xref: /openbmc/linux/fs/nfs/dir.c (revision f7c85868)
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 *, int, struct nameidata *);
51 static int nfs_mkdir(struct inode *, struct dentry *, int);
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 *, int, dev_t);
57 static int nfs_rename(struct inode *, struct dentry *,
58 		      struct inode *, struct dentry *);
59 static int nfs_fsync_dir(struct file *, 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 struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
115 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
116 const struct inode_operations nfs4_dir_inode_operations = {
117 	.create		= nfs_open_create,
118 	.lookup		= nfs_atomic_lookup,
119 	.link		= nfs_link,
120 	.unlink		= nfs_unlink,
121 	.symlink	= nfs_symlink,
122 	.mkdir		= nfs_mkdir,
123 	.rmdir		= nfs_rmdir,
124 	.mknod		= nfs_mknod,
125 	.rename		= nfs_rename,
126 	.permission	= nfs_permission,
127 	.getattr	= nfs_getattr,
128 	.setattr	= nfs_setattr,
129 	.getxattr	= generic_getxattr,
130 	.setxattr	= generic_setxattr,
131 	.listxattr	= generic_listxattr,
132 	.removexattr	= generic_removexattr,
133 };
134 
135 #endif /* CONFIG_NFS_V4 */
136 
137 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct rpc_cred *cred)
138 {
139 	struct nfs_open_dir_context *ctx;
140 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
141 	if (ctx != NULL) {
142 		ctx->duped = 0;
143 		ctx->dir_cookie = 0;
144 		ctx->dup_cookie = 0;
145 		ctx->cred = get_rpccred(cred);
146 	} else
147 		ctx = ERR_PTR(-ENOMEM);
148 	return ctx;
149 }
150 
151 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
152 {
153 	put_rpccred(ctx->cred);
154 	kfree(ctx);
155 }
156 
157 /*
158  * Open file
159  */
160 static int
161 nfs_opendir(struct inode *inode, struct file *filp)
162 {
163 	int res = 0;
164 	struct nfs_open_dir_context *ctx;
165 	struct rpc_cred *cred;
166 
167 	dfprintk(FILE, "NFS: open dir(%s/%s)\n",
168 			filp->f_path.dentry->d_parent->d_name.name,
169 			filp->f_path.dentry->d_name.name);
170 
171 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
172 
173 	cred = rpc_lookup_cred();
174 	if (IS_ERR(cred))
175 		return PTR_ERR(cred);
176 	ctx = alloc_nfs_open_dir_context(cred);
177 	if (IS_ERR(ctx)) {
178 		res = PTR_ERR(ctx);
179 		goto out;
180 	}
181 	filp->private_data = ctx;
182 	if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
183 		/* This is a mountpoint, so d_revalidate will never
184 		 * have been called, so we need to refresh the
185 		 * inode (for close-open consistency) ourselves.
186 		 */
187 		__nfs_revalidate_inode(NFS_SERVER(inode), inode);
188 	}
189 out:
190 	put_rpccred(cred);
191 	return res;
192 }
193 
194 static int
195 nfs_closedir(struct inode *inode, struct file *filp)
196 {
197 	put_nfs_open_dir_context(filp->private_data);
198 	return 0;
199 }
200 
201 struct nfs_cache_array_entry {
202 	u64 cookie;
203 	u64 ino;
204 	struct qstr string;
205 	unsigned char d_type;
206 };
207 
208 struct nfs_cache_array {
209 	unsigned int size;
210 	int eof_index;
211 	u64 last_cookie;
212 	struct nfs_cache_array_entry array[0];
213 };
214 
215 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
216 typedef struct {
217 	struct file	*file;
218 	struct page	*page;
219 	unsigned long	page_index;
220 	u64		*dir_cookie;
221 	u64		last_cookie;
222 	loff_t		current_index;
223 	decode_dirent_t	decode;
224 
225 	unsigned long	timestamp;
226 	unsigned long	gencount;
227 	unsigned int	cache_entry_index;
228 	unsigned int	plus:1;
229 	unsigned int	eof:1;
230 } nfs_readdir_descriptor_t;
231 
232 /*
233  * The caller is responsible for calling nfs_readdir_release_array(page)
234  */
235 static
236 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
237 {
238 	void *ptr;
239 	if (page == NULL)
240 		return ERR_PTR(-EIO);
241 	ptr = kmap(page);
242 	if (ptr == NULL)
243 		return ERR_PTR(-ENOMEM);
244 	return ptr;
245 }
246 
247 static
248 void nfs_readdir_release_array(struct page *page)
249 {
250 	kunmap(page);
251 }
252 
253 /*
254  * we are freeing strings created by nfs_add_to_readdir_array()
255  */
256 static
257 void nfs_readdir_clear_array(struct page *page)
258 {
259 	struct nfs_cache_array *array;
260 	int i;
261 
262 	array = kmap_atomic(page, KM_USER0);
263 	for (i = 0; i < array->size; i++)
264 		kfree(array->array[i].string.name);
265 	kunmap_atomic(array, KM_USER0);
266 }
267 
268 /*
269  * the caller is responsible for freeing qstr.name
270  * when called by nfs_readdir_add_to_array, the strings will be freed in
271  * nfs_clear_readdir_array()
272  */
273 static
274 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
275 {
276 	string->len = len;
277 	string->name = kmemdup(name, len, GFP_KERNEL);
278 	if (string->name == NULL)
279 		return -ENOMEM;
280 	/*
281 	 * Avoid a kmemleak false positive. The pointer to the name is stored
282 	 * in a page cache page which kmemleak does not scan.
283 	 */
284 	kmemleak_not_leak(string->name);
285 	string->hash = full_name_hash(name, len);
286 	return 0;
287 }
288 
289 static
290 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
291 {
292 	struct nfs_cache_array *array = nfs_readdir_get_array(page);
293 	struct nfs_cache_array_entry *cache_entry;
294 	int ret;
295 
296 	if (IS_ERR(array))
297 		return PTR_ERR(array);
298 
299 	cache_entry = &array->array[array->size];
300 
301 	/* Check that this entry lies within the page bounds */
302 	ret = -ENOSPC;
303 	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
304 		goto out;
305 
306 	cache_entry->cookie = entry->prev_cookie;
307 	cache_entry->ino = entry->ino;
308 	cache_entry->d_type = entry->d_type;
309 	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
310 	if (ret)
311 		goto out;
312 	array->last_cookie = entry->cookie;
313 	array->size++;
314 	if (entry->eof != 0)
315 		array->eof_index = array->size;
316 out:
317 	nfs_readdir_release_array(page);
318 	return ret;
319 }
320 
321 static
322 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
323 {
324 	loff_t diff = desc->file->f_pos - desc->current_index;
325 	unsigned int index;
326 	struct nfs_open_dir_context *ctx = desc->file->private_data;
327 
328 	if (diff < 0)
329 		goto out_eof;
330 	if (diff >= array->size) {
331 		if (array->eof_index >= 0)
332 			goto out_eof;
333 		return -EAGAIN;
334 	}
335 
336 	index = (unsigned int)diff;
337 	*desc->dir_cookie = array->array[index].cookie;
338 	desc->cache_entry_index = index;
339 	ctx->duped = 0;
340 	return 0;
341 out_eof:
342 	desc->eof = 1;
343 	return -EBADCOOKIE;
344 }
345 
346 static
347 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
348 {
349 	int i;
350 	loff_t new_pos;
351 	int status = -EAGAIN;
352 	struct nfs_open_dir_context *ctx = desc->file->private_data;
353 
354 	for (i = 0; i < array->size; i++) {
355 		if (array->array[i].cookie == *desc->dir_cookie) {
356 			new_pos = desc->current_index + i;
357 			if (new_pos < desc->file->f_pos) {
358 				ctx->dup_cookie = *desc->dir_cookie;
359 				ctx->duped = 1;
360 			}
361 			desc->file->f_pos = new_pos;
362 			desc->cache_entry_index = i;
363 			return 0;
364 		}
365 	}
366 	if (array->eof_index >= 0) {
367 		status = -EBADCOOKIE;
368 		if (*desc->dir_cookie == array->last_cookie)
369 			desc->eof = 1;
370 	}
371 	return status;
372 }
373 
374 static
375 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
376 {
377 	struct nfs_cache_array *array;
378 	int status;
379 
380 	array = nfs_readdir_get_array(desc->page);
381 	if (IS_ERR(array)) {
382 		status = PTR_ERR(array);
383 		goto out;
384 	}
385 
386 	if (*desc->dir_cookie == 0)
387 		status = nfs_readdir_search_for_pos(array, desc);
388 	else
389 		status = nfs_readdir_search_for_cookie(array, desc);
390 
391 	if (status == -EAGAIN) {
392 		desc->last_cookie = array->last_cookie;
393 		desc->current_index += array->size;
394 		desc->page_index++;
395 	}
396 	nfs_readdir_release_array(desc->page);
397 out:
398 	return status;
399 }
400 
401 /* Fill a page with xdr information before transferring to the cache page */
402 static
403 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
404 			struct nfs_entry *entry, struct file *file, struct inode *inode)
405 {
406 	struct nfs_open_dir_context *ctx = file->private_data;
407 	struct rpc_cred	*cred = ctx->cred;
408 	unsigned long	timestamp, gencount;
409 	int		error;
410 
411  again:
412 	timestamp = jiffies;
413 	gencount = nfs_inc_attr_generation_counter();
414 	error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
415 					  NFS_SERVER(inode)->dtsize, desc->plus);
416 	if (error < 0) {
417 		/* We requested READDIRPLUS, but the server doesn't grok it */
418 		if (error == -ENOTSUPP && desc->plus) {
419 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
420 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
421 			desc->plus = 0;
422 			goto again;
423 		}
424 		goto error;
425 	}
426 	desc->timestamp = timestamp;
427 	desc->gencount = gencount;
428 error:
429 	return error;
430 }
431 
432 static int xdr_decode(nfs_readdir_descriptor_t *desc,
433 		      struct nfs_entry *entry, struct xdr_stream *xdr)
434 {
435 	int error;
436 
437 	error = desc->decode(xdr, entry, desc->plus);
438 	if (error)
439 		return error;
440 	entry->fattr->time_start = desc->timestamp;
441 	entry->fattr->gencount = desc->gencount;
442 	return 0;
443 }
444 
445 static
446 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
447 {
448 	if (dentry->d_inode == NULL)
449 		goto different;
450 	if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
451 		goto different;
452 	return 1;
453 different:
454 	return 0;
455 }
456 
457 static
458 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
459 {
460 	struct qstr filename = {
461 		.len = entry->len,
462 		.name = entry->name,
463 	};
464 	struct dentry *dentry;
465 	struct dentry *alias;
466 	struct inode *dir = parent->d_inode;
467 	struct inode *inode;
468 
469 	if (filename.name[0] == '.') {
470 		if (filename.len == 1)
471 			return;
472 		if (filename.len == 2 && filename.name[1] == '.')
