Linux Audio

Check our new training course

Loading...
v3.1
 
   1/*
   2 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3 * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4 *
   5 * This copyrighted material is made available to anyone wishing to use,
   6 * modify, copy, or redistribute it subject to the terms and conditions
   7 * of the GNU General Public License version 2.
   8 */
   9
  10#include <linux/slab.h>
  11#include <linux/spinlock.h>
 
  12#include <linux/completion.h>
  13#include <linux/buffer_head.h>
  14#include <linux/pagemap.h>
  15#include <linux/uio.h>
  16#include <linux/blkdev.h>
  17#include <linux/mm.h>
  18#include <linux/mount.h>
  19#include <linux/fs.h>
 
  20#include <linux/gfs2_ondisk.h>
  21#include <linux/ext2_fs.h>
  22#include <linux/falloc.h>
  23#include <linux/swap.h>
  24#include <linux/crc32.h>
  25#include <linux/writeback.h>
  26#include <asm/uaccess.h>
  27#include <linux/dlm.h>
  28#include <linux/dlm_plock.h>
 
 
 
  29
  30#include "gfs2.h"
  31#include "incore.h"
  32#include "bmap.h"
 
  33#include "dir.h"
  34#include "glock.h"
  35#include "glops.h"
  36#include "inode.h"
  37#include "log.h"
  38#include "meta_io.h"
  39#include "quota.h"
  40#include "rgrp.h"
  41#include "trans.h"
  42#include "util.h"
  43
  44/**
  45 * gfs2_llseek - seek to a location in a file
  46 * @file: the file
  47 * @offset: the offset
  48 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
  49 *
  50 * SEEK_END requires the glock for the file because it references the
  51 * file's size.
  52 *
  53 * Returns: The new offset, or errno
  54 */
  55
  56static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
  57{
  58	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
  59	struct gfs2_holder i_gh;
  60	loff_t error;
  61
  62	if (origin == 2) {
 
  63		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
  64					   &i_gh);
  65		if (!error) {
  66			error = generic_file_llseek_unlocked(file, offset, origin);
  67			gfs2_glock_dq_uninit(&i_gh);
  68		}
  69	} else
  70		error = generic_file_llseek_unlocked(file, offset, origin);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  71
  72	return error;
  73}
  74
  75/**
  76 * gfs2_readdir - Read directory entries from a directory
  77 * @file: The directory to read from
  78 * @dirent: Buffer for dirents
  79 * @filldir: Function used to do the copying
  80 *
  81 * Returns: errno
  82 */
  83
  84static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
  85{
  86	struct inode *dir = file->f_mapping->host;
  87	struct gfs2_inode *dip = GFS2_I(dir);
  88	struct gfs2_holder d_gh;
  89	u64 offset = file->f_pos;
  90	int error;
  91
  92	gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
  93	error = gfs2_glock_nq(&d_gh);
  94	if (error) {
  95		gfs2_holder_uninit(&d_gh);
  96		return error;
  97	}
  98
  99	error = gfs2_dir_read(dir, &offset, dirent, filldir);
 100
 101	gfs2_glock_dq_uninit(&d_gh);
 102
 103	file->f_pos = offset;
 104
 105	return error;
 106}
 107
 108/**
 109 * fsflags_cvt
 110 * @table: A table of 32 u32 flags
 111 * @val: a 32 bit value to convert
 112 *
 113 * This function can be used to convert between fsflags values and
 114 * GFS2's own flags values.
 115 *
 116 * Returns: the converted flags
 
 117 */
 118static u32 fsflags_cvt(const u32 *table, u32 val)
 119{
 120	u32 res = 0;
 121	while(val) {
 122		if (val & 1)
 123			res |= *table;
 124		table++;
 125		val >>= 1;
 126	}
 127	return res;
 128}
 129
 130static const u32 fsflags_to_gfs2[32] = {
 131	[3] = GFS2_DIF_SYNC,
 132	[4] = GFS2_DIF_IMMUTABLE,
 133	[5] = GFS2_DIF_APPENDONLY,
 134	[7] = GFS2_DIF_NOATIME,
 135	[12] = GFS2_DIF_EXHASH,
 136	[14] = GFS2_DIF_INHERIT_JDATA,
 137};
 138
 139static const u32 gfs2_to_fsflags[32] = {
 140	[gfs2fl_Sync] = FS_SYNC_FL,
 141	[gfs2fl_Immutable] = FS_IMMUTABLE_FL,
 142	[gfs2fl_AppendOnly] = FS_APPEND_FL,
 143	[gfs2fl_NoAtime] = FS_NOATIME_FL,
 144	[gfs2fl_ExHash] = FS_INDEX_FL,
 145	[gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
 146};
 147
 148static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
 
 
 
 
 
 
 
 
 
 
 
 149{
 150	struct inode *inode = filp->f_path.dentry->d_inode;
 151	struct gfs2_inode *ip = GFS2_I(inode);
 152	struct gfs2_holder gh;
 153	int error;
 154	u32 fsflags;
 155
 
 
 
 156	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 157	error = gfs2_glock_nq(&gh);
 158	if (error)
 159		return error;
 160
 161	fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
 162	if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
 163		fsflags |= FS_JOURNAL_DATA_FL;
 164	if (put_user(fsflags, ptr))
 165		error = -EFAULT;
 166
 167	gfs2_glock_dq(&gh);
 
 168	gfs2_holder_uninit(&gh);
 169	return error;
 170}
 171
 172void gfs2_set_inode_flags(struct inode *inode)
 173{
 174	struct gfs2_inode *ip = GFS2_I(inode);
 175	unsigned int flags = inode->i_flags;
 176
 177	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
 178	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
 179		inode->i_flags |= S_NOSEC;
 180	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
 181		flags |= S_IMMUTABLE;
 182	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
 183		flags |= S_APPEND;
 184	if (ip->i_diskflags & GFS2_DIF_NOATIME)
 185		flags |= S_NOATIME;
 186	if (ip->i_diskflags & GFS2_DIF_SYNC)
 187		flags |= S_SYNC;
 188	inode->i_flags = flags;
 189}
 190
 191/* Flags that can be set by user space */
 192#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
 193			     GFS2_DIF_IMMUTABLE|		\
 194			     GFS2_DIF_APPENDONLY|		\
 195			     GFS2_DIF_NOATIME|			\
 196			     GFS2_DIF_SYNC|			\
 197			     GFS2_DIF_SYSTEM|			\
 198			     GFS2_DIF_INHERIT_JDATA)
 199
 200/**
 201 * gfs2_set_flags - set flags on an inode
 202 * @inode: The inode
 203 * @flags: The flags to set
 204 * @mask: Indicates which flags are valid
 205 *
 206 */
 207static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
 208{
 209	struct inode *inode = filp->f_path.dentry->d_inode;
 210	struct gfs2_inode *ip = GFS2_I(inode);
 211	struct gfs2_sbd *sdp = GFS2_SB(inode);
 212	struct buffer_head *bh;
 213	struct gfs2_holder gh;
 214	int error;
 215	u32 new_flags, flags;
 216
 217	error = mnt_want_write(filp->f_path.mnt);
 218	if (error)
 219		return error;
 220
 221	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 222	if (error)
 223		goto out_drop_write;
 224
 225	error = -EACCES;
 226	if (!inode_owner_or_capable(inode))
 227		goto out;
 228
 229	error = 0;
 230	flags = ip->i_diskflags;
 231	new_flags = (flags & ~mask) | (reqflags & mask);
 232	if ((new_flags ^ flags) == 0)
 233		goto out;
 234
 235	error = -EINVAL;
 236	if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
 237		goto out;
 238
 239	error = -EPERM;
 240	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
 241		goto out;
 242	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
 243		goto out;
 244	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
 245	    !capable(CAP_LINUX_IMMUTABLE))
 246		goto out;
 247	if (!IS_IMMUTABLE(inode)) {
 248		error = gfs2_permission(inode, MAY_WRITE);
 249		if (error)
 250			goto out;
 251	}
 252	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
 253		if (flags & GFS2_DIF_JDATA)
 254			gfs2_log_flush(sdp, ip->i_gl);
 
 
 255		error = filemap_fdatawrite(inode->i_mapping);
 256		if (error)
 257			goto out;
 258		error = filemap_fdatawait(inode->i_mapping);
 259		if (error)
 260			goto out;
 
 
 
 261	}
 262	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 263	if (error)
 264		goto out;
 265	error = gfs2_meta_inode_buffer(ip, &bh);
 266	if (error)
 267		goto out_trans_end;
 268	gfs2_trans_add_bh(ip->i_gl, bh, 1);
 
 269	ip->i_diskflags = new_flags;
 270	gfs2_dinode_out(ip, bh->b_data);
 271	brelse(bh);
 272	gfs2_set_inode_flags(inode);
 273	gfs2_set_aops(inode);
 274out_trans_end:
 275	gfs2_trans_end(sdp);
 276out:
 277	gfs2_glock_dq_uninit(&gh);
 278out_drop_write:
 279	mnt_drop_write(filp->f_path.mnt);
 280	return error;
 281}
 282
 283static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
 
 284{
 285	struct inode *inode = filp->f_path.dentry->d_inode;
 286	u32 fsflags, gfsflags;
 
 
 287
 288	if (get_user(fsflags, ptr))
 289		return -EFAULT;
 
 
 
 290
 291	gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
 292	if (!S_ISDIR(inode->i_mode)) {
 293		if (gfsflags & GFS2_DIF_INHERIT_JDATA)
 294			gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
 295		return do_gfs2_set_flags(filp, gfsflags, ~0);
 296	}
 297	return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 298}
 299
 300static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 301{
 302	switch(cmd) {
 303	case FS_IOC_GETFLAGS:
 304		return gfs2_get_flags(filp, (u32 __user *)arg);
 305	case FS_IOC_SETFLAGS:
 306		return gfs2_set_flags(filp, (u32 __user *)arg);
 307	}
 
 308	return -ENOTTY;
 309}
 310
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 311/**
 312 * gfs2_allocate_page_backing - Use bmap to allocate blocks
 313 * @page: The (locked) page to allocate backing for
 
 
 
 
 
