xref: /linux/fs/nfs/dir.c (revision 26b0d14106954ae46d2f4f7eec3481828a210f7d)
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 bool nfs_use_readdirplus(struct inode *dir, struct file *filp)
479 {
480 	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
481 		return false;
482 	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
483 		return true;
484 	if (filp->f_pos == 0)
485 		return true;
486 	return false;
487 }
488 
489 /*
490  * This function is called by the lookup code to request the use of
491  * readdirplus to accelerate any future lookups in the same
492  * directory.
493  */
494 static
495 void nfs_advise_use_readdirplus(struct inode *dir)
496 {
497 	set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
498 }
499 
500 static
501 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
502 {
503 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
504 	struct dentry *dentry;
505 	struct dentry *alias;
506 	struct inode *dir = parent->d_inode;
507 	struct inode *inode;
508 
509 	if (filename.name[0] == '.') {
510 		if (filename.len == 1)
511 			return;
512 		if (filename.len == 2 && filename.name[1] == '.')
513 			return;
514 	}
515 	filename.hash = full_name_hash(filename.name, filename.len);
516 
517 	dentry = d_lookup(parent, &filename);
518 	if (dentry != NULL) {
519 		if (nfs_same_file(dentry, entry)) {
520 			nfs_refresh_inode(dentry->d_inode, entry->fattr);
521 			goto out;
522 		} else {
523 			d_drop(dentry);
524 			dput(dentry);
525 		}
526 	}
527 
528 	dentry = d_alloc(parent, &filename);
529 	if (dentry == NULL)
530 		return;
531 
532 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
533 	if (IS_ERR(inode))
534 		goto out;
535 
536 	alias = d_materialise_unique(dentry, inode);
537 	if (IS_ERR(alias))
538 		goto out;
539 	else if (alias) {
540 		nfs_set_verifier(alias, nfs_save_change_attribute(dir));
541 		dput(alias);
542 	} else
543 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
544 
545 out:
546 	dput(dentry);
547 }
548 
549 /* Perform conversion from xdr to cache array */
550 static
551 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
552 				struct page **xdr_pages, struct page *page, unsigned int buflen)
553 {
554 	struct xdr_stream stream;
555 	struct xdr_buf buf;
556 	struct page *scratch;
557 	struct nfs_cache_array *array;
558 	unsigned int count = 0;
559 	int status;
560 
561 	scratch = alloc_page(GFP_KERNEL);
562 	if (scratch == NULL)
563 		return -ENOMEM;
564 
565 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
566 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
567 
568 	do {
569 		status = xdr_decode(desc, entry, &stream);
570 		if (status != 0) {
571 			if (status == -EAGAIN)
572 				status = 0;
573 			break;
574 		}
575 
576 		count++;
577 
578 		if (desc->plus != 0)
579 			nfs_prime_dcache(desc->file->f_path.dentry, entry);
580 
581 		status = nfs_readdir_add_to_array(entry, page);
582 		if (status != 0)
583 			break;
584 	} while (!entry->eof);
585 
586 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
587 		array = nfs_readdir_get_array(page);
588 		if (!IS_ERR(array)) {
589 			array->eof_index = array->size;
590 			status = 0;
591 			nfs_readdir_release_array(page);
592 		} else
593 			status = PTR_ERR(array);
594 	}
595 
596 	put_page(scratch);
597 	return status;
598 }
599 
600 static
601 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
602 {
603 	unsigned int i;
604 	for (i = 0; i < npages; i++)
605 		put_page(pages[i]);
606 }
607 
608 static
609 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
610 		unsigned int npages)
611 {
612 	nfs_readdir_free_pagearray(pages, npages);
613 }
614 
615 /*
616  * nfs_readdir_large_page will allocate pages that must be freed with a call
617  * to nfs_readdir_free_large_page
618  */
619 static
620 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
621 {
622 	unsigned int i;
623 
624 	for (i = 0; i < npages; i++) {
