Linux Audio

Check our new training course

In-person Linux kernel drivers training

Jun 16-20, 2025
Register
Loading...
v3.1
   1/*
   2 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3 * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4 *
   5 * This copyrighted material is made available to anyone wishing to use,
   6 * modify, copy, or redistribute it subject to the terms and conditions
   7 * of the GNU General Public License version 2.
   8 */
   9
  10#include <linux/slab.h>
  11#include <linux/spinlock.h>
  12#include <linux/completion.h>
  13#include <linux/buffer_head.h>
  14#include <linux/pagemap.h>
  15#include <linux/uio.h>
  16#include <linux/blkdev.h>
  17#include <linux/mm.h>
  18#include <linux/mount.h>
  19#include <linux/fs.h>
  20#include <linux/gfs2_ondisk.h>
  21#include <linux/ext2_fs.h>
  22#include <linux/falloc.h>
  23#include <linux/swap.h>
  24#include <linux/crc32.h>
  25#include <linux/writeback.h>
  26#include <asm/uaccess.h>
  27#include <linux/dlm.h>
  28#include <linux/dlm_plock.h>
  29
  30#include "gfs2.h"
  31#include "incore.h"
  32#include "bmap.h"
  33#include "dir.h"
  34#include "glock.h"
  35#include "glops.h"
  36#include "inode.h"
  37#include "log.h"
  38#include "meta_io.h"
  39#include "quota.h"
  40#include "rgrp.h"
  41#include "trans.h"
  42#include "util.h"
  43
  44/**
  45 * gfs2_llseek - seek to a location in a file
  46 * @file: the file
  47 * @offset: the offset
  48 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
  49 *
  50 * SEEK_END requires the glock for the file because it references the
  51 * file's size.
  52 *
  53 * Returns: The new offset, or errno
  54 */
  55
  56static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
  57{
  58	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
  59	struct gfs2_holder i_gh;
  60	loff_t error;
  61
  62	if (origin == 2) {
 
 
 
  63		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
  64					   &i_gh);
  65		if (!error) {
  66			error = generic_file_llseek_unlocked(file, offset, origin);
  67			gfs2_glock_dq_uninit(&i_gh);
  68		}
  69	} else
  70		error = generic_file_llseek_unlocked(file, offset, origin);
 
 
 
 
 
 
  71
  72	return error;
  73}
  74
  75/**
  76 * gfs2_readdir - Read directory entries from a directory
  77 * @file: The directory to read from
  78 * @dirent: Buffer for dirents
  79 * @filldir: Function used to do the copying
  80 *
  81 * Returns: errno
  82 */
  83
  84static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
  85{
  86	struct inode *dir = file->f_mapping->host;
  87	struct gfs2_inode *dip = GFS2_I(dir);
  88	struct gfs2_holder d_gh;
  89	u64 offset = file->f_pos;
  90	int error;
  91
  92	gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
  93	error = gfs2_glock_nq(&d_gh);
  94	if (error) {
  95		gfs2_holder_uninit(&d_gh);
  96		return error;
  97	}
  98
  99	error = gfs2_dir_read(dir, &offset, dirent, filldir);
 100
 101	gfs2_glock_dq_uninit(&d_gh);
 102
 103	file->f_pos = offset;
 104
 105	return error;
 106}
 107
 108/**
 109 * fsflags_cvt
 110 * @table: A table of 32 u32 flags
 111 * @val: a 32 bit value to convert
 112 *
 113 * This function can be used to convert between fsflags values and
 114 * GFS2's own flags values.
 115 *
 116 * Returns: the converted flags
 117 */
 118static u32 fsflags_cvt(const u32 *table, u32 val)
 119{
 120	u32 res = 0;
 121	while(val) {
 122		if (val & 1)
 123			res |= *table;
 124		table++;
 125		val >>= 1;
 126	}
 127	return res;
 128}
 129
 130static const u32 fsflags_to_gfs2[32] = {
 131	[3] = GFS2_DIF_SYNC,
 132	[4] = GFS2_DIF_IMMUTABLE,
 133	[5] = GFS2_DIF_APPENDONLY,
 134	[7] = GFS2_DIF_NOATIME,
 135	[12] = GFS2_DIF_EXHASH,
 136	[14] = GFS2_DIF_INHERIT_JDATA,
 137};
 138
 139static const u32 gfs2_to_fsflags[32] = {
 140	[gfs2fl_Sync] = FS_SYNC_FL,
 141	[gfs2fl_Immutable] = FS_IMMUTABLE_FL,
 142	[gfs2fl_AppendOnly] = FS_APPEND_FL,
 143	[gfs2fl_NoAtime] = FS_NOATIME_FL,
 144	[gfs2fl_ExHash] = FS_INDEX_FL,
 145	[gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
 146};
 147
 148static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
 149{
 150	struct inode *inode = filp->f_path.dentry->d_inode;
 151	struct gfs2_inode *ip = GFS2_I(inode);
 152	struct gfs2_holder gh;
 153	int error;
 154	u32 fsflags;
 155
 156	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 157	error = gfs2_glock_nq(&gh);
 158	if (error)
 159		return error;
 160
 161	fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
 162	if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
 163		fsflags |= FS_JOURNAL_DATA_FL;
 164	if (put_user(fsflags, ptr))
 165		error = -EFAULT;
 166
 167	gfs2_glock_dq(&gh);
 168	gfs2_holder_uninit(&gh);
 169	return error;
 170}
 171
 172void gfs2_set_inode_flags(struct inode *inode)
 173{
 174	struct gfs2_inode *ip = GFS2_I(inode);
 175	unsigned int flags = inode->i_flags;
 176
 177	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
 178	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
 179		inode->i_flags |= S_NOSEC;
 180	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
 181		flags |= S_IMMUTABLE;
 182	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
 183		flags |= S_APPEND;
 184	if (ip->i_diskflags & GFS2_DIF_NOATIME)
 185		flags |= S_NOATIME;
 186	if (ip->i_diskflags & GFS2_DIF_SYNC)
 187		flags |= S_SYNC;
 188	inode->i_flags = flags;
 189}
 190
 191/* Flags that can be set by user space */
 192#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
 193			     GFS2_DIF_IMMUTABLE|		\
 194			     GFS2_DIF_APPENDONLY|		\
 195			     GFS2_DIF_NOATIME|			\
 196			     GFS2_DIF_SYNC|			\
 197			     GFS2_DIF_SYSTEM|			\
 198			     GFS2_DIF_INHERIT_JDATA)
 199
 200/**
 201 * gfs2_set_flags - set flags on an inode
 202 * @inode: The inode
 203 * @flags: The flags to set
 204 * @mask: Indicates which flags are valid
 205 *
 206 */
 207static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
 208{
 209	struct inode *inode = filp->f_path.dentry->d_inode;
 210	struct gfs2_inode *ip = GFS2_I(inode);
 211	struct gfs2_sbd *sdp = GFS2_SB(inode);
 212	struct buffer_head *bh;
 213	struct gfs2_holder gh;
 214	int error;
 215	u32 new_flags, flags;
 216
 217	error = mnt_want_write(filp->f_path.mnt);
 218	if (error)
 219		return error;
 220
 221	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 222	if (error)
 223		goto out_drop_write;
 224
 225	error = -EACCES;
 226	if (!inode_owner_or_capable(inode))
 227		goto out;
 228
 229	error = 0;
 230	flags = ip->i_diskflags;
 231	new_flags = (flags & ~mask) | (reqflags & mask);
 232	if ((new_flags ^ flags) == 0)
 233		goto out;
 234
 235	error = -EINVAL;
 236	if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
 237		goto out;
 238
 239	error = -EPERM;
 240	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
 241		goto out;
 242	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
 243		goto out;
 244	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
 245	    !capable(CAP_LINUX_IMMUTABLE))
 246		goto out;
 247	if (!IS_IMMUTABLE(inode)) {
 248		error = gfs2_permission(inode, MAY_WRITE);
 249		if (error)
 250			goto out;
 251	}
 252	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
 253		if (flags & GFS2_DIF_JDATA)
 254			gfs2_log_flush(sdp, ip->i_gl);
 255		error = filemap_fdatawrite(inode->i_mapping);
 256		if (error)
 257			goto out;
 258		error = filemap_fdatawait(inode->i_mapping);
 259		if (error)
 260			goto out;
 261	}
 262	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 263	if (error)
 264		goto out;
 265	error = gfs2_meta_inode_buffer(ip, &bh);
 266	if (error)
 267		goto out_trans_end;
 268	gfs2_trans_add_bh(ip->i_gl, bh, 1);
 269	ip->i_diskflags = new_flags;
 270	gfs2_dinode_out(ip, bh->b_data);
 271	brelse(bh);
 272	gfs2_set_inode_flags(inode);
 273	gfs2_set_aops(inode);
 274out_trans_end:
 275	gfs2_trans_end(sdp);
 276out:
 277	gfs2_glock_dq_uninit(&gh);
 278out_drop_write:
 279	mnt_drop_write(filp->f_path.mnt);
 280	return error;
 281}
 282
 283static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
 284{
 285	struct inode *inode = filp->f_path.dentry->d_inode;
 286	u32 fsflags, gfsflags;
 287
 288	if (get_user(fsflags, ptr))
 289		return -EFAULT;
 290
 291	gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
 292	if (!S_ISDIR(inode->i_mode)) {
 293		if (gfsflags & GFS2_DIF_INHERIT_JDATA)
 294			gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
 295		return do_gfs2_set_flags(filp, gfsflags, ~0);
 296	}
 297	return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
 298}
 299
 300static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 301{
 302	switch(cmd) {
 303	case FS_IOC_GETFLAGS:
 304		return gfs2_get_flags(filp, (u32 __user *)arg);
 305	case FS_IOC_SETFLAGS:
 306		return gfs2_set_flags(filp, (u32 __user *)arg);
 