473 			return;
474 	}
475 	filename.hash = full_name_hash(filename.name, filename.len);
476 
477 	dentry = d_lookup(parent, &filename);
478 	if (dentry != NULL) {
479 		if (nfs_same_file(dentry, entry)) {
480 			nfs_refresh_inode(dentry->d_inode, entry->fattr);
481 			goto out;
482 		} else {
483 			d_drop(dentry);
484 			dput(dentry);
485 		}
486 	}
487 
488 	dentry = d_alloc(parent, &filename);
489 	if (dentry == NULL)
490 		return;
491 
492 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
493 	if (IS_ERR(inode))
494 		goto out;
495 
496 	alias = d_materialise_unique(dentry, inode);
497 	if (IS_ERR(alias))
498 		goto out;
499 	else if (alias) {
500 		nfs_set_verifier(alias, nfs_save_change_attribute(dir));
501 		dput(alias);
502 	} else
503 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
504 
505 out:
506 	dput(dentry);
507 }
508 
509 /* Perform conversion from xdr to cache array */
510 static
511 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
512 				struct page **xdr_pages, struct page *page, unsigned int buflen)
513 {
514 	struct xdr_stream stream;
515 	struct xdr_buf buf;
516 	struct page *scratch;
517 	struct nfs_cache_array *array;
518 	unsigned int count = 0;
519 	int status;
520 
521 	scratch = alloc_page(GFP_KERNEL);
522 	if (scratch == NULL)
523 		return -ENOMEM;
524 
525 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
526 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
527 
528 	do {
529 		status = xdr_decode(desc, entry, &stream);
530 		if (status != 0) {
531 			if (status == -EAGAIN)
532 				status = 0;
533 			break;
534 		}
535 
536 		count++;
537 
538 		if (desc->plus != 0)
539 			nfs_prime_dcache(desc->file->f_path.dentry, entry);
540 
541 		status = nfs_readdir_add_to_array(entry, page);
542 		if (status != 0)
543 			break;
544 	} while (!entry->eof);
545 
546 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
547 		array = nfs_readdir_get_array(page);
548 		if (!IS_ERR(array)) {
549 			array->eof_index = array->size;
550 			status = 0;
551 			nfs_readdir_release_array(page);
552 		} else
553 			status = PTR_ERR(array);
554 	}
555 
556 	put_page(scratch);
557 	return status;
558 }
559 
560 static
561 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
562 {
563 	unsigned int i;
564 	for (i = 0; i < npages; i++)
565 		put_page(pages[i]);
566 }
567 
568 static
569 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
570 		unsigned int npages)
571 {
572 	nfs_readdir_free_pagearray(pages, npages);
573 }
574 
575 /*
576  * nfs_readdir_large_page will allocate pages that must be freed with a call
577  * to nfs_readdir_free_large_page
578  */
579 static
580 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
581 {
582 	unsigned int i;
583 
584 	for (i = 0; i < npages; i++) {
585 		struct page *page = alloc_page(GFP_KERNEL);
586 		if (page == NULL)
587 			goto out_freepages;
588 		pages[i] = page;
589 	}
590 	return 0;
591 
592 out_freepages:
593 	nfs_readdir_free_pagearray(pages, i);
594 	return -ENOMEM;
595 }
596 
597 static
598 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
599 {
600 	struct page *pages[NFS_MAX_READDIR_PAGES];
601 	void *pages_ptr = NULL;
602 	struct nfs_entry entry;
603 	struct file	*file = desc->file;
604 	struct nfs_cache_array *array;
605 	int status = -ENOMEM;
606 	unsigned int array_size = ARRAY_SIZE(pages);
607 
608 	entry.prev_cookie = 0;
609 	entry.cookie = desc->last_cookie;
610 	entry.eof = 0;
611 	entry.fh = nfs_alloc_fhandle();
612 	entry.fattr = nfs_alloc_fattr();
613 	entry.server = NFS_SERVER(inode);
614 	if (entry.fh == NULL || entry.fattr == NULL)
615 		goto out;
616 
617 	array = nfs_readdir_get_array(page);
618 	if (IS_ERR(array)) {
619 		status = PTR_ERR(array);
620 		goto out;
621 	}
622 	memset(array, 0, sizeof(struct nfs_cache_array));
623 	array->eof_index = -1;
624 
625 	status = nfs_readdir_large_page(pages, array_size);
626 	if (status < 0)
627 		goto out_release_array;
628 	do {
629 		unsigned int pglen;
630 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
631 
632 		if (status < 0)
633 			break;
634 		pglen = status;
635 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
636 		if (status < 0) {
637 			if (status == -ENOSPC)
638 				status = 0;
639 			break;
640 		}
641 	} while (array->eof_index < 0);
642 
643 	nfs_readdir_free_large_page(pages_ptr, pages, array_size);
644 out_release_array:
645 	nfs_readdir_release_array(page);
646 out:
647 	nfs_free_fattr(entry.fattr);
648 	nfs_free_fhandle(entry.fh);
649 	return status;
650 }
651 
652 /*
653  * Now we cache directories properly, by converting xdr information
654  * to an array that can be used for lookups later.  This results in
655  * fewer cache pages, since we can store more information on each page.
