xref: /linux/fs/nfs/dir.c (revision 5148fa52a12fa1b97c730b2fe321f2aad7ea041c)
1 /*
2  *  linux/fs/nfs/dir.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  nfs directory handling functions
7  *
8  * 10 Apr 1996	Added silly rename for unlink	--okir
9  * 28 Sep 1996	Improved directory cache --okir
10  * 23 Aug 1997  Claus Heine claus@momo.math.rwth-aachen.de
11  *              Re-implemented silly rename for unlink, newly implemented
12  *              silly rename for nfs_rename() following the suggestions
13  *              of Olaf Kirch (okir) found in this file.
14  *              Following Linus comments on my original hack, this version
15  *              depends only on the dcache stuff and doesn't touch the inode
16  *              layer (iput() and friends).
17  *  6 Jun 1999	Cache readdir lookups in the page cache. -DaveM
18  */
19 
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
27 #include <linux/mm.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/kmemleak.h>
37 #include <linux/xattr.h>
38 
39 #include "delegation.h"
40 #include "iostat.h"
41 #include "internal.h"
42 #include "fscache.h"
43 
44 /* #define NFS_DEBUG_VERBOSE 1 */
45 
46 static int nfs_opendir(struct inode *, struct file *);
47 static int nfs_closedir(struct inode *, struct file *);
48 static int nfs_readdir(struct file *, void *, filldir_t);
49 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
50 static int nfs_create(struct inode *, struct dentry *, umode_t, struct nameidata *);
51 static int nfs_mkdir(struct inode *, struct dentry *, umode_t);
52 static int nfs_rmdir(struct inode *, struct dentry *);
53 static int nfs_unlink(struct inode *, struct dentry *);
54 static int nfs_symlink(struct inode *, struct dentry *, const char *);
55 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
56 static int nfs_mknod(struct inode *, struct dentry *, umode_t, dev_t);
57 static int nfs_rename(struct inode *, struct dentry *,
58 		      struct inode *, struct dentry *);
59 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
60 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
61 static void nfs_readdir_clear_array(struct page*);
62 
63 const struct file_operations nfs_dir_operations = {
64 	.llseek		= nfs_llseek_dir,
65 	.read		= generic_read_dir,
66 	.readdir	= nfs_readdir,
67 	.open		= nfs_opendir,
68 	.release	= nfs_closedir,
69 	.fsync		= nfs_fsync_dir,
70 };
71 
72 const struct inode_operations nfs_dir_inode_operations = {
73 	.create		= nfs_create,
74 	.lookup		= nfs_lookup,
75 	.link		= nfs_link,
76 	.unlink		= nfs_unlink,
77 	.symlink	= nfs_symlink,
78 	.mkdir		= nfs_mkdir,
79 	.rmdir		= nfs_rmdir,
80 	.mknod		= nfs_mknod,
81 	.rename		= nfs_rename,
82 	.permission	= nfs_permission,
83 	.getattr	= nfs_getattr,
84 	.setattr	= nfs_setattr,
85 };
86 
87 const struct address_space_operations nfs_dir_aops = {
88 	.freepage = nfs_readdir_clear_array,
89 };
90 
91 #ifdef CONFIG_NFS_V3
92 const struct inode_operations nfs3_dir_inode_operations = {
93 	.create		= nfs_create,
94 	.lookup		= nfs_lookup,
95 	.link		= nfs_link,
96 	.unlink		= nfs_unlink,
97 	.symlink	= nfs_symlink,
98 	.mkdir		= nfs_mkdir,
99 	.rmdir		= nfs_rmdir,
100 	.mknod		= nfs_mknod,
101 	.rename		= nfs_rename,
102 	.permission	= nfs_permission,
103 	.getattr	= nfs_getattr,
104 	.setattr	= nfs_setattr,
105 	.listxattr	= nfs3_listxattr,
106 	.getxattr	= nfs3_getxattr,
107 	.setxattr	= nfs3_setxattr,
108 	.removexattr	= nfs3_removexattr,
109 };
110 #endif  /* CONFIG_NFS_V3 */
111 
112 #ifdef CONFIG_NFS_V4
113 
114 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
115 static int nfs_open_create(struct inode *dir, struct dentry *dentry, umode_t 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 inode *dir, 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->attr_gencount = NFS_I(dir)->attr_gencount;
144 		ctx->dir_cookie = 0;
145 		ctx->dup_cookie = 0;
146 		ctx->cred = get_rpccred(cred);
147 		return ctx;
148 	}
149 	return  ERR_PTR(-ENOMEM);
150 }
151 
152 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
153 {
154 	put_rpccred(ctx->cred);
155 	kfree(ctx);
156 }
157 
158 /*
159  * Open file
160  */
161 static int
162 nfs_opendir(struct inode *inode, struct file *filp)
163 {
164 	int res = 0;
165 	struct nfs_open_dir_context *ctx;
166 	struct rpc_cred *cred;
167 
168 	dfprintk(FILE, "NFS: open dir(%s/%s)\n",
169 			filp->f_path.dentry->d_parent->d_name.name,
170 			filp->f_path.dentry->d_name.name);
171 
172 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
173 
174 	cred = rpc_lookup_cred();
175 	if (IS_ERR(cred))
176 		return PTR_ERR(cred);
177 	ctx = alloc_nfs_open_dir_context(inode, cred);
178 	if (IS_ERR(ctx)) {
179 		res = PTR_ERR(ctx);
180 		goto out;
181 	}
182 	filp->private_data = ctx;
183 	if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
184 		/* This is a mountpoint, so d_revalidate will never
185 		 * have been called, so we need to refresh the
186 		 * inode (for close-open consistency) ourselves.
187 		 */
188 		__nfs_revalidate_inode(NFS_SERVER(inode), inode);
189 	}
190 out:
191 	put_rpccred(cred);
192 	return res;
193 }
194 
195 static int
196 nfs_closedir(struct inode *inode, struct file *filp)
197 {
198 	put_nfs_open_dir_context(filp->private_data);
199 	return 0;
200 }
201 
202 struct nfs_cache_array_entry {
203 	u64 cookie;
204 	u64 ino;
205 	struct qstr string;
206 	unsigned char d_type;
207 };
208 
209 struct nfs_cache_array {
210 	int size;
211 	int eof_index;
212 	u64 last_cookie;
213 	struct nfs_cache_array_entry array[0];
214 };
215 
216 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
217 typedef struct {
218 	struct file	*file;
219 	struct page	*page;
220 	unsigned long	page_index;
221 	u64		*dir_cookie;
222 	u64		last_cookie;
223 	loff_t		current_index;
224 	decode_dirent_t	decode;
225 
226 	unsigned long	timestamp;
227 	unsigned long	gencount;
228 	unsigned int	cache_entry_index;
229 	unsigned int	plus:1;
230 	unsigned int	eof:1;
231 } nfs_readdir_descriptor_t;
232 
233 /*
234  * The caller is responsible for calling nfs_readdir_release_array(page)
235  */
236 static
237 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
238 {
239 	void *ptr;
240 	if (page == NULL)
241 		return ERR_PTR(-EIO);
242 	ptr = kmap(page);
243 	if (ptr == NULL)
244 		return ERR_PTR(-ENOMEM);
245 	return ptr;
246 }
247 
248 static
249 void nfs_readdir_release_array(struct page *page)
250 {
251 	kunmap(page);
252 }
253 
254 /*
255  * we are freeing strings created by nfs_add_to_readdir_array()
256  */
257 static
258 void nfs_readdir_clear_array(struct page *page)
259 {
260 	struct nfs_cache_array *array;
261 	int i;
262 
263 	array = kmap_atomic(page);
264 	for (i = 0; i < array->size; i++)
265 		kfree(array->array[i].string.name);
266 	kunmap_atomic(array);
267 }
268 
269 /*
270  * the caller is responsible for freeing qstr.name
271  * when called by nfs_readdir_add_to_array, the strings will be freed in
272  * nfs_clear_readdir_array()
273  */
274 static
275 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
276 {
277 	string->len = len;
278 	string->name = kmemdup(name, len, GFP_KERNEL);
279 	if (string->name == NULL)
280 		return -ENOMEM;
281 	/*
282 	 * Avoid a kmemleak false positive. The pointer to the name is stored
283 	 * in a page cache page which kmemleak does not scan.
