linux/fs/ocfs2/file.c
<<
>>
Prefs
   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
  41#include <cluster/masklog.h>
  42
  43#include "ocfs2.h"
  44
  45#include "alloc.h"
  46#include "aops.h"
  47#include "dir.h"
  48#include "dlmglue.h"
  49#include "extent_map.h"
  50#include "file.h"
  51#include "sysfile.h"
  52#include "inode.h"
  53#include "ioctl.h"
  54#include "journal.h"
  55#include "locks.h"
  56#include "mmap.h"
  57#include "suballoc.h"
  58#include "super.h"
  59#include "xattr.h"
  60#include "acl.h"
  61#include "quota.h"
  62#include "refcounttree.h"
  63#include "ocfs2_trace.h"
  64
  65#include "buffer_head_io.h"
  66
  67static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  68{
  69        struct ocfs2_file_private *fp;
  70
  71        fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  72        if (!fp)
  73                return -ENOMEM;
  74
  75        fp->fp_file = file;
  76        mutex_init(&fp->fp_mutex);
  77        ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  78        file->private_data = fp;
  79
  80        return 0;
  81}
  82
  83static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  84{
  85        struct ocfs2_file_private *fp = file->private_data;
  86        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  87
  88        if (fp) {
  89                ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  90                ocfs2_lock_res_free(&fp->fp_flock);
  91                kfree(fp);
  92                file->private_data = NULL;
  93        }
  94}
  95
  96static int ocfs2_file_open(struct inode *inode, struct file *file)
  97{
  98        int status;
  99        int mode = file->f_flags;
 100        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 101
 102        trace_ocfs2_file_open(inode, file, file->f_path.dentry,
 103                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 104                              file->f_path.dentry->d_name.len,
 105                              file->f_path.dentry->d_name.name, mode);
 106
 107        if (file->f_mode & FMODE_WRITE)
 108                dquot_initialize(inode);
 109
 110        spin_lock(&oi->ip_lock);
 111
 112        /* Check that the inode hasn't been wiped from disk by another
 113         * node. If it hasn't then we're safe as long as we hold the
 114         * spin lock until our increment of open count. */
 115        if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
 116                spin_unlock(&oi->ip_lock);
 117
 118                status = -ENOENT;
 119                goto leave;
 120        }
 121
 122        if (mode & O_DIRECT)
 123                oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 124
 125        oi->ip_open_count++;
 126        spin_unlock(&oi->ip_lock);
 127
 128        status = ocfs2_init_file_private(inode, file);
 129        if (status) {
 130                /*
 131                 * We want to set open count back if we're failing the
 132                 * open.
 133                 */
 134                spin_lock(&oi->ip_lock);
 135                oi->ip_open_count--;
 136                spin_unlock(&oi->ip_lock);
 137        }
 138
 139leave:
 140        return status;
 141}
 142
 143static int ocfs2_file_release(struct inode *inode, struct file *file)
 144{
 145        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 146
 147        spin_lock(&oi->ip_lock);
 148        if (!--oi->ip_open_count)
 149                oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 150
 151        trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 152                                 oi->ip_blkno,
 153                                 file->f_path.dentry->d_name.len,
 154                                 file->f_path.dentry->d_name.name,
 155                                 oi->ip_open_count);
 156        spin_unlock(&oi->ip_lock);
 157
 158        ocfs2_free_file_private(inode, file);
 159
 160        return 0;
 161}
 162
 163static int ocfs2_dir_open(struct inode *inode, struct file *file)
 164{
 165        return ocfs2_init_file_private(inode, file);
 166}
 167
 168static int ocfs2_dir_release(struct inode *inode, struct file *file)
 169{
 170        ocfs2_free_file_private(inode, file);
 171        return 0;
 172}
 173
 174static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 175                           int datasync)
 176{
 177        int err = 0;
 178        journal_t *journal;
 179        struct inode *inode = file->f_mapping->host;
 180        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 181
 182        trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 183                              OCFS2_I(inode)->ip_blkno,
 184                              file->f_path.dentry->d_name.len,
 185                              file->f_path.dentry->d_name.name,
 186                              (unsigned long long)datasync);
 187
 188        err = filemap_write_and_wait_range(inode->i_mapping, start, end);
 189        if (err)
 190                return err;
 191
 192        /*
 193         * Probably don't need the i_mutex at all in here, just putting it here
 194         * to be consistent with how fsync used to be called, someone more
 195         * familiar with the fs could possibly remove it.
 196         */
 197        mutex_lock(&inode->i_mutex);
 198        if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
 199                /*
 200                 * We still have to flush drive's caches to get data to the
 201                 * platter
 202                 */
 203                if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
 204                        blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 205                goto bail;
 206        }
 207
 208        journal = osb->journal->j_journal;
 209        err = jbd2_journal_force_commit(journal);
 210
 211bail:
 212        if (err)
 213                mlog_errno(err);
 214        mutex_unlock(&inode->i_mutex);
 215
 216        return (err < 0) ? -EIO : 0;
 217}
 218
 219int ocfs2_should_update_atime(struct inode *inode,
 220                              struct vfsmount *vfsmnt)
 221{
 222        struct timespec now;
 223        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 224
 225        if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 226                return 0;
 227
 228        if ((inode->i_flags & S_NOATIME) ||
 229            ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 230                return 0;
 231
 232        /*
 233         * We can be called with no vfsmnt structure - NFSD will
 234         * sometimes do this.
 235         *
 236         * Note that our action here is different than touch_atime() -
 237         * if we can't tell whether this is a noatime mount, then we
 238         * don't know whether to trust the value of s_atime_quantum.
 239         */
 240        if (vfsmnt == NULL)
 241                return 0;
 242
 243        if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 244            ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 245                return 0;
 246
 247        if (vfsmnt->mnt_flags & MNT_RELATIME) {
 248                if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 249                    (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 250                        return 1;
 251
 252                return 0;
 253        }
 254
 255        now = CURRENT_TIME;
 256        if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 257                return 0;
 258        else
 259                return 1;
 260}
 261
 262int ocfs2_update_inode_atime(struct inode *inode,
 263                             struct buffer_head *bh)
 264{
 265        int ret;
 266        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 267        handle_t *handle;
 268        struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 269
 270        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 271        if (IS_ERR(handle)) {
 272                ret = PTR_ERR(handle);
 273                mlog_errno(ret);
 274                goto out;
 275        }
 276
 277        ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 278                                      OCFS2_JOURNAL_ACCESS_WRITE);
 279        if (ret) {
 280                mlog_errno(ret);
 281                goto out_commit;
 282        }
 283
 284        /*
 285         * Don't use ocfs2_mark_inode_dirty() here as we don't always
 286         * have i_mutex to guard against concurrent changes to other
 287         * inode fields.
 288         */
 289        inode->i_atime = CURRENT_TIME;
 290        di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 291        di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 292        ocfs2_journal_dirty(handle, bh);
 293
 294out_commit:
 295        ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 296out:
 297        return ret;
 298}
 299
 300static int ocfs2_set_inode_size(handle_t *handle,
 301                                struct inode *inode,
 302                                struct buffer_head *fe_bh,
 303                                u64 new_i_size)
 304{
 305        int status;
 306
 307        i_size_write(inode, new_i_size);
 308        inode->i_blocks = ocfs2_inode_sector_count(inode);
 309        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 310
 311        status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 312        if (status < 0) {
 313                mlog_errno(status);
 314                goto bail;
 315        }
 316
 317bail:
 318        return status;
 319}
 320
 321int ocfs2_simple_size_update(struct inode *inode,
 322                             struct buffer_head *di_bh,
 323                             u64 new_i_size)
 324{
 325        int ret;
 326        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 327        handle_t *handle = NULL;
 328
 329        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 330        if (IS_ERR(handle)) {
 331                ret = PTR_ERR(handle);
 332                mlog_errno(ret);
 333                goto out;
 334        }
 335
 336        ret = ocfs2_set_inode_size(handle, inode, di_bh,
 337                                   new_i_size);
 338        if (ret < 0)
 339                mlog_errno(ret);
 340
 341        ocfs2_commit_trans(osb, handle);
 342out:
 343        return ret;
 344}
 345
 346static int ocfs2_cow_file_pos(struct inode *inode,
 347                              struct buffer_head *fe_bh,
 348                              u64 offset)
 349{
 350        int status;
 351        u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 352        unsigned int num_clusters = 0;
 353        unsigned int ext_flags = 0;
 354
 355        /*
 356         * If the new offset is aligned to the range of the cluster, there is
 357         * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 358         * CoW either.
 359         */
 360        if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 361                return 0;
 362
 363        status = ocfs2_get_clusters(inode, cpos, &phys,
 364                                    &num_clusters, &ext_flags);
 365        if (status) {
 366                mlog_errno(status);
 367                goto out;
 368        }
 369
 370        if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 371                goto out;
 372
 373        return ocfs2_refcount_cow(inode, NULL, fe_bh, cpos, 1, cpos+1);
 374
 375out:
 376        return status;
 377}
 378
 379static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 380                                     struct inode *inode,
 381                                     struct buffer_head *fe_bh,
 382                                     u64 new_i_size)
 383{
 384        int status;
 385        handle_t *handle;
 386        struct ocfs2_dinode *di;
 387        u64 cluster_bytes;
 388
 389        /*
 390         * We need to CoW the cluster contains the offset if it is reflinked
 391         * since we will call ocfs2_zero_range_for_truncate later which will
 392         * write "0" from offset to the end of the cluster.
 393         */
 394        status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 395        if (status) {
 396                mlog_errno(status);
 397                return status;
 398        }
 399
 400        /* TODO: This needs to actually orphan the inode in this
 401         * transaction. */
 402
 403        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 404        if (IS_ERR(handle)) {
 405                status = PTR_ERR(handle);
 406                mlog_errno(status);
 407                goto out;
 408        }
 409
 410        status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 411                                         OCFS2_JOURNAL_ACCESS_WRITE);
 412        if (status < 0) {
 413                mlog_errno(status);
 414                goto out_commit;
 415        }
 416
 417        /*
 418         * Do this before setting i_size.