656  * We only need to convert from xdr once so future lookups are much simpler
657  */
658 static
659 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
660 {
661 	struct inode	*inode = desc->file->f_path.dentry->d_inode;
662 	int ret;
663 
664 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
665 	if (ret < 0)
666 		goto error;
667 	SetPageUptodate(page);
668 
669 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
670 		/* Should never happen */
671 		nfs_zap_mapping(inode, inode->i_mapping);
672 	}
673 	unlock_page(page);
674 	return 0;
675  error:
676 	unlock_page(page);
677 	return ret;
678 }
679 
680 static
681 void cache_page_release(nfs_readdir_descriptor_t *desc)
682 {
683 	if (!desc->page->mapping)
684 		nfs_readdir_clear_array(desc->page);
685 	page_cache_release(desc->page);
686 	desc->page = NULL;
687 }
688 
689 static
690 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
691 {
692 	return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
693 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
694 }
695 
696 /*
697  * Returns 0 if desc->dir_cookie was found on page desc->page_index
698  */
699 static
700 int find_cache_page(nfs_readdir_descriptor_t *desc)
701 {
702 	int res;
703 
704 	desc->page = get_cache_page(desc);
705 	if (IS_ERR(desc->page))
706 		return PTR_ERR(desc->page);
707 
708 	res = nfs_readdir_search_array(desc);
709 	if (res != 0)
710 		cache_page_release(desc);
711 	return res;
712 }
713 
714 /* Search for desc->dir_cookie from the beginning of the page cache */
715 static inline
716 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
717 {
718 	int res;
719 
720 	if (desc->page_index == 0) {
721 		desc->current_index = 0;
722 		desc->last_cookie = 0;
723 	}
724 	do {
725 		res = find_cache_page(desc);
726 	} while (res == -EAGAIN);
727 	return res;
728 }
729 
730 /*
731  * Once we've found the start of the dirent within a page: fill 'er up...
732  */
733 static
734 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
735 		   filldir_t filldir)
736 {
737 	struct file	*file = desc->file;
738 	int i = 0;
739 	int res = 0;
740 	struct nfs_cache_array *array = NULL;
741 	struct nfs_open_dir_context *ctx = file->private_data;
742 
743 	if (ctx->duped != 0 && ctx->dup_cookie == *desc->dir_cookie) {
744 		if (printk_ratelimit()) {
745 			pr_notice("NFS: directory %s/%s contains a readdir loop.  "
746 				"Please contact your server vendor.  "
747 				"Offending cookie: %llu\n",
748 				file->f_dentry->d_parent->d_name.name,
749 				file->f_dentry->d_name.name,
750 				*desc->dir_cookie);
751 		}
752 		res = -ELOOP;
753 		goto out;
754 	}
755 
756 	array = nfs_readdir_get_array(desc->page);
757 	if (IS_ERR(array)) {
758 		res = PTR_ERR(array);
759 		goto out;
760 	}
761 
762 	for (i = desc->cache_entry_index; i < array->size; i++) {
763 		struct nfs_cache_array_entry *ent;
764 
765 		ent = &array->array[i];
766 		if (filldir(dirent, ent->string.name, ent->string.len,
767 		    file->f_pos, nfs_compat_user_ino64(ent->ino),
768 		    ent->d_type) < 0) {
769 			desc->eof = 1;
770 			break;
771 		}
772 		file->f_pos++;
773 		if (i < (array->size-1))
774 			*desc->dir_cookie = array->array[i+1].cookie;
775 		else
776 			*desc->dir_cookie = array->last_cookie;
777 	}
778 	if (array->eof_index >= 0)
779 		desc->eof = 1;
780 
781 	nfs_readdir_release_array(desc->page);
782 out:
783 	cache_page_release(desc);
784 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
785 			(unsigned long long)*desc->dir_cookie, res);
786 	return res;
787 }
788 
789 /*
790  * If we cannot find a cookie in our cache, we suspect that this is
791  * because it points to a deleted file, so we ask the server to return
792  * whatever it thinks is the next entry. We then feed this to filldir.
793  * If all goes well, we should then be able to find our way round the
794  * cache on the next call to readdir_search_pagecache();
795  *
796  * NOTE: we cannot add the anonymous page to the pagecache because
797  *	 the data it contains might not be page aligned. Besides,
798  *	 we should already have a complete representation of the
799  *	 directory in the page cache by the time we get here.
800  */
801 static inline
802 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
803 		     filldir_t filldir)
804 {
805 	struct page	*page = NULL;
806 	int		status;
807 	struct inode *inode = desc->file->f_path.dentry->d_inode;
808 
809 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
810 			(unsigned long long)*desc->dir_cookie);
811 
812 	page = alloc_page(GFP_HIGHUSER);
813 	if (!page) {
814 		status = -ENOMEM;
815 		goto out;
816 	}
817 
818 	desc->page_index = 0;
819 	desc->last_cookie = *desc->dir_cookie;
820 	desc->page = page;
821 
822 	status = nfs_readdir_xdr_to_array(desc, page, inode);
823 	if (status < 0)
824 		goto out_release;
825 
826 	status = nfs_do_filldir(desc, dirent, filldir);
827 
828  out:
829 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
830 			__func__, status);
831 	return status;
832  out_release:
833 	cache_page_release(desc);
834 	goto out;
835 }
836 
837 /* The file offset position represents the dirent entry number.  A
838    last cookie cache takes care of the common case of reading the
839    whole directory.
840  */
841 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
842 {
843 	struct dentry	*dentry = filp->f_path.dentry;
844 	struct inode	*inode = dentry->d_inode;
845 	nfs_readdir_descriptor_t my_desc,
846 			*desc = &my_desc;
847 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
848 	int res;
849 
850 	dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
851 			dentry->d_parent->d_name.name, dentry->d_name.name,
852 			(long long)filp->f_pos);
853 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
854 
855 	/*
856 	 * filp->f_pos points to the dirent entry number.
857 	 * *desc->dir_cookie has the cookie for the next entry. We have
858 	 * to either find the entry with the appropriate number or
859 	 * revalidate the cookie.
860 	 */
861 	memset(desc, 0, sizeof(*desc));
862 
863 	desc->file = filp;
864 	desc->dir_cookie = &dir_ctx->dir_cookie;
865 	desc->decode = NFS_PROTO(inode)->decode_dirent;
866 	desc->plus = NFS_USE_READDIRPLUS(inode);
867 
868 	nfs_block_sillyrename(dentry);
869 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
870 	if (res < 0)
871 		goto out;
872 
873 	do {
874 		res = readdir_search_pagecache(desc);
875 
876 		if (res == -EBADCOOKIE) {
877 			res = 0;
878 			/* This means either end of directory */
879 			if (*desc->dir_cookie && desc->eof == 0) {
880 				/* Or that the server has 'lost' a cookie */
881 				res = uncached_readdir(desc, dirent, filldir);
882 				if (res == 0)
883 					continue;
884 			}
885 			break;
886 		}
887 		if (res == -ETOOSMALL && desc->plus) {
888 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
889 			nfs_zap_caches(inode);
890 			desc->page_index = 0;
891 			desc->plus = 0;
892 			desc->eof = 0;
893 			continue;
894 		}
895 		if (res < 0)
896 			break;
897 
898 		res = nfs_do_filldir(desc, dirent, filldir);
899 		if (res < 0)
900 			break;
901 	} while (!desc->eof);
902 out:
903 	nfs_unblock_sillyrename(dentry);
904 	if (res > 0)
905 		res = 0;
906 	dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
907 			dentry->d_parent->d_name.name, dentry->d_name.name,
908 			res);
909 	return res;
910 }
911 
912 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
913 {
914 	struct dentry *dentry = filp->f_path.dentry;
915 	struct inode *inode = dentry->d_inode;
916 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
917 
918 	dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
919 			dentry->d_parent->d_name.name,
920 			dentry->d_name.name,
921 			offset, origin);
922 
923 	mutex_lock(&inode->i_mutex);
924 	switch (origin) {
925 		case 1:
926 			offset += filp->f_pos;
927 		case 0:
928 			if (offset >= 0)
929 				break;
930 		default:
931 			offset = -EINVAL;
932 			goto out;
933 	}
934 	if (offset != filp->f_pos) {
935 		filp->f_pos = offset;
936 		dir_ctx->dir_cookie = 0;
937 		dir_ctx->duped = 0;
938 	}
939 out:
940 	mutex_unlock(&inode->i_mutex);
941 	return offset;
942 }
943 
944 /*
945  * All directory operations under NFS are synchronous, so fsync()
946  * is a dummy operation.
947  */
948 static int nfs_fsync_dir(struct file *filp, int datasync)
949 {
950 	struct dentry *dentry = filp->f_path.dentry;
951 
952 	dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
953 			dentry->d_parent->d_name.name, dentry->d_name.name,
954 			datasync);
955 
956 	nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
957 	return 0;
958 }
959 
960 /**
961  * nfs_force_lookup_revalidate - Mark the directory as having changed
962  * @dir - pointer to directory inode
963  *
964  * This forces the revalidation code in nfs_lookup_revalidate() to do a
965  * full lookup on all child dentries of 'dir' whenever a change occurs
966  * on the server that might have invalidated our dcache.
967  *
968  * The caller should be holding dir->i_lock
969  */
970 void nfs_force_lookup_revalidate(struct inode *dir)
971 {
972 	NFS_I(dir)->cache_change_attribute++;
973 }
974 
975 /*
976  * A check for whether or not the parent directory has changed.
