fs/ocfs2/file.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * file.c
   6  *
   7  * File open, close, extend, truncate
   8  *
   9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
  10  */
  11
  12 #include <linux/capability.h>
  13 #include <linux/fs.h>
  14 #include <linux/types.h>
  15 #include <linux/slab.h>
  16 #include <linux/highmem.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/uio.h>
  19 #include <linux/sched.h>
  20 #include <linux/splice.h>
  21 #include <linux/mount.h>
  22 #include <linux/writeback.h>
  23 #include <linux/falloc.h>
  24 #include <linux/quotaops.h>
  25 #include <linux/blkdev.h>
  26 #include <linux/backing-dev.h>
  27
  28 #include <cluster/masklog.h>
  29
  30 #include "ocfs2.h"
  31
  32 #include "alloc.h"
  33 #include "aops.h"
  34 #include "dir.h"
  35 #include "dlmglue.h"
  36 #include "extent_map.h"
  37 #include "file.h"
  38 #include "sysfile.h"
  39 #include "inode.h"
  40 #include "ioctl.h"
  41 #include "journal.h"
  42 #include "locks.h"
  43 #include "mmap.h"
  44 #include "suballoc.h"
  45 #include "super.h"
  46 #include "xattr.h"
  47 #include "acl.h"
  48 #include "quota.h"
  49 #include "refcounttree.h"
  50 #include "ocfs2_trace.h"
  51
  52 #include "buffer_head_io.h"
  53
  54 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  55 {
  56         struct ocfs2_file_private *fp;
  57
  58         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  59         if (!fp)
  60                 return -ENOMEM;
  61
  62         fp->fp_file = file;
  63         mutex_init(&fp->fp_mutex);
  64         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  65         file->private_data = fp;
  66
  67         return 0;
  68 }
  69
  70 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  71 {
  72         struct ocfs2_file_private *fp = file->private_data;
  73         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  74
  75         if (fp) {
  76                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  77                 ocfs2_lock_res_free(&fp->fp_flock);
  78                 kfree(fp);
  79                 file->private_data = NULL;
  80         }
  81 }
  82
  83 static int ocfs2_file_open(struct inode *inode, struct file *file)
  84 {
  85         int status;
  86         int mode = file->f_flags;
  87         struct ocfs2_inode_info *oi = OCFS2_I(inode);
  88
  89         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
  90                               (unsigned long long)oi->ip_blkno,
  91                               file->f_path.dentry->d_name.len,
  92                               file->f_path.dentry->d_name.name, mode);
  93
  94         if (file->f_mode & FMODE_WRITE) {
  95                 status = dquot_initialize(inode);
  96                 if (status)
  97                         goto leave;
  98         }
  99
 100         spin_lock(&oi->ip_lock);
 101
 102         /* Check that the inode hasn't been wiped from disk by another
 103          * node. If it hasn't then we're safe as long as we hold the
 104          * spin lock until our increment of open count. */
 105         if (oi->ip_flags & OCFS2_INODE_DELETED) {
 106                 spin_unlock(&oi->ip_lock);
 107
 108                 status = -ENOENT;
 109                 goto leave;
 110         }
 111
 112         if (mode & O_DIRECT)
 113                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 114
 115         oi->ip_open_count++;
 116         spin_unlock(&oi->ip_lock);
 117
 118         status = ocfs2_init_file_private(inode, file);
 119         if (status) {
 120                 /*
 121                  * We want to set open count back if we're failing the
 122                  * open.
 123                  */
 124                 spin_lock(&oi->ip_lock);
 125                 oi->ip_open_count--;
 126                 spin_unlock(&oi->ip_lock);
 127         }
 128
 129         file->f_mode |= FMODE_NOWAIT;
 130
 131 leave:
 132         return status;
 133 }
 134
 135 static int ocfs2_file_release(struct inode *inode, struct file *file)
 136 {
 137         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 138
 139         spin_lock(&oi->ip_lock);
 140         if (!--oi->ip_open_count)
 141                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 142
 143         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 144                                  oi->ip_blkno,
 145                                  file->f_path.dentry->d_name.len,
 146                                  file->f_path.dentry->d_name.name,
 147                                  oi->ip_open_count);
 148         spin_unlock(&oi->ip_lock);
 149
 150         ocfs2_free_file_private(inode, file);
 151
 152         return 0;
 153 }
 154
 155 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 156 {
 157         return ocfs2_init_file_private(inode, file);
 158 }
 159
 160 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 161 {
 162         ocfs2_free_file_private(inode, file);
 163         return 0;
 164 }
 165
 166 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 167                            int datasync)
 168 {
 169         int err = 0;
 170         struct inode *inode = file->f_mapping->host;
 171         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 172         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 173         journal_t *journal = osb->journal->j_journal;
 174         int ret;
 175         tid_t commit_tid;
 176         bool needs_barrier = false;
 177
 178         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 179                               oi->ip_blkno,
 180                               file->f_path.dentry->d_name.len,
 181                               file->f_path.dentry->d_name.name,
 182                               (unsigned long long)datasync);
 183
 184         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 185                 return -EROFS;
 186
 187         err = file_write_and_wait_range(file, start, end);
 188         if (err)
 189                 return err;
 190
 191         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 192         if (journal->j_flags & JBD2_BARRIER &&
 193             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 194                 needs_barrier = true;
 195         err = jbd2_complete_transaction(journal, commit_tid);
 196         if (needs_barrier) {
 197                 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 198                 if (!err)
 199                         err = ret;
 200         }
 201
 202         if (err)
 203                 mlog_errno(err);
 204
 205         return (err < 0) ? -EIO : 0;
 206 }
 207
 208 int ocfs2_should_update_atime(struct inode *inode,
 209                               struct vfsmount *vfsmnt)
 210 {
 211         struct timespec64 now;
 212         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 213
 214         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 215                 return 0;
 216
 217         if ((inode->i_flags & S_NOATIME) ||
 218             ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
 219                 return 0;
 220
 221         /*
 222          * We can be called with no vfsmnt structure - NFSD will
 223          * sometimes do this.
 224          *
 225          * Note that our action here is different than touch_atime() -
 226          * if we can't tell whether this is a noatime mount, then we
 227          * don't know whether to trust the value of s_atime_quantum.
 228          */
 229         if (vfsmnt == NULL)
 230                 return 0;
 231
 232         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 233             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 234                 return 0;
 235
 236         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 237                 if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 238                     (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 239                         return 1;
 240
 241                 return 0;
 242         }
 243
 244         now = current_time(inode);
 245         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 246                 return 0;
 247         else
 248                 return 1;
 249 }
 250
 251 int ocfs2_update_inode_atime(struct inode *inode,
 252                              struct buffer_head *bh)
 253 {
 254         int ret;
 255         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 256         handle_t *handle;
 257         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 258
 259         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 260         if (IS_ERR(handle)) {
 261                 ret = PTR_ERR(handle);
 262                 mlog_errno(ret);
 263                 goto out;
 264         }
 265
 266         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 267                                       OCFS2_JOURNAL_ACCESS_WRITE);
 268         if (ret) {
 269                 mlog_errno(ret);
 270                 goto out_commit;
 271         }
 272
 273         /*
 274          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 275          * have i_mutex to guard against concurrent changes to other
 276          * inode fields.
 277          */
 278         inode->i_atime = current_time(inode);
 279         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 280         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 281         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 282         ocfs2_journal_dirty(handle, bh);
 283
 284 out_commit:
 285         ocfs2_commit_trans(osb, handle);
 286 out:
 287         return ret;
 288 }
 289
 290 int ocfs2_set_inode_size(handle_t *handle,
 291                                 struct inode *inode,
 292                                 struct buffer_head *fe_bh,
 293                                 u64 new_i_size)
 294 {
 295         int status;
 296
 297         i_size_write(inode, new_i_size);
 298         inode->i_blocks = ocfs2_inode_sector_count(inode);
 299         inode->i_ctime = inode->i_mtime = current_time(inode);
 300
 301         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 302         if (status < 0) {
 303                 mlog_errno(status);
 304                 goto bail;
 305         }
 306
 307 bail:
 308         return status;
 309 }
 310
 311 int ocfs2_simple_size_update(struct inode *inode,
 312                              struct buffer_head *di_bh,
 313                              u64 new_i_size)
 314 {
 315         int ret;
 316         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 317         handle_t *handle = NULL;
 318
 319         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 320         if (IS_ERR(handle)) {
 321                 ret = PTR_ERR(handle);
 322                 mlog_errno(ret);
 323                 goto out;
 324         }
 325
 326         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 327                                    new_i_size);
 328         if (ret < 0)
 329                 mlog_errno(ret);
 330
 331         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 332         ocfs2_commit_trans(osb, handle);
 333 out:
 334         return ret;
 335 }
 336
 337 static int ocfs2_cow_file_pos(struct inode *inode,
 338                               struct buffer_head *fe_bh,
 339                               u64 offset)
 340 {
 341         int status;
 342         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 343         unsigned int num_clusters = 0;
 344         unsigned int ext_flags = 0;
 345
 346         /*
 347          * If the new offset is aligned to the range of the cluster, there is
 348          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 349          * CoW either.
 350          */
 351         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 352                 return 0;
 353
 354         status = ocfs2_get_clusters(inode, cpos, &phys,
 355                                     &num_clusters, &ext_flags);
 356         if (status) {
 357                 mlog_errno(status);
 358                 goto out;
 359         }
 360
 361         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 362                 goto out;
 363
 364         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 365
 366 out:
 367         return status;
 368 }
 369
 370 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 371                                      struct inode *inode,
 372                                      struct buffer_head *fe_bh,
 373                                      u64 new_i_size)
 374 {
 375         int status;
 376         handle_t *handle;
 377         struct ocfs2_dinode *di;
 378         u64 cluster_bytes;
 379
 380         /*
 381          * We need to CoW the cluster contains the offset if it is reflinked
 382          * since we will call ocfs2_zero_range_for_truncate later which will
 383          * write "0" from offset to the end of the cluster.
