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/signal.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 #include <linux/uaccess.h>
42 #include <linux/iversion.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 */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62 
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65 
66 static struct kmem_cache * nfs_inode_cachep;
67 
68 static inline unsigned long
nfs_fattr_to_ino_t(struct nfs_fattr * fattr)69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70 {
71 	return nfs_fileid_to_ino_t(fattr->fileid);
72 }
73 
nfs_wait_killable(int mode)74 static int nfs_wait_killable(int mode)
75 {
76 	freezable_schedule_unsafe();
77 	if (signal_pending_state(mode, current))
78 		return -ERESTARTSYS;
79 	return 0;
80 }
81 
nfs_wait_bit_killable(struct wait_bit_key * key,int mode)82 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
83 {
84 	return nfs_wait_killable(mode);
85 }
86 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
87 
88 /**
89  * nfs_compat_user_ino64 - returns the user-visible inode number
90  * @fileid: 64-bit fileid
91  *
92  * This function returns a 32-bit inode number if the boot parameter
93  * nfs.enable_ino64 is zero.
94  */
nfs_compat_user_ino64(u64 fileid)95 u64 nfs_compat_user_ino64(u64 fileid)
96 {
97 #ifdef CONFIG_COMPAT
98 	compat_ulong_t ino;
99 #else
100 	unsigned long ino;
101 #endif
102 
103 	if (enable_ino64)
104 		return fileid;
105 	ino = fileid;
106 	if (sizeof(ino) < sizeof(fileid))
107 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
108 	return ino;
109 }
110 
nfs_drop_inode(struct inode * inode)111 int nfs_drop_inode(struct inode *inode)
112 {
113 	return NFS_STALE(inode) || generic_drop_inode(inode);
114 }
115 EXPORT_SYMBOL_GPL(nfs_drop_inode);
116 
nfs_clear_inode(struct inode * inode)117 void nfs_clear_inode(struct inode *inode)
118 {
119 	/*
120 	 * The following should never happen...
121 	 */
122 	WARN_ON_ONCE(nfs_have_writebacks(inode));
123 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
124 	nfs_zap_acl_cache(inode);
125 	nfs_access_zap_cache(inode);
126 	nfs_fscache_clear_inode(inode);
127 }
128 EXPORT_SYMBOL_GPL(nfs_clear_inode);
129 
nfs_evict_inode(struct inode * inode)130 void nfs_evict_inode(struct inode *inode)
131 {
132 	truncate_inode_pages_final(&inode->i_data);
133 	clear_inode(inode);
134 	nfs_clear_inode(inode);
135 }
136 
nfs_sync_inode(struct inode * inode)137 int nfs_sync_inode(struct inode *inode)
138 {
139 	inode_dio_wait(inode);
140 	return nfs_wb_all(inode);
141 }
142 EXPORT_SYMBOL_GPL(nfs_sync_inode);
143 
144 /**
145  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
146  */
nfs_sync_mapping(struct address_space * mapping)147 int nfs_sync_mapping(struct address_space *mapping)
148 {
149 	int ret = 0;
150 
151 	if (mapping->nrpages != 0) {
152 		unmap_mapping_range(mapping, 0, 0, 0);
153 		ret = nfs_wb_all(mapping->host);
154 	}
155 	return ret;
156 }
157 
nfs_attribute_timeout(struct inode * inode)158 static int nfs_attribute_timeout(struct inode *inode)
159 {
160 	struct nfs_inode *nfsi = NFS_I(inode);
161 
162 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
163 }
164 
nfs_check_cache_invalid_delegated(struct inode * inode,unsigned long flags)165 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
166 {
167 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
168 
169 	/* Special case for the pagecache or access cache */
170 	if (flags == NFS_INO_REVAL_PAGECACHE &&
171 	    !(cache_validity & NFS_INO_REVAL_FORCED))
172 		return false;
173 	return (cache_validity & flags) != 0;
174 }
175 
nfs_check_cache_invalid_not_delegated(struct inode * inode,unsigned long flags)176 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
177 {
178 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
179 
180 	if ((cache_validity & flags) != 0)
181 		return true;
182 	if (nfs_attribute_timeout(inode))
183 		return true;
184 	return false;
185 }
186 
nfs_check_cache_invalid(struct inode * inode,unsigned long flags)187 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
188 {
189 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
190 		return nfs_check_cache_invalid_delegated(inode, flags);
191 
192 	return nfs_check_cache_invalid_not_delegated(inode, flags);
193 }
194 
nfs_set_cache_invalid(struct inode * inode,unsigned long flags)195 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
196 {
197 	struct nfs_inode *nfsi = NFS_I(inode);
198 	bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
199 
200 	if (have_delegation) {
201 		if (!(flags & NFS_INO_REVAL_FORCED))
202 			flags &= ~NFS_INO_INVALID_OTHER;
203 		flags &= ~(NFS_INO_INVALID_CHANGE
204 				| NFS_INO_INVALID_SIZE
205 				| NFS_INO_REVAL_PAGECACHE);
206 	}
207 
208 	if (inode->i_mapping->nrpages == 0)
209 		flags &= ~NFS_INO_INVALID_DATA;
210 	nfsi->cache_validity |= flags;
211 	if (flags & NFS_INO_INVALID_DATA)
212 		nfs_fscache_invalidate(inode);
213 }
214 
215 /*
216  * Invalidate the local caches
217  */
nfs_zap_caches_locked(struct inode * inode)218 static void nfs_zap_caches_locked(struct inode *inode)
219 {
220 	struct nfs_inode *nfsi = NFS_I(inode);
221 	int mode = inode->i_mode;
222 
223 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
224 
225 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
226 	nfsi->attrtimeo_timestamp = jiffies;
227 
228 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
229 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
230 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
231 					| NFS_INO_INVALID_DATA
232 					| NFS_INO_INVALID_ACCESS
233 					| NFS_INO_INVALID_ACL
234 					| NFS_INO_REVAL_PAGECACHE);
235 	} else
236 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
237 					| NFS_INO_INVALID_ACCESS
238 					| NFS_INO_INVALID_ACL
239 					| NFS_INO_REVAL_PAGECACHE);
240 	nfs_zap_label_cache_locked(nfsi);
241 }
242 
nfs_zap_caches(struct inode * inode)243 void nfs_zap_caches(struct inode *inode)
244 {
245 	spin_lock(&inode->i_lock);
246 	nfs_zap_caches_locked(inode);
247 	spin_unlock(&inode->i_lock);
248 }
249 
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)250 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
251 {
252 	if (mapping->nrpages != 0) {
253 		spin_lock(&inode->i_lock);
254 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
255 		spin_unlock(&inode->i_lock);
256 	}
257 }
258 
nfs_zap_acl_cache(struct inode * inode)259 void nfs_zap_acl_cache(struct inode *inode)
260 {
261 	void (*clear_acl_cache)(struct inode *);
262 
263 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
264 	if (clear_acl_cache != NULL)
265 		clear_acl_cache(inode);
266 	spin_lock(&inode->i_lock);
267 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
268 	spin_unlock(&inode->i_lock);
269 }
270 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
271 
nfs_invalidate_atime(struct inode * inode)272 void nfs_invalidate_atime(struct inode *inode)
273 {
274 	spin_lock(&inode->i_lock);
275 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
276 	spin_unlock(&inode->i_lock);
277 }
278 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
279 
280 /*
281  * Invalidate, but do not unhash, the inode.
282  * NB: must be called with inode->i_lock held!
283  */
nfs_invalidate_inode(struct inode * inode)284 static void nfs_invalidate_inode(struct inode *inode)
285 {
286 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
287 	nfs_zap_caches_locked(inode);
288 }
289 
290 struct nfs_find_desc {
291 	struct nfs_fh		*fh;
292 	struct nfs_fattr	*fattr;
293 };
294 
295 /*
296  * In NFSv3 we can have 64bit inode numbers. In order to support
297  * this, and re-exported directories (also seen in NFSv2)
298  * we are forced to allow 2 different inodes to have the same
299  * i_ino.