284 	 */
285 	kmemleak_not_leak(string->name);
286 	string->hash = full_name_hash(name, len);
287 	return 0;
288 }
289 
290 static
291 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
292 {
293 	struct nfs_cache_array *array = nfs_readdir_get_array(page);
294 	struct nfs_cache_array_entry *cache_entry;
295 	int ret;
296 
297 	if (IS_ERR(array))
298 		return PTR_ERR(array);
299 
300 	cache_entry = &array->array[array->size];
301 
302 	/* Check that this entry lies within the page bounds */
303 	ret = -ENOSPC;
304 	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
305 		goto out;
306 
307 	cache_entry->cookie = entry->prev_cookie;
308 	cache_entry->ino = entry->ino;
309 	cache_entry->d_type = entry->d_type;
310 	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
311 	if (ret)
312 		goto out;
313 	array->last_cookie = entry->cookie;
314 	array->size++;
315 	if (entry->eof != 0)
316 		array->eof_index = array->size;
317 out:
318 	nfs_readdir_release_array(page);
319 	return ret;
320 }
321 
322 static
323 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
324 {
325 	loff_t diff = desc->file->f_pos - desc->current_index;
326 	unsigned int index;
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 	return 0;
340 out_eof:
341 	desc->eof = 1;
342 	return -EBADCOOKIE;
343 }
344 
345 static
346 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
347 {
348 	int i;
349 	loff_t new_pos;
350 	int status = -EAGAIN;
351 
352 	for (i = 0; i < array->size; i++) {
353 		if (array->array[i].cookie == *desc->dir_cookie) {
354 			struct nfs_inode *nfsi = NFS_I(desc->file->f_path.dentry->d_inode);
355 			struct nfs_open_dir_context *ctx = desc->file->private_data;
356 
357 			new_pos = desc->current_index + i;
358 			if (ctx->attr_gencount != nfsi->attr_gencount
359 			    || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
360 				ctx->duped = 0;
361 				ctx->attr_gencount = nfsi->attr_gencount;
362 			} else if (new_pos < desc->file->f_pos) {
363 				if (ctx->duped > 0
364 				    && ctx->dup_cookie == *desc->dir_cookie) {
365 					if (printk_ratelimit()) {
366 						pr_notice("NFS: directory %s/%s contains a readdir loop."
367 								"Please contact your server vendor.  "
368 								"The file: %s has duplicate cookie %llu\n",
369 								desc->file->f_dentry->d_parent->d_name.name,
370 								desc->file->f_dentry->d_name.name,
371 								array->array[i].string.name,
372 								*desc->dir_cookie);
373 					}
374 					status = -ELOOP;
375 					goto out;
376 				}
377 				ctx->dup_cookie = *desc->dir_cookie;
378 				ctx->duped = -1;
379 			}
380 			desc->file->f_pos = new_pos;
381 			desc->cache_entry_index = i;
382 			return 0;
383 		}
384 	}
385 	if (array->eof_index >= 0) {
386 		status = -EBADCOOKIE;
387 		if (*desc->dir_cookie == array->last_cookie)
388 			desc->eof = 1;
389 	}
390 out:
391 	return status;
392 }
393 
394 static
395 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
396 {
397 	struct nfs_cache_array *array;
398 	int status;
399 
400 	array = nfs_readdir_get_array(desc->page);
401 	if (IS_ERR(array)) {
402 		status = PTR_ERR(array);
403 		goto out;
404 	}
405 
406 	if (*desc->dir_cookie == 0)
407 		status = nfs_readdir_search_for_pos(array, desc);
408 	else
409 		status = nfs_readdir_search_for_cookie(array, desc);
410 
411 	if (status == -EAGAIN) {
412 		desc->last_cookie = array->last_cookie;
413 		desc->current_index += array->size;
414 		desc->page_index++;
415 	}
416 	nfs_readdir_release_array(desc->page);
417 out:
418 	return status;
419 }
420 
421 /* Fill a page with xdr information before transferring to the cache page */
422 static
423 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
424 			struct nfs_entry *entry, struct file *file, struct inode *inode)
425 {
426 	struct nfs_open_dir_context *ctx = file->private_data;
427 	struct rpc_cred	*cred = ctx->cred;
428 	unsigned long	timestamp, gencount;
429 	int		error;
430 
431  again:
432 	timestamp = jiffies;
433 	gencount = nfs_inc_attr_generation_counter();
434 	error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
435 					  NFS_SERVER(inode)->dtsize, desc->plus);
436 	if (error < 0) {
437 		/* We requested READDIRPLUS, but the server doesn't grok it */
438 		if (error == -ENOTSUPP && desc->plus) {
439 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
440 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
441 			desc->plus = 0;
442 			goto again;
443 		}
444 		goto error;
445 	}
446 	desc->timestamp = timestamp;
447 	desc->gencount = gencount;
448 error:
449 	return error;
450 }
451 
452 static int xdr_decode(nfs_readdir_descriptor_t *desc,
453 		      struct nfs_entry *entry, struct xdr_stream *xdr)
454 {
455 	int error;
456 
457 	error = desc->decode(xdr, entry, desc->plus);
458 	if (error)
459 		return error;
460 	entry->fattr->time_start = desc->timestamp;
461 	entry->fattr->gencount = desc->gencount;
462 	return 0;
463 }
464 
465 static
466 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
467 {
468 	if (dentry->d_inode == NULL)
469 		goto different;
470 	if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
471 		goto different;
472 	return 1;
473 different:
474 	return 0;
475 }
476 
477 static
478 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
479 {
480 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
481 	struct dentry *dentry;
482 	struct dentry *alias;
483 	struct inode *dir = parent->d_inode;
484 	struct inode *inode;
485 
486 	if (filename.name[0] == '.') {
487 		if (filename.len == 1)
488 			return;
489 		if (filename.len == 2 && filename.name[1] == '.')
490 			return;
491 	}
492 	filename.hash = full_name_hash(filename.name, filename.len);
493 
494 	dentry = d_lookup(parent, &filename);
495 	if (dentry != NULL) {
496 		if (nfs_same_file(dentry, entry)) {
497 			nfs_refresh_inode(dentry->d_inode, entry->fattr);
498 			goto out;
499 		} else {
500 			d_drop(dentry);
501 			dput(dentry);
502 		}
503 	}
504 
505 	dentry = d_alloc(parent, &filename);
506 	if (dentry == NULL)
507 		return;
508 
509 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
510 	if (IS_ERR(inode))
511 		goto out;
512 
513 	alias = d_materialise_unique(dentry, inode);
514 	if (IS_ERR(alias))
515 		goto out;
516 	else if (alias) {
517 		nfs_set_verifier(alias, nfs_save_change_attribute(dir));
518 		dput(alias);
519 	} else
520 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
521 
522 out:
523 	dput(dentry);
524 }
525 
526 /* Perform conversion from xdr to cache array */
527 static
528 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
529 				struct page **xdr_pages, struct page *page, unsigned int buflen)
530 {
531 	struct xdr_stream stream;
532 	struct xdr_buf buf;
533 	struct page *scratch;
534 	struct nfs_cache_array *array;
535 	unsigned int count = 0;
536 	int status;
537 
538 	scratch = alloc_page(GFP_KERNEL);
539 	if (scratch == NULL)
540 		return -ENOMEM;
541 
542 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
543 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
544 
545 	do {
546 		status = xdr_decode(desc, entry, &stream);
547 		if (status != 0) {
548 			if (status == -EAGAIN)
549 				status = 0;
550 			break;
551 		}
552 
553 		count++;
554 
555 		if (desc->plus != 0)
556 			nfs_prime_dcache(desc->file->f_path.dentry, entry);
557 
558 		status = nfs_readdir_add_to_array(entry, page);
559 		if (status != 0)
560 			break;
561 	} while (!entry->eof);
562 
563 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
564 		array = nfs_readdir_get_array(page);
565 		if (!IS_ERR(array)) {
566 			array->eof_index = array->size;
567 			status = 0;
568 			nfs_readdir_release_array(page);
569 		} else
570 			status = PTR_ERR(array);
571 	}
572 
573 	put_page(scratch);
574 	return status;
575 }
576 
577 static
578 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
579 {
580 	unsigned int i;
581 	for (i = 0; i < npages; i++)
582 		put_page(pages[i]);
583 }
584 
585 static
586 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
587 		unsigned int npages)
588 {
589 	nfs_readdir_free_pagearray(pages, npages);
590 }
591 
592 /*
593  * nfs_readdir_large_page will allocate pages that must be freed with a call
594  * to nfs_readdir_free_large_page
595  */
596 static
597 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
598 {
599 	unsigned int i;
600 
601 	for (i = 0; i < npages; i++) {
602 		struct page *page = alloc_page(GFP_KERNEL);
603 		if (page == NULL)
604 			goto out_freepages;
605 		pages[i] = page;
606 	}
607 	return 0;
608 
609 out_freepages:
610 	nfs_readdir_free_pagearray(pages, i);
611 	return -ENOMEM;
612 }
613 
614 static
615 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
616 {
617 	struct page *pages[NFS_MAX_READDIR_PAGES];
618 	void *pages_ptr = NULL;
619 	struct nfs_entry entry;
620 	struct file	*file = desc->file;
621 	struct nfs_cache_array *array;
622 	int status = -ENOMEM;
623 	unsigned int array_size = ARRAY_SIZE(pages);
624 
625 	entry.prev_cookie = 0;
626 	entry.cookie = desc->last_cookie;
627 	entry.eof = 0;
628 	entry.fh = nfs_alloc_fhandle();
629 	entry.fattr = nfs_alloc_fattr();
630 	entry.server = NFS_SERVER(inode);
631 	if (entry.fh == NULL || entry.fattr == NULL)
632 		goto out;
633 
634 	array = nfs_readdir_get_array(page);
635 	if (IS_ERR(array)) {
636 		status = PTR_ERR(array);
637 		goto out;
638 	}
639 	memset(array, 0, sizeof(struct nfs_cache_array));
640 	array->eof_index = -1;
641 
642 	status = nfs_readdir_large_page(pages, array_size);
643 	if (status < 0)
644 		goto out_release_array;
645 	do {
646 		unsigned int pglen;
647 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
648 
649 		if (status < 0)
650 			break;
651 		pglen = status;
652 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
653 		if (status < 0) {
654 			if (status == -ENOSPC)
655 				status = 0;
656 			break;
657 		}
658 	} while (array->eof_index < 0);
659 
660 	nfs_readdir_free_large_page(pages_ptr, pages, array_size);
661 out_release_array:
662 	nfs_readdir_release_array(page);
663 out:
664 	nfs_free_fattr(entry.fattr);
665 	nfs_free_fhandle(entry.fh);
666 	return status;
667 }
668 
669 /*
670  * Now we cache directories properly, by converting xdr information
671  * to an array that can be used for lookups later.  This results in
672  * fewer cache pages, since we can store more information on each page.
