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 */
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 
74 /**
75  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76  * @word: long word containing the bit lock
77  */
nfs_wait_bit_killable(struct wait_bit_key * key)78 int nfs_wait_bit_killable(struct wait_bit_key *key)
79 {
80 	if (fatal_signal_pending(current))
81 		return -ERESTARTSYS;
82 	freezable_schedule_unsafe();
83 	return 0;
84 }
85 EXPORT_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  */
nfs_compat_user_ino64(u64 fileid)94 u64 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 
nfs_drop_inode(struct inode * inode)110 int nfs_drop_inode(struct inode *inode)
111 {
112 	return NFS_STALE(inode) || generic_drop_inode(inode);
113 }
114 EXPORT_SYMBOL_GPL(nfs_drop_inode);
115 
nfs_clear_inode(struct inode * inode)116 void 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 }
127 EXPORT_SYMBOL_GPL(nfs_clear_inode);
128 
nfs_evict_inode(struct inode * inode)129 void 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 
nfs_sync_inode(struct inode * inode)136 int nfs_sync_inode(struct inode *inode)
137 {
138 	nfs_inode_dio_wait(inode);
139 	return nfs_wb_all(inode);
140 }
141 EXPORT_SYMBOL_GPL(nfs_sync_inode);
142 
143 /**
144  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
145  */
nfs_sync_mapping(struct address_space * mapping)146 int 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 
nfs_set_cache_invalid(struct inode * inode,unsigned long flags)157 static 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  */
nfs_zap_caches_locked(struct inode * inode)171 static 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 
nfs_zap_caches(struct inode * inode)196 void 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 
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)203 void 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 
nfs_zap_acl_cache(struct inode * inode)212 void 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 }
223 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
224 
nfs_invalidate_atime(struct inode * inode)225 void 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 }
231 EXPORT_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  */
nfs_invalidate_inode(struct inode * inode)237 static 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 
243 struct 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  */
254 static int
nfs_find_actor(struct inode * inode,void * opaque)255 nfs_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 
272 static int
nfs_init_locked(struct inode * inode,void * opaque)273 nfs_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
nfs_clear_label_invalid(struct inode * inode)284 static 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 
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr,struct nfs4_label * label)291 void 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 
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)312 struct 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 }
336 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
337 #else
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr,struct nfs4_label * label)338 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
339 					struct nfs4_label *label)
340 {
341 }
342 #endif
343 EXPORT_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  */
349 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr,struct nfs4_label * label)350 nfs_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 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
448 			inode->i_size = nfs_size_to_loff_t(fattr->size);
449 		else
450 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
451 				| NFS_INO_REVAL_PAGECACHE);
452 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
453 			set_nlink(inode, fattr->nlink);
454 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
455 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
456 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
457 			inode->i_uid = fattr->uid;
458 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
459 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
460 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
461 			inode->i_gid = fattr->gid;
462 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
463 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
464 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
465 			inode->i_blocks = fattr->du.nfs2.blocks;
466 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
467 			/*
468 			 * report the blocks in 512byte units
469 			 */
470 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
471 		}
472 
473 		nfs_setsecurity(inode, fattr, label);
474 
475 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
476 		nfsi->attrtimeo_timestamp = now;
477 		nfsi->access_cache = RB_ROOT;
478 
479 		nfs_fscache_init_inode(inode);
480 
481 		unlock_new_inode(inode);
482 	} else
483 		nfs_refresh_inode(inode, fattr);
484 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
485 		inode->i_sb->s_id,
486 		(unsigned long long)NFS_FILEID(inode),
487 		nfs_display_fhandle_hash(fh),
488 		atomic_read(&inode->i_count));
489 
490 out:
491 	return inode;
492 
493 out_no_inode:
494 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
495 	goto out;
496 }
497 EXPORT_SYMBOL_GPL(nfs_fhget);
498 
499 #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)
500 
501 int
nfs_setattr(struct dentry * dentry,struct iattr * attr)502 nfs_setattr(struct dentry *dentry, struct iattr *attr)
503 {
504 	struct inode *inode = d_inode(dentry);
505 	struct nfs_fattr *fattr;
506 	int error = -ENOMEM;
507 
508 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
509 
510 	/* skip mode change if it's just for clearing setuid/setgid */
511 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
512 		attr->ia_valid &= ~ATTR_MODE;
513 
514 	if (attr->ia_valid & ATTR_SIZE) {
515 		loff_t i_size;
516 
517 		BUG_ON(!S_ISREG(inode->i_mode));
518 
519 		i_size = i_size_read(inode);
520 		if (attr->ia_size == i_size)
521 			attr->ia_valid &= ~ATTR_SIZE;
522 		else if (attr->ia_size < i_size && IS_SWAPFILE(inode))
523 			return -ETXTBSY;
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 		goto out;
540 	/*
541 	 * Return any delegations if we're going to change ACLs
542 	 */
543 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
544 		NFS_PROTO(inode)->return_delegation(inode);
545 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
546 	if (error == 0)
547 		error = nfs_refresh_inode(inode, fattr);
548 	nfs_free_fattr(fattr);
549 out:
550 	trace_nfs_setattr_exit(inode, error);
551 	return error;
552 }
553 EXPORT_SYMBOL_GPL(nfs_setattr);
554 
555 /**
556  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
557  * @inode: inode of the file used
558  * @offset: file offset to start truncating
559  *
560  * This is a copy of the common vmtruncate, but with the locking
561  * corrected to take into account the fact that NFS requires
562  * inode->i_size to be updated under the inode->i_lock.
563  * Note: must be called with inode->i_lock held!
564  */
nfs_vmtruncate(struct inode * inode,loff_t offset)565 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
566 {
567 	int err;
568 
569 	err = inode_newsize_ok(inode, offset);
570 	if (err)
571 		goto out;
572 
573 	i_size_write(inode, offset);
574 	/* Optimisation */
575 	if (offset == 0)
576 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
577 
578 	spin_unlock(&inode->i_lock);
579 	truncate_pagecache(inode, offset);
580 	spin_lock(&inode->i_lock);
581 out:
582 	return err;
583 }
584 
585 /**
586  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
587  * @inode: pointer to struct inode
588  * @attr: pointer to struct iattr
589  *
590  * Note: we do this in the *proc.c in order to ensure that
591  *       it works for things like exclusive creates too.
