1/* -*- mode: c; c-basic-offset: 8; -*-
   2 * vim: noexpandtab sw=8 ts=8 sts=0:
   3 *
   4 * file.c
   5 *
   6 * File open, close, extend, truncate
   7 *
   8 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9 *
  10 * This program is free software; you can redistribute it and/or
  11 * modify it under the terms of the GNU General Public
  12 * License as published by the Free Software Foundation; either
  13 * version 2 of the License, or (at your option) any later version.
  14 *
  15 * This program is distributed in the hope that it will be useful,
  16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18 * General Public License for more details.
  19 *
  20 * You should have received a copy of the GNU General Public
  21 * License along with this program; if not, write to the
  22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23 * Boston, MA 021110-1307, USA.
  24 */
  25
  26#include <linux/capability.h>
  27#include <linux/fs.h>
  28#include <linux/types.h>
  29#include <linux/slab.h>
  30#include <linux/highmem.h>
  31#include <linux/pagemap.h>
  32#include <linux/uio.h>
  33#include <linux/sched.h>
  34#include <linux/splice.h>
  35#include <linux/mount.h>
  36#include <linux/writeback.h>
  37#include <linux/falloc.h>
  38#include <linux/quotaops.h>
  39#include <linux/blkdev.h>
  40#include <linux/backing-dev.h>
  41
  42#include <cluster/masklog.h>
  43
  44#include "ocfs2.h"
  45
  46#include "alloc.h"
  47#include "aops.h"
  48#include "dir.h"
  49#include "dlmglue.h"
  50#include "extent_map.h"
  51#include "file.h"
  52#include "sysfile.h"
  53#include "inode.h"
  54#include "ioctl.h"
  55#include "journal.h"
  56#include "locks.h"
  57#include "mmap.h"
  58#include "suballoc.h"
  59#include "super.h"
  60#include "xattr.h"
  61#include "acl.h"
  62#include "quota.h"
  63#include "refcounttree.h"
  64#include "ocfs2_trace.h"
  65
  66#include "buffer_head_io.h"
  67
  68static int ocfs2_init_file_private(struct inode *inode, struct file *file)
  69{
  70	struct ocfs2_file_private *fp;
  71
  72	fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
  73	if (!fp)
  74		return -ENOMEM;
  75
  76	fp->fp_file = file;
  77	mutex_init(&fp->fp_mutex);
  78	ocfs2_file_lock_res_init(&fp->fp_flock, fp);
  79	file->private_data = fp;
  80
  81	return 0;
  82}
  83
  84static void ocfs2_free_file_private(struct inode *inode, struct file *file)
  85{
  86	struct ocfs2_file_private *fp = file->private_data;
  87	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  88
  89	if (fp) {
  90		ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
  91		ocfs2_lock_res_free(&fp->fp_flock);
  92		kfree(fp);
  93		file->private_data = NULL;
  94	}
  95}
  96
  97static int ocfs2_file_open(struct inode *inode, struct file *file)
  98{
  99	int status;
 100	int mode = file->f_flags;
 101	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 102
 103	trace_ocfs2_file_open(inode, file, file->f_path.dentry,
 104			      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 105			      file->f_path.dentry->d_name.len,
 106			      file->f_path.dentry->d_name.name, mode);
 107
 108	if (file->f_mode & FMODE_WRITE) {
 109		status = dquot_initialize(inode);
 110		if (status)
 111			goto leave;
 112	}
 113
 114	spin_lock(&oi->ip_lock);
 115
 116	/* Check that the inode hasn't been wiped from disk by another
 117	 * node. If it hasn't then we're safe as long as we hold the
 118	 * spin lock until our increment of open count. */
 119	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
 120		spin_unlock(&oi->ip_lock);
 121
 122		status = -ENOENT;
 123		goto leave;
 124	}
 125
 126	if (mode & O_DIRECT)
 127		oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
 128
 129	oi->ip_open_count++;
 130	spin_unlock(&oi->ip_lock);
 131
 132	status = ocfs2_init_file_private(inode, file);
 133	if (status) {
 134		/*
 135		 * We want to set open count back if we're failing the
 136		 * open.
 137		 */
 138		spin_lock(&oi->ip_lock);
 139		oi->ip_open_count--;
 140		spin_unlock(&oi->ip_lock);
 141	}
 142
 143leave:
 144	return status;
 145}
 146
 147static int ocfs2_file_release(struct inode *inode, struct file *file)
 148{
 149	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 150
 151	spin_lock(&oi->ip_lock);
 152	if (!--oi->ip_open_count)
 153		oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
 154
 155	trace_ocfs2_file_release(inode, file, file->f_path.dentry,
 156				 oi->ip_blkno,
 157				 file->f_path.dentry->d_name.len,
 158				 file->f_path.dentry->d_name.name,
 159				 oi->ip_open_count);
 160	spin_unlock(&oi->ip_lock);
 161
 162	ocfs2_free_file_private(inode, file);
 163
 164	return 0;
 165}
 166
 167static int ocfs2_dir_open(struct inode *inode, struct file *file)
 168{
 169	return ocfs2_init_file_private(inode, file);
 170}
 171
 172static int ocfs2_dir_release(struct inode *inode, struct file *file)
 173{
 174	ocfs2_free_file_private(inode, file);
 175	return 0;
 176}
 177
 178static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
 179			   int datasync)
 180{
 181	int err = 0;
 
 182	struct inode *inode = file->f_mapping->host;
 183	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 184	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 185	journal_t *journal = osb->journal->j_journal;
 186	int ret;
 187	tid_t commit_tid;
 188	bool needs_barrier = false;
 189
 190	trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
 191			      OCFS2_I(inode)->ip_blkno,
 192			      file->f_path.dentry->d_name.len,
 193			      file->f_path.dentry->d_name.name,
 194			      (unsigned long long)datasync);
 195
 196	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 197		return -EROFS;
 198
 199	err = filemap_write_and_wait_range(inode->i_mapping, start, end);
 200	if (err)
 201		return err;
 202
 203	commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
 204	if (journal->j_flags & JBD2_BARRIER &&
 205	    !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 206		needs_barrier = true;
 207	err = jbd2_complete_transaction(journal, commit_tid);
 208	if (needs_barrier) {
 209		ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 210		if (!err)
 211			err = ret;
 
 
 
 
 
 212	}
 213
 
 
 
 
 214	if (err)
 215		mlog_errno(err);
 
