fs/ocfs2/file.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * file.c
   5  *
   6  * File open, close, extend, truncate
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/capability.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/uio.h>
  33 #include <linux/sched.h>
  34 #include <linux/splice.h>
  35 #include <linux/mount.h>
  36 #include <linux/writeback.h>
  37 #include <linux/falloc.h>
  38 #include <linux/quotaops.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/backing-dev.h>
  41
  42 #include <cluster/masklog.h>
  43
  44 #include "ocfs2.h"
  45
  46 #include "alloc.h"
  47 #include "aops.h"
  48 #include "dir.h"
  49 #include "dlmglue.h"
  50 #include "extent_map.h"
  51 #include "file.h"
  52 #include "sysfile.h"
  53 #include "inode.h"
  54 #include "ioctl.h"
  55 #include "journal.h"
  56 #include "locks.h"
  57 #include "mmap.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "xattr.h"
  61 #include "acl.h"
  62 #include "quota.h"
  63 #include "refcounttree.h"
  64 #include "ocfs2_trace.h"
  65
  66 #include "buffer_head_io.h"
  67
  68 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  69 {
  70         struct ocfs2_file_private *fp;
  71
  72         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  73         if (!fp)
  74                 return -ENOMEM;
  75
  76         fp->fp_file = file;
  77         mutex_init(&fp->fp_mutex);
  78         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  79         file->private_data = fp;
  80
  81         return 0;
  82 }
  83
  84 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  85 {
  86         struct ocfs2_file_private *fp = file->private_data;
  87         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  88
  89         if (fp) {
  90                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  91                 ocfs2_lock_res_free(&fp->fp_flock);
  92                 kfree(fp);
  93                 file->private_data = NULL;
  94         }
  95 }
  96
  97 static int ocfs2_file_open(struct inode *inode, struct file *file)
  98 {
  99         int status;
 100         int mode = file->f_flags;
 101         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 102
 103         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
 104                               (unsigned long long)OCFS2_I(inode)->ip_blkno,
 105                               file->f_path.dentry->d_name.len,
 106                               file->f_path.dentry->d_name.name, mode);
 107
 108         if (file->f_mode & FMODE_WRITE) {
 109                 status = dquot_initialize(inode);
 110                 if (status)
 111                         goto leave;
 112         }
 113
 114         spin_lock(&oi->ip_lock);
 115
 116         /* Check that the inode hasn't been wiped from disk by another
 117          * node. If it hasn't then we're safe as long as we hold the
 118          * spin lock until our increment of open count. */
 119         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
 120                 spin_unlock(&oi->ip_lock);
 121
 122                 status = -ENOENT;
 123                 goto leave;
 124         }
 125
 126         if (mode & O_DIRECT)
 127                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 128
 129         oi->ip_open_count++;
 130         spin_unlock(&oi->ip_lock);
 131
 132         status = ocfs2_init_file_private(inode, file);
 133         if (status) {
 134                 /*
 135                  * We want to set open count back if we're failing the
 136                  * open.
 137                  */
 138                 spin_lock(&oi->ip_lock);
 139                 oi->ip_open_count--;
 140                 spin_unlock(&oi->ip_lock);
 141         }
 142
 143 leave:
 144         return status;
 145 }
 146
 147 static int ocfs2_file_release(struct inode *inode, struct file *file)
 148 {
 149         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 150
 151         spin_lock(&oi->ip_lock);
 152         if (!--oi->ip_open_count)
 153                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 154
 155         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 156                                  oi->ip_blkno,
 157                                  file->f_path.dentry->d_name.len,
 158                                  file->f_path.dentry->d_name.name,
 159                                  oi->ip_open_count);
 160         spin_unlock(&oi->ip_lock);
 161
 162         ocfs2_free_file_private(inode, file);
 163
 164         return 0;
 165 }
 166
 167 static int ocfs2_dir_open(struct inode *inode, struct file *file)
 168 {
 169         return ocfs2_init_file_private(inode, file);
 170 }
 171
 172 static int ocfs2_dir_release(struct inode *inode, struct file *file)
 173 {
 174         ocfs2_free_file_private(inode, file);
 175         return 0;
 176 }
 177
 178 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 179                            int datasync)
 180 {
 181         int err = 0;
 182         struct inode *inode = file->f_mapping->host;
 183         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 184         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 185         journal_t *journal = osb->journal->j_journal;
 186         int ret;
 187         tid_t commit_tid;
 188         bool needs_barrier = false;
 189
 190         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 191                               OCFS2_I(inode)->ip_blkno,
 192                               file->f_path.dentry->d_name.len,
 193                               file->f_path.dentry->d_name.name,
 194                               (unsigned long long)datasync);
 195
 196         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 197                 return -EROFS;
 198
 199         err = filemap_write_and_wait_range(inode->i_mapping, start, end);
 200         if (err)
 201                 return err;
 202
 203         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 204         if (journal->j_flags & JBD2_BARRIER &&
 205             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 206                 needs_barrier = true;
 207         err = jbd2_complete_transaction(journal, commit_tid);
 208         if (needs_barrier) {
 209                 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 210                 if (!err)
 211                         err = ret;
 212         }
 213
 214         if (err)
 215                 mlog_errno(err);
 216
 217         return (err < 0) ? -EIO : 0;
 218 }
 219
 220 int ocfs2_should_update_atime(struct inode *inode,
 221                               struct vfsmount *vfsmnt)
 222 {
 223         struct timespec now;
 224         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 225
 226         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 227                 return 0;
 228
 229         if ((inode->i_flags & S_NOATIME) ||
 230             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 231                 return 0;
 232
 233         /*
 234          * We can be called with no vfsmnt structure - NFSD will
 235          * sometimes do this.
 236          *
 237          * Note that our action here is different than touch_atime() -
 238          * if we can't tell whether this is a noatime mount, then we
 239          * don't know whether to trust the value of s_atime_quantum.
 240          */
 241         if (vfsmnt == NULL)
 242                 return 0;
 243
 244         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 245             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 246                 return 0;
 247
 248         if (vfsmnt->mnt_flags & MNT_RELATIME) {
 249                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 250                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 251                         return 1;
 252
 253                 return 0;
 254         }
 255
 256         now = current_time(inode);
 257         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 258                 return 0;
 259         else
 260                 return 1;
 261 }
 262
 263 int ocfs2_update_inode_atime(struct inode *inode,
 264                              struct buffer_head *bh)
 265 {
 266         int ret;
 267         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 268         handle_t *handle;
 269         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 270
 271         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 272         if (IS_ERR(handle)) {
 273                 ret = PTR_ERR(handle);
 274                 mlog_errno(ret);
 275                 goto out;
 276         }
 277
 278         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 279                                       OCFS2_JOURNAL_ACCESS_WRITE);
 280         if (ret) {
 281                 mlog_errno(ret);
 282                 goto out_commit;
 283         }
 284
 285         /*
 286          * Don't use ocfs2_mark_inode_dirty() here as we don't always
 287          * have i_mutex to guard against concurrent changes to other
 288          * inode fields.
 289          */
 290         inode->i_atime = current_time(inode);
 291         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 292         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 293         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 294         ocfs2_journal_dirty(handle, bh);
 295
 296 out_commit:
 297         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 298 out:
 299         return ret;
 300 }
 301
 302 int ocfs2_set_inode_size(handle_t *handle,
 303                                 struct inode *inode,
 304                                 struct buffer_head *fe_bh,
 305                                 u64 new_i_size)
 306 {
 307         int status;
 308
 309         i_size_write(inode, new_i_size);
 310         inode->i_blocks = ocfs2_inode_sector_count(inode);
 311         inode->i_ctime = inode->i_mtime = current_time(inode);
 312
 313         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 314         if (status < 0) {
 315                 mlog_errno(status);
 316                 goto bail;
 317         }
 318
 319 bail:
 320         return status;
 321 }
 322
 323 int ocfs2_simple_size_update(struct inode *inode,
 324                              struct buffer_head *di_bh,
 325                              u64 new_i_size)
 326 {
 327         int ret;
 328         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 329         handle_t *handle = NULL;
 330
 331         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 332         if (IS_ERR(handle)) {
 333                 ret = PTR_ERR(handle);
 334                 mlog_errno(ret);
 335                 goto out;
 336         }
 337
 338         ret = ocfs2_set_inode_size(handle, inode, di_bh,
 339                                    new_i_size);
 340         if (ret < 0)
 341                 mlog_errno(ret);
 342
 343         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 344         ocfs2_commit_trans(osb, handle);
 345 out:
 346         return ret;
 347 }
 348
 349 static int ocfs2_cow_file_pos(struct inode *inode,
 350                               struct buffer_head *fe_bh,
 351                               u64 offset)
 352 {
 353         int status;
 354         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 355         unsigned int num_clusters = 0;
 356         unsigned int ext_flags = 0;
 357
 358         /*
 359          * If the new offset is aligned to the range of the cluster, there is
 360          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 361          * CoW either.
 362          */
 363         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 364                 return 0;
 365
 366         status = ocfs2_get_clusters(inode, cpos, &phys,
 367                                     &num_clusters, &ext_flags);
 368         if (status) {
 369                 mlog_errno(status);
 370                 goto out;
 371         }
 372
 373         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 374                 goto out;
 375
 376         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 377
 378 out:
 379         return status;
 380 }
 381
 382 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 383                                      struct inode *inode,
 384                                      struct buffer_head *fe_bh,
 385                                      u64 new_i_size)
 386 {
 387         int status;
 388         handle_t *handle;
 389         struct ocfs2_dinode *di;
 390         u64 cluster_bytes;
 391
 392         /*
 393          * We need to CoW the cluster contains the offset if it is reflinked
 394          * since we will call ocfs2_zero_range_for_truncate later which will
 395          * write "0" from offset to the end of the cluster.
