fs/ocfs2/file.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * file.c
   4  *
   5  * File open, close, extend, truncate
   6  *
   7  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   8  */
   9
  10 #include <linux/capability.h>
  11 #include <linux/fs.h>
  12 #include <linux/types.h>
  13 #include <linux/slab.h>
  14 #include <linux/highmem.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/uio.h>
  17 #include <linux/sched.h>
  18 #include <linux/splice.h>
  19 #include <linux/mount.h>
  20 #include <linux/writeback.h>
  21 #include <linux/falloc.h>
  22 #include <linux/quotaops.h>
  23 #include <linux/blkdev.h>
  24 #include <linux/backing-dev.h>
  25
  26 #include <cluster/masklog.h>
  27
  28 #include "ocfs2.h"
  29
  30 #include "alloc.h"
  31 #include "aops.h"
  32 #include "dir.h"
  33 #include "dlmglue.h"
  34 #include "extent_map.h"
  35 #include "file.h"
  36 #include "sysfile.h"
  37 #include "inode.h"
  38 #include "ioctl.h"
  39 #include "journal.h"
  40 #include "locks.h"
  41 #include "mmap.h"
  42 #include "suballoc.h"
  43 #include "super.h"
  44 #include "xattr.h"
  45 #include "acl.h"
  46 #include "quota.h"
  47 #include "refcounttree.h"
  48 #include "ocfs2_trace.h"
  49
  50 #include "buffer_head_io.h"
  51
  52 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  53 {
  54         struct ocfs2_file_private *fp;
  55
  56         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  57         if (!fp)
  58                 return -ENOMEM;
  59
  60         fp->fp_file = file;
  61         mutex_init(&fp->fp_mutex);
  62         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  63         file->private_data = fp;
  64
  65         return 0;
  66 }
  67
  68 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  69 {
  70         struct ocfs2_file_private *fp = file->private_data;
  71         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  72
  73         if (fp) {
  74                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  75                 ocfs2_lock_res_free(&fp->fp_flock);
  76                 kfree(fp);
  77                 file->private_data = NULL;
  78         }
  79 }
  80
  81 static int ocfs2_file_open(struct inode *inode, struct file *file)
  82 {
  83         int status;
  84         int mode = file->f_flags;
  85         struct ocfs2_inode_info *oi = OCFS2_I(inode);
  86
  87         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
  88                               (unsigned long long)oi->ip_blkno,
  89                               file->f_path.dentry->d_name.len,
  90                               file->f_path.dentry->d_name.name, mode);
  91
  92         if (file->f_mode & FMODE_WRITE) {
  93                 status = dquot_initialize(inode);
  94                 if (status)
  95                         goto leave;
  96         }
  97
  98         spin_lock(&oi->ip_lock);
  99
 100         /* Check that the inode hasn't been wiped from disk by another
 101          * node. If it hasn't then we're safe as long as we hold the
 102          * spin lock until our increment of open count. */
 103         if (oi->ip_flags & OCFS2_INODE_DELETED) {
 104                 spin_unlock(&oi->ip_lock);
 105
 106                 status = -ENOENT;
 107                 goto leave;
 108         }
 109
 110         if (mode & O_DIRECT)
 111                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 112
 113         oi->ip_open_count++;
 114         spin_unlock(&oi->ip_lock);
 115
 116         status = ocfs2_init_file_private(inode, file);
 117         if (status) {
 118                 /*
 119                  * We want to set open count back if we're failing the
 120                  * open.
 121                  */
 122                 spin_lock(&oi->ip_lock);
 123                 oi->ip_open_count--;
 124                 spin_unlock(&oi->ip_lock);
 125         }
 126
 127         file->f_mode |= FMODE_NOWAIT;
 128
 129 leave:
 130         return status;
 131 }
 132
 133 static int ocfs2_file_release(struct inode *inode, struct file *file)
 134 {
 135         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 136
 137         spin_lock(&oi->ip_lock);
 138         if (!--oi->ip_open_count)
 139                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 140
 141         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 142                                  oi->ip_blkno,
 143                                  file->f_path.dentry->d_name.len,
 144                                  file->f_path.dentry->d_name.name,
 145                                  oi->ip_open_count);
 146         spin_unlock(&oi->ip_lock);
 147
 148         ocfs2_free_file_private(inode, file);
 149
 150         return 0;
 151 }
 152
 153 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 154 {
 155         return ocfs2_init_file_private(inode, file);
 156 }
 157
 158 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 159 {
 160         ocfs2_free_file_private(inode, file);
 161         return 0;
 162 }
 163
 164 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 165                            int datasync)
 166 {
 167         int err = 0;
 168         struct inode *inode = file->f_mapping->host;
 169         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 170         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 171         journal_t *journal = osb->journal->j_journal;
 172         int ret;
 173         tid_t commit_tid;
 174         bool needs_barrier = false;
 175
 176         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 177                               oi->ip_blkno,
 178                               file->f_path.dentry->d_name.len,
 179                               file->f_path.dentry->d_name.name,
 180                               (unsigned long long)datasync);
 181
 182         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 183                 return -EROFS;
 184
 185         err = file_write_and_wait_range(file, start, end);
 186         if (err)
 187                 return err;
 188
 189         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 190         if (journal->j_flags & JBD2_BARRIER &&
 191             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 192                 needs_barrier = true;
 193         err = jbd2_complete_transaction(journal, commit_tid);
 194         if (needs_barrier) {
 195                 ret = blkdev_issue_flush(inode->i_sb->s_bdev);
 196                 if (!err)
 197                         err = ret;
 198         }
 199
 200         if (err)
 201                 mlog_errno(err);
 202
 203         return (err < 0) ? -EIO : 0;
 204 }
 205
 206 int ocfs2_should_update_atime(struct inode *inode,
 207                               struct vfsmount *vfsmnt)
 208 {
 209         struct timespec64 now;
 210         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 211
 212         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 213                 return 0;
 214
 215         if ((inode->i_flags & S_NOATIME) ||
 216             ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
 217                 return 0;
 218
 219         /*
 220          * We can be called with no vfsmnt structure - NFSD will
 221          * sometimes do this.
 222          *
 223          * Note that our action here is different than touch_atime() -
 224          * if we can't tell whether this is a noatime mount, then we
 225          * don't know whether to trust the value of s_atime_quantum.
 226          */
 227         if (vfsmnt == NULL)
 228                 return 0;
 229
 230         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 231             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 232                 return 0;
 233
 234         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 235                 struct timespec64 ctime = inode_get_ctime(inode);
 236                 struct timespec64 atime = inode_get_atime(inode);
 237                 struct timespec64 mtime = inode_get_mtime(inode);
 238
 239                 if ((timespec64_compare(&atime, &mtime) <= 0) ||
 240                     (timespec64_compare(&atime, &ctime) <= 0))
 241                         return 1;
 242
 243                 return 0;
 244         }
 245
 246         now = current_time(inode);
 247         if ((now.tv_sec - inode_get_atime_sec(inode) <= osb->s_atime_quantum))
 248                 return 0;
 249         else
 250                 return 1;
 251 }
 252
 253 int ocfs2_update_inode_atime(struct inode *inode,
 254                              struct buffer_head *bh)
 255 {
 256         int ret;
 257         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 258         handle_t *handle;
 259         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 260
 261         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 262         if (IS_ERR(handle)) {
 263                 ret = PTR_ERR(handle);
 264                 mlog_errno(ret);
 265                 goto out;
 266         }
 267
 268         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 269                                       OCFS2_JOURNAL_ACCESS_WRITE);
 270         if (ret) {
 271                 mlog_errno(ret);
 272                 goto out_commit;
 273         }
 274
 275         /*
 276          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 277          * have i_rwsem to guard against concurrent changes to other
 278          * inode fields.
 279          */
 280         inode_set_atime_to_ts(inode, current_time(inode));
 281         di->i_atime = cpu_to_le64(inode_get_atime_sec(inode));
 282         di->i_atime_nsec = cpu_to_le32(inode_get_atime_nsec(inode));
 283         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 284         ocfs2_journal_dirty(handle, bh);
 285
 286 out_commit:
 287         ocfs2_commit_trans(osb, handle);
 288 out:
 289         return ret;
 290 }
 291
 292 int ocfs2_set_inode_size(handle_t *handle,
 293                                 struct inode *inode,
 294                                 struct buffer_head *fe_bh,
 295                                 u64 new_i_size)
 296 {
 297         int status;
 298
 299         i_size_write(inode, new_i_size);
 300         inode->i_blocks = ocfs2_inode_sector_count(inode);
 301         inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
 302
 303         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 304         if (status < 0) {
 305                 mlog_errno(status);
 306                 goto bail;
 307         }
 308
 309 bail:
 310         return status;
 311 }
 312
 313 int ocfs2_simple_size_update(struct inode *inode,
 314                              struct buffer_head *di_bh,
 315                              u64 new_i_size)
 316 {
 317         int ret;
 318         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 319         handle_t *handle = NULL;
 320
 321         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 322         if (IS_ERR(handle)) {
 323                 ret = PTR_ERR(handle);
 324                 mlog_errno(ret);
 325                 goto out;
 326         }
 327
 328         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 329                                    new_i_size);
 330         if (ret < 0)
 331                 mlog_errno(ret);
 332
 333         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 334         ocfs2_commit_trans(osb, handle);
 335 out:
 336         return ret;
 337 }
 338
 339 static int ocfs2_cow_file_pos(struct inode *inode,
 340                               struct buffer_head *fe_bh,
 341                               u64 offset)
 342 {
 343         int status;
 344         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 345         unsigned int num_clusters = 0;
 346         unsigned int ext_flags = 0;
 347
 348         /*
 349          * If the new offset is aligned to the range of the cluster, there is
 350          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 351          * CoW either.
 352          */
 353         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 354                 return 0;
 355
 356         status = ocfs2_get_clusters(inode, cpos, &phys,
 357                                     &num_clusters, &ext_flags);
 358         if (status) {
 359                 mlog_errno(status);
 360                 goto out;
 361         }
 362
 363         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 364                 goto out;
 365
 366         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 367
 368 out:
 369         return status;
 370 }
 371
 372 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 373                                      struct inode *inode,
 374                                      struct buffer_head *fe_bh,
 375                                      u64 new_i_size)
 376 {
 377         int status;
 378         handle_t *handle;
 379         struct ocfs2_dinode *di;
 380         u64 cluster_bytes;
 381
 382         /*
 383          * We need to CoW the cluster contains the offset if it is reflinked
 384          * since we will call ocfs2_zero_range_for_truncate later which will
 385          * write "0" from offset to the end of the cluster.
