xref: /openbmc/linux/fs/nfs/dir.c (revision facc3530fb5c89a40bc83045422add392b8db4a1)
1 /*
2  *  linux/fs/nfs/dir.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  nfs directory handling functions
7  *
8  * 10 Apr 1996	Added silly rename for unlink	--okir
9  * 28 Sep 1996	Improved directory cache --okir
10  * 23 Aug 1997  Claus Heine claus@momo.math.rwth-aachen.de
11  *              Re-implemented silly rename for unlink, newly implemented
12  *              silly rename for nfs_rename() following the suggestions
13  *              of Olaf Kirch (okir) found in this file.
14  *              Following Linus comments on my original hack, this version
15  *              depends only on the dcache stuff and doesn't touch the inode
16  *              layer (iput() and friends).
17  *  6 Jun 1999	Cache readdir lookups in the page cache. -DaveM
18  */
19 
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
27 #include <linux/mm.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/kmemleak.h>
37 #include <linux/xattr.h>
38 
39 #include "delegation.h"
40 #include "iostat.h"
41 #include "internal.h"
42 #include "fscache.h"
43 
44 /* #define NFS_DEBUG_VERBOSE 1 */
45 
46 static int nfs_opendir(struct inode *, struct file *);
47 static int nfs_closedir(struct inode *, struct file *);
48 static int nfs_readdir(struct file *, void *, filldir_t);
49 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
50 static int nfs_create(struct inode *, struct dentry *, umode_t, struct nameidata *);
51 static int nfs_mkdir(struct inode *, struct dentry *, umode_t);
52 static int nfs_rmdir(struct inode *, struct dentry *);
53 static int nfs_unlink(struct inode *, struct dentry *);
54 static int nfs_symlink(struct inode *, struct dentry *, const char *);
55 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
56 static int nfs_mknod(struct inode *, struct dentry *, umode_t, dev_t);
57 static int nfs_rename(struct inode *, struct dentry *,
58 		      struct inode *, struct dentry *);
59 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
60 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
61 static void nfs_readdir_clear_array(struct page*);
62 
63 const struct file_operations nfs_dir_operations = {
64 	.llseek		= nfs_llseek_dir,
65 	.read		= generic_read_dir,
66 	.readdir	= nfs_readdir,
67 	.open		= nfs_opendir,
68 	.release	= nfs_closedir,
69 	.fsync		= nfs_fsync_dir,
70 };
71 
72 const struct inode_operations nfs_dir_inode_operations = {
73 	.create		= nfs_create,
74 	.lookup		= nfs_lookup,
75 	.link		= nfs_link,
76 	.unlink		= nfs_unlink,
77 	.symlink	= nfs_symlink,
78 	.mkdir		= nfs_mkdir,
79 	.rmdir		= nfs_rmdir,
80 	.mknod		= nfs_mknod,
81 	.rename		= nfs_rename,
82 	.permission	= nfs_permission,
83 	.getattr	= nfs_getattr,
84 	.setattr	= nfs_setattr,
85 };
86 
87 const struct address_space_operations nfs_dir_aops = {
88 	.freepage = nfs_readdir_clear_array,
89 };
90 
91 #ifdef CONFIG_NFS_V3
92 const struct inode_operations nfs3_dir_inode_operations = {
93 	.create		= nfs_create,
94 	.lookup		= nfs_lookup,
95 	.link		= nfs_link,
96 	.unlink		= nfs_unlink,
97 	.symlink	= nfs_symlink,
98 	.mkdir		= nfs_mkdir,
99 	.rmdir		= nfs_rmdir,
100 	.mknod		= nfs_mknod,
101 	.rename		= nfs_rename,
102 	.permission	= nfs_permission,
103 	.getattr	= nfs_getattr,
104 	.setattr	= nfs_setattr,
105 	.listxattr	= nfs3_listxattr,
106 	.getxattr	= nfs3_getxattr,
107 	.setxattr	= nfs3_setxattr,
108 	.removexattr	= nfs3_removexattr,
109 };
110 #endif  /* CONFIG_NFS_V3 */
111 
112 #ifdef CONFIG_NFS_V4
113 
114 static int nfs_atomic_open(struct inode *, struct dentry *,
115 			   struct file *, unsigned, umode_t,
116 			   int *);
117 const struct inode_operations nfs4_dir_inode_operations = {
118 	.create		= nfs_create,
119 	.lookup		= nfs_lookup,
120 	.atomic_open	= nfs_atomic_open,
121 	.link		= nfs_link,
122 	.unlink		= nfs_unlink,
123 	.symlink	= nfs_symlink,
124 	.mkdir		= nfs_mkdir,
125 	.rmdir		= nfs_rmdir,
126 	.mknod		= nfs_mknod,
127 	.rename		= nfs_rename,
128 	.permission	= nfs_permission,
129 	.getattr	= nfs_getattr,
130 	.setattr	= nfs_setattr,
131 	.getxattr	= generic_getxattr,
132 	.setxattr	= generic_setxattr,
133 	.listxattr	= generic_listxattr,
134 	.removexattr	= generic_removexattr,
135 };
136 
137 #endif /* CONFIG_NFS_V4 */
138 
139 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
140 {
141 	struct nfs_open_dir_context *ctx;
142 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
143 	if (ctx != NULL) {
144 		ctx->duped = 0;
145 		ctx->attr_gencount = NFS_I(dir)->attr_gencount;
146 		ctx->dir_cookie = 0;
147 		ctx->dup_cookie = 0;
148 		ctx->cred = get_rpccred(cred);
149 		return ctx;
150 	}
151 	return  ERR_PTR(-ENOMEM);
152 }
153 
154 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
155 {
156 	put_rpccred(ctx->cred);
157 	kfree(ctx);
158 }
159 
160 /*
161  * Open file
162  */
163 static int
164 nfs_opendir(struct inode *inode, struct file *filp)
165 {
166 	int res = 0;
167 	struct nfs_open_dir_context *ctx;
168 	struct rpc_cred *cred;
169 
170 	dfprintk(FILE, "NFS: open dir(%s/%s)\n",
171 			filp->f_path.dentry->d_parent->d_name.name,
172 			filp->f_path.dentry->d_name.name);
173 
174 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
175 
176 	cred = rpc_lookup_cred();
177 	if (IS_ERR(cred))
178 		return PTR_ERR(cred);
179 	ctx = alloc_nfs_open_dir_context(inode, cred);
180 	if (IS_ERR(ctx)) {
181 		res = PTR_ERR(ctx);
182 		goto out;
183 	}
184 	filp->private_data = ctx;
185 	if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
186 		/* This is a mountpoint, so d_revalidate will never
187 		 * have been called, so we need to refresh the
188 		 * inode (for close-open consistency) ourselves.
189 		 */
190 		__nfs_revalidate_inode(NFS_SERVER(inode), inode);
191 	}
192 out:
193 	put_rpccred(cred);
194 	return res;
195 }
196 
197 static int
198 nfs_closedir(struct inode *inode, struct file *filp)
199 {
200 	put_nfs_open_dir_context(filp->private_data);
201 	return 0;
202 }
203 
204 struct nfs_cache_array_entry {
205 	u64 cookie;
206 	u64 ino;
207 	struct qstr string;
208 	unsigned char d_type;
209 };
210 
211 struct nfs_cache_array {
212 	int size;
213 	int eof_index;
214 	u64 last_cookie;
215 	struct nfs_cache_array_entry array[0];
216 };
217 
218 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
219 typedef struct {
220 	struct file	*file;
221 	struct page	*page;
222 	unsigned long	page_index;
223 	u64		*dir_cookie;
224 	u64		last_cookie;
225 	loff_t		current_index;
226 	decode_dirent_t	decode;
227 
228 	unsigned long	timestamp;
229 	unsigned long	gencount;
230 	unsigned int	cache_entry_index;
231 	unsigned int	plus:1;
232 	unsigned int	eof:1;
233 } nfs_readdir_descriptor_t;
234 
235 /*
236  * The caller is responsible for calling nfs_readdir_release_array(page)
237  */
238 static
239 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
240 {
241 	void *ptr;
242 	if (page == NULL)
243 		return ERR_PTR(-EIO);
244 	ptr = kmap(page);
245 	if (ptr == NULL)
246 		return ERR_PTR(-ENOMEM);
247 	return ptr;
248 }
249 
250 static
251 void nfs_readdir_release_array(struct page *page)
252 {
253 	kunmap(page);
254 }
255 
256 /*
257  * we are freeing strings created by nfs_add_to_readdir_array()
258  */
259 static
260 void nfs_readdir_clear_array(struct page *page)
261 {
262 	struct nfs_cache_array *array;
263 	int i;
264 
265 	array = kmap_atomic(page);
266 	for (i = 0; i < array->size; i++)
267 		kfree(array->array[i].string.name);
268 	kunmap_atomic(array);
269 }
270 
271 /*
272  * the caller is responsible for freeing qstr.name
273  * when called by nfs_readdir_add_to_array, the strings will be freed in
274  * nfs_clear_readdir_array()
275  */
276 static
277 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
278 {
279 	string->len = len;
280 	string->name = kmemdup(name, len, GFP_KERNEL);
281 	if (string->name == NULL)
282 		return -ENOMEM;
283 	/*
284 	 * Avoid a kmemleak false positive. The pointer to the name is stored
285 	 * in a page cache page which kmemleak does not scan.
