inode.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/nfs/inode.c
   4 *
   5 *  Copyright (C) 1992  Rick Sladkey
   6 *
   7 *  nfs inode and superblock handling functions
   8 *
   9 *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
  10 *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
  11 *
  12 *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
  13 *  J.S.Peatfield@damtp.cam.ac.uk
  14 *
  15 */
  16
  17#include <linux/module.h>
  18#include <linux/init.h>
  19#include <linux/sched/signal.h>
  20#include <linux/time.h>
  21#include <linux/kernel.h>
  22#include <linux/mm.h>
  23#include <linux/string.h>
  24#include <linux/stat.h>
  25#include <linux/errno.h>
  26#include <linux/unistd.h>
  27#include <linux/sunrpc/clnt.h>
  28#include <linux/sunrpc/stats.h>
  29#include <linux/sunrpc/metrics.h>
  30#include <linux/nfs_fs.h>
  31#include <linux/nfs_mount.h>
  32#include <linux/nfs4_mount.h>
  33#include <linux/lockd/bind.h>
  34#include <linux/seq_file.h>
  35#include <linux/mount.h>
  36#include <linux/vfs.h>
  37#include <linux/inet.h>
  38#include <linux/nfs_xdr.h>
  39#include <linux/slab.h>
  40#include <linux/compat.h>
  41#include <linux/freezer.h>
  42#include <linux/uaccess.h>
  43#include <linux/iversion.h>
  44
  45#include "nfs4_fs.h"
  46#include "callback.h"
  47#include "delegation.h"
  48#include "iostat.h"
  49#include "internal.h"
  50#include "fscache.h"
  51#include "pnfs.h"
  52#include "nfs.h"
  53#include "netns.h"
  54#include "sysfs.h"
  55
  56#include "nfstrace.h"
  57
  58#define NFSDBG_FACILITY		NFSDBG_VFS
  59
  60#define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
  61
  62/* Default is to see 64-bit inode numbers */
  63static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
  64
  65static int nfs_update_inode(struct inode *, struct nfs_fattr *);
  66
  67static struct kmem_cache * nfs_inode_cachep;
  68
  69static inline unsigned long
  70nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
  71{
  72	return nfs_fileid_to_ino_t(fattr->fileid);
  73}
  74
  75int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
  76{
  77	if (unlikely(nfs_current_task_exiting()))
  78		return -EINTR;
  79	schedule();
  80	if (signal_pending_state(mode, current))
  81		return -ERESTARTSYS;
  82	return 0;
  83}
  84EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
  85
  86/**
  87 * nfs_compat_user_ino64 - returns the user-visible inode number
  88 * @fileid: 64-bit fileid
  89 *
  90 * This function returns a 32-bit inode number if the boot parameter
  91 * nfs.enable_ino64 is zero.
  92 */
  93u64 nfs_compat_user_ino64(u64 fileid)
  94{
  95#ifdef CONFIG_COMPAT
  96	compat_ulong_t ino;
  97#else	
  98	unsigned long ino;
  99#endif
 100
 101	if (enable_ino64)
 102		return fileid;
 103	ino = fileid;
 104	if (sizeof(ino) < sizeof(fileid))
 105		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
 106	return ino;
 107}
 108
 109int nfs_drop_inode(struct inode *inode)
 110{
 111	return NFS_STALE(inode) || inode_generic_drop(inode);
 112}
 113EXPORT_SYMBOL_GPL(nfs_drop_inode);
 114
 115void nfs_clear_inode(struct inode *inode)
 116{
 117	/*
 118	 * The following should never happen...
 119	 */
 120	WARN_ON_ONCE(nfs_have_writebacks(inode));
 121	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
 122	nfs_zap_acl_cache(inode);
 123	nfs_access_zap_cache(inode);
 124	nfs_fscache_clear_inode(inode);
 125}
 126EXPORT_SYMBOL_GPL(nfs_clear_inode);
 127
 128void nfs_evict_inode(struct inode *inode)
 129{
 130	truncate_inode_pages_final(&inode->i_data);
 131	clear_inode(inode);
 132	nfs_clear_inode(inode);
 133}
 134
 135int nfs_sync_inode(struct inode *inode)
 136{
 137	inode_dio_wait(inode);
 138	return nfs_wb_all(inode);
 139}
 140EXPORT_SYMBOL_GPL(nfs_sync_inode);
 141
 142/**
 143 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
 144 * @mapping: pointer to struct address_space
 145 */
 146int nfs_sync_mapping(struct address_space *mapping)
 147{
 148	int ret = 0;
 149
 150	if (mapping->nrpages != 0) {
 151		unmap_mapping_range(mapping, 0, 0, 0);
 152		ret = nfs_wb_all(mapping->host);
 153	}
 154	return ret;
 155}
 156
 157static int nfs_attribute_timeout(struct inode *inode)
 158{
 159	struct nfs_inode *nfsi = NFS_I(inode);
 160
 161	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
 162}
 163
 164static bool nfs_check_cache_flags_invalid(struct inode *inode,
 165					  unsigned long flags)
 166{
 167	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
 168
 169	return (cache_validity & flags) != 0;
 170}
 171
 172bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
 173{
 174	if (nfs_check_cache_flags_invalid(inode, flags))
 175		return true;
 176	return nfs_attribute_cache_expired(inode);
 177}
 178EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
 179
 180#ifdef CONFIG_NFS_V4_2
 181static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
 182{
 183	return nfsi->xattr_cache != NULL;
 184}
 185#else
 186static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
 187{
 188	return false;
 189}
 190#endif
 191
 192void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
 193{
 194	struct nfs_inode *nfsi = NFS_I(inode);
 195
 196	if (nfs_have_delegated_attributes(inode)) {
 197		if (!(flags & NFS_INO_REVAL_FORCED))
 198			flags &= ~(NFS_INO_INVALID_MODE |
 199				   NFS_INO_INVALID_OTHER |
 200				   NFS_INO_INVALID_BTIME |
 201				   NFS_INO_INVALID_XATTR);
 202		flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
 203	}
 204
 205	if (!nfs_has_xattr_cache(nfsi))
 206		flags &= ~NFS_INO_INVALID_XATTR;
 207	if (flags & NFS_INO_INVALID_DATA)
 208		nfs_fscache_invalidate(inode, 0);
 209	flags &= ~NFS_INO_REVAL_FORCED;
 210
 211	flags |= nfsi->cache_validity;
 212	if (inode->i_mapping->nrpages == 0)
 213		flags &= ~NFS_INO_INVALID_DATA;
 214
 215	/* pairs with nfs_clear_invalid_mapping()'s smp_load_acquire() */
 216	smp_store_release(&nfsi->cache_validity, flags);
 217
 218	if (inode->i_mapping->nrpages == 0 ||
 219	    nfsi->cache_validity & NFS_INO_INVALID_DATA) {
 220		nfs_ooo_clear(nfsi);
 221	}
 222	trace_nfs_set_cache_invalid(inode, 0);
 223}
 224EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
 225
 226/*
 227 * Invalidate the local caches
 228 */
 229static void nfs_zap_caches_locked(struct inode *inode)
 230{
 231	struct nfs_inode *nfsi = NFS_I(inode);
 232	int mode = inode->i_mode;
 233
 234	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
 235
 236	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
 237	nfsi->attrtimeo_timestamp = jiffies;
 238
 239	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
 240		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
 241						     NFS_INO_INVALID_DATA |
 242						     NFS_INO_INVALID_ACCESS |
 243						     NFS_INO_INVALID_ACL |
 244						     NFS_INO_INVALID_XATTR);
 245	else
 246		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
 247						     NFS_INO_INVALID_ACCESS |
 248						     NFS_INO_INVALID_ACL |
 249						     NFS_INO_INVALID_XATTR);
 250	nfs_zap_label_cache_locked(nfsi);
 251}
 252
 253void nfs_zap_caches(struct inode *inode)
 254{
 255	spin_lock(&inode->i_lock);
 256	nfs_zap_caches_locked(inode);
 257	spin_unlock(&inode->i_lock);
 258}
 259
 260void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
 261{
 262	if (mapping->nrpages != 0) {
 263		spin_lock(&inode->i_lock);
 264		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
 265		spin_unlock(&inode->i_lock);
 266	}
 267}
 268
 269void nfs_zap_acl_cache(struct inode *inode)
 270{
 271	void (*clear_acl_cache)(struct inode *);
 272
 273	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
 274	if (clear_acl_cache != NULL)
 275		clear_acl_cache(inode);
 276	spin_lock(&inode->i_lock);
 277	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
 278	spin_unlock(&inode->i_lock);
 279}
 280EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
 281
 282void nfs_invalidate_atime(struct inode *inode)
 283{
 284	if (nfs_have_delegated_atime(inode))
 285		return;
 286	spin_lock(&inode->i_lock);
 287	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
 288	spin_unlock(&inode->i_lock);
 289}
 290EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
 291
 292/*
 293 * Invalidate, but do not unhash, the inode.
 294 * NB: must be called with inode->i_lock held!
 295 */
 296static void nfs_set_inode_stale_locked(struct inode *inode)
 297{
 298	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 299	nfs_zap_caches_locked(inode);
 300	trace_nfs_set_inode_stale(inode);
 301}
 302
 303void nfs_set_inode_stale(struct inode *inode)
 304{
 305	spin_lock(&inode->i_lock);
 306	nfs_set_inode_stale_locked(inode);
 307	spin_unlock(&inode->i_lock);
 308}
 309
 310struct nfs_find_desc {
 311	struct nfs_fh		*fh;
 312	struct nfs_fattr	*fattr;
 313};
 314
 315/*
 316 * In NFSv3 we can have 64bit inode numbers. In order to support
 317 * this, and re-exported directories (also seen in NFSv2)
 318 * we are forced to allow 2 different inodes to have the same
 319 * i_ino.