 386          */
 387         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 388         if (status) {
 389                 mlog_errno(status);
 390                 return status;
 391         }
 392
 393         /* TODO: This needs to actually orphan the inode in this
 394          * transaction. */
 395
 396         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 397         if (IS_ERR(handle)) {
 398                 status = PTR_ERR(handle);
 399                 mlog_errno(status);
 400                 goto out;
 401         }
 402
 403         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 404                                          OCFS2_JOURNAL_ACCESS_WRITE);
 405         if (status < 0) {
 406                 mlog_errno(status);
 407                 goto out_commit;
 408         }
 409
 410         /*
 411          * Do this before setting i_size.
 412          */
 413         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 414         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 415                                                cluster_bytes);
 416         if (status) {
 417                 mlog_errno(status);
 418                 goto out_commit;
 419         }
 420
 421         i_size_write(inode, new_i_size);
 422         inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
 423
 424         di = (struct ocfs2_dinode *) fe_bh->b_data;
 425         di->i_size = cpu_to_le64(new_i_size);
 426         di->i_ctime = di->i_mtime = cpu_to_le64(inode_get_ctime_sec(inode));
 427         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
 428         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 429
 430         ocfs2_journal_dirty(handle, fe_bh);
 431
 432 out_commit:
 433         ocfs2_commit_trans(osb, handle);
 434 out:
 435         return status;
 436 }
 437
 438 int ocfs2_truncate_file(struct inode *inode,
 439                                struct buffer_head *di_bh,
 440                                u64 new_i_size)
 441 {
 442         int status = 0;
 443         struct ocfs2_dinode *fe = NULL;
 444         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 445
 446         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 447          * already validated it */
 448         fe = (struct ocfs2_dinode *) di_bh->b_data;
 449
 450         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 451                                   (unsigned long long)le64_to_cpu(fe->i_size),
 452                                   (unsigned long long)new_i_size);
 453
 454         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 455                         "Inode %llu, inode i_size = %lld != di "
 456                         "i_size = %llu, i_flags = 0x%x\n",
 457                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 458                         i_size_read(inode),
 459                         (unsigned long long)le64_to_cpu(fe->i_size),
 460                         le32_to_cpu(fe->i_flags));
 461
 462         if (new_i_size > le64_to_cpu(fe->i_size)) {
 463                 trace_ocfs2_truncate_file_error(
 464                         (unsigned long long)le64_to_cpu(fe->i_size),
 465                         (unsigned long long)new_i_size);
 466                 status = -EINVAL;
 467                 mlog_errno(status);
 468                 goto bail;
 469         }
 470
 471         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 472
 473         ocfs2_resv_discard(&osb->osb_la_resmap,
 474                            &OCFS2_I(inode)->ip_la_data_resv);
 475
 476         /*
 477          * The inode lock forced other nodes to sync and drop their
 478          * pages, which (correctly) happens even if we have a truncate
 479          * without allocation change - ocfs2 cluster sizes can be much
 480          * greater than page size, so we have to truncate them
 481          * anyway.
 482          */
 483
 484         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 485                 unmap_mapping_range(inode->i_mapping,
 486                                     new_i_size + PAGE_SIZE - 1, 0, 1);
 487                 truncate_inode_pages(inode->i_mapping, new_i_size);
 488                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 489                                                i_size_read(inode), 1);
 490                 if (status)
 491                         mlog_errno(status);
 492
 493                 goto bail_unlock_sem;
 494         }
 495
 496         /* alright, we're going to need to do a full blown alloc size
 497          * change. Orphan the inode so that recovery can complete the
 498          * truncate if necessary. This does the task of marking
 499          * i_size. */
 500         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 501         if (status < 0) {
 502                 mlog_errno(status);
 503                 goto bail_unlock_sem;
 504         }
 505
 506         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 507         truncate_inode_pages(inode->i_mapping, new_i_size);
 508
 509         status = ocfs2_commit_truncate(osb, inode, di_bh);
 510         if (status < 0) {
 511                 mlog_errno(status);
 512                 goto bail_unlock_sem;
 513         }
 514
 515         /* TODO: orphan dir cleanup here. */
 516 bail_unlock_sem:
 517         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 518
 519 bail:
 520         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 521                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 522
 523         return status;
 524 }
 525
 526 /*
 527  * extend file allocation only here.
 528  * we'll update all the disk stuff, and oip->alloc_size
 529  *
 530  * expect stuff to be locked, a transaction started and enough data /
 531  * metadata reservations in the contexts.
 532  *
 533  * Will return -EAGAIN, and a reason if a restart is needed.
 534  * If passed in, *reason will always be set, even in error.
 535  */
 536 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 537                          struct inode *inode,
 538                          u32 *logical_offset,
 539                          u32 clusters_to_add,
 540                          int mark_unwritten,
 541                          struct buffer_head *fe_bh,
 542                          handle_t *handle,
 543                          struct ocfs2_alloc_context *data_ac,
 544                          struct ocfs2_alloc_context *meta_ac,
 545                          enum ocfs2_alloc_restarted *reason_ret)
 546 {
 547         struct ocfs2_extent_tree et;
 548
 549         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 550         return ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 551                                            clusters_to_add, mark_unwritten,
 552                                            data_ac, meta_ac, reason_ret);
 553 }
 554
 555 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 556                                    u32 clusters_to_add, int mark_unwritten)
 557 {
 558         int status = 0;
 559         int restart_func = 0;
 560         int credits;
 561         u32 prev_clusters;
 562         struct buffer_head *bh = NULL;
 563         struct ocfs2_dinode *fe = NULL;
 564         handle_t *handle = NULL;
 565         struct ocfs2_alloc_context *data_ac = NULL;
 566         struct ocfs2_alloc_context *meta_ac = NULL;
 567         enum ocfs2_alloc_restarted why = RESTART_NONE;
 568         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 569         struct ocfs2_extent_tree et;
 570         int did_quota = 0;
 571
 572         /*
 573          * Unwritten extent only exists for file systems which
 574          * support holes.
 575          */
 576         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 577
 578         status = ocfs2_read_inode_block(inode, &bh);
 579         if (status < 0) {
 580                 mlog_errno(status);
 581                 goto leave;
 582         }
 583         fe = (struct ocfs2_dinode *) bh->b_data;
 584
 585 restart_all:
 586         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 587
 588         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 589         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 590                                        &data_ac, &meta_ac);
 591         if (status) {
 592                 mlog_errno(status);
 593                 goto leave;
 594         }
 595
 596         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 597         handle = ocfs2_start_trans(osb, credits);
 598         if (IS_ERR(handle)) {
 599                 status = PTR_ERR(handle);
 600                 handle = NULL;
 601                 mlog_errno(status);
 602                 goto leave;
 603         }
 604
 605 restarted_transaction:
 606         trace_ocfs2_extend_allocation(
 607                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 608                 (unsigned long long)i_size_read(inode),
 609                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 610                 why, restart_func);
 611
 612         status = dquot_alloc_space_nodirty(inode,
 613                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 614         if (status)
 615                 goto leave;
 616         did_quota = 1;
 617
 618         /* reserve a write to the file entry early on - that we if we
 619          * run out of credits in the allocation path, we can still
 620          * update i_size. */
 621         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 622                                          OCFS2_JOURNAL_ACCESS_WRITE);
 623         if (status < 0) {
 624                 mlog_errno(status);
 625                 goto leave;
 626         }
 627
 628         prev_clusters = OCFS2_I(inode)->ip_clusters;
 629
 630         status = ocfs2_add_inode_data(osb,
 631                                       inode,
 632                                       &logical_start,
 633                                       clusters_to_add,
 634                                       mark_unwritten,
 635                                       bh,
 636                                       handle,
 637                                       data_ac,
 638                                       meta_ac,
 639                                       &why);
 640         if ((status < 0) && (status != -EAGAIN)) {
 641                 if (status != -ENOSPC)
 642                         mlog_errno(status);
 643                 goto leave;
 644         }
 645         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 646         ocfs2_journal_dirty(handle, bh);
 647
 648         spin_lock(&OCFS2_I(inode)->ip_lock);
 649         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 650         spin_unlock(&OCFS2_I(inode)->ip_lock);
 651         /* Release unused quota reservation */
 652         dquot_free_space(inode,
 653                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 654         did_quota = 0;
 655
 656         if (why != RESTART_NONE && clusters_to_add) {
 657                 if (why == RESTART_META) {
 658                         restart_func = 1;
 659                         status = 0;
 660                 } else {
 661                         BUG_ON(why != RESTART_TRANS);
 662
 663                         status = ocfs2_allocate_extend_trans(handle, 1);
 664                         if (status < 0) {
 665                                 /* handle still has to be committed at
 666                                  * this point. */
 667                                 status = -ENOMEM;
 668                                 mlog_errno(status);
 669                                 goto leave;
 670                         }
 671                         goto restarted_transaction;
 672                 }
 673         }
 674
 675         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 676              le32_to_cpu(fe->i_clusters),
 677              (unsigned long long)le64_to_cpu(fe->i_size),
 678              OCFS2_I(inode)->ip_clusters,
 679              (unsigned long long)i_size_read(inode));
 680
 681 leave:
 682         if (status < 0 && did_quota)
 683                 dquot_free_space(inode,
 684                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 685         if (handle) {
 686                 ocfs2_commit_trans(osb, handle);
 687                 handle = NULL;
 688         }
 689         if (data_ac) {
 690                 ocfs2_free_alloc_context(data_ac);
 691                 data_ac = NULL;
 692         }
 693         if (meta_ac) {
 694                 ocfs2_free_alloc_context(meta_ac);
 695                 meta_ac = NULL;
 696         }
 697         if ((!status) && restart_func) {
 698                 restart_func = 0;
 699                 goto restart_all;
 700         }
 701         brelse(bh);
 702         bh = NULL;
 703
 704         return status;
 705 }
 706
 707 /*
 708  * While a write will already be ordering the data, a truncate will not.
 709  * Thus, we need to explicitly order the zeroed pages.