 
 307	}
 308	return -ENOTTY;
 309}
 310
 311/**
 312 * gfs2_allocate_page_backing - Use bmap to allocate blocks
 313 * @page: The (locked) page to allocate backing for
 314 *
 315 * We try to allocate all the blocks required for the page in
 316 * one go. This might fail for various reasons, so we keep
 317 * trying until all the blocks to back this page are allocated.
 318 * If some of the blocks are already allocated, thats ok too.
 319 */
 320
 321static int gfs2_allocate_page_backing(struct page *page)
 322{
 323	struct inode *inode = page->mapping->host;
 324	struct buffer_head bh;
 325	unsigned long size = PAGE_CACHE_SIZE;
 326	u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 327
 328	do {
 329		bh.b_state = 0;
 330		bh.b_size = size;
 331		gfs2_block_map(inode, lblock, &bh, 1);
 332		if (!buffer_mapped(&bh))
 333			return -EIO;
 334		size -= bh.b_size;
 335		lblock += (bh.b_size >> inode->i_blkbits);
 336	} while(size > 0);
 337	return 0;
 338}
 339
 340/**
 341 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
 342 * @vma: The virtual memory area
 343 * @page: The page which is about to become writable
 344 *
 345 * When the page becomes writable, we need to ensure that we have
 346 * blocks allocated on disk to back that page.
 347 */
 348
 349static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 350{
 351	struct page *page = vmf->page;
 352	struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
 353	struct gfs2_inode *ip = GFS2_I(inode);
 354	struct gfs2_sbd *sdp = GFS2_SB(inode);
 355	unsigned long last_index;
 356	u64 pos = page->index << PAGE_CACHE_SHIFT;
 357	unsigned int data_blocks, ind_blocks, rblocks;
 358	struct gfs2_holder gh;
 359	struct gfs2_alloc *al;
 
 360	int ret;
 361
 
 
 
 
 
 
 362	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 363	ret = gfs2_glock_nq(&gh);
 364	if (ret)
 365		goto out;
 366
 367	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
 368	set_bit(GIF_SW_PAGED, &ip->i_flags);
 369
 370	if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE))
 
 
 
 
 
 371		goto out_unlock;
 
 
 372	ret = -ENOMEM;
 373	al = gfs2_alloc_get(ip);
 374	if (al == NULL)
 375		goto out_unlock;
 376
 377	ret = gfs2_quota_lock_check(ip);
 378	if (ret)
 379		goto out_alloc_put;
 380	gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
 381	al->al_requested = data_blocks + ind_blocks;
 382	ret = gfs2_inplace_reserve(ip);
 383	if (ret)
 384		goto out_quota_unlock;
 385
 386	rblocks = RES_DINODE + ind_blocks;
 387	if (gfs2_is_jdata(ip))
 388		rblocks += data_blocks ? data_blocks : 1;
 389	if (ind_blocks || data_blocks) {
 390		rblocks += RES_STATFS + RES_QUOTA;
 391		rblocks += gfs2_rg_blocks(al);
 392	}
 393	ret = gfs2_trans_begin(sdp, rblocks, 0);
 394	if (ret)
 395		goto out_trans_fail;
 396
 397	lock_page(page);
 398	ret = -EINVAL;
 399	last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT;
 400	if (page->index > last_index)
 401		goto out_unlock_page;
 
 
 
 
 
 
 
 
 
 
 
 402	ret = 0;
 403	if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping)
 404		goto out_unlock_page;
 405	if (gfs2_is_stuffed(ip)) {
 406		ret = gfs2_unstuff_dinode(ip, page);
 407		if (ret)
 408			goto out_unlock_page;
 409	}
 410	ret = gfs2_allocate_page_backing(page);
 411
 412out_unlock_page:
 413	unlock_page(page);
 
 414	gfs2_trans_end(sdp);
 415out_trans_fail:
 416	gfs2_inplace_release(ip);
 417out_quota_unlock:
 418	gfs2_quota_unlock(ip);
 419out_alloc_put:
 420	gfs2_alloc_put(ip);
 421out_unlock:
 422	gfs2_glock_dq(&gh);
 423out:
 424	gfs2_holder_uninit(&gh);
 425	if (ret == -ENOMEM)
 426		ret = VM_FAULT_OOM;
 427	else if (ret)
 428		ret = VM_FAULT_SIGBUS;
 429	return ret;
 
 
 
 
 
 
 430}
 431
 432static const struct vm_operations_struct gfs2_vm_ops = {
 433	.fault = filemap_fault,
 434	.page_mkwrite = gfs2_page_mkwrite,
 435};
 436
 437/**
 438 * gfs2_mmap -
 439 * @file: The file to map
 440 * @vma: The VMA which described the mapping
 441 *
 442 * There is no need to get a lock here unless we should be updating
 443 * atime. We ignore any locking errors since the only consequence is
 444 * a missed atime update (which will just be deferred until later).
 445 *
 446 * Returns: 0
 447 */
 448
 449static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 450{
 451	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 452
 453	if (!(file->f_flags & O_NOATIME) &&
 454	    !IS_NOATIME(&ip->i_inode)) {
 455		struct gfs2_holder i_gh;
 456		int error;
 457
 458		gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 459		error = gfs2_glock_nq(&i_gh);
 460		if (error == 0) {
 461			file_accessed(file);
 462			gfs2_glock_dq(&i_gh);
 463		}
 464		gfs2_holder_uninit(&i_gh);
 465		if (error)
 466			return error;
 467	}
 468	vma->vm_ops = &gfs2_vm_ops;
 469	vma->vm_flags |= VM_CAN_NONLINEAR;
 470
 471	return 0;
 472}
 473
 474/**
 475 * gfs2_open - open a file
 476 * @inode: the inode to open
 477 * @file: the struct file for this opening
 478 *
 479 * Returns: errno
 480 */
 481
 482static int gfs2_open(struct inode *inode, struct file *file)
 483{
 484	struct gfs2_inode *ip = GFS2_I(inode);
 485	struct gfs2_holder i_gh;
 486	struct gfs2_file *fp;
 487	int error;
 488
 489	fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
 490	if (!fp)
 491		return -ENOMEM;
 492
 493	mutex_init(&fp->f_fl_mutex);
 494
 495	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
 496	file->private_data = fp;
 497
 498	if (S_ISREG(ip->i_inode.i_mode)) {
 499		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 500					   &i_gh);
 501		if (error)
 502			goto fail;
 503
 504		if (!(file->f_flags & O_LARGEFILE) &&
 505		    i_size_read(inode) > MAX_NON_LFS) {
 506			error = -EOVERFLOW;
 507			goto fail_gunlock;
 508		}
 509
 510		gfs2_glock_dq_uninit(&i_gh);
 511	}
 512
 513	return 0;
 514
 515fail_gunlock:
 516	gfs2_glock_dq_uninit(&i_gh);
 517fail:
 518	file->private_data = NULL;
 519	kfree(fp);
 520	return error;
 521}
 522
 523/**
 524 * gfs2_close - called to close a struct file
 525 * @inode: the inode the struct file belongs to
 526 * @file: the struct file being closed
 527 *
 528 * Returns: errno
 529 */
 530
 531static int gfs2_close(struct inode *inode, struct file *file)
 532{
 533	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
 534	struct gfs2_file *fp;
 535
 536	fp = file->private_data;
 537	file->private_data = NULL;
 538
 539	if (gfs2_assert_warn(sdp, fp))
 540		return -EIO;
 541
 542	kfree(fp);
 543
 544	return 0;
 545}
 546
 547/**
 548 * gfs2_fsync - sync the dirty data for a file (across the cluster)
 549 * @file: the file that points to the dentry
 550 * @start: the start position in the file to sync
 551 * @end: the end position in the file to sync
 552 * @datasync: set if we can ignore timestamp changes
 553 *
 554 * The VFS will flush data for us. We only need to worry
 555 * about metadata here.
 
