xref: /openbmc/linux/fs/nfs/dir.c (revision ff56535d)
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 		.pages = xdr_pages,
517 		.page_len = buflen,
518 		.buflen = buflen,
519 		.len = buflen,
520 	};
521 	struct page *scratch;
522 	struct nfs_cache_array *array;
523 	unsigned int count = 0;
524 	int status;
525 
526 	scratch = alloc_page(GFP_KERNEL);
527 	if (scratch == NULL)
528 		return -ENOMEM;
529 
530 	xdr_init_decode(&stream, &buf, NULL);
531 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
532 
533 	do {
534 		status = xdr_decode(desc, entry, &stream);
535 		if (status != 0) {
536 			if (status == -EAGAIN)
537 				status = 0;
538 			break;
539 		}
540 
541 		count++;
542 
543 		if (desc->plus != 0)
544 			nfs_prime_dcache(desc->file->f_path.dentry, entry);
545 
546 		status = nfs_readdir_add_to_array(entry, page);
547 		if (status != 0)
548 			break;
549 	} while (!entry->eof);
550 
551 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
552 		array = nfs_readdir_get_array(page);
553 		if (!IS_ERR(array)) {
554 			array->eof_index = array->size;
555 			status = 0;
556 			nfs_readdir_release_array(page);
557 		} else
558 			status = PTR_ERR(array);
559 	}
560 
561 	put_page(scratch);
562 	return status;
563 }
564 
565 static
566 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
567 {
568 	unsigned int i;
569 	for (i = 0; i < npages; i++)
570 		put_page(pages[i]);
571 }
572 
573 static
574 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
575 		unsigned int npages)
576 {
577 	nfs_readdir_free_pagearray(pages, npages);
578 }
579 
580 /*
581  * nfs_readdir_large_page will allocate pages that must be freed with a call
582  * to nfs_readdir_free_large_page
583  */
584 static
585 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
586 {
587 	unsigned int i;
588 
589 	for (i = 0; i < npages; i++) {
590 		struct page *page = alloc_page(GFP_KERNEL);
591 		if (page == NULL)
592 			goto out_freepages;
593 		pages[i] = page;
594 	}
595 	return 0;
596 
597 out_freepages:
598 	nfs_readdir_free_pagearray(pages, i);
599 	return -ENOMEM;
600 }
601 
602 static
603 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
604 {
605 	struct page *pages[NFS_MAX_READDIR_PAGES];
606 	void *pages_ptr = NULL;
607 	struct nfs_entry entry;
608 	struct file	*file = desc->file;
609 	struct nfs_cache_array *array;
610 	int status = -ENOMEM;
611 	unsigned int array_size = ARRAY_SIZE(pages);
612 
613 	entry.prev_cookie = 0;
614 	entry.cookie = desc->last_cookie;
615 	entry.eof = 0;
616 	entry.fh = nfs_alloc_fhandle();
617 	entry.fattr = nfs_alloc_fattr();
618 	entry.server = NFS_SERVER(inode);
619 	if (entry.fh == NULL || entry.fattr == NULL)
620 		goto out;
621 
622 	array = nfs_readdir_get_array(page);
623 	if (IS_ERR(array)) {
624 		status = PTR_ERR(array);
625 		goto out;
626 	}
627 	memset(array, 0, sizeof(struct nfs_cache_array));
628 	array->eof_index = -1;
629 
630 	status = nfs_readdir_large_page(pages, array_size);
631 	if (status < 0)
632 		goto out_release_array;
633 	do {
634 		unsigned int pglen;
635 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
636 
637 		if (status < 0)
638 			break;
639 		pglen = status;
640 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
641 		if (status < 0) {
642 			if (status == -ENOSPC)
643 				status = 0;
644 			break;
645 		}
646 	} while (array->eof_index < 0);
647 
648 	nfs_readdir_free_large_page(pages_ptr, pages, array_size);
649 out_release_array:
650 	nfs_readdir_release_array(page);
651 out:
652 	nfs_free_fattr(entry.fattr);
653 	nfs_free_fhandle(entry.fh);
654 	return status;
655 }
656 
657 /*
658  * Now we cache directories properly, by converting xdr information
659  * to an array that can be used for lookups later.  This results in
660  * fewer cache pages, since we can store more information on each page.
661  * We only need to convert from xdr once so future lookups are much simpler
662  */
663 static
664 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
665 {
666 	struct inode	*inode = desc->file->f_path.dentry->d_inode;
667 	int ret;
668 
669 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
670 	if (ret < 0)
671 		goto error;
672 	SetPageUptodate(page);
673 
674 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
675 		/* Should never happen */
676 		nfs_zap_mapping(inode, inode->i_mapping);
677 	}
678 	unlock_page(page);
679 	return 0;
680  error:
681 	unlock_page(page);
682 	return ret;
683 }
684 
685 static
686 void cache_page_release(nfs_readdir_descriptor_t *desc)
687 {
688 	if (!desc->page->mapping)
689 		nfs_readdir_clear_array(desc->page);
690 	page_cache_release(desc->page);
691 	desc->page = NULL;
692 }
693 
694 static
695 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
696 {
697 	return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
698 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
699 }
700 
701 /*
702  * Returns 0 if desc->dir_cookie was found on page desc->page_index
703  */
704 static
705 int find_cache_page(nfs_readdir_descriptor_t *desc)
706 {
707 	int res;
708 
709 	desc->page = get_cache_page(desc);
710 	if (IS_ERR(desc->page))
711 		return PTR_ERR(desc->page);
712 
713 	res = nfs_readdir_search_array(desc);
714 	if (res != 0)
715 		cache_page_release(desc);
716 	return res;
717 }
718 
719 /* Search for desc->dir_cookie from the beginning of the page cache */
720 static inline
721 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
722 {
723 	int res;
724 
725 	if (desc->page_index == 0) {
726 		desc->current_index = 0;
727 		desc->last_cookie = 0;
728 	}
729 	do {
730 		res = find_cache_page(desc);
731 	} while (res == -EAGAIN);
732 	return res;
733 }
734 
735 /*
736  * Once we've found the start of the dirent within a page: fill 'er up...