673  * We only need to convert from xdr once so future lookups are much simpler
674  */
675 static
676 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
677 {
678 	struct inode	*inode = desc->file->f_path.dentry->d_inode;
679 	int ret;
680 
681 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
682 	if (ret < 0)
683 		goto error;
684 	SetPageUptodate(page);
685 
686 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
687 		/* Should never happen */
688 		nfs_zap_mapping(inode, inode->i_mapping);
689 	}
690 	unlock_page(page);
691 	return 0;
692  error:
693 	unlock_page(page);
694 	return ret;
695 }
696 
697 static
698 void cache_page_release(nfs_readdir_descriptor_t *desc)
699 {
700 	if (!desc->page->mapping)
701 		nfs_readdir_clear_array(desc->page);
702 	page_cache_release(desc->page);
703 	desc->page = NULL;
704 }
705 
706 static
707 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
708 {
709 	return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
710 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
711 }
712 
713 /*
714  * Returns 0 if desc->dir_cookie was found on page desc->page_index
715  */
716 static
717 int find_cache_page(nfs_readdir_descriptor_t *desc)
718 {
719 	int res;
720 
721 	desc->page = get_cache_page(desc);
722 	if (IS_ERR(desc->page))
723 		return PTR_ERR(desc->page);
724 
725 	res = nfs_readdir_search_array(desc);
726 	if (res != 0)
727 		cache_page_release(desc);
728 	return res;
729 }
730 
731 /* Search for desc->dir_cookie from the beginning of the page cache */
732 static inline
733 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
734 {
735 	int res;
736 
737 	if (desc->page_index == 0) {
738 		desc->current_index = 0;
739 		desc->last_cookie = 0;
740 	}
741 	do {
742 		res = find_cache_page(desc);
743 	} while (res == -EAGAIN);
744 	return res;
745 }
746 
747 /*
748  * Once we've found the start of the dirent within a page: fill 'er up...
749  */
750 static
751 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
752 		   filldir_t filldir)
753 {
754 	struct file	*file = desc->file;
755 	int i = 0;
756 	int res = 0;
757 	struct nfs_cache_array *array = NULL;
758 	struct nfs_open_dir_context *ctx = file->private_data;
759 
760 	array = nfs_readdir_get_array(desc->page);
761 	if (IS_ERR(array)) {
762 		res = PTR_ERR(array);
763 		goto out;
764 	}
765 
766 	for (i = desc->cache_entry_index; i < array->size; i++) {
767 		struct nfs_cache_array_entry *ent;
768 
769 		ent = &array->array[i];
770 		if (filldir(dirent, ent->string.name, ent->string.len,
771 		    file->f_pos, nfs_compat_user_ino64(ent->ino),
772 		    ent->d_type) < 0) {
773 			desc->eof = 1;
774 			break;
775 		}
776 		file->f_pos++;
777 		if (i < (array->size-1))
778 			*desc->dir_cookie = array->array[i+1].cookie;
779 		else
780 			*desc->dir_cookie = array->last_cookie;
781 		if (ctx->duped != 0)
782 			ctx->duped = 1;
783 	}
784 	if (array->eof_index >= 0)
785 		desc->eof = 1;
786 
787 	nfs_readdir_release_array(desc->page);
788 out:
789 	cache_page_release(desc);
790 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
791 			(unsigned long long)*desc->dir_cookie, res);
792 	return res;
793 }
794 
795 /*
796  * If we cannot find a cookie in our cache, we suspect that this is
797  * because it points to a deleted file, so we ask the server to return
798  * whatever it thinks is the next entry. We then feed this to filldir.
799  * If all goes well, we should then be able to find our way round the
800  * cache on the next call to readdir_search_pagecache();
801  *
802  * NOTE: we cannot add the anonymous page to the pagecache because
803  *	 the data it contains might not be page aligned. Besides,
804  *	 we should already have a complete representation of the
805  *	 directory in the page cache by the time we get here.
806  */
807 static inline
808 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
809 		     filldir_t filldir)
810 {
811 	struct page	*page = NULL;
812 	int		status;
813 	struct inode *inode = desc->file->f_path.dentry->d_inode;
814 	struct nfs_open_dir_context *ctx = desc->file->private_data;
815 
816 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
817 			(unsigned long long)*desc->dir_cookie);
818 
819 	page = alloc_page(GFP_HIGHUSER);
820 	if (!page) {
821 		status = -ENOMEM;
822 		goto out;
823 	}
824 
825 	desc->page_index = 0;
826 	desc->last_cookie = *desc->dir_cookie;
827 	desc->page = page;
828 	ctx->duped = 0;
829 
830 	status = nfs_readdir_xdr_to_array(desc, page, inode);
831 	if (status < 0)
832 		goto out_release;
833 
834 	status = nfs_do_filldir(desc, dirent, filldir);
835 
836  out:
837 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
838 			__func__, status);
839 	return status;
840  out_release:
841 	cache_page_release(desc);
842 	goto out;
843 }
844 
845 /* The file offset position represents the dirent entry number.  A
846    last cookie cache takes care of the common case of reading the
847    whole directory.
848  */
849 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
850 {
851 	struct dentry	*dentry = filp->f_path.dentry;
852 	struct inode	*inode = dentry->d_inode;
853 	nfs_readdir_descriptor_t my_desc,
854 			*desc = &my_desc;
855 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
856 	int res;
857 
858 	dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
859 			dentry->d_parent->d_name.name, dentry->d_name.name,
860 			(long long)filp->f_pos);
861 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
862 
863 	/*
864 	 * filp->f_pos points to the dirent entry number.
865 	 * *desc->dir_cookie has the cookie for the next entry. We have
866 	 * to either find the entry with the appropriate number or
867 	 * revalidate the cookie.
868 	 */
869 	memset(desc, 0, sizeof(*desc));
870 
871 	desc->file = filp;
872 	desc->dir_cookie = &dir_ctx->dir_cookie;
873 	desc->decode = NFS_PROTO(inode)->decode_dirent;
874 	desc->plus = NFS_USE_READDIRPLUS(inode);
875 
876 	nfs_block_sillyrename(dentry);
877 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
878 	if (res < 0)
879 		goto out;
880 
881 	do {
882 		res = readdir_search_pagecache(desc);
883 
884 		if (res == -EBADCOOKIE) {
885 			res = 0;
886 			/* This means either end of directory */
887 			if (*desc->dir_cookie && desc->eof == 0) {
888 				/* Or that the server has 'lost' a cookie */
889 				res = uncached_readdir(desc, dirent, filldir);
890 				if (res == 0)
891 					continue;
892 			}
893 			break;
894 		}
895 		if (res == -ETOOSMALL && desc->plus) {
896 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
897 			nfs_zap_caches(inode);
898 			desc->page_index = 0;
899 			desc->plus = 0;
900 			desc->eof = 0;
901 			continue;
902 		}
903 		if (res < 0)
904 			break;
905 
906 		res = nfs_do_filldir(desc, dirent, filldir);
907 		if (res < 0)
908 			break;
909 	} while (!desc->eof);
910 out:
911 	nfs_unblock_sillyrename(dentry);
912 	if (res > 0)
913 		res = 0;
914 	dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
915 			dentry->d_parent->d_name.name, dentry->d_name.name,
916 			res);
917 	return res;
918 }
919 
920 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
921 {
922 	struct dentry *dentry = filp->f_path.dentry;
923 	struct inode *inode = dentry->d_inode;
924 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
925 
926 	dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
927 			dentry->d_parent->d_name.name,
928 			dentry->d_name.name,
929 			offset, origin);
930 
931 	mutex_lock(&inode->i_mutex);
932 	switch (origin) {
933 		case 1:
934 			offset += filp->f_pos;
935 		case 0:
936 			if (offset >= 0)
937 				break;
938 		default:
939 			offset = -EINVAL;
940 			goto out;
941 	}
942 	if (offset != filp->f_pos) {
943 		filp->f_pos = offset;
944 		dir_ctx->dir_cookie = 0;
945 		dir_ctx->duped = 0;
946 	}
947 out:
948 	mutex_unlock(&inode->i_mutex);
949 	return offset;
950 }
951 
952 /*
953  * All directory operations under NFS are synchronous, so fsync()
954  * is a dummy operation.