 
 
 
 
 
 
 
 556 *
 557 * Returns: errno
 558 */
 559
 560static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 561		      int datasync)
 562{
 563	struct inode *inode = file->f_mapping->host;
 
 564	int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
 565	struct gfs2_inode *ip = GFS2_I(inode);
 566	int ret;
 567
 568	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
 569	if (ret)
 570		return ret;
 571	mutex_lock(&inode->i_mutex);
 
 572
 573	if (datasync)
 574		sync_state &= ~I_DIRTY_SYNC;
 575
 576	if (sync_state) {
 577		ret = sync_inode_metadata(inode, 1);
 578		if (ret) {
 579			mutex_unlock(&inode->i_mutex);
 580			return ret;
 581		}
 582		gfs2_ail_flush(ip->i_gl);
 
 583	}
 584
 585	mutex_unlock(&inode->i_mutex);
 586	return 0;
 
 
 587}
 588
 589/**
 590 * gfs2_file_aio_write - Perform a write to a file
 591 * @iocb: The io context
 592 * @iov: The data to write
 593 * @nr_segs: Number of @iov segments
 594 * @pos: The file position
 595 *
 596 * We have to do a lock/unlock here to refresh the inode size for
 597 * O_APPEND writes, otherwise we can land up writing at the wrong
 598 * offset. There is still a race, but provided the app is using its
 599 * own file locking, this will make O_APPEND work as expected.
 600 *
 601 */
 602
 603static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
 604				   unsigned long nr_segs, loff_t pos)
 605{
 606	struct file *file = iocb->ki_filp;
 607
 608	if (file->f_flags & O_APPEND) {
 609		struct dentry *dentry = file->f_dentry;
 610		struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
 611		struct gfs2_holder gh;
 612		int ret;
 613
 614		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 615		if (ret)
 616			return ret;
 617		gfs2_glock_dq_uninit(&gh);
 618	}
 619
 620	return generic_file_aio_write(iocb, iov, nr_segs, pos);
 621}
 622
 623static int empty_write_end(struct page *page, unsigned from,
 624			   unsigned to, int mode)
 625{
 626	struct inode *inode = page->mapping->host;
 627	struct gfs2_inode *ip = GFS2_I(inode);
 628	struct buffer_head *bh;
 629	unsigned offset, blksize = 1 << inode->i_blkbits;
 630	pgoff_t end_index = i_size_read(inode) >> PAGE_CACHE_SHIFT;
 631
 632	zero_user(page, from, to-from);
 633	mark_page_accessed(page);
 634
 635	if (page->index < end_index || !(mode & FALLOC_FL_KEEP_SIZE)) {
 636		if (!gfs2_is_writeback(ip))
 637			gfs2_page_add_databufs(ip, page, from, to);
 638
 639		block_commit_write(page, from, to);
 640		return 0;
 641	}
 642
 643	offset = 0;
 644	bh = page_buffers(page);
 645	while (offset < to) {
 646		if (offset >= from) {
 647			set_buffer_uptodate(bh);
 648			mark_buffer_dirty(bh);
 649			clear_buffer_new(bh);
 650			write_dirty_buffer(bh, WRITE);
 651		}
 652		offset += blksize;
 653		bh = bh->b_this_page;
 654	}
 655
 656	offset = 0;
 657	bh = page_buffers(page);
 658	while (offset < to) {
 659		if (offset >= from) {
 660			wait_on_buffer(bh);
 661			if (!buffer_uptodate(bh))
 662				return -EIO;
 663		}
 664		offset += blksize;
 665		bh = bh->b_this_page;
 666	}
 667	return 0;
 668}
 669
 670static int needs_empty_write(sector_t block, struct inode *inode)
 671{
 672	int error;
 673	struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 674
 675	bh_map.b_size = 1 << inode->i_blkbits;
 676	error = gfs2_block_map(inode, block, &bh_map, 0);
 677	if (unlikely(error))
 678		return error;
 679	return !buffer_mapped(&bh_map);
 680}
 681
 682static int write_empty_blocks(struct page *page, unsigned from, unsigned to,
 683			      int mode)
 684{
 685	struct inode *inode = page->mapping->host;
 686	unsigned start, end, next, blksize;
 687	sector_t block = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 688	int ret;
 689
 690	blksize = 1 << inode->i_blkbits;
 691	next = end = 0;
 692	while (next < from) {
 693		next += blksize;
 694		block++;
 695	}
 696	start = next;
 697	do {
 698		next += blksize;
 699		ret = needs_empty_write(block, inode);
 700		if (unlikely(ret < 0))
 701			return ret;
 702		if (ret == 0) {
 703			if (end) {
 704				ret = __block_write_begin(page, start, end - start,
 705							  gfs2_block_map);
 706				if (unlikely(ret))
 707					return ret;
 708				ret = empty_write_end(page, start, end, mode);
 709				if (unlikely(ret))
 710					return ret;
 711				end = 0;
 712			}
 713			start = next;
 714		}
 715		else
 716			end = next;
 717		block++;
 718	} while (next < to);
 719
 720	if (end) {
 721		ret = __block_write_begin(page, start, end - start, gfs2_block_map);
 722		if (unlikely(ret))
 723			return ret;
 724		ret = empty_write_end(page, start, end, mode);
 725		if (unlikely(ret))
 726			return ret;
 727	}
 728
 729	return 0;
 730}
 731
 732static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
 733			   int mode)
 734{
 735	struct gfs2_inode *ip = GFS2_I(inode);
 736	struct buffer_head *dibh;
 737	int error;
 738	u64 start = offset >> PAGE_CACHE_SHIFT;
 739	unsigned int start_offset = offset & ~PAGE_CACHE_MASK;
 740	u64 end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
 741	pgoff_t curr;
 742	struct page *page;
 743	unsigned int end_offset = (offset + len) & ~PAGE_CACHE_MASK;
 744	unsigned int from, to;
 745
 746	if (!end_offset)
 747		end_offset = PAGE_CACHE_SIZE;
 748
 749	error = gfs2_meta_inode_buffer(ip, &dibh);
 750	if (unlikely(error))
 751		goto out;
 752
 753	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 754
 755	if (gfs2_is_stuffed(ip)) {
 756		error = gfs2_unstuff_dinode(ip, NULL);
 757		if (unlikely(error))
 758			goto out;
 759	}
 760
 761	curr = start;
 762	offset = start << PAGE_CACHE_SHIFT;
 763	from = start_offset;
 764	to = PAGE_CACHE_SIZE;
 765	while (curr <= end) {
 766		page = grab_cache_page_write_begin(inode->i_mapping, curr,
 767						   AOP_FLAG_NOFS);
 768		if (unlikely(!page)) {
 769			error = -ENOMEM;
 770			goto out;
 771		}
 772
 773		if (curr == end)
 774			to = end_offset;
 775		error = write_empty_blocks(page, from, to, mode);
 776		if (!error && offset + to > inode->i_size &&
 777		    !(mode & FALLOC_FL_KEEP_SIZE)) {
 778			i_size_write(inode, offset + to);
 779		}
 780		unlock_page(page);
 781		page_cache_release(page);
 782		if (error)
 783			goto out;
 784		curr++;
 785		offset += PAGE_CACHE_SIZE;
 786		from = 0;
 787	}
 
 
 788
 789	gfs2_dinode_out(ip, dibh->b_data);
 790	mark_inode_dirty(inode);
 791
 792	brelse(dibh);
 793
 794out:
 
 795	return error;
 796}
 797
 798static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len,
 799			    unsigned int *data_blocks, unsigned int *ind_blocks)
 800{
 801	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 802	unsigned int max_blocks = ip->i_alloc->al_rgd->rd_free_clone;
 803	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
 804
 805	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
 806		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
 807		max_data -= tmp;
 808	}
 809	/* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
 810	   so it might end up with fewer data blocks */
 811	if (max_data <= *data_blocks)
 812		return;
 813	*data_blocks = max_data;
 814	*ind_blocks = max_blocks - max_data;
 815	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
 816	if (*len > max) {
 817		*len = max;
 818		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
 819	}
 820}
 821
 822static long gfs2_fallocate(struct file *file, int mode, loff_t offset,
 823			   loff_t len)
 824{
 825	struct inode *inode = file->f_path.dentry->d_inode;
 826	struct gfs2_sbd *sdp = GFS2_SB(inode);
 827	struct gfs2_inode *ip = GFS2_I(inode);
 828	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
 829	loff_t bytes, max_bytes;
 830	struct gfs2_alloc *al;
 831	int error;
 