737  */
738 static
739 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
740 		   filldir_t filldir)
741 {
742 	struct file	*file = desc->file;
743 	int i = 0;
744 	int res = 0;
745 	struct nfs_cache_array *array = NULL;
746 	struct nfs_open_dir_context *ctx = file->private_data;
747 
748 	if (ctx->duped != 0 && ctx->dup_cookie == *desc->dir_cookie) {
749 		if (printk_ratelimit()) {
750 			pr_notice("NFS: directory %s/%s contains a readdir loop.  "
751 				"Please contact your server vendor.  "
752 				"Offending cookie: %llu\n",
753 				file->f_dentry->d_parent->d_name.name,
754 				file->f_dentry->d_name.name,
755 				*desc->dir_cookie);
756 		}
757 		res = -ELOOP;
758 		goto out;
759 	}
760 
761 	array = nfs_readdir_get_array(desc->page);
762 	if (IS_ERR(array)) {
763 		res = PTR_ERR(array);
764 		goto out;
765 	}
766 
767 	for (i = desc->cache_entry_index; i < array->size; i++) {
768 		struct nfs_cache_array_entry *ent;
769 
770 		ent = &array->array[i];
771 		if (filldir(dirent, ent->string.name, ent->string.len,
772 		    file->f_pos, nfs_compat_user_ino64(ent->ino),
773 		    ent->d_type) < 0) {
774 			desc->eof = 1;
775 			break;
776 		}
777 		file->f_pos++;
778 		if (i < (array->size-1))
779 			*desc->dir_cookie = array->array[i+1].cookie;
780 		else
781 			*desc->dir_cookie = array->last_cookie;
782 	}
783 	if (array->eof_index >= 0)
784 		desc->eof = 1;
785 
786 	nfs_readdir_release_array(desc->page);
787 out:
788 	cache_page_release(desc);
789 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
790 			(unsigned long long)*desc->dir_cookie, res);
791 	return res;
792 }
793 
794 /*
795  * If we cannot find a cookie in our cache, we suspect that this is
796  * because it points to a deleted file, so we ask the server to return
797  * whatever it thinks is the next entry. We then feed this to filldir.
798  * If all goes well, we should then be able to find our way round the
799  * cache on the next call to readdir_search_pagecache();
800  *
801  * NOTE: we cannot add the anonymous page to the pagecache because
802  *	 the data it contains might not be page aligned. Besides,
803  *	 we should already have a complete representation of the
804  *	 directory in the page cache by the time we get here.
805  */
806 static inline
807 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
808 		     filldir_t filldir)
809 {
810 	struct page	*page = NULL;
811 	int		status;
812 	struct inode *inode = desc->file->f_path.dentry->d_inode;
813 
814 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
815 			(unsigned long long)*desc->dir_cookie);
816 
817 	page = alloc_page(GFP_HIGHUSER);
818 	if (!page) {
819 		status = -ENOMEM;
820 		goto out;
821 	}
822 
823 	desc->page_index = 0;
824 	desc->last_cookie = *desc->dir_cookie;
825 	desc->page = page;
826 
827 	status = nfs_readdir_xdr_to_array(desc, page, inode);
828 	if (status < 0)
829 		goto out_release;
830 
831 	status = nfs_do_filldir(desc, dirent, filldir);
832 
833  out:
834 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
835 			__func__, status);
836 	return status;
837  out_release:
838 	cache_page_release(desc);
839 	goto out;
840 }
841 
842 /* The file offset position represents the dirent entry number.  A
843    last cookie cache takes care of the common case of reading the
844    whole directory.
845  */
846 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
847 {
848 	struct dentry	*dentry = filp->f_path.dentry;
849 	struct inode	*inode = dentry->d_inode;
850 	nfs_readdir_descriptor_t my_desc,
851 			*desc = &my_desc;
852 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
853 	int res;
854 
855 	dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
856 			dentry->d_parent->d_name.name, dentry->d_name.name,
857 			(long long)filp->f_pos);
858 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
859 
860 	/*
861 	 * filp->f_pos points to the dirent entry number.
862 	 * *desc->dir_cookie has the cookie for the next entry. We have
863 	 * to either find the entry with the appropriate number or
864 	 * revalidate the cookie.
865 	 */
866 	memset(desc, 0, sizeof(*desc));
867 
868 	desc->file = filp;
869 	desc->dir_cookie = &dir_ctx->dir_cookie;
870 	desc->decode = NFS_PROTO(inode)->decode_dirent;
871 	desc->plus = NFS_USE_READDIRPLUS(inode);
872 
873 	nfs_block_sillyrename(dentry);
874 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
875 	if (res < 0)
876 		goto out;
877 
878 	do {
879 		res = readdir_search_pagecache(desc);
880 
881 		if (res == -EBADCOOKIE) {
882 			res = 0;
883 			/* This means either end of directory */
884 			if (*desc->dir_cookie && desc->eof == 0) {
885 				/* Or that the server has 'lost' a cookie */
886 				res = uncached_readdir(desc, dirent, filldir);
887 				if (res == 0)
888 					continue;
889 			}
890 			break;
891 		}
892 		if (res == -ETOOSMALL && desc->plus) {
893 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
894 			nfs_zap_caches(inode);
895 			desc->page_index = 0;
896 			desc->plus = 0;
897 			desc->eof = 0;
898 			continue;
899 		}
900 		if (res < 0)
901 			break;
902 
903 		res = nfs_do_filldir(desc, dirent, filldir);
904 		if (res < 0)
905 			break;
906 	} while (!desc->eof);
907 out:
908 	nfs_unblock_sillyrename(dentry);
909 	if (res > 0)
910 		res = 0;
911 	dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
912 			dentry->d_parent->d_name.name, dentry->d_name.name,
913 			res);
914 	return res;
915 }
916 
917 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
918 {
919 	struct dentry *dentry = filp->f_path.dentry;
920 	struct inode *inode = dentry->d_inode;
921 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
922 
923 	dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
924 			dentry->d_parent->d_name.name,
925 			dentry->d_name.name,
926 			offset, origin);
927 
928 	mutex_lock(&inode->i_mutex);
929 	switch (origin) {
930 		case 1:
931 			offset += filp->f_pos;
932 		case 0:
933 			if (offset >= 0)
934 				break;
935 		default:
936 			offset = -EINVAL;
937 			goto out;
938 	}
939 	if (offset != filp->f_pos) {
940 		filp->f_pos = offset;
941 		dir_ctx->dir_cookie = 0;
942 		dir_ctx->duped = 0;
943 	}
944 out:
945 	mutex_unlock(&inode->i_mutex);
946 	return offset;
947 }
948 
949 /*
950  * All directory operations under NFS are synchronous, so fsync()
951  * is a dummy operation.
952  */
953 static int nfs_fsync_dir(struct file *filp, int datasync)
954 {
955 	struct dentry *dentry = filp->f_path.dentry;
956 
957 	dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
958 			dentry->d_parent->d_name.name, dentry->d_name.name,
959 			datasync);
960 
961 	nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
962 	return 0;
963 }
964 
965 /**
966  * nfs_force_lookup_revalidate - Mark the directory as having changed
967  * @dir - pointer to directory inode
968  *
969  * This forces the revalidation code in nfs_lookup_revalidate() to do a
970  * full lookup on all child dentries of 'dir' whenever a change occurs
971  * on the server that might have invalidated our dcache.
972  *
973  * The caller should be holding dir->i_lock
974  */
975 void nfs_force_lookup_revalidate(struct inode *dir)
976 {
977 	NFS_I(dir)->cache_change_attribute++;
978 }
979 
980 /*
981  * A check for whether or not the parent directory has changed.
982  * In the case it has, we assume that the dentries are untrustworthy
983  * and may need to be looked up again.
984  */
985 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
986 {
987 	if (IS_ROOT(dentry))
988 		return 1;
989 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
990 		return 0;
991 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
992 		return 0;
993 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
994 	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
995 		return 0;
996 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
997 		return 0;
998 	return 1;
999 }
1000 
1001 /*
1002  * Return the intent data that applies to this particular path component
1003  *
1004  * Note that the current set of intents only apply to the very last
1005  * component of the path.