592  */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)593 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
594 		struct nfs_fattr *fattr)
595 {
596 	/* Barrier: bump the attribute generation count. */
597 	nfs_fattr_set_barrier(fattr);
598 
599 	spin_lock(&inode->i_lock);
600 	NFS_I(inode)->attr_gencount = fattr->gencount;
601 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
602 		if ((attr->ia_valid & ATTR_MODE) != 0) {
603 			int mode = attr->ia_mode & S_IALLUGO;
604 			mode |= inode->i_mode & ~S_IALLUGO;
605 			inode->i_mode = mode;
606 		}
607 		if ((attr->ia_valid & ATTR_UID) != 0)
608 			inode->i_uid = attr->ia_uid;
609 		if ((attr->ia_valid & ATTR_GID) != 0)
610 			inode->i_gid = attr->ia_gid;
611 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
612 				| NFS_INO_INVALID_ACL);
613 	}
614 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
615 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
616 		nfs_vmtruncate(inode, attr->ia_size);
617 	}
618 	nfs_update_inode(inode, fattr);
619 	spin_unlock(&inode->i_lock);
620 }
621 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
622 
nfs_request_parent_use_readdirplus(struct dentry * dentry)623 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
624 {
625 	struct dentry *parent;
626 
627 	parent = dget_parent(dentry);
628 	nfs_force_use_readdirplus(d_inode(parent));
629 	dput(parent);
630 }
631 
nfs_need_revalidate_inode(struct inode * inode)632 static bool nfs_need_revalidate_inode(struct inode *inode)
633 {
634 	if (NFS_I(inode)->cache_validity &
635 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
636 		return true;
637 	if (nfs_attribute_cache_expired(inode))
638 		return true;
639 	return false;
640 }
641 
nfs_getattr(struct vfsmount * mnt,struct dentry * dentry,struct kstat * stat)642 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
643 {
644 	struct inode *inode = d_inode(dentry);
645 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
646 	int err = 0;
647 
648 	trace_nfs_getattr_enter(inode);
649 	/* Flush out writes to the server in order to update c/mtime.  */
650 	if (S_ISREG(inode->i_mode)) {
651 		mutex_lock(&inode->i_mutex);
652 		err = nfs_sync_inode(inode);
653 		mutex_unlock(&inode->i_mutex);
654 		if (err)
655 			goto out;
656 	}
657 
658 	/*
659 	 * We may force a getattr if the user cares about atime.
660 	 *
661 	 * Note that we only have to check the vfsmount flags here:
662 	 *  - NFS always sets S_NOATIME by so checking it would give a
663 	 *    bogus result
664 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
665 	 *    no point in checking those.
666 	 */
667  	if ((mnt->mnt_flags & MNT_NOATIME) ||
668  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
669 		need_atime = 0;
670 
671 	if (need_atime || nfs_need_revalidate_inode(inode)) {
672 		struct nfs_server *server = NFS_SERVER(inode);
673 
674 		if (server->caps & NFS_CAP_READDIRPLUS)
675 			nfs_request_parent_use_readdirplus(dentry);
676 		err = __nfs_revalidate_inode(server, inode);
677 	}
678 	if (!err) {
679 		generic_fillattr(inode, stat);
680 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
681 	}
682 out:
683 	trace_nfs_getattr_exit(inode, err);
684 	return err;
685 }
686 EXPORT_SYMBOL_GPL(nfs_getattr);
687 
nfs_init_lock_context(struct nfs_lock_context * l_ctx)688 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
689 {
690 	atomic_set(&l_ctx->count, 1);
691 	l_ctx->lockowner.l_owner = current->files;
692 	l_ctx->lockowner.l_pid = current->tgid;
693 	INIT_LIST_HEAD(&l_ctx->list);
694 	nfs_iocounter_init(&l_ctx->io_count);
695 }
696 
__nfs_find_lock_context(struct nfs_open_context * ctx)697 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
698 {
699 	struct nfs_lock_context *head = &ctx->lock_context;
700 	struct nfs_lock_context *pos = head;
701 
702 	do {
703 		if (pos->lockowner.l_owner != current->files)
704 			continue;
705 		if (pos->lockowner.l_pid != current->tgid)
706 			continue;
707 		atomic_inc(&pos->count);
708 		return pos;
709 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
710 	return NULL;
711 }
712 
nfs_get_lock_context(struct nfs_open_context * ctx)713 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
714 {
715 	struct nfs_lock_context *res, *new = NULL;
716 	struct inode *inode = d_inode(ctx->dentry);
717 
718 	spin_lock(&inode->i_lock);
719 	res = __nfs_find_lock_context(ctx);
720 	if (res == NULL) {
721 		spin_unlock(&inode->i_lock);
722 		new = kmalloc(sizeof(*new), GFP_KERNEL);
723 		if (new == NULL)
724 			return ERR_PTR(-ENOMEM);
725 		nfs_init_lock_context(new);
726 		spin_lock(&inode->i_lock);
727 		res = __nfs_find_lock_context(ctx);
728 		if (res == NULL) {
729 			list_add_tail(&new->list, &ctx->lock_context.list);
730 			new->open_context = ctx;
731 			res = new;
732 			new = NULL;
733 		}
734 	}
735 	spin_unlock(&inode->i_lock);
736 	kfree(new);
737 	return res;
738 }
739 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
740 
nfs_put_lock_context(struct nfs_lock_context * l_ctx)741 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
742 {
743 	struct nfs_open_context *ctx = l_ctx->open_context;
744 	struct inode *inode = d_inode(ctx->dentry);
745 
746 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
747 		return;
748 	list_del(&l_ctx->list);
749 	spin_unlock(&inode->i_lock);
750 	kfree(l_ctx);
751 }
752 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
753 
754 /**
755  * nfs_close_context - Common close_context() routine NFSv2/v3
756  * @ctx: pointer to context
757  * @is_sync: is this a synchronous close
758  *
759  * always ensure that the attributes are up to date if we're mounted
760  * with close-to-open semantics
761  */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)762 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
763 {
764 	struct inode *inode;
765 	struct nfs_server *server;
766 
767 	if (!(ctx->mode & FMODE_WRITE))
768 		return;
769 	if (!is_sync)
770 		return;
771 	inode = d_inode(ctx->dentry);
772 	if (!list_empty(&NFS_I(inode)->open_files))
773 		return;
774 	server = NFS_SERVER(inode);
775 	if (server->flags & NFS_MOUNT_NOCTO)
776 		return;
777 	nfs_revalidate_inode(server, inode);
778 }
779 EXPORT_SYMBOL_GPL(nfs_close_context);
780 
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode)781 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
782 {
783 	struct nfs_open_context *ctx;
784 	struct rpc_cred *cred = rpc_lookup_cred();
785 	if (IS_ERR(cred))
786 		return ERR_CAST(cred);
787 
788 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
789 	if (!ctx) {
790 		put_rpccred(cred);
791 		return ERR_PTR(-ENOMEM);
792 	}
793 	nfs_sb_active(dentry->d_sb);
794 	ctx->dentry = dget(dentry);
795 	ctx->cred = cred;
796 	ctx->state = NULL;
797 	ctx->mode = f_mode;
798 	ctx->flags = 0;
799 	ctx->error = 0;
800 	nfs_init_lock_context(&ctx->lock_context);
