Linux Audio

Check our new training course

Linux kernel drivers training

May 6-19, 2025
Register
Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4 * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
 
 
 
 
   5 */
   6
   7#include <linux/slab.h>
   8#include <linux/spinlock.h>
   9#include <linux/compat.h>
  10#include <linux/completion.h>
  11#include <linux/buffer_head.h>
  12#include <linux/pagemap.h>
  13#include <linux/uio.h>
  14#include <linux/blkdev.h>
  15#include <linux/mm.h>
  16#include <linux/mount.h>
  17#include <linux/fs.h>
  18#include <linux/gfs2_ondisk.h>
  19#include <linux/falloc.h>
  20#include <linux/swap.h>
  21#include <linux/crc32.h>
  22#include <linux/writeback.h>
  23#include <linux/uaccess.h>
  24#include <linux/dlm.h>
  25#include <linux/dlm_plock.h>
  26#include <linux/delay.h>
  27#include <linux/backing-dev.h>
  28#include <linux/fileattr.h>
  29
  30#include "gfs2.h"
  31#include "incore.h"
  32#include "bmap.h"
  33#include "aops.h"
  34#include "dir.h"
  35#include "glock.h"
  36#include "glops.h"
  37#include "inode.h"
  38#include "log.h"
  39#include "meta_io.h"
  40#include "quota.h"
  41#include "rgrp.h"
  42#include "trans.h"
  43#include "util.h"
  44
  45/**
  46 * gfs2_llseek - seek to a location in a file
  47 * @file: the file
  48 * @offset: the offset
  49 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
  50 *
  51 * SEEK_END requires the glock for the file because it references the
  52 * file's size.
  53 *
  54 * Returns: The new offset, or errno
  55 */
  56
  57static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
  58{
  59	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
  60	struct gfs2_holder i_gh;
  61	loff_t error;
  62
  63	switch (whence) {
  64	case SEEK_END:
  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, whence);
  69			gfs2_glock_dq_uninit(&i_gh);
  70		}
  71		break;
  72
  73	case SEEK_DATA:
  74		error = gfs2_seek_data(file, offset);
  75		break;
  76
  77	case SEEK_HOLE:
  78		error = gfs2_seek_hole(file, offset);
  79		break;
  80
  81	case SEEK_CUR:
  82	case SEEK_SET:
  83		/*
  84		 * These don't reference inode->i_size and don't depend on the
  85		 * block mapping, so we don't need the glock.
  86		 */
  87		error = generic_file_llseek(file, offset, whence);
  88		break;
  89	default:
  90		error = -EINVAL;
  91	}
  92
  93	return error;
  94}
  95
  96/**
  97 * gfs2_readdir - Iterator for a directory
  98 * @file: The directory to read from
  99 * @ctx: What to feed directory entries to
 100 *
 101 * Returns: errno
 102 */
 103
 104static int gfs2_readdir(struct file *file, struct dir_context *ctx)
 105{
 106	struct inode *dir = file->f_mapping->host;
 107	struct gfs2_inode *dip = GFS2_I(dir);
 108	struct gfs2_holder d_gh;
 109	int error;
 110
 111	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
 112	if (error)
 113		return error;
 114
 115	error = gfs2_dir_read(dir, ctx, &file->f_ra);
 116
 117	gfs2_glock_dq_uninit(&d_gh);
 118
 119	return error;
 120}
 121
 122/*
 123 * struct fsflag_gfs2flag
 124 *
 125 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
 126 * and to GFS2_DIF_JDATA for non-directories.
 127 */
 128static struct {
 129	u32 fsflag;
 130	u32 gfsflag;
 131} fsflag_gfs2flag[] = {
 132	{FS_SYNC_FL, GFS2_DIF_SYNC},
 133	{FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
 134	{FS_APPEND_FL, GFS2_DIF_APPENDONLY},
 135	{FS_NOATIME_FL, GFS2_DIF_NOATIME},
 136	{FS_INDEX_FL, GFS2_DIF_EXHASH},
 137	{FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
 138	{FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
 139};
 140
 141static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
 142{
 143	int i;
 144	u32 fsflags = 0;
 145
 146	if (S_ISDIR(inode->i_mode))
 147		gfsflags &= ~GFS2_DIF_JDATA;
 148	else
 149		gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
 150
 151	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
 152		if (gfsflags & fsflag_gfs2flag[i].gfsflag)
 153			fsflags |= fsflag_gfs2flag[i].fsflag;
 154	return fsflags;
 155}
 156
 157int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 158{
 159	struct inode *inode = d_inode(dentry);
 160	struct gfs2_inode *ip = GFS2_I(inode);
 161	struct gfs2_holder gh;
 162	int error;
 163	u32 fsflags;
 164
 165	if (d_is_special(dentry))
 166		return -ENOTTY;
 167
 168	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 169	error = gfs2_glock_nq(&gh);
 170	if (error)
 171		goto out_uninit;
 172
 173	fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
 
 
 
 
 
 
 
 174
 175	fileattr_fill_flags(fa, fsflags);
 
 176
 177	gfs2_glock_dq(&gh);
 178out_uninit:
 179	gfs2_holder_uninit(&gh);
 180	return error;
 181}
 182
 183void gfs2_set_inode_flags(struct inode *inode)
 184{
 185	struct gfs2_inode *ip = GFS2_I(inode);
 186	unsigned int flags = inode->i_flags;
 187
 188	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
 189	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
 190		flags |= S_NOSEC;
 191	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
 192		flags |= S_IMMUTABLE;
 193	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
 194		flags |= S_APPEND;
 195	if (ip->i_diskflags & GFS2_DIF_NOATIME)
 196		flags |= S_NOATIME;
 197	if (ip->i_diskflags & GFS2_DIF_SYNC)
 198		flags |= S_SYNC;
 199	inode->i_flags = flags;
 200}
 201
 202/* Flags that can be set by user space */
 203#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
 204			     GFS2_DIF_IMMUTABLE|		\
 205			     GFS2_DIF_APPENDONLY|		\
 206			     GFS2_DIF_NOATIME|			\
 207			     GFS2_DIF_SYNC|			\
 208			     GFS2_DIF_TOPDIR|			\
 209			     GFS2_DIF_INHERIT_JDATA)
 210
 211/**
 212 * do_gfs2_set_flags - set flags on an inode
 213 * @inode: The inode
 214 * @reqflags: The flags to set
 215 * @mask: Indicates which flags are valid
 216 *
 217 */
 218static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask)
 219{
 
 220	struct gfs2_inode *ip = GFS2_I(inode);
 221	struct gfs2_sbd *sdp = GFS2_SB(inode);
 222	struct buffer_head *bh;
 223	struct gfs2_holder gh;
 224	int error;
 225	u32 new_flags, flags;
 226
 227	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 228	if (error)
 229		return error;
 230
 
 
 
 
 
 
 
 
 231	error = 0;
 232	flags = ip->i_diskflags;
 233	new_flags = (flags & ~mask) | (reqflags & mask);
 234	if ((new_flags ^ flags) == 0)
 235		goto out;
 236
 
 
 
 
 
 
 
 
 237	if (!IS_IMMUTABLE(inode)) {
 238		error = gfs2_permission(&init_user_ns, inode, MAY_WRITE);
 239		if (error)
 240			goto out;
 241	}
 242	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
 243		if (new_flags & GFS2_DIF_JDATA)
 244			gfs2_log_flush(sdp, ip->i_gl,
 245				       GFS2_LOG_HEAD_FLUSH_NORMAL |
 246				       GFS2_LFC_SET_FLAGS);
 247		error = filemap_fdatawrite(inode->i_mapping);
 248		if (error)
 249			goto out;
 250		error = filemap_fdatawait(inode->i_mapping);
 251		if (error)
 252			goto out;
 253		if (new_flags & GFS2_DIF_JDATA)
 254			gfs2_ordered_del_inode(ip);
 255	}
 256	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 257	if (error)
 258		goto out;
 259	error = gfs2_meta_inode_buffer(ip, &bh);
 260	if (error)
 261		goto out_trans_end;
 262	inode->i_ctime = current_time(inode);
 263	gfs2_trans_add_meta(ip->i_gl, bh);
 264	ip->i_diskflags = new_flags;
 265	gfs2_dinode_out(ip, bh->b_data);
 266	brelse(bh);
 267	gfs2_set_inode_flags(inode);
 268	gfs2_set_aops(inode);
 269out_trans_end:
 270	gfs2_trans_end(sdp);
 271out:
 272	gfs2_glock_dq_uninit(&gh);
 
 
 273	return error;
 274}
 275
 276int gfs2_fileattr_set(struct user_namespace *mnt_userns,
 277		      struct dentry *dentry, struct fileattr *fa)
 278{
 279	struct inode *inode = d_inode(dentry);
 280	u32 fsflags = fa->flags, gfsflags = 0;
 281	u32 mask;
 282	int i;
 283
 284	if (d_is_special(dentry))
 285		return -ENOTTY;
 286
 287	if (fileattr_has_fsx(fa))
 288		return -EOPNOTSUPP;
 289
 290	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
 291		if (fsflags & fsflag_gfs2flag[i].fsflag) {
 292			fsflags &= ~fsflag_gfs2flag[i].fsflag;
 293			gfsflags |= fsflag_gfs2flag[i].gfsflag;
 294		}
 295	}
 296	if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
 297		return -EINVAL;
 298
 299	mask = GFS2_FLAGS_USER_SET;
 300	if (S_ISDIR(inode->i_mode)) {
 301		mask &= ~GFS2_DIF_JDATA;
 302	} else {
 303		/* The GFS2_DIF_TOPDIR flag is only valid for directories. */
 304		if (gfsflags & GFS2_DIF_TOPDIR)
 305			return -EINVAL;
 306		mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
 307	}
 308
 309	return do_gfs2_set_flags(inode, gfsflags, mask);
 310}
 311
 312static int gfs2_getlabel(struct file *filp, char __user *label)
 313{
 314	struct inode *inode = file_inode(filp);
 315	struct gfs2_sbd *sdp = GFS2_SB(inode);
 316
 317	if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
 318		return -EFAULT;
 319
 320	return 0;
 321}
 322
 323static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 324{
 325	switch(cmd) {
 
 
 