1006  * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
1007  */
1008 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1009 						unsigned int mask)
1010 {
1011 	if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
1012 		return 0;
1013 	return nd->flags & mask;
1014 }
1015 
1016 /*
1017  * Use intent information to check whether or not we're going to do
1018  * an O_EXCL create using this path component.
1019  */
1020 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1021 {
1022 	if (NFS_PROTO(dir)->version == 2)
1023 		return 0;
1024 	return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1025 }
1026 
1027 /*
1028  * Inode and filehandle revalidation for lookups.
1029  *
1030  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1031  * or if the intent information indicates that we're about to open this
1032  * particular file and the "nocto" mount flag is not set.
1033  *
1034  */
1035 static inline
1036 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1037 {
1038 	struct nfs_server *server = NFS_SERVER(inode);
1039 
1040 	if (IS_AUTOMOUNT(inode))
1041 		return 0;
1042 	if (nd != NULL) {
1043 		/* VFS wants an on-the-wire revalidation */
1044 		if (nd->flags & LOOKUP_REVAL)
1045 			goto out_force;
1046 		/* This is an open(2) */
1047 		if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1048 				!(server->flags & NFS_MOUNT_NOCTO) &&
1049 				(S_ISREG(inode->i_mode) ||
1050 				 S_ISDIR(inode->i_mode)))
1051 			goto out_force;
1052 		return 0;
1053 	}
1054 	return nfs_revalidate_inode(server, inode);
1055 out_force:
1056 	return __nfs_revalidate_inode(server, inode);
1057 }
1058 
1059 /*
1060  * We judge how long we want to trust negative
1061  * dentries by looking at the parent inode mtime.
1062  *
1063  * If parent mtime has changed, we revalidate, else we wait for a
1064  * period corresponding to the parent's attribute cache timeout value.
1065  */
1066 static inline
1067 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1068 		       struct nameidata *nd)
1069 {
1070 	/* Don't revalidate a negative dentry if we're creating a new file */
1071 	if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1072 		return 0;
1073 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1074 		return 1;
1075 	return !nfs_check_verifier(dir, dentry);
1076 }
1077 
1078 /*
1079  * This is called every time the dcache has a lookup hit,
1080  * and we should check whether we can really trust that
1081  * lookup.
1082  *
1083  * NOTE! The hit can be a negative hit too, don't assume
1084  * we have an inode!
1085  *
1086  * If the parent directory is seen to have changed, we throw out the
1087  * cached dentry and do a new lookup.
1088  */
1089 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1090 {
1091 	struct inode *dir;
1092 	struct inode *inode;
1093 	struct dentry *parent;
1094 	struct nfs_fh *fhandle = NULL;
1095 	struct nfs_fattr *fattr = NULL;
1096 	int error;
1097 
1098 	if (nd->flags & LOOKUP_RCU)
1099 		return -ECHILD;
1100 
1101 	parent = dget_parent(dentry);
1102 	dir = parent->d_inode;
1103 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1104 	inode = dentry->d_inode;
1105 
1106 	if (!inode) {
1107 		if (nfs_neg_need_reval(dir, dentry, nd))
1108 			goto out_bad;
1109 		goto out_valid;
1110 	}
1111 
1112 	if (is_bad_inode(inode)) {
1113 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1114 				__func__, dentry->d_parent->d_name.name,
1115 				dentry->d_name.name);
1116 		goto out_bad;
1117 	}
1118 
1119 	if (nfs_have_delegation(inode, FMODE_READ))
1120 		goto out_set_verifier;
1121 
1122 	/* Force a full look up iff the parent directory has changed */
1123 	if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1124 		if (nfs_lookup_verify_inode(inode, nd))
1125 			goto out_zap_parent;
1126 		goto out_valid;
1127 	}
1128 
1129 	if (NFS_STALE(inode))
1130 		goto out_bad;
1131 
1132 	error = -ENOMEM;
1133 	fhandle = nfs_alloc_fhandle();
1134 	fattr = nfs_alloc_fattr();
1135 	if (fhandle == NULL || fattr == NULL)
1136 		goto out_error;
1137 
1138 	error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1139 	if (error)
1140 		goto out_bad;
1141 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1142 		goto out_bad;
1143 	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1144 		goto out_bad;
1145 
1146 	nfs_free_fattr(fattr);
1147 	nfs_free_fhandle(fhandle);
1148 out_set_verifier:
1149 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1150  out_valid:
1151 	dput(parent);
1152 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1153 			__func__, dentry->d_parent->d_name.name,
1154 			dentry->d_name.name);
1155 	return 1;
1156 out_zap_parent:
1157 	nfs_zap_caches(dir);
1158  out_bad:
1159 	nfs_mark_for_revalidate(dir);
1160 	if (inode && S_ISDIR(inode->i_mode)) {
1161 		/* Purge readdir caches. */
1162 		nfs_zap_caches(inode);
1163 		/* If we have submounts, don't unhash ! */
1164 		if (have_submounts(dentry))
1165 			goto out_valid;
1166 		if (dentry->d_flags & DCACHE_DISCONNECTED)
1167 			goto out_valid;
1168 		shrink_dcache_parent(dentry);
1169 	}
1170 	d_drop(dentry);
1171 	nfs_free_fattr(fattr);
1172 	nfs_free_fhandle(fhandle);
1173 	dput(parent);
1174 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1175 			__func__, dentry->d_parent->d_name.name,
1176 			dentry->d_name.name);
1177 	return 0;
1178 out_error:
1179 	nfs_free_fattr(fattr);
1180 	nfs_free_fhandle(fhandle);
1181 	dput(parent);
1182 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1183 			__func__, dentry->d_parent->d_name.name,
1184 			dentry->d_name.name, error);
1185 	return error;
1186 }
1187 
1188 /*
1189  * This is called from dput() when d_count is going to 0.
1190  */
1191 static int nfs_dentry_delete(const struct dentry *dentry)
1192 {
1193 	dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1194 		dentry->d_parent->d_name.name, dentry->d_name.name,
1195 		dentry->d_flags);
1196 
1197 	/* Unhash any dentry with a stale inode */
1198 	if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1199 		return 1;
1200 
1201 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1202 		/* Unhash it, so that ->d_iput() would be called */
1203 		return 1;
1204 	}
1205 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1206 		/* Unhash it, so that ancestors of killed async unlink
1207 		 * files will be cleaned up during umount */
1208 		return 1;
1209 	}
1210 	return 0;
1211 
1212 }
1213 
1214 static void nfs_drop_nlink(struct inode *inode)
1215 {
1216 	spin_lock(&inode->i_lock);
1217 	if (inode->i_nlink > 0)
1218 		drop_nlink(inode);
1219 	spin_unlock(&inode->i_lock);
1220 }
1221 
1222 /*
1223  * Called when the dentry loses inode.
1224  * We use it to clean up silly-renamed files.
