fs/ocfs2/file.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * file.c
   4  *
   5  * File open, close, extend, truncate
   6  *
   7  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   8  */
   9
  10 #include <linux/capability.h>
  11 #include <linux/fs.h>
  12 #include <linux/types.h>
  13 #include <linux/slab.h>
  14 #include <linux/highmem.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/uio.h>
  17 #include <linux/sched.h>
  18 #include <linux/splice.h>
  19 #include <linux/mount.h>
  20 #include <linux/writeback.h>
  21 #include <linux/falloc.h>
  22 #include <linux/quotaops.h>
  23 #include <linux/blkdev.h>
  24 #include <linux/backing-dev.h>
  25
  26 #include <cluster/masklog.h>
  27
  28 #include "ocfs2.h"
  29
  30 #include "alloc.h"
  31 #include "aops.h"
  32 #include "dir.h"
  33 #include "dlmglue.h"
  34 #include "extent_map.h"
  35 #include "file.h"
  36 #include "sysfile.h"
  37 #include "inode.h"
  38 #include "ioctl.h"
  39 #include "journal.h"
  40 #include "locks.h"
  41 #include "mmap.h"
  42 #include "suballoc.h"
  43 #include "super.h"
  44 #include "xattr.h"
  45 #include "acl.h"
  46 #include "quota.h"
  47 #include "refcounttree.h"
  48 #include "ocfs2_trace.h"
  49
  50 #include "buffer_head_io.h"
  51
  52 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  53 {
  54         struct ocfs2_file_private *fp;
  55
  56         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  57         if (!fp)
  58                 return -ENOMEM;
  59
  60         fp->fp_file = file;
  61         mutex_init(&fp->fp_mutex);
  62         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  63         file->private_data = fp;
  64
  65         return 0;
  66 }
  67
  68 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  69 {
  70         struct ocfs2_file_private *fp = file->private_data;
  71         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  72
  73         if (fp) {
  74                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  75                 ocfs2_lock_res_free(&fp->fp_flock);
  76                 kfree(fp);
  77                 file->private_data = NULL;
  78         }
  79 }
  80
  81 static int ocfs2_file_open(struct inode *inode, struct file *file)
  82 {
  83         int status;
  84         int mode = file->f_flags;
  85         struct ocfs2_inode_info *oi = OCFS2_I(inode);
  86
  87         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
  88                               (unsigned long long)oi->ip_blkno,
  89                               file->f_path.dentry->d_name.len,
  90                               file->f_path.dentry->d_name.name, mode);
  91
  92         if (file->f_mode & FMODE_WRITE) {
  93                 status = dquot_initialize(inode);
  94                 if (status)
  95                         goto leave;
  96         }
  97
  98         spin_lock(&oi->ip_lock);
  99
 100         /* Check that the inode hasn't been wiped from disk by another
 101          * node. If it hasn't then we're safe as long as we hold the
 102          * spin lock until our increment of open count. */
 103         if (oi->ip_flags & OCFS2_INODE_DELETED) {
 104                 spin_unlock(&oi->ip_lock);
 105
 106                 status = -ENOENT;
 107                 goto leave;
 108         }
 109
 110         if (mode & O_DIRECT)
 111                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 112
 113         oi->ip_open_count++;
 114         spin_unlock(&oi->ip_lock);
 115
 116         status = ocfs2_init_file_private(inode, file);
 117         if (status) {
 118                 /*
 119                  * We want to set open count back if we're failing the
 120                  * open.
 121                  */
 122                 spin_lock(&oi->ip_lock);
 123                 oi->ip_open_count--;
 124                 spin_unlock(&oi->ip_lock);
 125         }
 126
 127         file->f_mode |= FMODE_NOWAIT;
 128
 129 leave:
 130         return status;
 131 }
 132
 133 static int ocfs2_file_release(struct inode *inode, struct file *file)
 134 {
 135         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 136
 137         spin_lock(&oi->ip_lock);
 138         if (!--oi->ip_open_count)
 139                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 140
 141         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 142                                  oi->ip_blkno,
 143                                  file->f_path.dentry->d_name.len,
 144                                  file->f_path.dentry->d_name.name,
 145                                  oi->ip_open_count);
 146         spin_unlock(&oi->ip_lock);
 147
 148         ocfs2_free_file_private(inode, file);
 149
 150         return 0;
 151 }
 152
 153 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 154 {
 155         return ocfs2_init_file_private(inode, file);
 156 }
 157
 158 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 159 {
 160         ocfs2_free_file_private(inode, file);
 161         return 0;
 162 }
 163
 164 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 165                            int datasync)
 166 {
 167         int err = 0;
 168         struct inode *inode = file->f_mapping->host;
 169         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 170         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 171         journal_t *journal = osb->journal->j_journal;
 172         int ret;
 173         tid_t commit_tid;
 174         bool needs_barrier = false;
 175
 176         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 177                               oi->ip_blkno,
 178                               file->f_path.dentry->d_name.len,
 179                               file->f_path.dentry->d_name.name,
 180                               (unsigned long long)datasync);
 181
 182         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 183                 return -EROFS;
 184
 185         err = file_write_and_wait_range(file, start, end);
 186         if (err)
 187                 return err;
 188
 189         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 190         if (journal->j_flags & JBD2_BARRIER &&
 191             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 192                 needs_barrier = true;
 193         err = jbd2_complete_transaction(journal, commit_tid);
 194         if (needs_barrier) {
 195                 ret = blkdev_issue_flush(inode->i_sb->s_bdev);
 196                 if (!err)
 197                         err = ret;
 198         }
 199
 200         if (err)
 201                 mlog_errno(err);
 202
 203         return (err < 0) ? -EIO : 0;
 204 }
 205
 206 int ocfs2_should_update_atime(struct inode *inode,
 207                               struct vfsmount *vfsmnt)
 208 {
 209         struct timespec64 now;
 210         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 211
 212         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 213                 return 0;
 214
 215         if ((inode->i_flags & S_NOATIME) ||
 216             ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
 217                 return 0;
 218
 219         /*
 220          * We can be called with no vfsmnt structure - NFSD will
 221          * sometimes do this.
 222          *
 223          * Note that our action here is different than touch_atime() -
 224          * if we can't tell whether this is a noatime mount, then we
 225          * don't know whether to trust the value of s_atime_quantum.
 226          */
 227         if (vfsmnt == NULL)
 228                 return 0;
 229
 230         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 231             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 232                 return 0;
 233
 234         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 235                 if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 236                     (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 237                         return 1;
 238
 239                 return 0;
 240         }
 241
 242         now = current_time(inode);
 243         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 244                 return 0;
 245         else
 246                 return 1;
 247 }
 248
 249 int ocfs2_update_inode_atime(struct inode *inode,
 250                              struct buffer_head *bh)
 251 {
 252         int ret;
 253         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 254         handle_t *handle;
 255         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 256
 257         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 258         if (IS_ERR(handle)) {
 259                 ret = PTR_ERR(handle);
 260                 mlog_errno(ret);
 261                 goto out;
 262         }
 263
 264         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 265                                       OCFS2_JOURNAL_ACCESS_WRITE);
 266         if (ret) {
 267                 mlog_errno(ret);
 268                 goto out_commit;
 269         }
 270
 271         /*
 272          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 273          * have i_mutex to guard against concurrent changes to other
 274          * inode fields.
 275          */
 276         inode->i_atime = current_time(inode);
 277         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 278         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 279         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 280         ocfs2_journal_dirty(handle, bh);
 281
 282 out_commit:
 283         ocfs2_commit_trans(osb, handle);
 284 out:
 285         return ret;
 286 }
 287
 288 int ocfs2_set_inode_size(handle_t *handle,
 289                                 struct inode *inode,
 290                                 struct buffer_head *fe_bh,
 291                                 u64 new_i_size)
 292 {
 293         int status;
 294
 295         i_size_write(inode, new_i_size);
 296         inode->i_blocks = ocfs2_inode_sector_count(inode);
 297         inode->i_ctime = inode->i_mtime = current_time(inode);
 298
 299         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 300         if (status < 0) {
 301                 mlog_errno(status);
 302                 goto bail;
 303         }
 304
 305 bail:
 306         return status;
 307 }
 308
 309 int ocfs2_simple_size_update(struct inode *inode,
 310                              struct buffer_head *di_bh,
 311                              u64 new_i_size)
 312 {
 313         int ret;
 314         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 315         handle_t *handle = NULL;
 316
 317         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 318         if (IS_ERR(handle)) {
 319                 ret = PTR_ERR(handle);
 320                 mlog_errno(ret);
 321                 goto out;
 322         }
 323
 324         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 325                                    new_i_size);
 326         if (ret < 0)
 327                 mlog_errno(ret);
 328
 329         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 330         ocfs2_commit_trans(osb, handle);
 331 out:
 332         return ret;
 333 }
 334
 335 static int ocfs2_cow_file_pos(struct inode *inode,
 336                               struct buffer_head *fe_bh,
 337                               u64 offset)
 338 {
 339         int status;
 340         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 341         unsigned int num_clusters = 0;
 342         unsigned int ext_flags = 0;
 343
 344         /*
 345          * If the new offset is aligned to the range of the cluster, there is
 346          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 347          * CoW either.
 348          */
 349         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 350                 return 0;
 351
 352         status = ocfs2_get_clusters(inode, cpos, &phys,
 353                                     &num_clusters, &ext_flags);
 354         if (status) {
 355                 mlog_errno(status);
 356                 goto out;
 357         }
 358
 359         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 360                 goto out;
 361
 362         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 363
 364 out:
 365         return status;
 366 }
 367
 368 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 369                                      struct inode *inode,
 370                                      struct buffer_head *fe_bh,
 371                                      u64 new_i_size)
 372 {
 373         int status;
 374         handle_t *handle;
 375         struct ocfs2_dinode *di;
 376         u64 cluster_bytes;
 377
 378         /*
 379          * We need to CoW the cluster contains the offset if it is reflinked
 380          * since we will call ocfs2_zero_range_for_truncate later which will
 381          * write "0" from offset to the end of the cluster.