 
 326	case FITRIM:
 327		return gfs2_fitrim(filp, (void __user *)arg);
 328	case FS_IOC_GETFSLABEL:
 329		return gfs2_getlabel(filp, (char __user *)arg);
 330	}
 331
 332	return -ENOTTY;
 333}
 334
 335#ifdef CONFIG_COMPAT
 336static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 337{
 338	switch(cmd) {
 339	/* Keep this list in sync with gfs2_ioctl */
 340	case FITRIM:
 341	case FS_IOC_GETFSLABEL:
 342		break;
 343	default:
 344		return -ENOIOCTLCMD;
 345	}
 346
 347	return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
 348}
 349#else
 350#define gfs2_compat_ioctl NULL
 351#endif
 352
 353/**
 354 * gfs2_size_hint - Give a hint to the size of a write request
 355 * @filep: The struct file
 356 * @offset: The file offset of the write
 357 * @size: The length of the write
 358 *
 359 * When we are about to do a write, this function records the total
 360 * write size in order to provide a suitable hint to the lower layers
 361 * about how many blocks will be required.
 362 *
 363 */
 364
 365static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
 366{
 367	struct inode *inode = file_inode(filep);
 368	struct gfs2_sbd *sdp = GFS2_SB(inode);
 369	struct gfs2_inode *ip = GFS2_I(inode);
 370	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
 371	int hint = min_t(size_t, INT_MAX, blks);
 372
 373	if (hint > atomic_read(&ip->i_sizehint))
 374		atomic_set(&ip->i_sizehint, hint);
 375}
 376
 377/**
 378 * gfs2_allocate_page_backing - Allocate blocks for a write fault
 379 * @page: The (locked) page to allocate backing for
 380 * @length: Size of the allocation
 381 *
 382 * We try to allocate all the blocks required for the page in one go.  This
 383 * might fail for various reasons, so we keep trying until all the blocks to
 384 * back this page are allocated.  If some of the blocks are already allocated,
 385 * that is ok too.
 386 */
 387static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
 
 388{
 389	u64 pos = page_offset(page);
 
 
 
 390
 391	do {
 392		struct iomap iomap = { };
 393
 394		if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
 
 395			return -EIO;
 396
 397		if (length < iomap.length)
 398			iomap.length = length;
 399		length -= iomap.length;
 400		pos += iomap.length;
 401	} while (length > 0);
 402
 403	return 0;
 404}
 405
 406/**
 407 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
 
 408 * @vmf: The virtual memory fault containing the page to become writable
 409 *
 410 * When the page becomes writable, we need to ensure that we have
 411 * blocks allocated on disk to back that page.
 412 */
 413
 414static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
 415{
 416	struct page *page = vmf->page;
 417	struct inode *inode = file_inode(vmf->vma->vm_file);
 418	struct gfs2_inode *ip = GFS2_I(inode);
 419	struct gfs2_sbd *sdp = GFS2_SB(inode);
 420	struct gfs2_alloc_parms ap = { .aflags = 0, };
 421	u64 offset = page_offset(page);
 
 422	unsigned int data_blocks, ind_blocks, rblocks;
 423	vm_fault_t ret = VM_FAULT_LOCKED;
 424	struct gfs2_holder gh;
 425	unsigned int length;
 426	loff_t size;
 427	int err;
 428
 429	sb_start_pagefault(inode->i_sb);
 430
 
 
 
 
 
 
 431	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 432	err = gfs2_glock_nq(&gh);
 433	if (err) {
 434		ret = block_page_mkwrite_return(err);
 435		goto out_uninit;
 436	}
 437
 438	/* Check page index against inode size */
 439	size = i_size_read(inode);
 440	if (offset >= size) {
 441		ret = VM_FAULT_SIGBUS;
 442		goto out_unlock;
 443	}
 444
 445	/* Update file times before taking page lock */
 446	file_update_time(vmf->vma->vm_file);
 447
 448	/* page is wholly or partially inside EOF */
 449	if (size - offset < PAGE_SIZE)
 450		length = size - offset;
 451	else
 452		length = PAGE_SIZE;
 453
 454	gfs2_size_hint(vmf->vma->vm_file, offset, length);
 455
 456	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
 457	set_bit(GIF_SW_PAGED, &ip->i_flags);
 458
 459	/*
 460	 * iomap_writepage / iomap_writepages currently don't support inline
 461	 * files, so always unstuff here.
 462	 */
 463
 464	if (!gfs2_is_stuffed(ip) &&
 465	    !gfs2_write_alloc_required(ip, offset, length)) {
 466		lock_page(page);
 467		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
 468			ret = VM_FAULT_NOPAGE;
 469			unlock_page(page);
 470		}
 471		goto out_unlock;
 472	}
 473
 474	err = gfs2_rindex_update(sdp);
 475	if (err) {
 476		ret = block_page_mkwrite_return(err);
 477		goto out_unlock;
 478	}
 479
 480	gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
 481	ap.target = data_blocks + ind_blocks;
 482	err = gfs2_quota_lock_check(ip, &ap);
 483	if (err) {
 484		ret = block_page_mkwrite_return(err);
 485		goto out_unlock;
 486	}
 487	err = gfs2_inplace_reserve(ip, &ap);
 488	if (err) {
 489		ret = block_page_mkwrite_return(err);
 490		goto out_quota_unlock;
 491	}
 492
 493	rblocks = RES_DINODE + ind_blocks;
 494	if (gfs2_is_jdata(ip))
 495		rblocks += data_blocks ? data_blocks : 1;
 496	if (ind_blocks || data_blocks) {
 497		rblocks += RES_STATFS + RES_QUOTA;
 498		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
 499	}
 500	err = gfs2_trans_begin(sdp, rblocks, 0);
 501	if (err) {
 502		ret = block_page_mkwrite_return(err);
 503		goto out_trans_fail;
 504	}
 505
 506	/* Unstuff, if required, and allocate backing blocks for page */
 507	if (gfs2_is_stuffed(ip)) {
 508		err = gfs2_unstuff_dinode(ip);
 509		if (err) {
 510			ret = block_page_mkwrite_return(err);
 511			goto out_trans_end;
 512		}
 513	}
 514
 515	lock_page(page);
 
 
 
 
 
 
 
 
 516	/* If truncated, we must retry the operation, we may have raced
 517	 * with the glock demotion code.
 518	 */
 519	if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
 520		ret = VM_FAULT_NOPAGE;
 521		goto out_page_locked;
 522	}
 523
 524	err = gfs2_allocate_page_backing(page, length);
 525	if (err)
 526		ret = block_page_mkwrite_return(err);
 
 
 
 527
 528out_page_locked:
 529	if (ret != VM_FAULT_LOCKED)
 530		unlock_page(page);
 531out_trans_end:
 
 
 532	gfs2_trans_end(sdp);
 533out_trans_fail:
 534	gfs2_inplace_release(ip);
 535out_quota_unlock:
 536	gfs2_quota_unlock(ip);
 537out_unlock:
 538	gfs2_glock_dq(&gh);
 539out_uninit:
 540	gfs2_holder_uninit(&gh);
 541	if (ret == VM_FAULT_LOCKED) {
 542		set_page_dirty(page);
 543		wait_for_stable_page(page);
 544	}
 