1225  */
1226 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1227 {
1228 	if (S_ISDIR(inode->i_mode))
1229 		/* drop any readdir cache as it could easily be old */
1230 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1231 
1232 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1233 		drop_nlink(inode);
1234 		nfs_complete_unlink(dentry, inode);
1235 	}
1236 	iput(inode);
1237 }
1238 
1239 static void nfs_d_release(struct dentry *dentry)
1240 {
1241 	/* free cached devname value, if it survived that far */
1242 	if (unlikely(dentry->d_fsdata)) {
1243 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1244 			WARN_ON(1);
1245 		else
1246 			kfree(dentry->d_fsdata);
1247 	}
1248 }
1249 
1250 const struct dentry_operations nfs_dentry_operations = {
1251 	.d_revalidate	= nfs_lookup_revalidate,
1252 	.d_delete	= nfs_dentry_delete,
1253 	.d_iput		= nfs_dentry_iput,
1254 	.d_automount	= nfs_d_automount,
1255 	.d_release	= nfs_d_release,
1256 };
1257 
1258 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1259 {
1260 	struct dentry *res;
1261 	struct dentry *parent;
1262 	struct inode *inode = NULL;
1263 	struct nfs_fh *fhandle = NULL;
1264 	struct nfs_fattr *fattr = NULL;
1265 	int error;
1266 
1267 	dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1268 		dentry->d_parent->d_name.name, dentry->d_name.name);
1269 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1270 
1271 	res = ERR_PTR(-ENAMETOOLONG);
1272 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1273 		goto out;
1274 
1275 	/*
1276 	 * If we're doing an exclusive create, optimize away the lookup
1277 	 * but don't hash the dentry.
1278 	 */
1279 	if (nfs_is_exclusive_create(dir, nd)) {
1280 		d_instantiate(dentry, NULL);
1281 		res = NULL;
1282 		goto out;
1283 	}
1284 
1285 	res = ERR_PTR(-ENOMEM);
1286 	fhandle = nfs_alloc_fhandle();
1287 	fattr = nfs_alloc_fattr();
1288 	if (fhandle == NULL || fattr == NULL)
1289 		goto out;
1290 
1291 	parent = dentry->d_parent;
1292 	/* Protect against concurrent sillydeletes */
1293 	nfs_block_sillyrename(parent);
1294 	error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1295 	if (error == -ENOENT)
1296 		goto no_entry;
1297 	if (error < 0) {
1298 		res = ERR_PTR(error);
1299 		goto out_unblock_sillyrename;
1300 	}
1301 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1302 	res = ERR_CAST(inode);
1303 	if (IS_ERR(res))
1304 		goto out_unblock_sillyrename;
1305 
1306 no_entry:
1307 	res = d_materialise_unique(dentry, inode);
1308 	if (res != NULL) {
1309 		if (IS_ERR(res))
1310 			goto out_unblock_sillyrename;
1311 		dentry = res;
1312 	}
1313 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1314 out_unblock_sillyrename:
1315 	nfs_unblock_sillyrename(parent);
1316 out:
1317 	nfs_free_fattr(fattr);
1318 	nfs_free_fhandle(fhandle);
1319 	return res;
1320 }
1321 
1322 #ifdef CONFIG_NFS_V4
1323 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1324 
1325 const struct dentry_operations nfs4_dentry_operations = {
1326 	.d_revalidate	= nfs_open_revalidate,
1327 	.d_delete	= nfs_dentry_delete,
1328 	.d_iput		= nfs_dentry_iput,
1329 	.d_automount	= nfs_d_automount,
1330 	.d_release	= nfs_d_release,
1331 };
1332 
1333 /*
1334  * Use intent information to determine whether we need to substitute
1335  * the NFSv4-style stateful OPEN for the LOOKUP call
1336  */
1337 static int is_atomic_open(struct nameidata *nd)
1338 {
1339 	if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1340 		return 0;
1341 	/* NFS does not (yet) have a stateful open for directories */
1342 	if (nd->flags & LOOKUP_DIRECTORY)
1343 		return 0;
1344 	/* Are we trying to write to a read only partition? */
1345 	if (__mnt_is_readonly(nd->path.mnt) &&
1346 	    (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1347 		return 0;
1348 	return 1;
1349 }
1350 
1351 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1352 {
1353 	struct path path = {
1354 		.mnt = nd->path.mnt,
1355 		.dentry = dentry,
1356 	};
1357 	struct nfs_open_context *ctx;
1358 	struct rpc_cred *cred;
1359 	fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1360 
1361 	cred = rpc_lookup_cred();
1362 	if (IS_ERR(cred))
1363 		return ERR_CAST(cred);
1364 	ctx = alloc_nfs_open_context(&path, cred, fmode);
1365 	put_rpccred(cred);
1366 	if (ctx == NULL)
1367 		return ERR_PTR(-ENOMEM);
1368 	return ctx;
1369 }
1370 
1371 static int do_open(struct inode *inode, struct file *filp)
1372 {
1373 	nfs_fscache_set_inode_cookie(inode, filp);
1374 	return 0;
1375 }
1376 
1377 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1378 {
1379 	struct file *filp;
1380 	int ret = 0;
1381 
1382 	/* If the open_intent is for execute, we have an extra check to make */
1383 	if (ctx->mode & FMODE_EXEC) {
1384 		ret = nfs_may_open(ctx->path.dentry->d_inode,
1385 				ctx->cred,
1386 				nd->intent.open.flags);
1387 		if (ret < 0)
1388 			goto out;
1389 	}
1390 	filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1391 	if (IS_ERR(filp))
1392 		ret = PTR_ERR(filp);
1393 	else
1394 		nfs_file_set_open_context(filp, ctx);
1395 out:
1396 	put_nfs_open_context(ctx);
1397 	return ret;
1398 }
1399 
1400 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1401 {
1402 	struct nfs_open_context *ctx;
1403 	struct iattr attr;
1404 	struct dentry *res = NULL;
1405 	struct inode *inode;
1406 	int open_flags;
1407 	int err;
1408 
1409 	dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1410 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1411 
1412 	/* Check that we are indeed trying to open this file */
1413 	if (!is_atomic_open(nd))
1414 		goto no_open;
1415 
1416 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1417 		res = ERR_PTR(-ENAMETOOLONG);
1418 		goto out;
1419 	}
1420 
1421 	/* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1422 	 * the dentry. */
1423 	if (nd->flags & LOOKUP_EXCL) {
1424 		d_instantiate(dentry, NULL);
1425 		goto out;
1426 	}
1427 
1428 	ctx = nameidata_to_nfs_open_context(dentry, nd);
1429 	res = ERR_CAST(ctx);
1430 	if (IS_ERR(ctx))
1431 		goto out;
1432 
1433 	open_flags = nd->intent.open.flags;
1434 	if (nd->flags & LOOKUP_CREATE) {
1435 		attr.ia_mode = nd->intent.open.create_mode;
1436 		attr.ia_valid = ATTR_MODE;
1437 		attr.ia_mode &= ~current_umask();
