fs/ocfs2/file.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * file.c
   5  *
   6  * File open, close, extend, truncate
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/capability.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/uio.h>
  33 #include <linux/sched.h>
  34 #include <linux/splice.h>
  35 #include <linux/mount.h>
  36 #include <linux/writeback.h>
  37 #include <linux/falloc.h>
  38 #include <linux/quotaops.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/backing-dev.h>
  41
  42 #include <cluster/masklog.h>
  43
  44 #include "ocfs2.h"
  45
  46 #include "alloc.h"
  47 #include "aops.h"
  48 #include "dir.h"
  49 #include "dlmglue.h"
  50 #include "extent_map.h"
  51 #include "file.h"
  52 #include "sysfile.h"
  53 #include "inode.h"
  54 #include "ioctl.h"
  55 #include "journal.h"
  56 #include "locks.h"
  57 #include "mmap.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "xattr.h"
  61 #include "acl.h"
  62 #include "quota.h"
  63 #include "refcounttree.h"
  64 #include "ocfs2_trace.h"
  65
  66 #include "buffer_head_io.h"
  67
  68 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  69 {
  70         struct ocfs2_file_private *fp;
  71
  72         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  73         if (!fp)
  74                 return -ENOMEM;
  75
  76         fp->fp_file = file;
  77         mutex_init(&fp->fp_mutex);
  78         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  79         file->private_data = fp;
  80
  81         return 0;
  82 }
  83
  84 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  85 {
  86         struct ocfs2_file_private *fp = file->private_data;
  87         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  88
  89         if (fp) {
  90                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  91                 ocfs2_lock_res_free(&fp->fp_flock);
  92                 kfree(fp);
  93                 file->private_data = NULL;
  94         }
  95 }
  96
  97 static int ocfs2_file_open(struct inode *inode, struct file *file)
  98 {
  99         int status;
 100         int mode = file->f_flags;
 101         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 102
 103         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
 104                               (unsigned long long)oi->ip_blkno,
 105                               file->f_path.dentry->d_name.len,
 106                               file->f_path.dentry->d_name.name, mode);
 107
 108         if (file->f_mode & FMODE_WRITE) {
 109                 status = dquot_initialize(inode);
 110                 if (status)
 111                         goto leave;
 112         }
 113
 114         spin_lock(&oi->ip_lock);
 115
 116         /* Check that the inode hasn't been wiped from disk by another
 117          * node. If it hasn't then we're safe as long as we hold the
 118          * spin lock until our increment of open count. */
 119         if (oi->ip_flags & OCFS2_INODE_DELETED) {
 120                 spin_unlock(&oi->ip_lock);
 121
 122                 status = -ENOENT;
 123                 goto leave;
 124         }
 125
 126         if (mode & O_DIRECT)
 127                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 128
 129         oi->ip_open_count++;
 130         spin_unlock(&oi->ip_lock);
 131
 132         status = ocfs2_init_file_private(inode, file);
 133         if (status) {
 134                 /*
 135                  * We want to set open count back if we're failing the
 136                  * open.
 137                  */
 138                 spin_lock(&oi->ip_lock);
 139                 oi->ip_open_count--;
 140                 spin_unlock(&oi->ip_lock);
 141         }
 142
 143         file->f_mode |= FMODE_NOWAIT;
 144
 145 leave:
 146         return status;
 147 }
 148
 149 static int ocfs2_file_release(struct inode *inode, struct file *file)
 150 {
 151         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 152
 153         spin_lock(&oi->ip_lock);
 154         if (!--oi->ip_open_count)
 155                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 156
 157         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 158                                  oi->ip_blkno,
 159                                  file->f_path.dentry->d_name.len,
 160                                  file->f_path.dentry->d_name.name,
 161                                  oi->ip_open_count);
 162         spin_unlock(&oi->ip_lock);
 163
 164         ocfs2_free_file_private(inode, file);
 165
 166         return 0;
 167 }
 168
 169 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 170 {
 171         return ocfs2_init_file_private(inode, file);
 172 }
 173
 174 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 175 {
 176         ocfs2_free_file_private(inode, file);
 177         return 0;
 178 }
 179
 180 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 181                            int datasync)
 182 {
 183         int err = 0;
 184         struct inode *inode = file->f_mapping->host;
 185         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 186         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 187         journal_t *journal = osb->journal->j_journal;
 188         int ret;
 189         tid_t commit_tid;
 190         bool needs_barrier = false;
 191
 192         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 193                               oi->ip_blkno,
 194                               file->f_path.dentry->d_name.len,
 195                               file->f_path.dentry->d_name.name,
 196                               (unsigned long long)datasync);
 197
 198         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 199                 return -EROFS;
 200
 201         err = file_write_and_wait_range(file, start, end);
 202         if (err)
 203                 return err;
 204
 205         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 206         if (journal->j_flags & JBD2_BARRIER &&
 207             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 208                 needs_barrier = true;
 209         err = jbd2_complete_transaction(journal, commit_tid);
 210         if (needs_barrier) {
 211                 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 212                 if (!err)
 213                         err = ret;
 214         }
 215
 216         if (err)
 217                 mlog_errno(err);
 218
 219         return (err < 0) ? -EIO : 0;
 220 }
 221
 222 int ocfs2_should_update_atime(struct inode *inode,
 223                               struct vfsmount *vfsmnt)
 224 {
 225         struct timespec64 now;
 226         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 227
 228         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 229                 return 0;
 230
 231         if ((inode->i_flags & S_NOATIME) ||
 232             ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
 233                 return 0;
 234
 235         /*
 236          * We can be called with no vfsmnt structure - NFSD will
 237          * sometimes do this.
 238          *
 239          * Note that our action here is different than touch_atime() -
 240          * if we can't tell whether this is a noatime mount, then we
 241          * don't know whether to trust the value of s_atime_quantum.
 242          */
 243         if (vfsmnt == NULL)
 244                 return 0;
 245
 246         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 247             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 248                 return 0;
 249
 250         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 251                 if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 252                     (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 253                         return 1;
 254
 255                 return 0;
 256         }
 257
 258         now = current_time(inode);
 259         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 260                 return 0;
 261         else
 262                 return 1;
 263 }
 264
 265 int ocfs2_update_inode_atime(struct inode *inode,
 266                              struct buffer_head *bh)
 267 {
 268         int ret;
 269         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 270         handle_t *handle;
 271         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 272
 273         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 274         if (IS_ERR(handle)) {
 275                 ret = PTR_ERR(handle);
 276                 mlog_errno(ret);
 277                 goto out;
 278         }
 279
 280         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 281                                       OCFS2_JOURNAL_ACCESS_WRITE);
 282         if (ret) {
 283                 mlog_errno(ret);
 284                 goto out_commit;
 285         }
 286
 287         /*
 288          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 289          * have i_mutex to guard against concurrent changes to other
 290          * inode fields.
 291          */
 292         inode->i_atime = current_time(inode);
 293         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 294         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 295         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 296         ocfs2_journal_dirty(handle, bh);
 297
 298 out_commit:
 299         ocfs2_commit_trans(osb, handle);
 300 out:
 301         return ret;
 302 }
 303
 304 int ocfs2_set_inode_size(handle_t *handle,
 305                                 struct inode *inode,
 306                                 struct buffer_head *fe_bh,
 307                                 u64 new_i_size)
 308 {
 309         int status;
 310
 311         i_size_write(inode, new_i_size);
 312         inode->i_blocks = ocfs2_inode_sector_count(inode);
 313         inode->i_ctime = inode->i_mtime = current_time(inode);
 314
 315         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 316         if (status < 0) {
 317                 mlog_errno(status);
 318                 goto bail;
 319         }
 320
 321 bail:
 322         return status;
 323 }
 324
 325 int ocfs2_simple_size_update(struct inode *inode,
 326                              struct buffer_head *di_bh,
 327                              u64 new_i_size)
 328 {
 329         int ret;
 330         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 331         handle_t *handle = NULL;
 332
 333         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 334         if (IS_ERR(handle)) {
 335                 ret = PTR_ERR(handle);
 336                 mlog_errno(ret);
 337                 goto out;
 338         }
 339
 340         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 341                                    new_i_size);
 342         if (ret < 0)
 343                 mlog_errno(ret);
 344
 345         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 346         ocfs2_commit_trans(osb, handle);
 347 out:
 348         return ret;
 349 }
 350
 351 static int ocfs2_cow_file_pos(struct inode *inode,
 352                               struct buffer_head *fe_bh,
 353                               u64 offset)
 354 {
 355         int status;
 356         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 357         unsigned int num_clusters = 0;
 358         unsigned int ext_flags = 0;
 359
 360         /*
 361          * If the new offset is aligned to the range of the cluster, there is
 362          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 363          * CoW either.
 364          */
 365         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 366                 return 0;
 367
 368         status = ocfs2_get_clusters(inode, cpos, &phys,
 369                                     &num_clusters, &ext_flags);
 370         if (status) {
 371                 mlog_errno(status);
 372                 goto out;
 373         }
 374
 375         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 376                 goto out;
 377
 378         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 379
 380 out:
 381         return status;
 382 }
 383
 384 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 385                                      struct inode *inode,
 386                                      struct buffer_head *fe_bh,
 387                                      u64 new_i_size)
 388 {
 389         int status;
 390         handle_t *handle;
 391         struct ocfs2_dinode *di;
 392         u64 cluster_bytes;
 393
 394         /*
 395          * We need to CoW the cluster contains the offset if it is reflinked
 396          * since we will call ocfs2_zero_range_for_truncate later which will
 397          * write "0" from offset to the end of the cluster.
 398          */
 399         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 400         if (status) {
 401                 mlog_errno(status);
 402                 return status;
 403         }
 404
 405         /* TODO: This needs to actually orphan the inode in this
 406          * transaction. */
 407
 408         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 409         if (IS_ERR(handle)) {
 410                 status = PTR_ERR(handle);
 411                 mlog_errno(status);
 412                 goto out;
 413         }
 414
 415         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 416                                          OCFS2_JOURNAL_ACCESS_WRITE);
 417         if (status < 0) {
 418                 mlog_errno(status);
 419                 goto out_commit;
 420         }
 421
 422         /*
 423          * Do this before setting i_size.