 710  */
 711 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 712                                                       struct buffer_head *di_bh,
 713                                                       loff_t start_byte,
 714                                                       loff_t length)
 715 {
 716         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 717         handle_t *handle = NULL;
 718         int ret = 0;
 719
 720         if (!ocfs2_should_order_data(inode))
 721                 goto out;
 722
 723         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 724         if (IS_ERR(handle)) {
 725                 ret = -ENOMEM;
 726                 mlog_errno(ret);
 727                 goto out;
 728         }
 729
 730         ret = ocfs2_jbd2_inode_add_write(handle, inode, start_byte, length);
 731         if (ret < 0) {
 732                 mlog_errno(ret);
 733                 goto out;
 734         }
 735
 736         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 737                                       OCFS2_JOURNAL_ACCESS_WRITE);
 738         if (ret)
 739                 mlog_errno(ret);
 740         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 741
 742 out:
 743         if (ret) {
 744                 if (!IS_ERR(handle))
 745                         ocfs2_commit_trans(osb, handle);
 746                 handle = ERR_PTR(ret);
 747         }
 748         return handle;
 749 }
 750
 751 /* Some parts of this taken from generic_cont_expand, which turned out
 752  * to be too fragile to do exactly what we need without us having to
 753  * worry about recursive locking in ->write_begin() and ->write_end(). */
 754 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 755                                  u64 abs_to, struct buffer_head *di_bh)
 756 {
 757         struct address_space *mapping = inode->i_mapping;
 758         struct page *page;
 759         unsigned long index = abs_from >> PAGE_SHIFT;
 760         handle_t *handle;
 761         int ret = 0;
 762         unsigned zero_from, zero_to, block_start, block_end;
 763         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 764
 765         BUG_ON(abs_from >= abs_to);
 766         BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 767         BUG_ON(abs_from & (inode->i_blkbits - 1));
 768
 769         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh,
 770                                                       abs_from,
 771                                                       abs_to - abs_from);
 772         if (IS_ERR(handle)) {
 773                 ret = PTR_ERR(handle);
 774                 goto out;
 775         }
 776
 777         page = find_or_create_page(mapping, index, GFP_NOFS);
 778         if (!page) {
 779                 ret = -ENOMEM;
 780                 mlog_errno(ret);
 781                 goto out_commit_trans;
 782         }
 783
 784         /* Get the offsets within the page that we want to zero */
 785         zero_from = abs_from & (PAGE_SIZE - 1);
 786         zero_to = abs_to & (PAGE_SIZE - 1);
 787         if (!zero_to)
 788                 zero_to = PAGE_SIZE;
 789
 790         trace_ocfs2_write_zero_page(
 791                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 792                         (unsigned long long)abs_from,
 793                         (unsigned long long)abs_to,
 794                         index, zero_from, zero_to);
 795
 796         /* We know that zero_from is block aligned */
 797         for (block_start = zero_from; block_start < zero_to;
 798              block_start = block_end) {
 799                 block_end = block_start + i_blocksize(inode);
 800
 801                 /*
 802                  * block_start is block-aligned.  Bump it by one to force
 803                  * __block_write_begin and block_commit_write to zero the
 804                  * whole block.
 805                  */
 806                 ret = __block_write_begin(page, block_start + 1, 0,
 807                                           ocfs2_get_block);
 808                 if (ret < 0) {
 809                         mlog_errno(ret);
 810                         goto out_unlock;
 811                 }
 812
 813
 814                 /* must not update i_size! */
 815                 block_commit_write(page, block_start + 1, block_start + 1);
 816         }
 817
 818         /*
 819          * fs-writeback will release the dirty pages without page lock
 820          * whose offset are over inode size, the release happens at
 821          * block_write_full_folio().
 822          */
 823         i_size_write(inode, abs_to);
 824         inode->i_blocks = ocfs2_inode_sector_count(inode);
 825         di->i_size = cpu_to_le64((u64)i_size_read(inode));
 826         inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
 827         di->i_mtime = di->i_ctime = cpu_to_le64(inode_get_mtime_sec(inode));
 828         di->i_ctime_nsec = cpu_to_le32(inode_get_mtime_nsec(inode));
 829         di->i_mtime_nsec = di->i_ctime_nsec;
 830         if (handle) {
 831                 ocfs2_journal_dirty(handle, di_bh);
 832                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 833         }
 834
 835 out_unlock:
 836         unlock_page(page);
 837         put_page(page);
 838 out_commit_trans:
 839         if (handle)
 840                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 841 out:
 842         return ret;
 843 }
 844
 845 /*
 846  * Find the next range to zero.  We do this in terms of bytes because
 847  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 848  * pagecache.  We may return multiple extents.
 849  *
 850  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 851  * needs to be zeroed.  range_start and range_end return the next zeroing
 852  * range.  A subsequent call should pass the previous range_end as its
 853  * zero_start.  If range_end is 0, there's nothing to do.
 854  *
 855  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 856  */
 857 static int ocfs2_zero_extend_get_range(struct inode *inode,
 858                                        struct buffer_head *di_bh,
 859                                        u64 zero_start, u64 zero_end,
 860                                        u64 *range_start, u64 *range_end)
 861 {
 862         int rc = 0, needs_cow = 0;
 863         u32 p_cpos, zero_clusters = 0;
 864         u32 zero_cpos =
 865                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 866         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 867         unsigned int num_clusters = 0;
 868         unsigned int ext_flags = 0;
 869
 870         while (zero_cpos < last_cpos) {
 871                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 872                                         &num_clusters, &ext_flags);
 873                 if (rc) {
 874                         mlog_errno(rc);
 875                         goto out;
 876                 }
 877
 878                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 879                         zero_clusters = num_clusters;
 880                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 881                                 needs_cow = 1;
 882                         break;
 883                 }
 884
 885                 zero_cpos += num_clusters;
 886         }
 887         if (!zero_clusters) {
 888                 *range_end = 0;
 889                 goto out;
 890         }
 891
 892         while ((zero_cpos + zero_clusters) < last_cpos) {
 893                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 894                                         &p_cpos, &num_clusters,
 895                                         &ext_flags);
 896                 if (rc) {
 897                         mlog_errno(rc);
 898                         goto out;
 899                 }
 900
 901                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 902                         break;
 903                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 904                         needs_cow = 1;
 905                 zero_clusters += num_clusters;
 906         }
 907         if ((zero_cpos + zero_clusters) > last_cpos)
 908                 zero_clusters = last_cpos - zero_cpos;
 909
 910         if (needs_cow) {
 911                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 912                                         zero_clusters, UINT_MAX);
 913                 if (rc) {
 914                         mlog_errno(rc);
 915                         goto out;
 916                 }
 917         }
 918
 919         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 920         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 921                                              zero_cpos + zero_clusters);
 922
 923 out:
 924         return rc;
 925 }
 926
 927 /*
 928  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 929  * has made sure that the entire range needs zeroing.
 930  */
 931 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 932                                    u64 range_end, struct buffer_head *di_bh)
 933 {
 934         int rc = 0;
 935         u64 next_pos;
 936         u64 zero_pos = range_start;
 937
 938         trace_ocfs2_zero_extend_range(
 939                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 940                         (unsigned long long)range_start,
 941                         (unsigned long long)range_end);
 942         BUG_ON(range_start >= range_end);
 943
 944         while (zero_pos < range_end) {
 945                 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 946                 if (next_pos > range_end)
 947                         next_pos = range_end;
 948                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 949                 if (rc < 0) {
 950                         mlog_errno(rc);
 951                         break;
 952                 }
 953                 zero_pos = next_pos;
 954
 955                 /*
 956                  * Very large extends have the potential to lock up
 957                  * the cpu for extended periods of time.
 958                  */
 959                 cond_resched();
 960         }
 961
 962         return rc;
 963 }
 964
 965 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 966                       loff_t zero_to_size)
 967 {
 968         int ret = 0;
 969         u64 zero_start, range_start = 0, range_end = 0;
 970         struct super_block *sb = inode->i_sb;
 971
 972         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 973         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 974                                 (unsigned long long)zero_start,
 975                                 (unsigned long long)i_size_read(inode));
 976         while (zero_start < zero_to_size) {
 977                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 978                                                   zero_to_size,
 979                                                   &range_start,
 980                                                   &range_end);
 981                 if (ret) {
 982                         mlog_errno(ret);
 983                         break;
 984                 }
 985                 if (!range_end)
 986                         break;
 987                 /* Trim the ends */
 988                 if (range_start < zero_start)
 989                         range_start = zero_start;
 990                 if (range_end > zero_to_size)
 991                         range_end = zero_to_size;
 992
 993                 ret = ocfs2_zero_extend_range(inode, range_start,
 994                                               range_end, di_bh);
 995                 if (ret) {
 996                         mlog_errno(ret);
 997                         break;
 998                 }
 999                 zero_start = range_end;
1000         }
1001
1002         return ret;
1003 }
1004
1005 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1006                           u64 new_i_size, u64 zero_to)
1007 {
1008         int ret;
1009         u32 clusters_to_add;
1010         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1011
1012         /*
1013          * Only quota files call this without a bh, and they can't be
1014          * refcounted.
1015          */
1016         BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1017         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1018
1019         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1020         if (clusters_to_add < oi->ip_clusters)
1021                 clusters_to_add = 0;
1022         else
1023                 clusters_to_add -= oi->ip_clusters;
1024
1025         if (clusters_to_add) {
1026                 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1027                                               clusters_to_add, 0);
1028                 if (ret) {
1029                         mlog_errno(ret);
1030                         goto out;
1031                 }
1032         }
1033
1034         /*
1035          * Call this even if we don't add any clusters to the tree. We
1036          * still need to zero the area between the old i_size and the
1037          * new i_size.
1038          */
1039         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1040         if (ret < 0)
1041                 mlog_errno(ret);
1042
1043 out:
1044         return ret;
1045 }
1046
1047 static int ocfs2_extend_file(struct inode *inode,
1048                              struct buffer_head *di_bh,
1049                              u64 new_i_size)
1050 {
1051         int ret = 0;
1052         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1053
1054         BUG_ON(!di_bh);
1055
1056         /* setattr sometimes calls us like this. */
1057         if (new_i_size == 0)
1058                 goto out;
1059
1060         if (i_size_read(inode) == new_i_size)
1061                 goto out;
1062         BUG_ON(new_i_size < i_size_read(inode));
1063
1064         /*
1065          * The alloc sem blocks people in read/write from reading our
1066          * allocation until we're done changing it. We depend on
1067          * i_rwsem to block other extend/truncate calls while we're
1068          * here.  We even have to hold it for sparse files because there
1069          * might be some tail zeroing.
1070          */
1071         down_write(&oi->ip_alloc_sem);
1072
1073         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1074                 /*
1075                  * We can optimize small extends by keeping the inodes
1076                  * inline data.