801 	ctx->lock_context.open_context = ctx;
802 	INIT_LIST_HEAD(&ctx->list);
803 	ctx->mdsthreshold = NULL;
804 	return ctx;
805 }
806 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
807 
get_nfs_open_context(struct nfs_open_context * ctx)808 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
809 {
810 	if (ctx != NULL)
811 		atomic_inc(&ctx->lock_context.count);
812 	return ctx;
813 }
814 EXPORT_SYMBOL_GPL(get_nfs_open_context);
815 
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)816 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
817 {
818 	struct inode *inode = d_inode(ctx->dentry);
819 	struct super_block *sb = ctx->dentry->d_sb;
820 
821 	if (!list_empty(&ctx->list)) {
822 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
823 			return;
824 		list_del(&ctx->list);
825 		spin_unlock(&inode->i_lock);
826 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
827 		return;
828 	if (inode != NULL)
829 		NFS_PROTO(inode)->close_context(ctx, is_sync);
830 	if (ctx->cred != NULL)
831 		put_rpccred(ctx->cred);
832 	dput(ctx->dentry);
833 	nfs_sb_deactive(sb);
834 	kfree(ctx->mdsthreshold);
835 	kfree(ctx);
836 }
837 
put_nfs_open_context(struct nfs_open_context * ctx)838 void put_nfs_open_context(struct nfs_open_context *ctx)
839 {
840 	__put_nfs_open_context(ctx, 0);
841 }
842 EXPORT_SYMBOL_GPL(put_nfs_open_context);
843 
844 /*
845  * Ensure that mmap has a recent RPC credential for use when writing out
846  * shared pages
847  */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)848 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
849 {
850 	struct inode *inode = d_inode(ctx->dentry);
851 	struct nfs_inode *nfsi = NFS_I(inode);
852 
853 	spin_lock(&inode->i_lock);
854 	list_add(&ctx->list, &nfsi->open_files);
855 	spin_unlock(&inode->i_lock);
856 }
857 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
858 
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)859 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
860 {
861 	filp->private_data = get_nfs_open_context(ctx);
862 	if (list_empty(&ctx->list))
863 		nfs_inode_attach_open_context(ctx);
864 }
865 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
866 
867 /*
868  * Given an inode, search for an open context with the desired characteristics
869  */
nfs_find_open_context(struct inode * inode,struct rpc_cred * cred,fmode_t mode)870 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
871 {
872 	struct nfs_inode *nfsi = NFS_I(inode);
873 	struct nfs_open_context *pos, *ctx = NULL;
874 
875 	spin_lock(&inode->i_lock);
876 	list_for_each_entry(pos, &nfsi->open_files, list) {
877 		if (cred != NULL && pos->cred != cred)
878 			continue;
879 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
880 			continue;
881 		ctx = get_nfs_open_context(pos);
882 		break;
883 	}
884 	spin_unlock(&inode->i_lock);
885 	return ctx;
886 }
887 
nfs_file_clear_open_context(struct file * filp)888 static void nfs_file_clear_open_context(struct file *filp)
889 {
890 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
891 
892 	if (ctx) {
893 		struct inode *inode = d_inode(ctx->dentry);
894 
895 		filp->private_data = NULL;
896 		spin_lock(&inode->i_lock);
897 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
898 		spin_unlock(&inode->i_lock);
899 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
900 	}
901 }
902 
903 /*
904  * These allocate and release file read/write context information.
905  */
nfs_open(struct inode * inode,struct file * filp)906 int nfs_open(struct inode *inode, struct file *filp)
907 {
908 	struct nfs_open_context *ctx;
909 
910 	ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
911 	if (IS_ERR(ctx))
912 		return PTR_ERR(ctx);
913 	nfs_file_set_open_context(filp, ctx);
914 	put_nfs_open_context(ctx);
915 	nfs_fscache_open_file(inode, filp);
916 	return 0;
917 }
918 
nfs_release(struct inode * inode,struct file * filp)919 int nfs_release(struct inode *inode, struct file *filp)
920 {
921 	nfs_file_clear_open_context(filp);
922 	return 0;
923 }
924 
925 /*
926  * This function is called whenever some part of NFS notices that
927  * the cached attributes have to be refreshed.
928  */
929 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)930 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
931 {
932 	int		 status = -ESTALE;
933 	struct nfs4_label *label = NULL;
934 	struct nfs_fattr *fattr = NULL;
935 	struct nfs_inode *nfsi = NFS_I(inode);
936 
937 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
938 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
939 
940 	trace_nfs_revalidate_inode_enter(inode);
941 
942 	if (is_bad_inode(inode))
943 		goto out;
944 	if (NFS_STALE(inode))
945 		goto out;
946 
947 	status = -ENOMEM;
948 	fattr = nfs_alloc_fattr();
949 	if (fattr == NULL)
950 		goto out;
951 
952 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
953 
954 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
955 	if (IS_ERR(label)) {
956 		status = PTR_ERR(label);
957 		goto out;
958 	}
959 
960 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
961 	if (status != 0) {
962 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
963 			 inode->i_sb->s_id,
964 			 (unsigned long long)NFS_FILEID(inode), status);
965 		if (status == -ESTALE) {
966 			nfs_zap_caches(inode);
967 			if (!S_ISDIR(inode->i_mode))
968 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
969 		}
970 		goto err_out;
971 	}
972 
973 	status = nfs_refresh_inode(inode, fattr);
974 	if (status) {
975 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
976 			 inode->i_sb->s_id,
977 			 (unsigned long long)NFS_FILEID(inode), status);
978 		goto err_out;
979 	}
980 
981 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
982 		nfs_zap_acl_cache(inode);
983 
984 	nfs_setsecurity(inode, fattr, label);
985 
986 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
987 		inode->i_sb->s_id,
988 		(unsigned long long)NFS_FILEID(inode));
989 
990 err_out:
991 	nfs4_label_free(label);
992 out:
993 	nfs_free_fattr(fattr);
994 	trace_nfs_revalidate_inode_exit(inode, status);
995 	return status;
996 }
997 
nfs_attribute_timeout(struct inode * inode)998 int nfs_attribute_timeout(struct inode *inode)
999 {
1000 	struct nfs_inode *nfsi = NFS_I(inode);
1001 
1002 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1003 }
1004 
nfs_attribute_cache_expired(struct inode * inode)1005 int nfs_attribute_cache_expired(struct inode *inode)
1006 {
1007 	if (nfs_have_delegated_attributes(inode))
1008 		return 0;
1009 	return nfs_attribute_timeout(inode);
1010 }
1011 
1012 /**
1013  * nfs_revalidate_inode - Revalidate the inode attributes
1014  * @server - pointer to nfs_server struct
1015  * @inode - pointer to inode struct
1016  *
1017  * Updates inode attribute information by retrieving the data from the server.