 396          */
 397         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 398         if (status) {
 399                 mlog_errno(status);
 400                 return status;
 401         }
 402
 403         /* TODO: This needs to actually orphan the inode in this
 404          * transaction. */
 405
 406         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 407         if (IS_ERR(handle)) {
 408                 status = PTR_ERR(handle);
 409                 mlog_errno(status);
 410                 goto out;
 411         }
 412
 413         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 414                                          OCFS2_JOURNAL_ACCESS_WRITE);
 415         if (status < 0) {
 416                 mlog_errno(status);
 417                 goto out_commit;
 418         }
 419
 420         /*
 421          * Do this before setting i_size.
 422          */
 423         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 424         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 425                                                cluster_bytes);
 426         if (status) {
 427                 mlog_errno(status);
 428                 goto out_commit;
 429         }
 430
 431         i_size_write(inode, new_i_size);
 432         inode->i_ctime = inode->i_mtime = current_time(inode);
 433
 434         di = (struct ocfs2_dinode *) fe_bh->b_data;
 435         di->i_size = cpu_to_le64(new_i_size);
 436         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 437         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 438         ocfs2_update_inode_fsync_trans(handle, inode, 0);
 439
 440         ocfs2_journal_dirty(handle, fe_bh);
 441
 442 out_commit:
 443         ocfs2_commit_trans(osb, handle);
 444 out:
 445         return status;
 446 }
 447
 448 int ocfs2_truncate_file(struct inode *inode,
 449                                struct buffer_head *di_bh,
 450                                u64 new_i_size)
 451 {
 452         int status = 0;
 453         struct ocfs2_dinode *fe = NULL;
 454         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 455
 456         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 457          * already validated it */
 458         fe = (struct ocfs2_dinode *) di_bh->b_data;
 459
 460         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 461                                   (unsigned long long)le64_to_cpu(fe->i_size),
 462                                   (unsigned long long)new_i_size);
 463
 464         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 465                         "Inode %llu, inode i_size = %lld != di "
 466                         "i_size = %llu, i_flags = 0x%x\n",
 467                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 468                         i_size_read(inode),
 469                         (unsigned long long)le64_to_cpu(fe->i_size),
 470                         le32_to_cpu(fe->i_flags));
 471
 472         if (new_i_size > le64_to_cpu(fe->i_size)) {
 473                 trace_ocfs2_truncate_file_error(
 474                         (unsigned long long)le64_to_cpu(fe->i_size),
 475                         (unsigned long long)new_i_size);
 476                 status = -EINVAL;
 477                 mlog_errno(status);
 478                 goto bail;
 479         }
 480
 481         down_write(&OCFS2_I(inode)->ip_alloc_sem);
 482
 483         ocfs2_resv_discard(&osb->osb_la_resmap,
 484                            &OCFS2_I(inode)->ip_la_data_resv);
 485
 486         /*
 487          * The inode lock forced other nodes to sync and drop their
 488          * pages, which (correctly) happens even if we have a truncate
 489          * without allocation change - ocfs2 cluster sizes can be much
 490          * greater than page size, so we have to truncate them
 491          * anyway.
 492          */
 493         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 494         truncate_inode_pages(inode->i_mapping, new_i_size);
 495
 496         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 497                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 498                                                i_size_read(inode), 1);
 499                 if (status)
 500                         mlog_errno(status);
 501
 502                 goto bail_unlock_sem;
 503         }
 504
 505         /* alright, we're going to need to do a full blown alloc size
 506          * change. Orphan the inode so that recovery can complete the
 507          * truncate if necessary. This does the task of marking
 508          * i_size. */
 509         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 510         if (status < 0) {
 511                 mlog_errno(status);
 512                 goto bail_unlock_sem;
 513         }
 514
 515         status = ocfs2_commit_truncate(osb, inode, di_bh);
 516         if (status < 0) {
 517                 mlog_errno(status);
 518                 goto bail_unlock_sem;
 519         }
 520
 521         /* TODO: orphan dir cleanup here. */
 522 bail_unlock_sem:
 523         up_write(&OCFS2_I(inode)->ip_alloc_sem);
 524
 525 bail:
 526         if (!status && OCFS2_I(inode)->ip_clusters == 0)
 527                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 528
 529         return status;
 530 }
 531
 532 /*
 533  * extend file allocation only here.
 534  * we'll update all the disk stuff, and oip->alloc_size
 535  *
 536  * expect stuff to be locked, a transaction started and enough data /
 537  * metadata reservations in the contexts.
 538  *
 539  * Will return -EAGAIN, and a reason if a restart is needed.
 540  * If passed in, *reason will always be set, even in error.
 541  */
 542 int ocfs2_add_inode_data(struct ocfs2_super *osb,
 543                          struct inode *inode,
 544                          u32 *logical_offset,
 545                          u32 clusters_to_add,
 546                          int mark_unwritten,
 547                          struct buffer_head *fe_bh,
 548                          handle_t *handle,
 549                          struct ocfs2_alloc_context *data_ac,
 550                          struct ocfs2_alloc_context *meta_ac,
 551                          enum ocfs2_alloc_restarted *reason_ret)
 552 {
 553         int ret;
 554         struct ocfs2_extent_tree et;
 555
 556         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 557         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 558                                           clusters_to_add, mark_unwritten,
 559                                           data_ac, meta_ac, reason_ret);
 560
 561         return ret;
 562 }
 563
 564 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 565                                      u32 clusters_to_add, int mark_unwritten)
 566 {
 567         int status = 0;
 568         int restart_func = 0;
 569         int credits;
 570         u32 prev_clusters;
 571         struct buffer_head *bh = NULL;
 572         struct ocfs2_dinode *fe = NULL;
 573         handle_t *handle = NULL;
 574         struct ocfs2_alloc_context *data_ac = NULL;
 575         struct ocfs2_alloc_context *meta_ac = NULL;
 576         enum ocfs2_alloc_restarted why = RESTART_NONE;
 577         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 578         struct ocfs2_extent_tree et;
 579         int did_quota = 0;
 580
 581         /*
 582          * Unwritten extent only exists for file systems which
 583          * support holes.
 584          */
 585         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 586
 587         status = ocfs2_read_inode_block(inode, &bh);
 588         if (status < 0) {
 589                 mlog_errno(status);
 590                 goto leave;
 591         }
 592         fe = (struct ocfs2_dinode *) bh->b_data;
 593
 594 restart_all:
 595         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 596
 597         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 598         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 599                                        &data_ac, &meta_ac);
 600         if (status) {
 601                 mlog_errno(status);
 602                 goto leave;
 603         }
 604
 605         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 606         handle = ocfs2_start_trans(osb, credits);
 607         if (IS_ERR(handle)) {
 608                 status = PTR_ERR(handle);
 609                 handle = NULL;
 610                 mlog_errno(status);
 611                 goto leave;
 612         }
 613
 614 restarted_transaction:
 615         trace_ocfs2_extend_allocation(
 616                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 617                 (unsigned long long)i_size_read(inode),
 618                 le32_to_cpu(fe->i_clusters), clusters_to_add,
 619                 why, restart_func);
 620
 621         status = dquot_alloc_space_nodirty(inode,
 622                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 623         if (status)
 624                 goto leave;
 625         did_quota = 1;
 626
 627         /* reserve a write to the file entry early on - that we if we
 628          * run out of credits in the allocation path, we can still
 629          * update i_size. */
 630         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 631                                          OCFS2_JOURNAL_ACCESS_WRITE);
 632         if (status < 0) {
 633                 mlog_errno(status);
 634                 goto leave;
 635         }
 636
 637         prev_clusters = OCFS2_I(inode)->ip_clusters;
 638
 639         status = ocfs2_add_inode_data(osb,
 640                                       inode,
 641                                       &logical_start,
 642                                       clusters_to_add,
 643                                       mark_unwritten,
 644                                       bh,
 645                                       handle,
 646                                       data_ac,
 647                                       meta_ac,
 648                                       &why);
 649         if ((status < 0) && (status != -EAGAIN)) {
 650                 if (status != -ENOSPC)
 651                         mlog_errno(status);
 652                 goto leave;
 653         }
 654         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 655         ocfs2_journal_dirty(handle, bh);
 656
 657         spin_lock(&OCFS2_I(inode)->ip_lock);
 658         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 659         spin_unlock(&OCFS2_I(inode)->ip_lock);
 660         /* Release unused quota reservation */
 661         dquot_free_space(inode,
 662                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 663         did_quota = 0;
 664
 665         if (why != RESTART_NONE && clusters_to_add) {
 666                 if (why == RESTART_META) {
 667                         restart_func = 1;
 668                         status = 0;
 669                 } else {
 670                         BUG_ON(why != RESTART_TRANS);
 671
 672                         status = ocfs2_allocate_extend_trans(handle, 1);
 673                         if (status < 0) {
 674                                 /* handle still has to be committed at
 675                                  * this point. */
 676                                 status = -ENOMEM;
 677                                 mlog_errno(status);
 678                                 goto leave;
 679                         }
 680                         goto restarted_transaction;
 681                 }
 682         }
 683
 684         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 685              le32_to_cpu(fe->i_clusters),
 686              (unsigned long long)le64_to_cpu(fe->i_size),
 687              OCFS2_I(inode)->ip_clusters,
 688              (unsigned long long)i_size_read(inode));
 689
 690 leave:
 691         if (status < 0 && did_quota)
 692                 dquot_free_space(inode,
 693                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 694         if (handle) {
 695                 ocfs2_commit_trans(osb, handle);
 696                 handle = NULL;
 697         }
 698         if (data_ac) {
 699                 ocfs2_free_alloc_context(data_ac);
 700                 data_ac = NULL;
 701         }
 702         if (meta_ac) {
 703                 ocfs2_free_alloc_context(meta_ac);
 704                 meta_ac = NULL;
 705         }
 706         if ((!status) && restart_func) {
 707                 restart_func = 0;
 708                 goto restart_all;
 709         }
 710         brelse(bh);
 711         bh = NULL;
 712
 713         return status;
 714 }
 715
 716 int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 717                 u32 clusters_to_add, int mark_unwritten)
 718 {
 719         return __ocfs2_extend_allocation(inode, logical_start,
 720                         clusters_to_add, mark_unwritten);
 721 }
 722
 723 /*
 724  * While a write will already be ordering the data, a truncate will not.