955  */
956 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
957 			 int datasync)
958 {
959 	struct dentry *dentry = filp->f_path.dentry;
960 	struct inode *inode = dentry->d_inode;
961 
962 	dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
963 			dentry->d_parent->d_name.name, dentry->d_name.name,
964 			datasync);
965 
966 	mutex_lock(&inode->i_mutex);
967 	nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
968 	mutex_unlock(&inode->i_mutex);
969 	return 0;
970 }
971 
972 /**
973  * nfs_force_lookup_revalidate - Mark the directory as having changed
974  * @dir - pointer to directory inode
975  *
976  * This forces the revalidation code in nfs_lookup_revalidate() to do a
977  * full lookup on all child dentries of 'dir' whenever a change occurs
978  * on the server that might have invalidated our dcache.
979  *
980  * The caller should be holding dir->i_lock
981  */
982 void nfs_force_lookup_revalidate(struct inode *dir)
983 {
984 	NFS_I(dir)->cache_change_attribute++;
985 }
986 
987 /*
988  * A check for whether or not the parent directory has changed.
989  * In the case it has, we assume that the dentries are untrustworthy
990  * and may need to be looked up again.
991  */
992 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
993 {
994 	if (IS_ROOT(dentry))
995 		return 1;
996 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
997 		return 0;
998 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
999 		return 0;
1000 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1001 	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1002 		return 0;
1003 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1004 		return 0;
1005 	return 1;
1006 }
1007 
1008 /*
1009  * Return the intent data that applies to this particular path component
1010  *
1011  * Note that the current set of intents only apply to the very last
1012  * component of the path and none of them is set before that last
1013  * component.
1014  */
1015 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1016 						unsigned int mask)
1017 {
1018 	return nd->flags & mask;
1019 }
1020 
1021 /*
1022  * Use intent information to check whether or not we're going to do
1023  * an O_EXCL create using this path component.
1024  */
1025 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1026 {
1027 	if (NFS_PROTO(dir)->version == 2)
1028 		return 0;
1029 	return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1030 }
1031 
1032 /*
1033  * Inode and filehandle revalidation for lookups.
1034  *
1035  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1036  * or if the intent information indicates that we're about to open this
1037  * particular file and the "nocto" mount flag is not set.
1038  *
1039  */
1040 static inline
1041 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1042 {
1043 	struct nfs_server *server = NFS_SERVER(inode);
1044 
1045 	if (IS_AUTOMOUNT(inode))
1046 		return 0;
1047 	if (nd != NULL) {
1048 		/* VFS wants an on-the-wire revalidation */
1049 		if (nd->flags & LOOKUP_REVAL)
1050 			goto out_force;
1051 		/* This is an open(2) */
1052 		if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1053 				!(server->flags & NFS_MOUNT_NOCTO) &&
1054 				(S_ISREG(inode->i_mode) ||
1055 				 S_ISDIR(inode->i_mode)))
1056 			goto out_force;
1057 		return 0;
1058 	}
1059 	return nfs_revalidate_inode(server, inode);
1060 out_force:
1061 	return __nfs_revalidate_inode(server, inode);
1062 }
1063 
1064 /*
1065  * We judge how long we want to trust negative
1066  * dentries by looking at the parent inode mtime.
1067  *
1068  * If parent mtime has changed, we revalidate, else we wait for a
1069  * period corresponding to the parent's attribute cache timeout value.
1070  */
1071 static inline
1072 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1073 		       struct nameidata *nd)
1074 {
1075 	/* Don't revalidate a negative dentry if we're creating a new file */
1076 	if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1077 		return 0;
1078 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1079 		return 1;
1080 	return !nfs_check_verifier(dir, dentry);
1081 }
1082 
1083 /*
1084  * This is called every time the dcache has a lookup hit,
1085  * and we should check whether we can really trust that
1086  * lookup.
1087  *
1088  * NOTE! The hit can be a negative hit too, don't assume
1089  * we have an inode!
1090  *
1091  * If the parent directory is seen to have changed, we throw out the
1092  * cached dentry and do a new lookup.
1093  */
1094 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1095 {
1096 	struct inode *dir;
1097 	struct inode *inode;
1098 	struct dentry *parent;
1099 	struct nfs_fh *fhandle = NULL;
1100 	struct nfs_fattr *fattr = NULL;
1101 	int error;
1102 
1103 	if (nd->flags & LOOKUP_RCU)
1104 		return -ECHILD;
1105 
1106 	parent = dget_parent(dentry);
1107 	dir = parent->d_inode;
1108 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1109 	inode = dentry->d_inode;
1110 
1111 	if (!inode) {
1112 		if (nfs_neg_need_reval(dir, dentry, nd))
1113 			goto out_bad;
1114 		goto out_valid;
1115 	}
1116 
1117 	if (is_bad_inode(inode)) {
1118 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1119 				__func__, dentry->d_parent->d_name.name,
1120 				dentry->d_name.name);
1121 		goto out_bad;
1122 	}
1123 
1124 	if (nfs_have_delegation(inode, FMODE_READ))
1125 		goto out_set_verifier;
1126 
1127 	/* Force a full look up iff the parent directory has changed */
1128 	if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1129 		if (nfs_lookup_verify_inode(inode, nd))
1130 			goto out_zap_parent;
1131 		goto out_valid;
1132 	}
1133 
1134 	if (NFS_STALE(inode))
1135 		goto out_bad;
1136 
1137 	error = -ENOMEM;
1138 	fhandle = nfs_alloc_fhandle();
1139 	fattr = nfs_alloc_fattr();
1140 	if (fhandle == NULL || fattr == NULL)
1141 		goto out_error;
1142 
1143 	error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1144 	if (error)
1145 		goto out_bad;
1146 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1147 		goto out_bad;
1148 	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1149 		goto out_bad;
1150 
1151 	nfs_free_fattr(fattr);
1152 	nfs_free_fhandle(fhandle);
1153 out_set_verifier:
1154 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1155  out_valid:
1156 	dput(parent);
1157 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1158 			__func__, dentry->d_parent->d_name.name,
1159 			dentry->d_name.name);
1160 	return 1;
1161 out_zap_parent:
1162 	nfs_zap_caches(dir);
1163  out_bad:
1164 	nfs_mark_for_revalidate(dir);
1165 	if (inode && S_ISDIR(inode->i_mode)) {
1166 		/* Purge readdir caches. */
1167 		nfs_zap_caches(inode);
1168 		/* If we have submounts, don't unhash ! */
1169 		if (have_submounts(dentry))
1170 			goto out_valid;
1171 		if (dentry->d_flags & DCACHE_DISCONNECTED)
1172 			goto out_valid;
1173 		shrink_dcache_parent(dentry);
1174 	}
1175 	d_drop(dentry);
1176 	nfs_free_fattr(fattr);
1177 	nfs_free_fhandle(fhandle);
1178 	dput(parent);
1179 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1180 			__func__, dentry->d_parent->d_name.name,
1181 			dentry->d_name.name);
1182 	return 0;
1183 out_error:
1184 	nfs_free_fattr(fattr);
1185 	nfs_free_fhandle(fhandle);
1186 	dput(parent);
1187 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1188 			__func__, dentry->d_parent->d_name.name,
1189 			dentry->d_name.name, error);
1190 	return error;
1191 }
1192 
1193 /*
1194  * This is called from dput() when d_count is going to 0.