1018  */
nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1019 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1020 {
1021 	if (!nfs_need_revalidate_inode(inode))
1022 		return NFS_STALE(inode) ? -ESTALE : 0;
1023 	return __nfs_revalidate_inode(server, inode);
1024 }
1025 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1026 
nfs_revalidate_inode_rcu(struct nfs_server * server,struct inode * inode)1027 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1028 {
1029 	if (!(NFS_I(inode)->cache_validity &
1030 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1031 			&& !nfs_attribute_cache_expired(inode))
1032 		return NFS_STALE(inode) ? -ESTALE : 0;
1033 	return -ECHILD;
1034 }
1035 
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1036 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1037 {
1038 	struct nfs_inode *nfsi = NFS_I(inode);
1039 	int ret;
1040 
1041 	if (mapping->nrpages != 0) {
1042 		if (S_ISREG(inode->i_mode)) {
1043 			unmap_mapping_range(mapping, 0, 0, 0);
1044 			ret = nfs_sync_mapping(mapping);
1045 			if (ret < 0)
1046 				return ret;
1047 		}
1048 		ret = invalidate_inode_pages2(mapping);
1049 		if (ret < 0)
1050 			return ret;
1051 	}
1052 	if (S_ISDIR(inode->i_mode)) {
1053 		spin_lock(&inode->i_lock);
1054 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1055 		spin_unlock(&inode->i_lock);
1056 	}
1057 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1058 	nfs_fscache_wait_on_invalidate(inode);
1059 
1060 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1061 			inode->i_sb->s_id,
1062 			(unsigned long long)NFS_FILEID(inode));
1063 	return 0;
1064 }
1065 
nfs_mapping_need_revalidate_inode(struct inode * inode)1066 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1067 {
1068 	if (nfs_have_delegated_attributes(inode))
1069 		return false;
1070 	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1071 		|| nfs_attribute_timeout(inode)
1072 		|| NFS_STALE(inode);
1073 }
1074 
1075 /**
1076  * __nfs_revalidate_mapping - Revalidate the pagecache
1077  * @inode - pointer to host inode
1078  * @mapping - pointer to mapping
1079  * @may_lock - take inode->i_mutex?
1080  */
__nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping,bool may_lock)1081 static int __nfs_revalidate_mapping(struct inode *inode,
1082 		struct address_space *mapping,
1083 		bool may_lock)
1084 {
1085 	struct nfs_inode *nfsi = NFS_I(inode);
1086 	unsigned long *bitlock = &nfsi->flags;
1087 	int ret = 0;
1088 
1089 	/* swapfiles are not supposed to be shared. */
1090 	if (IS_SWAPFILE(inode))
1091 		goto out;
1092 
1093 	if (nfs_mapping_need_revalidate_inode(inode)) {
1094 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1095 		if (ret < 0)
1096 			goto out;
1097 	}
1098 
1099 	/*
1100 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1101 	 * invalidations that come in while we're shooting down the mappings
1102 	 * are respected. But, that leaves a race window where one revalidator
1103 	 * can clear the flag, and then another checks it before the mapping
1104 	 * gets invalidated. Fix that by serializing access to this part of
1105 	 * the function.
1106 	 *
1107 	 * At the same time, we need to allow other tasks to see whether we
1108 	 * might be in the middle of invalidating the pages, so we only set
1109 	 * the bit lock here if it looks like we're going to be doing that.
1110 	 */
1111 	for (;;) {
1112 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1113 					 nfs_wait_bit_killable, TASK_KILLABLE);
1114 		if (ret)
1115 			goto out;
1116 		spin_lock(&inode->i_lock);
1117 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1118 			spin_unlock(&inode->i_lock);
1119 			continue;
1120 		}
1121 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1122 			break;
1123 		spin_unlock(&inode->i_lock);
1124 		goto out;
1125 	}
1126 
1127 	set_bit(NFS_INO_INVALIDATING, bitlock);
1128 	smp_wmb();
1129 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1130 	spin_unlock(&inode->i_lock);
1131 	trace_nfs_invalidate_mapping_enter(inode);
1132 	if (may_lock) {
1133 		mutex_lock(&inode->i_mutex);
1134 		ret = nfs_invalidate_mapping(inode, mapping);
1135 		mutex_unlock(&inode->i_mutex);
1136 	} else
1137 		ret = nfs_invalidate_mapping(inode, mapping);
1138 	trace_nfs_invalidate_mapping_exit(inode, ret);
1139 
1140 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1141 	smp_mb__after_atomic();
1142 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1143 out:
1144 	return ret;
1145 }
1146 
1147 /**
1148  * nfs_revalidate_mapping - Revalidate the pagecache
1149  * @inode - pointer to host inode
1150  * @mapping - pointer to mapping
1151  */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1152 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1153 {
1154 	return __nfs_revalidate_mapping(inode, mapping, false);
1155 }
1156 
1157 /**
1158  * nfs_revalidate_mapping_protected - Revalidate the pagecache
1159  * @inode - pointer to host inode
1160  * @mapping - pointer to mapping
1161  *
1162  * Differs from nfs_revalidate_mapping() in that it grabs the inode->i_mutex
1163  * while invalidating the mapping.
1164  */
nfs_revalidate_mapping_protected(struct inode * inode,struct address_space * mapping)1165 int nfs_revalidate_mapping_protected(struct inode *inode, struct address_space *mapping)
1166 {
1167 	return __nfs_revalidate_mapping(inode, mapping, true);
1168 }
1169 
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1170 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1171 {
1172 	struct nfs_inode *nfsi = NFS_I(inode);
1173 	unsigned long ret = 0;
1174 
1175 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1176 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1177 			&& inode->i_version == fattr->pre_change_attr) {
1178 		inode->i_version = fattr->change_attr;
1179 		if (S_ISDIR(inode->i_mode))
1180 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1181 		ret |= NFS_INO_INVALID_ATTR;
1182 	}
1183 	/* If we have atomic WCC data, we may update some attributes */
1184 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1185 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1186 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1187 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1188 		ret |= NFS_INO_INVALID_ATTR;
1189 	}
1190 
1191 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1192 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1193 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1194 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1195 		if (S_ISDIR(inode->i_mode))
1196 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1197 		ret |= NFS_INO_INVALID_ATTR;
1198 	}
1199 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1200 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1201 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1202 			&& nfsi->nrequests == 0) {
1203 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1204 		ret |= NFS_INO_INVALID_ATTR;
1205 	}
1206 
1207 	return ret;
1208 }
1209 
1210 /**
1211  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1212  * @inode - pointer to inode
1213  * @fattr - updated attributes
1214  *
1215  * Verifies the attribute cache. If we have just changed the attributes,
1216  * so that fattr carries weak cache consistency data, then it may
1217  * also update the ctime/mtime/change_attribute.