286 	 */
287 	kmemleak_not_leak(string->name);
288 	string->hash = full_name_hash(name, len);
289 	return 0;
290 }
291 
292 static
293 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
294 {
295 	struct nfs_cache_array *array = nfs_readdir_get_array(page);
296 	struct nfs_cache_array_entry *cache_entry;
297 	int ret;
298 
299 	if (IS_ERR(array))
300 		return PTR_ERR(array);
301 
302 	cache_entry = &array->array[array->size];
303 
304 	/* Check that this entry lies within the page bounds */
305 	ret = -ENOSPC;
306 	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
307 		goto out;
308 
309 	cache_entry->cookie = entry->prev_cookie;
310 	cache_entry->ino = entry->ino;
311 	cache_entry->d_type = entry->d_type;
312 	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
313 	if (ret)
314 		goto out;
315 	array->last_cookie = entry->cookie;
316 	array->size++;
317 	if (entry->eof != 0)
318 		array->eof_index = array->size;
319 out:
320 	nfs_readdir_release_array(page);
321 	return ret;
322 }
323 
324 static
325 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
326 {
327 	loff_t diff = desc->file->f_pos - desc->current_index;
328 	unsigned int index;
329 
330 	if (diff < 0)
331 		goto out_eof;
332 	if (diff >= array->size) {
333 		if (array->eof_index >= 0)
334 			goto out_eof;
335 		return -EAGAIN;
336 	}
337 
338 	index = (unsigned int)diff;
339 	*desc->dir_cookie = array->array[index].cookie;
340 	desc->cache_entry_index = index;
341 	return 0;
342 out_eof:
343 	desc->eof = 1;
344 	return -EBADCOOKIE;
345 }
346 
347 static
348 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
349 {
350 	int i;
351 	loff_t new_pos;
352 	int status = -EAGAIN;
353 
354 	for (i = 0; i < array->size; i++) {
355 		if (array->array[i].cookie == *desc->dir_cookie) {
356 			struct nfs_inode *nfsi = NFS_I(desc->file->f_path.dentry->d_inode);
357 			struct nfs_open_dir_context *ctx = desc->file->private_data;
358 
359 			new_pos = desc->current_index + i;
360 			if (ctx->attr_gencount != nfsi->attr_gencount
361 			    || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
362 				ctx->duped = 0;
363 				ctx->attr_gencount = nfsi->attr_gencount;
364 			} else if (new_pos < desc->file->f_pos) {
365 				if (ctx->duped > 0
366 				    && ctx->dup_cookie == *desc->dir_cookie) {
367 					if (printk_ratelimit()) {
368 						pr_notice("NFS: directory %s/%s contains a readdir loop."
369 								"Please contact your server vendor.  "
370 								"The file: %s has duplicate cookie %llu\n",
371 								desc->file->f_dentry->d_parent->d_name.name,
372 								desc->file->f_dentry->d_name.name,
373 								array->array[i].string.name,
374 								*desc->dir_cookie);
375 					}
376 					status = -ELOOP;
377 					goto out;
378 				}
379 				ctx->dup_cookie = *desc->dir_cookie;
380 				ctx->duped = -1;
381 			}
382 			desc->file->f_pos = new_pos;
383 			desc->cache_entry_index = i;
384 			return 0;
385 		}
386 	}
387 	if (array->eof_index >= 0) {
388 		status = -EBADCOOKIE;
389 		if (*desc->dir_cookie == array->last_cookie)
390 			desc->eof = 1;
391 	}
392 out:
393 	return status;
394 }
395 
396 static
397 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
398 {
399 	struct nfs_cache_array *array;
400 	int status;
401 
402 	array = nfs_readdir_get_array(desc->page);
403 	if (IS_ERR(array)) {
404 		status = PTR_ERR(array);
405 		goto out;
406 	}
407 
408 	if (*desc->dir_cookie == 0)
409 		status = nfs_readdir_search_for_pos(array, desc);
410 	else
411 		status = nfs_readdir_search_for_cookie(array, desc);
412 
413 	if (status == -EAGAIN) {
414 		desc->last_cookie = array->last_cookie;
415 		desc->current_index += array->size;
416 		desc->page_index++;
417 	}
418 	nfs_readdir_release_array(desc->page);
419 out:
420 	return status;
421 }
422 
423 /* Fill a page with xdr information before transferring to the cache page */
424 static
425 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
426 			struct nfs_entry *entry, struct file *file, struct inode *inode)
427 {
428 	struct nfs_open_dir_context *ctx = file->private_data;
429 	struct rpc_cred	*cred = ctx->cred;
430 	unsigned long	timestamp, gencount;
431 	int		error;
432 
433  again:
434 	timestamp = jiffies;
435 	gencount = nfs_inc_attr_generation_counter();
436 	error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
437 					  NFS_SERVER(inode)->dtsize, desc->plus);
438 	if (error < 0) {
439 		/* We requested READDIRPLUS, but the server doesn't grok it */
440 		if (error == -ENOTSUPP && desc->plus) {
441 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
442 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
443 			desc->plus = 0;
444 			goto again;
445 		}
446 		goto error;
447 	}
448 	desc->timestamp = timestamp;
449 	desc->gencount = gencount;
450 error:
451 	return error;
452 }
453 
454 static int xdr_decode(nfs_readdir_descriptor_t *desc,
455 		      struct nfs_entry *entry, struct xdr_stream *xdr)
456 {
457 	int error;
458 
459 	error = desc->decode(xdr, entry, desc->plus);
460 	if (error)
461 		return error;
462 	entry->fattr->time_start = desc->timestamp;
463 	entry->fattr->gencount = desc->gencount;
464 	return 0;
465 }
466 
467 static
468 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
469 {
470 	if (dentry->d_inode == NULL)
471 		goto different;
472 	if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
473 		goto different;
474 	return 1;
475 different:
476 	return 0;
477 }
478 
479 static
480 bool nfs_use_readdirplus(struct inode *dir, struct file *filp)
481 {
482 	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
483 		return false;
484 	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
485 		return true;
486 	if (filp->f_pos == 0)
487 		return true;
488 	return false;
489 }
490 
491 /*
492  * This function is called by the lookup code to request the use of
493  * readdirplus to accelerate any future lookups in the same
494  * directory.
495  */
496 static
497 void nfs_advise_use_readdirplus(struct inode *dir)
498 {
499 	set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
500 }
501 
502 static
503 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
504 {
505 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
506 	struct dentry *dentry;
507 	struct dentry *alias;
508 	struct inode *dir = parent->d_inode;
509 	struct inode *inode;
510 
511 	if (filename.name[0] == '.') {
512 		if (filename.len == 1)
513 			return;
514 		if (filename.len == 2 && filename.name[1] == '.')