 424          */
 425         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 426         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 427                                                cluster_bytes);
 428         if (status) {
 429                 mlog_errno(status);
 430                 goto out_commit;
 431         }
 432
 433         i_size_write(inode, new_i_size);
 434         inode->i_ctime = inode->i_mtime = current_time(inode);
 435
 436         di = (struct ocfs2_dinode *) fe_bh->b_data;
 437         di->i_size = cpu_to_le64(new_i_size);
 438         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 439         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 440         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 441
 442         ocfs2_journal_dirty(handle, fe_bh);
 443
 444 out_commit:
 445         ocfs2_commit_trans(osb, handle);
 446 out:
 447         return status;
 448 }
 449
 450 int ocfs2_truncate_file(struct inode *inode,
 451                                struct buffer_head *di_bh,
 452                                u64 new_i_size)
 453 {
 454         int status = 0;
 455         struct ocfs2_dinode *fe = NULL;
 456         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 457
 458         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 459          * already validated it */
 460         fe = (struct ocfs2_dinode *) di_bh->b_data;
 461
 462         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 463                                   (unsigned long long)le64_to_cpu(fe->i_size),
 464                                   (unsigned long long)new_i_size);
 465
 466         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 467                         "Inode %llu, inode i_size = %lld != di "
 468                         "i_size = %llu, i_flags = 0x%x\n",
 469                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 470                         i_size_read(inode),
 471                         (unsigned long long)le64_to_cpu(fe->i_size),
 472                         le32_to_cpu(fe->i_flags));
 473
 474         if (new_i_size > le64_to_cpu(fe->i_size)) {
 475                 trace_ocfs2_truncate_file_error(
 476                         (unsigned long long)le64_to_cpu(fe->i_size),
 477                         (unsigned long long)new_i_size);
 478                 status = -EINVAL;
 479                 mlog_errno(status);
 480                 goto bail;
 481         }
 482
 483         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 484
 485         ocfs2_resv_discard(&osb->osb_la_resmap,
 486                            &OCFS2_I(inode)->ip_la_data_resv);
 487
 488         /*
 489          * The inode lock forced other nodes to sync and drop their
 490          * pages, which (correctly) happens even if we have a truncate
 491          * without allocation change - ocfs2 cluster sizes can be much
 492          * greater than page size, so we have to truncate them
 493          * anyway.
 494          */
 495
 496         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 497                 unmap_mapping_range(inode->i_mapping,
 498                                     new_i_size + PAGE_SIZE - 1, 0, 1);
 499                 truncate_inode_pages(inode->i_mapping, new_i_size);
 500                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 501                                                i_size_read(inode), 1);
 502                 if (status)
 503                         mlog_errno(status);
 504
 505                 goto bail_unlock_sem;
 506         }
 507
 508         /* alright, we're going to need to do a full blown alloc size
 509          * change. Orphan the inode so that recovery can complete the
 510          * truncate if necessary. This does the task of marking
 511          * i_size. */
 512         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 513         if (status < 0) {
 514                 mlog_errno(status);
 515                 goto bail_unlock_sem;
 516         }
 517
 518         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 519         truncate_inode_pages(inode->i_mapping, new_i_size);
 520
 521         status = ocfs2_commit_truncate(osb, inode, di_bh);
 522         if (status < 0) {
 523                 mlog_errno(status);
 524                 goto bail_unlock_sem;
 525         }
 526
 527         /* TODO: orphan dir cleanup here. */
 528 bail_unlock_sem:
 529         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 530
 531 bail:
 532         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 533                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 534
 535         return status;
 536 }
 537
 538 /*
 539  * extend file allocation only here.
 540  * we'll update all the disk stuff, and oip->alloc_size
 541  *
 542  * expect stuff to be locked, a transaction started and enough data /
 543  * metadata reservations in the contexts.
 544  *
 545  * Will return -EAGAIN, and a reason if a restart is needed.
 546  * If passed in, *reason will always be set, even in error.
 547  */
 548 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 549                          struct inode *inode,
 550                          u32 *logical_offset,
 551                          u32 clusters_to_add,
 552                          int mark_unwritten,
 553                          struct buffer_head *fe_bh,
 554                          handle_t *handle,
 555                          struct ocfs2_alloc_context *data_ac,
 556                          struct ocfs2_alloc_context *meta_ac,
 557                          enum ocfs2_alloc_restarted *reason_ret)
 558 {
 559         int ret;
 560         struct ocfs2_extent_tree et;
 561
 562         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 563         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 564                                           clusters_to_add, mark_unwritten,
 565                                           data_ac, meta_ac, reason_ret);
 566
 567         return ret;
 568 }
 569
 570 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 571                                    u32 clusters_to_add, int mark_unwritten)
 572 {
 573         int status = 0;
 574         int restart_func = 0;
 575         int credits;
 576         u32 prev_clusters;
 577         struct buffer_head *bh = NULL;
 578         struct ocfs2_dinode *fe = NULL;
 579         handle_t *handle = NULL;
 580         struct ocfs2_alloc_context *data_ac = NULL;
 581         struct ocfs2_alloc_context *meta_ac = NULL;
 582         enum ocfs2_alloc_restarted why = RESTART_NONE;
 583         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 584         struct ocfs2_extent_tree et;
 585         int did_quota = 0;
 586
 587         /*
 588          * Unwritten extent only exists for file systems which
 589          * support holes.
 590          */
 591         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 592
 593         status = ocfs2_read_inode_block(inode, &bh);
 594         if (status < 0) {
 595                 mlog_errno(status);
 596                 goto leave;
 597         }
 598         fe = (struct ocfs2_dinode *) bh->b_data;
 599
 600 restart_all:
 601         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 602
 603         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 604         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 605                                        &data_ac, &meta_ac);
 606         if (status) {
 607                 mlog_errno(status);
 608                 goto leave;
 609         }
 610
 611         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 612         handle = ocfs2_start_trans(osb, credits);
 613         if (IS_ERR(handle)) {
 614                 status = PTR_ERR(handle);
 615                 handle = NULL;
 616                 mlog_errno(status);
 617                 goto leave;
 618         }
 619
 620 restarted_transaction:
 621         trace_ocfs2_extend_allocation(
 622                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 623                 (unsigned long long)i_size_read(inode),
 624                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 625                 why, restart_func);
 626
 627         status = dquot_alloc_space_nodirty(inode,
 628                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 629         if (status)
 630                 goto leave;
 631         did_quota = 1;
 632
 633         /* reserve a write to the file entry early on - that we if we
 634          * run out of credits in the allocation path, we can still
 635          * update i_size. */
 636         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 637                                          OCFS2_JOURNAL_ACCESS_WRITE);
 638         if (status < 0) {
 639                 mlog_errno(status);
 640                 goto leave;
 641         }
 642
 643         prev_clusters = OCFS2_I(inode)->ip_clusters;
 644
 645         status = ocfs2_add_inode_data(osb,
 646                                       inode,
 647                                       &logical_start,
 648                                       clusters_to_add,
 649                                       mark_unwritten,
 650                                       bh,
 651                                       handle,
 652                                       data_ac,
 653                                       meta_ac,
 654                                       &why);
 655         if ((status < 0) && (status != -EAGAIN)) {
 656                 if (status != -ENOSPC)
 657                         mlog_errno(status);
 658                 goto leave;
 659         }
 660         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 661         ocfs2_journal_dirty(handle, bh);
 662
 663         spin_lock(&OCFS2_I(inode)->ip_lock);
 664         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 665         spin_unlock(&OCFS2_I(inode)->ip_lock);
 666         /* Release unused quota reservation */
 667         dquot_free_space(inode,
 668                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 669         did_quota = 0;
 670
 671         if (why != RESTART_NONE && clusters_to_add) {
 672                 if (why == RESTART_META) {
 673                         restart_func = 1;
 674                         status = 0;
 675                 } else {
 676                         BUG_ON(why != RESTART_TRANS);
 677
 678                         status = ocfs2_allocate_extend_trans(handle, 1);
 679                         if (status < 0) {
 680                                 /* handle still has to be committed at
 681                                  * this point. */
 682                                 status = -ENOMEM;
 683                                 mlog_errno(status);
 684                                 goto leave;
 685                         }
 686                         goto restarted_transaction;
 687                 }
 688         }
 689
 690         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 691              le32_to_cpu(fe->i_clusters),
 692              (unsigned long long)le64_to_cpu(fe->i_size),
 693              OCFS2_I(inode)->ip_clusters,
 694              (unsigned long long)i_size_read(inode));
 695
 696 leave:
 697         if (status < 0 && did_quota)
 698                 dquot_free_space(inode,
 699                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 700         if (handle) {
 701                 ocfs2_commit_trans(osb, handle);
 702                 handle = NULL;
 703         }
 704         if (data_ac) {
 705                 ocfs2_free_alloc_context(data_ac);
 706                 data_ac = NULL;
 707         }
 708         if (meta_ac) {
 709                 ocfs2_free_alloc_context(meta_ac);
 710                 meta_ac = NULL;
 711         }
 712         if ((!status) && restart_func) {
 713                 restart_func = 0;
 714                 goto restart_all;
 715         }
 716         brelse(bh);
 717         bh = NULL;
 718
 719         return status;
 720 }
 721
 722 /*
 723  * While a write will already be ordering the data, a truncate will not.
 724  * Thus, we need to explicitly order the zeroed pages.