 382          */
 383         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 384         if (status) {
 385                 mlog_errno(status);
 386                 return status;
 387         }
 388
 389         /* TODO: This needs to actually orphan the inode in this
 390          * transaction. */
 391
 392         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 393         if (IS_ERR(handle)) {
 394                 status = PTR_ERR(handle);
 395                 mlog_errno(status);
 396                 goto out;
 397         }
 398
 399         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 400                                          OCFS2_JOURNAL_ACCESS_WRITE);
 401         if (status < 0) {
 402                 mlog_errno(status);
 403                 goto out_commit;
 404         }
 405
 406         /*
 407          * Do this before setting i_size.
 408          */
 409         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 410         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 411                                                cluster_bytes);
 412         if (status) {
 413                 mlog_errno(status);
 414                 goto out_commit;
 415         }
 416
 417         i_size_write(inode, new_i_size);
 418         inode->i_ctime = inode->i_mtime = current_time(inode);
 419
 420         di = (struct ocfs2_dinode *) fe_bh->b_data;
 421         di->i_size = cpu_to_le64(new_i_size);
 422         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 423         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 424         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 425
 426         ocfs2_journal_dirty(handle, fe_bh);
 427
 428 out_commit:
 429         ocfs2_commit_trans(osb, handle);
 430 out:
 431         return status;
 432 }
 433
 434 int ocfs2_truncate_file(struct inode *inode,
 435                                struct buffer_head *di_bh,
 436                                u64 new_i_size)
 437 {
 438         int status = 0;
 439         struct ocfs2_dinode *fe = NULL;
 440         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 441
 442         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 443          * already validated it */
 444         fe = (struct ocfs2_dinode *) di_bh->b_data;
 445
 446         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 447                                   (unsigned long long)le64_to_cpu(fe->i_size),
 448                                   (unsigned long long)new_i_size);
 449
 450         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 451                         "Inode %llu, inode i_size = %lld != di "
 452                         "i_size = %llu, i_flags = 0x%x\n",
 453                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 454                         i_size_read(inode),
 455                         (unsigned long long)le64_to_cpu(fe->i_size),
 456                         le32_to_cpu(fe->i_flags));
 457
 458         if (new_i_size > le64_to_cpu(fe->i_size)) {
 459                 trace_ocfs2_truncate_file_error(
 460                         (unsigned long long)le64_to_cpu(fe->i_size),
 461                         (unsigned long long)new_i_size);
 462                 status = -EINVAL;
 463                 mlog_errno(status);
 464                 goto bail;
 465         }
 466
 467         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 468
 469         ocfs2_resv_discard(&osb->osb_la_resmap,
 470                            &OCFS2_I(inode)->ip_la_data_resv);
 471
 472         /*
 473          * The inode lock forced other nodes to sync and drop their
 474          * pages, which (correctly) happens even if we have a truncate
 475          * without allocation change - ocfs2 cluster sizes can be much
 476          * greater than page size, so we have to truncate them
 477          * anyway.
 478          */
 479
 480         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 481                 unmap_mapping_range(inode->i_mapping,
 482                                     new_i_size + PAGE_SIZE - 1, 0, 1);
 483                 truncate_inode_pages(inode->i_mapping, new_i_size);
 484                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 485                                                i_size_read(inode), 1);
 486                 if (status)
 487                         mlog_errno(status);
 488
 489                 goto bail_unlock_sem;
 490         }
 491
 492         /* alright, we're going to need to do a full blown alloc size
 493          * change. Orphan the inode so that recovery can complete the
 494          * truncate if necessary. This does the task of marking
 495          * i_size. */
 496         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 497         if (status < 0) {
 498                 mlog_errno(status);
 499                 goto bail_unlock_sem;
 500         }
 501
 502         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 503         truncate_inode_pages(inode->i_mapping, new_i_size);
 504
 505         status = ocfs2_commit_truncate(osb, inode, di_bh);
 506         if (status < 0) {
 507                 mlog_errno(status);
 508                 goto bail_unlock_sem;
 509         }
 510
 511         /* TODO: orphan dir cleanup here. */
 512 bail_unlock_sem:
 513         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 514
 515 bail:
 516         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 517                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 518
 519         return status;
 520 }
 521
 522 /*
 523  * extend file allocation only here.
 524  * we'll update all the disk stuff, and oip->alloc_size
 525  *
 526  * expect stuff to be locked, a transaction started and enough data /
 527  * metadata reservations in the contexts.
 528  *
 529  * Will return -EAGAIN, and a reason if a restart is needed.
 530  * If passed in, *reason will always be set, even in error.
 531  */
 532 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 533                          struct inode *inode,
 534                          u32 *logical_offset,
 535                          u32 clusters_to_add,
 536                          int mark_unwritten,
 537                          struct buffer_head *fe_bh,
 538                          handle_t *handle,
 539                          struct ocfs2_alloc_context *data_ac,
 540                          struct ocfs2_alloc_context *meta_ac,
 541                          enum ocfs2_alloc_restarted *reason_ret)
 542 {
 543         int ret;
 544         struct ocfs2_extent_tree et;
 545
 546         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 547         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 548                                           clusters_to_add, mark_unwritten,
 549                                           data_ac, meta_ac, reason_ret);
 550
 551         return ret;
 552 }
 553
 554 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 555                                    u32 clusters_to_add, int mark_unwritten)
 556 {
 557         int status = 0;
 558         int restart_func = 0;
 559         int credits;
 560         u32 prev_clusters;
 561         struct buffer_head *bh = NULL;
 562         struct ocfs2_dinode *fe = NULL;
 563         handle_t *handle = NULL;
 564         struct ocfs2_alloc_context *data_ac = NULL;
 565         struct ocfs2_alloc_context *meta_ac = NULL;
 566         enum ocfs2_alloc_restarted why = RESTART_NONE;
 567         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 568         struct ocfs2_extent_tree et;
 569         int did_quota = 0;
 570
 571         /*
 572          * Unwritten extent only exists for file systems which
 573          * support holes.
 574          */
 575         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 576
 577         status = ocfs2_read_inode_block(inode, &bh);
 578         if (status < 0) {
 579                 mlog_errno(status);
 580                 goto leave;
 581         }
 582         fe = (struct ocfs2_dinode *) bh->b_data;
 583
 584 restart_all:
 585         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 586
 587         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 588         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 589                                        &data_ac, &meta_ac);
 590         if (status) {
 591                 mlog_errno(status);
 592                 goto leave;
 593         }
 594
 595         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 596         handle = ocfs2_start_trans(osb, credits);
 597         if (IS_ERR(handle)) {
 598                 status = PTR_ERR(handle);
 599                 handle = NULL;
 600                 mlog_errno(status);
 601                 goto leave;
 602         }
 603
 604 restarted_transaction:
 605         trace_ocfs2_extend_allocation(
 606                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 607                 (unsigned long long)i_size_read(inode),
 608                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 609                 why, restart_func);
 610
 611         status = dquot_alloc_space_nodirty(inode,
 612                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 613         if (status)
 614                 goto leave;
 615         did_quota = 1;
 616
 617         /* reserve a write to the file entry early on - that we if we
 618          * run out of credits in the allocation path, we can still
 619          * update i_size. */
 620         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 621                                          OCFS2_JOURNAL_ACCESS_WRITE);
 622         if (status < 0) {
 623                 mlog_errno(status);
 624                 goto leave;
 625         }
 626
 627         prev_clusters = OCFS2_I(inode)->ip_clusters;
 628
 629         status = ocfs2_add_inode_data(osb,
 630                                       inode,
 631                                       &logical_start,
 632                                       clusters_to_add,
 633                                       mark_unwritten,
 634                                       bh,
 635                                       handle,
 636                                       data_ac,
 637                                       meta_ac,
 638                                       &why);
 639         if ((status < 0) && (status != -EAGAIN)) {
 640                 if (status != -ENOSPC)
 641                         mlog_errno(status);
 642                 goto leave;
 643         }
 644         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 645         ocfs2_journal_dirty(handle, bh);
 646
 647         spin_lock(&OCFS2_I(inode)->ip_lock);
 648         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 649         spin_unlock(&OCFS2_I(inode)->ip_lock);
 650         /* Release unused quota reservation */
 651         dquot_free_space(inode,
 652                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 653         did_quota = 0;
 654
 655         if (why != RESTART_NONE && clusters_to_add) {
 656                 if (why == RESTART_META) {
 657                         restart_func = 1;
 658                         status = 0;
 659                 } else {
 660                         BUG_ON(why != RESTART_TRANS);
 661
 662                         status = ocfs2_allocate_extend_trans(handle, 1);
 663                         if (status < 0) {
 664                                 /* handle still has to be committed at
 665                                  * this point. */
 666                                 status = -ENOMEM;
 667                                 mlog_errno(status);
 668                                 goto leave;
 669                         }
 670                         goto restarted_transaction;
 671                 }
 672         }
 673
 674         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 675              le32_to_cpu(fe->i_clusters),
 676              (unsigned long long)le64_to_cpu(fe->i_size),
 677              OCFS2_I(inode)->ip_clusters,
 678              (unsigned long long)i_size_read(inode));
 679
 680 leave:
 681         if (status < 0 && did_quota)
 682                 dquot_free_space(inode,
 683                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 684         if (handle) {
 685                 ocfs2_commit_trans(osb, handle);
 686                 handle = NULL;
 687         }
 688         if (data_ac) {
 689                 ocfs2_free_alloc_context(data_ac);
 690                 data_ac = NULL;
 691         }
 692         if (meta_ac) {
 693                 ocfs2_free_alloc_context(meta_ac);
 694                 meta_ac = NULL;
 695         }
 696         if ((!status) && restart_func) {
 697                 restart_func = 0;
 698                 goto restart_all;
 699         }
 700         brelse(bh);
 701         bh = NULL;
 702
 703         return status;
 704 }
 705
 706 /*
 707  * While a write will already be ordering the data, a truncate will not.
 708  * Thus, we need to explicitly order the zeroed pages.
