1/*
2 *  linux/fs/nfs/inode.c
3 *
4 *  Copyright (C) 1992  Rick Sladkey
5 *
6 *  nfs inode and superblock handling functions
7 *
8 *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9 *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
10 *
11 *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 *  J.S.Peatfield@damtp.cam.ac.uk
13 *
14 */
15
16#include <linux/module.h>
17#include <linux/init.h>
18#include <linux/sched.h>
19#include <linux/time.h>
20#include <linux/kernel.h>
21#include <linux/mm.h>
22#include <linux/string.h>
23#include <linux/stat.h>
24#include <linux/errno.h>
25#include <linux/unistd.h>
26#include <linux/sunrpc/clnt.h>
27#include <linux/sunrpc/stats.h>
28#include <linux/sunrpc/metrics.h>
29#include <linux/nfs_fs.h>
30#include <linux/nfs_mount.h>
31#include <linux/nfs4_mount.h>
32#include <linux/lockd/bind.h>
33#include <linux/seq_file.h>
34#include <linux/mount.h>
35#include <linux/vfs.h>
36#include <linux/inet.h>
37#include <linux/nfs_xdr.h>
38#include <linux/slab.h>
39#include <linux/compat.h>
40#include <linux/freezer.h>
41
42#include <asm/uaccess.h>
43
44#include "nfs4_fs.h"
45#include "callback.h"
46#include "delegation.h"
47#include "iostat.h"
48#include "internal.h"
49#include "fscache.h"
50#include "pnfs.h"
51#include "nfs.h"
52#include "netns.h"
53
54#include "nfstrace.h"
55
56#define NFSDBG_FACILITY		NFSDBG_VFS
57
58#define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
59
60/* Default is to see 64-bit inode numbers */
61static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62
63static void nfs_invalidate_inode(struct inode *);
64static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65
66static struct kmem_cache * nfs_inode_cachep;
67
68static inline unsigned long
69nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70{
71	return nfs_fileid_to_ino_t(fattr->fileid);
72}
73
74/**
75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76 * @word: long word containing the bit lock
77 */
78int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
79{
80	freezable_schedule_unsafe();
81	if (signal_pending_state(mode, current))
82		return -ERESTARTSYS;
83	return 0;
84}
85EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
86
87/**
88 * nfs_compat_user_ino64 - returns the user-visible inode number
89 * @fileid: 64-bit fileid
90 *
91 * This function returns a 32-bit inode number if the boot parameter
92 * nfs.enable_ino64 is zero.
93 */
94u64 nfs_compat_user_ino64(u64 fileid)
95{
96#ifdef CONFIG_COMPAT
97	compat_ulong_t ino;
98#else
99	unsigned long ino;
100#endif
101
102	if (enable_ino64)
103		return fileid;
104	ino = fileid;
105	if (sizeof(ino) < sizeof(fileid))
106		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
107	return ino;
108}
109
110int nfs_drop_inode(struct inode *inode)
111{
112	return NFS_STALE(inode) || generic_drop_inode(inode);
113}
114EXPORT_SYMBOL_GPL(nfs_drop_inode);
115
116void nfs_clear_inode(struct inode *inode)
117{
118	/*
119	 * The following should never happen...
120	 */
121	WARN_ON_ONCE(nfs_have_writebacks(inode));
122	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
123	nfs_zap_acl_cache(inode);
124	nfs_access_zap_cache(inode);
125	nfs_fscache_clear_inode(inode);
126}
127EXPORT_SYMBOL_GPL(nfs_clear_inode);
128
129void nfs_evict_inode(struct inode *inode)
130{
131	truncate_inode_pages_final(&inode->i_data);
132	clear_inode(inode);
133	nfs_clear_inode(inode);
134}
135
136int nfs_sync_inode(struct inode *inode)
137{
138	nfs_inode_dio_wait(inode);
139	return nfs_wb_all(inode);
140}
141EXPORT_SYMBOL_GPL(nfs_sync_inode);
142
143/**
144 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
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 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
158{
159	struct nfs_inode *nfsi = NFS_I(inode);
160
161	if (inode->i_mapping->nrpages == 0)
162		flags &= ~NFS_INO_INVALID_DATA;
163	nfsi->cache_validity |= flags;
164	if (flags & NFS_INO_INVALID_DATA)
165		nfs_fscache_invalidate(inode);
166}
167
168/*
169 * Invalidate the local caches
170 */
171static void nfs_zap_caches_locked(struct inode *inode)
172{
173	struct nfs_inode *nfsi = NFS_I(inode);
174	int mode = inode->i_mode;
175
176	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
177
178	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
179	nfsi->attrtimeo_timestamp = jiffies;
180
181	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
182	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
183		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
184					| NFS_INO_INVALID_DATA
185					| NFS_INO_INVALID_ACCESS
186					| NFS_INO_INVALID_ACL
187					| NFS_INO_REVAL_PAGECACHE);
188	} else
189		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
190					| NFS_INO_INVALID_ACCESS
191					| NFS_INO_INVALID_ACL
192					| NFS_INO_REVAL_PAGECACHE);
193	nfs_zap_label_cache_locked(nfsi);
194}
195
196void nfs_zap_caches(struct inode *inode)
197{
198	spin_lock(&inode->i_lock);
199	nfs_zap_caches_locked(inode);
200	spin_unlock(&inode->i_lock);
201}
202
203void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
204{
205	if (mapping->nrpages != 0) {
206		spin_lock(&inode->i_lock);
207		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
208		spin_unlock(&inode->i_lock);
209	}
210}
211
212void nfs_zap_acl_cache(struct inode *inode)
213{
214	void (*clear_acl_cache)(struct inode *);
215
216	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
217	if (clear_acl_cache != NULL)
218		clear_acl_cache(inode);
219	spin_lock(&inode->i_lock);
220	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
221	spin_unlock(&inode->i_lock);
222}
223EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
224
225void nfs_invalidate_atime(struct inode *inode)
226{
227	spin_lock(&inode->i_lock);
228	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
229	spin_unlock(&inode->i_lock);
230}
231EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
232
233/*
234 * Invalidate, but do not unhash, the inode.
235 * NB: must be called with inode->i_lock held!
236 */
237static void nfs_invalidate_inode(struct inode *inode)
238{
239	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
240	nfs_zap_caches_locked(inode);
241}
242
243struct nfs_find_desc {
244	struct nfs_fh		*fh;
245	struct nfs_fattr	*fattr;
246};
247
248/*
249 * In NFSv3 we can have 64bit inode numbers. In order to support
250 * this, and re-exported directories (also seen in NFSv2)
251 * we are forced to allow 2 different inodes to have the same
252 * i_ino.