 384          */
 385         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 386         if (status) {
 387                 mlog_errno(status);
 388                 return status;
 389         }
 390
 391         /* TODO: This needs to actually orphan the inode in this
 392          * transaction. */
 393
 394         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 395         if (IS_ERR(handle)) {
 396                 status = PTR_ERR(handle);
 397                 mlog_errno(status);
 398                 goto out;
 399         }
 400
 401         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 402                                          OCFS2_JOURNAL_ACCESS_WRITE);
 403         if (status < 0) {
 404                 mlog_errno(status);
 405                 goto out_commit;
 406         }
 407
 408         /*
 409          * Do this before setting i_size.
 410          */
 411         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 412         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 413                                                cluster_bytes);
 414         if (status) {
 415                 mlog_errno(status);
 416                 goto out_commit;
 417         }
 418
 419         i_size_write(inode, new_i_size);
 420         inode->i_ctime = inode->i_mtime = current_time(inode);
 421
 422         di = (struct ocfs2_dinode *) fe_bh->b_data;
 423         di->i_size = cpu_to_le64(new_i_size);
 424         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 425         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 426         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 427
 428         ocfs2_journal_dirty(handle, fe_bh);
 429
 430 out_commit:
 431         ocfs2_commit_trans(osb, handle);
 432 out:
 433         return status;
 434 }
 435
 436 int ocfs2_truncate_file(struct inode *inode,
 437                                struct buffer_head *di_bh,
 438                                u64 new_i_size)
 439 {
 440         int status = 0;
 441         struct ocfs2_dinode *fe = NULL;
 442         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 443
 444         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 445          * already validated it */
 446         fe = (struct ocfs2_dinode *) di_bh->b_data;
 447
 448         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 449                                   (unsigned long long)le64_to_cpu(fe->i_size),
 450                                   (unsigned long long)new_i_size);
 451
 452         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 453                         "Inode %llu, inode i_size = %lld != di "
 454                         "i_size = %llu, i_flags = 0x%x\n",
 455                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 456                         i_size_read(inode),
 457                         (unsigned long long)le64_to_cpu(fe->i_size),
 458                         le32_to_cpu(fe->i_flags));
 459
 460         if (new_i_size > le64_to_cpu(fe->i_size)) {
 461                 trace_ocfs2_truncate_file_error(
 462                         (unsigned long long)le64_to_cpu(fe->i_size),
 463                         (unsigned long long)new_i_size);
 464                 status = -EINVAL;
 465                 mlog_errno(status);
 466                 goto bail;
 467         }
 468
 469         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 470
 471         ocfs2_resv_discard(&osb->osb_la_resmap,
 472                            &OCFS2_I(inode)->ip_la_data_resv);
 473
 474         /*
 475          * The inode lock forced other nodes to sync and drop their
 476          * pages, which (correctly) happens even if we have a truncate
 477          * without allocation change - ocfs2 cluster sizes can be much
 478          * greater than page size, so we have to truncate them
 479          * anyway.
 480          */
 481
 482         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 483                 unmap_mapping_range(inode->i_mapping,
 484                                     new_i_size + PAGE_SIZE - 1, 0, 1);
 485                 truncate_inode_pages(inode->i_mapping, new_i_size);
 486                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 487                                                i_size_read(inode), 1);
 488                 if (status)
 489                         mlog_errno(status);
 490
 491                 goto bail_unlock_sem;
 492         }
 493
 494         /* alright, we're going to need to do a full blown alloc size
 495          * change. Orphan the inode so that recovery can complete the
 496          * truncate if necessary. This does the task of marking
 497          * i_size. */
 498         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 499         if (status < 0) {
 500                 mlog_errno(status);
 501                 goto bail_unlock_sem;
 502         }
 503
 504         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 505         truncate_inode_pages(inode->i_mapping, new_i_size);
 506
 507         status = ocfs2_commit_truncate(osb, inode, di_bh);
 508         if (status < 0) {
 509                 mlog_errno(status);
 510                 goto bail_unlock_sem;
 511         }
 512
 513         /* TODO: orphan dir cleanup here. */
 514 bail_unlock_sem:
 515         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 516
 517 bail:
 518         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 519                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 520
 521         return status;
 522 }
 523
 524 /*
 525  * extend file allocation only here.
 526  * we'll update all the disk stuff, and oip->alloc_size
 527  *
 528  * expect stuff to be locked, a transaction started and enough data /
 529  * metadata reservations in the contexts.
 530  *
 531  * Will return -EAGAIN, and a reason if a restart is needed.
 532  * If passed in, *reason will always be set, even in error.
 533  */
 534 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 535                          struct inode *inode,
 536                          u32 *logical_offset,
 537                          u32 clusters_to_add,
 538                          int mark_unwritten,
 539                          struct buffer_head *fe_bh,
 540                          handle_t *handle,
 541                          struct ocfs2_alloc_context *data_ac,
 542                          struct ocfs2_alloc_context *meta_ac,
 543                          enum ocfs2_alloc_restarted *reason_ret)
 544 {
 545         int ret;
 546         struct ocfs2_extent_tree et;
 547
 548         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 549         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 550                                           clusters_to_add, mark_unwritten,
 551                                           data_ac, meta_ac, reason_ret);
 552
 553         return ret;
 554 }
 555
 556 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 557                                    u32 clusters_to_add, int mark_unwritten)
 558 {
 559         int status = 0;
 560         int restart_func = 0;
 561         int credits;
 562         u32 prev_clusters;
 563         struct buffer_head *bh = NULL;
 564         struct ocfs2_dinode *fe = NULL;
 565         handle_t *handle = NULL;
 566         struct ocfs2_alloc_context *data_ac = NULL;
 567         struct ocfs2_alloc_context *meta_ac = NULL;
 568         enum ocfs2_alloc_restarted why = RESTART_NONE;
 569         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 570         struct ocfs2_extent_tree et;
 571         int did_quota = 0;
 572
 573         /*
 574          * Unwritten extent only exists for file systems which
 575          * support holes.
 576          */
 577         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 578
 579         status = ocfs2_read_inode_block(inode, &bh);
 580         if (status < 0) {
 581                 mlog_errno(status);
 582                 goto leave;
 583         }
 584         fe = (struct ocfs2_dinode *) bh->b_data;
 585
 586 restart_all:
 587         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 588
 589         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 590         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 591                                        &data_ac, &meta_ac);
 592         if (status) {
 593                 mlog_errno(status);
 594                 goto leave;
 595         }
 596
 597         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 598         handle = ocfs2_start_trans(osb, credits);
 599         if (IS_ERR(handle)) {
 600                 status = PTR_ERR(handle);
 601                 handle = NULL;
 602                 mlog_errno(status);
 603                 goto leave;
 604         }
 605
 606 restarted_transaction:
 607         trace_ocfs2_extend_allocation(
 608                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 609                 (unsigned long long)i_size_read(inode),
 610                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 611                 why, restart_func);
 612
 613         status = dquot_alloc_space_nodirty(inode,
 614                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 615         if (status)
 616                 goto leave;
 617         did_quota = 1;
 618
 619         /* reserve a write to the file entry early on - that we if we
 620          * run out of credits in the allocation path, we can still
 621          * update i_size. */
 622         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 623                                          OCFS2_JOURNAL_ACCESS_WRITE);
 624         if (status < 0) {
 625                 mlog_errno(status);
 626                 goto leave;
 627         }
 628
 629         prev_clusters = OCFS2_I(inode)->ip_clusters;
 630
 631         status = ocfs2_add_inode_data(osb,
 632                                       inode,
 633                                       &logical_start,
 634                                       clusters_to_add,
 635                                       mark_unwritten,
 636                                       bh,
 637                                       handle,
 638                                       data_ac,
 639                                       meta_ac,
 640                                       &why);
 641         if ((status < 0) && (status != -EAGAIN)) {
 642                 if (status != -ENOSPC)
 643                         mlog_errno(status);
 644                 goto leave;
 645         }
 646         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 647         ocfs2_journal_dirty(handle, bh);
 648
 649         spin_lock(&OCFS2_I(inode)->ip_lock);
 650         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 651         spin_unlock(&OCFS2_I(inode)->ip_lock);
 652         /* Release unused quota reservation */
 653         dquot_free_space(inode,
 654                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 655         did_quota = 0;
 656
 657         if (why != RESTART_NONE && clusters_to_add) {
 658                 if (why == RESTART_META) {
 659                         restart_func = 1;
 660                         status = 0;
 661                 } else {
 662                         BUG_ON(why != RESTART_TRANS);
 663
 664                         status = ocfs2_allocate_extend_trans(handle, 1);
 665                         if (status < 0) {
 666                                 /* handle still has to be committed at
 667                                  * this point. */
 668                                 status = -ENOMEM;
 669                                 mlog_errno(status);
 670                                 goto leave;
 671                         }
 672                         goto restarted_transaction;
 673                 }
 674         }
 675
 676         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 677              le32_to_cpu(fe->i_clusters),
 678              (unsigned long long)le64_to_cpu(fe->i_size),
 679              OCFS2_I(inode)->ip_clusters,
 680              (unsigned long long)i_size_read(inode));
 681
 682 leave:
 683         if (status < 0 && did_quota)
 684                 dquot_free_space(inode,
 685                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 686         if (handle) {
 687                 ocfs2_commit_trans(osb, handle);
 688                 handle = NULL;
 689         }
 690         if (data_ac) {
 691                 ocfs2_free_alloc_context(data_ac);
 692                 data_ac = NULL;
 693         }
 694         if (meta_ac) {
 695                 ocfs2_free_alloc_context(meta_ac);
 696                 meta_ac = NULL;
 697         }
 698         if ((!status) && restart_func) {
 699                 restart_func = 0;
 700                 goto restart_all;
 701         }
 702         brelse(bh);
 703         bh = NULL;
 704
 705         return status;
 706 }
 707
 708 /*
 709  * While a write will already be ordering the data, a truncate will not.
 710  * Thus, we need to explicitly order the zeroed pages.