1218  */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1219 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1220 {
1221 	struct nfs_inode *nfsi = NFS_I(inode);
1222 	loff_t cur_size, new_isize;
1223 	unsigned long invalid = 0;
1224 
1225 
1226 	if (nfs_have_delegated_attributes(inode))
1227 		return 0;
1228 	/* Has the inode gone and changed behind our back? */
1229 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1230 		return -EIO;
1231 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1232 		return -EIO;
1233 
1234 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1235 			inode->i_version != fattr->change_attr)
1236 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1237 
1238 	/* Verify a few of the more important attributes */
1239 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1240 		invalid |= NFS_INO_INVALID_ATTR;
1241 
1242 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1243 		cur_size = i_size_read(inode);
1244 		new_isize = nfs_size_to_loff_t(fattr->size);
1245 		if (cur_size != new_isize)
1246 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1247 	}
1248 	if (nfsi->nrequests != 0)
1249 		invalid &= ~NFS_INO_REVAL_PAGECACHE;
1250 
1251 	/* Have any file permissions changed? */
1252 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1253 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1254 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1255 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1256 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1257 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1258 
1259 	/* Has the link count changed? */
1260 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1261 		invalid |= NFS_INO_INVALID_ATTR;
1262 
1263 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1264 		invalid |= NFS_INO_INVALID_ATIME;
1265 
1266 	if (invalid != 0)
1267 		nfs_set_cache_invalid(inode, invalid);
1268 
1269 	nfsi->read_cache_jiffies = fattr->time_start;
1270 	return 0;
1271 }
1272 
1273 static atomic_long_t nfs_attr_generation_counter;
1274 
nfs_read_attr_generation_counter(void)1275 static unsigned long nfs_read_attr_generation_counter(void)
1276 {
1277 	return atomic_long_read(&nfs_attr_generation_counter);
1278 }
1279 
nfs_inc_attr_generation_counter(void)1280 unsigned long nfs_inc_attr_generation_counter(void)
1281 {
1282 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1283 }
1284 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1285 
nfs_fattr_init(struct nfs_fattr * fattr)1286 void nfs_fattr_init(struct nfs_fattr *fattr)
1287 {
1288 	fattr->valid = 0;
1289 	fattr->time_start = jiffies;
1290 	fattr->gencount = nfs_inc_attr_generation_counter();
1291 	fattr->owner_name = NULL;
1292 	fattr->group_name = NULL;
1293 }
1294 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1295 
1296 /**
1297  * nfs_fattr_set_barrier
1298  * @fattr: attributes
1299  *
1300  * Used to set a barrier after an attribute was updated. This
1301  * barrier ensures that older attributes from RPC calls that may
1302  * have raced with our update cannot clobber these new values.
1303  * Note that you are still responsible for ensuring that other
1304  * operations which change the attribute on the server do not
1305  * collide.
1306  */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1307 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1308 {
1309 	fattr->gencount = nfs_inc_attr_generation_counter();
1310 }
1311 
nfs_alloc_fattr(void)1312 struct nfs_fattr *nfs_alloc_fattr(void)
1313 {
1314 	struct nfs_fattr *fattr;
1315 
1316 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1317 	if (fattr != NULL)
1318 		nfs_fattr_init(fattr);
1319 	return fattr;
1320 }
1321 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1322 
nfs_alloc_fhandle(void)1323 struct nfs_fh *nfs_alloc_fhandle(void)
1324 {
1325 	struct nfs_fh *fh;
1326 
1327 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1328 	if (fh != NULL)
1329 		fh->size = 0;
1330 	return fh;
1331 }
1332 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1333 
1334 #ifdef NFS_DEBUG
1335 /*
1336  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1337  *                             in the same way that wireshark does
1338  *
1339  * @fh: file handle
1340  *
1341  * For debugging only.
1342  */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1343 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1344 {
1345 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1346 	 * not on the result */
1347 	return nfs_fhandle_hash(fh);
1348 }
1349 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1350 
1351 /*
1352  * _nfs_display_fhandle - display an NFS file handle on the console
1353  *
1354  * @fh: file handle to display
1355  * @caption: display caption
1356  *
1357  * For debugging only.
1358  */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1359 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1360 {
1361 	unsigned short i;
1362 
1363 	if (fh == NULL || fh->size == 0) {
1364 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1365 		return;
1366 	}
1367 
1368 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1369 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1370 	for (i = 0; i < fh->size; i += 16) {
1371 		__be32 *pos = (__be32 *)&fh->data[i];
1372 
1373 		switch ((fh->size - i - 1) >> 2) {
1374 		case 0:
1375 			printk(KERN_DEFAULT " %08x\n",
1376 				be32_to_cpup(pos));
1377 			break;
1378 		case 1:
1379 			printk(KERN_DEFAULT " %08x %08x\n",
1380 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1381 			break;
1382 		case 2:
1383 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1384 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1385 				be32_to_cpup(pos + 2));
1386 			break;
1387 		default:
1388 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1389 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1390 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1391 		}
1392 	}
1393 }
1394 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1395 #endif
1396 
1397 /**
1398  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1399  * @inode - pointer to inode
1400  * @fattr - attributes
1401  *
1402  * Attempt to divine whether or not an RPC call reply carrying stale
1403  * attributes got scheduled after another call carrying updated ones.
1404  *
1405  * To do so, the function first assumes that a more recent ctime means
1406  * that the attributes in fattr are newer, however it also attempt to
1407  * catch the case where ctime either didn't change, or went backwards
1408  * (if someone reset the clock on the server) by looking at whether
1409  * or not this RPC call was started after the inode was last updated.
1410  * Note also the check for wraparound of 'attr_gencount'
1411  *
1412  * The function returns 'true' if it thinks the attributes in 'fattr' are
1413  * more recent than the ones cached in the inode.
1414  *
1415  */
nfs_inode_attrs_need_update(const struct inode * inode,const struct nfs_fattr * fattr)1416 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1417 {
1418 	const struct nfs_inode *nfsi = NFS_I(inode);
1419 
1420 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1421 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1422 }
1423 
1424 /*
1425  * Don't trust the change_attribute, mtime, ctime or size if
1426  * a pnfs LAYOUTCOMMIT is outstanding
1427  */
nfs_inode_attrs_handle_layoutcommit(struct inode * inode,struct nfs_fattr * fattr)1428 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1429 		struct nfs_fattr *fattr)
1430 {
1431 	if (pnfs_layoutcommit_outstanding(inode))
1432 		fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1433 				NFS_ATTR_FATTR_MTIME |
1434 				NFS_ATTR_FATTR_CTIME |
1435 				NFS_ATTR_FATTR_SIZE);
1436 }
1437 
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1438 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1439 {
1440 	int ret;
1441 
1442 	trace_nfs_refresh_inode_enter(inode);
1443 
1444 	nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1445 
1446 	if (nfs_inode_attrs_need_update(inode, fattr))
1447 		ret = nfs_update_inode(inode, fattr);
1448 	else
1449 		ret = nfs_check_inode_attributes(inode, fattr);
1450 
1451 	trace_nfs_refresh_inode_exit(inode, ret);
1452 	return ret;
1453 }
1454 
1455 /**
1456  * nfs_refresh_inode - try to update the inode attribute cache
1457  * @inode - pointer to inode
1458  * @fattr - updated attributes
1459  *
1460  * Check that an RPC call that returned attributes has not overlapped with
1461  * other recent updates of the inode metadata, then decide whether it is
1462  * safe to do a full update of the inode attributes, or whether just to
1463  * call nfs_check_inode_attributes.