253 */
254static int
255nfs_find_actor(struct inode *inode, void *opaque)
256{
257	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
258	struct nfs_fh		*fh = desc->fh;
259	struct nfs_fattr	*fattr = desc->fattr;
260
261	if (NFS_FILEID(inode) != fattr->fileid)
262		return 0;
263	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
264		return 0;
265	if (nfs_compare_fh(NFS_FH(inode), fh))
266		return 0;
267	if (is_bad_inode(inode) || NFS_STALE(inode))
268		return 0;
269	return 1;
270}
271
272static int
273nfs_init_locked(struct inode *inode, void *opaque)
274{
275	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
276	struct nfs_fattr	*fattr = desc->fattr;
277
278	set_nfs_fileid(inode, fattr->fileid);
279	nfs_copy_fh(NFS_FH(inode), desc->fh);
280	return 0;
281}
282
283#ifdef CONFIG_NFS_V4_SECURITY_LABEL
284static void nfs_clear_label_invalid(struct inode *inode)
285{
286	spin_lock(&inode->i_lock);
287	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
288	spin_unlock(&inode->i_lock);
289}
290
291void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
292					struct nfs4_label *label)
293{
294	int error;
295
296	if (label == NULL)
297		return;
298
299	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
300		error = security_inode_notifysecctx(inode, label->label,
301				label->len);
302		if (error)
303			printk(KERN_ERR "%s() %s %d "
304					"security_inode_notifysecctx() %d\n",
305					__func__,
306					(char *)label->label,
307					label->len, error);
308		nfs_clear_label_invalid(inode);
309	}
310}
311
312struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
313{
314	struct nfs4_label *label = NULL;
315	int minor_version = server->nfs_client->cl_minorversion;
316
317	if (minor_version < 2)
318		return label;
319
320	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
321		return label;
322
323	label = kzalloc(sizeof(struct nfs4_label), flags);
324	if (label == NULL)
325		return ERR_PTR(-ENOMEM);
326
327	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
328	if (label->label == NULL) {
329		kfree(label);
330		return ERR_PTR(-ENOMEM);
331	}
332	label->len = NFS4_MAXLABELLEN;
333
334	return label;
335}
336EXPORT_SYMBOL_GPL(nfs4_label_alloc);
337#else
338void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
339					struct nfs4_label *label)
340{
341}
342#endif
343EXPORT_SYMBOL_GPL(nfs_setsecurity);
344
345/*
346 * This is our front-end to iget that looks up inodes by file handle
347 * instead of inode number.
348 */
349struct inode *
350nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
351{
352	struct nfs_find_desc desc = {
353		.fh	= fh,
354		.fattr	= fattr
355	};
356	struct inode *inode = ERR_PTR(-ENOENT);
357	unsigned long hash;
358
359	nfs_attr_check_mountpoint(sb, fattr);
360
361	if (nfs_attr_use_mounted_on_fileid(fattr))
362		fattr->fileid = fattr->mounted_on_fileid;
363	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
364		goto out_no_inode;
365	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
366		goto out_no_inode;
367
368	hash = nfs_fattr_to_ino_t(fattr);
369
370	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
371	if (inode == NULL) {
372		inode = ERR_PTR(-ENOMEM);
373		goto out_no_inode;
374	}
375
376	if (inode->i_state & I_NEW) {
377		struct nfs_inode *nfsi = NFS_I(inode);
378		unsigned long now = jiffies;
379
380		/* We set i_ino for the few things that still rely on it,
381		 * such as stat(2) */
382		inode->i_ino = hash;
383
384		/* We can't support update_atime(), since the server will reset it */
385		inode->i_flags |= S_NOATIME|S_NOCMTIME;
386		inode->i_mode = fattr->mode;
387		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
388				&& nfs_server_capable(inode, NFS_CAP_MODE))
389			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
390		/* Why so? Because we want revalidate for devices/FIFOs, and
391		 * that's precisely what we have in nfs_file_inode_operations.
392		 */
393		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
394		if (S_ISREG(inode->i_mode)) {
395			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
396			inode->i_data.a_ops = &nfs_file_aops;
397		} else if (S_ISDIR(inode->i_mode)) {
398			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
399			inode->i_fop = &nfs_dir_operations;
400			inode->i_data.a_ops = &nfs_dir_aops;
401			/* Deal with crossing mountpoints */
402			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
403					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
404				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
405					inode->i_op = &nfs_referral_inode_operations;
406				else
407					inode->i_op = &nfs_mountpoint_inode_operations;
408				inode->i_fop = NULL;
409				inode->i_flags |= S_AUTOMOUNT;
410			}
411		} else if (S_ISLNK(inode->i_mode))
412			inode->i_op = &nfs_symlink_inode_operations;
413		else
414			init_special_inode(inode, inode->i_mode, fattr->rdev);
415
416		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
417		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
418		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
419		inode->i_version = 0;
420		inode->i_size = 0;
421		clear_nlink(inode);
422		inode->i_uid = make_kuid(&init_user_ns, -2);
423		inode->i_gid = make_kgid(&init_user_ns, -2);
424		inode->i_blocks = 0;
425		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
426		nfsi->write_io = 0;
427		nfsi->read_io = 0;
428
429		nfsi->read_cache_jiffies = fattr->time_start;
430		nfsi->attr_gencount = fattr->gencount;
431		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
432			inode->i_atime = fattr->atime;
433		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
434			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
435		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
436			inode->i_mtime = fattr->mtime;
437		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
438			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
439		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
440			inode->i_ctime = fattr->ctime;
441		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
442			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
443		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
444			inode->i_version = fattr->change_attr;
445		else
446			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
447				| NFS_INO_REVAL_PAGECACHE);
448		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
449			inode->i_size = nfs_size_to_loff_t(fattr->size);
450		else
451			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
452				| NFS_INO_REVAL_PAGECACHE);
453		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
454			set_nlink(inode, fattr->nlink);
455		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
456			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
457		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
458			inode->i_uid = fattr->uid;
459		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
460			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
461		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
462			inode->i_gid = fattr->gid;
463		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
464			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
465		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
466			inode->i_blocks = fattr->du.nfs2.blocks;
467		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
468			/*
469			 * report the blocks in 512byte units
470			 */
471			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
472		}
473
474		nfs_setsecurity(inode, fattr, label);
475
476		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
477		nfsi->attrtimeo_timestamp = now;
478		nfsi->access_cache = RB_ROOT;
479
480		nfs_fscache_init_inode(inode);
481
482		unlock_new_inode(inode);
483	} else
484		nfs_refresh_inode(inode, fattr);
485	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
486		inode->i_sb->s_id,
487		(unsigned long long)NFS_FILEID(inode),
488		nfs_display_fhandle_hash(fh),
489		atomic_read(&inode->i_count));
490
491out:
492	return inode;
493
494out_no_inode:
495	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
496	goto out;
497}
498EXPORT_SYMBOL_GPL(nfs_fhget);
499
500#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)
501
502int
503nfs_setattr(struct dentry *dentry, struct iattr *attr)
504{
505	struct inode *inode = d_inode(dentry);
506	struct nfs_fattr *fattr;
507	int error = 0;
508
509	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
510
511	/* skip mode change if it's just for clearing setuid/setgid */
512	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
513		attr->ia_valid &= ~ATTR_MODE;
514
515	if (attr->ia_valid & ATTR_SIZE) {
516		BUG_ON(!S_ISREG(inode->i_mode));
517
518		error = inode_newsize_ok(inode, attr->ia_size);
519		if (error)
520			return error;
521
522		if (attr->ia_size == i_size_read(inode))
523			attr->ia_valid &= ~ATTR_SIZE;
524	}
525
526	/* Optimization: if the end result is no change, don't RPC */
527	attr->ia_valid &= NFS_VALID_ATTRS;
528	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
529		return 0;
530
531	trace_nfs_setattr_enter(inode);
532
533	/* Write all dirty data */
534	if (S_ISREG(inode->i_mode))
535		nfs_sync_inode(inode);
536
537	fattr = nfs_alloc_fattr();
538	if (fattr == NULL) {
539		error = -ENOMEM;
540		goto out;
541	}
542
543	/*
544	 * Return any delegations if we're going to change ACLs
545	 */
546	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
547		NFS_PROTO(inode)->return_delegation(inode);
548	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
549	if (error == 0)
550		error = nfs_refresh_inode(inode, fattr);
551	nfs_free_fattr(fattr);
552out:
553	trace_nfs_setattr_exit(inode, error);
554	return error;
555}
556EXPORT_SYMBOL_GPL(nfs_setattr);
557
558/**
559 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
560 * @inode: inode of the file used
561 * @offset: file offset to start truncating
562 *
563 * This is a copy of the common vmtruncate, but with the locking
564 * corrected to take into account the fact that NFS requires
565 * inode->i_size to be updated under the inode->i_lock.
