Linux Audio

Check our new training course

Buildroot integration, development and maintenance

Need a Buildroot system for your embedded project?
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
v6.8
   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/filelock.h>
  19#include <linux/gfs2_ondisk.h>
  20#include <linux/falloc.h>
  21#include <linux/swap.h>
  22#include <linux/crc32.h>
  23#include <linux/writeback.h>
  24#include <linux/uaccess.h>
  25#include <linux/dlm.h>
  26#include <linux/dlm_plock.h>
  27#include <linux/delay.h>
  28#include <linux/backing-dev.h>
  29#include <linux/fileattr.h>
  30
  31#include "gfs2.h"
  32#include "incore.h"
  33#include "bmap.h"
  34#include "aops.h"
  35#include "dir.h"
  36#include "glock.h"
  37#include "glops.h"
  38#include "inode.h"
  39#include "log.h"
  40#include "meta_io.h"
  41#include "quota.h"
  42#include "rgrp.h"
  43#include "trans.h"
  44#include "util.h"
  45
  46/**
  47 * gfs2_llseek - seek to a location in a file
  48 * @file: the file
  49 * @offset: the offset
  50 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
  51 *
  52 * SEEK_END requires the glock for the file because it references the
  53 * file's size.
  54 *
  55 * Returns: The new offset, or errno
  56 */
  57
  58static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
  59{
  60	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
  61	struct gfs2_holder i_gh;
  62	loff_t error;
  63
  64	switch (whence) {
  65	case SEEK_END:
  66		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
  67					   &i_gh);
  68		if (!error) {
  69			error = generic_file_llseek(file, offset, whence);
  70			gfs2_glock_dq_uninit(&i_gh);
  71		}
  72		break;
  73
  74	case SEEK_DATA:
  75		error = gfs2_seek_data(file, offset);
  76		break;
  77
  78	case SEEK_HOLE:
  79		error = gfs2_seek_hole(file, offset);
  80		break;
  81
  82	case SEEK_CUR:
  83	case SEEK_SET:
  84		/*
  85		 * These don't reference inode->i_size and don't depend on the
  86		 * block mapping, so we don't need the glock.
  87		 */
  88		error = generic_file_llseek(file, offset, whence);
  89		break;
  90	default:
  91		error = -EINVAL;
  92	}
  93
  94	return error;
  95}
  96
  97/**
  98 * gfs2_readdir - Iterator for a directory
  99 * @file: The directory to read from
 100 * @ctx: What to feed directory entries to
 101 *
 102 * Returns: errno
 103 */
 104
 105static int gfs2_readdir(struct file *file, struct dir_context *ctx)
 106{
 107	struct inode *dir = file->f_mapping->host;
 108	struct gfs2_inode *dip = GFS2_I(dir);
 109	struct gfs2_holder d_gh;
 110	int error;
 111
 112	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
 113	if (error)
 114		return error;
 115
 116	error = gfs2_dir_read(dir, ctx, &file->f_ra);
 117
 118	gfs2_glock_dq_uninit(&d_gh);
 119
 120	return error;
 121}
 122
 123/*
 124 * struct fsflag_gfs2flag
 125 *
 126 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
 127 * and to GFS2_DIF_JDATA for non-directories.
 128 */
 129static struct {
 130	u32 fsflag;
 131	u32 gfsflag;
 132} fsflag_gfs2flag[] = {
 133	{FS_SYNC_FL, GFS2_DIF_SYNC},
 134	{FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
 135	{FS_APPEND_FL, GFS2_DIF_APPENDONLY},
 136	{FS_NOATIME_FL, GFS2_DIF_NOATIME},
 137	{FS_INDEX_FL, GFS2_DIF_EXHASH},
 138	{FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
 139	{FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
 140};
 141
 142static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
 143{
 144	int i;
 145	u32 fsflags = 0;
 146
 147	if (S_ISDIR(inode->i_mode))
 148		gfsflags &= ~GFS2_DIF_JDATA;
 149	else
 150		gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
 151
 152	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
 153		if (gfsflags & fsflag_gfs2flag[i].gfsflag)
 154			fsflags |= fsflag_gfs2flag[i].fsflag;
 155	return fsflags;
 156}
 157
 158int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 159{
 160	struct inode *inode = d_inode(dentry);
 161	struct gfs2_inode *ip = GFS2_I(inode);
 162	struct gfs2_holder gh;
 163	int error;
 164	u32 fsflags;
 165
 166	if (d_is_special(dentry))
 167		return -ENOTTY;
 168
 169	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 170	error = gfs2_glock_nq(&gh);
 171	if (error)
 172		goto out_uninit;
 173
 174	fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
 175
 176	fileattr_fill_flags(fa, fsflags);
 177
 178	gfs2_glock_dq(&gh);
 179out_uninit:
 180	gfs2_holder_uninit(&gh);
 181	return error;
 182}
 183
 184void gfs2_set_inode_flags(struct inode *inode)
 185{
 186	struct gfs2_inode *ip = GFS2_I(inode);
 187	unsigned int flags = inode->i_flags;
 188
 189	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
 190	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
 191		flags |= S_NOSEC;
 192	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
 193		flags |= S_IMMUTABLE;
 194	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
 195		flags |= S_APPEND;
 196	if (ip->i_diskflags & GFS2_DIF_NOATIME)
 197		flags |= S_NOATIME;
 198	if (ip->i_diskflags & GFS2_DIF_SYNC)
 199		flags |= S_SYNC;
 200	inode->i_flags = flags;
 201}
 202
 203/* Flags that can be set by user space */
 204#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
 205			     GFS2_DIF_IMMUTABLE|		\
 206			     GFS2_DIF_APPENDONLY|		\
 207			     GFS2_DIF_NOATIME|			\
 208			     GFS2_DIF_SYNC|			\
 209			     GFS2_DIF_TOPDIR|			\
 210			     GFS2_DIF_INHERIT_JDATA)
 211
 212/**
 213 * do_gfs2_set_flags - set flags on an inode
 214 * @inode: The inode
 215 * @reqflags: The flags to set
 216 * @mask: Indicates which flags are valid
 217 *
 218 */