 419         */
 420        cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 421        status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 422                                               cluster_bytes);
 423        if (status) {
 424                mlog_errno(status);
 425                goto out_commit;
 426        }
 427
 428        i_size_write(inode, new_i_size);
 429        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 430
 431        di = (struct ocfs2_dinode *) fe_bh->b_data;
 432        di->i_size = cpu_to_le64(new_i_size);
 433        di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 434        di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 435
 436        ocfs2_journal_dirty(handle, fe_bh);
 437
 438out_commit:
 439        ocfs2_commit_trans(osb, handle);
 440out:
 441        return status;
 442}
 443
 444static int ocfs2_truncate_file(struct inode *inode,
 445                               struct buffer_head *di_bh,
 446                               u64 new_i_size)
 447{
 448        int status = 0;
 449        struct ocfs2_dinode *fe = NULL;
 450        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 451
 452        /* We trust di_bh because it comes from ocfs2_inode_lock(), which
 453         * already validated it */
 454        fe = (struct ocfs2_dinode *) di_bh->b_data;
 455
 456        trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 457                                  (unsigned long long)le64_to_cpu(fe->i_size),
 458                                  (unsigned long long)new_i_size);
 459
 460        mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 461                        "Inode %llu, inode i_size = %lld != di "
 462                        "i_size = %llu, i_flags = 0x%x\n",
 463                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
 464                        i_size_read(inode),
 465                        (unsigned long long)le64_to_cpu(fe->i_size),
 466                        le32_to_cpu(fe->i_flags));
 467
 468        if (new_i_size > le64_to_cpu(fe->i_size)) {
 469                trace_ocfs2_truncate_file_error(
 470                        (unsigned long long)le64_to_cpu(fe->i_size),
 471                        (unsigned long long)new_i_size);
 472                status = -EINVAL;
 473                mlog_errno(status);
 474                goto bail;
 475        }
 476
 477        /* lets handle the simple truncate cases before doing any more
 478         * cluster locking. */
 479        if (new_i_size == le64_to_cpu(fe->i_size))
 480                goto bail;
 481
 482        down_write(&OCFS2_I(inode)->ip_alloc_sem);
 483
 484        ocfs2_resv_discard(&osb->osb_la_resmap,
 485                           &OCFS2_I(inode)->ip_la_data_resv);
 486
 487        /*
 488         * The inode lock forced other nodes to sync and drop their
 489         * pages, which (correctly) happens even if we have a truncate
 490         * without allocation change - ocfs2 cluster sizes can be much
 491         * greater than page size, so we have to truncate them
 492         * anyway.
 493         */
 494        unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 495        truncate_inode_pages(inode->i_mapping, new_i_size);
 496
 497        if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 498                status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 499                                               i_size_read(inode), 1);
 500                if (status)
 501                        mlog_errno(status);
 502
 503                goto bail_unlock_sem;
 504        }
 505
 506        /* alright, we're going to need to do a full blown alloc size
 507         * change. Orphan the inode so that recovery can complete the
 508         * truncate if necessary. This does the task of marking
 509         * i_size. */
 510        status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 511        if (status < 0) {
 512                mlog_errno(status);
 513                goto bail_unlock_sem;
 514        }
 515
 516        status = ocfs2_commit_truncate(osb, inode, di_bh);
 517        if (status < 0) {
 518                mlog_errno(status);
 519                goto bail_unlock_sem;
 520        }
 521
 522        /* TODO: orphan dir cleanup here. */
 523bail_unlock_sem:
 524        up_write(&OCFS2_I(inode)->ip_alloc_sem);
 525
 526bail:
 527        if (!status && OCFS2_I(inode)->ip_clusters == 0)
 528                status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 529
 530        return status;
 531}
 532
 533/*
 534 * extend file allocation only here.
 535 * we'll update all the disk stuff, and oip->alloc_size
 536 *
 537 * expect stuff to be locked, a transaction started and enough data /
 538 * metadata reservations in the contexts.
 539 *
 540 * Will return -EAGAIN, and a reason if a restart is needed.
 541 * If passed in, *reason will always be set, even in error.
 542 */
 543int ocfs2_add_inode_data(struct ocfs2_super *osb,
 544                         struct inode *inode,
 545                         u32 *logical_offset,
 546                         u32 clusters_to_add,
 547                         int mark_unwritten,
 548                         struct buffer_head *fe_bh,
 549                         handle_t *handle,
 550                         struct ocfs2_alloc_context *data_ac,
 551                         struct ocfs2_alloc_context *meta_ac,
 552                         enum ocfs2_alloc_restarted *reason_ret)
 553{
 554        int ret;
 555        struct ocfs2_extent_tree et;
 556
 557        ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 558        ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 559                                          clusters_to_add, mark_unwritten,
 560                                          data_ac, meta_ac, reason_ret);
 561
 562        return ret;
 563}
 564
 565static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 566                                     u32 clusters_to_add, int mark_unwritten)
 567{
 568        int status = 0;
 569        int restart_func = 0;
 570        int credits;
 571        u32 prev_clusters;
 572        struct buffer_head *bh = NULL;
 573        struct ocfs2_dinode *fe = NULL;
 574        handle_t *handle = NULL;
 575        struct ocfs2_alloc_context *data_ac = NULL;
 576        struct ocfs2_alloc_context *meta_ac = NULL;
 577        enum ocfs2_alloc_restarted why;
 578        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 579        struct ocfs2_extent_tree et;
 580        int did_quota = 0;
 581
 582        /*
 583         * This function only exists for file systems which don't
 584         * support holes.
 585         */
 586        BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 587
 588        status = ocfs2_read_inode_block(inode, &bh);
 589        if (status < 0) {
 590                mlog_errno(status);
 591                goto leave;
 592        }
 593        fe = (struct ocfs2_dinode *) bh->b_data;
 594
 595restart_all:
 596        BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 597
 598        ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 599        status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 600                                       &data_ac, &meta_ac);
 601        if (status) {
 602                mlog_errno(status);
 603                goto leave;
 604        }
 605
 606        credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
 607                                            clusters_to_add);
 608        handle = ocfs2_start_trans(osb, credits);
 609        if (IS_ERR(handle)) {
 610                status = PTR_ERR(handle);
 611                handle = NULL;
 612                mlog_errno(status);
 613                goto leave;
 614        }
 615
 616restarted_transaction:
 617        trace_ocfs2_extend_allocation(
 618                (unsigned long long)OCFS2_I(inode)->ip_blkno,
 619                (unsigned long long)i_size_read(inode),
 620                le32_to_cpu(fe->i_clusters), clusters_to_add,
 621                why, restart_func);
 622
 623        status = dquot_alloc_space_nodirty(inode,
 624                        ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 625        if (status)
 626                goto leave;
 627        did_quota = 1;
 628
 629        /* reserve a write to the file entry early on - that we if we
 630         * run out of credits in the allocation path, we can still
 631         * update i_size. */
 632        status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 633                                         OCFS2_JOURNAL_ACCESS_WRITE);
 634        if (status < 0) {
 635                mlog_errno(status);
 636                goto leave;
 637        }
 638
 639        prev_clusters = OCFS2_I(inode)->ip_clusters;
 640
 641        status = ocfs2_add_inode_data(osb,
 642                                      inode,
 643                                      &logical_start,
 644                                      clusters_to_add,
 645                                      mark_unwritten,
 646                                      bh,
 647                                      handle,
 648                                      data_ac,
 649                                      meta_ac,
 650                                      &why);
 651        if ((status < 0) && (status != -EAGAIN)) {
 652                if (status != -ENOSPC)
 653                        mlog_errno(status);
 654                goto leave;
 655        }
 656
 657        ocfs2_journal_dirty(handle, bh);
 658
 659        spin_lock(&OCFS2_I(inode)->ip_lock);
 660        clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 661        spin_unlock(&OCFS2_I(inode)->ip_lock);
 662        /* Release unused quota reservation */
 663        dquot_free_space(inode,
 664                        ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 665        did_quota = 0;
 666
 667        if (why != RESTART_NONE && clusters_to_add) {
 668                if (why == RESTART_META) {
 669                        restart_func = 1;
 670                        status = 0;
 671                } else {
 672                        BUG_ON(why != RESTART_TRANS);
 673
 674                        /* TODO: This can be more intelligent. */
 675                        credits = ocfs2_calc_extend_credits(osb->sb,
 676                                                            &fe->id2.i_list,
 677                                                            clusters_to_add);
 678                        status = ocfs2_extend_trans(handle, credits);
 679                        if (status < 0) {
 680                                /* handle still has to be committed at
 681                                 * this point. */
 682                                status = -ENOMEM;
 683                                mlog_errno(status);
 684                                goto leave;
 685                        }
 686                        goto restarted_transaction;
 687                }
 688        }
 689
 690        trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 691             le32_to_cpu(fe->i_clusters),
 692             (unsigned long long)le64_to_cpu(fe->i_size),
 693             OCFS2_I(inode)->ip_clusters,
 694             (unsigned long long)i_size_read(inode));
 695
 696leave:
 697        if (status < 0 && did_quota)
 698                dquot_free_space(inode,
 699                        ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 700        if (handle) {
 701                ocfs2_commit_trans(osb, handle);
 702                handle = NULL;
 703        }
 704        if (data_ac) {
 705                ocfs2_free_alloc_context(data_ac);
 706                data_ac = NULL;
 707        }
 708        if (meta_ac) {
 709                ocfs2_free_alloc_context(meta_ac);
 710                meta_ac = NULL;
 711        }
 712        if ((!status) && restart_func) {
 713                restart_func = 0;
 714                goto restart_all;
 715        }
 716        brelse(bh);
 717        bh = NULL;
 718
 719        return status;
 720}
 721
 722/*
 723 * While a write will already be ordering the data, a truncate will not.
 724 * Thus, we need to explicitly order the zeroed pages.