 725  * Thus, we need to explicitly order the zeroed pages.
 726  */
 727 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 728                                                 struct buffer_head *di_bh)
 729 {
 730         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 731         handle_t *handle = NULL;
 732         int ret = 0;
 733
 734         if (!ocfs2_should_order_data(inode))
 735                 goto out;
 736
 737         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 738         if (IS_ERR(handle)) {
 739                 ret = -ENOMEM;
 740                 mlog_errno(ret);
 741                 goto out;
 742         }
 743
 744         ret = ocfs2_jbd2_file_inode(handle, inode);
 745         if (ret < 0) {
 746                 mlog_errno(ret);
 747                 goto out;
 748         }
 749
 750         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 751                                       OCFS2_JOURNAL_ACCESS_WRITE);
 752         if (ret)
 753                 mlog_errno(ret);
 754         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 755
 756 out:
 757         if (ret) {
 758                 if (!IS_ERR(handle))
 759                         ocfs2_commit_trans(osb, handle);
 760                 handle = ERR_PTR(ret);
 761         }
 762         return handle;
 763 }
 764
 765 /* Some parts of this taken from generic_cont_expand, which turned out
 766  * to be too fragile to do exactly what we need without us having to
 767  * worry about recursive locking in ->write_begin() and ->write_end(). */
 768 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 769                                  u64 abs_to, struct buffer_head *di_bh)
 770 {
 771         struct address_space *mapping = inode->i_mapping;
 772         struct page *page;
 773         unsigned long index = abs_from >> PAGE_SHIFT;
 774         handle_t *handle;
 775         int ret = 0;
 776         unsigned zero_from, zero_to, block_start, block_end;
 777         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 778
 779         BUG_ON(abs_from >= abs_to);
 780         BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 781         BUG_ON(abs_from & (inode->i_blkbits - 1));
 782
 783         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
 784         if (IS_ERR(handle)) {
 785                 ret = PTR_ERR(handle);
 786                 goto out;
 787         }
 788
 789         page = find_or_create_page(mapping, index, GFP_NOFS);
 790         if (!page) {
 791                 ret = -ENOMEM;
 792                 mlog_errno(ret);
 793                 goto out_commit_trans;
 794         }
 795
 796         /* Get the offsets within the page that we want to zero */
 797         zero_from = abs_from & (PAGE_SIZE - 1);
 798         zero_to = abs_to & (PAGE_SIZE - 1);
 799         if (!zero_to)
 800                 zero_to = PAGE_SIZE;
 801
 802         trace_ocfs2_write_zero_page(
 803                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 804                         (unsigned long long)abs_from,
 805                         (unsigned long long)abs_to,
 806                         index, zero_from, zero_to);
 807
 808         /* We know that zero_from is block aligned */
 809         for (block_start = zero_from; block_start < zero_to;
 810              block_start = block_end) {
 811                 block_end = block_start + i_blocksize(inode);
 812
 813                 /*
 814                  * block_start is block-aligned.  Bump it by one to force
 815                  * __block_write_begin and block_commit_write to zero the
 816                  * whole block.
 817                  */
 818                 ret = __block_write_begin(page, block_start + 1, 0,
 819                                           ocfs2_get_block);
 820                 if (ret < 0) {
 821                         mlog_errno(ret);
 822                         goto out_unlock;
 823                 }
 824
 825
 826                 /* must not update i_size! */
 827                 ret = block_commit_write(page, block_start + 1,
 828                                          block_start + 1);
 829                 if (ret < 0)
 830                         mlog_errno(ret);
 831                 else
 832                         ret = 0;
 833         }
 834
 835         /*
 836          * fs-writeback will release the dirty pages without page lock
 837          * whose offset are over inode size, the release happens at
 838          * block_write_full_page().
 839          */
 840         i_size_write(inode, abs_to);
 841         inode->i_blocks = ocfs2_inode_sector_count(inode);
 842         di->i_size = cpu_to_le64((u64)i_size_read(inode));
 843         inode->i_mtime = inode->i_ctime = current_time(inode);
 844         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 845         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 846         di->i_mtime_nsec = di->i_ctime_nsec;
 847         if (handle) {
 848                 ocfs2_journal_dirty(handle, di_bh);
 849                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
 850         }
 851
 852 out_unlock:
 853         unlock_page(page);
 854         put_page(page);
 855 out_commit_trans:
 856         if (handle)
 857                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 858 out:
 859         return ret;
 860 }
 861
 862 /*
 863  * Find the next range to zero.  We do this in terms of bytes because
 864  * that's what ocfs2_zero_extend() wants, and it is dealing with the
 865  * pagecache.  We may return multiple extents.
 866  *
 867  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 868  * needs to be zeroed.  range_start and range_end return the next zeroing
 869  * range.  A subsequent call should pass the previous range_end as its
 870  * zero_start.  If range_end is 0, there's nothing to do.
 871  *
 872  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 873  */
 874 static int ocfs2_zero_extend_get_range(struct inode *inode,
 875                                        struct buffer_head *di_bh,
 876                                        u64 zero_start, u64 zero_end,
 877                                        u64 *range_start, u64 *range_end)
 878 {
 879         int rc = 0, needs_cow = 0;
 880         u32 p_cpos, zero_clusters = 0;
 881         u32 zero_cpos =
 882                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 883         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 884         unsigned int num_clusters = 0;
 885         unsigned int ext_flags = 0;
 886
 887         while (zero_cpos < last_cpos) {
 888                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 889                                         &num_clusters, &ext_flags);
 890                 if (rc) {
 891                         mlog_errno(rc);
 892                         goto out;
 893                 }
 894
 895                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 896                         zero_clusters = num_clusters;
 897                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 898                                 needs_cow = 1;
 899                         break;
 900                 }
 901
 902                 zero_cpos += num_clusters;
 903         }
 904         if (!zero_clusters) {
 905                 *range_end = 0;
 906                 goto out;
 907         }
 908
 909         while ((zero_cpos + zero_clusters) < last_cpos) {
 910                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 911                                         &p_cpos, &num_clusters,
 912                                         &ext_flags);
 913                 if (rc) {
 914                         mlog_errno(rc);
 915                         goto out;
 916                 }
 917
 918                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 919                         break;
 920                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
 921                         needs_cow = 1;
 922                 zero_clusters += num_clusters;
 923         }
 924         if ((zero_cpos + zero_clusters) > last_cpos)
 925                 zero_clusters = last_cpos - zero_cpos;
 926
 927         if (needs_cow) {
 928                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 929                                         zero_clusters, UINT_MAX);
 930                 if (rc) {
 931                         mlog_errno(rc);
 932                         goto out;
 933                 }
 934         }
 935
 936         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 937         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 938                                              zero_cpos + zero_clusters);
 939
 940 out:
 941         return rc;
 942 }
 943
 944 /*
 945  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 946  * has made sure that the entire range needs zeroing.
 947  */
 948 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 949                                    u64 range_end, struct buffer_head *di_bh)
 950 {
 951         int rc = 0;
 952         u64 next_pos;
 953         u64 zero_pos = range_start;
 954
 955         trace_ocfs2_zero_extend_range(
 956                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
 957                         (unsigned long long)range_start,
 958                         (unsigned long long)range_end);
 959         BUG_ON(range_start >= range_end);
 960
 961         while (zero_pos < range_end) {
 962                 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 963                 if (next_pos > range_end)
 964                         next_pos = range_end;
 965                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 966                 if (rc < 0) {
 967                         mlog_errno(rc);
 968                         break;
 969                 }
 970                 zero_pos = next_pos;
 971
 972                 /*
 973                  * Very large extends have the potential to lock up
 974                  * the cpu for extended periods of time.