 545	sb_end_pagefault(inode->i_sb);
 546	return ret;
 547}
 548
 549static vm_fault_t gfs2_fault(struct vm_fault *vmf)
 550{
 551	struct inode *inode = file_inode(vmf->vma->vm_file);
 552	struct gfs2_inode *ip = GFS2_I(inode);
 553	struct gfs2_holder gh;
 554	vm_fault_t ret;
 555	int err;
 556
 557	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 558	err = gfs2_glock_nq(&gh);
 559	if (err) {
 560		ret = block_page_mkwrite_return(err);
 561		goto out_uninit;
 562	}
 563	ret = filemap_fault(vmf);
 564	gfs2_glock_dq(&gh);
 565out_uninit:
 566	gfs2_holder_uninit(&gh);
 567	return ret;
 568}
 569
 570static const struct vm_operations_struct gfs2_vm_ops = {
 571	.fault = gfs2_fault,
 572	.map_pages = filemap_map_pages,
 573	.page_mkwrite = gfs2_page_mkwrite,
 574};
 575
 576/**
 577 * gfs2_mmap
 578 * @file: The file to map
 579 * @vma: The VMA which described the mapping
 580 *
 581 * There is no need to get a lock here unless we should be updating
 582 * atime. We ignore any locking errors since the only consequence is
 583 * a missed atime update (which will just be deferred until later).
 584 *
 585 * Returns: 0
 586 */
 587
 588static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 589{
 590	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 591
 592	if (!(file->f_flags & O_NOATIME) &&
 593	    !IS_NOATIME(&ip->i_inode)) {
 594		struct gfs2_holder i_gh;
 595		int error;
 596
 597		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 598					   &i_gh);
 599		if (error)
 600			return error;
 601		/* grab lock to update inode */
 602		gfs2_glock_dq_uninit(&i_gh);
 603		file_accessed(file);
 604	}
 605	vma->vm_ops = &gfs2_vm_ops;
 606
 607	return 0;
 608}
 609
 610/**
 611 * gfs2_open_common - This is common to open and atomic_open
 612 * @inode: The inode being opened
 613 * @file: The file being opened
 614 *
 615 * This maybe called under a glock or not depending upon how it has
 616 * been called. We must always be called under a glock for regular
 617 * files, however. For other file types, it does not matter whether
 618 * we hold the glock or not.
 619 *
 620 * Returns: Error code or 0 for success
 621 */
 622
 623int gfs2_open_common(struct inode *inode, struct file *file)
 624{
 625	struct gfs2_file *fp;
 626	int ret;
 627
 628	if (S_ISREG(inode->i_mode)) {
 629		ret = generic_file_open(inode, file);
 630		if (ret)
 631			return ret;
 632	}
 633
 634	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
 635	if (!fp)
 636		return -ENOMEM;
 637
 638	mutex_init(&fp->f_fl_mutex);
 639
 640	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
 641	file->private_data = fp;
 642	if (file->f_mode & FMODE_WRITE) {
 643		ret = gfs2_qa_get(GFS2_I(inode));
 644		if (ret)
 645			goto fail;
 646	}
 647	return 0;
 648
 649fail:
 650	kfree(file->private_data);
 651	file->private_data = NULL;
 652	return ret;
 653}
 654
 655/**
 656 * gfs2_open - open a file
 657 * @inode: the inode to open
 658 * @file: the struct file for this opening
 659 *
 660 * After atomic_open, this function is only used for opening files
 661 * which are already cached. We must still get the glock for regular
 662 * files to ensure that we have the file size uptodate for the large
 663 * file check which is in the common code. That is only an issue for
 664 * regular files though.
 665 *
 666 * Returns: errno
 667 */
 668
 669static int gfs2_open(struct inode *inode, struct file *file)
 670{
 671	struct gfs2_inode *ip = GFS2_I(inode);
 672	struct gfs2_holder i_gh;
 673	int error;
 674	bool need_unlock = false;
 675
 676	if (S_ISREG(ip->i_inode.i_mode)) {
 677		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 678					   &i_gh);
 679		if (error)
 680			return error;
 681		need_unlock = true;
 682	}
 683
 684	error = gfs2_open_common(inode, file);
 685
 686	if (need_unlock)
 687		gfs2_glock_dq_uninit(&i_gh);
 688
 689	return error;
 690}
 691
 692/**
 693 * gfs2_release - called to close a struct file
 694 * @inode: the inode the struct file belongs to
 695 * @file: the struct file being closed
 696 *
 697 * Returns: errno
 698 */
 699
 700static int gfs2_release(struct inode *inode, struct file *file)
 701{
 702	struct gfs2_inode *ip = GFS2_I(inode);
 703
 704	kfree(file->private_data);
 705	file->private_data = NULL;
 706
 707	if (file->f_mode & FMODE_WRITE) {
 708		if (gfs2_rs_active(&ip->i_res))
 709			gfs2_rs_delete(ip);
 710		gfs2_qa_put(ip);
 711	}
 712	return 0;
 713}
 714
 715/**
 716 * gfs2_fsync - sync the dirty data for a file (across the cluster)
 717 * @file: the file that points to the dentry
 718 * @start: the start position in the file to sync
 719 * @end: the end position in the file to sync
 720 * @datasync: set if we can ignore timestamp changes
 721 *
 722 * We split the data flushing here so that we don't wait for the data
 723 * until after we've also sent the metadata to disk. Note that for
 724 * data=ordered, we will write & wait for the data at the log flush
 725 * stage anyway, so this is unlikely to make much of a difference
 726 * except in the data=writeback case.
 727 *
 728 * If the fdatawrite fails due to any reason except -EIO, we will
 729 * continue the remainder of the fsync, although we'll still report
 730 * the error at the end. This is to match filemap_write_and_wait_range()
 731 * behaviour.
 732 *
 733 * Returns: errno
 734 */
 735
 736static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 737		      int datasync)
 738{
 739	struct address_space *mapping = file->f_mapping;
 740	struct inode *inode = mapping->host;
 741	int sync_state = inode->i_state & I_DIRTY;
 742	struct gfs2_inode *ip = GFS2_I(inode);
 743	int ret = 0, ret1 = 0;
 744
 745	if (mapping->nrpages) {
 746		ret1 = filemap_fdatawrite_range(mapping, start, end);
 747		if (ret1 == -EIO)
 748			return ret1;
 749	}
 750
 751	if (!gfs2_is_jdata(ip))
 752		sync_state &= ~I_DIRTY_PAGES;
 753	if (datasync)
 754		sync_state &= ~I_DIRTY_SYNC;
 755
 756	if (sync_state) {
 757		ret = sync_inode_metadata(inode, 1);
 758		if (ret)
 759			return ret;
 760		if (gfs2_is_jdata(ip))
 761			ret = file_write_and_wait(file);
 762		if (ret)
 763			return ret;
 764		gfs2_ail_flush(ip->i_gl, 1);
 765	}
 766
 767	if (mapping->nrpages)
 768		ret = file_fdatawait_range(file, start, end);
 769
 770	return ret ? ret : ret1;
 771}
 772
 773static inline bool should_fault_in_pages(struct iov_iter *i,
 774					 struct kiocb *iocb,
 775					 size_t *prev_count,
 776					 size_t *window_size)
 777{
 778	size_t count = iov_iter_count(i);
 779	size_t size, offs;
 780
 781	if (!count)
 782		return false;
 783	if (!user_backed_iter(i))
 784		return false;
 785
 786	size = PAGE_SIZE;
 787	offs = offset_in_page(iocb->ki_pos);
 788	if (*prev_count != count || !*window_size) {
 789		size_t nr_dirtied;
 790
 791		nr_dirtied = max(current->nr_dirtied_pause -
 792				 current->nr_dirtied, 8);
 793		size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT);
 794	}
 795
 796	*prev_count = count;
 797	*window_size = size - offs;
 798	return true;
 799}
 800
 801static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
 802				     struct gfs2_holder *gh)
 803{
 804	struct file *file = iocb->ki_filp;
 805	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 806	size_t prev_count = 0, window_size = 0;
 807	size_t read = 0;
 808	ssize_t ret;
 809
 810	/*
 811	 * In this function, we disable page faults when we're holding the
 812	 * inode glock while doing I/O.  If a page fault occurs, we indicate
 813	 * that the inode glock may be dropped, fault in the pages manually,
 814	 * and retry.
 815	 *
 816	 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
 817	 * physical as well as manual page faults, and we need to disable both
 818	 * kinds.
 819	 *
 820	 * For direct I/O, gfs2 takes the inode glock in deferred mode.  This
 821	 * locking mode is compatible with other deferred holders, so multiple
 822	 * processes and nodes can do direct I/O to a file at the same time.
 823	 * There's no guarantee that reads or writes will be atomic.  Any
 824	 * coordination among readers and writers needs to happen externally.
 825	 */
 826
 827	if (!iov_iter_count(to))
 828		return 0; /* skip atime */
 829
 830	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
 831retry:
 832	ret = gfs2_glock_nq(gh);
 833	if (ret)
 834		goto out_uninit;
 835	pagefault_disable();
 836	to->nofault = true;
 837	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
 838			   IOMAP_DIO_PARTIAL, NULL, read);
 839	to->nofault = false;
 840	pagefault_enable();
 841	if (ret <= 0 && ret != -EFAULT)
 842		goto out_unlock;
 843	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
 844	if (ret > 0)
 845		read = ret;
 846
 847	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
 848		gfs2_glock_dq(gh);
 849		window_size -= fault_in_iov_iter_writeable(to, window_size);
 850		if (window_size)
 851			goto retry;
 852	}
 853out_unlock:
 854	if (gfs2_holder_queued(gh))
 855		gfs2_glock_dq(gh);
 856out_uninit:
 857	gfs2_holder_uninit(gh);
 858	/* User space doesn't expect partial success. */
 859	if (ret < 0)
 860		return ret;
 861	return read;
 862}
 863
 864static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
 865				      struct gfs2_holder *gh)
 866{
 867	struct file *file = iocb->ki_filp;
 868	struct inode *inode = file->f_mapping->host;
 869	struct gfs2_inode *ip = GFS2_I(inode);
 870	size_t prev_count = 0, window_size = 0;
 871	size_t written = 0;
 872	ssize_t ret;
 873
 874	/*
 875	 * In this function, we disable page faults when we're holding the
 876	 * inode glock while doing I/O.  If a page fault occurs, we indicate
 877	 * that the inode glock may be dropped, fault in the pages manually,
 878	 * and retry.
 879	 *
 880	 * For writes, iomap_dio_rw only triggers manual page faults, so we
 881	 * don't need to disable physical ones.
 882	 */
 883
 884	/*
 885	 * Deferred lock, even if its a write, since we do no allocation on
 886	 * this path. All we need to change is the atime, and this lock mode
 887	 * ensures that other nodes have flushed their buffered read caches
 888	 * (i.e. their page cache entries for this inode). We do not,
 889	 * unfortunately, have the option of only flushing a range like the
 890	 * VFS does.
 891	 */
 892	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
 893retry:
 894	ret = gfs2_glock_nq(gh);
 895	if (ret)
 896		goto out_uninit;
 897	/* Silently fall back to buffered I/O when writing beyond EOF */
 898	if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
 899		goto out_unlock;
 900
 901	from->nofault = true;
 902	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
 903			   IOMAP_DIO_PARTIAL, NULL, written);
 904	from->nofault = false;
 905	if (ret <= 0) {
 906		if (ret == -ENOTBLK)
 907			ret = 0;
 908		if (ret != -EFAULT)
 909			goto out_unlock;
 910	}
 911	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
 912	if (ret > 0)
 913		written = ret;
 914
 915	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
 916		gfs2_glock_dq(gh);
 917		window_size -= fault_in_iov_iter_readable(from, window_size);
 918		if (window_size)
 919			goto retry;
 920	}
 921out_unlock:
 922	if (gfs2_holder_queued(gh))
 923		gfs2_glock_dq(gh);
 924out_uninit:
 925	gfs2_holder_uninit(gh);
 926	/* User space doesn't expect partial success. */
 927	if (ret < 0)
 928		return ret;
 929	return written;
 930}
 931
 932static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 933{
 934	struct gfs2_inode *ip;
 935	struct gfs2_holder gh;
 936	size_t prev_count = 0, window_size = 0;
 937	size_t read = 0;
 938	ssize_t ret;
 939
 940	/*
 941	 * In this function, we disable page faults when we're holding the
 942	 * inode glock while doing I/O.  If a page fault occurs, we indicate
 943	 * that the inode glock may be dropped, fault in the pages manually,
 944	 * and retry.
 945	 */
 946
 947	if (iocb->ki_flags & IOCB_DIRECT)
 948		return gfs2_file_direct_read(iocb, to, &gh);
 949
 950	pagefault_disable();
 951	iocb->ki_flags |= IOCB_NOIO;
 952	ret = generic_file_read_iter(iocb, to);
 953	iocb->ki_flags &= ~IOCB_NOIO;
 954	pagefault_enable();
 955	if (ret >= 0) {
 956		if (!iov_iter_count(to))
 957			return ret;
 958		read = ret;
 959	} else if (ret != -EFAULT) {
 960		if (ret != -EAGAIN)
 961			return ret;
 962		if (iocb->ki_flags & IOCB_NOWAIT)
 963			return ret;
 964	}
 965	ip = GFS2_I(iocb->ki_filp->f_mapping->host);
 966	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 967retry:
 968	ret = gfs2_glock_nq(&gh);
 969	if (ret)
 970		goto out_uninit;
 971	pagefault_disable();
 972	ret = generic_file_read_iter(iocb, to);
 973	pagefault_enable();
 974	if (ret <= 0 && ret != -EFAULT)
 975		goto out_unlock;
 976	if (ret > 0)
 977		read += ret;
 978
 979	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
 980		gfs2_glock_dq(&gh);
 981		window_size -= fault_in_iov_iter_writeable(to, window_size);
 982		if (window_size)
 983			goto retry;
 984	}
 985out_unlock:
 986	if (gfs2_holder_queued(&gh))
 987		gfs2_glock_dq(&gh);
 988out_uninit:
 989	gfs2_holder_uninit(&gh);
 990	return read ? read : ret;
 991}
 992
 993static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
 994					struct iov_iter *from,
 995					struct gfs2_holder *gh)
 996{
 997	struct file *file = iocb->ki_filp;
 998	struct inode *inode = file_inode(file);
 999	struct gfs2_inode *ip = GFS2_I(inode);
1000	struct gfs2_sbd *sdp = GFS2_SB(inode);
1001	struct gfs2_holder *statfs_gh = NULL;
1002	size_t prev_count = 0, window_size = 0;
1003	size_t orig_count = iov_iter_count(from);
1004	size_t written = 0;
1005	ssize_t ret;
1006
1007	/*
1008	 * In this function, we disable page faults when we're holding the
1009	 * inode glock while doing I/O.  If a page fault occurs, we indicate
1010	 * that the inode glock may be dropped, fault in the pages manually,
1011	 * and retry.
1012	 */
1013
1014	if (inode == sdp->sd_rindex) {
1015		statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1016		if (!statfs_gh)
1017			return -ENOMEM;
1018	}
1019
1020	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
1021retry:
1022	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1023		window_size -= fault_in_iov_iter_readable(from, window_size);
1024		if (!window_size) {
1025			ret = -EFAULT;
1026			goto out_uninit;
1027		}
1028		from->count = min(from->count, window_size);
1029	}
1030	ret = gfs2_glock_nq(gh);
1031	if (ret)
1032		goto out_uninit;
1033
1034	if (inode == sdp->sd_rindex) {
1035		struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1036
1037		ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1038					 GL_NOCACHE, statfs_gh);
1039		if (ret)
1040			goto out_unlock;
1041	}
1042
1043	current->backing_dev_info = inode_to_bdi(inode);
1044	pagefault_disable();
1045	ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
1046	pagefault_enable();
1047	current->backing_dev_info = NULL;
1048	if (ret > 0) {
1049		iocb->ki_pos += ret;
1050		written += ret;
1051	}
1052
1053	if (inode == sdp->sd_rindex)
1054		gfs2_glock_dq_uninit(statfs_gh);
1055
1056	if (ret <= 0 && ret != -EFAULT)
1057		goto out_unlock;
1058
1059	from->count = orig_count - written;
1060	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1061		gfs2_glock_dq(gh);
1062		goto retry;
1063	}
1064out_unlock:
1065	if (gfs2_holder_queued(gh))
1066		gfs2_glock_dq(gh);
1067out_uninit:
1068	gfs2_holder_uninit(gh);
1069	kfree(statfs_gh);
1070	from->count = orig_count - written;
1071	return written ? written : ret;
1072}
1073
1074/**
1075 * gfs2_file_write_iter - Perform a write to a file
1076 * @iocb: The io context
1077 * @from: The data to write
 