1464  */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)1465 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1466 {
1467 	int status;
1468 
1469 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1470 		return 0;
1471 	spin_lock(&inode->i_lock);
1472 	status = nfs_refresh_inode_locked(inode, fattr);
1473 	spin_unlock(&inode->i_lock);
1474 
1475 	return status;
1476 }
1477 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1478 
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1479 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1480 {
1481 	unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1482 
1483 	/*
1484 	 * Don't revalidate the pagecache if we hold a delegation, but do
1485 	 * force an attribute update
1486 	 */
1487 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1488 		invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_FORCED;
1489 
1490 	if (S_ISDIR(inode->i_mode))
1491 		invalid |= NFS_INO_INVALID_DATA;
1492 	nfs_set_cache_invalid(inode, invalid);
1493 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1494 		return 0;
1495 	return nfs_refresh_inode_locked(inode, fattr);
1496 }
1497 
1498 /**
1499  * nfs_post_op_update_inode - try to update the inode attribute cache
1500  * @inode - pointer to inode
1501  * @fattr - updated attributes
1502  *
1503  * After an operation that has changed the inode metadata, mark the
1504  * attribute cache as being invalid, then try to update it.
1505  *
1506  * NB: if the server didn't return any post op attributes, this
1507  * function will force the retrieval of attributes before the next
1508  * NFS request.  Thus it should be used only for operations that
1509  * are expected to change one or more attributes, to avoid
1510  * unnecessary NFS requests and trips through nfs_update_inode().
1511  */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)1512 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1513 {
1514 	int status;
1515 
1516 	spin_lock(&inode->i_lock);
1517 	nfs_fattr_set_barrier(fattr);
1518 	status = nfs_post_op_update_inode_locked(inode, fattr);
1519 	spin_unlock(&inode->i_lock);
1520 
1521 	return status;
1522 }
1523 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1524 
1525 /**
1526  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1527  * @inode - pointer to inode
1528  * @fattr - updated attributes
1529  *
1530  * After an operation that has changed the inode metadata, mark the
1531  * attribute cache as being invalid, then try to update it. Fake up
1532  * weak cache consistency data, if none exist.
1533  *
1534  * This function is mainly designed to be used by the ->write_done() functions.
1535  */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)1536 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1537 {
1538 	int status;
1539 
1540 	/* Don't do a WCC update if these attributes are already stale */
1541 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1542 			!nfs_inode_attrs_need_update(inode, fattr)) {
1543 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1544 				| NFS_ATTR_FATTR_PRESIZE
1545 				| NFS_ATTR_FATTR_PREMTIME
1546 				| NFS_ATTR_FATTR_PRECTIME);
1547 		goto out_noforce;
1548 	}
1549 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1550 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1551 		fattr->pre_change_attr = inode->i_version;
1552 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1553 	}
1554 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1555 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1556 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1557 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1558 	}
1559 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1560 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1561 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1562 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1563 	}
1564 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1565 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1566 		fattr->pre_size = i_size_read(inode);
1567 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1568 	}
1569 out_noforce:
1570 	status = nfs_post_op_update_inode_locked(inode, fattr);
1571 	return status;
1572 }
1573 
1574 /**
1575  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1576  * @inode - pointer to inode
1577  * @fattr - updated attributes
1578  *
1579  * After an operation that has changed the inode metadata, mark the
1580  * attribute cache as being invalid, then try to update it. Fake up
1581  * weak cache consistency data, if none exist.
1582  *
1583  * This function is mainly designed to be used by the ->write_done() functions.
1584  */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)1585 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1586 {
1587 	int status;
1588 
1589 	spin_lock(&inode->i_lock);
1590 	nfs_fattr_set_barrier(fattr);
1591 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1592 	spin_unlock(&inode->i_lock);
1593 	return status;
1594 }
1595 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1596 
1597 
nfs_fileid_valid(struct nfs_inode * nfsi,struct nfs_fattr * fattr)1598 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1599 				    struct nfs_fattr *fattr)
1600 {
1601 	bool ret1 = true, ret2 = true;
1602 
1603 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1604 		ret1 = (nfsi->fileid == fattr->fileid);
1605 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1606 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1607 	return ret1 || ret2;
1608 }
1609 
1610 /*
1611  * Many nfs protocol calls return the new file attributes after
1612  * an operation.  Here we update the inode to reflect the state
1613  * of the server's inode.
1614  *
1615  * This is a bit tricky because we have to make sure all dirty pages
1616  * have been sent off to the server before calling invalidate_inode_pages.
1617  * To make sure no other process adds more write requests while we try
1618  * our best to flush them, we make them sleep during the attribute refresh.
1619  *
1620  * A very similar scenario holds for the dir cache.
1621  */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)1622 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1623 {
1624 	struct nfs_server *server;
1625 	struct nfs_inode *nfsi = NFS_I(inode);
1626 	loff_t cur_isize, new_isize;
1627 	unsigned long invalid = 0;
1628 	unsigned long now = jiffies;
1629 	unsigned long save_cache_validity;
1630 	bool cache_revalidated = true;
1631 
1632 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1633 			__func__, inode->i_sb->s_id, inode->i_ino,
1634 			nfs_display_fhandle_hash(NFS_FH(inode)),
1635 			atomic_read(&inode->i_count), fattr->valid);
1636 
1637 	if (!nfs_fileid_valid(nfsi, fattr)) {
1638 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1639 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1640 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1641 			inode->i_sb->s_id, (long long)nfsi->fileid,
1642 			(long long)fattr->fileid);
1643 		goto out_err;
1644 	}
1645 
1646 	/*
1647 	 * Make sure the inode's type hasn't changed.
1648 	 */
1649 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1650 		/*
1651 		* Big trouble! The inode has become a different object.
1652 		*/
1653 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1654 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1655 		goto out_err;
1656 	}
1657 
1658 	server = NFS_SERVER(inode);
1659 	/* Update the fsid? */
1660 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1661 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1662 			!IS_AUTOMOUNT(inode))
1663 		server->fsid = fattr->fsid;
1664 
1665 	/*
1666 	 * Update the read time so we don't revalidate too often.