 975                  */
 976                 cond_resched();
 977         }
 978
 979         return rc;
 980 }
 981
 982 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 983                       loff_t zero_to_size)
 984 {
 985         int ret = 0;
 986         u64 zero_start, range_start = 0, range_end = 0;
 987         struct super_block *sb = inode->i_sb;
 988
 989         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 990         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 991                                 (unsigned long long)zero_start,
 992                                 (unsigned long long)i_size_read(inode));
 993         while (zero_start < zero_to_size) {
 994                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 995                                                   zero_to_size,
 996                                                   &range_start,
 997                                                   &range_end);
 998                 if (ret) {
 999                         mlog_errno(ret);
1000                         break;
1001                 }
1002                 if (!range_end)
1003                         break;
1004                 /* Trim the ends */
1005                 if (range_start < zero_start)
1006                         range_start = zero_start;
1007                 if (range_end > zero_to_size)
1008                         range_end = zero_to_size;
1009
1010                 ret = ocfs2_zero_extend_range(inode, range_start,
1011                                               range_end, di_bh);
1012                 if (ret) {
1013                         mlog_errno(ret);
1014                         break;
1015                 }
1016                 zero_start = range_end;
1017         }
1018
1019         return ret;
1020 }
1021
1022 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1023                           u64 new_i_size, u64 zero_to)
1024 {
1025         int ret;
1026         u32 clusters_to_add;
1027         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1028
1029         /*
1030          * Only quota files call this without a bh, and they can't be
1031          * refcounted.
1032          */
1033         BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1034         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1035
1036         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1037         if (clusters_to_add < oi->ip_clusters)
1038                 clusters_to_add = 0;
1039         else
1040                 clusters_to_add -= oi->ip_clusters;
1041
1042         if (clusters_to_add) {
1043                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1044                                                 clusters_to_add, 0);
1045                 if (ret) {
1046                         mlog_errno(ret);
1047                         goto out;
1048                 }
1049         }
1050
1051         /*
1052          * Call this even if we don't add any clusters to the tree. We
1053          * still need to zero the area between the old i_size and the
1054          * new i_size.
1055          */
1056         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1057         if (ret < 0)
1058                 mlog_errno(ret);
1059
1060 out:
1061         return ret;
1062 }
1063
1064 static int ocfs2_extend_file(struct inode *inode,
1065                              struct buffer_head *di_bh,
1066                              u64 new_i_size)
1067 {
1068         int ret = 0;
1069         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1070
1071         BUG_ON(!di_bh);
1072
1073         /* setattr sometimes calls us like this. */
1074         if (new_i_size == 0)
1075                 goto out;
1076
1077         if (i_size_read(inode) == new_i_size)
1078                 goto out;
1079         BUG_ON(new_i_size < i_size_read(inode));
1080
1081         /*
1082          * The alloc sem blocks people in read/write from reading our
1083          * allocation until we're done changing it. We depend on
1084          * i_mutex to block other extend/truncate calls while we're
1085          * here.  We even have to hold it for sparse files because there
1086          * might be some tail zeroing.
1087          */
1088         down_write(&oi->ip_alloc_sem);
1089
1090         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1091                 /*
1092                  * We can optimize small extends by keeping the inodes
1093                  * inline data.
1094                  */
1095                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1096                         up_write(&oi->ip_alloc_sem);
1097                         goto out_update_size;
1098                 }
1099
1100                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1101                 if (ret) {
1102                         up_write(&oi->ip_alloc_sem);
1103                         mlog_errno(ret);
1104                         goto out;
1105                 }
1106         }
1107
1108         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1109                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1110         else
1111                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1112                                             new_i_size);
1113
1114         up_write(&oi->ip_alloc_sem);
1115
1116         if (ret < 0) {
1117                 mlog_errno(ret);
1118                 goto out;
1119         }
1120
1121 out_update_size:
1122         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1123         if (ret < 0)
1124                 mlog_errno(ret);
1125
1126 out:
1127         return ret;
1128 }
1129
1130 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1131 {
1132         int status = 0, size_change;
1133         int inode_locked = 0;
1134         struct inode *inode = d_inode(dentry);
1135         struct super_block *sb = inode->i_sb;
1136         struct ocfs2_super *osb = OCFS2_SB(sb);
1137         struct buffer_head *bh = NULL;
1138         handle_t *handle = NULL;
1139         struct dquot *transfer_to[MAXQUOTAS] = { };
1140         int qtype;
1141
1142         trace_ocfs2_setattr(inode, dentry,
1143                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1144                             dentry->d_name.len, dentry->d_name.name,
1145                             attr->ia_valid, attr->ia_mode,
1146                             from_kuid(&init_user_ns, attr->ia_uid),
1147                             from_kgid(&init_user_ns, attr->ia_gid));
1148
1149         /* ensuring we don't even attempt to truncate a symlink */
1150         if (S_ISLNK(inode->i_mode))
1151                 attr->ia_valid &= ~ATTR_SIZE;
1152
1153 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1154                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1155         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1156                 return 0;
1157
1158         status = setattr_prepare(dentry, attr);
1159         if (status)
1160                 return status;
1161
1162         if (is_quota_modification(inode, attr)) {
1163                 status = dquot_initialize(inode);
1164                 if (status)
1165                         return status;
1166         }
1167         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1168         if (size_change) {
1169                 /*
1170                  * Here we should wait dio to finish before inode lock
1171                  * to avoid a deadlock between ocfs2_setattr() and
1172                  * ocfs2_dio_end_io_write()
1173                  */
1174                 inode_dio_wait(inode);
1175
1176                 status = ocfs2_rw_lock(inode, 1);
1177                 if (status < 0) {
1178                         mlog_errno(status);
1179                         goto bail;
1180                 }
1181         }
1182
1183         status = ocfs2_inode_lock(inode, &bh, 1);
1184         if (status < 0) {
1185                 if (status != -ENOENT)
1186                         mlog_errno(status);
1187                 goto bail_unlock_rw;
1188         }
1189         inode_locked = 1;
1190
1191         if (size_change) {
1192                 status = inode_newsize_ok(inode, attr->ia_size);
1193                 if (status)
1194                         goto bail_unlock;
1195
1196                 if (i_size_read(inode) >= attr->ia_size) {
1197                         if (ocfs2_should_order_data(inode)) {
1198                                 status = ocfs2_begin_ordered_truncate(inode,
1199                                                                       attr->ia_size);
1200                                 if (status)
1201                                         goto bail_unlock;
1202                         }
1203                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1204                 } else
1205                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1206                 if (status < 0) {
1207                         if (status != -ENOSPC)
1208                                 mlog_errno(status);
1209                         status = -ENOSPC;
1210                         goto bail_unlock;
1211                 }
1212         }
1213
1214         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1215             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1216                 /*
1217                  * Gather pointers to quota structures so that allocation /
1218                  * freeing of quota structures happens here and not inside
1219                  * dquot_transfer() where we have problems with lock ordering
1220                  */
1221                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1222                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1223                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1224                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1225                         if (IS_ERR(transfer_to[USRQUOTA])) {
1226                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1227                                 goto bail_unlock;
1228                         }
1229                 }
1230                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1231                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1232                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1233                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1234                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1235                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1236                                 goto bail_unlock;
1237                         }
1238                 }
1239                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1240                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1241                                            2 * ocfs2_quota_trans_credits(sb));
1242                 if (IS_ERR(handle)) {
1243                         status = PTR_ERR(handle);
1244                         mlog_errno(status);
1245                         goto bail_unlock_alloc;
1246                 }
1247                 status = __dquot_transfer(inode, transfer_to);
1248                 if (status < 0)
1249                         goto bail_commit;
1250         } else {
1251                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1252                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1253                 if (IS_ERR(handle)) {
1254                         status = PTR_ERR(handle);
1255                         mlog_errno(status);
1256                         goto bail_unlock_alloc;
1257                 }
1258         }
1259
1260         setattr_copy(inode, attr);
1261         mark_inode_dirty(inode);
1262
1263         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1264         if (status < 0)
1265                 mlog_errno(status);
1266
1267 bail_commit:
1268         ocfs2_commit_trans(osb, handle);
1269 bail_unlock_alloc:
1270         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1271 bail_unlock:
1272         if (status) {
1273                 ocfs2_inode_unlock(inode, 1);
1274                 inode_locked = 0;
1275         }
1276 bail_unlock_rw:
1277         if (size_change)
1278                 ocfs2_rw_unlock(inode, 1);
1279 bail:
1280
1281         /* Release quota pointers in case we acquired them */
1282         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1283                 dqput(transfer_to[qtype]);
1284
1285         if (!status && attr->ia_valid & ATTR_MODE) {
1286                 status = ocfs2_acl_chmod(inode, bh);
1287                 if (status < 0)
1288                         mlog_errno(status);
1289         }
1290         if (inode_locked)
1291                 ocfs2_inode_unlock(inode, 1);
1292
1293         brelse(bh);
1294         return status;
1295 }
1296
1297 int ocfs2_getattr(struct vfsmount *mnt,
1298                   struct dentry *dentry,
1299                   struct kstat *stat)
1300 {
1301         struct inode *inode = d_inode(dentry);
1302         struct super_block *sb = dentry->d_sb;
1303         struct ocfs2_super *osb = sb->s_fs_info;
1304         int err;
1305
1306         err = ocfs2_inode_revalidate(dentry);
1307         if (err) {
1308                 if (err != -ENOENT)
1309                         mlog_errno(err);
1310                 goto bail;
1311         }
1312
1313         generic_fillattr(inode, stat);
1314         /*
1315          * If there is inline data in the inode, the inode will normally not
1316          * have data blocks allocated (it may have an external xattr block).
1317          * Report at least one sector for such files, so tools like tar, rsync,
1318          * others don't incorrectly think the file is completely sparse.