 709  */
 710 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 711                                                       struct buffer_head *di_bh,
 712                                                       loff_t start_byte,
 713                                                       loff_t length)
 714 {
 715         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 716         handle_t *handle = NULL;
 717         int ret = 0;
 718
 719         if (!ocfs2_should_order_data(inode))
 720                 goto out;
 721
 722         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 723         if (IS_ERR(handle)) {
 724                 ret = -ENOMEM;
 725                 mlog_errno(ret);
 726                 goto out;
 727         }
 728
 729         ret = ocfs2_jbd2_inode_add_write(handle, inode, start_byte, length);
 730         if (ret < 0) {
 731                 mlog_errno(ret);
 732                 goto out;
 733         }
 734
 735         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 736                                       OCFS2_JOURNAL_ACCESS_WRITE);
 737         if (ret)
 738                 mlog_errno(ret);
 739         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 740
 741 out:
 742         if (ret) {
 743                 if (!IS_ERR(handle))
 744                         ocfs2_commit_trans(osb, handle);
 745                 handle = ERR_PTR(ret);
 746         }
 747         return handle;
 748 }
 749
 750 /* Some parts of this taken from generic_cont_expand, which turned out
 751  * to be too fragile to do exactly what we need without us having to
 752  * worry about recursive locking in ->write_begin() and ->write_end(). */
 753 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 754                                  u64 abs_to, struct buffer_head *di_bh)
 755 {
 756         struct address_space *mapping = inode->i_mapping;
 757         struct page *page;
 758         unsigned long index = abs_from >> PAGE_SHIFT;
 759         handle_t *handle;
 760         int ret = 0;
 761         unsigned zero_from, zero_to, block_start, block_end;
 762         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 763
 764         BUG_ON(abs_from >= abs_to);
 765         BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 766         BUG_ON(abs_from & (inode->i_blkbits - 1));
 767
 768         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh,
 769                                                       abs_from,
 770                                                       abs_to - abs_from);
 771         if (IS_ERR(handle)) {
 772                 ret = PTR_ERR(handle);
 773                 goto out;
 774         }
 775
 776         page = find_or_create_page(mapping, index, GFP_NOFS);
 777         if (!page) {
 778                 ret = -ENOMEM;
 779                 mlog_errno(ret);
 780                 goto out_commit_trans;
 781         }
 782
 783         /* Get the offsets within the page that we want to zero */
 784         zero_from = abs_from & (PAGE_SIZE - 1);
 785         zero_to = abs_to & (PAGE_SIZE - 1);
 786         if (!zero_to)
 787                 zero_to = PAGE_SIZE;
 788
 789         trace_ocfs2_write_zero_page(
 790                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 791                         (unsigned long long)abs_from,
 792                         (unsigned long long)abs_to,
 793                         index, zero_from, zero_to);
 794
 795         /* We know that zero_from is block aligned */
 796         for (block_start = zero_from; block_start < zero_to;
 797              block_start = block_end) {
 798                 block_end = block_start + i_blocksize(inode);
 799
 800                 /*
 801                  * block_start is block-aligned.  Bump it by one to force
 802                  * __block_write_begin and block_commit_write to zero the
 803                  * whole block.
 804                  */
 805                 ret = __block_write_begin(page, block_start + 1, 0,
 806                                           ocfs2_get_block);
 807                 if (ret < 0) {
 808                         mlog_errno(ret);
 809                         goto out_unlock;
 810                 }
 811
 812
 813                 /* must not update i_size! */
 814                 ret = block_commit_write(page, block_start + 1,
 815                                          block_start + 1);
 816                 if (ret < 0)
 817                         mlog_errno(ret);
 818                 else
 819                         ret = 0;
 820         }
 821
 822         /*
 823          * fs-writeback will release the dirty pages without page lock
 824          * whose offset are over inode size, the release happens at
 825          * block_write_full_page().
 826          */
 827         i_size_write(inode, abs_to);
 828         inode->i_blocks = ocfs2_inode_sector_count(inode);
 829         di->i_size = cpu_to_le64((u64)i_size_read(inode));
 830         inode->i_mtime = inode->i_ctime = current_time(inode);
 831         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 832         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 833         di->i_mtime_nsec = di->i_ctime_nsec;
 834         if (handle) {
 835                 ocfs2_journal_dirty(handle, di_bh);
 836                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 837         }
 838
 839 out_unlock:
 840         unlock_page(page);
 841         put_page(page);
 842 out_commit_trans:
 843         if (handle)
 844                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 845 out:
 846         return ret;
 847 }
 848
 849 /*
 850  * Find the next range to zero.  We do this in terms of bytes because
 851  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 852  * pagecache.  We may return multiple extents.
 853  *
 854  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 855  * needs to be zeroed.  range_start and range_end return the next zeroing
 856  * range.  A subsequent call should pass the previous range_end as its
 857  * zero_start.  If range_end is 0, there's nothing to do.
 858  *
 859  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 860  */
 861 static int ocfs2_zero_extend_get_range(struct inode *inode,
 862                                        struct buffer_head *di_bh,
 863                                        u64 zero_start, u64 zero_end,
 864                                        u64 *range_start, u64 *range_end)
 865 {
 866         int rc = 0, needs_cow = 0;
 867         u32 p_cpos, zero_clusters = 0;
 868         u32 zero_cpos =
 869                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 870         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 871         unsigned int num_clusters = 0;
 872         unsigned int ext_flags = 0;
 873
 874         while (zero_cpos < last_cpos) {
 875                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 876                                         &num_clusters, &ext_flags);
 877                 if (rc) {
 878                         mlog_errno(rc);
 879                         goto out;
 880                 }
 881
 882                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 883                         zero_clusters = num_clusters;
 884                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 885                                 needs_cow = 1;
 886                         break;
 887                 }
 888
 889                 zero_cpos += num_clusters;
 890         }
 891         if (!zero_clusters) {
 892                 *range_end = 0;
 893                 goto out;
 894         }
 895
 896         while ((zero_cpos + zero_clusters) < last_cpos) {
 897                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 898                                         &p_cpos, &num_clusters,
 899                                         &ext_flags);
 900                 if (rc) {
 901                         mlog_errno(rc);
 902                         goto out;
 903                 }
 904
 905                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 906                         break;
 907                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 908                         needs_cow = 1;
 909                 zero_clusters += num_clusters;
 910         }
 911         if ((zero_cpos + zero_clusters) > last_cpos)
 912                 zero_clusters = last_cpos - zero_cpos;
 913
 914         if (needs_cow) {
 915                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 916                                         zero_clusters, UINT_MAX);
 917                 if (rc) {
 918                         mlog_errno(rc);
 919                         goto out;
 920                 }
 921         }
 922
 923         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 924         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 925                                              zero_cpos + zero_clusters);
 926
 927 out:
 928         return rc;
 929 }
 930
 931 /*
 932  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 933  * has made sure that the entire range needs zeroing.
 934  */
 935 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 936                                    u64 range_end, struct buffer_head *di_bh)
 937 {
 938         int rc = 0;
 939         u64 next_pos;
 940         u64 zero_pos = range_start;
 941
 942         trace_ocfs2_zero_extend_range(
 943                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 944                         (unsigned long long)range_start,
 945                         (unsigned long long)range_end);
 946         BUG_ON(range_start >= range_end);
 947
 948         while (zero_pos < range_end) {
 949                 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 950                 if (next_pos > range_end)
 951                         next_pos = range_end;
 952                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 953                 if (rc < 0) {
 954                         mlog_errno(rc);
 955                         break;
 956                 }
 957                 zero_pos = next_pos;
 958
 959                 /*
 960                  * Very large extends have the potential to lock up
 961                  * the cpu for extended periods of time.
 962                  */
 963                 cond_resched();
 964         }
 965
 966         return rc;
 967 }
 968
 969 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 970                       loff_t zero_to_size)
 971 {
 972         int ret = 0;
 973         u64 zero_start, range_start = 0, range_end = 0;
 974         struct super_block *sb = inode->i_sb;
 975
 976         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 977         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 978                                 (unsigned long long)zero_start,
 979                                 (unsigned long long)i_size_read(inode));
 980         while (zero_start < zero_to_size) {
 981                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 982                                                   zero_to_size,
 983                                                   &range_start,
 984                                                   &range_end);
 985                 if (ret) {
 986                         mlog_errno(ret);
 987                         break;
 988                 }
 989                 if (!range_end)
 990                         break;
 991                 /* Trim the ends */
 992                 if (range_start < zero_start)
 993                         range_start = zero_start;
 994                 if (range_end > zero_to_size)
 995                         range_end = zero_to_size;
 996
 997                 ret = ocfs2_zero_extend_range(inode, range_start,
 998                                               range_end, di_bh);
 999                 if (ret) {
1000                         mlog_errno(ret);
1001                         break;
1002                 }
1003                 zero_start = range_end;
1004         }
1005
1006         return ret;
1007 }
1008
1009 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1010                           u64 new_i_size, u64 zero_to)
1011 {
1012         int ret;
1013         u32 clusters_to_add;
1014         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1015
1016         /*
1017          * Only quota files call this without a bh, and they can't be
1018          * refcounted.
1019          */
1020         BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1021         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1022
1023         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1024         if (clusters_to_add < oi->ip_clusters)
1025                 clusters_to_add = 0;
1026         else
1027                 clusters_to_add -= oi->ip_clusters;
1028
1029         if (clusters_to_add) {
1030                 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1031                                               clusters_to_add, 0);
1032                 if (ret) {
1033                         mlog_errno(ret);
1034                         goto out;
1035                 }
1036         }
1037
1038         /*
1039          * Call this even if we don't add any clusters to the tree. We
1040          * still need to zero the area between the old i_size and the
1041          * new i_size.
1042          */
1043         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1044         if (ret < 0)
1045                 mlog_errno(ret);
1046
1047 out:
1048         return ret;
1049 }
1050
1051 static int ocfs2_extend_file(struct inode *inode,
1052                              struct buffer_head *di_bh,
1053                              u64 new_i_size)
1054 {
1055         int ret = 0;
1056         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1057
1058         BUG_ON(!di_bh);
1059
1060         /* setattr sometimes calls us like this. */
1061         if (new_i_size == 0)
1062                 goto out;
1063
1064         if (i_size_read(inode) == new_i_size)
1065                 goto out;
1066         BUG_ON(new_i_size < i_size_read(inode));
1067
1068         /*
1069          * The alloc sem blocks people in read/write from reading our
1070          * allocation until we're done changing it. We depend on
1071          * i_mutex to block other extend/truncate calls while we're
1072          * here.  We even have to hold it for sparse files because there
1073          * might be some tail zeroing.
1074          */
1075         down_write(&oi->ip_alloc_sem);
1076
1077         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1078                 /*
1079                  * We can optimize small extends by keeping the inodes
1080                  * inline data.