625 		struct page *page = alloc_page(GFP_KERNEL);
626 		if (page == NULL)
627 			goto out_freepages;
628 		pages[i] = page;
629 	}
630 	return 0;
631 
632 out_freepages:
633 	nfs_readdir_free_pagearray(pages, i);
634 	return -ENOMEM;
635 }
636 
637 static
638 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
639 {
640 	struct page *pages[NFS_MAX_READDIR_PAGES];
641 	void *pages_ptr = NULL;
642 	struct nfs_entry entry;
643 	struct file	*file = desc->file;
644 	struct nfs_cache_array *array;
645 	int status = -ENOMEM;
646 	unsigned int array_size = ARRAY_SIZE(pages);
647 
648 	entry.prev_cookie = 0;
649 	entry.cookie = desc->last_cookie;
650 	entry.eof = 0;
651 	entry.fh = nfs_alloc_fhandle();
652 	entry.fattr = nfs_alloc_fattr();
653 	entry.server = NFS_SERVER(inode);
654 	if (entry.fh == NULL || entry.fattr == NULL)
655 		goto out;
656 
657 	array = nfs_readdir_get_array(page);
658 	if (IS_ERR(array)) {
659 		status = PTR_ERR(array);
660 		goto out;
661 	}
662 	memset(array, 0, sizeof(struct nfs_cache_array));
663 	array->eof_index = -1;
664 
665 	status = nfs_readdir_large_page(pages, array_size);
666 	if (status < 0)
667 		goto out_release_array;
668 	do {
669 		unsigned int pglen;
670 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
671 
672 		if (status < 0)
673 			break;
674 		pglen = status;
675 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
676 		if (status < 0) {
677 			if (status == -ENOSPC)
678 				status = 0;
679 			break;
680 		}
681 	} while (array->eof_index < 0);
682 
683 	nfs_readdir_free_large_page(pages_ptr, pages, array_size);
684 out_release_array:
685 	nfs_readdir_release_array(page);
686 out:
687 	nfs_free_fattr(entry.fattr);
688 	nfs_free_fhandle(entry.fh);
689 	return status;
690 }
691 
692 /*
693  * Now we cache directories properly, by converting xdr information
694  * to an array that can be used for lookups later.  This results in
695  * fewer cache pages, since we can store more information on each page.
696  * We only need to convert from xdr once so future lookups are much simpler
697  */
698 static
699 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
700 {
701 	struct inode	*inode = desc->file->f_path.dentry->d_inode;
702 	int ret;
703 
704 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
705 	if (ret < 0)
706 		goto error;
707 	SetPageUptodate(page);
708 
709 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
710 		/* Should never happen */
711 		nfs_zap_mapping(inode, inode->i_mapping);
712 	}
713 	unlock_page(page);
714 	return 0;
715  error:
716 	unlock_page(page);
717 	return ret;
718 }
719 
720 static
721 void cache_page_release(nfs_readdir_descriptor_t *desc)
722 {
723 	if (!desc->page->mapping)
724 		nfs_readdir_clear_array(desc->page);
725 	page_cache_release(desc->page);
726 	desc->page = NULL;
727 }
728 
729 static
730 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
731 {
732 	return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
733 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
734 }
735 
736 /*
737  * Returns 0 if desc->dir_cookie was found on page desc->page_index
738  */
739 static
740 int find_cache_page(nfs_readdir_descriptor_t *desc)
741 {
742 	int res;
743 
744 	desc->page = get_cache_page(desc);
745 	if (IS_ERR(desc->page))
746 		return PTR_ERR(desc->page);
747 
748 	res = nfs_readdir_search_array(desc);
749 	if (res != 0)
750 		cache_page_release(desc);
751 	return res;
752 }
753 
754 /* Search for desc->dir_cookie from the beginning of the page cache */
755 static inline
756 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
757 {
758 	int res;
759 
760 	if (desc->page_index == 0) {
761 		desc->current_index = 0;
762 		desc->last_cookie = 0;
763 	}
764 	do {
765 		res = find_cache_page(desc);
766 	} while (res == -EAGAIN);
767 	return res;
768 }
769 
770 /*
771  * Once we've found the start of the dirent within a page: fill 'er up...