1195  */
1196 static int nfs_dentry_delete(const struct dentry *dentry)
1197 {
1198 	dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1199 		dentry->d_parent->d_name.name, dentry->d_name.name,
1200 		dentry->d_flags);
1201 
1202 	/* Unhash any dentry with a stale inode */
1203 	if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1204 		return 1;
1205 
1206 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1207 		/* Unhash it, so that ->d_iput() would be called */
1208 		return 1;
1209 	}
1210 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1211 		/* Unhash it, so that ancestors of killed async unlink
1212 		 * files will be cleaned up during umount */
1213 		return 1;
1214 	}
1215 	return 0;
1216 
1217 }
1218 
1219 static void nfs_drop_nlink(struct inode *inode)
1220 {
1221 	spin_lock(&inode->i_lock);
1222 	if (inode->i_nlink > 0)
1223 		drop_nlink(inode);
1224 	spin_unlock(&inode->i_lock);
1225 }
1226 
1227 /*
1228  * Called when the dentry loses inode.
1229  * We use it to clean up silly-renamed files.
1230  */
1231 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1232 {
1233 	if (S_ISDIR(inode->i_mode))
1234 		/* drop any readdir cache as it could easily be old */
1235 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1236 
1237 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1238 		drop_nlink(inode);
1239 		nfs_complete_unlink(dentry, inode);
1240 	}
1241 	iput(inode);
1242 }
1243 
1244 static void nfs_d_release(struct dentry *dentry)
1245 {
1246 	/* free cached devname value, if it survived that far */
1247 	if (unlikely(dentry->d_fsdata)) {
1248 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1249 			WARN_ON(1);
1250 		else
1251 			kfree(dentry->d_fsdata);
1252 	}
1253 }
1254 
1255 const struct dentry_operations nfs_dentry_operations = {
1256 	.d_revalidate	= nfs_lookup_revalidate,
1257 	.d_delete	= nfs_dentry_delete,
1258 	.d_iput		= nfs_dentry_iput,
1259 	.d_automount	= nfs_d_automount,
1260 	.d_release	= nfs_d_release,
1261 };
1262 
1263 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1264 {
1265 	struct dentry *res;
1266 	struct dentry *parent;
1267 	struct inode *inode = NULL;
1268 	struct nfs_fh *fhandle = NULL;
1269 	struct nfs_fattr *fattr = NULL;
1270 	int error;
1271 
1272 	dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1273 		dentry->d_parent->d_name.name, dentry->d_name.name);
1274 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1275 
1276 	res = ERR_PTR(-ENAMETOOLONG);
1277 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1278 		goto out;
1279 
1280 	/*
1281 	 * If we're doing an exclusive create, optimize away the lookup
1282 	 * but don't hash the dentry.
1283 	 */
1284 	if (nfs_is_exclusive_create(dir, nd)) {
1285 		d_instantiate(dentry, NULL);
1286 		res = NULL;
1287 		goto out;
1288 	}
1289 
1290 	res = ERR_PTR(-ENOMEM);
1291 	fhandle = nfs_alloc_fhandle();
1292 	fattr = nfs_alloc_fattr();
1293 	if (fhandle == NULL || fattr == NULL)
1294 		goto out;
1295 
1296 	parent = dentry->d_parent;
1297 	/* Protect against concurrent sillydeletes */
1298 	nfs_block_sillyrename(parent);
1299 	error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1300 	if (error == -ENOENT)
1301 		goto no_entry;
1302 	if (error < 0) {
1303 		res = ERR_PTR(error);
1304 		goto out_unblock_sillyrename;
1305 	}
1306 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1307 	res = ERR_CAST(inode);
1308 	if (IS_ERR(res))
1309 		goto out_unblock_sillyrename;
1310 
1311 no_entry:
1312 	res = d_materialise_unique(dentry, inode);
1313 	if (res != NULL) {
1314 		if (IS_ERR(res))
1315 			goto out_unblock_sillyrename;
1316 		dentry = res;
1317 	}
1318 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1319 out_unblock_sillyrename:
1320 	nfs_unblock_sillyrename(parent);
1321 out:
1322 	nfs_free_fattr(fattr);
1323 	nfs_free_fhandle(fhandle);
1324 	return res;
1325 }
1326 
1327 #ifdef CONFIG_NFS_V4
1328 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1329 
1330 const struct dentry_operations nfs4_dentry_operations = {
1331 	.d_revalidate	= nfs_open_revalidate,
1332 	.d_delete	= nfs_dentry_delete,
1333 	.d_iput		= nfs_dentry_iput,
1334 	.d_automount	= nfs_d_automount,
1335 	.d_release	= nfs_d_release,
1336 };
1337 
1338 /*
1339  * Use intent information to determine whether we need to substitute
1340  * the NFSv4-style stateful OPEN for the LOOKUP call
1341  */
1342 static int is_atomic_open(struct nameidata *nd)
1343 {
1344 	if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1345 		return 0;
1346 	/* NFS does not (yet) have a stateful open for directories */
1347 	if (nd->flags & LOOKUP_DIRECTORY)
1348 		return 0;
1349 	/* Are we trying to write to a read only partition? */
1350 	if (__mnt_is_readonly(nd->path.mnt) &&
1351 	    (nd->intent.open.flags & (O_CREAT|O_TRUNC|O_ACCMODE)))
1352 		return 0;
1353 	return 1;
1354 }
1355 
1356 static fmode_t flags_to_mode(int flags)
1357 {
1358 	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1359 	if ((flags & O_ACCMODE) != O_WRONLY)
1360 		res |= FMODE_READ;
1361 	if ((flags & O_ACCMODE) != O_RDONLY)
1362 		res |= FMODE_WRITE;
1363 	return res;
1364 }
1365 
1366 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1367 {
1368 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
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->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->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 	open_flags = nd->intent.open.flags;
1429 	attr.ia_valid = ATTR_OPEN;
1430 
1431 	ctx = create_nfs_open_context(dentry, open_flags);
1432 	res = ERR_CAST(ctx);
1433 	if (IS_ERR(ctx))
1434 		goto out;
1435 
1436 	if (nd->flags & LOOKUP_CREATE) {
1437 		attr.ia_mode = nd->intent.open.create_mode;
1438 		attr.ia_valid |= ATTR_MODE;
1439 		attr.ia_mode &= ~current_umask();
1440 	} else
1441 		open_flags &= ~(O_EXCL | O_CREAT);
1442 
1443 	if (open_flags & O_TRUNC) {
1444 		attr.ia_valid |= ATTR_SIZE;
1445 		attr.ia_size = 0;
1446 	}
1447 
1448 	/* Open the file on the server */
1449 	nfs_block_sillyrename(dentry->d_parent);
1450 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1451 	if (IS_ERR(inode)) {
1452 		nfs_unblock_sillyrename(dentry->d_parent);
1453 		put_nfs_open_context(ctx);
1454 		switch (PTR_ERR(inode)) {
1455 			/* Make a negative dentry */
1456 			case -ENOENT:
1457 				d_add(dentry, NULL);
1458 				res = NULL;
1459 				goto out;
1460 			/* This turned out not to be a regular file */
1461 			case -EISDIR:
1462 			case -ENOTDIR:
1463 				goto no_open;
1464 			case -ELOOP:
1465 				if (!(nd->intent.open.flags & O_NOFOLLOW))
1466 					goto no_open;
1467 			/* case -EINVAL: */
1468 			default:
1469 				res = ERR_CAST(inode);
1470 				goto out;
1471 		}
1472 	}
1473 	res = d_add_unique(dentry, inode);
1474 	nfs_unblock_sillyrename(dentry->d_parent);
1475 	if (res != NULL) {
1476 		dput(ctx->dentry);
1477 		ctx->dentry = dget(res);
1478 		dentry = res;
1479 	}
1480 	err = nfs_intent_set_file(nd, ctx);
1481 	if (err < 0) {
1482 		if (res != NULL)
1483 			dput(res);
1484 		return ERR_PTR(err);
1485 	}
1486 out:
1487 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1488 	return res;
1489 no_open:
1490 	return nfs_lookup(dir, dentry, nd);
1491 }
1492 
1493 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1494 {
1495 	struct dentry *parent = NULL;
1496 	struct inode *inode;
1497 	struct inode *dir;
1498 	struct nfs_open_context *ctx;
1499 	struct iattr attr;
1500 	int openflags, ret = 0;
1501 
1502 	if (nd->flags & LOOKUP_RCU)
1503 		return -ECHILD;
1504 
1505 	inode = dentry->d_inode;
1506 	if (!is_atomic_open(nd) || d_mountpoint(dentry))
1507 		goto no_open;
1508 
1509 	parent = dget_parent(dentry);
1510 	dir = parent->d_inode;
1511 
1512 	/* We can't create new files in nfs_open_revalidate(), so we
1513 	 * optimize away revalidation of negative dentries.