 711  */
 712 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 713                                                       struct buffer_head *di_bh,
 714                                                       loff_t start_byte,
 715                                                       loff_t length)
 716 {
 717         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 718         handle_t *handle = NULL;
 719         int ret = 0;
 720
 721         if (!ocfs2_should_order_data(inode))
 722                 goto out;
 723
 724         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 725         if (IS_ERR(handle)) {
 726                 ret = -ENOMEM;
 727                 mlog_errno(ret);
 728                 goto out;
 729         }
 730
 731         ret = ocfs2_jbd2_inode_add_write(handle, inode, start_byte, length);
 732         if (ret < 0) {
 733                 mlog_errno(ret);
 734                 goto out;
 735         }
 736
 737         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 738                                       OCFS2_JOURNAL_ACCESS_WRITE);
 739         if (ret)
 740                 mlog_errno(ret);
 741         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 742
 743 out:
 744         if (ret) {
 745                 if (!IS_ERR(handle))
 746                         ocfs2_commit_trans(osb, handle);
 747                 handle = ERR_PTR(ret);
 748         }
 749         return handle;
 750 }
 751
 752 /* Some parts of this taken from generic_cont_expand, which turned out
 753  * to be too fragile to do exactly what we need without us having to
 754  * worry about recursive locking in ->write_begin() and ->write_end(). */
 755 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 756                                  u64 abs_to, struct buffer_head *di_bh)
 757 {
 758         struct address_space *mapping = inode->i_mapping;
 759         struct page *page;
 760         unsigned long index = abs_from >> PAGE_SHIFT;
 761         handle_t *handle;
 762         int ret = 0;
 763         unsigned zero_from, zero_to, block_start, block_end;
 764         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 765
 766         BUG_ON(abs_from >= abs_to);
 767         BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 768         BUG_ON(abs_from & (inode->i_blkbits - 1));
 769
 770         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh,
 771                                                       abs_from,
 772                                                       abs_to - abs_from);
 773         if (IS_ERR(handle)) {
 774                 ret = PTR_ERR(handle);
 775                 goto out;
 776         }
 777
 778         page = find_or_create_page(mapping, index, GFP_NOFS);
 779         if (!page) {
 780                 ret = -ENOMEM;
 781                 mlog_errno(ret);
 782                 goto out_commit_trans;
 783         }
 784
 785         /* Get the offsets within the page that we want to zero */
 786         zero_from = abs_from & (PAGE_SIZE - 1);
 787         zero_to = abs_to & (PAGE_SIZE - 1);
 788         if (!zero_to)
 789                 zero_to = PAGE_SIZE;
 790
 791         trace_ocfs2_write_zero_page(
 792                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 793                         (unsigned long long)abs_from,
 794                         (unsigned long long)abs_to,
 795                         index, zero_from, zero_to);
 796
 797         /* We know that zero_from is block aligned */
 798         for (block_start = zero_from; block_start < zero_to;
 799              block_start = block_end) {
 800                 block_end = block_start + i_blocksize(inode);
 801
 802                 /*
 803                  * block_start is block-aligned.  Bump it by one to force
 804                  * __block_write_begin and block_commit_write to zero the
 805                  * whole block.
 806                  */
 807                 ret = __block_write_begin(page, block_start + 1, 0,
 808                                           ocfs2_get_block);
 809                 if (ret < 0) {
 810                         mlog_errno(ret);
 811                         goto out_unlock;
 812                 }
 813
 814
 815                 /* must not update i_size! */
 816                 ret = block_commit_write(page, block_start + 1,
 817                                          block_start + 1);
 818                 if (ret < 0)
 819                         mlog_errno(ret);
 820                 else
 821                         ret = 0;
 822         }
 823
 824         /*
 825          * fs-writeback will release the dirty pages without page lock
 826          * whose offset are over inode size, the release happens at
 827          * block_write_full_page().
 828          */
 829         i_size_write(inode, abs_to);
 830         inode->i_blocks = ocfs2_inode_sector_count(inode);
 831         di->i_size = cpu_to_le64((u64)i_size_read(inode));
 832         inode->i_mtime = inode->i_ctime = current_time(inode);
 833         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 834         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 835         di->i_mtime_nsec = di->i_ctime_nsec;
 836         if (handle) {
 837                 ocfs2_journal_dirty(handle, di_bh);
 838                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 839         }
 840
 841 out_unlock:
 842         unlock_page(page);
 843         put_page(page);
 844 out_commit_trans:
 845         if (handle)
 846                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 847 out:
 848         return ret;
 849 }
 850
 851 /*
 852  * Find the next range to zero.  We do this in terms of bytes because
 853  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 854  * pagecache.  We may return multiple extents.
 855  *
 856  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 857  * needs to be zeroed.  range_start and range_end return the next zeroing
 858  * range.  A subsequent call should pass the previous range_end as its
 859  * zero_start.  If range_end is 0, there's nothing to do.
 860  *
 861  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 862  */
 863 static int ocfs2_zero_extend_get_range(struct inode *inode,
 864                                        struct buffer_head *di_bh,
 865                                        u64 zero_start, u64 zero_end,
 866                                        u64 *range_start, u64 *range_end)
 867 {
 868         int rc = 0, needs_cow = 0;
 869         u32 p_cpos, zero_clusters = 0;
 870         u32 zero_cpos =
 871                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 872         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 873         unsigned int num_clusters = 0;
 874         unsigned int ext_flags = 0;
 875
 876         while (zero_cpos < last_cpos) {
 877                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 878                                         &num_clusters, &ext_flags);
 879                 if (rc) {
 880                         mlog_errno(rc);
 881                         goto out;
 882                 }
 883
 884                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 885                         zero_clusters = num_clusters;
 886                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 887                                 needs_cow = 1;
 888                         break;
 889                 }
 890
 891                 zero_cpos += num_clusters;
 892         }
 893         if (!zero_clusters) {
 894                 *range_end = 0;
 895                 goto out;
 896         }
 897
 898         while ((zero_cpos + zero_clusters) < last_cpos) {
 899                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 900                                         &p_cpos, &num_clusters,
 901                                         &ext_flags);
 902                 if (rc) {
 903                         mlog_errno(rc);
 904                         goto out;
 905                 }
 906
 907                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 908                         break;
 909                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 910                         needs_cow = 1;
 911                 zero_clusters += num_clusters;
 912         }
 913         if ((zero_cpos + zero_clusters) > last_cpos)
 914                 zero_clusters = last_cpos - zero_cpos;
 915
 916         if (needs_cow) {
 917                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 918                                         zero_clusters, UINT_MAX);
 919                 if (rc) {
 920                         mlog_errno(rc);
 921                         goto out;
 922                 }
 923         }
 924
 925         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 926         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 927                                              zero_cpos + zero_clusters);
 928
 929 out:
 930         return rc;
 931 }
 932
 933 /*
 934  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 935  * has made sure that the entire range needs zeroing.
 936  */
 937 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 938                                    u64 range_end, struct buffer_head *di_bh)
 939 {
 940         int rc = 0;
 941         u64 next_pos;
 942         u64 zero_pos = range_start;
 943
 944         trace_ocfs2_zero_extend_range(
 945                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 946                         (unsigned long long)range_start,
 947                         (unsigned long long)range_end);
 948         BUG_ON(range_start >= range_end);
 949
 950         while (zero_pos < range_end) {
 951                 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 952                 if (next_pos > range_end)
 953                         next_pos = range_end;
 954                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 955                 if (rc < 0) {
 956                         mlog_errno(rc);
 957                         break;
 958                 }
 959                 zero_pos = next_pos;
 960
 961                 /*
 962                  * Very large extends have the potential to lock up
 963                  * the cpu for extended periods of time.
 964                  */
 965                 cond_resched();
 966         }
 967
 968         return rc;
 969 }
 970
 971 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 972                       loff_t zero_to_size)
 973 {
 974         int ret = 0;
 975         u64 zero_start, range_start = 0, range_end = 0;
 976         struct super_block *sb = inode->i_sb;
 977
 978         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 979         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 980                                 (unsigned long long)zero_start,
 981                                 (unsigned long long)i_size_read(inode));
 982         while (zero_start < zero_to_size) {
 983                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 984                                                   zero_to_size,
 985                                                   &range_start,
 986                                                   &range_end);
 987                 if (ret) {
 988                         mlog_errno(ret);
 989                         break;
 990                 }
 991                 if (!range_end)
 992                         break;
 993                 /* Trim the ends */
 994                 if (range_start < zero_start)
 995                         range_start = zero_start;
 996                 if (range_end > zero_to_size)
 997                         range_end = zero_to_size;
 998
 999                 ret = ocfs2_zero_extend_range(inode, range_start,
1000                                               range_end, di_bh);
1001                 if (ret) {
1002                         mlog_errno(ret);
1003                         break;
1004                 }
1005                 zero_start = range_end;
1006         }
1007
1008         return ret;
1009 }
1010
1011 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1012                           u64 new_i_size, u64 zero_to)
1013 {
1014         int ret;
1015         u32 clusters_to_add;
1016         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1017
1018         /*
1019          * Only quota files call this without a bh, and they can't be
1020          * refcounted.
1021          */
1022         BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1023         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1024
1025         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1026         if (clusters_to_add < oi->ip_clusters)
1027                 clusters_to_add = 0;
1028         else
1029                 clusters_to_add -= oi->ip_clusters;
1030
1031         if (clusters_to_add) {
1032                 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1033                                               clusters_to_add, 0);
1034                 if (ret) {
1035                         mlog_errno(ret);
1036                         goto out;
1037                 }
1038         }
1039
1040         /*
1041          * Call this even if we don't add any clusters to the tree. We
1042          * still need to zero the area between the old i_size and the
1043          * new i_size.
1044          */
1045         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1046         if (ret < 0)
1047                 mlog_errno(ret);
1048
1049 out:
1050         return ret;
1051 }
1052
1053 static int ocfs2_extend_file(struct inode *inode,
1054                              struct buffer_head *di_bh,
1055                              u64 new_i_size)
1056 {
1057         int ret = 0;
1058         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1059
1060         BUG_ON(!di_bh);
1061
1062         /* setattr sometimes calls us like this. */
1063         if (new_i_size == 0)
1064                 goto out;
1065
1066         if (i_size_read(inode) == new_i_size)
1067                 goto out;
1068         BUG_ON(new_i_size < i_size_read(inode));
1069
1070         /*
1071          * The alloc sem blocks people in read/write from reading our
1072          * allocation until we're done changing it. We depend on
1073          * i_mutex to block other extend/truncate calls while we're
1074          * here.  We even have to hold it for sparse files because there
1075          * might be some tail zeroing.
1076          */
1077         down_write(&oi->ip_alloc_sem);
1078
1079         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1080                 /*
1081                  * We can optimize small extends by keeping the inodes
1082                  * inline data.