 725  */
 726 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 727                                                 struct buffer_head *di_bh)
 728 {
 729         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 730         handle_t *handle = NULL;
 731         int ret = 0;
 732
 733         if (!ocfs2_should_order_data(inode))
 734                 goto out;
 735
 736         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 737         if (IS_ERR(handle)) {
 738                 ret = -ENOMEM;
 739                 mlog_errno(ret);
 740                 goto out;
 741         }
 742
 743         ret = ocfs2_jbd2_file_inode(handle, inode);
 744         if (ret < 0) {
 745                 mlog_errno(ret);
 746                 goto out;
 747         }
 748
 749         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 750                                       OCFS2_JOURNAL_ACCESS_WRITE);
 751         if (ret)
 752                 mlog_errno(ret);
 753         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 754
 755 out:
 756         if (ret) {
 757                 if (!IS_ERR(handle))
 758                         ocfs2_commit_trans(osb, handle);
 759                 handle = ERR_PTR(ret);
 760         }
 761         return handle;
 762 }
 763
 764 /* Some parts of this taken from generic_cont_expand, which turned out
 765  * to be too fragile to do exactly what we need without us having to
 766  * worry about recursive locking in ->write_begin() and ->write_end(). */
 767 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 768                                  u64 abs_to, struct buffer_head *di_bh)
 769 {
 770         struct address_space *mapping = inode->i_mapping;
 771         struct page *page;
 772         unsigned long index = abs_from >> PAGE_SHIFT;
 773         handle_t *handle;
 774         int ret = 0;
 775         unsigned zero_from, zero_to, block_start, block_end;
 776         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 777
 778         BUG_ON(abs_from >= abs_to);
 779         BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 780         BUG_ON(abs_from & (inode->i_blkbits - 1));
 781
 782         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
 783         if (IS_ERR(handle)) {
 784                 ret = PTR_ERR(handle);
 785                 goto out;
 786         }
 787
 788         page = find_or_create_page(mapping, index, GFP_NOFS);
 789         if (!page) {
 790                 ret = -ENOMEM;
 791                 mlog_errno(ret);
 792                 goto out_commit_trans;
 793         }
 794
 795         /* Get the offsets within the page that we want to zero */
 796         zero_from = abs_from & (PAGE_SIZE - 1);
 797         zero_to = abs_to & (PAGE_SIZE - 1);
 798         if (!zero_to)
 799                 zero_to = PAGE_SIZE;
 800
 801         trace_ocfs2_write_zero_page(
 802                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 803                         (unsigned long long)abs_from,
 804                         (unsigned long long)abs_to,
 805                         index, zero_from, zero_to);
 806
 807         /* We know that zero_from is block aligned */
 808         for (block_start = zero_from; block_start < zero_to;
 809              block_start = block_end) {
 810                 block_end = block_start + i_blocksize(inode);
 811
 812                 /*
 813                  * block_start is block-aligned.  Bump it by one to force
 814                  * __block_write_begin and block_commit_write to zero the
 815                  * whole block.
 816                  */
 817                 ret = __block_write_begin(page, block_start + 1, 0,
 818                                           ocfs2_get_block);
 819                 if (ret < 0) {
 820                         mlog_errno(ret);
 821                         goto out_unlock;
 822                 }
 823
 824
 825                 /* must not update i_size! */
 826                 ret = block_commit_write(page, block_start + 1,
 827                                          block_start + 1);
 828                 if (ret < 0)
 829                         mlog_errno(ret);
 830                 else
 831                         ret = 0;
 832         }
 833
 834         /*
 835          * fs-writeback will release the dirty pages without page lock
 836          * whose offset are over inode size, the release happens at
 837          * block_write_full_page().
 838          */
 839         i_size_write(inode, abs_to);
 840         inode->i_blocks = ocfs2_inode_sector_count(inode);
 841         di->i_size = cpu_to_le64((u64)i_size_read(inode));
 842         inode->i_mtime = inode->i_ctime = current_time(inode);
 843         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 844         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 845         di->i_mtime_nsec = di->i_ctime_nsec;
 846         if (handle) {
 847                 ocfs2_journal_dirty(handle, di_bh);
 848                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 849         }
 850
 851 out_unlock:
 852         unlock_page(page);
 853         put_page(page);
 854 out_commit_trans:
 855         if (handle)
 856                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 857 out:
 858         return ret;
 859 }
 860
 861 /*
 862  * Find the next range to zero.  We do this in terms of bytes because
 863  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 864  * pagecache.  We may return multiple extents.
 865  *
 866  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 867  * needs to be zeroed.  range_start and range_end return the next zeroing
 868  * range.  A subsequent call should pass the previous range_end as its
 869  * zero_start.  If range_end is 0, there's nothing to do.
 870  *
 871  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 872  */
 873 static int ocfs2_zero_extend_get_range(struct inode *inode,
 874                                        struct buffer_head *di_bh,
 875                                        u64 zero_start, u64 zero_end,
 876                                        u64 *range_start, u64 *range_end)
 877 {
 878         int rc = 0, needs_cow = 0;
 879         u32 p_cpos, zero_clusters = 0;
 880         u32 zero_cpos =
 881                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 882         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 883         unsigned int num_clusters = 0;
 884         unsigned int ext_flags = 0;
 885
 886         while (zero_cpos < last_cpos) {
 887                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 888                                         &num_clusters, &ext_flags);
 889                 if (rc) {
 890                         mlog_errno(rc);
 891                         goto out;
 892                 }
 893
 894                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 895                         zero_clusters = num_clusters;
 896                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 897                                 needs_cow = 1;
 898                         break;
 899                 }
 900
 901                 zero_cpos += num_clusters;
 902         }
 903         if (!zero_clusters) {
 904                 *range_end = 0;
 905                 goto out;
 906         }
 907
 908         while ((zero_cpos + zero_clusters) < last_cpos) {
 909                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 910                                         &p_cpos, &num_clusters,
 911                                         &ext_flags);
 912                 if (rc) {
 913                         mlog_errno(rc);
 914                         goto out;
 915                 }
 916
 917                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 918                         break;
 919                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 920                         needs_cow = 1;
 921                 zero_clusters += num_clusters;
 922         }
 923         if ((zero_cpos + zero_clusters) > last_cpos)
 924                 zero_clusters = last_cpos - zero_cpos;
 925
 926         if (needs_cow) {
 927                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 928                                         zero_clusters, UINT_MAX);
 929                 if (rc) {
 930                         mlog_errno(rc);
 931                         goto out;
 932                 }
 933         }
 934
 935         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 936         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 937                                              zero_cpos + zero_clusters);
 938
 939 out:
 940         return rc;
 941 }
 942
 943 /*
 944  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 945  * has made sure that the entire range needs zeroing.
 946  */
 947 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 948                                    u64 range_end, struct buffer_head *di_bh)
 949 {
 950         int rc = 0;
 951         u64 next_pos;
 952         u64 zero_pos = range_start;
 953
 954         trace_ocfs2_zero_extend_range(
 955                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 956                         (unsigned long long)range_start,
 957                         (unsigned long long)range_end);
 958         BUG_ON(range_start >= range_end);
 959
 960         while (zero_pos < range_end) {
 961                 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 962                 if (next_pos > range_end)
 963                         next_pos = range_end;
 964                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 965                 if (rc < 0) {
 966                         mlog_errno(rc);
 967                         break;
 968                 }
 969                 zero_pos = next_pos;
 970
 971                 /*
 972                  * Very large extends have the potential to lock up
 973                  * the cpu for extended periods of time.
 974                  */
 975                 cond_resched();
 976         }
 977
 978         return rc;
 979 }
 980
 981 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 982                       loff_t zero_to_size)
 983 {
 984         int ret = 0;
 985         u64 zero_start, range_start = 0, range_end = 0;
 986         struct super_block *sb = inode->i_sb;
 987
 988         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 989         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 990                                 (unsigned long long)zero_start,
 991                                 (unsigned long long)i_size_read(inode));
 992         while (zero_start < zero_to_size) {
 993                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 994                                                   zero_to_size,
 995                                                   &range_start,
 996                                                   &range_end);
 997                 if (ret) {
 998                         mlog_errno(ret);
 999                         break;
1000                 }
1001                 if (!range_end)
1002                         break;
1003                 /* Trim the ends */
1004                 if (range_start < zero_start)
1005                         range_start = zero_start;
1006                 if (range_end > zero_to_size)
1007                         range_end = zero_to_size;
1008
1009                 ret = ocfs2_zero_extend_range(inode, range_start,
1010                                               range_end, di_bh);
1011                 if (ret) {
1012                         mlog_errno(ret);
1013                         break;
1014                 }
1015                 zero_start = range_end;
1016         }
1017
1018         return ret;
1019 }
1020
1021 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1022                           u64 new_i_size, u64 zero_to)
1023 {
1024         int ret;
1025         u32 clusters_to_add;
1026         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1027
1028         /*
1029          * Only quota files call this without a bh, and they can't be
1030          * refcounted.
1031          */
1032         BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1033         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1034
1035         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1036         if (clusters_to_add < oi->ip_clusters)
1037                 clusters_to_add = 0;
1038         else
1039                 clusters_to_add -= oi->ip_clusters;
1040
1041         if (clusters_to_add) {
1042                 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1043                                               clusters_to_add, 0);
1044                 if (ret) {
1045                         mlog_errno(ret);
1046                         goto out;
1047                 }
1048         }
1049
1050         /*
1051          * Call this even if we don't add any clusters to the tree. We
1052          * still need to zero the area between the old i_size and the
1053          * new i_size.
1054          */
1055         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1056         if (ret < 0)
1057                 mlog_errno(ret);
1058
1059 out:
1060         return ret;
1061 }
1062
1063 static int ocfs2_extend_file(struct inode *inode,
1064                              struct buffer_head *di_bh,
1065                              u64 new_i_size)
1066 {
1067         int ret = 0;
1068         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1069
1070         BUG_ON(!di_bh);
1071
1072         /* setattr sometimes calls us like this. */
1073         if (new_i_size == 0)
1074                 goto out;
1075
1076         if (i_size_read(inode) == new_i_size)
1077                 goto out;
1078         BUG_ON(new_i_size < i_size_read(inode));
1079
1080         /*
1081          * The alloc sem blocks people in read/write from reading our
1082          * allocation until we're done changing it. We depend on
1083          * i_mutex to block other extend/truncate calls while we're
1084          * here.  We even have to hold it for sparse files because there
1085          * might be some tail zeroing.