 320 */
 321static int
 322nfs_find_actor(struct inode *inode, void *opaque)
 323{
 324	struct nfs_find_desc	*desc = opaque;
 325	struct nfs_fh		*fh = desc->fh;
 326	struct nfs_fattr	*fattr = desc->fattr;
 327
 328	if (NFS_FILEID(inode) != fattr->fileid)
 329		return 0;
 330	if (inode_wrong_type(inode, fattr->mode))
 331		return 0;
 332	if (nfs_compare_fh(NFS_FH(inode), fh))
 333		return 0;
 334	if (is_bad_inode(inode) || NFS_STALE(inode))
 335		return 0;
 336	return 1;
 337}
 338
 339static int
 340nfs_init_locked(struct inode *inode, void *opaque)
 341{
 342	struct nfs_find_desc	*desc = opaque;
 343	struct nfs_fattr	*fattr = desc->fattr;
 344
 345	set_nfs_fileid(inode, fattr->fileid);
 346	inode->i_mode = fattr->mode;
 347	nfs_copy_fh(NFS_FH(inode), desc->fh);
 348	return 0;
 349}
 350
 351#ifdef CONFIG_NFS_V4_SECURITY_LABEL
 352static void nfs_clear_label_invalid(struct inode *inode)
 353{
 354	spin_lock(&inode->i_lock);
 355	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
 356	spin_unlock(&inode->i_lock);
 357}
 358
 359void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
 360{
 361	int error;
 362
 363	if (fattr->label == NULL)
 364		return;
 365
 366	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
 367		error = security_inode_notifysecctx(inode, fattr->label->label,
 368				fattr->label->len);
 369		if (error)
 370			printk(KERN_ERR "%s() %s %d "
 371					"security_inode_notifysecctx() %d\n",
 372					__func__,
 373					(char *)fattr->label->label,
 374					fattr->label->len, error);
 375		nfs_clear_label_invalid(inode);
 376	}
 377}
 378
 379struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
 380{
 381	struct nfs4_label *label;
 382
 383	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
 384		return NULL;
 385
 386	label = kzalloc(sizeof(struct nfs4_label), flags);
 387	if (label == NULL)
 388		return ERR_PTR(-ENOMEM);
 389
 390	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
 391	if (label->label == NULL) {
 392		kfree(label);
 393		return ERR_PTR(-ENOMEM);
 394	}
 395	label->len = NFS4_MAXLABELLEN;
 396
 397	return label;
 398}
 399EXPORT_SYMBOL_GPL(nfs4_label_alloc);
 400#else
 401void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
 402{
 403}
 404#endif
 405EXPORT_SYMBOL_GPL(nfs_setsecurity);
 406
 407/* Search for inode identified by fh, fileid and i_mode in inode cache. */
 408struct inode *
 409nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
 410{
 411	struct nfs_find_desc desc = {
 412		.fh	= fh,
 413		.fattr	= fattr,
 414	};
 415	struct inode *inode;
 416	unsigned long hash;
 417
 418	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
 419	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
 420		return NULL;
 421
 422	hash = nfs_fattr_to_ino_t(fattr);
 423	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
 424
 425	dprintk("%s: returning %p\n", __func__, inode);
 426	return inode;
 427}
 428
 429static void nfs_inode_init_regular(struct nfs_inode *nfsi)
 430{
 431	atomic_long_set(&nfsi->nrequests, 0);
 432	atomic_long_set(&nfsi->redirtied_pages, 0);
 433	INIT_LIST_HEAD(&nfsi->commit_info.list);
 434	atomic_long_set(&nfsi->commit_info.ncommit, 0);
 435	atomic_set(&nfsi->commit_info.rpcs_out, 0);
 436	mutex_init(&nfsi->commit_mutex);
 437}
 438
 439static void nfs_inode_init_dir(struct nfs_inode *nfsi)
 440{
 441	nfsi->cache_change_attribute = 0;
 442	memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
 443	init_rwsem(&nfsi->rmdir_sem);
 444}
 445
 446/*
 447 * This is our front-end to iget that looks up inodes by file handle
 448 * instead of inode number.
 449 */
 450struct inode *
 451nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
 452{
 453	struct nfs_find_desc desc = {
 454		.fh	= fh,
 455		.fattr	= fattr
 456	};
 457	struct inode *inode = ERR_PTR(-ENOENT);
 458	u64 fattr_supported = NFS_SB(sb)->fattr_valid;
 459	unsigned long hash;
 460
 461	nfs_attr_check_mountpoint(sb, fattr);
 462
 463	if (nfs_attr_use_mounted_on_fileid(fattr))
 464		fattr->fileid = fattr->mounted_on_fileid;
 465	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
 466		goto out_no_inode;
 467	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
 468		goto out_no_inode;
 469
 470	hash = nfs_fattr_to_ino_t(fattr);
 471
 472	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
 473	if (inode == NULL) {
 474		inode = ERR_PTR(-ENOMEM);
 475		goto out_no_inode;
 476	}
 477
 478	if (inode_state_read_once(inode) & I_NEW) {
 479		struct nfs_inode *nfsi = NFS_I(inode);
 480		unsigned long now = jiffies;
 481
 482		/* We set i_ino for the few things that still rely on it,
 483		 * such as stat(2) */
 484		inode->i_ino = hash;
 485
 486		/* We can't support update_atime(), since the server will reset it */
 487		inode->i_flags |= S_NOATIME|S_NOCMTIME;
 488		inode->i_mode = fattr->mode;
 489		nfsi->cache_validity = 0;
 490		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
 491				&& (fattr_supported & NFS_ATTR_FATTR_MODE))
 492			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
 493		/* Why so? Because we want revalidate for devices/FIFOs, and
 494		 * that's precisely what we have in nfs_file_inode_operations.
 495		 */
 496		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
 497		if (S_ISREG(inode->i_mode)) {
 498			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
 499			inode->i_data.a_ops = &nfs_file_aops;
 500			nfs_inode_init_regular(nfsi);
 501			mapping_set_large_folios(inode->i_mapping);
 502		} else if (S_ISDIR(inode->i_mode)) {
 503			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
 504			inode->i_fop = &nfs_dir_operations;
 505			inode->i_data.a_ops = &nfs_dir_aops;
 506			nfs_inode_init_dir(nfsi);
 507			/* Deal with crossing mountpoints */
 508			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
 509					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
 510				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
 511					inode->i_op = &nfs_referral_inode_operations;
 512				else
 513					inode->i_op = &nfs_mountpoint_inode_operations;
 514				inode->i_fop = NULL;
 515				inode->i_flags |= S_AUTOMOUNT;
 516			}
 517		} else if (S_ISLNK(inode->i_mode)) {
 518			inode->i_op = &nfs_symlink_inode_operations;
 519			inode_nohighmem(inode);
 520		} else
 521			init_special_inode(inode, inode->i_mode, fattr->rdev);
 522
 523		inode_set_atime(inode, 0, 0);
 524		inode_set_mtime(inode, 0, 0);
 525		inode_set_ctime(inode, 0, 0);
 526		memset(&nfsi->btime, 0, sizeof(nfsi->btime));
 527		inode_set_iversion_raw(inode, 0);
 528		inode->i_size = 0;
 529		clear_nlink(inode);
 530		inode->i_uid = make_kuid(&init_user_ns, -2);
 531		inode->i_gid = make_kgid(&init_user_ns, -2);
 532		inode->i_blocks = 0;
 533		nfsi->write_io = 0;
 534		nfsi->read_io = 0;
 535
 536		nfsi->read_cache_jiffies = fattr->time_start;
 537		nfsi->attr_gencount = fattr->gencount;
 538		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
 539			inode_set_atime_to_ts(inode, fattr->atime);
 540		else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
 541			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
 542		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
 543			inode_set_mtime_to_ts(inode, fattr->mtime);
 544		else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
 545			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
 546		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 547			inode_set_ctime_to_ts(inode, fattr->ctime);
 548		else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
 549			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
 550		if (fattr->valid & NFS_ATTR_FATTR_BTIME)
 551			nfsi->btime = fattr->btime;
 552		else if (fattr_supported & NFS_ATTR_FATTR_BTIME)
 553			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BTIME);
 554		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
 555			inode_set_iversion_raw(inode, fattr->change_attr);
 556		else
 557			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
 558		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
 559			inode->i_size = nfs_size_to_loff_t(fattr->size);
 560		else
 561			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
 562		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
 563			set_nlink(inode, fattr->nlink);
 564		else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
 565			nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
 566		else
 567			set_nlink(inode, 1);
 568		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
 569			inode->i_uid = fattr->uid;
 570		else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
 571			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
 572		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
 573			inode->i_gid = fattr->gid;
 574		else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
 575			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
 576		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
 577			inode->i_blocks = fattr->du.nfs2.blocks;
 578		else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
 579			 fattr->size != 0)
 580			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
 581		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
 582			/*
 583			 * report the blocks in 512byte units
 584			 */
 585			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
 586		} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
 587			   fattr->size != 0)
 588			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
 589
 590		nfs_setsecurity(inode, fattr);
 591
 592		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
 593		nfsi->attrtimeo_timestamp = now;
 594		nfsi->access_cache = RB_ROOT;
 595
 596		nfs_fscache_init_inode(inode);
 597
 598		unlock_new_inode(inode);
 599	} else {
 600		int err = nfs_refresh_inode(inode, fattr);
 601		if (err < 0) {
 602			iput(inode);
 603			inode = ERR_PTR(err);
 604			goto out_no_inode;
 605		}
 606	}
 607	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
 608		inode->i_sb->s_id,
 609		(unsigned long long)NFS_FILEID(inode),
 610		nfs_display_fhandle_hash(fh),
 611		icount_read(inode));
 612
 613out:
 614	return inode;
 615
 616out_no_inode:
 617	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
 618	goto out;
 619}
 620EXPORT_SYMBOL_GPL(nfs_fhget);
 621
 622static void
 623nfs_fattr_fixup_delegated(struct inode *inode, struct nfs_fattr *fattr)
 624{
 625	unsigned long cache_validity = NFS_I(inode)->cache_validity;
 626
 627	if (nfs_have_delegated_mtime(inode)) {
 628		if (!(cache_validity & NFS_INO_INVALID_CTIME))
 629			fattr->valid &= ~(NFS_ATTR_FATTR_PRECTIME |
 630					  NFS_ATTR_FATTR_CTIME);
 631
 632		if (!(cache_validity & NFS_INO_INVALID_MTIME))
 633			fattr->valid &= ~(NFS_ATTR_FATTR_PREMTIME |
 634					  NFS_ATTR_FATTR_MTIME);
 635
 636		if (!(cache_validity & NFS_INO_INVALID_ATIME))
 637			fattr->valid &= ~NFS_ATTR_FATTR_ATIME;
 638	} else if (nfs_have_delegated_atime(inode)) {
 639		if (!(cache_validity & NFS_INO_INVALID_ATIME))
 640			fattr->valid &= ~NFS_ATTR_FATTR_ATIME;
 641	}
 642}
 643
 644static void nfs_set_timestamps_to_ts(struct inode *inode, struct iattr *attr)
 645{
 646	unsigned int cache_flags = 0;
 647
 648	if (attr->ia_valid & ATTR_MTIME_SET) {
 649		struct timespec64 ctime = inode_get_ctime(inode);
 650		struct timespec64 mtime = inode_get_mtime(inode);
 651		struct timespec64 now;
 652		int updated = 0;
 653
 654		now = inode_set_ctime_current(inode);
 655		if (!timespec64_equal(&now, &ctime))
 656			updated |= S_CTIME;
 657
 658		inode_set_mtime_to_ts(inode, attr->ia_mtime);
 659		if (!timespec64_equal(&now, &mtime))
 660			updated |= S_MTIME;
 661
 662		inode_maybe_inc_iversion(inode, updated);
 663		cache_flags |= NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME;
 664	}
 665	if (attr->ia_valid & ATTR_ATIME_SET) {
 666		inode_set_atime_to_ts(inode, attr->ia_atime);
 667		cache_flags |= NFS_INO_INVALID_ATIME;
 668	}
 669	NFS_I(inode)->cache_validity &= ~cache_flags;
 670}
 671