 216
 217	return (err < 0) ? -EIO : 0;
 218}
 219
 220int ocfs2_should_update_atime(struct inode *inode,
 221			      struct vfsmount *vfsmnt)
 222{
 223	struct timespec now;
 224	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 225
 226	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
 227		return 0;
 228
 229	if ((inode->i_flags & S_NOATIME) ||
 230	    ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
 231		return 0;
 232
 233	/*
 234	 * We can be called with no vfsmnt structure - NFSD will
 235	 * sometimes do this.
 236	 *
 237	 * Note that our action here is different than touch_atime() -
 238	 * if we can't tell whether this is a noatime mount, then we
 239	 * don't know whether to trust the value of s_atime_quantum.
 240	 */
 241	if (vfsmnt == NULL)
 242		return 0;
 243
 244	if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
 245	    ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 246		return 0;
 247
 248	if (vfsmnt->mnt_flags & MNT_RELATIME) {
 249		if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
 250		    (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
 251			return 1;
 252
 253		return 0;
 254	}
 255
 256	now = CURRENT_TIME;
 257	if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
 258		return 0;
 259	else
 260		return 1;
 261}
 262
 263int ocfs2_update_inode_atime(struct inode *inode,
 264			     struct buffer_head *bh)
 265{
 266	int ret;
 267	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 268	handle_t *handle;
 269	struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 270
 271	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 272	if (IS_ERR(handle)) {
 273		ret = PTR_ERR(handle);
 274		mlog_errno(ret);
 275		goto out;
 276	}
 277
 278	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 279				      OCFS2_JOURNAL_ACCESS_WRITE);
 280	if (ret) {
 281		mlog_errno(ret);
 282		goto out_commit;
 283	}
 284
 285	/*
 286	 * Don't use ocfs2_mark_inode_dirty() here as we don't always
 287	 * have i_mutex to guard against concurrent changes to other
 288	 * inode fields.
 289	 */
 290	inode->i_atime = CURRENT_TIME;
 291	di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 292	di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 293	ocfs2_update_inode_fsync_trans(handle, inode, 0);
 294	ocfs2_journal_dirty(handle, bh);
 295
 296out_commit:
 297	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 298out:
 299	return ret;
 300}
 301
 302int ocfs2_set_inode_size(handle_t *handle,
 303				struct inode *inode,
 304				struct buffer_head *fe_bh,
 305				u64 new_i_size)
 306{
 307	int status;
 308
 309	i_size_write(inode, new_i_size);
 310	inode->i_blocks = ocfs2_inode_sector_count(inode);
 311	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 312
 313	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 314	if (status < 0) {
 315		mlog_errno(status);
 316		goto bail;
 317	}
 318
 319bail:
 320	return status;
 321}
 322
 323int ocfs2_simple_size_update(struct inode *inode,
 324			     struct buffer_head *di_bh,
 325			     u64 new_i_size)
 326{
 327	int ret;
 328	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 329	handle_t *handle = NULL;
 330
 331	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 332	if (IS_ERR(handle)) {
 333		ret = PTR_ERR(handle);
 334		mlog_errno(ret);
 335		goto out;
 336	}
 337
 338	ret = ocfs2_set_inode_size(handle, inode, di_bh,
 339				   new_i_size);
 340	if (ret < 0)
 341		mlog_errno(ret);
 342
 343	ocfs2_update_inode_fsync_trans(handle, inode, 0);
 344	ocfs2_commit_trans(osb, handle);
 345out:
 346	return ret;
 347}
 348
 349static int ocfs2_cow_file_pos(struct inode *inode,
 350			      struct buffer_head *fe_bh,
 351			      u64 offset)
 352{
 353	int status;
 354	u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 355	unsigned int num_clusters = 0;
 356	unsigned int ext_flags = 0;
 357
 358	/*
 359	 * If the new offset is aligned to the range of the cluster, there is
 360	 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
 361	 * CoW either.
 362	 */
 363	if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
 364		return 0;
 365
 366	status = ocfs2_get_clusters(inode, cpos, &phys,
 367				    &num_clusters, &ext_flags);
 368	if (status) {
 369		mlog_errno(status);
 370		goto out;
 371	}
 372
 373	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
 374		goto out;
 375
 376	return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
 377
 378out:
 379	return status;
 380}
 381
 382static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 383				     struct inode *inode,
 384				     struct buffer_head *fe_bh,
 385				     u64 new_i_size)
 386{
 387	int status;
 388	handle_t *handle;
 389	struct ocfs2_dinode *di;
 390	u64 cluster_bytes;
 391
 392	/*
 393	 * We need to CoW the cluster contains the offset if it is reflinked
 394	 * since we will call ocfs2_zero_range_for_truncate later which will
 395	 * write "0" from offset to the end of the cluster.
 396	 */
 397	status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
 398	if (status) {
 399		mlog_errno(status);
 400		return status;
 401	}
 402
 403	/* TODO: This needs to actually orphan the inode in this
 404	 * transaction. */
 405
 406	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 407	if (IS_ERR(handle)) {
 408		status = PTR_ERR(handle);
 409		mlog_errno(status);
 410		goto out;
 411	}
 412
 413	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 414					 OCFS2_JOURNAL_ACCESS_WRITE);
 415	if (status < 0) {
 416		mlog_errno(status);
 417		goto out_commit;
 418	}
 419
 420	/*
 421	 * Do this before setting i_size.
 422	 */
 423	cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
 424	status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
 425					       cluster_bytes);
 426	if (status) {
 427		mlog_errno(status);
 428		goto out_commit;
 429	}
 430
 431	i_size_write(inode, new_i_size);
 432	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 433
 434	di = (struct ocfs2_dinode *) fe_bh->b_data;
 435	di->i_size = cpu_to_le64(new_i_size);
 436	di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
 437	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 438	ocfs2_update_inode_fsync_trans(handle, inode, 0);
 439
 440	ocfs2_journal_dirty(handle, fe_bh);
 441
 442out_commit:
 443	ocfs2_commit_trans(osb, handle);
 444out:
 445	return status;
 446}
 447
 448int ocfs2_truncate_file(struct inode *inode,
 449			       struct buffer_head *di_bh,
 450			       u64 new_i_size)
 451{
 452	int status = 0;
 453	struct ocfs2_dinode *fe = NULL;
 454	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 455
 456	/* We trust di_bh because it comes from ocfs2_inode_lock(), which
 457	 * already validated it */
 458	fe = (struct ocfs2_dinode *) di_bh->b_data;
 459
 460	trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
 461				  (unsigned long long)le64_to_cpu(fe->i_size),
 462				  (unsigned long long)new_i_size);
 463
 464	mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
 465			"Inode %llu, inode i_size = %lld != di "
 466			"i_size = %llu, i_flags = 0x%x\n",
 467			(unsigned long long)OCFS2_I(inode)->ip_blkno,
 468			i_size_read(inode),
 469			(unsigned long long)le64_to_cpu(fe->i_size),
 470			le32_to_cpu(fe->i_flags));
 471
 472	if (new_i_size > le64_to_cpu(fe->i_size)) {
 473		trace_ocfs2_truncate_file_error(
 474			(unsigned long long)le64_to_cpu(fe->i_size),
 475			(unsigned long long)new_i_size);
 476		status = -EINVAL;
 477		mlog_errno(status);
 478		goto bail;
 479	}
 480
 
 
 
 
 
 481	down_write(&OCFS2_I(inode)->ip_alloc_sem);
 482
 483	ocfs2_resv_discard(&osb->osb_la_resmap,
 484			   &OCFS2_I(inode)->ip_la_data_resv);
 485
 486	/*
 487	 * The inode lock forced other nodes to sync and drop their
 488	 * pages, which (correctly) happens even if we have a truncate
 489	 * without allocation change - ocfs2 cluster sizes can be much
 490	 * greater than page size, so we have to truncate them
 491	 * anyway.
 492	 */
 493	unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
 494	truncate_inode_pages(inode->i_mapping, new_i_size);
 495
 496	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 497		status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
 498					       i_size_read(inode), 1);
 499		if (status)
 500			mlog_errno(status);
 501
 502		goto bail_unlock_sem;
 503	}
 504
 505	/* alright, we're going to need to do a full blown alloc size
 506	 * change. Orphan the inode so that recovery can complete the
 507	 * truncate if necessary. This does the task of marking
 508	 * i_size. */
 509	status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 510	if (status < 0) {
 511		mlog_errno(status);
 512		goto bail_unlock_sem;
 513	}
 514
 515	status = ocfs2_commit_truncate(osb, inode, di_bh);
 516	if (status < 0) {
 517		mlog_errno(status);
 518		goto bail_unlock_sem;
 519	}
 520
 521	/* TODO: orphan dir cleanup here. */
 522bail_unlock_sem:
 523	up_write(&OCFS2_I(inode)->ip_alloc_sem);
 524
 525bail:
 526	if (!status && OCFS2_I(inode)->ip_clusters == 0)
 527		status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 528
 529	return status;
 530}
 531
 532/*
 533 * extend file allocation only here.
 534 * we'll update all the disk stuff, and oip->alloc_size
 535 *
 536 * expect stuff to be locked, a transaction started and enough data /
 537 * metadata reservations in the contexts.
 538 *
 539 * Will return -EAGAIN, and a reason if a restart is needed.
 540 * If passed in, *reason will always be set, even in error.
 541 */
 542int ocfs2_add_inode_data(struct ocfs2_super *osb,
 543			 struct inode *inode,
 544			 u32 *logical_offset,
 545			 u32 clusters_to_add,
 546			 int mark_unwritten,
 547			 struct buffer_head *fe_bh,
 548			 handle_t *handle,
 549			 struct ocfs2_alloc_context *data_ac,
 550			 struct ocfs2_alloc_context *meta_ac,
 551			 enum ocfs2_alloc_restarted *reason_ret)
 552{
 553	int ret;
 554	struct ocfs2_extent_tree et;
 555
 556	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
 557	ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
 558					  clusters_to_add, mark_unwritten,
 559					  data_ac, meta_ac, reason_ret);
 560
 561	return ret;
 562}
 563
 564static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 565				     u32 clusters_to_add, int mark_unwritten)
 566{
 567	int status = 0;
 568	int restart_func = 0;
 569	int credits;
 570	u32 prev_clusters;
 571	struct buffer_head *bh = NULL;
 572	struct ocfs2_dinode *fe = NULL;
 573	handle_t *handle = NULL;
 574	struct ocfs2_alloc_context *data_ac = NULL;
 575	struct ocfs2_alloc_context *meta_ac = NULL;
 576	enum ocfs2_alloc_restarted why = RESTART_NONE;
 577	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 578	struct ocfs2_extent_tree et;
 579	int did_quota = 0;
 580
 581	/*
 582	 * Unwritten extent only exists for file systems which
 583	 * support holes.
 584	 */
 585	BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 586
 587	status = ocfs2_read_inode_block(inode, &bh);
 588	if (status < 0) {
 589		mlog_errno(status);
 590		goto leave;
 591	}
 592	fe = (struct ocfs2_dinode *) bh->b_data;
 593
 594restart_all:
 595	BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 596
 597	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 598	status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 599				       &data_ac, &meta_ac);
 600	if (status) {
 601		mlog_errno(status);
 602		goto leave;
 603	}
 604
 605	credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
 
 606	handle = ocfs2_start_trans(osb, credits);
 607	if (IS_ERR(handle)) {
 608		status = PTR_ERR(handle);
 609		handle = NULL;
 610		mlog_errno(status);
 611		goto leave;
 612	}
 613
 614restarted_transaction:
 615	trace_ocfs2_extend_allocation(
 616		(unsigned long long)OCFS2_I(inode)->ip_blkno,
 617		(unsigned long long)i_size_read(inode),
 618		le32_to_cpu(fe->i_clusters), clusters_to_add,
 619		why, restart_func);
 620
 621	status = dquot_alloc_space_nodirty(inode,
 622			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 623	if (status)
 624		goto leave;
 625	did_quota = 1;
 626
 627	/* reserve a write to the file entry early on - that we if we
 628	 * run out of credits in the allocation path, we can still
 629	 * update i_size. */
 630	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 631					 OCFS2_JOURNAL_ACCESS_WRITE);
 632	if (status < 0) {
 633		mlog_errno(status);
 634		goto leave;
 635	}
 636
 637	prev_clusters = OCFS2_I(inode)->ip_clusters;
 638
 639	status = ocfs2_add_inode_data(osb,
 640				      inode,
 641				      &logical_start,
 642				      clusters_to_add,
 643				      mark_unwritten,
 644				      bh,
 645				      handle,
 646				      data_ac,
 647				      meta_ac,
 648				      &why);
 649	if ((status < 0) && (status != -EAGAIN)) {
 650		if (status != -ENOSPC)
 651			mlog_errno(status);
 652		goto leave;
 653	}
 654	ocfs2_update_inode_fsync_trans(handle, inode, 1);
 655	ocfs2_journal_dirty(handle, bh);
 656
 657	spin_lock(&OCFS2_I(inode)->ip_lock);
 658	clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
 659	spin_unlock(&OCFS2_I(inode)->ip_lock);
 660	/* Release unused quota reservation */
 661	dquot_free_space(inode,
 662			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 663	did_quota = 0;
 664
 665	if (why != RESTART_NONE && clusters_to_add) {
 666		if (why == RESTART_META) {
 667			restart_func = 1;
 668			status = 0;
 669		} else {
 670			BUG_ON(why != RESTART_TRANS);
 671
 672			status = ocfs2_allocate_extend_trans(handle, 1);
 