566 * Note: must be called with inode->i_lock held!
567 */
568static int nfs_vmtruncate(struct inode * inode, loff_t offset)
569{
570	int err;
571
572	err = inode_newsize_ok(inode, offset);
573	if (err)
574		goto out;
575
576	i_size_write(inode, offset);
577	/* Optimisation */
578	if (offset == 0)
579		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
580
581	spin_unlock(&inode->i_lock);
582	truncate_pagecache(inode, offset);
583	spin_lock(&inode->i_lock);
584out:
585	return err;
586}
587
588/**
589 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
590 * @inode: pointer to struct inode
591 * @attr: pointer to struct iattr
592 *
593 * Note: we do this in the *proc.c in order to ensure that
594 *       it works for things like exclusive creates too.
595 */
596void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
597		struct nfs_fattr *fattr)
598{
599	/* Barrier: bump the attribute generation count. */
600	nfs_fattr_set_barrier(fattr);
601
602	spin_lock(&inode->i_lock);
603	NFS_I(inode)->attr_gencount = fattr->gencount;
604	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
605		if ((attr->ia_valid & ATTR_MODE) != 0) {
606			int mode = attr->ia_mode & S_IALLUGO;
607			mode |= inode->i_mode & ~S_IALLUGO;
608			inode->i_mode = mode;
609		}
610		if ((attr->ia_valid & ATTR_UID) != 0)
611			inode->i_uid = attr->ia_uid;
612		if ((attr->ia_valid & ATTR_GID) != 0)
613			inode->i_gid = attr->ia_gid;
614		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
615				| NFS_INO_INVALID_ACL);
616	}
617	if ((attr->ia_valid & ATTR_SIZE) != 0) {
618		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
619		nfs_vmtruncate(inode, attr->ia_size);
620	}
621	if (fattr->valid)
622		nfs_update_inode(inode, fattr);
623	else
624		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
625	spin_unlock(&inode->i_lock);
626}
627EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
628
629static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
630{
631	struct dentry *parent;
632
633	parent = dget_parent(dentry);
634	nfs_force_use_readdirplus(d_inode(parent));
635	dput(parent);
636}
637
638static bool nfs_need_revalidate_inode(struct inode *inode)
639{
640	if (NFS_I(inode)->cache_validity &
641			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
642		return true;
643	if (nfs_attribute_cache_expired(inode))
644		return true;
645	return false;
646}
647
648int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
649{
650	struct inode *inode = d_inode(dentry);
651	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
652	int err = 0;
653
654	trace_nfs_getattr_enter(inode);
655	/* Flush out writes to the server in order to update c/mtime.  */
656	if (S_ISREG(inode->i_mode)) {
657		mutex_lock(&inode->i_mutex);
658		err = nfs_sync_inode(inode);
659		mutex_unlock(&inode->i_mutex);
660		if (err)
661			goto out;
662	}
663
664	/*
665	 * We may force a getattr if the user cares about atime.
666	 *
667	 * Note that we only have to check the vfsmount flags here:
668	 *  - NFS always sets S_NOATIME by so checking it would give a
669	 *    bogus result
670	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
671	 *    no point in checking those.
672	 */
673 	if ((mnt->mnt_flags & MNT_NOATIME) ||
674 	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
675		need_atime = 0;
676
677	if (need_atime || nfs_need_revalidate_inode(inode)) {
678		struct nfs_server *server = NFS_SERVER(inode);
679
680		if (server->caps & NFS_CAP_READDIRPLUS)
681			nfs_request_parent_use_readdirplus(dentry);
682		err = __nfs_revalidate_inode(server, inode);
683	}
684	if (!err) {
685		generic_fillattr(inode, stat);
686		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
687		if (S_ISDIR(inode->i_mode))
688			stat->blksize = NFS_SERVER(inode)->dtsize;
689	}
690out:
691	trace_nfs_getattr_exit(inode, err);
692	return err;
693}
694EXPORT_SYMBOL_GPL(nfs_getattr);
695
696static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
697{
698	atomic_set(&l_ctx->count, 1);
699	l_ctx->lockowner.l_owner = current->files;
700	l_ctx->lockowner.l_pid = current->tgid;
701	INIT_LIST_HEAD(&l_ctx->list);
702	nfs_iocounter_init(&l_ctx->io_count);
703}
704
705static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
706{
707	struct nfs_lock_context *head = &ctx->lock_context;
708	struct nfs_lock_context *pos = head;
709
710	do {
711		if (pos->lockowner.l_owner != current->files)
712			continue;
713		if (pos->lockowner.l_pid != current->tgid)
714			continue;
715		atomic_inc(&pos->count);
716		return pos;
717	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
718	return NULL;
719}
720
721struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
722{
723	struct nfs_lock_context *res, *new = NULL;
724	struct inode *inode = d_inode(ctx->dentry);
725
726	spin_lock(&inode->i_lock);
727	res = __nfs_find_lock_context(ctx);
728	if (res == NULL) {
729		spin_unlock(&inode->i_lock);
730		new = kmalloc(sizeof(*new), GFP_KERNEL);
731		if (new == NULL)
732			return ERR_PTR(-ENOMEM);
733		nfs_init_lock_context(new);
734		spin_lock(&inode->i_lock);
735		res = __nfs_find_lock_context(ctx);
736		if (res == NULL) {
737			list_add_tail(&new->list, &ctx->lock_context.list);
738			new->open_context = ctx;
739			res = new;
740			new = NULL;
741		}
742	}
743	spin_unlock(&inode->i_lock);
744	kfree(new);
745	return res;
746}
747EXPORT_SYMBOL_GPL(nfs_get_lock_context);
748
749void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
750{
751	struct nfs_open_context *ctx = l_ctx->open_context;
752	struct inode *inode = d_inode(ctx->dentry);
753
754	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
755		return;
756	list_del(&l_ctx->list);
757	spin_unlock(&inode->i_lock);
758	kfree(l_ctx);
759}
760EXPORT_SYMBOL_GPL(nfs_put_lock_context);
761
762/**
763 * nfs_close_context - Common close_context() routine NFSv2/v3
764 * @ctx: pointer to context
765 * @is_sync: is this a synchronous close
766 *
767 * Ensure that the attributes are up to date if we're mounted
768 * with close-to-open semantics and we have cached data that will
769 * need to be revalidated on open.