1077                  */
1078                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1079                         up_write(&oi->ip_alloc_sem);
1080                         goto out_update_size;
1081                 }
1082
1083                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1084                 if (ret) {
1085                         up_write(&oi->ip_alloc_sem);
1086                         mlog_errno(ret);
1087                         goto out;
1088                 }
1089         }
1090
1091         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1092                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1093         else
1094                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1095                                             new_i_size);
1096
1097         up_write(&oi->ip_alloc_sem);
1098
1099         if (ret < 0) {
1100                 mlog_errno(ret);
1101                 goto out;
1102         }
1103
1104 out_update_size:
1105         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1106         if (ret < 0)
1107                 mlog_errno(ret);
1108
1109 out:
1110         return ret;
1111 }
1112
1113 int ocfs2_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
1114                   struct iattr *attr)
1115 {
1116         int status = 0, size_change;
1117         int inode_locked = 0;
1118         struct inode *inode = d_inode(dentry);
1119         struct super_block *sb = inode->i_sb;
1120         struct ocfs2_super *osb = OCFS2_SB(sb);
1121         struct buffer_head *bh = NULL;
1122         handle_t *handle = NULL;
1123         struct dquot *transfer_to[MAXQUOTAS] = { };
1124         int qtype;
1125         int had_lock;
1126         struct ocfs2_lock_holder oh;
1127
1128         trace_ocfs2_setattr(inode, dentry,
1129                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1130                             dentry->d_name.len, dentry->d_name.name,
1131                             attr->ia_valid, attr->ia_mode,
1132                             from_kuid(&init_user_ns, attr->ia_uid),
1133                             from_kgid(&init_user_ns, attr->ia_gid));
1134
1135         /* ensuring we don't even attempt to truncate a symlink */
1136         if (S_ISLNK(inode->i_mode))
1137                 attr->ia_valid &= ~ATTR_SIZE;
1138
1139 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1140                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1141         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1142                 return 0;
1143
1144         status = setattr_prepare(&nop_mnt_idmap, dentry, attr);
1145         if (status)
1146                 return status;
1147
1148         if (is_quota_modification(&nop_mnt_idmap, inode, attr)) {
1149                 status = dquot_initialize(inode);
1150                 if (status)
1151                         return status;
1152         }
1153         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1154         if (size_change) {
1155                 /*
1156                  * Here we should wait dio to finish before inode lock
1157                  * to avoid a deadlock between ocfs2_setattr() and
1158                  * ocfs2_dio_end_io_write()
1159                  */
1160                 inode_dio_wait(inode);
1161
1162                 status = ocfs2_rw_lock(inode, 1);
1163                 if (status < 0) {
1164                         mlog_errno(status);
1165                         goto bail;
1166                 }
1167         }
1168
1169         had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1170         if (had_lock < 0) {
1171                 status = had_lock;
1172                 goto bail_unlock_rw;
1173         } else if (had_lock) {
1174                 /*
1175                  * As far as we know, ocfs2_setattr() could only be the first
1176                  * VFS entry point in the call chain of recursive cluster
1177                  * locking issue.
1178                  *
1179                  * For instance:
1180                  * chmod_common()
1181                  *  notify_change()
1182                  *   ocfs2_setattr()
1183                  *    posix_acl_chmod()
1184                  *     ocfs2_iop_get_acl()
1185                  *
1186                  * But, we're not 100% sure if it's always true, because the
1187                  * ordering of the VFS entry points in the call chain is out
1188                  * of our control. So, we'd better dump the stack here to
1189                  * catch the other cases of recursive locking.
1190                  */
1191                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1192                 dump_stack();
1193         }
1194         inode_locked = 1;
1195
1196         if (size_change) {
1197                 status = inode_newsize_ok(inode, attr->ia_size);
1198                 if (status)
1199                         goto bail_unlock;
1200
1201                 if (i_size_read(inode) >= attr->ia_size) {
1202                         if (ocfs2_should_order_data(inode)) {
1203                                 status = ocfs2_begin_ordered_truncate(inode,
1204                                                                       attr->ia_size);
1205                                 if (status)
1206                                         goto bail_unlock;
1207                         }
1208                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1209                 } else
1210                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1211                 if (status < 0) {
1212                         if (status != -ENOSPC)
1213                                 mlog_errno(status);
1214                         status = -ENOSPC;
1215                         goto bail_unlock;
1216                 }
1217         }
1218
1219         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1220             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1221                 /*
1222                  * Gather pointers to quota structures so that allocation /
1223                  * freeing of quota structures happens here and not inside
1224                  * dquot_transfer() where we have problems with lock ordering
1225                  */
1226                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1227                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1228                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1229                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1230                         if (IS_ERR(transfer_to[USRQUOTA])) {
1231                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1232                                 transfer_to[USRQUOTA] = NULL;
1233                                 goto bail_unlock;
1234                         }
1235                 }
1236                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1237                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1238                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1239                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1240                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1241                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1242                                 transfer_to[GRPQUOTA] = NULL;
1243                                 goto bail_unlock;
1244                         }
1245                 }
1246                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1247                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1248                                            2 * ocfs2_quota_trans_credits(sb));
1249                 if (IS_ERR(handle)) {
1250                         status = PTR_ERR(handle);
1251                         mlog_errno(status);
1252                         goto bail_unlock_alloc;
1253                 }
1254                 status = __dquot_transfer(inode, transfer_to);
1255                 if (status < 0)
1256                         goto bail_commit;
1257         } else {
1258                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1259                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1260                 if (IS_ERR(handle)) {
1261                         status = PTR_ERR(handle);
1262                         mlog_errno(status);
1263                         goto bail_unlock_alloc;
1264                 }
1265         }
1266
1267         setattr_copy(&nop_mnt_idmap, inode, attr);
1268         mark_inode_dirty(inode);
1269
1270         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1271         if (status < 0)
1272                 mlog_errno(status);
1273
1274 bail_commit:
1275         ocfs2_commit_trans(osb, handle);
1276 bail_unlock_alloc:
1277         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1278 bail_unlock:
1279         if (status && inode_locked) {
1280                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1281                 inode_locked = 0;
1282         }
1283 bail_unlock_rw:
1284         if (size_change)
1285                 ocfs2_rw_unlock(inode, 1);
1286 bail:
1287
1288         /* Release quota pointers in case we acquired them */
1289         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1290                 dqput(transfer_to[qtype]);
1291
1292         if (!status && attr->ia_valid & ATTR_MODE) {
1293                 status = ocfs2_acl_chmod(inode, bh);
1294                 if (status < 0)
1295                         mlog_errno(status);
1296         }
1297         if (inode_locked)
1298                 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1299
1300         brelse(bh);
1301         return status;
1302 }
1303
1304 int ocfs2_getattr(struct mnt_idmap *idmap, const struct path *path,
1305                   struct kstat *stat, u32 request_mask, unsigned int flags)
1306 {
1307         struct inode *inode = d_inode(path->dentry);
1308         struct super_block *sb = path->dentry->d_sb;
1309         struct ocfs2_super *osb = sb->s_fs_info;
1310         int err;
1311
1312         err = ocfs2_inode_revalidate(path->dentry);
1313         if (err) {
1314                 if (err != -ENOENT)
1315                         mlog_errno(err);
1316                 goto bail;
1317         }
1318
1319         generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1320         /*
1321          * If there is inline data in the inode, the inode will normally not
1322          * have data blocks allocated (it may have an external xattr block).
1323          * Report at least one sector for such files, so tools like tar, rsync,
1324          * others don't incorrectly think the file is completely sparse.
1325          */
1326         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1327                 stat->blocks += (stat->size + 511)>>9;
1328
1329         /* We set the blksize from the cluster size for performance */
1330         stat->blksize = osb->s_clustersize;
1331
1332 bail:
1333         return err;
1334 }
1335
1336 int ocfs2_permission(struct mnt_idmap *idmap, struct inode *inode,
1337                      int mask)
1338 {
1339         int ret, had_lock;
1340         struct ocfs2_lock_holder oh;
1341
1342         if (mask & MAY_NOT_BLOCK)
1343                 return -ECHILD;
1344
1345         had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1346         if (had_lock < 0) {
1347                 ret = had_lock;
1348                 goto out;
1349         } else if (had_lock) {
1350                 /* See comments in ocfs2_setattr() for details.
1351                  * The call chain of this case could be:
1352                  * do_sys_open()
1353                  *  may_open()
1354                  *   inode_permission()
1355                  *    ocfs2_permission()
1356                  *     ocfs2_iop_get_acl()
1357                  */
1358                 mlog(ML_ERROR, "Another case of recursive locking:\n");
1359                 dump_stack();
1360         }
1361
1362         ret = generic_permission(&nop_mnt_idmap, inode, mask);
1363
1364         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1365 out:
1366         return ret;
1367 }
1368
1369 static int __ocfs2_write_remove_suid(struct inode *inode,
1370                                      struct buffer_head *bh)
1371 {
1372         int ret;
1373         handle_t *handle;
1374         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1375         struct ocfs2_dinode *di;
1376
1377         trace_ocfs2_write_remove_suid(
1378                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1379                         inode->i_mode);
1380
1381         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1382         if (IS_ERR(handle)) {
1383                 ret = PTR_ERR(handle);
1384                 mlog_errno(ret);
1385                 goto out;
1386         }
1387
1388         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1389                                       OCFS2_JOURNAL_ACCESS_WRITE);
1390         if (ret < 0) {
1391                 mlog_errno(ret);
1392                 goto out_trans;
1393         }
1394
1395         inode->i_mode &= ~S_ISUID;
1396         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1397                 inode->i_mode &= ~S_ISGID;
1398
1399         di = (struct ocfs2_dinode *) bh->b_data;
1400         di->i_mode = cpu_to_le16(inode->i_mode);
1401         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1402
1403         ocfs2_journal_dirty(handle, bh);
1404
1405 out_trans:
1406         ocfs2_commit_trans(osb, handle);
1407 out:
1408         return ret;
1409 }
1410
1411 static int ocfs2_write_remove_suid(struct inode *inode)
1412 {
1413         int ret;
1414         struct buffer_head *bh = NULL;
1415
1416         ret = ocfs2_read_inode_block(inode, &bh);
1417         if (ret < 0) {
1418                 mlog_errno(ret);
1419                 goto out;
1420         }
1421
1422         ret =  __ocfs2_write_remove_suid(inode, bh);
1423 out:
1424         brelse(bh);
1425         return ret;
1426 }
1427
1428 /*
1429  * Allocate enough extents to cover the region starting at byte offset
1430  * start for len bytes. Existing extents are skipped, any extents
1431  * added are marked as "unwritten".
1432  */
1433 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1434                                             u64 start, u64 len)
1435 {
1436         int ret;
1437         u32 cpos, phys_cpos, clusters, alloc_size;
1438         u64 end = start + len;
1439         struct buffer_head *di_bh = NULL;
1440
1441         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1442                 ret = ocfs2_read_inode_block(inode, &di_bh);
1443                 if (ret) {
1444                         mlog_errno(ret);
1445                         goto out;
1446                 }
1447
1448                 /*
1449                  * Nothing to do if the requested reservation range
1450                  * fits within the inode.