1083                  */
1084                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1085                         up_write(&oi->ip_alloc_sem);
1086                         goto out_update_size;
1087                 }
1088
1089                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1090                 if (ret) {
1091                         up_write(&oi->ip_alloc_sem);
1092                         mlog_errno(ret);
1093                         goto out;
1094                 }
1095         }
1096
1097         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1098                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1099         else
1100                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1101                                             new_i_size);
1102
1103         up_write(&oi->ip_alloc_sem);
1104
1105         if (ret < 0) {
1106                 mlog_errno(ret);
1107                 goto out;
1108         }
1109
1110 out_update_size:
1111         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1112         if (ret < 0)
1113                 mlog_errno(ret);
1114
1115 out:
1116         return ret;
1117 }
1118
1119 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1120 {
1121         int status = 0, size_change;
1122         int inode_locked = 0;
1123         struct inode *inode = d_inode(dentry);
1124         struct super_block *sb = inode->i_sb;
1125         struct ocfs2_super *osb = OCFS2_SB(sb);
1126         struct buffer_head *bh = NULL;
1127         handle_t *handle = NULL;
1128         struct dquot *transfer_to[MAXQUOTAS] = { };
1129         int qtype;
1130         int had_lock;
1131         struct ocfs2_lock_holder oh;
1132
1133         trace_ocfs2_setattr(inode, dentry,
1134                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1135                             dentry->d_name.len, dentry->d_name.name,
1136                             attr->ia_valid, attr->ia_mode,
1137                             from_kuid(&init_user_ns, attr->ia_uid),
1138                             from_kgid(&init_user_ns, attr->ia_gid));
1139
1140         /* ensuring we don't even attempt to truncate a symlink */
1141         if (S_ISLNK(inode->i_mode))
1142                 attr->ia_valid &= ~ATTR_SIZE;
1143
1144 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1145                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1146         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1147                 return 0;
1148
1149         status = setattr_prepare(dentry, attr);
1150         if (status)
1151                 return status;
1152
1153         if (is_quota_modification(inode, attr)) {
1154                 status = dquot_initialize(inode);
1155                 if (status)
1156                         return status;
1157         }
1158         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1159         if (size_change) {
1160                 /*
1161                  * Here we should wait dio to finish before inode lock
1162                  * to avoid a deadlock between ocfs2_setattr() and
1163                  * ocfs2_dio_end_io_write()
1164                  */
1165                 inode_dio_wait(inode);
1166
1167                 status = ocfs2_rw_lock(inode, 1);
1168                 if (status < 0) {
1169                         mlog_errno(status);
1170                         goto bail;
1171                 }
1172         }
1173
1174         had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1175         if (had_lock < 0) {
1176                 status = had_lock;
1177                 goto bail_unlock_rw;
1178         } else if (had_lock) {
1179                 /*
1180                  * As far as we know, ocfs2_setattr() could only be the first
1181                  * VFS entry point in the call chain of recursive cluster
1182                  * locking issue.
1183                  *
1184                  * For instance:
1185                  * chmod_common()
1186                  *  notify_change()
1187                  *   ocfs2_setattr()
1188                  *    posix_acl_chmod()
1189                  *     ocfs2_iop_get_acl()
1190                  *
1191                  * But, we're not 100% sure if it's always true, because the
1192                  * ordering of the VFS entry points in the call chain is out
1193                  * of our control. So, we'd better dump the stack here to
1194                  * catch the other cases of recursive locking.
1195                  */
1196                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1197                 dump_stack();
1198         }
1199         inode_locked = 1;
1200
1201         if (size_change) {
1202                 status = inode_newsize_ok(inode, attr->ia_size);
1203                 if (status)
1204                         goto bail_unlock;
1205
1206                 if (i_size_read(inode) >= attr->ia_size) {
1207                         if (ocfs2_should_order_data(inode)) {
1208                                 status = ocfs2_begin_ordered_truncate(inode,
1209                                                                       attr->ia_size);
1210                                 if (status)
1211                                         goto bail_unlock;
1212                         }
1213                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1214                 } else
1215                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1216                 if (status < 0) {
1217                         if (status != -ENOSPC)
1218                                 mlog_errno(status);
1219                         status = -ENOSPC;
1220                         goto bail_unlock;
1221                 }
1222         }
1223
1224         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1225             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1226                 /*
1227                  * Gather pointers to quota structures so that allocation /
1228                  * freeing of quota structures happens here and not inside
1229                  * dquot_transfer() where we have problems with lock ordering
1230                  */
1231                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1232                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1233                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1234                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1235                         if (IS_ERR(transfer_to[USRQUOTA])) {
1236                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1237                                 transfer_to[USRQUOTA] = NULL;
1238                                 goto bail_unlock;
1239                         }
1240                 }
1241                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1242                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1243                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1244                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1245                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1246                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1247                                 transfer_to[GRPQUOTA] = NULL;
1248                                 goto bail_unlock;
1249                         }
1250                 }
1251                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1252                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1253                                            2 * ocfs2_quota_trans_credits(sb));
1254                 if (IS_ERR(handle)) {
1255                         status = PTR_ERR(handle);
1256                         mlog_errno(status);
1257                         goto bail_unlock_alloc;
1258                 }
1259                 status = __dquot_transfer(inode, transfer_to);
1260                 if (status < 0)
1261                         goto bail_commit;
1262         } else {
1263                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1264                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1265                 if (IS_ERR(handle)) {
1266                         status = PTR_ERR(handle);
1267                         mlog_errno(status);
1268                         goto bail_unlock_alloc;
1269                 }
1270         }
1271
1272         setattr_copy(inode, attr);
1273         mark_inode_dirty(inode);
1274
1275         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1276         if (status < 0)
1277                 mlog_errno(status);
1278
1279 bail_commit:
1280         ocfs2_commit_trans(osb, handle);
1281 bail_unlock_alloc:
1282         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1283 bail_unlock:
1284         if (status && inode_locked) {
1285                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1286                 inode_locked = 0;
1287         }
1288 bail_unlock_rw:
1289         if (size_change)
1290                 ocfs2_rw_unlock(inode, 1);
1291 bail:
1292
1293         /* Release quota pointers in case we acquired them */
1294         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1295                 dqput(transfer_to[qtype]);
1296
1297         if (!status && attr->ia_valid & ATTR_MODE) {
1298                 status = ocfs2_acl_chmod(inode, bh);
1299                 if (status < 0)
1300                         mlog_errno(status);
1301         }
1302         if (inode_locked)
1303                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1304
1305         brelse(bh);
1306         return status;
1307 }
1308
1309 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1310                   u32 request_mask, unsigned int flags)
1311 {
1312         struct inode *inode = d_inode(path->dentry);
1313         struct super_block *sb = path->dentry->d_sb;
1314         struct ocfs2_super *osb = sb->s_fs_info;
1315         int err;
1316
1317         err = ocfs2_inode_revalidate(path->dentry);
1318         if (err) {
1319                 if (err != -ENOENT)
1320                         mlog_errno(err);
1321                 goto bail;
1322         }
1323
1324         generic_fillattr(inode, stat);
1325         /*
1326          * If there is inline data in the inode, the inode will normally not
1327          * have data blocks allocated (it may have an external xattr block).
1328          * Report at least one sector for such files, so tools like tar, rsync,
1329          * others don't incorrectly think the file is completely sparse.
1330          */
1331         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1332                 stat->blocks += (stat->size + 511)>>9;
1333
1334         /* We set the blksize from the cluster size for performance */
1335         stat->blksize = osb->s_clustersize;
1336
1337 bail:
1338         return err;
1339 }
1340
1341 int ocfs2_permission(struct inode *inode, int mask)
1342 {
1343         int ret, had_lock;
1344         struct ocfs2_lock_holder oh;
1345
1346         if (mask & MAY_NOT_BLOCK)
1347                 return -ECHILD;
1348
1349         had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1350         if (had_lock < 0) {
1351                 ret = had_lock;
1352                 goto out;
1353         } else if (had_lock) {
1354                 /* See comments in ocfs2_setattr() for details.
1355                  * The call chain of this case could be:
1356                  * do_sys_open()
1357                  *  may_open()
1358                  *   inode_permission()
1359                  *    ocfs2_permission()
1360                  *     ocfs2_iop_get_acl()
1361                  */
1362                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1363                 dump_stack();
1364         }
1365
1366         ret = generic_permission(inode, mask);
1367
1368         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1369 out:
1370         return ret;
1371 }
1372
1373 static int __ocfs2_write_remove_suid(struct inode *inode,
1374                                      struct buffer_head *bh)
1375 {
1376         int ret;
1377         handle_t *handle;
1378         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1379         struct ocfs2_dinode *di;
1380
1381         trace_ocfs2_write_remove_suid(
1382                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1383                         inode->i_mode);
1384
1385         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1386         if (IS_ERR(handle)) {
1387                 ret = PTR_ERR(handle);
1388                 mlog_errno(ret);
1389                 goto out;
1390         }
1391
1392         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1393                                       OCFS2_JOURNAL_ACCESS_WRITE);
1394         if (ret < 0) {
1395                 mlog_errno(ret);
1396                 goto out_trans;
1397         }
1398
1399         inode->i_mode &= ~S_ISUID;
1400         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1401                 inode->i_mode &= ~S_ISGID;
1402
1403         di = (struct ocfs2_dinode *) bh->b_data;
1404         di->i_mode = cpu_to_le16(inode->i_mode);
1405         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1406
1407         ocfs2_journal_dirty(handle, bh);
1408
1409 out_trans:
1410         ocfs2_commit_trans(osb, handle);
1411 out:
1412         return ret;
1413 }
1414
1415 static int ocfs2_write_remove_suid(struct inode *inode)
1416 {
1417         int ret;
1418         struct buffer_head *bh = NULL;
1419
1420         ret = ocfs2_read_inode_block(inode, &bh);
1421         if (ret < 0) {
1422                 mlog_errno(ret);
1423                 goto out;
1424         }
1425
1426         ret =  __ocfs2_write_remove_suid(inode, bh);
1427 out:
1428         brelse(bh);
1429         return ret;
1430 }
1431
1432 /*
1433  * Allocate enough extents to cover the region starting at byte offset
1434  * start for len bytes. Existing extents are skipped, any extents
1435  * added are marked as "unwritten".