300  */
301 static int
nfs_find_actor(struct inode * inode,void * opaque)302 nfs_find_actor(struct inode *inode, void *opaque)
303 {
304 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
305 	struct nfs_fh		*fh = desc->fh;
306 	struct nfs_fattr	*fattr = desc->fattr;
307 
308 	if (NFS_FILEID(inode) != fattr->fileid)
309 		return 0;
310 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
311 		return 0;
312 	if (nfs_compare_fh(NFS_FH(inode), fh))
313 		return 0;
314 	if (is_bad_inode(inode) || NFS_STALE(inode))
315 		return 0;
316 	return 1;
317 }
318 
319 static int
nfs_init_locked(struct inode * inode,void * opaque)320 nfs_init_locked(struct inode *inode, void *opaque)
321 {
322 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
323 	struct nfs_fattr	*fattr = desc->fattr;
324 
325 	set_nfs_fileid(inode, fattr->fileid);
326 	inode->i_mode = fattr->mode;
327 	nfs_copy_fh(NFS_FH(inode), desc->fh);
328 	return 0;
329 }
330 
331 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
nfs_clear_label_invalid(struct inode * inode)332 static void nfs_clear_label_invalid(struct inode *inode)
333 {
334 	spin_lock(&inode->i_lock);
335 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
336 	spin_unlock(&inode->i_lock);
337 }
338 
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr,struct nfs4_label * label)339 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
340 					struct nfs4_label *label)
341 {
342 	int error;
343 
344 	if (label == NULL)
345 		return;
346 
347 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
348 		error = security_inode_notifysecctx(inode, label->label,
349 				label->len);
350 		if (error)
351 			printk(KERN_ERR "%s() %s %d "
352 					"security_inode_notifysecctx() %d\n",
353 					__func__,
354 					(char *)label->label,
355 					label->len, error);
356 		nfs_clear_label_invalid(inode);
357 	}
358 }
359 
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)360 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
361 {
362 	struct nfs4_label *label = NULL;
363 	int minor_version = server->nfs_client->cl_minorversion;
364 
365 	if (minor_version < 2)
366 		return label;
367 
368 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
369 		return label;
370 
371 	label = kzalloc(sizeof(struct nfs4_label), flags);
372 	if (label == NULL)
373 		return ERR_PTR(-ENOMEM);
374 
375 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
376 	if (label->label == NULL) {
377 		kfree(label);
378 		return ERR_PTR(-ENOMEM);
379 	}
380 	label->len = NFS4_MAXLABELLEN;
381 
382 	return label;
383 }
384 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
385 #else
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr,struct nfs4_label * label)386 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
387 					struct nfs4_label *label)
388 {
389 }
390 #endif
391 EXPORT_SYMBOL_GPL(nfs_setsecurity);
392 
393 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
394 struct inode *
nfs_ilookup(struct super_block * sb,struct nfs_fattr * fattr,struct nfs_fh * fh)395 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
396 {
397 	struct nfs_find_desc desc = {
398 		.fh	= fh,
399 		.fattr	= fattr,
400 	};
401 	struct inode *inode;
402 	unsigned long hash;
403 
404 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
405 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
406 		return NULL;
407 
408 	hash = nfs_fattr_to_ino_t(fattr);
409 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
410 
411 	dprintk("%s: returning %p\n", __func__, inode);
412 	return inode;
413 }
414 
415 /*
416  * This is our front-end to iget that looks up inodes by file handle
417  * instead of inode number.
418  */
419 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr,struct nfs4_label * label)420 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
421 {
422 	struct nfs_find_desc desc = {
423 		.fh	= fh,
424 		.fattr	= fattr
425 	};
426 	struct inode *inode = ERR_PTR(-ENOENT);
427 	unsigned long hash;
428 
429 	nfs_attr_check_mountpoint(sb, fattr);
430 
431 	if (nfs_attr_use_mounted_on_fileid(fattr))
432 		fattr->fileid = fattr->mounted_on_fileid;
433 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
434 		goto out_no_inode;
435 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
436 		goto out_no_inode;
437 
438 	hash = nfs_fattr_to_ino_t(fattr);
439 
440 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
441 	if (inode == NULL) {
442 		inode = ERR_PTR(-ENOMEM);
443 		goto out_no_inode;
444 	}
445 
446 	if (inode->i_state & I_NEW) {
447 		struct nfs_inode *nfsi = NFS_I(inode);
448 		unsigned long now = jiffies;
449 
450 		/* We set i_ino for the few things that still rely on it,
451 		 * such as stat(2) */
452 		inode->i_ino = hash;
453 
454 		/* We can't support update_atime(), since the server will reset it */
455 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
456 		inode->i_mode = fattr->mode;
457 		nfsi->cache_validity = 0;
458 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
459 				&& nfs_server_capable(inode, NFS_CAP_MODE))
460 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
461 		/* Why so? Because we want revalidate for devices/FIFOs, and
462 		 * that's precisely what we have in nfs_file_inode_operations.
463 		 */
464 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
465 		if (S_ISREG(inode->i_mode)) {
466 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
467 			inode->i_data.a_ops = &nfs_file_aops;
468 		} else if (S_ISDIR(inode->i_mode)) {
469 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
470 			inode->i_fop = &nfs_dir_operations;
471 			inode->i_data.a_ops = &nfs_dir_aops;
472 			/* Deal with crossing mountpoints */
473 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
474 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
475 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
476 					inode->i_op = &nfs_referral_inode_operations;
477 				else
478 					inode->i_op = &nfs_mountpoint_inode_operations;
479 				inode->i_fop = NULL;
480 				inode->i_flags |= S_AUTOMOUNT;
481 			}
482 		} else if (S_ISLNK(inode->i_mode)) {
483 			inode->i_op = &nfs_symlink_inode_operations;
484 			inode_nohighmem(inode);
485 		} else
486 			init_special_inode(inode, inode->i_mode, fattr->rdev);
487 
488 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
489 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
490 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
491 		inode_set_iversion_raw(inode, 0);
492 		inode->i_size = 0;
493 		clear_nlink(inode);
494 		inode->i_uid = make_kuid(&init_user_ns, -2);
495 		inode->i_gid = make_kgid(&init_user_ns, -2);
496 		inode->i_blocks = 0;
497 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
498 		nfsi->write_io = 0;
499 		nfsi->read_io = 0;
500 
501 		nfsi->read_cache_jiffies = fattr->time_start;
502 		nfsi->attr_gencount = fattr->gencount;
503 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
504 			inode->i_atime = timespec_to_timespec64(fattr->atime);
505 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
506 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
507 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
508 			inode->i_mtime = timespec_to_timespec64(fattr->mtime);
509 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
510 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
511 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
512 			inode->i_ctime = timespec_to_timespec64(fattr->ctime);
513 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
514 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
515 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
516 			inode_set_iversion_raw(inode, fattr->change_attr);
517 		else
518 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
519 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
520 			inode->i_size = nfs_size_to_loff_t(fattr->size);
521 		else
522 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
523 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
524 			set_nlink(inode, fattr->nlink);
525 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
526 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
527 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
528 			inode->i_uid = fattr->uid;
529 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
530 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
531 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
532 			inode->i_gid = fattr->gid;
533 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
534 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
535 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
536 			inode->i_blocks = fattr->du.nfs2.blocks;
537 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
538 			/*
539 			 * report the blocks in 512byte units
540 			 */
541 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
542 		}
543 
544 		if (nfsi->cache_validity != 0)
545 			nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
546 
547 		nfs_setsecurity(inode, fattr, label);
548 
549 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
550 		nfsi->attrtimeo_timestamp = now;
551 		nfsi->access_cache = RB_ROOT;
552 
553 		nfs_fscache_init_inode(inode);
554 
555 		unlock_new_inode(inode);
556 	} else {
557 		int err = nfs_refresh_inode(inode, fattr);
558 		if (err < 0) {
559 			iput(inode);
560 			inode = ERR_PTR(err);
561 			goto out_no_inode;
562 		}
563 	}
564 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
565 		inode->i_sb->s_id,
566 		(unsigned long long)NFS_FILEID(inode),
567 		nfs_display_fhandle_hash(fh),
568 		atomic_read(&inode->i_count));
569 
570 out:
571 	return inode;
572 
573 out_no_inode:
574 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
575 	goto out;
576 }
577 EXPORT_SYMBOL_GPL(nfs_fhget);
578 
579 #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)
580 
581 int
nfs_setattr(struct dentry * dentry,struct iattr * attr)582 nfs_setattr(struct dentry *dentry, struct iattr *attr)
583 {
584 	struct inode *inode = d_inode(dentry);
585 	struct nfs_fattr *fattr;
586 	int error = 0;
587 
588 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
589 
590 	/* skip mode change if it's just for clearing setuid/setgid */
591 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
592 		attr->ia_valid &= ~ATTR_MODE;
593 
594 	if (attr->ia_valid & ATTR_SIZE) {
595 		BUG_ON(!S_ISREG(inode->i_mode));
596 
597 		error = inode_newsize_ok(inode, attr->ia_size);
598 		if (error)
599 			return error;
600 
601 		if (attr->ia_size == i_size_read(inode))
602 			attr->ia_valid &= ~ATTR_SIZE;
603 	}
604 
605 	/* Optimization: if the end result is no change, don't RPC */
606 	attr->ia_valid &= NFS_VALID_ATTRS;
607 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
608 		return 0;
609 
610 	trace_nfs_setattr_enter(inode);
611 
612 	/* Write all dirty data */
613 	if (S_ISREG(inode->i_mode))
614 		nfs_sync_inode(inode);
615 
616 	fattr = nfs_alloc_fattr();
617 	if (fattr == NULL) {
618 		error = -ENOMEM;
619 		goto out;
620 	}
621 
622 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
623 	if (error == 0)
624 		error = nfs_refresh_inode(inode, fattr);
625 	nfs_free_fattr(fattr);
626 out:
627 	trace_nfs_setattr_exit(inode, error);
628 	return error;
629 }
630 EXPORT_SYMBOL_GPL(nfs_setattr);
631 
632 /**
633  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
634  * @inode: inode of the file used
635  * @offset: file offset to start truncating
636  *
637  * This is a copy of the common vmtruncate, but with the locking
638  * corrected to take into account the fact that NFS requires
639  * inode->i_size to be updated under the inode->i_lock.