770 */
771void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
772{
773	struct nfs_inode *nfsi;
774	struct inode *inode;
775	struct nfs_server *server;
776
777	if (!(ctx->mode & FMODE_WRITE))
778		return;
779	if (!is_sync)
780		return;
781	inode = d_inode(ctx->dentry);
782	nfsi = NFS_I(inode);
783	if (inode->i_mapping->nrpages == 0)
784		return;
785	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
786		return;
787	if (!list_empty(&nfsi->open_files))
788		return;
789	server = NFS_SERVER(inode);
790	if (server->flags & NFS_MOUNT_NOCTO)
791		return;
792	nfs_revalidate_inode(server, inode);
793}
794EXPORT_SYMBOL_GPL(nfs_close_context);
795
796struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
797{
798	struct nfs_open_context *ctx;
799	struct rpc_cred *cred = rpc_lookup_cred();
800	if (IS_ERR(cred))
801		return ERR_CAST(cred);
802
803	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
804	if (!ctx) {
805		put_rpccred(cred);
806		return ERR_PTR(-ENOMEM);
807	}
808	nfs_sb_active(dentry->d_sb);
809	ctx->dentry = dget(dentry);
810	ctx->cred = cred;
811	ctx->state = NULL;
812	ctx->mode = f_mode;
813	ctx->flags = 0;
814	ctx->error = 0;
815	nfs_init_lock_context(&ctx->lock_context);
816	ctx->lock_context.open_context = ctx;
817	INIT_LIST_HEAD(&ctx->list);
818	ctx->mdsthreshold = NULL;
819	return ctx;
820}
821EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
822
823struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
824{
825	if (ctx != NULL)
826		atomic_inc(&ctx->lock_context.count);
827	return ctx;
828}
829EXPORT_SYMBOL_GPL(get_nfs_open_context);
830
831static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
832{
833	struct inode *inode = d_inode(ctx->dentry);
834	struct super_block *sb = ctx->dentry->d_sb;
835
836	if (!list_empty(&ctx->list)) {
837		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
838			return;
839		list_del(&ctx->list);
840		spin_unlock(&inode->i_lock);
841	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
842		return;
843	if (inode != NULL)
844		NFS_PROTO(inode)->close_context(ctx, is_sync);
845	if (ctx->cred != NULL)
846		put_rpccred(ctx->cred);
847	dput(ctx->dentry);
848	nfs_sb_deactive(sb);
849	kfree(ctx->mdsthreshold);
850	kfree(ctx);
851}
852
853void put_nfs_open_context(struct nfs_open_context *ctx)
854{
855	__put_nfs_open_context(ctx, 0);
856}
857EXPORT_SYMBOL_GPL(put_nfs_open_context);
858
859static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
860{
861	__put_nfs_open_context(ctx, 1);
862}
863
864/*
865 * Ensure that mmap has a recent RPC credential for use when writing out
866 * shared pages
867 */
868void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
869{
870	struct inode *inode = d_inode(ctx->dentry);
871	struct nfs_inode *nfsi = NFS_I(inode);
872
873	spin_lock(&inode->i_lock);
874	list_add(&ctx->list, &nfsi->open_files);
875	spin_unlock(&inode->i_lock);
876}
877EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
878
879void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
880{
881	filp->private_data = get_nfs_open_context(ctx);
882	if (list_empty(&ctx->list))
883		nfs_inode_attach_open_context(ctx);
884}
885EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
886
887/*
888 * Given an inode, search for an open context with the desired characteristics
889 */
890struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
891{
892	struct nfs_inode *nfsi = NFS_I(inode);
893	struct nfs_open_context *pos, *ctx = NULL;
894
895	spin_lock(&inode->i_lock);
896	list_for_each_entry(pos, &nfsi->open_files, list) {
897		if (cred != NULL && pos->cred != cred)
898			continue;
899		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
900			continue;
901		ctx = get_nfs_open_context(pos);
902		break;
903	}
904	spin_unlock(&inode->i_lock);
905	return ctx;
906}
907
908void nfs_file_clear_open_context(struct file *filp)
909{
910	struct nfs_open_context *ctx = nfs_file_open_context(filp);
911
912	if (ctx) {
913		struct inode *inode = d_inode(ctx->dentry);
914
915		filp->private_data = NULL;
916		spin_lock(&inode->i_lock);
917		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
918		spin_unlock(&inode->i_lock);
919		put_nfs_open_context_sync(ctx);
920	}
921}
922
923/*
924 * These allocate and release file read/write context information.
925 */
926int nfs_open(struct inode *inode, struct file *filp)
927{
928	struct nfs_open_context *ctx;
929
930	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode);
931	if (IS_ERR(ctx))
932		return PTR_ERR(ctx);
933	nfs_file_set_open_context(filp, ctx);
934	put_nfs_open_context(ctx);
935	nfs_fscache_open_file(inode, filp);
936	return 0;
937}
938
939/*
940 * This function is called whenever some part of NFS notices that
941 * the cached attributes have to be refreshed.
942 */
943int
944__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
945{
946	int		 status = -ESTALE;
947	struct nfs4_label *label = NULL;
948	struct nfs_fattr *fattr = NULL;
949	struct nfs_inode *nfsi = NFS_I(inode);
950
951	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
952		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
953
954	trace_nfs_revalidate_inode_enter(inode);
955
956	if (is_bad_inode(inode))
957		goto out;
958	if (NFS_STALE(inode))
959		goto out;
960
961	status = -ENOMEM;
962	fattr = nfs_alloc_fattr();
963	if (fattr == NULL)
964		goto out;
965
966	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
967
968	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
969	if (IS_ERR(label)) {
970		status = PTR_ERR(label);
971		goto out;
972	}
973
974	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
975	if (status != 0) {
976		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
977			 inode->i_sb->s_id,
978			 (unsigned long long)NFS_FILEID(inode), status);
979		if (status == -ESTALE) {
980			nfs_zap_caches(inode);
981			if (!S_ISDIR(inode->i_mode))
982				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
983		}
984		goto err_out;
985	}
986
987	status = nfs_refresh_inode(inode, fattr);
988	if (status) {
989		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
990			 inode->i_sb->s_id,
991			 (unsigned long long)NFS_FILEID(inode), status);
992		goto err_out;
993	}
994
995	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
996		nfs_zap_acl_cache(inode);
997
998	nfs_setsecurity(inode, fattr, label);
999
1000	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1001		inode->i_sb->s_id,
1002		(unsigned long long)NFS_FILEID(inode));
1003
1004err_out:
1005	nfs4_label_free(label);
1006out:
1007	nfs_free_fattr(fattr);
1008	trace_nfs_revalidate_inode_exit(inode, status);
1009	return status;
1010}
1011
1012int nfs_attribute_timeout(struct inode *inode)
1013{
1014	struct nfs_inode *nfsi = NFS_I(inode);
1015
1016	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1017}
1018
1019int nfs_attribute_cache_expired(struct inode *inode)
1020{
1021	if (nfs_have_delegated_attributes(inode))
1022		return 0;
1023	return nfs_attribute_timeout(inode);
1024}
1025
1026/**
1027 * nfs_revalidate_inode - Revalidate the inode attributes
1028 * @server - pointer to nfs_server struct
1029 * @inode - pointer to inode struct
1030 *
1031 * Updates inode attribute information by retrieving the data from the server.
1032 */
1033int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1034{
1035	if (!nfs_need_revalidate_inode(inode))
1036		return NFS_STALE(inode) ? -ESTALE : 0;
1037	return __nfs_revalidate_inode(server, inode);
1038}
1039EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1040
1041int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1042{
1043	if (!(NFS_I(inode)->cache_validity &
1044			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1045			&& !nfs_attribute_cache_expired(inode))
1046		return NFS_STALE(inode) ? -ESTALE : 0;
1047	return -ECHILD;
1048}
1049
1050static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1051{
1052	struct nfs_inode *nfsi = NFS_I(inode);
1053	int ret;
1054
1055	if (mapping->nrpages != 0) {
1056		if (S_ISREG(inode->i_mode)) {
1057			unmap_mapping_range(mapping, 0, 0, 0);
1058			ret = nfs_sync_mapping(mapping);
1059			if (ret < 0)
1060				return ret;
1061		}
1062		ret = invalidate_inode_pages2(mapping);
1063		if (ret < 0)
1064			return ret;
1065	}
1066	if (S_ISDIR(inode->i_mode)) {
1067		spin_lock(&inode->i_lock);
1068		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1069		spin_unlock(&inode->i_lock);
1070	}
1071	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1072	nfs_fscache_wait_on_invalidate(inode);
1073
1074	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1075			inode->i_sb->s_id,
1076			(unsigned long long)NFS_FILEID(inode));
1077	return 0;
1078}
1079
1080static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1081{
1082	if (nfs_have_delegated_attributes(inode))
1083		return false;
1084	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1085		|| nfs_attribute_timeout(inode)
1086		|| NFS_STALE(inode);
1087}
1088
1089/**
1090 * __nfs_revalidate_mapping - Revalidate the pagecache
1091 * @inode - pointer to host inode
1092 * @mapping - pointer to mapping
1093 * @may_lock - take inode->i_mutex?