1436  */
1437 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1438                                             u64 start, u64 len)
1439 {
1440         int ret;
1441         u32 cpos, phys_cpos, clusters, alloc_size;
1442         u64 end = start + len;
1443         struct buffer_head *di_bh = NULL;
1444
1445         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1446                 ret = ocfs2_read_inode_block(inode, &di_bh);
1447                 if (ret) {
1448                         mlog_errno(ret);
1449                         goto out;
1450                 }
1451
1452                 /*
1453                  * Nothing to do if the requested reservation range
1454                  * fits within the inode.
1455                  */
1456                 if (ocfs2_size_fits_inline_data(di_bh, end))
1457                         goto out;
1458
1459                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1460                 if (ret) {
1461                         mlog_errno(ret);
1462                         goto out;
1463                 }
1464         }
1465
1466         /*
1467          * We consider both start and len to be inclusive.
1468          */
1469         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1470         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1471         clusters -= cpos;
1472
1473         while (clusters) {
1474                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1475                                          &alloc_size, NULL);
1476                 if (ret) {
1477                         mlog_errno(ret);
1478                         goto out;
1479                 }
1480
1481                 /*
1482                  * Hole or existing extent len can be arbitrary, so
1483                  * cap it to our own allocation request.
1484                  */
1485                 if (alloc_size > clusters)
1486                         alloc_size = clusters;
1487
1488                 if (phys_cpos) {
1489                         /*
1490                          * We already have an allocation at this
1491                          * region so we can safely skip it.
1492                          */
1493                         goto next;
1494                 }
1495
1496                 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1497                 if (ret) {
1498                         if (ret != -ENOSPC)
1499                                 mlog_errno(ret);
1500                         goto out;
1501                 }
1502
1503 next:
1504                 cpos += alloc_size;
1505                 clusters -= alloc_size;
1506         }
1507
1508         ret = 0;
1509 out:
1510
1511         brelse(di_bh);
1512         return ret;
1513 }
1514
1515 /*
1516  * Truncate a byte range, avoiding pages within partial clusters. This
1517  * preserves those pages for the zeroing code to write to.
1518  */
1519 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1520                                          u64 byte_len)
1521 {
1522         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1523         loff_t start, end;
1524         struct address_space *mapping = inode->i_mapping;
1525
1526         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1527         end = byte_start + byte_len;
1528         end = end & ~(osb->s_clustersize - 1);
1529
1530         if (start < end) {
1531                 unmap_mapping_range(mapping, start, end - start, 0);
1532                 truncate_inode_pages_range(mapping, start, end - 1);
1533         }
1534 }
1535
1536 /*
1537  * zero out partial blocks of one cluster.
1538  *
1539  * start: file offset where zero starts, will be made upper block aligned.
1540  * len: it will be trimmed to the end of current cluster if "start + len"
1541  *      is bigger than it.
1542  */
1543 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1544                                         u64 start, u64 len)
1545 {
1546         int ret;
1547         u64 start_block, end_block, nr_blocks;
1548         u64 p_block, offset;
1549         u32 cluster, p_cluster, nr_clusters;
1550         struct super_block *sb = inode->i_sb;
1551         u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1552
1553         if (start + len < end)
1554                 end = start + len;
1555
1556         start_block = ocfs2_blocks_for_bytes(sb, start);
1557         end_block = ocfs2_blocks_for_bytes(sb, end);
1558         nr_blocks = end_block - start_block;
1559         if (!nr_blocks)
1560                 return 0;
1561
1562         cluster = ocfs2_bytes_to_clusters(sb, start);
1563         ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1564                                 &nr_clusters, NULL);
1565         if (ret)
1566                 return ret;
1567         if (!p_cluster)
1568                 return 0;
1569
1570         offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1571         p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1572         return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1573 }
1574
1575 static int ocfs2_zero_partial_clusters(struct inode *inode,
1576                                        u64 start, u64 len)
1577 {
1578         int ret = 0;
1579         u64 tmpend = 0;
1580         u64 end = start + len;
1581         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1582         unsigned int csize = osb->s_clustersize;
1583         handle_t *handle;
1584         loff_t isize = i_size_read(inode);
1585
1586         /*
1587          * The "start" and "end" values are NOT necessarily part of
1588          * the range whose allocation is being deleted. Rather, this
1589          * is what the user passed in with the request. We must zero
1590          * partial clusters here. There's no need to worry about
1591          * physical allocation - the zeroing code knows to skip holes.
1592          */
1593         trace_ocfs2_zero_partial_clusters(
1594                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1595                 (unsigned long long)start, (unsigned long long)end);
1596
1597         /*
1598          * If both edges are on a cluster boundary then there's no
1599          * zeroing required as the region is part of the allocation to
1600          * be truncated.
1601          */
1602         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1603                 goto out;
1604
1605         /* No page cache for EOF blocks, issue zero out to disk. */
1606         if (end > isize) {
1607                 /*
1608                  * zeroout eof blocks in last cluster starting from
1609                  * "isize" even "start" > "isize" because it is
1610                  * complicated to zeroout just at "start" as "start"
1611                  * may be not aligned with block size, buffer write
1612                  * would be required to do that, but out of eof buffer
1613                  * write is not supported.
1614                  */
1615                 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1616                                         end - isize);
1617                 if (ret) {
1618                         mlog_errno(ret);
1619                         goto out;
1620                 }
1621                 if (start >= isize)
1622                         goto out;
1623                 end = isize;
1624         }
1625         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1626         if (IS_ERR(handle)) {
1627                 ret = PTR_ERR(handle);
1628                 mlog_errno(ret);
1629                 goto out;
1630         }
1631
1632         /*
1633          * If start is on a cluster boundary and end is somewhere in another
1634          * cluster, we have not COWed the cluster starting at start, unless
1635          * end is also within the same cluster. So, in this case, we skip this
1636          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1637          * to the next one.
1638          */
1639         if ((start & (csize - 1)) != 0) {
1640                 /*
1641                  * We want to get the byte offset of the end of the 1st
1642                  * cluster.
1643                  */
1644                 tmpend = (u64)osb->s_clustersize +
1645                         (start & ~(osb->s_clustersize - 1));
1646                 if (tmpend > end)
1647                         tmpend = end;
1648
1649                 trace_ocfs2_zero_partial_clusters_range1(
1650                         (unsigned long long)start,
1651                         (unsigned long long)tmpend);
1652
1653                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1654                                                     tmpend);
1655                 if (ret)
1656                         mlog_errno(ret);
1657         }
1658
1659         if (tmpend < end) {
1660                 /*
1661                  * This may make start and end equal, but the zeroing
1662                  * code will skip any work in that case so there's no
1663                  * need to catch it up here.
1664                  */
1665                 start = end & ~(osb->s_clustersize - 1);
1666
1667                 trace_ocfs2_zero_partial_clusters_range2(
1668                         (unsigned long long)start, (unsigned long long)end);
1669
1670                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1671                 if (ret)
1672                         mlog_errno(ret);
1673         }
1674         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1675
1676         ocfs2_commit_trans(osb, handle);
1677 out:
1678         return ret;
1679 }
1680
1681 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1682 {
1683         int i;
1684         struct ocfs2_extent_rec *rec = NULL;
1685
1686         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1687
1688                 rec = &el->l_recs[i];
1689
1690                 if (le32_to_cpu(rec->e_cpos) < pos)
1691                         break;
1692         }
1693
1694         return i;
1695 }
1696
1697 /*
1698  * Helper to calculate the punching pos and length in one run, we handle the
1699  * following three cases in order:
1700  *
1701  * - remove the entire record
1702  * - remove a partial record
1703  * - no record needs to be removed (hole-punching completed)
1704 */
1705 static void ocfs2_calc_trunc_pos(struct inode *inode,
1706                                  struct ocfs2_extent_list *el,
1707                                  struct ocfs2_extent_rec *rec,
1708                                  u32 trunc_start, u32 *trunc_cpos,
1709                                  u32 *trunc_len, u32 *trunc_end,
1710                                  u64 *blkno, int *done)
1711 {
1712         int ret = 0;
1713         u32 coff, range;
1714
1715         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1716
1717         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1718                 /*
1719                  * remove an entire extent record.
1720                  */
1721                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1722                 /*
1723                  * Skip holes if any.
1724                  */
1725                 if (range < *trunc_end)
1726                         *trunc_end = range;
1727                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1728                 *blkno = le64_to_cpu(rec->e_blkno);
1729                 *trunc_end = le32_to_cpu(rec->e_cpos);
1730         } else if (range > trunc_start) {
1731                 /*
1732                  * remove a partial extent record, which means we're
1733                  * removing the last extent record.
1734                  */
1735                 *trunc_cpos = trunc_start;
1736                 /*
1737                  * skip hole if any.
1738                  */
1739                 if (range < *trunc_end)
1740                         *trunc_end = range;
1741                 *trunc_len = *trunc_end - trunc_start;
1742                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1743                 *blkno = le64_to_cpu(rec->e_blkno) +
1744                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1745                 *trunc_end = trunc_start;
1746         } else {
1747                 /*
1748                  * It may have two following possibilities:
1749                  *
1750                  * - last record has been removed
1751                  * - trunc_start was within a hole
1752                  *
1753                  * both two cases mean the completion of hole punching.
1754                  */
1755                 ret = 1;
1756         }
1757
1758         *done = ret;
1759 }
1760
1761 int ocfs2_remove_inode_range(struct inode *inode,
1762                              struct buffer_head *di_bh, u64 byte_start,
1763                              u64 byte_len)
1764 {
1765         int ret = 0, flags = 0, done = 0, i;
1766         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1767         u32 cluster_in_el;
1768         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1769         struct ocfs2_cached_dealloc_ctxt dealloc;
1770         struct address_space *mapping = inode->i_mapping;
1771         struct ocfs2_extent_tree et;
1772         struct ocfs2_path *path = NULL;
1773         struct ocfs2_extent_list *el = NULL;
1774         struct ocfs2_extent_rec *rec = NULL;
1775         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1776         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1777
1778         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1779         ocfs2_init_dealloc_ctxt(&dealloc);
1780
1781         trace_ocfs2_remove_inode_range(
1782                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1783                         (unsigned long long)byte_start,
1784                         (unsigned long long)byte_len);
1785
1786         if (byte_len == 0)
1787                 return 0;
1788
1789         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1790                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1791                                             byte_start + byte_len, 0);
1792                 if (ret) {
1793                         mlog_errno(ret);
1794                         goto out;
1795                 }
1796                 /*
1797                  * There's no need to get fancy with the page cache
1798                  * truncate of an inline-data inode. We're talking
1799                  * about less than a page here, which will be cached
1800                  * in the dinode buffer anyway.