1081                  */
1082                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1083                         up_write(&oi->ip_alloc_sem);
1084                         goto out_update_size;
1085                 }
1086
1087                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1088                 if (ret) {
1089                         up_write(&oi->ip_alloc_sem);
1090                         mlog_errno(ret);
1091                         goto out;
1092                 }
1093         }
1094
1095         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1096                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1097         else
1098                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1099                                             new_i_size);
1100
1101         up_write(&oi->ip_alloc_sem);
1102
1103         if (ret < 0) {
1104                 mlog_errno(ret);
1105                 goto out;
1106         }
1107
1108 out_update_size:
1109         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1110         if (ret < 0)
1111                 mlog_errno(ret);
1112
1113 out:
1114         return ret;
1115 }
1116
1117 int ocfs2_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
1118                   struct iattr *attr)
1119 {
1120         int status = 0, size_change;
1121         int inode_locked = 0;
1122         struct inode *inode = d_inode(dentry);
1123         struct super_block *sb = inode->i_sb;
1124         struct ocfs2_super *osb = OCFS2_SB(sb);
1125         struct buffer_head *bh = NULL;
1126         handle_t *handle = NULL;
1127         struct dquot *transfer_to[MAXQUOTAS] = { };
1128         int qtype;
1129         int had_lock;
1130         struct ocfs2_lock_holder oh;
1131
1132         trace_ocfs2_setattr(inode, dentry,
1133                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1134                             dentry->d_name.len, dentry->d_name.name,
1135                             attr->ia_valid, attr->ia_mode,
1136                             from_kuid(&init_user_ns, attr->ia_uid),
1137                             from_kgid(&init_user_ns, attr->ia_gid));
1138
1139         /* ensuring we don't even attempt to truncate a symlink */
1140         if (S_ISLNK(inode->i_mode))
1141                 attr->ia_valid &= ~ATTR_SIZE;
1142
1143 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1144                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1145         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1146                 return 0;
1147
1148         status = setattr_prepare(&init_user_ns, dentry, attr);
1149         if (status)
1150                 return status;
1151
1152         if (is_quota_modification(inode, attr)) {
1153                 status = dquot_initialize(inode);
1154                 if (status)
1155                         return status;
1156         }
1157         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1158         if (size_change) {
1159                 /*
1160                  * Here we should wait dio to finish before inode lock
1161                  * to avoid a deadlock between ocfs2_setattr() and
1162                  * ocfs2_dio_end_io_write()
1163                  */
1164                 inode_dio_wait(inode);
1165
1166                 status = ocfs2_rw_lock(inode, 1);
1167                 if (status < 0) {
1168                         mlog_errno(status);
1169                         goto bail;
1170                 }
1171         }
1172
1173         had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1174         if (had_lock < 0) {
1175                 status = had_lock;
1176                 goto bail_unlock_rw;
1177         } else if (had_lock) {
1178                 /*
1179                  * As far as we know, ocfs2_setattr() could only be the first
1180                  * VFS entry point in the call chain of recursive cluster
1181                  * locking issue.
1182                  *
1183                  * For instance:
1184                  * chmod_common()
1185                  *  notify_change()
1186                  *   ocfs2_setattr()
1187                  *    posix_acl_chmod()
1188                  *     ocfs2_iop_get_acl()
1189                  *
1190                  * But, we're not 100% sure if it's always true, because the
1191                  * ordering of the VFS entry points in the call chain is out
1192                  * of our control. So, we'd better dump the stack here to
1193                  * catch the other cases of recursive locking.
1194                  */
1195                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1196                 dump_stack();
1197         }
1198         inode_locked = 1;
1199
1200         if (size_change) {
1201                 status = inode_newsize_ok(inode, attr->ia_size);
1202                 if (status)
1203                         goto bail_unlock;
1204
1205                 if (i_size_read(inode) >= attr->ia_size) {
1206                         if (ocfs2_should_order_data(inode)) {
1207                                 status = ocfs2_begin_ordered_truncate(inode,
1208                                                                       attr->ia_size);
1209                                 if (status)
1210                                         goto bail_unlock;
1211                         }
1212                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1213                 } else
1214                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1215                 if (status < 0) {
1216                         if (status != -ENOSPC)
1217                                 mlog_errno(status);
1218                         status = -ENOSPC;
1219                         goto bail_unlock;
1220                 }
1221         }
1222
1223         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1224             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1225                 /*
1226                  * Gather pointers to quota structures so that allocation /
1227                  * freeing of quota structures happens here and not inside
1228                  * dquot_transfer() where we have problems with lock ordering
1229                  */
1230                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1231                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1232                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1233                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1234                         if (IS_ERR(transfer_to[USRQUOTA])) {
1235                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1236                                 transfer_to[USRQUOTA] = NULL;
1237                                 goto bail_unlock;
1238                         }
1239                 }
1240                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1241                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1242                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1243                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1244                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1245                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1246                                 transfer_to[GRPQUOTA] = NULL;
1247                                 goto bail_unlock;
1248                         }
1249                 }
1250                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1251                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1252                                            2 * ocfs2_quota_trans_credits(sb));
1253                 if (IS_ERR(handle)) {
1254                         status = PTR_ERR(handle);
1255                         mlog_errno(status);
1256                         goto bail_unlock_alloc;
1257                 }
1258                 status = __dquot_transfer(inode, transfer_to);
1259                 if (status < 0)
1260                         goto bail_commit;
1261         } else {
1262                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1263                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1264                 if (IS_ERR(handle)) {
1265                         status = PTR_ERR(handle);
1266                         mlog_errno(status);
1267                         goto bail_unlock_alloc;
1268                 }
1269         }
1270
1271         setattr_copy(&init_user_ns, inode, attr);
1272         mark_inode_dirty(inode);
1273
1274         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1275         if (status < 0)
1276                 mlog_errno(status);
1277
1278 bail_commit:
1279         ocfs2_commit_trans(osb, handle);
1280 bail_unlock_alloc:
1281         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1282 bail_unlock:
1283         if (status && inode_locked) {
1284                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1285                 inode_locked = 0;
1286         }
1287 bail_unlock_rw:
1288         if (size_change)
1289                 ocfs2_rw_unlock(inode, 1);
1290 bail:
1291
1292         /* Release quota pointers in case we acquired them */
1293         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1294                 dqput(transfer_to[qtype]);
1295
1296         if (!status && attr->ia_valid & ATTR_MODE) {
1297                 status = ocfs2_acl_chmod(inode, bh);
1298                 if (status < 0)
1299                         mlog_errno(status);
1300         }
1301         if (inode_locked)
1302                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1303
1304         brelse(bh);
1305         return status;
1306 }
1307
1308 int ocfs2_getattr(struct user_namespace *mnt_userns, const struct path *path,
1309                   struct kstat *stat, u32 request_mask, unsigned int flags)
1310 {
1311         struct inode *inode = d_inode(path->dentry);
1312         struct super_block *sb = path->dentry->d_sb;
1313         struct ocfs2_super *osb = sb->s_fs_info;
1314         int err;
1315
1316         err = ocfs2_inode_revalidate(path->dentry);
1317         if (err) {
1318                 if (err != -ENOENT)
1319                         mlog_errno(err);
1320                 goto bail;
1321         }
1322
1323         generic_fillattr(&init_user_ns, inode, stat);
1324         /*
1325          * If there is inline data in the inode, the inode will normally not
1326          * have data blocks allocated (it may have an external xattr block).
1327          * Report at least one sector for such files, so tools like tar, rsync,
1328          * others don't incorrectly think the file is completely sparse.
1329          */
1330         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1331                 stat->blocks += (stat->size + 511)>>9;
1332
1333         /* We set the blksize from the cluster size for performance */
1334         stat->blksize = osb->s_clustersize;
1335
1336 bail:
1337         return err;
1338 }
1339
1340 int ocfs2_permission(struct user_namespace *mnt_userns, struct inode *inode,
1341                      int mask)
1342 {
1343         int ret, had_lock;
1344         struct ocfs2_lock_holder oh;
1345
1346         if (mask & MAY_NOT_BLOCK)
1347                 return -ECHILD;
1348
1349         had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1350         if (had_lock < 0) {
1351                 ret = had_lock;
1352                 goto out;
1353         } else if (had_lock) {
1354                 /* See comments in ocfs2_setattr() for details.
1355                  * The call chain of this case could be:
1356                  * do_sys_open()
1357                  *  may_open()
1358                  *   inode_permission()
1359                  *    ocfs2_permission()
1360                  *     ocfs2_iop_get_acl()
1361                  */
1362                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1363                 dump_stack();
1364         }
1365
1366         ret = generic_permission(&init_user_ns, inode, mask);
1367
1368         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1369 out:
1370         return ret;
1371 }
1372
1373 static int __ocfs2_write_remove_suid(struct inode *inode,
1374                                      struct buffer_head *bh)
1375 {
1376         int ret;
1377         handle_t *handle;
1378         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1379         struct ocfs2_dinode *di;
1380
1381         trace_ocfs2_write_remove_suid(
1382                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1383                         inode->i_mode);
1384
1385         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1386         if (IS_ERR(handle)) {
1387                 ret = PTR_ERR(handle);
1388                 mlog_errno(ret);
1389                 goto out;
1390         }
1391
1392         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1393                                       OCFS2_JOURNAL_ACCESS_WRITE);
1394         if (ret < 0) {
1395                 mlog_errno(ret);
1396                 goto out_trans;
1397         }
1398
1399         inode->i_mode &= ~S_ISUID;
1400         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1401                 inode->i_mode &= ~S_ISGID;
1402
1403         di = (struct ocfs2_dinode *) bh->b_data;
1404         di->i_mode = cpu_to_le16(inode->i_mode);
1405         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1406
1407         ocfs2_journal_dirty(handle, bh);
1408
1409 out_trans:
1410         ocfs2_commit_trans(osb, handle);
1411 out:
1412         return ret;
1413 }
1414
1415 static int ocfs2_write_remove_suid(struct inode *inode)
1416 {
1417         int ret;
1418         struct buffer_head *bh = NULL;
1419
1420         ret = ocfs2_read_inode_block(inode, &bh);
1421         if (ret < 0) {
1422                 mlog_errno(ret);
1423                 goto out;
1424         }
1425
1426         ret =  __ocfs2_write_remove_suid(inode, bh);
1427 out:
1428         brelse(bh);
1429         return ret;
1430 }
1431
1432 /*
1433  * Allocate enough extents to cover the region starting at byte offset
1434  * start for len bytes. Existing extents are skipped, any extents
1435  * added are marked as "unwritten".