 
 
 
 673			if (status < 0) {
 674				/* handle still has to be committed at
 675				 * this point. */
 676				status = -ENOMEM;
 677				mlog_errno(status);
 678				goto leave;
 679			}
 680			goto restarted_transaction;
 681		}
 682	}
 683
 684	trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
 685	     le32_to_cpu(fe->i_clusters),
 686	     (unsigned long long)le64_to_cpu(fe->i_size),
 687	     OCFS2_I(inode)->ip_clusters,
 688	     (unsigned long long)i_size_read(inode));
 689
 690leave:
 691	if (status < 0 && did_quota)
 692		dquot_free_space(inode,
 693			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
 694	if (handle) {
 695		ocfs2_commit_trans(osb, handle);
 696		handle = NULL;
 697	}
 698	if (data_ac) {
 699		ocfs2_free_alloc_context(data_ac);
 700		data_ac = NULL;
 701	}
 702	if (meta_ac) {
 703		ocfs2_free_alloc_context(meta_ac);
 704		meta_ac = NULL;
 705	}
 706	if ((!status) && restart_func) {
 707		restart_func = 0;
 708		goto restart_all;
 709	}
 710	brelse(bh);
 711	bh = NULL;
 712
 713	return status;
 714}
 715
 716int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
 717		u32 clusters_to_add, int mark_unwritten)
 718{
 719	return __ocfs2_extend_allocation(inode, logical_start,
 720			clusters_to_add, mark_unwritten);
 721}
 722
 723/*
 724 * While a write will already be ordering the data, a truncate will not.
 725 * Thus, we need to explicitly order the zeroed pages.
 726 */
 727static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
 728						struct buffer_head *di_bh)
 729{
 730	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 731	handle_t *handle = NULL;
 732	int ret = 0;
 733
 734	if (!ocfs2_should_order_data(inode))
 735		goto out;
 736
 737	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 738	if (IS_ERR(handle)) {
 739		ret = -ENOMEM;
 740		mlog_errno(ret);
 741		goto out;
 742	}
 743
 744	ret = ocfs2_jbd2_file_inode(handle, inode);
 745	if (ret < 0) {
 746		mlog_errno(ret);
 747		goto out;
 748	}
 749
 750	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 751				      OCFS2_JOURNAL_ACCESS_WRITE);
 752	if (ret)
 753		mlog_errno(ret);
 754	ocfs2_update_inode_fsync_trans(handle, inode, 1);
 755
 756out:
 757	if (ret) {
 758		if (!IS_ERR(handle))
 759			ocfs2_commit_trans(osb, handle);
 760		handle = ERR_PTR(ret);
 761	}
 762	return handle;
 763}
 764
 765/* Some parts of this taken from generic_cont_expand, which turned out
 766 * to be too fragile to do exactly what we need without us having to
 767 * worry about recursive locking in ->write_begin() and ->write_end(). */
 768static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
 769				 u64 abs_to, struct buffer_head *di_bh)
 770{
 771	struct address_space *mapping = inode->i_mapping;
 772	struct page *page;
 773	unsigned long index = abs_from >> PAGE_SHIFT;
 774	handle_t *handle;
 775	int ret = 0;
 776	unsigned zero_from, zero_to, block_start, block_end;
 777	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 778
 779	BUG_ON(abs_from >= abs_to);
 780	BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
 781	BUG_ON(abs_from & (inode->i_blkbits - 1));
 782
 783	handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
 784	if (IS_ERR(handle)) {
 785		ret = PTR_ERR(handle);
 786		goto out;
 787	}
 788
 789	page = find_or_create_page(mapping, index, GFP_NOFS);
 790	if (!page) {
 791		ret = -ENOMEM;
 792		mlog_errno(ret);
 793		goto out_commit_trans;
 794	}
 795
 796	/* Get the offsets within the page that we want to zero */
 797	zero_from = abs_from & (PAGE_SIZE - 1);
 798	zero_to = abs_to & (PAGE_SIZE - 1);
 799	if (!zero_to)
 800		zero_to = PAGE_SIZE;
 801
 802	trace_ocfs2_write_zero_page(
 803			(unsigned long long)OCFS2_I(inode)->ip_blkno,
 804			(unsigned long long)abs_from,
 805			(unsigned long long)abs_to,
 806			index, zero_from, zero_to);
 807
 808	/* We know that zero_from is block aligned */
 809	for (block_start = zero_from; block_start < zero_to;
 810	     block_start = block_end) {
 811		block_end = block_start + (1 << inode->i_blkbits);
 812
 813		/*
 814		 * block_start is block-aligned.  Bump it by one to force
 815		 * __block_write_begin and block_commit_write to zero the
 816		 * whole block.
 817		 */
 818		ret = __block_write_begin(page, block_start + 1, 0,
 819					  ocfs2_get_block);
 820		if (ret < 0) {
 821			mlog_errno(ret);
 822			goto out_unlock;
 823		}
 824
 
 
 
 
 
 
 