977  * In the case it has, we assume that the dentries are untrustworthy
978  * and may need to be looked up again.
979  */
980 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
981 {
982 	if (IS_ROOT(dentry))
983 		return 1;
984 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
985 		return 0;
986 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
987 		return 0;
988 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
989 	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
990 		return 0;
991 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
992 		return 0;
993 	return 1;
994 }
995 
996 /*
997  * Return the intent data that applies to this particular path component
998  *
999  * Note that the current set of intents only apply to the very last
1000  * component of the path.
1001  * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
1002  */
1003 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1004 						unsigned int mask)
1005 {
1006 	if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
1007 		return 0;
1008 	return nd->flags & mask;
1009 }
1010 
1011 /*
1012  * Use intent information to check whether or not we're going to do
1013  * an O_EXCL create using this path component.
1014  */
1015 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1016 {
1017 	if (NFS_PROTO(dir)->version == 2)
1018 		return 0;
1019 	return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1020 }
1021 
1022 /*
1023  * Inode and filehandle revalidation for lookups.
1024  *
1025  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1026  * or if the intent information indicates that we're about to open this
1027  * particular file and the "nocto" mount flag is not set.
1028  *
1029  */
1030 static inline
1031 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1032 {
1033 	struct nfs_server *server = NFS_SERVER(inode);
1034 
1035 	if (IS_AUTOMOUNT(inode))
1036 		return 0;
1037 	if (nd != NULL) {
1038 		/* VFS wants an on-the-wire revalidation */
1039 		if (nd->flags & LOOKUP_REVAL)
1040 			goto out_force;
1041 		/* This is an open(2) */
1042 		if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1043 				!(server->flags & NFS_MOUNT_NOCTO) &&
1044 				(S_ISREG(inode->i_mode) ||
1045 				 S_ISDIR(inode->i_mode)))
1046 			goto out_force;
1047 		return 0;
1048 	}
1049 	return nfs_revalidate_inode(server, inode);
1050 out_force:
1051 	return __nfs_revalidate_inode(server, inode);
1052 }
1053 
1054 /*
1055  * We judge how long we want to trust negative
1056  * dentries by looking at the parent inode mtime.
1057  *
1058  * If parent mtime has changed, we revalidate, else we wait for a
1059  * period corresponding to the parent's attribute cache timeout value.
1060  */
1061 static inline
1062 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1063 		       struct nameidata *nd)
1064 {
1065 	/* Don't revalidate a negative dentry if we're creating a new file */
1066 	if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1067 		return 0;
1068 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1069 		return 1;
1070 	return !nfs_check_verifier(dir, dentry);
1071 }
1072 
1073 /*
1074  * This is called every time the dcache has a lookup hit,
1075  * and we should check whether we can really trust that
1076  * lookup.
1077  *
1078  * NOTE! The hit can be a negative hit too, don't assume
1079  * we have an inode!
1080  *
1081  * If the parent directory is seen to have changed, we throw out the
1082  * cached dentry and do a new lookup.
1083  */
1084 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1085 {
1086 	struct inode *dir;
1087 	struct inode *inode;
1088 	struct dentry *parent;
1089 	struct nfs_fh *fhandle = NULL;
1090 	struct nfs_fattr *fattr = NULL;
1091 	int error;
1092 
1093 	if (nd->flags & LOOKUP_RCU)
1094 		return -ECHILD;
1095 
1096 	parent = dget_parent(dentry);
1097 	dir = parent->d_inode;
1098 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1099 	inode = dentry->d_inode;
1100 
1101 	if (!inode) {
1102 		if (nfs_neg_need_reval(dir, dentry, nd))
1103 			goto out_bad;
1104 		goto out_valid;
1105 	}
1106 
1107 	if (is_bad_inode(inode)) {
1108 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1109 				__func__, dentry->d_parent->d_name.name,
1110 				dentry->d_name.name);
1111 		goto out_bad;
1112 	}
1113 
1114 	if (nfs_have_delegation(inode, FMODE_READ))
1115 		goto out_set_verifier;
1116 
1117 	/* Force a full look up iff the parent directory has changed */
1118 	if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1119 		if (nfs_lookup_verify_inode(inode, nd))
1120 			goto out_zap_parent;
1121 		goto out_valid;
1122 	}
1123 
1124 	if (NFS_STALE(inode))
1125 		goto out_bad;
1126 
1127 	error = -ENOMEM;
1128 	fhandle = nfs_alloc_fhandle();
1129 	fattr = nfs_alloc_fattr();
1130 	if (fhandle == NULL || fattr == NULL)
1131 		goto out_error;
1132 
1133 	error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1134 	if (error)
1135 		goto out_bad;
1136 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1137 		goto out_bad;
1138 	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1139 		goto out_bad;
1140 
1141 	nfs_free_fattr(fattr);
1142 	nfs_free_fhandle(fhandle);
1143 out_set_verifier:
1144 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1145  out_valid:
1146 	dput(parent);
1147 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1148 			__func__, dentry->d_parent->d_name.name,
1149 			dentry->d_name.name);
1150 	return 1;
1151 out_zap_parent:
1152 	nfs_zap_caches(dir);
1153  out_bad:
1154 	nfs_mark_for_revalidate(dir);
1155 	if (inode && S_ISDIR(inode->i_mode)) {
1156 		/* Purge readdir caches. */
1157 		nfs_zap_caches(inode);
1158 		/* If we have submounts, don't unhash ! */
1159 		if (have_submounts(dentry))
1160 			goto out_valid;
1161 		if (dentry->d_flags & DCACHE_DISCONNECTED)
1162 			goto out_valid;
1163 		shrink_dcache_parent(dentry);
1164 	}
1165 	d_drop(dentry);
1166 	nfs_free_fattr(fattr);
1167 	nfs_free_fhandle(fhandle);
1168 	dput(parent);
1169 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1170 			__func__, dentry->d_parent->d_name.name,
1171 			dentry->d_name.name);
1172 	return 0;
1173 out_error:
1174 	nfs_free_fattr(fattr);
1175 	nfs_free_fhandle(fhandle);
1176 	dput(parent);
1177 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1178 			__func__, dentry->d_parent->d_name.name,
1179 			dentry->d_name.name, error);
1180 	return error;
1181 }
1182 
1183 /*
1184  * This is called from dput() when d_count is going to 0.
1185  */
1186 static int nfs_dentry_delete(const struct dentry *dentry)
1187 {
1188 	dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1189 		dentry->d_parent->d_name.name, dentry->d_name.name,
1190 		dentry->d_flags);
1191 
1192 	/* Unhash any dentry with a stale inode */
1193 	if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1194 		return 1;
1195 
1196 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1197 		/* Unhash it, so that ->d_iput() would be called */
1198 		return 1;
1199 	}
1200 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1201 		/* Unhash it, so that ancestors of killed async unlink
1202 		 * files will be cleaned up during umount */
1203 		return 1;
1204 	}
1205 	return 0;
1206 
1207 }
1208 
1209 static void nfs_drop_nlink(struct inode *inode)
1210 {
1211 	spin_lock(&inode->i_lock);
1212 	if (inode->i_nlink > 0)
1213 		drop_nlink(inode);
1214 	spin_unlock(&inode->i_lock);
1215 }
1216 
1217 /*
1218  * Called when the dentry loses inode.
1219  * We use it to clean up silly-renamed files.
1220  */
1221 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1222 {
1223 	if (S_ISDIR(inode->i_mode))
1224 		/* drop any readdir cache as it could easily be old */
1225 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1226 
1227 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1228 		drop_nlink(inode);
1229 		nfs_complete_unlink(dentry, inode);
1230 	}
1231 	iput(inode);
1232 }
1233 
1234 static void nfs_d_release(struct dentry *dentry)
1235 {
1236 	/* free cached devname value, if it survived that far */
1237 	if (unlikely(dentry->d_fsdata)) {
1238 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1239 			WARN_ON(1);
1240 		else
1241 			kfree(dentry->d_fsdata);
1242 	}
1243 }
1244 
1245 const struct dentry_operations nfs_dentry_operations = {
1246 	.d_revalidate	= nfs_lookup_revalidate,
1247 	.d_delete	= nfs_dentry_delete,
1248 	.d_iput		= nfs_dentry_iput,
1249 	.d_automount	= nfs_d_automount,
1250 	.d_release	= nfs_d_release,
1251 };
1252 
1253 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1254 {
1255 	struct dentry *res;
1256 	struct dentry *parent;
1257 	struct inode *inode = NULL;
1258 	struct nfs_fh *fhandle = NULL;
1259 	struct nfs_fattr *fattr = NULL;
1260 	int error;
1261 
1262 	dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1263 		dentry->d_parent->d_name.name, dentry->d_name.name);
1264 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1265 
1266 	res = ERR_PTR(-ENAMETOOLONG);
1267 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1268 		goto out;
1269 
1270 	/*
1271 	 * If we're doing an exclusive create, optimize away the lookup
1272 	 * but don't hash the dentry.