640  * Note: must be called with inode->i_lock held!
641  */
nfs_vmtruncate(struct inode * inode,loff_t offset)642 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
643 {
644 	int err;
645 
646 	err = inode_newsize_ok(inode, offset);
647 	if (err)
648 		goto out;
649 
650 	i_size_write(inode, offset);
651 	/* Optimisation */
652 	if (offset == 0)
653 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
654 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
655 
656 	spin_unlock(&inode->i_lock);
657 	truncate_pagecache(inode, offset);
658 	spin_lock(&inode->i_lock);
659 out:
660 	return err;
661 }
662 
663 /**
664  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
665  * @inode: pointer to struct inode
666  * @attr: pointer to struct iattr
667  * @fattr: pointer to struct nfs_fattr
668  *
669  * Note: we do this in the *proc.c in order to ensure that
670  *       it works for things like exclusive creates too.
671  */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)672 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
673 		struct nfs_fattr *fattr)
674 {
675 	/* Barrier: bump the attribute generation count. */
676 	nfs_fattr_set_barrier(fattr);
677 
678 	spin_lock(&inode->i_lock);
679 	NFS_I(inode)->attr_gencount = fattr->gencount;
680 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
681 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
682 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
683 		nfs_vmtruncate(inode, attr->ia_size);
684 	}
685 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
686 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
687 		if ((attr->ia_valid & ATTR_MODE) != 0) {
688 			int mode = attr->ia_mode & S_IALLUGO;
689 			mode |= inode->i_mode & ~S_IALLUGO;
690 			inode->i_mode = mode;
691 		}
692 		if ((attr->ia_valid & ATTR_UID) != 0)
693 			inode->i_uid = attr->ia_uid;
694 		if ((attr->ia_valid & ATTR_GID) != 0)
695 			inode->i_gid = attr->ia_gid;
696 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
697 			inode->i_ctime = timespec_to_timespec64(fattr->ctime);
698 		else
699 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
700 					| NFS_INO_INVALID_CTIME);
701 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
702 				| NFS_INO_INVALID_ACL);
703 	}
704 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
705 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
706 				| NFS_INO_INVALID_CTIME);
707 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
708 			inode->i_atime = timespec_to_timespec64(fattr->atime);
709 		else if (attr->ia_valid & ATTR_ATIME_SET)
710 			inode->i_atime = attr->ia_atime;
711 		else
712 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
713 
714 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
715 			inode->i_ctime = timespec_to_timespec64(fattr->ctime);
716 		else
717 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
718 					| NFS_INO_INVALID_CTIME);
719 	}
720 	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
721 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
722 				| NFS_INO_INVALID_CTIME);
723 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
724 			inode->i_mtime = timespec_to_timespec64(fattr->mtime);
725 		else if (attr->ia_valid & ATTR_MTIME_SET)
726 			inode->i_mtime = attr->ia_mtime;
727 		else
728 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
729 
730 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
731 			inode->i_ctime = timespec_to_timespec64(fattr->ctime);
732 		else
733 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
734 					| NFS_INO_INVALID_CTIME);
735 	}
736 	if (fattr->valid)
737 		nfs_update_inode(inode, fattr);
738 	spin_unlock(&inode->i_lock);
739 }
740 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
741 
nfs_readdirplus_parent_cache_miss(struct dentry * dentry)742 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
743 {
744 	struct dentry *parent;
745 
746 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
747 		return;
748 	parent = dget_parent(dentry);
749 	nfs_force_use_readdirplus(d_inode(parent));
750 	dput(parent);
751 }
752 
nfs_readdirplus_parent_cache_hit(struct dentry * dentry)753 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
754 {
755 	struct dentry *parent;
756 
757 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
758 		return;
759 	parent = dget_parent(dentry);
760 	nfs_advise_use_readdirplus(d_inode(parent));
761 	dput(parent);
762 }
763 
nfs_need_revalidate_inode(struct inode * inode)764 static bool nfs_need_revalidate_inode(struct inode *inode)
765 {
766 	if (NFS_I(inode)->cache_validity &
767 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
768 		return true;
769 	if (nfs_attribute_cache_expired(inode))
770 		return true;
771 	return false;
772 }
773 
nfs_getattr(const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)774 int nfs_getattr(const struct path *path, struct kstat *stat,
775 		u32 request_mask, unsigned int query_flags)
776 {
777 	struct inode *inode = d_inode(path->dentry);
778 	struct nfs_server *server = NFS_SERVER(inode);
779 	unsigned long cache_validity;
780 	int err = 0;
781 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
782 	bool do_update = false;
783 
784 	trace_nfs_getattr_enter(inode);
785 
786 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
787 		goto out_no_update;
788 
789 	/* Flush out writes to the server in order to update c/mtime.  */
790 	if ((request_mask & (STATX_CTIME | STATX_MTIME)) &&
791 	    S_ISREG(inode->i_mode))
792 		filemap_write_and_wait(inode->i_mapping);
793 
794 	/*
795 	 * We may force a getattr if the user cares about atime.
796 	 *
797 	 * Note that we only have to check the vfsmount flags here:
798 	 *  - NFS always sets S_NOATIME by so checking it would give a
799 	 *    bogus result
800 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
801 	 *    no point in checking those.
802 	 */
803 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
804 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
805 		request_mask &= ~STATX_ATIME;
806 
807 	/* Is the user requesting attributes that might need revalidation? */
808 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
809 					STATX_MTIME|STATX_UID|STATX_GID|
810 					STATX_SIZE|STATX_BLOCKS)))
811 		goto out_no_revalidate;
812 
813 	/* Check whether the cached attributes are stale */
814 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
815 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
816 	do_update |= cache_validity &
817 		(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
818 	if (request_mask & STATX_ATIME)
819 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
820 	if (request_mask & (STATX_CTIME|STATX_MTIME))
821 		do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
822 	if (do_update) {
823 		/* Update the attribute cache */
824 		if (!(server->flags & NFS_MOUNT_NOAC))
825 			nfs_readdirplus_parent_cache_miss(path->dentry);
826 		else
827 			nfs_readdirplus_parent_cache_hit(path->dentry);
828 		err = __nfs_revalidate_inode(server, inode);
829 		if (err)
830 			goto out;
831 	} else
832 		nfs_readdirplus_parent_cache_hit(path->dentry);
833 out_no_revalidate:
834 	/* Only return attributes that were revalidated. */
835 	stat->result_mask &= request_mask;
836 out_no_update:
837 	generic_fillattr(inode, stat);
838 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
839 	if (S_ISDIR(inode->i_mode))
840 		stat->blksize = NFS_SERVER(inode)->dtsize;
841 out:
842 	trace_nfs_getattr_exit(inode, err);
843 	return err;
844 }
845 EXPORT_SYMBOL_GPL(nfs_getattr);
846 
nfs_init_lock_context(struct nfs_lock_context * l_ctx)847 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
848 {
849 	refcount_set(&l_ctx->count, 1);
850 	l_ctx->lockowner = current->files;
851 	INIT_LIST_HEAD(&l_ctx->list);
852 	atomic_set(&l_ctx->io_count, 0);
853 }
854 
__nfs_find_lock_context(struct nfs_open_context * ctx)855 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
856 {
857 	struct nfs_lock_context *head = &ctx->lock_context;
858 	struct nfs_lock_context *pos = head;
859 
860 	do {
861 		if (pos->lockowner != current->files)
862 			continue;
863 		refcount_inc(&pos->count);
864 		return pos;
865 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
866 	return NULL;
867 }
868 
nfs_get_lock_context(struct nfs_open_context * ctx)869 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
870 {
871 	struct nfs_lock_context *res, *new = NULL;
872 	struct inode *inode = d_inode(ctx->dentry);
873 
874 	spin_lock(&inode->i_lock);
875 	res = __nfs_find_lock_context(ctx);
876 	if (res == NULL) {
877 		spin_unlock(&inode->i_lock);
878 		new = kmalloc(sizeof(*new), GFP_KERNEL);
879 		if (new == NULL)
880 			return ERR_PTR(-ENOMEM);
881 		nfs_init_lock_context(new);
882 		spin_lock(&inode->i_lock);
883 		res = __nfs_find_lock_context(ctx);
884 		if (res == NULL) {
885 			list_add_tail(&new->list, &ctx->lock_context.list);
886 			new->open_context = ctx;
887 			res = new;
888 			new = NULL;
889 		}
890 	}
891 	spin_unlock(&inode->i_lock);
892 	kfree(new);
893 	return res;
894 }
895 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
896 
nfs_put_lock_context(struct nfs_lock_context * l_ctx)897 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
898 {
899 	struct nfs_open_context *ctx = l_ctx->open_context;
900 	struct inode *inode = d_inode(ctx->dentry);
901 
902 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
903 		return;
904 	list_del(&l_ctx->list);
905 	spin_unlock(&inode->i_lock);
906 	kfree(l_ctx);
907 }
908 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
909 
910 /**
911  * nfs_close_context - Common close_context() routine NFSv2/v3
912  * @ctx: pointer to context
913  * @is_sync: is this a synchronous close
914  *
915  * Ensure that the attributes are up to date if we're mounted
916  * with close-to-open semantics and we have cached data that will
917  * need to be revalidated on open.