1319          */
1320         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1321                 stat->blocks += (stat->size + 511)>>9;
1322
1323         /* We set the blksize from the cluster size for performance */
1324         stat->blksize = osb->s_clustersize;
1325
1326 bail:
1327         return err;
1328 }
1329
1330 int ocfs2_permission(struct inode *inode, int mask)
1331 {
1332         int ret;
1333
1334         if (mask & MAY_NOT_BLOCK)
1335                 return -ECHILD;
1336
1337         ret = ocfs2_inode_lock(inode, NULL, 0);
1338         if (ret) {
1339                 if (ret != -ENOENT)
1340                         mlog_errno(ret);
1341                 goto out;
1342         }
1343
1344         ret = generic_permission(inode, mask);
1345
1346         ocfs2_inode_unlock(inode, 0);
1347 out:
1348         return ret;
1349 }
1350
1351 static int __ocfs2_write_remove_suid(struct inode *inode,
1352                                      struct buffer_head *bh)
1353 {
1354         int ret;
1355         handle_t *handle;
1356         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1357         struct ocfs2_dinode *di;
1358
1359         trace_ocfs2_write_remove_suid(
1360                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1361                         inode->i_mode);
1362
1363         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1364         if (IS_ERR(handle)) {
1365                 ret = PTR_ERR(handle);
1366                 mlog_errno(ret);
1367                 goto out;
1368         }
1369
1370         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1371                                       OCFS2_JOURNAL_ACCESS_WRITE);
1372         if (ret < 0) {
1373                 mlog_errno(ret);
1374                 goto out_trans;
1375         }
1376
1377         inode->i_mode &= ~S_ISUID;
1378         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1379                 inode->i_mode &= ~S_ISGID;
1380
1381         di = (struct ocfs2_dinode *) bh->b_data;
1382         di->i_mode = cpu_to_le16(inode->i_mode);
1383         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1384
1385         ocfs2_journal_dirty(handle, bh);
1386
1387 out_trans:
1388         ocfs2_commit_trans(osb, handle);
1389 out:
1390         return ret;
1391 }
1392
1393 static int ocfs2_write_remove_suid(struct inode *inode)
1394 {
1395         int ret;
1396         struct buffer_head *bh = NULL;
1397
1398         ret = ocfs2_read_inode_block(inode, &bh);
1399         if (ret < 0) {
1400                 mlog_errno(ret);
1401                 goto out;
1402         }
1403
1404         ret =  __ocfs2_write_remove_suid(inode, bh);
1405 out:
1406         brelse(bh);
1407         return ret;
1408 }
1409
1410 /*
1411  * Allocate enough extents to cover the region starting at byte offset
1412  * start for len bytes. Existing extents are skipped, any extents
1413  * added are marked as "unwritten".
1414  */
1415 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1416                                             u64 start, u64 len)
1417 {
1418         int ret;
1419         u32 cpos, phys_cpos, clusters, alloc_size;
1420         u64 end = start + len;
1421         struct buffer_head *di_bh = NULL;
1422
1423         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1424                 ret = ocfs2_read_inode_block(inode, &di_bh);
1425                 if (ret) {
1426                         mlog_errno(ret);
1427                         goto out;
1428                 }
1429
1430                 /*
1431                  * Nothing to do if the requested reservation range
1432                  * fits within the inode.
1433                  */
1434                 if (ocfs2_size_fits_inline_data(di_bh, end))
1435                         goto out;
1436
1437                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1438                 if (ret) {
1439                         mlog_errno(ret);
1440                         goto out;
1441                 }
1442         }
1443
1444         /*
1445          * We consider both start and len to be inclusive.
1446          */
1447         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1448         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1449         clusters -= cpos;
1450
1451         while (clusters) {
1452                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1453                                          &alloc_size, NULL);
1454                 if (ret) {
1455                         mlog_errno(ret);
1456                         goto out;
1457                 }
1458
1459                 /*
1460                  * Hole or existing extent len can be arbitrary, so
1461                  * cap it to our own allocation request.
1462                  */
1463                 if (alloc_size > clusters)
1464                         alloc_size = clusters;
1465
1466                 if (phys_cpos) {
1467                         /*
1468                          * We already have an allocation at this
1469                          * region so we can safely skip it.
1470                          */
1471                         goto next;
1472                 }
1473
1474                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1475                 if (ret) {
1476                         if (ret != -ENOSPC)
1477                                 mlog_errno(ret);
1478                         goto out;
1479                 }
1480
1481 next:
1482                 cpos += alloc_size;
1483                 clusters -= alloc_size;
1484         }
1485
1486         ret = 0;
1487 out:
1488
1489         brelse(di_bh);
1490         return ret;
1491 }
1492
1493 /*
1494  * Truncate a byte range, avoiding pages within partial clusters. This
1495  * preserves those pages for the zeroing code to write to.
1496  */
1497 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1498                                          u64 byte_len)
1499 {
1500         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1501         loff_t start, end;
1502         struct address_space *mapping = inode->i_mapping;
1503
1504         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1505         end = byte_start + byte_len;
1506         end = end & ~(osb->s_clustersize - 1);
1507
1508         if (start < end) {
1509                 unmap_mapping_range(mapping, start, end - start, 0);
1510                 truncate_inode_pages_range(mapping, start, end - 1);
1511         }
1512 }
1513
1514 /*
1515  * zero out partial blocks of one cluster.
1516  *
1517  * start: file offset where zero starts, will be made upper block aligned.
1518  * len: it will be trimmed to the end of current cluster if "start + len"
1519  *      is bigger than it.
1520  */
1521 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1522                                         u64 start, u64 len)
1523 {
1524         int ret;
1525         u64 start_block, end_block, nr_blocks;
1526         u64 p_block, offset;
1527         u32 cluster, p_cluster, nr_clusters;
1528         struct super_block *sb = inode->i_sb;
1529         u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1530
1531         if (start + len < end)
1532                 end = start + len;
1533
1534         start_block = ocfs2_blocks_for_bytes(sb, start);
1535         end_block = ocfs2_blocks_for_bytes(sb, end);
1536         nr_blocks = end_block - start_block;
1537         if (!nr_blocks)
1538                 return 0;
1539
1540         cluster = ocfs2_bytes_to_clusters(sb, start);
1541         ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1542                                 &nr_clusters, NULL);
1543         if (ret)
1544                 return ret;
1545         if (!p_cluster)
1546                 return 0;
1547
1548         offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1549         p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1550         return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1551 }
1552
1553 static int ocfs2_zero_partial_clusters(struct inode *inode,
1554                                        u64 start, u64 len)
1555 {
1556         int ret = 0;
1557         u64 tmpend = 0;
1558         u64 end = start + len;
1559         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1560         unsigned int csize = osb->s_clustersize;
1561         handle_t *handle;
1562         loff_t isize = i_size_read(inode);
1563
1564         /*
1565          * The "start" and "end" values are NOT necessarily part of
1566          * the range whose allocation is being deleted. Rather, this
1567          * is what the user passed in with the request. We must zero
1568          * partial clusters here. There's no need to worry about
1569          * physical allocation - the zeroing code knows to skip holes.
1570          */
1571         trace_ocfs2_zero_partial_clusters(
1572                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1573                 (unsigned long long)start, (unsigned long long)end);
1574
1575         /*
1576          * If both edges are on a cluster boundary then there's no
1577          * zeroing required as the region is part of the allocation to
1578          * be truncated.
1579          */
1580         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1581                 goto out;
1582
1583         /* No page cache for EOF blocks, issue zero out to disk. */
1584         if (end > isize) {
1585                 /*
1586                  * zeroout eof blocks in last cluster starting from
1587                  * "isize" even "start" > "isize" because it is
1588                  * complicated to zeroout just at "start" as "start"
1589                  * may be not aligned with block size, buffer write
1590                  * would be required to do that, but out of eof buffer
1591                  * write is not supported.
1592                  */
1593                 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1594                                         end - isize);
1595                 if (ret) {
1596                         mlog_errno(ret);
1597                         goto out;
1598                 }
1599                 if (start >= isize)
1600                         goto out;
1601                 end = isize;
1602         }
1603         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1604         if (IS_ERR(handle)) {
1605                 ret = PTR_ERR(handle);
1606                 mlog_errno(ret);
1607                 goto out;
1608         }
1609
1610         /*
1611          * If start is on a cluster boundary and end is somewhere in another
1612          * cluster, we have not COWed the cluster starting at start, unless
1613          * end is also within the same cluster. So, in this case, we skip this
1614          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1615          * to the next one.
1616          */
1617         if ((start & (csize - 1)) != 0) {
1618                 /*
1619                  * We want to get the byte offset of the end of the 1st
1620                  * cluster.
1621                  */
1622                 tmpend = (u64)osb->s_clustersize +
1623                         (start & ~(osb->s_clustersize - 1));
1624                 if (tmpend > end)
1625                         tmpend = end;
1626
1627                 trace_ocfs2_zero_partial_clusters_range1(
1628                         (unsigned long long)start,
1629                         (unsigned long long)tmpend);
1630
1631                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1632                                                     tmpend);
1633                 if (ret)
1634                         mlog_errno(ret);
1635         }
1636
1637         if (tmpend < end) {
1638                 /*
1639                  * This may make start and end equal, but the zeroing
1640                  * code will skip any work in that case so there's no
1641                  * need to catch it up here.