1438 	} else {
1439 		open_flags &= ~(O_EXCL | O_CREAT);
1440 		attr.ia_valid = 0;
1441 	}
1442 
1443 	/* Open the file on the server */
1444 	nfs_block_sillyrename(dentry->d_parent);
1445 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1446 	if (IS_ERR(inode)) {
1447 		nfs_unblock_sillyrename(dentry->d_parent);
1448 		put_nfs_open_context(ctx);
1449 		switch (PTR_ERR(inode)) {
1450 			/* Make a negative dentry */
1451 			case -ENOENT:
1452 				d_add(dentry, NULL);
1453 				res = NULL;
1454 				goto out;
1455 			/* This turned out not to be a regular file */
1456 			case -ENOTDIR:
1457 				goto no_open;
1458 			case -ELOOP:
1459 				if (!(nd->intent.open.flags & O_NOFOLLOW))
1460 					goto no_open;
1461 			/* case -EISDIR: */
1462 			/* case -EINVAL: */
1463 			default:
1464 				res = ERR_CAST(inode);
1465 				goto out;
1466 		}
1467 	}
1468 	res = d_add_unique(dentry, inode);
1469 	nfs_unblock_sillyrename(dentry->d_parent);
1470 	if (res != NULL) {
1471 		dput(ctx->path.dentry);
1472 		ctx->path.dentry = dget(res);
1473 		dentry = res;
1474 	}
1475 	err = nfs_intent_set_file(nd, ctx);
1476 	if (err < 0) {
1477 		if (res != NULL)
1478 			dput(res);
1479 		return ERR_PTR(err);
1480 	}
1481 out:
1482 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1483 	return res;
1484 no_open:
1485 	return nfs_lookup(dir, dentry, nd);
1486 }
1487 
1488 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1489 {
1490 	struct dentry *parent = NULL;
1491 	struct inode *inode;
1492 	struct inode *dir;
1493 	struct nfs_open_context *ctx;
1494 	int openflags, ret = 0;
1495 
1496 	if (nd->flags & LOOKUP_RCU)
1497 		return -ECHILD;
1498 
1499 	inode = dentry->d_inode;
1500 	if (!is_atomic_open(nd) || d_mountpoint(dentry))
1501 		goto no_open;
1502 
1503 	parent = dget_parent(dentry);
1504 	dir = parent->d_inode;
1505 
1506 	/* We can't create new files in nfs_open_revalidate(), so we
1507 	 * optimize away revalidation of negative dentries.
1508 	 */
1509 	if (inode == NULL) {
1510 		if (!nfs_neg_need_reval(dir, dentry, nd))
1511 			ret = 1;
1512 		goto out;
1513 	}
1514 
1515 	/* NFS only supports OPEN on regular files */
1516 	if (!S_ISREG(inode->i_mode))
1517 		goto no_open_dput;
1518 	openflags = nd->intent.open.flags;
1519 	/* We cannot do exclusive creation on a positive dentry */
1520 	if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1521 		goto no_open_dput;
1522 	/* We can't create new files, or truncate existing ones here */
1523 	openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1524 
1525 	ctx = nameidata_to_nfs_open_context(dentry, nd);
1526 	ret = PTR_ERR(ctx);
1527 	if (IS_ERR(ctx))
1528 		goto out;
1529 	/*
1530 	 * Note: we're not holding inode->i_mutex and so may be racing with
1531 	 * operations that change the directory. We therefore save the
1532 	 * change attribute *before* we do the RPC call.
1533 	 */
1534 	inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1535 	if (IS_ERR(inode)) {
1536 		ret = PTR_ERR(inode);
1537 		switch (ret) {
1538 		case -EPERM:
1539 		case -EACCES:
1540 		case -EDQUOT:
1541 		case -ENOSPC:
1542 		case -EROFS:
1543 			goto out_put_ctx;
1544 		default:
1545 			goto out_drop;
1546 		}
1547 	}
1548 	iput(inode);
1549 	if (inode != dentry->d_inode)
1550 		goto out_drop;
1551 
1552 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1553 	ret = nfs_intent_set_file(nd, ctx);
1554 	if (ret >= 0)
1555 		ret = 1;
1556 out:
1557 	dput(parent);
1558 	return ret;
1559 out_drop:
1560 	d_drop(dentry);
1561 	ret = 0;
1562 out_put_ctx:
1563 	put_nfs_open_context(ctx);
1564 	goto out;
1565 
1566 no_open_dput:
1567 	dput(parent);
1568 no_open:
1569 	return nfs_lookup_revalidate(dentry, nd);
1570 }
1571 
1572 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1573 		struct nameidata *nd)
1574 {
1575 	struct nfs_open_context *ctx = NULL;
1576 	struct iattr attr;
1577 	int error;
1578 	int open_flags = 0;
1579 
1580 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1581 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1582 
1583 	attr.ia_mode = mode;
1584 	attr.ia_valid = ATTR_MODE;
1585 
1586 	if ((nd->flags & LOOKUP_CREATE) != 0) {
1587 		open_flags = nd->intent.open.flags;
1588 
1589 		ctx = nameidata_to_nfs_open_context(dentry, nd);
1590 		error = PTR_ERR(ctx);
1591 		if (IS_ERR(ctx))
1592 			goto out_err_drop;
1593 	}
1594 
1595 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1596 	if (error != 0)
1597 		goto out_put_ctx;
1598 	if (ctx != NULL) {
1599 		error = nfs_intent_set_file(nd, ctx);
1600 		if (error < 0)
1601 			goto out_err;
1602 	}
1603 	return 0;
1604 out_put_ctx:
1605 	if (ctx != NULL)
1606 		put_nfs_open_context(ctx);
1607 out_err_drop:
1608 	d_drop(dentry);
1609 out_err:
1610 	return error;
1611 }
1612 
1613 #endif /* CONFIG_NFSV4 */
1614 
1615 /*
1616  * Code common to create, mkdir, and mknod.
1617  */
1618 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1619 				struct nfs_fattr *fattr)
1620 {
1621 	struct dentry *parent = dget_parent(dentry);
1622 	struct inode *dir = parent->d_inode;
1623 	struct inode *inode;
1624 	int error = -EACCES;
1625 
1626 	d_drop(dentry);
1627 
1628 	/* We may have been initialized further down */
1629 	if (dentry->d_inode)
1630 		goto out;
1631 	if (fhandle->size == 0) {
1632 		error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1633 		if (error)
1634 			goto out_error;
1635 	}
1636 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1637 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1638 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1639 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1640 		if (error < 0)
1641 			goto out_error;
1642 	}
1643 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1644 	error = PTR_ERR(inode);
1645 	if (IS_ERR(inode))
1646 		goto out_error;
1647 	d_add(dentry, inode);
1648 out:
1649 	dput(parent);
1650 	return 0;
1651 out_error:
1652 	nfs_mark_for_revalidate(dir);
1653 	dput(parent);
1654 	return error;
1655 }
1656 
1657 /*
1658  * Following a failed create operation, we drop the dentry rather
1659  * than retain a negative dentry. This avoids a problem in the event
1660  * that the operation succeeded on the server, but an error in the
1661  * reply path made it appear to have failed.