515 			return;
516 	}
517 	filename.hash = full_name_hash(filename.name, filename.len);
518 
519 	dentry = d_lookup(parent, &filename);
520 	if (dentry != NULL) {
521 		if (nfs_same_file(dentry, entry)) {
522 			nfs_refresh_inode(dentry->d_inode, entry->fattr);
523 			goto out;
524 		} else {
525 			d_drop(dentry);
526 			dput(dentry);
527 		}
528 	}
529 
530 	dentry = d_alloc(parent, &filename);
531 	if (dentry == NULL)
532 		return;
533 
534 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
535 	if (IS_ERR(inode))
536 		goto out;
537 
538 	alias = d_materialise_unique(dentry, inode);
539 	if (IS_ERR(alias))
540 		goto out;
541 	else if (alias) {
542 		nfs_set_verifier(alias, nfs_save_change_attribute(dir));
543 		dput(alias);
544 	} else
545 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
546 
547 out:
548 	dput(dentry);
549 }
550 
551 /* Perform conversion from xdr to cache array */
552 static
553 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
554 				struct page **xdr_pages, struct page *page, unsigned int buflen)
555 {
556 	struct xdr_stream stream;
557 	struct xdr_buf buf;
558 	struct page *scratch;
559 	struct nfs_cache_array *array;
560 	unsigned int count = 0;
561 	int status;
562 
563 	scratch = alloc_page(GFP_KERNEL);
564 	if (scratch == NULL)
565 		return -ENOMEM;
566 
567 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
568 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
569 
570 	do {
571 		status = xdr_decode(desc, entry, &stream);
572 		if (status != 0) {
573 			if (status == -EAGAIN)
574 				status = 0;
575 			break;
576 		}
577 
578 		count++;
579 
580 		if (desc->plus != 0)
581 			nfs_prime_dcache(desc->file->f_path.dentry, entry);
582 
583 		status = nfs_readdir_add_to_array(entry, page);
584 		if (status != 0)
585 			break;
586 	} while (!entry->eof);
587 
588 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
589 		array = nfs_readdir_get_array(page);
590 		if (!IS_ERR(array)) {
591 			array->eof_index = array->size;
592 			status = 0;
593 			nfs_readdir_release_array(page);
594 		} else
595 			status = PTR_ERR(array);
596 	}
597 
598 	put_page(scratch);
599 	return status;
600 }
601 
602 static
603 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
604 {
605 	unsigned int i;
606 	for (i = 0; i < npages; i++)
607 		put_page(pages[i]);
608 }
609 
610 static
611 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
612 		unsigned int npages)
613 {
614 	nfs_readdir_free_pagearray(pages, npages);
615 }
616 
617 /*
618  * nfs_readdir_large_page will allocate pages that must be freed with a call
619  * to nfs_readdir_free_large_page
620  */
621 static
622 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
623 {
624 	unsigned int i;
625 
626 	for (i = 0; i < npages; i++) {
627 		struct page *page = alloc_page(GFP_KERNEL);
628 		if (page == NULL)
629 			goto out_freepages;
630 		pages[i] = page;
631 	}
632 	return 0;
633 
634 out_freepages:
635 	nfs_readdir_free_pagearray(pages, i);
636 	return -ENOMEM;
637 }
638 
639 static
640 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
641 {
642 	struct page *pages[NFS_MAX_READDIR_PAGES];
643 	void *pages_ptr = NULL;
644 	struct nfs_entry entry;
645 	struct file	*file = desc->file;
646 	struct nfs_cache_array *array;
647 	int status = -ENOMEM;
648 	unsigned int array_size = ARRAY_SIZE(pages);
649 
650 	entry.prev_cookie = 0;
651 	entry.cookie = desc->last_cookie;
652 	entry.eof = 0;
653 	entry.fh = nfs_alloc_fhandle();
654 	entry.fattr = nfs_alloc_fattr();
655 	entry.server = NFS_SERVER(inode);
656 	if (entry.fh == NULL || entry.fattr == NULL)
657 		goto out;
658 
659 	array = nfs_readdir_get_array(page);
660 	if (IS_ERR(array)) {
661 		status = PTR_ERR(array);
662 		goto out;
663 	}
664 	memset(array, 0, sizeof(struct nfs_cache_array));
665 	array->eof_index = -1;
666 
667 	status = nfs_readdir_large_page(pages, array_size);
668 	if (status < 0)
669 		goto out_release_array;
670 	do {
671 		unsigned int pglen;
672 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
673 
674 		if (status < 0)
675 			break;
676 		pglen = status;
677 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
678 		if (status < 0) {
679 			if (status == -ENOSPC)
680 				status = 0;
681 			break;
682 		}
683 	} while (array->eof_index < 0);
684 
685 	nfs_readdir_free_large_page(pages_ptr, pages, array_size);
686 out_release_array:
687 	nfs_readdir_release_array(page);
688 out:
689 	nfs_free_fattr(entry.fattr);
690 	nfs_free_fhandle(entry.fh);
691 	return status;
692 }
693 
694 /*
695  * Now we cache directories properly, by converting xdr information
696  * to an array that can be used for lookups later.  This results in
697  * fewer cache pages, since we can store more information on each page.
698  * We only need to convert from xdr once so future lookups are much simpler
699  */
700 static
701 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
702 {
703 	struct inode	*inode = desc->file->f_path.dentry->d_inode;
704 	int ret;
705 
706 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
707 	if (ret < 0)
708 		goto error;
709 	SetPageUptodate(page);
710 
711 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
712 		/* Should never happen */
713 		nfs_zap_mapping(inode, inode->i_mapping);
714 	}
715 	unlock_page(page);
716 	return 0;
717  error:
718 	unlock_page(page);
719 	return ret;
720 }
721 
722 static
723 void cache_page_release(nfs_readdir_descriptor_t *desc)
724 {
725 	if (!desc->page->mapping)
726 		nfs_readdir_clear_array(desc->page);
727 	page_cache_release(desc->page);
728 	desc->page = NULL;
729 }
730 
731 static
732 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
733 {
734 	return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
735 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
736 }
737 
738 /*
739  * Returns 0 if desc->dir_cookie was found on page desc->page_index
740  */
741 static
742 int find_cache_page(nfs_readdir_descriptor_t *desc)
743 {
744 	int res;
745 
746 	desc->page = get_cache_page(desc);
747 	if (IS_ERR(desc->page))
748 		return PTR_ERR(desc->page);
749 
750 	res = nfs_readdir_search_array(desc);
751 	if (res != 0)
752 		cache_page_release(desc);
753 	return res;
754 }
755 
756 /* Search for desc->dir_cookie from the beginning of the page cache */
757 static inline
758 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
759 {
760 	int res;
761 
762 	if (desc->page_index == 0) {
763 		desc->current_index = 0;
764 		desc->last_cookie = 0;
765 	}
766 	do {
767 		res = find_cache_page(desc);
768 	} while (res == -EAGAIN);
769 	return res;
770 }
771 
772 /*
773  * Once we've found the start of the dirent within a page: fill 'er up...
774  */
775 static
776 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
777 		   filldir_t filldir)
778 {
779 	struct file	*file = desc->file;
780 	int i = 0;
781 	int res = 0;
782 	struct nfs_cache_array *array = NULL;
783 	struct nfs_open_dir_context *ctx = file->private_data;
784 
785 	array = nfs_readdir_get_array(desc->page);
786 	if (IS_ERR(array)) {
787 		res = PTR_ERR(array);
788 		goto out;
789 	}
790 
791 	for (i = desc->cache_entry_index; i < array->size; i++) {
792 		struct nfs_cache_array_entry *ent;
793 
794 		ent = &array->array[i];
795 		if (filldir(dirent, ent->string.name, ent->string.len,
796 		    file->f_pos, nfs_compat_user_ino64(ent->ino),
797 		    ent->d_type) < 0) {
798 			desc->eof = 1;
799 			break;
800 		}
801 		file->f_pos++;
802 		if (i < (array->size-1))
803 			*desc->dir_cookie = array->array[i+1].cookie;
804 		else
805 			*desc->dir_cookie = array->last_cookie;
806 		if (ctx->duped != 0)
807 			ctx->duped = 1;
808 	}
809 	if (array->eof_index >= 0)
810 		desc->eof = 1;
811 
812 	nfs_readdir_release_array(desc->page);
813 out:
814 	cache_page_release(desc);
815 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
816 			(unsigned long long)*desc->dir_cookie, res);
817 	return res;
818 }
819 
820 /*
821  * If we cannot find a cookie in our cache, we suspect that this is
822  * because it points to a deleted file, so we ask the server to return
823  * whatever it thinks is the next entry. We then feed this to filldir.
824  * If all goes well, we should then be able to find our way round the
825  * cache on the next call to readdir_search_pagecache();
826  *
827  * NOTE: we cannot add the anonymous page to the pagecache because
828  *	 the data it contains might not be page aligned. Besides,
829  *	 we should already have a complete representation of the
830  *	 directory in the page cache by the time we get here.
831  */
832 static inline
833 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
834 		     filldir_t filldir)
835 {
836 	struct page	*page = NULL;
837 	int		status;
838 	struct inode *inode = desc->file->f_path.dentry->d_inode;
839 	struct nfs_open_dir_context *ctx = desc->file->private_data;
840 
841 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
842 			(unsigned long long)*desc->dir_cookie);
843 
844 	page = alloc_page(GFP_HIGHUSER);
845 	if (!page) {
846 		status = -ENOMEM;
847 		goto out;
848 	}
849 
850 	desc->page_index = 0;
851 	desc->last_cookie = *desc->dir_cookie;
852 	desc->page = page;
853 	ctx->duped = 0;
854 
855 	status = nfs_readdir_xdr_to_array(desc, page, inode);
856 	if (status < 0)
857 		goto out_release;
858 
859 	status = nfs_do_filldir(desc, dirent, filldir);
860 
861  out:
862 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
863 			__func__, status);
864 	return status;
865  out_release:
866 	cache_page_release(desc);
867 	goto out;
868 }
869 
870 /* The file offset position represents the dirent entry number.  A
871    last cookie cache takes care of the common case of reading the
872    whole directory.
873  */
874 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
875 {
876 	struct dentry	*dentry = filp->f_path.dentry;
877 	struct inode	*inode = dentry->d_inode;
878 	nfs_readdir_descriptor_t my_desc,
879 			*desc = &my_desc;
880 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
881 	int res;
882 
883 	dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
884 			dentry->d_parent->d_name.name, dentry->d_name.name,
885 			(long long)filp->f_pos);
886 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
887 
888 	/*
889 	 * filp->f_pos points to the dirent entry number.