 672static void nfs_update_timestamps(struct inode *inode, unsigned int ia_valid)
 673{
 674	enum file_time_flags time_flags = 0;
 675	unsigned int cache_flags = 0;
 676
 677	if (ia_valid & ATTR_MTIME) {
 678		time_flags |= S_MTIME | S_CTIME;
 679		cache_flags |= NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME;
 680	}
 681	if (ia_valid & ATTR_ATIME) {
 682		time_flags |= S_ATIME;
 683		cache_flags |= NFS_INO_INVALID_ATIME;
 684	}
 685	inode_update_timestamps(inode, time_flags);
 686	NFS_I(inode)->cache_validity &= ~cache_flags;
 687}
 688
 689void nfs_update_delegated_atime(struct inode *inode)
 690{
 691	spin_lock(&inode->i_lock);
 692	if (nfs_have_delegated_atime(inode))
 693		nfs_update_timestamps(inode, ATTR_ATIME);
 694	spin_unlock(&inode->i_lock);
 695}
 696
 697void nfs_update_delegated_mtime_locked(struct inode *inode)
 698{
 699	if (nfs_have_delegated_mtime(inode))
 700		nfs_update_timestamps(inode, ATTR_MTIME);
 701}
 702
 703void nfs_update_delegated_mtime(struct inode *inode)
 704{
 705	spin_lock(&inode->i_lock);
 706	nfs_update_delegated_mtime_locked(inode);
 707	spin_unlock(&inode->i_lock);
 708}
 709EXPORT_SYMBOL_GPL(nfs_update_delegated_mtime);
 710
 711#define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
 712
 713int
 714nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 715	    struct iattr *attr)
 716{
 717	struct inode *inode = d_inode(dentry);
 718	struct nfs_fattr *fattr;
 719	loff_t oldsize;
 720	int error = 0;
 721	kuid_t task_uid = current_fsuid();
 722	kuid_t owner_uid = inode->i_uid;
 723
 724	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
 725
 726	/* skip mode change if it's just for clearing setuid/setgid */
 727	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
 728		attr->ia_valid &= ~ATTR_MODE;
 729
 730	if (S_ISREG(inode->i_mode))
 731		nfs_file_block_o_direct(NFS_I(inode));
 732
 733	oldsize = i_size_read(inode);
 734	if (attr->ia_valid & ATTR_SIZE) {
 735		BUG_ON(!S_ISREG(inode->i_mode));
 736
 737		error = inode_newsize_ok(inode, attr->ia_size);
 738		if (error)
 739			return error;
 740
 741		if (attr->ia_size == oldsize)
 742			attr->ia_valid &= ~ATTR_SIZE;
 743	}
 744
 745	if (nfs_have_delegated_mtime(inode) && attr->ia_valid & ATTR_MTIME) {
 746		spin_lock(&inode->i_lock);
 747		if (attr->ia_valid & ATTR_MTIME_SET) {
 748			if (uid_eq(task_uid, owner_uid)) {
 749				nfs_set_timestamps_to_ts(inode, attr);
 750				attr->ia_valid &= ~(ATTR_MTIME|ATTR_MTIME_SET|
 751						ATTR_ATIME|ATTR_ATIME_SET);
 752			}
 753		} else {
 754			nfs_update_timestamps(inode, attr->ia_valid);
 755			attr->ia_valid &= ~(ATTR_MTIME|ATTR_ATIME);
 756		}
 757		spin_unlock(&inode->i_lock);
 758	} else if (nfs_have_delegated_atime(inode) &&
 759		   attr->ia_valid & ATTR_ATIME &&
 760		   !(attr->ia_valid & ATTR_MTIME)) {
 761		if (attr->ia_valid & ATTR_ATIME_SET) {
 762			if (uid_eq(task_uid, owner_uid)) {
 763				spin_lock(&inode->i_lock);
 764				nfs_set_timestamps_to_ts(inode, attr);
 765				spin_unlock(&inode->i_lock);
 766				attr->ia_valid &= ~(ATTR_ATIME|ATTR_ATIME_SET);
 767			}
 768		} else {
 769			nfs_update_delegated_atime(inode);
 770			attr->ia_valid &= ~ATTR_ATIME;
 771		}
 772	}
 773
 774	/* Optimization: if the end result is no change, don't RPC */
 775	if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
 776		return 0;
 777
 778	trace_nfs_setattr_enter(inode);
 779
 780	/* Write all dirty data */
 781	if (S_ISREG(inode->i_mode))
 782		nfs_sync_inode(inode);
 783
 784	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
 785	if (fattr == NULL) {
 786		error = -ENOMEM;
 787		goto out;
 788	}
 789
 790	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
 791	if (error == 0) {
 792		if (attr->ia_valid & ATTR_SIZE)
 793			nfs_truncate_last_folio(inode->i_mapping, oldsize,
 794						attr->ia_size);
 795		error = nfs_refresh_inode(inode, fattr);
 796	}
 797	nfs_free_fattr(fattr);
 798out:
 799	trace_nfs_setattr_exit(inode, error);
 800	return error;
 801}
 802EXPORT_SYMBOL_GPL(nfs_setattr);
 803
 804/**
 805 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
 806 * @inode: inode of the file used
 807 * @offset: file offset to start truncating
 808 *
 809 * This is a copy of the common vmtruncate, but with the locking
 810 * corrected to take into account the fact that NFS requires
 811 * inode->i_size to be updated under the inode->i_lock.
 812 * Note: must be called with inode->i_lock held!
 813 */
 814static int nfs_vmtruncate(struct inode * inode, loff_t offset)
 815{
 816	int err;
 817
 818	err = inode_newsize_ok(inode, offset);
 819	if (err)
 820		goto out;
 821
 822	trace_nfs_size_truncate(inode, offset);
 823	i_size_write(inode, offset);
 824	/* Optimisation */
 825	if (offset == 0) {
 826		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
 827		nfs_ooo_clear(NFS_I(inode));
 828	}
 829	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
 830
 831	spin_unlock(&inode->i_lock);
 832	truncate_pagecache(inode, offset);
 833	nfs_update_delegated_mtime_locked(inode);
 834	spin_lock(&inode->i_lock);
 835out:
 836	return err;
 837}
 838
 839/**
 840 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
 841 * @inode: pointer to struct inode
 842 * @attr: pointer to struct iattr
 843 * @fattr: pointer to struct nfs_fattr
 844 *
 845 * Note: we do this in the *proc.c in order to ensure that
 846 *       it works for things like exclusive creates too.
 847 */
 848void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
 849		struct nfs_fattr *fattr)
 850{
 851	/* Barrier: bump the attribute generation count. */
 852	nfs_fattr_set_barrier(fattr);
 853
 854	spin_lock(&inode->i_lock);
 855	NFS_I(inode)->attr_gencount = fattr->gencount;
 856	if ((attr->ia_valid & ATTR_SIZE) != 0) {
 857		if (!nfs_have_delegated_mtime(inode))
 858			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
 859		nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
 860		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
 861		nfs_vmtruncate(inode, attr->ia_size);
 862	}
 863	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
 864		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
 865		if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
 866		    inode->i_mode & S_ISUID)
 867			inode->i_mode &= ~S_ISUID;
 868		if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
 869			inode->i_mode &= ~S_ISGID;
 870		if ((attr->ia_valid & ATTR_MODE) != 0) {
 871			int mode = attr->ia_mode & S_IALLUGO;
 872			mode |= inode->i_mode & ~S_IALLUGO;
 873			inode->i_mode = mode;
 874		}
 875		if ((attr->ia_valid & ATTR_UID) != 0)
 876			inode->i_uid = attr->ia_uid;
 877		if ((attr->ia_valid & ATTR_GID) != 0)
 878			inode->i_gid = attr->ia_gid;
 879		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 880			inode_set_ctime_to_ts(inode, fattr->ctime);
 881		else
 882			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
 883					| NFS_INO_INVALID_CTIME);
 884		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
 885				| NFS_INO_INVALID_ACL);
 886	}
 887	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
 888		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
 889				| NFS_INO_INVALID_CTIME);
 890		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
 891			inode_set_atime_to_ts(inode, fattr->atime);
 892		else if (attr->ia_valid & ATTR_ATIME_SET)
 893			inode_set_atime_to_ts(inode, attr->ia_atime);
 894		else
 895			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
 896
 897		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 898			inode_set_ctime_to_ts(inode, fattr->ctime);
 899		else
 900			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
 901					| NFS_INO_INVALID_CTIME);
 902	}
 903	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
 904		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
 905				| NFS_INO_INVALID_CTIME);
 906		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
 907			inode_set_mtime_to_ts(inode, fattr->mtime);
 908		else if (attr->ia_valid & ATTR_MTIME_SET)
 909			inode_set_mtime_to_ts(inode, attr->ia_mtime);
 910		else
 911			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
 912
 913		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 914			inode_set_ctime_to_ts(inode, fattr->ctime);
 915		else
 916			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
 917					| NFS_INO_INVALID_CTIME);
 918	}
 919	if (fattr->valid)
 920		nfs_update_inode(inode, fattr);
 921	spin_unlock(&inode->i_lock);
 922}
 923EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
 924
 925/*
 926 * Don't request help from readdirplus if the file is being written to,
 927 * or if attribute caching is turned off
 928 */
 929static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
 930{
 931	return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
 932	       !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
 933}
 934
 935static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
 936{
 937	if (!IS_ROOT(dentry)) {
 938		struct dentry *parent = dget_parent(dentry);
 939		nfs_readdir_record_entry_cache_miss(d_inode(parent));
 940		dput(parent);
 941	}
 942}
 943
 944static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
 945{
 946	if (!IS_ROOT(dentry)) {
 947		struct dentry *parent = dget_parent(dentry);
 948		nfs_readdir_record_entry_cache_hit(d_inode(parent));
 949		dput(parent);
 950	}
 951}
 952
 953static u32 nfs_get_valid_attrmask(struct inode *inode)
 954{
 955	u64 fattr_valid = NFS_SERVER(inode)->fattr_valid;
 956	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
 957	u32 reply_mask = STATX_INO | STATX_TYPE;
 958
 959	if (!(cache_validity & NFS_INO_INVALID_ATIME))
 960		reply_mask |= STATX_ATIME;
 961	if (!(cache_validity & NFS_INO_INVALID_CTIME))
 962		reply_mask |= STATX_CTIME;
 963	if (!(cache_validity & NFS_INO_INVALID_MTIME))
 964		reply_mask |= STATX_MTIME;
 965	if (!(cache_validity & NFS_INO_INVALID_SIZE))
 966		reply_mask |= STATX_SIZE;
 967	if (!(cache_validity & NFS_INO_INVALID_NLINK))
 968		reply_mask |= STATX_NLINK;
 969	if (!(cache_validity & NFS_INO_INVALID_MODE))
 970		reply_mask |= STATX_MODE;
 971	if (!(cache_validity & NFS_INO_INVALID_OTHER))
 972		reply_mask |= STATX_UID | STATX_GID;
 973	if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
 974		reply_mask |= STATX_BLOCKS;
 975	if (!(cache_validity & NFS_INO_INVALID_BTIME) &&
 976	    (fattr_valid & NFS_ATTR_FATTR_BTIME))
 977		reply_mask |= STATX_BTIME;
 978	if (!(cache_validity & NFS_INO_INVALID_CHANGE))
 979		reply_mask |= STATX_CHANGE_COOKIE;
 980	return reply_mask;
 981}
 982
 983int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
 984		struct kstat *stat, u32 request_mask, unsigned int query_flags)
 985{
 986	struct inode *inode = d_inode(path->dentry);
 987	struct nfs_server *server = NFS_SERVER(inode);
 988	u64 fattr_valid = server->fattr_valid;
 989	unsigned long cache_validity;
 990	int err = 0;
 991	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
 992	bool do_update = false;
 993	bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
 994
 995	trace_nfs_getattr_enter(inode);
 996
 997	request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
 998			STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
 999			STATX_INO | STATX_SIZE | STATX_BLOCKS | STATX_BTIME |
1000			STATX_CHANGE_COOKIE;
1001
1002	if (!(fattr_valid & NFS_ATTR_FATTR_BTIME))
1003		request_mask &= ~STATX_BTIME;
1004
1005	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
1006		if (readdirplus_enabled)
1007			nfs_readdirplus_parent_cache_hit(path->dentry);
1008		goto out_no_revalidate;
1009	}
1010
1011	/* Flush out writes to the server in order to update c/mtime/version.  */
1012	if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
1013	    S_ISREG(inode->i_mode)) {
1014		if (nfs_have_delegated_mtime(inode))
1015			filemap_fdatawrite(inode->i_mapping);
1016		else
1017			filemap_write_and_wait(inode->i_mapping);
1018	}
1019
1020	/*
1021	 * We may force a getattr if the user cares about atime.