1662  */
1663 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1664 		struct nameidata *nd)
1665 {
1666 	struct iattr attr;
1667 	int error;
1668 	int open_flags = 0;
1669 
1670 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1671 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1672 
1673 	attr.ia_mode = mode;
1674 	attr.ia_valid = ATTR_MODE;
1675 
1676 	if ((nd->flags & LOOKUP_CREATE) != 0)
1677 		open_flags = nd->intent.open.flags;
1678 
1679 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1680 	if (error != 0)
1681 		goto out_err;
1682 	return 0;
1683 out_err:
1684 	d_drop(dentry);
1685 	return error;
1686 }
1687 
1688 /*
1689  * See comments for nfs_proc_create regarding failed operations.
1690  */
1691 static int
1692 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1693 {
1694 	struct iattr attr;
1695 	int status;
1696 
1697 	dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1698 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1699 
1700 	if (!new_valid_dev(rdev))
1701 		return -EINVAL;
1702 
1703 	attr.ia_mode = mode;
1704 	attr.ia_valid = ATTR_MODE;
1705 
1706 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1707 	if (status != 0)
1708 		goto out_err;
1709 	return 0;
1710 out_err:
1711 	d_drop(dentry);
1712 	return status;
1713 }
1714 
1715 /*
1716  * See comments for nfs_proc_create regarding failed operations.
1717  */
1718 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1719 {
1720 	struct iattr attr;
1721 	int error;
1722 
1723 	dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1724 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1725 
1726 	attr.ia_valid = ATTR_MODE;
1727 	attr.ia_mode = mode | S_IFDIR;
1728 
1729 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1730 	if (error != 0)
1731 		goto out_err;
1732 	return 0;
1733 out_err:
1734 	d_drop(dentry);
1735 	return error;
1736 }
1737 
1738 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1739 {
1740 	if (dentry->d_inode != NULL && !d_unhashed(dentry))
1741 		d_delete(dentry);
1742 }
1743 
1744 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1745 {
1746 	int error;
1747 
1748 	dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1749 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1750 
1751 	error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1752 	/* Ensure the VFS deletes this inode */
1753 	if (error == 0 && dentry->d_inode != NULL)
1754 		clear_nlink(dentry->d_inode);
1755 	else if (error == -ENOENT)
1756 		nfs_dentry_handle_enoent(dentry);
1757 
1758 	return error;
1759 }
1760 
1761 /*
1762  * Remove a file after making sure there are no pending writes,
1763  * and after checking that the file has only one user.
1764  *
1765  * We invalidate the attribute cache and free the inode prior to the operation
1766  * to avoid possible races if the server reuses the inode.
1767  */
1768 static int nfs_safe_remove(struct dentry *dentry)
1769 {
1770 	struct inode *dir = dentry->d_parent->d_inode;
1771 	struct inode *inode = dentry->d_inode;
1772 	int error = -EBUSY;
1773 
1774 	dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1775 		dentry->d_parent->d_name.name, dentry->d_name.name);
1776 
1777 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1778 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1779 		error = 0;
1780 		goto out;
1781 	}
1782 
1783 	if (inode != NULL) {
1784 		nfs_inode_return_delegation(inode);
1785 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1786 		/* The VFS may want to delete this inode */
1787 		if (error == 0)
1788 			nfs_drop_nlink(inode);
1789 		nfs_mark_for_revalidate(inode);
1790 	} else
1791 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1792 	if (error == -ENOENT)
1793 		nfs_dentry_handle_enoent(dentry);
1794 out:
1795 	return error;
1796 }
1797 
1798 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1799  *  belongs to an active ".nfs..." file and we return -EBUSY.
1800  *
1801  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1802  */
1803 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1804 {
1805 	int error;
1806 	int need_rehash = 0;
1807 
1808 	dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1809 		dir->i_ino, dentry->d_name.name);
1810 
1811 	spin_lock(&dentry->d_lock);
1812 	if (dentry->d_count > 1) {
1813 		spin_unlock(&dentry->d_lock);
1814 		/* Start asynchronous writeout of the inode */
1815 		write_inode_now(dentry->d_inode, 0);
1816 		error = nfs_sillyrename(dir, dentry);
1817 		return error;
1818 	}
1819 	if (!d_unhashed(dentry)) {
1820 		__d_drop(dentry);
1821 		need_rehash = 1;
1822 	}
1823 	spin_unlock(&dentry->d_lock);
1824 	error = nfs_safe_remove(dentry);
1825 	if (!error || error == -ENOENT) {
1826 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1827 	} else if (need_rehash)
1828 		d_rehash(dentry);
1829 	return error;
1830 }
1831 
1832 /*
1833  * To create a symbolic link, most file systems instantiate a new inode,
1834  * add a page to it containing the path, then write it out to the disk
1835  * using prepare_write/commit_write.
1836  *
1837  * Unfortunately the NFS client can't create the in-core inode first
1838  * because it needs a file handle to create an in-core inode (see
1839  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1840  * symlink request has completed on the server.
1841  *
1842  * So instead we allocate a raw page, copy the symname into it, then do
1843  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1844  * now have a new file handle and can instantiate an in-core NFS inode
1845  * and move the raw page into its mapping.
1846  */
1847 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1848 {
1849 	struct pagevec lru_pvec;
1850 	struct page *page;
1851 	char *kaddr;
1852 	struct iattr attr;
1853 	unsigned int pathlen = strlen(symname);
1854 	int error;
1855 
1856 	dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1857 		dir->i_ino, dentry->d_name.name, symname);
1858 
1859 	if (pathlen > PAGE_SIZE)
1860 		return -ENAMETOOLONG;
1861 
1862 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1863 	attr.ia_valid = ATTR_MODE;
1864 
1865 	page = alloc_page(GFP_HIGHUSER);
1866 	if (!page)
1867 		return -ENOMEM;
1868 
1869 	kaddr = kmap_atomic(page, KM_USER0);
1870 	memcpy(kaddr, symname, pathlen);
1871 	if (pathlen < PAGE_SIZE)
1872 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1873 	kunmap_atomic(kaddr, KM_USER0);
1874 
1875 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1876 	if (error != 0) {
1877 		dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1878 			dir->i_sb->s_id, dir->i_ino,
1879 			dentry->d_name.name, symname, error);
1880 		d_drop(dentry);
1881 		__free_page(page);
1882 		return error;
1883 	}
1884 
1885 	/*
1886 	 * No big deal if we can't add this page to the page cache here.
1887 	 * READLINK will get the missing page from the server if needed.