1642                  */
1643                 start = end & ~(osb->s_clustersize - 1);
1644
1645                 trace_ocfs2_zero_partial_clusters_range2(
1646                         (unsigned long long)start, (unsigned long long)end);
1647
1648                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1649                 if (ret)
1650                         mlog_errno(ret);
1651         }
1652         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1653
1654         ocfs2_commit_trans(osb, handle);
1655 out:
1656         return ret;
1657 }
1658
1659 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1660 {
1661         int i;
1662         struct ocfs2_extent_rec *rec = NULL;
1663
1664         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1665
1666                 rec = &el->l_recs[i];
1667
1668                 if (le32_to_cpu(rec->e_cpos) < pos)
1669                         break;
1670         }
1671
1672         return i;
1673 }
1674
1675 /*
1676  * Helper to calculate the punching pos and length in one run, we handle the
1677  * following three cases in order:
1678  *
1679  * - remove the entire record
1680  * - remove a partial record
1681  * - no record needs to be removed (hole-punching completed)
1682 */
1683 static void ocfs2_calc_trunc_pos(struct inode *inode,
1684                                  struct ocfs2_extent_list *el,
1685                                  struct ocfs2_extent_rec *rec,
1686                                  u32 trunc_start, u32 *trunc_cpos,
1687                                  u32 *trunc_len, u32 *trunc_end,
1688                                  u64 *blkno, int *done)
1689 {
1690         int ret = 0;
1691         u32 coff, range;
1692
1693         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1694
1695         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1696                 /*
1697                  * remove an entire extent record.
1698                  */
1699                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1700                 /*
1701                  * Skip holes if any.
1702                  */
1703                 if (range < *trunc_end)
1704                         *trunc_end = range;
1705                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1706                 *blkno = le64_to_cpu(rec->e_blkno);
1707                 *trunc_end = le32_to_cpu(rec->e_cpos);
1708         } else if (range > trunc_start) {
1709                 /*
1710                  * remove a partial extent record, which means we're
1711                  * removing the last extent record.
1712                  */
1713                 *trunc_cpos = trunc_start;
1714                 /*
1715                  * skip hole if any.
1716                  */
1717                 if (range < *trunc_end)
1718                         *trunc_end = range;
1719                 *trunc_len = *trunc_end - trunc_start;
1720                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1721                 *blkno = le64_to_cpu(rec->e_blkno) +
1722                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1723                 *trunc_end = trunc_start;
1724         } else {
1725                 /*
1726                  * It may have two following possibilities:
1727                  *
1728                  * - last record has been removed
1729                  * - trunc_start was within a hole
1730                  *
1731                  * both two cases mean the completion of hole punching.
1732                  */
1733                 ret = 1;
1734         }
1735
1736         *done = ret;
1737 }
1738
1739 static int ocfs2_remove_inode_range(struct inode *inode,
1740                                     struct buffer_head *di_bh, u64 byte_start,
1741                                     u64 byte_len)
1742 {
1743         int ret = 0, flags = 0, done = 0, i;
1744         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1745         u32 cluster_in_el;
1746         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1747         struct ocfs2_cached_dealloc_ctxt dealloc;
1748         struct address_space *mapping = inode->i_mapping;
1749         struct ocfs2_extent_tree et;
1750         struct ocfs2_path *path = NULL;
1751         struct ocfs2_extent_list *el = NULL;
1752         struct ocfs2_extent_rec *rec = NULL;
1753         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1754         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1755
1756         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1757         ocfs2_init_dealloc_ctxt(&dealloc);
1758
1759         trace_ocfs2_remove_inode_range(
1760                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1761                         (unsigned long long)byte_start,
1762                         (unsigned long long)byte_len);
1763
1764         if (byte_len == 0)
1765                 return 0;
1766
1767         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1768                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1769                                             byte_start + byte_len, 0);
1770                 if (ret) {
1771                         mlog_errno(ret);
1772                         goto out;
1773                 }
1774                 /*
1775                  * There's no need to get fancy with the page cache
1776                  * truncate of an inline-data inode. We're talking
1777                  * about less than a page here, which will be cached
1778                  * in the dinode buffer anyway.
1779                  */
1780                 unmap_mapping_range(mapping, 0, 0, 0);
1781                 truncate_inode_pages(mapping, 0);
1782                 goto out;
1783         }
1784
1785         /*
1786          * For reflinks, we may need to CoW 2 clusters which might be
1787          * partially zero'd later, if hole's start and end offset were
1788          * within one cluster(means is not exactly aligned to clustersize).
1789          */
1790
1791         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1792
1793                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1794                 if (ret) {
1795                         mlog_errno(ret);
1796                         goto out;
1797                 }
1798
1799                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1800                 if (ret) {
1801                         mlog_errno(ret);
1802                         goto out;
1803                 }
1804         }
1805
1806         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1807         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1808         cluster_in_el = trunc_end;
1809
1810         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1811         if (ret) {
1812                 mlog_errno(ret);
1813                 goto out;
1814         }
1815
1816         path = ocfs2_new_path_from_et(&et);
1817         if (!path) {
1818                 ret = -ENOMEM;
1819                 mlog_errno(ret);
1820                 goto out;
1821         }
1822
1823         while (trunc_end > trunc_start) {
1824
1825                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1826                                       cluster_in_el);
1827                 if (ret) {
1828                         mlog_errno(ret);
1829                         goto out;
1830                 }
1831
1832                 el = path_leaf_el(path);
1833
1834                 i = ocfs2_find_rec(el, trunc_end);
1835                 /*
1836                  * Need to go to previous extent block.
1837                  */
1838                 if (i < 0) {
1839                         if (path->p_tree_depth == 0)
1840                                 break;
1841
1842                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1843                                                             path,
1844                                                             &cluster_in_el);
1845                         if (ret) {
1846                                 mlog_errno(ret);
1847                                 goto out;
1848                         }
1849
1850                         /*
1851                          * We've reached the leftmost extent block,
1852                          * it's safe to leave.
1853                          */
1854                         if (cluster_in_el == 0)
1855                                 break;
1856
1857                         /*
1858                          * The 'pos' searched for previous extent block is
1859                          * always one cluster less than actual trunc_end.
1860                          */
1861                         trunc_end = cluster_in_el + 1;
1862
1863                         ocfs2_reinit_path(path, 1);
1864
1865                         continue;
1866
1867                 } else
1868                         rec = &el->l_recs[i];
1869
1870                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1871                                      &trunc_len, &trunc_end, &blkno, &done);
1872                 if (done)
1873                         break;
1874
1875                 flags = rec->e_flags;
1876                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1877
1878                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1879                                                phys_cpos, trunc_len, flags,
1880                                                &dealloc, refcount_loc, false);
1881                 if (ret < 0) {
1882                         mlog_errno(ret);
1883                         goto out;
1884                 }
1885
1886                 cluster_in_el = trunc_end;
1887
1888                 ocfs2_reinit_path(path, 1);
1889         }
1890
1891         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1892
1893 out:
1894         ocfs2_free_path(path);
1895         ocfs2_schedule_truncate_log_flush(osb, 1);
1896         ocfs2_run_deallocs(osb, &dealloc);
1897
1898         return ret;
1899 }
1900
1901 /*
1902  * Parts of this function taken from xfs_change_file_space()
1903  */
1904 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1905                                      loff_t f_pos, unsigned int cmd,
1906                                      struct ocfs2_space_resv *sr,
1907                                      int change_size)
1908 {
1909         int ret;
1910         s64 llen;
1911         loff_t size, orig_isize;
1912         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1913         struct buffer_head *di_bh = NULL;
1914         handle_t *handle;
1915         unsigned long long max_off = inode->i_sb->s_maxbytes;
1916
1917         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1918                 return -EROFS;
1919
1920         inode_lock(inode);
1921
1922         /*
1923          * This prevents concurrent writes on other nodes
1924          */
1925         ret = ocfs2_rw_lock(inode, 1);
1926         if (ret) {
1927                 mlog_errno(ret);
1928                 goto out;
1929         }
1930
1931         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1932         if (ret) {
1933                 mlog_errno(ret);
1934                 goto out_rw_unlock;
1935         }
1936
1937         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1938                 ret = -EPERM;
1939                 goto out_inode_unlock;
1940         }
1941
1942         switch (sr->l_whence) {
1943         case 0: /*SEEK_SET*/
1944                 break;
1945         case 1: /*SEEK_CUR*/
1946                 sr->l_start += f_pos;
1947                 break;
1948         case 2: /*SEEK_END*/
1949                 sr->l_start += i_size_read(inode);
1950                 break;
1951         default:
1952                 ret = -EINVAL;
1953                 goto out_inode_unlock;
1954         }
1955         sr->l_whence = 0;
1956
1957         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1958
1959         if (sr->l_start < 0
1960             || sr->l_start > max_off
1961             || (sr->l_start + llen) < 0
1962             || (sr->l_start + llen) > max_off) {
1963                 ret = -EINVAL;
1964                 goto out_inode_unlock;
1965         }
1966         size = sr->l_start + sr->l_len;
1967
1968         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1969             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1970                 if (sr->l_len <= 0) {
1971                         ret = -EINVAL;
1972                         goto out_inode_unlock;
1973                 }
1974         }
1975
1976         if (file && should_remove_suid(file->f_path.dentry)) {
1977                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1978                 if (ret) {
1979                         mlog_errno(ret);
1980                         goto out_inode_unlock;
1981                 }
1982         }
1983
1984         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1985         switch (cmd) {
1986         case OCFS2_IOC_RESVSP:
1987         case OCFS2_IOC_RESVSP64:
1988                 /*
1989                  * This takes unsigned offsets, but the signed ones we
1990                  * pass have been checked against overflow above.