1022	 *
1023	 * Note that we only have to check the vfsmount flags here:
1024	 *  - NFS always sets S_NOATIME by so checking it would give a
1025	 *    bogus result
1026	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
1027	 *    no point in checking those.
1028	 */
1029	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
1030	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
1031		request_mask &= ~STATX_ATIME;
1032
1033	/* Is the user requesting attributes that might need revalidation? */
1034	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
1035					STATX_MTIME|STATX_UID|STATX_GID|
1036					STATX_SIZE|STATX_BLOCKS|STATX_BTIME|
1037					STATX_CHANGE_COOKIE)))
1038		goto out_no_revalidate;
1039
1040	/* Check whether the cached attributes are stale */
1041	do_update |= force_sync || nfs_attribute_cache_expired(inode);
1042	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
1043	do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
1044	if (request_mask & STATX_ATIME)
1045		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
1046	if (request_mask & STATX_CTIME)
1047		do_update |= cache_validity & NFS_INO_INVALID_CTIME;
1048	if (request_mask & STATX_MTIME)
1049		do_update |= cache_validity & NFS_INO_INVALID_MTIME;
1050	if (request_mask & STATX_SIZE)
1051		do_update |= cache_validity & NFS_INO_INVALID_SIZE;
1052	if (request_mask & STATX_NLINK)
1053		do_update |= cache_validity & NFS_INO_INVALID_NLINK;
1054	if (request_mask & STATX_MODE)
1055		do_update |= cache_validity & NFS_INO_INVALID_MODE;
1056	if (request_mask & (STATX_UID | STATX_GID))
1057		do_update |= cache_validity & NFS_INO_INVALID_OTHER;
1058	if (request_mask & STATX_BLOCKS)
1059		do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
1060	if (request_mask & STATX_BTIME)
1061		do_update |= cache_validity & NFS_INO_INVALID_BTIME;
1062
1063	if (do_update) {
1064		if (readdirplus_enabled)
1065			nfs_readdirplus_parent_cache_miss(path->dentry);
1066		err = __nfs_revalidate_inode(server, inode);
1067		if (err)
1068			goto out;
1069	} else if (readdirplus_enabled)
1070		nfs_readdirplus_parent_cache_hit(path->dentry);
1071out_no_revalidate:
1072	/* Only return attributes that were revalidated. */
1073	stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
1074
1075	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1076	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
1077	stat->change_cookie = inode_peek_iversion_raw(inode);
1078	stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
1079	if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
1080		stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
1081	if (S_ISDIR(inode->i_mode))
1082		stat->blksize = NFS_SERVER(inode)->dtsize;
1083	stat->btime = NFS_I(inode)->btime;
1084
1085	/* Special handling for STATX_DIOALIGN and STATX_DIO_READ_ALIGN
1086	 * - NFS doesn't have DIO alignment constraints, avoid getting
1087	 *   these DIO attrs from remote and just respond with most
1088	 *   accommodating limits (so client will issue supported DIO).
1089	 * - this is unintuitive, but the most coarse-grained
1090	 *   dio_offset_align is the most accommodating.
1091	 */
1092	if ((request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN)) &&
1093	    S_ISREG(inode->i_mode)) {
1094		stat->result_mask |= STATX_DIOALIGN | STATX_DIO_READ_ALIGN;
1095		stat->dio_mem_align = 4; /* 4-byte alignment */
1096		stat->dio_offset_align = PAGE_SIZE;
1097		stat->dio_read_offset_align = stat->dio_offset_align;
1098	}
1099out:
1100	trace_nfs_getattr_exit(inode, err);
1101	return err;
1102}
1103EXPORT_SYMBOL_GPL(nfs_getattr);
1104
1105static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
1106{
1107	refcount_set(&l_ctx->count, 1);
1108	l_ctx->lockowner = current->files;
1109	INIT_LIST_HEAD(&l_ctx->list);
1110	atomic_set(&l_ctx->io_count, 0);
1111}
1112
1113static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
1114{
1115	struct nfs_lock_context *pos;
1116
1117	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
1118		if (pos->lockowner != current->files)
1119			continue;
1120		if (refcount_inc_not_zero(&pos->count))
1121			return pos;
1122	}
1123	return NULL;
1124}
1125
1126struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
1127{
1128	struct nfs_lock_context *res, *new = NULL;
1129	struct inode *inode = d_inode(ctx->dentry);
1130
1131	rcu_read_lock();
1132	res = __nfs_find_lock_context(ctx);
1133	rcu_read_unlock();
1134	if (res == NULL) {
1135		new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1136		if (new == NULL)
1137			return ERR_PTR(-ENOMEM);
1138		nfs_init_lock_context(new);
1139		spin_lock(&inode->i_lock);
1140		res = __nfs_find_lock_context(ctx);
1141		if (res == NULL) {
1142			new->open_context = get_nfs_open_context(ctx);
1143			if (new->open_context) {
1144				list_add_tail_rcu(&new->list,
1145						&ctx->lock_context.list);
1146				res = new;
1147				new = NULL;
1148			} else
1149				res = ERR_PTR(-EBADF);
1150		}
1151		spin_unlock(&inode->i_lock);
1152		kfree(new);
1153	}
1154	return res;
1155}
1156EXPORT_SYMBOL_GPL(nfs_get_lock_context);
1157
1158void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
1159{
1160	struct nfs_open_context *ctx = l_ctx->open_context;
1161	struct inode *inode = d_inode(ctx->dentry);
1162
1163	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
1164		return;
1165	list_del_rcu(&l_ctx->list);
1166	spin_unlock(&inode->i_lock);
1167	put_nfs_open_context(ctx);
1168	kfree_rcu(l_ctx, rcu_head);
1169}
1170EXPORT_SYMBOL_GPL(nfs_put_lock_context);
1171
1172/**
1173 * nfs_close_context - Common close_context() routine NFSv2/v3
1174 * @ctx: pointer to context
1175 * @is_sync: is this a synchronous close
1176 *
1177 * Ensure that the attributes are up to date if we're mounted
1178 * with close-to-open semantics and we have cached data that will
1179 * need to be revalidated on open.
1180 */
1181void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1182{
1183	struct nfs_inode *nfsi;
1184	struct inode *inode;
1185
1186	if (!(ctx->mode & FMODE_WRITE))
1187		return;
1188	if (!is_sync)
1189		return;
1190	inode = d_inode(ctx->dentry);
1191	if (nfs_have_read_or_write_delegation(inode))
1192		return;
1193	nfsi = NFS_I(inode);
1194	if (inode->i_mapping->nrpages == 0)
1195		return;
1196	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1197		return;
1198	if (!list_empty(&nfsi->open_files))
1199		return;
1200	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1201		return;
1202	nfs_revalidate_inode(inode,
1203			     NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1204}
1205EXPORT_SYMBOL_GPL(nfs_close_context);
1206
1207struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1208						fmode_t f_mode,
1209						struct file *filp)
1210{
1211	struct nfs_open_context *ctx;
1212
1213	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1214	if (!ctx)
1215		return ERR_PTR(-ENOMEM);
1216	nfs_sb_active(dentry->d_sb);
1217	ctx->dentry = dget(dentry);
1218	if (filp)
1219		ctx->cred = get_cred(filp->f_cred);
1220	else
1221		ctx->cred = get_current_cred();
1222	rcu_assign_pointer(ctx->ll_cred, NULL);
1223	ctx->state = NULL;
1224	ctx->mode = f_mode;
1225	ctx->flags = 0;
1226	ctx->error = 0;
1227	ctx->flock_owner = (fl_owner_t)filp;
1228	nfs_init_lock_context(&ctx->lock_context);
1229	ctx->lock_context.open_context = ctx;
1230	INIT_LIST_HEAD(&ctx->list);
1231	ctx->mdsthreshold = NULL;
1232	nfs_localio_file_init(&ctx->nfl);
1233
1234	return ctx;
1235}
1236EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1237
1238struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1239{
1240	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1241		return ctx;
1242	return NULL;
1243}
1244EXPORT_SYMBOL_GPL(get_nfs_open_context);
1245
1246static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1247{
1248	struct inode *inode = d_inode(ctx->dentry);
1249	struct super_block *sb = ctx->dentry->d_sb;
1250
1251	if (!refcount_dec_and_test(&ctx->lock_context.count))
1252		return;
1253	if (!list_empty(&ctx->list)) {
1254		spin_lock(&inode->i_lock);
1255		list_del_rcu(&ctx->list);
1256		spin_unlock(&inode->i_lock);
1257	}
1258	if (inode != NULL)
1259		NFS_PROTO(inode)->close_context(ctx, is_sync);
1260	put_cred(ctx->cred);
1261	dput(ctx->dentry);
1262	nfs_sb_deactive(sb);
1263	put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1264	kfree(ctx->mdsthreshold);
1265	nfs_close_local_fh(&ctx->nfl);
1266	kfree_rcu(ctx, rcu_head);
1267}
1268
1269void put_nfs_open_context(struct nfs_open_context *ctx)
1270{
1271	__put_nfs_open_context(ctx, 0);
1272}
1273EXPORT_SYMBOL_GPL(put_nfs_open_context);
1274
1275static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1276{
1277	__put_nfs_open_context(ctx, 1);
1278}
1279
1280/*
1281 * Ensure that mmap has a recent RPC credential for use when writing out
1282 * shared pages
1283 */
1284void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1285{
1286	struct inode *inode = d_inode(ctx->dentry);
1287	struct nfs_inode *nfsi = NFS_I(inode);
1288
1289	spin_lock(&inode->i_lock);
1290	if (list_empty(&nfsi->open_files) &&
1291	    nfs_ooo_test(nfsi))
1292		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1293						     NFS_INO_REVAL_FORCED);
1294	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1295	spin_unlock(&inode->i_lock);
1296}
1297EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1298
1299void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1300{
1301	filp->private_data = get_nfs_open_context(ctx);
1302	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1303	if (list_empty(&ctx->list))
1304		nfs_inode_attach_open_context(ctx);
1305}
1306EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1307
1308/*
1309 * Given an inode, search for an open context with the desired characteristics
1310 */
1311struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1312{
1313	struct nfs_inode *nfsi = NFS_I(inode);
1314	struct nfs_open_context *pos, *ctx = NULL;
1315
1316	rcu_read_lock();
1317	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1318		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1319			continue;
1320		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1321			continue;
1322		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1323			continue;
1324		ctx = get_nfs_open_context(pos);
1325		if (ctx)
1326			break;
1327	}
1328	rcu_read_unlock();
1329	return ctx;
1330}
1331
1332void nfs_file_clear_open_context(struct file *filp)
1333{
1334	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1335
1336	if (ctx) {
1337		struct inode *inode = d_inode(ctx->dentry);
1338
1339		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1340		/*
1341		 * We fatal error on write before. Try to writeback
1342		 * every page again.
