nowait) {
2387                 if (!inode_trylock(inode))
2388                         return -EAGAIN;
2389         } else
2390                 inode_lock(inode);
2391
2392         /*
2393          * Concurrent O_DIRECT writes are allowed with
2394          * mount_option "coherency=buffered".
2395          * For append write, we must take rw EX.
2396          */
2397         rw_level = (!direct_io || full_coherency || append_write);
2398
2399         if (nowait)
2400                 ret = ocfs2_try_rw_lock(inode, rw_level);
2401         else
2402                 ret = ocfs2_rw_lock(inode, rw_level);
2403         if (ret < 0) {
2404                 if (ret != -EAGAIN)
2405                         mlog_errno(ret);
2406                 goto out_mutex;
2407         }
2408
2409         /*
2410          * O_DIRECT writes with "coherency=full" need to take EX cluster
2411          * inode_lock to guarantee coherency.
2412          */
2413         if (direct_io && full_coherency) {
2414                 /*
2415                  * We need to take and drop the inode lock to force
2416                  * other nodes to drop their caches.  Buffered I/O
2417                  * already does this in write_begin().
2418                  */
2419                 if (nowait)
2420                         ret = ocfs2_try_inode_lock(inode, NULL, 1);
2421                 else
2422                         ret = ocfs2_inode_lock(inode, NULL, 1);
2423                 if (ret < 0) {
2424                         if (ret != -EAGAIN)
2425                                 mlog_errno(ret);
2426                         goto out;
2427                 }
2428
2429                 ocfs2_inode_unlock(inode, 1);
2430         }
2431
2432         ret = generic_write_checks(iocb, from);
2433         if (ret <= 0) {
2434                 if (ret)
2435                         mlog_errno(ret);
2436                 goto out;
2437         }
2438         count = ret;
2439
2440         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2441         if (ret < 0) {
2442                 if (ret != -EAGAIN)
2443                         mlog_errno(ret);
2444                 goto out;
2445         }
2446
2447         if (direct_io && !is_sync_kiocb(iocb) &&
2448             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2449                 /*
2450                  * Make it a sync io if it's an unaligned aio.
2451                  */
2452                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2453         }
2454
2455         /* communicate with ocfs2_dio_end_io */
2456         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2457
2458         written = __generic_file_write_iter(iocb, from);
2459         /* buffered aio wouldn't have proper lock coverage today */
2460         BUG_ON(written == -EIOCBQUEUED && !direct_io);
2461
2462         /*
2463          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2464          * function pointer which is called when o_direct io completes so that
2465          * it can unlock our rw lock.
2466          * Unfortunately there are error cases which call end_io and others
2467          * that don't.  so we don't have to unlock the rw_lock if either an
2468          * async dio is going to do it in the future or an end_io after an
2469          * error has already done it.
2470          */
2471         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2472                 rw_level = -1;
2473         }
2474
2475         if (unlikely(written <= 0))
2476                 goto out;
2477
2478         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2479             IS_SYNC(inode)) {
2480                 ret = filemap_fdatawrite_range(file->f_mapping,
2481                                                iocb->ki_pos - written,
2482                                                iocb->ki_pos - 1);
2483                 if (ret < 0)
2484                         written = ret;
2485
2486                 if (!ret) {
2487                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2488                         if (ret < 0)
2489                                 written = ret;
2490                 }
2491
2492                 if (!ret)
2493                         ret = filemap_fdatawait_range(file->f_mapping,
2494                                                       iocb->ki_pos - written,
2495                                                       iocb->ki_pos - 1);
2496         }
2497
2498 out:
2499         if (saved_ki_complete)
2500                 xchg(&iocb->ki_complete, saved_ki_complete);
2501
2502         if (rw_level != -1)
2503                 ocfs2_rw_unlock(inode, rw_level);
2504
2505 out_mutex:
2506         inode_unlock(inode);
2507
2508         if (written)
2509                 ret = written;
2510         return ret;
2511 }
2512
2513 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2514                                    struct iov_iter *to)
2515 {
2516         int ret = 0, rw_level = -1, lock_level = 0;
2517         struct file *filp = iocb->ki_filp;
2518         struct inode *inode = file_inode(filp);
2519         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2520         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2521
2522         trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2523                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2524                         filp->f_path.dentry->d_name.len,
2525                         filp->f_path.dentry->d_name.name,
2526                         to->nr_segs);   /* GRRRRR */
2527
2528
2529         if (!inode) {
2530                 ret = -EINVAL;
2531                 mlog_errno(ret);
2532                 goto bail;
2533         }
2534
2535         if (!direct_io && nowait)
2536                 return -EOPNOTSUPP;
2537
2538         /*
2539          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2540          * need locks to protect pending reads from racing with truncate.