1436  */
1437 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1438                                             u64 start, u64 len)
1439 {
1440         int ret;
1441         u32 cpos, phys_cpos, clusters, alloc_size;
1442         u64 end = start + len;
1443         struct buffer_head *di_bh = NULL;
1444
1445         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1446                 ret = ocfs2_read_inode_block(inode, &di_bh);
1447                 if (ret) {
1448                         mlog_errno(ret);
1449                         goto out;
1450                 }
1451
1452                 /*
1453                  * Nothing to do if the requested reservation range
1454                  * fits within the inode.
1455                  */
1456                 if (ocfs2_size_fits_inline_data(di_bh, end))
1457                         goto out;
1458
1459                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1460                 if (ret) {
1461                         mlog_errno(ret);
1462                         goto out;
1463                 }
1464         }
1465
1466         /*
1467          * We consider both start and len to be inclusive.
1468          */
1469         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1470         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1471         clusters -= cpos;
1472
1473         while (clusters) {
1474                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1475                                          &alloc_size, NULL);
1476                 if (ret) {
1477                         mlog_errno(ret);
1478                         goto out;
1479                 }
1480
1481                 /*
1482                  * Hole or existing extent len can be arbitrary, so
1483                  * cap it to our own allocation request.
1484                  */
1485                 if (alloc_size > clusters)
1486                         alloc_size = clusters;
1487
1488                 if (phys_cpos) {
1489                         /*
1490                          * We already have an allocation at this
1491                          * region so we can safely skip it.
1492                          */
1493                         goto next;
1494                 }
1495
1496                 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1497                 if (ret) {
1498                         if (ret != -ENOSPC)
1499                                 mlog_errno(ret);
1500                         goto out;
1501                 }
1502
1503 next:
1504                 cpos += alloc_size;
1505                 clusters -= alloc_size;
1506         }
1507
1508         ret = 0;
1509 out:
1510
1511         brelse(di_bh);
1512         return ret;
1513 }
1514
1515 /*
1516  * Truncate a byte range, avoiding pages within partial clusters. This
1517  * preserves those pages for the zeroing code to write to.
1518  */
1519 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1520                                          u64 byte_len)
1521 {
1522         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1523         loff_t start, end;
1524         struct address_space *mapping = inode->i_mapping;
1525
1526         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1527         end = byte_start + byte_len;
1528         end = end & ~(osb->s_clustersize - 1);
1529
1530         if (start < end) {
1531                 unmap_mapping_range(mapping, start, end - start, 0);
1532                 truncate_inode_pages_range(mapping, start, end - 1);
1533         }
1534 }
1535
1536 /*
1537  * zero out partial blocks of one cluster.
1538  *
1539  * start: file offset where zero starts, will be made upper block aligned.
1540  * len: it will be trimmed to the end of current cluster if "start + len"
1541  *      is bigger than it.
1542  */
1543 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1544                                         u64 start, u64 len)
1545 {
1546         int ret;
1547         u64 start_block, end_block, nr_blocks;
1548         u64 p_block, offset;
1549         u32 cluster, p_cluster, nr_clusters;
1550         struct super_block *sb = inode->i_sb;
1551         u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1552
1553         if (start + len < end)
1554                 end = start + len;
1555
1556         start_block = ocfs2_blocks_for_bytes(sb, start);
1557         end_block = ocfs2_blocks_for_bytes(sb, end);
1558         nr_blocks = end_block - start_block;
1559         if (!nr_blocks)
1560                 return 0;
1561
1562         cluster = ocfs2_bytes_to_clusters(sb, start);
1563         ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1564                                 &nr_clusters, NULL);
1565         if (ret)
1566                 return ret;
1567         if (!p_cluster)
1568                 return 0;
1569
1570         offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1571         p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1572         return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1573 }
1574
1575 static int ocfs2_zero_partial_clusters(struct inode *inode,
1576                                        u64 start, u64 len)
1577 {
1578         int ret = 0;
1579         u64 tmpend = 0;
1580         u64 end = start + len;
1581         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1582         unsigned int csize = osb->s_clustersize;
1583         handle_t *handle;
1584         loff_t isize = i_size_read(inode);
1585
1586         /*
1587          * The "start" and "end" values are NOT necessarily part of
1588          * the range whose allocation is being deleted. Rather, this
1589          * is what the user passed in with the request. We must zero
1590          * partial clusters here. There's no need to worry about
1591          * physical allocation - the zeroing code knows to skip holes.
1592          */
1593         trace_ocfs2_zero_partial_clusters(
1594                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1595                 (unsigned long long)start, (unsigned long long)end);
1596
1597         /*
1598          * If both edges are on a cluster boundary then there's no
1599          * zeroing required as the region is part of the allocation to
1600          * be truncated.
1601          */
1602         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1603                 goto out;
1604
1605         /* No page cache for EOF blocks, issue zero out to disk. */
1606         if (end > isize) {
1607                 /*
1608                  * zeroout eof blocks in last cluster starting from
1609                  * "isize" even "start" > "isize" because it is
1610                  * complicated to zeroout just at "start" as "start"
1611                  * may be not aligned with block size, buffer write
1612                  * would be required to do that, but out of eof buffer
1613                  * write is not supported.
1614                  */
1615                 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1616                                         end - isize);
1617                 if (ret) {
1618                         mlog_errno(ret);
1619                         goto out;
1620                 }
1621                 if (start >= isize)
1622                         goto out;
1623                 end = isize;
1624         }
1625         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1626         if (IS_ERR(handle)) {
1627                 ret = PTR_ERR(handle);
1628                 mlog_errno(ret);
1629                 goto out;
1630         }
1631
1632         /*
1633          * If start is on a cluster boundary and end is somewhere in another
1634          * cluster, we have not COWed the cluster starting at start, unless
1635          * end is also within the same cluster. So, in this case, we skip this
1636          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1637          * to the next one.
1638          */
1639         if ((start & (csize - 1)) != 0) {
1640                 /*
1641                  * We want to get the byte offset of the end of the 1st
1642                  * cluster.
1643                  */
1644                 tmpend = (u64)osb->s_clustersize +
1645                         (start & ~(osb->s_clustersize - 1));
1646                 if (tmpend > end)
1647                         tmpend = end;
1648
1649                 trace_ocfs2_zero_partial_clusters_range1(
1650                         (unsigned long long)start,
1651                         (unsigned long long)tmpend);
1652
1653                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1654                                                     tmpend);
1655                 if (ret)
1656                         mlog_errno(ret);
1657         }
1658
1659         if (tmpend < end) {
1660                 /*
1661                  * This may make start and end equal, but the zeroing
1662                  * code will skip any work in that case so there's no
1663                  * need to catch it up here.
1664                  */
1665                 start = end & ~(osb->s_clustersize - 1);
1666
1667                 trace_ocfs2_zero_partial_clusters_range2(
1668                         (unsigned long long)start, (unsigned long long)end);
1669
1670                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1671                 if (ret)
1672                         mlog_errno(ret);
1673         }
1674         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1675
1676         ocfs2_commit_trans(osb, handle);
1677 out:
1678         return ret;
1679 }
1680
1681 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1682 {
1683         int i;
1684         struct ocfs2_extent_rec *rec = NULL;
1685
1686         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1687
1688                 rec = &el->l_recs[i];
1689
1690                 if (le32_to_cpu(rec->e_cpos) < pos)
1691                         break;
1692         }
1693
1694         return i;
1695 }
1696
1697 /*
1698  * Helper to calculate the punching pos and length in one run, we handle the
1699  * following three cases in order:
1700  *
1701  * - remove the entire record
1702  * - remove a partial record
1703  * - no record needs to be removed (hole-punching completed)
1704 */
1705 static void ocfs2_calc_trunc_pos(struct inode *inode,
1706                                  struct ocfs2_extent_list *el,
1707                                  struct ocfs2_extent_rec *rec,
1708                                  u32 trunc_start, u32 *trunc_cpos,
1709                                  u32 *trunc_len, u32 *trunc_end,
1710                                  u64 *blkno, int *done)
1711 {
1712         int ret = 0;
1713         u32 coff, range;
1714
1715         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1716
1717         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1718                 /*
1719                  * remove an entire extent record.
1720                  */
1721                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1722                 /*
1723                  * Skip holes if any.
1724                  */
1725                 if (range < *trunc_end)
1726                         *trunc_end = range;
1727                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1728                 *blkno = le64_to_cpu(rec->e_blkno);
1729                 *trunc_end = le32_to_cpu(rec->e_cpos);
1730         } else if (range > trunc_start) {
1731                 /*
1732                  * remove a partial extent record, which means we're
1733                  * removing the last extent record.
1734                  */
1735                 *trunc_cpos = trunc_start;
1736                 /*
1737                  * skip hole if any.
1738                  */
1739                 if (range < *trunc_end)
1740                         *trunc_end = range;
1741                 *trunc_len = *trunc_end - trunc_start;
1742                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1743                 *blkno = le64_to_cpu(rec->e_blkno) +
1744                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1745                 *trunc_end = trunc_start;
1746         } else {
1747                 /*
1748                  * It may have two following possibilities:
1749                  *
1750                  * - last record has been removed
1751                  * - trunc_start was within a hole
1752                  *
1753                  * both two cases mean the completion of hole punching.
1754                  */
1755                 ret = 1;
1756         }
1757
1758         *done = ret;
1759 }
1760
1761 int ocfs2_remove_inode_range(struct inode *inode,
1762                              struct buffer_head *di_bh, u64 byte_start,
1763                              u64 byte_len)
1764 {
1765         int ret = 0, flags = 0, done = 0, i;
1766         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1767         u32 cluster_in_el;
1768         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1769         struct ocfs2_cached_dealloc_ctxt dealloc;
1770         struct address_space *mapping = inode->i_mapping;
1771         struct ocfs2_extent_tree et;
1772         struct ocfs2_path *path = NULL;
1773         struct ocfs2_extent_list *el = NULL;
1774         struct ocfs2_extent_rec *rec = NULL;
1775         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1776         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1777
1778         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1779         ocfs2_init_dealloc_ctxt(&dealloc);
1780
1781         trace_ocfs2_remove_inode_range(
1782                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1783                         (unsigned long long)byte_start,
1784                         (unsigned long long)byte_len);
1785
1786         if (byte_len == 0)
1787                 return 0;
1788
1789         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1790                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1791                                             byte_start + byte_len, 0);
1792                 if (ret) {
1793                         mlog_errno(ret);
1794                         goto out;
1795                 }
1796                 /*
1797                  * There's no need to get fancy with the page cache
1798                  * truncate of an inline-data inode. We're talking
1799                  * about less than a page here, which will be cached
1800                  * in the dinode buffer anyway.