1086          */
1087         down_write(&oi->ip_alloc_sem);
1088
1089         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1090                 /*
1091                  * We can optimize small extends by keeping the inodes
1092                  * inline data.
1093                  */
1094                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1095                         up_write(&oi->ip_alloc_sem);
1096                         goto out_update_size;
1097                 }
1098
1099                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1100                 if (ret) {
1101                         up_write(&oi->ip_alloc_sem);
1102                         mlog_errno(ret);
1103                         goto out;
1104                 }
1105         }
1106
1107         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1108                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1109         else
1110                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1111                                             new_i_size);
1112
1113         up_write(&oi->ip_alloc_sem);
1114
1115         if (ret < 0) {
1116                 mlog_errno(ret);
1117                 goto out;
1118         }
1119
1120 out_update_size:
1121         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1122         if (ret < 0)
1123                 mlog_errno(ret);
1124
1125 out:
1126         return ret;
1127 }
1128
1129 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1130 {
1131         int status = 0, size_change;
1132         int inode_locked = 0;
1133         struct inode *inode = d_inode(dentry);
1134         struct super_block *sb = inode->i_sb;
1135         struct ocfs2_super *osb = OCFS2_SB(sb);
1136         struct buffer_head *bh = NULL;
1137         handle_t *handle = NULL;
1138         struct dquot *transfer_to[MAXQUOTAS] = { };
1139         int qtype;
1140         int had_lock;
1141         struct ocfs2_lock_holder oh;
1142
1143         trace_ocfs2_setattr(inode, dentry,
1144                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1145                             dentry->d_name.len, dentry->d_name.name,
1146                             attr->ia_valid, attr->ia_mode,
1147                             from_kuid(&init_user_ns, attr->ia_uid),
1148                             from_kgid(&init_user_ns, attr->ia_gid));
1149
1150         /* ensuring we don't even attempt to truncate a symlink */
1151         if (S_ISLNK(inode->i_mode))
1152                 attr->ia_valid &= ~ATTR_SIZE;
1153
1154 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1155                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1156         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1157                 return 0;
1158
1159         status = setattr_prepare(dentry, attr);
1160         if (status)
1161                 return status;
1162
1163         if (is_quota_modification(inode, attr)) {
1164                 status = dquot_initialize(inode);
1165                 if (status)
1166                         return status;
1167         }
1168         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1169         if (size_change) {
1170                 /*
1171                  * Here we should wait dio to finish before inode lock
1172                  * to avoid a deadlock between ocfs2_setattr() and
1173                  * ocfs2_dio_end_io_write()
1174                  */
1175                 inode_dio_wait(inode);
1176
1177                 status = ocfs2_rw_lock(inode, 1);
1178                 if (status < 0) {
1179                         mlog_errno(status);
1180                         goto bail;
1181                 }
1182         }
1183
1184         had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1185         if (had_lock < 0) {
1186                 status = had_lock;
1187                 goto bail_unlock_rw;
1188         } else if (had_lock) {
1189                 /*
1190                  * As far as we know, ocfs2_setattr() could only be the first
1191                  * VFS entry point in the call chain of recursive cluster
1192                  * locking issue.
1193                  *
1194                  * For instance:
1195                  * chmod_common()
1196                  *  notify_change()
1197                  *   ocfs2_setattr()
1198                  *    posix_acl_chmod()
1199                  *     ocfs2_iop_get_acl()
1200                  *
1201                  * But, we're not 100% sure if it's always true, because the
1202                  * ordering of the VFS entry points in the call chain is out
1203                  * of our control. So, we'd better dump the stack here to
1204                  * catch the other cases of recursive locking.
1205                  */
1206                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1207                 dump_stack();
1208         }
1209         inode_locked = 1;
1210
1211         if (size_change) {
1212                 status = inode_newsize_ok(inode, attr->ia_size);
1213                 if (status)
1214                         goto bail_unlock;
1215
1216                 if (i_size_read(inode) >= attr->ia_size) {
1217                         if (ocfs2_should_order_data(inode)) {
1218                                 status = ocfs2_begin_ordered_truncate(inode,
1219                                                                       attr->ia_size);
1220                                 if (status)
1221                                         goto bail_unlock;
1222                         }
1223                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1224                 } else
1225                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1226                 if (status < 0) {
1227                         if (status != -ENOSPC)
1228                                 mlog_errno(status);
1229                         status = -ENOSPC;
1230                         goto bail_unlock;
1231                 }
1232         }
1233
1234         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1235             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1236                 /*
1237                  * Gather pointers to quota structures so that allocation /
1238                  * freeing of quota structures happens here and not inside
1239                  * dquot_transfer() where we have problems with lock ordering
1240                  */
1241                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1242                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1243                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1244                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1245                         if (IS_ERR(transfer_to[USRQUOTA])) {
1246                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1247                                 goto bail_unlock;
1248                         }
1249                 }
1250                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1251                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1252                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1253                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1254                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1255                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1256                                 goto bail_unlock;
1257                         }
1258                 }
1259                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1260                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1261                                            2 * ocfs2_quota_trans_credits(sb));
1262                 if (IS_ERR(handle)) {
1263                         status = PTR_ERR(handle);
1264                         mlog_errno(status);
1265                         goto bail_unlock_alloc;
1266                 }
1267                 status = __dquot_transfer(inode, transfer_to);
1268                 if (status < 0)
1269                         goto bail_commit;
1270         } else {
1271                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1272                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1273                 if (IS_ERR(handle)) {
1274                         status = PTR_ERR(handle);
1275                         mlog_errno(status);
1276                         goto bail_unlock_alloc;
1277                 }
1278         }
1279
1280         setattr_copy(inode, attr);
1281         mark_inode_dirty(inode);
1282
1283         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1284         if (status < 0)
1285                 mlog_errno(status);
1286
1287 bail_commit:
1288         ocfs2_commit_trans(osb, handle);
1289 bail_unlock_alloc:
1290         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1291 bail_unlock:
1292         if (status && inode_locked) {
1293                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1294                 inode_locked = 0;
1295         }
1296 bail_unlock_rw:
1297         if (size_change)
1298                 ocfs2_rw_unlock(inode, 1);
1299 bail:
1300
1301         /* Release quota pointers in case we acquired them */
1302         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1303                 dqput(transfer_to[qtype]);
1304
1305         if (!status && attr->ia_valid & ATTR_MODE) {
1306                 status = ocfs2_acl_chmod(inode, bh);
1307                 if (status < 0)
1308                         mlog_errno(status);
1309         }
1310         if (inode_locked)
1311                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1312
1313         brelse(bh);
1314         return status;
1315 }
1316
1317 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1318                   u32 request_mask, unsigned int flags)
1319 {
1320         struct inode *inode = d_inode(path->dentry);
1321         struct super_block *sb = path->dentry->d_sb;
1322         struct ocfs2_super *osb = sb->s_fs_info;
1323         int err;
1324
1325         err = ocfs2_inode_revalidate(path->dentry);
1326         if (err) {
1327                 if (err != -ENOENT)
1328                         mlog_errno(err);
1329                 goto bail;
1330         }
1331
1332         generic_fillattr(inode, stat);
1333         /*
1334          * If there is inline data in the inode, the inode will normally not
1335          * have data blocks allocated (it may have an external xattr block).
1336          * Report at least one sector for such files, so tools like tar, rsync,
1337          * others don't incorrectly think the file is completely sparse.
1338          */
1339         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1340                 stat->blocks += (stat->size + 511)>>9;
1341
1342         /* We set the blksize from the cluster size for performance */
1343         stat->blksize = osb->s_clustersize;
1344
1345 bail:
1346         return err;
1347 }
1348
1349 int ocfs2_permission(struct inode *inode, int mask)
1350 {
1351         int ret, had_lock;
1352         struct ocfs2_lock_holder oh;
1353
1354         if (mask & MAY_NOT_BLOCK)
1355                 return -ECHILD;
1356
1357         had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1358         if (had_lock < 0) {
1359                 ret = had_lock;
1360                 goto out;
1361         } else if (had_lock) {
1362                 /* See comments in ocfs2_setattr() for details.
1363                  * The call chain of this case could be:
1364                  * do_sys_open()
1365                  *  may_open()
1366                  *   inode_permission()
1367                  *    ocfs2_permission()
1368                  *     ocfs2_iop_get_acl()
1369                  */
1370                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1371                 dump_stack();
1372         }
1373
1374         ret = generic_permission(inode, mask);
1375
1376         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1377 out:
1378         return ret;
1379 }
1380
1381 static int __ocfs2_write_remove_suid(struct inode *inode,
1382                                      struct buffer_head *bh)
1383 {
1384         int ret;
1385         handle_t *handle;
1386         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1387         struct ocfs2_dinode *di;
1388
1389         trace_ocfs2_write_remove_suid(
1390                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1391                         inode->i_mode);
1392
1393         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1394         if (IS_ERR(handle)) {
1395                 ret = PTR_ERR(handle);
1396                 mlog_errno(ret);
1397                 goto out;
1398         }
1399
1400         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1401                                       OCFS2_JOURNAL_ACCESS_WRITE);
1402         if (ret < 0) {
1403                 mlog_errno(ret);
1404                 goto out_trans;
1405         }
1406
1407         inode->i_mode &= ~S_ISUID;
1408         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1409                 inode->i_mode &= ~S_ISGID;
1410
1411         di = (struct ocfs2_dinode *) bh->b_data;
1412         di->i_mode = cpu_to_le16(inode->i_mode);
1413         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1414
1415         ocfs2_journal_dirty(handle, bh);
1416
1417 out_trans:
1418         ocfs2_commit_trans(osb, handle);
1419 out:
1420         return ret;
1421 }
1422
1423 static int ocfs2_write_remove_suid(struct inode *inode)
1424 {
1425         int ret;
1426         struct buffer_head *bh = NULL;
1427
1428         ret = ocfs2_read_inode_block(inode, &bh);
1429         if (ret < 0) {
1430                 mlog_errno(ret);
1431                 goto out;
1432         }
1433
1434         ret =  __ocfs2_write_remove_suid(inode, bh);
1435 out:
1436         brelse(bh);
1437         return ret;
1438 }
1439
1440 /*
1441  * Allocate enough extents to cover the region starting at byte offset
1442  * start for len bytes. Existing extents are skipped, any extents
1443  * added are marked as "unwritten".