 
 314 *
 315 * We try to allocate all the blocks required for the page in
 316 * one go. This might fail for various reasons, so we keep
 317 * trying until all the blocks to back this page are allocated.
 318 * If some of the blocks are already allocated, thats ok too.
 319 */
 320
 321static int gfs2_allocate_page_backing(struct page *page)
 322{
 323	struct inode *inode = page->mapping->host;
 324	struct buffer_head bh;
 325	unsigned long size = PAGE_CACHE_SIZE;
 326	u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 327
 328	do {
 329		bh.b_state = 0;
 330		bh.b_size = size;
 331		gfs2_block_map(inode, lblock, &bh, 1);
 332		if (!buffer_mapped(&bh))
 333			return -EIO;
 334		size -= bh.b_size;
 335		lblock += (bh.b_size >> inode->i_blkbits);
 336	} while(size > 0);
 
 
 
 
 337	return 0;
 338}
 339
 340/**
 341 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
 342 * @vma: The virtual memory area
 343 * @page: The page which is about to become writable
 344 *
 345 * When the page becomes writable, we need to ensure that we have
 346 * blocks allocated on disk to back that page.
 347 */
 348
 349static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 350{
 351	struct page *page = vmf->page;
 352	struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
 353	struct gfs2_inode *ip = GFS2_I(inode);
 354	struct gfs2_sbd *sdp = GFS2_SB(inode);
 355	unsigned long last_index;
 356	u64 pos = page->index << PAGE_CACHE_SHIFT;
 357	unsigned int data_blocks, ind_blocks, rblocks;
 
 358	struct gfs2_holder gh;
 359	struct gfs2_alloc *al;
 360	int ret;
 
 
 
 361
 362	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 363	ret = gfs2_glock_nq(&gh);
 364	if (ret)
 365		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 366
 367	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
 368	set_bit(GIF_SW_PAGED, &ip->i_flags);
 369
 370	if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE))
 
 
 
 
 
 
 
 
 
 
 
 
 371		goto out_unlock;
 372	ret = -ENOMEM;
 373	al = gfs2_alloc_get(ip);
 374	if (al == NULL)
 
 
 375		goto out_unlock;
 
 376
 377	ret = gfs2_quota_lock_check(ip);
 378	if (ret)
 379		goto out_alloc_put;
 380	gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
 381	al->al_requested = data_blocks + ind_blocks;
 382	ret = gfs2_inplace_reserve(ip);
 383	if (ret)
 
 
 
 384		goto out_quota_unlock;
 
 385
 386	rblocks = RES_DINODE + ind_blocks;
 387	if (gfs2_is_jdata(ip))
 388		rblocks += data_blocks ? data_blocks : 1;
 389	if (ind_blocks || data_blocks) {
 390		rblocks += RES_STATFS + RES_QUOTA;
 391		rblocks += gfs2_rg_blocks(al);
 392	}
 393	ret = gfs2_trans_begin(sdp, rblocks, 0);
 394	if (ret)
 
 395		goto out_trans_fail;
 
 396
 397	lock_page(page);
 398	ret = -EINVAL;
 399	last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT;
 400	if (page->index > last_index)
 401		goto out_unlock_page;
 402	ret = 0;
 403	if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping)
 404		goto out_unlock_page;
 405	if (gfs2_is_stuffed(ip)) {
 406		ret = gfs2_unstuff_dinode(ip, page);
 407		if (ret)
 408			goto out_unlock_page;
 
 
 409	}
 410	ret = gfs2_allocate_page_backing(page);
 411
 412out_unlock_page:
 413	unlock_page(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 414	gfs2_trans_end(sdp);
 415out_trans_fail:
 416	gfs2_inplace_release(ip);
 417out_quota_unlock:
 418	gfs2_quota_unlock(ip);
 419out_alloc_put:
 420	gfs2_alloc_put(ip);
 421out_unlock:
 422	gfs2_glock_dq(&gh);
 423out:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 424	gfs2_holder_uninit(&gh);
 425	if (ret == -ENOMEM)
 426		ret = VM_FAULT_OOM;
 427	else if (ret)
 428		ret = VM_FAULT_SIGBUS;
 429	return ret;
 430}
 431
 432static const struct vm_operations_struct gfs2_vm_ops = {
 433	.fault = filemap_fault,
 
 434	.page_mkwrite = gfs2_page_mkwrite,
 435};
 436
 437/**
 438 * gfs2_mmap -
 439 * @file: The file to map
 440 * @vma: The VMA which described the mapping
 441 *
 442 * There is no need to get a lock here unless we should be updating
 443 * atime. We ignore any locking errors since the only consequence is
 444 * a missed atime update (which will just be deferred until later).
 445 *
 446 * Returns: 0
 447 */
 448
 449static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 450{
 451	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 452
 453	if (!(file->f_flags & O_NOATIME) &&
 454	    !IS_NOATIME(&ip->i_inode)) {
 455		struct gfs2_holder i_gh;
 456		int error;
 457
 458		gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 459		error = gfs2_glock_nq(&i_gh);
 460		if (error == 0) {
 461			file_accessed(file);
 462			gfs2_glock_dq(&i_gh);
 463		}
 464		gfs2_holder_uninit(&i_gh);
 465		if (error)
 466			return error;
 
 
 
 467	}
 468	vma->vm_ops = &gfs2_vm_ops;
 469	vma->vm_flags |= VM_CAN_NONLINEAR;
 470
 471	return 0;
 472}
 473
 474/**
 475 * gfs2_open - open a file
 476 * @inode: the inode to open
 477 * @file: the struct file for this opening
 
 
 
 
 
 478 *
 479 * Returns: errno
 480 */
 481
 482static int gfs2_open(struct inode *inode, struct file *file)
 483{
 484	struct gfs2_inode *ip = GFS2_I(inode);
 485	struct gfs2_holder i_gh;
 486	struct gfs2_file *fp;
 487	int error;
 488
 489	fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
 
 
 
 
 
 
 
 
 
 490	if (!fp)
 491		return -ENOMEM;
 492
 493	mutex_init(&fp->f_fl_mutex);
 494
 495	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
 496	file->private_data = fp;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 497
 498	if (S_ISREG(ip->i_inode.i_mode)) {
 499		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 500					   &i_gh);
 501		if (error)
 502			goto fail;
 
 
 503
 504		if (!(file->f_flags & O_LARGEFILE) &&
 505		    i_size_read(inode) > MAX_NON_LFS) {
 506			error = -EOVERFLOW;
 507			goto fail_gunlock;
 508		}
 509
 
 510		gfs2_glock_dq_uninit(&i_gh);
 511	}
 512
 513	return 0;
 514
 515fail_gunlock:
 516	gfs2_glock_dq_uninit(&i_gh);
 517fail:
 518	file->private_data = NULL;
 519	kfree(fp);
 520	return error;
 521}
 522
 523/**
 524 * gfs2_close - called to close a struct file
 525 * @inode: the inode the struct file belongs to
 526 * @file: the struct file being closed
 527 *
 528 * Returns: errno
 529 */
 530
 531static int gfs2_close(struct inode *inode, struct file *file)
 532{
 533	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
 534	struct gfs2_file *fp;
 535
 536	fp = file->private_data;
 537	file->private_data = NULL;
 538
 539	if (gfs2_assert_warn(sdp, fp))
 540		return -EIO;
 541
 542	kfree(fp);
 543
 544	return 0;
 545}
 546
 547/**
 548 * gfs2_fsync - sync the dirty data for a file (across the cluster)
 549 * @file: the file that points to the dentry
 550 * @start: the start position in the file to sync
 551 * @end: the end position in the file to sync
 552 * @datasync: set if we can ignore timestamp changes
 553 *
 554 * The VFS will flush data for us. We only need to worry
 555 * about metadata here.
 
 
 
 
 
 
 
 
 556 *
 557 * Returns: errno
 558 */
 559
 560static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 561		      int datasync)
 562{
 563	struct inode *inode = file->f_mapping->host;
 564	int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
 
 565	struct gfs2_inode *ip = GFS2_I(inode);
 566	int ret;
 567
 568	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
 569	if (ret)
 570		return ret;
 571	mutex_lock(&inode->i_mutex);
 
 572
 
 
 573	if (datasync)
 574		sync_state &= ~I_DIRTY_SYNC;
 575
 576	if (sync_state) {
 577		ret = sync_inode_metadata(inode, 1);
 578		if (ret) {
 579			mutex_unlock(&inode->i_mutex);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 580			return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 581		}
 582		gfs2_ail_flush(ip->i_gl);
 583	}
 
 
 
 584
 585	mutex_unlock(&inode->i_mutex);
 586	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 587}
 588
 589/**
 590 * gfs2_file_aio_write - Perform a write to a file
 591 * @iocb: The io context
 592 * @iov: The data to write
 593 * @nr_segs: Number of @iov segments
 594 * @pos: The file position
 595 *
 596 * We have to do a lock/unlock here to refresh the inode size for
 597 * O_APPEND writes, otherwise we can land up writing at the wrong
 598 * offset. There is still a race, but provided the app is using its
 599 * own file locking, this will make O_APPEND work as expected.
 600 *
 601 */
 602
 603static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
 604				   unsigned long nr_segs, loff_t pos)
 605{
 606	struct file *file = iocb->ki_filp;
 
 
 
 
 607
 608	if (file->f_flags & O_APPEND) {
 609		struct dentry *dentry = file->f_dentry;
 610		struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
 611		struct gfs2_holder gh;
 612		int ret;
 613
 
 614		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 615		if (ret)
 616			return ret;
 617		gfs2_glock_dq_uninit(&gh);
 618	}
 619
 620	return generic_file_aio_write(iocb, iov, nr_segs, pos);
 621}
 622
 623static int empty_write_end(struct page *page, unsigned from,
 624			   unsigned to, int mode)
 625{
 626	struct inode *inode = page->mapping->host;
 627	struct gfs2_inode *ip = GFS2_I(inode);
 628	struct buffer_head *bh;
 629	unsigned offset, blksize = 1 << inode->i_blkbits;
 630	pgoff_t end_index = i_size_read(inode) >> PAGE_CACHE_SHIFT;
 631
 632	zero_user(page, from, to-from);
 633	mark_page_accessed(page);
 634
 635	if (page->index < end_index || !(mode & FALLOC_FL_KEEP_SIZE)) {
 636		if (!gfs2_is_writeback(ip))
 637			gfs2_page_add_databufs(ip, page, from, to);
 638
 639		block_commit_write(page, from, to);
 640		return 0;
 641	}
 