1801                  */
1802                 unmap_mapping_range(mapping, 0, 0, 0);
1803                 truncate_inode_pages(mapping, 0);
1804                 goto out;
1805         }
1806
1807         /*
1808          * For reflinks, we may need to CoW 2 clusters which might be
1809          * partially zero'd later, if hole's start and end offset were
1810          * within one cluster(means is not exactly aligned to clustersize).
1811          */
1812
1813         if (ocfs2_is_refcount_inode(inode)) {
1814                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1815                 if (ret) {
1816                         mlog_errno(ret);
1817                         goto out;
1818                 }
1819
1820                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1821                 if (ret) {
1822                         mlog_errno(ret);
1823                         goto out;
1824                 }
1825         }
1826
1827         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1828         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1829         cluster_in_el = trunc_end;
1830
1831         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1832         if (ret) {
1833                 mlog_errno(ret);
1834                 goto out;
1835         }
1836
1837         path = ocfs2_new_path_from_et(&et);
1838         if (!path) {
1839                 ret = -ENOMEM;
1840                 mlog_errno(ret);
1841                 goto out;
1842         }
1843
1844         while (trunc_end > trunc_start) {
1845
1846                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1847                                       cluster_in_el);
1848                 if (ret) {
1849                         mlog_errno(ret);
1850                         goto out;
1851                 }
1852
1853                 el = path_leaf_el(path);
1854
1855                 i = ocfs2_find_rec(el, trunc_end);
1856                 /*
1857                  * Need to go to previous extent block.
1858                  */
1859                 if (i < 0) {
1860                         if (path->p_tree_depth == 0)
1861                                 break;
1862
1863                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1864                                                             path,
1865                                                             &cluster_in_el);
1866                         if (ret) {
1867                                 mlog_errno(ret);
1868                                 goto out;
1869                         }
1870
1871                         /*
1872                          * We've reached the leftmost extent block,
1873                          * it's safe to leave.
1874                          */
1875                         if (cluster_in_el == 0)
1876                                 break;
1877
1878                         /*
1879                          * The 'pos' searched for previous extent block is
1880                          * always one cluster less than actual trunc_end.
1881                          */
1882                         trunc_end = cluster_in_el + 1;
1883
1884                         ocfs2_reinit_path(path, 1);
1885
1886                         continue;
1887
1888                 } else
1889                         rec = &el->l_recs[i];
1890
1891                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1892                                      &trunc_len, &trunc_end, &blkno, &done);
1893                 if (done)
1894                         break;
1895
1896                 flags = rec->e_flags;
1897                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1898
1899                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1900                                                phys_cpos, trunc_len, flags,
1901                                                &dealloc, refcount_loc, false);
1902                 if (ret < 0) {
1903                         mlog_errno(ret);
1904                         goto out;
1905                 }
1906
1907                 cluster_in_el = trunc_end;
1908
1909                 ocfs2_reinit_path(path, 1);
1910         }
1911
1912         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1913
1914 out:
1915         ocfs2_free_path(path);
1916         ocfs2_schedule_truncate_log_flush(osb, 1);
1917         ocfs2_run_deallocs(osb, &dealloc);
1918
1919         return ret;
1920 }
1921
1922 /*
1923  * Parts of this function taken from xfs_change_file_space()
1924  */
1925 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1926                                      loff_t f_pos, unsigned int cmd,
1927                                      struct ocfs2_space_resv *sr,
1928                                      int change_size)
1929 {
1930         int ret;
1931         s64 llen;
1932         loff_t size, orig_isize;
1933         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1934         struct buffer_head *di_bh = NULL;
1935         handle_t *handle;
1936         unsigned long long max_off = inode->i_sb->s_maxbytes;
1937
1938         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1939                 return -EROFS;
1940
1941         inode_lock(inode);
1942
1943         /*
1944          * This prevents concurrent writes on other nodes
1945          */
1946         ret = ocfs2_rw_lock(inode, 1);
1947         if (ret) {
1948                 mlog_errno(ret);
1949                 goto out;
1950         }
1951
1952         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1953         if (ret) {
1954                 mlog_errno(ret);
1955                 goto out_rw_unlock;
1956         }
1957
1958         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1959                 ret = -EPERM;
1960                 goto out_inode_unlock;
1961         }
1962
1963         switch (sr->l_whence) {
1964         case 0: /*SEEK_SET*/
1965                 break;
1966         case 1: /*SEEK_CUR*/
1967                 sr->l_start += f_pos;
1968                 break;
1969         case 2: /*SEEK_END*/
1970                 sr->l_start += i_size_read(inode);
1971                 break;
1972         default:
1973                 ret = -EINVAL;
1974                 goto out_inode_unlock;
1975         }
1976         sr->l_whence = 0;
1977
1978         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1979
1980         if (sr->l_start < 0
1981             || sr->l_start > max_off
1982             || (sr->l_start + llen) < 0
1983             || (sr->l_start + llen) > max_off) {
1984                 ret = -EINVAL;
1985                 goto out_inode_unlock;
1986         }
1987         size = sr->l_start + sr->l_len;
1988
1989         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1990             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1991                 if (sr->l_len <= 0) {
1992                         ret = -EINVAL;
1993                         goto out_inode_unlock;
1994                 }
1995         }
1996
1997         if (file && should_remove_suid(file->f_path.dentry)) {
1998                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1999                 if (ret) {
2000                         mlog_errno(ret);
2001                         goto out_inode_unlock;
2002                 }
2003         }
2004
2005         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2006         switch (cmd) {
2007         case OCFS2_IOC_RESVSP:
2008         case OCFS2_IOC_RESVSP64:
2009                 /*
2010                  * This takes unsigned offsets, but the signed ones we
2011                  * pass have been checked against overflow above.
2012                  */
2013                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2014                                                        sr->l_len);
2015                 break;
2016         case OCFS2_IOC_UNRESVSP:
2017         case OCFS2_IOC_UNRESVSP64:
2018                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2019                                                sr->l_len);
2020                 break;
2021         default:
2022                 ret = -EINVAL;
2023         }
2024
2025         orig_isize = i_size_read(inode);
2026         /* zeroout eof blocks in the cluster. */
2027         if (!ret && change_size && orig_isize < size) {
2028                 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2029                                         size - orig_isize);
2030                 if (!ret)
2031                         i_size_write(inode, size);
2032         }
2033         up_write(&OCFS2_I(inode)->ip_alloc_sem);
2034         if (ret) {
2035                 mlog_errno(ret);
2036                 goto out_inode_unlock;
2037         }
2038
2039         /*
2040          * We update c/mtime for these changes
2041          */
2042         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2043         if (IS_ERR(handle)) {
2044                 ret = PTR_ERR(handle);
2045                 mlog_errno(ret);
2046                 goto out_inode_unlock;
2047         }
2048
2049         inode->i_ctime = inode->i_mtime = current_time(inode);
2050         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2051         if (ret < 0)
2052                 mlog_errno(ret);
2053
2054         if (file && (file->f_flags & O_SYNC))
2055                 handle->h_sync = 1;
2056
2057         ocfs2_commit_trans(osb, handle);
2058
2059 out_inode_unlock:
2060         brelse(di_bh);
2061         ocfs2_inode_unlock(inode, 1);
2062 out_rw_unlock:
2063         ocfs2_rw_unlock(inode, 1);
2064
2065 out:
2066         inode_unlock(inode);
2067         return ret;
2068 }
2069
2070 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2071                             struct ocfs2_space_resv *sr)
2072 {
2073         struct inode *inode = file_inode(file);
2074         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2075         int ret;
2076
2077         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2078             !ocfs2_writes_unwritten_extents(osb))
2079                 return -ENOTTY;
2080         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2081                  !ocfs2_sparse_alloc(osb))
2082                 return -ENOTTY;
2083
2084         if (!S_ISREG(inode->i_mode))
2085                 return -EINVAL;
2086
2087         if (!(file->f_mode & FMODE_WRITE))
2088                 return -EBADF;
2089
2090         ret = mnt_want_write_file(file);
2091         if (ret)
2092                 return ret;
2093         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2094         mnt_drop_write_file(file);
2095         return ret;
2096 }
2097
2098 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2099                             loff_t len)
2100 {
2101         struct inode *inode = file_inode(file);
2102         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2103         struct ocfs2_space_resv sr;
2104         int change_size = 1;
2105         int cmd = OCFS2_IOC_RESVSP64;
2106
2107         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2108                 return -EOPNOTSUPP;
2109         if (!ocfs2_writes_unwritten_extents(osb))
2110                 return -EOPNOTSUPP;
2111
2112         if (mode & FALLOC_FL_KEEP_SIZE)
2113                 change_size = 0;
2114
2115         if (mode & FALLOC_FL_PUNCH_HOLE)
2116                 cmd = OCFS2_IOC_UNRESVSP64;
2117
2118         sr.l_whence = 0;
2119         sr.l_start = (s64)offset;
2120         sr.l_len = (s64)len;
2121
2122         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2123                                          change_size);
2124 }
2125
2126 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2127                                    size_t count)
2128 {
2129         int ret = 0;
2130         unsigned int extent_flags;
2131         u32 cpos, clusters, extent_len, phys_cpos;
2132         struct super_block *sb = inode->i_sb;
2133
2134         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2135             !ocfs2_is_refcount_inode(inode) ||
2136             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2137                 return 0;
2138
2139         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2140         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2141
2142         while (clusters) {
2143                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2144                                          &extent_flags);
2145                 if (ret < 0) {
2146                         mlog_errno(ret);
2147                         goto out;
2148                 }
2149
2150                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2151                         ret = 1;
2152                         break;
2153                 }
2154
2155                 if (extent_len > clusters)
2156                         extent_len = clusters;
2157
2158                 clusters -= extent_len;
2159                 cpos += extent_len;
2160         }
2161 out:
2162         return ret;
2163 }
2164
2165 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2166 {
2167         int blockmask = inode->i_sb->s_blocksize - 1;
2168         loff_t final_size = pos + count;
2169
2170         if ((pos & blockmask) || (final_size & blockmask))
2171                 return 1;
2172         return 0;
2173 }
2174
2175 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2176                                             struct buffer_head **di_bh,
2177                                             int meta_level,
2178                                             int write_sem,
2179                                             int wait)
2180 {
2181         int ret = 0;
2182
2183         if (wait)
2184                 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2185         else
2186                 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2187         if (ret < 0)
2188                 goto out;
2189
2190         if (wait) {
2191                 if (write_sem)
2192                         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2193                 else
2194                         down_read(&OCFS2_I(inode)->ip_alloc_sem);
2195         } else {
2196                 if (write_sem)
2197                         ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2198                 else
2199                         ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2200
2201                 if (!ret) {
2202                         ret = -EAGAIN;
2203                         goto out_unlock;
2204                 }
2205         }
2206
2207         return ret;
2208
2209 out_unlock:
2210         brelse(*di_bh);
2211         *di_bh = NULL;
2212         ocfs2_inode_unlock(inode, meta_level);
2213 out:
2214         return ret;
2215 }
2216
2217 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2218                                                struct buffer_head **di_bh,
2219                                                int meta_level,
2220                                                int write_sem)
2221 {
2222         if (write_sem)
2223                 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2224         else
2225                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2226
2227         brelse(*di_bh);
2228         *di_bh = NULL;
2229
2230         if (meta_level >= 0)
2231                 ocfs2_inode_unlock(inode, meta_level);
2232 }
2233
2234 static int ocfs2_prepare_inode_for_write(struct file *file,
2235                                          loff_t pos, size_t count, int wait)
2236 {
2237         int ret = 0, meta_level = 0, overwrite_io = 0;
2238         int write_sem = 0;
2239         struct dentry *dentry = file->f_path.dentry;
2240         struct inode *inode = d_inode(dentry);
2241         struct buffer_head *di_bh = NULL;
2242         u32 cpos;
2243         u32 clusters;
2244
2245         /*
2246          * We start with a read level meta lock and only jump to an ex
2247          * if we need to make modifications here.