1444  */
1445 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1446                                             u64 start, u64 len)
1447 {
1448         int ret;
1449         u32 cpos, phys_cpos, clusters, alloc_size;
1450         u64 end = start + len;
1451         struct buffer_head *di_bh = NULL;
1452
1453         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1454                 ret = ocfs2_read_inode_block(inode, &di_bh);
1455                 if (ret) {
1456                         mlog_errno(ret);
1457                         goto out;
1458                 }
1459
1460                 /*
1461                  * Nothing to do if the requested reservation range
1462                  * fits within the inode.
1463                  */
1464                 if (ocfs2_size_fits_inline_data(di_bh, end))
1465                         goto out;
1466
1467                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1468                 if (ret) {
1469                         mlog_errno(ret);
1470                         goto out;
1471                 }
1472         }
1473
1474         /*
1475          * We consider both start and len to be inclusive.
1476          */
1477         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1478         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1479         clusters -= cpos;
1480
1481         while (clusters) {
1482                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1483                                          &alloc_size, NULL);
1484                 if (ret) {
1485                         mlog_errno(ret);
1486                         goto out;
1487                 }
1488
1489                 /*
1490                  * Hole or existing extent len can be arbitrary, so
1491                  * cap it to our own allocation request.
1492                  */
1493                 if (alloc_size > clusters)
1494                         alloc_size = clusters;
1495
1496                 if (phys_cpos) {
1497                         /*
1498                          * We already have an allocation at this
1499                          * region so we can safely skip it.
1500                          */
1501                         goto next;
1502                 }
1503
1504                 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1505                 if (ret) {
1506                         if (ret != -ENOSPC)
1507                                 mlog_errno(ret);
1508                         goto out;
1509                 }
1510
1511 next:
1512                 cpos += alloc_size;
1513                 clusters -= alloc_size;
1514         }
1515
1516         ret = 0;
1517 out:
1518
1519         brelse(di_bh);
1520         return ret;
1521 }
1522
1523 /*
1524  * Truncate a byte range, avoiding pages within partial clusters. This
1525  * preserves those pages for the zeroing code to write to.
1526  */
1527 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1528                                          u64 byte_len)
1529 {
1530         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1531         loff_t start, end;
1532         struct address_space *mapping = inode->i_mapping;
1533
1534         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1535         end = byte_start + byte_len;
1536         end = end & ~(osb->s_clustersize - 1);
1537
1538         if (start < end) {
1539                 unmap_mapping_range(mapping, start, end - start, 0);
1540                 truncate_inode_pages_range(mapping, start, end - 1);
1541         }
1542 }
1543
1544 /*
1545  * zero out partial blocks of one cluster.
1546  *
1547  * start: file offset where zero starts, will be made upper block aligned.
1548  * len: it will be trimmed to the end of current cluster if "start + len"
1549  *      is bigger than it.
1550  */
1551 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1552                                         u64 start, u64 len)
1553 {
1554         int ret;
1555         u64 start_block, end_block, nr_blocks;
1556         u64 p_block, offset;
1557         u32 cluster, p_cluster, nr_clusters;
1558         struct super_block *sb = inode->i_sb;
1559         u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1560
1561         if (start + len < end)
1562                 end = start + len;
1563
1564         start_block = ocfs2_blocks_for_bytes(sb, start);
1565         end_block = ocfs2_blocks_for_bytes(sb, end);
1566         nr_blocks = end_block - start_block;
1567         if (!nr_blocks)
1568                 return 0;
1569
1570         cluster = ocfs2_bytes_to_clusters(sb, start);
1571         ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1572                                 &nr_clusters, NULL);
1573         if (ret)
1574                 return ret;
1575         if (!p_cluster)
1576                 return 0;
1577
1578         offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1579         p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1580         return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1581 }
1582
1583 static int ocfs2_zero_partial_clusters(struct inode *inode,
1584                                        u64 start, u64 len)
1585 {
1586         int ret = 0;
1587         u64 tmpend = 0;
1588         u64 end = start + len;
1589         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1590         unsigned int csize = osb->s_clustersize;
1591         handle_t *handle;
1592         loff_t isize = i_size_read(inode);
1593
1594         /*
1595          * The "start" and "end" values are NOT necessarily part of
1596          * the range whose allocation is being deleted. Rather, this
1597          * is what the user passed in with the request. We must zero
1598          * partial clusters here. There's no need to worry about
1599          * physical allocation - the zeroing code knows to skip holes.
1600          */
1601         trace_ocfs2_zero_partial_clusters(
1602                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1603                 (unsigned long long)start, (unsigned long long)end);
1604
1605         /*
1606          * If both edges are on a cluster boundary then there's no
1607          * zeroing required as the region is part of the allocation to
1608          * be truncated.
1609          */
1610         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1611                 goto out;
1612
1613         /* No page cache for EOF blocks, issue zero out to disk. */
1614         if (end > isize) {
1615                 /*
1616                  * zeroout eof blocks in last cluster starting from
1617                  * "isize" even "start" > "isize" because it is
1618                  * complicated to zeroout just at "start" as "start"
1619                  * may be not aligned with block size, buffer write
1620                  * would be required to do that, but out of eof buffer
1621                  * write is not supported.
1622                  */
1623                 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1624                                         end - isize);
1625                 if (ret) {
1626                         mlog_errno(ret);
1627                         goto out;
1628                 }
1629                 if (start >= isize)
1630                         goto out;
1631                 end = isize;
1632         }
1633         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1634         if (IS_ERR(handle)) {
1635                 ret = PTR_ERR(handle);
1636                 mlog_errno(ret);
1637                 goto out;
1638         }
1639
1640         /*
1641          * If start is on a cluster boundary and end is somewhere in another
1642          * cluster, we have not COWed the cluster starting at start, unless
1643          * end is also within the same cluster. So, in this case, we skip this
1644          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1645          * to the next one.
1646          */
1647         if ((start & (csize - 1)) != 0) {
1648                 /*
1649                  * We want to get the byte offset of the end of the 1st
1650                  * cluster.
1651                  */
1652                 tmpend = (u64)osb->s_clustersize +
1653                         (start & ~(osb->s_clustersize - 1));
1654                 if (tmpend > end)
1655                         tmpend = end;
1656
1657                 trace_ocfs2_zero_partial_clusters_range1(
1658                         (unsigned long long)start,
1659                         (unsigned long long)tmpend);
1660
1661                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1662                                                     tmpend);
1663                 if (ret)
1664                         mlog_errno(ret);
1665         }
1666
1667         if (tmpend < end) {
1668                 /*
1669                  * This may make start and end equal, but the zeroing
1670                  * code will skip any work in that case so there's no
1671                  * need to catch it up here.
1672                  */
1673                 start = end & ~(osb->s_clustersize - 1);
1674
1675                 trace_ocfs2_zero_partial_clusters_range2(
1676                         (unsigned long long)start, (unsigned long long)end);
1677
1678                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1679                 if (ret)
1680                         mlog_errno(ret);
1681         }
1682         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1683
1684         ocfs2_commit_trans(osb, handle);
1685 out:
1686         return ret;
1687 }
1688
1689 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1690 {
1691         int i;
1692         struct ocfs2_extent_rec *rec = NULL;
1693
1694         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1695
1696                 rec = &el->l_recs[i];
1697
1698                 if (le32_to_cpu(rec->e_cpos) < pos)
1699                         break;
1700         }
1701
1702         return i;
1703 }
1704
1705 /*
1706  * Helper to calculate the punching pos and length in one run, we handle the
1707  * following three cases in order:
1708  *
1709  * - remove the entire record
1710  * - remove a partial record
1711  * - no record needs to be removed (hole-punching completed)
1712 */
1713 static void ocfs2_calc_trunc_pos(struct inode *inode,
1714                                  struct ocfs2_extent_list *el,
1715                                  struct ocfs2_extent_rec *rec,
1716                                  u32 trunc_start, u32 *trunc_cpos,
1717                                  u32 *trunc_len, u32 *trunc_end,
1718                                  u64 *blkno, int *done)
1719 {
1720         int ret = 0;
1721         u32 coff, range;
1722
1723         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1724
1725         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1726                 /*
1727                  * remove an entire extent record.
1728                  */
1729                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1730                 /*
1731                  * Skip holes if any.
1732                  */
1733                 if (range < *trunc_end)
1734                         *trunc_end = range;
1735                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1736                 *blkno = le64_to_cpu(rec->e_blkno);
1737                 *trunc_end = le32_to_cpu(rec->e_cpos);
1738         } else if (range > trunc_start) {
1739                 /*
1740                  * remove a partial extent record, which means we're
1741                  * removing the last extent record.
1742                  */
1743                 *trunc_cpos = trunc_start;
1744                 /*
1745                  * skip hole if any.
1746                  */
1747                 if (range < *trunc_end)
1748                         *trunc_end = range;
1749                 *trunc_len = *trunc_end - trunc_start;
1750                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1751                 *blkno = le64_to_cpu(rec->e_blkno) +
1752                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1753                 *trunc_end = trunc_start;
1754         } else {
1755                 /*
1756                  * It may have two following possibilities:
1757                  *
1758                  * - last record has been removed
1759                  * - trunc_start was within a hole
1760                  *
1761                  * both two cases mean the completion of hole punching.