 
 825
 826		/* must not update i_size! */
 827		ret = block_commit_write(page, block_start + 1,
 828					 block_start + 1);
 829		if (ret < 0)
 830			mlog_errno(ret);
 831		else
 832			ret = 0;
 833	}
 834
 835	/*
 836	 * fs-writeback will release the dirty pages without page lock
 837	 * whose offset are over inode size, the release happens at
 838	 * block_write_full_page().
 839	 */
 840	i_size_write(inode, abs_to);
 841	inode->i_blocks = ocfs2_inode_sector_count(inode);
 842	di->i_size = cpu_to_le64((u64)i_size_read(inode));
 843	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 844	di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
 845	di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 846	di->i_mtime_nsec = di->i_ctime_nsec;
 847	if (handle) {
 848		ocfs2_journal_dirty(handle, di_bh);
 849		ocfs2_update_inode_fsync_trans(handle, inode, 1);
 850	}
 851
 852out_unlock:
 853	unlock_page(page);
 854	put_page(page);
 855out_commit_trans:
 856	if (handle)
 857		ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 858out:
 859	return ret;
 860}
 861
 862/*
 863 * Find the next range to zero.  We do this in terms of bytes because
 864 * that's what ocfs2_zero_extend() wants, and it is dealing with the
 865 * pagecache.  We may return multiple extents.
 866 *
 867 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
 868 * needs to be zeroed.  range_start and range_end return the next zeroing
 869 * range.  A subsequent call should pass the previous range_end as its
 870 * zero_start.  If range_end is 0, there's nothing to do.
 871 *
 872 * Unwritten extents are skipped over.  Refcounted extents are CoWd.
 873 */
 874static int ocfs2_zero_extend_get_range(struct inode *inode,
 875				       struct buffer_head *di_bh,
 876				       u64 zero_start, u64 zero_end,
 877				       u64 *range_start, u64 *range_end)
 878{
 879	int rc = 0, needs_cow = 0;
 880	u32 p_cpos, zero_clusters = 0;
 881	u32 zero_cpos =
 882		zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 883	u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
 884	unsigned int num_clusters = 0;
 885	unsigned int ext_flags = 0;
 886
 887	while (zero_cpos < last_cpos) {
 888		rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
 889					&num_clusters, &ext_flags);
 890		if (rc) {
 891			mlog_errno(rc);
 892			goto out;
 893		}
 894
 895		if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
 896			zero_clusters = num_clusters;
 897			if (ext_flags & OCFS2_EXT_REFCOUNTED)
 898				needs_cow = 1;
 899			break;
 900		}
 901
 902		zero_cpos += num_clusters;
 903	}
 904	if (!zero_clusters) {
 905		*range_end = 0;
 906		goto out;
 907	}
 908
 909	while ((zero_cpos + zero_clusters) < last_cpos) {
 910		rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
 911					&p_cpos, &num_clusters,
 912					&ext_flags);
 913		if (rc) {
 914			mlog_errno(rc);
 915			goto out;
 916		}
 917
 918		if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
 919			break;
 920		if (ext_flags & OCFS2_EXT_REFCOUNTED)
 921			needs_cow = 1;
 922		zero_clusters += num_clusters;
 923	}
 924	if ((zero_cpos + zero_clusters) > last_cpos)
 925		zero_clusters = last_cpos - zero_cpos;
 926
 927	if (needs_cow) {
 928		rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
 929					zero_clusters, UINT_MAX);
 930		if (rc) {
 931			mlog_errno(rc);
 932			goto out;
 933		}
 934	}
 935
 936	*range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
 937	*range_end = ocfs2_clusters_to_bytes(inode->i_sb,
 938					     zero_cpos + zero_clusters);
 939
 940out:
 941	return rc;
 942}
 943
 944/*
 945 * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
 946 * has made sure that the entire range needs zeroing.
 947 */
 948static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
 949				   u64 range_end, struct buffer_head *di_bh)
 950{
 951	int rc = 0;
 952	u64 next_pos;
 953	u64 zero_pos = range_start;
 954
 955	trace_ocfs2_zero_extend_range(
 956			(unsigned long long)OCFS2_I(inode)->ip_blkno,
 957			(unsigned long long)range_start,
 958			(unsigned long long)range_end);
 959	BUG_ON(range_start >= range_end);
 960
 961	while (zero_pos < range_end) {
 962		next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
 963		if (next_pos > range_end)
 964			next_pos = range_end;
 965		rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
 966		if (rc < 0) {
 967			mlog_errno(rc);
 968			break;
 969		}
 970		zero_pos = next_pos;
 971
 972		/*
 973		 * Very large extends have the potential to lock up
 974		 * the cpu for extended periods of time.
 975		 */
 976		cond_resched();
 977	}
 978
 979	return rc;
 980}
 981
 982int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
 983		      loff_t zero_to_size)
 984{
 985	int ret = 0;
 986	u64 zero_start, range_start = 0, range_end = 0;
 987	struct super_block *sb = inode->i_sb;
 988
 989	zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
 990	trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
 991				(unsigned long long)zero_start,
 992				(unsigned long long)i_size_read(inode));
 993	while (zero_start < zero_to_size) {
 994		ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
 995						  zero_to_size,
 996						  &range_start,
 997						  &range_end);
 998		if (ret) {
 999			mlog_errno(ret);
1000			break;
1001		}
1002		if (!range_end)
1003			break;
1004		/* Trim the ends */
1005		if (range_start < zero_start)
1006			range_start = zero_start;
1007		if (range_end > zero_to_size)
1008			range_end = zero_to_size;
1009
1010		ret = ocfs2_zero_extend_range(inode, range_start,
1011					      range_end, di_bh);
1012		if (ret) {
1013			mlog_errno(ret);
1014			break;
1015		}
1016		zero_start = range_end;
1017	}
1018
1019	return ret;
1020}
1021
1022int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1023			  u64 new_i_size, u64 zero_to)
1024{
1025	int ret;
1026	u32 clusters_to_add;
1027	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1028
1029	/*
1030	 * Only quota files call this without a bh, and they can't be
1031	 * refcounted.
1032	 */
1033	BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1034	BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1035
1036	clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1037	if (clusters_to_add < oi->ip_clusters)
1038		clusters_to_add = 0;
1039	else
1040		clusters_to_add -= oi->ip_clusters;
1041
1042	if (clusters_to_add) {
1043		ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1044						clusters_to_add, 0);
1045		if (ret) {
1046			mlog_errno(ret);
1047			goto out;
1048		}
1049	}
1050
1051	/*
1052	 * Call this even if we don't add any clusters to the tree. We
1053	 * still need to zero the area between the old i_size and the
1054	 * new i_size.
1055	 */
1056	ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1057	if (ret < 0)
1058		mlog_errno(ret);
1059
1060out:
1061	return ret;
1062}
1063
1064static int ocfs2_extend_file(struct inode *inode,
1065			     struct buffer_head *di_bh,
1066			     u64 new_i_size)
1067{
1068	int ret = 0;
1069	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1070
1071	BUG_ON(!di_bh);
1072
1073	/* setattr sometimes calls us like this. */
1074	if (new_i_size == 0)
1075		goto out;
1076
1077	if (i_size_read(inode) == new_i_size)
1078		goto out;
1079	BUG_ON(new_i_size < i_size_read(inode));
1080
1081	/*
1082	 * The alloc sem blocks people in read/write from reading our
1083	 * allocation until we're done changing it. We depend on
1084	 * i_mutex to block other extend/truncate calls while we're
1085	 * here.  We even have to hold it for sparse files because there
1086	 * might be some tail zeroing.
1087	 */
1088	down_write(&oi->ip_alloc_sem);
1089
1090	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1091		/*
1092		 * We can optimize small extends by keeping the inodes
1093		 * inline data.
1094		 */
1095		if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1096			up_write(&oi->ip_alloc_sem);
1097			goto out_update_size;
1098		}
1099
1100		ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1101		if (ret) {
1102			up_write(&oi->ip_alloc_sem);
1103			mlog_errno(ret);
1104			goto out;
1105		}
1106	}
1107
1108	if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1109		ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1110	else
1111		ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1112					    new_i_size);
1113
1114	up_write(&oi->ip_alloc_sem);
1115
1116	if (ret < 0) {
1117		mlog_errno(ret);
1118		goto out;
1119	}
1120
1121out_update_size:
1122	ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1123	if (ret < 0)
1124		mlog_errno(ret);
1125
1126out:
1127	return ret;
1128}
1129
1130int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1131{
1132	int status = 0, size_change;
1133	int inode_locked = 0;
1134	struct inode *inode = d_inode(dentry);
1135	struct super_block *sb = inode->i_sb;
1136	struct ocfs2_super *osb = OCFS2_SB(sb);
1137	struct buffer_head *bh = NULL;
1138	handle_t *handle = NULL;
1139	struct dquot *transfer_to[MAXQUOTAS] = { };
1140	int qtype;
1141
1142	trace_ocfs2_setattr(inode, dentry,
1143			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
1144			    dentry->d_name.len, dentry->d_name.name,
1145			    attr->ia_valid, attr->ia_mode,
1146			    from_kuid(&init_user_ns, attr->ia_uid),
1147			    from_kgid(&init_user_ns, attr->ia_gid));
1148
1149	/* ensuring we don't even attempt to truncate a symlink */
1150	if (S_ISLNK(inode->i_mode))
1151		attr->ia_valid &= ~ATTR_SIZE;
1152
1153#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1154			   | ATTR_GID | ATTR_UID | ATTR_MODE)
1155	if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1156		return 0;
1157
1158	status = inode_change_ok(inode, attr);
1159	if (status)
1160		return status;
1161
1162	if (is_quota_modification(inode, attr)) {
1163		status = dquot_initialize(inode);
1164		if (status)
1165			return status;
1166	}
1167	size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1168	if (size_change) {
1169		status = ocfs2_rw_lock(inode, 1);
1170		if (status < 0) {
1171			mlog_errno(status);
1172			goto bail;
1173		}
1174	}
1175
1176	status = ocfs2_inode_lock(inode, &bh, 1);
1177	if (status < 0) {
1178		if (status != -ENOENT)
1179			mlog_errno(status);
1180		goto bail_unlock_rw;
1181	}
1182	inode_locked = 1;
1183
1184	if (size_change) {
1185		status = inode_newsize_ok(inode, attr->ia_size);
1186		if (status)
1187			goto bail_unlock;
1188
1189		inode_dio_wait(inode);
1190
1191		if (i_size_read(inode) >= attr->ia_size) {
1192			if (ocfs2_should_order_data(inode)) {
1193				status = ocfs2_begin_ordered_truncate(inode,
1194								      attr->ia_size);
1195				if (status)
1196					goto bail_unlock;
1197			}
1198			status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1199		} else
1200			status = ocfs2_extend_file(inode, bh, attr->ia_size);
1201		if (status < 0) {
1202			if (status != -ENOSPC)
1203				mlog_errno(status);
1204			status = -ENOSPC;
1205			goto bail_unlock;
1206		}
1207	}
1208
1209	if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1210	    (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1211		/*
1212		 * Gather pointers to quota structures so that allocation /
1213		 * freeing of quota structures happens here and not inside
1214		 * dquot_transfer() where we have problems with lock ordering
1215		 */
1216		if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1217		    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1218		    OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1219			transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1220			if (IS_ERR(transfer_to[USRQUOTA])) {
1221				status = PTR_ERR(transfer_to[USRQUOTA]);
 
1222				goto bail_unlock;
1223			}
1224		}
1225		if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1226		    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1227		    OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1228			transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1229			if (IS_ERR(transfer_to[GRPQUOTA])) {
1230				status = PTR_ERR(transfer_to[GRPQUOTA]);
 
1231				goto bail_unlock;
1232			}
1233		}
1234		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1235					   2 * ocfs2_quota_trans_credits(sb));
1236		if (IS_ERR(handle)) {
1237			status = PTR_ERR(handle);
1238			mlog_errno(status);
1239			goto bail_unlock;
1240		}
1241		status = __dquot_transfer(inode, transfer_to);
1242		if (status < 0)
1243			goto bail_commit;
1244	} else {
1245		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1246		if (IS_ERR(handle)) {
1247			status = PTR_ERR(handle);
1248			mlog_errno(status);
1249			goto bail_unlock;
1250		}
1251	}
1252
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1253	setattr_copy(inode, attr);
1254	mark_inode_dirty(inode);
1255
1256	status = ocfs2_mark_inode_dirty(handle, inode, bh);
1257	if (status < 0)
1258		mlog_errno(status);
1259
1260bail_commit:
1261	ocfs2_commit_trans(osb, handle);
1262bail_unlock:
1263	if (status) {
1264		ocfs2_inode_unlock(inode, 1);
1265		inode_locked = 0;
1266	}
1267bail_unlock_rw:
1268	if (size_change)
1269		ocfs2_rw_unlock(inode, 1);
1270bail:
 