 725 */
 726static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
 727{
 728        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 729        handle_t *handle = NULL;
 730        int ret = 0;
 731
 732        if (!ocfs2_should_order_data(inode))
 733                goto out;
 734
 735        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 736        if (IS_ERR(handle)) {
 737                ret = -ENOMEM;
 738                mlog_errno(ret);
 739                goto out;
 740        }
 741
 742        ret = ocfs2_jbd2_file_inode(handle, inode);
 743        if (ret < 0)
 744                mlog_errno(ret);
 745
 746out:
 747        if (ret) {
 748                if (!IS_ERR(handle))
 749                        ocfs2_commit_trans(osb, handle);
 750                handle = ERR_PTR(ret);
 751        }
 752        return handle;
 753}
 754
 755/* Some parts of this taken from generic_cont_expand, which turned out
 756 * to be too fragile to do exactly what we need without us having to
 757 * worry about recursive locking in ->write_begin() and ->write_end(). */
 758static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 759                                 u64 abs_to)
 760{
 761        struct address_space *mapping = inode->i_mapping;
 762        struct page *page;
 763        unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
 764        handle_t *handle = NULL;
 765        int ret = 0;
 766        unsigned zero_from, zero_to, block_start, block_end;
 767
 768        BUG_ON(abs_from >= abs_to);
 769        BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
 770        BUG_ON(abs_from & (inode->i_blkbits - 1));
 771
 772        page = find_or_create_page(mapping, index, GFP_NOFS);
 773        if (!page) {
 774                ret = -ENOMEM;
 775                mlog_errno(ret);
 776                goto out;
 777        }
 778
 779        /* Get the offsets within the page that we want to zero */
 780        zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
 781        zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
 782        if (!zero_to)
 783                zero_to = PAGE_CACHE_SIZE;
 784
 785        trace_ocfs2_write_zero_page(
 786                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
 787                        (unsigned long long)abs_from,
 788                        (unsigned long long)abs_to,
 789                        index, zero_from, zero_to);
 790
 791        /* We know that zero_from is block aligned */
 792        for (block_start = zero_from; block_start < zero_to;
 793             block_start = block_end) {
 794                block_end = block_start + (1 << inode->i_blkbits);
 795
 796                /*
 797                 * block_start is block-aligned.  Bump it by one to force
 798                 * __block_write_begin and block_commit_write to zero the
 799                 * whole block.
 800                 */
 801                ret = __block_write_begin(page, block_start + 1, 0,
 802                                          ocfs2_get_block);
 803                if (ret < 0) {
 804                        mlog_errno(ret);
 805                        goto out_unlock;
 806                }
 807
 808                if (!handle) {
 809                        handle = ocfs2_zero_start_ordered_transaction(inode);
 810                        if (IS_ERR(handle)) {
 811                                ret = PTR_ERR(handle);
 812                                handle = NULL;
 813                                break;
 814                        }
 815                }
 816
 817                /* must not update i_size! */
 818                ret = block_commit_write(page, block_start + 1,
 819                                         block_start + 1);
 820                if (ret < 0)
 821                        mlog_errno(ret);
 822                else
 823                        ret = 0;
 824        }
 825
 826        if (handle)
 827                ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 828
 829out_unlock:
 830        unlock_page(page);
 831        page_cache_release(page);
 832out:
 833        return ret;
 834}
 835
 836/*
 837 * Find the next range to zero.  We do this in terms of bytes because
 838 * that's what ocfs2_zero_extend() wants, and it is dealing with the
 839 * pagecache.  We may return multiple extents.
 840 *
 841 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 842 * needs to be zeroed.  range_start and range_end return the next zeroing
 843 * range.  A subsequent call should pass the previous range_end as its
 844 * zero_start.  If range_end is 0, there's nothing to do.
 845 *
 846 * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 847 */
 848static int ocfs2_zero_extend_get_range(struct inode *inode,
 849                                       struct buffer_head *di_bh,
 850                                       u64 zero_start, u64 zero_end,
 851                                       u64 *range_start, u64 *range_end)
 852{
 853        int rc = 0, needs_cow = 0;
 854        u32 p_cpos, zero_clusters = 0;
 855        u32 zero_cpos =
 856                zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 857        u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 858        unsigned int num_clusters = 0;
 859        unsigned int ext_flags = 0;
 860
 861        while (zero_cpos < last_cpos) {
 862                rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 863                                        &num_clusters, &ext_flags);
 864                if (rc) {
 865                        mlog_errno(rc);
 866                        goto out;
 867                }
 868
 869                if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 870                        zero_clusters = num_clusters;
 871                        if (ext_flags & OCFS2_EXT_REFCOUNTED)
 872                                needs_cow = 1;
 873                        break;
 874                }
 875
 876                zero_cpos += num_clusters;
 877        }
 878        if (!zero_clusters) {
 879                *range_end = 0;
 880                goto out;
 881        }
 882
 883        while ((zero_cpos + zero_clusters) < last_cpos) {
 884                rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 885                                        &p_cpos, &num_clusters,
 886                                        &ext_flags);
 887                if (rc) {
 888                        mlog_errno(rc);
 889                        goto out;
 890                }
 891
 892                if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 893                        break;
 894                if (ext_flags & OCFS2_EXT_REFCOUNTED)
 895                        needs_cow = 1;
 896                zero_clusters += num_clusters;
 897        }
 898        if ((zero_cpos + zero_clusters) > last_cpos)
 899                zero_clusters = last_cpos - zero_cpos;
 900
 901        if (needs_cow) {
 902                rc = ocfs2_refcount_cow(inode, NULL, di_bh, zero_cpos,
 903                                        zero_clusters, UINT_MAX);
 904                if (rc) {
 905                        mlog_errno(rc);
 906                        goto out;
 907                }
 908        }
 909
 910        *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 911        *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 912                                             zero_cpos + zero_clusters);
 913
 914out:
 915        return rc;
 916}
 917
 918/*
 919 * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 920 * has made sure that the entire range needs zeroing.
 921 */
 922static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 923                                   u64 range_end)
 924{
 925        int rc = 0;
 926        u64 next_pos;
 927        u64 zero_pos = range_start;
 928
 929        trace_ocfs2_zero_extend_range(
 930                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
 931                        (unsigned long long)range_start,
 932                        (unsigned long long)range_end);
 933        BUG_ON(range_start >= range_end);
 934
 935        while (zero_pos < range_end) {
 936                next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
 937                if (next_pos > range_end)
 938                        next_pos = range_end;
 939                rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
 940                if (rc < 0) {
 941                        mlog_errno(rc);
 942                        break;
 943                }
 944                zero_pos = next_pos;
 945
 946                /*
 947                 * Very large extends have the potential to lock up
 948                 * the cpu for extended periods of time.
 949                 */
 950                cond_resched();
 951        }
 952
 953        return rc;
 954}
 955
 956int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 957                      loff_t zero_to_size)
 958{
 959        int ret = 0;
 960        u64 zero_start, range_start = 0, range_end = 0;
 961        struct super_block *sb = inode->i_sb;
 962
 963        zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 964        trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 965                                (unsigned long long)zero_start,
 966                                (unsigned long long)i_size_read(inode));
 967        while (zero_start < zero_to_size) {
 968                ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 969                                                  zero_to_size,
 970                                                  &range_start,
 971                                                  &range_end);
 972                if (ret) {
 973                        mlog_errno(ret);
 974                        break;
 975                }
 976                if (!range_end)
 977                        break;
 978                /* Trim the ends */
 979                if (range_start < zero_start)
 980                        range_start = zero_start;
 981                if (range_end > zero_to_size)
 982                        range_end = zero_to_size;
 983
 984                ret = ocfs2_zero_extend_range(inode, range_start,
 985                                              range_end);
 986                if (ret) {
 987                        mlog_errno(ret);
 988                        break;
 989                }
 990                zero_start = range_end;
 991        }
 992
 993        return ret;
 994}
 995
 996int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
 997                          u64 new_i_size, u64 zero_to)
 998{
 999        int ret;
1000        u32 clusters_to_add;
1001        struct ocfs2_inode_info *oi = OCFS2_I(inode);
1002
1003        /*
1004         * Only quota files call this without a bh, and they can't be
1005         * refcounted.
1006         */
1007        BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1008        BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1009
1010        clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1011        if (clusters_to_add < oi->ip_clusters)
1012                clusters_to_add = 0;
1013        else
1014                clusters_to_add -= oi->ip_clusters;
1015
1016        if (clusters_to_add) {
1017                ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1018                                                clusters_to_add, 0);
1019                if (ret) {
1020                        mlog_errno(ret);
1021                        goto out;
1022                }
1023        }
1024
1025        /*
1026         * Call this even if we don't add any clusters to the tree. We
1027         * still need to zero the area between the old i_size and the
1028         * new i_size.
1029         */
1030        ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1031        if (ret < 0)
1032                mlog_errno(ret);
1033
1034out:
1035        return ret;
1036}
1037
1038static int ocfs2_extend_file(struct inode *inode,
1039                             struct buffer_head *di_bh,
1040                             u64 new_i_size)
1041{
1042        int ret = 0;
1043        struct ocfs2_inode_info *oi = OCFS2_I(inode);
1044
1045        BUG_ON(!di_bh);
1046
1047        /* setattr sometimes calls us like this. */
1048        if (new_i_size == 0)
1049                goto out;
1050
1051        if (i_size_read(inode) == new_i_size)
1052                goto out;
1053        BUG_ON(new_i_size < i_size_read(inode));
1054
1055        /*
1056         * The alloc sem blocks people in read/write from reading our
1057         * allocation until we're done changing it. We depend on
1058         * i_mutex to block other extend/truncate calls while we're
1059         * here.  We even have to hold it for sparse files because there
1060         * might be some tail zeroing.
1061         */
1062        down_write(&oi->ip_alloc_sem);
1063
1064        if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1065                /*
1066                 * We can optimize small extends by keeping the inodes
1067                 * inline data.