 
 
 
 
 
 642
 643	offset = 0;
 644	bh = page_buffers(page);
 645	while (offset < to) {
 646		if (offset >= from) {
 647			set_buffer_uptodate(bh);
 648			mark_buffer_dirty(bh);
 649			clear_buffer_new(bh);
 650			write_dirty_buffer(bh, WRITE);
 651		}
 652		offset += blksize;
 653		bh = bh->b_this_page;
 654	}
 655
 656	offset = 0;
 657	bh = page_buffers(page);
 658	while (offset < to) {
 659		if (offset >= from) {
 660			wait_on_buffer(bh);
 661			if (!buffer_uptodate(bh))
 662				return -EIO;
 663		}
 664		offset += blksize;
 665		bh = bh->b_this_page;
 666	}
 667	return 0;
 668}
 669
 670static int needs_empty_write(sector_t block, struct inode *inode)
 671{
 672	int error;
 673	struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 674
 675	bh_map.b_size = 1 << inode->i_blkbits;
 676	error = gfs2_block_map(inode, block, &bh_map, 0);
 677	if (unlikely(error))
 678		return error;
 679	return !buffer_mapped(&bh_map);
 680}
 681
 682static int write_empty_blocks(struct page *page, unsigned from, unsigned to,
 683			      int mode)
 684{
 685	struct inode *inode = page->mapping->host;
 686	unsigned start, end, next, blksize;
 687	sector_t block = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 688	int ret;
 
 
 
 689
 690	blksize = 1 << inode->i_blkbits;
 691	next = end = 0;
 692	while (next < from) {
 693		next += blksize;
 694		block++;
 695	}
 696	start = next;
 697	do {
 698		next += blksize;
 699		ret = needs_empty_write(block, inode);
 700		if (unlikely(ret < 0))
 701			return ret;
 702		if (ret == 0) {
 703			if (end) {
 704				ret = __block_write_begin(page, start, end - start,
 705							  gfs2_block_map);
 706				if (unlikely(ret))
 707					return ret;
 708				ret = empty_write_end(page, start, end, mode);
 709				if (unlikely(ret))
 710					return ret;
 711				end = 0;
 712			}
 713			start = next;
 714		}
 715		else
 716			end = next;
 717		block++;
 718	} while (next < to);
 719
 720	if (end) {
 721		ret = __block_write_begin(page, start, end - start, gfs2_block_map);
 722		if (unlikely(ret))
 723			return ret;
 724		ret = empty_write_end(page, start, end, mode);
 725		if (unlikely(ret))
 726			return ret;
 727	}
 728
 729	return 0;
 
 
 730}
 731
 732static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
 733			   int mode)
 734{
 
 735	struct gfs2_inode *ip = GFS2_I(inode);
 
 736	struct buffer_head *dibh;
 737	int error;
 738	u64 start = offset >> PAGE_CACHE_SHIFT;
 739	unsigned int start_offset = offset & ~PAGE_CACHE_MASK;
 740	u64 end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
 741	pgoff_t curr;
 742	struct page *page;
 743	unsigned int end_offset = (offset + len) & ~PAGE_CACHE_MASK;
 744	unsigned int from, to;
 745
 746	if (!end_offset)
 747		end_offset = PAGE_CACHE_SIZE;
 748
 749	error = gfs2_meta_inode_buffer(ip, &dibh);
 750	if (unlikely(error))
 751		goto out;
 752
 753	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 754
 755	if (gfs2_is_stuffed(ip)) {
 756		error = gfs2_unstuff_dinode(ip, NULL);
 757		if (unlikely(error))
 758			goto out;
 759	}
 760
 761	curr = start;
 762	offset = start << PAGE_CACHE_SHIFT;
 763	from = start_offset;
 764	to = PAGE_CACHE_SIZE;
 765	while (curr <= end) {
 766		page = grab_cache_page_write_begin(inode->i_mapping, curr,
 767						   AOP_FLAG_NOFS);
 768		if (unlikely(!page)) {
 769			error = -ENOMEM;
 770			goto out;
 771		}
 772
 773		if (curr == end)
 774			to = end_offset;
 775		error = write_empty_blocks(page, from, to, mode);
 776		if (!error && offset + to > inode->i_size &&
 777		    !(mode & FALLOC_FL_KEEP_SIZE)) {
 778			i_size_write(inode, offset + to);
 779		}
 780		unlock_page(page);
 781		page_cache_release(page);
 782		if (error)
 783			goto out;
 784		curr++;
 785		offset += PAGE_CACHE_SIZE;
 786		from = 0;
 
 
 
 
 
 
 
 787	}
 788
 789	gfs2_dinode_out(ip, dibh->b_data);
 790	mark_inode_dirty(inode);
 791
 792	brelse(dibh);
 793
 794out:
 
 795	return error;
 796}
 797
 798static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len,
 799			    unsigned int *data_blocks, unsigned int *ind_blocks)
 
 
 
 
 
 
 
 
 
 
 
 
 800{
 
 801	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 802	unsigned int max_blocks = ip->i_alloc->al_rgd->rd_free_clone;
 803	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
 804
 805	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
 806		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
 807		max_data -= tmp;
 808	}
 809	/* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
 810	   so it might end up with fewer data blocks */
 811	if (max_data <= *data_blocks)
 812		return;
 813	*data_blocks = max_data;
 814	*ind_blocks = max_blocks - max_data;
 815	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
 816	if (*len > max) {
 817		*len = max;
 818		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
 819	}
 820}
 821
 822static long gfs2_fallocate(struct file *file, int mode, loff_t offset,
 823			   loff_t len)
 824{
 825	struct inode *inode = file->f_path.dentry->d_inode;
 826	struct gfs2_sbd *sdp = GFS2_SB(inode);
 827	struct gfs2_inode *ip = GFS2_I(inode);
 
 828	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
 829	loff_t bytes, max_bytes;
 830	struct gfs2_alloc *al;
 831	int error;
 
 
 832	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
 833	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
 834	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
 835
 836	/* We only support the FALLOC_FL_KEEP_SIZE mode */
 837	if (mode & ~FALLOC_FL_KEEP_SIZE)
 838		return -EOPNOTSUPP;
 839
 840	offset &= bsize_mask;
 841
 842	len = next - offset;
 843	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
 844	if (!bytes)
 845		bytes = UINT_MAX;
 846	bytes &= bsize_mask;
 847	if (bytes == 0)
 848		bytes = sdp->sd_sb.sb_bsize;
 849
 850	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
 851	error = gfs2_glock_nq(&ip->i_gh);
 852	if (unlikely(error))
 853		goto out_uninit;
 854
 855	if (!gfs2_write_alloc_required(ip, offset, len))
 856		goto out_unlock;
 857
 858	while (len > 0) {
 859		if (len < bytes)
 860			bytes = len;
 861		al = gfs2_alloc_get(ip);
 862		if (!al) {
 863			error = -ENOMEM;
 864			goto out_unlock;
 865		}
 866
 867		error = gfs2_quota_lock_check(ip);
 868		if (error)
 869			goto out_alloc_put;
 870
 871retry:
 
 
 
 
 
 872		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
 
 873
 874		al->al_requested = data_blocks + ind_blocks;
 875		error = gfs2_inplace_reserve(ip);
 876		if (error) {
 877			if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) {
 878				bytes >>= 1;
 879				bytes &= bsize_mask;
 880				if (bytes == 0)
 881					bytes = sdp->sd_sb.sb_bsize;
 882				goto retry;
 883			}
 
 884			goto out_qunlock;
 885		}
 886		max_bytes = bytes;
 887		calc_max_reserv(ip, len, &max_bytes, &data_blocks, &ind_blocks);
 888		al->al_requested = data_blocks + ind_blocks;
 
 
 
 
 
 
 889
 890		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
 891			  RES_RG_HDR + gfs2_rg_blocks(al);
 892		if (gfs2_is_jdata(ip))
 893			rblocks += data_blocks ? data_blocks : 1;
 894
 895		error = gfs2_trans_begin(sdp, rblocks,
 896					 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
 897		if (error)
 898			goto out_trans_fail;
 899
 900		error = fallocate_chunk(inode, offset, max_bytes, mode);
 901		gfs2_trans_end(sdp);
 902
 903		if (error)
 904			goto out_trans_fail;
 905
 906		len -= max_bytes;
 907		offset += max_bytes;
 908		gfs2_inplace_release(ip);
 909		gfs2_quota_unlock(ip);
 910		gfs2_alloc_put(ip);
 911	}
 912	goto out_unlock;
 
 
 
 
 
 
 
 
 
 913
 914out_trans_fail:
 915	gfs2_inplace_release(ip);
 916out_qunlock:
 917	gfs2_quota_unlock(ip);
 918out_alloc_put:
 919	gfs2_alloc_put(ip);
 920out_unlock:
 921	gfs2_glock_dq(&ip->i_gh);
 922out_uninit:
 923	gfs2_holder_uninit(&ip->i_gh);
 924	return error;
 925}
 926
 927#ifdef CONFIG_GFS2_FS_LOCKING_DLM
 
 
 
 
 
 
 928
 929/**
 930 * gfs2_setlease - acquire/release a file lease
 931 * @file: the file pointer
 932 * @arg: lease type
 933 * @fl: file lock
 934 *
 935 * We don't currently have a way to enforce a lease across the whole
 936 * cluster; until we do, disable leases (by just returning -EINVAL),
 937 * unless the administrator has requested purely local locking.
 938 *
 939 * Locking: called under lock_flocks
 940 *
 941 * Returns: errno
 942 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 943
 944static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
 
 
 
 
 
 
 
 
 
 
 
 945{
 946	return -EINVAL;
 
 
 
 
 
 947}
 948
 
 
 949/**
 950 * gfs2_lock - acquire/release a posix lock on a file
 951 * @file: the file pointer
 952 * @cmd: either modify or retrieve lock state, possibly wait
 953 * @fl: type and range of lock
 954 *
 955 * Returns: errno
 956 */
 957
 958static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
 959{
 960	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 961	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
 962	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 963
 964	if (!(fl->fl_flags & FL_POSIX))
 965		return -ENOLCK;
 966	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
 967		return -ENOLCK;
 968
 969	if (cmd == F_CANCELLK) {
 970		/* Hack: */
 971		cmd = F_SETLK;
 972		fl->fl_type = F_UNLCK;
 973	}
 974	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 975		return -EIO;
 976	if (IS_GETLK(cmd))
 