1888 	 */
1889 	pagevec_init(&lru_pvec, 0);
1890 	if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1891 							GFP_KERNEL)) {
1892 		pagevec_add(&lru_pvec, page);
1893 		pagevec_lru_add_file(&lru_pvec);
1894 		SetPageUptodate(page);
1895 		unlock_page(page);
1896 	} else
1897 		__free_page(page);
1898 
1899 	return 0;
1900 }
1901 
1902 static int
1903 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1904 {
1905 	struct inode *inode = old_dentry->d_inode;
1906 	int error;
1907 
1908 	dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1909 		old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1910 		dentry->d_parent->d_name.name, dentry->d_name.name);
1911 
1912 	nfs_inode_return_delegation(inode);
1913 
1914 	d_drop(dentry);
1915 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1916 	if (error == 0) {
1917 		ihold(inode);
1918 		d_add(dentry, inode);
1919 	}
1920 	return error;
1921 }
1922 
1923 /*
1924  * RENAME
1925  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1926  * different file handle for the same inode after a rename (e.g. when
1927  * moving to a different directory). A fail-safe method to do so would
1928  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1929  * rename the old file using the sillyrename stuff. This way, the original
1930  * file in old_dir will go away when the last process iput()s the inode.
1931  *
1932  * FIXED.
1933  *
1934  * It actually works quite well. One needs to have the possibility for
1935  * at least one ".nfs..." file in each directory the file ever gets
1936  * moved or linked to which happens automagically with the new
1937  * implementation that only depends on the dcache stuff instead of
1938  * using the inode layer
1939  *
1940  * Unfortunately, things are a little more complicated than indicated
1941  * above. For a cross-directory move, we want to make sure we can get
1942  * rid of the old inode after the operation.  This means there must be
1943  * no pending writes (if it's a file), and the use count must be 1.
1944  * If these conditions are met, we can drop the dentries before doing
1945  * the rename.
1946  */
1947 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1948 		      struct inode *new_dir, struct dentry *new_dentry)
1949 {
1950 	struct inode *old_inode = old_dentry->d_inode;
1951 	struct inode *new_inode = new_dentry->d_inode;
1952 	struct dentry *dentry = NULL, *rehash = NULL;
1953 	int error = -EBUSY;
1954 
1955 	dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1956 		 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1957 		 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1958 		 new_dentry->d_count);
1959 
1960 	/*
1961 	 * For non-directories, check whether the target is busy and if so,
1962 	 * make a copy of the dentry and then do a silly-rename. If the
1963 	 * silly-rename succeeds, the copied dentry is hashed and becomes
1964 	 * the new target.
1965 	 */
1966 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1967 		/*
1968 		 * To prevent any new references to the target during the
1969 		 * rename, we unhash the dentry in advance.
1970 		 */
1971 		if (!d_unhashed(new_dentry)) {
1972 			d_drop(new_dentry);
1973 			rehash = new_dentry;
1974 		}
1975 
1976 		if (new_dentry->d_count > 2) {
1977 			int err;
1978 
1979 			/* copy the target dentry's name */
1980 			dentry = d_alloc(new_dentry->d_parent,
1981 					 &new_dentry->d_name);
1982 			if (!dentry)
1983 				goto out;
1984 
1985 			/* silly-rename the existing target ... */
1986 			err = nfs_sillyrename(new_dir, new_dentry);
1987 			if (err)
1988 				goto out;
1989 
1990 			new_dentry = dentry;
1991 			rehash = NULL;
1992 			new_inode = NULL;
1993 		}
1994 	}
1995 
1996 	nfs_inode_return_delegation(old_inode);
1997 	if (new_inode != NULL)
1998 		nfs_inode_return_delegation(new_inode);
1999 
2000 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
2001 					   new_dir, &new_dentry->d_name);
2002 	nfs_mark_for_revalidate(old_inode);
2003 out:
2004 	if (rehash)
2005 		d_rehash(rehash);
2006 	if (!error) {
2007 		if (new_inode != NULL)
2008 			nfs_drop_nlink(new_inode);
2009 		d_move(old_dentry, new_dentry);
2010 		nfs_set_verifier(new_dentry,
2011 					nfs_save_change_attribute(new_dir));
2012 	} else if (error == -ENOENT)
2013 		nfs_dentry_handle_enoent(old_dentry);
2014 
2015 	/* new dentry created? */
2016 	if (dentry)
2017 		dput(dentry);
2018 	return error;
2019 }
2020 
2021 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2022 static LIST_HEAD(nfs_access_lru_list);
2023 static atomic_long_t nfs_access_nr_entries;
2024 
2025 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2026 {
2027 	put_rpccred(entry->cred);
2028 	kfree(entry);
2029 	smp_mb__before_atomic_dec();
2030 	atomic_long_dec(&nfs_access_nr_entries);
2031 	smp_mb__after_atomic_dec();
2032 }
2033 
2034 static void nfs_access_free_list(struct list_head *head)
2035 {
2036 	struct nfs_access_entry *cache;
2037 
2038 	while (!list_empty(head)) {
2039 		cache = list_entry(head->next, struct nfs_access_entry, lru);
2040 		list_del(&cache->lru);
2041 		nfs_access_free_entry(cache);
2042 	}
2043 }
2044 
2045 int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
2046 {
2047 	LIST_HEAD(head);
2048 	struct nfs_inode *nfsi, *next;
2049 	struct nfs_access_entry *cache;
2050 
2051 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2052 		return (nr_to_scan == 0) ? 0 : -1;
2053 
2054 	spin_lock(&nfs_access_lru_lock);
2055 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2056 		struct inode *inode;
2057 
2058 		if (nr_to_scan-- == 0)
2059 			break;
2060 		inode = &nfsi->vfs_inode;
2061 		spin_lock(&inode->i_lock);
2062 		if (list_empty(&nfsi->access_cache_entry_lru))
2063 			goto remove_lru_entry;
2064 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2065 				struct nfs_access_entry, lru);
2066 		list_move(&cache->lru, &head);
2067 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2068 		if (!list_empty(&nfsi->access_cache_entry_lru))
2069 			list_move_tail(&nfsi->access_cache_inode_lru,
2070 					&nfs_access_lru_list);
2071 		else {
2072 remove_lru_entry:
2073 			list_del_init(&nfsi->access_cache_inode_lru);
2074 			smp_mb__before_clear_bit();
2075 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2076 			smp_mb__after_clear_bit();
2077 		}
2078 		spin_unlock(&inode->i_lock);
2079 	}
2080 	spin_unlock(&nfs_access_lru_lock);
2081 	nfs_access_free_list(&head);
2082 	return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2083 }
2084 
2085 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2086 {
2087 	struct rb_root *root_node = &nfsi->access_cache;
2088 	struct rb_node *n;
2089 	struct nfs_access_entry *entry;
2090 
2091 	/* Unhook entries from the cache */
2092 	while ((n = rb_first(root_node)) != NULL) {
2093 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2094 		rb_erase(n, root_node);
2095 		list_move(&entry->lru, head);
2096 	}
2097 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2098 }
2099 
2100 void nfs_access_zap_cache(struct inode *inode)
2101 {