1273 	 */
1274 	if (nfs_is_exclusive_create(dir, nd)) {
1275 		d_instantiate(dentry, NULL);
1276 		res = NULL;
1277 		goto out;
1278 	}
1279 
1280 	res = ERR_PTR(-ENOMEM);
1281 	fhandle = nfs_alloc_fhandle();
1282 	fattr = nfs_alloc_fattr();
1283 	if (fhandle == NULL || fattr == NULL)
1284 		goto out;
1285 
1286 	parent = dentry->d_parent;
1287 	/* Protect against concurrent sillydeletes */
1288 	nfs_block_sillyrename(parent);
1289 	error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1290 	if (error == -ENOENT)
1291 		goto no_entry;
1292 	if (error < 0) {
1293 		res = ERR_PTR(error);
1294 		goto out_unblock_sillyrename;
1295 	}
1296 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1297 	res = ERR_CAST(inode);
1298 	if (IS_ERR(res))
1299 		goto out_unblock_sillyrename;
1300 
1301 no_entry:
1302 	res = d_materialise_unique(dentry, inode);
1303 	if (res != NULL) {
1304 		if (IS_ERR(res))
1305 			goto out_unblock_sillyrename;
1306 		dentry = res;
1307 	}
1308 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1309 out_unblock_sillyrename:
1310 	nfs_unblock_sillyrename(parent);
1311 out:
1312 	nfs_free_fattr(fattr);
1313 	nfs_free_fhandle(fhandle);
1314 	return res;
1315 }
1316 
1317 #ifdef CONFIG_NFS_V4
1318 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1319 
1320 const struct dentry_operations nfs4_dentry_operations = {
1321 	.d_revalidate	= nfs_open_revalidate,
1322 	.d_delete	= nfs_dentry_delete,
1323 	.d_iput		= nfs_dentry_iput,
1324 	.d_automount	= nfs_d_automount,
1325 	.d_release	= nfs_d_release,
1326 };
1327 
1328 /*
1329  * Use intent information to determine whether we need to substitute
1330  * the NFSv4-style stateful OPEN for the LOOKUP call
1331  */
1332 static int is_atomic_open(struct nameidata *nd)
1333 {
1334 	if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1335 		return 0;
1336 	/* NFS does not (yet) have a stateful open for directories */
1337 	if (nd->flags & LOOKUP_DIRECTORY)
1338 		return 0;
1339 	/* Are we trying to write to a read only partition? */
1340 	if (__mnt_is_readonly(nd->path.mnt) &&
1341 	    (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1342 		return 0;
1343 	return 1;
1344 }
1345 
1346 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1347 {
1348 	struct nfs_open_context *ctx;
1349 	struct rpc_cred *cred;
1350 	fmode_t fmode = open_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1351 
1352 	cred = rpc_lookup_cred();
1353 	if (IS_ERR(cred))
1354 		return ERR_CAST(cred);
1355 	ctx = alloc_nfs_open_context(dentry, cred, fmode);
1356 	put_rpccred(cred);
1357 	if (ctx == NULL)
1358 		return ERR_PTR(-ENOMEM);
1359 	return ctx;
1360 }
1361 
1362 static int do_open(struct inode *inode, struct file *filp)
1363 {
1364 	nfs_fscache_set_inode_cookie(inode, filp);
1365 	return 0;
1366 }
1367 
1368 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1369 {
1370 	struct file *filp;
1371 	int ret = 0;
1372 
1373 	/* If the open_intent is for execute, we have an extra check to make */
1374 	if (ctx->mode & FMODE_EXEC) {
1375 		ret = nfs_may_open(ctx->dentry->d_inode,
1376 				ctx->cred,
1377 				nd->intent.open.flags);
1378 		if (ret < 0)
1379 			goto out;
1380 	}
1381 	filp = lookup_instantiate_filp(nd, ctx->dentry, do_open);
1382 	if (IS_ERR(filp))
1383 		ret = PTR_ERR(filp);
1384 	else
1385 		nfs_file_set_open_context(filp, ctx);
1386 out:
1387 	put_nfs_open_context(ctx);
1388 	return ret;
1389 }
1390 
1391 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1392 {
1393 	struct nfs_open_context *ctx;
1394 	struct iattr attr;
1395 	struct dentry *res = NULL;
1396 	struct inode *inode;
1397 	int open_flags;
1398 	int err;
1399 
1400 	dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1401 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1402 
1403 	/* Check that we are indeed trying to open this file */
1404 	if (!is_atomic_open(nd))
1405 		goto no_open;
1406 
1407 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1408 		res = ERR_PTR(-ENAMETOOLONG);
1409 		goto out;
1410 	}
1411 
1412 	/* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1413 	 * the dentry. */
1414 	if (nd->flags & LOOKUP_EXCL) {
1415 		d_instantiate(dentry, NULL);
1416 		goto out;
1417 	}
1418 
1419 	open_flags = nd->intent.open.flags;
1420 
1421 	ctx = create_nfs_open_context(dentry, open_flags);
1422 	res = ERR_CAST(ctx);
1423 	if (IS_ERR(ctx))
1424 		goto out;
1425 
1426 	if (nd->flags & LOOKUP_CREATE) {
1427 		attr.ia_mode = nd->intent.open.create_mode;
1428 		attr.ia_valid = ATTR_MODE;
1429 		attr.ia_mode &= ~current_umask();
1430 	} else {
1431 		open_flags &= ~(O_EXCL | O_CREAT);
1432 		attr.ia_valid = 0;
1433 	}
1434 
1435 	/* Open the file on the server */
1436 	nfs_block_sillyrename(dentry->d_parent);
1437 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1438 	if (IS_ERR(inode)) {
1439 		nfs_unblock_sillyrename(dentry->d_parent);
1440 		put_nfs_open_context(ctx);
1441 		switch (PTR_ERR(inode)) {
1442 			/* Make a negative dentry */
1443 			case -ENOENT:
1444 				d_add(dentry, NULL);
1445 				res = NULL;
1446 				goto out;
1447 			/* This turned out not to be a regular file */
1448 			case -ENOTDIR:
1449 				goto no_open;
1450 			case -ELOOP:
1451 				if (!(nd->intent.open.flags & O_NOFOLLOW))
1452 					goto no_open;
1453 			/* case -EISDIR: */
1454 			/* case -EINVAL: */
1455 			default:
1456 				res = ERR_CAST(inode);
1457 				goto out;
1458 		}
1459 	}
1460 	res = d_add_unique(dentry, inode);
1461 	nfs_unblock_sillyrename(dentry->d_parent);
1462 	if (res != NULL) {
1463 		dput(ctx->dentry);
1464 		ctx->dentry = dget(res);
1465 		dentry = res;
1466 	}
1467 	err = nfs_intent_set_file(nd, ctx);
1468 	if (err < 0) {
1469 		if (res != NULL)
1470 			dput(res);
1471 		return ERR_PTR(err);
1472 	}
1473 out:
1474 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1475 	return res;
1476 no_open:
1477 	return nfs_lookup(dir, dentry, nd);
1478 }
1479 
1480 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1481 {
1482 	struct dentry *parent = NULL;
1483 	struct inode *inode;
1484 	struct inode *dir;
1485 	struct nfs_open_context *ctx;
1486 	int openflags, ret = 0;
1487 
1488 	if (nd->flags & LOOKUP_RCU)
1489 		return -ECHILD;
1490 
1491 	inode = dentry->d_inode;
1492 	if (!is_atomic_open(nd) || d_mountpoint(dentry))
1493 		goto no_open;
1494 
1495 	parent = dget_parent(dentry);
1496 	dir = parent->d_inode;
1497 
1498 	/* We can't create new files in nfs_open_revalidate(), so we
1499 	 * optimize away revalidation of negative dentries.