918  */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)919 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
920 {
921 	struct nfs_inode *nfsi;
922 	struct inode *inode;
923 	struct nfs_server *server;
924 
925 	if (!(ctx->mode & FMODE_WRITE))
926 		return;
927 	if (!is_sync)
928 		return;
929 	inode = d_inode(ctx->dentry);
930 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
931 		return;
932 	nfsi = NFS_I(inode);
933 	if (inode->i_mapping->nrpages == 0)
934 		return;
935 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
936 		return;
937 	if (!list_empty(&nfsi->open_files))
938 		return;
939 	server = NFS_SERVER(inode);
940 	if (server->flags & NFS_MOUNT_NOCTO)
941 		return;
942 	nfs_revalidate_inode(server, inode);
943 }
944 EXPORT_SYMBOL_GPL(nfs_close_context);
945 
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)946 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
947 						fmode_t f_mode,
948 						struct file *filp)
949 {
950 	struct nfs_open_context *ctx;
951 	struct rpc_cred *cred = rpc_lookup_cred();
952 	if (IS_ERR(cred))
953 		return ERR_CAST(cred);
954 
955 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
956 	if (!ctx) {
957 		put_rpccred(cred);
958 		return ERR_PTR(-ENOMEM);
959 	}
960 	nfs_sb_active(dentry->d_sb);
961 	ctx->dentry = dget(dentry);
962 	ctx->cred = cred;
963 	ctx->state = NULL;
964 	ctx->mode = f_mode;
965 	ctx->flags = 0;
966 	ctx->error = 0;
967 	ctx->flock_owner = (fl_owner_t)filp;
968 	nfs_init_lock_context(&ctx->lock_context);
969 	ctx->lock_context.open_context = ctx;
970 	INIT_LIST_HEAD(&ctx->list);
971 	ctx->mdsthreshold = NULL;
972 	return ctx;
973 }
974 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
975 
get_nfs_open_context(struct nfs_open_context * ctx)976 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
977 {
978 	if (ctx != NULL)
979 		refcount_inc(&ctx->lock_context.count);
980 	return ctx;
981 }
982 EXPORT_SYMBOL_GPL(get_nfs_open_context);
983 
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)984 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
985 {
986 	struct inode *inode = d_inode(ctx->dentry);
987 	struct super_block *sb = ctx->dentry->d_sb;
988 
989 	if (!list_empty(&ctx->list)) {
990 		if (!refcount_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
991 			return;
992 		list_del(&ctx->list);
993 		spin_unlock(&inode->i_lock);
994 	} else if (!refcount_dec_and_test(&ctx->lock_context.count))
995 		return;
996 	if (inode != NULL)
997 		NFS_PROTO(inode)->close_context(ctx, is_sync);
998 	if (ctx->cred != NULL)
999 		put_rpccred(ctx->cred);
1000 	dput(ctx->dentry);
1001 	nfs_sb_deactive(sb);
1002 	kfree(ctx->mdsthreshold);
1003 	kfree(ctx);
1004 }
1005 
put_nfs_open_context(struct nfs_open_context * ctx)1006 void put_nfs_open_context(struct nfs_open_context *ctx)
1007 {
1008 	__put_nfs_open_context(ctx, 0);
1009 }
1010 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1011 
put_nfs_open_context_sync(struct nfs_open_context * ctx)1012 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1013 {
1014 	__put_nfs_open_context(ctx, 1);
1015 }
1016 
1017 /*
1018  * Ensure that mmap has a recent RPC credential for use when writing out
1019  * shared pages
1020  */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1021 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1022 {
1023 	struct inode *inode = d_inode(ctx->dentry);
1024 	struct nfs_inode *nfsi = NFS_I(inode);
1025 
1026 	spin_lock(&inode->i_lock);
1027 	if (ctx->mode & FMODE_WRITE)
1028 		list_add(&ctx->list, &nfsi->open_files);
1029 	else
1030 		list_add_tail(&ctx->list, &nfsi->open_files);
1031 	spin_unlock(&inode->i_lock);
1032 }
1033 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1034 
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1035 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1036 {
1037 	filp->private_data = get_nfs_open_context(ctx);
1038 	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1039 	if (list_empty(&ctx->list))
1040 		nfs_inode_attach_open_context(ctx);
1041 }
1042 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1043 
1044 /*
1045  * Given an inode, search for an open context with the desired characteristics
1046  */
nfs_find_open_context(struct inode * inode,struct rpc_cred * cred,fmode_t mode)1047 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
1048 {
1049 	struct nfs_inode *nfsi = NFS_I(inode);
1050 	struct nfs_open_context *pos, *ctx = NULL;
1051 
1052 	spin_lock(&inode->i_lock);
1053 	list_for_each_entry(pos, &nfsi->open_files, list) {
1054 		if (cred != NULL && pos->cred != cred)
1055 			continue;
1056 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1057 			continue;
1058 		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1059 			continue;
1060 		ctx = get_nfs_open_context(pos);
1061 		break;
1062 	}
1063 	spin_unlock(&inode->i_lock);
1064 	return ctx;
1065 }
1066 
nfs_file_clear_open_context(struct file * filp)1067 void nfs_file_clear_open_context(struct file *filp)
1068 {
1069 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1070 
1071 	if (ctx) {
1072 		struct inode *inode = d_inode(ctx->dentry);
1073 
1074 		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1075 		/*
1076 		 * We fatal error on write before. Try to writeback
1077 		 * every page again.
1078 		 */
1079 		if (ctx->error < 0)
1080 			invalidate_inode_pages2(inode->i_mapping);
1081 		filp->private_data = NULL;
1082 		spin_lock(&inode->i_lock);
1083 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1084 		spin_unlock(&inode->i_lock);
1085 		put_nfs_open_context_sync(ctx);
1086 	}
1087 }
1088 
1089 /*
1090  * These allocate and release file read/write context information.
1091  */
nfs_open(struct inode * inode,struct file * filp)1092 int nfs_open(struct inode *inode, struct file *filp)
1093 {
1094 	struct nfs_open_context *ctx;
1095 
1096 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1097 	if (IS_ERR(ctx))
1098 		return PTR_ERR(ctx);
1099 	nfs_file_set_open_context(filp, ctx);
1100 	put_nfs_open_context(ctx);
1101 	nfs_fscache_open_file(inode, filp);
1102 	return 0;
1103 }
1104 EXPORT_SYMBOL_GPL(nfs_open);
1105 
1106 /*
1107  * This function is called whenever some part of NFS notices that
1108  * the cached attributes have to be refreshed.
1109  */
1110 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1111 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1112 {
1113 	int		 status = -ESTALE;
1114 	struct nfs4_label *label = NULL;
1115 	struct nfs_fattr *fattr = NULL;
1116 	struct nfs_inode *nfsi = NFS_I(inode);
1117 
1118 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1119 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1120 
1121 	trace_nfs_revalidate_inode_enter(inode);
1122 
1123 	if (is_bad_inode(inode))
1124 		goto out;
1125 	if (NFS_STALE(inode))
1126 		goto out;
1127 
1128 	/* pNFS: Attributes aren't updated until we layoutcommit */
1129 	if (S_ISREG(inode->i_mode)) {
1130 		status = pnfs_sync_inode(inode, false);
1131 		if (status)
1132 			goto out;
1133 	}
1134 
1135 	status = -ENOMEM;
1136 	fattr = nfs_alloc_fattr();
1137 	if (fattr == NULL)
1138 		goto out;
1139 
1140 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1141 
1142 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1143 	if (IS_ERR(label)) {
1144 		status = PTR_ERR(label);
1145 		goto out;
1146 	}
1147 
1148 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1149 			label, inode);
1150 	if (status != 0) {
1151 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1152 			 inode->i_sb->s_id,
1153 			 (unsigned long long)NFS_FILEID(inode), status);
1154 		if (status == -ESTALE) {
1155 			nfs_zap_caches(inode);
1156 			if (!S_ISDIR(inode->i_mode))
1157 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1158 		}
1159 		goto err_out;
1160 	}
1161 
1162 	status = nfs_refresh_inode(inode, fattr);
1163 	if (status) {
1164 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1165 			 inode->i_sb->s_id,
1166 			 (unsigned long long)NFS_FILEID(inode), status);
1167 		goto err_out;
1168 	}
1169 
1170 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1171 		nfs_zap_acl_cache(inode);
1172 
1173 	nfs_setsecurity(inode, fattr, label);
1174 
1175 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1176 		inode->i_sb->s_id,
1177 		(unsigned long long)NFS_FILEID(inode));
1178 
1179 err_out:
1180 	nfs4_label_free(label);
1181 out:
1182 	nfs_free_fattr(fattr);
1183 	trace_nfs_revalidate_inode_exit(inode, status);
1184 	return status;
1185 }
1186 
nfs_attribute_cache_expired(struct inode * inode)1187 int nfs_attribute_cache_expired(struct inode *inode)
1188 {
1189 	if (nfs_have_delegated_attributes(inode))
1190 		return 0;
1191 	return nfs_attribute_timeout(inode);
1192 }
1193 
1194 /**
1195  * nfs_revalidate_inode - Revalidate the inode attributes
1196  * @server - pointer to nfs_server struct
1197  * @inode - pointer to inode struct
1198  *
1199  * Updates inode attribute information by retrieving the data from the server.