1271
1272	/* Release quota pointers in case we acquired them */
1273	for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1274		dqput(transfer_to[qtype]);
1275
1276	if (!status && attr->ia_valid & ATTR_MODE) {
1277		status = ocfs2_acl_chmod(inode, bh);
1278		if (status < 0)
1279			mlog_errno(status);
1280	}
1281	if (inode_locked)
1282		ocfs2_inode_unlock(inode, 1);
1283
1284	brelse(bh);
1285	return status;
1286}
1287
1288int ocfs2_getattr(struct vfsmount *mnt,
1289		  struct dentry *dentry,
1290		  struct kstat *stat)
1291{
1292	struct inode *inode = d_inode(dentry);
1293	struct super_block *sb = d_inode(dentry)->i_sb;
1294	struct ocfs2_super *osb = sb->s_fs_info;
1295	int err;
1296
1297	err = ocfs2_inode_revalidate(dentry);
1298	if (err) {
1299		if (err != -ENOENT)
1300			mlog_errno(err);
1301		goto bail;
1302	}
1303
1304	generic_fillattr(inode, stat);
1305	/*
1306	 * If there is inline data in the inode, the inode will normally not
1307	 * have data blocks allocated (it may have an external xattr block).
1308	 * Report at least one sector for such files, so tools like tar, rsync,
1309	 * others don't incorrectly think the file is completely sparse.
1310	 */
1311	if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1312		stat->blocks += (stat->size + 511)>>9;
1313
1314	/* We set the blksize from the cluster size for performance */
1315	stat->blksize = osb->s_clustersize;
1316
1317bail:
1318	return err;
1319}
1320
1321int ocfs2_permission(struct inode *inode, int mask)
1322{
1323	int ret;
1324
1325	if (mask & MAY_NOT_BLOCK)
1326		return -ECHILD;
1327
1328	ret = ocfs2_inode_lock(inode, NULL, 0);
1329	if (ret) {
1330		if (ret != -ENOENT)
1331			mlog_errno(ret);
1332		goto out;
1333	}
1334
1335	ret = generic_permission(inode, mask);
1336
1337	ocfs2_inode_unlock(inode, 0);
1338out:
1339	return ret;
1340}
1341
1342static int __ocfs2_write_remove_suid(struct inode *inode,
1343				     struct buffer_head *bh)
1344{
1345	int ret;
1346	handle_t *handle;
1347	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1348	struct ocfs2_dinode *di;
1349
1350	trace_ocfs2_write_remove_suid(
1351			(unsigned long long)OCFS2_I(inode)->ip_blkno,
1352			inode->i_mode);
1353
1354	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1355	if (IS_ERR(handle)) {
1356		ret = PTR_ERR(handle);
1357		mlog_errno(ret);
1358		goto out;
1359	}
1360
1361	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1362				      OCFS2_JOURNAL_ACCESS_WRITE);
1363	if (ret < 0) {
1364		mlog_errno(ret);
1365		goto out_trans;
1366	}
1367
1368	inode->i_mode &= ~S_ISUID;
1369	if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1370		inode->i_mode &= ~S_ISGID;
1371
1372	di = (struct ocfs2_dinode *) bh->b_data;
1373	di->i_mode = cpu_to_le16(inode->i_mode);
1374	ocfs2_update_inode_fsync_trans(handle, inode, 0);
1375
1376	ocfs2_journal_dirty(handle, bh);
1377
1378out_trans:
1379	ocfs2_commit_trans(osb, handle);
1380out:
1381	return ret;
1382}
1383
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1384static int ocfs2_write_remove_suid(struct inode *inode)
1385{
1386	int ret;
1387	struct buffer_head *bh = NULL;
1388
1389	ret = ocfs2_read_inode_block(inode, &bh);
1390	if (ret < 0) {
1391		mlog_errno(ret);
1392		goto out;
1393	}
1394
1395	ret =  __ocfs2_write_remove_suid(inode, bh);
1396out:
1397	brelse(bh);
1398	return ret;
1399}
1400
1401/*
1402 * Allocate enough extents to cover the region starting at byte offset
1403 * start for len bytes. Existing extents are skipped, any extents
1404 * added are marked as "unwritten".
1405 */
1406static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1407					    u64 start, u64 len)
1408{
1409	int ret;
1410	u32 cpos, phys_cpos, clusters, alloc_size;
1411	u64 end = start + len;
1412	struct buffer_head *di_bh = NULL;
1413
1414	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1415		ret = ocfs2_read_inode_block(inode, &di_bh);
1416		if (ret) {
1417			mlog_errno(ret);
1418			goto out;
1419		}
1420
1421		/*
1422		 * Nothing to do if the requested reservation range
1423		 * fits within the inode.
1424		 */
1425		if (ocfs2_size_fits_inline_data(di_bh, end))
1426			goto out;
1427
1428		ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1429		if (ret) {
1430			mlog_errno(ret);
1431			goto out;
1432		}
1433	}
1434
1435	/*
1436	 * We consider both start and len to be inclusive.
1437	 */
1438	cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1439	clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1440	clusters -= cpos;
1441
1442	while (clusters) {
1443		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1444					 &alloc_size, NULL);
1445		if (ret) {
1446			mlog_errno(ret);
1447			goto out;
1448		}
1449
1450		/*
1451		 * Hole or existing extent len can be arbitrary, so
1452		 * cap it to our own allocation request.
1453		 */
1454		if (alloc_size > clusters)
1455			alloc_size = clusters;
1456
1457		if (phys_cpos) {
1458			/*
1459			 * We already have an allocation at this
1460			 * region so we can safely skip it.
1461			 */
1462			goto next;
1463		}
1464
1465		ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1466		if (ret) {
1467			if (ret != -ENOSPC)
1468				mlog_errno(ret);
1469			goto out;
1470		}
1471
1472next:
1473		cpos += alloc_size;
1474		clusters -= alloc_size;
1475	}
1476
1477	ret = 0;
1478out:
1479
1480	brelse(di_bh);
1481	return ret;
1482}
1483
1484/*
1485 * Truncate a byte range, avoiding pages within partial clusters. This
1486 * preserves those pages for the zeroing code to write to.
1487 */
1488static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1489					 u64 byte_len)
1490{
1491	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1492	loff_t start, end;
1493	struct address_space *mapping = inode->i_mapping;
1494
1495	start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1496	end = byte_start + byte_len;
1497	end = end & ~(osb->s_clustersize - 1);
1498
1499	if (start < end) {
1500		unmap_mapping_range(mapping, start, end - start, 0);
1501		truncate_inode_pages_range(mapping, start, end - 1);
1502	}
1503}
1504
1505static int ocfs2_zero_partial_clusters(struct inode *inode,
1506				       u64 start, u64 len)
1507{
1508	int ret = 0;
1509	u64 tmpend, end = start + len;
1510	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1511	unsigned int csize = osb->s_clustersize;
1512	handle_t *handle;
1513
1514	/*
1515	 * The "start" and "end" values are NOT necessarily part of
1516	 * the range whose allocation is being deleted. Rather, this
1517	 * is what the user passed in with the request. We must zero
1518	 * partial clusters here. There's no need to worry about
1519	 * physical allocation - the zeroing code knows to skip holes.
1520	 */
1521	trace_ocfs2_zero_partial_clusters(
1522		(unsigned long long)OCFS2_I(inode)->ip_blkno,
1523		(unsigned long long)start, (unsigned long long)end);
1524
1525	/*
1526	 * If both edges are on a cluster boundary then there's no
1527	 * zeroing required as the region is part of the allocation to
1528	 * be truncated.
1529	 */
1530	if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1531		goto out;
1532
1533	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1534	if (IS_ERR(handle)) {
1535		ret = PTR_ERR(handle);
1536		mlog_errno(ret);
1537		goto out;
1538	}
1539
1540	/*
1541	 * We want to get the byte offset of the end of the 1st cluster.
1542	 */
1543	tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1544	if (tmpend > end)
1545		tmpend = end;
1546
1547	trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1548						 (unsigned long long)tmpend);
1549
1550	ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1551	if (ret)
1552		mlog_errno(ret);
1553
1554	if (tmpend < end) {
1555		/*
1556		 * This may make start and end equal, but the zeroing
1557		 * code will skip any work in that case so there's no
1558		 * need to catch it up here.
1559		 */
1560		start = end & ~(osb->s_clustersize - 1);
1561
1562		trace_ocfs2_zero_partial_clusters_range2(
1563			(unsigned long long)start, (unsigned long long)end);
1564
1565		ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1566		if (ret)
1567			mlog_errno(ret);
1568	}
1569	ocfs2_update_inode_fsync_trans(handle, inode, 1);
1570
1571	ocfs2_commit_trans(osb, handle);
1572out:
1573	return ret;
1574}
1575
1576static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1577{
1578	int i;
1579	struct ocfs2_extent_rec *rec = NULL;
1580
1581	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1582
1583		rec = &el->l_recs[i];
1584
1585		if (le32_to_cpu(rec->e_cpos) < pos)
1586			break;
1587	}
1588
1589	return i;
1590}
1591
1592/*
1593 * Helper to calculate the punching pos and length in one run, we handle the
1594 * following three cases in order:
1595 *
1596 * - remove the entire record
1597 * - remove a partial record
1598 * - no record needs to be removed (hole-punching completed)
1599*/
1600static void ocfs2_calc_trunc_pos(struct inode *inode,
1601				 struct ocfs2_extent_list *el,
1602				 struct ocfs2_extent_rec *rec,
1603				 u32 trunc_start, u32 *trunc_cpos,
1604				 u32 *trunc_len, u32 *trunc_end,
1605				 u64 *blkno, int *done)
1606{
1607	int ret = 0;
1608	u32 coff, range;
1609
1610	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1611
1612	if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1613		/*
1614		 * remove an entire extent record.
1615		 */
1616		*trunc_cpos = le32_to_cpu(rec->e_cpos);
1617		/*
1618		 * Skip holes if any.
1619		 */
1620		if (range < *trunc_end)
1621			*trunc_end = range;
1622		*trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1623		*blkno = le64_to_cpu(rec->e_blkno);
1624		*trunc_end = le32_to_cpu(rec->e_cpos);
1625	} else if (range > trunc_start) {
1626		/*
1627		 * remove a partial extent record, which means we're
1628		 * removing the last extent record.
1629		 */
1630		*trunc_cpos = trunc_start;
1631		/*
1632		 * skip hole if any.
1633		 */
1634		if (range < *trunc_end)
1635			*trunc_end = range;
1636		*trunc_len = *trunc_end - trunc_start;
1637		coff = trunc_start - le32_to_cpu(rec->e_cpos);
1638		*blkno = le64_to_cpu(rec->e_blkno) +
1639				ocfs2_clusters_to_blocks(inode->i_sb, coff);
1640		*trunc_end = trunc_start;
1641	} else {
1642		/*
1643		 * It may have two following possibilities:
1644		 *
1645		 * - last record has been removed
1646		 * - trunc_start was within a hole
1647		 *
1648		 * both two cases mean the completion of hole punching.
1649		 */
1650		ret = 1;
1651	}
1652
1653	*done = ret;
1654}
1655
1656static int ocfs2_remove_inode_range(struct inode *inode,