1500 	 */
1501 	if (inode == NULL) {
1502 		if (!nfs_neg_need_reval(dir, dentry, nd))
1503 			ret = 1;
1504 		goto out;
1505 	}
1506 
1507 	/* NFS only supports OPEN on regular files */
1508 	if (!S_ISREG(inode->i_mode))
1509 		goto no_open_dput;
1510 	openflags = nd->intent.open.flags;
1511 	/* We cannot do exclusive creation on a positive dentry */
1512 	if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1513 		goto no_open_dput;
1514 	/* We can't create new files, or truncate existing ones here */
1515 	openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1516 
1517 	ctx = create_nfs_open_context(dentry, openflags);
1518 	ret = PTR_ERR(ctx);
1519 	if (IS_ERR(ctx))
1520 		goto out;
1521 	/*
1522 	 * Note: we're not holding inode->i_mutex and so may be racing with
1523 	 * operations that change the directory. We therefore save the
1524 	 * change attribute *before* we do the RPC call.
1525 	 */
1526 	inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1527 	if (IS_ERR(inode)) {
1528 		ret = PTR_ERR(inode);
1529 		switch (ret) {
1530 		case -EPERM:
1531 		case -EACCES:
1532 		case -EDQUOT:
1533 		case -ENOSPC:
1534 		case -EROFS:
1535 			goto out_put_ctx;
1536 		default:
1537 			goto out_drop;
1538 		}
1539 	}
1540 	iput(inode);
1541 	if (inode != dentry->d_inode)
1542 		goto out_drop;
1543 
1544 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1545 	ret = nfs_intent_set_file(nd, ctx);
1546 	if (ret >= 0)
1547 		ret = 1;
1548 out:
1549 	dput(parent);
1550 	return ret;
1551 out_drop:
1552 	d_drop(dentry);
1553 	ret = 0;
1554 out_put_ctx:
1555 	put_nfs_open_context(ctx);
1556 	goto out;
1557 
1558 no_open_dput:
1559 	dput(parent);
1560 no_open:
1561 	return nfs_lookup_revalidate(dentry, nd);
1562 }
1563 
1564 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1565 		struct nameidata *nd)
1566 {
1567 	struct nfs_open_context *ctx = NULL;
1568 	struct iattr attr;
1569 	int error;
1570 	int open_flags = O_CREAT|O_EXCL|FMODE_READ;
1571 
1572 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1573 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1574 
1575 	attr.ia_mode = mode;
1576 	attr.ia_valid = ATTR_MODE;
1577 
1578 	if (nd && (nd->flags & LOOKUP_OPEN) != 0)
1579 		open_flags = nd->intent.open.flags;
1580 
1581 	ctx = create_nfs_open_context(dentry, open_flags);
1582 	error = PTR_ERR(ctx);
1583 	if (IS_ERR(ctx))
1584 		goto out_err_drop;
1585 
1586 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1587 	if (error != 0)
1588 		goto out_put_ctx;
1589 	if (nd && (nd->flags & LOOKUP_OPEN) != 0) {
1590 		error = nfs_intent_set_file(nd, ctx);
1591 		if (error < 0)
1592 			goto out_err;
1593 	} else {
1594 		put_nfs_open_context(ctx);
1595 	}
1596 	return 0;
1597 out_put_ctx:
1598 	put_nfs_open_context(ctx);
1599 out_err_drop:
1600 	d_drop(dentry);
1601 out_err:
1602 	return error;
1603 }
1604 
1605 #endif /* CONFIG_NFSV4 */
1606 
1607 /*
1608  * Code common to create, mkdir, and mknod.
1609  */
1610 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1611 				struct nfs_fattr *fattr)
1612 {
1613 	struct dentry *parent = dget_parent(dentry);
1614 	struct inode *dir = parent->d_inode;
1615 	struct inode *inode;
1616 	int error = -EACCES;
1617 
1618 	d_drop(dentry);
1619 
1620 	/* We may have been initialized further down */
1621 	if (dentry->d_inode)
1622 		goto out;
1623 	if (fhandle->size == 0) {
1624 		error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1625 		if (error)
1626 			goto out_error;
1627 	}
1628 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1629 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1630 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1631 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1632 		if (error < 0)
1633 			goto out_error;
1634 	}
1635 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1636 	error = PTR_ERR(inode);
1637 	if (IS_ERR(inode))
1638 		goto out_error;
1639 	d_add(dentry, inode);
1640 out:
1641 	dput(parent);
1642 	return 0;
1643 out_error:
1644 	nfs_mark_for_revalidate(dir);
1645 	dput(parent);
1646 	return error;
1647 }
1648 
1649 /*
1650  * Following a failed create operation, we drop the dentry rather
1651  * than retain a negative dentry. This avoids a problem in the event
1652  * that the operation succeeded on the server, but an error in the
1653  * reply path made it appear to have failed.
1654  */
1655 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1656 		struct nameidata *nd)
1657 {
1658 	struct iattr attr;
1659 	int error;
1660 	int open_flags = O_CREAT|O_EXCL|FMODE_READ;
1661 
1662 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1663 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1664 
1665 	attr.ia_mode = mode;
1666 	attr.ia_valid = ATTR_MODE;
1667 
1668 	if (nd && (nd->flags & LOOKUP_OPEN) != 0)
1669 		open_flags = nd->intent.open.flags;
1670 
1671 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1672 	if (error != 0)
1673 		goto out_err;
1674 	return 0;
1675 out_err:
1676 	d_drop(dentry);
1677 	return error;
1678 }
1679 
1680 /*
1681  * See comments for nfs_proc_create regarding failed operations.
1682  */
1683 static int
1684 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1685 {
1686 	struct iattr attr;
1687 	int status;
1688 
1689 	dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1690 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1691 
1692 	if (!new_valid_dev(rdev))
1693 		return -EINVAL;
1694 
1695 	attr.ia_mode = mode;
1696 	attr.ia_valid = ATTR_MODE;
1697 
1698 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1699 	if (status != 0)
1700 		goto out_err;
1701 	return 0;
1702 out_err:
1703 	d_drop(dentry);
1704 	return status;
1705 }
1706 
1707 /*
1708  * See comments for nfs_proc_create regarding failed operations.
1709  */
1710 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1711 {
1712 	struct iattr attr;
1713 	int error;
1714 
1715 	dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1716 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1717 
1718 	attr.ia_valid = ATTR_MODE;
1719 	attr.ia_mode = mode | S_IFDIR;
1720 
1721 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1722 	if (error != 0)
1723 		goto out_err;
1724 	return 0;
1725 out_err:
1726 	d_drop(dentry);
1727 	return error;
1728 }
1729 
1730 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1731 {
1732 	if (dentry->d_inode != NULL && !d_unhashed(dentry))
1733 		d_delete(dentry);
1734 }
1735 
1736 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1737 {
1738 	int error;
1739 
1740 	dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1741 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1742 
1743 	error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1744 	/* Ensure the VFS deletes this inode */
1745 	if (error == 0 && dentry->d_inode != NULL)
1746 		clear_nlink(dentry->d_inode);
1747 	else if (error == -ENOENT)
1748 		nfs_dentry_handle_enoent(dentry);
1749 
1750 	return error;
1751 }
1752 
1753 /*
1754  * Remove a file after making sure there are no pending writes,
1755  * and after checking that the file has only one user.
1756  *
1757  * We invalidate the attribute cache and free the inode prior to the operation
1758  * to avoid possible races if the server reuses the inode.
1759  */
1760 static int nfs_safe_remove(struct dentry *dentry)
1761 {
1762 	struct inode *dir = dentry->d_parent->d_inode;
1763 	struct inode *inode = dentry->d_inode;
1764 	int error = -EBUSY;
1765 
1766 	dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1767 		dentry->d_parent->d_name.name, dentry->d_name.name);
1768 
1769 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1770 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1771 		error = 0;
1772 		goto out;
1773 	}
1774 
1775 	if (inode != NULL) {
1776 		nfs_inode_return_delegation(inode);
1777 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1778 		/* The VFS may want to delete this inode */
1779 		if (error == 0)
1780 			nfs_drop_nlink(inode);
1781 		nfs_mark_for_revalidate(inode);
1782 	} else
1783 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1784 	if (error == -ENOENT)
1785 		nfs_dentry_handle_enoent(dentry);
1786 out:
1787 	return error;
1788 }
1789 
1790 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1791  *  belongs to an active ".nfs..." file and we return -EBUSY.