890 	 * *desc->dir_cookie has the cookie for the next entry. We have
891 	 * to either find the entry with the appropriate number or
892 	 * revalidate the cookie.
893 	 */
894 	memset(desc, 0, sizeof(*desc));
895 
896 	desc->file = filp;
897 	desc->dir_cookie = &dir_ctx->dir_cookie;
898 	desc->decode = NFS_PROTO(inode)->decode_dirent;
899 	desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0;
900 
901 	nfs_block_sillyrename(dentry);
902 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
903 	if (res < 0)
904 		goto out;
905 
906 	do {
907 		res = readdir_search_pagecache(desc);
908 
909 		if (res == -EBADCOOKIE) {
910 			res = 0;
911 			/* This means either end of directory */
912 			if (*desc->dir_cookie && desc->eof == 0) {
913 				/* Or that the server has 'lost' a cookie */
914 				res = uncached_readdir(desc, dirent, filldir);
915 				if (res == 0)
916 					continue;
917 			}
918 			break;
919 		}
920 		if (res == -ETOOSMALL && desc->plus) {
921 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
922 			nfs_zap_caches(inode);
923 			desc->page_index = 0;
924 			desc->plus = 0;
925 			desc->eof = 0;
926 			continue;
927 		}
928 		if (res < 0)
929 			break;
930 
931 		res = nfs_do_filldir(desc, dirent, filldir);
932 		if (res < 0)
933 			break;
934 	} while (!desc->eof);
935 out:
936 	nfs_unblock_sillyrename(dentry);
937 	if (res > 0)
938 		res = 0;
939 	dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
940 			dentry->d_parent->d_name.name, dentry->d_name.name,
941 			res);
942 	return res;
943 }
944 
945 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
946 {
947 	struct dentry *dentry = filp->f_path.dentry;
948 	struct inode *inode = dentry->d_inode;
949 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
950 
951 	dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
952 			dentry->d_parent->d_name.name,
953 			dentry->d_name.name,
954 			offset, origin);
955 
956 	mutex_lock(&inode->i_mutex);
957 	switch (origin) {
958 		case 1:
959 			offset += filp->f_pos;
960 		case 0:
961 			if (offset >= 0)
962 				break;
963 		default:
964 			offset = -EINVAL;
965 			goto out;
966 	}
967 	if (offset != filp->f_pos) {
968 		filp->f_pos = offset;
969 		dir_ctx->dir_cookie = 0;
970 		dir_ctx->duped = 0;
971 	}
972 out:
973 	mutex_unlock(&inode->i_mutex);
974 	return offset;
975 }
976 
977 /*
978  * All directory operations under NFS are synchronous, so fsync()
979  * is a dummy operation.
980  */
981 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
982 			 int datasync)
983 {
984 	struct dentry *dentry = filp->f_path.dentry;
985 	struct inode *inode = dentry->d_inode;
986 
987 	dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
988 			dentry->d_parent->d_name.name, dentry->d_name.name,
989 			datasync);
990 
991 	mutex_lock(&inode->i_mutex);
992 	nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
993 	mutex_unlock(&inode->i_mutex);
994 	return 0;
995 }
996 
997 /**
998  * nfs_force_lookup_revalidate - Mark the directory as having changed
999  * @dir - pointer to directory inode
1000  *
1001  * This forces the revalidation code in nfs_lookup_revalidate() to do a
1002  * full lookup on all child dentries of 'dir' whenever a change occurs
1003  * on the server that might have invalidated our dcache.
1004  *
1005  * The caller should be holding dir->i_lock
1006  */
1007 void nfs_force_lookup_revalidate(struct inode *dir)
1008 {
1009 	NFS_I(dir)->cache_change_attribute++;
1010 }
1011 
1012 /*
1013  * A check for whether or not the parent directory has changed.
1014  * In the case it has, we assume that the dentries are untrustworthy
1015  * and may need to be looked up again.
1016  */
1017 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
1018 {
1019 	if (IS_ROOT(dentry))
1020 		return 1;
1021 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1022 		return 0;
1023 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1024 		return 0;
1025 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1026 	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1027 		return 0;
1028 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1029 		return 0;
1030 	return 1;
1031 }
1032 
1033 /*
1034  * Return the intent data that applies to this particular path component
1035  *
1036  * Note that the current set of intents only apply to the very last
1037  * component of the path and none of them is set before that last
1038  * component.
1039  */
1040 static inline unsigned int nfs_lookup_check_intent(unsigned int flags,
1041 						unsigned int mask)
1042 {
1043 	return flags & mask;
1044 }
1045 
1046 /*
1047  * Use intent information to check whether or not we're going to do
1048  * an O_EXCL create using this path component.
1049  */
1050 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1051 {
1052 	if (NFS_PROTO(dir)->version == 2)
1053 		return 0;
1054 	return nd && nfs_lookup_check_intent(nd->flags, LOOKUP_EXCL);
1055 }
1056 
1057 /*
1058  * Inode and filehandle revalidation for lookups.
1059  *
1060  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1061  * or if the intent information indicates that we're about to open this
1062  * particular file and the "nocto" mount flag is not set.
1063  *
1064  */
1065 static inline
1066 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1067 {
1068 	struct nfs_server *server = NFS_SERVER(inode);
1069 
1070 	if (IS_AUTOMOUNT(inode))
1071 		return 0;
1072 	/* VFS wants an on-the-wire revalidation */
1073 	if (nd->flags & LOOKUP_REVAL)
1074 		goto out_force;
1075 	/* This is an open(2) */
1076 	if (nfs_lookup_check_intent(nd->flags, LOOKUP_OPEN) != 0 &&
1077 			!(server->flags & NFS_MOUNT_NOCTO) &&
1078 			(S_ISREG(inode->i_mode) ||
1079 			 S_ISDIR(inode->i_mode)))
1080 		goto out_force;
1081 	return 0;
1082 out_force:
1083 	return __nfs_revalidate_inode(server, inode);
1084 }
1085 
1086 /*
1087  * We judge how long we want to trust negative
1088  * dentries by looking at the parent inode mtime.
1089  *
1090  * If parent mtime has changed, we revalidate, else we wait for a
1091  * period corresponding to the parent's attribute cache timeout value.
1092  */
1093 static inline
1094 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1095 		       struct nameidata *nd)
1096 {
1097 	/* Don't revalidate a negative dentry if we're creating a new file */
1098 	if (nd != NULL && nfs_lookup_check_intent(nd->flags, LOOKUP_CREATE) != 0)
1099 		return 0;
1100 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1101 		return 1;
1102 	return !nfs_check_verifier(dir, dentry);
1103 }
1104 
1105 /*
1106  * This is called every time the dcache has a lookup hit,
1107  * and we should check whether we can really trust that
1108  * lookup.
1109  *
1110  * NOTE! The hit can be a negative hit too, don't assume
1111  * we have an inode!
1112  *
1113  * If the parent directory is seen to have changed, we throw out the
1114  * cached dentry and do a new lookup.