1657				    struct buffer_head *di_bh, u64 byte_start,
1658				    u64 byte_len)
1659{
1660	int ret = 0, flags = 0, done = 0, i;
1661	u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1662	u32 cluster_in_el;
1663	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1664	struct ocfs2_cached_dealloc_ctxt dealloc;
1665	struct address_space *mapping = inode->i_mapping;
1666	struct ocfs2_extent_tree et;
1667	struct ocfs2_path *path = NULL;
1668	struct ocfs2_extent_list *el = NULL;
1669	struct ocfs2_extent_rec *rec = NULL;
1670	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1671	u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1672
1673	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1674	ocfs2_init_dealloc_ctxt(&dealloc);
1675
1676	trace_ocfs2_remove_inode_range(
1677			(unsigned long long)OCFS2_I(inode)->ip_blkno,
1678			(unsigned long long)byte_start,
1679			(unsigned long long)byte_len);
1680
1681	if (byte_len == 0)
1682		return 0;
1683
1684	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1685		ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1686					    byte_start + byte_len, 0);
1687		if (ret) {
1688			mlog_errno(ret);
1689			goto out;
1690		}
1691		/*
1692		 * There's no need to get fancy with the page cache
1693		 * truncate of an inline-data inode. We're talking
1694		 * about less than a page here, which will be cached
1695		 * in the dinode buffer anyway.
1696		 */
1697		unmap_mapping_range(mapping, 0, 0, 0);
1698		truncate_inode_pages(mapping, 0);
1699		goto out;
1700	}
1701
1702	/*
1703	 * For reflinks, we may need to CoW 2 clusters which might be
1704	 * partially zero'd later, if hole's start and end offset were
1705	 * within one cluster(means is not exactly aligned to clustersize).
1706	 */
1707
1708	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1709
1710		ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1711		if (ret) {
1712			mlog_errno(ret);
1713			goto out;
1714		}
1715
1716		ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1717		if (ret) {
1718			mlog_errno(ret);
1719			goto out;
1720		}
1721	}
1722
1723	trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1724	trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1725	cluster_in_el = trunc_end;
1726
1727	ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1728	if (ret) {
1729		mlog_errno(ret);
1730		goto out;
1731	}
1732
1733	path = ocfs2_new_path_from_et(&et);
1734	if (!path) {
1735		ret = -ENOMEM;
1736		mlog_errno(ret);
1737		goto out;
1738	}
1739
1740	while (trunc_end > trunc_start) {
1741
1742		ret = ocfs2_find_path(INODE_CACHE(inode), path,
1743				      cluster_in_el);
1744		if (ret) {
1745			mlog_errno(ret);
1746			goto out;
1747		}
1748
1749		el = path_leaf_el(path);
1750
1751		i = ocfs2_find_rec(el, trunc_end);
1752		/*
1753		 * Need to go to previous extent block.
1754		 */
1755		if (i < 0) {
1756			if (path->p_tree_depth == 0)
1757				break;
1758
1759			ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1760							    path,
1761							    &cluster_in_el);
1762			if (ret) {
1763				mlog_errno(ret);
1764				goto out;
1765			}
1766
1767			/*
1768			 * We've reached the leftmost extent block,
1769			 * it's safe to leave.
1770			 */
1771			if (cluster_in_el == 0)
1772				break;
1773
1774			/*
1775			 * The 'pos' searched for previous extent block is
1776			 * always one cluster less than actual trunc_end.
1777			 */
1778			trunc_end = cluster_in_el + 1;
1779
1780			ocfs2_reinit_path(path, 1);
1781
1782			continue;
1783
1784		} else
1785			rec = &el->l_recs[i];
1786
1787		ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1788				     &trunc_len, &trunc_end, &blkno, &done);
1789		if (done)
1790			break;
1791
1792		flags = rec->e_flags;
1793		phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1794
1795		ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1796					       phys_cpos, trunc_len, flags,
1797					       &dealloc, refcount_loc, false);
1798		if (ret < 0) {
1799			mlog_errno(ret);
1800			goto out;
1801		}
1802
1803		cluster_in_el = trunc_end;
1804
1805		ocfs2_reinit_path(path, 1);
1806	}
1807
1808	ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1809
1810out:
1811	ocfs2_free_path(path);
1812	ocfs2_schedule_truncate_log_flush(osb, 1);
1813	ocfs2_run_deallocs(osb, &dealloc);
1814
1815	return ret;
1816}
1817
1818/*
1819 * Parts of this function taken from xfs_change_file_space()
1820 */
1821static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1822				     loff_t f_pos, unsigned int cmd,
1823				     struct ocfs2_space_resv *sr,
1824				     int change_size)
1825{
1826	int ret;
1827	s64 llen;
1828	loff_t size;
1829	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1830	struct buffer_head *di_bh = NULL;
1831	handle_t *handle;
1832	unsigned long long max_off = inode->i_sb->s_maxbytes;
1833
1834	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1835		return -EROFS;
1836
1837	inode_lock(inode);
1838
1839	/*
1840	 * This prevents concurrent writes on other nodes
1841	 */
1842	ret = ocfs2_rw_lock(inode, 1);
1843	if (ret) {
1844		mlog_errno(ret);
1845		goto out;
1846	}
1847
1848	ret = ocfs2_inode_lock(inode, &di_bh, 1);
1849	if (ret) {
1850		mlog_errno(ret);
1851		goto out_rw_unlock;
1852	}
1853
1854	if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1855		ret = -EPERM;
1856		goto out_inode_unlock;
1857	}
1858
1859	switch (sr->l_whence) {
1860	case 0: /*SEEK_SET*/
1861		break;
1862	case 1: /*SEEK_CUR*/
1863		sr->l_start += f_pos;
1864		break;
1865	case 2: /*SEEK_END*/
1866		sr->l_start += i_size_read(inode);
1867		break;
1868	default:
1869		ret = -EINVAL;
1870		goto out_inode_unlock;
1871	}
1872	sr->l_whence = 0;
1873
1874	llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1875
1876	if (sr->l_start < 0
1877	    || sr->l_start > max_off
1878	    || (sr->l_start + llen) < 0
1879	    || (sr->l_start + llen) > max_off) {
1880		ret = -EINVAL;
1881		goto out_inode_unlock;
1882	}
1883	size = sr->l_start + sr->l_len;
1884
1885	if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1886	    cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1887		if (sr->l_len <= 0) {
1888			ret = -EINVAL;
1889			goto out_inode_unlock;
1890		}
1891	}
1892
1893	if (file && should_remove_suid(file->f_path.dentry)) {
1894		ret = __ocfs2_write_remove_suid(inode, di_bh);
1895		if (ret) {
1896			mlog_errno(ret);
1897			goto out_inode_unlock;
1898		}
1899	}
1900
1901	down_write(&OCFS2_I(inode)->ip_alloc_sem);
1902	switch (cmd) {
1903	case OCFS2_IOC_RESVSP:
1904	case OCFS2_IOC_RESVSP64:
1905		/*
1906		 * This takes unsigned offsets, but the signed ones we
1907		 * pass have been checked against overflow above.
1908		 */
1909		ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1910						       sr->l_len);
1911		break;
1912	case OCFS2_IOC_UNRESVSP:
1913	case OCFS2_IOC_UNRESVSP64:
1914		ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1915					       sr->l_len);
1916		break;
1917	default:
1918		ret = -EINVAL;
1919	}
1920	up_write(&OCFS2_I(inode)->ip_alloc_sem);
1921	if (ret) {
1922		mlog_errno(ret);
1923		goto out_inode_unlock;
1924	}
1925
1926	/*
1927	 * We update c/mtime for these changes
1928	 */
1929	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1930	if (IS_ERR(handle)) {
1931		ret = PTR_ERR(handle);
1932		mlog_errno(ret);
1933		goto out_inode_unlock;
1934	}
1935
1936	if (change_size && i_size_read(inode) < size)
1937		i_size_write(inode, size);
1938
1939	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1940	ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1941	if (ret < 0)
1942		mlog_errno(ret);
1943
1944	if (file && (file->f_flags & O_SYNC))
1945		handle->h_sync = 1;
1946
1947	ocfs2_commit_trans(osb, handle);
1948
1949out_inode_unlock:
1950	brelse(di_bh);
1951	ocfs2_inode_unlock(inode, 1);
1952out_rw_unlock:
1953	ocfs2_rw_unlock(inode, 1);
1954
1955out:
1956	inode_unlock(inode);
1957	return ret;
1958}
1959
1960int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1961			    struct ocfs2_space_resv *sr)
1962{
1963	struct inode *inode = file_inode(file);
1964	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1965	int ret;
1966
1967	if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1968	    !ocfs2_writes_unwritten_extents(osb))
1969		return -ENOTTY;
1970	else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1971		 !ocfs2_sparse_alloc(osb))
1972		return -ENOTTY;
1973
1974	if (!S_ISREG(inode->i_mode))
1975		return -EINVAL;
1976
1977	if (!(file->f_mode & FMODE_WRITE))
1978		return -EBADF;
1979
1980	ret = mnt_want_write_file(file);
1981	if (ret)
1982		return ret;
1983	ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1984	mnt_drop_write_file(file);
1985	return ret;
1986}
1987
1988static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1989			    loff_t len)
1990{
1991	struct inode *inode = file_inode(file);
1992	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1993	struct ocfs2_space_resv sr;
1994	int change_size = 1;
1995	int cmd = OCFS2_IOC_RESVSP64;
1996
1997	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1998		return -EOPNOTSUPP;
1999	if (!ocfs2_writes_unwritten_extents(osb))
2000		return -EOPNOTSUPP;
2001
2002	if (mode & FALLOC_FL_KEEP_SIZE)
2003		change_size = 0;
2004
2005	if (mode & FALLOC_FL_PUNCH_HOLE)
2006		cmd = OCFS2_IOC_UNRESVSP64;
2007
2008	sr.l_whence = 0;
2009	sr.l_start = (s64)offset;
2010	sr.l_len = (s64)len;
2011
2012	return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2013					 change_size);
2014}
2015
2016int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2017				   size_t count)
2018{
2019	int ret = 0;
2020	unsigned int extent_flags;
2021	u32 cpos, clusters, extent_len, phys_cpos;
2022	struct super_block *sb = inode->i_sb;
2023
2024	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2025	    !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2026	    OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2027		return 0;
2028
2029	cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2030	clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2031
2032	while (clusters) {
2033		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2034					 &extent_flags);
2035		if (ret < 0) {
2036			mlog_errno(ret);
2037			goto out;
2038		}
2039
2040		if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2041			ret = 1;
2042			break;
2043		}
2044
2045		if (extent_len > clusters)
2046			extent_len = clusters;
2047
2048		clusters -= extent_len;
2049		cpos += extent_len;
2050	}
2051out:
2052	return ret;
2053}
2054
 