1343		 */
1344		if (ctx->error < 0)
1345			invalidate_inode_pages2(inode->i_mapping);
1346		filp->private_data = NULL;
1347		put_nfs_open_context_sync(ctx);
1348	}
1349}
1350
1351/*
1352 * These allocate and release file read/write context information.
1353 */
1354int nfs_open(struct inode *inode, struct file *filp)
1355{
1356	struct nfs_open_context *ctx;
1357
1358	ctx = alloc_nfs_open_context(file_dentry(filp),
1359				     flags_to_mode(filp->f_flags), filp);
1360	if (IS_ERR(ctx))
1361		return PTR_ERR(ctx);
1362	nfs_file_set_open_context(filp, ctx);
1363	put_nfs_open_context(ctx);
1364	nfs_fscache_open_file(inode, filp);
1365	return 0;
1366}
1367
1368/*
1369 * This function is called whenever some part of NFS notices that
1370 * the cached attributes have to be refreshed.
1371 */
1372int
1373__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1374{
1375	int		 status = -ESTALE;
1376	struct nfs_fattr *fattr = NULL;
1377	struct nfs_inode *nfsi = NFS_I(inode);
1378
1379	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1380		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1381
1382	trace_nfs_revalidate_inode_enter(inode);
1383
1384	if (is_bad_inode(inode))
1385		goto out;
1386	if (NFS_STALE(inode))
1387		goto out;
1388
1389	/* pNFS: Attributes aren't updated until we layoutcommit */
1390	if (S_ISREG(inode->i_mode)) {
1391		status = pnfs_sync_inode(inode, false);
1392		if (status)
1393			goto out;
1394	} else if (nfs_have_directory_delegation(inode)) {
1395		status = 0;
1396		goto out;
1397	}
1398
1399	status = -ENOMEM;
1400	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1401	if (fattr == NULL)
1402		goto out;
1403
1404	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1405
1406	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1407	if (status != 0) {
1408		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1409			 inode->i_sb->s_id,
1410			 (unsigned long long)NFS_FILEID(inode), status);
1411		switch (status) {
1412		case -ETIMEDOUT:
1413			/* A soft timeout occurred. Use cached information? */
1414			if (server->flags & NFS_MOUNT_SOFTREVAL)
1415				status = 0;
1416			break;
1417		case -ESTALE:
1418			if (!S_ISDIR(inode->i_mode))
1419				nfs_set_inode_stale(inode);
1420			else
1421				nfs_zap_caches(inode);
1422		}
1423		goto out;
1424	}
1425
1426	status = nfs_refresh_inode(inode, fattr);
1427	if (status) {
1428		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1429			 inode->i_sb->s_id,
1430			 (unsigned long long)NFS_FILEID(inode), status);
1431		goto out;
1432	}
1433
1434	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1435		nfs_zap_acl_cache(inode);
1436
1437	nfs_setsecurity(inode, fattr);
1438
1439	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1440		inode->i_sb->s_id,
1441		(unsigned long long)NFS_FILEID(inode));
1442
1443out:
1444	nfs_free_fattr(fattr);
1445	trace_nfs_revalidate_inode_exit(inode, status);
1446	return status;
1447}
1448
1449int nfs_attribute_cache_expired(struct inode *inode)
1450{
1451	if (nfs_have_delegated_attributes(inode))
1452		return 0;
1453	return nfs_attribute_timeout(inode);
1454}
1455
1456/**
1457 * nfs_revalidate_inode - Revalidate the inode attributes
1458 * @inode: pointer to inode struct
1459 * @flags: cache flags to check
1460 *
1461 * Updates inode attribute information by retrieving the data from the server.
1462 */
1463int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1464{
1465	if (!nfs_check_cache_invalid(inode, flags))
1466		return NFS_STALE(inode) ? -ESTALE : 0;
1467	return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1468}
1469EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1470
1471static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1472{
1473	int ret;
1474
1475	nfs_fscache_invalidate(inode, 0);
1476	if (mapping->nrpages != 0) {
1477		if (S_ISREG(inode->i_mode)) {
1478			ret = nfs_sync_mapping(mapping);
1479			if (ret < 0)
1480				return ret;
1481		}
1482		ret = invalidate_inode_pages2(mapping);
1483		if (ret < 0)
1484			return ret;
1485	}
1486	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1487
1488	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1489			inode->i_sb->s_id,
1490			(unsigned long long)NFS_FILEID(inode));
1491	return 0;
1492}
1493
1494/**
1495 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1496 * @mapping: pointer to mapping
1497 *
1498 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1499 */
1500int nfs_clear_invalid_mapping(struct address_space *mapping)
1501{
1502	struct inode *inode = mapping->host;
1503	struct nfs_inode *nfsi = NFS_I(inode);
1504	unsigned long *bitlock = &nfsi->flags;
1505	int ret = 0;
1506
1507	/*
1508	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1509	 * invalidations that come in while we're shooting down the mappings
1510	 * are respected. But, that leaves a race window where one revalidator
1511	 * can clear the flag, and then another checks it before the mapping
1512	 * gets invalidated. Fix that by serializing access to this part of
1513	 * the function.
1514	 *
1515	 * At the same time, we need to allow other tasks to see whether we
1516	 * might be in the middle of invalidating the pages, so we only set
1517	 * the bit lock here if it looks like we're going to be doing that.
1518	 */
1519	for (;;) {
1520		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1521					 nfs_wait_bit_killable,
1522					 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1523		if (ret)
1524			goto out;
1525		smp_rmb(); /* pairs with smp_wmb() below */
1526		if (test_bit(NFS_INO_INVALIDATING, bitlock))
1527			continue;
1528		/* pairs with nfs_set_cache_invalid()'s smp_store_release() */
1529		if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA))
1530			goto out;
1531		/* Slow-path that double-checks with spinlock held */
1532		spin_lock(&inode->i_lock);
1533		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1534			spin_unlock(&inode->i_lock);
1535			continue;
1536		}
1537		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1538			break;
1539		spin_unlock(&inode->i_lock);
1540		goto out;
1541	}
1542
1543	set_bit(NFS_INO_INVALIDATING, bitlock);
1544	smp_wmb();
1545	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1546	nfs_ooo_clear(nfsi);
1547	spin_unlock(&inode->i_lock);
1548	trace_nfs_invalidate_mapping_enter(inode);
1549	ret = nfs_invalidate_mapping(inode, mapping);
1550	trace_nfs_invalidate_mapping_exit(inode, ret);
1551
1552	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1553	smp_mb__after_atomic();
1554	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1555out:
1556	return ret;
1557}
1558
1559bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1560{
1561	return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1562		NFS_STALE(inode);
1563}
1564
1565int nfs_revalidate_mapping_rcu(struct inode *inode)
1566{
1567	struct nfs_inode *nfsi = NFS_I(inode);
1568	unsigned long *bitlock = &nfsi->flags;
1569	int ret = 0;
1570
1571	if (IS_SWAPFILE(inode))
1572		goto out;
1573	if (nfs_mapping_need_revalidate_inode(inode)) {
1574		ret = -ECHILD;
1575		goto out;
1576	}
1577	spin_lock(&inode->i_lock);
1578	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1579	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1580		ret = -ECHILD;
1581	spin_unlock(&inode->i_lock);
1582out:
1583	return ret;
1584}
1585
1586/**
1587 * nfs_revalidate_mapping - Revalidate the pagecache
1588 * @inode: pointer to host inode
1589 * @mapping: pointer to mapping
1590 */
1591int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1592{
1593	/* swapfiles are not supposed to be shared. */
1594	if (IS_SWAPFILE(inode))
1595		return 0;
1596
1597	if (nfs_mapping_need_revalidate_inode(inode)) {
1598		int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1599		if (ret < 0)
1600			return ret;
1601	}
1602
1603	return nfs_clear_invalid_mapping(mapping);
1604}
1605
1606static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1607{
1608	struct inode *inode = &nfsi->vfs_inode;
1609
1610	if (!S_ISREG(inode->i_mode))
1611		return false;
1612	if (list_empty(&nfsi->open_files))
1613		return false;
1614	return inode_is_open_for_write(inode);
1615}
1616
1617static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1618{
1619	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1620}
1621
1622static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1623{
1624	struct timespec64 ts;
1625
1626	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1627			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1628			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1629		inode_set_iversion_raw(inode, fattr->change_attr);
1630		if (S_ISDIR(inode->i_mode))
1631			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1632		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1633			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1634	}
1635	/* If we have atomic WCC data, we may update some attributes */
1636	ts = inode_get_ctime(inode);
1637	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1638			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1639			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1640		inode_set_ctime_to_ts(inode, fattr->ctime);
1641	}
1642
1643	ts = inode_get_mtime(inode);
1644	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1645			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1646			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1647		inode_set_mtime_to_ts(inode, fattr->mtime);
1648	}
1649	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1650			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1651			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1652			&& !nfs_have_writebacks(inode)) {
1653		trace_nfs_size_wcc(inode, fattr->size);
1654		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1655	}
1656}
1657
1658/**
1659 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1660 * @inode: pointer to inode
1661 * @fattr: updated attributes
1662 *
1663 * Verifies the attribute cache. If we have just changed the attributes,
1664 * so that fattr carries weak cache consistency data, then it may
1665 * also update the ctime/mtime/change_attribute.
1666 */
1667static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1668{
1669	struct nfs_inode *nfsi = NFS_I(inode);
1670	loff_t cur_size, new_isize;
1671	unsigned long invalid = 0;
1672	struct timespec64 ts;
1673
1674	if (nfs_have_delegated_attributes(inode))
1675		return 0;
1676
1677	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1678		/* Only a mounted-on-fileid? Just exit */
1679		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1680			return 0;
1681	/* Has the inode gone and changed behind our back? */
1682	} else if (nfsi->fileid != fattr->fileid) {
1683		/* Is this perhaps the mounted-on fileid? */
1684		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1685		    nfsi->fileid == fattr->mounted_on_fileid)
1686			return 0;
1687		return -ESTALE;
1688	}
1689	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1690		return -ESTALE;
1691
1692
1693	if (!nfs_file_has_buffered_writers(nfsi)) {
1694		/* Verify a few of the more important attributes */
1695		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1696			invalid |= NFS_INO_INVALID_CHANGE;
1697
1698		ts = inode_get_mtime(inode);
1699		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1700			invalid |= NFS_INO_INVALID_MTIME;
1701
1702		ts = inode_get_ctime(inode);
1703		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1704			invalid |= NFS_INO_INVALID_CTIME;
1705
1706		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1707			cur_size = i_size_read(inode);
1708			new_isize = nfs_size_to_loff_t(fattr->size);
1709			if (cur_size != new_isize)
1710				invalid |= NFS_INO_INVALID_SIZE;
1711		}
1712	}
1713
1714	/* Have any file permissions changed? */
1715	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1716		invalid |= NFS_INO_INVALID_MODE;
1717	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1718		invalid |= NFS_INO_INVALID_OTHER;
1719	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1720		invalid |= NFS_INO_INVALID_OTHER;
1721
1722	/* Has the link count changed? */
1723	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1724		invalid |= NFS_INO_INVALID_NLINK;
1725
1726	ts = inode_get_atime(inode);
1727	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1728		invalid |= NFS_INO_INVALID_ATIME;
1729
1730	if (invalid != 0)
1731		nfs_set_cache_invalid(inode, invalid);
1732
1733	nfsi->read_cache_jiffies = fattr->time_start;
1734	return 0;
1735}
1736
1737static atomic_long_t nfs_attr_generation_counter;
1738
1739static unsigned long nfs_read_attr_generation_counter(void)
1740{
1741	return atomic_long_read(&nfs_attr_generation_counter);
1742}
1743
1744unsigned long nfs_inc_attr_generation_counter(void)
1745{
1746	return atomic_long_inc_return(&nfs_attr_generation_counter);
1747}
1748EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1749
1750void nfs_fattr_init(struct nfs_fattr *fattr)
1751{
1752	fattr->valid = 0;
1753	fattr->time_start = jiffies;
1754	fattr->gencount = nfs_inc_attr_generation_counter();
1755	fattr->owner_name = NULL;
1756	fattr->group_name = NULL;
1757	fattr->mdsthreshold = NULL;
1758}
1759EXPORT_SYMBOL_GPL(nfs_fattr_init);
1760
1761/**
1762 * nfs_fattr_set_barrier
1763 * @fattr: attributes
1764 *
1765 * Used to set a barrier after an attribute was updated. This
1766 * barrier ensures that older attributes from RPC calls that may
1767 * have raced with our update cannot clobber these new values.