1068                 */
1069                if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1070                        up_write(&oi->ip_alloc_sem);
1071                        goto out_update_size;
1072                }
1073
1074                ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1075                if (ret) {
1076                        up_write(&oi->ip_alloc_sem);
1077                        mlog_errno(ret);
1078                        goto out;
1079                }
1080        }
1081
1082        if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1083                ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1084        else
1085                ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1086                                            new_i_size);
1087
1088        up_write(&oi->ip_alloc_sem);
1089
1090        if (ret < 0) {
1091                mlog_errno(ret);
1092                goto out;
1093        }
1094
1095out_update_size:
1096        ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1097        if (ret < 0)
1098                mlog_errno(ret);
1099
1100out:
1101        return ret;
1102}
1103
1104int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1105{
1106        int status = 0, size_change;
1107        struct inode *inode = dentry->d_inode;
1108        struct super_block *sb = inode->i_sb;
1109        struct ocfs2_super *osb = OCFS2_SB(sb);
1110        struct buffer_head *bh = NULL;
1111        handle_t *handle = NULL;
1112        struct dquot *transfer_to[MAXQUOTAS] = { };
1113        int qtype;
1114
1115        trace_ocfs2_setattr(inode, dentry,
1116                            (unsigned long long)OCFS2_I(inode)->ip_blkno,
1117                            dentry->d_name.len, dentry->d_name.name,
1118                            attr->ia_valid, attr->ia_mode,
1119                            attr->ia_uid, attr->ia_gid);
1120
1121        /* ensuring we don't even attempt to truncate a symlink */
1122        if (S_ISLNK(inode->i_mode))
1123                attr->ia_valid &= ~ATTR_SIZE;
1124
1125#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1126                           | ATTR_GID | ATTR_UID | ATTR_MODE)
1127        if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1128                return 0;
1129
1130        status = inode_change_ok(inode, attr);
1131        if (status)
1132                return status;
1133
1134        if (is_quota_modification(inode, attr))
1135                dquot_initialize(inode);
1136        size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1137        if (size_change) {
1138                status = ocfs2_rw_lock(inode, 1);
1139                if (status < 0) {
1140                        mlog_errno(status);
1141                        goto bail;
1142                }
1143        }
1144
1145        status = ocfs2_inode_lock(inode, &bh, 1);
1146        if (status < 0) {
1147                if (status != -ENOENT)
1148                        mlog_errno(status);
1149                goto bail_unlock_rw;
1150        }
1151
1152        if (size_change && attr->ia_size != i_size_read(inode)) {
1153                status = inode_newsize_ok(inode, attr->ia_size);
1154                if (status)
1155                        goto bail_unlock;
1156
1157                inode_dio_wait(inode);
1158
1159                if (i_size_read(inode) > attr->ia_size) {
1160                        if (ocfs2_should_order_data(inode)) {
1161                                status = ocfs2_begin_ordered_truncate(inode,
1162                                                                      attr->ia_size);
1163                                if (status)
1164                                        goto bail_unlock;
1165                        }
1166                        status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1167                } else
1168                        status = ocfs2_extend_file(inode, bh, attr->ia_size);
1169                if (status < 0) {
1170                        if (status != -ENOSPC)
1171                                mlog_errno(status);
1172                        status = -ENOSPC;
1173                        goto bail_unlock;
1174                }
1175        }
1176
1177        if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
1178            (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
1179                /*
1180                 * Gather pointers to quota structures so that allocation /
1181                 * freeing of quota structures happens here and not inside
1182                 * dquot_transfer() where we have problems with lock ordering
1183                 */
1184                if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
1185                    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1186                    OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1187                        transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1188                        if (!transfer_to[USRQUOTA]) {
1189                                status = -ESRCH;
1190                                goto bail_unlock;
1191                        }
1192                }
1193                if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
1194                    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1195                    OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1196                        transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1197                        if (!transfer_to[GRPQUOTA]) {
1198                                status = -ESRCH;
1199                                goto bail_unlock;
1200                        }
1201                }
1202                handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1203                                           2 * ocfs2_quota_trans_credits(sb));
1204                if (IS_ERR(handle)) {
1205                        status = PTR_ERR(handle);
1206                        mlog_errno(status);
1207                        goto bail_unlock;
1208                }
1209                status = __dquot_transfer(inode, transfer_to);
1210                if (status < 0)
1211                        goto bail_commit;
1212        } else {
1213                handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1214                if (IS_ERR(handle)) {
1215                        status = PTR_ERR(handle);
1216                        mlog_errno(status);
1217                        goto bail_unlock;
1218                }
1219        }
1220
1221        /*
1222         * This will intentionally not wind up calling truncate_setsize(),
1223         * since all the work for a size change has been done above.
1224         * Otherwise, we could get into problems with truncate as
1225         * ip_alloc_sem is used there to protect against i_size
1226         * changes.
1227         *
1228         * XXX: this means the conditional below can probably be removed.
1229         */
1230        if ((attr->ia_valid & ATTR_SIZE) &&
1231            attr->ia_size != i_size_read(inode)) {
1232                status = vmtruncate(inode, attr->ia_size);
1233                if (status) {
1234                        mlog_errno(status);
1235                        goto bail_commit;
1236                }
1237        }
1238
1239        setattr_copy(inode, attr);
1240        mark_inode_dirty(inode);
1241
1242        status = ocfs2_mark_inode_dirty(handle, inode, bh);
1243        if (status < 0)
1244                mlog_errno(status);
1245
1246bail_commit:
1247        ocfs2_commit_trans(osb, handle);
1248bail_unlock:
1249        ocfs2_inode_unlock(inode, 1);
1250bail_unlock_rw:
1251        if (size_change)
1252                ocfs2_rw_unlock(inode, 1);
1253bail:
1254        brelse(bh);
1255
1256        /* Release quota pointers in case we acquired them */
1257        for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1258                dqput(transfer_to[qtype]);
1259
1260        if (!status && attr->ia_valid & ATTR_MODE) {
1261                status = ocfs2_acl_chmod(inode);
1262                if (status < 0)
1263                        mlog_errno(status);
1264        }
1265
1266        return status;
1267}
1268
1269int ocfs2_getattr(struct vfsmount *mnt,
1270                  struct dentry *dentry,
1271                  struct kstat *stat)
1272{
1273        struct inode *inode = dentry->d_inode;
1274        struct super_block *sb = dentry->d_inode->i_sb;
1275        struct ocfs2_super *osb = sb->s_fs_info;
1276        int err;
1277
1278        err = ocfs2_inode_revalidate(dentry);
1279        if (err) {
1280                if (err != -ENOENT)
1281                        mlog_errno(err);
1282                goto bail;
1283        }
1284
1285        generic_fillattr(inode, stat);
1286
1287        /* We set the blksize from the cluster size for performance */
1288        stat->blksize = osb->s_clustersize;
1289
1290bail:
1291        return err;
1292}
1293
1294int ocfs2_permission(struct inode *inode, int mask)
1295{
1296        int ret;
1297
1298        if (mask & MAY_NOT_BLOCK)
1299                return -ECHILD;
1300
1301        ret = ocfs2_inode_lock(inode, NULL, 0);
1302        if (ret) {
1303                if (ret != -ENOENT)
1304                        mlog_errno(ret);
1305                goto out;
1306        }
1307
1308        ret = generic_permission(inode, mask);
1309
1310        ocfs2_inode_unlock(inode, 0);
1311out:
1312        return ret;
1313}
1314
1315static int __ocfs2_write_remove_suid(struct inode *inode,
1316                                     struct buffer_head *bh)
1317{
1318        int ret;
1319        handle_t *handle;
1320        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1321        struct ocfs2_dinode *di;
1322
1323        trace_ocfs2_write_remove_suid(
1324                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
1325                        inode->i_mode);
1326
1327        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1328        if (IS_ERR(handle)) {
1329                ret = PTR_ERR(handle);
1330                mlog_errno(ret);
1331                goto out;
1332        }
1333
1334        ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1335                                      OCFS2_JOURNAL_ACCESS_WRITE);
1336        if (ret < 0) {
1337                mlog_errno(ret);
1338                goto out_trans;
1339        }
1340
1341        inode->i_mode &= ~S_ISUID;
1342        if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1343                inode->i_mode &= ~S_ISGID;
1344
1345        di = (struct ocfs2_dinode *) bh->b_data;
1346        di->i_mode = cpu_to_le16(inode->i_mode);
1347
1348        ocfs2_journal_dirty(handle, bh);
1349
1350out_trans:
1351        ocfs2_commit_trans(osb, handle);
1352out:
1353        return ret;
1354}
1355
1356/*
1357 * Will look for holes and unwritten extents in the range starting at
1358 * pos for count bytes (inclusive).
1359 */
1360static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1361                                       size_t count)
1362{
1363        int ret = 0;
1364        unsigned int extent_flags;
1365        u32 cpos, clusters, extent_len, phys_cpos;
1366        struct super_block *sb = inode->i_sb;
1367
1368        cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1369        clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1370
1371        while (clusters) {
1372                ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1373                                         &extent_flags);
1374                if (ret < 0) {
1375                        mlog_errno(ret);
1376                        goto out;
1377                }
1378
1379                if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1380                        ret = 1;
1381                        break;
1382                }
1383
1384                if (extent_len > clusters)
1385                        extent_len = clusters;
1386
1387                clusters -= extent_len;
1388                cpos += extent_len;
1389        }
1390out:
1391        return ret;
1392}
1393
1394static int ocfs2_write_remove_suid(struct inode *inode)
1395{
1396        int ret;
1397        struct buffer_head *bh = NULL;
1398
1399        ret = ocfs2_read_inode_block(inode, &bh);
1400        if (ret < 0) {
1401                mlog_errno(ret);
1402                goto out;
1403        }
1404
1405        ret =  __ocfs2_write_remove_suid(inode, bh);
1406out:
1407        brelse(bh);
1408        return ret;
1409}
1410
1411/*
1412 * Allocate enough extents to cover the region starting at byte offset
1413 * start for len bytes. Existing extents are skipped, any extents
1414 * added are marked as "unwritten".
1415 */
1416static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1417                                            u64 start, u64 len)
1418{
1419        int ret;
1420        u32 cpos, phys_cpos, clusters, alloc_size;
1421        u64 end = start + len;
1422        struct buffer_head *di_bh = NULL;
1423
1424        if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1425                ret = ocfs2_read_inode_block(inode, &di_bh);
1426                if (ret) {
1427                        mlog_errno(ret);
1428                        goto out;
1429                }
1430
1431                /*
1432                 * Nothing to do if the requested reservation range
1433                 * fits within the inode.
1434                 */
1435                if (ocfs2_size_fits_inline_data(di_bh, end))
1436                        goto out;
1437
1438                ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1439                if (ret) {
1440                        mlog_errno(ret);
1441                        goto out;
1442                }
1443        }
1444
1445        /*
1446         * We consider both start and len to be inclusive.