1115  */
1116 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1117 {
1118 	struct inode *dir;
1119 	struct inode *inode;
1120 	struct dentry *parent;
1121 	struct nfs_fh *fhandle = NULL;
1122 	struct nfs_fattr *fattr = NULL;
1123 	int error;
1124 
1125 	if (nd->flags & LOOKUP_RCU)
1126 		return -ECHILD;
1127 
1128 	parent = dget_parent(dentry);
1129 	dir = parent->d_inode;
1130 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1131 	inode = dentry->d_inode;
1132 
1133 	if (!inode) {
1134 		if (nfs_neg_need_reval(dir, dentry, nd))
1135 			goto out_bad;
1136 		goto out_valid_noent;
1137 	}
1138 
1139 	if (is_bad_inode(inode)) {
1140 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1141 				__func__, dentry->d_parent->d_name.name,
1142 				dentry->d_name.name);
1143 		goto out_bad;
1144 	}
1145 
1146 	if (nfs_have_delegation(inode, FMODE_READ))
1147 		goto out_set_verifier;
1148 
1149 	/* Force a full look up iff the parent directory has changed */
1150 	if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1151 		if (nfs_lookup_verify_inode(inode, nd))
1152 			goto out_zap_parent;
1153 		goto out_valid;
1154 	}
1155 
1156 	if (NFS_STALE(inode))
1157 		goto out_bad;
1158 
1159 	error = -ENOMEM;
1160 	fhandle = nfs_alloc_fhandle();
1161 	fattr = nfs_alloc_fattr();
1162 	if (fhandle == NULL || fattr == NULL)
1163 		goto out_error;
1164 
1165 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1166 	if (error)
1167 		goto out_bad;
1168 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1169 		goto out_bad;
1170 	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1171 		goto out_bad;
1172 
1173 	nfs_free_fattr(fattr);
1174 	nfs_free_fhandle(fhandle);
1175 out_set_verifier:
1176 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1177  out_valid:
1178 	/* Success: notify readdir to use READDIRPLUS */
1179 	nfs_advise_use_readdirplus(dir);
1180  out_valid_noent:
1181 	dput(parent);
1182 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1183 			__func__, dentry->d_parent->d_name.name,
1184 			dentry->d_name.name);
1185 	return 1;
1186 out_zap_parent:
1187 	nfs_zap_caches(dir);
1188  out_bad:
1189 	nfs_mark_for_revalidate(dir);
1190 	if (inode && S_ISDIR(inode->i_mode)) {
1191 		/* Purge readdir caches. */
1192 		nfs_zap_caches(inode);
1193 		/* If we have submounts, don't unhash ! */
1194 		if (have_submounts(dentry))
1195 			goto out_valid;
1196 		if (dentry->d_flags & DCACHE_DISCONNECTED)
1197 			goto out_valid;
1198 		shrink_dcache_parent(dentry);
1199 	}
1200 	d_drop(dentry);
1201 	nfs_free_fattr(fattr);
1202 	nfs_free_fhandle(fhandle);
1203 	dput(parent);
1204 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1205 			__func__, dentry->d_parent->d_name.name,
1206 			dentry->d_name.name);
1207 	return 0;
1208 out_error:
1209 	nfs_free_fattr(fattr);
1210 	nfs_free_fhandle(fhandle);
1211 	dput(parent);
1212 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1213 			__func__, dentry->d_parent->d_name.name,
1214 			dentry->d_name.name, error);
1215 	return error;
1216 }
1217 
1218 /*
1219  * This is called from dput() when d_count is going to 0.
1220  */
1221 static int nfs_dentry_delete(const struct dentry *dentry)
1222 {
1223 	dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1224 		dentry->d_parent->d_name.name, dentry->d_name.name,
1225 		dentry->d_flags);
1226 
1227 	/* Unhash any dentry with a stale inode */
1228 	if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1229 		return 1;
1230 
1231 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1232 		/* Unhash it, so that ->d_iput() would be called */
1233 		return 1;
1234 	}
1235 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1236 		/* Unhash it, so that ancestors of killed async unlink
1237 		 * files will be cleaned up during umount */
1238 		return 1;
1239 	}
1240 	return 0;
1241 
1242 }
1243 
1244 static void nfs_drop_nlink(struct inode *inode)
1245 {
1246 	spin_lock(&inode->i_lock);
1247 	if (inode->i_nlink > 0)
1248 		drop_nlink(inode);
1249 	spin_unlock(&inode->i_lock);
1250 }
1251 
1252 /*
1253  * Called when the dentry loses inode.
1254  * We use it to clean up silly-renamed files.
1255  */
1256 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1257 {
1258 	if (S_ISDIR(inode->i_mode))
1259 		/* drop any readdir cache as it could easily be old */
1260 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1261 
1262 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1263 		drop_nlink(inode);
1264 		nfs_complete_unlink(dentry, inode);
1265 	}
1266 	iput(inode);
1267 }
1268 
1269 static void nfs_d_release(struct dentry *dentry)
1270 {
1271 	/* free cached devname value, if it survived that far */
1272 	if (unlikely(dentry->d_fsdata)) {
1273 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1274 			WARN_ON(1);
1275 		else
1276 			kfree(dentry->d_fsdata);
1277 	}
1278 }
1279 
1280 const struct dentry_operations nfs_dentry_operations = {
1281 	.d_revalidate	= nfs_lookup_revalidate,
1282 	.d_delete	= nfs_dentry_delete,
1283 	.d_iput		= nfs_dentry_iput,
1284 	.d_automount	= nfs_d_automount,
1285 	.d_release	= nfs_d_release,
1286 };
1287 
1288 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1289 {
1290 	struct dentry *res;
1291 	struct dentry *parent;
1292 	struct inode *inode = NULL;
1293 	struct nfs_fh *fhandle = NULL;
1294 	struct nfs_fattr *fattr = NULL;
1295 	int error;
1296 
1297 	dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1298 		dentry->d_parent->d_name.name, dentry->d_name.name);
1299 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1300 
1301 	res = ERR_PTR(-ENAMETOOLONG);
1302 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1303 		goto out;
1304 
1305 	/*
1306 	 * If we're doing an exclusive create, optimize away the lookup
1307 	 * but don't hash the dentry.
1308 	 */
1309 	if (nfs_is_exclusive_create(dir, nd)) {
1310 		d_instantiate(dentry, NULL);
1311 		res = NULL;
1312 		goto out;
1313 	}
1314 
1315 	res = ERR_PTR(-ENOMEM);
1316 	fhandle = nfs_alloc_fhandle();
1317 	fattr = nfs_alloc_fattr();
1318 	if (fhandle == NULL || fattr == NULL)
1319 		goto out;
1320 
1321 	parent = dentry->d_parent;
1322 	/* Protect against concurrent sillydeletes */
1323 	nfs_block_sillyrename(parent);
1324 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1325 	if (error == -ENOENT)
1326 		goto no_entry;
1327 	if (error < 0) {
1328 		res = ERR_PTR(error);
1329 		goto out_unblock_sillyrename;
1330 	}
1331 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1332 	res = ERR_CAST(inode);
1333 	if (IS_ERR(res))
1334 		goto out_unblock_sillyrename;
1335 
1336 	/* Success: notify readdir to use READDIRPLUS */
1337 	nfs_advise_use_readdirplus(dir);
1338 
1339 no_entry:
1340 	res = d_materialise_unique(dentry, inode);
1341 	if (res != NULL) {
1342 		if (IS_ERR(res))
1343 			goto out_unblock_sillyrename;
1344 		dentry = res;
1345 	}
1346 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1347 out_unblock_sillyrename:
1348 	nfs_unblock_sillyrename(parent);
1349 out:
1350 	nfs_free_fattr(fattr);
1351 	nfs_free_fhandle(fhandle);
1352 	return res;
1353 }
1354 
1355 #ifdef CONFIG_NFS_V4
1356 static int nfs4_lookup_revalidate(struct dentry *, struct nameidata *);
1357 
1358 const struct dentry_operations nfs4_dentry_operations = {
1359 	.d_revalidate	= nfs4_lookup_revalidate,
1360 	.d_delete	= nfs_dentry_delete,
1361 	.d_iput		= nfs_dentry_iput,
1362 	.d_automount	= nfs_d_automount,
1363 	.d_release	= nfs_d_release,
1364 };
1365 
1366 static fmode_t flags_to_mode(int flags)
1367 {
1368 	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1369 	if ((flags & O_ACCMODE) != O_WRONLY)
1370 		res |= FMODE_READ;
1371 	if ((flags & O_ACCMODE) != O_RDONLY)
1372 		res |= FMODE_WRITE;
1373 	return res;
1374 }
1375 
1376 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1377 {
1378 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1379 }
1380 
1381 static int do_open(struct inode *inode, struct file *filp)
1382 {
1383 	nfs_fscache_set_inode_cookie(inode, filp);
1384 	return 0;
1385 }
1386 
1387 static int nfs_finish_open(struct nfs_open_context *ctx,
1388 			   struct dentry *dentry,
1389 			   struct file *file, unsigned open_flags,
1390 			   int *opened)
1391 {
1392 	int err;
1393 
1394 	if (ctx->dentry != dentry) {
1395 		dput(ctx->dentry);
1396 		ctx->dentry = dget(dentry);
1397 	}
1398 
1399 	/* If the open_intent is for execute, we have an extra check to make */
1400 	if (ctx->mode & FMODE_EXEC) {
1401 		err = nfs_may_open(dentry->d_inode, ctx->cred, open_flags);
1402 		if (err < 0)
1403 			goto out;
1404 	}
1405 
1406 	err = finish_open(file, dentry, do_open, opened);
1407 	if (err)
1408 		goto out;
1409 	nfs_file_set_open_context(file, ctx);
1410 
1411 out:
1412 	put_nfs_open_context(ctx);
1413 	return err;
1414 }
1415 
1416 static int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1417 			    struct file *file, unsigned open_flags,
1418 			    umode_t mode, int *opened)
1419 {
1420 	struct nfs_open_context *ctx;