1768 * Note that you are still responsible for ensuring that other
1769 * operations which change the attribute on the server do not
1770 * collide.
1771 */
1772void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1773{
1774	fattr->gencount = nfs_inc_attr_generation_counter();
1775}
1776
1777struct nfs_fattr *nfs_alloc_fattr(void)
1778{
1779	struct nfs_fattr *fattr;
1780
1781	fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1782	if (fattr != NULL) {
1783		nfs_fattr_init(fattr);
1784		fattr->label = NULL;
1785	}
1786	return fattr;
1787}
1788EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1789
1790struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1791{
1792	struct nfs_fattr *fattr = nfs_alloc_fattr();
1793
1794	if (!fattr)
1795		return NULL;
1796
1797	fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1798	if (IS_ERR(fattr->label)) {
1799		kfree(fattr);
1800		return NULL;
1801	}
1802
1803	return fattr;
1804}
1805EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1806
1807struct nfs_fh *nfs_alloc_fhandle(void)
1808{
1809	struct nfs_fh *fh;
1810
1811	fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1812	if (fh != NULL)
1813		fh->size = 0;
1814	return fh;
1815}
1816EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1817
1818#ifdef NFS_DEBUG
1819/*
1820 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1821 *                             in the same way that wireshark does
1822 *
1823 * @fh: file handle
1824 *
1825 * For debugging only.
1826 */
1827u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1828{
1829	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1830	 * not on the result */
1831	return nfs_fhandle_hash(fh);
1832}
1833EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1834
1835/*
1836 * _nfs_display_fhandle - display an NFS file handle on the console
1837 *
1838 * @fh: file handle to display
1839 * @caption: display caption
1840 *
1841 * For debugging only.
1842 */
1843void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1844{
1845	unsigned short i;
1846
1847	if (fh == NULL || fh->size == 0) {
1848		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1849		return;
1850	}
1851
1852	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1853	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1854	for (i = 0; i < fh->size; i += 16) {
1855		__be32 *pos = (__be32 *)&fh->data[i];
1856
1857		switch ((fh->size - i - 1) >> 2) {
1858		case 0:
1859			printk(KERN_DEFAULT " %08x\n",
1860				be32_to_cpup(pos));
1861			break;
1862		case 1:
1863			printk(KERN_DEFAULT " %08x %08x\n",
1864				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1865			break;
1866		case 2:
1867			printk(KERN_DEFAULT " %08x %08x %08x\n",
1868				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1869				be32_to_cpup(pos + 2));
1870			break;
1871		default:
1872			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1873				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1874				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1875		}
1876	}
1877}
1878EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1879#endif
1880
1881/**
1882 * nfs_inode_attrs_cmp_generic - compare attributes
1883 * @fattr: attributes
1884 * @inode: pointer to inode
1885 *
1886 * Attempt to divine whether or not an RPC call reply carrying stale
1887 * attributes got scheduled after another call carrying updated ones.
1888 * Note also the check for wraparound of 'attr_gencount'
1889 *
1890 * The function returns '1' if it thinks the attributes in @fattr are
1891 * more recent than the ones cached in @inode. Otherwise it returns
1892 * the value '0'.
1893 */
1894static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1895				       const struct inode *inode)
1896{
1897	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1898
1899	return (long)(fattr->gencount - attr_gencount) > 0 ||
1900	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1901}
1902
1903/**
1904 * nfs_inode_attrs_cmp_monotonic - compare attributes
1905 * @fattr: attributes
1906 * @inode: pointer to inode
1907 *
1908 * Attempt to divine whether or not an RPC call reply carrying stale
1909 * attributes got scheduled after another call carrying updated ones.
1910 *
1911 * We assume that the server observes monotonic semantics for
1912 * the change attribute, so a larger value means that the attributes in
1913 * @fattr are more recent, in which case the function returns the
1914 * value '1'.
1915 * A return value of '0' indicates no measurable change
1916 * A return value of '-1' means that the attributes in @inode are
1917 * more recent.
1918 */
1919static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1920					 const struct inode *inode)
1921{
1922	s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1923	if (diff > 0)
1924		return 1;
1925	return diff == 0 ? 0 : -1;
1926}
1927
1928/**
1929 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1930 * @fattr: attributes
1931 * @inode: pointer to inode
1932 *
1933 * Attempt to divine whether or not an RPC call reply carrying stale
1934 * attributes got scheduled after another call carrying updated ones.
1935 *
1936 * We assume that the server observes strictly monotonic semantics for
1937 * the change attribute, so a larger value means that the attributes in
1938 * @fattr are more recent, in which case the function returns the
1939 * value '1'.
1940 * A return value of '-1' means that the attributes in @inode are
1941 * more recent or unchanged.
1942 */
1943static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1944						const struct inode *inode)
1945{
1946	return  nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1947}
1948
1949/**
1950 * nfs_inode_attrs_cmp - compare attributes
1951 * @fattr: attributes
1952 * @inode: pointer to inode
1953 *
1954 * This function returns '1' if it thinks the attributes in @fattr are
1955 * more recent than the ones cached in @inode. It returns '-1' if
1956 * the attributes in @inode are more recent than the ones in @fattr,
1957 * and it returns 0 if not sure.
1958 */
1959static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1960			       const struct inode *inode)
1961{
1962	if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1963		return 1;
1964	switch (NFS_SERVER(inode)->change_attr_type) {
1965	case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1966		break;
1967	case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1968		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1969			break;
1970		return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1971	default:
1972		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1973			break;
1974		return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1975	}
1976	return 0;
1977}
1978
1979/**
1980 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1981 * @fattr: attributes
1982 * @inode: pointer to inode
1983 *
1984 * Returns '1' if the last attribute update left the inode cached
1985 * attributes in a partially unrevalidated state, and @fattr
1986 * matches the change attribute of that partial update.
1987 * Otherwise returns '0'.
1988 */
1989static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1990						const struct inode *inode)
1991{
1992	const unsigned long check_valid =
1993		NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1994		NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1995		NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1996		NFS_INO_INVALID_NLINK | NFS_INO_INVALID_BTIME;
1997	unsigned long cache_validity = NFS_I(inode)->cache_validity;
1998	enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1999
2000	if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
2001	    !(cache_validity & NFS_INO_INVALID_CHANGE) &&
2002	    (cache_validity & check_valid) != 0 &&
2003	    (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2004	    nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
2005		return 1;
2006	return 0;
2007}
2008
2009static void nfs_ooo_merge(struct nfs_inode *nfsi,
2010			  u64 start, u64 end)
2011{
2012	int i, cnt;
2013
2014	if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
2015		/* No point merging anything */
2016		return;
2017
2018	if (!nfsi->ooo) {
2019		nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
2020		if (!nfsi->ooo) {
2021			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2022			return;
2023		}
2024		nfsi->ooo->cnt = 0;
2025	}
2026
2027	/* add this range, merging if possible */
2028	cnt = nfsi->ooo->cnt;
2029	for (i = 0; i < cnt; i++) {
2030		if (end == nfsi->ooo->gap[i].start)
2031			end = nfsi->ooo->gap[i].end;
2032		else if (start == nfsi->ooo->gap[i].end)
2033			start = nfsi->ooo->gap[i].start;
2034		else
2035			continue;
2036		/* Remove 'i' from table and loop to insert the new range */
2037		cnt -= 1;
2038		nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
2039		i = -1;
2040	}
2041	if (start != end) {
2042		if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
2043			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2044			kfree(nfsi->ooo);
2045			nfsi->ooo = NULL;
2046			return;
2047		}
2048		nfsi->ooo->gap[cnt].start = start;
2049		nfsi->ooo->gap[cnt].end = end;
2050		cnt += 1;
2051	}
2052	nfsi->ooo->cnt = cnt;
2053}
2054
2055static void nfs_ooo_record(struct nfs_inode *nfsi,
2056			   struct nfs_fattr *fattr)
2057{
2058	/* This reply was out-of-order, so record in the
2059	 * pre/post change id, possibly cancelling
2060	 * gaps created when iversion was jumpped forward.
2061	 */
2062	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
2063	    (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
2064		nfs_ooo_merge(nfsi,
2065			      fattr->change_attr,
2066			      fattr->pre_change_attr);
2067}
2068
2069static int nfs_refresh_inode_locked(struct inode *inode,
2070				    struct nfs_fattr *fattr)
2071{
2072	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
2073	int ret = 0;
2074
2075	trace_nfs_refresh_inode_enter(inode);
2076
2077	if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
2078		ret = nfs_update_inode(inode, fattr);
2079	else {
2080		nfs_ooo_record(NFS_I(inode), fattr);
2081
2082		if (attr_cmp == 0)
2083			ret = nfs_check_inode_attributes(inode, fattr);
2084	}
2085
2086	trace_nfs_refresh_inode_exit(inode, ret);
2087	return ret;
2088}
2089
2090/**
2091 * nfs_refresh_inode - try to update the inode attribute cache
2092 * @inode: pointer to inode
2093 * @fattr: updated attributes
2094 *
2095 * Check that an RPC call that returned attributes has not overlapped with
2096 * other recent updates of the inode metadata, then decide whether it is
2097 * safe to do a full update of the inode attributes, or whether just to
2098 * call nfs_check_inode_attributes.
2099 */
2100int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
2101{
2102	int status;
2103
2104	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
2105		return 0;
2106	spin_lock(&inode->i_lock);
2107	status = nfs_refresh_inode_locked(inode, fattr);
2108	spin_unlock(&inode->i_lock);
2109
2110	return status;
2111}
2112EXPORT_SYMBOL_GPL(nfs_refresh_inode);
2113
2114static int nfs_post_op_update_inode_locked(struct inode *inode,
2115		struct nfs_fattr *fattr, unsigned int invalid)
2116{
2117	if (S_ISDIR(inode->i_mode))
2118		invalid |= NFS_INO_INVALID_DATA;
2119	nfs_set_cache_invalid(inode, invalid);
2120	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
2121		return 0;
2122	return nfs_refresh_inode_locked(inode, fattr);
2123}
2124
2125/**
2126 * nfs_post_op_update_inode - try to update the inode attribute cache
2127 * @inode: pointer to inode
2128 * @fattr: updated attributes
2129 *
2130 * After an operation that has changed the inode metadata, mark the
2131 * attribute cache as being invalid, then try to update it.