 
 
 
 
 
 
2055static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2056{
2057	int blockmask = inode->i_sb->s_blocksize - 1;
2058	loff_t final_size = pos + count;
2059
2060	if ((pos & blockmask) || (final_size & blockmask))
2061		return 1;
2062	return 0;
2063}
2064
2065static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2066					    struct file *file,
2067					    loff_t pos, size_t count,
2068					    int *meta_level)
2069{
2070	int ret;
2071	struct buffer_head *di_bh = NULL;
2072	u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2073	u32 clusters =
2074		ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2075
2076	ret = ocfs2_inode_lock(inode, &di_bh, 1);
2077	if (ret) {
2078		mlog_errno(ret);
2079		goto out;
2080	}
2081
2082	*meta_level = 1;
2083
2084	ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2085	if (ret)
2086		mlog_errno(ret);
2087out:
2088	brelse(di_bh);
2089	return ret;
2090}
2091
2092static int ocfs2_prepare_inode_for_write(struct file *file,
2093					 loff_t pos,
2094					 size_t count)
 
 
 
2095{
2096	int ret = 0, meta_level = 0;
2097	struct dentry *dentry = file->f_path.dentry;
2098	struct inode *inode = d_inode(dentry);
2099	loff_t end;
2100
2101	/*
2102	 * We start with a read level meta lock and only jump to an ex
2103	 * if we need to make modifications here.
2104	 */
2105	for(;;) {
2106		ret = ocfs2_inode_lock(inode, NULL, meta_level);
2107		if (ret < 0) {
2108			meta_level = -1;
2109			mlog_errno(ret);
2110			goto out;
2111		}
2112
2113		/* Clear suid / sgid if necessary. We do this here
2114		 * instead of later in the write path because
2115		 * remove_suid() calls ->setattr without any hint that
2116		 * we may have already done our cluster locking. Since
2117		 * ocfs2_setattr() *must* take cluster locks to
2118		 * proceed, this will lead us to recursively lock the
2119		 * inode. There's also the dinode i_size state which
2120		 * can be lost via setattr during extending writes (we
2121		 * set inode->i_size at the end of a write. */
2122		if (should_remove_suid(dentry)) {
2123			if (meta_level == 0) {
2124				ocfs2_inode_unlock(inode, meta_level);
2125				meta_level = 1;
2126				continue;
2127			}
2128
2129			ret = ocfs2_write_remove_suid(inode);
2130			if (ret < 0) {
2131				mlog_errno(ret);
2132				goto out_unlock;
2133			}
2134		}
2135
2136		end = pos + count;
 
 
 
 
 
 
 
2137
2138		ret = ocfs2_check_range_for_refcount(inode, pos, count);
2139		if (ret == 1) {
2140			ocfs2_inode_unlock(inode, meta_level);
2141			meta_level = -1;
2142
2143			ret = ocfs2_prepare_inode_for_refcount(inode,
2144							       file,
2145							       pos,
2146							       count,
2147							       &meta_level);
 
 
 
 
2148		}
2149
2150		if (ret < 0) {
2151			mlog_errno(ret);
2152			goto out_unlock;
2153		}
2154
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2155		break;
2156	}
2157
 
 
 
2158out_unlock:
2159	trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2160					    pos, count);
 
2161
2162	if (meta_level >= 0)
2163		ocfs2_inode_unlock(inode, meta_level);
2164
2165out:
2166	return ret;
2167}
2168
2169static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2170				    struct iov_iter *from)
 
 
2171{
2172	int direct_io, rw_level;
 
2173	ssize_t written = 0;
2174	ssize_t ret;
2175	size_t count = iov_iter_count(from);
 
 
2176	struct file *file = iocb->ki_filp;
2177	struct inode *inode = file_inode(file);
2178	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2179	int full_coherency = !(osb->s_mount_opt &
2180			       OCFS2_MOUNT_COHERENCY_BUFFERED);
2181	void *saved_ki_complete = NULL;
2182	int append_write = ((iocb->ki_pos + count) >=
2183			i_size_read(inode) ? 1 : 0);
2184
2185	trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2186		(unsigned long long)OCFS2_I(inode)->ip_blkno,
2187		file->f_path.dentry->d_name.len,
2188		file->f_path.dentry->d_name.name,
2189		(unsigned int)from->nr_segs);	/* GRRRRR */
2190
2191	if (count == 0)
2192		return 0;
2193
2194	direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
 
 
 
2195
2196	inode_lock(inode);
 
 
 
 
 
 
 
 
 
 
2197
2198	/*
2199	 * Concurrent O_DIRECT writes are allowed with
2200	 * mount_option "coherency=buffered".
2201	 * For append write, we must take rw EX.
2202	 */
2203	rw_level = (!direct_io || full_coherency || append_write);
2204
2205	ret = ocfs2_rw_lock(inode, rw_level);
2206	if (ret < 0) {
2207		mlog_errno(ret);
2208		goto out_mutex;
2209	}
2210
2211	/*
2212	 * O_DIRECT writes with "coherency=full" need to take EX cluster
2213	 * inode_lock to guarantee coherency.
2214	 */
2215	if (direct_io && full_coherency) {
2216		/*
2217		 * We need to take and drop the inode lock to force
2218		 * other nodes to drop their caches.  Buffered I/O
2219		 * already does this in write_begin().
2220		 */
2221		ret = ocfs2_inode_lock(inode, NULL, 1);
2222		if (ret < 0) {
2223			mlog_errno(ret);
2224			goto out;
2225		}
2226
2227		ocfs2_inode_unlock(inode, 1);
2228	}
2229
2230	ret = generic_write_checks(iocb, from);
2231	if (ret <= 0) {
2232		if (ret)
2233			mlog_errno(ret);
2234		goto out;
2235	}
2236	count = ret;
2237
2238	ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count);
2239	if (ret < 0) {
2240		mlog_errno(ret);
2241		goto out;
2242	}
2243
2244	if (direct_io && !is_sync_kiocb(iocb) &&
2245	    ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2246		/*
2247		 * Make it a sync io if it's an unaligned aio.
 