 
 
 977		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
 978	else if (fl->fl_type == F_UNLCK)
 979		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
 980	else
 981		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
 982}
 983
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 984static int do_flock(struct file *file, int cmd, struct file_lock *fl)
 985{
 986	struct gfs2_file *fp = file->private_data;
 987	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
 988	struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
 989	struct gfs2_glock *gl;
 990	unsigned int state;
 991	int flags;
 992	int error = 0;
 
 993
 994	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
 995	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
 
 
 996
 997	mutex_lock(&fp->f_fl_mutex);
 998
 999	gl = fl_gh->gh_gl;
1000	if (gl) {
1001		if (fl_gh->gh_state == state)
1002			goto out;
1003		flock_lock_file_wait(file,
1004				     &(struct file_lock){.fl_type = F_UNLCK});
1005		gfs2_glock_dq_wait(fl_gh);
 
 
1006		gfs2_holder_reinit(state, flags, fl_gh);
1007	} else {
1008		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1009				       &gfs2_flock_glops, CREATE, &gl);
1010		if (error)
1011			goto out;
 
1012		gfs2_holder_init(gl, state, flags, fl_gh);
 
1013		gfs2_glock_put(gl);
1014	}
1015	error = gfs2_glock_nq(fl_gh);
 
 
 
 
 
 
 
1016	if (error) {
1017		gfs2_holder_uninit(fl_gh);
1018		if (error == GLR_TRYFAILED)
1019			error = -EAGAIN;
1020	} else {
1021		error = flock_lock_file_wait(file, fl);
1022		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1023	}
1024
1025out:
1026	mutex_unlock(&fp->f_fl_mutex);
1027	return error;
1028}
1029
1030static void do_unflock(struct file *file, struct file_lock *fl)
1031{
1032	struct gfs2_file *fp = file->private_data;
1033	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1034
1035	mutex_lock(&fp->f_fl_mutex);
1036	flock_lock_file_wait(file, fl);
1037	if (fl_gh->gh_gl) {
1038		gfs2_glock_dq_wait(fl_gh);
1039		gfs2_holder_uninit(fl_gh);
1040	}
1041	mutex_unlock(&fp->f_fl_mutex);
1042}
1043
1044/**
1045 * gfs2_flock - acquire/release a flock lock on a file
1046 * @file: the file pointer
1047 * @cmd: either modify or retrieve lock state, possibly wait
1048 * @fl: type and range of lock
1049 *
1050 * Returns: errno
1051 */
1052
1053static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1054{
1055	if (!(fl->fl_flags & FL_FLOCK))
1056		return -ENOLCK;
1057	if (fl->fl_type & LOCK_MAND)
1058		return -EOPNOTSUPP;
1059
1060	if (fl->fl_type == F_UNLCK) {
1061		do_unflock(file, fl);
1062		return 0;
1063	} else {
1064		return do_flock(file, cmd, fl);
1065	}
1066}
1067
1068const struct file_operations gfs2_file_fops = {
1069	.llseek		= gfs2_llseek,
1070	.read		= do_sync_read,
1071	.aio_read	= generic_file_aio_read,
1072	.write		= do_sync_write,
1073	.aio_write	= gfs2_file_aio_write,
1074	.unlocked_ioctl	= gfs2_ioctl,
 
1075	.mmap		= gfs2_mmap,
1076	.open		= gfs2_open,
1077	.release	= gfs2_close,
1078	.fsync		= gfs2_fsync,
1079	.lock		= gfs2_lock,
1080	.flock		= gfs2_flock,
1081	.splice_read	= generic_file_splice_read,
1082	.splice_write	= generic_file_splice_write,
1083	.setlease	= gfs2_setlease,
1084	.fallocate	= gfs2_fallocate,
 
1085};
1086
1087const struct file_operations gfs2_dir_fops = {
1088	.readdir	= gfs2_readdir,
1089	.unlocked_ioctl	= gfs2_ioctl,
 
1090	.open		= gfs2_open,
1091	.release	= gfs2_close,
1092	.fsync		= gfs2_fsync,
1093	.lock		= gfs2_lock,
1094	.flock		= gfs2_flock,
1095	.llseek		= default_llseek,
 
1096};
1097
1098#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1099
1100const struct file_operations gfs2_file_fops_nolock = {
1101	.llseek		= gfs2_llseek,
1102	.read		= do_sync_read,
1103	.aio_read	= generic_file_aio_read,
1104	.write		= do_sync_write,
1105	.aio_write	= gfs2_file_aio_write,
1106	.unlocked_ioctl	= gfs2_ioctl,
 
1107	.mmap		= gfs2_mmap,
1108	.open		= gfs2_open,
1109	.release	= gfs2_close,
1110	.fsync		= gfs2_fsync,
1111	.splice_read	= generic_file_splice_read,
1112	.splice_write	= generic_file_splice_write,
1113	.setlease	= generic_setlease,
1114	.fallocate	= gfs2_fallocate,
1115};
1116
1117const struct file_operations gfs2_dir_fops_nolock = {
1118	.readdir	= gfs2_readdir,
1119	.unlocked_ioctl	= gfs2_ioctl,
 
1120	.open		= gfs2_open,
1121	.release	= gfs2_close,
1122	.fsync		= gfs2_fsync,
1123	.llseek		= default_llseek,
1124};
1125
v6.13.7
   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		truncate_inode_pages(inode->i_mapping, 0);
 255		if (new_flags & GFS2_DIF_JDATA)
 256			gfs2_ordered_del_inode(ip);
 257	}
 258	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 259	if (error)
 260		goto out;
 261	error = gfs2_meta_inode_buffer(ip, &bh);
 262	if (error)
 263		goto out_trans_end;
 264	inode_set_ctime_current(inode);
 265	gfs2_trans_add_meta(ip->i_gl, bh);
 266	ip->i_diskflags = new_flags;
 267	gfs2_dinode_out(ip, bh->b_data);
 268	brelse(bh);
 269	gfs2_set_inode_flags(inode);
 270	gfs2_set_aops(inode);
 271out_trans_end:
 272	gfs2_trans_end(sdp);
 273out:
 274	gfs2_glock_dq_uninit(&gh);
 
 
 275	return error;
 276}
 277
 278int gfs2_fileattr_set(struct mnt_idmap *idmap,
 279		      struct dentry *dentry, struct fileattr *fa)
 280{
 281	struct inode *inode = d_inode(dentry);
 282	u32 fsflags = fa->flags, gfsflags = 0;
 283	u32 mask;
 284	int i;
 285
 286	if (d_is_special(dentry))
 287		return -ENOTTY;
 288
 289	if (fileattr_has_fsx(fa))
 290		return -EOPNOTSUPP;
 291
 292	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
 293		if (fsflags & fsflag_gfs2flag[i].fsflag) {
 294			fsflags &= ~fsflag_gfs2flag[i].fsflag;
 295			gfsflags |= fsflag_gfs2flag[i].gfsflag;
 296		}
 297	}
 298	if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
 299		return -EINVAL;
 300
 301	mask = GFS2_FLAGS_USER_SET;
 302	if (S_ISDIR(inode->i_mode)) {
 303		mask &= ~GFS2_DIF_JDATA;
 304	} else {
 305		/* The GFS2_DIF_TOPDIR flag is only valid for directories. */
 306		if (gfsflags & GFS2_DIF_TOPDIR)
 307			return -EINVAL;
 308		mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
 309	}
 310
 311	return do_gfs2_set_flags(inode, gfsflags, mask);
 312}
 313
 314static int gfs2_getlabel(struct file *filp, char __user *label)
 315{
 316	struct inode *inode = file_inode(filp);
 317	struct gfs2_sbd *sdp = GFS2_SB(inode);
 318
 319	if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
 320		return -EFAULT;
 321
 322	return 0;
 323}
 324
 325static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 326{
 327	switch(cmd) {
 328	case FITRIM:
 329		return gfs2_fitrim(filp, (void __user *)arg);
 330	case FS_IOC_GETFSLABEL:
 331		return gfs2_getlabel(filp, (char __user *)arg);
 332	}
 333
 334	return -ENOTTY;
 335}
 336
 337#ifdef CONFIG_COMPAT
 338static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 339{
 340	switch(cmd) {
 341	/* Keep this list in sync with gfs2_ioctl */
 342	case FITRIM:
 343	case FS_IOC_GETFSLABEL:
 344		break;
 345	default:
 346		return -ENOIOCTLCMD;
 347	}
 348
 349	return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
 350}
 351#else
 352#define gfs2_compat_ioctl NULL
 353#endif
 354
 355/**
 356 * gfs2_size_hint - Give a hint to the size of a write request
 357 * @filep: The struct file
 358 * @offset: The file offset of the write
 359 * @size: The length of the write
 360 *
 361 * When we are about to do a write, this function records the total
 362 * write size in order to provide a suitable hint to the lower layers
 363 * about how many blocks will be required.
 364 *
 
 
 
 
 365 */
 366
 367static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
 368{
 369	struct inode *inode = file_inode(filep);
 370	struct gfs2_sbd *sdp = GFS2_SB(inode);
 371	struct gfs2_inode *ip = GFS2_I(inode);
 372	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
 373	int hint = min_t(size_t, INT_MAX, blks);
 374
 375	if (hint > atomic_read(&ip->i_sizehint))
 376		atomic_set(&ip->i_sizehint, hint);
 377}
 378
 379/**
 380 * gfs2_allocate_folio_backing - Allocate blocks for a write fault
 381 * @folio: The (locked) folio to allocate backing for
 382 * @length: Size of the allocation
 383 *
 384 * We try to allocate all the blocks required for the folio in one go.  This
 385 * might fail for various reasons, so we keep trying until all the blocks to
 386 * back this folio are allocated.  If some of the blocks are already allocated,
 387 * that is ok too.
 388 */
 389static int gfs2_allocate_folio_backing(struct folio *folio, size_t length)
 390{
 391	u64 pos = folio_pos(folio);
 392
 393	do {
 394		struct iomap iomap = { };
 395
 396		if (gfs2_iomap_alloc(folio->mapping->host, pos, length, &iomap))
 