1451                  */
1452                 if (ocfs2_size_fits_inline_data(di_bh, end))
1453                         goto out;
1454
1455                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1456                 if (ret) {
1457                         mlog_errno(ret);
1458                         goto out;
1459                 }
1460         }
1461
1462         /*
1463          * We consider both start and len to be inclusive.
1464          */
1465         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1466         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1467         clusters -= cpos;
1468
1469         while (clusters) {
1470                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1471                                          &alloc_size, NULL);
1472                 if (ret) {
1473                         mlog_errno(ret);
1474                         goto out;
1475                 }
1476
1477                 /*
1478                  * Hole or existing extent len can be arbitrary, so
1479                  * cap it to our own allocation request.
1480                  */
1481                 if (alloc_size > clusters)
1482                         alloc_size = clusters;
1483
1484                 if (phys_cpos) {
1485                         /*
1486                          * We already have an allocation at this
1487                          * region so we can safely skip it.
1488                          */
1489                         goto next;
1490                 }
1491
1492                 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1493                 if (ret) {
1494                         if (ret != -ENOSPC)
1495                                 mlog_errno(ret);
1496                         goto out;
1497                 }
1498
1499 next:
1500                 cpos += alloc_size;
1501                 clusters -= alloc_size;
1502         }
1503
1504         ret = 0;
1505 out:
1506
1507         brelse(di_bh);
1508         return ret;
1509 }
1510
1511 /*
1512  * Truncate a byte range, avoiding pages within partial clusters. This
1513  * preserves those pages for the zeroing code to write to.
1514  */
1515 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1516                                          u64 byte_len)
1517 {
1518         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1519         loff_t start, end;
1520         struct address_space *mapping = inode->i_mapping;
1521
1522         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1523         end = byte_start + byte_len;
1524         end = end & ~(osb->s_clustersize - 1);
1525
1526         if (start < end) {
1527                 unmap_mapping_range(mapping, start, end - start, 0);
1528                 truncate_inode_pages_range(mapping, start, end - 1);
1529         }
1530 }
1531
1532 /*
1533  * zero out partial blocks of one cluster.
1534  *
1535  * start: file offset where zero starts, will be made upper block aligned.
1536  * len: it will be trimmed to the end of current cluster if "start + len"
1537  *      is bigger than it.
1538  */
1539 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1540                                         u64 start, u64 len)
1541 {
1542         int ret;
1543         u64 start_block, end_block, nr_blocks;
1544         u64 p_block, offset;
1545         u32 cluster, p_cluster, nr_clusters;
1546         struct super_block *sb = inode->i_sb;
1547         u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1548
1549         if (start + len < end)
1550                 end = start + len;
1551
1552         start_block = ocfs2_blocks_for_bytes(sb, start);
1553         end_block = ocfs2_blocks_for_bytes(sb, end);
1554         nr_blocks = end_block - start_block;
1555         if (!nr_blocks)
1556                 return 0;
1557
1558         cluster = ocfs2_bytes_to_clusters(sb, start);
1559         ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1560                                 &nr_clusters, NULL);
1561         if (ret)
1562                 return ret;
1563         if (!p_cluster)
1564                 return 0;
1565
1566         offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1567         p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1568         return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1569 }
1570
1571 static int ocfs2_zero_partial_clusters(struct inode *inode,
1572                                        u64 start, u64 len)
1573 {
1574         int ret = 0;
1575         u64 tmpend = 0;
1576         u64 end = start + len;
1577         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1578         unsigned int csize = osb->s_clustersize;
1579         handle_t *handle;
1580         loff_t isize = i_size_read(inode);
1581
1582         /*
1583          * The "start" and "end" values are NOT necessarily part of
1584          * the range whose allocation is being deleted. Rather, this
1585          * is what the user passed in with the request. We must zero
1586          * partial clusters here. There's no need to worry about
1587          * physical allocation - the zeroing code knows to skip holes.
1588          */
1589         trace_ocfs2_zero_partial_clusters(
1590                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1591                 (unsigned long long)start, (unsigned long long)end);
1592
1593         /*
1594          * If both edges are on a cluster boundary then there's no
1595          * zeroing required as the region is part of the allocation to
1596          * be truncated.
1597          */
1598         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1599                 goto out;
1600
1601         /* No page cache for EOF blocks, issue zero out to disk. */
1602         if (end > isize) {
1603                 /*
1604                  * zeroout eof blocks in last cluster starting from
1605                  * "isize" even "start" > "isize" because it is
1606                  * complicated to zeroout just at "start" as "start"
1607                  * may be not aligned with block size, buffer write
1608                  * would be required to do that, but out of eof buffer
1609                  * write is not supported.
1610                  */
1611                 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1612                                         end - isize);
1613                 if (ret) {
1614                         mlog_errno(ret);
1615                         goto out;
1616                 }
1617                 if (start >= isize)
1618                         goto out;
1619                 end = isize;
1620         }
1621         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1622         if (IS_ERR(handle)) {
1623                 ret = PTR_ERR(handle);
1624                 mlog_errno(ret);
1625                 goto out;
1626         }
1627
1628         /*
1629          * If start is on a cluster boundary and end is somewhere in another
1630          * cluster, we have not COWed the cluster starting at start, unless
1631          * end is also within the same cluster. So, in this case, we skip this
1632          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1633          * to the next one.
1634          */
1635         if ((start & (csize - 1)) != 0) {
1636                 /*
1637                  * We want to get the byte offset of the end of the 1st
1638                  * cluster.
1639                  */
1640                 tmpend = (u64)osb->s_clustersize +
1641                         (start & ~(osb->s_clustersize - 1));
1642                 if (tmpend > end)
1643                         tmpend = end;
1644
1645                 trace_ocfs2_zero_partial_clusters_range1(
1646                         (unsigned long long)start,
1647                         (unsigned long long)tmpend);
1648
1649                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1650                                                     tmpend);
1651                 if (ret)
1652                         mlog_errno(ret);
1653         }
1654
1655         if (tmpend < end) {
1656                 /*
1657                  * This may make start and end equal, but the zeroing
1658                  * code will skip any work in that case so there's no
1659                  * need to catch it up here.
1660                  */
1661                 start = end & ~(osb->s_clustersize - 1);
1662
1663                 trace_ocfs2_zero_partial_clusters_range2(
1664                         (unsigned long long)start, (unsigned long long)end);
1665
1666                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1667                 if (ret)
1668                         mlog_errno(ret);
1669         }
1670         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1671
1672         ocfs2_commit_trans(osb, handle);
1673 out:
1674         return ret;
1675 }
1676
1677 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1678 {
1679         int i;
1680         struct ocfs2_extent_rec *rec = NULL;
1681
1682         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1683
1684                 rec = &el->l_recs[i];
1685
1686                 if (le32_to_cpu(rec->e_cpos) < pos)
1687                         break;
1688         }
1689
1690         return i;
1691 }
1692
1693 /*
1694  * Helper to calculate the punching pos and length in one run, we handle the
1695  * following three cases in order:
1696  *
1697  * - remove the entire record
1698  * - remove a partial record
1699  * - no record needs to be removed (hole-punching completed)
1700 */
1701 static void ocfs2_calc_trunc_pos(struct inode *inode,
1702                                  struct ocfs2_extent_list *el,
1703                                  struct ocfs2_extent_rec *rec,
1704                                  u32 trunc_start, u32 *trunc_cpos,
1705                                  u32 *trunc_len, u32 *trunc_end,
1706                                  u64 *blkno, int *done)
1707 {
1708         int ret = 0;
1709         u32 coff, range;
1710
1711         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1712
1713         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1714                 /*
1715                  * remove an entire extent record.
1716                  */
1717                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1718                 /*
1719                  * Skip holes if any.
1720                  */
1721                 if (range < *trunc_end)
1722                         *trunc_end = range;
1723                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1724                 *blkno = le64_to_cpu(rec->e_blkno);
1725                 *trunc_end = le32_to_cpu(rec->e_cpos);
1726         } else if (range > trunc_start) {
1727                 /*
1728                  * remove a partial extent record, which means we're
1729                  * removing the last extent record.
1730                  */
1731                 *trunc_cpos = trunc_start;
1732                 /*
1733                  * skip hole if any.
1734                  */
1735                 if (range < *trunc_end)
1736                         *trunc_end = range;
1737                 *trunc_len = *trunc_end - trunc_start;
1738                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1739                 *blkno = le64_to_cpu(rec->e_blkno) +
1740                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1741                 *trunc_end = trunc_start;
1742         } else {
1743                 /*
1744                  * It may have two following possibilities:
1745                  *
1746                  * - last record has been removed
1747                  * - trunc_start was within a hole
1748                  *
1749                  * both two cases mean the completion of hole punching.
1750                  */
1751                 ret = 1;
1752         }
1753
1754         *done = ret;
1755 }
1756
1757 int ocfs2_remove_inode_range(struct inode *inode,
1758                              struct buffer_head *di_bh, u64 byte_start,
1759                              u64 byte_len)
1760 {
1761         int ret = 0, flags = 0, done = 0, i;
1762         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1763         u32 cluster_in_el;
1764         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1765         struct ocfs2_cached_dealloc_ctxt dealloc;
1766         struct address_space *mapping = inode->i_mapping;
1767         struct ocfs2_extent_tree et;
1768         struct ocfs2_path *path = NULL;
1769         struct ocfs2_extent_list *el = NULL;
1770         struct ocfs2_extent_rec *rec = NULL;
1771         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1772         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1773
1774         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1775         ocfs2_init_dealloc_ctxt(&dealloc);
1776
1777         trace_ocfs2_remove_inode_range(
1778                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1779                         (unsigned long long)byte_start,
1780                         (unsigned long long)byte_len);
1781
1782         if (byte_len == 0)
1783                 return 0;
1784
1785         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1786                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1787                                             byte_start + byte_len, 0);
1788                 if (ret) {
1789                         mlog_errno(ret);
1790                         goto out;
1791                 }
1792                 /*
1793                  * There's no need to get fancy with the page cache
1794                  * truncate of an inline-data inode. We're talking
1795                  * about less than a page here, which will be cached
1796                  * in the dinode buffer anyway.
1797                  */
1798                 unmap_mapping_range(mapping, 0, 0, 0);
1799                 truncate_inode_pages(mapping, 0);
1800                 goto out;
1801         }
1802
1803         /*
1804          * For reflinks, we may need to CoW 2 clusters which might be
1805          * partially zero'd later, if hole's start and end offset were
1806          * within one cluster(means is not exactly aligned to clustersize).