 219static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask)
 220{
 221	struct gfs2_inode *ip = GFS2_I(inode);
 222	struct gfs2_sbd *sdp = GFS2_SB(inode);
 223	struct buffer_head *bh;
 224	struct gfs2_holder gh;
 225	int error;
 226	u32 new_flags, flags;
 227
 228	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 229	if (error)
 230		return error;
 231
 232	error = 0;
 233	flags = ip->i_diskflags;
 234	new_flags = (flags & ~mask) | (reqflags & mask);
 235	if ((new_flags ^ flags) == 0)
 236		goto out;
 237
 238	if (!IS_IMMUTABLE(inode)) {
 239		error = gfs2_permission(&nop_mnt_idmap, inode, MAY_WRITE);
 240		if (error)
 241			goto out;
 242	}
 243	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
 244		if (new_flags & GFS2_DIF_JDATA)
 245			gfs2_log_flush(sdp, ip->i_gl,
 246				       GFS2_LOG_HEAD_FLUSH_NORMAL |
 247				       GFS2_LFC_SET_FLAGS);
 248		error = filemap_fdatawrite(inode->i_mapping);
 249		if (error)
 250			goto out;
 251		error = filemap_fdatawait(inode->i_mapping);
 252		if (error)
 253			goto out;
 254		if (new_flags & GFS2_DIF_JDATA)
 255			gfs2_ordered_del_inode(ip);
 256	}
 257	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 258	if (error)
 259		goto out;
 260	error = gfs2_meta_inode_buffer(ip, &bh);
 261	if (error)
 262		goto out_trans_end;
 263	inode_set_ctime_current(inode);
 264	gfs2_trans_add_meta(ip->i_gl, bh);
 265	ip->i_diskflags = new_flags;
 266	gfs2_dinode_out(ip, bh->b_data);
 267	brelse(bh);
 268	gfs2_set_inode_flags(inode);
 269	gfs2_set_aops(inode);
 270out_trans_end:
 271	gfs2_trans_end(sdp);
 272out:
 273	gfs2_glock_dq_uninit(&gh);
 274	return error;
 275}
 276
 277int gfs2_fileattr_set(struct mnt_idmap *idmap,
 278		      struct dentry *dentry, struct fileattr *fa)
 279{
 280	struct inode *inode = d_inode(dentry);
 281	u32 fsflags = fa->flags, gfsflags = 0;
 282	u32 mask;
 283	int i;
 284
 285	if (d_is_special(dentry))
 286		return -ENOTTY;
 287
 288	if (fileattr_has_fsx(fa))
 289		return -EOPNOTSUPP;
 290
 291	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
 292		if (fsflags & fsflag_gfs2flag[i].fsflag) {
 293			fsflags &= ~fsflag_gfs2flag[i].fsflag;
 294			gfsflags |= fsflag_gfs2flag[i].gfsflag;
 295		}
 296	}
 297	if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
 298		return -EINVAL;
 299
 300	mask = GFS2_FLAGS_USER_SET;
 301	if (S_ISDIR(inode->i_mode)) {
 302		mask &= ~GFS2_DIF_JDATA;
 303	} else {
 304		/* The GFS2_DIF_TOPDIR flag is only valid for directories. */
 305		if (gfsflags & GFS2_DIF_TOPDIR)
 306			return -EINVAL;
 307		mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
 308	}
 309
 310	return do_gfs2_set_flags(inode, gfsflags, mask);
 311}
 312
 313static int gfs2_getlabel(struct file *filp, char __user *label)
 314{
 315	struct inode *inode = file_inode(filp);
 316	struct gfs2_sbd *sdp = GFS2_SB(inode);
 317
 318	if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
 319		return -EFAULT;
 320
 321	return 0;
 322}
 323
 324static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 325{
 326	switch(cmd) {
 327	case FITRIM:
 328		return gfs2_fitrim(filp, (void __user *)arg);
 329	case FS_IOC_GETFSLABEL:
 330		return gfs2_getlabel(filp, (char __user *)arg);
 331	}
 332
 333	return -ENOTTY;
 334}
 335
 336#ifdef CONFIG_COMPAT
 337static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 338{
 339	switch(cmd) {
 340	/* Keep this list in sync with gfs2_ioctl */
 341	case FITRIM:
 342	case FS_IOC_GETFSLABEL:
 343		break;
 344	default:
 345		return -ENOIOCTLCMD;
 346	}
 347
 348	return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
 349}
 350#else
 351#define gfs2_compat_ioctl NULL
 352#endif
 353
 354/**
 355 * gfs2_size_hint - Give a hint to the size of a write request
 356 * @filep: The struct file
 357 * @offset: The file offset of the write
 358 * @size: The length of the write
 359 *
 360 * When we are about to do a write, this function records the total
 361 * write size in order to provide a suitable hint to the lower layers
 362 * about how many blocks will be required.
 363 *
 364 */
 365
 366static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
 367{
 368	struct inode *inode = file_inode(filep);
 369	struct gfs2_sbd *sdp = GFS2_SB(inode);
 370	struct gfs2_inode *ip = GFS2_I(inode);
 371	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
 372	int hint = min_t(size_t, INT_MAX, blks);
 373
 374	if (hint > atomic_read(&ip->i_sizehint))
 375		atomic_set(&ip->i_sizehint, hint);
 376}
 377
 378/**
 379 * gfs2_allocate_page_backing - Allocate blocks for a write fault
 380 * @page: The (locked) page to allocate backing for
 381 * @length: Size of the allocation
 382 *
 383 * We try to allocate all the blocks required for the page in one go.  This
 384 * might fail for various reasons, so we keep trying until all the blocks to
 385 * back this page are allocated.  If some of the blocks are already allocated,
 386 * that is ok too.
 387 */
 388static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
 389{
 390	u64 pos = page_offset(page);
 391
 392	do {
 393		struct iomap iomap = { };
 394
 395		if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
 396			return -EIO;
 397
 398		if (length < iomap.length)
 399			iomap.length = length;
 400		length -= iomap.length;
 401		pos += iomap.length;
 402	} while (length > 0);
 403
 404	return 0;
 405}
 406
 407/**
 408 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
 409 * @vmf: The virtual memory fault containing the page to become writable
 410 *
 411 * When the page becomes writable, we need to ensure that we have
 412 * blocks allocated on disk to back that page.