1514 	 */
1515 	if (inode == NULL) {
1516 		if (!nfs_neg_need_reval(dir, dentry, nd))
1517 			ret = 1;
1518 		goto out;
1519 	}
1520 
1521 	/* NFS only supports OPEN on regular files */
1522 	if (!S_ISREG(inode->i_mode))
1523 		goto no_open_dput;
1524 	openflags = nd->intent.open.flags;
1525 	/* We cannot do exclusive creation on a positive dentry */
1526 	if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1527 		goto no_open_dput;
1528 	/* We can't create new files here */
1529 	openflags &= ~(O_CREAT|O_EXCL);
1530 
1531 	ctx = create_nfs_open_context(dentry, openflags);
1532 	ret = PTR_ERR(ctx);
1533 	if (IS_ERR(ctx))
1534 		goto out;
1535 
1536 	attr.ia_valid = ATTR_OPEN;
1537 	if (openflags & O_TRUNC) {
1538 		attr.ia_valid |= ATTR_SIZE;
1539 		attr.ia_size = 0;
1540 		nfs_wb_all(inode);
1541 	}
1542 
1543 	/*
1544 	 * Note: we're not holding inode->i_mutex and so may be racing with
1545 	 * operations that change the directory. We therefore save the
1546 	 * change attribute *before* we do the RPC call.
1547 	 */
1548 	inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr);
1549 	if (IS_ERR(inode)) {
1550 		ret = PTR_ERR(inode);
1551 		switch (ret) {
1552 		case -EPERM:
1553 		case -EACCES:
1554 		case -EDQUOT:
1555 		case -ENOSPC:
1556 		case -EROFS:
1557 			goto out_put_ctx;
1558 		default:
1559 			goto out_drop;
1560 		}
1561 	}
1562 	iput(inode);
1563 	if (inode != dentry->d_inode)
1564 		goto out_drop;
1565 
1566 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1567 	ret = nfs_intent_set_file(nd, ctx);
1568 	if (ret >= 0)
1569 		ret = 1;
1570 out:
1571 	dput(parent);
1572 	return ret;
1573 out_drop:
1574 	d_drop(dentry);
1575 	ret = 0;
1576 out_put_ctx:
1577 	put_nfs_open_context(ctx);
1578 	goto out;
1579 
1580 no_open_dput:
1581 	dput(parent);
1582 no_open:
1583 	return nfs_lookup_revalidate(dentry, nd);
1584 }
1585 
1586 static int nfs_open_create(struct inode *dir, struct dentry *dentry,
1587 		umode_t mode, struct nameidata *nd)
1588 {
1589 	struct nfs_open_context *ctx = NULL;
1590 	struct iattr attr;
1591 	int error;
1592 	int open_flags = O_CREAT|O_EXCL;
1593 
1594 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1595 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1596 
1597 	attr.ia_mode = mode;
1598 	attr.ia_valid = ATTR_MODE;
1599 
1600 	if (nd)
1601 		open_flags = nd->intent.open.flags;
1602 
1603 	ctx = create_nfs_open_context(dentry, open_flags);
1604 	error = PTR_ERR(ctx);
1605 	if (IS_ERR(ctx))
1606 		goto out_err_drop;
1607 
1608 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1609 	if (error != 0)
1610 		goto out_put_ctx;
1611 	if (nd) {
1612 		error = nfs_intent_set_file(nd, ctx);
1613 		if (error < 0)
1614 			goto out_err;
1615 	} else {
1616 		put_nfs_open_context(ctx);
1617 	}
1618 	return 0;
1619 out_put_ctx:
1620 	put_nfs_open_context(ctx);
1621 out_err_drop:
1622 	d_drop(dentry);
1623 out_err:
1624 	return error;
1625 }
1626 
1627 #endif /* CONFIG_NFSV4 */
1628 
1629 /*
1630  * Code common to create, mkdir, and mknod.
1631  */
1632 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1633 				struct nfs_fattr *fattr)
1634 {
1635 	struct dentry *parent = dget_parent(dentry);
1636 	struct inode *dir = parent->d_inode;
1637 	struct inode *inode;
1638 	int error = -EACCES;
1639 
1640 	d_drop(dentry);
1641 
1642 	/* We may have been initialized further down */
1643 	if (dentry->d_inode)
1644 		goto out;
1645 	if (fhandle->size == 0) {
1646 		error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1647 		if (error)
1648 			goto out_error;
1649 	}
1650 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1651 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1652 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1653 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1654 		if (error < 0)
1655 			goto out_error;
1656 	}
1657 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1658 	error = PTR_ERR(inode);
1659 	if (IS_ERR(inode))
1660 		goto out_error;
1661 	d_add(dentry, inode);
1662 out:
1663 	dput(parent);
1664 	return 0;
1665 out_error:
1666 	nfs_mark_for_revalidate(dir);
1667 	dput(parent);
1668 	return error;
1669 }
1670 
1671 /*
1672  * Following a failed create operation, we drop the dentry rather
1673  * than retain a negative dentry. This avoids a problem in the event
1674  * that the operation succeeded on the server, but an error in the
1675  * reply path made it appear to have failed.
1676  */
1677 static int nfs_create(struct inode *dir, struct dentry *dentry,
1678 		umode_t mode, struct nameidata *nd)
1679 {
1680 	struct iattr attr;
1681 	int error;
1682 	int open_flags = O_CREAT|O_EXCL;
1683 
1684 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1685 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1686 
1687 	attr.ia_mode = mode;
1688 	attr.ia_valid = ATTR_MODE;
1689 
1690 	if (nd)
1691 		open_flags = nd->intent.open.flags;
1692 
1693 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1694 	if (error != 0)
1695 		goto out_err;
1696 	return 0;
1697 out_err:
1698 	d_drop(dentry);
1699 	return error;
1700 }
1701 
1702 /*
1703  * See comments for nfs_proc_create regarding failed operations.
1704  */
1705 static int
1706 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1707 {
1708 	struct iattr attr;
1709 	int status;
1710 
1711 	dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1712 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1713 
1714 	if (!new_valid_dev(rdev))
1715 		return -EINVAL;
1716 
1717 	attr.ia_mode = mode;
1718 	attr.ia_valid = ATTR_MODE;
1719 
1720 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1721 	if (status != 0)
1722 		goto out_err;
1723 	return 0;
1724 out_err:
1725 	d_drop(dentry);
1726 	return status;
1727 }
1728 
1729 /*
1730  * See comments for nfs_proc_create regarding failed operations.
1731  */
1732 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1733 {
1734 	struct iattr attr;
1735 	int error;
1736 
1737 	dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1738 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1739 
1740 	attr.ia_valid = ATTR_MODE;
1741 	attr.ia_mode = mode | S_IFDIR;
1742 
1743 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1744 	if (error != 0)
1745 		goto out_err;
1746 	return 0;
1747 out_err:
1748 	d_drop(dentry);
1749 	return error;
1750 }
1751 
1752 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1753 {
1754 	if (dentry->d_inode != NULL && !d_unhashed(dentry))
1755 		d_delete(dentry);
1756 }
1757 
1758 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1759 {
1760 	int error;
1761 
1762 	dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1763 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1764 
1765 	error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1766 	/* Ensure the VFS deletes this inode */
1767 	if (error == 0 && dentry->d_inode != NULL)
1768 		clear_nlink(dentry->d_inode);
1769 	else if (error == -ENOENT)
1770 		nfs_dentry_handle_enoent(dentry);
1771 
1772 	return error;
1773 }
1774 
1775 /*
1776  * Remove a file after making sure there are no pending writes,
1777  * and after checking that the file has only one user.
1778  *
1779  * We invalidate the attribute cache and free the inode prior to the operation
1780  * to avoid possible races if the server reuses the inode.
1781  */
1782 static int nfs_safe_remove(struct dentry *dentry)
1783 {
1784 	struct inode *dir = dentry->d_parent->d_inode;
1785 	struct inode *inode = dentry->d_inode;
1786 	int error = -EBUSY;
1787 
1788 	dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1789 		dentry->d_parent->d_name.name, dentry->d_name.name);
1790 
1791 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1792 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1793 		error = 0;
1794 		goto out;
1795 	}
1796 
1797 	if (inode != NULL) {
1798 		nfs_inode_return_delegation(inode);
1799 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1800 		/* The VFS may want to delete this inode */
1801 		if (error == 0)
1802 			nfs_drop_nlink(inode);
1803 		nfs_mark_for_revalidate(inode);
1804 	} else
1805 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1806 	if (error == -ENOENT)
1807 		nfs_dentry_handle_enoent(dentry);
1808 out:
1809 	return error;
1810 }
1811 
1812 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1813  *  belongs to an active ".nfs..." file and we return -EBUSY.