1447         */
1448        cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1449        clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1450        clusters -= cpos;
1451
1452        while (clusters) {
1453                ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1454                                         &alloc_size, NULL);
1455                if (ret) {
1456                        mlog_errno(ret);
1457                        goto out;
1458                }
1459
1460                /*
1461                 * Hole or existing extent len can be arbitrary, so
1462                 * cap it to our own allocation request.
1463                 */
1464                if (alloc_size > clusters)
1465                        alloc_size = clusters;
1466
1467                if (phys_cpos) {
1468                        /*
1469                         * We already have an allocation at this
1470                         * region so we can safely skip it.
1471                         */
1472                        goto next;
1473                }
1474
1475                ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1476                if (ret) {
1477                        if (ret != -ENOSPC)
1478                                mlog_errno(ret);
1479                        goto out;
1480                }
1481
1482next:
1483                cpos += alloc_size;
1484                clusters -= alloc_size;
1485        }
1486
1487        ret = 0;
1488out:
1489
1490        brelse(di_bh);
1491        return ret;
1492}
1493
1494/*
1495 * Truncate a byte range, avoiding pages within partial clusters. This
1496 * preserves those pages for the zeroing code to write to.
1497 */
1498static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1499                                         u64 byte_len)
1500{
1501        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1502        loff_t start, end;
1503        struct address_space *mapping = inode->i_mapping;
1504
1505        start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1506        end = byte_start + byte_len;
1507        end = end & ~(osb->s_clustersize - 1);
1508
1509        if (start < end) {
1510                unmap_mapping_range(mapping, start, end - start, 0);
1511                truncate_inode_pages_range(mapping, start, end - 1);
1512        }
1513}
1514
1515static int ocfs2_zero_partial_clusters(struct inode *inode,
1516                                       u64 start, u64 len)
1517{
1518        int ret = 0;
1519        u64 tmpend, end = start + len;
1520        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1521        unsigned int csize = osb->s_clustersize;
1522        handle_t *handle;
1523
1524        /*
1525         * The "start" and "end" values are NOT necessarily part of
1526         * the range whose allocation is being deleted. Rather, this
1527         * is what the user passed in with the request. We must zero
1528         * partial clusters here. There's no need to worry about
1529         * physical allocation - the zeroing code knows to skip holes.
1530         */
1531        trace_ocfs2_zero_partial_clusters(
1532                (unsigned long long)OCFS2_I(inode)->ip_blkno,
1533                (unsigned long long)start, (unsigned long long)end);
1534
1535        /*
1536         * If both edges are on a cluster boundary then there's no
1537         * zeroing required as the region is part of the allocation to
1538         * be truncated.
1539         */
1540        if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1541                goto out;
1542
1543        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1544        if (IS_ERR(handle)) {
1545                ret = PTR_ERR(handle);
1546                mlog_errno(ret);
1547                goto out;
1548        }
1549
1550        /*
1551         * We want to get the byte offset of the end of the 1st cluster.
1552         */
1553        tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1554        if (tmpend > end)
1555                tmpend = end;
1556
1557        trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1558                                                 (unsigned long long)tmpend);
1559
1560        ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1561        if (ret)
1562                mlog_errno(ret);
1563
1564        if (tmpend < end) {
1565                /*
1566                 * This may make start and end equal, but the zeroing
1567                 * code will skip any work in that case so there's no
1568                 * need to catch it up here.
1569                 */
1570                start = end & ~(osb->s_clustersize - 1);
1571
1572                trace_ocfs2_zero_partial_clusters_range2(
1573                        (unsigned long long)start, (unsigned long long)end);
1574
1575                ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1576                if (ret)
1577                        mlog_errno(ret);
1578        }
1579
1580        ocfs2_commit_trans(osb, handle);
1581out:
1582        return ret;
1583}
1584
1585static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1586{
1587        int i;
1588        struct ocfs2_extent_rec *rec = NULL;
1589
1590        for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1591
1592                rec = &el->l_recs[i];
1593
1594                if (le32_to_cpu(rec->e_cpos) < pos)
1595                        break;
1596        }
1597
1598        return i;
1599}
1600
1601/*
1602 * Helper to calculate the punching pos and length in one run, we handle the
1603 * following three cases in order:
1604 *
1605 * - remove the entire record
1606 * - remove a partial record
1607 * - no record needs to be removed (hole-punching completed)
1608*/
1609static void ocfs2_calc_trunc_pos(struct inode *inode,
1610                                 struct ocfs2_extent_list *el,
1611                                 struct ocfs2_extent_rec *rec,
1612                                 u32 trunc_start, u32 *trunc_cpos,
1613                                 u32 *trunc_len, u32 *trunc_end,
1614                                 u64 *blkno, int *done)
1615{
1616        int ret = 0;
1617        u32 coff, range;
1618
1619        range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1620
1621        if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1622                /*
1623                 * remove an entire extent record.
1624                 */
1625                *trunc_cpos = le32_to_cpu(rec->e_cpos);
1626                /*
1627                 * Skip holes if any.
1628                 */
1629                if (range < *trunc_end)
1630                        *trunc_end = range;
1631                *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1632                *blkno = le64_to_cpu(rec->e_blkno);
1633                *trunc_end = le32_to_cpu(rec->e_cpos);
1634        } else if (range > trunc_start) {
1635                /*
1636                 * remove a partial extent record, which means we're
1637                 * removing the last extent record.
1638                 */
1639                *trunc_cpos = trunc_start;
1640                /*
1641                 * skip hole if any.
1642                 */
1643                if (range < *trunc_end)
1644                        *trunc_end = range;
1645                *trunc_len = *trunc_end - trunc_start;
1646                coff = trunc_start - le32_to_cpu(rec->e_cpos);
1647                *blkno = le64_to_cpu(rec->e_blkno) +
1648                                ocfs2_clusters_to_blocks(inode->i_sb, coff);
1649                *trunc_end = trunc_start;
1650        } else {
1651                /*
1652                 * It may have two following possibilities:
1653                 *
1654                 * - last record has been removed
1655                 * - trunc_start was within a hole
1656                 *
1657                 * both two cases mean the completion of hole punching.
1658                 */
1659                ret = 1;
1660        }
1661
1662        *done = ret;
1663}
1664
1665static int ocfs2_remove_inode_range(struct inode *inode,
1666                                    struct buffer_head *di_bh, u64 byte_start,
1667                                    u64 byte_len)
1668{
1669        int ret = 0, flags = 0, done = 0, i;
1670        u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1671        u32 cluster_in_el;
1672        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1673        struct ocfs2_cached_dealloc_ctxt dealloc;
1674        struct address_space *mapping = inode->i_mapping;
1675        struct ocfs2_extent_tree et;
1676        struct ocfs2_path *path = NULL;
1677        struct ocfs2_extent_list *el = NULL;
1678        struct ocfs2_extent_rec *rec = NULL;
1679        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1680        u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1681
1682        ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1683        ocfs2_init_dealloc_ctxt(&dealloc);
1684
1685        trace_ocfs2_remove_inode_range(
1686                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
1687                        (unsigned long long)byte_start,
1688                        (unsigned long long)byte_len);
1689
1690        if (byte_len == 0)
1691                return 0;
1692
1693        if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1694                ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1695                                            byte_start + byte_len, 0);
1696                if (ret) {
1697                        mlog_errno(ret);
1698                        goto out;
1699                }
1700                /*
1701                 * There's no need to get fancy with the page cache
1702                 * truncate of an inline-data inode. We're talking
1703                 * about less than a page here, which will be cached
1704                 * in the dinode buffer anyway.
1705                 */
1706                unmap_mapping_range(mapping, 0, 0, 0);
1707                truncate_inode_pages(mapping, 0);
1708                goto out;
1709        }
1710
1711        /*
1712         * For reflinks, we may need to CoW 2 clusters which might be
1713         * partially zero'd later, if hole's start and end offset were
1714         * within one cluster(means is not exactly aligned to clustersize).
1715         */
1716
1717        if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1718
1719                ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1720                if (ret) {
1721                        mlog_errno(ret);
1722                        goto out;
1723                }
1724
1725                ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1726                if (ret) {
1727                        mlog_errno(ret);
1728                        goto out;
1729                }
1730        }
1731
1732        trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1733        trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1734        cluster_in_el = trunc_end;
1735
1736        ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1737        if (ret) {
1738                mlog_errno(ret);
1739                goto out;
1740        }
1741
1742        path = ocfs2_new_path_from_et(&et);
1743        if (!path) {
1744                ret = -ENOMEM;
1745                mlog_errno(ret);
1746                goto out;
1747        }
1748
1749        while (trunc_end > trunc_start) {
1750
1751                ret = ocfs2_find_path(INODE_CACHE(inode), path,
1752                                      cluster_in_el);
1753                if (ret) {
1754                        mlog_errno(ret);
1755                        goto out;
1756                }
1757
1758                el = path_leaf_el(path);
1759
1760                i = ocfs2_find_rec(el, trunc_end);
1761                /*
1762                 * Need to go to previous extent block.
1763                 */
1764                if (i < 0) {
1765                        if (path->p_tree_depth == 0)
1766                                break;
1767
1768                        ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1769                                                            path,
1770                                                            &cluster_in_el);
1771                        if (ret) {
1772                                mlog_errno(ret);
1773                                goto out;
1774                        }
1775
1776                        /*
1777                         * We've reached the leftmost extent block,
1778                         * it's safe to leave.
1779                         */
1780                        if (cluster_in_el == 0)
1781                                break;
1782
1783                        /*
1784                         * The 'pos' searched for previous extent block is
1785                         * always one cluster less than actual trunc_end.