2132 *
2133 * NB: if the server didn't return any post op attributes, this
2134 * function will force the retrieval of attributes before the next
2135 * NFS request.  Thus it should be used only for operations that
2136 * are expected to change one or more attributes, to avoid
2137 * unnecessary NFS requests and trips through nfs_update_inode().
2138 */
2139int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2140{
2141	int status;
2142
2143	spin_lock(&inode->i_lock);
2144	nfs_fattr_set_barrier(fattr);
2145	status = nfs_post_op_update_inode_locked(inode, fattr,
2146			NFS_INO_INVALID_CHANGE
2147			| NFS_INO_INVALID_CTIME
2148			| NFS_INO_REVAL_FORCED);
2149	spin_unlock(&inode->i_lock);
2150
2151	return status;
2152}
2153EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
2154
2155/**
2156 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
2157 * @inode: pointer to inode
2158 * @fattr: updated attributes
2159 *
2160 * After an operation that has changed the inode metadata, mark the
2161 * attribute cache as being invalid, then try to update it. Fake up
2162 * weak cache consistency data, if none exist.
2163 *
2164 * This function is mainly designed to be used by the ->write_done() functions.
2165 */
2166int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
2167{
2168	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
2169	int status;
2170
2171	/* Don't do a WCC update if these attributes are already stale */
2172	if (attr_cmp < 0)
2173		return 0;
2174	if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
2175		/* Record the pre/post change info before clearing PRECHANGE */
2176		nfs_ooo_record(NFS_I(inode), fattr);
2177		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
2178				| NFS_ATTR_FATTR_PRESIZE
2179				| NFS_ATTR_FATTR_PREMTIME
2180				| NFS_ATTR_FATTR_PRECTIME);
2181		goto out_noforce;
2182	}
2183	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2184			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
2185		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
2186		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
2187	}
2188	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
2189			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2190		fattr->pre_ctime = inode_get_ctime(inode);
2191		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2192	}
2193	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2194			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2195		fattr->pre_mtime = inode_get_mtime(inode);
2196		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2197	}
2198	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2199			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2200		fattr->pre_size = i_size_read(inode);
2201		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2202	}
2203out_noforce:
2204	status = nfs_post_op_update_inode_locked(inode, fattr,
2205			NFS_INO_INVALID_CHANGE
2206			| NFS_INO_INVALID_CTIME
2207			| NFS_INO_INVALID_MTIME
2208			| NFS_INO_INVALID_BLOCKS);
2209	return status;
2210}
2211
2212/**
2213 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2214 * @inode: pointer to inode
2215 * @fattr: updated attributes
2216 *
2217 * After an operation that has changed the inode metadata, mark the
2218 * attribute cache as being invalid, then try to update it. Fake up
2219 * weak cache consistency data, if none exist.
2220 *
2221 * This function is mainly designed to be used by the ->write_done() functions.
2222 */
2223int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2224{
2225	int status;
2226
2227	spin_lock(&inode->i_lock);
2228	nfs_fattr_set_barrier(fattr);
2229	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2230	spin_unlock(&inode->i_lock);
2231	return status;
2232}
2233EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2234
2235
2236/*
2237 * Many nfs protocol calls return the new file attributes after
2238 * an operation.  Here we update the inode to reflect the state
2239 * of the server's inode.
2240 *
2241 * This is a bit tricky because we have to make sure all dirty pages
2242 * have been sent off to the server before calling invalidate_inode_pages.
2243 * To make sure no other process adds more write requests while we try
2244 * our best to flush them, we make them sleep during the attribute refresh.
2245 *
2246 * A very similar scenario holds for the dir cache.
2247 */
2248static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2249{
2250	struct nfs_server *server = NFS_SERVER(inode);
2251	struct nfs_inode *nfsi = NFS_I(inode);
2252	loff_t cur_isize, new_isize;
2253	u64 fattr_supported = server->fattr_valid;
2254	unsigned long invalid = 0;
2255	unsigned long now = jiffies;
2256	unsigned long save_cache_validity;
2257	bool have_writers = nfs_file_has_buffered_writers(nfsi);
2258	bool cache_revalidated = true;
2259	bool attr_changed = false;
2260	bool have_delegation;
2261
2262	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%llx)\n",
2263			__func__, inode->i_sb->s_id, inode->i_ino,
2264			nfs_display_fhandle_hash(NFS_FH(inode)),
2265			icount_read(inode), fattr->valid);
2266
2267	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2268		/* Only a mounted-on-fileid? Just exit */
2269		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2270			return 0;
2271	/* Has the inode gone and changed behind our back? */
2272	} else if (nfsi->fileid != fattr->fileid) {
2273		/* Is this perhaps the mounted-on fileid? */
2274		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2275		    nfsi->fileid == fattr->mounted_on_fileid)
2276			return 0;
2277		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2278			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2279			NFS_SERVER(inode)->nfs_client->cl_hostname,
2280			inode->i_sb->s_id, (long long)nfsi->fileid,
2281			(long long)fattr->fileid);
2282		goto out_err;
2283	}
2284
2285	/*
2286	 * Make sure the inode's type hasn't changed.
2287	 */
2288	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2289		/*
2290		* Big trouble! The inode has become a different object.
2291		*/
2292		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2293				__func__, inode->i_ino, inode->i_mode, fattr->mode);
2294		goto out_err;
2295	}
2296
2297	/* Update the fsid? */
2298	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2299			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2300			!IS_AUTOMOUNT(inode))
2301		server->fsid = fattr->fsid;
2302
2303	/* Save the delegation state before clearing cache_validity */
2304	have_delegation = nfs_have_delegated_attributes(inode);
2305
2306	/*
2307	 * Update the read time so we don't revalidate too often.
2308	 */
2309	nfsi->read_cache_jiffies = fattr->time_start;
2310
2311	/* Fix up any delegated attributes in the struct nfs_fattr */
2312	nfs_fattr_fixup_delegated(inode, fattr);
2313
2314	save_cache_validity = nfsi->cache_validity;
2315	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2316			| NFS_INO_INVALID_ATIME
2317			| NFS_INO_REVAL_FORCED
2318			| NFS_INO_INVALID_BLOCKS);
2319
2320	/* Do atomic weak cache consistency updates */
2321	nfs_wcc_update_inode(inode, fattr);
2322
2323	if (pnfs_layoutcommit_outstanding(inode)) {
2324		nfsi->cache_validity |=
2325			save_cache_validity &
2326			(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2327			 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2328			 NFS_INO_INVALID_BLOCKS);
2329		cache_revalidated = false;
2330	}
2331
2332	/* More cache consistency checks */
2333	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2334		if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2335		    nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2336			/* There is one remaining gap that hasn't been
2337			 * merged into iversion - do that now.
2338			 */
2339			inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2340			kfree(nfsi->ooo);
2341			nfsi->ooo = NULL;
2342		}
2343		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2344			/* Could it be a race with writeback? */
2345			if (!(have_writers || have_delegation)) {
2346				invalid |= NFS_INO_INVALID_DATA
2347					| NFS_INO_INVALID_ACCESS
2348					| NFS_INO_INVALID_ACL
2349					| NFS_INO_INVALID_XATTR;
2350				/* Force revalidate of all attributes */
2351				save_cache_validity |= NFS_INO_INVALID_CTIME
2352					| NFS_INO_INVALID_MTIME
2353					| NFS_INO_INVALID_SIZE
2354					| NFS_INO_INVALID_BLOCKS
2355					| NFS_INO_INVALID_NLINK
2356					| NFS_INO_INVALID_MODE
2357					| NFS_INO_INVALID_OTHER
2358					| NFS_INO_INVALID_BTIME;
2359				if (S_ISDIR(inode->i_mode))
2360					nfs_force_lookup_revalidate(inode);
2361				attr_changed = true;
2362				dprintk("NFS: change_attr change on server for file %s/%ld\n",
2363						inode->i_sb->s_id,
2364						inode->i_ino);
2365			} else if (!have_delegation) {
2366				nfs_ooo_record(nfsi, fattr);
2367				nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2368					      fattr->change_attr);
2369			}
2370			inode_set_iversion_raw(inode, fattr->change_attr);
2371		}
2372	} else {
2373		nfsi->cache_validity |=
2374			save_cache_validity & NFS_INO_INVALID_CHANGE;
2375		if (!have_delegation ||
2376		    (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2377			cache_revalidated = false;
2378	}
2379
2380	if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2381		inode_set_mtime_to_ts(inode, fattr->mtime);
2382	else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2383		nfsi->cache_validity |=
2384			save_cache_validity & NFS_INO_INVALID_MTIME;
2385
2386	if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2387		inode_set_ctime_to_ts(inode, fattr->ctime);
2388	else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2389		nfsi->cache_validity |=
2390			save_cache_validity & NFS_INO_INVALID_CTIME;
2391
2392	if (fattr->valid & NFS_ATTR_FATTR_BTIME)
2393		nfsi->btime = fattr->btime;
2394	else if (fattr_supported & NFS_ATTR_FATTR_BTIME)
2395		nfsi->cache_validity |=
2396			save_cache_validity & NFS_INO_INVALID_BTIME;
2397
2398	/* Check if our cached file size is stale */
2399	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2400		new_isize = nfs_size_to_loff_t(fattr->size);
2401		cur_isize = i_size_read(inode);
2402		if (new_isize != cur_isize && !have_delegation) {
2403			/* Do we perhaps have any outstanding writes, or has
2404			 * the file grown beyond our last write? */
2405			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2406				trace_nfs_size_update(inode, new_isize);
2407				i_size_write(inode, new_isize);
2408				if (!have_writers)
2409					invalid |= NFS_INO_INVALID_DATA;
2410			}
2411		}
2412		if (new_isize == 0 &&
2413		    !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2414				      NFS_ATTR_FATTR_BLOCKS_USED))) {
2415			fattr->du.nfs3.used = 0;
2416			fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2417		}
2418	} else
2419		nfsi->cache_validity |=
2420			save_cache_validity & NFS_INO_INVALID_SIZE;
2421
2422	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2423		inode_set_atime_to_ts(inode, fattr->atime);
2424	else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2425		nfsi->cache_validity |=
2426			save_cache_validity & NFS_INO_INVALID_ATIME;
2427
2428	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2429		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2430			umode_t newmode = inode->i_mode & S_IFMT;
2431			newmode |= fattr->mode & S_IALLUGO;
2432			inode->i_mode = newmode;
2433			invalid |= NFS_INO_INVALID_ACCESS
2434				| NFS_INO_INVALID_ACL;
2435		}
2436	} else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2437		nfsi->cache_validity |=
2438			save_cache_validity & NFS_INO_INVALID_MODE;
2439
2440	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2441		if (!uid_eq(inode->i_uid, fattr->uid)) {
2442			invalid |= NFS_INO_INVALID_ACCESS
2443				| NFS_INO_INVALID_ACL;
2444			inode->i_uid = fattr->uid;
2445		}
2446	} else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2447		nfsi->cache_validity |=
2448			save_cache_validity & NFS_INO_INVALID_OTHER;
2449
2450	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2451		if (!gid_eq(inode->i_gid, fattr->gid)) {
2452			invalid |= NFS_INO_INVALID_ACCESS
2453				| NFS_INO_INVALID_ACL;
2454			inode->i_gid = fattr->gid;
2455		}
2456	} else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2457		nfsi->cache_validity |=
2458			save_cache_validity & NFS_INO_INVALID_OTHER;
2459
2460	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2461		if (inode->i_nlink != fattr->nlink)
2462			set_nlink(inode, fattr->nlink);
2463	} else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2464		nfsi->cache_validity |=
2465			save_cache_validity & NFS_INO_INVALID_NLINK;
2466
2467	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2468		/*
2469		 * report the blocks in 512byte units
2470		 */
2471		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2472	} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2473		nfsi->cache_validity |=
2474			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2475
2476	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2477		inode->i_blocks = fattr->du.nfs2.blocks;
2478	else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2479		nfsi->cache_validity |=
2480			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2481
2482	/* Update attrtimeo value if we're out of the unstable period */
2483	if (attr_changed) {
2484		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2485		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2486		nfsi->attrtimeo_timestamp = now;
2487		/* Set barrier to be more recent than all outstanding updates */
2488		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2489	} else {
2490		if (cache_revalidated) {
2491			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2492				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2493				nfsi->attrtimeo <<= 1;
2494				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2495					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2496			}
2497			nfsi->attrtimeo_timestamp = now;
2498		}
2499		/* Set the barrier to be more recent than this fattr */
2500		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2501			nfsi->attr_gencount = fattr->gencount;
2502	}
2503
2504	/* Don't invalidate the data if we were to blame */
2505	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2506				|| S_ISLNK(inode->i_mode)))
2507		invalid &= ~NFS_INO_INVALID_DATA;
2508	nfs_set_cache_invalid(inode, invalid);
2509
2510	return 0;
2511 out_err:
2512	/*
2513	 * No need to worry about unhashing the dentry, as the
2514	 * lookup validation will know that the inode is bad.