1991                  */
1992                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1993                                                        sr->l_len);
1994                 break;
1995         case OCFS2_IOC_UNRESVSP:
1996         case OCFS2_IOC_UNRESVSP64:
1997                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1998                                                sr->l_len);
1999                 break;
2000         default:
2001                 ret = -EINVAL;
2002         }
2003
2004         orig_isize = i_size_read(inode);
2005         /* zeroout eof blocks in the cluster. */
2006         if (!ret && change_size && orig_isize < size) {
2007                 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2008                                         size - orig_isize);
2009                 if (!ret)
2010                         i_size_write(inode, size);
2011         }
2012         up_write(&OCFS2_I(inode)->ip_alloc_sem);
2013         if (ret) {
2014                 mlog_errno(ret);
2015                 goto out_inode_unlock;
2016         }
2017
2018         /*
2019          * We update c/mtime for these changes
2020          */
2021         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2022         if (IS_ERR(handle)) {
2023                 ret = PTR_ERR(handle);
2024                 mlog_errno(ret);
2025                 goto out_inode_unlock;
2026         }
2027
2028         inode->i_ctime = inode->i_mtime = current_time(inode);
2029         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2030         if (ret < 0)
2031                 mlog_errno(ret);
2032
2033         if (file && (file->f_flags & O_SYNC))
2034                 handle->h_sync = 1;
2035
2036         ocfs2_commit_trans(osb, handle);
2037
2038 out_inode_unlock:
2039         brelse(di_bh);
2040         ocfs2_inode_unlock(inode, 1);
2041 out_rw_unlock:
2042         ocfs2_rw_unlock(inode, 1);
2043
2044 out:
2045         inode_unlock(inode);
2046         return ret;
2047 }
2048
2049 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2050                             struct ocfs2_space_resv *sr)
2051 {
2052         struct inode *inode = file_inode(file);
2053         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2054         int ret;
2055
2056         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2057             !ocfs2_writes_unwritten_extents(osb))
2058                 return -ENOTTY;
2059         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2060                  !ocfs2_sparse_alloc(osb))
2061                 return -ENOTTY;
2062
2063         if (!S_ISREG(inode->i_mode))
2064                 return -EINVAL;
2065
2066         if (!(file->f_mode & FMODE_WRITE))
2067                 return -EBADF;
2068
2069         ret = mnt_want_write_file(file);
2070         if (ret)
2071                 return ret;
2072         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2073         mnt_drop_write_file(file);
2074         return ret;
2075 }
2076
2077 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2078                             loff_t len)
2079 {
2080         struct inode *inode = file_inode(file);
2081         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2082         struct ocfs2_space_resv sr;
2083         int change_size = 1;
2084         int cmd = OCFS2_IOC_RESVSP64;
2085
2086         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2087                 return -EOPNOTSUPP;
2088         if (!ocfs2_writes_unwritten_extents(osb))
2089                 return -EOPNOTSUPP;
2090
2091         if (mode & FALLOC_FL_KEEP_SIZE)
2092                 change_size = 0;
2093
2094         if (mode & FALLOC_FL_PUNCH_HOLE)
2095                 cmd = OCFS2_IOC_UNRESVSP64;
2096
2097         sr.l_whence = 0;
2098         sr.l_start = (s64)offset;
2099         sr.l_len = (s64)len;
2100
2101         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2102                                          change_size);
2103 }
2104
2105 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2106                                    size_t count)
2107 {
2108         int ret = 0;
2109         unsigned int extent_flags;
2110         u32 cpos, clusters, extent_len, phys_cpos;
2111         struct super_block *sb = inode->i_sb;
2112
2113         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2114             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2115             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2116                 return 0;
2117
2118         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2119         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2120
2121         while (clusters) {
2122                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2123                                          &extent_flags);
2124                 if (ret < 0) {
2125                         mlog_errno(ret);
2126                         goto out;
2127                 }
2128
2129                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2130                         ret = 1;
2131                         break;
2132                 }
2133
2134                 if (extent_len > clusters)
2135                         extent_len = clusters;
2136
2137                 clusters -= extent_len;
2138                 cpos += extent_len;
2139         }
2140 out:
2141         return ret;
2142 }
2143
2144 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2145 {
2146         int blockmask = inode->i_sb->s_blocksize - 1;
2147         loff_t final_size = pos + count;
2148
2149         if ((pos & blockmask) || (final_size & blockmask))
2150                 return 1;
2151         return 0;
2152 }
2153
2154 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2155                                             struct file *file,
2156                                             loff_t pos, size_t count,
2157                                             int *meta_level)
2158 {
2159         int ret;
2160         struct buffer_head *di_bh = NULL;
2161         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2162         u32 clusters =
2163                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2164
2165         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2166         if (ret) {
2167                 mlog_errno(ret);
2168                 goto out;
2169         }
2170
2171         *meta_level = 1;
2172
2173         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2174         if (ret)
2175                 mlog_errno(ret);
2176 out:
2177         brelse(di_bh);
2178         return ret;
2179 }
2180
2181 static int ocfs2_prepare_inode_for_write(struct file *file,
2182                                          loff_t pos,
2183                                          size_t count)
2184 {
2185         int ret = 0, meta_level = 0;
2186         struct dentry *dentry = file->f_path.dentry;
2187         struct inode *inode = d_inode(dentry);
2188         loff_t end;
2189
2190         /*
2191          * We start with a read level meta lock and only jump to an ex
2192          * if we need to make modifications here.
2193          */
2194         for(;;) {
2195                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2196                 if (ret < 0) {
2197                         meta_level = -1;
2198                         mlog_errno(ret);
2199                         goto out;
2200                 }
2201
2202                 /* Clear suid / sgid if necessary. We do this here
2203                  * instead of later in the write path because
2204                  * remove_suid() calls ->setattr without any hint that
2205                  * we may have already done our cluster locking. Since
2206                  * ocfs2_setattr() *must* take cluster locks to
2207                  * proceed, this will lead us to recursively lock the
2208                  * inode. There's also the dinode i_size state which
2209                  * can be lost via setattr during extending writes (we
2210                  * set inode->i_size at the end of a write. */
2211                 if (should_remove_suid(dentry)) {
2212                         if (meta_level == 0) {
2213                                 ocfs2_inode_unlock(inode, meta_level);
2214                                 meta_level = 1;
2215                                 continue;
2216                         }
2217
2218                         ret = ocfs2_write_remove_suid(inode);
2219                         if (ret < 0) {
2220                                 mlog_errno(ret);
2221                                 goto out_unlock;
2222                         }
2223                 }
2224
2225                 end = pos + count;
2226
2227                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2228                 if (ret == 1) {
2229                         ocfs2_inode_unlock(inode, meta_level);
2230                         meta_level = -1;
2231
2232                         ret = ocfs2_prepare_inode_for_refcount(inode,
2233                                                                file,
2234                                                                pos,
2235                                                                count,
2236                                                                &meta_level);
2237                 }
2238
2239                 if (ret < 0) {
2240                         mlog_errno(ret);
2241                         goto out_unlock;
2242                 }
2243
2244                 break;
2245         }
2246
2247 out_unlock:
2248         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2249                                             pos, count);
2250
2251         if (meta_level >= 0)
2252                 ocfs2_inode_unlock(inode, meta_level);
2253
2254 out:
2255         return ret;
2256 }
2257
2258 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2259                                     struct iov_iter *from)
2260 {
2261         int direct_io, rw_level;
2262         ssize_t written = 0;
2263         ssize_t ret;
2264         size_t count = iov_iter_count(from);
2265         struct file *file = iocb->ki_filp;
2266         struct inode *inode = file_inode(file);
2267         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2268         int full_coherency = !(osb->s_mount_opt &
2269                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2270         void *saved_ki_complete = NULL;
2271         int append_write = ((iocb->ki_pos + count) >=
2272                         i_size_read(inode) ? 1 : 0);
2273
2274         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2275                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2276                 file->f_path.dentry->d_name.len,
2277                 file->f_path.dentry->d_name.name,
2278                 (unsigned int)from->nr_segs);   /* GRRRRR */
2279
2280         if (count == 0)
2281                 return 0;
2282
2283         direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2284
2285         inode_lock(inode);
2286
2287         /*
2288          * Concurrent O_DIRECT writes are allowed with
2289          * mount_option "coherency=buffered".
2290          * For append write, we must take rw EX.
2291          */
2292         rw_level = (!direct_io || full_coherency || append_write);
2293
2294         ret = ocfs2_rw_lock(inode, rw_level);
2295         if (ret < 0) {
2296                 mlog_errno(ret);
2297                 goto out_mutex;
2298         }
2299
2300         /*
2301          * O_DIRECT writes with "coherency=full" need to take EX cluster
2302          * inode_lock to guarantee coherency.
2303          */
2304         if (direct_io && full_coherency) {
2305                 /*
2306                  * We need to take and drop the inode lock to force
2307                  * other nodes to drop their caches.  Buffered I/O
2308                  * already does this in write_begin().
2309                  */
2310                 ret = ocfs2_inode_lock(inode, NULL, 1);
2311                 if (ret < 0) {
2312                         mlog_errno(ret);
2313                         goto out;
2314                 }
2315
2316                 ocfs2_inode_unlock(inode, 1);
2317         }
2318
2319         ret = generic_write_checks(iocb, from);
2320         if (ret <= 0) {
2321                 if (ret)
2322                         mlog_errno(ret);
2323                 goto out;
2324         }
2325         count = ret;
2326
2327         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count);
2328         if (ret < 0) {
2329                 mlog_errno(ret);
2330                 goto out;
2331         }
2332
2333         if (direct_io && !is_sync_kiocb(iocb) &&
2334             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2335                 /*
2336                  * Make it a sync io if it's an unaligned aio.