 413 */
 414
 415static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
 416{
 417	struct page *page = vmf->page;
 418	struct inode *inode = file_inode(vmf->vma->vm_file);
 419	struct gfs2_inode *ip = GFS2_I(inode);
 420	struct gfs2_sbd *sdp = GFS2_SB(inode);
 421	struct gfs2_alloc_parms ap = {};
 422	u64 offset = page_offset(page);
 423	unsigned int data_blocks, ind_blocks, rblocks;
 424	vm_fault_t ret = VM_FAULT_LOCKED;
 425	struct gfs2_holder gh;
 426	unsigned int length;
 427	loff_t size;
 428	int err;
 429
 430	sb_start_pagefault(inode->i_sb);
 431
 432	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 433	err = gfs2_glock_nq(&gh);
 434	if (err) {
 435		ret = vmf_fs_error(err);
 436		goto out_uninit;
 437	}
 438
 439	/* Check page index against inode size */
 440	size = i_size_read(inode);
 441	if (offset >= size) {
 442		ret = VM_FAULT_SIGBUS;
 443		goto out_unlock;
 444	}
 445
 446	/* Update file times before taking page lock */
 447	file_update_time(vmf->vma->vm_file);
 448
 449	/* page is wholly or partially inside EOF */
 450	if (size - offset < PAGE_SIZE)
 451		length = size - offset;
 452	else
 453		length = PAGE_SIZE;
 454
 455	gfs2_size_hint(vmf->vma->vm_file, offset, length);
 456
 457	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
 458	set_bit(GIF_SW_PAGED, &ip->i_flags);
 459
 460	/*
 461	 * iomap_writepage / iomap_writepages currently don't support inline
 462	 * files, so always unstuff here.
 463	 */
 464
 465	if (!gfs2_is_stuffed(ip) &&
 466	    !gfs2_write_alloc_required(ip, offset, length)) {
 467		lock_page(page);
 468		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
 469			ret = VM_FAULT_NOPAGE;
 470			unlock_page(page);
 471		}
 472		goto out_unlock;
 473	}
 474
 475	err = gfs2_rindex_update(sdp);
 476	if (err) {
 477		ret = vmf_fs_error(err);
 478		goto out_unlock;
 479	}
 480
 481	gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
 482	ap.target = data_blocks + ind_blocks;
 483	err = gfs2_quota_lock_check(ip, &ap);
 484	if (err) {
 485		ret = vmf_fs_error(err);
 486		goto out_unlock;
 487	}
 488	err = gfs2_inplace_reserve(ip, &ap);
 489	if (err) {
 490		ret = vmf_fs_error(err);
 491		goto out_quota_unlock;
 492	}
 493
 494	rblocks = RES_DINODE + ind_blocks;
 495	if (gfs2_is_jdata(ip))
 496		rblocks += data_blocks ? data_blocks : 1;
 497	if (ind_blocks || data_blocks) {
 498		rblocks += RES_STATFS + RES_QUOTA;
 499		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
 500	}
 501	err = gfs2_trans_begin(sdp, rblocks, 0);
 502	if (err) {
 503		ret = vmf_fs_error(err);
 504		goto out_trans_fail;
 505	}
 506
 507	/* Unstuff, if required, and allocate backing blocks for page */
 508	if (gfs2_is_stuffed(ip)) {
 509		err = gfs2_unstuff_dinode(ip);
 510		if (err) {
 511			ret = vmf_fs_error(err);
 512			goto out_trans_end;
 513		}
 514	}
 515
 516	lock_page(page);
 517	/* If truncated, we must retry the operation, we may have raced
 518	 * with the glock demotion code.
 519	 */
 520	if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
 521		ret = VM_FAULT_NOPAGE;
 522		goto out_page_locked;
 523	}
 524
 525	err = gfs2_allocate_page_backing(page, length);
 526	if (err)
 527		ret = vmf_fs_error(err);
 528
 529out_page_locked:
 530	if (ret != VM_FAULT_LOCKED)
 531		unlock_page(page);
 532out_trans_end:
 533	gfs2_trans_end(sdp);
 534out_trans_fail:
 535	gfs2_inplace_release(ip);
 536out_quota_unlock:
 537	gfs2_quota_unlock(ip);
 538out_unlock:
 539	gfs2_glock_dq(&gh);
 540out_uninit:
 541	gfs2_holder_uninit(&gh);
 542	if (ret == VM_FAULT_LOCKED) {
 543		set_page_dirty(page);
 544		wait_for_stable_page(page);
 545	}
 546	sb_end_pagefault(inode->i_sb);
 547	return ret;
 548}
 549
 550static vm_fault_t gfs2_fault(struct vm_fault *vmf)
 551{
 552	struct inode *inode = file_inode(vmf->vma->vm_file);
 553	struct gfs2_inode *ip = GFS2_I(inode);
 554	struct gfs2_holder gh;
 555	vm_fault_t ret;
 556	int err;
 557
 558	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 559	err = gfs2_glock_nq(&gh);
 560	if (err) {
 561		ret = vmf_fs_error(err);
 562		goto out_uninit;
 563	}
 564	ret = filemap_fault(vmf);
 565	gfs2_glock_dq(&gh);
 566out_uninit:
 567	gfs2_holder_uninit(&gh);
 568	return ret;
 569}
 570
 571static const struct vm_operations_struct gfs2_vm_ops = {
 572	.fault = gfs2_fault,
 573	.map_pages = filemap_map_pages,
 574	.page_mkwrite = gfs2_page_mkwrite,
 575};
 576
 577/**
 578 * gfs2_mmap
 579 * @file: The file to map
 580 * @vma: The VMA which described the mapping
 581 *
 582 * There is no need to get a lock here unless we should be updating
 583 * atime. We ignore any locking errors since the only consequence is
 584 * a missed atime update (which will just be deferred until later).
 585 *
 586 * Returns: 0
 587 */
 588
 589static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 590{
 591	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 592
 593	if (!(file->f_flags & O_NOATIME) &&
 594	    !IS_NOATIME(&ip->i_inode)) {
 595		struct gfs2_holder i_gh;
 596		int error;
 597
 598		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 599					   &i_gh);
 600		if (error)
 601			return error;
 602		/* grab lock to update inode */
 603		gfs2_glock_dq_uninit(&i_gh);
 604		file_accessed(file);
 605	}
 606	vma->vm_ops = &gfs2_vm_ops;
 607
 608	return 0;
 609}
 610
 611/**
 612 * gfs2_open_common - This is common to open and atomic_open
 613 * @inode: The inode being opened
 614 * @file: The file being opened
 615 *
 616 * This maybe called under a glock or not depending upon how it has
 617 * been called. We must always be called under a glock for regular
 618 * files, however. For other file types, it does not matter whether
 619 * we hold the glock or not.
