fs/ocfs2/file.c

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