1094 */
1095static int __nfs_revalidate_mapping(struct inode *inode,
1096		struct address_space *mapping,
1097		bool may_lock)
1098{
1099	struct nfs_inode *nfsi = NFS_I(inode);
1100	unsigned long *bitlock = &nfsi->flags;
1101	int ret = 0;
1102
1103	/* swapfiles are not supposed to be shared. */
1104	if (IS_SWAPFILE(inode))
1105		goto out;
1106
1107	if (nfs_mapping_need_revalidate_inode(inode)) {
1108		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1109		if (ret < 0)
1110			goto out;
1111	}
1112
1113	/*
1114	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1115	 * invalidations that come in while we're shooting down the mappings
1116	 * are respected. But, that leaves a race window where one revalidator
1117	 * can clear the flag, and then another checks it before the mapping
1118	 * gets invalidated. Fix that by serializing access to this part of
1119	 * the function.
1120	 *
1121	 * At the same time, we need to allow other tasks to see whether we
1122	 * might be in the middle of invalidating the pages, so we only set
1123	 * the bit lock here if it looks like we're going to be doing that.
1124	 */
1125	for (;;) {
1126		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1127					 nfs_wait_bit_killable, TASK_KILLABLE);
1128		if (ret)
1129			goto out;
1130		spin_lock(&inode->i_lock);
1131		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1132			spin_unlock(&inode->i_lock);
1133			continue;
1134		}
1135		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1136			break;
1137		spin_unlock(&inode->i_lock);
1138		goto out;
1139	}
1140
1141	set_bit(NFS_INO_INVALIDATING, bitlock);
1142	smp_wmb();
1143	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1144	spin_unlock(&inode->i_lock);
1145	trace_nfs_invalidate_mapping_enter(inode);
1146	if (may_lock) {
1147		mutex_lock(&inode->i_mutex);
1148		ret = nfs_invalidate_mapping(inode, mapping);
1149		mutex_unlock(&inode->i_mutex);
1150	} else
1151		ret = nfs_invalidate_mapping(inode, mapping);
1152	trace_nfs_invalidate_mapping_exit(inode, ret);
1153
1154	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1155	smp_mb__after_atomic();
1156	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1157out:
1158	return ret;
1159}
1160
1161/**
1162 * nfs_revalidate_mapping - Revalidate the pagecache
1163 * @inode - pointer to host inode
1164 * @mapping - pointer to mapping
1165 */
1166int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1167{
1168	return __nfs_revalidate_mapping(inode, mapping, false);
1169}
1170
1171/**
1172 * nfs_revalidate_mapping_protected - Revalidate the pagecache
1173 * @inode - pointer to host inode
1174 * @mapping - pointer to mapping
1175 *
1176 * Differs from nfs_revalidate_mapping() in that it grabs the inode->i_mutex
1177 * while invalidating the mapping.
1178 */
1179int nfs_revalidate_mapping_protected(struct inode *inode, struct address_space *mapping)
1180{
1181	return __nfs_revalidate_mapping(inode, mapping, true);
1182}
1183
1184static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1185{
1186	struct nfs_inode *nfsi = NFS_I(inode);
1187	unsigned long ret = 0;
1188
1189	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1190			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1191			&& inode->i_version == fattr->pre_change_attr) {
1192		inode->i_version = fattr->change_attr;
1193		if (S_ISDIR(inode->i_mode))
1194			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1195		ret |= NFS_INO_INVALID_ATTR;
1196	}
1197	/* If we have atomic WCC data, we may update some attributes */
1198	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1199			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1200			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1201		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1202		ret |= NFS_INO_INVALID_ATTR;
1203	}
1204
1205	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1206			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1207			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1208		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1209		if (S_ISDIR(inode->i_mode))
1210			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1211		ret |= NFS_INO_INVALID_ATTR;
1212	}
1213	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1214			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1215			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1216			&& nfsi->nrequests == 0) {
1217		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1218		ret |= NFS_INO_INVALID_ATTR;
1219	}
1220
1221	return ret;
1222}
1223
1224/**
1225 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1226 * @inode - pointer to inode
1227 * @fattr - updated attributes
1228 *
1229 * Verifies the attribute cache. If we have just changed the attributes,
1230 * so that fattr carries weak cache consistency data, then it may
1231 * also update the ctime/mtime/change_attribute.
1232 */
1233static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1234{
1235	struct nfs_inode *nfsi = NFS_I(inode);
1236	loff_t cur_size, new_isize;
1237	unsigned long invalid = 0;
1238
1239
1240	if (nfs_have_delegated_attributes(inode))
1241		return 0;
1242	/* Has the inode gone and changed behind our back? */
1243	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1244		return -EIO;
1245	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1246		return -EIO;
1247
1248	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1249			inode->i_version != fattr->change_attr)
1250		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1251
1252	/* Verify a few of the more important attributes */
1253	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1254		invalid |= NFS_INO_INVALID_ATTR;
1255
1256	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1257		cur_size = i_size_read(inode);
1258		new_isize = nfs_size_to_loff_t(fattr->size);
1259		if (cur_size != new_isize)
1260			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1261	}
1262	if (nfsi->nrequests != 0)
1263		invalid &= ~NFS_INO_REVAL_PAGECACHE;
1264
1265	/* Have any file permissions changed? */
1266	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1267		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1268	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1269		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1270	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1271		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1272
1273	/* Has the link count changed? */
1274	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1275		invalid |= NFS_INO_INVALID_ATTR;
1276
1277	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1278		invalid |= NFS_INO_INVALID_ATIME;
1279
1280	if (invalid != 0)
1281		nfs_set_cache_invalid(inode, invalid);
1282
1283	nfsi->read_cache_jiffies = fattr->time_start;
1284	return 0;
1285}
1286
1287static atomic_long_t nfs_attr_generation_counter;
1288
1289static unsigned long nfs_read_attr_generation_counter(void)
1290{
1291	return atomic_long_read(&nfs_attr_generation_counter);
1292}
1293
1294unsigned long nfs_inc_attr_generation_counter(void)
1295{
1296	return atomic_long_inc_return(&nfs_attr_generation_counter);
1297}
1298EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1299
1300void nfs_fattr_init(struct nfs_fattr *fattr)
1301{
1302	fattr->valid = 0;
1303	fattr->time_start = jiffies;
1304	fattr->gencount = nfs_inc_attr_generation_counter();
1305	fattr->owner_name = NULL;
1306	fattr->group_name = NULL;
1307}
1308EXPORT_SYMBOL_GPL(nfs_fattr_init);
1309
1310/**
1311 * nfs_fattr_set_barrier
1312 * @fattr: attributes
1313 *
1314 * Used to set a barrier after an attribute was updated. This
1315 * barrier ensures that older attributes from RPC calls that may
1316 * have raced with our update cannot clobber these new values.
1317 * Note that you are still responsible for ensuring that other
1318 * operations which change the attribute on the server do not
1319 * collide.
1320 */
1321void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1322{
1323	fattr->gencount = nfs_inc_attr_generation_counter();
1324}
1325
1326struct nfs_fattr *nfs_alloc_fattr(void)
1327{
1328	struct nfs_fattr *fattr;
1329
1330	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1331	if (fattr != NULL)
1332		nfs_fattr_init(fattr);
1333	return fattr;
1334}
1335EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1336
1337struct nfs_fh *nfs_alloc_fhandle(void)
1338{
1339	struct nfs_fh *fh;
1340
1341	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1342	if (fh != NULL)
1343		fh->size = 0;
1344	return fh;
1345}
1346EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1347
1348#ifdef NFS_DEBUG
1349/*
1350 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1351 *                             in the same way that wireshark does
1352 *
1353 * @fh: file handle
1354 *
1355 * For debugging only.