 620 *
 621 * Returns: Error code or 0 for success
 622 */
 623
 624int gfs2_open_common(struct inode *inode, struct file *file)
 625{
 626	struct gfs2_file *fp;
 627	int ret;
 628
 629	if (S_ISREG(inode->i_mode)) {
 630		ret = generic_file_open(inode, file);
 631		if (ret)
 632			return ret;
 633
 634		if (!gfs2_is_jdata(GFS2_I(inode)))
 635			file->f_mode |= FMODE_CAN_ODIRECT;
 636	}
 637
 638	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
 639	if (!fp)
 640		return -ENOMEM;
 641
 642	mutex_init(&fp->f_fl_mutex);
 643
 644	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
 645	file->private_data = fp;
 646	if (file->f_mode & FMODE_WRITE) {
 647		ret = gfs2_qa_get(GFS2_I(inode));
 648		if (ret)
 649			goto fail;
 650	}
 651	return 0;
 652
 653fail:
 654	kfree(file->private_data);
 655	file->private_data = NULL;
 656	return ret;
 657}
 658
 659/**
 660 * gfs2_open - open a file
 661 * @inode: the inode to open
 662 * @file: the struct file for this opening
 663 *
 664 * After atomic_open, this function is only used for opening files
 665 * which are already cached. We must still get the glock for regular
 666 * files to ensure that we have the file size uptodate for the large
 667 * file check which is in the common code. That is only an issue for
 668 * regular files though.
 669 *
 670 * Returns: errno
 671 */
 672
 673static int gfs2_open(struct inode *inode, struct file *file)
 674{
 675	struct gfs2_inode *ip = GFS2_I(inode);
 676	struct gfs2_holder i_gh;
 677	int error;
 678	bool need_unlock = false;
 679
 680	if (S_ISREG(ip->i_inode.i_mode)) {
 681		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 682					   &i_gh);
 683		if (error)
 684			return error;
 685		need_unlock = true;
 686	}
 687
 688	error = gfs2_open_common(inode, file);
 689
 690	if (need_unlock)
 691		gfs2_glock_dq_uninit(&i_gh);
 692
 693	return error;
 694}
 695
 696/**
 697 * gfs2_release - called to close a struct file
 698 * @inode: the inode the struct file belongs to
 699 * @file: the struct file being closed
 700 *
 701 * Returns: errno
 702 */
 703
 704static int gfs2_release(struct inode *inode, struct file *file)
 705{
 706	struct gfs2_inode *ip = GFS2_I(inode);
 707
 708	kfree(file->private_data);
 709	file->private_data = NULL;
 710
 711	if (file->f_mode & FMODE_WRITE) {
 712		if (gfs2_rs_active(&ip->i_res))
 713			gfs2_rs_delete(ip);
 714		gfs2_qa_put(ip);
 715	}
 716	return 0;
 717}
 718
 719/**
 720 * gfs2_fsync - sync the dirty data for a file (across the cluster)
 721 * @file: the file that points to the dentry
 722 * @start: the start position in the file to sync
 723 * @end: the end position in the file to sync
 724 * @datasync: set if we can ignore timestamp changes
 725 *
 726 * We split the data flushing here so that we don't wait for the data
 727 * until after we've also sent the metadata to disk. Note that for
 728 * data=ordered, we will write & wait for the data at the log flush
 729 * stage anyway, so this is unlikely to make much of a difference
 730 * except in the data=writeback case.
 731 *
 732 * If the fdatawrite fails due to any reason except -EIO, we will
 733 * continue the remainder of the fsync, although we'll still report
 734 * the error at the end. This is to match filemap_write_and_wait_range()
 735 * behaviour.
 736 *
 737 * Returns: errno
 738 */
 739
 740static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 741		      int datasync)
 742{
 743	struct address_space *mapping = file->f_mapping;
 744	struct inode *inode = mapping->host;
 745	int sync_state = inode->i_state & I_DIRTY;
 746	struct gfs2_inode *ip = GFS2_I(inode);
 747	int ret = 0, ret1 = 0;
 748
 749	if (mapping->nrpages) {
 750		ret1 = filemap_fdatawrite_range(mapping, start, end);
 751		if (ret1 == -EIO)
 752			return ret1;
 753	}
 754
 755	if (!gfs2_is_jdata(ip))
 756		sync_state &= ~I_DIRTY_PAGES;
 757	if (datasync)
 758		sync_state &= ~I_DIRTY_SYNC;
 759
 760	if (sync_state) {
 761		ret = sync_inode_metadata(inode, 1);
 762		if (ret)
 763			return ret;
 764		if (gfs2_is_jdata(ip))
 765			ret = file_write_and_wait(file);
 766		if (ret)
 767			return ret;
 768		gfs2_ail_flush(ip->i_gl, 1);
 769	}
 770
 771	if (mapping->nrpages)
 772		ret = file_fdatawait_range(file, start, end);
 773
 774	return ret ? ret : ret1;
 775}
 776
 777static inline bool should_fault_in_pages(struct iov_iter *i,
 778					 struct kiocb *iocb,
 779					 size_t *prev_count,
 780					 size_t *window_size)
 781{
 782	size_t count = iov_iter_count(i);
 783	size_t size, offs;
 784
 785	if (!count)
 786		return false;
 787	if (!user_backed_iter(i))
 788		return false;
 789
 790	/*
 791	 * Try to fault in multiple pages initially.  When that doesn't result
 792	 * in any progress, fall back to a single page.
 793	 */
 794	size = PAGE_SIZE;
 795	offs = offset_in_page(iocb->ki_pos);
 796	if (*prev_count != count) {
 797		size_t nr_dirtied;
 798
 799		nr_dirtied = max(current->nr_dirtied_pause -
 800				 current->nr_dirtied, 8);
 801		size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT);
 802	}
 803
 804	*prev_count = count;
 805	*window_size = size - offs;
 806	return true;
 807}
 808
 809static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
 810				     struct gfs2_holder *gh)
 811{
 812	struct file *file = iocb->ki_filp;
 813	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 814	size_t prev_count = 0, window_size = 0;
 815	size_t read = 0;
 816	ssize_t ret;
 817
 818	/*
 819	 * In this function, we disable page faults when we're holding the
 820	 * inode glock while doing I/O.  If a page fault occurs, we indicate
 821	 * that the inode glock may be dropped, fault in the pages manually,
 822	 * and retry.
 823	 *
 824	 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
 825	 * physical as well as manual page faults, and we need to disable both
 826	 * kinds.
 827	 *
 828	 * For direct I/O, gfs2 takes the inode glock in deferred mode.  This
 829	 * locking mode is compatible with other deferred holders, so multiple
 830	 * processes and nodes can do direct I/O to a file at the same time.
 831	 * There's no guarantee that reads or writes will be atomic.  Any
 832	 * coordination among readers and writers needs to happen externally.