1762                  */
1763                 ret = 1;
1764         }
1765
1766         *done = ret;
1767 }
1768
1769 int ocfs2_remove_inode_range(struct inode *inode,
1770                              struct buffer_head *di_bh, u64 byte_start,
1771                              u64 byte_len)
1772 {
1773         int ret = 0, flags = 0, done = 0, i;
1774         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1775         u32 cluster_in_el;
1776         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1777         struct ocfs2_cached_dealloc_ctxt dealloc;
1778         struct address_space *mapping = inode->i_mapping;
1779         struct ocfs2_extent_tree et;
1780         struct ocfs2_path *path = NULL;
1781         struct ocfs2_extent_list *el = NULL;
1782         struct ocfs2_extent_rec *rec = NULL;
1783         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1784         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1785
1786         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1787         ocfs2_init_dealloc_ctxt(&dealloc);
1788
1789         trace_ocfs2_remove_inode_range(
1790                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1791                         (unsigned long long)byte_start,
1792                         (unsigned long long)byte_len);
1793
1794         if (byte_len == 0)
1795                 return 0;
1796
1797         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1798                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1799                                             byte_start + byte_len, 0);
1800                 if (ret) {
1801                         mlog_errno(ret);
1802                         goto out;
1803                 }
1804                 /*
1805                  * There's no need to get fancy with the page cache
1806                  * truncate of an inline-data inode. We're talking
1807                  * about less than a page here, which will be cached
1808                  * in the dinode buffer anyway.
1809                  */
1810                 unmap_mapping_range(mapping, 0, 0, 0);
1811                 truncate_inode_pages(mapping, 0);
1812                 goto out;
1813         }
1814
1815         /*
1816          * For reflinks, we may need to CoW 2 clusters which might be
1817          * partially zero'd later, if hole's start and end offset were
1818          * within one cluster(means is not exactly aligned to clustersize).
1819          */
1820
1821         if (ocfs2_is_refcount_inode(inode)) {
1822                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1823                 if (ret) {
1824                         mlog_errno(ret);
1825                         goto out;
1826                 }
1827
1828                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1829                 if (ret) {
1830                         mlog_errno(ret);
1831                         goto out;
1832                 }
1833         }
1834
1835         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1836         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1837         cluster_in_el = trunc_end;
1838
1839         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1840         if (ret) {
1841                 mlog_errno(ret);
1842                 goto out;
1843         }
1844
1845         path = ocfs2_new_path_from_et(&et);
1846         if (!path) {
1847                 ret = -ENOMEM;
1848                 mlog_errno(ret);
1849                 goto out;
1850         }
1851
1852         while (trunc_end > trunc_start) {
1853
1854                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1855                                       cluster_in_el);
1856                 if (ret) {
1857                         mlog_errno(ret);
1858                         goto out;
1859                 }
1860
1861                 el = path_leaf_el(path);
1862
1863                 i = ocfs2_find_rec(el, trunc_end);
1864                 /*
1865                  * Need to go to previous extent block.
1866                  */
1867                 if (i < 0) {
1868                         if (path->p_tree_depth == 0)
1869                                 break;
1870
1871                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1872                                                             path,
1873                                                             &cluster_in_el);
1874                         if (ret) {
1875                                 mlog_errno(ret);
1876                                 goto out;
1877                         }
1878
1879                         /*
1880                          * We've reached the leftmost extent block,
1881                          * it's safe to leave.
1882                          */
1883                         if (cluster_in_el == 0)
1884                                 break;
1885
1886                         /*
1887                          * The 'pos' searched for previous extent block is
1888                          * always one cluster less than actual trunc_end.
1889                          */
1890                         trunc_end = cluster_in_el + 1;
1891
1892                         ocfs2_reinit_path(path, 1);
1893
1894                         continue;
1895
1896                 } else
1897                         rec = &el->l_recs[i];
1898
1899                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1900                                      &trunc_len, &trunc_end, &blkno, &done);
1901                 if (done)
1902                         break;
1903
1904                 flags = rec->e_flags;
1905                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1906
1907                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1908                                                phys_cpos, trunc_len, flags,
1909                                                &dealloc, refcount_loc, false);
1910                 if (ret < 0) {
1911                         mlog_errno(ret);
1912                         goto out;
1913                 }
1914
1915                 cluster_in_el = trunc_end;
1916
1917                 ocfs2_reinit_path(path, 1);
1918         }
1919
1920         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1921
1922 out:
1923         ocfs2_free_path(path);
1924         ocfs2_schedule_truncate_log_flush(osb, 1);
1925         ocfs2_run_deallocs(osb, &dealloc);
1926
1927         return ret;
1928 }
1929
1930 /*
1931  * Parts of this function taken from xfs_change_file_space()
1932  */
1933 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1934                                      loff_t f_pos, unsigned int cmd,
1935                                      struct ocfs2_space_resv *sr,
1936                                      int change_size)
1937 {
1938         int ret;
1939         s64 llen;
1940         loff_t size, orig_isize;
1941         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1942         struct buffer_head *di_bh = NULL;
1943         handle_t *handle;
1944         unsigned long long max_off = inode->i_sb->s_maxbytes;
1945
1946         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1947                 return -EROFS;
1948
1949         inode_lock(inode);
1950
1951         /*
1952          * This prevents concurrent writes on other nodes
1953          */
1954         ret = ocfs2_rw_lock(inode, 1);
1955         if (ret) {
1956                 mlog_errno(ret);
1957                 goto out;
1958         }
1959
1960         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1961         if (ret) {
1962                 mlog_errno(ret);
1963                 goto out_rw_unlock;
1964         }
1965
1966         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1967                 ret = -EPERM;
1968                 goto out_inode_unlock;
1969         }
1970
1971         switch (sr->l_whence) {
1972         case 0: /*SEEK_SET*/
1973                 break;
1974         case 1: /*SEEK_CUR*/
1975                 sr->l_start += f_pos;
1976                 break;
1977         case 2: /*SEEK_END*/
1978                 sr->l_start += i_size_read(inode);
1979                 break;
1980         default:
1981                 ret = -EINVAL;
1982                 goto out_inode_unlock;
1983         }
1984         sr->l_whence = 0;
1985
1986         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1987
1988         if (sr->l_start < 0
1989             || sr->l_start > max_off
1990             || (sr->l_start + llen) < 0
1991             || (sr->l_start + llen) > max_off) {
1992                 ret = -EINVAL;
1993                 goto out_inode_unlock;
1994         }
1995         size = sr->l_start + sr->l_len;
1996
1997         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1998             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1999                 if (sr->l_len <= 0) {
2000                         ret = -EINVAL;
2001                         goto out_inode_unlock;
2002                 }
2003         }
2004
2005         if (file && should_remove_suid(file->f_path.dentry)) {
2006                 ret = __ocfs2_write_remove_suid(inode, di_bh);
2007                 if (ret) {
2008                         mlog_errno(ret);
2009                         goto out_inode_unlock;
2010                 }
2011         }
2012
2013         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2014         switch (cmd) {
2015         case OCFS2_IOC_RESVSP:
2016         case OCFS2_IOC_RESVSP64:
2017                 /*
2018                  * This takes unsigned offsets, but the signed ones we
2019                  * pass have been checked against overflow above.
2020                  */
2021                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2022                                                        sr->l_len);
2023                 break;
2024         case OCFS2_IOC_UNRESVSP:
2025         case OCFS2_IOC_UNRESVSP64:
2026                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2027                                                sr->l_len);
2028                 break;
2029         default:
2030                 ret = -EINVAL;
2031         }
2032
2033         orig_isize = i_size_read(inode);
2034         /* zeroout eof blocks in the cluster. */
2035         if (!ret && change_size && orig_isize < size) {
2036                 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2037                                         size - orig_isize);
2038                 if (!ret)
2039                         i_size_write(inode, size);
2040         }
2041         up_write(&OCFS2_I(inode)->ip_alloc_sem);
2042         if (ret) {
2043                 mlog_errno(ret);
2044                 goto out_inode_unlock;
2045         }
2046
2047         /*
2048          * We update c/mtime for these changes
2049          */
2050         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2051         if (IS_ERR(handle)) {
2052                 ret = PTR_ERR(handle);
2053                 mlog_errno(ret);
2054                 goto out_inode_unlock;
2055         }
2056
2057         inode->i_ctime = inode->i_mtime = current_time(inode);
2058         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2059         if (ret < 0)
2060                 mlog_errno(ret);
2061
2062         if (file && (file->f_flags & O_SYNC))
2063                 handle->h_sync = 1;
2064
2065         ocfs2_commit_trans(osb, handle);
2066
2067 out_inode_unlock:
2068         brelse(di_bh);
2069         ocfs2_inode_unlock(inode, 1);
2070 out_rw_unlock:
2071         ocfs2_rw_unlock(inode, 1);
2072
2073 out:
2074         inode_unlock(inode);
2075         return ret;
2076 }
2077
2078 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2079                             struct ocfs2_space_resv *sr)
2080 {
2081         struct inode *inode = file_inode(file);
2082         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2083         int ret;
2084
2085         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2086             !ocfs2_writes_unwritten_extents(osb))
2087                 return -ENOTTY;
2088         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2089                  !ocfs2_sparse_alloc(osb))
2090                 return -ENOTTY;
2091
2092         if (!S_ISREG(inode->i_mode))
2093                 return -EINVAL;
2094
2095         if (!(file->f_mode & FMODE_WRITE))
2096                 return -EBADF;
2097
2098         ret = mnt_want_write_file(file);
2099         if (ret)
2100                 return ret;
2101         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2102         mnt_drop_write_file(file);
2103         return ret;
2104 }
2105
2106 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2107                             loff_t len)
2108 {
2109         struct inode *inode = file_inode(file);
2110         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2111         struct ocfs2_space_resv sr;
2112         int change_size = 1;
2113         int cmd = OCFS2_IOC_RESVSP64;
2114         int ret = 0;
2115
2116         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2117                 return -EOPNOTSUPP;
2118         if (!ocfs2_writes_unwritten_extents(osb))
2119                 return -EOPNOTSUPP;
2120
2121         if (mode & FALLOC_FL_KEEP_SIZE) {
2122                 change_size = 0;
2123         } else {
2124                 ret = inode_newsize_ok(inode, offset + len);
2125                 if (ret)
2126                         return ret;
2127         }
2128
2129         if (mode & FALLOC_FL_PUNCH_HOLE)
2130                 cmd = OCFS2_IOC_UNRESVSP64;
2131
2132         sr.l_whence = 0;
2133         sr.l_start = (s64)offset;
2134         sr.l_len = (s64)len;
2135
2136         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2137                                          change_size);
2138 }
2139