1200  */
nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1201 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1202 {
1203 	if (!nfs_need_revalidate_inode(inode))
1204 		return NFS_STALE(inode) ? -ESTALE : 0;
1205 	return __nfs_revalidate_inode(server, inode);
1206 }
1207 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1208 
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1209 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1210 {
1211 	struct nfs_inode *nfsi = NFS_I(inode);
1212 	int ret;
1213 
1214 	if (mapping->nrpages != 0) {
1215 		if (S_ISREG(inode->i_mode)) {
1216 			ret = nfs_sync_mapping(mapping);
1217 			if (ret < 0)
1218 				return ret;
1219 		}
1220 		ret = invalidate_inode_pages2(mapping);
1221 		if (ret < 0)
1222 			return ret;
1223 	}
1224 	if (S_ISDIR(inode->i_mode)) {
1225 		spin_lock(&inode->i_lock);
1226 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1227 		spin_unlock(&inode->i_lock);
1228 	}
1229 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1230 	nfs_fscache_wait_on_invalidate(inode);
1231 
1232 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1233 			inode->i_sb->s_id,
1234 			(unsigned long long)NFS_FILEID(inode));
1235 	return 0;
1236 }
1237 
nfs_mapping_need_revalidate_inode(struct inode * inode)1238 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1239 {
1240 	return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1241 		NFS_STALE(inode);
1242 }
1243 
nfs_revalidate_mapping_rcu(struct inode * inode)1244 int nfs_revalidate_mapping_rcu(struct inode *inode)
1245 {
1246 	struct nfs_inode *nfsi = NFS_I(inode);
1247 	unsigned long *bitlock = &nfsi->flags;
1248 	int ret = 0;
1249 
1250 	if (IS_SWAPFILE(inode))
1251 		goto out;
1252 	if (nfs_mapping_need_revalidate_inode(inode)) {
1253 		ret = -ECHILD;
1254 		goto out;
1255 	}
1256 	spin_lock(&inode->i_lock);
1257 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1258 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1259 		ret = -ECHILD;
1260 	spin_unlock(&inode->i_lock);
1261 out:
1262 	return ret;
1263 }
1264 
1265 /**
1266  * nfs_revalidate_mapping - Revalidate the pagecache
1267  * @inode - pointer to host inode
1268  * @mapping - pointer to mapping
1269  */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1270 int nfs_revalidate_mapping(struct inode *inode,
1271 		struct address_space *mapping)
1272 {
1273 	struct nfs_inode *nfsi = NFS_I(inode);
1274 	unsigned long *bitlock = &nfsi->flags;
1275 	int ret = 0;
1276 
1277 	/* swapfiles are not supposed to be shared. */
1278 	if (IS_SWAPFILE(inode))
1279 		goto out;
1280 
1281 	if (nfs_mapping_need_revalidate_inode(inode)) {
1282 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1283 		if (ret < 0)
1284 			goto out;
1285 	}
1286 
1287 	/*
1288 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1289 	 * invalidations that come in while we're shooting down the mappings
1290 	 * are respected. But, that leaves a race window where one revalidator
1291 	 * can clear the flag, and then another checks it before the mapping
1292 	 * gets invalidated. Fix that by serializing access to this part of
1293 	 * the function.
1294 	 *
1295 	 * At the same time, we need to allow other tasks to see whether we
1296 	 * might be in the middle of invalidating the pages, so we only set
1297 	 * the bit lock here if it looks like we're going to be doing that.
1298 	 */
1299 	for (;;) {
1300 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1301 					 nfs_wait_bit_killable, TASK_KILLABLE);
1302 		if (ret)
1303 			goto out;
1304 		spin_lock(&inode->i_lock);
1305 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1306 			spin_unlock(&inode->i_lock);
1307 			continue;
1308 		}
1309 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1310 			break;
1311 		spin_unlock(&inode->i_lock);
1312 		goto out;
1313 	}
1314 
1315 	set_bit(NFS_INO_INVALIDATING, bitlock);
1316 	smp_wmb();
1317 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1318 	spin_unlock(&inode->i_lock);
1319 	trace_nfs_invalidate_mapping_enter(inode);
1320 	ret = nfs_invalidate_mapping(inode, mapping);
1321 	trace_nfs_invalidate_mapping_exit(inode, ret);
1322 
1323 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1324 	smp_mb__after_atomic();
1325 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1326 out:
1327 	return ret;
1328 }
1329 
nfs_file_has_writers(struct nfs_inode * nfsi)1330 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1331 {
1332 	struct inode *inode = &nfsi->vfs_inode;
1333 
1334 	assert_spin_locked(&inode->i_lock);
1335 
1336 	if (!S_ISREG(inode->i_mode))
1337 		return false;
1338 	if (list_empty(&nfsi->open_files))
1339 		return false;
1340 	/* Note: This relies on nfsi->open_files being ordered with writers
1341 	 *       being placed at the head of the list.
1342 	 *       See nfs_inode_attach_open_context()
1343 	 */
1344 	return (list_first_entry(&nfsi->open_files,
1345 			struct nfs_open_context,
1346 			list)->mode & FMODE_WRITE) == FMODE_WRITE;
1347 }
1348 
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1349 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1350 {
1351 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1352 }
1353 
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1354 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1355 {
1356 	struct timespec ts;
1357 
1358 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1359 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1360 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1361 		inode_set_iversion_raw(inode, fattr->change_attr);
1362 		if (S_ISDIR(inode->i_mode))
1363 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1364 	}
1365 	/* If we have atomic WCC data, we may update some attributes */
1366 	ts = timespec64_to_timespec(inode->i_ctime);
1367 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1368 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1369 			&& timespec_equal(&ts, &fattr->pre_ctime)) {
1370 		inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1371 	}
1372 
1373 	ts = timespec64_to_timespec(inode->i_mtime);
1374 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1375 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1376 			&& timespec_equal(&ts, &fattr->pre_mtime)) {
1377 		inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1378 		if (S_ISDIR(inode->i_mode))
1379 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1380 	}
1381 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1382 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1383 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1384 			&& !nfs_have_writebacks(inode)) {
1385 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1386 	}
1387 }
1388 
1389 /**
1390  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1391  * @inode - pointer to inode
1392  * @fattr - updated attributes
1393  *
1394  * Verifies the attribute cache. If we have just changed the attributes,
1395  * so that fattr carries weak cache consistency data, then it may
1396  * also update the ctime/mtime/change_attribute.
1397  */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1398 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1399 {
1400 	struct nfs_inode *nfsi = NFS_I(inode);
1401 	loff_t cur_size, new_isize;
1402 	unsigned long invalid = 0;
1403 	struct timespec ts;
1404 
1405 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1406 		return 0;
1407 
1408 	/* Has the inode gone and changed behind our back? */
1409 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1410 		return -ESTALE;
1411 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1412 		return -ESTALE;
1413 
1414 	if (!nfs_file_has_buffered_writers(nfsi)) {
1415 		/* Verify a few of the more important attributes */
1416 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1417 			invalid |= NFS_INO_INVALID_CHANGE
1418 				| NFS_INO_REVAL_PAGECACHE;
1419 
1420 		ts = timespec64_to_timespec(inode->i_mtime);
1421 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&ts, &fattr->mtime))
1422 			invalid |= NFS_INO_INVALID_MTIME;
1423 
1424 		ts = timespec64_to_timespec(inode->i_ctime);
1425 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&ts, &fattr->ctime))
1426 			invalid |= NFS_INO_INVALID_CTIME;
1427 
1428 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1429 			cur_size = i_size_read(inode);
1430 			new_isize = nfs_size_to_loff_t(fattr->size);
1431 			if (cur_size != new_isize)
1432 				invalid |= NFS_INO_INVALID_SIZE
1433 					| NFS_INO_REVAL_PAGECACHE;
1434 		}
1435 	}
1436 
1437 	/* Have any file permissions changed? */
1438 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1439 		invalid |= NFS_INO_INVALID_ACCESS
1440 			| NFS_INO_INVALID_ACL
1441 			| NFS_INO_INVALID_OTHER;
1442 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1443 		invalid |= NFS_INO_INVALID_ACCESS
1444 			| NFS_INO_INVALID_ACL
1445 			| NFS_INO_INVALID_OTHER;
1446 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1447 		invalid |= NFS_INO_INVALID_ACCESS
1448 			| NFS_INO_INVALID_ACL
1449 			| NFS_INO_INVALID_OTHER;
1450 
1451 	/* Has the link count changed? */
1452 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1453 		invalid |= NFS_INO_INVALID_OTHER;
1454 
1455 	ts = timespec64_to_timespec(inode->i_atime);
1456 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&ts, &fattr->atime))
1457 		invalid |= NFS_INO_INVALID_ATIME;
1458 
1459 	if (invalid != 0)
1460 		nfs_set_cache_invalid(inode, invalid);
1461 
1462 	nfsi->read_cache_jiffies = fattr->time_start;
1463 	return 0;
1464 }
1465 
1466 static atomic_long_t nfs_attr_generation_counter;
1467 
nfs_read_attr_generation_counter(void)1468 static unsigned long nfs_read_attr_generation_counter(void)
1469 {
1470 	return atomic_long_read(&nfs_attr_generation_counter);
1471 }
1472 
nfs_inc_attr_generation_counter(void)1473 unsigned long nfs_inc_attr_generation_counter(void)
1474 {
1475 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1476 }
1477 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1478 
nfs_fattr_init(struct nfs_fattr * fattr)1479 void nfs_fattr_init(struct nfs_fattr *fattr)
1480 {
1481 	fattr->valid = 0;
1482 	fattr->time_start = jiffies;
1483 	fattr->gencount = nfs_inc_attr_generation_counter();
1484 	fattr->owner_name = NULL;
1485 	fattr->group_name = NULL;
1486 }
1487 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1488 
1489 /**
1490  * nfs_fattr_set_barrier
1491  * @fattr: attributes
1492  *
1493  * Used to set a barrier after an attribute was updated. This
1494  * barrier ensures that older attributes from RPC calls that may
1495  * have raced with our update cannot clobber these new values.