 833	 */
 834
 835	if (!iov_iter_count(to))
 836		return 0; /* skip atime */
 837
 838	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
 839retry:
 840	ret = gfs2_glock_nq(gh);
 841	if (ret)
 842		goto out_uninit;
 843	pagefault_disable();
 844	to->nofault = true;
 845	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
 846			   IOMAP_DIO_PARTIAL, NULL, read);
 847	to->nofault = false;
 848	pagefault_enable();
 849	if (ret <= 0 && ret != -EFAULT)
 850		goto out_unlock;
 851	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
 852	if (ret > 0)
 853		read = ret;
 854
 855	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
 856		gfs2_glock_dq(gh);
 857		window_size -= fault_in_iov_iter_writeable(to, window_size);
 858		if (window_size)
 859			goto retry;
 860	}
 861out_unlock:
 862	if (gfs2_holder_queued(gh))
 863		gfs2_glock_dq(gh);
 864out_uninit:
 865	gfs2_holder_uninit(gh);
 866	/* User space doesn't expect partial success. */
 867	if (ret < 0)
 868		return ret;
 869	return read;
 870}
 871
 872static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
 873				      struct gfs2_holder *gh)
 874{
 875	struct file *file = iocb->ki_filp;
 876	struct inode *inode = file->f_mapping->host;
 877	struct gfs2_inode *ip = GFS2_I(inode);
 878	size_t prev_count = 0, window_size = 0;
 879	size_t written = 0;
 880	bool enough_retries;
 881	ssize_t ret;
 882
 883	/*
 884	 * In this function, we disable page faults when we're holding the
 885	 * inode glock while doing I/O.  If a page fault occurs, we indicate
 886	 * that the inode glock may be dropped, fault in the pages manually,
 887	 * and retry.
 888	 *
 889	 * For writes, iomap_dio_rw only triggers manual page faults, so we
 890	 * don't need to disable physical ones.
 891	 */
 892
 893	/*
 894	 * Deferred lock, even if its a write, since we do no allocation on
 895	 * this path. All we need to change is the atime, and this lock mode
 896	 * ensures that other nodes have flushed their buffered read caches
 897	 * (i.e. their page cache entries for this inode). We do not,
 898	 * unfortunately, have the option of only flushing a range like the
 899	 * VFS does.
 900	 */
 901	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
 902retry:
 903	ret = gfs2_glock_nq(gh);
 904	if (ret)
 905		goto out_uninit;
 906	/* Silently fall back to buffered I/O when writing beyond EOF */
 907	if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
 908		goto out_unlock;
 909
 910	from->nofault = true;
 911	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
 912			   IOMAP_DIO_PARTIAL, NULL, written);
 913	from->nofault = false;
 914	if (ret <= 0) {
 915		if (ret == -ENOTBLK)
 916			ret = 0;
 917		if (ret != -EFAULT)
 918			goto out_unlock;
 919	}
 920	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
 921	if (ret > 0)
 922		written = ret;
 923
 924	enough_retries = prev_count == iov_iter_count(from) &&
 925			 window_size <= PAGE_SIZE;
 926	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
 927		gfs2_glock_dq(gh);
 928		window_size -= fault_in_iov_iter_readable(from, window_size);
 929		if (window_size) {
 930			if (!enough_retries)
 931				goto retry;
 932			/* fall back to buffered I/O */
 933			ret = 0;
 934		}
 935	}
 936out_unlock:
 937	if (gfs2_holder_queued(gh))
 938		gfs2_glock_dq(gh);
 939out_uninit:
 940	gfs2_holder_uninit(gh);
 941	/* User space doesn't expect partial success. */
 942	if (ret < 0)
 943		return ret;
 944	return written;
 945}
 946
 947static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 948{
 949	struct gfs2_inode *ip;
 950	struct gfs2_holder gh;
 951	size_t prev_count = 0, window_size = 0;
 952	size_t read = 0;
 953	ssize_t ret;
 954
 955	/*
 956	 * In this function, we disable page faults when we're holding the
 957	 * inode glock while doing I/O.  If a page fault occurs, we indicate
 958	 * that the inode glock may be dropped, fault in the pages manually,
 959	 * and retry.
 960	 */
 961
 962	if (iocb->ki_flags & IOCB_DIRECT)
 963		return gfs2_file_direct_read(iocb, to, &gh);
 964
 965	pagefault_disable();
 966	iocb->ki_flags |= IOCB_NOIO;
 967	ret = generic_file_read_iter(iocb, to);
 968	iocb->ki_flags &= ~IOCB_NOIO;
 969	pagefault_enable();
 970	if (ret >= 0) {
 971		if (!iov_iter_count(to))
 972			return ret;
 973		read = ret;
 974	} else if (ret != -EFAULT) {
 975		if (ret != -EAGAIN)
 976			return ret;
 977		if (iocb->ki_flags & IOCB_NOWAIT)
 978			return ret;
 979	}
 980	ip = GFS2_I(iocb->ki_filp->f_mapping->host);
 981	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 982retry:
 983	ret = gfs2_glock_nq(&gh);
 984	if (ret)
 985		goto out_uninit;
 986	pagefault_disable();
 987	ret = generic_file_read_iter(iocb, to);
 988	pagefault_enable();
 989	if (ret <= 0 && ret != -EFAULT)
 990		goto out_unlock;
 991	if (ret > 0)
 992		read += ret;
 993
 994	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
 995		gfs2_glock_dq(&gh);
 996		window_size -= fault_in_iov_iter_writeable(to, window_size);
 997		if (window_size)
 998			goto retry;
 999	}
1000out_unlock:
1001	if (gfs2_holder_queued(&gh))
1002		gfs2_glock_dq(&gh);
1003out_uninit:
1004	gfs2_holder_uninit(&gh);
1005	return read ? read : ret;
1006}
1007
1008static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
1009					struct iov_iter *from,
1010					struct gfs2_holder *gh)
1011{
1012	struct file *file = iocb->ki_filp;
1013	struct inode *inode = file_inode(file);
1014	struct gfs2_inode *ip = GFS2_I(inode);
1015	struct gfs2_sbd *sdp = GFS2_SB(inode);
1016	struct gfs2_holder *statfs_gh = NULL;
1017	size_t prev_count = 0, window_size = 0;
1018	size_t orig_count = iov_iter_count(from);
1019	size_t written = 0;
1020	ssize_t ret;
1021
1022	/*
1023	 * In this function, we disable page faults when we're holding the
1024	 * inode glock while doing I/O.  If a page fault occurs, we indicate
1025	 * that the inode glock may be dropped, fault in the pages manually,
1026	 * and retry.