1786                         */
1787                        trunc_end = cluster_in_el + 1;
1788
1789                        ocfs2_reinit_path(path, 1);
1790
1791                        continue;
1792
1793                } else
1794                        rec = &el->l_recs[i];
1795
1796                ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1797                                     &trunc_len, &trunc_end, &blkno, &done);
1798                if (done)
1799                        break;
1800
1801                flags = rec->e_flags;
1802                phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1803
1804                ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1805                                               phys_cpos, trunc_len, flags,
1806                                               &dealloc, refcount_loc);
1807                if (ret < 0) {
1808                        mlog_errno(ret);
1809                        goto out;
1810                }
1811
1812                cluster_in_el = trunc_end;
1813
1814                ocfs2_reinit_path(path, 1);
1815        }
1816
1817        ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1818
1819out:
1820        ocfs2_schedule_truncate_log_flush(osb, 1);
1821        ocfs2_run_deallocs(osb, &dealloc);
1822
1823        return ret;
1824}
1825
1826/*
1827 * Parts of this function taken from xfs_change_file_space()
1828 */
1829static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1830                                     loff_t f_pos, unsigned int cmd,
1831                                     struct ocfs2_space_resv *sr,
1832                                     int change_size)
1833{
1834        int ret;
1835        s64 llen;
1836        loff_t size;
1837        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1838        struct buffer_head *di_bh = NULL;
1839        handle_t *handle;
1840        unsigned long long max_off = inode->i_sb->s_maxbytes;
1841
1842        if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1843                return -EROFS;
1844
1845        mutex_lock(&inode->i_mutex);
1846
1847        /*
1848         * This prevents concurrent writes on other nodes
1849         */
1850        ret = ocfs2_rw_lock(inode, 1);
1851        if (ret) {
1852                mlog_errno(ret);
1853                goto out;
1854        }
1855
1856        ret = ocfs2_inode_lock(inode, &di_bh, 1);
1857        if (ret) {
1858                mlog_errno(ret);
1859                goto out_rw_unlock;
1860        }
1861
1862        if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1863                ret = -EPERM;
1864                goto out_inode_unlock;
1865        }
1866
1867        switch (sr->l_whence) {
1868        case 0: /*SEEK_SET*/
1869                break;
1870        case 1: /*SEEK_CUR*/
1871                sr->l_start += f_pos;
1872                break;
1873        case 2: /*SEEK_END*/
1874                sr->l_start += i_size_read(inode);
1875                break;
1876        default:
1877                ret = -EINVAL;
1878                goto out_inode_unlock;
1879        }
1880        sr->l_whence = 0;
1881
1882        llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1883
1884        if (sr->l_start < 0
1885            || sr->l_start > max_off
1886            || (sr->l_start + llen) < 0
1887            || (sr->l_start + llen) > max_off) {
1888                ret = -EINVAL;
1889                goto out_inode_unlock;
1890        }
1891        size = sr->l_start + sr->l_len;
1892
1893        if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1894                if (sr->l_len <= 0) {
1895                        ret = -EINVAL;
1896                        goto out_inode_unlock;
1897                }
1898        }
1899
1900        if (file && should_remove_suid(file->f_path.dentry)) {
1901                ret = __ocfs2_write_remove_suid(inode, di_bh);
1902                if (ret) {
1903                        mlog_errno(ret);
1904                        goto out_inode_unlock;
1905                }
1906        }
1907
1908        down_write(&OCFS2_I(inode)->ip_alloc_sem);
1909        switch (cmd) {
1910        case OCFS2_IOC_RESVSP:
1911        case OCFS2_IOC_RESVSP64:
1912                /*
1913                 * This takes unsigned offsets, but the signed ones we
1914                 * pass have been checked against overflow above.
1915                 */
1916                ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1917                                                       sr->l_len);
1918                break;
1919        case OCFS2_IOC_UNRESVSP:
1920        case OCFS2_IOC_UNRESVSP64:
1921                ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1922                                               sr->l_len);
1923                break;
1924        default:
1925                ret = -EINVAL;
1926        }
1927        up_write(&OCFS2_I(inode)->ip_alloc_sem);
1928        if (ret) {
1929                mlog_errno(ret);
1930                goto out_inode_unlock;
1931        }
1932
1933        /*
1934         * We update c/mtime for these changes
1935         */
1936        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1937        if (IS_ERR(handle)) {
1938                ret = PTR_ERR(handle);
1939                mlog_errno(ret);
1940                goto out_inode_unlock;
1941        }
1942
1943        if (change_size && i_size_read(inode) < size)
1944                i_size_write(inode, size);
1945
1946        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1947        ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1948        if (ret < 0)
1949                mlog_errno(ret);
1950
1951        if (file && (file->f_flags & O_SYNC))
1952                handle->h_sync = 1;
1953
1954        ocfs2_commit_trans(osb, handle);
1955
1956out_inode_unlock:
1957        brelse(di_bh);
1958        ocfs2_inode_unlock(inode, 1);
1959out_rw_unlock:
1960        ocfs2_rw_unlock(inode, 1);
1961
1962out:
1963        mutex_unlock(&inode->i_mutex);
1964        return ret;
1965}
1966
1967int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1968                            struct ocfs2_space_resv *sr)
1969{
1970        struct inode *inode = file->f_path.dentry->d_inode;
1971        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1972        int ret;
1973
1974        if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1975            !ocfs2_writes_unwritten_extents(osb))
1976                return -ENOTTY;
1977        else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1978                 !ocfs2_sparse_alloc(osb))
1979                return -ENOTTY;
1980
1981        if (!S_ISREG(inode->i_mode))
1982                return -EINVAL;
1983
1984        if (!(file->f_mode & FMODE_WRITE))
1985                return -EBADF;
1986
1987        ret = mnt_want_write_file(file);
1988        if (ret)
1989                return ret;
1990        ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1991        mnt_drop_write_file(file);
1992        return ret;
1993}
1994
1995static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1996                            loff_t len)
1997{
1998        struct inode *inode = file->f_path.dentry->d_inode;
1999        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2000        struct ocfs2_space_resv sr;
2001        int change_size = 1;
2002        int cmd = OCFS2_IOC_RESVSP64;
2003
2004        if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2005                return -EOPNOTSUPP;
2006        if (!ocfs2_writes_unwritten_extents(osb))
2007                return -EOPNOTSUPP;
2008
2009        if (mode & FALLOC_FL_KEEP_SIZE)
2010                change_size = 0;
2011
2012        if (mode & FALLOC_FL_PUNCH_HOLE)
2013                cmd = OCFS2_IOC_UNRESVSP64;
2014
2015        sr.l_whence = 0;
2016        sr.l_start = (s64)offset;
2017        sr.l_len = (s64)len;
2018
2019        return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2020                                         change_size);
2021}
2022
2023int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2024                                   size_t count)
2025{
2026        int ret = 0;
2027        unsigned int extent_flags;
2028        u32 cpos, clusters, extent_len, phys_cpos;
2029        struct super_block *sb = inode->i_sb;
2030
2031        if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2032            !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2033            OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2034                return 0;
2035
2036        cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2037        clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2038
2039        while (clusters) {
2040                ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2041                                         &extent_flags);
2042                if (ret < 0) {
2043                        mlog_errno(ret);
2044                        goto out;
2045                }
2046
2047                if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2048                        ret = 1;
2049                        break;
2050                }
2051
2052                if (extent_len > clusters)
2053                        extent_len = clusters;
2054
2055                clusters -= extent_len;
2056                cpos += extent_len;
2057        }
2058out:
2059        return ret;
2060}
2061
2062static void ocfs2_aiodio_wait(struct inode *inode)
2063{
2064        wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
2065
2066        wait_event(*wq, (atomic_read(&OCFS2_I(inode)->ip_unaligned_aio) == 0));
2067}
2068
2069static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2070{
2071        int blockmask = inode->i_sb->s_blocksize - 1;
2072        loff_t final_size = pos + count;
2073
2074        if ((pos & blockmask) || (final_size & blockmask))
2075                return 1;
2076        return 0;
2077}
2078
2079static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2080                                            struct file *file,
2081                                            loff_t pos, size_t count,
2082                                            int *meta_level)
2083{
2084        int ret;
2085        struct buffer_head *di_bh = NULL;
2086        u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2087        u32 clusters =
2088                ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2089
2090        ret = ocfs2_inode_lock(inode, &di_bh, 1);
2091        if (ret) {
2092                mlog_errno(ret);
2093                goto out;
2094        }
2095
2096        *meta_level = 1;
2097
2098        ret = ocfs2_refcount_cow(inode, file, di_bh, cpos, clusters, UINT_MAX);
2099        if (ret)
2100                mlog_errno(ret);
2101out:
2102        brelse(di_bh);
2103        return ret;
2104}
2105
2106static int ocfs2_prepare_inode_for_write(struct file *file,
2107                                         loff_t *ppos,
2108                                         size_t count,
2109                                         int appending,
2110                                         int *direct_io,
2111                                         int *has_refcount)
2112{
2113        int ret = 0, meta_level = 0;
2114        struct dentry *dentry = file->f_path.dentry;
2115        struct inode *inode = dentry->d_inode;
2116        loff_t saved_pos = 0, end;
2117
2118        /*
2119         * We start with a read level meta lock and only jump to an ex
2120         * if we need to make modifications here.
2121         */
2122        for(;;) {
2123                ret = ocfs2_inode_lock(inode, NULL, meta_level);
2124                if (ret < 0) {
2125                        meta_level = -1;
2126                        mlog_errno(ret);
2127                        goto out;
2128                }
2129
2130                /* Clear suid / sgid if necessary. We do this here
2131                 * instead of later in the write path because
2132                 * remove_suid() calls ->setattr without any hint that
2133                 * we may have already done our cluster locking. Since
2134                 * ocfs2_setattr() *must* take cluster locks to
2135                 * proceed, this will lead us to recursively lock the
2136                 * inode. There's also the dinode i_size state which
2137                 * can be lost via setattr during extending writes (we
2138                 * set inode->i_size at the end of a write. */
2139                if (should_remove_suid(dentry)) {
2140                        if (meta_level == 0) {
2141                                ocfs2_inode_unlock(inode, meta_level);
2142                                meta_level = 1;
2143                                continue;
2144                        }
2145
2146                        ret = ocfs2_write_remove_suid(inode);
2147                        if (ret < 0) {
2148                                mlog_errno(ret);
2149                                goto out_unlock;
2150                        }
2151                }
2152
2153                /* work on a copy of ppos until we're sure that we won't have
2154                 * to recalculate it due to relocking. */
2155                if (appending)
2156                        saved_pos = i_size_read(inode);
2157                else
2158                        saved_pos = *ppos;
2159
2160                end = saved_pos + count;
2161
2162                ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2163                if (ret == 1) {
2164                        ocfs2_inode_unlock(inode, meta_level);
2165                        meta_level = -1;
2166
2167                        ret = ocfs2_prepare_inode_for_refcount(inode,
2168                                                               file,
2169                                                               saved_pos,
2170                                                               count,
2171                                                               &meta_level);
2172                        if (has_refcount)
2173                                *has_refcount = 1;
2174                        if (direct_io)
2175                                *direct_io = 0;
2176                }
2177
2178                if (ret < 0) {
2179                        mlog_errno(ret);
2180                        goto out_unlock;
2181                }
2182
2183                /*
2184                 * Skip the O_DIRECT checks if we don't need
2185                 * them.