1421 	struct dentry *res;
1422 	struct iattr attr = { .ia_valid = ATTR_OPEN };
1423 	struct inode *inode;
1424 	int err;
1425 
1426 	/* Expect a negative dentry */
1427 	BUG_ON(dentry->d_inode);
1428 
1429 	dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n",
1430 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1431 
1432 	/* NFS only supports OPEN on regular files */
1433 	if ((open_flags & O_DIRECTORY)) {
1434 		if (!d_unhashed(dentry)) {
1435 			/*
1436 			 * Hashed negative dentry with O_DIRECTORY: dentry was
1437 			 * revalidated and is fine, no need to perform lookup
1438 			 * again
1439 			 */
1440 			return -ENOENT;
1441 		}
1442 		goto no_open;
1443 	}
1444 
1445 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1446 		return -ENAMETOOLONG;
1447 
1448 	if (open_flags & O_CREAT) {
1449 		attr.ia_valid |= ATTR_MODE;
1450 		attr.ia_mode = mode & ~current_umask();
1451 	}
1452 	if (open_flags & O_TRUNC) {
1453 		attr.ia_valid |= ATTR_SIZE;
1454 		attr.ia_size = 0;
1455 	}
1456 
1457 	ctx = create_nfs_open_context(dentry, open_flags);
1458 	err = PTR_ERR(ctx);
1459 	if (IS_ERR(ctx))
1460 		goto out;
1461 
1462 	nfs_block_sillyrename(dentry->d_parent);
1463 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1464 	d_drop(dentry);
1465 	if (IS_ERR(inode)) {
1466 		nfs_unblock_sillyrename(dentry->d_parent);
1467 		put_nfs_open_context(ctx);
1468 		err = PTR_ERR(inode);
1469 		switch (err) {
1470 		case -ENOENT:
1471 			d_add(dentry, NULL);
1472 			break;
1473 		case -EISDIR:
1474 		case -ENOTDIR:
1475 			goto no_open;
1476 		case -ELOOP:
1477 			if (!(open_flags & O_NOFOLLOW))
1478 				goto no_open;
1479 			break;
1480 			/* case -EINVAL: */
1481 		default:
1482 			break;
1483 		}
1484 		goto out;
1485 	}
1486 	res = d_add_unique(dentry, inode);
1487 	if (res != NULL)
1488 		dentry = res;
1489 
1490 	nfs_unblock_sillyrename(dentry->d_parent);
1491 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1492 
1493 	err = nfs_finish_open(ctx, dentry, file, open_flags, opened);
1494 
1495 	dput(res);
1496 out:
1497 	return err;
1498 
1499 no_open:
1500 	res = nfs_lookup(dir, dentry, NULL);
1501 	err = PTR_ERR(res);
1502 	if (IS_ERR(res))
1503 		goto out;
1504 
1505 	return finish_no_open(file, res);
1506 }
1507 
1508 static int nfs4_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1509 {
1510 	struct dentry *parent = NULL;
1511 	struct inode *inode;
1512 	struct inode *dir;
1513 	int ret = 0;
1514 
1515 	if (nd->flags & LOOKUP_RCU)
1516 		return -ECHILD;
1517 
1518 	if (!(nd->flags & LOOKUP_OPEN) || (nd->flags & LOOKUP_DIRECTORY))
1519 		goto no_open;
1520 	if (d_mountpoint(dentry))
1521 		goto no_open;
1522 
1523 	inode = dentry->d_inode;
1524 	parent = dget_parent(dentry);
1525 	dir = parent->d_inode;
1526 
1527 	/* We can't create new files in nfs_open_revalidate(), so we
1528 	 * optimize away revalidation of negative dentries.
1529 	 */
1530 	if (inode == NULL) {
1531 		if (!nfs_neg_need_reval(dir, dentry, nd))
1532 			ret = 1;
1533 		goto out;
1534 	}
1535 
1536 	/* NFS only supports OPEN on regular files */
1537 	if (!S_ISREG(inode->i_mode))
1538 		goto no_open_dput;
1539 	/* We cannot do exclusive creation on a positive dentry */
1540 	if (nd && nd->flags & LOOKUP_EXCL)
1541 		goto no_open_dput;
1542 
1543 	/* Let f_op->open() actually open (and revalidate) the file */
1544 	ret = 1;
1545 
1546 out:
1547 	dput(parent);
1548 	return ret;
1549 
1550 no_open_dput:
1551 	dput(parent);
1552 no_open:
1553 	return nfs_lookup_revalidate(dentry, nd);
1554 }
1555 
1556 #endif /* CONFIG_NFSV4 */
1557 
1558 /*
1559  * Code common to create, mkdir, and mknod.
1560  */
1561 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1562 				struct nfs_fattr *fattr)
1563 {
1564 	struct dentry *parent = dget_parent(dentry);
1565 	struct inode *dir = parent->d_inode;
1566 	struct inode *inode;
1567 	int error = -EACCES;
1568 
1569 	d_drop(dentry);
1570 
1571 	/* We may have been initialized further down */
1572 	if (dentry->d_inode)
1573 		goto out;
1574 	if (fhandle->size == 0) {
1575 		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1576 		if (error)
1577 			goto out_error;
1578 	}
1579 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1580 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1581 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1582 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1583 		if (error < 0)
1584 			goto out_error;
1585 	}
1586 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1587 	error = PTR_ERR(inode);
1588 	if (IS_ERR(inode))
1589 		goto out_error;
1590 	d_add(dentry, inode);
1591 out:
1592 	dput(parent);
1593 	return 0;
1594 out_error:
1595 	nfs_mark_for_revalidate(dir);
1596 	dput(parent);
1597 	return error;
1598 }
1599 
1600 /*
1601  * Following a failed create operation, we drop the dentry rather
1602  * than retain a negative dentry. This avoids a problem in the event
1603  * that the operation succeeded on the server, but an error in the
1604  * reply path made it appear to have failed.
1605  */
1606 static int nfs_create(struct inode *dir, struct dentry *dentry,
1607 		umode_t mode, struct nameidata *nd)
1608 {
1609 	struct iattr attr;
1610 	int error;
1611 	int open_flags = O_CREAT|O_EXCL;
1612 
1613 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1614 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1615 
1616 	attr.ia_mode = mode;
1617 	attr.ia_valid = ATTR_MODE;
1618 
1619 	if (nd && !(nd->flags & LOOKUP_EXCL))
1620 		open_flags = O_CREAT;
1621 
1622 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1623 	if (error != 0)
1624 		goto out_err;
1625 	return 0;
1626 out_err:
1627 	d_drop(dentry);
1628 	return error;
1629 }
1630 
1631 /*
1632  * See comments for nfs_proc_create regarding failed operations.
1633  */
1634 static int
1635 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1636 {
1637 	struct iattr attr;
1638 	int status;
1639 
1640 	dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1641 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1642 
1643 	if (!new_valid_dev(rdev))
1644 		return -EINVAL;
1645 
1646 	attr.ia_mode = mode;
1647 	attr.ia_valid = ATTR_MODE;
1648 
1649 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1650 	if (status != 0)
1651 		goto out_err;
1652 	return 0;
1653 out_err:
1654 	d_drop(dentry);
1655 	return status;
1656 }
1657 
1658 /*
1659  * See comments for nfs_proc_create regarding failed operations.
1660  */
1661 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1662 {
1663 	struct iattr attr;
1664 	int error;
1665 
1666 	dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1667 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1668 
1669 	attr.ia_valid = ATTR_MODE;
1670 	attr.ia_mode = mode | S_IFDIR;
1671 
1672 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1673 	if (error != 0)
1674 		goto out_err;
1675 	return 0;
1676 out_err:
1677 	d_drop(dentry);
1678 	return error;
1679 }
1680 
1681 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1682 {
1683 	if (dentry->d_inode != NULL && !d_unhashed(dentry))
1684 		d_delete(dentry);
1685 }
1686 
1687 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1688 {
1689 	int error;
1690 
1691 	dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1692 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1693 
1694 	error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1695 	/* Ensure the VFS deletes this inode */
1696 	if (error == 0 && dentry->d_inode != NULL)
1697 		clear_nlink(dentry->d_inode);
1698 	else if (error == -ENOENT)
1699 		nfs_dentry_handle_enoent(dentry);
1700 
1701 	return error;
1702 }
1703 
1704 /*
1705  * Remove a file after making sure there are no pending writes,
1706  * and after checking that the file has only one user.
1707  *
1708  * We invalidate the attribute cache and free the inode prior to the operation
1709  * to avoid possible races if the server reuses the inode.
1710  */
1711 static int nfs_safe_remove(struct dentry *dentry)
1712 {
1713 	struct inode *dir = dentry->d_parent->d_inode;
1714 	struct inode *inode = dentry->d_inode;
1715 	int error = -EBUSY;
1716 
1717 	dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1718 		dentry->d_parent->d_name.name, dentry->d_name.name);
1719 
1720 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1721 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1722 		error = 0;
1723 		goto out;
1724 	}
1725 
1726 	if (inode != NULL) {
1727 		nfs_inode_return_delegation(inode);
1728 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1729 		/* The VFS may want to delete this inode */
1730 		if (error == 0)
1731 			nfs_drop_nlink(inode);
1732 		nfs_mark_for_revalidate(inode);
1733 	} else
1734 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1735 	if (error == -ENOENT)
1736 		nfs_dentry_handle_enoent(dentry);
1737 out:
1738 	return error;
1739 }
1740 
1741 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1742  *  belongs to an active ".nfs..." file and we return -EBUSY.