1801                  */
1802                 unmap_mapping_range(mapping, 0, 0, 0);
1803                 truncate_inode_pages(mapping, 0);
1804                 goto out;
1805         }
1806
1807         /*
1808          * For reflinks, we may need to CoW 2 clusters which might be
1809          * partially zero'd later, if hole's start and end offset were
1810          * within one cluster(means is not exactly aligned to clustersize).
1811          */
1812
1813         if (ocfs2_is_refcount_inode(inode)) {
1814                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1815                 if (ret) {
1816                         mlog_errno(ret);
1817                         goto out;
1818                 }
1819
1820                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1821                 if (ret) {
1822                         mlog_errno(ret);
1823                         goto out;
1824                 }
1825         }
1826
1827         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1828         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1829         cluster_in_el = trunc_end;
1830
1831         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1832         if (ret) {
1833                 mlog_errno(ret);
1834                 goto out;
1835         }
1836
1837         path = ocfs2_new_path_from_et(&et);
1838         if (!path) {
1839                 ret = -ENOMEM;
1840                 mlog_errno(ret);
1841                 goto out;
1842         }
1843
1844         while (trunc_end > trunc_start) {
1845
1846                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1847                                       cluster_in_el);
1848                 if (ret) {
1849                         mlog_errno(ret);
1850                         goto out;
1851                 }
1852
1853                 el = path_leaf_el(path);
1854
1855                 i = ocfs2_find_rec(el, trunc_end);
1856                 /*
1857                  * Need to go to previous extent block.
1858                  */
1859                 if (i < 0) {
1860                         if (path->p_tree_depth == 0)
1861                                 break;
1862
1863                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1864                                                             path,
1865                                                             &cluster_in_el);
1866                         if (ret) {
1867                                 mlog_errno(ret);
1868                                 goto out;
1869                         }
1870
1871                         /*
1872                          * We've reached the leftmost extent block,
1873                          * it's safe to leave.
1874                          */
1875                         if (cluster_in_el == 0)
1876                                 break;
1877
1878                         /*
1879                          * The 'pos' searched for previous extent block is
1880                          * always one cluster less than actual trunc_end.
1881                          */
1882                         trunc_end = cluster_in_el + 1;
1883
1884                         ocfs2_reinit_path(path, 1);
1885
1886                         continue;
1887
1888                 } else
1889                         rec = &el->l_recs[i];
1890
1891                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1892                                      &trunc_len, &trunc_end, &blkno, &done);
1893                 if (done)
1894                         break;
1895
1896                 flags = rec->e_flags;
1897                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1898
1899                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1900                                                phys_cpos, trunc_len, flags,
1901                                                &dealloc, refcount_loc, false);
1902                 if (ret < 0) {
1903                         mlog_errno(ret);
1904                         goto out;
1905                 }
1906
1907                 cluster_in_el = trunc_end;
1908
1909                 ocfs2_reinit_path(path, 1);
1910         }
1911
1912         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1913
1914 out:
1915         ocfs2_free_path(path);
1916         ocfs2_schedule_truncate_log_flush(osb, 1);
1917         ocfs2_run_deallocs(osb, &dealloc);
1918
1919         return ret;
1920 }
1921
1922 /*
1923  * Parts of this function taken from xfs_change_file_space()
1924  */
1925 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1926                                      loff_t f_pos, unsigned int cmd,
1927                                      struct ocfs2_space_resv *sr,
1928                                      int change_size)
1929 {
1930         int ret;
1931         s64 llen;
1932         loff_t size, orig_isize;
1933         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1934         struct buffer_head *di_bh = NULL;
1935         handle_t *handle;
1936         unsigned long long max_off = inode->i_sb->s_maxbytes;
1937
1938         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1939                 return -EROFS;
1940
1941         inode_lock(inode);
1942
1943         /*
1944          * This prevents concurrent writes on other nodes
1945          */
1946         ret = ocfs2_rw_lock(inode, 1);
1947         if (ret) {
1948                 mlog_errno(ret);
1949                 goto out;
1950         }
1951
1952         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1953         if (ret) {
1954                 mlog_errno(ret);
1955                 goto out_rw_unlock;
1956         }
1957
1958         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1959                 ret = -EPERM;
1960                 goto out_inode_unlock;
1961         }
1962
1963         switch (sr->l_whence) {
1964         case 0: /*SEEK_SET*/
1965                 break;
1966         case 1: /*SEEK_CUR*/
1967                 sr->l_start += f_pos;
1968                 break;
1969         case 2: /*SEEK_END*/
1970                 sr->l_start += i_size_read(inode);
1971                 break;
1972         default:
1973                 ret = -EINVAL;
1974                 goto out_inode_unlock;
1975         }
1976         sr->l_whence = 0;
1977
1978         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1979
1980         if (sr->l_start < 0
1981             || sr->l_start > max_off
1982             || (sr->l_start + llen) < 0
1983             || (sr->l_start + llen) > max_off) {
1984                 ret = -EINVAL;
1985                 goto out_inode_unlock;
1986         }
1987         size = sr->l_start + sr->l_len;
1988
1989         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1990             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1991                 if (sr->l_len <= 0) {
1992                         ret = -EINVAL;
1993                         goto out_inode_unlock;
1994                 }
1995         }
1996
1997         if (file && should_remove_suid(file->f_path.dentry)) {
1998                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1999                 if (ret) {
2000                         mlog_errno(ret);
2001                         goto out_inode_unlock;
2002                 }
2003         }
2004
2005         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2006         switch (cmd) {
2007         case OCFS2_IOC_RESVSP:
2008         case OCFS2_IOC_RESVSP64:
2009                 /*
2010                  * This takes unsigned offsets, but the signed ones we
2011                  * pass have been checked against overflow above.
2012                  */
2013                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2014                                                        sr->l_len);
2015                 break;
2016         case OCFS2_IOC_UNRESVSP:
2017         case OCFS2_IOC_UNRESVSP64:
2018                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2019                                                sr->l_len);
2020                 break;
2021         default:
2022                 ret = -EINVAL;
2023         }
2024
2025         orig_isize = i_size_read(inode);
2026         /* zeroout eof blocks in the cluster. */
2027         if (!ret && change_size && orig_isize < size) {
2028                 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2029                                         size - orig_isize);
2030                 if (!ret)
2031                         i_size_write(inode, size);
2032         }
2033         up_write(&OCFS2_I(inode)->ip_alloc_sem);
2034         if (ret) {
2035                 mlog_errno(ret);
2036                 goto out_inode_unlock;
2037         }
2038
2039         /*
2040          * We update c/mtime for these changes
2041          */
2042         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2043         if (IS_ERR(handle)) {
2044                 ret = PTR_ERR(handle);
2045                 mlog_errno(ret);
2046                 goto out_inode_unlock;
2047         }
2048
2049         inode->i_ctime = inode->i_mtime = current_time(inode);
2050         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2051         if (ret < 0)
2052                 mlog_errno(ret);
2053
2054         if (file && (file->f_flags & O_SYNC))
2055                 handle->h_sync = 1;
2056
2057         ocfs2_commit_trans(osb, handle);
2058
2059 out_inode_unlock:
2060         brelse(di_bh);
2061         ocfs2_inode_unlock(inode, 1);
2062 out_rw_unlock:
2063         ocfs2_rw_unlock(inode, 1);
2064
2065 out:
2066         inode_unlock(inode);
2067         return ret;
2068 }
2069
2070 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2071                             struct ocfs2_space_resv *sr)
2072 {
2073         struct inode *inode = file_inode(file);
2074         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2075         int ret;
2076
2077         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2078             !ocfs2_writes_unwritten_extents(osb))
2079                 return -ENOTTY;
2080         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2081                  !ocfs2_sparse_alloc(osb))
2082                 return -ENOTTY;
2083
2084         if (!S_ISREG(inode->i_mode))
2085                 return -EINVAL;
2086
2087         if (!(file->f_mode & FMODE_WRITE))
2088                 return -EBADF;
2089
2090         ret = mnt_want_write_file(file);
2091         if (ret)
2092                 return ret;
2093         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2094         mnt_drop_write_file(file);
2095         return ret;
2096 }
2097
2098 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2099                             loff_t len)
2100 {
2101         struct inode *inode = file_inode(file);
2102         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2103         struct ocfs2_space_resv sr;
2104         int change_size = 1;
2105         int cmd = OCFS2_IOC_RESVSP64;
2106         int ret = 0;
2107
2108         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2109                 return -EOPNOTSUPP;
2110         if (!ocfs2_writes_unwritten_extents(osb))
2111                 return -EOPNOTSUPP;
2112
2113         if (mode & FALLOC_FL_KEEP_SIZE) {
2114                 change_size = 0;
2115         } else {
2116                 ret = inode_newsize_ok(inode, offset + len);
2117                 if (ret)
2118                         return ret;
2119         }
2120
2121         if (mode & FALLOC_FL_PUNCH_HOLE)
2122                 cmd = OCFS2_IOC_UNRESVSP64;
2123
2124         sr.l_whence = 0;
2125         sr.l_start = (s64)offset;
2126         sr.l_len = (s64)len;
2127
2128         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2129                                          change_size);
2130 }
2131
2132 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2133                                    size_t count)
2134 {
2135         int ret = 0;
2136         unsigned int extent_flags;
2137         u32 cpos, clusters, extent_len, phys_cpos;
2138         struct super_block *sb = inode->i_sb;
2139
2140         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2141             !ocfs2_is_refcount_inode(inode) ||
2142             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2143                 return 0;
2144
2145         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2146         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2147
2148         while (clusters) {
2149                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2150                                          &extent_flags);
2151                 if (ret < 0) {
2152                         mlog_errno(ret);
2153                         goto out;
2154                 }
2155
2156                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2157                         ret = 1;
2158                         break;
2159                 }
2160
2161                 if (extent_len > clusters)
2162                         extent_len = clusters;
2163
2164                 clusters -= extent_len;
2165                 cpos += extent_len;
2166         }
2167 out:
2168         return ret;
2169 }
2170
2171 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2172 {
2173         int blockmask = inode->i_sb->s_blocksize - 1;
2174         loff_t final_size = pos + count;
2175
2176         if ((pos & blockmask) || (final_size & blockmask))
2177                 return 1;
2178         return 0;
2179 }
2180
2181 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2182                                             struct buffer_head **di_bh,
2183                                             int meta_level,
2184                                             int write_sem,
2185                                             int wait)
2186 {
2187         int ret = 0;
2188
2189         if (wait)
2190                 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2191         else
2192                 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2193         if (ret < 0)
2194                 goto out;
2195
2196         if (wait) {
2197                 if (write_sem)
2198                         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2199                 else
2200                         down_read(&OCFS2_I(inode)->ip_alloc_sem);
2201         } else {
2202                 if (write_sem)
2203                         ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2204                 else
2205                         ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2206
2207                 if (!ret) {
2208                         ret = -EAGAIN;
2209                         goto out_unlock;
2210                 }
2211         }
2212
2213         return ret;
2214
2215 out_unlock:
2216         brelse(*di_bh);
2217         *di_bh = NULL;
2218         ocfs2_inode_unlock(inode, meta_level);
2219 out:
2220         return ret;
2221 }
2222
2223 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2224                                                struct buffer_head **di_bh,
2225                                                int meta_level,
2226                                                int write_sem)
2227 {
2228         if (write_sem)
2229                 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2230         else
2231                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2232
2233         brelse(*di_bh);
2234         *di_bh = NULL;
2235
2236         if (meta_level >= 0)
2237                 ocfs2_inode_unlock(inode, meta_level);
2238 }
2239
2240 static int ocfs2_prepare_inode_for_write(struct file *file,
2241                                          loff_t pos, size_t count, int wait)
2242 {
2243         int ret = 0, meta_level = 0, overwrite_io = 0;
2244         int write_sem = 0;
2245         struct dentry *dentry = file->f_path.dentry;
2246         struct inode *inode = d_inode(dentry);
2247         struct buffer_head *di_bh = NULL;
2248         u32 cpos;
2249         u32 clusters;
2250
2251         /*
2252          * We start with a read level meta lock and only jump to an ex
2253          * if we need to make modifications here.