2186                 */
2187                if (!direct_io || !(*direct_io))
2188                        break;
2189
2190                /*
2191                 * There's no sane way to do direct writes to an inode
2192                 * with inline data.
2193                 */
2194                if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2195                        *direct_io = 0;
2196                        break;
2197                }
2198
2199                /*
2200                 * Allowing concurrent direct writes means
2201                 * i_size changes wouldn't be synchronized, so
2202                 * one node could wind up truncating another
2203                 * nodes writes.
2204                 */
2205                if (end > i_size_read(inode)) {
2206                        *direct_io = 0;
2207                        break;
2208                }
2209
2210                /*
2211                 * We don't fill holes during direct io, so
2212                 * check for them here. If any are found, the
2213                 * caller will have to retake some cluster
2214                 * locks and initiate the io as buffered.
2215                 */
2216                ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2217                if (ret == 1) {
2218                        *direct_io = 0;
2219                        ret = 0;
2220                } else if (ret < 0)
2221                        mlog_errno(ret);
2222                break;
2223        }
2224
2225        if (appending)
2226                *ppos = saved_pos;
2227
2228out_unlock:
2229        trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2230                                            saved_pos, appending, count,
2231                                            direct_io, has_refcount);
2232
2233        if (meta_level >= 0)
2234                ocfs2_inode_unlock(inode, meta_level);
2235
2236out:
2237        return ret;
2238}
2239
2240static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2241                                    const struct iovec *iov,
2242                                    unsigned long nr_segs,
2243                                    loff_t pos)
2244{
2245        int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
2246        int can_do_direct, has_refcount = 0;
2247        ssize_t written = 0;
2248        size_t ocount;          /* original count */
2249        size_t count;           /* after file limit checks */
2250        loff_t old_size, *ppos = &iocb->ki_pos;
2251        u32 old_clusters;
2252        struct file *file = iocb->ki_filp;
2253        struct inode *inode = file->f_path.dentry->d_inode;
2254        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2255        int full_coherency = !(osb->s_mount_opt &
2256                               OCFS2_MOUNT_COHERENCY_BUFFERED);
2257        int unaligned_dio = 0;
2258
2259        trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2260                (unsigned long long)OCFS2_I(inode)->ip_blkno,
2261                file->f_path.dentry->d_name.len,
2262                file->f_path.dentry->d_name.name,
2263                (unsigned int)nr_segs);
2264
2265        if (iocb->ki_left == 0)
2266                return 0;
2267
2268        sb_start_write(inode->i_sb);
2269
2270        appending = file->f_flags & O_APPEND ? 1 : 0;
2271        direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2272
2273        mutex_lock(&inode->i_mutex);
2274
2275        ocfs2_iocb_clear_sem_locked(iocb);
2276
2277relock:
2278        /* to match setattr's i_mutex -> rw_lock ordering */
2279        if (direct_io) {
2280                have_alloc_sem = 1;
2281                /* communicate with ocfs2_dio_end_io */
2282                ocfs2_iocb_set_sem_locked(iocb);
2283        }
2284
2285        /*
2286         * Concurrent O_DIRECT writes are allowed with
2287         * mount_option "coherency=buffered".
2288         */
2289        rw_level = (!direct_io || full_coherency);
2290
2291        ret = ocfs2_rw_lock(inode, rw_level);
2292        if (ret < 0) {
2293                mlog_errno(ret);
2294                goto out_sems;
2295        }
2296
2297        /*
2298         * O_DIRECT writes with "coherency=full" need to take EX cluster
2299         * inode_lock to guarantee coherency.
2300         */
2301        if (direct_io && full_coherency) {
2302                /*
2303                 * We need to take and drop the inode lock to force
2304                 * other nodes to drop their caches.  Buffered I/O
2305                 * already does this in write_begin().
2306                 */
2307                ret = ocfs2_inode_lock(inode, NULL, 1);
2308                if (ret < 0) {
2309                        mlog_errno(ret);
2310                        goto out_sems;
2311                }
2312
2313                ocfs2_inode_unlock(inode, 1);
2314        }
2315
2316        can_do_direct = direct_io;
2317        ret = ocfs2_prepare_inode_for_write(file, ppos,
2318                                            iocb->ki_left, appending,
2319                                            &can_do_direct, &has_refcount);
2320        if (ret < 0) {
2321                mlog_errno(ret);
2322                goto out;
2323        }
2324
2325        if (direct_io && !is_sync_kiocb(iocb))
2326                unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_left,
2327                                                      *ppos);
2328
2329        /*
2330         * We can't complete the direct I/O as requested, fall back to
2331         * buffered I/O.
2332         */
2333        if (direct_io && !can_do_direct) {
2334                ocfs2_rw_unlock(inode, rw_level);
2335
2336                have_alloc_sem = 0;
2337                rw_level = -1;
2338
2339                direct_io = 0;
2340                goto relock;
2341        }
2342
2343        if (unaligned_dio) {
2344                /*
2345                 * Wait on previous unaligned aio to complete before
2346                 * proceeding.
2347                 */
2348                ocfs2_aiodio_wait(inode);
2349
2350                /* Mark the iocb as needing a decrement in ocfs2_dio_end_io */
2351                atomic_inc(&OCFS2_I(inode)->ip_unaligned_aio);
2352                ocfs2_iocb_set_unaligned_aio(iocb);
2353        }
2354
2355        /*
2356         * To later detect whether a journal commit for sync writes is
2357         * necessary, we sample i_size, and cluster count here.
2358         */
2359        old_size = i_size_read(inode);
2360        old_clusters = OCFS2_I(inode)->ip_clusters;
2361
2362        /* communicate with ocfs2_dio_end_io */
2363        ocfs2_iocb_set_rw_locked(iocb, rw_level);
2364
2365        ret = generic_segment_checks(iov, &nr_segs, &ocount,
2366                                     VERIFY_READ);
2367        if (ret)
2368                goto out_dio;
2369
2370        count = ocount;
2371        ret = generic_write_checks(file, ppos, &count,
2372                                   S_ISBLK(inode->i_mode));
2373        if (ret)
2374                goto out_dio;
2375
2376        if (direct_io) {
2377                written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2378                                                    ppos, count, ocount);
2379                if (written < 0) {
2380                        ret = written;
2381                        goto out_dio;
2382                }
2383        } else {
2384                current->backing_dev_info = file->f_mapping->backing_dev_info;
2385                written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2386                                                      ppos, count, 0);
2387                current->backing_dev_info = NULL;
2388        }
2389
2390out_dio:
2391        /* buffered aio wouldn't have proper lock coverage today */
2392        BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2393
2394        if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2395            ((file->f_flags & O_DIRECT) && !direct_io)) {
2396                ret = filemap_fdatawrite_range(file->f_mapping, pos,
2397                                               pos + count - 1);
2398                if (ret < 0)
2399                        written = ret;
2400
2401                if (!ret && ((old_size != i_size_read(inode)) ||
2402                             (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2403                             has_refcount)) {
2404                        ret = jbd2_journal_force_commit(osb->journal->j_journal);
2405                        if (ret < 0)
2406                                written = ret;
2407                }
2408
2409                if (!ret)
2410                        ret = filemap_fdatawait_range(file->f_mapping, pos,
2411                                                      pos + count - 1);
2412        }
2413
2414        /*
2415         * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2416         * function pointer which is called when o_direct io completes so that
2417         * it can unlock our rw lock.
2418         * Unfortunately there are error cases which call end_io and others
2419         * that don't.  so we don't have to unlock the rw_lock if either an
2420         * async dio is going to do it in the future or an end_io after an
2421         * error has already done it.