 
1078 *
1079 * We have to do a lock/unlock here to refresh the inode size for
1080 * O_APPEND writes, otherwise we can land up writing at the wrong
1081 * offset. There is still a race, but provided the app is using its
1082 * own file locking, this will make O_APPEND work as expected.
1083 *
1084 */
1085
1086static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1087{
1088	struct file *file = iocb->ki_filp;
1089	struct inode *inode = file_inode(file);
1090	struct gfs2_inode *ip = GFS2_I(inode);
1091	struct gfs2_holder gh;
1092	ssize_t ret;
 
 
1093
1094	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
1095
1096	if (iocb->ki_flags & IOCB_APPEND) {
 
 
1097		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1098		if (ret)
1099			return ret;
1100		gfs2_glock_dq_uninit(&gh);
1101	}
1102
1103	inode_lock(inode);
1104	ret = generic_write_checks(iocb, from);
1105	if (ret <= 0)
1106		goto out_unlock;
1107
1108	ret = file_remove_privs(file);
1109	if (ret)
1110		goto out_unlock;
1111
1112	ret = file_update_time(file);
1113	if (ret)
1114		goto out_unlock;
1115
1116	if (iocb->ki_flags & IOCB_DIRECT) {
1117		struct address_space *mapping = file->f_mapping;
1118		ssize_t buffered, ret2;
1119
1120		ret = gfs2_file_direct_write(iocb, from, &gh);
1121		if (ret < 0 || !iov_iter_count(from))
1122			goto out_unlock;
1123
1124		iocb->ki_flags |= IOCB_DSYNC;
1125		buffered = gfs2_file_buffered_write(iocb, from, &gh);
1126		if (unlikely(buffered <= 0)) {
1127			if (!ret)
1128				ret = buffered;
1129			goto out_unlock;
1130		}
1131
1132		/*
1133		 * We need to ensure that the page cache pages are written to
1134		 * disk and invalidated to preserve the expected O_DIRECT
1135		 * semantics.  If the writeback or invalidate fails, only report
1136		 * the direct I/O range as we don't know if the buffered pages
1137		 * made it to disk.
1138		 */
1139		ret2 = generic_write_sync(iocb, buffered);
1140		invalidate_mapping_pages(mapping,
1141				(iocb->ki_pos - buffered) >> PAGE_SHIFT,
1142				(iocb->ki_pos - 1) >> PAGE_SHIFT);
1143		if (!ret || ret2 > 0)
1144			ret += ret2;
1145	} else {
1146		ret = gfs2_file_buffered_write(iocb, from, &gh);
1147		if (likely(ret > 0))
1148			ret = generic_write_sync(iocb, ret);
1149	}
1150
1151out_unlock:
1152	inode_unlock(inode);
1153	return ret;
1154}
1155
1156static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
1157			   int mode)
1158{
1159	struct super_block *sb = inode->i_sb;
1160	struct gfs2_inode *ip = GFS2_I(inode);
1161	loff_t end = offset + len;
1162	struct buffer_head *dibh;
 
1163	int error;
1164
1165	error = gfs2_meta_inode_buffer(ip, &dibh);
1166	if (unlikely(error))
1167		return error;
1168
1169	gfs2_trans_add_meta(ip->i_gl, dibh);
1170
1171	if (gfs2_is_stuffed(ip)) {
1172		error = gfs2_unstuff_dinode(ip);
1173		if (unlikely(error))
1174			goto out;
1175	}
1176
1177	while (offset < end) {
1178		struct iomap iomap = { };
1179
1180		error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
1181		if (error)
1182			goto out;
1183		offset = iomap.offset + iomap.length;
1184		if (!(iomap.flags & IOMAP_F_NEW))
1185			continue;
1186		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1187					 iomap.length >> inode->i_blkbits,
1188					 GFP_NOFS);
1189		if (error) {
1190			fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1191			goto out;
1192		}
1193	}
1194out:
1195	brelse(dibh);
1196	return error;
1197}
1198
1199/**
1200 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1201 *                     blocks, determine how many bytes can be written.
1202 * @ip:          The inode in question.
1203 * @len:         Max cap of bytes. What we return in *len must be <= this.
1204 * @data_blocks: Compute and return the number of data blocks needed
1205 * @ind_blocks:  Compute and return the number of indirect blocks needed
1206 * @max_blocks:  The total blocks available to work with.
1207 *
1208 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1209 */
1210static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1211			    unsigned int *data_blocks, unsigned int *ind_blocks,
1212			    unsigned int max_blocks)
1213{
1214	loff_t max = *len;
1215	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1216	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1217
1218	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1219		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1220		max_data -= tmp;
1221	}
1222
1223	*data_blocks = max_data;
1224	*ind_blocks = max_blocks - max_data;
1225	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1226	if (*len > max) {
1227		*len = max;
1228		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1229	}
1230}
1231
1232static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1233{
1234	struct inode *inode = file_inode(file);
1235	struct gfs2_sbd *sdp = GFS2_SB(inode);
1236	struct gfs2_inode *ip = GFS2_I(inode);
1237	struct gfs2_alloc_parms ap = { .aflags = 0, };
1238	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1239	loff_t bytes, max_bytes, max_blks;
1240	int error;
1241	const loff_t pos = offset;
1242	const loff_t count = len;
1243	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1244	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1245	loff_t max_chunk_size = UINT_MAX & bsize_mask;
1246
1247	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1248
1249	offset &= bsize_mask;
1250
1251	len = next - offset;
1252	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1253	if (!bytes)
1254		bytes = UINT_MAX;
1255	bytes &= bsize_mask;
1256	if (bytes == 0)
1257		bytes = sdp->sd_sb.sb_bsize;
1258
1259	gfs2_size_hint(file, offset, len);
1260
1261	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1262	ap.min_target = data_blocks + ind_blocks;
1263
1264	while (len > 0) {
1265		if (len < bytes)
1266			bytes = len;
1267		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1268			len -= bytes;
1269			offset += bytes;
1270			continue;
1271		}
1272
1273		/* We need to determine how many bytes we can actually
1274		 * fallocate without exceeding quota or going over the
1275		 * end of the fs. We start off optimistically by assuming
1276		 * we can write max_bytes */
1277		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1278
1279		/* Since max_bytes is most likely a theoretical max, we
1280		 * calculate a more realistic 'bytes' to serve as a good
1281		 * starting point for the number of bytes we may be able
1282		 * to write */
1283		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1284		ap.target = data_blocks + ind_blocks;
1285
1286		error = gfs2_quota_lock_check(ip, &ap);
1287		if (error)
1288			return error;
1289		/* ap.allowed tells us how many blocks quota will allow
1290		 * us to write. Check if this reduces max_blks */
1291		max_blks = UINT_MAX;
1292		if (ap.allowed)
1293			max_blks = ap.allowed;
1294
1295		error = gfs2_inplace_reserve(ip, &ap);
1296		if (error)
1297			goto out_qunlock;
1298
1299		/* check if the selected rgrp limits our max_blks further */
1300		if (ip->i_res.rs_reserved < max_blks)
1301			max_blks = ip->i_res.rs_reserved;
1302
1303		/* Almost done. Calculate bytes that can be written using
1304		 * max_blks. We also recompute max_bytes, data_blocks and
1305		 * ind_blocks */
1306		calc_max_reserv(ip, &max_bytes, &data_blocks,
1307				&ind_blocks, max_blks);
1308
1309		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1310			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1311		if (gfs2_is_jdata(ip))
1312			rblocks += data_blocks ? data_blocks : 1;
1313
1314		error = gfs2_trans_begin(sdp, rblocks,
1315					 PAGE_SIZE >> inode->i_blkbits);
1316		if (error)
1317			goto out_trans_fail;
1318
1319		error = fallocate_chunk(inode, offset, max_bytes, mode);
1320		gfs2_trans_end(sdp);
1321
1322		if (error)
1323			goto out_trans_fail;
1324
1325		len -= max_bytes;
1326		offset += max_bytes;
1327		gfs2_inplace_release(ip);
1328		gfs2_quota_unlock(ip);
1329	}
1330
1331	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1332		i_size_write(inode, pos + count);
1333	file_update_time(file);
1334	mark_inode_dirty(inode);
 
1335
1336	if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1337		return vfs_fsync_range(file, pos, pos + count - 1,
1338			       (file->f_flags & __O_SYNC) ? 0 : 1);
1339	return 0;
1340
1341out_trans_fail:
1342	gfs2_inplace_release(ip);
1343out_qunlock:
1344	gfs2_quota_unlock(ip);
1345	return error;
1346}
1347
1348static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1349{
1350	struct inode *inode = file_inode(file);
1351	struct gfs2_sbd *sdp = GFS2_SB(inode);
1352	struct gfs2_inode *ip = GFS2_I(inode);
1353	struct gfs2_holder gh;
1354	int ret;
1355
1356	if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1357		return -EOPNOTSUPP;
1358	/* fallocate is needed by gfs2_grow to reserve space in the rindex */
1359	if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1360		return -EOPNOTSUPP;
1361
1362	inode_lock(inode);
1363
1364	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1365	ret = gfs2_glock_nq(&gh);
1366	if (ret)
1367		goto out_uninit;
1368
1369	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1370	    (offset + len) > inode->i_size) {
1371		ret = inode_newsize_ok(inode, offset + len);
1372		if (ret)
1373			goto out_unlock;
1374	}
1375
1376	ret = get_write_access(inode);
1377	if (ret)
1378		goto out_unlock;
1379
1380	if (mode & FALLOC_FL_PUNCH_HOLE) {
1381		ret = __gfs2_punch_hole(file, offset, len);
1382	} else {
 
 
 
 
1383		ret = __gfs2_fallocate(file, mode, offset, len);
 