2248          */
2249         for(;;) {
2250                 ret = ocfs2_inode_lock_for_extent_tree(inode,
2251                                                        &di_bh,
2252                                                        meta_level,
2253                                                        write_sem,
2254                                                        wait);
2255                 if (ret < 0) {
2256                         if (ret != -EAGAIN)
2257                                 mlog_errno(ret);
2258                         goto out;
2259                 }
2260
2261                 /*
2262                  * Check if IO will overwrite allocated blocks in case
2263                  * IOCB_NOWAIT flag is set.
2264                  */
2265                 if (!wait && !overwrite_io) {
2266                         overwrite_io = 1;
2267
2268                         ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2269                         if (ret < 0) {
2270                                 if (ret != -EAGAIN)
2271                                         mlog_errno(ret);
2272                                 goto out_unlock;
2273                         }
2274                 }
2275
2276                 /* Clear suid / sgid if necessary. We do this here
2277                  * instead of later in the write path because
2278                  * remove_suid() calls ->setattr without any hint that
2279                  * we may have already done our cluster locking. Since
2280                  * ocfs2_setattr() *must* take cluster locks to
2281                  * proceed, this will lead us to recursively lock the
2282                  * inode. There's also the dinode i_size state which
2283                  * can be lost via setattr during extending writes (we
2284                  * set inode->i_size at the end of a write. */
2285                 if (should_remove_suid(dentry)) {
2286                         if (meta_level == 0) {
2287                                 ocfs2_inode_unlock_for_extent_tree(inode,
2288                                                                    &di_bh,
2289                                                                    meta_level,
2290                                                                    write_sem);
2291                                 meta_level = 1;
2292                                 continue;
2293                         }
2294
2295                         ret = ocfs2_write_remove_suid(inode);
2296                         if (ret < 0) {
2297                                 mlog_errno(ret);
2298                                 goto out_unlock;
2299                         }
2300                 }
2301
2302                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2303                 if (ret == 1) {
2304                         ocfs2_inode_unlock_for_extent_tree(inode,
2305                                                            &di_bh,
2306                                                            meta_level,
2307                                                            write_sem);
2308                         meta_level = 1;
2309                         write_sem = 1;
2310                         ret = ocfs2_inode_lock_for_extent_tree(inode,
2311                                                                &di_bh,
2312                                                                meta_level,
2313                                                                write_sem,
2314                                                                wait);
2315                         if (ret < 0) {
2316                                 if (ret != -EAGAIN)
2317                                         mlog_errno(ret);
2318                                 goto out;
2319                         }
2320
2321                         cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2322                         clusters =
2323                                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2324                         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2325                 }
2326
2327                 if (ret < 0) {
2328                         if (ret != -EAGAIN)
2329                                 mlog_errno(ret);
2330                         goto out_unlock;
2331                 }
2332
2333                 break;
2334         }
2335
2336 out_unlock:
2337         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2338                                             pos, count, wait);
2339
2340         ocfs2_inode_unlock_for_extent_tree(inode,
2341                                            &di_bh,
2342                                            meta_level,
2343                                            write_sem);
2344
2345 out:
2346         return ret;
2347 }
2348
2349 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2350                                     struct iov_iter *from)
2351 {
2352         int rw_level;
2353         ssize_t written = 0;
2354         ssize_t ret;
2355         size_t count = iov_iter_count(from);
2356         struct file *file = iocb->ki_filp;
2357         struct inode *inode = file_inode(file);
2358         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2359         int full_coherency = !(osb->s_mount_opt &
2360                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2361         void *saved_ki_complete = NULL;
2362         int append_write = ((iocb->ki_pos + count) >=
2363                         i_size_read(inode) ? 1 : 0);
2364         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2365         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2366
2367         trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2368                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2369                 file->f_path.dentry->d_name.len,
2370                 file->f_path.dentry->d_name.name,
2371                 (unsigned int)from->nr_segs);   /* GRRRRR */
2372
2373         if (!direct_io && nowait)
2374                 return -EOPNOTSUPP;
2375
2376         if (count == 0)
2377                 return 0;
2378
2379         if (nowait) {
2380                 if (!inode_trylock(inode))
2381                         return -EAGAIN;
2382         } else
2383                 inode_lock(inode);
2384
2385         /*
2386          * Concurrent O_DIRECT writes are allowed with
2387          * mount_option "coherency=buffered".
2388          * For append write, we must take rw EX.
2389          */
2390         rw_level = (!direct_io || full_coherency || append_write);
2391
2392         if (nowait)
2393                 ret = ocfs2_try_rw_lock(inode, rw_level);
2394         else
2395                 ret = ocfs2_rw_lock(inode, rw_level);
2396         if (ret < 0) {
2397                 if (ret != -EAGAIN)
2398                         mlog_errno(ret);
2399                 goto out_mutex;
2400         }
2401
2402         /*
2403          * O_DIRECT writes with "coherency=full" need to take EX cluster
2404          * inode_lock to guarantee coherency.
2405          */
2406         if (direct_io && full_coherency) {
2407                 /*
2408                  * We need to take and drop the inode lock to force
2409                  * other nodes to drop their caches.  Buffered I/O
2410                  * already does this in write_begin().
2411                  */
2412                 if (nowait)
2413                         ret = ocfs2_try_inode_lock(inode, NULL, 1);
2414                 else
2415                         ret = ocfs2_inode_lock(inode, NULL, 1);
2416                 if (ret < 0) {
2417                         if (ret != -EAGAIN)
2418                                 mlog_errno(ret);
2419                         goto out;
2420                 }
2421
2422                 ocfs2_inode_unlock(inode, 1);
2423         }
2424
2425         ret = generic_write_checks(iocb, from);
2426         if (ret <= 0) {
2427                 if (ret)
2428                         mlog_errno(ret);
2429                 goto out;
2430         }
2431         count = ret;
2432
2433         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2434         if (ret < 0) {
2435                 if (ret != -EAGAIN)
2436                         mlog_errno(ret);
2437                 goto out;
2438         }
2439
2440         if (direct_io && !is_sync_kiocb(iocb) &&
2441             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2442                 /*
2443                  * Make it a sync io if it's an unaligned aio.
2444                  */
2445                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2446         }
2447
2448         /* communicate with ocfs2_dio_end_io */
2449         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2450
2451         written = __generic_file_write_iter(iocb, from);
2452         /* buffered aio wouldn't have proper lock coverage today */
2453         BUG_ON(written == -EIOCBQUEUED && !direct_io);
2454
2455         /*
2456          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2457          * function pointer which is called when o_direct io completes so that
2458          * it can unlock our rw lock.
2459          * Unfortunately there are error cases which call end_io and others
2460          * that don't.  so we don't have to unlock the rw_lock if either an
2461          * async dio is going to do it in the future or an end_io after an
2462          * error has already done it.
2463          */
2464         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2465                 rw_level = -1;
2466         }
2467
2468         if (unlikely(written <= 0))
2469                 goto out;
2470
2471         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2472             IS_SYNC(inode)) {
2473                 ret = filemap_fdatawrite_range(file->f_mapping,
2474                                                iocb->ki_pos - written,
2475                                                iocb->ki_pos - 1);
2476                 if (ret < 0)
2477                         written = ret;
2478
2479                 if (!ret) {
2480                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2481                         if (ret < 0)
2482                                 written = ret;
2483                 }
2484
2485                 if (!ret)
2486                         ret = filemap_fdatawait_range(file->f_mapping,
2487                                                       iocb->ki_pos - written,
2488                                                       iocb->ki_pos - 1);
2489         }
2490
2491 out:
2492         if (saved_ki_complete)
2493                 xchg(&iocb->ki_complete, saved_ki_complete);
2494
2495         if (rw_level != -1)
2496                 ocfs2_rw_unlock(inode, rw_level);
2497
2498 out_mutex:
2499         inode_unlock(inode);
2500
2501         if (written)
2502                 ret = written;
2503         return ret;
2504 }
2505
2506 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2507                                    struct iov_iter *to)
2508 {
2509         int ret = 0, rw_level = -1, lock_level = 0;
2510         struct file *filp = iocb->ki_filp;
2511         struct inode *inode = file_inode(filp);
2512         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2513         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2514
2515         trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2516                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2517                         filp->f_path.dentry->d_name.len,
2518                         filp->f_path.dentry->d_name.name,
2519                         to->nr_segs);   /* GRRRRR */
2520
2521
2522         if (!inode) {
2523                 ret = -EINVAL;
2524                 mlog_errno(ret);
2525                 goto bail;
2526         }
2527
2528         if (!direct_io && nowait)
2529                 return -EOPNOTSUPP;
2530
2531         /*
2532          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2533          * need locks to protect pending reads from racing with truncate.