2422         */
2423        if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2424                rw_level = -1;
2425                have_alloc_sem = 0;
2426                unaligned_dio = 0;
2427        }
2428
2429        if (unaligned_dio) {
2430                ocfs2_iocb_clear_unaligned_aio(iocb);
2431                atomic_dec(&OCFS2_I(inode)->ip_unaligned_aio);
2432        }
2433
2434out:
2435        if (rw_level != -1)
2436                ocfs2_rw_unlock(inode, rw_level);
2437
2438out_sems:
2439        if (have_alloc_sem)
2440                ocfs2_iocb_clear_sem_locked(iocb);
2441
2442        mutex_unlock(&inode->i_mutex);
2443        sb_end_write(inode->i_sb);
2444
2445        if (written)
2446                ret = written;
2447        return ret;
2448}
2449
2450static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2451                                struct file *out,
2452                                struct splice_desc *sd)
2453{
2454        int ret;
2455
2456        ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2457                                            sd->total_len, 0, NULL, NULL);
2458        if (ret < 0) {
2459                mlog_errno(ret);
2460                return ret;
2461        }
2462
2463        return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2464}
2465
2466static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2467                                       struct file *out,
2468                                       loff_t *ppos,
2469                                       size_t len,
2470                                       unsigned int flags)
2471{
2472        int ret;
2473        struct address_space *mapping = out->f_mapping;
2474        struct inode *inode = mapping->host;
2475        struct splice_desc sd = {
2476                .total_len = len,
2477                .flags = flags,
2478                .pos = *ppos,
2479                .u.file = out,
2480        };
2481
2482
2483        trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2484                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
2485                        out->f_path.dentry->d_name.len,
2486                        out->f_path.dentry->d_name.name, len);
2487
2488        if (pipe->inode)
2489                mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
2490
2491        splice_from_pipe_begin(&sd);
2492        do {
2493                ret = splice_from_pipe_next(pipe, &sd);
2494                if (ret <= 0)
2495                        break;
2496
2497                mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2498                ret = ocfs2_rw_lock(inode, 1);
2499                if (ret < 0)
2500                        mlog_errno(ret);
2501                else {
2502                        ret = ocfs2_splice_to_file(pipe, out, &sd);
2503                        ocfs2_rw_unlock(inode, 1);
2504                }
2505                mutex_unlock(&inode->i_mutex);
2506        } while (ret > 0);
2507        splice_from_pipe_end(pipe, &sd);
2508
2509        if (pipe->inode)
2510                mutex_unlock(&pipe->inode->i_mutex);
2511
2512        if (sd.num_spliced)
2513                ret = sd.num_spliced;
2514
2515        if (ret > 0) {
2516                unsigned long nr_pages;
2517                int err;
2518
2519                nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2520
2521                err = generic_write_sync(out, *ppos, ret);
2522                if (err)
2523                        ret = err;
2524                else
2525                        *ppos += ret;
2526
2527                balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2528        }
2529
2530        return ret;
2531}
2532
2533static ssize_t ocfs2_file_splice_read(struct file *in,
2534                                      loff_t *ppos,
2535                                      struct pipe_inode_info *pipe,
2536                                      size_t len,
2537                                      unsigned int flags)
2538{
2539        int ret = 0, lock_level = 0;
2540        struct inode *inode = in->f_path.dentry->d_inode;
2541
2542        trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2543                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
2544                        in->f_path.dentry->d_name.len,
2545                        in->f_path.dentry->d_name.name, len);
2546
2547        /*
2548         * See the comment in ocfs2_file_aio_read()
2549         */
2550        ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
2551        if (ret < 0) {
2552                mlog_errno(ret);
2553                goto bail;
2554        }
2555        ocfs2_inode_unlock(inode, lock_level);
2556
2557        ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2558
2559bail:
2560        return ret;
2561}
2562
2563static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2564                                   const struct iovec *iov,
2565                                   unsigned long nr_segs,
2566                                   loff_t pos)
2567{
2568        int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2569        struct file *filp = iocb->ki_filp;
2570        struct inode *inode = filp->f_path.dentry->d_inode;
2571
2572        trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2573                        (unsigned long long)OCFS2_I(inode)->ip_blkno,
2574                        filp->f_path.dentry->d_name.len,
2575                        filp->f_path.dentry->d_name.name, nr_segs);
2576
2577
2578        if (!inode) {
2579                ret = -EINVAL;
2580                mlog_errno(ret);
2581                goto bail;
2582        }
2583
2584        ocfs2_iocb_clear_sem_locked(iocb);
2585
2586        /*
2587         * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2588         * need locks to protect pending reads from racing with truncate.
2589         */
2590        if (filp->f_flags & O_DIRECT) {
2591                have_alloc_sem = 1;
2592                ocfs2_iocb_set_sem_locked(iocb);
2593
2594                ret = ocfs2_rw_lock(inode, 0);
2595                if (ret < 0) {
2596                        mlog_errno(ret);
2597                        goto bail;
2598                }
2599                rw_level = 0;
2600                /* communicate with ocfs2_dio_end_io */
2601                ocfs2_iocb_set_rw_locked(iocb, rw_level);
2602        }
2603
2604        /*
2605         * We're fine letting folks race truncates and extending
2606         * writes with read across the cluster, just like they can
2607         * locally. Hence no rw_lock during read.
2608         *
2609         * Take and drop the meta data lock to update inode fields
2610         * like i_size. This allows the checks down below
2611         * generic_file_aio_read() a chance of actually working.
2612         */
2613        ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
2614        if (ret < 0) {
2615                mlog_errno(ret);
2616                goto bail;
2617        }
2618        ocfs2_inode_unlock(inode, lock_level);
2619
2620        ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2621        trace_generic_file_aio_read_ret(ret);
2622
2623        /* buffered aio wouldn't have proper lock coverage today */
2624        BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2625
2626        /* see ocfs2_file_aio_write */
2627        if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2628                rw_level = -1;
2629                have_alloc_sem = 0;
2630        }
2631
2632bail:
2633        if (have_alloc_sem)
2634                ocfs2_iocb_clear_sem_locked(iocb);
2635
2636        if (rw_level != -1)
2637                ocfs2_rw_unlock(inode, rw_level);
2638
2639        return ret;
2640}
2641
2642/* Refer generic_file_llseek_unlocked() */
2643static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int origin)
2644{
2645        struct inode *inode = file->f_mapping->host;
2646        int ret = 0;
2647
2648        mutex_lock(&inode->i_mutex);
2649
2650        switch (origin) {
2651        case SEEK_SET:
2652                break;
2653        case SEEK_END:
2654                offset += inode->i_size;
2655                break;
2656        case SEEK_CUR:
2657                if (offset == 0) {
2658                        offset = file->f_pos;
2659                        goto out;
2660                }
2661                offset += file->f_pos;
2662                break;
2663        case SEEK_DATA:
2664        case SEEK_HOLE:
2665                ret = ocfs2_seek_data_hole_offset(file, &offset, origin);
2666                if (ret)
2667                        goto out;
2668                break;
2669        default:
2670                ret = -EINVAL;
2671                goto out;
2672        }
2673
2674        if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2675                ret = -EINVAL;
2676        if (!ret && offset > inode->i_sb->s_maxbytes)
2677                ret = -EINVAL;
2678        if (ret)
2679                goto out;
2680
2681        if (offset != file->f_pos) {
2682                file->f_pos = offset;
2683                file->f_version = 0;
2684        }
2685
2686out:
2687        mutex_unlock(&inode->i_mutex);
2688        if (ret)
2689                return ret;
2690        return offset;
2691}
2692
2693const struct inode_operations ocfs2_file_iops = {
2694        .setattr        = ocfs2_setattr,
2695        .getattr        = ocfs2_getattr,
2696        .permission     = ocfs2_permission,
2697        .setxattr       = generic_setxattr,
2698        .getxattr       = generic_getxattr,
2699        .listxattr      = ocfs2_listxattr,
2700        .removexattr    = generic_removexattr,
2701        .fiemap         = ocfs2_fiemap,
2702        .get_acl        = ocfs2_iop_get_acl,
2703};
2704
2705const struct inode_operations ocfs2_special_file_iops = {
2706        .setattr        = ocfs2_setattr,
2707        .getattr        = ocfs2_getattr,
2708        .permission     = ocfs2_permission,
2709        .get_acl        = ocfs2_iop_get_acl,
2710};
2711
2712/*
2713 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2714 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2715 */
2716const struct file_operations ocfs2_fops = {
2717        .llseek         = ocfs2_file_llseek,
2718        .read           = do_sync_read,
2719        .write          = do_sync_write,
2720        .mmap           = ocfs2_mmap,
2721        .fsync          = ocfs2_sync_file,
2722        .release        = ocfs2_file_release,
2723        .open           = ocfs2_file_open,
2724        .aio_read       = ocfs2_file_aio_read,
2725        .aio_write      = ocfs2_file_aio_write,
2726        .unlocked_ioctl = ocfs2_ioctl,
2727#ifdef CONFIG_COMPAT
2728        .compat_ioctl   = ocfs2_compat_ioctl,
2729#endif
2730        .lock           = ocfs2_lock,
2731        .flock          = ocfs2_flock,
2732        .splice_read    = ocfs2_file_splice_read,
2733        .splice_write   = ocfs2_file_splice_write,
2734        .fallocate      = ocfs2_fallocate,
2735};
2736
2737const struct file_operations ocfs2_dops = {
2738        .llseek         = generic_file_llseek,
2739        .read           = generic_read_dir,
2740        .readdir        = ocfs2_readdir,
2741        .fsync          = ocfs2_sync_file,
2742        .release        = ocfs2_dir_release,
2743        .open           = ocfs2_dir_open,
2744        .unlocked_ioctl = ocfs2_ioctl,
2745#ifdef CONFIG_COMPAT
2746        .compat_ioctl   = ocfs2_compat_ioctl,
2747#endif
2748        .lock           = ocfs2_lock,
2749        .flock          = ocfs2_flock,
2750};
2751
2752/*
2753 * POSIX-lockless variants of our file_operations.
2754 *
2755 * These will be used if the underlying cluster stack does not support
2756 * posix file locking, if the user passes the "localflocks" mount
2757 * option, or if we have a local-only fs.
2758 *
2759 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2760 * so we still want it in the case of no stack support for
2761 * plocks. Internally, it will do the right thing when asked to ignore
2762 * the cluster.
2763 */
2764const struct file_operations ocfs2_fops_no_plocks = {
2765        .llseek         = ocfs2_file_llseek,
2766        .read           = do_sync_read,
2767        .write          = do_sync_write,
2768        .mmap           = ocfs2_mmap,
2769        .fsync          = ocfs2_sync_file,
2770        .release        = ocfs2_file_release,
2771        .open           = ocfs2_file_open,
2772        .aio_read       = ocfs2_file_aio_read,
2773        .aio_write      = ocfs2_file_aio_write,
2774        .unlocked_ioctl = ocfs2_ioctl,
2775#ifdef CONFIG_COMPAT
2776        .compat_ioctl   = ocfs2_compat_ioctl,
2777#endif
2778        .flock          = ocfs2_flock,
2779        .splice_read    = ocfs2_file_splice_read,
2780        .splice_write   = ocfs2_file_splice_write,
2781        .fallocate      = ocfs2_fallocate,
2782};
2783
2784const struct file_operations ocfs2_dops_no_plocks = {
2785        .llseek         = generic_file_llseek,
2786        .read           = generic_read_dir,
2787        .readdir        = ocfs2_readdir,
2788        .fsync          = ocfs2_sync_file,
2789        .release        = ocfs2_dir_release,
2790        .open           = ocfs2_dir_open,
2791        .unlocked_ioctl = ocfs2_ioctl,
2792#ifdef CONFIG_COMPAT
2793        .compat_ioctl   = ocfs2_compat_ioctl,
2794#endif
2795        .flock          = ocfs2_flock,
2796};
2797
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.