1814  *
1815  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1816  */
1817 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1818 {
1819 	int error;
1820 	int need_rehash = 0;
1821 
1822 	dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1823 		dir->i_ino, dentry->d_name.name);
1824 
1825 	spin_lock(&dentry->d_lock);
1826 	if (dentry->d_count > 1) {
1827 		spin_unlock(&dentry->d_lock);
1828 		/* Start asynchronous writeout of the inode */
1829 		write_inode_now(dentry->d_inode, 0);
1830 		error = nfs_sillyrename(dir, dentry);
1831 		return error;
1832 	}
1833 	if (!d_unhashed(dentry)) {
1834 		__d_drop(dentry);
1835 		need_rehash = 1;
1836 	}
1837 	spin_unlock(&dentry->d_lock);
1838 	error = nfs_safe_remove(dentry);
1839 	if (!error || error == -ENOENT) {
1840 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1841 	} else if (need_rehash)
1842 		d_rehash(dentry);
1843 	return error;
1844 }
1845 
1846 /*
1847  * To create a symbolic link, most file systems instantiate a new inode,
1848  * add a page to it containing the path, then write it out to the disk
1849  * using prepare_write/commit_write.
1850  *
1851  * Unfortunately the NFS client can't create the in-core inode first
1852  * because it needs a file handle to create an in-core inode (see
1853  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1854  * symlink request has completed on the server.
1855  *
1856  * So instead we allocate a raw page, copy the symname into it, then do
1857  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1858  * now have a new file handle and can instantiate an in-core NFS inode
1859  * and move the raw page into its mapping.
1860  */
1861 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1862 {
1863 	struct pagevec lru_pvec;
1864 	struct page *page;
1865 	char *kaddr;
1866 	struct iattr attr;
1867 	unsigned int pathlen = strlen(symname);
1868 	int error;
1869 
1870 	dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1871 		dir->i_ino, dentry->d_name.name, symname);
1872 
1873 	if (pathlen > PAGE_SIZE)
1874 		return -ENAMETOOLONG;
1875 
1876 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1877 	attr.ia_valid = ATTR_MODE;
1878 
1879 	page = alloc_page(GFP_HIGHUSER);
1880 	if (!page)
1881 		return -ENOMEM;
1882 
1883 	kaddr = kmap_atomic(page);
1884 	memcpy(kaddr, symname, pathlen);
1885 	if (pathlen < PAGE_SIZE)
1886 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1887 	kunmap_atomic(kaddr);
1888 
1889 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1890 	if (error != 0) {
1891 		dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1892 			dir->i_sb->s_id, dir->i_ino,
1893 			dentry->d_name.name, symname, error);
1894 		d_drop(dentry);
1895 		__free_page(page);
1896 		return error;
1897 	}
1898 
1899 	/*
1900 	 * No big deal if we can't add this page to the page cache here.
1901 	 * READLINK will get the missing page from the server if needed.
1902 	 */
1903 	pagevec_init(&lru_pvec, 0);
1904 	if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1905 							GFP_KERNEL)) {
1906 		pagevec_add(&lru_pvec, page);
1907 		pagevec_lru_add_file(&lru_pvec);
1908 		SetPageUptodate(page);
1909 		unlock_page(page);
1910 	} else
1911 		__free_page(page);
1912 
1913 	return 0;
1914 }
1915 
1916 static int
1917 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1918 {
1919 	struct inode *inode = old_dentry->d_inode;
1920 	int error;
1921 
1922 	dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1923 		old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1924 		dentry->d_parent->d_name.name, dentry->d_name.name);
1925 
1926 	nfs_inode_return_delegation(inode);
1927 
1928 	d_drop(dentry);
1929 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1930 	if (error == 0) {
1931 		ihold(inode);
1932 		d_add(dentry, inode);
1933 	}
1934 	return error;
1935 }
1936 
1937 /*
1938  * RENAME
1939  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1940  * different file handle for the same inode after a rename (e.g. when
1941  * moving to a different directory). A fail-safe method to do so would
1942  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1943  * rename the old file using the sillyrename stuff. This way, the original
1944  * file in old_dir will go away when the last process iput()s the inode.
1945  *
1946  * FIXED.
1947  *
1948  * It actually works quite well. One needs to have the possibility for
1949  * at least one ".nfs..." file in each directory the file ever gets
1950  * moved or linked to which happens automagically with the new
1951  * implementation that only depends on the dcache stuff instead of
1952  * using the inode layer
1953  *
1954  * Unfortunately, things are a little more complicated than indicated
1955  * above. For a cross-directory move, we want to make sure we can get
1956  * rid of the old inode after the operation.  This means there must be
1957  * no pending writes (if it's a file), and the use count must be 1.
1958  * If these conditions are met, we can drop the dentries before doing
1959  * the rename.
1960  */
1961 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1962 		      struct inode *new_dir, struct dentry *new_dentry)
1963 {
1964 	struct inode *old_inode = old_dentry->d_inode;
1965 	struct inode *new_inode = new_dentry->d_inode;
1966 	struct dentry *dentry = NULL, *rehash = NULL;
1967 	int error = -EBUSY;
1968 
1969 	dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1970 		 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1971 		 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1972 		 new_dentry->d_count);
1973 
1974 	/*
1975 	 * For non-directories, check whether the target is busy and if so,
1976 	 * make a copy of the dentry and then do a silly-rename. If the
1977 	 * silly-rename succeeds, the copied dentry is hashed and becomes
1978 	 * the new target.
1979 	 */
1980 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1981 		/*
1982 		 * To prevent any new references to the target during the
1983 		 * rename, we unhash the dentry in advance.
1984 		 */
1985 		if (!d_unhashed(new_dentry)) {
1986 			d_drop(new_dentry);
1987 			rehash = new_dentry;
1988 		}
1989 
1990 		if (new_dentry->d_count > 2) {
1991 			int err;
1992 
1993 			/* copy the target dentry's name */
1994 			dentry = d_alloc(new_dentry->d_parent,
1995 					 &new_dentry->d_name);
1996 			if (!dentry)
1997 				goto out;
1998 
1999 			/* silly-rename the existing target ... */
2000 			err = nfs_sillyrename(new_dir, new_dentry);
2001 			if (err)
2002 				goto out;
2003 
2004 			new_dentry = dentry;
2005 			rehash = NULL;
2006 			new_inode = NULL;
2007 		}
2008 	}
2009 
2010 	nfs_inode_return_delegation(old_inode);
2011 	if (new_inode != NULL)
2012 		nfs_inode_return_delegation(new_inode);
2013 
2014 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
2015 					   new_dir, &new_dentry->d_name);
2016 	nfs_mark_for_revalidate(old_inode);
2017 out:
2018 	if (rehash)
2019 		d_rehash(rehash);
2020 	if (!error) {
2021 		if (new_inode != NULL)
2022 			nfs_drop_nlink(new_inode);
2023 		d_move(old_dentry, new_dentry);
2024 		nfs_set_verifier(new_dentry,
2025 					nfs_save_change_attribute(new_dir));
2026 	} else if (error == -ENOENT)
2027 		nfs_dentry_handle_enoent(old_dentry);
2028 
2029 	/* new dentry created? */
2030 	if (dentry)
2031 		dput(dentry);
2032 	return error;
2033 }
2034 
2035 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2036 static LIST_HEAD(nfs_access_lru_list);
2037 static atomic_long_t nfs_access_nr_entries;
2038 
2039 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2040 {
2041 	put_rpccred(entry->cred);
2042 	kfree(entry);
2043 	smp_mb__before_atomic_dec();
2044 	atomic_long_dec(&nfs_access_nr_entries);
2045 	smp_mb__after_atomic_dec();
2046 }
2047 
2048 static void nfs_access_free_list(struct list_head *head)
2049 {
2050 	struct nfs_access_entry *cache;
2051 
2052 	while (!list_empty(head)) {
2053 		cache = list_entry(head->next, struct nfs_access_entry, lru);
2054 		list_del(&cache->lru);
2055 		nfs_access_free_entry(cache);
2056 	}
2057 }
2058 
2059 int nfs_access_cache_shrinker(struct shrinker *shrink,
2060 			      struct shrink_control *sc)
2061 {
2062 	LIST_HEAD(head);
2063 	struct nfs_inode *nfsi, *next;
2064 	struct nfs_access_entry *cache;
2065 	int nr_to_scan = sc->nr_to_scan;
2066 	gfp_t gfp_mask = sc->gfp_mask;
2067 
2068 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2069 		return (nr_to_scan == 0) ? 0 : -1;
2070 
2071 	spin_lock(&nfs_access_lru_lock);
2072 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2073 		struct inode *inode;