1384		if (ret)
1385			gfs2_rs_deltree(&ip->i_res);
1386	}
1387
 
1388	put_write_access(inode);
1389out_unlock:
1390	gfs2_glock_dq(&gh);
1391out_uninit:
1392	gfs2_holder_uninit(&gh);
1393	inode_unlock(inode);
1394	return ret;
1395}
1396
1397static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1398				      struct file *out, loff_t *ppos,
1399				      size_t len, unsigned int flags)
1400{
1401	ssize_t ret;
 
 
 
 
 
1402
1403	gfs2_size_hint(out, *ppos, len);
1404
1405	ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1406	return ret;
1407}
1408
1409#ifdef CONFIG_GFS2_FS_LOCKING_DLM
1410
1411/**
1412 * gfs2_lock - acquire/release a posix lock on a file
1413 * @file: the file pointer
1414 * @cmd: either modify or retrieve lock state, possibly wait
1415 * @fl: type and range of lock
1416 *
1417 * Returns: errno
1418 */
1419
1420static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1421{
1422	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1423	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1424	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1425
1426	if (!(fl->fl_flags & FL_POSIX))
1427		return -ENOLCK;
 
 
 
1428	if (cmd == F_CANCELLK) {
1429		/* Hack: */
1430		cmd = F_SETLK;
1431		fl->fl_type = F_UNLCK;
1432	}
1433	if (unlikely(gfs2_withdrawn(sdp))) {
1434		if (fl->fl_type == F_UNLCK)
1435			locks_lock_file_wait(file, fl);
1436		return -EIO;
1437	}
1438	if (IS_GETLK(cmd))
1439		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1440	else if (fl->fl_type == F_UNLCK)
1441		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1442	else
1443		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1444}
1445
1446static void __flock_holder_uninit(struct file *file, struct gfs2_holder *fl_gh)
1447{
1448	struct gfs2_glock *gl = gfs2_glock_hold(fl_gh->gh_gl);
1449
1450	/*
1451	 * Make sure gfs2_glock_put() won't sleep under the file->f_lock
1452	 * spinlock.
1453	 */
1454
1455	spin_lock(&file->f_lock);
1456	gfs2_holder_uninit(fl_gh);
1457	spin_unlock(&file->f_lock);
1458	gfs2_glock_put(gl);
1459}
1460
1461static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1462{
1463	struct gfs2_file *fp = file->private_data;
1464	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1465	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1466	struct gfs2_glock *gl;
1467	unsigned int state;
1468	u16 flags;
1469	int error = 0;
1470	int sleeptime;
1471
1472	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1473	flags = GL_EXACT | GL_NOPID;
1474	if (!IS_SETLKW(cmd))
1475		flags |= LM_FLAG_TRY_1CB;
1476
1477	mutex_lock(&fp->f_fl_mutex);
1478
1479	if (gfs2_holder_initialized(fl_gh)) {
1480		struct file_lock request;
1481		if (fl_gh->gh_state == state)
1482			goto out;
1483		locks_init_lock(&request);
1484		request.fl_type = F_UNLCK;
1485		request.fl_flags = FL_FLOCK;
1486		locks_lock_file_wait(file, &request);
 
1487		gfs2_glock_dq(fl_gh);
1488		gfs2_holder_reinit(state, flags, fl_gh);
1489	} else {
1490		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1491				       &gfs2_flock_glops, CREATE, &gl);
1492		if (error)
1493			goto out;
1494		spin_lock(&file->f_lock);
1495		gfs2_holder_init(gl, state, flags, fl_gh);
1496		spin_unlock(&file->f_lock);
1497		gfs2_glock_put(gl);
1498	}
1499	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1500		error = gfs2_glock_nq(fl_gh);
1501		if (error != GLR_TRYFAILED)
1502			break;
1503		fl_gh->gh_flags &= ~LM_FLAG_TRY_1CB;
1504		fl_gh->gh_flags |= LM_FLAG_TRY;
1505		msleep(sleeptime);
1506	}
1507	if (error) {
1508		__flock_holder_uninit(file, fl_gh);
1509		if (error == GLR_TRYFAILED)
1510			error = -EAGAIN;
1511	} else {
1512		error = locks_lock_file_wait(file, fl);
1513		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1514	}
1515
1516out:
1517	mutex_unlock(&fp->f_fl_mutex);
1518	return error;
1519}
1520
1521static void do_unflock(struct file *file, struct file_lock *fl)
1522{
1523	struct gfs2_file *fp = file->private_data;
1524	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1525
1526	mutex_lock(&fp->f_fl_mutex);
1527	locks_lock_file_wait(file, fl);
1528	if (gfs2_holder_initialized(fl_gh)) {
1529		gfs2_glock_dq(fl_gh);
1530		__flock_holder_uninit(file, fl_gh);
1531	}
1532	mutex_unlock(&fp->f_fl_mutex);
1533}
1534
1535/**
1536 * gfs2_flock - acquire/release a flock lock on a file
1537 * @file: the file pointer
1538 * @cmd: either modify or retrieve lock state, possibly wait
1539 * @fl: type and range of lock
1540 *
1541 * Returns: errno
1542 */
1543
1544static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1545{
1546	if (!(fl->fl_flags & FL_FLOCK))
1547		return -ENOLCK;
 
 
1548
1549	if (fl->fl_type == F_UNLCK) {
1550		do_unflock(file, fl);
1551		return 0;
1552	} else {
1553		return do_flock(file, cmd, fl);
1554	}
1555}
1556
1557const struct file_operations gfs2_file_fops = {
1558	.llseek		= gfs2_llseek,
1559	.read_iter	= gfs2_file_read_iter,
1560	.write_iter	= gfs2_file_write_iter,
1561	.iopoll		= iocb_bio_iopoll,
1562	.unlocked_ioctl	= gfs2_ioctl,
1563	.compat_ioctl	= gfs2_compat_ioctl,
1564	.mmap		= gfs2_mmap,
1565	.open		= gfs2_open,
1566	.release	= gfs2_release,
1567	.fsync		= gfs2_fsync,
1568	.lock		= gfs2_lock,
1569	.flock		= gfs2_flock,
1570	.splice_read	= generic_file_splice_read,
1571	.splice_write	= gfs2_file_splice_write,
1572	.setlease	= simple_nosetlease,
1573	.fallocate	= gfs2_fallocate,
1574};
1575
1576const struct file_operations gfs2_dir_fops = {
1577	.iterate_shared	= gfs2_readdir,
1578	.unlocked_ioctl	= gfs2_ioctl,
1579	.compat_ioctl	= gfs2_compat_ioctl,
1580	.open		= gfs2_open,
1581	.release	= gfs2_release,
1582	.fsync		= gfs2_fsync,
1583	.lock		= gfs2_lock,
1584	.flock		= gfs2_flock,
1585	.llseek		= default_llseek,
1586};
1587
1588#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1589
1590const struct file_operations gfs2_file_fops_nolock = {
1591	.llseek		= gfs2_llseek,
1592	.read_iter	= gfs2_file_read_iter,
1593	.write_iter	= gfs2_file_write_iter,
1594	.iopoll		= iocb_bio_iopoll,
1595	.unlocked_ioctl	= gfs2_ioctl,
1596	.compat_ioctl	= gfs2_compat_ioctl,
1597	.mmap		= gfs2_mmap,
1598	.open		= gfs2_open,
1599	.release	= gfs2_release,
1600	.fsync		= gfs2_fsync,
1601	.splice_read	= generic_file_splice_read,
1602	.splice_write	= gfs2_file_splice_write,
1603	.setlease	= generic_setlease,
1604	.fallocate	= gfs2_fallocate,
1605};
1606
1607const struct file_operations gfs2_dir_fops_nolock = {
1608	.iterate_shared	= gfs2_readdir,
1609	.unlocked_ioctl	= gfs2_ioctl,
1610	.compat_ioctl	= gfs2_compat_ioctl,
1611	.open		= gfs2_open,
1612	.release	= gfs2_release,
1613	.fsync		= gfs2_fsync,
1614	.llseek		= default_llseek,
1615};
1616
v4.17
 
   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 <linux/uaccess.h>
  26#include <linux/dlm.h>
  27#include <linux/dlm_plock.h>
  28#include <linux/delay.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 * @whence: 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 whence)
  57{
  58	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
  59	struct gfs2_holder i_gh;
  60	loff_t error;
  61
  62	switch (whence) {
  63	case SEEK_END:
  64		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
  65					   &i_gh);
  66		if (!error) {
  67			error = generic_file_llseek(file, offset, whence);
  68			gfs2_glock_dq_uninit(&i_gh);
  69		}
  70		break;
  71
  72	case SEEK_DATA:
  73		error = gfs2_seek_data(file, offset);
  74		break;
  75
  76	case SEEK_HOLE:
  77		error = gfs2_seek_hole(file, offset);
  78		break;
  79
  80	case SEEK_CUR:
  81	case SEEK_SET:
  82		/*
  83		 * These don't reference inode->i_size and don't depend on the
  84		 * block mapping, so we don't need the glock.
  85		 */
  86		error = generic_file_llseek(file, offset, whence);
  87		break;
  88	default:
  89		error = -EINVAL;
  90	}
  91
  92	return error;
  93}
  94
  95/**
  96 * gfs2_readdir - Iterator for a directory
  97 * @file: The directory to read from
  98 * @ctx: What to feed directory entries to
  99 *
 100 * Returns: errno
 101 */
 102
 103static int gfs2_readdir(struct file *file, struct dir_context *ctx)
 104{
 105	struct inode *dir = file->f_mapping->host;
 106	struct gfs2_inode *dip = GFS2_I(dir);
 107	struct gfs2_holder d_gh;
 108	int error;
 109
 110	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
 111	if (error)
 112		return error;
 113
 114	error = gfs2_dir_read(dir, ctx, &file->f_ra);
 115
 116	gfs2_glock_dq_uninit(&d_gh);
 117
 118	return error;
 119}
 120
 121/**
 122 * fsflag_gfs2flag
 123 *
 124 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
 125 * and to GFS2_DIF_JDATA for non-directories.
 126 */
 127static struct {
 128	u32 fsflag;
 129	u32 gfsflag;
 130} fsflag_gfs2flag[] = {
 131	{FS_SYNC_FL, GFS2_DIF_SYNC},
 132	{FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
 133	{FS_APPEND_FL, GFS2_DIF_APPENDONLY},
 134	{FS_NOATIME_FL, GFS2_DIF_NOATIME},
 135	{FS_INDEX_FL, GFS2_DIF_EXHASH},
 136	{FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
 137	{FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
 138};
 139
 140static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 141{
 142	struct inode *inode = file_inode(filp);
 143	struct gfs2_inode *ip = GFS2_I(inode);
 144	struct gfs2_holder gh;
 145	int i, error;
 146	u32 gfsflags, fsflags = 0;
 
 
 