 397			return -EIO;
 398
 399		if (length < iomap.length)
 400			iomap.length = length;
 401		length -= iomap.length;
 402		pos += iomap.length;
 403	} while (length > 0);
 404
 405	return 0;
 406}
 407
 408/**
 409 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
 410 * @vmf: The virtual memory fault containing the page to become writable
 
 411 *
 412 * When the page becomes writable, we need to ensure that we have
 413 * blocks allocated on disk to back that page.
 414 */
 415
 416static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
 417{
 418	struct folio *folio = page_folio(vmf->page);
 419	struct inode *inode = file_inode(vmf->vma->vm_file);
 420	struct gfs2_inode *ip = GFS2_I(inode);
 421	struct gfs2_sbd *sdp = GFS2_SB(inode);
 422	struct gfs2_alloc_parms ap = {};
 423	u64 pos = folio_pos(folio);
 424	unsigned int data_blocks, ind_blocks, rblocks;
 425	vm_fault_t ret = VM_FAULT_LOCKED;
 426	struct gfs2_holder gh;
 427	size_t length;
 428	loff_t size;
 429	int err;
 430
 431	sb_start_pagefault(inode->i_sb);
 432
 433	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 434	err = gfs2_glock_nq(&gh);
 435	if (err) {
 436		ret = vmf_fs_error(err);
 437		goto out_uninit;
 438	}
 439
 440	/* Check folio index against inode size */
 441	size = i_size_read(inode);
 442	if (pos >= size) {
 443		ret = VM_FAULT_SIGBUS;
 444		goto out_unlock;
 445	}
 446
 447	/* Update file times before taking folio lock */
 448	file_update_time(vmf->vma->vm_file);
 449
 450	/* folio is wholly or partially inside EOF */
 451	if (size - pos < folio_size(folio))
 452		length = size - pos;
 453	else
 454		length = folio_size(folio);
 455
 456	gfs2_size_hint(vmf->vma->vm_file, pos, length);
 457
 458	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
 459	set_bit(GIF_SW_PAGED, &ip->i_flags);
 460
 461	/*
 462	 * iomap_writepage / iomap_writepages currently don't support inline
 463	 * files, so always unstuff here.
 464	 */
 465
 466	if (!gfs2_is_stuffed(ip) &&
 467	    !gfs2_write_alloc_required(ip, pos, length)) {
 468		folio_lock(folio);
 469		if (!folio_test_uptodate(folio) ||
 470		    folio->mapping != inode->i_mapping) {
 471			ret = VM_FAULT_NOPAGE;
 472			folio_unlock(folio);
 473		}
 474		goto out_unlock;
 475	}
 476
 477	err = gfs2_rindex_update(sdp);
 478	if (err) {
 479		ret = vmf_fs_error(err);
 480		goto out_unlock;
 481	}
 482
 483	gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
 484	ap.target = data_blocks + ind_blocks;
 485	err = gfs2_quota_lock_check(ip, &ap);
 486	if (err) {
 487		ret = vmf_fs_error(err);
 488		goto out_unlock;
 489	}
 490	err = gfs2_inplace_reserve(ip, &ap);
 491	if (err) {
 492		ret = vmf_fs_error(err);
 493		goto out_quota_unlock;
 494	}
 495
 496	rblocks = RES_DINODE + ind_blocks;
 497	if (gfs2_is_jdata(ip))
 498		rblocks += data_blocks ? data_blocks : 1;
 499	if (ind_blocks || data_blocks) {
 500		rblocks += RES_STATFS + RES_QUOTA;
 501		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
 502	}
 503	err = gfs2_trans_begin(sdp, rblocks, 0);
 504	if (err) {
 505		ret = vmf_fs_error(err);
 506		goto out_trans_fail;
 507	}
 508
 509	/* Unstuff, if required, and allocate backing blocks for folio */
 
 
 
 
 
 
 
 510	if (gfs2_is_stuffed(ip)) {
 511		err = gfs2_unstuff_dinode(ip);
 512		if (err) {
 513			ret = vmf_fs_error(err);
 514			goto out_trans_end;
 515		}
 516	}
 
 517
 518	folio_lock(folio);
 519	/* If truncated, we must retry the operation, we may have raced
 520	 * with the glock demotion code.
 521	 */
 522	if (!folio_test_uptodate(folio) || folio->mapping != inode->i_mapping) {
 523		ret = VM_FAULT_NOPAGE;
 524		goto out_page_locked;
 525	}
 526
 527	err = gfs2_allocate_folio_backing(folio, length);
 528	if (err)
 529		ret = vmf_fs_error(err);
 530
 531out_page_locked:
 532	if (ret != VM_FAULT_LOCKED)
 533		folio_unlock(folio);
 534out_trans_end:
 535	gfs2_trans_end(sdp);
 536out_trans_fail:
 537	gfs2_inplace_release(ip);
 538out_quota_unlock:
 539	gfs2_quota_unlock(ip);
 
 
 540out_unlock:
 541	gfs2_glock_dq(&gh);
 542out_uninit:
 543	gfs2_holder_uninit(&gh);
 544	if (ret == VM_FAULT_LOCKED) {
 545		folio_mark_dirty(folio);
 546		folio_wait_stable(folio);
 547	}
 548	sb_end_pagefault(inode->i_sb);
 549	return ret;
 550}
 551
 552static vm_fault_t gfs2_fault(struct vm_fault *vmf)
 553{
 554	struct inode *inode = file_inode(vmf->vma->vm_file);
 555	struct gfs2_inode *ip = GFS2_I(inode);
 556	struct gfs2_holder gh;
 557	vm_fault_t ret;
 558	int err;
 559
 560	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 561	err = gfs2_glock_nq(&gh);
 562	if (err) {
 563		ret = vmf_fs_error(err);
 564		goto out_uninit;
 565	}
 566	ret = filemap_fault(vmf);
 567	gfs2_glock_dq(&gh);
 568out_uninit:
 569	gfs2_holder_uninit(&gh);
 
 
 
 
 570	return ret;
 571}
 572
 573static const struct vm_operations_struct gfs2_vm_ops = {
 574	.fault = gfs2_fault,
 575	.map_pages = filemap_map_pages,
 576	.page_mkwrite = gfs2_page_mkwrite,
 577};
 578
 579/**
 580 * gfs2_mmap
 581 * @file: The file to map
 582 * @vma: The VMA which described the mapping
 583 *
 584 * There is no need to get a lock here unless we should be updating
 585 * atime. We ignore any locking errors since the only consequence is
 586 * a missed atime update (which will just be deferred until later).
 587 *
 588 * Returns: 0
 589 */
 590
 591static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 592{
 593	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 594
 595	if (!(file->f_flags & O_NOATIME) &&
 596	    !IS_NOATIME(&ip->i_inode)) {
 597		struct gfs2_holder i_gh;
 598		int error;
 599
 600		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 601					   &i_gh);
 
 
 
 
 
 602		if (error)
 603			return error;
 604		/* grab lock to update inode */
 605		gfs2_glock_dq_uninit(&i_gh);
 606		file_accessed(file);
 607	}
 608	vma->vm_ops = &gfs2_vm_ops;
 
 609
 610	return 0;
 611}
 612
 613/**
 614 * gfs2_open_common - This is common to open and atomic_open
 615 * @inode: The inode being opened
 616 * @file: The file being opened
 617 *
 618 * This maybe called under a glock or not depending upon how it has
 619 * been called. We must always be called under a glock for regular
 620 * files, however. For other file types, it does not matter whether
 621 * we hold the glock or not.
 622 *
 623 * Returns: Error code or 0 for success
 624 */
 625
 626int gfs2_open_common(struct inode *inode, struct file *file)
 627{
 
 
 628	struct gfs2_file *fp;
 629	int ret;
 630
 631	if (S_ISREG(inode->i_mode)) {
 632		ret = generic_file_open(inode, file);
 633		if (ret)
 634			return ret;
 635
 636		if (!gfs2_is_jdata(GFS2_I(inode)))
 637			file->f_mode |= FMODE_CAN_ODIRECT;
 638	}
 639
 640	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
 641	if (!fp)
 642		return -ENOMEM;
 643
 644	mutex_init(&fp->f_fl_mutex);
 645
 646	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
 647	file->private_data = fp;
 648	if (file->f_mode & FMODE_WRITE) {
 649		ret = gfs2_qa_get(GFS2_I(inode));
 650		if (ret)
 651			goto fail;
 652	}
 653	return 0;
 654
 655fail:
 656	kfree(file->private_data);
 657	file->private_data = NULL;
 658	return ret;
 659}
 660
 661/**
 662 * gfs2_open - open a file
 663 * @inode: the inode to open
 664 * @file: the struct file for this opening
 665 *
 666 * After atomic_open, this function is only used for opening files
 667 * which are already cached. We must still get the glock for regular
 668 * files to ensure that we have the file size uptodate for the large
 669 * file check which is in the common code. That is only an issue for
 670 * regular files though.
 671 *
 672 * Returns: errno
 673 */
 674
 675static int gfs2_open(struct inode *inode, struct file *file)
 676{
 677	struct gfs2_inode *ip = GFS2_I(inode);
 678	struct gfs2_holder i_gh;
 679	int error;
 680	bool need_unlock = false;
 681
 682	if (S_ISREG(ip->i_inode.i_mode)) {
 683		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 684					   &i_gh);
 685		if (error)
 686			return error;
 687		need_unlock = true;
 688	}
 689
 690	error = gfs2_open_common(inode, file);
 
 
 
 
 691
 692	if (need_unlock)
 693		gfs2_glock_dq_uninit(&i_gh);
 
 694
 
 
 
 
 
 
 
 695	return error;
 696}
 697
 698/**
 699 * gfs2_release - called to close a struct file
 700 * @inode: the inode the struct file belongs to
 701 * @file: the struct file being closed
 702 *
 703 * Returns: errno
 704 */
 705
 706static int gfs2_release(struct inode *inode, struct file *file)
 707{
 708	struct gfs2_inode *ip = GFS2_I(inode);
 