1027	 */
1028
1029	if (inode == sdp->sd_rindex) {
1030		statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1031		if (!statfs_gh)
1032			return -ENOMEM;
1033	}
1034
1035	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
 
1036	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1037retry:
1038		window_size -= fault_in_iov_iter_readable(from, window_size);
1039		if (!window_size) {
1040			ret = -EFAULT;
1041			goto out_uninit;
1042		}
1043		from->count = min(from->count, window_size);
1044	}
1045	ret = gfs2_glock_nq(gh);
1046	if (ret)
1047		goto out_uninit;
1048
1049	if (inode == sdp->sd_rindex) {
1050		struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1051
1052		ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1053					 GL_NOCACHE, statfs_gh);
1054		if (ret)
1055			goto out_unlock;
1056	}
1057
 
1058	pagefault_disable();
1059	ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
1060	pagefault_enable();
1061	if (ret > 0)
 
 
1062		written += ret;
 
1063
1064	if (inode == sdp->sd_rindex)
1065		gfs2_glock_dq_uninit(statfs_gh);
1066
1067	if (ret <= 0 && ret != -EFAULT)
1068		goto out_unlock;
1069
1070	from->count = orig_count - written;
1071	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1072		gfs2_glock_dq(gh);
1073		goto retry;
1074	}
1075out_unlock:
1076	if (gfs2_holder_queued(gh))
1077		gfs2_glock_dq(gh);
1078out_uninit:
1079	gfs2_holder_uninit(gh);
1080	kfree(statfs_gh);
1081	from->count = orig_count - written;
1082	return written ? written : ret;
1083}
1084
1085/**
1086 * gfs2_file_write_iter - Perform a write to a file
1087 * @iocb: The io context
1088 * @from: The data to write
1089 *
1090 * We have to do a lock/unlock here to refresh the inode size for
1091 * O_APPEND writes, otherwise we can land up writing at the wrong
1092 * offset. There is still a race, but provided the app is using its
1093 * own file locking, this will make O_APPEND work as expected.
1094 *
1095 */
1096
1097static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1098{
1099	struct file *file = iocb->ki_filp;
1100	struct inode *inode = file_inode(file);
1101	struct gfs2_inode *ip = GFS2_I(inode);
1102	struct gfs2_holder gh;
1103	ssize_t ret;
1104
1105	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
1106
1107	if (iocb->ki_flags & IOCB_APPEND) {
1108		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1109		if (ret)
1110			return ret;
1111		gfs2_glock_dq_uninit(&gh);
1112	}
1113
1114	inode_lock(inode);
1115	ret = generic_write_checks(iocb, from);
1116	if (ret <= 0)
1117		goto out_unlock;
1118
1119	ret = file_remove_privs(file);
1120	if (ret)
1121		goto out_unlock;
1122
 
 
 
 
1123	if (iocb->ki_flags & IOCB_DIRECT) {
1124		struct address_space *mapping = file->f_mapping;
1125		ssize_t buffered, ret2;
1126
1127		/*
1128		 * Note that under direct I/O, we don't allow and inode
1129		 * timestamp updates, so we're not calling file_update_time()
1130		 * here.
1131		 */
1132
1133		ret = gfs2_file_direct_write(iocb, from, &gh);
1134		if (ret < 0 || !iov_iter_count(from))
1135			goto out_unlock;
1136
1137		iocb->ki_flags |= IOCB_DSYNC;
1138		buffered = gfs2_file_buffered_write(iocb, from, &gh);
1139		if (unlikely(buffered <= 0)) {
1140			if (!ret)
1141				ret = buffered;
1142			goto out_unlock;
1143		}
1144
1145		/*
1146		 * We need to ensure that the page cache pages are written to
1147		 * disk and invalidated to preserve the expected O_DIRECT
1148		 * semantics.  If the writeback or invalidate fails, only report
1149		 * the direct I/O range as we don't know if the buffered pages
1150		 * made it to disk.
1151		 */
1152		ret2 = generic_write_sync(iocb, buffered);
1153		invalidate_mapping_pages(mapping,
1154				(iocb->ki_pos - buffered) >> PAGE_SHIFT,
1155				(iocb->ki_pos - 1) >> PAGE_SHIFT);
1156		if (!ret || ret2 > 0)
1157			ret += ret2;
1158	} else {
1159		ret = file_update_time(file);
1160		if (ret)
1161			goto out_unlock;
1162
1163		ret = gfs2_file_buffered_write(iocb, from, &gh);
1164		if (likely(ret > 0))
1165			ret = generic_write_sync(iocb, ret);
1166	}
1167
1168out_unlock:
1169	inode_unlock(inode);
1170	return ret;
1171}
1172
1173static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
1174			   int mode)
1175{
1176	struct super_block *sb = inode->i_sb;
1177	struct gfs2_inode *ip = GFS2_I(inode);
1178	loff_t end = offset + len;
1179	struct buffer_head *dibh;
1180	int error;
1181
1182	error = gfs2_meta_inode_buffer(ip, &dibh);
1183	if (unlikely(error))
1184		return error;
1185
1186	gfs2_trans_add_meta(ip->i_gl, dibh);
1187
1188	if (gfs2_is_stuffed(ip)) {
1189		error = gfs2_unstuff_dinode(ip);
1190		if (unlikely(error))
1191			goto out;
1192	}
1193
1194	while (offset < end) {
1195		struct iomap iomap = { };
1196
1197		error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
1198		if (error)
1199			goto out;
1200		offset = iomap.offset + iomap.length;
1201		if (!(iomap.flags & IOMAP_F_NEW))
1202			continue;
1203		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1204					 iomap.length >> inode->i_blkbits,
1205					 GFP_NOFS);
1206		if (error) {
1207			fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1208			goto out;
1209		}
1210	}
1211out:
1212	brelse(dibh);
1213	return error;
1214}
1215
1216/**
1217 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1218 *                     blocks, determine how many bytes can be written.
1219 * @ip:          The inode in question.
1220 * @len:         Max cap of bytes. What we return in *len must be <= this.
1221 * @data_blocks: Compute and return the number of data blocks needed
1222 * @ind_blocks:  Compute and return the number of indirect blocks needed
1223 * @max_blocks:  The total blocks available to work with.