1743  *
1744  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1745  */
1746 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1747 {
1748 	int error;
1749 	int need_rehash = 0;
1750 
1751 	dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1752 		dir->i_ino, dentry->d_name.name);
1753 
1754 	spin_lock(&dentry->d_lock);
1755 	if (dentry->d_count > 1) {
1756 		spin_unlock(&dentry->d_lock);
1757 		/* Start asynchronous writeout of the inode */
1758 		write_inode_now(dentry->d_inode, 0);
1759 		error = nfs_sillyrename(dir, dentry);
1760 		return error;
1761 	}
1762 	if (!d_unhashed(dentry)) {
1763 		__d_drop(dentry);
1764 		need_rehash = 1;
1765 	}
1766 	spin_unlock(&dentry->d_lock);
1767 	error = nfs_safe_remove(dentry);
1768 	if (!error || error == -ENOENT) {
1769 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1770 	} else if (need_rehash)
1771 		d_rehash(dentry);
1772 	return error;
1773 }
1774 
1775 /*
1776  * To create a symbolic link, most file systems instantiate a new inode,
1777  * add a page to it containing the path, then write it out to the disk
1778  * using prepare_write/commit_write.
1779  *
1780  * Unfortunately the NFS client can't create the in-core inode first
1781  * because it needs a file handle to create an in-core inode (see
1782  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1783  * symlink request has completed on the server.
1784  *
1785  * So instead we allocate a raw page, copy the symname into it, then do
1786  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1787  * now have a new file handle and can instantiate an in-core NFS inode
1788  * and move the raw page into its mapping.
1789  */
1790 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1791 {
1792 	struct pagevec lru_pvec;
1793 	struct page *page;
1794 	char *kaddr;
1795 	struct iattr attr;
1796 	unsigned int pathlen = strlen(symname);
1797 	int error;
1798 
1799 	dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1800 		dir->i_ino, dentry->d_name.name, symname);
1801 
1802 	if (pathlen > PAGE_SIZE)
1803 		return -ENAMETOOLONG;
1804 
1805 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1806 	attr.ia_valid = ATTR_MODE;
1807 
1808 	page = alloc_page(GFP_HIGHUSER);
1809 	if (!page)
1810 		return -ENOMEM;
1811 
1812 	kaddr = kmap_atomic(page);
1813 	memcpy(kaddr, symname, pathlen);
1814 	if (pathlen < PAGE_SIZE)
1815 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1816 	kunmap_atomic(kaddr);
1817 
1818 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1819 	if (error != 0) {
1820 		dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1821 			dir->i_sb->s_id, dir->i_ino,
1822 			dentry->d_name.name, symname, error);
1823 		d_drop(dentry);
1824 		__free_page(page);
1825 		return error;
1826 	}
1827 
1828 	/*
1829 	 * No big deal if we can't add this page to the page cache here.
1830 	 * READLINK will get the missing page from the server if needed.
1831 	 */
1832 	pagevec_init(&lru_pvec, 0);
1833 	if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1834 							GFP_KERNEL)) {
1835 		pagevec_add(&lru_pvec, page);
1836 		pagevec_lru_add_file(&lru_pvec);
1837 		SetPageUptodate(page);
1838 		unlock_page(page);
1839 	} else
1840 		__free_page(page);
1841 
1842 	return 0;
1843 }
1844 
1845 static int
1846 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1847 {
1848 	struct inode *inode = old_dentry->d_inode;
1849 	int error;
1850 
1851 	dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1852 		old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1853 		dentry->d_parent->d_name.name, dentry->d_name.name);
1854 
1855 	nfs_inode_return_delegation(inode);
1856 
1857 	d_drop(dentry);
1858 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1859 	if (error == 0) {
1860 		ihold(inode);
1861 		d_add(dentry, inode);
1862 	}
1863 	return error;
1864 }
1865 
1866 /*
1867  * RENAME
1868  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1869  * different file handle for the same inode after a rename (e.g. when
1870  * moving to a different directory). A fail-safe method to do so would
1871  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1872  * rename the old file using the sillyrename stuff. This way, the original
1873  * file in old_dir will go away when the last process iput()s the inode.
1874  *
1875  * FIXED.
1876  *
1877  * It actually works quite well. One needs to have the possibility for
1878  * at least one ".nfs..." file in each directory the file ever gets
1879  * moved or linked to which happens automagically with the new
1880  * implementation that only depends on the dcache stuff instead of
1881  * using the inode layer
1882  *
1883  * Unfortunately, things are a little more complicated than indicated
1884  * above. For a cross-directory move, we want to make sure we can get
1885  * rid of the old inode after the operation.  This means there must be
1886  * no pending writes (if it's a file), and the use count must be 1.
1887  * If these conditions are met, we can drop the dentries before doing
1888  * the rename.
1889  */
1890 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1891 		      struct inode *new_dir, struct dentry *new_dentry)
1892 {
1893 	struct inode *old_inode = old_dentry->d_inode;
1894 	struct inode *new_inode = new_dentry->d_inode;
1895 	struct dentry *dentry = NULL, *rehash = NULL;
1896 	int error = -EBUSY;
1897 
1898 	dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1899 		 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1900 		 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1901 		 new_dentry->d_count);
1902 
1903 	/*
1904 	 * For non-directories, check whether the target is busy and if so,
1905 	 * make a copy of the dentry and then do a silly-rename. If the
1906 	 * silly-rename succeeds, the copied dentry is hashed and becomes
1907 	 * the new target.
1908 	 */
1909 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1910 		/*
1911 		 * To prevent any new references to the target during the
1912 		 * rename, we unhash the dentry in advance.
1913 		 */
1914 		if (!d_unhashed(new_dentry)) {
1915 			d_drop(new_dentry);
1916 			rehash = new_dentry;
1917 		}
1918 
1919 		if (new_dentry->d_count > 2) {
1920 			int err;
1921 
1922 			/* copy the target dentry's name */
1923 			dentry = d_alloc(new_dentry->d_parent,
1924 					 &new_dentry->d_name);
1925 			if (!dentry)
1926 				goto out;
1927 
1928 			/* silly-rename the existing target ... */
1929 			err = nfs_sillyrename(new_dir, new_dentry);
1930 			if (err)
1931 				goto out;
1932 
1933 			new_dentry = dentry;
1934 			rehash = NULL;
1935 			new_inode = NULL;
1936 		}
1937 	}
1938 
1939 	nfs_inode_return_delegation(old_inode);
1940 	if (new_inode != NULL)
1941 		nfs_inode_return_delegation(new_inode);
1942 
1943 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1944 					   new_dir, &new_dentry->d_name);
1945 	nfs_mark_for_revalidate(old_inode);
1946 out:
1947 	if (rehash)
1948 		d_rehash(rehash);
1949 	if (!error) {
1950 		if (new_inode != NULL)
1951 			nfs_drop_nlink(new_inode);
1952 		d_move(old_dentry, new_dentry);
1953 		nfs_set_verifier(new_dentry,
1954 					nfs_save_change_attribute(new_dir));
1955 	} else if (error == -ENOENT)
1956 		nfs_dentry_handle_enoent(old_dentry);
1957 
1958 	/* new dentry created? */
1959 	if (dentry)
1960 		dput(dentry);
1961 	return error;
1962 }
1963 
1964 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1965 static LIST_HEAD(nfs_access_lru_list);
1966 static atomic_long_t nfs_access_nr_entries;
1967 
1968 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1969 {
1970 	put_rpccred(entry->cred);
1971 	kfree(entry);
1972 	smp_mb__before_atomic_dec();
1973 	atomic_long_dec(&nfs_access_nr_entries);
1974 	smp_mb__after_atomic_dec();
1975 }
1976 
1977 static void nfs_access_free_list(struct list_head *head)
1978 {
1979 	struct nfs_access_entry *cache;
1980 
1981 	while (!list_empty(head)) {
1982 		cache = list_entry(head->next, struct nfs_access_entry, lru);
1983 		list_del(&cache->lru);
1984 		nfs_access_free_entry(cache);
1985 	}
1986 }
1987 
1988 int nfs_access_cache_shrinker(struct shrinker *shrink,
1989 			      struct shrink_control *sc)
1990 {
1991 	LIST_HEAD(head);
1992 	struct nfs_inode *nfsi, *next;
1993 	struct nfs_access_entry *cache;
1994 	int nr_to_scan = sc->nr_to_scan;
1995 	gfp_t gfp_mask = sc->gfp_mask;
1996 
1997 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1998 		return (nr_to_scan == 0) ? 0 : -1;
1999 
2000 	spin_lock(&nfs_access_lru_lock);
2001 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2002 		struct inode *inode;