1807          */
1808
1809         if (ocfs2_is_refcount_inode(inode)) {
1810                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1811                 if (ret) {
1812                         mlog_errno(ret);
1813                         goto out;
1814                 }
1815
1816                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1817                 if (ret) {
1818                         mlog_errno(ret);
1819                         goto out;
1820                 }
1821         }
1822
1823         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1824         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1825         cluster_in_el = trunc_end;
1826
1827         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1828         if (ret) {
1829                 mlog_errno(ret);
1830                 goto out;
1831         }
1832
1833         path = ocfs2_new_path_from_et(&et);
1834         if (!path) {
1835                 ret = -ENOMEM;
1836                 mlog_errno(ret);
1837                 goto out;
1838         }
1839
1840         while (trunc_end > trunc_start) {
1841
1842                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1843                                       cluster_in_el);
1844                 if (ret) {
1845                         mlog_errno(ret);
1846                         goto out;
1847                 }
1848
1849                 el = path_leaf_el(path);
1850
1851                 i = ocfs2_find_rec(el, trunc_end);
1852                 /*
1853                  * Need to go to previous extent block.
1854                  */
1855                 if (i < 0) {
1856                         if (path->p_tree_depth == 0)
1857                                 break;
1858
1859                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1860                                                             path,
1861                                                             &cluster_in_el);
1862                         if (ret) {
1863                                 mlog_errno(ret);
1864                                 goto out;
1865                         }
1866
1867                         /*
1868                          * We've reached the leftmost extent block,
1869                          * it's safe to leave.
1870                          */
1871                         if (cluster_in_el == 0)
1872                                 break;
1873
1874                         /*
1875                          * The 'pos' searched for previous extent block is
1876                          * always one cluster less than actual trunc_end.
1877                          */
1878                         trunc_end = cluster_in_el + 1;
1879
1880                         ocfs2_reinit_path(path, 1);
1881
1882                         continue;
1883
1884                 } else
1885                         rec = &el->l_recs[i];
1886
1887                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1888                                      &trunc_len, &trunc_end, &blkno, &done);
1889                 if (done)
1890                         break;
1891
1892                 flags = rec->e_flags;
1893                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1894
1895                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1896                                                phys_cpos, trunc_len, flags,
1897                                                &dealloc, refcount_loc, false);
1898                 if (ret < 0) {
1899                         mlog_errno(ret);
1900                         goto out;
1901                 }
1902
1903                 cluster_in_el = trunc_end;
1904
1905                 ocfs2_reinit_path(path, 1);
1906         }
1907
1908         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1909
1910 out:
1911         ocfs2_free_path(path);
1912         ocfs2_schedule_truncate_log_flush(osb, 1);
1913         ocfs2_run_deallocs(osb, &dealloc);
1914
1915         return ret;
1916 }
1917
1918 /*
1919  * Parts of this function taken from xfs_change_file_space()
1920  */
1921 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1922                                      loff_t f_pos, unsigned int cmd,
1923                                      struct ocfs2_space_resv *sr,
1924                                      int change_size)
1925 {
1926         int ret;
1927         s64 llen;
1928         loff_t size, orig_isize;
1929         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1930         struct buffer_head *di_bh = NULL;
1931         handle_t *handle;
1932         unsigned long long max_off = inode->i_sb->s_maxbytes;
1933
1934         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1935                 return -EROFS;
1936
1937         inode_lock(inode);
1938
1939         /*
1940          * This prevents concurrent writes on other nodes
1941          */
1942         ret = ocfs2_rw_lock(inode, 1);
1943         if (ret) {
1944                 mlog_errno(ret);
1945                 goto out;
1946         }
1947
1948         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1949         if (ret) {
1950                 mlog_errno(ret);
1951                 goto out_rw_unlock;
1952         }
1953
1954         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1955                 ret = -EPERM;
1956                 goto out_inode_unlock;
1957         }
1958
1959         switch (sr->l_whence) {
1960         case 0: /*SEEK_SET*/
1961                 break;
1962         case 1: /*SEEK_CUR*/
1963                 sr->l_start += f_pos;
1964                 break;
1965         case 2: /*SEEK_END*/
1966                 sr->l_start += i_size_read(inode);
1967                 break;
1968         default:
1969                 ret = -EINVAL;
1970                 goto out_inode_unlock;
1971         }
1972         sr->l_whence = 0;
1973
1974         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1975
1976         if (sr->l_start < 0
1977             || sr->l_start > max_off
1978             || (sr->l_start + llen) < 0
1979             || (sr->l_start + llen) > max_off) {
1980                 ret = -EINVAL;
1981                 goto out_inode_unlock;
1982         }
1983         size = sr->l_start + sr->l_len;
1984
1985         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1986             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1987                 if (sr->l_len <= 0) {
1988                         ret = -EINVAL;
1989                         goto out_inode_unlock;
1990                 }
1991         }
1992
1993         if (file && setattr_should_drop_suidgid(&nop_mnt_idmap, file_inode(file))) {
1994                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1995                 if (ret) {
1996                         mlog_errno(ret);
1997                         goto out_inode_unlock;
1998                 }
1999         }
2000
2001         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2002         switch (cmd) {
2003         case OCFS2_IOC_RESVSP:
2004         case OCFS2_IOC_RESVSP64:
2005                 /*
2006                  * This takes unsigned offsets, but the signed ones we
2007                  * pass have been checked against overflow above.
2008                  */
2009                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2010                                                        sr->l_len);
2011                 break;
2012         case OCFS2_IOC_UNRESVSP:
2013         case OCFS2_IOC_UNRESVSP64:
2014                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2015                                                sr->l_len);
2016                 break;
2017         default:
2018                 ret = -EINVAL;
2019         }
2020
2021         orig_isize = i_size_read(inode);
2022         /* zeroout eof blocks in the cluster. */
2023         if (!ret && change_size && orig_isize < size) {
2024                 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2025                                         size - orig_isize);
2026                 if (!ret)
2027                         i_size_write(inode, size);
2028         }
2029         up_write(&OCFS2_I(inode)->ip_alloc_sem);
2030         if (ret) {
2031                 mlog_errno(ret);
2032                 goto out_inode_unlock;
2033         }
2034
2035         /*
2036          * We update c/mtime for these changes
2037          */
2038         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2039         if (IS_ERR(handle)) {
2040                 ret = PTR_ERR(handle);
2041                 mlog_errno(ret);
2042                 goto out_inode_unlock;
2043         }
2044
2045         inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
2046         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2047         if (ret < 0)
2048                 mlog_errno(ret);
2049
2050         if (file && (file->f_flags & O_SYNC))
2051                 handle->h_sync = 1;
2052
2053         ocfs2_commit_trans(osb, handle);
2054
2055 out_inode_unlock:
2056         brelse(di_bh);
2057         ocfs2_inode_unlock(inode, 1);
2058 out_rw_unlock:
2059         ocfs2_rw_unlock(inode, 1);
2060
2061 out:
2062         inode_unlock(inode);
2063         return ret;
2064 }
2065
2066 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2067                             struct ocfs2_space_resv *sr)
2068 {
2069         struct inode *inode = file_inode(file);
2070         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2071         int ret;
2072
2073         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2074             !ocfs2_writes_unwritten_extents(osb))
2075                 return -ENOTTY;
2076         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2077                  !ocfs2_sparse_alloc(osb))
2078                 return -ENOTTY;
2079
2080         if (!S_ISREG(inode->i_mode))
2081                 return -EINVAL;
2082
2083         if (!(file->f_mode & FMODE_WRITE))
2084                 return -EBADF;
2085
2086         ret = mnt_want_write_file(file);
2087         if (ret)
2088                 return ret;
2089         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2090         mnt_drop_write_file(file);
2091         return ret;
2092 }
2093
2094 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2095                             loff_t len)
2096 {
2097         struct inode *inode = file_inode(file);
2098         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2099         struct ocfs2_space_resv sr;
2100         int change_size = 1;
2101         int cmd = OCFS2_IOC_RESVSP64;
2102         int ret = 0;
2103
2104         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2105                 return -EOPNOTSUPP;
2106         if (!ocfs2_writes_unwritten_extents(osb))
2107                 return -EOPNOTSUPP;
2108
2109         if (mode & FALLOC_FL_KEEP_SIZE) {
2110                 change_size = 0;
2111         } else {
2112                 ret = inode_newsize_ok(inode, offset + len);
2113                 if (ret)
2114                         return ret;
2115         }
2116
2117         if (mode & FALLOC_FL_PUNCH_HOLE)
2118                 cmd = OCFS2_IOC_UNRESVSP64;
2119
2120         sr.l_whence = 0;
2121         sr.l_start = (s64)offset;
2122         sr.l_len = (s64)len;
2123
2124         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2125                                          change_size);
2126 }
2127
2128 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2129                                    size_t count)
2130 {
2131         int ret = 0;
2132         unsigned int extent_flags;
2133         u32 cpos, clusters, extent_len, phys_cpos;
2134         struct super_block *sb = inode->i_sb;
2135
2136         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2137             !ocfs2_is_refcount_inode(inode) ||
2138             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2139                 return 0;
2140
2141         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2142         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2143
2144         while (clusters) {
2145                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2146                                          &extent_flags);
2147                 if (ret < 0) {
2148                         mlog_errno(ret);
2149                         goto out;
2150                 }
2151
2152                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2153                         ret = 1;
2154                         break;
2155                 }
2156
2157                 if (extent_len > clusters)
2158                         extent_len = clusters;
2159
2160                 clusters -= extent_len;
2161                 cpos += extent_len;
2162         }
2163 out:
2164         return ret;
2165 }
2166
2167 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2168 {
2169         int blockmask = inode->i_sb->s_blocksize - 1;
2170         loff_t final_size = pos + count;
2171
2172         if ((pos & blockmask) || (final_size & blockmask))
2173                 return 1;
2174         return 0;
2175 }
2176
2177 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2178                                             struct buffer_head **di_bh,
2179                                             int meta_level,
2180                                             int write_sem,
2181                                             int wait)
2182 {
2183         int ret = 0;
2184
2185         if (wait)
2186                 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2187         else
2188                 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2189         if (ret < 0)
2190                 goto out;
2191
2192         if (wait) {
2193                 if (write_sem)
2194                         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2195                 else
2196                         down_read(&OCFS2_I(inode)->ip_alloc_sem);
2197         } else {
2198                 if (write_sem)
2199                         ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2200                 else
2201                         ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2202
2203                 if (!ret) {
2204                         ret = -EAGAIN;
2205                         goto out_unlock;
2206                 }
2207         }
2208
2209         return ret;
2210
2211 out_unlock:
2212         brelse(*di_bh);
2213         *di_bh = NULL;
2214         ocfs2_inode_unlock(inode, meta_level);
2215 out:
2216         return ret;
2217 }
2218
2219 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2220                                                struct buffer_head **di_bh,
2221                                                int meta_level,
2222                                                int write_sem)
2223 {
2224         if (write_sem)
2225                 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2226         else
2227                 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2228
2229         brelse(*di_bh);
2230         *di_bh = NULL;
2231
2232         if (meta_level >= 0)
2233                 ocfs2_inode_unlock(inode, meta_level);
2234 }
2235
2236 static int ocfs2_prepare_inode_for_write(struct file *file,
2237                                          loff_t pos, size_t count, int wait)
2238 {
2239         int ret = 0, meta_level = 0, overwrite_io = 0;
2240         int write_sem = 0;
2241         struct dentry *dentry = file->f_path.dentry;
2242         struct inode *inode = d_inode(dentry);
2243         struct buffer_head *di_bh = NULL;
2244         u32 cpos;
2245         u32 clusters;
2246
2247         /*
2248          * We start with a read level meta lock and only jump to an ex
2249          * if we need to make modifications here.