1667 	 */
1668 	nfsi->read_cache_jiffies = fattr->time_start;
1669 
1670 	save_cache_validity = nfsi->cache_validity;
1671 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1672 			| NFS_INO_INVALID_ATIME
1673 			| NFS_INO_REVAL_FORCED
1674 			| NFS_INO_REVAL_PAGECACHE);
1675 
1676 	/* Do atomic weak cache consistency updates */
1677 	invalid |= nfs_wcc_update_inode(inode, fattr);
1678 
1679 	/* More cache consistency checks */
1680 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1681 		if (inode->i_version != fattr->change_attr) {
1682 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1683 					inode->i_sb->s_id, inode->i_ino);
1684 			invalid |= NFS_INO_INVALID_ATTR
1685 				| NFS_INO_INVALID_DATA
1686 				| NFS_INO_INVALID_ACCESS
1687 				| NFS_INO_INVALID_ACL;
1688 			if (S_ISDIR(inode->i_mode))
1689 				nfs_force_lookup_revalidate(inode);
1690 			inode->i_version = fattr->change_attr;
1691 		}
1692 	} else {
1693 		nfsi->cache_validity |= save_cache_validity;
1694 		cache_revalidated = false;
1695 	}
1696 
1697 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1698 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1699 	} else if (server->caps & NFS_CAP_MTIME) {
1700 		nfsi->cache_validity |= save_cache_validity &
1701 				(NFS_INO_INVALID_ATTR
1702 				| NFS_INO_REVAL_FORCED);
1703 		cache_revalidated = false;
1704 	}
1705 
1706 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1707 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1708 	} else if (server->caps & NFS_CAP_CTIME) {
1709 		nfsi->cache_validity |= save_cache_validity &
1710 				(NFS_INO_INVALID_ATTR
1711 				| NFS_INO_REVAL_FORCED);
1712 		cache_revalidated = false;
1713 	}
1714 
1715 	/* Check if our cached file size is stale */
1716 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1717 		new_isize = nfs_size_to_loff_t(fattr->size);
1718 		cur_isize = i_size_read(inode);
1719 		if (new_isize != cur_isize) {
1720 			/* Do we perhaps have any outstanding writes, or has
1721 			 * the file grown beyond our last write? */
1722 			if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1723 				i_size_write(inode, new_isize);
1724 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1725 			}
1726 			dprintk("NFS: isize change on server for file %s/%ld "
1727 					"(%Ld to %Ld)\n",
1728 					inode->i_sb->s_id,
1729 					inode->i_ino,
1730 					(long long)cur_isize,
1731 					(long long)new_isize);
1732 		}
1733 	} else {
1734 		nfsi->cache_validity |= save_cache_validity &
1735 				(NFS_INO_INVALID_ATTR
1736 				| NFS_INO_REVAL_PAGECACHE
1737 				| NFS_INO_REVAL_FORCED);
1738 		cache_revalidated = false;
1739 	}
1740 
1741 
1742 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1743 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1744 	else if (server->caps & NFS_CAP_ATIME) {
1745 		nfsi->cache_validity |= save_cache_validity &
1746 				(NFS_INO_INVALID_ATIME
1747 				| NFS_INO_REVAL_FORCED);
1748 		cache_revalidated = false;
1749 	}
1750 
1751 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1752 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1753 			umode_t newmode = inode->i_mode & S_IFMT;
1754 			newmode |= fattr->mode & S_IALLUGO;
1755 			inode->i_mode = newmode;
1756 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1757 		}
1758 	} else if (server->caps & NFS_CAP_MODE) {
1759 		nfsi->cache_validity |= save_cache_validity &
1760 				(NFS_INO_INVALID_ATTR
1761 				| NFS_INO_INVALID_ACCESS
1762 				| NFS_INO_INVALID_ACL
1763 				| NFS_INO_REVAL_FORCED);
1764 		cache_revalidated = false;
1765 	}
1766 
1767 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1768 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1769 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1770 			inode->i_uid = fattr->uid;
1771 		}
1772 	} else if (server->caps & NFS_CAP_OWNER) {
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_GROUP) {
1782 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1783 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1784 			inode->i_gid = fattr->gid;
1785 		}
1786 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
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_NLINK) {
1796 		if (inode->i_nlink != fattr->nlink) {
1797 			invalid |= NFS_INO_INVALID_ATTR;
1798 			if (S_ISDIR(inode->i_mode))
1799 				invalid |= NFS_INO_INVALID_DATA;
1800 			set_nlink(inode, fattr->nlink);
1801 		}
1802 	} else if (server->caps & NFS_CAP_NLINK) {
1803 		nfsi->cache_validity |= save_cache_validity &
1804 				(NFS_INO_INVALID_ATTR
1805 				| NFS_INO_REVAL_FORCED);
1806 		cache_revalidated = false;
1807 	}
1808 
1809 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1810 		/*
1811 		 * report the blocks in 512byte units
1812 		 */
1813 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1814 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1815 		inode->i_blocks = fattr->du.nfs2.blocks;
1816 	else
1817 		cache_revalidated = false;
1818 
1819 	/* Update attrtimeo value if we're out of the unstable period */
1820 	if (invalid & NFS_INO_INVALID_ATTR) {
1821 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1822 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1823 		nfsi->attrtimeo_timestamp = now;
1824 		/* Set barrier to be more recent than all outstanding updates */
1825 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1826 	} else {
1827 		if (cache_revalidated) {
1828 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1829 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1830 				nfsi->attrtimeo <<= 1;
1831 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1832 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1833 			}
1834 			nfsi->attrtimeo_timestamp = now;
1835 		}
1836 		/* Set the barrier to be more recent than this fattr */
1837 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1838 			nfsi->attr_gencount = fattr->gencount;
1839 	}
1840 
1841 	/* Don't declare attrcache up to date if there were no attrs! */
1842 	if (cache_revalidated)
1843 		invalid &= ~NFS_INO_INVALID_ATTR;
1844 
1845 	/* Don't invalidate the data if we were to blame */
1846 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1847 				|| S_ISLNK(inode->i_mode)))
1848 		invalid &= ~NFS_INO_INVALID_DATA;
1849 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1850 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1851 		nfs_set_cache_invalid(inode, invalid);
1852 
1853 	return 0;
1854  out_err:
1855 	/*
1856 	 * No need to worry about unhashing the dentry, as the
1857 	 * lookup validation will know that the inode is bad.
1858 	 * (But we fall through to invalidate the caches.)