1496  * Note that you are still responsible for ensuring that other
1497  * operations which change the attribute on the server do not
1498  * collide.
1499  */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1500 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1501 {
1502 	fattr->gencount = nfs_inc_attr_generation_counter();
1503 }
1504 
nfs_alloc_fattr(void)1505 struct nfs_fattr *nfs_alloc_fattr(void)
1506 {
1507 	struct nfs_fattr *fattr;
1508 
1509 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1510 	if (fattr != NULL)
1511 		nfs_fattr_init(fattr);
1512 	return fattr;
1513 }
1514 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1515 
nfs_alloc_fhandle(void)1516 struct nfs_fh *nfs_alloc_fhandle(void)
1517 {
1518 	struct nfs_fh *fh;
1519 
1520 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1521 	if (fh != NULL)
1522 		fh->size = 0;
1523 	return fh;
1524 }
1525 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1526 
1527 #ifdef NFS_DEBUG
1528 /*
1529  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1530  *                             in the same way that wireshark does
1531  *
1532  * @fh: file handle
1533  *
1534  * For debugging only.
1535  */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1536 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1537 {
1538 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1539 	 * not on the result */
1540 	return nfs_fhandle_hash(fh);
1541 }
1542 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1543 
1544 /*
1545  * _nfs_display_fhandle - display an NFS file handle on the console
1546  *
1547  * @fh: file handle to display
1548  * @caption: display caption
1549  *
1550  * For debugging only.
1551  */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1552 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1553 {
1554 	unsigned short i;
1555 
1556 	if (fh == NULL || fh->size == 0) {
1557 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1558 		return;
1559 	}
1560 
1561 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1562 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1563 	for (i = 0; i < fh->size; i += 16) {
1564 		__be32 *pos = (__be32 *)&fh->data[i];
1565 
1566 		switch ((fh->size - i - 1) >> 2) {
1567 		case 0:
1568 			printk(KERN_DEFAULT " %08x\n",
1569 				be32_to_cpup(pos));
1570 			break;
1571 		case 1:
1572 			printk(KERN_DEFAULT " %08x %08x\n",
1573 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1574 			break;
1575 		case 2:
1576 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1577 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1578 				be32_to_cpup(pos + 2));
1579 			break;
1580 		default:
1581 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1582 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1583 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1584 		}
1585 	}
1586 }
1587 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1588 #endif
1589 
1590 /**
1591  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1592  * @inode - pointer to inode
1593  * @fattr - attributes
1594  *
1595  * Attempt to divine whether or not an RPC call reply carrying stale
1596  * attributes got scheduled after another call carrying updated ones.
1597  *
1598  * To do so, the function first assumes that a more recent ctime means
1599  * that the attributes in fattr are newer, however it also attempt to
1600  * catch the case where ctime either didn't change, or went backwards
1601  * (if someone reset the clock on the server) by looking at whether
1602  * or not this RPC call was started after the inode was last updated.
1603  * Note also the check for wraparound of 'attr_gencount'
1604  *
1605  * The function returns 'true' if it thinks the attributes in 'fattr' are
1606  * more recent than the ones cached in the inode.
1607  *
1608  */
nfs_inode_attrs_need_update(const struct inode * inode,const struct nfs_fattr * fattr)1609 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1610 {
1611 	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1612 
1613 	return (long)(fattr->gencount - attr_gencount) > 0 ||
1614 	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1615 }
1616 
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1617 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1618 {
1619 	int ret;
1620 
1621 	trace_nfs_refresh_inode_enter(inode);
1622 
1623 	if (nfs_inode_attrs_need_update(inode, fattr))
1624 		ret = nfs_update_inode(inode, fattr);
1625 	else
1626 		ret = nfs_check_inode_attributes(inode, fattr);
1627 
1628 	trace_nfs_refresh_inode_exit(inode, ret);
1629 	return ret;
1630 }
1631 
1632 /**
1633  * nfs_refresh_inode - try to update the inode attribute cache
1634  * @inode - pointer to inode
1635  * @fattr - updated attributes
1636  *
1637  * Check that an RPC call that returned attributes has not overlapped with
1638  * other recent updates of the inode metadata, then decide whether it is
1639  * safe to do a full update of the inode attributes, or whether just to
1640  * call nfs_check_inode_attributes.
1641  */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)1642 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1643 {
1644 	int status;
1645 
1646 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1647 		return 0;
1648 	spin_lock(&inode->i_lock);
1649 	status = nfs_refresh_inode_locked(inode, fattr);
1650 	spin_unlock(&inode->i_lock);
1651 
1652 	return status;
1653 }
1654 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1655 
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)1656 static int nfs_post_op_update_inode_locked(struct inode *inode,
1657 		struct nfs_fattr *fattr, unsigned int invalid)
1658 {
1659 	if (S_ISDIR(inode->i_mode))
1660 		invalid |= NFS_INO_INVALID_DATA;
1661 	nfs_set_cache_invalid(inode, invalid);
1662 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1663 		return 0;
1664 	return nfs_refresh_inode_locked(inode, fattr);
1665 }
1666 
1667 /**
1668  * nfs_post_op_update_inode - try to update the inode attribute cache
1669  * @inode - pointer to inode
1670  * @fattr - updated attributes
1671  *
1672  * After an operation that has changed the inode metadata, mark the
1673  * attribute cache as being invalid, then try to update it.
1674  *
1675  * NB: if the server didn't return any post op attributes, this
1676  * function will force the retrieval of attributes before the next
1677  * NFS request.  Thus it should be used only for operations that
1678  * are expected to change one or more attributes, to avoid
1679  * unnecessary NFS requests and trips through nfs_update_inode().
1680  */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)1681 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1682 {
1683 	int status;
1684 
1685 	spin_lock(&inode->i_lock);
1686 	nfs_fattr_set_barrier(fattr);
1687 	status = nfs_post_op_update_inode_locked(inode, fattr,
1688 			NFS_INO_INVALID_CHANGE
1689 			| NFS_INO_INVALID_CTIME
1690 			| NFS_INO_REVAL_FORCED);
1691 	spin_unlock(&inode->i_lock);
1692 
1693 	return status;
1694 }
1695 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1696 
1697 /**
1698  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1699  * @inode - pointer to inode
1700  * @fattr - updated attributes
1701  *
1702  * After an operation that has changed the inode metadata, mark the
1703  * attribute cache as being invalid, then try to update it. Fake up
1704  * weak cache consistency data, if none exist.
1705  *
1706  * This function is mainly designed to be used by the ->write_done() functions.
1707  */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)1708 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1709 {
1710 	int status;
1711 
1712 	/* Don't do a WCC update if these attributes are already stale */
1713 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1714 			!nfs_inode_attrs_need_update(inode, fattr)) {
1715 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1716 				| NFS_ATTR_FATTR_PRESIZE
1717 				| NFS_ATTR_FATTR_PREMTIME
1718 				| NFS_ATTR_FATTR_PRECTIME);
1719 		goto out_noforce;
1720 	}
1721 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1722 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1723 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1724 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1725 	}
1726 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1727 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1728 		fattr->pre_ctime = timespec64_to_timespec(inode->i_ctime);
1729 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1730 	}
1731 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1732 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1733 		fattr->pre_mtime = timespec64_to_timespec(inode->i_mtime);
1734 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1735 	}
1736 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1737 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1738 		fattr->pre_size = i_size_read(inode);
1739 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1740 	}
1741 out_noforce:
1742 	status = nfs_post_op_update_inode_locked(inode, fattr,
1743 			NFS_INO_INVALID_CHANGE
1744 			| NFS_INO_INVALID_CTIME
1745 			| NFS_INO_INVALID_MTIME);
1746 	return status;
1747 }
1748 
1749 /**
1750  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1751  * @inode - pointer to inode
1752  * @fattr - updated attributes
1753  *
1754  * After an operation that has changed the inode metadata, mark the
1755  * attribute cache as being invalid, then try to update it. Fake up
1756  * weak cache consistency data, if none exist.
1757  *
1758  * This function is mainly designed to be used by the ->write_done() functions.