 147
 148	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 149	error = gfs2_glock_nq(&gh);
 150	if (error)
 151		goto out_uninit;
 152
 153	gfsflags = ip->i_diskflags;
 154	if (S_ISDIR(inode->i_mode))
 155		gfsflags &= ~GFS2_DIF_JDATA;
 156	else
 157		gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
 158	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
 159		if (gfsflags & fsflag_gfs2flag[i].gfsflag)
 160			fsflags |= fsflag_gfs2flag[i].fsflag;
 161
 162	if (put_user(fsflags, ptr))
 163		error = -EFAULT;
 164
 165	gfs2_glock_dq(&gh);
 166out_uninit:
 167	gfs2_holder_uninit(&gh);
 168	return error;
 169}
 170
 171void gfs2_set_inode_flags(struct inode *inode)
 172{
 173	struct gfs2_inode *ip = GFS2_I(inode);
 174	unsigned int flags = inode->i_flags;
 175
 176	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
 177	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
 178		flags |= S_NOSEC;
 179	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
 180		flags |= S_IMMUTABLE;
 181	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
 182		flags |= S_APPEND;
 183	if (ip->i_diskflags & GFS2_DIF_NOATIME)
 184		flags |= S_NOATIME;
 185	if (ip->i_diskflags & GFS2_DIF_SYNC)
 186		flags |= S_SYNC;
 187	inode->i_flags = flags;
 188}
 189
 190/* Flags that can be set by user space */
 191#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
 192			     GFS2_DIF_IMMUTABLE|		\
 193			     GFS2_DIF_APPENDONLY|		\
 194			     GFS2_DIF_NOATIME|			\
 195			     GFS2_DIF_SYNC|			\
 196			     GFS2_DIF_TOPDIR|			\
 197			     GFS2_DIF_INHERIT_JDATA)
 198
 199/**
 200 * do_gfs2_set_flags - set flags on an inode
 201 * @filp: file pointer
 202 * @reqflags: The flags to set
 203 * @mask: Indicates which flags are valid
 204 *
 205 */
 206static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
 207{
 208	struct inode *inode = file_inode(filp);
 209	struct gfs2_inode *ip = GFS2_I(inode);
 210	struct gfs2_sbd *sdp = GFS2_SB(inode);
 211	struct buffer_head *bh;
 212	struct gfs2_holder gh;
 213	int error;
 214	u32 new_flags, flags;
 215
 216	error = mnt_want_write_file(filp);
 217	if (error)
 218		return error;
 219
 220	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 221	if (error)
 222		goto out_drop_write;
 223
 224	error = -EACCES;
 225	if (!inode_owner_or_capable(inode))
 226		goto out;
 227
 228	error = 0;
 229	flags = ip->i_diskflags;
 230	new_flags = (flags & ~mask) | (reqflags & mask);
 231	if ((new_flags ^ flags) == 0)
 232		goto out;
 233
 234	error = -EPERM;
 235	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
 236		goto out;
 237	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
 238		goto out;
 239	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
 240	    !capable(CAP_LINUX_IMMUTABLE))
 241		goto out;
 242	if (!IS_IMMUTABLE(inode)) {
 243		error = gfs2_permission(inode, MAY_WRITE);
 244		if (error)
 245			goto out;
 246	}
 247	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
 248		if (new_flags & GFS2_DIF_JDATA)
 249			gfs2_log_flush(sdp, ip->i_gl,
 250				       GFS2_LOG_HEAD_FLUSH_NORMAL |
 251				       GFS2_LFC_SET_FLAGS);
 252		error = filemap_fdatawrite(inode->i_mapping);
 253		if (error)
 254			goto out;
 255		error = filemap_fdatawait(inode->i_mapping);
 256		if (error)
 257			goto out;
 258		if (new_flags & GFS2_DIF_JDATA)
 259			gfs2_ordered_del_inode(ip);
 260	}
 261	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 262	if (error)
 263		goto out;
 264	error = gfs2_meta_inode_buffer(ip, &bh);
 265	if (error)
 266		goto out_trans_end;
 267	inode->i_ctime = current_time(inode);
 268	gfs2_trans_add_meta(ip->i_gl, bh);
 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_file(filp);
 280	return error;
 281}
 282
 283static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
 
 284{
 285	struct inode *inode = file_inode(filp);
 286	u32 fsflags, gfsflags = 0;
 287	u32 mask;
 288	int i;
 289
 290	if (get_user(fsflags, ptr))
 291		return -EFAULT;
 
 
 
 292
 293	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
 294		if (fsflags & fsflag_gfs2flag[i].fsflag) {
 295			fsflags &= ~fsflag_gfs2flag[i].fsflag;
 296			gfsflags |= fsflag_gfs2flag[i].gfsflag;
 297		}
 298	}
 299	if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
 300		return -EINVAL;
 301
 302	mask = GFS2_FLAGS_USER_SET;
 303	if (S_ISDIR(inode->i_mode)) {
 304		mask &= ~GFS2_DIF_JDATA;
 305	} else {
 306		/* The GFS2_DIF_TOPDIR flag is only valid for directories. */
 307		if (gfsflags & GFS2_DIF_TOPDIR)
 308			return -EINVAL;
 309		mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
 310	}
 311
 312	return do_gfs2_set_flags(filp, gfsflags, mask);
 
 
 
 
 
 
 
 
 
 
 
 313}
 314
 315static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 316{
 317	switch(cmd) {
 318	case FS_IOC_GETFLAGS:
 319		return gfs2_get_flags(filp, (u32 __user *)arg);
 320	case FS_IOC_SETFLAGS:
 321		return gfs2_set_flags(filp, (u32 __user *)arg);
 322	case FITRIM:
 323		return gfs2_fitrim(filp, (void __user *)arg);
 
 
 324	}
 
 325	return -ENOTTY;
 326}
 327
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 328/**
 329 * gfs2_size_hint - Give a hint to the size of a write request
 330 * @filep: The struct file
 331 * @offset: The file offset of the write
 332 * @size: The length of the write
 333 *
 334 * When we are about to do a write, this function records the total
 335 * write size in order to provide a suitable hint to the lower layers
 336 * about how many blocks will be required.
 337 *
 338 */
 339
 340static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
 341{
 342	struct inode *inode = file_inode(filep);
 343	struct gfs2_sbd *sdp = GFS2_SB(inode);
 344	struct gfs2_inode *ip = GFS2_I(inode);
 345	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
 346	int hint = min_t(size_t, INT_MAX, blks);
 347
 348	if (hint > atomic_read(&ip->i_res.rs_sizehint))
 349		atomic_set(&ip->i_res.rs_sizehint, hint);
 350}
 351
 352/**
 353 * gfs2_allocate_page_backing - Use bmap to allocate blocks
 354 * @page: The (locked) page to allocate backing for
 
 355 *
 356 * We try to allocate all the blocks required for the page in
 357 * one go. This might fail for various reasons, so we keep
 358 * trying until all the blocks to back this page are allocated.
 359 * If some of the blocks are already allocated, thats ok too.
 360 */
 361
 362static int gfs2_allocate_page_backing(struct page *page)
 363{
 364	struct inode *inode = page->mapping->host;
 365	struct buffer_head bh;
 366	unsigned long size = PAGE_SIZE;
 367	u64 lblock = page->index << (PAGE_SHIFT - inode->i_blkbits);
 368
 369	do {
 370		bh.b_state = 0;
 371		bh.b_size = size;
 372		gfs2_block_map(inode, lblock, &bh, 1);
 373		if (!buffer_mapped(&bh))
 374			return -EIO;
 375		size -= bh.b_size;
 376		lblock += (bh.b_size >> inode->i_blkbits);
 377	} while(size > 0);
 
 
 
 
 378	return 0;
 379}
 380
 381/**
 382 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
 383 * @vma: The virtual memory area
 384 * @vmf: The virtual memory fault containing the page to become writable
 385 *
 386 * When the page becomes writable, we need to ensure that we have
 387 * blocks allocated on disk to back that page.
 388 */
 389
 390static int gfs2_page_mkwrite(struct vm_fault *vmf)
 391{
 392	struct page *page = vmf->page;
 393	struct inode *inode = file_inode(vmf->vma->vm_file);
 394	struct gfs2_inode *ip = GFS2_I(inode);
 395	struct gfs2_sbd *sdp = GFS2_SB(inode);
 396	struct gfs2_alloc_parms ap = { .aflags = 0, };
 397	unsigned long last_index;
 398	u64 pos = page->index << PAGE_SHIFT;
 399	unsigned int data_blocks, ind_blocks, rblocks;
 
 400	struct gfs2_holder gh;
 
 401	loff_t size;
 402	int ret;
 403
 404	sb_start_pagefault(inode->i_sb);
 405
 406	ret = gfs2_rsqa_alloc(ip);
 407	if (ret)
 408		goto out;
 409
 410	gfs2_size_hint(vmf->vma->vm_file, pos, PAGE_SIZE);
 411
 412	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 413	ret = gfs2_glock_nq(&gh);
 414	if (ret)
 
 415		goto out_uninit;
 
 
 
 
 
 
 
 
 416
 417	/* Update file times before taking page lock */
 418	file_update_time(vmf->vma->vm_file);
 419
 
 
 
 
 
 
 
 
 420	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
 421	set_bit(GIF_SW_PAGED, &ip->i_flags);
 422
 423	if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) {
 
 
 
 
 
 
 424		lock_page(page);
 425		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
 426			ret = -EAGAIN;
 427			unlock_page(page);
 428		}
 429		goto out_unlock;
 430	}
 431
 432	ret = gfs2_rindex_update(sdp);
 433	if (ret)
 
 434		goto out_unlock;
 
 435
 436	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
 437	ap.target = data_blocks + ind_blocks;
 438	ret = gfs2_quota_lock_check(ip, &ap);
 439	if (ret)
 
 440		goto out_unlock;
 441	ret = gfs2_inplace_reserve(ip, &ap);
 442	if (ret)
 
 
 443		goto out_quota_unlock;
 
 444
 445	rblocks = RES_DINODE + ind_blocks;
 446	if (gfs2_is_jdata(ip))
 447		rblocks += data_blocks ? data_blocks : 1;
 448	if (ind_blocks || data_blocks) {
 449		rblocks += RES_STATFS + RES_QUOTA;
 450		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
 451	}
 452	ret = gfs2_trans_begin(sdp, rblocks, 0);
 453	if (ret)
 
 454		goto out_trans_fail;
 
 
 
 
 
 
 
 
 
 
 455
 456	lock_page(page);
 457	ret = -EINVAL;
 458	size = i_size_read(inode);
 459	last_index = (size - 1) >> PAGE_SHIFT;
 460	/* Check page index against inode size */
 461	if (size == 0 || (page->index > last_index))
 462		goto out_trans_end;
 463
 464	ret = -EAGAIN;
 465	/* If truncated, we must retry the operation, we may have raced
 466	 * with the glock demotion code.
 467	 */
 468	if (!PageUptodate(page) || page->mapping != inode->i_mapping)
 469		goto out_trans_end;
 
 
 470
 471	/* Unstuff, if required, and allocate backing blocks for page */
 472	ret = 0;
 473	if (gfs2_is_stuffed(ip))
 474		ret = gfs2_unstuff_dinode(ip, page);
 475	if (ret == 0)
 476		ret = gfs2_allocate_page_backing(page);
 477
 
