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