1759  */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)1760 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1761 {
1762 	int status;
1763 
1764 	spin_lock(&inode->i_lock);
1765 	nfs_fattr_set_barrier(fattr);
1766 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1767 	spin_unlock(&inode->i_lock);
1768 	return status;
1769 }
1770 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1771 
1772 
nfs_fileid_valid(struct nfs_inode * nfsi,struct nfs_fattr * fattr)1773 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1774 				    struct nfs_fattr *fattr)
1775 {
1776 	bool ret1 = true, ret2 = true;
1777 
1778 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1779 		ret1 = (nfsi->fileid == fattr->fileid);
1780 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1781 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1782 	return ret1 || ret2;
1783 }
1784 
1785 /*
1786  * Many nfs protocol calls return the new file attributes after
1787  * an operation.  Here we update the inode to reflect the state
1788  * of the server's inode.
1789  *
1790  * This is a bit tricky because we have to make sure all dirty pages
1791  * have been sent off to the server before calling invalidate_inode_pages.
1792  * To make sure no other process adds more write requests while we try
1793  * our best to flush them, we make them sleep during the attribute refresh.
1794  *
1795  * A very similar scenario holds for the dir cache.
1796  */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)1797 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1798 {
1799 	struct nfs_server *server;
1800 	struct nfs_inode *nfsi = NFS_I(inode);
1801 	loff_t cur_isize, new_isize;
1802 	unsigned long invalid = 0;
1803 	unsigned long now = jiffies;
1804 	unsigned long save_cache_validity;
1805 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1806 	bool cache_revalidated = true;
1807 	bool attr_changed = false;
1808 	bool have_delegation;
1809 
1810 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1811 			__func__, inode->i_sb->s_id, inode->i_ino,
1812 			nfs_display_fhandle_hash(NFS_FH(inode)),
1813 			atomic_read(&inode->i_count), fattr->valid);
1814 
1815 	if (!nfs_fileid_valid(nfsi, fattr)) {
1816 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1817 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1818 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1819 			inode->i_sb->s_id, (long long)nfsi->fileid,
1820 			(long long)fattr->fileid);
1821 		goto out_err;
1822 	}
1823 
1824 	/*
1825 	 * Make sure the inode's type hasn't changed.
1826 	 */
1827 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1828 		/*
1829 		* Big trouble! The inode has become a different object.
1830 		*/
1831 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1832 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1833 		goto out_err;
1834 	}
1835 
1836 	server = NFS_SERVER(inode);
1837 	/* Update the fsid? */
1838 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1839 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1840 			!IS_AUTOMOUNT(inode))
1841 		server->fsid = fattr->fsid;
1842 
1843 	/* Save the delegation state before clearing cache_validity */
1844 	have_delegation = nfs_have_delegated_attributes(inode);
1845 
1846 	/*
1847 	 * Update the read time so we don't revalidate too often.
1848 	 */
1849 	nfsi->read_cache_jiffies = fattr->time_start;
1850 
1851 	save_cache_validity = nfsi->cache_validity;
1852 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1853 			| NFS_INO_INVALID_ATIME
1854 			| NFS_INO_REVAL_FORCED
1855 			| NFS_INO_REVAL_PAGECACHE);
1856 
1857 	/* Do atomic weak cache consistency updates */
1858 	nfs_wcc_update_inode(inode, fattr);
1859 
1860 	if (pnfs_layoutcommit_outstanding(inode)) {
1861 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1862 		cache_revalidated = false;
1863 	}
1864 
1865 	/* More cache consistency checks */
1866 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1867 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1868 			/* Could it be a race with writeback? */
1869 			if (!(have_writers || have_delegation)) {
1870 				invalid |= NFS_INO_INVALID_DATA
1871 					| NFS_INO_INVALID_ACCESS
1872 					| NFS_INO_INVALID_ACL;
1873 				/* Force revalidate of all attributes */
1874 				save_cache_validity |= NFS_INO_INVALID_CTIME
1875 					| NFS_INO_INVALID_MTIME
1876 					| NFS_INO_INVALID_SIZE
1877 					| NFS_INO_INVALID_OTHER;
1878 				if (S_ISDIR(inode->i_mode))
1879 					nfs_force_lookup_revalidate(inode);
1880 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
1881 						inode->i_sb->s_id,
1882 						inode->i_ino);
1883 			}
1884 			inode_set_iversion_raw(inode, fattr->change_attr);
1885 			attr_changed = true;
1886 		}
1887 	} else {
1888 		nfsi->cache_validity |= save_cache_validity &
1889 				(NFS_INO_INVALID_CHANGE
1890 				| NFS_INO_REVAL_PAGECACHE
1891 				| NFS_INO_REVAL_FORCED);
1892 		cache_revalidated = false;
1893 	}
1894 
1895 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1896 		inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1897 	} else if (server->caps & NFS_CAP_MTIME) {
1898 		nfsi->cache_validity |= save_cache_validity &
1899 				(NFS_INO_INVALID_MTIME
1900 				| NFS_INO_REVAL_FORCED);
1901 		cache_revalidated = false;
1902 	}
1903 
1904 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1905 		inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1906 	} else if (server->caps & NFS_CAP_CTIME) {
1907 		nfsi->cache_validity |= save_cache_validity &
1908 				(NFS_INO_INVALID_CTIME
1909 				| NFS_INO_REVAL_FORCED);
1910 		cache_revalidated = false;
1911 	}
1912 
1913 	/* Check if our cached file size is stale */
1914 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1915 		new_isize = nfs_size_to_loff_t(fattr->size);
1916 		cur_isize = i_size_read(inode);
1917 		if (new_isize != cur_isize && !have_delegation) {
1918 			/* Do we perhaps have any outstanding writes, or has
1919 			 * the file grown beyond our last write? */
1920 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1921 				i_size_write(inode, new_isize);
1922 				if (!have_writers)
1923 					invalid |= NFS_INO_INVALID_DATA;
1924 				attr_changed = true;
1925 			}
1926 			dprintk("NFS: isize change on server for file %s/%ld "
1927 					"(%Ld to %Ld)\n",
1928 					inode->i_sb->s_id,
1929 					inode->i_ino,
1930 					(long long)cur_isize,
1931 					(long long)new_isize);
1932 		}
1933 	} else {
1934 		nfsi->cache_validity |= save_cache_validity &
1935 				(NFS_INO_INVALID_SIZE
1936 				| NFS_INO_REVAL_PAGECACHE
1937 				| NFS_INO_REVAL_FORCED);
1938 		cache_revalidated = false;
1939 	}
1940 
1941 
1942 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1943 		inode->i_atime = timespec_to_timespec64(fattr->atime);
1944 	else if (server->caps & NFS_CAP_ATIME) {
1945 		nfsi->cache_validity |= save_cache_validity &
1946 				(NFS_INO_INVALID_ATIME
1947 				| NFS_INO_REVAL_FORCED);
1948 		cache_revalidated = false;
1949 	}
1950 
1951 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1952 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1953 			umode_t newmode = inode->i_mode & S_IFMT;
1954 			newmode |= fattr->mode & S_IALLUGO;
1955 			inode->i_mode = newmode;
1956 			invalid |= NFS_INO_INVALID_ACCESS
1957 				| NFS_INO_INVALID_ACL;
1958 			attr_changed = true;
1959 		}
1960 	} else if (server->caps & NFS_CAP_MODE) {
1961 		nfsi->cache_validity |= save_cache_validity &
1962 				(NFS_INO_INVALID_OTHER
1963 				| NFS_INO_REVAL_FORCED);
1964 		cache_revalidated = false;
1965 	}
1966 
1967 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1968 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1969 			invalid |= NFS_INO_INVALID_ACCESS
1970 				| NFS_INO_INVALID_ACL;
1971 			inode->i_uid = fattr->uid;
1972 			attr_changed = true;
1973 		}
1974 	} else if (server->caps & NFS_CAP_OWNER) {
1975 		nfsi->cache_validity |= save_cache_validity &
1976 				(NFS_INO_INVALID_OTHER
1977 				| NFS_INO_REVAL_FORCED);
1978 		cache_revalidated = false;
1979 	}
1980 
1981 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1982 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1983 			invalid |= NFS_INO_INVALID_ACCESS
1984 				| NFS_INO_INVALID_ACL;
1985 			inode->i_gid = fattr->gid;
1986 			attr_changed = true;
1987 		}
1988 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1989 		nfsi->cache_validity |= save_cache_validity &
1990 				(NFS_INO_INVALID_OTHER
1991 				| NFS_INO_REVAL_FORCED);
1992 		cache_revalidated = false;
1993 	}
1994 
1995 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1996 		if (inode->i_nlink != fattr->nlink) {
1997 			if (S_ISDIR(inode->i_mode))
1998 				invalid |= NFS_INO_INVALID_DATA;
1999 			set_nlink(inode, fattr->nlink);
2000 			attr_changed = true;
2001 		}
2002 	} else if (server->caps & NFS_CAP_NLINK) {
2003 		nfsi->cache_validity |= save_cache_validity &
2004 				(NFS_INO_INVALID_OTHER
2005 				| NFS_INO_REVAL_FORCED);
2006 		cache_revalidated = false;
2007 	}
2008 
2009 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2010 		/*
2011 		 * report the blocks in 512byte units
2012 		 */
2013 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2014 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2015 		inode->i_blocks = fattr->du.nfs2.blocks;
2016 	else
2017 		cache_revalidated = false;
2018 
2019 	/* Update attrtimeo value if we're out of the unstable period */
2020 	if (attr_changed) {
2021 		invalid &= ~NFS_INO_INVALID_ATTR;
2022 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2023 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2024 		nfsi->attrtimeo_timestamp = now;
2025 		/* Set barrier to be more recent than all outstanding updates */
2026 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2027 	} else {
2028 		if (cache_revalidated) {
2029 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2030 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2031 				nfsi->attrtimeo <<= 1;
2032 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2033 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2034 			}
2035 			nfsi->attrtimeo_timestamp = now;
2036 		}
2037 		/* Set the barrier to be more recent than this fattr */
2038 		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2039 			nfsi->attr_gencount = fattr->gencount;
2040 	}
2041 
2042 	/* Don't invalidate the data if we were to blame */
2043 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2044 				|| S_ISLNK(inode->i_mode)))
2045 		invalid &= ~NFS_INO_INVALID_DATA;
2046 	nfs_set_cache_invalid(inode, invalid);
2047 
2048 	return 0;
2049  out_err:
2050 	/*
2051 	 * No need to worry about unhashing the dentry, as the
2052 	 * lookup validation will know that the inode is bad.