 
 
 478out_trans_end:
 479	if (ret)
 480		unlock_page(page);
 481	gfs2_trans_end(sdp);
 482out_trans_fail:
 483	gfs2_inplace_release(ip);
 484out_quota_unlock:
 485	gfs2_quota_unlock(ip);
 486out_unlock:
 487	gfs2_glock_dq(&gh);
 488out_uninit:
 489	gfs2_holder_uninit(&gh);
 490	if (ret == 0) {
 491		set_page_dirty(page);
 492		wait_for_stable_page(page);
 493	}
 494out:
 495	sb_end_pagefault(inode->i_sb);
 496	return block_page_mkwrite_return(ret);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 497}
 498
 499static const struct vm_operations_struct gfs2_vm_ops = {
 500	.fault = filemap_fault,
 501	.map_pages = filemap_map_pages,
 502	.page_mkwrite = gfs2_page_mkwrite,
 503};
 504
 505/**
 506 * gfs2_mmap -
 507 * @file: The file to map
 508 * @vma: The VMA which described the mapping
 509 *
 510 * There is no need to get a lock here unless we should be updating
 511 * atime. We ignore any locking errors since the only consequence is
 512 * a missed atime update (which will just be deferred until later).
 513 *
 514 * Returns: 0
 515 */
 516
 517static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 518{
 519	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 520
 521	if (!(file->f_flags & O_NOATIME) &&
 522	    !IS_NOATIME(&ip->i_inode)) {
 523		struct gfs2_holder i_gh;
 524		int error;
 525
 526		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 527					   &i_gh);
 528		if (error)
 529			return error;
 530		/* grab lock to update inode */
 531		gfs2_glock_dq_uninit(&i_gh);
 532		file_accessed(file);
 533	}
 534	vma->vm_ops = &gfs2_vm_ops;
 535
 536	return 0;
 537}
 538
 539/**
 540 * gfs2_open_common - This is common to open and atomic_open
 541 * @inode: The inode being opened
 542 * @file: The file being opened
 543 *
 544 * This maybe called under a glock or not depending upon how it has
 545 * been called. We must always be called under a glock for regular
 546 * files, however. For other file types, it does not matter whether
 547 * we hold the glock or not.
 548 *
 549 * Returns: Error code or 0 for success
 550 */
 551
 552int gfs2_open_common(struct inode *inode, struct file *file)
 553{
 554	struct gfs2_file *fp;
 555	int ret;
 556
 557	if (S_ISREG(inode->i_mode)) {
 558		ret = generic_file_open(inode, file);
 559		if (ret)
 560			return ret;
 561	}
 562
 563	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
 564	if (!fp)
 565		return -ENOMEM;
 566
 567	mutex_init(&fp->f_fl_mutex);
 568
 569	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
 570	file->private_data = fp;
 
 
 
 
 
 571	return 0;
 
 
 
 
 
 572}
 573
 574/**
 575 * gfs2_open - open a file
 576 * @inode: the inode to open
 577 * @file: the struct file for this opening
 578 *
 579 * After atomic_open, this function is only used for opening files
 580 * which are already cached. We must still get the glock for regular
 581 * files to ensure that we have the file size uptodate for the large
 582 * file check which is in the common code. That is only an issue for
 583 * regular files though.
 584 *
 585 * Returns: errno
 586 */
 587
 588static int gfs2_open(struct inode *inode, struct file *file)
 589{
 590	struct gfs2_inode *ip = GFS2_I(inode);
 591	struct gfs2_holder i_gh;
 592	int error;
 593	bool need_unlock = false;
 594
 595	if (S_ISREG(ip->i_inode.i_mode)) {
 596		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 597					   &i_gh);
 598		if (error)
 599			return error;
 600		need_unlock = true;
 601	}
 602
 603	error = gfs2_open_common(inode, file);
 604
 605	if (need_unlock)
 606		gfs2_glock_dq_uninit(&i_gh);
 607
 608	return error;
 609}
 610
 611/**
 612 * gfs2_release - called to close a struct file
 613 * @inode: the inode the struct file belongs to
 614 * @file: the struct file being closed
 615 *
 616 * Returns: errno
 617 */
 618
 619static int gfs2_release(struct inode *inode, struct file *file)
 620{
 621	struct gfs2_inode *ip = GFS2_I(inode);
 622
 623	kfree(file->private_data);
 624	file->private_data = NULL;
 625
 626	if (!(file->f_mode & FMODE_WRITE))
 627		return 0;
 628
 629	gfs2_rsqa_delete(ip, &inode->i_writecount);
 
 630	return 0;
 631}
 632
 633/**
 634 * gfs2_fsync - sync the dirty data for a file (across the cluster)
 635 * @file: the file that points to the dentry
 636 * @start: the start position in the file to sync
 637 * @end: the end position in the file to sync
 638 * @datasync: set if we can ignore timestamp changes
 639 *
 640 * We split the data flushing here so that we don't wait for the data
 641 * until after we've also sent the metadata to disk. Note that for
 642 * data=ordered, we will write & wait for the data at the log flush
 643 * stage anyway, so this is unlikely to make much of a difference
 644 * except in the data=writeback case.
 645 *
 646 * If the fdatawrite fails due to any reason except -EIO, we will
 647 * continue the remainder of the fsync, although we'll still report
 648 * the error at the end. This is to match filemap_write_and_wait_range()
 649 * behaviour.
 650 *
 651 * Returns: errno
 652 */
 653
 654static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 655		      int datasync)
 656{
 657	struct address_space *mapping = file->f_mapping;
 658	struct inode *inode = mapping->host;
 659	int sync_state = inode->i_state & I_DIRTY_ALL;
 660	struct gfs2_inode *ip = GFS2_I(inode);
 661	int ret = 0, ret1 = 0;
 662
 663	if (mapping->nrpages) {
 664		ret1 = filemap_fdatawrite_range(mapping, start, end);
 665		if (ret1 == -EIO)
 666			return ret1;
 667	}
 668
 669	if (!gfs2_is_jdata(ip))
 670		sync_state &= ~I_DIRTY_PAGES;
 671	if (datasync)
 672		sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
 673
 674	if (sync_state) {
 675		ret = sync_inode_metadata(inode, 1);
 676		if (ret)
 677			return ret;
 678		if (gfs2_is_jdata(ip))
 679			ret = file_write_and_wait(file);
 680		if (ret)
 681			return ret;
 682		gfs2_ail_flush(ip->i_gl, 1);
 683	}
 684
 685	if (mapping->nrpages)
 686		ret = file_fdatawait_range(file, start, end);
 687
 688	return ret ? ret : ret1;
 689}
 690
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 691/**
 692 * gfs2_file_write_iter - Perform a write to a file
 693 * @iocb: The io context
 694 * @iov: The data to write
 695 * @nr_segs: Number of @iov segments
 696 * @pos: The file position
 697 *
 698 * We have to do a lock/unlock here to refresh the inode size for
 699 * O_APPEND writes, otherwise we can land up writing at the wrong
 700 * offset. There is still a race, but provided the app is using its
 701 * own file locking, this will make O_APPEND work as expected.
 702 *
 703 */
 704
 705static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 706{
 707	struct file *file = iocb->ki_filp;
 708	struct gfs2_inode *ip = GFS2_I(file_inode(file));
 709	int ret;
 710
 711	ret = gfs2_rsqa_alloc(ip);
 712	if (ret)
 713		return ret;
 714
 715	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
 716
 717	if (iocb->ki_flags & IOCB_APPEND) {
 718		struct gfs2_holder gh;
 719
 720		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 721		if (ret)
 722			return ret;
 723		gfs2_glock_dq_uninit(&gh);
 724	}
 725
 726	return generic_file_write_iter(iocb, from);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 727}
 728
 729static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
 730			   int mode)
 731{
 732	struct super_block *sb = inode->i_sb;
 733	struct gfs2_inode *ip = GFS2_I(inode);
 734	loff_t end = offset + len;
 735	struct buffer_head *dibh;
 736	struct iomap iomap;
 737	int error;
 738
 739	error = gfs2_meta_inode_buffer(ip, &dibh);
 740	if (unlikely(error))
 741		return error;
 742
 743	gfs2_trans_add_meta(ip->i_gl, dibh);
 744
 745	if (gfs2_is_stuffed(ip)) {
 746		error = gfs2_unstuff_dinode(ip, NULL);
 747		if (unlikely(error))
 748			goto out;
 749	}
 750
 751	while (offset < end) {
 752		error = gfs2_iomap_begin(inode, offset, end - offset,
 753					 IOMAP_WRITE, &iomap);
 
 754		if (error)
 755			goto out;
 756		offset = iomap.offset + iomap.length;
 757		if (iomap.type != IOMAP_HOLE)
 758			continue;
 759		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
 760					 iomap.length >> inode->i_blkbits,
 761					 GFP_NOFS);
 762		if (error) {
 763			fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
 764			goto out;
 765		}
 766	}
 767out:
 768	brelse(dibh);
 769	return error;
 770}
 