1356 */
1357u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1358{
1359	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1360	 * not on the result */
1361	return nfs_fhandle_hash(fh);
1362}
1363EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1364
1365/*
1366 * _nfs_display_fhandle - display an NFS file handle on the console
1367 *
1368 * @fh: file handle to display
1369 * @caption: display caption
1370 *
1371 * For debugging only.
1372 */
1373void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1374{
1375	unsigned short i;
1376
1377	if (fh == NULL || fh->size == 0) {
1378		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1379		return;
1380	}
1381
1382	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1383	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1384	for (i = 0; i < fh->size; i += 16) {
1385		__be32 *pos = (__be32 *)&fh->data[i];
1386
1387		switch ((fh->size - i - 1) >> 2) {
1388		case 0:
1389			printk(KERN_DEFAULT " %08x\n",
1390				be32_to_cpup(pos));
1391			break;
1392		case 1:
1393			printk(KERN_DEFAULT " %08x %08x\n",
1394				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1395			break;
1396		case 2:
1397			printk(KERN_DEFAULT " %08x %08x %08x\n",
1398				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1399				be32_to_cpup(pos + 2));
1400			break;
1401		default:
1402			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1403				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1404				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1405		}
1406	}
1407}
1408EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1409#endif
1410
1411/**
1412 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1413 * @inode - pointer to inode
1414 * @fattr - attributes
1415 *
1416 * Attempt to divine whether or not an RPC call reply carrying stale
1417 * attributes got scheduled after another call carrying updated ones.
1418 *
1419 * To do so, the function first assumes that a more recent ctime means
1420 * that the attributes in fattr are newer, however it also attempt to
1421 * catch the case where ctime either didn't change, or went backwards
1422 * (if someone reset the clock on the server) by looking at whether
1423 * or not this RPC call was started after the inode was last updated.
1424 * Note also the check for wraparound of 'attr_gencount'
1425 *
1426 * The function returns 'true' if it thinks the attributes in 'fattr' are
1427 * more recent than the ones cached in the inode.
1428 *
1429 */
1430static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1431{
1432	const struct nfs_inode *nfsi = NFS_I(inode);
1433
1434	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1435		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1436}
1437
1438/*
1439 * Don't trust the change_attribute, mtime, ctime or size if
1440 * a pnfs LAYOUTCOMMIT is outstanding
1441 */
1442static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1443		struct nfs_fattr *fattr)
1444{
1445	if (pnfs_layoutcommit_outstanding(inode))
1446		fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1447				NFS_ATTR_FATTR_MTIME |
1448				NFS_ATTR_FATTR_CTIME |
1449				NFS_ATTR_FATTR_SIZE);
1450}
1451
1452static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1453{
1454	int ret;
1455
1456	trace_nfs_refresh_inode_enter(inode);
1457
1458	nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1459
1460	if (nfs_inode_attrs_need_update(inode, fattr))
1461		ret = nfs_update_inode(inode, fattr);
1462	else
1463		ret = nfs_check_inode_attributes(inode, fattr);
1464
1465	trace_nfs_refresh_inode_exit(inode, ret);
1466	return ret;
1467}
1468
1469/**
1470 * nfs_refresh_inode - try to update the inode attribute cache
1471 * @inode - pointer to inode
1472 * @fattr - updated attributes
1473 *
1474 * Check that an RPC call that returned attributes has not overlapped with
1475 * other recent updates of the inode metadata, then decide whether it is
1476 * safe to do a full update of the inode attributes, or whether just to
1477 * call nfs_check_inode_attributes.
1478 */
1479int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1480{
1481	int status;
1482
1483	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1484		return 0;
1485	spin_lock(&inode->i_lock);
1486	status = nfs_refresh_inode_locked(inode, fattr);
1487	spin_unlock(&inode->i_lock);
1488
1489	return status;
1490}
1491EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1492
1493static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1494{
1495	unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1496
1497	/*
1498	 * Don't revalidate the pagecache if we hold a delegation, but do
1499	 * force an attribute update
1500	 */
1501	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1502		invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_FORCED;
1503
1504	if (S_ISDIR(inode->i_mode))
1505		invalid |= NFS_INO_INVALID_DATA;
1506	nfs_set_cache_invalid(inode, invalid);
1507	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1508		return 0;
1509	return nfs_refresh_inode_locked(inode, fattr);
1510}
1511
1512/**
1513 * nfs_post_op_update_inode - try to update the inode attribute cache
1514 * @inode - pointer to inode
1515 * @fattr - updated attributes
1516 *
1517 * After an operation that has changed the inode metadata, mark the
1518 * attribute cache as being invalid, then try to update it.
1519 *
1520 * NB: if the server didn't return any post op attributes, this
1521 * function will force the retrieval of attributes before the next
1522 * NFS request.  Thus it should be used only for operations that
1523 * are expected to change one or more attributes, to avoid
1524 * unnecessary NFS requests and trips through nfs_update_inode().
1525 */
1526int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1527{
1528	int status;
1529
1530	spin_lock(&inode->i_lock);
1531	nfs_fattr_set_barrier(fattr);
1532	status = nfs_post_op_update_inode_locked(inode, fattr);
1533	spin_unlock(&inode->i_lock);
1534
1535	return status;
1536}
1537EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1538
1539/**
1540 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1541 * @inode - pointer to inode
1542 * @fattr - updated attributes
1543 *
1544 * After an operation that has changed the inode metadata, mark the
1545 * attribute cache as being invalid, then try to update it. Fake up
1546 * weak cache consistency data, if none exist.
1547 *
1548 * This function is mainly designed to be used by the ->write_done() functions.
1549 */
1550int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1551{
1552	int status;
1553
1554	/* Don't do a WCC update if these attributes are already stale */
1555	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1556			!nfs_inode_attrs_need_update(inode, fattr)) {
1557		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1558				| NFS_ATTR_FATTR_PRESIZE
1559				| NFS_ATTR_FATTR_PREMTIME
1560				| NFS_ATTR_FATTR_PRECTIME);
1561		goto out_noforce;
1562	}
1563	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1564			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1565		fattr->pre_change_attr = inode->i_version;
1566		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1567	}
1568	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1569			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1570		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1571		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1572	}
1573	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1574			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1575		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1576		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1577	}
1578	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1579			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1580		fattr->pre_size = i_size_read(inode);
1581		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1582	}
1583out_noforce:
1584	status = nfs_post_op_update_inode_locked(inode, fattr);
1585	return status;
1586}
1587
1588/**
1589 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1590 * @inode - pointer to inode
1591 * @fattr - updated attributes
1592 *
1593 * After an operation that has changed the inode metadata, mark the
1594 * attribute cache as being invalid, then try to update it. Fake up
1595 * weak cache consistency data, if none exist.
1596 *
1597 * This function is mainly designed to be used by the ->write_done() functions.
1598 */
1599int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1600{
1601	int status;
1602
1603	spin_lock(&inode->i_lock);
1604	nfs_fattr_set_barrier(fattr);
1605	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1606	spin_unlock(&inode->i_lock);
1607	return status;
1608}
1609EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1610
1611
1612static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1613				    struct nfs_fattr *fattr)
1614{
1615	bool ret1 = true, ret2 = true;
1616
1617	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1618		ret1 = (nfsi->fileid == fattr->fileid);
1619	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1620		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1621	return ret1 || ret2;
1622}
1623
1624/*
1625 * Many nfs protocol calls return the new file attributes after
1626 * an operation.  Here we update the inode to reflect the state
1627 * of the server's inode.
1628 *
1629 * This is a bit tricky because we have to make sure all dirty pages
1630 * have been sent off to the server before calling invalidate_inode_pages.