2003 
2004 		if (nr_to_scan-- == 0)
2005 			break;
2006 		inode = &nfsi->vfs_inode;
2007 		spin_lock(&inode->i_lock);
2008 		if (list_empty(&nfsi->access_cache_entry_lru))
2009 			goto remove_lru_entry;
2010 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2011 				struct nfs_access_entry, lru);
2012 		list_move(&cache->lru, &head);
2013 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2014 		if (!list_empty(&nfsi->access_cache_entry_lru))
2015 			list_move_tail(&nfsi->access_cache_inode_lru,
2016 					&nfs_access_lru_list);
2017 		else {
2018 remove_lru_entry:
2019 			list_del_init(&nfsi->access_cache_inode_lru);
2020 			smp_mb__before_clear_bit();
2021 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2022 			smp_mb__after_clear_bit();
2023 		}
2024 		spin_unlock(&inode->i_lock);
2025 	}
2026 	spin_unlock(&nfs_access_lru_lock);
2027 	nfs_access_free_list(&head);
2028 	return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2029 }
2030 
2031 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2032 {
2033 	struct rb_root *root_node = &nfsi->access_cache;
2034 	struct rb_node *n;
2035 	struct nfs_access_entry *entry;
2036 
2037 	/* Unhook entries from the cache */
2038 	while ((n = rb_first(root_node)) != NULL) {
2039 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2040 		rb_erase(n, root_node);
2041 		list_move(&entry->lru, head);
2042 	}
2043 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2044 }
2045 
2046 void nfs_access_zap_cache(struct inode *inode)
2047 {
2048 	LIST_HEAD(head);
2049 
2050 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2051 		return;
2052 	/* Remove from global LRU init */
2053 	spin_lock(&nfs_access_lru_lock);
2054 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2055 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2056 
2057 	spin_lock(&inode->i_lock);
2058 	__nfs_access_zap_cache(NFS_I(inode), &head);
2059 	spin_unlock(&inode->i_lock);
2060 	spin_unlock(&nfs_access_lru_lock);
2061 	nfs_access_free_list(&head);
2062 }
2063 
2064 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2065 {
2066 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2067 	struct nfs_access_entry *entry;
2068 
2069 	while (n != NULL) {
2070 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2071 
2072 		if (cred < entry->cred)
2073 			n = n->rb_left;
2074 		else if (cred > entry->cred)
2075 			n = n->rb_right;
2076 		else
2077 			return entry;
2078 	}
2079 	return NULL;
2080 }
2081 
2082 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2083 {
2084 	struct nfs_inode *nfsi = NFS_I(inode);
2085 	struct nfs_access_entry *cache;
2086 	int err = -ENOENT;
2087 
2088 	spin_lock(&inode->i_lock);
2089 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2090 		goto out_zap;
2091 	cache = nfs_access_search_rbtree(inode, cred);
2092 	if (cache == NULL)
2093 		goto out;
2094 	if (!nfs_have_delegated_attributes(inode) &&
2095 	    !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2096 		goto out_stale;
2097 	res->jiffies = cache->jiffies;
2098 	res->cred = cache->cred;
2099 	res->mask = cache->mask;
2100 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2101 	err = 0;
2102 out:
2103 	spin_unlock(&inode->i_lock);
2104 	return err;
2105 out_stale:
2106 	rb_erase(&cache->rb_node, &nfsi->access_cache);
2107 	list_del(&cache->lru);
2108 	spin_unlock(&inode->i_lock);
2109 	nfs_access_free_entry(cache);
2110 	return -ENOENT;
2111 out_zap:
2112 	spin_unlock(&inode->i_lock);
2113 	nfs_access_zap_cache(inode);
2114 	return -ENOENT;
2115 }
2116 
2117 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2118 {
2119 	struct nfs_inode *nfsi = NFS_I(inode);
2120 	struct rb_root *root_node = &nfsi->access_cache;
2121 	struct rb_node **p = &root_node->rb_node;
2122 	struct rb_node *parent = NULL;
2123 	struct nfs_access_entry *entry;
2124 
2125 	spin_lock(&inode->i_lock);
2126 	while (*p != NULL) {
2127 		parent = *p;
2128 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2129 
2130 		if (set->cred < entry->cred)
2131 			p = &parent->rb_left;
2132 		else if (set->cred > entry->cred)
2133 			p = &parent->rb_right;
2134 		else
2135 			goto found;
2136 	}
2137 	rb_link_node(&set->rb_node, parent, p);
2138 	rb_insert_color(&set->rb_node, root_node);
2139 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2140 	spin_unlock(&inode->i_lock);
2141 	return;
2142 found:
2143 	rb_replace_node(parent, &set->rb_node, root_node);
2144 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2145 	list_del(&entry->lru);
2146 	spin_unlock(&inode->i_lock);
2147 	nfs_access_free_entry(entry);
2148 }
2149 
2150 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2151 {
2152 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2153 	if (cache == NULL)
2154 		return;
2155 	RB_CLEAR_NODE(&cache->rb_node);
2156 	cache->jiffies = set->jiffies;
2157 	cache->cred = get_rpccred(set->cred);
2158 	cache->mask = set->mask;
2159 
2160 	nfs_access_add_rbtree(inode, cache);
2161 
2162 	/* Update accounting */
2163 	smp_mb__before_atomic_inc();
2164 	atomic_long_inc(&nfs_access_nr_entries);
2165 	smp_mb__after_atomic_inc();
2166 
2167 	/* Add inode to global LRU list */
2168 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2169 		spin_lock(&nfs_access_lru_lock);
2170 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2171 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2172 					&nfs_access_lru_list);
2173 		spin_unlock(&nfs_access_lru_lock);
2174 	}
2175 }
2176 
2177 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2178 {
2179 	struct nfs_access_entry cache;
2180 	int status;
2181 
2182 	status = nfs_access_get_cached(inode, cred, &cache);
2183 	if (status == 0)
2184 		goto out;
2185 
2186 	/* Be clever: ask server to check for all possible rights */
2187 	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2188 	cache.cred = cred;
2189 	cache.jiffies = jiffies;
2190 	status = NFS_PROTO(inode)->access(inode, &cache);
2191 	if (status != 0) {
2192 		if (status == -ESTALE) {
2193 			nfs_zap_caches(inode);
2194 			if (!S_ISDIR(inode->i_mode))
2195 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2196 		}
2197 		return status;
2198 	}
2199 	nfs_access_add_cache(inode, &cache);
2200 out:
2201 	if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2202 		return 0;
2203 	return -EACCES;
2204 }
2205 
2206 static int nfs_open_permission_mask(int openflags)
2207 {
2208 	int mask = 0;
2209 
2210 	if ((openflags & O_ACCMODE) != O_WRONLY)
2211 		mask |= MAY_READ;
2212 	if ((openflags & O_ACCMODE) != O_RDONLY)
2213 		mask |= MAY_WRITE;
2214 	if (openflags & __FMODE_EXEC)
2215 		mask |= MAY_EXEC;
2216 	return mask;
2217 }
2218 
2219 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2220 {
2221 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2222 }
2223 
2224 int nfs_permission(struct inode *inode, int mask)
2225 {
2226 	struct rpc_cred *cred;
2227 	int res = 0;
2228 
2229 	if (mask & MAY_NOT_BLOCK)
2230 		return -ECHILD;
2231 
2232 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2233 
2234 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2235 		goto out;
2236 	/* Is this sys_access() ? */
2237 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2238 		goto force_lookup;
2239 
2240 	switch (inode->i_mode & S_IFMT) {
2241 		case S_IFLNK:
2242 			goto out;
2243 		case S_IFREG:
2244 			/* NFSv4 has atomic_open... */
2245 			if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2246 					&& (mask & MAY_OPEN)
2247 					&& !(mask & MAY_EXEC))
2248 				goto out;
2249 			break;
2250 		case S_IFDIR:
2251 			/*
2252 			 * Optimize away all write operations, since the server
2253 			 * will check permissions when we perform the op.
2254 			 */
2255 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2256 				goto out;
2257 	}
2258 
2259 force_lookup:
2260 	if (!NFS_PROTO(inode)->access)
2261 		goto out_notsup;
2262 
2263 	cred = rpc_lookup_cred();
2264 	if (!IS_ERR(cred)) {
2265 		res = nfs_do_access(inode, cred, mask);
2266 		put_rpccred(cred);
2267 	} else
2268 		res = PTR_ERR(cred);
2269 out:
2270 	if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2271 		res = -EACCES;
2272 
2273 	dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2274 		inode->i_sb->s_id, inode->i_ino, mask, res);
2275 	return res;
2276 out_notsup:
2277 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2278 	if (res == 0)
2279 		res = generic_permission(inode, mask);
2280 	goto out;
2281 }
2282 
2283 /*
2284  * Local variables:
2285  *  version-control: t
2286  *  kept-new-versions: 5
2287  * End:
2288  */
2289