2102 	LIST_HEAD(head);
2103 
2104 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2105 		return;
2106 	/* Remove from global LRU init */
2107 	spin_lock(&nfs_access_lru_lock);
2108 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2109 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2110 
2111 	spin_lock(&inode->i_lock);
2112 	__nfs_access_zap_cache(NFS_I(inode), &head);
2113 	spin_unlock(&inode->i_lock);
2114 	spin_unlock(&nfs_access_lru_lock);
2115 	nfs_access_free_list(&head);
2116 }
2117 
2118 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2119 {
2120 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2121 	struct nfs_access_entry *entry;
2122 
2123 	while (n != NULL) {
2124 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2125 
2126 		if (cred < entry->cred)
2127 			n = n->rb_left;
2128 		else if (cred > entry->cred)
2129 			n = n->rb_right;
2130 		else
2131 			return entry;
2132 	}
2133 	return NULL;
2134 }
2135 
2136 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2137 {
2138 	struct nfs_inode *nfsi = NFS_I(inode);
2139 	struct nfs_access_entry *cache;
2140 	int err = -ENOENT;
2141 
2142 	spin_lock(&inode->i_lock);
2143 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2144 		goto out_zap;
2145 	cache = nfs_access_search_rbtree(inode, cred);
2146 	if (cache == NULL)
2147 		goto out;
2148 	if (!nfs_have_delegated_attributes(inode) &&
2149 	    !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2150 		goto out_stale;
2151 	res->jiffies = cache->jiffies;
2152 	res->cred = cache->cred;
2153 	res->mask = cache->mask;
2154 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2155 	err = 0;
2156 out:
2157 	spin_unlock(&inode->i_lock);
2158 	return err;
2159 out_stale:
2160 	rb_erase(&cache->rb_node, &nfsi->access_cache);
2161 	list_del(&cache->lru);
2162 	spin_unlock(&inode->i_lock);
2163 	nfs_access_free_entry(cache);
2164 	return -ENOENT;
2165 out_zap:
2166 	spin_unlock(&inode->i_lock);
2167 	nfs_access_zap_cache(inode);
2168 	return -ENOENT;
2169 }
2170 
2171 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2172 {
2173 	struct nfs_inode *nfsi = NFS_I(inode);
2174 	struct rb_root *root_node = &nfsi->access_cache;
2175 	struct rb_node **p = &root_node->rb_node;
2176 	struct rb_node *parent = NULL;
2177 	struct nfs_access_entry *entry;
2178 
2179 	spin_lock(&inode->i_lock);
2180 	while (*p != NULL) {
2181 		parent = *p;
2182 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2183 
2184 		if (set->cred < entry->cred)
2185 			p = &parent->rb_left;
2186 		else if (set->cred > entry->cred)
2187 			p = &parent->rb_right;
2188 		else
2189 			goto found;
2190 	}
2191 	rb_link_node(&set->rb_node, parent, p);
2192 	rb_insert_color(&set->rb_node, root_node);
2193 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2194 	spin_unlock(&inode->i_lock);
2195 	return;
2196 found:
2197 	rb_replace_node(parent, &set->rb_node, root_node);
2198 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2199 	list_del(&entry->lru);
2200 	spin_unlock(&inode->i_lock);
2201 	nfs_access_free_entry(entry);
2202 }
2203 
2204 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2205 {
2206 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2207 	if (cache == NULL)
2208 		return;
2209 	RB_CLEAR_NODE(&cache->rb_node);
2210 	cache->jiffies = set->jiffies;
2211 	cache->cred = get_rpccred(set->cred);
2212 	cache->mask = set->mask;
2213 
2214 	nfs_access_add_rbtree(inode, cache);
2215 
2216 	/* Update accounting */
2217 	smp_mb__before_atomic_inc();
2218 	atomic_long_inc(&nfs_access_nr_entries);
2219 	smp_mb__after_atomic_inc();
2220 
2221 	/* Add inode to global LRU list */
2222 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2223 		spin_lock(&nfs_access_lru_lock);
2224 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2225 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2226 					&nfs_access_lru_list);
2227 		spin_unlock(&nfs_access_lru_lock);
2228 	}
2229 }
2230 
2231 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2232 {
2233 	struct nfs_access_entry cache;
2234 	int status;
2235 
2236 	status = nfs_access_get_cached(inode, cred, &cache);
2237 	if (status == 0)
2238 		goto out;
2239 
2240 	/* Be clever: ask server to check for all possible rights */
2241 	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2242 	cache.cred = cred;
2243 	cache.jiffies = jiffies;
2244 	status = NFS_PROTO(inode)->access(inode, &cache);
2245 	if (status != 0) {
2246 		if (status == -ESTALE) {
2247 			nfs_zap_caches(inode);
2248 			if (!S_ISDIR(inode->i_mode))
2249 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2250 		}
2251 		return status;
2252 	}
2253 	nfs_access_add_cache(inode, &cache);
2254 out:
2255 	if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2256 		return 0;
2257 	return -EACCES;
2258 }
2259 
2260 static int nfs_open_permission_mask(int openflags)
2261 {
2262 	int mask = 0;
2263 
2264 	if (openflags & FMODE_READ)
2265 		mask |= MAY_READ;
2266 	if (openflags & FMODE_WRITE)
2267 		mask |= MAY_WRITE;
2268 	if (openflags & FMODE_EXEC)
2269 		mask |= MAY_EXEC;
2270 	return mask;
2271 }
2272 
2273 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2274 {
2275 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2276 }
2277 
2278 int nfs_permission(struct inode *inode, int mask, unsigned int flags)
2279 {
2280 	struct rpc_cred *cred;
2281 	int res = 0;
2282 
2283 	if (flags & IPERM_FLAG_RCU)
2284 		return -ECHILD;
2285 
2286 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2287 
2288 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2289 		goto out;
2290 	/* Is this sys_access() ? */
2291 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2292 		goto force_lookup;
2293 
2294 	switch (inode->i_mode & S_IFMT) {
2295 		case S_IFLNK:
2296 			goto out;
2297 		case S_IFREG:
2298 			/* NFSv4 has atomic_open... */
2299 			if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2300 					&& (mask & MAY_OPEN)
2301 					&& !(mask & MAY_EXEC))
2302 				goto out;
2303 			break;
2304 		case S_IFDIR:
2305 			/*
2306 			 * Optimize away all write operations, since the server
2307 			 * will check permissions when we perform the op.
2308 			 */
2309 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2310 				goto out;
2311 	}
2312 
2313 force_lookup:
2314 	if (!NFS_PROTO(inode)->access)
2315 		goto out_notsup;
2316 
2317 	cred = rpc_lookup_cred();
2318 	if (!IS_ERR(cred)) {
2319 		res = nfs_do_access(inode, cred, mask);
2320 		put_rpccred(cred);
2321 	} else
2322 		res = PTR_ERR(cred);
2323 out:
2324 	if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2325 		res = -EACCES;
2326 
2327 	dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2328 		inode->i_sb->s_id, inode->i_ino, mask, res);
2329 	return res;
2330 out_notsup:
2331 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2332 	if (res == 0)
2333 		res = generic_permission(inode, mask, flags, NULL);
2334 	goto out;
2335 }
2336 
2337 /*
2338  * Local variables:
2339  *  version-control: t
2340  *  kept-new-versions: 5
2341  * End:
2342  */
2343