 709
 710	kfree(file->private_data);
 711	file->private_data = NULL;
 712
 713	if (file->f_mode & FMODE_WRITE) {
 714		if (gfs2_rs_active(&ip->i_res))
 715			gfs2_rs_delete(ip);
 716		gfs2_qa_put(ip);
 717	}
 718	return 0;
 719}
 720
 721/**
 722 * gfs2_fsync - sync the dirty data for a file (across the cluster)
 723 * @file: the file that points to the dentry
 724 * @start: the start position in the file to sync
 725 * @end: the end position in the file to sync
 726 * @datasync: set if we can ignore timestamp changes
 727 *
 728 * We split the data flushing here so that we don't wait for the data
 729 * until after we've also sent the metadata to disk. Note that for
 730 * data=ordered, we will write & wait for the data at the log flush
 731 * stage anyway, so this is unlikely to make much of a difference
 732 * except in the data=writeback case.
 733 *
 734 * If the fdatawrite fails due to any reason except -EIO, we will
 735 * continue the remainder of the fsync, although we'll still report
 736 * the error at the end. This is to match filemap_write_and_wait_range()
 737 * behaviour.
 738 *
 739 * Returns: errno
 740 */
 741
 742static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 743		      int datasync)
 744{
 745	struct address_space *mapping = file->f_mapping;
 746	struct inode *inode = mapping->host;
 747	int sync_state = inode->i_state & I_DIRTY;
 748	struct gfs2_inode *ip = GFS2_I(inode);
 749	int ret = 0, ret1 = 0;
 750
 751	if (mapping->nrpages) {
 752		ret1 = filemap_fdatawrite_range(mapping, start, end);
 753		if (ret1 == -EIO)
 754			return ret1;
 755	}
 756
 757	if (!gfs2_is_jdata(ip))
 758		sync_state &= ~I_DIRTY_PAGES;
 759	if (datasync)
 760		sync_state &= ~I_DIRTY_SYNC;
 761
 762	if (sync_state) {
 763		ret = sync_inode_metadata(inode, 1);
 764		if (ret)
 765			return ret;
 766		if (gfs2_is_jdata(ip))
 767			ret = file_write_and_wait(file);
 768		if (ret)
 769			return ret;
 770		gfs2_ail_flush(ip->i_gl, 1);
 771	}
 772
 773	if (mapping->nrpages)
 774		ret = file_fdatawait_range(file, start, end);
 775
 776	return ret ? ret : ret1;
 777}
 778
 779static inline bool should_fault_in_pages(struct iov_iter *i,
 780					 struct kiocb *iocb,
 781					 size_t *prev_count,
 782					 size_t *window_size)
 783{
 784	size_t count = iov_iter_count(i);
 785	size_t size, offs;
 786
 787	if (!count)
 788		return false;
 789	if (!user_backed_iter(i))
 790		return false;
 791
 792	/*
 793	 * Try to fault in multiple pages initially.  When that doesn't result
 794	 * in any progress, fall back to a single page.
 795	 */
 796	size = PAGE_SIZE;
 797	offs = offset_in_page(iocb->ki_pos);
 798	if (*prev_count != count) {
 799		size_t nr_dirtied;
 800
 801		nr_dirtied = max(current->nr_dirtied_pause -
 802				 current->nr_dirtied, 8);
 803		size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT);
 804	}
 805
 806	*prev_count = count;
 807	*window_size = size - offs;
 808	return true;
 809}
 810
 811static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
 812				     struct gfs2_holder *gh)
 813{
 814	struct file *file = iocb->ki_filp;
 815	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 816	size_t prev_count = 0, window_size = 0;
 817	size_t read = 0;
 818	ssize_t ret;
 819
 820	/*
 821	 * In this function, we disable page faults when we're holding the
 822	 * inode glock while doing I/O.  If a page fault occurs, we indicate
 823	 * that the inode glock may be dropped, fault in the pages manually,
 824	 * and retry.
 825	 *
 826	 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
 827	 * physical as well as manual page faults, and we need to disable both
 828	 * kinds.
 829	 *
 830	 * For direct I/O, gfs2 takes the inode glock in deferred mode.  This
 831	 * locking mode is compatible with other deferred holders, so multiple
 832	 * processes and nodes can do direct I/O to a file at the same time.
 833	 * There's no guarantee that reads or writes will be atomic.  Any
 834	 * coordination among readers and writers needs to happen externally.
 835	 */
 836
 837	if (!iov_iter_count(to))
 838		return 0; /* skip atime */
 839
 840	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
 841retry:
 842	ret = gfs2_glock_nq(gh);
 843	if (ret)
 844		goto out_uninit;
 845	pagefault_disable();
 846	to->nofault = true;
 847	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
 848			   IOMAP_DIO_PARTIAL, NULL, read);
 849	to->nofault = false;
 850	pagefault_enable();
 851	if (ret <= 0 && ret != -EFAULT)
 852		goto out_unlock;
 853	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
 854	if (ret > 0)
 855		read = ret;
 856
 857	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
 858		gfs2_glock_dq(gh);
 859		window_size -= fault_in_iov_iter_writeable(to, window_size);
 860		if (window_size)
 861			goto retry;
 862	}
 863out_unlock:
 864	if (gfs2_holder_queued(gh))
 865		gfs2_glock_dq(gh);
 866out_uninit:
 867	gfs2_holder_uninit(gh);
 868	/* User space doesn't expect partial success. */
 869	if (ret < 0)
 870		return ret;
 871	return read;
 872}
 873
 874static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
 875				      struct gfs2_holder *gh)
 876{
 877	struct file *file = iocb->ki_filp;
 878	struct inode *inode = file->f_mapping->host;
 879	struct gfs2_inode *ip = GFS2_I(inode);
 880	size_t prev_count = 0, window_size = 0;
 881	size_t written = 0;
 882	bool enough_retries;
 883	ssize_t ret;
 884
 885	/*
 886	 * In this function, we disable page faults when we're holding the
 887	 * inode glock while doing I/O.  If a page fault occurs, we indicate
 888	 * that the inode glock may be dropped, fault in the pages manually,
 889	 * and retry.
 890	 *
 891	 * For writes, iomap_dio_rw only triggers manual page faults, so we
 892	 * don't need to disable physical ones.
 893	 */
 894
 895	/*
 896	 * Deferred lock, even if its a write, since we do no allocation on
 897	 * this path. All we need to change is the atime, and this lock mode
 898	 * ensures that other nodes have flushed their buffered read caches
 899	 * (i.e. their page cache entries for this inode). We do not,
 900	 * unfortunately, have the option of only flushing a range like the
 901	 * VFS does.
 902	 */
 903	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
 904retry:
 905	ret = gfs2_glock_nq(gh);
 906	if (ret)
 907		goto out_uninit;
 908	/* Silently fall back to buffered I/O when writing beyond EOF */
 909	if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
 910		goto out_unlock;
 911
 912	from->nofault = true;
 913	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
 914			   IOMAP_DIO_PARTIAL, NULL, written);
 915	from->nofault = false;
 916	if (ret <= 0) {
 917		if (ret == -ENOTBLK)
 918			ret = 0;
 919		if (ret != -EFAULT)
 920			goto out_unlock;
 921	}
 922	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
 923	if (ret > 0)
 924		written = ret;
 925
 926	enough_retries = prev_count == iov_iter_count(from) &&
 927			 window_size <= PAGE_SIZE;
 928	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
 929		gfs2_glock_dq(gh);
 930		window_size -= fault_in_iov_iter_readable(from, window_size);
 931		if (window_size) {
 932			if (!enough_retries)
 933				goto retry;
 934			/* fall back to buffered I/O */
 935			ret = 0;
 936		}
 937	}
 938out_unlock:
 939	if (gfs2_holder_queued(gh))
 940		gfs2_glock_dq(gh);
 941out_uninit:
 942	gfs2_holder_uninit(gh);
 943	/* User space doesn't expect partial success. */
 944	if (ret < 0)
 945		return ret;
 946	return written;
 947}
 948
 949static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 950{
 951	struct gfs2_inode *ip;
 952	struct gfs2_holder gh;
 953	size_t prev_count = 0, window_size = 0;
 954	size_t read = 0;
 955	ssize_t ret;
 956
 957	/*
 958	 * In this function, we disable page faults when we're holding the
 959	 * inode glock while doing I/O.  If a page fault occurs, we indicate
 960	 * that the inode glock may be dropped, fault in the pages manually,
 961	 * and retry.
 962	 */
 963
 964	if (iocb->ki_flags & IOCB_DIRECT)
 965		return gfs2_file_direct_read(iocb, to, &gh);
 966
 967	pagefault_disable();
 968	iocb->ki_flags |= IOCB_NOIO;
 969	ret = generic_file_read_iter(iocb, to);
 970	iocb->ki_flags &= ~IOCB_NOIO;
 971	pagefault_enable();
 972	if (ret >= 0) {
 973		if (!iov_iter_count(to))
 974			return ret;
 975		read = ret;
 976	} else if (ret != -EFAULT) {
 977		if (ret != -EAGAIN)
 978			return ret;
 979		if (iocb->ki_flags & IOCB_NOWAIT)
 980			return ret;
 981	}
 982	ip = GFS2_I(iocb->ki_filp->f_mapping->host);
 983	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 984retry:
 985	ret = gfs2_glock_nq(&gh);
 986	if (ret)
 987		goto out_uninit;
 988	pagefault_disable();
 989	ret = generic_file_read_iter(iocb, to);
 990	pagefault_enable();
 991	if (ret <= 0 && ret != -EFAULT)
 992		goto out_unlock;
 993	if (ret > 0)
 994		read += ret;
 995
 996	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
 997		gfs2_glock_dq(&gh);
 998		window_size -= fault_in_iov_iter_writeable(to, window_size);
 999		if (window_size)
1000			goto retry;
1001	}
1002out_unlock:
1003	if (gfs2_holder_queued(&gh))
1004		gfs2_glock_dq(&gh);
1005out_uninit:
1006	gfs2_holder_uninit(&gh);
1007	return read ? read : ret;
1008}
1009
1010static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
1011					struct iov_iter *from,
1012					struct gfs2_holder *gh)
1013{
1014	struct file *file = iocb->ki_filp;
1015	struct inode *inode = file_inode(file);
1016	struct gfs2_inode *ip = GFS2_I(inode);
1017	struct gfs2_sbd *sdp = GFS2_SB(inode);
1018	struct gfs2_holder *statfs_gh = NULL;
1019	size_t prev_count = 0, window_size = 0;
1020	size_t orig_count = iov_iter_count(from);
1021	size_t written = 0;
1022	ssize_t ret;
1023
1024	/*
1025	 * In this function, we disable page faults when we're holding the
1026	 * inode glock while doing I/O.  If a page fault occurs, we indicate
1027	 * that the inode glock may be dropped, fault in the pages manually,
1028	 * and retry.
1029	 */
1030
1031	if (inode == sdp->sd_rindex) {
1032		statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1033		if (!statfs_gh)
1034			return -ENOMEM;
1035	}
1036
1037	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
1038	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1039retry:
1040		window_size -= fault_in_iov_iter_readable(from, window_size);
1041		if (!window_size) {
1042			ret = -EFAULT;
1043			goto out_uninit;
1044		}
1045		from->count = min(from->count, window_size);
1046	}
1047	ret = gfs2_glock_nq(gh);
1048	if (ret)
1049		goto out_uninit;
1050
1051	if (inode == sdp->sd_rindex) {
1052		struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1053
1054		ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1055					 GL_NOCACHE, statfs_gh);
1056		if (ret)
1057			goto out_unlock;
1058	}
1059
1060	pagefault_disable();
1061	ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops, NULL);
1062	pagefault_enable();
1063	if (ret > 0)
1064		written += ret;
1065
1066	if (inode == sdp->sd_rindex)
1067		gfs2_glock_dq_uninit(statfs_gh);
1068
1069	if (ret <= 0 && ret != -EFAULT)
1070		goto out_unlock;
1071
1072	from->count = orig_count - written;
1073	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1074		gfs2_glock_dq(gh);
1075		goto retry;
1076	}
1077out_unlock:
1078	if (gfs2_holder_queued(gh))
1079		gfs2_glock_dq(gh);
1080out_uninit:
1081	gfs2_holder_uninit(gh);
1082	kfree(statfs_gh);
1083	from->count = orig_count - written;
1084	return written ? written : ret;
1085}
1086
1087/**
1088 * gfs2_file_write_iter - Perform a write to a file
1089 * @iocb: The io context
1090 * @from: The data to write
 