 
 832	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
 833	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
 
 834	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
 835
 836	/* We only support the FALLOC_FL_KEEP_SIZE mode */
 837	if (mode & ~FALLOC_FL_KEEP_SIZE)
 838		return -EOPNOTSUPP;
 839
 840	offset &= bsize_mask;
 841
 842	len = next - offset;
 843	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
 844	if (!bytes)
 845		bytes = UINT_MAX;
 846	bytes &= bsize_mask;
 847	if (bytes == 0)
 848		bytes = sdp->sd_sb.sb_bsize;
 849
 850	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
 851	error = gfs2_glock_nq(&ip->i_gh);
 852	if (unlikely(error))
 853		goto out_uninit;
 854
 855	if (!gfs2_write_alloc_required(ip, offset, len))
 856		goto out_unlock;
 857
 858	while (len > 0) {
 859		if (len < bytes)
 860			bytes = len;
 861		al = gfs2_alloc_get(ip);
 862		if (!al) {
 
 
 
 
 
 863			error = -ENOMEM;
 864			goto out_unlock;
 865		}
 866
 867		error = gfs2_quota_lock_check(ip);
 868		if (error)
 869			goto out_alloc_put;
 870
 871retry:
 872		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
 873
 874		al->al_requested = data_blocks + ind_blocks;
 875		error = gfs2_inplace_reserve(ip);
 876		if (error) {
 877			if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) {
 878				bytes >>= 1;
 879				bytes &= bsize_mask;
 880				if (bytes == 0)
 881					bytes = sdp->sd_sb.sb_bsize;
 882				goto retry;
 883			}
 884			goto out_qunlock;
 885		}
 886		max_bytes = bytes;
 887		calc_max_reserv(ip, len, &max_bytes, &data_blocks, &ind_blocks);
 888		al->al_requested = data_blocks + ind_blocks;
 889
 890		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
 891			  RES_RG_HDR + gfs2_rg_blocks(al);
 892		if (gfs2_is_jdata(ip))
 893			rblocks += data_blocks ? data_blocks : 1;
 894
 895		error = gfs2_trans_begin(sdp, rblocks,
 896					 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
 897		if (error)
 898			goto out_trans_fail;
 899
 900		error = fallocate_chunk(inode, offset, max_bytes, mode);
 901		gfs2_trans_end(sdp);
 902
 903		if (error)
 904			goto out_trans_fail;
 905
 906		len -= max_bytes;
 907		offset += max_bytes;
 908		gfs2_inplace_release(ip);
 909		gfs2_quota_unlock(ip);
 910		gfs2_alloc_put(ip);
 911	}
 
 
 
 912	goto out_unlock;
 913
 914out_trans_fail:
 915	gfs2_inplace_release(ip);
 916out_qunlock:
 917	gfs2_quota_unlock(ip);
 918out_alloc_put:
 919	gfs2_alloc_put(ip);
 920out_unlock:
 921	gfs2_glock_dq(&ip->i_gh);
 922out_uninit:
 923	gfs2_holder_uninit(&ip->i_gh);
 924	return error;
 925}
 926
 927#ifdef CONFIG_GFS2_FS_LOCKING_DLM
 928
 929/**
 930 * gfs2_setlease - acquire/release a file lease
 931 * @file: the file pointer
 932 * @arg: lease type
 933 * @fl: file lock
 934 *
 935 * We don't currently have a way to enforce a lease across the whole
 936 * cluster; until we do, disable leases (by just returning -EINVAL),
 937 * unless the administrator has requested purely local locking.
 938 *
 939 * Locking: called under lock_flocks
 940 *
 941 * Returns: errno
 942 */
 943
 944static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
 945{
 946	return -EINVAL;
 947}
 948
 949/**
 950 * gfs2_lock - acquire/release a posix lock on a file
 951 * @file: the file pointer
 952 * @cmd: either modify or retrieve lock state, possibly wait
 953 * @fl: type and range of lock
 954 *
 955 * Returns: errno
 956 */
 957
 958static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
 959{
 960	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 961	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
 962	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 963
 964	if (!(fl->fl_flags & FL_POSIX))
 965		return -ENOLCK;
 966	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
 967		return -ENOLCK;
 968
 969	if (cmd == F_CANCELLK) {
 970		/* Hack: */
 971		cmd = F_SETLK;
 972		fl->fl_type = F_UNLCK;
 973	}
 974	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 975		return -EIO;
 976	if (IS_GETLK(cmd))
 977		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
 978	else if (fl->fl_type == F_UNLCK)
 979		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
 980	else
 981		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
 982}
 983
 984static int do_flock(struct file *file, int cmd, struct file_lock *fl)
 985{
 986	struct gfs2_file *fp = file->private_data;
 987	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
 988	struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
 989	struct gfs2_glock *gl;
 990	unsigned int state;
 991	int flags;
 992	int error = 0;
 993
 994	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
 995	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
 996
 997	mutex_lock(&fp->f_fl_mutex);
 998
 999	gl = fl_gh->gh_gl;
1000	if (gl) {
1001		if (fl_gh->gh_state == state)
1002			goto out;
1003		flock_lock_file_wait(file,
1004				     &(struct file_lock){.fl_type = F_UNLCK});
1005		gfs2_glock_dq_wait(fl_gh);
1006		gfs2_holder_reinit(state, flags, fl_gh);
1007	} else {
1008		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1009				       &gfs2_flock_glops, CREATE, &gl);
1010		if (error)
1011			goto out;
1012		gfs2_holder_init(gl, state, flags, fl_gh);
1013		gfs2_glock_put(gl);
1014	}
1015	error = gfs2_glock_nq(fl_gh);
1016	if (error) {
1017		gfs2_holder_uninit(fl_gh);
1018		if (error == GLR_TRYFAILED)
1019			error = -EAGAIN;
1020	} else {
1021		error = flock_lock_file_wait(file, fl);
1022		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1023	}
1024
1025out:
1026	mutex_unlock(&fp->f_fl_mutex);
1027	return error;
1028}
1029
1030static void do_unflock(struct file *file, struct file_lock *fl)
1031{
1032	struct gfs2_file *fp = file->private_data;
1033	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1034
1035	mutex_lock(&fp->f_fl_mutex);
1036	flock_lock_file_wait(file, fl);
1037	if (fl_gh->gh_gl) {
1038		gfs2_glock_dq_wait(fl_gh);
1039		gfs2_holder_uninit(fl_gh);
1040	}
1041	mutex_unlock(&fp->f_fl_mutex);
1042}
1043
1044/**
1045 * gfs2_flock - acquire/release a flock lock on a file
1046 * @file: the file pointer
1047 * @cmd: either modify or retrieve lock state, possibly wait
1048 * @fl: type and range of lock
1049 *
1050 * Returns: errno
1051 */
1052
1053static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1054{
1055	if (!(fl->fl_flags & FL_FLOCK))
1056		return -ENOLCK;
1057	if (fl->fl_type & LOCK_MAND)
1058		return -EOPNOTSUPP;
1059
1060	if (fl->fl_type == F_UNLCK) {
1061		do_unflock(file, fl);
1062		return 0;
1063	} else {
1064		return do_flock(file, cmd, fl);
1065	}
1066}
1067
1068const struct file_operations gfs2_file_fops = {
1069	.llseek		= gfs2_llseek,
1070	.read		= do_sync_read,
1071	.aio_read	= generic_file_aio_read,
1072	.write		= do_sync_write,
1073	.aio_write	= gfs2_file_aio_write,
1074	.unlocked_ioctl	= gfs2_ioctl,
1075	.mmap		= gfs2_mmap,
1076	.open		= gfs2_open,
1077	.release	= gfs2_close,
1078	.fsync		= gfs2_fsync,
1079	.lock		= gfs2_lock,
1080	.flock		= gfs2_flock,
1081	.splice_read	= generic_file_splice_read,
1082	.splice_write	= generic_file_splice_write,
1083	.setlease	= gfs2_setlease,
1084	.fallocate	= gfs2_fallocate,
1085};
1086
1087const struct file_operations gfs2_dir_fops = {
1088	.readdir	= gfs2_readdir,
1089	.unlocked_ioctl	= gfs2_ioctl,
1090	.open		= gfs2_open,
1091	.release	= gfs2_close,
1092	.fsync		= gfs2_fsync,
1093	.lock		= gfs2_lock,
1094	.flock		= gfs2_flock,
1095	.llseek		= default_llseek,
1096};
1097
1098#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1099
1100const struct file_operations gfs2_file_fops_nolock = {
1101	.llseek		= gfs2_llseek,
1102	.read		= do_sync_read,
1103	.aio_read	= generic_file_aio_read,
1104	.write		= do_sync_write,
1105	.aio_write	= gfs2_file_aio_write,
1106	.unlocked_ioctl	= gfs2_ioctl,
1107	.mmap		= gfs2_mmap,
1108	.open		= gfs2_open,
1109	.release	= gfs2_close,
1110	.fsync		= gfs2_fsync,
1111	.splice_read	= generic_file_splice_read,
1112	.splice_write	= generic_file_splice_write,
1113	.setlease	= generic_setlease,
1114	.fallocate	= gfs2_fallocate,
1115};
1116
1117const struct file_operations gfs2_dir_fops_nolock = {
1118	.readdir	= gfs2_readdir,
1119	.unlocked_ioctl	= gfs2_ioctl,
1120	.open		= gfs2_open,
1121	.release	= gfs2_close,
1122	.fsync		= gfs2_fsync,
1123	.llseek		= default_llseek,
1124};
1125
v3.5.6
   1/*
   2 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3 * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4 *
   5 * This copyrighted material is made available to anyone wishing to use,
   6 * modify, copy, or redistribute it subject to the terms and conditions
   7 * of the GNU General Public License version 2.
   8 */
   9
  10#include <linux/slab.h>
  11#include <linux/spinlock.h>
  12#include <linux/completion.h>
  13#include <linux/buffer_head.h>
  14#include <linux/pagemap.h>
  15#include <linux/uio.h>
  16#include <linux/blkdev.h>
  17#include <linux/mm.h>
  18#include <linux/mount.h>
  19#include <linux/fs.h>
  20#include <linux/gfs2_ondisk.h>
 