2140 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2141                                    size_t count)
2142 {
2143         int ret = 0;
2144         unsigned int extent_flags;
2145         u32 cpos, clusters, extent_len, phys_cpos;
2146         struct super_block *sb = inode->i_sb;
2147
2148         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2149             !ocfs2_is_refcount_inode(inode) ||
2150             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2151                 return 0;
2152
2153         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2154         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2155
2156         while (clusters) {
2157                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2158                                          &extent_flags);
2159                 if (ret < 0) {
2160                         mlog_errno(ret);
2161                         goto out;
2162                 }
2163
2164                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2165                         ret = 1;
2166                         break;
2167                 }
2168
2169                 if (extent_len > clusters)
2170                         extent_len = clusters;
2171
2172                 clusters -= extent_len;
2173                 cpos += extent_len;
2174         }
2175 out:
2176         return ret;
2177 }
2178
2179 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2180 {
2181         int blockmask = inode->i_sb->s_blocksize - 1;
2182         loff_t final_size = pos + count;
2183
2184         if ((pos & blockmask) || (final_size & blockmask))
2185                 return 1;
2186         return 0;
2187 }
2188
2189 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2190                                             struct buffer_head **di_bh,
2191                                             int meta_level,
2192                                             int write_sem,
2193                                             int wait)
2194 {
2195         int ret = 0;
2196
2197         if (wait)
2198                 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2199         else
2200                 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2201         if (ret < 0)
2202                 goto out;
2203
2204         if (wait) {
2205                 if (write_sem)
2206                         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2207                 else
2208                         down_read(&OCFS2_I(inode)->ip_alloc_sem);
2209         } else {
2210                 if (write_sem)
2211                         ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2212                 else
2213                         ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2214
2215                 if (!ret) {
2216                         ret = -EAGAIN;
2217                         goto out_unlock;
2218                 }
2219         }
2220
2221         return ret;
2222
2223 out_unlock:
2224         brelse(*di_bh);
2225         *di_bh = NULL;
2226         ocfs2_inode_unlock(inode, meta_level);
2227 out:
2228         return ret;
2229 }
2230
2231 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2232                                                struct buffer_head **di_bh,
2233                                                int meta_level,
2234                                                int write_sem)
2235 {
2236         if (write_sem)
2237                 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2238         else
2239                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2240
2241         brelse(*di_bh);
2242         *di_bh = NULL;
2243
2244         if (meta_level >= 0)
2245                 ocfs2_inode_unlock(inode, meta_level);
2246 }
2247
2248 static int ocfs2_prepare_inode_for_write(struct file *file,
2249                                          loff_t pos, size_t count, int wait)
2250 {
2251         int ret = 0, meta_level = 0, overwrite_io = 0;
2252         int write_sem = 0;
2253         struct dentry *dentry = file->f_path.dentry;
2254         struct inode *inode = d_inode(dentry);
2255         struct buffer_head *di_bh = NULL;
2256         loff_t end;
2257         u32 cpos;
2258         u32 clusters;
2259
2260         /*
2261          * We start with a read level meta lock and only jump to an ex
2262          * if we need to make modifications here.
2263          */
2264         for(;;) {
2265                 ret = ocfs2_inode_lock_for_extent_tree(inode,
2266                                                        &di_bh,
2267                                                        meta_level,
2268                                                        write_sem,
2269                                                        wait);
2270                 if (ret < 0) {
2271                         if (ret != -EAGAIN)
2272                                 mlog_errno(ret);
2273                         goto out;
2274                 }
2275
2276                 /*
2277                  * Check if IO will overwrite allocated blocks in case
2278                  * IOCB_NOWAIT flag is set.
2279                  */
2280                 if (!wait && !overwrite_io) {
2281                         overwrite_io = 1;
2282
2283                         ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2284                         if (ret < 0) {
2285                                 if (ret != -EAGAIN)
2286                                         mlog_errno(ret);
2287                                 goto out_unlock;
2288                         }
2289                 }
2290
2291                 /* Clear suid / sgid if necessary. We do this here
2292                  * instead of later in the write path because
2293                  * remove_suid() calls ->setattr without any hint that
2294                  * we may have already done our cluster locking. Since
2295                  * ocfs2_setattr() *must* take cluster locks to
2296                  * proceed, this will lead us to recursively lock the
2297                  * inode. There's also the dinode i_size state which
2298                  * can be lost via setattr during extending writes (we
2299                  * set inode->i_size at the end of a write. */
2300                 if (should_remove_suid(dentry)) {
2301                         if (meta_level == 0) {
2302                                 ocfs2_inode_unlock_for_extent_tree(inode,
2303                                                                    &di_bh,
2304                                                                    meta_level,
2305                                                                    write_sem);
2306                                 meta_level = 1;
2307                                 continue;
2308                         }
2309
2310                         ret = ocfs2_write_remove_suid(inode);
2311                         if (ret < 0) {
2312                                 mlog_errno(ret);
2313                                 goto out_unlock;
2314                         }
2315                 }
2316
2317                 end = pos + count;
2318
2319                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2320                 if (ret == 1) {
2321                         ocfs2_inode_unlock_for_extent_tree(inode,
2322                                                            &di_bh,
2323                                                            meta_level,
2324                                                            write_sem);
2325                         meta_level = 1;
2326                         write_sem = 1;
2327                         ret = ocfs2_inode_lock_for_extent_tree(inode,
2328                                                                &di_bh,
2329                                                                meta_level,
2330                                                                write_sem,
2331                                                                wait);
2332                         if (ret < 0) {
2333                                 if (ret != -EAGAIN)
2334                                         mlog_errno(ret);
2335                                 goto out;
2336                         }
2337
2338                         cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2339                         clusters =
2340                                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2341                         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2342                 }
2343
2344                 if (ret < 0) {
2345                         if (ret != -EAGAIN)
2346                                 mlog_errno(ret);
2347                         goto out_unlock;
2348                 }
2349
2350                 break;
2351         }
2352
2353 out_unlock:
2354         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2355                                             pos, count, wait);
2356
2357         ocfs2_inode_unlock_for_extent_tree(inode,
2358                                            &di_bh,
2359                                            meta_level,
2360                                            write_sem);
2361
2362 out:
2363         return ret;
2364 }
2365
2366 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2367                                     struct iov_iter *from)
2368 {
2369         int rw_level;
2370         ssize_t written = 0;
2371         ssize_t ret;
2372         size_t count = iov_iter_count(from);
2373         struct file *file = iocb->ki_filp;
2374         struct inode *inode = file_inode(file);
2375         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2376         int full_coherency = !(osb->s_mount_opt &
2377                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2378         void *saved_ki_complete = NULL;
2379         int append_write = ((iocb->ki_pos + count) >=
2380                         i_size_read(inode) ? 1 : 0);
2381         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2382         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2383
2384         trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2385                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2386                 file->f_path.dentry->d_name.len,
2387                 file->f_path.dentry->d_name.name,
2388                 (unsigned int)from->nr_segs);   /* GRRRRR */
2389
2390         if (!direct_io && nowait)
2391                 return -EOPNOTSUPP;
2392
2393         if (count == 0)
2394                 return 0;
2395
2396         if (nowait) {
2397                 if (!inode_trylock(inode))
2398                         return -EAGAIN;
2399         } else
2400                 inode_lock(inode);
2401
2402         /*
2403          * Concurrent O_DIRECT writes are allowed with
2404          * mount_option "coherency=buffered".
2405          * For append write, we must take rw EX.
2406          */
2407         rw_level = (!direct_io || full_coherency || append_write);
2408
2409         if (nowait)
2410                 ret = ocfs2_try_rw_lock(inode, rw_level);
2411         else
2412                 ret = ocfs2_rw_lock(inode, rw_level);
2413         if (ret < 0) {
2414                 if (ret != -EAGAIN)
2415                         mlog_errno(ret);
2416                 goto out_mutex;
2417         }
2418
2419         /*
2420          * O_DIRECT writes with "coherency=full" need to take EX cluster
2421          * inode_lock to guarantee coherency.
2422          */
2423         if (direct_io && full_coherency) {
2424                 /*
2425                  * We need to take and drop the inode lock to force
2426                  * other nodes to drop their caches.  Buffered I/O
2427                  * already does this in write_begin().
2428                  */
2429                 if (nowait)
2430                         ret = ocfs2_try_inode_lock(inode, NULL, 1);
2431                 else
2432                         ret = ocfs2_inode_lock(inode, NULL, 1);
2433                 if (ret < 0) {
2434                         if (ret != -EAGAIN)
2435                                 mlog_errno(ret);
2436                         goto out;
2437                 }
2438
2439                 ocfs2_inode_unlock(inode, 1);
2440         }
2441
2442         ret = generic_write_checks(iocb, from);
2443         if (ret <= 0) {
2444                 if (ret)
2445                         mlog_errno(ret);
2446                 goto out;
2447         }
2448         count = ret;
2449
2450         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2451         if (ret < 0) {
2452                 if (ret != -EAGAIN)
2453                         mlog_errno(ret);
2454                 goto out;
2455         }
2456
2457         if (direct_io && !is_sync_kiocb(iocb) &&
2458             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2459                 /*
2460                  * Make it a sync io if it's an unaligned aio.
2461                  */
2462                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2463         }
2464
2465         /* communicate with ocfs2_dio_end_io */
2466         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2467
2468         written = __generic_file_write_iter(iocb, from);
2469         /* buffered aio wouldn't have proper lock coverage today */
2470         BUG_ON(written == -EIOCBQUEUED && !direct_io);
2471
2472         /*
2473          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2474          * function pointer which is called when o_direct io completes so that
2475          * it can unlock our rw lock.
2476          * Unfortunately there are error cases which call end_io and others
2477          * that don't.  so we don't have to unlock the rw_lock if either an
2478          * async dio is going to do it in the future or an end_io after an
2479          * error has already done it.