2074 
2075 		if (nr_to_scan-- == 0)
2076 			break;
2077 		inode = &nfsi->vfs_inode;
2078 		spin_lock(&inode->i_lock);
2079 		if (list_empty(&nfsi->access_cache_entry_lru))
2080 			goto remove_lru_entry;
2081 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2082 				struct nfs_access_entry, lru);
2083 		list_move(&cache->lru, &head);
2084 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2085 		if (!list_empty(&nfsi->access_cache_entry_lru))
2086 			list_move_tail(&nfsi->access_cache_inode_lru,
2087 					&nfs_access_lru_list);
2088 		else {
2089 remove_lru_entry:
2090 			list_del_init(&nfsi->access_cache_inode_lru);
2091 			smp_mb__before_clear_bit();
2092 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2093 			smp_mb__after_clear_bit();
2094 		}
2095 		spin_unlock(&inode->i_lock);
2096 	}
2097 	spin_unlock(&nfs_access_lru_lock);
2098 	nfs_access_free_list(&head);
2099 	return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2100 }
2101 
2102 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2103 {
2104 	struct rb_root *root_node = &nfsi->access_cache;
2105 	struct rb_node *n;
2106 	struct nfs_access_entry *entry;
2107 
2108 	/* Unhook entries from the cache */
2109 	while ((n = rb_first(root_node)) != NULL) {
2110 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2111 		rb_erase(n, root_node);
2112 		list_move(&entry->lru, head);
2113 	}
2114 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2115 }
2116 
2117 void nfs_access_zap_cache(struct inode *inode)
2118 {
2119 	LIST_HEAD(head);
2120 
2121 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2122 		return;
2123 	/* Remove from global LRU init */
2124 	spin_lock(&nfs_access_lru_lock);
2125 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2126 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2127 
2128 	spin_lock(&inode->i_lock);
2129 	__nfs_access_zap_cache(NFS_I(inode), &head);
2130 	spin_unlock(&inode->i_lock);
2131 	spin_unlock(&nfs_access_lru_lock);
2132 	nfs_access_free_list(&head);
2133 }
2134 
2135 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2136 {
2137 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2138 	struct nfs_access_entry *entry;
2139 
2140 	while (n != NULL) {
2141 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2142 
2143 		if (cred < entry->cred)
2144 			n = n->rb_left;
2145 		else if (cred > entry->cred)
2146 			n = n->rb_right;
2147 		else
2148 			return entry;
2149 	}
2150 	return NULL;
2151 }
2152 
2153 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2154 {
2155 	struct nfs_inode *nfsi = NFS_I(inode);
2156 	struct nfs_access_entry *cache;
2157 	int err = -ENOENT;
2158 
2159 	spin_lock(&inode->i_lock);
2160 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2161 		goto out_zap;
2162 	cache = nfs_access_search_rbtree(inode, cred);
2163 	if (cache == NULL)
2164 		goto out;
2165 	if (!nfs_have_delegated_attributes(inode) &&
2166 	    !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2167 		goto out_stale;
2168 	res->jiffies = cache->jiffies;
2169 	res->cred = cache->cred;
2170 	res->mask = cache->mask;
2171 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2172 	err = 0;
2173 out:
2174 	spin_unlock(&inode->i_lock);
2175 	return err;
2176 out_stale:
2177 	rb_erase(&cache->rb_node, &nfsi->access_cache);
2178 	list_del(&cache->lru);
2179 	spin_unlock(&inode->i_lock);
2180 	nfs_access_free_entry(cache);
2181 	return -ENOENT;
2182 out_zap:
2183 	spin_unlock(&inode->i_lock);
2184 	nfs_access_zap_cache(inode);
2185 	return -ENOENT;
2186 }
2187 
2188 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2189 {
2190 	struct nfs_inode *nfsi = NFS_I(inode);
2191 	struct rb_root *root_node = &nfsi->access_cache;
2192 	struct rb_node **p = &root_node->rb_node;
2193 	struct rb_node *parent = NULL;
2194 	struct nfs_access_entry *entry;
2195 
2196 	spin_lock(&inode->i_lock);
2197 	while (*p != NULL) {
2198 		parent = *p;
2199 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2200 
2201 		if (set->cred < entry->cred)
2202 			p = &parent->rb_left;
2203 		else if (set->cred > entry->cred)
2204 			p = &parent->rb_right;
2205 		else
2206 			goto found;
2207 	}
2208 	rb_link_node(&set->rb_node, parent, p);
2209 	rb_insert_color(&set->rb_node, root_node);
2210 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2211 	spin_unlock(&inode->i_lock);
2212 	return;
2213 found:
2214 	rb_replace_node(parent, &set->rb_node, root_node);
2215 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2216 	list_del(&entry->lru);
2217 	spin_unlock(&inode->i_lock);
2218 	nfs_access_free_entry(entry);
2219 }
2220 
2221 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2222 {
2223 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2224 	if (cache == NULL)
2225 		return;
2226 	RB_CLEAR_NODE(&cache->rb_node);
2227 	cache->jiffies = set->jiffies;
2228 	cache->cred = get_rpccred(set->cred);
2229 	cache->mask = set->mask;
2230 
2231 	nfs_access_add_rbtree(inode, cache);
2232 
2233 	/* Update accounting */
2234 	smp_mb__before_atomic_inc();
2235 	atomic_long_inc(&nfs_access_nr_entries);
2236 	smp_mb__after_atomic_inc();
2237 
2238 	/* Add inode to global LRU list */
2239 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2240 		spin_lock(&nfs_access_lru_lock);
2241 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2242 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2243 					&nfs_access_lru_list);
2244 		spin_unlock(&nfs_access_lru_lock);
2245 	}
2246 }
2247 
2248 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2249 {
2250 	struct nfs_access_entry cache;
2251 	int status;
2252 
2253 	status = nfs_access_get_cached(inode, cred, &cache);
2254 	if (status == 0)
2255 		goto out;
2256 
2257 	/* Be clever: ask server to check for all possible rights */
2258 	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2259 	cache.cred = cred;
2260 	cache.jiffies = jiffies;
2261 	status = NFS_PROTO(inode)->access(inode, &cache);
2262 	if (status != 0) {
2263 		if (status == -ESTALE) {
2264 			nfs_zap_caches(inode);
2265 			if (!S_ISDIR(inode->i_mode))
2266 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2267 		}
2268 		return status;
2269 	}
2270 	nfs_access_add_cache(inode, &cache);
2271 out:
2272 	if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2273 		return 0;
2274 	return -EACCES;
2275 }
2276 
2277 static int nfs_open_permission_mask(int openflags)
2278 {
2279 	int mask = 0;
2280 
2281 	if ((openflags & O_ACCMODE) != O_WRONLY)
2282 		mask |= MAY_READ;
2283 	if ((openflags & O_ACCMODE) != O_RDONLY)
2284 		mask |= MAY_WRITE;
2285 	if (openflags & __FMODE_EXEC)
2286 		mask |= MAY_EXEC;
2287 	return mask;
2288 }
2289 
2290 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2291 {
2292 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2293 }
2294 
2295 int nfs_permission(struct inode *inode, int mask)
2296 {
2297 	struct rpc_cred *cred;
2298 	int res = 0;
2299 
2300 	if (mask & MAY_NOT_BLOCK)
2301 		return -ECHILD;
2302 
2303 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2304 
2305 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2306 		goto out;
2307 	/* Is this sys_access() ? */
2308 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2309 		goto force_lookup;
2310 
2311 	switch (inode->i_mode & S_IFMT) {
2312 		case S_IFLNK:
2313 			goto out;
2314 		case S_IFREG:
2315 			/* NFSv4 has atomic_open... */
2316 			if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2317 					&& (mask & MAY_OPEN)
2318 					&& !(mask & MAY_EXEC))
2319 				goto out;
2320 			break;
2321 		case S_IFDIR:
2322 			/*
2323 			 * Optimize away all write operations, since the server
2324 			 * will check permissions when we perform the op.
2325 			 */
2326 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2327 				goto out;
2328 	}
2329 
2330 force_lookup:
2331 	if (!NFS_PROTO(inode)->access)
2332 		goto out_notsup;
2333 
2334 	cred = rpc_lookup_cred();
2335 	if (!IS_ERR(cred)) {
2336 		res = nfs_do_access(inode, cred, mask);
2337 		put_rpccred(cred);
2338 	} else
2339 		res = PTR_ERR(cred);
2340 out:
2341 	if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2342 		res = -EACCES;
2343 
2344 	dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2345 		inode->i_sb->s_id, inode->i_ino, mask, res);
2346 	return res;
2347 out_notsup:
2348 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2349 	if (res == 0)
2350 		res = generic_permission(inode, mask);
2351 	goto out;
2352 }
2353 
2354 /*
2355  * Local variables:
2356  *  version-control: t
2357  *  kept-new-versions: 5
2358  * End:
2359  */
2360