 
1091 *
1092 * We have to do a lock/unlock here to refresh the inode size for
1093 * O_APPEND writes, otherwise we can land up writing at the wrong
1094 * offset. There is still a race, but provided the app is using its
1095 * own file locking, this will make O_APPEND work as expected.
1096 *
1097 */
1098
1099static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 
1100{
1101	struct file *file = iocb->ki_filp;
1102	struct inode *inode = file_inode(file);
1103	struct gfs2_inode *ip = GFS2_I(inode);
1104	struct gfs2_holder gh;
1105	ssize_t ret;
1106
1107	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
 
 
 
 
1108
1109	if (iocb->ki_flags & IOCB_APPEND) {
1110		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1111		if (ret)
1112			return ret;
1113		gfs2_glock_dq_uninit(&gh);
1114	}
1115
1116	inode_lock(inode);
1117	ret = generic_write_checks(iocb, from);
1118	if (ret <= 0)
1119		goto out_unlock;
 
 
 
 
 
 
 
 
 
 
1120
1121	ret = file_remove_privs(file);
1122	if (ret)
1123		goto out_unlock;
1124
1125	if (iocb->ki_flags & IOCB_DIRECT) {
1126		struct address_space *mapping = file->f_mapping;
1127		ssize_t buffered, ret2;
1128
1129		/*
1130		 * Note that under direct I/O, we don't allow and inode
1131		 * timestamp updates, so we're not calling file_update_time()
1132		 * here.
1133		 */
1134
1135		ret = gfs2_file_direct_write(iocb, from, &gh);
1136		if (ret < 0 || !iov_iter_count(from))
1137			goto out_unlock;
 
 
 
 
 
 
 
 
 
1138
1139		iocb->ki_flags |= IOCB_DSYNC;
1140		buffered = gfs2_file_buffered_write(iocb, from, &gh);
1141		if (unlikely(buffered <= 0)) {
1142			if (!ret)
1143				ret = buffered;
1144			goto out_unlock;
 
1145		}
 
 
 
 
 
 
 
 
 
 
1146
1147		/*
1148		 * We need to ensure that the page cache pages are written to
1149		 * disk and invalidated to preserve the expected O_DIRECT
1150		 * semantics.  If the writeback or invalidate fails, only report
1151		 * the direct I/O range as we don't know if the buffered pages
1152		 * made it to disk.
1153		 */
1154		ret2 = generic_write_sync(iocb, buffered);
1155		invalidate_mapping_pages(mapping,
1156				(iocb->ki_pos - buffered) >> PAGE_SHIFT,
1157				(iocb->ki_pos - 1) >> PAGE_SHIFT);
1158		if (!ret || ret2 > 0)
1159			ret += ret2;
1160	} else {
1161		ret = file_update_time(file);
1162		if (ret)
1163			goto out_unlock;
1164
1165		ret = gfs2_file_buffered_write(iocb, from, &gh);
1166		if (likely(ret > 0))
1167			ret = generic_write_sync(iocb, ret);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1168	}
1169
1170out_unlock:
1171	inode_unlock(inode);
1172	return ret;
1173}
1174
1175static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
1176			   int mode)
1177{
1178	struct super_block *sb = inode->i_sb;
1179	struct gfs2_inode *ip = GFS2_I(inode);
1180	loff_t end = offset + len;
1181	struct buffer_head *dibh;
1182	int error;
 
 
 
 
 
 
 
 
 
 
1183
1184	error = gfs2_meta_inode_buffer(ip, &dibh);
1185	if (unlikely(error))
1186		return error;
1187
1188	gfs2_trans_add_meta(ip->i_gl, dibh);
1189
1190	if (gfs2_is_stuffed(ip)) {
1191		error = gfs2_unstuff_dinode(ip);
1192		if (unlikely(error))
1193			goto out;
1194	}
1195
1196	while (offset < end) {
1197		struct iomap iomap = { };
 
 
 
 
 
 
 
 
 
1198
1199		error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
 
 
 
 
 
 
 
 
1200		if (error)
1201			goto out;
1202		offset = iomap.offset + iomap.length;
1203		if (!(iomap.flags & IOMAP_F_NEW))
1204			continue;
1205		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1206					 iomap.length >> inode->i_blkbits,
1207					 GFP_NOFS);
1208		if (error) {
1209			fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1210			goto out;
1211		}
1212	}
 
 
 
 
 
 
1213out:
1214	brelse(dibh);
1215	return error;
1216}
1217
1218/**
1219 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1220 *                     blocks, determine how many bytes can be written.
1221 * @ip:          The inode in question.
1222 * @len:         Max cap of bytes. What we return in *len must be <= this.
1223 * @data_blocks: Compute and return the number of data blocks needed
1224 * @ind_blocks:  Compute and return the number of indirect blocks needed
1225 * @max_blocks:  The total blocks available to work with.
1226 *
1227 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1228 */
1229static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1230			    unsigned int *data_blocks, unsigned int *ind_blocks,
1231			    unsigned int max_blocks)
1232{
1233	loff_t max = *len;
1234	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 
1235	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1236
1237	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1238		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1239		max_data -= tmp;
1240	}
1241
 
 
 
1242	*data_blocks = max_data;
1243	*ind_blocks = max_blocks - max_data;
1244	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1245	if (*len > max) {
1246		*len = max;
1247		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1248	}
1249}
1250
1251static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 
1252{
1253	struct inode *inode = file_inode(file);
1254	struct gfs2_sbd *sdp = GFS2_SB(inode);
1255	struct gfs2_inode *ip = GFS2_I(inode);
1256	struct gfs2_alloc_parms ap = {};
1257	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1258	loff_t bytes, max_bytes, max_blks;
 
1259	int error;
1260	const loff_t pos = offset;
1261	const loff_t count = len;
1262	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1263	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1264	loff_t max_chunk_size = UINT_MAX & bsize_mask;
1265
1266	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
 
 
1267
1268	offset &= bsize_mask;
1269
1270	len = next - offset;
1271	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1272	if (!bytes)
1273		bytes = UINT_MAX;
1274	bytes &= bsize_mask;
1275	if (bytes == 0)
1276		bytes = sdp->sd_sb.sb_bsize;
1277
1278	gfs2_size_hint(file, offset, len);
 
 
 
1279
1280	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1281	ap.min_target = data_blocks + ind_blocks;
1282
1283	while (len > 0) {
1284		if (len < bytes)
1285			bytes = len;
1286		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1287			len -= bytes;
1288			offset += bytes;
1289			continue;
1290		}
1291
1292		/* We need to determine how many bytes we can actually
1293		 * fallocate without exceeding quota or going over the
1294		 * end of the fs. We start off optimistically by assuming
1295		 * we can write max_bytes */
1296		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1297
1298		/* Since max_bytes is most likely a theoretical max, we
1299		 * calculate a more realistic 'bytes' to serve as a good
1300		 * starting point for the number of bytes we may be able
1301		 * to write */
1302		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1303		ap.target = data_blocks + ind_blocks;
1304
1305		error = gfs2_quota_lock_check(ip, &ap);
1306		if (error)
1307			return error;
1308		/* ap.allowed tells us how many blocks quota will allow
1309		 * us to write. Check if this reduces max_blks */
1310		max_blks = UINT_MAX;
1311		if (ap.allowed)
1312			max_blks = ap.allowed;
1313
1314		error = gfs2_inplace_reserve(ip, &ap);
1315		if (error)
1316			goto out_qunlock;
1317
1318		/* check if the selected rgrp limits our max_blks further */
1319		if (ip->i_res.rs_reserved < max_blks)
1320			max_blks = ip->i_res.rs_reserved;
1321
1322		/* Almost done. Calculate bytes that can be written using
1323		 * max_blks. We also recompute max_bytes, data_blocks and
1324		 * ind_blocks */
1325		calc_max_reserv(ip, &max_bytes, &data_blocks,
1326				&ind_blocks, max_blks);
1327
1328		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1329			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1330		if (gfs2_is_jdata(ip))
1331			rblocks += data_blocks ? data_blocks : 1;
1332
1333		error = gfs2_trans_begin(sdp, rblocks,
1334					 PAGE_SIZE >> inode->i_blkbits);
1335		if (error)
1336			goto out_trans_fail;
1337
1338		error = fallocate_chunk(inode, offset, max_bytes, mode);
1339		gfs2_trans_end(sdp);
1340
1341		if (error)
1342			goto out_trans_fail;
1343
1344		len -= max_bytes;
1345		offset += max_bytes;
1346		gfs2_inplace_release(ip);
1347		gfs2_quota_unlock(ip);
 