2480          */
2481         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2482                 rw_level = -1;
2483         }
2484
2485         if (unlikely(written <= 0))
2486                 goto out;
2487
2488         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2489             IS_SYNC(inode)) {
2490                 ret = filemap_fdatawrite_range(file->f_mapping,
2491                                                iocb->ki_pos - written,
2492                                                iocb->ki_pos - 1);
2493                 if (ret < 0)
2494                         written = ret;
2495
2496                 if (!ret) {
2497                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2498                         if (ret < 0)
2499                                 written = ret;
2500                 }
2501
2502                 if (!ret)
2503                         ret = filemap_fdatawait_range(file->f_mapping,
2504                                                       iocb->ki_pos - written,
2505                                                       iocb->ki_pos - 1);
2506         }
2507
2508 out:
2509         if (saved_ki_complete)
2510                 xchg(&iocb->ki_complete, saved_ki_complete);
2511
2512         if (rw_level != -1)
2513                 ocfs2_rw_unlock(inode, rw_level);
2514
2515 out_mutex:
2516         inode_unlock(inode);
2517
2518         if (written)
2519                 ret = written;
2520         return ret;
2521 }
2522
2523 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2524                                    struct iov_iter *to)
2525 {
2526         int ret = 0, rw_level = -1, lock_level = 0;
2527         struct file *filp = iocb->ki_filp;
2528         struct inode *inode = file_inode(filp);
2529         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2530         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2531
2532         trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2533                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2534                         filp->f_path.dentry->d_name.len,
2535                         filp->f_path.dentry->d_name.name,
2536                         to->nr_segs);   /* GRRRRR */
2537
2538
2539         if (!inode) {
2540                 ret = -EINVAL;
2541                 mlog_errno(ret);
2542                 goto bail;
2543         }
2544
2545         if (!direct_io && nowait)
2546                 return -EOPNOTSUPP;
2547
2548         /*
2549          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2550          * need locks to protect pending reads from racing with truncate.
2551          */
2552         if (direct_io) {
2553                 if (nowait)
2554                         ret = ocfs2_try_rw_lock(inode, 0);
2555                 else
2556                         ret = ocfs2_rw_lock(inode, 0);
2557
2558                 if (ret < 0) {
2559                         if (ret != -EAGAIN)
2560                                 mlog_errno(ret);
2561                         goto bail;
2562                 }
2563                 rw_level = 0;
2564                 /* communicate with ocfs2_dio_end_io */
2565                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2566         }
2567
2568         /*
2569          * We're fine letting folks race truncates and extending
2570          * writes with read across the cluster, just like they can
2571          * locally. Hence no rw_lock during read.
2572          *
2573          * Take and drop the meta data lock to update inode fields
2574          * like i_size. This allows the checks down below
2575          * generic_file_read_iter() a chance of actually working.
2576          */
2577         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2578                                      !nowait);
2579         if (ret < 0) {
2580                 if (ret != -EAGAIN)
2581                         mlog_errno(ret);
2582                 goto bail;
2583         }
2584         ocfs2_inode_unlock(inode, lock_level);
2585
2586         ret = generic_file_read_iter(iocb, to);
2587         trace_generic_file_read_iter_ret(ret);
2588
2589         /* buffered aio wouldn't have proper lock coverage today */
2590         BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2591
2592         /* see ocfs2_file_write_iter */
2593         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2594                 rw_level = -1;
2595         }
2596
2597 bail:
2598         if (rw_level != -1)
2599                 ocfs2_rw_unlock(inode, rw_level);
2600
2601         return ret;
2602 }
2603
2604 /* Refer generic_file_llseek_unlocked() */
2605 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2606 {
2607         struct inode *inode = file->f_mapping->host;
2608         int ret = 0;
2609
2610         inode_lock(inode);
2611
2612         switch (whence) {
2613         case SEEK_SET:
2614                 break;
2615         case SEEK_END:
2616                 /* SEEK_END requires the OCFS2 inode lock for the file
2617                  * because it references the file's size.
2618                  */
2619                 ret = ocfs2_inode_lock(inode, NULL, 0);
2620                 if (ret < 0) {
2621                         mlog_errno(ret);
2622                         goto out;
2623                 }
2624                 offset += i_size_read(inode);
2625                 ocfs2_inode_unlock(inode, 0);
2626                 break;
2627         case SEEK_CUR:
2628                 if (offset == 0) {
2629                         offset = file->f_pos;
2630                         goto out;
2631                 }
2632                 offset += file->f_pos;
2633                 break;
2634         case SEEK_DATA:
2635         case SEEK_HOLE:
2636                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2637                 if (ret)
2638                         goto out;
2639                 break;
2640         default:
2641                 ret = -EINVAL;
2642                 goto out;
2643         }
2644
2645         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2646
2647 out:
2648         inode_unlock(inode);
2649         if (ret)
2650                 return ret;
2651         return offset;
2652 }
2653
2654 static int ocfs2_file_clone_range(struct file *file_in,
2655                                   loff_t pos_in,
2656                                   struct file *file_out,
2657                                   loff_t pos_out,
2658                                   u64 len)
2659 {
2660         return ocfs2_reflink_remap_range(file_in, pos_in, file_out, pos_out,
2661                                          len, false);
2662 }
2663
2664 static int ocfs2_file_dedupe_range(struct file *file_in,
2665                                    loff_t pos_in,
2666                                    struct file *file_out,
2667                                    loff_t pos_out,
2668                                    u64 len)
2669 {
2670         return ocfs2_reflink_remap_range(file_in, pos_in, file_out, pos_out,
2671                                           len, true);
2672 }
2673
2674 const struct inode_operations ocfs2_file_iops = {
2675         .setattr        = ocfs2_setattr,
2676         .getattr        = ocfs2_getattr,
2677         .permission     = ocfs2_permission,
2678         .listxattr      = ocfs2_listxattr,
2679         .fiemap         = ocfs2_fiemap,
2680         .get_acl        = ocfs2_iop_get_acl,
2681         .set_acl        = ocfs2_iop_set_acl,
2682 };
2683
2684 const struct inode_operations ocfs2_special_file_iops = {
2685         .setattr        = ocfs2_setattr,
2686         .getattr        = ocfs2_getattr,
2687         .permission     = ocfs2_permission,
2688         .get_acl        = ocfs2_iop_get_acl,
2689         .set_acl        = ocfs2_iop_set_acl,
2690 };
2691
2692 /*
2693  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2694  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2695  */
2696 const struct file_operations ocfs2_fops = {
2697         .llseek         = ocfs2_file_llseek,
2698         .mmap           = ocfs2_mmap,
2699         .fsync          = ocfs2_sync_file,
2700         .release        = ocfs2_file_release,
2701         .open           = ocfs2_file_open,
2702         .read_iter      = ocfs2_file_read_iter,
2703         .write_iter     = ocfs2_file_write_iter,
2704         .unlocked_ioctl = ocfs2_ioctl,
2705 #ifdef CONFIG_COMPAT
2706         .compat_ioctl   = ocfs2_compat_ioctl,
2707 #endif
2708         .lock           = ocfs2_lock,
2709         .flock          = ocfs2_flock,
2710         .splice_read    = generic_file_splice_read,
2711         .splice_write   = iter_file_splice_write,
2712         .fallocate      = ocfs2_fallocate,
2713         .clone_file_range = ocfs2_file_clone_range,
2714         .dedupe_file_range = ocfs2_file_dedupe_range,
2715 };
2716
2717 const struct file_operations ocfs2_dops = {
2718         .llseek         = generic_file_llseek,
2719         .read           = generic_read_dir,
2720         .iterate        = ocfs2_readdir,
2721         .fsync          = ocfs2_sync_file,
2722         .release        = ocfs2_dir_release,
2723         .open           = ocfs2_dir_open,
2724         .unlocked_ioctl = ocfs2_ioctl,
2725 #ifdef CONFIG_COMPAT
2726         .compat_ioctl   = ocfs2_compat_ioctl,
2727 #endif
2728         .lock           = ocfs2_lock,
2729         .flock          = ocfs2_flock,
2730 };
2731
2732 /*
2733  * POSIX-lockless variants of our file_operations.
2734  *
2735  * These will be used if the underlying cluster stack does not support
2736  * posix file locking, if the user passes the "localflocks" mount
2737  * option, or if we have a local-only fs.
2738  *
2739  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2740  * so we still want it in the case of no stack support for
2741  * plocks. Internally, it will do the right thing when asked to ignore
2742  * the cluster.
2743  */
2744 const struct file_operations ocfs2_fops_no_plocks = {
2745         .llseek         = ocfs2_file_llseek,
2746         .mmap           = ocfs2_mmap,
2747         .fsync          = ocfs2_sync_file,
2748         .release        = ocfs2_file_release,
2749         .open           = ocfs2_file_open,
2750         .read_iter      = ocfs2_file_read_iter,
2751         .write_iter     = ocfs2_file_write_iter,
2752         .unlocked_ioctl = ocfs2_ioctl,
2753 #ifdef CONFIG_COMPAT
2754         .compat_ioctl   = ocfs2_compat_ioctl,
2755 #endif
2756         .flock          = ocfs2_flock,
2757         .splice_read    = generic_file_splice_read,
2758         .splice_write   = iter_file_splice_write,
2759         .fallocate      = ocfs2_fallocate,
2760         .clone_file_range = ocfs2_file_clone_range,
2761         .dedupe_file_range = ocfs2_file_dedupe_range,
2762 };
2763
2764 const struct file_operations ocfs2_dops_no_plocks = {
2765         .llseek         = generic_file_llseek,
2766         .read           = generic_read_dir,
2767         .iterate        = ocfs2_readdir,
2768         .fsync          = ocfs2_sync_file,
2769         .release        = ocfs2_dir_release,
2770         .open           = ocfs2_dir_open,
2771         .unlocked_ioctl = ocfs2_ioctl,
2772 #ifdef CONFIG_COMPAT
2773         .compat_ioctl   = ocfs2_compat_ioctl,
2774 #endif
2775         .flock          = ocfs2_flock,
2776 };