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