 771/**
 772 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
 773 *                     blocks, determine how many bytes can be written.
 774 * @ip:          The inode in question.
 775 * @len:         Max cap of bytes. What we return in *len must be <= this.
 776 * @data_blocks: Compute and return the number of data blocks needed
 777 * @ind_blocks:  Compute and return the number of indirect blocks needed
 778 * @max_blocks:  The total blocks available to work with.
 779 *
 780 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
 781 */
 782static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
 783			    unsigned int *data_blocks, unsigned int *ind_blocks,
 784			    unsigned int max_blocks)
 785{
 786	loff_t max = *len;
 787	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 788	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
 789
 790	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
 791		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
 792		max_data -= tmp;
 793	}
 794
 795	*data_blocks = max_data;
 796	*ind_blocks = max_blocks - max_data;
 797	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
 798	if (*len > max) {
 799		*len = max;
 800		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
 801	}
 802}
 803
 804static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 805{
 806	struct inode *inode = file_inode(file);
 807	struct gfs2_sbd *sdp = GFS2_SB(inode);
 808	struct gfs2_inode *ip = GFS2_I(inode);
 809	struct gfs2_alloc_parms ap = { .aflags = 0, };
 810	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
 811	loff_t bytes, max_bytes, max_blks;
 812	int error;
 813	const loff_t pos = offset;
 814	const loff_t count = len;
 815	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
 816	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
 817	loff_t max_chunk_size = UINT_MAX & bsize_mask;
 818
 819	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
 820
 821	offset &= bsize_mask;
 822
 823	len = next - offset;
 824	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
 825	if (!bytes)
 826		bytes = UINT_MAX;
 827	bytes &= bsize_mask;
 828	if (bytes == 0)
 829		bytes = sdp->sd_sb.sb_bsize;
 830
 831	gfs2_size_hint(file, offset, len);
 832
 833	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
 834	ap.min_target = data_blocks + ind_blocks;
 835
 836	while (len > 0) {
 837		if (len < bytes)
 838			bytes = len;
 839		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
 840			len -= bytes;
 841			offset += bytes;
 842			continue;
 843		}
 844
 845		/* We need to determine how many bytes we can actually
 846		 * fallocate without exceeding quota or going over the
 847		 * end of the fs. We start off optimistically by assuming
 848		 * we can write max_bytes */
 849		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
 850
 851		/* Since max_bytes is most likely a theoretical max, we
 852		 * calculate a more realistic 'bytes' to serve as a good
 853		 * starting point for the number of bytes we may be able
 854		 * to write */
 855		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
 856		ap.target = data_blocks + ind_blocks;
 857
 858		error = gfs2_quota_lock_check(ip, &ap);
 859		if (error)
 860			return error;
 861		/* ap.allowed tells us how many blocks quota will allow
 862		 * us to write. Check if this reduces max_blks */
 863		max_blks = UINT_MAX;
 864		if (ap.allowed)
 865			max_blks = ap.allowed;
 866
 867		error = gfs2_inplace_reserve(ip, &ap);
 868		if (error)
 869			goto out_qunlock;
 870
 871		/* check if the selected rgrp limits our max_blks further */
 872		if (ap.allowed && ap.allowed < max_blks)
 873			max_blks = ap.allowed;
 874
 875		/* Almost done. Calculate bytes that can be written using
 876		 * max_blks. We also recompute max_bytes, data_blocks and
 877		 * ind_blocks */
 878		calc_max_reserv(ip, &max_bytes, &data_blocks,
 879				&ind_blocks, max_blks);
 880
 881		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
 882			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
 883		if (gfs2_is_jdata(ip))
 884			rblocks += data_blocks ? data_blocks : 1;
 885
 886		error = gfs2_trans_begin(sdp, rblocks,
 887					 PAGE_SIZE/sdp->sd_sb.sb_bsize);
 888		if (error)
 889			goto out_trans_fail;
 890
 891		error = fallocate_chunk(inode, offset, max_bytes, mode);
 892		gfs2_trans_end(sdp);
 893
 894		if (error)
 895			goto out_trans_fail;
 896
 897		len -= max_bytes;
 898		offset += max_bytes;
 899		gfs2_inplace_release(ip);
 900		gfs2_quota_unlock(ip);
 901	}
 902
 903	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) {
 904		i_size_write(inode, pos + count);
 905		file_update_time(file);
 906		mark_inode_dirty(inode);
 907	}
 908
 909	if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
 910		return vfs_fsync_range(file, pos, pos + count - 1,
 911			       (file->f_flags & __O_SYNC) ? 0 : 1);
 912	return 0;
 913
 914out_trans_fail:
 915	gfs2_inplace_release(ip);
 916out_qunlock:
 917	gfs2_quota_unlock(ip);
 918	return error;
 919}
 920
 921static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 922{
 923	struct inode *inode = file_inode(file);
 924	struct gfs2_sbd *sdp = GFS2_SB(inode);
 925	struct gfs2_inode *ip = GFS2_I(inode);
 926	struct gfs2_holder gh;
 927	int ret;
 928
 929	if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
 930		return -EOPNOTSUPP;
 931	/* fallocate is needed by gfs2_grow to reserve space in the rindex */
 932	if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
 933		return -EOPNOTSUPP;
 934
 935	inode_lock(inode);
 936
 937	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 938	ret = gfs2_glock_nq(&gh);
 939	if (ret)
 940		goto out_uninit;
 941
 942	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
 943	    (offset + len) > inode->i_size) {
 944		ret = inode_newsize_ok(inode, offset + len);
 945		if (ret)
 946			goto out_unlock;
 947	}
 948
 949	ret = get_write_access(inode);
 950	if (ret)
 951		goto out_unlock;
 952
 953	if (mode & FALLOC_FL_PUNCH_HOLE) {
 954		ret = __gfs2_punch_hole(file, offset, len);
 955	} else {
 956		ret = gfs2_rsqa_alloc(ip);
 957		if (ret)
 958			goto out_putw;
 959
 960		ret = __gfs2_fallocate(file, mode, offset, len);
 961
 962		if (ret)
 963			gfs2_rs_deltree(&ip->i_res);
 964	}
 965
 966out_putw:
 967	put_write_access(inode);
 968out_unlock:
 969	gfs2_glock_dq(&gh);
 970out_uninit:
 971	gfs2_holder_uninit(&gh);
 972	inode_unlock(inode);
 973	return ret;
 974}
 975
 976static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
 977				      struct file *out, loff_t *ppos,
 978				      size_t len, unsigned int flags)
 979{
 980	int error;
 981	struct gfs2_inode *ip = GFS2_I(out->f_mapping->host);
 982
 983	error = gfs2_rsqa_alloc(ip);
 984	if (error)
 985		return (ssize_t)error;
 986
 987	gfs2_size_hint(out, *ppos, len);
 988
 989	return iter_file_splice_write(pipe, out, ppos, len, flags);
 
 990}
 991
 992#ifdef CONFIG_GFS2_FS_LOCKING_DLM
 993
 994/**
 995 * gfs2_lock - acquire/release a posix lock on a file
 996 * @file: the file pointer
 997 * @cmd: either modify or retrieve lock state, possibly wait
 998 * @fl: type and range of lock
 999 *
1000 * Returns: errno
1001 */
1002
1003static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1004{
1005	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1006	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1007	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1008
1009	if (!(fl->fl_flags & FL_POSIX))
1010		return -ENOLCK;
1011	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1012		return -ENOLCK;
1013
1014	if (cmd == F_CANCELLK) {
1015		/* Hack: */
1016		cmd = F_SETLK;
1017		fl->fl_type = F_UNLCK;
1018	}
1019	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1020		if (fl->fl_type == F_UNLCK)
1021			locks_lock_file_wait(file, fl);
1022		return -EIO;
1023	}
1024	if (IS_GETLK(cmd))
1025		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1026	else if (fl->fl_type == F_UNLCK)
1027		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1028	else
1029		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1030}
1031
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1032static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1033{
1034	struct gfs2_file *fp = file->private_data;
1035	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1036	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1037	struct gfs2_glock *gl;
1038	unsigned int state;
1039	u16 flags;
1040	int error = 0;
1041	int sleeptime;
1042
1043	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1044	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
 
 
1045
1046	mutex_lock(&fp->f_fl_mutex);
1047
1048	if (gfs2_holder_initialized(fl_gh)) {
 
1049		if (fl_gh->gh_state == state)
1050			goto out;
1051		locks_lock_file_wait(file,
1052				     &(struct file_lock) {
1053					     .fl_type = F_UNLCK,
1054					     .fl_flags = FL_FLOCK
1055				     });
1056		gfs2_glock_dq(fl_gh);
1057		gfs2_holder_reinit(state, flags, fl_gh);
1058	} else {
1059		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1060				       &gfs2_flock_glops, CREATE, &gl);
1061		if (error)
1062			goto out;
 
1063		gfs2_holder_init(gl, state, flags, fl_gh);
 
1064		gfs2_glock_put(gl);
1065	}
1066	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1067		error = gfs2_glock_nq(fl_gh);
1068		if (error != GLR_TRYFAILED)
1069			break;
1070		fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1071		fl_gh->gh_error = 0;
1072		msleep(sleeptime);
1073	}
1074	if (error) {
1075		gfs2_holder_uninit(fl_gh);
1076		if (error == GLR_TRYFAILED)
1077			error = -EAGAIN;
1078	} else {
1079		error = locks_lock_file_wait(file, fl);
1080		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1081	}
1082
1083out:
1084	mutex_unlock(&fp->f_fl_mutex);
1085	return error;
1086}
1087
1088static void do_unflock(struct file *file, struct file_lock *fl)
1089{
1090	struct gfs2_file *fp = file->private_data;
1091	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1092
1093	mutex_lock(&fp->f_fl_mutex);
1094	locks_lock_file_wait(file, fl);
1095	if (gfs2_holder_initialized(fl_gh)) {
1096		gfs2_glock_dq(fl_gh);
1097		gfs2_holder_uninit(fl_gh);
1098	}
1099	mutex_unlock(&fp->f_fl_mutex);
1100}
1101
1102/**
1103 * gfs2_flock - acquire/release a flock lock on a file
1104 * @file: the file pointer
1105 * @cmd: either modify or retrieve lock state, possibly wait
1106 * @fl: type and range of lock
1107 *
1108 * Returns: errno
1109 */
1110
1111static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1112{
1113	if (!(fl->fl_flags & FL_FLOCK))
1114		return -ENOLCK;
1115	if (fl->fl_type & LOCK_MAND)
1116		return -EOPNOTSUPP;
1117
1118	if (fl->fl_type == F_UNLCK) {
1119		do_unflock(file, fl);
1120		return 0;
1121	} else {
1122		return do_flock(file, cmd, fl);
1123	}
1124}
1125
1126const struct file_operations gfs2_file_fops = {
1127	.llseek		= gfs2_llseek,
1128	.read_iter	= generic_file_read_iter,
1129	.write_iter	= gfs2_file_write_iter,
 
1130	.unlocked_ioctl	= gfs2_ioctl,
 
1131	.mmap		= gfs2_mmap,
1132	.open		= gfs2_open,
1133	.release	= gfs2_release,
1134	.fsync		= gfs2_fsync,
1135	.lock		= gfs2_lock,
1136	.flock		= gfs2_flock,
1137	.splice_read	= generic_file_splice_read,
1138	.splice_write	= gfs2_file_splice_write,
1139	.setlease	= simple_nosetlease,
1140	.fallocate	= gfs2_fallocate,
1141};
1142
1143const struct file_operations gfs2_dir_fops = {
1144	.iterate_shared	= gfs2_readdir,
1145	.unlocked_ioctl	= gfs2_ioctl,
 
1146	.open		= gfs2_open,
1147	.release	= gfs2_release,
1148	.fsync		= gfs2_fsync,
1149	.lock		= gfs2_lock,
1150	.flock		= gfs2_flock,
1151	.llseek		= default_llseek,
1152};
1153
1154#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1155
1156const struct file_operations gfs2_file_fops_nolock = {
1157	.llseek		= gfs2_llseek,
1158	.read_iter	= generic_file_read_iter,
1159	.write_iter	= gfs2_file_write_iter,
 
1160	.unlocked_ioctl	= gfs2_ioctl,
 
1161	.mmap		= gfs2_mmap,
1162	.open		= gfs2_open,
1163	.release	= gfs2_release,
1164	.fsync		= gfs2_fsync,
1165	.splice_read	= generic_file_splice_read,
1166	.splice_write	= gfs2_file_splice_write,
1167	.setlease	= generic_setlease,
1168	.fallocate	= gfs2_fallocate,
1169};
1170
1171const struct file_operations gfs2_dir_fops_nolock = {
1172	.iterate_shared	= gfs2_readdir,
1173	.unlocked_ioctl	= gfs2_ioctl,
 
1174	.open		= gfs2_open,
1175	.release	= gfs2_release,
1176	.fsync		= gfs2_fsync,
1177	.llseek		= default_llseek,
1178};
1179