1859 	 */
1860 	nfs_invalidate_inode(inode);
1861 	return -ESTALE;
1862 }
1863 
nfs_alloc_inode(struct super_block * sb)1864 struct inode *nfs_alloc_inode(struct super_block *sb)
1865 {
1866 	struct nfs_inode *nfsi;
1867 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1868 	if (!nfsi)
1869 		return NULL;
1870 	nfsi->flags = 0UL;
1871 	nfsi->cache_validity = 0UL;
1872 #if IS_ENABLED(CONFIG_NFS_V4)
1873 	nfsi->nfs4_acl = NULL;
1874 #endif /* CONFIG_NFS_V4 */
1875 	return &nfsi->vfs_inode;
1876 }
1877 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1878 
nfs_i_callback(struct rcu_head * head)1879 static void nfs_i_callback(struct rcu_head *head)
1880 {
1881 	struct inode *inode = container_of(head, struct inode, i_rcu);
1882 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1883 }
1884 
nfs_destroy_inode(struct inode * inode)1885 void nfs_destroy_inode(struct inode *inode)
1886 {
1887 	call_rcu(&inode->i_rcu, nfs_i_callback);
1888 }
1889 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1890 
nfs4_init_once(struct nfs_inode * nfsi)1891 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1892 {
1893 #if IS_ENABLED(CONFIG_NFS_V4)
1894 	INIT_LIST_HEAD(&nfsi->open_states);
1895 	nfsi->delegation = NULL;
1896 	init_rwsem(&nfsi->rwsem);
1897 	nfsi->layout = NULL;
1898 #endif
1899 }
1900 
init_once(void * foo)1901 static void init_once(void *foo)
1902 {
1903 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1904 
1905 	inode_init_once(&nfsi->vfs_inode);
1906 	INIT_LIST_HEAD(&nfsi->open_files);
1907 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1908 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1909 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1910 	nfsi->nrequests = 0;
1911 	nfsi->commit_info.ncommit = 0;
1912 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1913 	atomic_set(&nfsi->silly_count, 1);
1914 	INIT_HLIST_HEAD(&nfsi->silly_list);
1915 	init_waitqueue_head(&nfsi->waitqueue);
1916 	nfs4_init_once(nfsi);
1917 }
1918 
nfs_init_inodecache(void)1919 static int __init nfs_init_inodecache(void)
1920 {
1921 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1922 					     sizeof(struct nfs_inode),
1923 					     0, (SLAB_RECLAIM_ACCOUNT|
1924 						SLAB_MEM_SPREAD),
1925 					     init_once);
1926 	if (nfs_inode_cachep == NULL)
1927 		return -ENOMEM;
1928 
1929 	return 0;
1930 }
1931 
nfs_destroy_inodecache(void)1932 static void nfs_destroy_inodecache(void)
1933 {
1934 	/*
1935 	 * Make sure all delayed rcu free inodes are flushed before we
1936 	 * destroy cache.
1937 	 */
1938 	rcu_barrier();
1939 	kmem_cache_destroy(nfs_inode_cachep);
1940 }
1941 
1942 struct workqueue_struct *nfsiod_workqueue;
1943 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1944 
1945 /*
1946  * start up the nfsiod workqueue
1947  */
nfsiod_start(void)1948 static int nfsiod_start(void)
1949 {
1950 	struct workqueue_struct *wq;
1951 	dprintk("RPC:       creating workqueue nfsiod\n");
1952 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1953 	if (wq == NULL)
1954 		return -ENOMEM;
1955 	nfsiod_workqueue = wq;
1956 	return 0;
1957 }
1958 
1959 /*
1960  * Destroy the nfsiod workqueue
1961  */
nfsiod_stop(void)1962 static void nfsiod_stop(void)
1963 {
1964 	struct workqueue_struct *wq;
1965 
1966 	wq = nfsiod_workqueue;
1967 	if (wq == NULL)
1968 		return;
1969 	nfsiod_workqueue = NULL;
1970 	destroy_workqueue(wq);
1971 }
1972 
1973 int nfs_net_id;
1974 EXPORT_SYMBOL_GPL(nfs_net_id);
1975 
nfs_net_init(struct net * net)1976 static int nfs_net_init(struct net *net)
1977 {
1978 	nfs_clients_init(net);
1979 	return nfs_fs_proc_net_init(net);
1980 }
1981 
nfs_net_exit(struct net * net)1982 static void nfs_net_exit(struct net *net)
1983 {
1984 	nfs_fs_proc_net_exit(net);
1985 	nfs_cleanup_cb_ident_idr(net);
1986 }
1987 
1988 static struct pernet_operations nfs_net_ops = {
1989 	.init = nfs_net_init,
1990 	.exit = nfs_net_exit,
1991 	.id   = &nfs_net_id,
1992 	.size = sizeof(struct nfs_net),
1993 };
1994 
1995 /*
1996  * Initialize NFS
1997  */
init_nfs_fs(void)1998 static int __init init_nfs_fs(void)
1999 {
2000 	int err;
2001 
2002 	err = register_pernet_subsys(&nfs_net_ops);
2003 	if (err < 0)
2004 		goto out9;
2005 
2006 	err = nfs_fscache_register();
2007 	if (err < 0)
2008 		goto out8;
2009 
2010 	err = nfsiod_start();
2011 	if (err)
2012 		goto out7;
2013 
2014 	err = nfs_fs_proc_init();
2015 	if (err)
2016 		goto out6;
2017 
2018 	err = nfs_init_nfspagecache();
2019 	if (err)
2020 		goto out5;
2021 
2022 	err = nfs_init_inodecache();
2023 	if (err)
2024 		goto out4;
2025 
2026 	err = nfs_init_readpagecache();
2027 	if (err)
2028 		goto out3;
2029 
2030 	err = nfs_init_writepagecache();
2031 	if (err)
2032 		goto out2;
2033 
2034 	err = nfs_init_directcache();
2035 	if (err)
2036 		goto out1;
2037 
2038 #ifdef CONFIG_PROC_FS
2039 	rpc_proc_register(&init_net, &nfs_rpcstat);
2040 #endif
2041 	if ((err = register_nfs_fs()) != 0)
2042 		goto out0;
2043 
2044 	return 0;
2045 out0:
2046 #ifdef CONFIG_PROC_FS
2047 	rpc_proc_unregister(&init_net, "nfs");
2048 #endif
2049 	nfs_destroy_directcache();
2050 out1:
2051 	nfs_destroy_writepagecache();
2052 out2:
2053 	nfs_destroy_readpagecache();
2054 out3:
2055 	nfs_destroy_inodecache();
2056 out4:
2057 	nfs_destroy_nfspagecache();
2058 out5:
2059 	nfs_fs_proc_exit();
2060 out6:
2061 	nfsiod_stop();
2062 out7:
2063 	nfs_fscache_unregister();
2064 out8:
2065 	unregister_pernet_subsys(&nfs_net_ops);
2066 out9:
2067 	return err;
2068 }
2069 
exit_nfs_fs(void)2070 static void __exit exit_nfs_fs(void)
2071 {
2072 	nfs_destroy_directcache();
2073 	nfs_destroy_writepagecache();
2074 	nfs_destroy_readpagecache();
2075 	nfs_destroy_inodecache();
2076 	nfs_destroy_nfspagecache();
2077 	nfs_fscache_unregister();
2078 	unregister_pernet_subsys(&nfs_net_ops);
2079 #ifdef CONFIG_PROC_FS
2080 	rpc_proc_unregister(&init_net, "nfs");
2081 #endif
2082 	unregister_nfs_fs();
2083 	nfs_fs_proc_exit();
2084 	nfsiod_stop();
2085 }
2086 
2087 /* Not quite true; I just maintain it */
2088 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2089 MODULE_LICENSE("GPL");
2090 module_param(enable_ino64, bool, 0644);
2091 
2092 module_init(init_nfs_fs)
2093 module_exit(exit_nfs_fs)
2094