2534          */
2535         if (direct_io) {
2536                 if (nowait)
2537                         ret = ocfs2_try_rw_lock(inode, 0);
2538                 else
2539                         ret = ocfs2_rw_lock(inode, 0);
2540
2541                 if (ret < 0) {
2542                         if (ret != -EAGAIN)
2543                                 mlog_errno(ret);
2544                         goto bail;
2545                 }
2546                 rw_level = 0;
2547                 /* communicate with ocfs2_dio_end_io */
2548                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2549         }
2550
2551         /*
2552          * We're fine letting folks race truncates and extending
2553          * writes with read across the cluster, just like they can
2554          * locally. Hence no rw_lock during read.
2555          *
2556          * Take and drop the meta data lock to update inode fields
2557          * like i_size. This allows the checks down below
2558          * generic_file_read_iter() a chance of actually working.
2559          */
2560         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2561                                      !nowait);
2562         if (ret < 0) {
2563                 if (ret != -EAGAIN)
2564                         mlog_errno(ret);
2565                 goto bail;
2566         }
2567         ocfs2_inode_unlock(inode, lock_level);
2568
2569         ret = generic_file_read_iter(iocb, to);
2570         trace_generic_file_read_iter_ret(ret);
2571
2572         /* buffered aio wouldn't have proper lock coverage today */
2573         BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2574
2575         /* see ocfs2_file_write_iter */
2576         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2577                 rw_level = -1;
2578         }
2579
2580 bail:
2581         if (rw_level != -1)
2582                 ocfs2_rw_unlock(inode, rw_level);
2583
2584         return ret;
2585 }
2586
2587 /* Refer generic_file_llseek_unlocked() */
2588 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2589 {
2590         struct inode *inode = file->f_mapping->host;
2591         int ret = 0;
2592
2593         inode_lock(inode);
2594
2595         switch (whence) {
2596         case SEEK_SET:
2597                 break;
2598         case SEEK_END:
2599                 /* SEEK_END requires the OCFS2 inode lock for the file
2600                  * because it references the file's size.
2601                  */
2602                 ret = ocfs2_inode_lock(inode, NULL, 0);
2603                 if (ret < 0) {
2604                         mlog_errno(ret);
2605                         goto out;
2606                 }
2607                 offset += i_size_read(inode);
2608                 ocfs2_inode_unlock(inode, 0);
2609                 break;
2610         case SEEK_CUR:
2611                 if (offset == 0) {
2612                         offset = file->f_pos;
2613                         goto out;
2614                 }
2615                 offset += file->f_pos;
2616                 break;
2617         case SEEK_DATA:
2618         case SEEK_HOLE:
2619                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2620                 if (ret)
2621                         goto out;
2622                 break;
2623         default:
2624                 ret = -EINVAL;
2625                 goto out;
2626         }
2627
2628         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2629
2630 out:
2631         inode_unlock(inode);
2632         if (ret)
2633                 return ret;
2634         return offset;
2635 }
2636
2637 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2638                                      struct file *file_out, loff_t pos_out,
2639                                      loff_t len, unsigned int remap_flags)
2640 {
2641         struct inode *inode_in = file_inode(file_in);
2642         struct inode *inode_out = file_inode(file_out);
2643         struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2644         struct buffer_head *in_bh = NULL, *out_bh = NULL;
2645         bool same_inode = (inode_in == inode_out);
2646         loff_t remapped = 0;
2647         ssize_t ret;
2648
2649         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2650                 return -EINVAL;
2651         if (!ocfs2_refcount_tree(osb))
2652                 return -EOPNOTSUPP;
2653         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2654                 return -EROFS;
2655
2656         /* Lock both files against IO */
2657         ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2658         if (ret)
2659                 return ret;
2660
2661         /* Check file eligibility and prepare for block sharing. */
2662         ret = -EINVAL;
2663         if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2664             (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2665                 goto out_unlock;
2666
2667         ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2668                         &len, remap_flags);
2669         if (ret < 0 || len == 0)
2670                 goto out_unlock;
2671
2672         /* Lock out changes to the allocation maps and remap. */
2673         down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2674         if (!same_inode)
2675                 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2676                                   SINGLE_DEPTH_NESTING);
2677
2678         /* Zap any page cache for the destination file's range. */
2679         truncate_inode_pages_range(&inode_out->i_data,
2680                                    round_down(pos_out, PAGE_SIZE),
2681                                    round_up(pos_out + len, PAGE_SIZE) - 1);
2682
2683         remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2684                         inode_out, out_bh, pos_out, len);
2685         up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2686         if (!same_inode)
2687                 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2688         if (remapped < 0) {
2689                 ret = remapped;
2690                 mlog_errno(ret);
2691                 goto out_unlock;
2692         }
2693
2694         /*
2695          * Empty the extent map so that we may get the right extent
2696          * record from the disk.
2697          */
2698         ocfs2_extent_map_trunc(inode_in, 0);
2699         ocfs2_extent_map_trunc(inode_out, 0);
2700
2701         ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2702         if (ret) {
2703                 mlog_errno(ret);
2704                 goto out_unlock;
2705         }
2706
2707 out_unlock:
2708         ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2709         return remapped > 0 ? remapped : ret;
2710 }
2711
2712 const struct inode_operations ocfs2_file_iops = {
2713         .setattr        = ocfs2_setattr,
2714         .getattr        = ocfs2_getattr,
2715         .permission     = ocfs2_permission,
2716         .listxattr      = ocfs2_listxattr,
2717         .fiemap         = ocfs2_fiemap,
2718         .get_acl        = ocfs2_iop_get_acl,
2719         .set_acl        = ocfs2_iop_set_acl,
2720         .fileattr_get   = ocfs2_fileattr_get,
2721         .fileattr_set   = ocfs2_fileattr_set,
2722 };
2723
2724 const struct inode_operations ocfs2_special_file_iops = {
2725         .setattr        = ocfs2_setattr,
2726         .getattr        = ocfs2_getattr,
2727         .permission     = ocfs2_permission,
2728         .get_acl        = ocfs2_iop_get_acl,
2729         .set_acl        = ocfs2_iop_set_acl,
2730 };
2731
2732 /*
2733  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2734  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2735  */
2736 const struct file_operations ocfs2_fops = {
2737         .llseek         = ocfs2_file_llseek,
2738         .mmap           = ocfs2_mmap,
2739         .fsync          = ocfs2_sync_file,
2740         .release        = ocfs2_file_release,
2741         .open           = ocfs2_file_open,
2742         .read_iter      = ocfs2_file_read_iter,
2743         .write_iter     = ocfs2_file_write_iter,
2744         .unlocked_ioctl = ocfs2_ioctl,
2745 #ifdef CONFIG_COMPAT
2746         .compat_ioctl   = ocfs2_compat_ioctl,
2747 #endif
2748         .lock           = ocfs2_lock,
2749         .flock          = ocfs2_flock,
2750         .splice_read    = generic_file_splice_read,
2751         .splice_write   = iter_file_splice_write,
2752         .fallocate      = ocfs2_fallocate,
2753         .remap_file_range = ocfs2_remap_file_range,
2754 };
2755
2756 const struct file_operations ocfs2_dops = {
2757         .llseek         = generic_file_llseek,
2758         .read           = generic_read_dir,
2759         .iterate        = ocfs2_readdir,
2760         .fsync          = ocfs2_sync_file,
2761         .release        = ocfs2_dir_release,
2762         .open           = ocfs2_dir_open,
2763         .unlocked_ioctl = ocfs2_ioctl,
2764 #ifdef CONFIG_COMPAT
2765         .compat_ioctl   = ocfs2_compat_ioctl,
2766 #endif
2767         .lock           = ocfs2_lock,
2768         .flock          = ocfs2_flock,
2769 };
2770
2771 /*
2772  * POSIX-lockless variants of our file_operations.
2773  *
2774  * These will be used if the underlying cluster stack does not support
2775  * posix file locking, if the user passes the "localflocks" mount
2776  * option, or if we have a local-only fs.
2777  *
2778  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2779  * so we still want it in the case of no stack support for
2780  * plocks. Internally, it will do the right thing when asked to ignore
2781  * the cluster.
2782  */
2783 const struct file_operations ocfs2_fops_no_plocks = {
2784         .llseek         = ocfs2_file_llseek,
2785         .mmap           = ocfs2_mmap,
2786         .fsync          = ocfs2_sync_file,
2787         .release        = ocfs2_file_release,
2788         .open           = ocfs2_file_open,
2789         .read_iter      = ocfs2_file_read_iter,
2790         .write_iter     = ocfs2_file_write_iter,
2791         .unlocked_ioctl = ocfs2_ioctl,
2792 #ifdef CONFIG_COMPAT
2793         .compat_ioctl   = ocfs2_compat_ioctl,
2794 #endif
2795         .flock          = ocfs2_flock,
2796         .splice_read    = generic_file_splice_read,
2797         .splice_write   = iter_file_splice_write,
2798         .fallocate      = ocfs2_fallocate,
2799         .remap_file_range = ocfs2_remap_file_range,
2800 };
2801
2802 const struct file_operations ocfs2_dops_no_plocks = {
2803         .llseek         = generic_file_llseek,
2804         .read           = generic_read_dir,
2805         .iterate        = ocfs2_readdir,
2806         .fsync          = ocfs2_sync_file,
2807         .release        = ocfs2_dir_release,
2808         .open           = ocfs2_dir_open,
2809         .unlocked_ioctl = ocfs2_ioctl,
2810 #ifdef CONFIG_COMPAT
2811         .compat_ioctl   = ocfs2_compat_ioctl,
2812 #endif
2813         .flock          = ocfs2_flock,
2814 };