1631 * To make sure no other process adds more write requests while we try
1632 * our best to flush them, we make them sleep during the attribute refresh.
1633 *
1634 * A very similar scenario holds for the dir cache.
1635 */
1636static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1637{
1638	struct nfs_server *server;
1639	struct nfs_inode *nfsi = NFS_I(inode);
1640	loff_t cur_isize, new_isize;
1641	unsigned long invalid = 0;
1642	unsigned long now = jiffies;
1643	unsigned long save_cache_validity;
1644	bool cache_revalidated = true;
1645
1646	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1647			__func__, inode->i_sb->s_id, inode->i_ino,
1648			nfs_display_fhandle_hash(NFS_FH(inode)),
1649			atomic_read(&inode->i_count), fattr->valid);
1650
1651	if (!nfs_fileid_valid(nfsi, fattr)) {
1652		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1653			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1654			NFS_SERVER(inode)->nfs_client->cl_hostname,
1655			inode->i_sb->s_id, (long long)nfsi->fileid,
1656			(long long)fattr->fileid);
1657		goto out_err;
1658	}
1659
1660	/*
1661	 * Make sure the inode's type hasn't changed.
1662	 */
1663	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1664		/*
1665		* Big trouble! The inode has become a different object.
1666		*/
1667		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1668				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1669		goto out_err;
1670	}
1671
1672	server = NFS_SERVER(inode);
1673	/* Update the fsid? */
1674	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1675			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1676			!IS_AUTOMOUNT(inode))
1677		server->fsid = fattr->fsid;
1678
1679	/*
1680	 * Update the read time so we don't revalidate too often.
1681	 */
1682	nfsi->read_cache_jiffies = fattr->time_start;
1683
1684	save_cache_validity = nfsi->cache_validity;
1685	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1686			| NFS_INO_INVALID_ATIME
1687			| NFS_INO_REVAL_FORCED
1688			| NFS_INO_REVAL_PAGECACHE);
1689
1690	/* Do atomic weak cache consistency updates */
1691	invalid |= nfs_wcc_update_inode(inode, fattr);
1692
1693	/* More cache consistency checks */
1694	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1695		if (inode->i_version != fattr->change_attr) {
1696			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1697					inode->i_sb->s_id, inode->i_ino);
1698			invalid |= NFS_INO_INVALID_ATTR
1699				| NFS_INO_INVALID_DATA
1700				| NFS_INO_INVALID_ACCESS
1701				| NFS_INO_INVALID_ACL;
1702			if (S_ISDIR(inode->i_mode))
1703				nfs_force_lookup_revalidate(inode);
1704			inode->i_version = fattr->change_attr;
1705		}
1706	} else {
1707		nfsi->cache_validity |= save_cache_validity;
1708		cache_revalidated = false;
1709	}
1710
1711	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1712		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1713	} else if (server->caps & NFS_CAP_MTIME) {
1714		nfsi->cache_validity |= save_cache_validity &
1715				(NFS_INO_INVALID_ATTR
1716				| NFS_INO_REVAL_FORCED);
1717		cache_revalidated = false;
1718	}
1719
1720	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1721		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1722	} else if (server->caps & NFS_CAP_CTIME) {
1723		nfsi->cache_validity |= save_cache_validity &
1724				(NFS_INO_INVALID_ATTR
1725				| NFS_INO_REVAL_FORCED);
1726		cache_revalidated = false;
1727	}
1728
1729	/* Check if our cached file size is stale */
1730	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1731		new_isize = nfs_size_to_loff_t(fattr->size);
1732		cur_isize = i_size_read(inode);
1733		if (new_isize != cur_isize) {
1734			/* Do we perhaps have any outstanding writes, or has
1735			 * the file grown beyond our last write? */
1736			if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1737				i_size_write(inode, new_isize);
1738				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1739			}
1740			dprintk("NFS: isize change on server for file %s/%ld "
1741					"(%Ld to %Ld)\n",
1742					inode->i_sb->s_id,
1743					inode->i_ino,
1744					(long long)cur_isize,
1745					(long long)new_isize);
1746		}
1747	} else {
1748		nfsi->cache_validity |= save_cache_validity &
1749				(NFS_INO_INVALID_ATTR
1750				| NFS_INO_REVAL_PAGECACHE
1751				| NFS_INO_REVAL_FORCED);
1752		cache_revalidated = false;
1753	}
1754
1755
1756	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1757		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1758	else if (server->caps & NFS_CAP_ATIME) {
1759		nfsi->cache_validity |= save_cache_validity &
1760				(NFS_INO_INVALID_ATIME
1761				| NFS_INO_REVAL_FORCED);
1762		cache_revalidated = false;
1763	}
1764
1765	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1766		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1767			umode_t newmode = inode->i_mode & S_IFMT;
1768			newmode |= fattr->mode & S_IALLUGO;
1769			inode->i_mode = newmode;
1770			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1771		}
1772	} else if (server->caps & NFS_CAP_MODE) {
1773		nfsi->cache_validity |= save_cache_validity &
1774				(NFS_INO_INVALID_ATTR
1775				| NFS_INO_INVALID_ACCESS
1776				| NFS_INO_INVALID_ACL
1777				| NFS_INO_REVAL_FORCED);
1778		cache_revalidated = false;
1779	}
1780
1781	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1782		if (!uid_eq(inode->i_uid, fattr->uid)) {
1783			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1784			inode->i_uid = fattr->uid;
1785		}
1786	} else if (server->caps & NFS_CAP_OWNER) {
1787		nfsi->cache_validity |= save_cache_validity &
1788				(NFS_INO_INVALID_ATTR
1789				| NFS_INO_INVALID_ACCESS
1790				| NFS_INO_INVALID_ACL
1791				| NFS_INO_REVAL_FORCED);
1792		cache_revalidated = false;
1793	}
1794
1795	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1796		if (!gid_eq(inode->i_gid, fattr->gid)) {
1797			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1798			inode->i_gid = fattr->gid;
1799		}
1800	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1801		nfsi->cache_validity |= save_cache_validity &
1802				(NFS_INO_INVALID_ATTR
1803				| NFS_INO_INVALID_ACCESS
1804				| NFS_INO_INVALID_ACL
1805				| NFS_INO_REVAL_FORCED);
1806		cache_revalidated = false;
1807	}
1808
1809	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1810		if (inode->i_nlink != fattr->nlink) {
1811			invalid |= NFS_INO_INVALID_ATTR;
1812			if (S_ISDIR(inode->i_mode))
1813				invalid |= NFS_INO_INVALID_DATA;
1814			set_nlink(inode, fattr->nlink);
1815		}
1816	} else if (server->caps & NFS_CAP_NLINK) {
1817		nfsi->cache_validity |= save_cache_validity &
1818				(NFS_INO_INVALID_ATTR
1819				| NFS_INO_REVAL_FORCED);
1820		cache_revalidated = false;
1821	}
1822
1823	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1824		/*
1825		 * report the blocks in 512byte units
1826		 */
1827		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1828	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1829		inode->i_blocks = fattr->du.nfs2.blocks;
1830	else
1831		cache_revalidated = false;
1832
1833	/* Update attrtimeo value if we're out of the unstable period */
1834	if (invalid & NFS_INO_INVALID_ATTR) {
1835		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1836		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1837		nfsi->attrtimeo_timestamp = now;
1838		/* Set barrier to be more recent than all outstanding updates */
1839		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1840	} else {
1841		if (cache_revalidated) {
1842			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1843				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1844				nfsi->attrtimeo <<= 1;
1845				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1846					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1847			}
1848			nfsi->attrtimeo_timestamp = now;
1849		}
1850		/* Set the barrier to be more recent than this fattr */
1851		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1852			nfsi->attr_gencount = fattr->gencount;
1853	}
1854
1855	/* Don't declare attrcache up to date if there were no attrs! */
1856	if (cache_revalidated)
1857		invalid &= ~NFS_INO_INVALID_ATTR;
1858
1859	/* Don't invalidate the data if we were to blame */
1860	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1861				|| S_ISLNK(inode->i_mode)))
1862		invalid &= ~NFS_INO_INVALID_DATA;
1863	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1864			(save_cache_validity & NFS_INO_REVAL_FORCED))
1865		nfs_set_cache_invalid(inode, invalid);
1866
1867	return 0;
1868 out_err:
1869	/*
1870	 * No need to worry about unhashing the dentry, as the
1871	 * lookup validation will know that the inode is bad.