  21#include <linux/falloc.h>
  22#include <linux/swap.h>
  23#include <linux/crc32.h>
  24#include <linux/writeback.h>
  25#include <asm/uaccess.h>
  26#include <linux/dlm.h>
  27#include <linux/dlm_plock.h>
  28
  29#include "gfs2.h"
  30#include "incore.h"
  31#include "bmap.h"
  32#include "dir.h"
  33#include "glock.h"
  34#include "glops.h"
  35#include "inode.h"
  36#include "log.h"
  37#include "meta_io.h"
  38#include "quota.h"
  39#include "rgrp.h"
  40#include "trans.h"
  41#include "util.h"
  42
  43/**
  44 * gfs2_llseek - seek to a location in a file
  45 * @file: the file
  46 * @offset: the offset
  47 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
  48 *
  49 * SEEK_END requires the glock for the file because it references the
  50 * file's size.
  51 *
  52 * Returns: The new offset, or errno
  53 */
  54
  55static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
  56{
  57	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
  58	struct gfs2_holder i_gh;
  59	loff_t error;
  60
  61	switch (origin) {
  62	case SEEK_END: /* These reference inode->i_size */
  63	case SEEK_DATA:
  64	case SEEK_HOLE:
  65		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
  66					   &i_gh);
  67		if (!error) {
  68			error = generic_file_llseek(file, offset, origin);
  69			gfs2_glock_dq_uninit(&i_gh);
  70		}
  71		break;
  72	case SEEK_CUR:
  73	case SEEK_SET:
  74		error = generic_file_llseek(file, offset, origin);
  75		break;
  76	default:
  77		error = -EINVAL;
  78	}
  79
  80	return error;
  81}
  82
  83/**
  84 * gfs2_readdir - Read directory entries from a directory
  85 * @file: The directory to read from
  86 * @dirent: Buffer for dirents
  87 * @filldir: Function used to do the copying
  88 *
  89 * Returns: errno
  90 */
  91
  92static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
  93{
  94	struct inode *dir = file->f_mapping->host;
  95	struct gfs2_inode *dip = GFS2_I(dir);
  96	struct gfs2_holder d_gh;
  97	u64 offset = file->f_pos;
  98	int error;
  99
 100	gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
 101	error = gfs2_glock_nq(&d_gh);
 102	if (error) {
 103		gfs2_holder_uninit(&d_gh);
 104		return error;
 105	}
 106
 107	error = gfs2_dir_read(dir, &offset, dirent, filldir, &file->f_ra);
 108
 109	gfs2_glock_dq_uninit(&d_gh);
 110
 111	file->f_pos = offset;
 112
 113	return error;
 114}
 115
 116/**
 117 * fsflags_cvt
 118 * @table: A table of 32 u32 flags
 119 * @val: a 32 bit value to convert
 120 *
 121 * This function can be used to convert between fsflags values and
 122 * GFS2's own flags values.
 123 *
 124 * Returns: the converted flags
 125 */
 126static u32 fsflags_cvt(const u32 *table, u32 val)
 127{
 128	u32 res = 0;
 129	while(val) {
 130		if (val & 1)
 131			res |= *table;
 132		table++;
 133		val >>= 1;
 134	}
 135	return res;
 136}
 137
 138static const u32 fsflags_to_gfs2[32] = {
 139	[3] = GFS2_DIF_SYNC,
 140	[4] = GFS2_DIF_IMMUTABLE,
 141	[5] = GFS2_DIF_APPENDONLY,
 142	[7] = GFS2_DIF_NOATIME,
 143	[12] = GFS2_DIF_EXHASH,
 144	[14] = GFS2_DIF_INHERIT_JDATA,
 145};
 146
 147static const u32 gfs2_to_fsflags[32] = {
 148	[gfs2fl_Sync] = FS_SYNC_FL,
 149	[gfs2fl_Immutable] = FS_IMMUTABLE_FL,
 150	[gfs2fl_AppendOnly] = FS_APPEND_FL,
 151	[gfs2fl_NoAtime] = FS_NOATIME_FL,
 152	[gfs2fl_ExHash] = FS_INDEX_FL,
 153	[gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
 154};
 155
 156static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
 157{
 158	struct inode *inode = filp->f_path.dentry->d_inode;
 159	struct gfs2_inode *ip = GFS2_I(inode);
 160	struct gfs2_holder gh;
 161	int error;
 162	u32 fsflags;
 163
 164	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 165	error = gfs2_glock_nq(&gh);
 166	if (error)
 167		return error;
 168
 169	fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
 170	if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
 171		fsflags |= FS_JOURNAL_DATA_FL;
 172	if (put_user(fsflags, ptr))
 173		error = -EFAULT;
 174
 175	gfs2_glock_dq(&gh);
 176	gfs2_holder_uninit(&gh);
 177	return error;
 178}
 179
 180void gfs2_set_inode_flags(struct inode *inode)
 181{
 182	struct gfs2_inode *ip = GFS2_I(inode);
 183	unsigned int flags = inode->i_flags;
 184
 185	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
 186	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
 187		inode->i_flags |= S_NOSEC;
 188	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
 189		flags |= S_IMMUTABLE;
 190	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
 191		flags |= S_APPEND;
 192	if (ip->i_diskflags & GFS2_DIF_NOATIME)
 193		flags |= S_NOATIME;
 194	if (ip->i_diskflags & GFS2_DIF_SYNC)
 195		flags |= S_SYNC;
 196	inode->i_flags = flags;
 197}
 198
 199/* Flags that can be set by user space */
 200#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
 201			     GFS2_DIF_IMMUTABLE|		\
 202			     GFS2_DIF_APPENDONLY|		\
 203			     GFS2_DIF_NOATIME|			\
 204			     GFS2_DIF_SYNC|			\
 205			     GFS2_DIF_SYSTEM|			\
 206			     GFS2_DIF_INHERIT_JDATA)
 207
 208/**
 209 * gfs2_set_flags - set flags on an inode
 210 * @inode: The inode
 211 * @flags: The flags to set
 212 * @mask: Indicates which flags are valid
 213 *
 214 */
 215static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
 216{
 217	struct inode *inode = filp->f_path.dentry->d_inode;
 218	struct gfs2_inode *ip = GFS2_I(inode);
 219	struct gfs2_sbd *sdp = GFS2_SB(inode);
 220	struct buffer_head *bh;
 221	struct gfs2_holder gh;
 222	int error;
 223	u32 new_flags, flags;
 224
 225	error = mnt_want_write_file(filp);
 226	if (error)
 227		return error;
 228
 229	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 230	if (error)
 231		goto out_drop_write;
 232
 233	error = -EACCES;
 234	if (!inode_owner_or_capable(inode))
 235		goto out;
 236
 237	error = 0;
 238	flags = ip->i_diskflags;
 239	new_flags = (flags & ~mask) | (reqflags & mask);
 240	if ((new_flags ^ flags) == 0)
 241		goto out;
 242
 243	error = -EINVAL;
 244	if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
 245		goto out;
 246
 247	error = -EPERM;
 248	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
 249		goto out;
 250	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
 251		goto out;
 252	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
 253	    !capable(CAP_LINUX_IMMUTABLE))
 254		goto out;
 255	if (!IS_IMMUTABLE(inode)) {
 256		error = gfs2_permission(inode, MAY_WRITE);
 257		if (error)
 258			goto out;
 259	}
 260	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
 261		if (flags & GFS2_DIF_JDATA)
 262			gfs2_log_flush(sdp, ip->i_gl);
 263		error = filemap_fdatawrite(inode->i_mapping);
 264		if (error)
 265			goto out;
 266		error = filemap_fdatawait(inode->i_mapping);
 267		if (error)
 268			goto out;
 269	}
 270	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 271	if (error)
 272		goto out;
 273	error = gfs2_meta_inode_buffer(ip, &bh);
 274	if (error)
 275		goto out_trans_end;
 276	gfs2_trans_add_bh(ip->i_gl, bh, 1);
 277	ip->i_diskflags = new_flags;
 278	gfs2_dinode_out(ip, bh->b_data);
 279	brelse(bh);
 280	gfs2_set_inode_flags(inode);
 281	gfs2_set_aops(inode);
 282out_trans_end:
 283	gfs2_trans_end(sdp);
 284out:
 285	gfs2_glock_dq_uninit(&gh);
 286out_drop_write:
 287	mnt_drop_write_file(filp);
 288	return error;
 289}
 290
 291static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
 292{
 293	struct inode *inode = filp->f_path.dentry->d_inode;
 294	u32 fsflags, gfsflags;
 295
 296	if (get_user(fsflags, ptr))
 297		return -EFAULT;
 298
 299	gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
 300	if (!S_ISDIR(inode->i_mode)) {
 301		if (gfsflags & GFS2_DIF_INHERIT_JDATA)
 302			gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
 303		return do_gfs2_set_flags(filp, gfsflags, ~0);
 304	}
 305	return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
 306}
 307
 308static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 309{
 310	switch(cmd) {
 311	case FS_IOC_GETFLAGS:
 312		return gfs2_get_flags(filp, (u32 __user *)arg);
 313	case FS_IOC_SETFLAGS:
 314		return gfs2_set_flags(filp, (u32 __user *)arg);
 315	case FITRIM:
 316		return gfs2_fitrim(filp, (void __user *)arg);
 317	}
 318	return -ENOTTY;
 319}
 320
 321/**
 322 * gfs2_allocate_page_backing - Use bmap to allocate blocks
 323 * @page: The (locked) page to allocate backing for
 324 *
 325 * We try to allocate all the blocks required for the page in
 326 * one go. This might fail for various reasons, so we keep
 327 * trying until all the blocks to back this page are allocated.
 328 * If some of the blocks are already allocated, thats ok too.
 329 */
 330
 331static int gfs2_allocate_page_backing(struct page *page)
 332{
 333	struct inode *inode = page->mapping->host;
 334	struct buffer_head bh;
 335	unsigned long size = PAGE_CACHE_SIZE;
 336	u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 337
 338	do {
 339		bh.b_state = 0;
 340		bh.b_size = size;
 341		gfs2_block_map(inode, lblock, &bh, 1);
 342		if (!buffer_mapped(&bh))
 343			return -EIO;
 344		size -= bh.b_size;
 345		lblock += (bh.b_size >> inode->i_blkbits);
 346	} while(size > 0);
 347	return 0;
 348}
 349
 350/**
 351 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
 352 * @vma: The virtual memory area
 353 * @page: The page which is about to become writable
 354 *
 355 * When the page becomes writable, we need to ensure that we have
 356 * blocks allocated on disk to back that page.
 357 */
 358
 359static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 360{
 361	struct page *page = vmf->page;
 362	struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
 363	struct gfs2_inode *ip = GFS2_I(inode);
 364	struct gfs2_sbd *sdp = GFS2_SB(inode);
 365	unsigned long last_index;
 366	u64 pos = page->index << PAGE_CACHE_SHIFT;
 367	unsigned int data_blocks, ind_blocks, rblocks;
 368	struct gfs2_holder gh;
 369	struct gfs2_qadata *qa;
 370	loff_t size;
 371	int ret;
 372
 373	/* Wait if fs is frozen. This is racy so we check again later on
 374	 * and retry if the fs has been frozen after the page lock has
 375	 * been acquired
 376	 */
 377	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 378
 379	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 380	ret = gfs2_glock_nq(&gh);
 381	if (ret)
 382		goto out;
 383
 384	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
 385	set_bit(GIF_SW_PAGED, &ip->i_flags);
 386
 387	if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) {
 388		lock_page(page);
 389		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
 390			ret = -EAGAIN;
 391			unlock_page(page);
 392		}
 393		goto out_unlock;
 394	}
 395
 396	ret = -ENOMEM;
 397	qa = gfs2_qadata_get(ip);
 398	if (qa == NULL)
 399		goto out_unlock;
 400
 401	ret = gfs2_quota_lock_check(ip);
 402	if (ret)
 403		goto out_alloc_put;
 404	gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
 405	ret = gfs2_inplace_reserve(ip, data_blocks + ind_blocks);
 