2541          */
2542         if (direct_io) {
2543                 if (nowait)
2544                         ret = ocfs2_try_rw_lock(inode, 0);
2545                 else
2546                         ret = ocfs2_rw_lock(inode, 0);
2547
2548                 if (ret < 0) {
2549                         if (ret != -EAGAIN)
2550                                 mlog_errno(ret);
2551                         goto bail;
2552                 }
2553                 rw_level = 0;
2554                 /* communicate with ocfs2_dio_end_io */
2555                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2556         }
2557
2558         /*
2559          * We're fine letting folks race truncates and extending
2560          * writes with read across the cluster, just like they can
2561          * locally. Hence no rw_lock during read.
2562          *
2563          * Take and drop the meta data lock to update inode fields
2564          * like i_size. This allows the checks down below
2565          * generic_file_read_iter() a chance of actually working.
2566          */
2567         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2568                                      !nowait);
2569         if (ret < 0) {
2570                 if (ret != -EAGAIN)
2571                         mlog_errno(ret);
2572                 goto bail;
2573         }
2574         ocfs2_inode_unlock(inode, lock_level);
2575
2576         ret = generic_file_read_iter(iocb, to);
2577         trace_generic_file_read_iter_ret(ret);
2578
2579         /* buffered aio wouldn't have proper lock coverage today */
2580         BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2581
2582         /* see ocfs2_file_write_iter */
2583         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2584                 rw_level = -1;
2585         }
2586
2587 bail:
2588         if (rw_level != -1)
2589                 ocfs2_rw_unlock(inode, rw_level);
2590
2591         return ret;
2592 }
2593
2594 /* Refer generic_file_llseek_unlocked() */
2595 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2596 {
2597         struct inode *inode = file->f_mapping->host;
2598         int ret = 0;
2599
2600         inode_lock(inode);
2601
2602         switch (whence) {
2603         case SEEK_SET:
2604                 break;
2605         case SEEK_END:
2606                 /* SEEK_END requires the OCFS2 inode lock for the file
2607                  * because it references the file's size.
2608                  */
2609                 ret = ocfs2_inode_lock(inode, NULL, 0);
2610                 if (ret < 0) {
2611                         mlog_errno(ret);
2612                         goto out;
2613                 }
2614                 offset += i_size_read(inode);
2615                 ocfs2_inode_unlock(inode, 0);
2616                 break;
2617         case SEEK_CUR:
2618                 if (offset == 0) {
2619                         offset = file->f_pos;
2620                         goto out;
2621                 }
2622                 offset += file->f_pos;
2623                 break;
2624         case SEEK_DATA:
2625         case SEEK_HOLE:
2626                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2627                 if (ret)
2628                         goto out;
2629                 break;
2630         default:
2631                 ret = -EINVAL;
2632                 goto out;
2633         }
2634
2635         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2636
2637 out:
2638         inode_unlock(inode);
2639         if (ret)
2640                 return ret;
2641         return offset;
2642 }
2643
2644 static int ocfs2_file_clone_range(struct file *file_in,
2645                                   loff_t pos_in,
2646                                   struct file *file_out,
2647                                   loff_t pos_out,
2648                                   u64 len)
2649 {
2650         return ocfs2_reflink_remap_range(file_in, pos_in, file_out, pos_out,
2651                                          len, false);
2652 }
2653
2654 static int ocfs2_file_dedupe_range(struct file *file_in,
2655                                    loff_t pos_in,
2656                                    struct file *file_out,
2657                                    loff_t pos_out,
2658                                    u64 len)
2659 {