1792  *
1793  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1794  */
1795 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1796 {
1797 	int error;
1798 	int need_rehash = 0;
1799 
1800 	dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1801 		dir->i_ino, dentry->d_name.name);
1802 
1803 	spin_lock(&dentry->d_lock);
1804 	if (dentry->d_count > 1) {
1805 		spin_unlock(&dentry->d_lock);
1806 		/* Start asynchronous writeout of the inode */
1807 		write_inode_now(dentry->d_inode, 0);
1808 		error = nfs_sillyrename(dir, dentry);
1809 		return error;
1810 	}
1811 	if (!d_unhashed(dentry)) {
1812 		__d_drop(dentry);
1813 		need_rehash = 1;
1814 	}
1815 	spin_unlock(&dentry->d_lock);
1816 	error = nfs_safe_remove(dentry);
1817 	if (!error || error == -ENOENT) {
1818 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1819 	} else if (need_rehash)
1820 		d_rehash(dentry);
1821 	return error;
1822 }
1823 
1824 /*
1825  * To create a symbolic link, most file systems instantiate a new inode,
1826  * add a page to it containing the path, then write it out to the disk
1827  * using prepare_write/commit_write.
1828  *
1829  * Unfortunately the NFS client can't create the in-core inode first
1830  * because it needs a file handle to create an in-core inode (see
1831  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1832  * symlink request has completed on the server.
1833  *
1834  * So instead we allocate a raw page, copy the symname into it, then do
1835  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1836  * now have a new file handle and can instantiate an in-core NFS inode
1837  * and move the raw page into its mapping.
1838  */
1839 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1840 {
1841 	struct pagevec lru_pvec;
1842 	struct page *page;
1843 	char *kaddr;
1844 	struct iattr attr;
1845 	unsigned int pathlen = strlen(symname);
1846 	int error;
1847 
1848 	dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1849 		dir->i_ino, dentry->d_name.name, symname);
1850 
1851 	if (pathlen > PAGE_SIZE)
1852 		return -ENAMETOOLONG;
1853 
1854 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1855 	attr.ia_valid = ATTR_MODE;
1856 
1857 	page = alloc_page(GFP_HIGHUSER);
1858 	if (!page)
1859 		return -ENOMEM;
1860 
1861 	kaddr = kmap_atomic(page, KM_USER0);
1862 	memcpy(kaddr, symname, pathlen);
1863 	if (pathlen < PAGE_SIZE)
1864 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1865 	kunmap_atomic(kaddr, KM_USER0);
1866 
1867 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1868 	if (error != 0) {
1869 		dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1870 			dir->i_sb->s_id, dir->i_ino,
1871 			dentry->d_name.name, symname, error);
1872 		d_drop(dentry);
1873 		__free_page(page);
1874 		return error;
1875 	}
1876 
1877 	/*
1878 	 * No big deal if we can't add this page to the page cache here.
1879 	 * READLINK will get the missing page from the server if needed.
1880 	 */
1881 	pagevec_init(&lru_pvec, 0);
1882 	if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1883 							GFP_KERNEL)) {
1884 		pagevec_add(&lru_pvec, page);
1885 		pagevec_lru_add_file(&lru_pvec);
1886 		SetPageUptodate(page);
1887 		unlock_page(page);
1888 	} else
1889 		__free_page(page);
1890 
1891 	return 0;
1892 }
1893 
1894 static int
1895 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1896 {
1897 	struct inode *inode = old_dentry->d_inode;
1898 	int error;
1899 
1900 	dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1901 		old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1902 		dentry->d_parent->d_name.name, dentry->d_name.name);
1903 
1904 	nfs_inode_return_delegation(inode);
1905 
1906 	d_drop(dentry);
1907 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1908 	if (error == 0) {
1909 		ihold(inode);
1910 		d_add(dentry, inode);
1911 	}
1912 	return error;
1913 }
1914 
1915 /*
1916  * RENAME
1917  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1918  * different file handle for the same inode after a rename (e.g. when
1919  * moving to a different directory). A fail-safe method to do so would
1920  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1921  * rename the old file using the sillyrename stuff. This way, the original
1922  * file in old_dir will go away when the last process iput()s the inode.
1923  *
1924  * FIXED.
1925  *
1926  * It actually works quite well. One needs to have the possibility for
1927  * at least one ".nfs..." file in each directory the file ever gets
1928  * moved or linked to which happens automagically with the new
1929  * implementation that only depends on the dcache stuff instead of
1930  * using the inode layer
1931  *
1932  * Unfortunately, things are a little more complicated than indicated
1933  * above. For a cross-directory move, we want to make sure we can get
1934  * rid of the old inode after the operation.  This means there must be
1935  * no pending writes (if it's a file), and the use count must be 1.
1936  * If these conditions are met, we can drop the dentries before doing
1937  * the rename.
1938  */
1939 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1940 		      struct inode *new_dir, struct dentry *new_dentry)
1941 {
1942 	struct inode *old_inode = old_dentry->d_inode;
1943 	struct inode *new_inode = new_dentry->d_inode;
1944 	struct dentry *dentry = NULL, *rehash = NULL;
1945 	int error = -EBUSY;
1946 
1947 	dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1948 		 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1949 		 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1950 		 new_dentry->d_count);
1951 
1952 	/*
1953 	 * For non-directories, check whether the target is busy and if so,
1954 	 * make a copy of the dentry and then do a silly-rename. If the
1955 	 * silly-rename succeeds, the copied dentry is hashed and becomes
1956 	 * the new target.
1957 	 */
1958 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1959 		/*
1960 		 * To prevent any new references to the target during the
1961 		 * rename, we unhash the dentry in advance.
1962 		 */
1963 		if (!d_unhashed(new_dentry)) {
1964 			d_drop(new_dentry);
1965 			rehash = new_dentry;
1966 		}
1967 
1968 		if (new_dentry->d_count > 2) {
1969 			int err;
1970 
1971 			/* copy the target dentry's name */
1972 			dentry = d_alloc(new_dentry->d_parent,
1973 					 &new_dentry->d_name);
1974 			if (!dentry)
1975 				goto out;
1976 
1977 			/* silly-rename the existing target ... */
1978 			err = nfs_sillyrename(new_dir, new_dentry);
1979 			if (err)
1980 				goto out;
1981 
1982 			new_dentry = dentry;
1983 			rehash = NULL;
1984 			new_inode = NULL;
1985 		}
1986 	}
1987 
1988 	nfs_inode_return_delegation(old_inode);
1989 	if (new_inode != NULL)
1990 		nfs_inode_return_delegation(new_inode);
1991 
1992 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1993 					   new_dir, &new_dentry->d_name);
1994 	nfs_mark_for_revalidate(old_inode);
1995 out:
1996 	if (rehash)
1997 		d_rehash(rehash);
1998 	if (!error) {
1999 		if (new_inode != NULL)
2000 			nfs_drop_nlink(new_inode);
2001 		d_move(old_dentry, new_dentry);
2002 		nfs_set_verifier(new_dentry,
2003 					nfs_save_change_attribute(new_dir));
2004 	} else if (error == -ENOENT)
2005 		nfs_dentry_handle_enoent(old_dentry);
2006 
2007 	/* new dentry created? */
2008 	if (dentry)
2009 		dput(dentry);
2010 	return error;
2011 }
2012 
2013 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2014 static LIST_HEAD(nfs_access_lru_list);
2015 static atomic_long_t nfs_access_nr_entries;
2016 
2017 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2018 {
2019 	put_rpccred(entry->cred);
2020 	kfree(entry);
2021 	smp_mb__before_atomic_dec();
2022 	atomic_long_dec(&nfs_access_nr_entries);
2023 	smp_mb__after_atomic_dec();
2024 }
2025 
2026 static void nfs_access_free_list(struct list_head *head)
2027 {
2028 	struct nfs_access_entry *cache;
2029 
2030 	while (!list_empty(head)) {
2031 		cache = list_entry(head->next, struct nfs_access_entry, lru);
2032 		list_del(&cache->lru);
2033 		nfs_access_free_entry(cache);
2034 	}
2035 }
2036 
2037 int nfs_access_cache_shrinker(struct shrinker *shrink,
2038 			      struct shrink_control *sc)
2039 {
2040 	LIST_HEAD(head);
2041 	struct nfs_inode *nfsi, *next;
2042 	struct nfs_access_entry *cache;
2043 	int nr_to_scan = sc->nr_to_scan;
2044 	gfp_t gfp_mask = sc->gfp_mask;
2045 
2046 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2047 		return (nr_to_scan == 0) ? 0 : -1;
2048 
2049 	spin_lock(&nfs_access_lru_lock);
2050 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2051 		struct inode *inode;