 406	if (ret)
 407		goto out_quota_unlock;
 408
 409	rblocks = RES_DINODE + ind_blocks;
 410	if (gfs2_is_jdata(ip))
 411		rblocks += data_blocks ? data_blocks : 1;
 412	if (ind_blocks || data_blocks) {
 413		rblocks += RES_STATFS + RES_QUOTA;
 414		rblocks += gfs2_rg_blocks(ip);
 415	}
 416	ret = gfs2_trans_begin(sdp, rblocks, 0);
 417	if (ret)
 418		goto out_trans_fail;
 419
 420	lock_page(page);
 421	ret = -EINVAL;
 422	size = i_size_read(inode);
 423	last_index = (size - 1) >> PAGE_CACHE_SHIFT;
 424	/* Check page index against inode size */
 425	if (size == 0 || (page->index > last_index))
 426		goto out_trans_end;
 427
 428	ret = -EAGAIN;
 429	/* If truncated, we must retry the operation, we may have raced
 430	 * with the glock demotion code.
 431	 */
 432	if (!PageUptodate(page) || page->mapping != inode->i_mapping)
 433		goto out_trans_end;
 434
 435	/* Unstuff, if required, and allocate backing blocks for page */
 436	ret = 0;
 437	if (gfs2_is_stuffed(ip))
 
 
 438		ret = gfs2_unstuff_dinode(ip, page);
 439	if (ret == 0)
 440		ret = gfs2_allocate_page_backing(page);
 