2248		 */
2249		saved_ki_complete = xchg(&iocb->ki_complete, NULL);
 
 
 
 
2250	}
2251
 
 
 
 
 
 
 
2252	/* communicate with ocfs2_dio_end_io */
2253	ocfs2_iocb_set_rw_locked(iocb, rw_level);
2254
2255	written = __generic_file_write_iter(iocb, from);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2256	/* buffered aio wouldn't have proper lock coverage today */
2257	BUG_ON(written == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2258
2259	/*
2260	 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2261	 * function pointer which is called when o_direct io completes so that
2262	 * it can unlock our rw lock.
2263	 * Unfortunately there are error cases which call end_io and others
2264	 * that don't.  so we don't have to unlock the rw_lock if either an
2265	 * async dio is going to do it in the future or an end_io after an
2266	 * error has already done it.
2267	 */
2268	if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2269		rw_level = -1;
 
 
2270	}
2271
2272	if (unlikely(written <= 0))
2273		goto out;
2274
2275	if (((file->f_flags & O_DSYNC) && !direct_io) ||
2276	    IS_SYNC(inode)) {
2277		ret = filemap_fdatawrite_range(file->f_mapping,
2278					       iocb->ki_pos - written,
2279					       iocb->ki_pos - 1);
2280		if (ret < 0)
2281			written = ret;
2282
2283		if (!ret) {
2284			ret = jbd2_journal_force_commit(osb->journal->j_journal);
2285			if (ret < 0)
2286				written = ret;
2287		}
2288
2289		if (!ret)
2290			ret = filemap_fdatawait_range(file->f_mapping,
2291						      iocb->ki_pos - written,
2292						      iocb->ki_pos - 1);
2293	}
2294
2295out:
2296	if (saved_ki_complete)
2297		xchg(&iocb->ki_complete, saved_ki_complete);
2298
2299	if (rw_level != -1)
2300		ocfs2_rw_unlock(inode, rw_level);
2301
2302out_mutex:
2303	inode_unlock(inode);
 
 
 
2304
2305	if (written)
2306		ret = written;
2307	return ret;
2308}
2309
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2310static ssize_t ocfs2_file_splice_read(struct file *in,
2311				      loff_t *ppos,
2312				      struct pipe_inode_info *pipe,
2313				      size_t len,
2314				      unsigned int flags)
2315{
2316	int ret = 0, lock_level = 0;
2317	struct inode *inode = file_inode(in);
2318
2319	trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2320			(unsigned long long)OCFS2_I(inode)->ip_blkno,
2321			in->f_path.dentry->d_name.len,
2322			in->f_path.dentry->d_name.name, len);
2323
2324	/*
2325	 * See the comment in ocfs2_file_read_iter()
2326	 */
2327	ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2328	if (ret < 0) {
2329		mlog_errno(ret);
2330		goto bail;
2331	}
2332	ocfs2_inode_unlock(inode, lock_level);
2333
2334	ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2335
2336bail:
2337	return ret;
2338}
2339
2340static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2341				   struct iov_iter *to)
 
 
2342{
2343	int ret = 0, rw_level = -1, lock_level = 0;
2344	struct file *filp = iocb->ki_filp;
2345	struct inode *inode = file_inode(filp);
2346
2347	trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2348			(unsigned long long)OCFS2_I(inode)->ip_blkno,
2349			filp->f_path.dentry->d_name.len,
2350			filp->f_path.dentry->d_name.name,
2351			to->nr_segs);	/* GRRRRR */
2352
2353
2354	if (!inode) {
2355		ret = -EINVAL;
2356		mlog_errno(ret);
2357		goto bail;
2358	}
2359
 
 
2360	/*
2361	 * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2362	 * need locks to protect pending reads from racing with truncate.
2363	 */
2364	if (iocb->ki_flags & IOCB_DIRECT) {
 
 
 
2365		ret = ocfs2_rw_lock(inode, 0);
2366		if (ret < 0) {
2367			mlog_errno(ret);
2368			goto bail;
2369		}
2370		rw_level = 0;
2371		/* communicate with ocfs2_dio_end_io */
2372		ocfs2_iocb_set_rw_locked(iocb, rw_level);
2373	}
2374
2375	/*
2376	 * We're fine letting folks race truncates and extending
2377	 * writes with read across the cluster, just like they can
2378	 * locally. Hence no rw_lock during read.
2379	 *
2380	 * Take and drop the meta data lock to update inode fields
2381	 * like i_size. This allows the checks down below
2382	 * generic_file_aio_read() a chance of actually working.
2383	 */
2384	ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2385	if (ret < 0) {
2386		mlog_errno(ret);
2387		goto bail;
2388	}
2389	ocfs2_inode_unlock(inode, lock_level);
2390
2391	ret = generic_file_read_iter(iocb, to);
2392	trace_generic_file_aio_read_ret(ret);
2393
2394	/* buffered aio wouldn't have proper lock coverage today */
2395	BUG_ON(ret == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2396
2397	/* see ocfs2_file_write_iter */
2398	if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2399		rw_level = -1;
 
2400	}
2401
2402bail:
 
 
 
2403	if (rw_level != -1)
2404		ocfs2_rw_unlock(inode, rw_level);
2405
2406	return ret;
2407}
2408
2409/* Refer generic_file_llseek_unlocked() */
2410static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2411{
2412	struct inode *inode = file->f_mapping->host;
2413	int ret = 0;
2414
2415	inode_lock(inode);
2416
2417	switch (whence) {
2418	case SEEK_SET:
2419		break;
2420	case SEEK_END:
2421		/* SEEK_END requires the OCFS2 inode lock for the file
2422		 * because it references the file's size.
2423		 */
2424		ret = ocfs2_inode_lock(inode, NULL, 0);
2425		if (ret < 0) {
2426			mlog_errno(ret);
2427			goto out;
2428		}
2429		offset += i_size_read(inode);
2430		ocfs2_inode_unlock(inode, 0);
2431		break;
2432	case SEEK_CUR:
2433		if (offset == 0) {
2434			offset = file->f_pos;
2435			goto out;
2436		}
2437		offset += file->f_pos;
2438		break;
2439	case SEEK_DATA:
2440	case SEEK_HOLE:
2441		ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2442		if (ret)
2443			goto out;
2444		break;
2445	default:
2446		ret = -EINVAL;
2447		goto out;
2448	}
2449
2450	offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
 
 
 
 
 
 
 
 
 
 
2451
2452out:
2453	inode_unlock(inode);
2454	if (ret)
2455		return ret;
2456	return offset;
2457}
2458
2459const struct inode_operations ocfs2_file_iops = {
2460	.setattr	= ocfs2_setattr,
2461	.getattr	= ocfs2_getattr,
2462	.permission	= ocfs2_permission,
2463	.setxattr	= generic_setxattr,
2464	.getxattr	= generic_getxattr,
2465	.listxattr	= ocfs2_listxattr,
2466	.removexattr	= generic_removexattr,
2467	.fiemap		= ocfs2_fiemap,
2468	.get_acl	= ocfs2_iop_get_acl,
2469	.set_acl	= ocfs2_iop_set_acl,
2470};
2471
2472const struct inode_operations ocfs2_special_file_iops = {
2473	.setattr	= ocfs2_setattr,
2474	.getattr	= ocfs2_getattr,
2475	.permission	= ocfs2_permission,
2476	.get_acl	= ocfs2_iop_get_acl,
2477	.set_acl	= ocfs2_iop_set_acl,
2478};
2479
2480/*
2481 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2482 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2483 */
2484const struct file_operations ocfs2_fops = {
2485	.llseek		= ocfs2_file_llseek,
 
 
2486	.mmap		= ocfs2_mmap,
2487	.fsync		= ocfs2_sync_file,
2488	.release	= ocfs2_file_release,
2489	.open		= ocfs2_file_open,
2490	.read_iter	= ocfs2_file_read_iter,
2491	.write_iter	= ocfs2_file_write_iter,
2492	.unlocked_ioctl	= ocfs2_ioctl,
2493#ifdef CONFIG_COMPAT
2494	.compat_ioctl   = ocfs2_compat_ioctl,
2495#endif
2496	.lock		= ocfs2_lock,
2497	.flock		= ocfs2_flock,
2498	.splice_read	= ocfs2_file_splice_read,
2499	.splice_write	= iter_file_splice_write,
2500	.fallocate	= ocfs2_fallocate,
2501};
2502
2503const struct file_operations ocfs2_dops = {
2504	.llseek		= generic_file_llseek,
2505	.read		= generic_read_dir,
2506	.iterate	= ocfs2_readdir,
2507	.fsync		= ocfs2_sync_file,
2508	.release	= ocfs2_dir_release,
2509	.open		= ocfs2_dir_open,
2510	.unlocked_ioctl	= ocfs2_ioctl,
2511#ifdef CONFIG_COMPAT
2512	.compat_ioctl   = ocfs2_compat_ioctl,
2513#endif
2514	.lock		= ocfs2_lock,
2515	.flock		= ocfs2_flock,
2516};
2517
2518/*
2519 * POSIX-lockless variants of our file_operations.
2520 *
2521 * These will be used if the underlying cluster stack does not support
2522 * posix file locking, if the user passes the "localflocks" mount
2523 * option, or if we have a local-only fs.
2524 *
2525 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2526 * so we still want it in the case of no stack support for
2527 * plocks. Internally, it will do the right thing when asked to ignore
2528 * the cluster.
2529 */
2530const struct file_operations ocfs2_fops_no_plocks = {
2531	.llseek		= ocfs2_file_llseek,
 
 
2532	.mmap		= ocfs2_mmap,
2533	.fsync		= ocfs2_sync_file,
2534	.release	= ocfs2_file_release,
2535	.open		= ocfs2_file_open,
2536	.read_iter	= ocfs2_file_read_iter,
2537	.write_iter	= ocfs2_file_write_iter,
2538	.unlocked_ioctl	= ocfs2_ioctl,
2539#ifdef CONFIG_COMPAT
2540	.compat_ioctl   = ocfs2_compat_ioctl,
2541#endif
2542	.flock		= ocfs2_flock,
2543	.splice_read	= ocfs2_file_splice_read,
2544	.splice_write	= iter_file_splice_write,
2545	.fallocate	= ocfs2_fallocate,
2546};
2547
2548const struct file_operations ocfs2_dops_no_plocks = {
2549	.llseek		= generic_file_llseek,
2550	.read		= generic_read_dir,
2551	.iterate	= ocfs2_readdir,
2552	.fsync		= ocfs2_sync_file,
2553	.release	= ocfs2_dir_release,
2554	.open		= ocfs2_dir_open,
2555	.unlocked_ioctl	= ocfs2_ioctl,
2556#ifdef CONFIG_COMPAT
2557	.compat_ioctl   = ocfs2_compat_ioctl,
2558#endif
2559	.flock		= ocfs2_flock,
2560};