2052 
2053 		if (nr_to_scan-- == 0)
2054 			break;
2055 		inode = &nfsi->vfs_inode;
2056 		spin_lock(&inode->i_lock);
2057 		if (list_empty(&nfsi->access_cache_entry_lru))
2058 			goto remove_lru_entry;
2059 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2060 				struct nfs_access_entry, lru);
2061 		list_move(&cache->lru, &head);
2062 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2063 		if (!list_empty(&nfsi->access_cache_entry_lru))
2064 			list_move_tail(&nfsi->access_cache_inode_lru,
2065 					&nfs_access_lru_list);
2066 		else {
2067 remove_lru_entry:
2068 			list_del_init(&nfsi->access_cache_inode_lru);
2069 			smp_mb__before_clear_bit();
2070 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2071 			smp_mb__after_clear_bit();
2072 		}
2073 		spin_unlock(&inode->i_lock);
2074 	}
2075 	spin_unlock(&nfs_access_lru_lock);
2076 	nfs_access_free_list(&head);
2077 	return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2078 }
2079 
2080 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2081 {
2082 	struct rb_root *root_node = &nfsi->access_cache;
2083 	struct rb_node *n;
2084 	struct nfs_access_entry *entry;
2085 
2086 	/* Unhook entries from the cache */
2087 	while ((n = rb_first(root_node)) != NULL) {
2088 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2089 		rb_erase(n, root_node);
2090 		list_move(&entry->lru, head);
2091 	}
2092 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2093 }
2094 
2095 void nfs_access_zap_cache(struct inode *inode)
2096 {
2097 	LIST_HEAD(head);
2098 
2099 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2100 		return;
2101 	/* Remove from global LRU init */
2102 	spin_lock(&nfs_access_lru_lock);
2103 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2104 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2105 
2106 	spin_lock(&inode->i_lock);
2107 	__nfs_access_zap_cache(NFS_I(inode), &head);
2108 	spin_unlock(&inode->i_lock);
2109 	spin_unlock(&nfs_access_lru_lock);
2110 	nfs_access_free_list(&head);
2111 }
2112 
2113 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2114 {
2115 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2116 	struct nfs_access_entry *entry;
2117 
2118 	while (n != NULL) {
2119 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2120 
2121 		if (cred < entry->cred)
2122 			n = n->rb_left;
2123 		else if (cred > entry->cred)
2124 			n = n->rb_right;
2125 		else
2126 			return entry;
2127 	}
2128 	return NULL;
2129 }
2130 
2131 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2132 {
2133 	struct nfs_inode *nfsi = NFS_I(inode);
2134 	struct nfs_access_entry *cache;
2135 	int err = -ENOENT;
2136 
2137 	spin_lock(&inode->i_lock);
2138 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2139 		goto out_zap;
2140 	cache = nfs_access_search_rbtree(inode, cred);
2141 	if (cache == NULL)
2142 		goto out;
2143 	if (!nfs_have_delegated_attributes(inode) &&
2144 	    !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2145 		goto out_stale;
2146 	res->jiffies = cache->jiffies;
2147 	res->cred = cache->cred;
2148 	res->mask = cache->mask;
2149 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2150 	err = 0;
2151 out:
2152 	spin_unlock(&inode->i_lock);
2153 	return err;
2154 out_stale:
2155 	rb_erase(&cache->rb_node, &nfsi->access_cache);
2156 	list_del(&cache->lru);
2157 	spin_unlock(&inode->i_lock);
2158 	nfs_access_free_entry(cache);
2159 	return -ENOENT;
2160 out_zap:
2161 	spin_unlock(&inode->i_lock);
2162 	nfs_access_zap_cache(inode);
2163 	return -ENOENT;
2164 }
2165 
2166 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2167 {
2168 	struct nfs_inode *nfsi = NFS_I(inode);
2169 	struct rb_root *root_node = &nfsi->access_cache;
2170 	struct rb_node **p = &root_node->rb_node;
2171 	struct rb_node *parent = NULL;
2172 	struct nfs_access_entry *entry;
2173 
2174 	spin_lock(&inode->i_lock);
2175 	while (*p != NULL) {
2176 		parent = *p;
2177 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2178 
2179 		if (set->cred < entry->cred)
2180 			p = &parent->rb_left;
2181 		else if (set->cred > entry->cred)
2182 			p = &parent->rb_right;
2183 		else
2184 			goto found;
2185 	}
2186 	rb_link_node(&set->rb_node, parent, p);
2187 	rb_insert_color(&set->rb_node, root_node);
2188 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2189 	spin_unlock(&inode->i_lock);
2190 	return;
2191 found:
2192 	rb_replace_node(parent, &set->rb_node, root_node);
2193 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2194 	list_del(&entry->lru);
2195 	spin_unlock(&inode->i_lock);
2196 	nfs_access_free_entry(entry);
2197 }
2198 
2199 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2200 {
2201 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2202 	if (cache == NULL)
2203 		return;
2204 	RB_CLEAR_NODE(&cache->rb_node);
2205 	cache->jiffies = set->jiffies;
2206 	cache->cred = get_rpccred(set->cred);
2207 	cache->mask = set->mask;
2208 
2209 	nfs_access_add_rbtree(inode, cache);
2210 
2211 	/* Update accounting */
2212 	smp_mb__before_atomic_inc();
2213 	atomic_long_inc(&nfs_access_nr_entries);
2214 	smp_mb__after_atomic_inc();
2215 
2216 	/* Add inode to global LRU list */
2217 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2218 		spin_lock(&nfs_access_lru_lock);
2219 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2220 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2221 					&nfs_access_lru_list);
2222 		spin_unlock(&nfs_access_lru_lock);
2223 	}
2224 }
2225 
2226 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2227 {
2228 	struct nfs_access_entry cache;
2229 	int status;
2230 
2231 	status = nfs_access_get_cached(inode, cred, &cache);
2232 	if (status == 0)
2233 		goto out;
2234 
2235 	/* Be clever: ask server to check for all possible rights */
2236 	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2237 	cache.cred = cred;
2238 	cache.jiffies = jiffies;
2239 	status = NFS_PROTO(inode)->access(inode, &cache);
2240 	if (status != 0) {
2241 		if (status == -ESTALE) {
2242 			nfs_zap_caches(inode);
2243 			if (!S_ISDIR(inode->i_mode))
2244 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2245 		}
2246 		return status;
2247 	}
2248 	nfs_access_add_cache(inode, &cache);
2249 out:
2250 	if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2251 		return 0;
2252 	return -EACCES;
2253 }
2254 
2255 static int nfs_open_permission_mask(int openflags)
2256 {
2257 	int mask = 0;
2258 
2259 	if (openflags & FMODE_READ)
2260 		mask |= MAY_READ;
2261 	if (openflags & FMODE_WRITE)
2262 		mask |= MAY_WRITE;
2263 	if (openflags & FMODE_EXEC)
2264 		mask |= MAY_EXEC;
2265 	return mask;
2266 }
2267 
2268 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2269 {
2270 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2271 }
2272 
2273 int nfs_permission(struct inode *inode, int mask)
2274 {
2275 	struct rpc_cred *cred;
2276 	int res = 0;
2277 
2278 	if (mask & MAY_NOT_BLOCK)
2279 		return -ECHILD;
2280 
2281 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2282 
2283 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2284 		goto out;
2285 	/* Is this sys_access() ? */
2286 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2287 		goto force_lookup;
2288 
2289 	switch (inode->i_mode & S_IFMT) {
2290 		case S_IFLNK:
2291 			goto out;
2292 		case S_IFREG:
2293 			/* NFSv4 has atomic_open... */
2294 			if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2295 					&& (mask & MAY_OPEN)
2296 					&& !(mask & MAY_EXEC))
2297 				goto out;
2298 			break;
2299 		case S_IFDIR:
2300 			/*
2301 			 * Optimize away all write operations, since the server
2302 			 * will check permissions when we perform the op.
2303 			 */
2304 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2305 				goto out;
2306 	}
2307 
2308 force_lookup:
2309 	if (!NFS_PROTO(inode)->access)
2310 		goto out_notsup;
2311 
2312 	cred = rpc_lookup_cred();
2313 	if (!IS_ERR(cred)) {
2314 		res = nfs_do_access(inode, cred, mask);
2315 		put_rpccred(cred);
2316 	} else
2317 		res = PTR_ERR(cred);
2318 out:
2319 	if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2320 		res = -EACCES;
2321 
2322 	dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2323 		inode->i_sb->s_id, inode->i_ino, mask, res);
2324 	return res;
2325 out_notsup:
2326 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2327 	if (res == 0)
2328 		res = generic_permission(inode, mask);
2329 	goto out;
2330 }
2331 
2332 /*
2333  * Local variables:
2334  *  version-control: t
2335  *  kept-new-versions: 5
2336  * End:
2337  */
2338