2250          */
2251         for(;;) {
2252                 ret = ocfs2_inode_lock_for_extent_tree(inode,
2253                                                        &di_bh,
2254                                                        meta_level,
2255                                                        write_sem,
2256                                                        wait);
2257                 if (ret < 0) {
2258                         if (ret != -EAGAIN)
2259                                 mlog_errno(ret);
2260                         goto out;
2261                 }
2262
2263                 /*
2264                  * Check if IO will overwrite allocated blocks in case
2265                  * IOCB_NOWAIT flag is set.
2266                  */
2267                 if (!wait && !overwrite_io) {
2268                         overwrite_io = 1;
2269
2270                         ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2271                         if (ret < 0) {
2272                                 if (ret != -EAGAIN)
2273                                         mlog_errno(ret);
2274                                 goto out_unlock;
2275                         }
2276                 }
2277
2278                 /* Clear suid / sgid if necessary. We do this here
2279                  * instead of later in the write path because
2280                  * remove_suid() calls ->setattr without any hint that
2281                  * we may have already done our cluster locking. Since
2282                  * ocfs2_setattr() *must* take cluster locks to
2283                  * proceed, this will lead us to recursively lock the
2284                  * inode. There's also the dinode i_size state which
2285                  * can be lost via setattr during extending writes (we
2286                  * set inode->i_size at the end of a write. */
2287                 if (setattr_should_drop_suidgid(&nop_mnt_idmap, inode)) {
2288                         if (meta_level == 0) {
2289                                 ocfs2_inode_unlock_for_extent_tree(inode,
2290                                                                    &di_bh,
2291                                                                    meta_level,
2292                                                                    write_sem);
2293                                 meta_level = 1;
2294                                 continue;
2295                         }
2296
2297                         ret = ocfs2_write_remove_suid(inode);
2298                         if (ret < 0) {
2299                                 mlog_errno(ret);
2300                                 goto out_unlock;
2301                         }
2302                 }
2303
2304                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2305                 if (ret == 1) {
2306                         ocfs2_inode_unlock_for_extent_tree(inode,
2307                                                            &di_bh,
2308                                                            meta_level,
2309                                                            write_sem);
2310                         meta_level = 1;
2311                         write_sem = 1;
2312                         ret = ocfs2_inode_lock_for_extent_tree(inode,
2313                                                                &di_bh,
2314                                                                meta_level,
2315                                                                write_sem,
2316                                                                wait);
2317                         if (ret < 0) {
2318                                 if (ret != -EAGAIN)
2319                                         mlog_errno(ret);
2320                                 goto out;
2321                         }
2322
2323                         cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2324                         clusters =
2325                                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2326                         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2327                 }
2328
2329                 if (ret < 0) {
2330                         if (ret != -EAGAIN)
2331                                 mlog_errno(ret);
2332                         goto out_unlock;
2333                 }
2334
2335                 break;
2336         }
2337
2338 out_unlock:
2339         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2340                                             pos, count, wait);
2341
2342         ocfs2_inode_unlock_for_extent_tree(inode,
2343                                            &di_bh,
2344                                            meta_level,
2345                                            write_sem);
2346
2347 out:
2348         return ret;
2349 }
2350
2351 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2352                                     struct iov_iter *from)
2353 {
2354         int rw_level;
2355         ssize_t written = 0;
2356         ssize_t ret;
2357         size_t count = iov_iter_count(from);
2358         struct file *file = iocb->ki_filp;
2359         struct inode *inode = file_inode(file);
2360         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2361         int full_coherency = !(osb->s_mount_opt &
2362                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2363         void *saved_ki_complete = NULL;
2364         int append_write = ((iocb->ki_pos + count) >=
2365                         i_size_read(inode) ? 1 : 0);
2366         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2367         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2368
2369         trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2370                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2371                 file->f_path.dentry->d_name.len,
2372                 file->f_path.dentry->d_name.name,
2373                 (unsigned int)from->nr_segs);   /* GRRRRR */
2374
2375         if (!direct_io && nowait)
2376                 return -EOPNOTSUPP;
2377
2378         if (count == 0)
2379                 return 0;
2380
2381         if (nowait) {
2382                 if (!inode_trylock(inode))
2383                         return -EAGAIN;
2384         } else
2385                 inode_lock(inode);
2386
2387         /*
2388          * Concurrent O_DIRECT writes are allowed with
2389          * mount_option "coherency=buffered".
2390          * For append write, we must take rw EX.
2391          */
2392         rw_level = (!direct_io || full_coherency || append_write);
2393
2394         if (nowait)
2395                 ret = ocfs2_try_rw_lock(inode, rw_level);
2396         else
2397                 ret = ocfs2_rw_lock(inode, rw_level);
2398         if (ret < 0) {
2399                 if (ret != -EAGAIN)
2400                         mlog_errno(ret);
2401                 goto out_mutex;
2402         }
2403
2404         /*
2405          * O_DIRECT writes with "coherency=full" need to take EX cluster
2406          * inode_lock to guarantee coherency.
2407          */
2408         if (direct_io && full_coherency) {
2409                 /*
2410                  * We need to take and drop the inode lock to force
2411                  * other nodes to drop their caches.  Buffered I/O
2412                  * already does this in write_begin().
2413                  */
2414                 if (nowait)
2415                         ret = ocfs2_try_inode_lock(inode, NULL, 1);
2416                 else
2417                         ret = ocfs2_inode_lock(inode, NULL, 1);
2418                 if (ret < 0) {
2419                         if (ret != -EAGAIN)
2420                                 mlog_errno(ret);
2421                         goto out;
2422                 }
2423
2424                 ocfs2_inode_unlock(inode, 1);
2425         }
2426
2427         ret = generic_write_checks(iocb, from);
2428         if (ret <= 0) {
2429                 if (ret)
2430                         mlog_errno(ret);
2431                 goto out;
2432         }
2433         count = ret;
2434
2435         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2436         if (ret < 0) {
2437                 if (ret != -EAGAIN)
2438                         mlog_errno(ret);
2439                 goto out;
2440         }
2441
2442         if (direct_io && !is_sync_kiocb(iocb) &&
2443             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2444                 /*
2445                  * Make it a sync io if it's an unaligned aio.
2446                  */
2447                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2448         }
2449
2450         /* communicate with ocfs2_dio_end_io */
2451         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2452
2453         written = __generic_file_write_iter(iocb, from);
2454         /* buffered aio wouldn't have proper lock coverage today */
2455         BUG_ON(written == -EIOCBQUEUED && !direct_io);
2456
2457         /*
2458          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2459          * function pointer which is called when o_direct io completes so that
2460          * it can unlock our rw lock.
2461          * Unfortunately there are error cases which call end_io and others
2462          * that don't.  so we don't have to unlock the rw_lock if either an
2463          * async dio is going to do it in the future or an end_io after an
2464          * error has already done it.
2465          */
2466         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2467                 rw_level = -1;
2468         }
2469
2470         if (unlikely(written <= 0))
2471                 goto out;
2472
2473         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2474             IS_SYNC(inode)) {
2475                 ret = filemap_fdatawrite_range(file->f_mapping,
2476                                                iocb->ki_pos - written,
2477                                                iocb->ki_pos - 1);
2478                 if (ret < 0)
2479                         written = ret;
2480
2481                 if (!ret) {
2482                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2483                         if (ret < 0)
2484                                 written = ret;
2485                 }
2486
2487                 if (!ret)
2488                         ret = filemap_fdatawait_range(file->f_mapping,
2489                                                       iocb->ki_pos - written,
2490                                                       iocb->ki_pos - 1);
2491         }
2492
2493 out:
2494         if (saved_ki_complete)
2495                 xchg(&iocb->ki_complete, saved_ki_complete);
2496
2497         if (rw_level != -1)
2498                 ocfs2_rw_unlock(inode, rw_level);
2499
2500 out_mutex:
2501         inode_unlock(inode);
2502
2503         if (written)
2504                 ret = written;
2505         return ret;
2506 }
2507
2508 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2509                                    struct iov_iter *to)
2510 {
2511         int ret = 0, rw_level = -1, lock_level = 0;
2512         struct file *filp = iocb->ki_filp;
2513         struct inode *inode = file_inode(filp);
2514         int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2515         int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2516
2517         trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2518                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2519                         filp->f_path.dentry->d_name.len,
2520                         filp->f_path.dentry->d_name.name,
2521                         to->nr_segs);   /* GRRRRR */
2522
2523
2524         if (!inode) {
2525                 ret = -EINVAL;
2526                 mlog_errno(ret);
2527                 goto bail;
2528         }
2529
2530         if (!direct_io && nowait)
2531                 return -EOPNOTSUPP;
2532
2533         /*
2534          * buffered reads protect themselves in ->read_folio().  O_DIRECT reads
2535          * need locks to protect pending reads from racing with truncate.
2536          */
2537         if (direct_io) {
2538                 if (nowait)
2539                         ret = ocfs2_try_rw_lock(inode, 0);
2540                 else
2541                         ret = ocfs2_rw_lock(inode, 0);
2542
2543                 if (ret < 0) {
2544                         if (ret != -EAGAIN)
2545                                 mlog_errno(ret);
2546                         goto bail;
2547                 }
2548                 rw_level = 0;
2549                 /* communicate with ocfs2_dio_end_io */
2550                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2551         }
2552
2553         /*
2554          * We're fine letting folks race truncates and extending
2555          * writes with read across the cluster, just like they can
2556          * locally. Hence no rw_lock during read.