2254          */
2255         for(;;) {
2256                 ret = ocfs2_inode_lock_for_extent_tree(inode,
2257                                                        &di_bh,
2258                                                        meta_level,
2259                                                        write_sem,
2260                                                        wait);
2261                 if (ret < 0) {
2262                         if (ret != -EAGAIN)
2263                                 mlog_errno(ret);
2264                         goto out;
2265                 }
2266
2267                 /*
2268                  * Check if IO will overwrite allocated blocks in case
2269                  * IOCB_NOWAIT flag is set.
2270                  */
2271                 if (!wait && !overwrite_io) {
2272                         overwrite_io = 1;
2273
2274                         ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2275                         if (ret < 0) {
2276                                 if (ret != -EAGAIN)
2277                                         mlog_errno(ret);
2278                                 goto out_unlock;
2279                         }
2280                 }
2281
2282                 /* Clear suid / sgid if necessary. We do this here
2283                  * instead of later in the write path because
2284                  * remove_suid() calls ->setattr without any hint that
2285                  * we may have already done our cluster locking. Since
2286                  * ocfs2_setattr() *must* take cluster locks to
2287                  * proceed, this will lead us to recursively lock the
2288                  * inode. There's also the dinode i_size state which
2289                  * can be lost via setattr during extending writes (we
2290                  * set inode->i_size at the end of a write. */
2291                 if (should_remove_suid(dentry)) {
2292                         if (meta_level == 0) {
2293                                 ocfs2_inode_unlock_for_extent_tree(inode,
2294                                                                    &di_bh,
2295                                                                    meta_level,
2296                                                                    write_sem);
2297                                 meta_level = 1;
2298                                 continue;
2299                         }
2300
2301                         ret = ocfs2_write_remove_suid(inode);
2302                         if (ret < 0) {
2303                                 mlog_errno(ret);
2304                                 goto out_unlock;
2305                         }
2306                 }
2307
2308                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2309                 if (ret == 1) {
2310                         ocfs2_inode_unlock_for_extent_tree(inode,
2311                                                            &di_bh,
2312                                                            meta_level,
2313                                                            write_sem);
2314                         meta_level = 1;
2315                         write_sem = 1;
2316                         ret = ocfs2_inode_lock_for_extent_tree(inode,
2317                                                                &di_bh,
2318                                                                meta_level,
2319                                                                write_sem,
2320                                                                wait);
2321                         if (ret < 0) {
2322                                 if (ret != -EAGAIN)
2323                                         mlog_errno(ret);
2324                                 goto out;
2325                         }
2326
2327                         cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2328                         clusters =
2329                                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2330                         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2331                 }
2332
2333                 if (ret < 0) {
2334                         if (ret != -EAGAIN)
2335                                 mlog_errno(ret);
2336                         goto out_unlock;
2337                 }
2338
2339                 break;
2340         }
2341
2342 out_unlock:
2343         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2344                                             pos, count, wait);
2345
2346         ocfs2_inode_unlock_for_extent_tree(inode,
2347                                            &di_bh,
2348                                            meta_level,
2349                                            write_sem);
2350
2351 out:
2352         return ret;
2353 }
2354
2355 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2356                                     struct iov_iter *from)
2357 {
2358         int rw_level;
2359         ssize_t written = 0;
2360         ssize_t ret;
2361         size_t count = iov_iter_count(from);
2362         struct file *file = iocb->ki_filp;
2363         struct inode *inode = file_inode(file);
2364         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2365         int full_coherency = !(osb->s_mount_opt &
2366                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2367         void *saved_ki_complete = NULL;
2368         int append_write = ((iocb->ki_pos + count) >=
2369                         i_size_read(inode) ? 1 : 0);
2370         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2371         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2372
2373         trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2374                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2375                 file->f_path.dentry->d_name.len,
2376                 file->f_path.dentry->d_name.name,
2377                 (unsigned int)from->nr_segs);   /* GRRRRR */
2378
2379         if (!direct_io && nowait)
2380                 return -EOPNOTSUPP;
2381
2382         if (count == 0)
2383                 return 0;
2384
2385         if (nowait) {
2386                 if (!inode_trylock(inode))
2387                         return -EAGAIN;
2388         } else
2389                 inode_lock(inode);
2390
2391         /*
2392          * Concurrent O_DIRECT writes are allowed with
2393          * mount_option "coherency=buffered".
2394          * For append write, we must take rw EX.
2395          */
2396         rw_level = (!direct_io || full_coherency || append_write);
2397
2398         if (nowait)
2399                 ret = ocfs2_try_rw_lock(inode, rw_level);
2400         else
2401                 ret = ocfs2_rw_lock(inode, rw_level);
2402         if (ret < 0) {
2403                 if (ret != -EAGAIN)
2404                         mlog_errno(ret);
2405                 goto out_mutex;
2406         }
2407
2408         /*
2409          * O_DIRECT writes with "coherency=full" need to take EX cluster
2410          * inode_lock to guarantee coherency.
2411          */
2412         if (direct_io && full_coherency) {
2413                 /*
2414                  * We need to take and drop the inode lock to force
2415                  * other nodes to drop their caches.  Buffered I/O
2416                  * already does this in write_begin().
2417                  */
2418                 if (nowait)
2419                         ret = ocfs2_try_inode_lock(inode, NULL, 1);
2420                 else
2421                         ret = ocfs2_inode_lock(inode, NULL, 1);
2422                 if (ret < 0) {
2423                         if (ret != -EAGAIN)
2424                                 mlog_errno(ret);
2425                         goto out;
2426                 }
2427
2428                 ocfs2_inode_unlock(inode, 1);
2429         }
2430
2431         ret = generic_write_checks(iocb, from);
2432         if (ret <= 0) {
2433                 if (ret)
2434                         mlog_errno(ret);
2435                 goto out;
2436         }
2437         count = ret;
2438
2439         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2440         if (ret < 0) {
2441                 if (ret != -EAGAIN)
2442                         mlog_errno(ret);
2443                 goto out;
2444         }
2445
2446         if (direct_io && !is_sync_kiocb(iocb) &&
2447             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2448                 /*
2449                  * Make it a sync io if it's an unaligned aio.
2450                  */
2451                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2452         }
2453
2454         /* communicate with ocfs2_dio_end_io */
2455         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2456
2457         written = __generic_file_write_iter(iocb, from);
2458         /* buffered aio wouldn't have proper lock coverage today */
2459         BUG_ON(written == -EIOCBQUEUED && !direct_io);
2460
2461         /*
2462          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2463          * function pointer which is called when o_direct io completes so that
2464          * it can unlock our rw lock.
2465          * Unfortunately there are error cases which call end_io and others
2466          * that don't.  so we don't have to unlock the rw_lock if either an
2467          * async dio is going to do it in the future or an end_io after an
2468          * error has already done it.
2469          */
2470         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2471                 rw_level = -1;
2472         }
2473
2474         if (unlikely(written <= 0))
2475                 goto out;
2476
2477         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2478             IS_SYNC(inode)) {
2479                 ret = filemap_fdatawrite_range(file->f_mapping,
2480                                                iocb->ki_pos - written,
2481                                                iocb->ki_pos - 1);
2482                 if (ret < 0)
2483                         written = ret;
2484
2485                 if (!ret) {
2486                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2487                         if (ret < 0)
2488                                 written = ret;
2489                 }
2490
2491                 if (!ret)
2492                         ret = filemap_fdatawait_range(file->f_mapping,
2493                                                       iocb->ki_pos - written,
2494                                                       iocb->ki_pos - 1);
2495         }
2496
2497 out:
2498         if (saved_ki_complete)
2499                 xchg(&iocb->ki_complete, saved_ki_complete);
2500
2501         if (rw_level != -1)
2502                 ocfs2_rw_unlock(inode, rw_level);
2503
2504 out_mutex:
2505         inode_unlock(inode);
2506
2507         if (written)
2508                 ret = written;
2509         return ret;
2510 }
2511
2512 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2513                                    struct iov_iter *to)
2514 {
2515         int ret = 0, rw_level = -1, lock_level = 0;
2516         struct file *filp = iocb->ki_filp;
2517         struct inode *inode = file_inode(filp);
2518         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2519         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2520
2521         trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2522                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2523                         filp->f_path.dentry->d_name.len,
2524                         filp->f_path.dentry->d_name.name,
2525                         to->nr_segs);   /* GRRRRR */
2526
2527
2528         if (!inode) {
2529                 ret = -EINVAL;
2530                 mlog_errno(ret);
2531                 goto bail;
2532         }
2533
2534         if (!direct_io && nowait)
2535                 return -EOPNOTSUPP;
2536
2537         /*
2538          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2539          * need locks to protect pending reads from racing with truncate.