2660         return ocfs2_reflink_remap_range(file_in, pos_in, file_out, pos_out,
2661                                           len, true);
2662 }
2663
2664 const struct inode_operations ocfs2_file_iops = {
2665         .setattr        = ocfs2_setattr,
2666         .getattr        = ocfs2_getattr,
2667         .permission     = ocfs2_permission,
2668         .listxattr      = ocfs2_listxattr,
2669         .fiemap         = ocfs2_fiemap,
2670         .get_acl        = ocfs2_iop_get_acl,
2671         .set_acl        = ocfs2_iop_set_acl,
2672 };
2673
2674 const struct inode_operations ocfs2_special_file_iops = {
2675         .setattr        = ocfs2_setattr,
2676         .getattr        = ocfs2_getattr,
2677         .permission     = ocfs2_permission,
2678         .get_acl        = ocfs2_iop_get_acl,
2679         .set_acl        = ocfs2_iop_set_acl,
2680 };
2681
2682 /*
2683  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2684  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2685  */
2686 const struct file_operations ocfs2_fops = {
2687         .llseek         = ocfs2_file_llseek,
2688         .mmap           = ocfs2_mmap,
2689         .fsync          = ocfs2_sync_file,
2690         .release        = ocfs2_file_release,
2691         .open           = ocfs2_file_open,
2692         .read_iter      = ocfs2_file_read_iter,
2693         .write_iter     = ocfs2_file_write_iter,
2694         .unlocked_ioctl = ocfs2_ioctl,
2695 #ifdef CONFIG_COMPAT
2696         .compat_ioctl   = ocfs2_compat_ioctl,
2697 #endif
2698         .lock           = ocfs2_lock,
2699         .flock          = ocfs2_flock,
2700         .splice_read    = generic_file_splice_read,
2701         .splice_write   = iter_file_splice_write,
2702         .fallocate      = ocfs2_fallocate,
2703         .clone_file_range = ocfs2_file_clone_range,
2704         .dedupe_file_range = ocfs2_file_dedupe_range,
2705 };
2706
2707 const struct file_operations ocfs2_dops = {
2708         .llseek         = generic_file_llseek,
2709         .read           = generic_read_dir,
2710         .iterate        = ocfs2_readdir,
2711         .fsync          = ocfs2_sync_file,
2712         .release        = ocfs2_dir_release,
2713         .open           = ocfs2_dir_open,
2714         .unlocked_ioctl = ocfs2_ioctl,
2715 #ifdef CONFIG_COMPAT
2716         .compat_ioctl   = ocfs2_compat_ioctl,
2717 #endif
2718         .lock           = ocfs2_lock,
2719         .flock          = ocfs2_flock,
2720 };
2721
2722 /*
2723  * POSIX-lockless variants of our file_operations.
2724  *
2725  * These will be used if the underlying cluster stack does not support
2726  * posix file locking, if the user passes the "localflocks" mount
2727  * option, or if we have a local-only fs.
2728  *
2729  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2730  * so we still want it in the case of no stack support for
2731  * plocks. Internally, it will do the right thing when asked to ignore
2732  * the cluster.
2733  */
2734 const struct file_operations ocfs2_fops_no_plocks = {
2735         .llseek         = ocfs2_file_llseek,
2736         .mmap           = ocfs2_mmap,
2737         .fsync          = ocfs2_sync_file,
2738         .release        = ocfs2_file_release,
2739         .open           = ocfs2_file_open,
2740         .read_iter      = ocfs2_file_read_iter,
2741         .write_iter     = ocfs2_file_write_iter,
2742         .unlocked_ioctl = ocfs2_ioctl,
2743 #ifdef CONFIG_COMPAT
2744         .compat_ioctl   = ocfs2_compat_ioctl,
2745 #endif
2746         .flock          = ocfs2_flock,
2747         .splice_read    = generic_file_splice_read,
2748         .splice_write   = iter_file_splice_write,
2749         .fallocate      = ocfs2_fallocate,
2750         .clone_file_range = ocfs2_file_clone_range,
2751         .dedupe_file_range = ocfs2_file_dedupe_range,
2752 };
2753
2754 const struct file_operations ocfs2_dops_no_plocks = {
2755         .llseek         = generic_file_llseek,
2756         .read           = generic_read_dir,
2757         .iterate        = ocfs2_readdir,
2758         .fsync          = ocfs2_sync_file,
2759         .release        = ocfs2_dir_release,
2760         .open           = ocfs2_dir_open,
2761         .unlocked_ioctl = ocfs2_ioctl,
2762 #ifdef CONFIG_COMPAT
2763         .compat_ioctl   = ocfs2_compat_ioctl,
2764 #endif
2765         .flock          = ocfs2_flock,
2766 };