1224 *
1225 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1226 */
1227static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1228			    unsigned int *data_blocks, unsigned int *ind_blocks,
1229			    unsigned int max_blocks)
1230{
1231	loff_t max = *len;
1232	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1233	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1234
1235	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1236		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1237		max_data -= tmp;
1238	}
1239
1240	*data_blocks = max_data;
1241	*ind_blocks = max_blocks - max_data;
1242	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1243	if (*len > max) {
1244		*len = max;
1245		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1246	}
1247}
1248
1249static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1250{
1251	struct inode *inode = file_inode(file);
1252	struct gfs2_sbd *sdp = GFS2_SB(inode);
1253	struct gfs2_inode *ip = GFS2_I(inode);
1254	struct gfs2_alloc_parms ap = {};
1255	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1256	loff_t bytes, max_bytes, max_blks;
1257	int error;
1258	const loff_t pos = offset;
1259	const loff_t count = len;
1260	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1261	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1262	loff_t max_chunk_size = UINT_MAX & bsize_mask;
1263
1264	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1265
1266	offset &= bsize_mask;
1267
1268	len = next - offset;
1269	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1270	if (!bytes)
1271		bytes = UINT_MAX;
1272	bytes &= bsize_mask;
1273	if (bytes == 0)
1274		bytes = sdp->sd_sb.sb_bsize;
1275
1276	gfs2_size_hint(file, offset, len);
1277
1278	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1279	ap.min_target = data_blocks + ind_blocks;
1280
1281	while (len > 0) {
1282		if (len < bytes)
1283			bytes = len;
1284		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1285			len -= bytes;
1286			offset += bytes;
1287			continue;
1288		}
1289
1290		/* We need to determine how many bytes we can actually
1291		 * fallocate without exceeding quota or going over the
1292		 * end of the fs. We start off optimistically by assuming
1293		 * we can write max_bytes */
1294		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1295
1296		/* Since max_bytes is most likely a theoretical max, we
1297		 * calculate a more realistic 'bytes' to serve as a good
1298		 * starting point for the number of bytes we may be able
1299		 * to write */
1300		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1301		ap.target = data_blocks + ind_blocks;
1302
1303		error = gfs2_quota_lock_check(ip, &ap);
1304		if (error)
1305			return error;
1306		/* ap.allowed tells us how many blocks quota will allow
1307		 * us to write. Check if this reduces max_blks */
1308		max_blks = UINT_MAX;
1309		if (ap.allowed)
1310			max_blks = ap.allowed;
1311
1312		error = gfs2_inplace_reserve(ip, &ap);
1313		if (error)
1314			goto out_qunlock;
1315
1316		/* check if the selected rgrp limits our max_blks further */
1317		if (ip->i_res.rs_reserved < max_blks)
1318			max_blks = ip->i_res.rs_reserved;
1319
1320		/* Almost done. Calculate bytes that can be written using
1321		 * max_blks. We also recompute max_bytes, data_blocks and
1322		 * ind_blocks */
1323		calc_max_reserv(ip, &max_bytes, &data_blocks,
1324				&ind_blocks, max_blks);
1325
1326		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1327			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1328		if (gfs2_is_jdata(ip))
1329			rblocks += data_blocks ? data_blocks : 1;
1330
1331		error = gfs2_trans_begin(sdp, rblocks,
1332					 PAGE_SIZE >> inode->i_blkbits);
1333		if (error)
1334			goto out_trans_fail;
1335
1336		error = fallocate_chunk(inode, offset, max_bytes, mode);
1337		gfs2_trans_end(sdp);
1338
1339		if (error)
1340			goto out_trans_fail;
1341
1342		len -= max_bytes;
1343		offset += max_bytes;
1344		gfs2_inplace_release(ip);
1345		gfs2_quota_unlock(ip);
1346	}
1347
1348	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1349		i_size_write(inode, pos + count);
1350	file_update_time(file);
1351	mark_inode_dirty(inode);
1352
1353	if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1354		return vfs_fsync_range(file, pos, pos + count - 1,
1355			       (file->f_flags & __O_SYNC) ? 0 : 1);
1356	return 0;
1357
1358out_trans_fail:
1359	gfs2_inplace_release(ip);
1360out_qunlock:
1361	gfs2_quota_unlock(ip);
1362	return error;
1363}
1364
1365static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1366{
1367	struct inode *inode = file_inode(file);
1368	struct gfs2_sbd *sdp = GFS2_SB(inode);
1369	struct gfs2_inode *ip = GFS2_I(inode);
1370	struct gfs2_holder gh;
1371	int ret;
1372
1373	if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1374		return -EOPNOTSUPP;
1375	/* fallocate is needed by gfs2_grow to reserve space in the rindex */
1376	if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1377		return -EOPNOTSUPP;
1378
1379	inode_lock(inode);
1380
1381	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1382	ret = gfs2_glock_nq(&gh);
1383	if (ret)
1384		goto out_uninit;
1385
1386	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1387	    (offset + len) > inode->i_size) {
1388		ret = inode_newsize_ok(inode, offset + len);
1389		if (ret)
1390			goto out_unlock;
1391	}
1392
1393	ret = get_write_access(inode);
1394	if (ret)
1395		goto out_unlock;
1396
1397	if (mode & FALLOC_FL_PUNCH_HOLE) {
1398		ret = __gfs2_punch_hole(file, offset, len);
1399	} else {
1400		ret = __gfs2_fallocate(file, mode, offset, len);
1401		if (ret)
1402			gfs2_rs_deltree(&ip->i_res);
1403	}
1404
1405	put_write_access(inode);
1406out_unlock:
1407	gfs2_glock_dq(&gh);
1408out_uninit:
1409	gfs2_holder_uninit(&gh);
1410	inode_unlock(inode);
1411	return ret;
1412}
1413
1414static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1415				      struct file *out, loff_t *ppos,
1416				      size_t len, unsigned int flags)
1417{
1418	ssize_t ret;
1419
1420	gfs2_size_hint(out, *ppos, len);
1421
1422	ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1423	return ret;
1424}
1425
1426#ifdef CONFIG_GFS2_FS_LOCKING_DLM
1427
1428/**
1429 * gfs2_lock - acquire/release a posix lock on a file
1430 * @file: the file pointer
1431 * @cmd: either modify or retrieve lock state, possibly wait
1432 * @fl: type and range of lock
1433 *
1434 * Returns: errno
1435 */
1436
1437static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1438{
1439	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1440	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1441	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1442
1443	if (!(fl->fl_flags & FL_POSIX))
1444		return -ENOLCK;
1445	if (gfs2_withdrawing_or_withdrawn(sdp)) {
 
 
 
 
 
1446		if (fl->fl_type == F_UNLCK)
1447			locks_lock_file_wait(file, fl);
1448		return -EIO;
1449	}
1450	if (cmd == F_CANCELLK)
1451		return dlm_posix_cancel(ls->ls_dlm, ip->i_no_addr, file, fl);
1452	else if (IS_GETLK(cmd))
1453		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1454	else if (fl->fl_type == F_UNLCK)
1455		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1456	else
1457		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1458}
1459
1460static void __flock_holder_uninit(struct file *file, struct gfs2_holder *fl_gh)
1461{
1462	struct gfs2_glock *gl = gfs2_glock_hold(fl_gh->gh_gl);
1463
1464	/*
1465	 * Make sure gfs2_glock_put() won't sleep under the file->f_lock
1466	 * spinlock.