2515	 * (But we fall through to invalidate the caches.)
2516	 */
2517	nfs_set_inode_stale_locked(inode);
2518	return -ESTALE;
2519}
2520
2521struct inode *nfs_alloc_inode(struct super_block *sb)
2522{
2523	struct nfs_inode *nfsi;
2524	nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2525	if (!nfsi)
2526		return NULL;
2527	nfsi->flags = 0UL;
2528	nfsi->cache_validity = 0UL;
2529	nfsi->ooo = NULL;
2530#if IS_ENABLED(CONFIG_NFS_V4)
2531	nfsi->nfs4_acl = NULL;
2532#endif /* CONFIG_NFS_V4 */
2533#ifdef CONFIG_NFS_V4_2
2534	nfsi->xattr_cache = NULL;
2535#endif
2536	nfs_netfs_inode_init(nfsi);
2537
2538	return &nfsi->vfs_inode;
2539}
2540EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2541
2542void nfs_free_inode(struct inode *inode)
2543{
2544	kfree(NFS_I(inode)->ooo);
2545	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2546}
2547EXPORT_SYMBOL_GPL(nfs_free_inode);
2548
2549static inline void nfs4_init_once(struct nfs_inode *nfsi)
2550{
2551#if IS_ENABLED(CONFIG_NFS_V4)
2552	INIT_LIST_HEAD(&nfsi->open_states);
2553	nfsi->delegation = NULL;
2554	init_rwsem(&nfsi->rwsem);
2555	nfsi->layout = NULL;
2556#endif
2557}
2558
2559static void init_once(void *foo)
2560{
2561	struct nfs_inode *nfsi = foo;
2562
2563	inode_init_once(&nfsi->vfs_inode);
2564	INIT_LIST_HEAD(&nfsi->open_files);
2565	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2566	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2567	nfs4_init_once(nfsi);
2568}
2569
2570static int __init nfs_init_inodecache(void)
2571{
2572	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2573					     sizeof(struct nfs_inode),
2574					     0, (SLAB_RECLAIM_ACCOUNT|
2575						SLAB_ACCOUNT),
2576					     init_once);
2577	if (nfs_inode_cachep == NULL)
2578		return -ENOMEM;
2579
2580	return 0;
2581}
2582
2583static void nfs_destroy_inodecache(void)
2584{
2585	/*
2586	 * Make sure all delayed rcu free inodes are flushed before we
2587	 * destroy cache.
2588	 */
2589	rcu_barrier();
2590	kmem_cache_destroy(nfs_inode_cachep);
2591}
2592
2593struct workqueue_struct *nfslocaliod_workqueue;
2594struct workqueue_struct *nfsiod_workqueue;
2595EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2596
2597/*
2598 * Destroy the nfsiod workqueues
2599 */
2600static void nfsiod_stop(void)
2601{
2602	struct workqueue_struct *wq;
2603
2604	wq = nfsiod_workqueue;
2605	if (wq != NULL) {
2606		nfsiod_workqueue = NULL;
2607		destroy_workqueue(wq);
2608	}
2609#if IS_ENABLED(CONFIG_NFS_LOCALIO)
2610	wq = nfslocaliod_workqueue;
2611	if (wq != NULL) {
2612		nfslocaliod_workqueue = NULL;
2613		destroy_workqueue(wq);
2614	}
2615#endif /* CONFIG_NFS_LOCALIO */
2616}
2617
2618/*
2619 * Start the nfsiod workqueues
2620 */
2621static int nfsiod_start(void)
2622{
2623	dprintk("RPC:       creating workqueue nfsiod\n");
2624	nfsiod_workqueue = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2625	if (nfsiod_workqueue == NULL)
2626		return -ENOMEM;
2627#if IS_ENABLED(CONFIG_NFS_LOCALIO)
2628	/*
2629	 * localio writes need to use a normal (non-memreclaim) workqueue.
2630	 * When we start getting low on space, XFS goes and calls flush_work() on
2631	 * a non-memreclaim work queue, which causes a priority inversion problem.
2632	 */
2633	dprintk("RPC:       creating workqueue nfslocaliod\n");
2634	nfslocaliod_workqueue = alloc_workqueue("nfslocaliod", WQ_UNBOUND, 0);
2635	if (unlikely(nfslocaliod_workqueue == NULL)) {
2636		nfsiod_stop();
2637		return -ENOMEM;
2638	}
2639#endif /* CONFIG_NFS_LOCALIO */
2640	return 0;
2641}
2642
2643unsigned int nfs_net_id;
2644EXPORT_SYMBOL_GPL(nfs_net_id);
2645
2646static int nfs_net_init(struct net *net)
2647{
2648	struct nfs_net *nn = net_generic(net, nfs_net_id);
2649	int err;
2650
2651	nfs_clients_init(net);
2652
2653	if (!rpc_proc_register(net, &nn->rpcstats)) {
2654		err = -ENOMEM;
2655		goto err_proc_rpc;
2656	}
2657
2658	err = nfs_fs_proc_net_init(net);
2659	if (err)
2660		goto err_proc_nfs;
2661
2662	return 0;
2663
2664err_proc_nfs:
2665	rpc_proc_unregister(net, "nfs");
2666err_proc_rpc:
2667	nfs_clients_exit(net);
2668	return err;
2669}
2670
2671static void nfs_net_exit(struct net *net)
2672{
2673	rpc_proc_unregister(net, "nfs");
2674	nfs_fs_proc_net_exit(net);
2675	nfs_clients_exit(net);
2676}
2677
2678static struct pernet_operations nfs_net_ops = {
2679	.init = nfs_net_init,
2680	.exit = nfs_net_exit,
2681	.id   = &nfs_net_id,
2682	.size = sizeof(struct nfs_net),
2683};
2684
2685#ifdef CONFIG_KEYS
2686static struct key *nfs_keyring;
2687
2688static int __init nfs_init_keyring(void)
2689{
2690	nfs_keyring = keyring_alloc(".nfs",
2691			     GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
2692			     current_cred(),
2693			     (KEY_POS_ALL & ~KEY_POS_SETATTR) |
2694			     (KEY_USR_ALL & ~KEY_USR_SETATTR),
2695			     KEY_ALLOC_NOT_IN_QUOTA, NULL, NULL);
2696	return PTR_ERR_OR_ZERO(nfs_keyring);
2697}
2698
2699static void nfs_exit_keyring(void)
2700{
2701	key_put(nfs_keyring);
2702}
2703#else
2704static inline int nfs_init_keyring(void)
2705{
2706	return 0;
2707}
2708
2709static inline void nfs_exit_keyring(void)
2710{
2711}
2712#endif /* CONFIG_KEYS */
2713
2714/*
2715 * Initialize NFS
2716 */
2717static int __init init_nfs_fs(void)
2718{
2719	int err;
2720
2721	err = nfs_init_keyring();
2722	if (err)
2723		return err;
2724
2725	err = nfs_sysfs_init();
2726	if (err < 0)
2727		goto out10;
2728
2729	err = register_pernet_subsys(&nfs_net_ops);
2730	if (err < 0)
2731		goto out9;
2732
2733	err = nfsiod_start();
2734	if (err)
2735		goto out7;
2736
2737	err = nfs_fs_proc_init();
2738	if (err)
2739		goto out6;
2740
2741	err = nfs_init_nfspagecache();
2742	if (err)
2743		goto out5;
2744
2745	err = nfs_init_inodecache();
2746	if (err)
2747		goto out4;
2748
2749	err = nfs_init_readpagecache();
2750	if (err)
2751		goto out3;
2752
2753	err = nfs_init_writepagecache();
2754	if (err)
2755		goto out2;
2756
2757	err = nfs_init_directcache();
2758	if (err)
2759		goto out1;
2760
2761	err = register_nfs_fs();
2762	if (err)
2763		goto out0;
2764
2765	return 0;
2766out0:
2767	nfs_destroy_directcache();
2768out1:
2769	nfs_destroy_writepagecache();
2770out2:
2771	nfs_destroy_readpagecache();
2772out3:
2773	nfs_destroy_inodecache();
2774out4:
2775	nfs_destroy_nfspagecache();
2776out5:
2777	nfs_fs_proc_exit();
2778out6:
2779	nfsiod_stop();
2780out7:
2781	unregister_pernet_subsys(&nfs_net_ops);
2782out9:
2783	nfs_sysfs_exit();
2784out10:
2785	nfs_exit_keyring();
2786	return err;
2787}
2788
2789static void __exit exit_nfs_fs(void)
2790{
2791	nfs_destroy_directcache();
2792	nfs_destroy_writepagecache();
2793	nfs_destroy_readpagecache();
2794	nfs_destroy_inodecache();
2795	nfs_destroy_nfspagecache();
2796	unregister_pernet_subsys(&nfs_net_ops);
2797	unregister_nfs_fs();
2798	nfs_fs_proc_exit();
2799	nfsiod_stop();
2800	nfs_sysfs_exit();
2801	nfs_exit_keyring();
2802}
2803
2804/* Not quite true; I just maintain it */
2805MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2806MODULE_DESCRIPTION("NFS client support");
2807MODULE_LICENSE("GPL");
2808module_param(enable_ino64, bool, 0644);
2809
2810module_init(init_nfs_fs)
2811module_exit(exit_nfs_fs)