2053 	 * (But we fall through to invalidate the caches.)
2054 	 */
2055 	nfs_invalidate_inode(inode);
2056 	return -ESTALE;
2057 }
2058 
nfs_alloc_inode(struct super_block * sb)2059 struct inode *nfs_alloc_inode(struct super_block *sb)
2060 {
2061 	struct nfs_inode *nfsi;
2062 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2063 	if (!nfsi)
2064 		return NULL;
2065 	nfsi->flags = 0UL;
2066 	nfsi->cache_validity = 0UL;
2067 #if IS_ENABLED(CONFIG_NFS_V4)
2068 	nfsi->nfs4_acl = NULL;
2069 #endif /* CONFIG_NFS_V4 */
2070 	return &nfsi->vfs_inode;
2071 }
2072 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2073 
nfs_i_callback(struct rcu_head * head)2074 static void nfs_i_callback(struct rcu_head *head)
2075 {
2076 	struct inode *inode = container_of(head, struct inode, i_rcu);
2077 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2078 }
2079 
nfs_destroy_inode(struct inode * inode)2080 void nfs_destroy_inode(struct inode *inode)
2081 {
2082 	call_rcu(&inode->i_rcu, nfs_i_callback);
2083 }
2084 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
2085 
nfs4_init_once(struct nfs_inode * nfsi)2086 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2087 {
2088 #if IS_ENABLED(CONFIG_NFS_V4)
2089 	INIT_LIST_HEAD(&nfsi->open_states);
2090 	nfsi->delegation = NULL;
2091 	init_rwsem(&nfsi->rwsem);
2092 	nfsi->layout = NULL;
2093 #endif
2094 }
2095 
init_once(void * foo)2096 static void init_once(void *foo)
2097 {
2098 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2099 
2100 	inode_init_once(&nfsi->vfs_inode);
2101 	INIT_LIST_HEAD(&nfsi->open_files);
2102 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2103 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2104 	INIT_LIST_HEAD(&nfsi->commit_info.list);
2105 	atomic_long_set(&nfsi->nrequests, 0);
2106 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
2107 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
2108 	init_rwsem(&nfsi->rmdir_sem);
2109 	mutex_init(&nfsi->commit_mutex);
2110 	nfs4_init_once(nfsi);
2111 }
2112 
nfs_init_inodecache(void)2113 static int __init nfs_init_inodecache(void)
2114 {
2115 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2116 					     sizeof(struct nfs_inode),
2117 					     0, (SLAB_RECLAIM_ACCOUNT|
2118 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2119 					     init_once);
2120 	if (nfs_inode_cachep == NULL)
2121 		return -ENOMEM;
2122 
2123 	return 0;
2124 }
2125 
nfs_destroy_inodecache(void)2126 static void nfs_destroy_inodecache(void)
2127 {
2128 	/*
2129 	 * Make sure all delayed rcu free inodes are flushed before we
2130 	 * destroy cache.
2131 	 */
2132 	rcu_barrier();
2133 	kmem_cache_destroy(nfs_inode_cachep);
2134 }
2135 
2136 struct workqueue_struct *nfsiod_workqueue;
2137 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2138 
2139 /*
2140  * start up the nfsiod workqueue
2141  */
nfsiod_start(void)2142 static int nfsiod_start(void)
2143 {
2144 	struct workqueue_struct *wq;
2145 	dprintk("RPC:       creating workqueue nfsiod\n");
2146 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2147 	if (wq == NULL)
2148 		return -ENOMEM;
2149 	nfsiod_workqueue = wq;
2150 	return 0;
2151 }
2152 
2153 /*
2154  * Destroy the nfsiod workqueue
2155  */
nfsiod_stop(void)2156 static void nfsiod_stop(void)
2157 {
2158 	struct workqueue_struct *wq;
2159 
2160 	wq = nfsiod_workqueue;
2161 	if (wq == NULL)
2162 		return;
2163 	nfsiod_workqueue = NULL;
2164 	destroy_workqueue(wq);
2165 }
2166 
2167 unsigned int nfs_net_id;
2168 EXPORT_SYMBOL_GPL(nfs_net_id);
2169 
nfs_net_init(struct net * net)2170 static int nfs_net_init(struct net *net)
2171 {
2172 	nfs_clients_init(net);
2173 	return nfs_fs_proc_net_init(net);
2174 }
2175 
nfs_net_exit(struct net * net)2176 static void nfs_net_exit(struct net *net)
2177 {
2178 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2179 
2180 	nfs_fs_proc_net_exit(net);
2181 	nfs_cleanup_cb_ident_idr(net);
2182 	WARN_ON_ONCE(!list_empty(&nn->nfs_client_list));
2183 	WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list));
2184 }
2185 
2186 static struct pernet_operations nfs_net_ops = {
2187 	.init = nfs_net_init,
2188 	.exit = nfs_net_exit,
2189 	.id   = &nfs_net_id,
2190 	.size = sizeof(struct nfs_net),
2191 };
2192 
2193 /*
2194  * Initialize NFS
2195  */
init_nfs_fs(void)2196 static int __init init_nfs_fs(void)
2197 {
2198 	int err;
2199 
2200 	err = register_pernet_subsys(&nfs_net_ops);
2201 	if (err < 0)
2202 		goto out9;
2203 
2204 	err = nfs_fscache_register();
2205 	if (err < 0)
2206 		goto out8;
2207 
2208 	err = nfsiod_start();
2209 	if (err)
2210 		goto out7;
2211 
2212 	err = nfs_fs_proc_init();
2213 	if (err)
2214 		goto out6;
2215 
2216 	err = nfs_init_nfspagecache();
2217 	if (err)
2218 		goto out5;
2219 
2220 	err = nfs_init_inodecache();
2221 	if (err)
2222 		goto out4;
2223 
2224 	err = nfs_init_readpagecache();
2225 	if (err)
2226 		goto out3;
2227 
2228 	err = nfs_init_writepagecache();
2229 	if (err)
2230 		goto out2;
2231 
2232 	err = nfs_init_directcache();
2233 	if (err)
2234 		goto out1;
2235 
2236 	rpc_proc_register(&init_net, &nfs_rpcstat);
2237 
2238 	err = register_nfs_fs();
2239 	if (err)
2240 		goto out0;
2241 
2242 	return 0;
2243 out0:
2244 	rpc_proc_unregister(&init_net, "nfs");
2245 	nfs_destroy_directcache();
2246 out1:
2247 	nfs_destroy_writepagecache();
2248 out2:
2249 	nfs_destroy_readpagecache();
2250 out3:
2251 	nfs_destroy_inodecache();
2252 out4:
2253 	nfs_destroy_nfspagecache();
2254 out5:
2255 	nfs_fs_proc_exit();
2256 out6:
2257 	nfsiod_stop();
2258 out7:
2259 	nfs_fscache_unregister();
2260 out8:
2261 	unregister_pernet_subsys(&nfs_net_ops);
2262 out9:
2263 	return err;
2264 }
2265 
exit_nfs_fs(void)2266 static void __exit exit_nfs_fs(void)
2267 {
2268 	nfs_destroy_directcache();
2269 	nfs_destroy_writepagecache();
2270 	nfs_destroy_readpagecache();
2271 	nfs_destroy_inodecache();
2272 	nfs_destroy_nfspagecache();
2273 	nfs_fscache_unregister();
2274 	unregister_pernet_subsys(&nfs_net_ops);
2275 	rpc_proc_unregister(&init_net, "nfs");
2276 	unregister_nfs_fs();
2277 	nfs_fs_proc_exit();
2278 	nfsiod_stop();
2279 }
2280 
2281 /* Not quite true; I just maintain it */
2282 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2283 MODULE_LICENSE("GPL");
2284 module_param(enable_ino64, bool, 0644);
2285 
2286 module_init(init_nfs_fs)
2287 module_exit(exit_nfs_fs)
2288