1872	 * (But we fall through to invalidate the caches.)
1873	 */
1874	nfs_invalidate_inode(inode);
1875	return -ESTALE;
1876}
1877
1878struct inode *nfs_alloc_inode(struct super_block *sb)
1879{
1880	struct nfs_inode *nfsi;
1881	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1882	if (!nfsi)
1883		return NULL;
1884	nfsi->flags = 0UL;
1885	nfsi->cache_validity = 0UL;
1886#if IS_ENABLED(CONFIG_NFS_V4)
1887	nfsi->nfs4_acl = NULL;
1888#endif /* CONFIG_NFS_V4 */
1889	return &nfsi->vfs_inode;
1890}
1891EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1892
1893static void nfs_i_callback(struct rcu_head *head)
1894{
1895	struct inode *inode = container_of(head, struct inode, i_rcu);
1896	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1897}
1898
1899void nfs_destroy_inode(struct inode *inode)
1900{
1901	call_rcu(&inode->i_rcu, nfs_i_callback);
1902}
1903EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1904
1905static inline void nfs4_init_once(struct nfs_inode *nfsi)
1906{
1907#if IS_ENABLED(CONFIG_NFS_V4)
1908	INIT_LIST_HEAD(&nfsi->open_states);
1909	nfsi->delegation = NULL;
1910	init_rwsem(&nfsi->rwsem);
1911	nfsi->layout = NULL;
1912#endif
1913}
1914
1915static void init_once(void *foo)
1916{
1917	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1918
1919	inode_init_once(&nfsi->vfs_inode);
1920	INIT_LIST_HEAD(&nfsi->open_files);
1921	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1922	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1923	INIT_LIST_HEAD(&nfsi->commit_info.list);
1924	nfsi->nrequests = 0;
1925	nfsi->commit_info.ncommit = 0;
1926	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1927	atomic_set(&nfsi->silly_count, 1);
1928	INIT_HLIST_HEAD(&nfsi->silly_list);
1929	init_waitqueue_head(&nfsi->waitqueue);
1930	nfs4_init_once(nfsi);
1931}
1932
1933static int __init nfs_init_inodecache(void)
1934{
1935	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1936					     sizeof(struct nfs_inode),
1937					     0, (SLAB_RECLAIM_ACCOUNT|
1938						SLAB_MEM_SPREAD),
1939					     init_once);
1940	if (nfs_inode_cachep == NULL)
1941		return -ENOMEM;
1942
1943	return 0;
1944}
1945
1946static void nfs_destroy_inodecache(void)
1947{
1948	/*
1949	 * Make sure all delayed rcu free inodes are flushed before we
1950	 * destroy cache.
1951	 */
1952	rcu_barrier();
1953	kmem_cache_destroy(nfs_inode_cachep);
1954}
1955
1956struct workqueue_struct *nfsiod_workqueue;
1957EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1958
1959/*
1960 * start up the nfsiod workqueue
1961 */
1962static int nfsiod_start(void)
1963{
1964	struct workqueue_struct *wq;
1965	dprintk("RPC:       creating workqueue nfsiod\n");
1966	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1967	if (wq == NULL)
1968		return -ENOMEM;
1969	nfsiod_workqueue = wq;
1970	return 0;
1971}
1972
1973/*
1974 * Destroy the nfsiod workqueue
1975 */
1976static void nfsiod_stop(void)
1977{
1978	struct workqueue_struct *wq;
1979
1980	wq = nfsiod_workqueue;
1981	if (wq == NULL)
1982		return;
1983	nfsiod_workqueue = NULL;
1984	destroy_workqueue(wq);
1985}
1986
1987int nfs_net_id;
1988EXPORT_SYMBOL_GPL(nfs_net_id);
1989
1990static int nfs_net_init(struct net *net)
1991{
1992	nfs_clients_init(net);
1993	return nfs_fs_proc_net_init(net);
1994}
1995
1996static void nfs_net_exit(struct net *net)
1997{
1998	nfs_fs_proc_net_exit(net);
1999	nfs_cleanup_cb_ident_idr(net);
2000}
2001
2002static struct pernet_operations nfs_net_ops = {
2003	.init = nfs_net_init,
2004	.exit = nfs_net_exit,
2005	.id   = &nfs_net_id,
2006	.size = sizeof(struct nfs_net),
2007};
2008
2009/*
2010 * Initialize NFS
2011 */
2012static int __init init_nfs_fs(void)
2013{
2014	int err;
2015
2016	err = register_pernet_subsys(&nfs_net_ops);
2017	if (err < 0)
2018		goto out9;
2019
2020	err = nfs_fscache_register();
2021	if (err < 0)
2022		goto out8;
2023
2024	err = nfsiod_start();
2025	if (err)
2026		goto out7;
2027
2028	err = nfs_fs_proc_init();
2029	if (err)
2030		goto out6;
2031
2032	err = nfs_init_nfspagecache();
2033	if (err)
2034		goto out5;
2035
2036	err = nfs_init_inodecache();
2037	if (err)
2038		goto out4;
2039
2040	err = nfs_init_readpagecache();
2041	if (err)
2042		goto out3;
2043
2044	err = nfs_init_writepagecache();
2045	if (err)
2046		goto out2;
2047
2048	err = nfs_init_directcache();
2049	if (err)
2050		goto out1;
2051
2052	rpc_proc_register(&init_net, &nfs_rpcstat);
2053
2054	err = register_nfs_fs();
2055	if (err)
2056		goto out0;
2057
2058	return 0;
2059out0:
2060	rpc_proc_unregister(&init_net, "nfs");
2061	nfs_destroy_directcache();
2062out1:
2063	nfs_destroy_writepagecache();
2064out2:
2065	nfs_destroy_readpagecache();
2066out3:
2067	nfs_destroy_inodecache();
2068out4:
2069	nfs_destroy_nfspagecache();
2070out5:
2071	nfs_fs_proc_exit();
2072out6:
2073	nfsiod_stop();
2074out7:
2075	nfs_fscache_unregister();
2076out8:
2077	unregister_pernet_subsys(&nfs_net_ops);
2078out9:
2079	return err;
2080}
2081
2082static void __exit exit_nfs_fs(void)
2083{
2084	nfs_destroy_directcache();
2085	nfs_destroy_writepagecache();
2086	nfs_destroy_readpagecache();
2087	nfs_destroy_inodecache();
2088	nfs_destroy_nfspagecache();
2089	nfs_fscache_unregister();
2090	unregister_pernet_subsys(&nfs_net_ops);
2091	rpc_proc_unregister(&init_net, "nfs");
2092	unregister_nfs_fs();
2093	nfs_fs_proc_exit();
2094	nfsiod_stop();
2095}
2096
2097/* Not quite true; I just maintain it */
2098MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2099MODULE_LICENSE("GPL");
2100module_param(enable_ino64, bool, 0644);
2101
2102module_init(init_nfs_fs)
2103module_exit(exit_nfs_fs)
2104