 
 441
 442out_trans_end:
 443	if (ret)
 444		unlock_page(page);
 445	gfs2_trans_end(sdp);
 446out_trans_fail:
 447	gfs2_inplace_release(ip);
 448out_quota_unlock:
 449	gfs2_quota_unlock(ip);
 450out_alloc_put:
 451	gfs2_qadata_put(ip);
 452out_unlock:
 453	gfs2_glock_dq(&gh);
 454out:
 455	gfs2_holder_uninit(&gh);
 456	if (ret == 0) {
 457		set_page_dirty(page);
 458		/* This check must be post dropping of transaction lock */
 459		if (inode->i_sb->s_frozen == SB_UNFROZEN) {
 460			wait_on_page_writeback(page);
 461		} else {
 462			ret = -EAGAIN;
 463			unlock_page(page);
 464		}
 465	}
 466	return block_page_mkwrite_return(ret);
 467}
 468
 469static const struct vm_operations_struct gfs2_vm_ops = {
 470	.fault = filemap_fault,
 471	.page_mkwrite = gfs2_page_mkwrite,
 472};
 473
 474/**
 475 * gfs2_mmap -
 476 * @file: The file to map
 477 * @vma: The VMA which described the mapping
 478 *
 479 * There is no need to get a lock here unless we should be updating
 480 * atime. We ignore any locking errors since the only consequence is
 481 * a missed atime update (which will just be deferred until later).
 482 *
 483 * Returns: 0
 484 */
 485
 486static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 487{
 488	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 489
 490	if (!(file->f_flags & O_NOATIME) &&
 491	    !IS_NOATIME(&ip->i_inode)) {
 492		struct gfs2_holder i_gh;
 493		int error;
 494
 495		gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 496		error = gfs2_glock_nq(&i_gh);
 497		if (error == 0) {
 498			file_accessed(file);
 499			gfs2_glock_dq(&i_gh);
 500		}
 501		gfs2_holder_uninit(&i_gh);
 502		if (error)
 503			return error;
 504	}
 505	vma->vm_ops = &gfs2_vm_ops;
 506	vma->vm_flags |= VM_CAN_NONLINEAR;
 507
 508	return 0;
 509}
 510
 511/**
 512 * gfs2_open - open a file
 513 * @inode: the inode to open
 514 * @file: the struct file for this opening
 515 *
 516 * Returns: errno
 517 */
 518
 519static int gfs2_open(struct inode *inode, struct file *file)
 520{
 521	struct gfs2_inode *ip = GFS2_I(inode);
 522	struct gfs2_holder i_gh;
 523	struct gfs2_file *fp;
 524	int error;
 525
 526	fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
 527	if (!fp)
 528		return -ENOMEM;
 529
 530	mutex_init(&fp->f_fl_mutex);
 531
 532	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
 533	file->private_data = fp;
 534
 535	if (S_ISREG(ip->i_inode.i_mode)) {
 536		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 537					   &i_gh);
 538		if (error)
 539			goto fail;
 540
 541		if (!(file->f_flags & O_LARGEFILE) &&
 542		    i_size_read(inode) > MAX_NON_LFS) {
 543			error = -EOVERFLOW;
 544			goto fail_gunlock;
 545		}
 546
 547		gfs2_glock_dq_uninit(&i_gh);
 548	}
 549
 550	return 0;
 551
 552fail_gunlock:
 553	gfs2_glock_dq_uninit(&i_gh);
 554fail:
 555	file->private_data = NULL;
 556	kfree(fp);
 557	return error;
 558}
 559
 560/**
 561 * gfs2_release - called to close a struct file
 562 * @inode: the inode the struct file belongs to
 563 * @file: the struct file being closed
 564 *
 565 * Returns: errno
 566 */
 567
 568static int gfs2_release(struct inode *inode, struct file *file)
 569{
 570	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
 571	struct gfs2_file *fp;
 572
 573	fp = file->private_data;
 574	file->private_data = NULL;
 575
 576	if (gfs2_assert_warn(sdp, fp))
 577		return -EIO;
 578
 579	kfree(fp);
 580
 581	return 0;
 582}
 583
 584/**
 585 * gfs2_fsync - sync the dirty data for a file (across the cluster)
 586 * @file: the file that points to the dentry
 587 * @start: the start position in the file to sync
 588 * @end: the end position in the file to sync
 589 * @datasync: set if we can ignore timestamp changes
 590 *
 591 * We split the data flushing here so that we don't wait for the data
 592 * until after we've also sent the metadata to disk. Note that for
 593 * data=ordered, we will write & wait for the data at the log flush
 594 * stage anyway, so this is unlikely to make much of a difference
 595 * except in the data=writeback case.
 596 *
 597 * If the fdatawrite fails due to any reason except -EIO, we will
 598 * continue the remainder of the fsync, although we'll still report
 599 * the error at the end. This is to match filemap_write_and_wait_range()
 600 * behaviour.
 601 *
 602 * Returns: errno
 603 */
 604
 605static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 606		      int datasync)
 607{
 608	struct address_space *mapping = file->f_mapping;
 609	struct inode *inode = mapping->host;
 610	int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
 611	struct gfs2_inode *ip = GFS2_I(inode);
 612	int ret = 0, ret1 = 0;
 613
 614	if (mapping->nrpages) {
 615		ret1 = filemap_fdatawrite_range(mapping, start, end);
 616		if (ret1 == -EIO)
 617			return ret1;
 618	}
 619
 620	if (datasync)
 621		sync_state &= ~I_DIRTY_SYNC;
 622
 623	if (sync_state) {
 624		ret = sync_inode_metadata(inode, 1);
 625		if (ret)
 
 626			return ret;
 627		if (gfs2_is_jdata(ip))
 628			filemap_write_and_wait(mapping);
 629		gfs2_ail_flush(ip->i_gl, 1);
 630	}
 631
 632	if (mapping->nrpages)
 633		ret = filemap_fdatawait_range(mapping, start, end);
 634
 635	return ret ? ret : ret1;
 636}
 637
 638/**
 639 * gfs2_file_aio_write - Perform a write to a file
 640 * @iocb: The io context
 641 * @iov: The data to write
 642 * @nr_segs: Number of @iov segments
 643 * @pos: The file position
 644 *
 645 * We have to do a lock/unlock here to refresh the inode size for
 646 * O_APPEND writes, otherwise we can land up writing at the wrong
 647 * offset. There is still a race, but provided the app is using its
 648 * own file locking, this will make O_APPEND work as expected.
 649 *
 650 */
 651
 652static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
 653				   unsigned long nr_segs, loff_t pos)
 654{
 655	struct file *file = iocb->ki_filp;
 656
 657	if (file->f_flags & O_APPEND) {
 658		struct dentry *dentry = file->f_dentry;
 659		struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
 660		struct gfs2_holder gh;
 661		int ret;
 662
 663		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 664		if (ret)
 665			return ret;
 666		gfs2_glock_dq_uninit(&gh);
 667	}
 668
 669	return generic_file_aio_write(iocb, iov, nr_segs, pos);
 670}
 671
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 672static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
 673			   int mode)
 674{
 675	struct gfs2_inode *ip = GFS2_I(inode);
 676	struct buffer_head *dibh;
 677	int error;
 678	loff_t size = len;
 679	unsigned int nr_blks;
 680	sector_t lblock = offset >> inode->i_blkbits;
 
 
 
 
 
 
 
 681
 682	error = gfs2_meta_inode_buffer(ip, &dibh);
 683	if (unlikely(error))
 684		return error;
 685
 686	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 687
 688	if (gfs2_is_stuffed(ip)) {
 689		error = gfs2_unstuff_dinode(ip, NULL);
 690		if (unlikely(error))
 691			goto out;
 692	}
 693
 694	while (len) {
 695		struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 696		bh_map.b_size = len;
 697		set_buffer_zeronew(&bh_map);
 
 
 
 
 
 
 
 698
 699		error = gfs2_block_map(inode, lblock, &bh_map, 1);
 700		if (unlikely(error))
 701			goto out;
 702		len -= bh_map.b_size;
 703		nr_blks = bh_map.b_size >> inode->i_blkbits;
 704		lblock += nr_blks;
 705		if (!buffer_new(&bh_map))
 706			continue;
 707		if (unlikely(!buffer_zeronew(&bh_map))) {
 708			error = -EIO;
 709			goto out;
 710		}
 
 
 711	}
 712	if (offset + size > inode->i_size && !(mode & FALLOC_FL_KEEP_SIZE))
 713		i_size_write(inode, offset + size);
 714
 
 715	mark_inode_dirty(inode);
 716
 
 
 717out:
 718	brelse(dibh);
 719	return error;
 720}
 721
 722static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len,
 723			    unsigned int *data_blocks, unsigned int *ind_blocks)
 724{
 725	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 726	unsigned int max_blocks = ip->i_rgd->rd_free_clone;
 727	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
 728
 729	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
 730		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
 731		max_data -= tmp;
 732	}
 733	/* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
 734	   so it might end up with fewer data blocks */
 735	if (max_data <= *data_blocks)
 736		return;
 737	*data_blocks = max_data;
 738	*ind_blocks = max_blocks - max_data;
 739	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
 740	if (*len > max) {
 741		*len = max;
 742		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
 743	}
 744}
 745
 746static long gfs2_fallocate(struct file *file, int mode, loff_t offset,
 747			   loff_t len)
 748{
 749	struct inode *inode = file->f_path.dentry->d_inode;
 750	struct gfs2_sbd *sdp = GFS2_SB(inode);
 751	struct gfs2_inode *ip = GFS2_I(inode);
 752	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
 753	loff_t bytes, max_bytes;
 754	struct gfs2_qadata *qa;
 755	int error;
 756	const loff_t pos = offset;
 757	const loff_t count = len;
 758	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
 759	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
 760	loff_t max_chunk_size = UINT_MAX & bsize_mask;
 761	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
 762
 763	/* We only support the FALLOC_FL_KEEP_SIZE mode */
 764	if (mode & ~FALLOC_FL_KEEP_SIZE)
 765		return -EOPNOTSUPP;
 766
 767	offset &= bsize_mask;
 768
 769	len = next - offset;
 770	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
 771	if (!bytes)
 772		bytes = UINT_MAX;
 773	bytes &= bsize_mask;
 774	if (bytes == 0)
 775		bytes = sdp->sd_sb.sb_bsize;
 776
 777	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
 778	error = gfs2_glock_nq(&ip->i_gh);
 779	if (unlikely(error))
 780		goto out_uninit;
 781
 
 
 