2557          *
2558          * Take and drop the meta data lock to update inode fields
2559          * like i_size. This allows the checks down below
2560          * copy_splice_read() a chance of actually working.
2561          */
2562         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2563                                      !nowait);
2564         if (ret < 0) {
2565                 if (ret != -EAGAIN)
2566                         mlog_errno(ret);
2567                 goto bail;
2568         }
2569         ocfs2_inode_unlock(inode, lock_level);
2570
2571         ret = generic_file_read_iter(iocb, to);
2572         trace_generic_file_read_iter_ret(ret);
2573
2574         /* buffered aio wouldn't have proper lock coverage today */
2575         BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2576
2577         /* see ocfs2_file_write_iter */
2578         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2579                 rw_level = -1;
2580         }
2581
2582 bail:
2583         if (rw_level != -1)
2584                 ocfs2_rw_unlock(inode, rw_level);
2585
2586         return ret;
2587 }
2588
2589 static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos,
2590                                       struct pipe_inode_info *pipe,
2591                                       size_t len, unsigned int flags)
2592 {
2593         struct inode *inode = file_inode(in);
2594         ssize_t ret = 0;
2595         int lock_level = 0;
2596
2597         trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2598                                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
2599                                      in->f_path.dentry->d_name.len,
2600                                      in->f_path.dentry->d_name.name,
2601                                      flags);
2602
2603         /*
2604          * We're fine letting folks race truncates and extending writes with
2605          * read across the cluster, just like they can locally.  Hence no
2606          * rw_lock during read.
2607          *
2608          * Take and drop the meta data lock to update inode fields like i_size.
2609          * This allows the checks down below filemap_splice_read() a chance of
2610          * actually working.
2611          */
2612         ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, 1);
2613         if (ret < 0) {
2614                 if (ret != -EAGAIN)
2615                         mlog_errno(ret);
2616                 goto bail;
2617         }
2618         ocfs2_inode_unlock(inode, lock_level);
2619
2620         ret = filemap_splice_read(in, ppos, pipe, len, flags);
2621         trace_filemap_splice_read_ret(ret);
2622 bail:
2623         return ret;
2624 }
2625
2626 /* Refer generic_file_llseek_unlocked() */
2627 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2628 {
2629         struct inode *inode = file->f_mapping->host;
2630         int ret = 0;
2631
2632         inode_lock(inode);
2633
2634         switch (whence) {
2635         case SEEK_SET:
2636                 break;
2637         case SEEK_END:
2638                 /* SEEK_END requires the OCFS2 inode lock for the file
2639                  * because it references the file's size.
2640                  */
2641                 ret = ocfs2_inode_lock(inode, NULL, 0);
2642                 if (ret < 0) {
2643                         mlog_errno(ret);
2644                         goto out;
2645                 }
2646                 offset += i_size_read(inode);
2647                 ocfs2_inode_unlock(inode, 0);
2648                 break;
2649         case SEEK_CUR:
2650                 if (offset == 0) {
2651                         offset = file->f_pos;
2652                         goto out;
2653                 }
2654                 offset += file->f_pos;
2655                 break;
2656         case SEEK_DATA:
2657         case SEEK_HOLE:
2658                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2659                 if (ret)
2660                         goto out;
2661                 break;
2662         default:
2663                 ret = -EINVAL;
2664                 goto out;
2665         }
2666
2667         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2668
2669 out:
2670         inode_unlock(inode);
2671         if (ret)
2672                 return ret;
2673         return offset;
2674 }
2675
2676 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2677                                      struct file *file_out, loff_t pos_out,
2678                                      loff_t len, unsigned int remap_flags)
2679 {
2680         struct inode *inode_in = file_inode(file_in);
2681         struct inode *inode_out = file_inode(file_out);
2682         struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2683         struct buffer_head *in_bh = NULL, *out_bh = NULL;
2684         bool same_inode = (inode_in == inode_out);
2685         loff_t remapped = 0;
2686         ssize_t ret;
2687
2688         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2689                 return -EINVAL;
2690         if (!ocfs2_refcount_tree(osb))
2691                 return -EOPNOTSUPP;
2692         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2693                 return -EROFS;
2694
2695         /* Lock both files against IO */
2696         ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2697         if (ret)
2698                 return ret;
2699
2700         /* Check file eligibility and prepare for block sharing. */
2701         ret = -EINVAL;
2702         if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2703             (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2704                 goto out_unlock;
2705
2706         ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2707                         &len, remap_flags);
2708         if (ret < 0 || len == 0)
2709                 goto out_unlock;
2710
2711         /* Lock out changes to the allocation maps and remap. */
2712         down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2713         if (!same_inode)
2714                 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2715                                   SINGLE_DEPTH_NESTING);
2716
2717         /* Zap any page cache for the destination file's range. */
2718         truncate_inode_pages_range(&inode_out->i_data,
2719                                    round_down(pos_out, PAGE_SIZE),
2720                                    round_up(pos_out + len, PAGE_SIZE) - 1);
2721
2722         remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2723                         inode_out, out_bh, pos_out, len);
2724         up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2725         if (!same_inode)
2726                 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2727         if (remapped < 0) {
2728                 ret = remapped;
2729                 mlog_errno(ret);
2730                 goto out_unlock;
2731         }
2732
2733         /*
2734          * Empty the extent map so that we may get the right extent
2735          * record from the disk.
2736          */
2737         ocfs2_extent_map_trunc(inode_in, 0);
2738         ocfs2_extent_map_trunc(inode_out, 0);
2739
2740         ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2741         if (ret) {
2742                 mlog_errno(ret);
2743                 goto out_unlock;
2744         }
2745
2746 out_unlock:
2747         ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2748         return remapped > 0 ? remapped : ret;
2749 }
2750
2751 const struct inode_operations ocfs2_file_iops = {
2752         .setattr        = ocfs2_setattr,
2753         .getattr        = ocfs2_getattr,
2754         .permission     = ocfs2_permission,
2755         .listxattr      = ocfs2_listxattr,
2756         .fiemap         = ocfs2_fiemap,
2757         .get_inode_acl  = ocfs2_iop_get_acl,
2758         .set_acl        = ocfs2_iop_set_acl,
2759         .fileattr_get   = ocfs2_fileattr_get,
2760         .fileattr_set   = ocfs2_fileattr_set,
2761 };
2762
2763 const struct inode_operations ocfs2_special_file_iops = {
2764         .setattr        = ocfs2_setattr,
2765         .getattr        = ocfs2_getattr,
2766         .listxattr      = ocfs2_listxattr,
2767         .permission     = ocfs2_permission,
2768         .get_inode_acl  = ocfs2_iop_get_acl,
2769         .set_acl        = ocfs2_iop_set_acl,
2770 };
2771
2772 /*
2773  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2774  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2775  */
2776 const struct file_operations ocfs2_fops = {
2777         .llseek         = ocfs2_file_llseek,
2778         .mmap           = ocfs2_mmap,
2779         .fsync          = ocfs2_sync_file,
2780         .release        = ocfs2_file_release,
2781         .open           = ocfs2_file_open,
2782         .read_iter      = ocfs2_file_read_iter,
2783         .write_iter     = ocfs2_file_write_iter,
2784         .unlocked_ioctl = ocfs2_ioctl,
2785 #ifdef CONFIG_COMPAT
2786         .compat_ioctl   = ocfs2_compat_ioctl,
2787 #endif
2788         .lock           = ocfs2_lock,
2789         .flock          = ocfs2_flock,
2790         .splice_read    = ocfs2_file_splice_read,
2791         .splice_write   = iter_file_splice_write,
2792         .fallocate      = ocfs2_fallocate,
2793         .remap_file_range = ocfs2_remap_file_range,
2794 };
2795
2796 WRAP_DIR_ITER(ocfs2_readdir) // FIXME!
2797 const struct file_operations ocfs2_dops = {
2798         .llseek         = generic_file_llseek,
2799         .read           = generic_read_dir,
2800         .iterate_shared = shared_ocfs2_readdir,
2801         .fsync          = ocfs2_sync_file,
2802         .release        = ocfs2_dir_release,
2803         .open           = ocfs2_dir_open,
2804         .unlocked_ioctl = ocfs2_ioctl,
2805 #ifdef CONFIG_COMPAT
2806         .compat_ioctl   = ocfs2_compat_ioctl,
2807 #endif
2808         .lock           = ocfs2_lock,
2809         .flock          = ocfs2_flock,
2810 };
2811
2812 /*
2813  * POSIX-lockless variants of our file_operations.
2814  *
2815  * These will be used if the underlying cluster stack does not support
2816  * posix file locking, if the user passes the "localflocks" mount
2817  * option, or if we have a local-only fs.
2818  *
2819  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2820  * so we still want it in the case of no stack support for
2821  * plocks. Internally, it will do the right thing when asked to ignore
2822  * the cluster.
2823  */
2824 const struct file_operations ocfs2_fops_no_plocks = {
2825         .llseek         = ocfs2_file_llseek,
2826         .mmap           = ocfs2_mmap,
2827         .fsync          = ocfs2_sync_file,
2828         .release        = ocfs2_file_release,
2829         .open           = ocfs2_file_open,
2830         .read_iter      = ocfs2_file_read_iter,
2831         .write_iter     = ocfs2_file_write_iter,
2832         .unlocked_ioctl = ocfs2_ioctl,
2833 #ifdef CONFIG_COMPAT
2834         .compat_ioctl   = ocfs2_compat_ioctl,
2835 #endif
2836         .flock          = ocfs2_flock,
2837         .splice_read    = filemap_splice_read,
2838         .splice_write   = iter_file_splice_write,
2839         .fallocate      = ocfs2_fallocate,
2840         .remap_file_range = ocfs2_remap_file_range,
2841 };
2842
2843 const struct file_operations ocfs2_dops_no_plocks = {
2844         .llseek         = generic_file_llseek,
2845         .read           = generic_read_dir,
2846         .iterate_shared = shared_ocfs2_readdir,
2847         .fsync          = ocfs2_sync_file,
2848         .release        = ocfs2_dir_release,
2849         .open           = ocfs2_dir_open,
2850         .unlocked_ioctl = ocfs2_ioctl,
2851 #ifdef CONFIG_COMPAT
2852         .compat_ioctl   = ocfs2_compat_ioctl,
2853 #endif
2854         .flock          = ocfs2_flock,
2855 };