1348	}
1349
1350	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1351		i_size_write(inode, pos + count);
1352	file_update_time(file);
1353	mark_inode_dirty(inode);
1354
1355	if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1356		return vfs_fsync_range(file, pos, pos + count - 1,
1357			       (file->f_flags & __O_SYNC) ? 0 : 1);
1358	return 0;
1359
1360out_trans_fail:
1361	gfs2_inplace_release(ip);
1362out_qunlock:
1363	gfs2_quota_unlock(ip);
 
 
 
 
 
 
1364	return error;
1365}
1366
1367static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1368{
1369	struct inode *inode = file_inode(file);
1370	struct gfs2_sbd *sdp = GFS2_SB(inode);
1371	struct gfs2_inode *ip = GFS2_I(inode);
1372	struct gfs2_holder gh;
1373	int ret;
1374
1375	if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1376		return -EOPNOTSUPP;
1377	/* fallocate is needed by gfs2_grow to reserve space in the rindex */
1378	if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1379		return -EOPNOTSUPP;
1380
1381	inode_lock(inode);
1382
1383	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1384	ret = gfs2_glock_nq(&gh);
1385	if (ret)
1386		goto out_uninit;
1387
1388	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1389	    (offset + len) > inode->i_size) {
1390		ret = inode_newsize_ok(inode, offset + len);
1391		if (ret)
1392			goto out_unlock;
1393	}
1394
1395	ret = get_write_access(inode);
1396	if (ret)
1397		goto out_unlock;
1398
1399	if (mode & FALLOC_FL_PUNCH_HOLE) {
1400		ret = __gfs2_punch_hole(file, offset, len);
1401	} else {
1402		ret = __gfs2_fallocate(file, mode, offset, len);
1403		if (ret)
1404			gfs2_rs_deltree(&ip->i_res);
1405	}
1406
1407	put_write_access(inode);
1408out_unlock:
1409	gfs2_glock_dq(&gh);
1410out_uninit:
1411	gfs2_holder_uninit(&gh);
1412	inode_unlock(inode);
1413	return ret;
1414}
1415
1416static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1417				      struct file *out, loff_t *ppos,
1418				      size_t len, unsigned int flags)
1419{
1420	ssize_t ret;
1421
1422	gfs2_size_hint(out, *ppos, len);
1423
1424	ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1425	return ret;
1426}
1427
1428#ifdef CONFIG_GFS2_FS_LOCKING_DLM
1429
1430/**
1431 * gfs2_lock - acquire/release a posix lock on a file
1432 * @file: the file pointer
1433 * @cmd: either modify or retrieve lock state, possibly wait
1434 * @fl: type and range of lock
1435 *
1436 * Returns: errno
1437 */
1438
1439static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1440{
1441	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1442	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1443	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1444
1445	if (!(fl->c.flc_flags & FL_POSIX))
 
 
1446		return -ENOLCK;
1447	if (gfs2_withdrawing_or_withdrawn(sdp)) {
1448		if (lock_is_unlock(fl))
1449			locks_lock_file_wait(file, fl);
 
 
 
 
1450		return -EIO;
1451	}
1452	if (cmd == F_CANCELLK)
1453		return dlm_posix_cancel(ls->ls_dlm, ip->i_no_addr, file, fl);
1454	else if (IS_GETLK(cmd))
1455		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1456	else if (lock_is_unlock(fl))
1457		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1458	else
1459		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1460}
1461
1462static void __flock_holder_uninit(struct file *file, struct gfs2_holder *fl_gh)
1463{
1464	struct gfs2_glock *gl = gfs2_glock_hold(fl_gh->gh_gl);
1465
1466	/*
1467	 * Make sure gfs2_glock_put() won't sleep under the file->f_lock
1468	 * spinlock.
1469	 */
1470
1471	spin_lock(&file->f_lock);
1472	gfs2_holder_uninit(fl_gh);
1473	spin_unlock(&file->f_lock);
1474	gfs2_glock_put(gl);
1475}
1476
1477static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1478{
1479	struct gfs2_file *fp = file->private_data;
1480	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1481	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1482	struct gfs2_glock *gl;
1483	unsigned int state;
1484	u16 flags;
1485	int error = 0;
1486	int sleeptime;
1487
1488	state = lock_is_write(fl) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1489	flags = GL_EXACT | GL_NOPID;
1490	if (!IS_SETLKW(cmd))
1491		flags |= LM_FLAG_TRY_1CB;
1492
1493	mutex_lock(&fp->f_fl_mutex);
1494
1495	if (gfs2_holder_initialized(fl_gh)) {
1496		struct file_lock request;
1497		if (fl_gh->gh_state == state)
1498			goto out;
1499		locks_init_lock(&request);
1500		request.c.flc_type = F_UNLCK;
1501		request.c.flc_flags = FL_FLOCK;
1502		locks_lock_file_wait(file, &request);
1503		gfs2_glock_dq(fl_gh);
1504		gfs2_holder_reinit(state, flags, fl_gh);
1505	} else {
1506		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1507				       &gfs2_flock_glops, CREATE, &gl);
1508		if (error)
1509			goto out;
1510		spin_lock(&file->f_lock);
1511		gfs2_holder_init(gl, state, flags, fl_gh);
1512		spin_unlock(&file->f_lock);
1513		gfs2_glock_put(gl);
1514	}
1515	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1516		error = gfs2_glock_nq(fl_gh);
1517		if (error != GLR_TRYFAILED)
1518			break;
1519		fl_gh->gh_flags &= ~LM_FLAG_TRY_1CB;
1520		fl_gh->gh_flags |= LM_FLAG_TRY;
1521		msleep(sleeptime);
1522	}
1523	if (error) {
1524		__flock_holder_uninit(file, fl_gh);
1525		if (error == GLR_TRYFAILED)
1526			error = -EAGAIN;
1527	} else {
1528		error = locks_lock_file_wait(file, fl);
1529		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1530	}
1531
1532out:
1533	mutex_unlock(&fp->f_fl_mutex);
1534	return error;
1535}
1536
1537static void do_unflock(struct file *file, struct file_lock *fl)
1538{
1539	struct gfs2_file *fp = file->private_data;
1540	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1541
1542	mutex_lock(&fp->f_fl_mutex);
1543	locks_lock_file_wait(file, fl);
1544	if (gfs2_holder_initialized(fl_gh)) {
1545		gfs2_glock_dq(fl_gh);
1546		__flock_holder_uninit(file, fl_gh);
1547	}
1548	mutex_unlock(&fp->f_fl_mutex);
1549}
1550
1551/**
1552 * gfs2_flock - acquire/release a flock lock on a file
1553 * @file: the file pointer
1554 * @cmd: either modify or retrieve lock state, possibly wait
1555 * @fl: type and range of lock
1556 *
1557 * Returns: errno
1558 */
1559
1560static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1561{
1562	if (!(fl->c.flc_flags & FL_FLOCK))
1563		return -ENOLCK;
 
 
1564
1565	if (lock_is_unlock(fl)) {
1566		do_unflock(file, fl);
1567		return 0;
1568	} else {
1569		return do_flock(file, cmd, fl);
1570	}
1571}
1572
1573const struct file_operations gfs2_file_fops = {
1574	.llseek		= gfs2_llseek,
1575	.read_iter	= gfs2_file_read_iter,
1576	.write_iter	= gfs2_file_write_iter,
1577	.iopoll		= iocb_bio_iopoll,
 
1578	.unlocked_ioctl	= gfs2_ioctl,
1579	.compat_ioctl	= gfs2_compat_ioctl,
1580	.mmap		= gfs2_mmap,
1581	.open		= gfs2_open,
1582	.release	= gfs2_release,
1583	.fsync		= gfs2_fsync,
1584	.lock		= gfs2_lock,
1585	.flock		= gfs2_flock,
1586	.splice_read	= copy_splice_read,
1587	.splice_write	= gfs2_file_splice_write,
1588	.setlease	= simple_nosetlease,
1589	.fallocate	= gfs2_fallocate,
1590	.fop_flags	= FOP_ASYNC_LOCK,
1591};
1592
1593const struct file_operations gfs2_dir_fops = {
1594	.iterate_shared	= gfs2_readdir,
1595	.unlocked_ioctl	= gfs2_ioctl,
1596	.compat_ioctl	= gfs2_compat_ioctl,
1597	.open		= gfs2_open,
1598	.release	= gfs2_release,
1599	.fsync		= gfs2_fsync,
1600	.lock		= gfs2_lock,
1601	.flock		= gfs2_flock,
1602	.llseek		= default_llseek,
1603	.fop_flags	= FOP_ASYNC_LOCK,
1604};
1605
1606#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1607
1608const struct file_operations gfs2_file_fops_nolock = {
1609	.llseek		= gfs2_llseek,
1610	.read_iter	= gfs2_file_read_iter,
1611	.write_iter	= gfs2_file_write_iter,
1612	.iopoll		= iocb_bio_iopoll,
 
1613	.unlocked_ioctl	= gfs2_ioctl,
1614	.compat_ioctl	= gfs2_compat_ioctl,
1615	.mmap		= gfs2_mmap,
1616	.open		= gfs2_open,
1617	.release	= gfs2_release,
1618	.fsync		= gfs2_fsync,
1619	.splice_read	= copy_splice_read,
1620	.splice_write	= gfs2_file_splice_write,
1621	.setlease	= generic_setlease,
1622	.fallocate	= gfs2_fallocate,
1623};
1624
1625const struct file_operations gfs2_dir_fops_nolock = {
1626	.iterate_shared	= gfs2_readdir,
1627	.unlocked_ioctl	= gfs2_ioctl,
1628	.compat_ioctl	= gfs2_compat_ioctl,
1629	.open		= gfs2_open,
1630	.release	= gfs2_release,
1631	.fsync		= gfs2_fsync,
1632	.llseek		= default_llseek,
1633};
1634