 782	while (len > 0) {
 783		if (len < bytes)
 784			bytes = len;
 785		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
 786			len -= bytes;
 787			offset += bytes;
 788			continue;
 789		}
 790		qa = gfs2_qadata_get(ip);
 791		if (!qa) {
 792			error = -ENOMEM;
 793			goto out_unlock;
 794		}
 795
 796		error = gfs2_quota_lock_check(ip);
 797		if (error)
 798			goto out_alloc_put;
 799
 800retry:
 801		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
 802
 803		error = gfs2_inplace_reserve(ip, data_blocks + ind_blocks);
 
 804		if (error) {
 805			if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) {
 806				bytes >>= 1;
 807				bytes &= bsize_mask;
 808				if (bytes == 0)
 809					bytes = sdp->sd_sb.sb_bsize;
 810				goto retry;
 811			}
 812			goto out_qunlock;
 813		}
 814		max_bytes = bytes;
 815		calc_max_reserv(ip, (len > max_chunk_size)? max_chunk_size: len,
 816				&max_bytes, &data_blocks, &ind_blocks);
 817
 818		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
 819			  RES_RG_HDR + gfs2_rg_blocks(ip);
 820		if (gfs2_is_jdata(ip))
 821			rblocks += data_blocks ? data_blocks : 1;
 822
 823		error = gfs2_trans_begin(sdp, rblocks,
 824					 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
 825		if (error)
 826			goto out_trans_fail;
 827
 828		error = fallocate_chunk(inode, offset, max_bytes, mode);
 829		gfs2_trans_end(sdp);
 830
 831		if (error)
 832			goto out_trans_fail;
 833
 834		len -= max_bytes;
 835		offset += max_bytes;
 836		gfs2_inplace_release(ip);
 837		gfs2_quota_unlock(ip);
 838		gfs2_qadata_put(ip);
 839	}
 840
 841	if (error == 0)
 842		error = generic_write_sync(file, pos, count);
 843	goto out_unlock;
 844
 845out_trans_fail:
 846	gfs2_inplace_release(ip);
 847out_qunlock:
 848	gfs2_quota_unlock(ip);
 849out_alloc_put:
 850	gfs2_qadata_put(ip);
 851out_unlock:
 852	gfs2_glock_dq(&ip->i_gh);
 853out_uninit:
 854	gfs2_holder_uninit(&ip->i_gh);
 855	return error;
 856}
 857
 858#ifdef CONFIG_GFS2_FS_LOCKING_DLM
 859
 860/**
 861 * gfs2_setlease - acquire/release a file lease
 862 * @file: the file pointer
 863 * @arg: lease type
 864 * @fl: file lock
 865 *
 866 * We don't currently have a way to enforce a lease across the whole
 867 * cluster; until we do, disable leases (by just returning -EINVAL),
 868 * unless the administrator has requested purely local locking.
 869 *
 870 * Locking: called under lock_flocks
 871 *
 872 * Returns: errno
 873 */
 874
 875static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
 876{
 877	return -EINVAL;
 878}
 879
 880/**
 881 * gfs2_lock - acquire/release a posix lock on a file
 882 * @file: the file pointer
 883 * @cmd: either modify or retrieve lock state, possibly wait
 884 * @fl: type and range of lock
 885 *
 886 * Returns: errno
 887 */
 888
 889static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
 890{
 891	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 892	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
 893	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 894
 895	if (!(fl->fl_flags & FL_POSIX))
 896		return -ENOLCK;
 897	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
 898		return -ENOLCK;
 899
 900	if (cmd == F_CANCELLK) {
 901		/* Hack: */
 902		cmd = F_SETLK;
 903		fl->fl_type = F_UNLCK;
 904	}
 905	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 906		return -EIO;
 907	if (IS_GETLK(cmd))
 908		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
 909	else if (fl->fl_type == F_UNLCK)
 910		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
 911	else
 912		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
 913}
 914
 915static int do_flock(struct file *file, int cmd, struct file_lock *fl)
 916{
 917	struct gfs2_file *fp = file->private_data;
 918	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
 919	struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
 920	struct gfs2_glock *gl;
 921	unsigned int state;
 922	int flags;
 923	int error = 0;
 924
 925	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
 926	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
 927
 928	mutex_lock(&fp->f_fl_mutex);
 929
 930	gl = fl_gh->gh_gl;
 931	if (gl) {
 932		if (fl_gh->gh_state == state)
 933			goto out;
 934		flock_lock_file_wait(file,
 935				     &(struct file_lock){.fl_type = F_UNLCK});
 936		gfs2_glock_dq_wait(fl_gh);
 937		gfs2_holder_reinit(state, flags, fl_gh);
 938	} else {
 939		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
 940				       &gfs2_flock_glops, CREATE, &gl);
 941		if (error)
 942			goto out;
 943		gfs2_holder_init(gl, state, flags, fl_gh);
 944		gfs2_glock_put(gl);
 945	}
 946	error = gfs2_glock_nq(fl_gh);
 947	if (error) {
 948		gfs2_holder_uninit(fl_gh);
 949		if (error == GLR_TRYFAILED)
 950			error = -EAGAIN;
 951	} else {
 952		error = flock_lock_file_wait(file, fl);
 953		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
 954	}
 955
 956out:
 957	mutex_unlock(&fp->f_fl_mutex);
 958	return error;
 959}
 960
 961static void do_unflock(struct file *file, struct file_lock *fl)
 962{
 963	struct gfs2_file *fp = file->private_data;
 964	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
 965
 966	mutex_lock(&fp->f_fl_mutex);
 967	flock_lock_file_wait(file, fl);
 968	if (fl_gh->gh_gl) {
 969		gfs2_glock_dq_wait(fl_gh);
 970		gfs2_holder_uninit(fl_gh);
 971	}
 972	mutex_unlock(&fp->f_fl_mutex);
 973}
 974
 975/**
 976 * gfs2_flock - acquire/release a flock lock on a file
 977 * @file: the file pointer
 978 * @cmd: either modify or retrieve lock state, possibly wait
 979 * @fl: type and range of lock
 980 *
 981 * Returns: errno
 982 */
 983
 984static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
 985{
 986	if (!(fl->fl_flags & FL_FLOCK))
 987		return -ENOLCK;
 988	if (fl->fl_type & LOCK_MAND)
 989		return -EOPNOTSUPP;
 990
 991	if (fl->fl_type == F_UNLCK) {
 992		do_unflock(file, fl);
 993		return 0;
 994	} else {
 995		return do_flock(file, cmd, fl);
 996	}
 997}
 998
 999const struct file_operations gfs2_file_fops = {
1000	.llseek		= gfs2_llseek,
1001	.read		= do_sync_read,
1002	.aio_read	= generic_file_aio_read,
1003	.write		= do_sync_write,
1004	.aio_write	= gfs2_file_aio_write,
1005	.unlocked_ioctl	= gfs2_ioctl,
1006	.mmap		= gfs2_mmap,
1007	.open		= gfs2_open,
1008	.release	= gfs2_release,
1009	.fsync		= gfs2_fsync,
1010	.lock		= gfs2_lock,
1011	.flock		= gfs2_flock,
1012	.splice_read	= generic_file_splice_read,
1013	.splice_write	= generic_file_splice_write,
1014	.setlease	= gfs2_setlease,
1015	.fallocate	= gfs2_fallocate,
1016};
1017
1018const struct file_operations gfs2_dir_fops = {
1019	.readdir	= gfs2_readdir,
1020	.unlocked_ioctl	= gfs2_ioctl,
1021	.open		= gfs2_open,
1022	.release	= gfs2_release,
1023	.fsync		= gfs2_fsync,
1024	.lock		= gfs2_lock,
1025	.flock		= gfs2_flock,
1026	.llseek		= default_llseek,
1027};
1028
1029#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1030
1031const struct file_operations gfs2_file_fops_nolock = {
1032	.llseek		= gfs2_llseek,
1033	.read		= do_sync_read,
1034	.aio_read	= generic_file_aio_read,
1035	.write		= do_sync_write,
1036	.aio_write	= gfs2_file_aio_write,
1037	.unlocked_ioctl	= gfs2_ioctl,
1038	.mmap		= gfs2_mmap,
1039	.open		= gfs2_open,
1040	.release	= gfs2_release,
1041	.fsync		= gfs2_fsync,
1042	.splice_read	= generic_file_splice_read,
1043	.splice_write	= generic_file_splice_write,
1044	.setlease	= generic_setlease,
1045	.fallocate	= gfs2_fallocate,
1046};
1047
1048const struct file_operations gfs2_dir_fops_nolock = {
1049	.readdir	= gfs2_readdir,
1050	.unlocked_ioctl	= gfs2_ioctl,
1051	.open		= gfs2_open,
1052	.release	= gfs2_release,
1053	.fsync		= gfs2_fsync,
1054	.llseek		= default_llseek,
1055};
1056