2337                  */
2338                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2339         }
2340
2341         /* communicate with ocfs2_dio_end_io */
2342         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2343
2344         written = __generic_file_write_iter(iocb, from);
2345         /* buffered aio wouldn't have proper lock coverage today */
2346         BUG_ON(written == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2347
2348         /*
2349          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2350          * function pointer which is called when o_direct io completes so that
2351          * it can unlock our rw lock.
2352          * Unfortunately there are error cases which call end_io and others
2353          * that don't.  so we don't have to unlock the rw_lock if either an
2354          * async dio is going to do it in the future or an end_io after an
2355          * error has already done it.
2356          */
2357         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2358                 rw_level = -1;
2359         }
2360
2361         if (unlikely(written <= 0))
2362                 goto out;
2363
2364         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2365             IS_SYNC(inode)) {
2366                 ret = filemap_fdatawrite_range(file->f_mapping,
2367                                                iocb->ki_pos - written,
2368                                                iocb->ki_pos - 1);
2369                 if (ret < 0)
2370                         written = ret;
2371
2372                 if (!ret) {
2373                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2374                         if (ret < 0)
2375                                 written = ret;
2376                 }
2377
2378                 if (!ret)
2379                         ret = filemap_fdatawait_range(file->f_mapping,
2380                                                       iocb->ki_pos - written,
2381                                                       iocb->ki_pos - 1);
2382         }
2383
2384 out:
2385         if (saved_ki_complete)
2386                 xchg(&iocb->ki_complete, saved_ki_complete);
2387
2388         if (rw_level != -1)
2389                 ocfs2_rw_unlock(inode, rw_level);
2390
2391 out_mutex:
2392         inode_unlock(inode);
2393
2394         if (written)
2395                 ret = written;
2396         return ret;
2397 }
2398
2399 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2400                                    struct iov_iter *to)
2401 {
2402         int ret = 0, rw_level = -1, lock_level = 0;
2403         struct file *filp = iocb->ki_filp;
2404         struct inode *inode = file_inode(filp);
2405
2406         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2407                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2408                         filp->f_path.dentry->d_name.len,
2409                         filp->f_path.dentry->d_name.name,
2410                         to->nr_segs);   /* GRRRRR */
2411
2412
2413         if (!inode) {
2414                 ret = -EINVAL;
2415                 mlog_errno(ret);
2416                 goto bail;
2417         }
2418
2419         /*
2420          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2421          * need locks to protect pending reads from racing with truncate.
2422          */
2423         if (iocb->ki_flags & IOCB_DIRECT) {
2424                 ret = ocfs2_rw_lock(inode, 0);
2425                 if (ret < 0) {
2426                         mlog_errno(ret);
2427                         goto bail;
2428                 }
2429                 rw_level = 0;
2430                 /* communicate with ocfs2_dio_end_io */
2431                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2432         }
2433
2434         /*
2435          * We're fine letting folks race truncates and extending
2436          * writes with read across the cluster, just like they can
2437          * locally. Hence no rw_lock during read.
2438          *
2439          * Take and drop the meta data lock to update inode fields
2440          * like i_size. This allows the checks down below
2441          * generic_file_aio_read() a chance of actually working.
2442          */
2443         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2444         if (ret < 0) {
2445                 mlog_errno(ret);
2446                 goto bail;
2447         }
2448         ocfs2_inode_unlock(inode, lock_level);
2449
2450         ret = generic_file_read_iter(iocb, to);
2451         trace_generic_file_aio_read_ret(ret);
2452
2453         /* buffered aio wouldn't have proper lock coverage today */
2454         BUG_ON(ret == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2455
2456         /* see ocfs2_file_write_iter */
2457         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2458                 rw_level = -1;
2459         }
2460
2461 bail:
2462         if (rw_level != -1)
2463                 ocfs2_rw_unlock(inode, rw_level);
2464
2465         return ret;
2466 }
2467
2468 /* Refer generic_file_llseek_unlocked() */
2469 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2470 {
2471         struct inode *inode = file->f_mapping->host;
2472         int ret = 0;
2473
2474         inode_lock(inode);
2475
2476         switch (whence) {
2477         case SEEK_SET:
2478                 break;
2479         case SEEK_END:
2480                 /* SEEK_END requires the OCFS2 inode lock for the file
2481                  * because it references the file's size.
2482                  */
2483                 ret = ocfs2_inode_lock(inode, NULL, 0);
2484                 if (ret < 0) {
2485                         mlog_errno(ret);
2486                         goto out;
2487                 }
2488                 offset += i_size_read(inode);
2489                 ocfs2_inode_unlock(inode, 0);
2490                 break;
2491         case SEEK_CUR:
2492                 if (offset == 0) {
2493                         offset = file->f_pos;
2494                         goto out;
2495                 }
2496                 offset += file->f_pos;
2497                 break;
2498         case SEEK_DATA:
2499         case SEEK_HOLE:
2500                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2501                 if (ret)
2502                         goto out;
2503                 break;
2504         default:
2505                 ret = -EINVAL;
2506                 goto out;
2507         }
2508
2509         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2510
2511 out:
2512         inode_unlock(inode);
2513         if (ret)
2514                 return ret;
2515         return offset;
2516 }
2517
2518 const struct inode_operations ocfs2_file_iops = {
2519         .setattr        = ocfs2_setattr,
2520         .getattr        = ocfs2_getattr,
2521         .permission     = ocfs2_permission,
2522         .listxattr      = ocfs2_listxattr,
2523         .fiemap         = ocfs2_fiemap,
2524         .get_acl        = ocfs2_iop_get_acl,
2525         .set_acl        = ocfs2_iop_set_acl,
2526 };
2527
2528 const struct inode_operations ocfs2_special_file_iops = {
2529         .setattr        = ocfs2_setattr,
2530         .getattr        = ocfs2_getattr,
2531         .permission     = ocfs2_permission,
2532         .get_acl        = ocfs2_iop_get_acl,
2533         .set_acl        = ocfs2_iop_set_acl,
2534 };
2535
2536 /*
2537  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2538  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2539  */
2540 const struct file_operations ocfs2_fops = {
2541         .llseek         = ocfs2_file_llseek,
2542         .mmap           = ocfs2_mmap,
2543         .fsync          = ocfs2_sync_file,
2544         .release        = ocfs2_file_release,
2545         .open           = ocfs2_file_open,
2546         .read_iter      = ocfs2_file_read_iter,
2547         .write_iter     = ocfs2_file_write_iter,
2548         .unlocked_ioctl = ocfs2_ioctl,
2549 #ifdef CONFIG_COMPAT
2550         .compat_ioctl   = ocfs2_compat_ioctl,
2551 #endif
2552         .lock           = ocfs2_lock,
2553         .flock          = ocfs2_flock,
2554         .splice_read    = generic_file_splice_read,
2555         .splice_write   = iter_file_splice_write,
2556         .fallocate      = ocfs2_fallocate,
2557 };
2558
2559 const struct file_operations ocfs2_dops = {
2560         .llseek         = generic_file_llseek,
2561         .read           = generic_read_dir,
2562         .iterate        = ocfs2_readdir,
2563         .fsync          = ocfs2_sync_file,
2564         .release        = ocfs2_dir_release,
2565         .open           = ocfs2_dir_open,
2566         .unlocked_ioctl = ocfs2_ioctl,
2567 #ifdef CONFIG_COMPAT
2568         .compat_ioctl   = ocfs2_compat_ioctl,
2569 #endif
2570         .lock           = ocfs2_lock,
2571         .flock          = ocfs2_flock,
2572 };
2573
2574 /*
2575  * POSIX-lockless variants of our file_operations.
2576  *
2577  * These will be used if the underlying cluster stack does not support
2578  * posix file locking, if the user passes the "localflocks" mount
2579  * option, or if we have a local-only fs.
2580  *
2581  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2582  * so we still want it in the case of no stack support for
2583  * plocks. Internally, it will do the right thing when asked to ignore
2584  * the cluster.
2585  */
2586 const struct file_operations ocfs2_fops_no_plocks = {
2587         .llseek         = ocfs2_file_llseek,
2588         .mmap           = ocfs2_mmap,
2589         .fsync          = ocfs2_sync_file,
2590         .release        = ocfs2_file_release,
2591         .open           = ocfs2_file_open,
2592         .read_iter      = ocfs2_file_read_iter,
2593         .write_iter     = ocfs2_file_write_iter,
2594         .unlocked_ioctl = ocfs2_ioctl,
2595 #ifdef CONFIG_COMPAT
2596         .compat_ioctl   = ocfs2_compat_ioctl,
2597 #endif
2598         .flock          = ocfs2_flock,
2599         .splice_read    = generic_file_splice_read,
2600         .splice_write   = iter_file_splice_write,
2601         .fallocate      = ocfs2_fallocate,
2602 };
2603
2604 const struct file_operations ocfs2_dops_no_plocks = {
2605         .llseek         = generic_file_llseek,
2606         .read           = generic_read_dir,
2607         .iterate        = ocfs2_readdir,
2608         .fsync          = ocfs2_sync_file,
2609         .release        = ocfs2_dir_release,
2610         .open           = ocfs2_dir_open,
2611         .unlocked_ioctl = ocfs2_ioctl,
2612 #ifdef CONFIG_COMPAT
2613         .compat_ioctl   = ocfs2_compat_ioctl,
2614 #endif
2615         .flock          = ocfs2_flock,
2616 };