2540          */
2541         if (direct_io) {
2542                 if (nowait)
2543                         ret = ocfs2_try_rw_lock(inode, 0);
2544                 else
2545                         ret = ocfs2_rw_lock(inode, 0);
2546
2547                 if (ret < 0) {
2548                         if (ret != -EAGAIN)
2549                                 mlog_errno(ret);
2550                         goto bail;
2551                 }
2552                 rw_level = 0;
2553                 /* communicate with ocfs2_dio_end_io */
2554                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2555         }
2556
2557         /*
2558          * We're fine letting folks race truncates and extending
2559          * writes with read across the cluster, just like they can
2560          * locally. Hence no rw_lock during read.
2561          *
2562          * Take and drop the meta data lock to update inode fields
2563          * like i_size. This allows the checks down below
2564          * generic_file_read_iter() a chance of actually working.
2565          */
2566         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2567                                      !nowait);
2568         if (ret < 0) {
2569                 if (ret != -EAGAIN)
2570                         mlog_errno(ret);
2571                 goto bail;
2572         }
2573         ocfs2_inode_unlock(inode, lock_level);
2574
2575         ret = generic_file_read_iter(iocb, to);
2576         trace_generic_file_read_iter_ret(ret);
2577
2578         /* buffered aio wouldn't have proper lock coverage today */
2579         BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2580
2581         /* see ocfs2_file_write_iter */
2582         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2583                 rw_level = -1;
2584         }
2585
2586 bail:
2587         if (rw_level != -1)
2588                 ocfs2_rw_unlock(inode, rw_level);
2589
2590         return ret;
2591 }
2592
2593 /* Refer generic_file_llseek_unlocked() */
2594 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2595 {
2596         struct inode *inode = file->f_mapping->host;
2597         int ret = 0;
2598
2599         inode_lock(inode);
2600
2601         switch (whence) {
2602         case SEEK_SET:
2603                 break;
2604         case SEEK_END:
2605                 /* SEEK_END requires the OCFS2 inode lock for the file
2606                  * because it references the file's size.
2607                  */
2608                 ret = ocfs2_inode_lock(inode, NULL, 0);
2609                 if (ret < 0) {
2610                         mlog_errno(ret);
2611                         goto out;
2612                 }
2613                 offset += i_size_read(inode);
2614                 ocfs2_inode_unlock(inode, 0);
2615                 break;
2616         case SEEK_CUR:
2617                 if (offset == 0) {
2618                         offset = file->f_pos;
2619                         goto out;
2620                 }
2621                 offset += file->f_pos;
2622                 break;
2623         case SEEK_DATA:
2624         case SEEK_HOLE:
2625                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2626                 if (ret)
2627                         goto out;
2628                 break;
2629         default:
2630                 ret = -EINVAL;
2631                 goto out;
2632         }
2633
2634         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2635
2636 out:
2637         inode_unlock(inode);
2638         if (ret)
2639                 return ret;
2640         return offset;
2641 }
2642
2643 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2644                                      struct file *file_out, loff_t pos_out,
2645                                      loff_t len, unsigned int remap_flags)
2646 {
2647         struct inode *inode_in = file_inode(file_in);
2648         struct inode *inode_out = file_inode(file_out);
2649         struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2650         struct buffer_head *in_bh = NULL, *out_bh = NULL;
2651         bool same_inode = (inode_in == inode_out);
2652         loff_t remapped = 0;
2653         ssize_t ret;
2654
2655         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2656                 return -EINVAL;
2657         if (!ocfs2_refcount_tree(osb))
2658                 return -EOPNOTSUPP;
2659         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2660                 return -EROFS;
2661
2662         /* Lock both files against IO */
2663         ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2664         if (ret)
2665                 return ret;
2666
2667         /* Check file eligibility and prepare for block sharing. */
2668         ret = -EINVAL;
2669         if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2670             (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2671                 goto out_unlock;
2672
2673         ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2674                         &len, remap_flags);
2675         if (ret < 0 || len == 0)
2676                 goto out_unlock;
2677
2678         /* Lock out changes to the allocation maps and remap. */
2679         down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2680         if (!same_inode)
2681                 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2682                                   SINGLE_DEPTH_NESTING);
2683
2684         /* Zap any page cache for the destination file's range. */
2685         truncate_inode_pages_range(&inode_out->i_data,
2686                                    round_down(pos_out, PAGE_SIZE),
2687                                    round_up(pos_out + len, PAGE_SIZE) - 1);
2688
2689         remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2690                         inode_out, out_bh, pos_out, len);
2691         up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2692         if (!same_inode)
2693                 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2694         if (remapped < 0) {
2695                 ret = remapped;
2696                 mlog_errno(ret);
2697                 goto out_unlock;
2698         }
2699
2700         /*
2701          * Empty the extent map so that we may get the right extent
2702          * record from the disk.
2703          */
2704         ocfs2_extent_map_trunc(inode_in, 0);
2705         ocfs2_extent_map_trunc(inode_out, 0);
2706
2707         ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2708         if (ret) {
2709                 mlog_errno(ret);
2710                 goto out_unlock;
2711         }
2712
2713 out_unlock:
2714         ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2715         return remapped > 0 ? remapped : ret;
2716 }
2717
2718 const struct inode_operations ocfs2_file_iops = {
2719         .setattr        = ocfs2_setattr,
2720         .getattr        = ocfs2_getattr,
2721         .permission     = ocfs2_permission,
2722         .listxattr      = ocfs2_listxattr,
2723         .fiemap         = ocfs2_fiemap,
2724         .get_acl        = ocfs2_iop_get_acl,
2725         .set_acl        = ocfs2_iop_set_acl,
2726 };
2727
2728 const struct inode_operations ocfs2_special_file_iops = {
2729         .setattr        = ocfs2_setattr,
2730         .getattr        = ocfs2_getattr,
2731         .permission     = ocfs2_permission,
2732         .get_acl        = ocfs2_iop_get_acl,
2733         .set_acl        = ocfs2_iop_set_acl,
2734 };
2735
2736 /*
2737  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2738  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2739  */
2740 const struct file_operations ocfs2_fops = {
2741         .llseek         = ocfs2_file_llseek,
2742         .mmap           = ocfs2_mmap,
2743         .fsync          = ocfs2_sync_file,
2744         .release        = ocfs2_file_release,
2745         .open           = ocfs2_file_open,
2746         .read_iter      = ocfs2_file_read_iter,
2747         .write_iter     = ocfs2_file_write_iter,
2748         .unlocked_ioctl = ocfs2_ioctl,
2749 #ifdef CONFIG_COMPAT
2750         .compat_ioctl   = ocfs2_compat_ioctl,
2751 #endif
2752         .lock           = ocfs2_lock,
2753         .flock          = ocfs2_flock,
2754         .splice_read    = generic_file_splice_read,
2755         .splice_write   = iter_file_splice_write,
2756         .fallocate      = ocfs2_fallocate,
2757         .remap_file_range = ocfs2_remap_file_range,
2758 };
2759
2760 const struct file_operations ocfs2_dops = {
2761         .llseek         = generic_file_llseek,
2762         .read           = generic_read_dir,
2763         .iterate        = ocfs2_readdir,
2764         .fsync          = ocfs2_sync_file,
2765         .release        = ocfs2_dir_release,
2766         .open           = ocfs2_dir_open,
2767         .unlocked_ioctl = ocfs2_ioctl,
2768 #ifdef CONFIG_COMPAT
2769         .compat_ioctl   = ocfs2_compat_ioctl,
2770 #endif
2771         .lock           = ocfs2_lock,
2772         .flock          = ocfs2_flock,
2773 };
2774
2775 /*
2776  * POSIX-lockless variants of our file_operations.
2777  *
2778  * These will be used if the underlying cluster stack does not support
2779  * posix file locking, if the user passes the "localflocks" mount
2780  * option, or if we have a local-only fs.
2781  *
2782  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2783  * so we still want it in the case of no stack support for
2784  * plocks. Internally, it will do the right thing when asked to ignore
2785  * the cluster.
2786  */
2787 const struct file_operations ocfs2_fops_no_plocks = {
2788         .llseek         = ocfs2_file_llseek,
2789         .mmap           = ocfs2_mmap,
2790         .fsync          = ocfs2_sync_file,
2791         .release        = ocfs2_file_release,
2792         .open           = ocfs2_file_open,
2793         .read_iter      = ocfs2_file_read_iter,
2794         .write_iter     = ocfs2_file_write_iter,
2795         .unlocked_ioctl = ocfs2_ioctl,
2796 #ifdef CONFIG_COMPAT
2797         .compat_ioctl   = ocfs2_compat_ioctl,
2798 #endif
2799         .flock          = ocfs2_flock,
2800         .splice_read    = generic_file_splice_read,
2801         .splice_write   = iter_file_splice_write,
2802         .fallocate      = ocfs2_fallocate,
2803         .remap_file_range = ocfs2_remap_file_range,
2804 };
2805
2806 const struct file_operations ocfs2_dops_no_plocks = {
2807         .llseek         = generic_file_llseek,
2808         .read           = generic_read_dir,
2809         .iterate        = ocfs2_readdir,
2810         .fsync          = ocfs2_sync_file,
2811         .release        = ocfs2_dir_release,
2812         .open           = ocfs2_dir_open,
2813         .unlocked_ioctl = ocfs2_ioctl,
2814 #ifdef CONFIG_COMPAT
2815         .compat_ioctl   = ocfs2_compat_ioctl,
2816 #endif
2817         .flock          = ocfs2_flock,
2818 };