1467	 */
1468
1469	spin_lock(&file->f_lock);
1470	gfs2_holder_uninit(fl_gh);
1471	spin_unlock(&file->f_lock);
1472	gfs2_glock_put(gl);
1473}
1474
1475static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1476{
1477	struct gfs2_file *fp = file->private_data;
1478	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1479	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1480	struct gfs2_glock *gl;
1481	unsigned int state;
1482	u16 flags;
1483	int error = 0;
1484	int sleeptime;
1485
1486	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1487	flags = GL_EXACT | GL_NOPID;
1488	if (!IS_SETLKW(cmd))
1489		flags |= LM_FLAG_TRY_1CB;
1490
1491	mutex_lock(&fp->f_fl_mutex);
1492
1493	if (gfs2_holder_initialized(fl_gh)) {
1494		struct file_lock request;
1495		if (fl_gh->gh_state == state)
1496			goto out;
1497		locks_init_lock(&request);
1498		request.fl_type = F_UNLCK;
1499		request.fl_flags = FL_FLOCK;
1500		locks_lock_file_wait(file, &request);
1501		gfs2_glock_dq(fl_gh);
1502		gfs2_holder_reinit(state, flags, fl_gh);
1503	} else {
1504		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1505				       &gfs2_flock_glops, CREATE, &gl);
1506		if (error)
1507			goto out;
1508		spin_lock(&file->f_lock);
1509		gfs2_holder_init(gl, state, flags, fl_gh);
1510		spin_unlock(&file->f_lock);
1511		gfs2_glock_put(gl);
1512	}
1513	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1514		error = gfs2_glock_nq(fl_gh);
1515		if (error != GLR_TRYFAILED)
1516			break;
1517		fl_gh->gh_flags &= ~LM_FLAG_TRY_1CB;
1518		fl_gh->gh_flags |= LM_FLAG_TRY;
1519		msleep(sleeptime);
1520	}
1521	if (error) {
1522		__flock_holder_uninit(file, fl_gh);
1523		if (error == GLR_TRYFAILED)
1524			error = -EAGAIN;
1525	} else {
1526		error = locks_lock_file_wait(file, fl);
1527		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1528	}
1529
1530out:
1531	mutex_unlock(&fp->f_fl_mutex);
1532	return error;
1533}
1534
1535static void do_unflock(struct file *file, struct file_lock *fl)
1536{
1537	struct gfs2_file *fp = file->private_data;
1538	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1539
1540	mutex_lock(&fp->f_fl_mutex);
1541	locks_lock_file_wait(file, fl);
1542	if (gfs2_holder_initialized(fl_gh)) {
1543		gfs2_glock_dq(fl_gh);
1544		__flock_holder_uninit(file, fl_gh);
1545	}
1546	mutex_unlock(&fp->f_fl_mutex);
1547}
1548
1549/**
1550 * gfs2_flock - acquire/release a flock lock on a file
1551 * @file: the file pointer
1552 * @cmd: either modify or retrieve lock state, possibly wait
1553 * @fl: type and range of lock
1554 *
1555 * Returns: errno
1556 */
1557
1558static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1559{
1560	if (!(fl->fl_flags & FL_FLOCK))
1561		return -ENOLCK;
1562
1563	if (fl->fl_type == F_UNLCK) {
1564		do_unflock(file, fl);
1565		return 0;
1566	} else {
1567		return do_flock(file, cmd, fl);
1568	}
1569}
1570
1571const struct file_operations gfs2_file_fops = {
1572	.llseek		= gfs2_llseek,
1573	.read_iter	= gfs2_file_read_iter,
1574	.write_iter	= gfs2_file_write_iter,
1575	.iopoll		= iocb_bio_iopoll,
1576	.unlocked_ioctl	= gfs2_ioctl,
1577	.compat_ioctl	= gfs2_compat_ioctl,
1578	.mmap		= gfs2_mmap,
1579	.open		= gfs2_open,
1580	.release	= gfs2_release,
1581	.fsync		= gfs2_fsync,
1582	.lock		= gfs2_lock,
1583	.flock		= gfs2_flock,
1584	.splice_read	= copy_splice_read,
1585	.splice_write	= gfs2_file_splice_write,
1586	.setlease	= simple_nosetlease,
1587	.fallocate	= gfs2_fallocate,
1588};
1589
1590const struct file_operations gfs2_dir_fops = {
1591	.iterate_shared	= gfs2_readdir,
1592	.unlocked_ioctl	= gfs2_ioctl,
1593	.compat_ioctl	= gfs2_compat_ioctl,
1594	.open		= gfs2_open,
1595	.release	= gfs2_release,
1596	.fsync		= gfs2_fsync,
1597	.lock		= gfs2_lock,
1598	.flock		= gfs2_flock,
1599	.llseek		= default_llseek,
1600};
1601
1602#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1603
1604const struct file_operations gfs2_file_fops_nolock = {
1605	.llseek		= gfs2_llseek,
1606	.read_iter	= gfs2_file_read_iter,
1607	.write_iter	= gfs2_file_write_iter,
1608	.iopoll		= iocb_bio_iopoll,
1609	.unlocked_ioctl	= gfs2_ioctl,
1610	.compat_ioctl	= gfs2_compat_ioctl,
1611	.mmap		= gfs2_mmap,
1612	.open		= gfs2_open,
1613	.release	= gfs2_release,
1614	.fsync		= gfs2_fsync,
1615	.splice_read	= copy_splice_read,
1616	.splice_write	= gfs2_file_splice_write,
1617	.setlease	= generic_setlease,
1618	.fallocate	= gfs2_fallocate,
1619};
1620
1621const struct file_operations gfs2_dir_fops_nolock = {
1622	.iterate_shared	= gfs2_readdir,
1623	.unlocked_ioctl	= gfs2_ioctl,
1624	.compat_ioctl	= gfs2_compat_ioctl,
1625	.open		= gfs2_open,
1626	.release	= gfs2_release,
1627	.fsync		= gfs2_fsync,
1628	.llseek		= default_llseek,
1629};
1630