772  */
773 static
774 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
775 		   filldir_t filldir)
776 {
777 	struct file	*file = desc->file;
778 	int i = 0;
779 	int res = 0;
780 	struct nfs_cache_array *array = NULL;
781 	struct nfs_open_dir_context *ctx = file->private_data;
782 
783 	array = nfs_readdir_get_array(desc->page);
784 	if (IS_ERR(array)) {
785 		res = PTR_ERR(array);
786 		goto out;
787 	}
788 
789 	for (i = desc->cache_entry_index; i < array->size; i++) {
790 		struct nfs_cache_array_entry *ent;
791 
792 		ent = &array->array[i];
793 		if (filldir(dirent, ent->string.name, ent->string.len,
794 		    file->f_pos, nfs_compat_user_ino64(ent->ino),
795 		    ent->d_type) < 0) {
796 			desc->eof = 1;
797 			break;
798 		}
799 		file->f_pos++;
800 		if (i < (array->size-1))
801 			*desc->dir_cookie = array->array[i+1].cookie;
802 		else
803 			*desc->dir_cookie = array->last_cookie;
804 		if (ctx->duped != 0)
805 			ctx->duped = 1;
806 	}
807 	if (array->eof_index >= 0)
808 		desc->eof = 1;
809 
810 	nfs_readdir_release_array(desc->page);
811 out:
812 	cache_page_release(desc);
813 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
814 			(unsigned long long)*desc->dir_cookie, res);
815 	return res;
816 }
817 
818 /*
819  * If we cannot find a cookie in our cache, we suspect that this is
820  * because it points to a deleted file, so we ask the server to return
821  * whatever it thinks is the next entry. We then feed this to filldir.
822  * If all goes well, we should then be able to find our way round the
823  * cache on the next call to readdir_search_pagecache();
824  *
825  * NOTE: we cannot add the anonymous page to the pagecache because
826  *	 the data it contains might not be page aligned. Besides,
827  *	 we should already have a complete representation of the
828  *	 directory in the page cache by the time we get here.
829  */
830 static inline
831 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
832 		     filldir_t filldir)
833 {
834 	struct page	*page = NULL;
835 	int		status;
836 	struct inode *inode = desc->file->f_path.dentry->d_inode;
837 	struct nfs_open_dir_context *ctx = desc->file->private_data;
838 
839 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
840 			(unsigned long long)*desc->dir_cookie);
841 
842 	page = alloc_page(GFP_HIGHUSER);
843 	if (!page) {
844 		status = -ENOMEM;
845 		goto out;
846 	}
847 
848 	desc->page_index = 0;
849 	desc->last_cookie = *desc->dir_cookie;
850 	desc->page = page;
851 	ctx->duped = 0;
852 
853 	status = nfs_readdir_xdr_to_array(desc, page, inode);
854 	if (status < 0)
855 		goto out_release;
856 
857 	status = nfs_do_filldir(desc, dirent, filldir);
858 
859  out:
860 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
861 			__func__, status);
862 	return status;
863  out_release:
864 	cache_page_release(desc);
865 	goto out;
866 }
867 
868 /* The file offset position represents the dirent entry number.  A
869    last cookie cache takes care of the common case of reading the
870    whole directory.
871  */
872 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
873 {
874 	struct dentry	*dentry = filp->f_path.dentry;
875 	struct inode	*inode = dentry->d_inode;
876 	nfs_readdir_descriptor_t my_desc,
877 			*desc = &my_desc;
878 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
879 	int res;
880 
881 	dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
882 			dentry->d_parent->d_name.name, dentry->d_name.name,
883 			(long long)filp->f_pos);
884 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
885 
886 	/*
887 	 * filp->f_pos points to the dirent entry number.
888 	 * *desc->dir_cookie has the cookie for the next entry. We have
889 	 * to either find the entry with the appropriate number or
890 	 * revalidate the cookie.
891 	 */
892 	memset(desc, 0, sizeof(*desc));
893 
894 	desc->file = filp;
895 	desc->dir_cookie = &dir_ctx->dir_cookie;
896 	desc->decode = NFS_PROTO(inode)->decode_dirent;
897 	desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0;
898 
899 	nfs_block_sillyrename(dentry);
900 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
901 	if (res < 0)
902 		goto out;
903 
904 	do {
905 		res = readdir_search_pagecache(desc);
906 
907 		if (res == -EBADCOOKIE) {
908 			res = 0;
909 			/* This means either end of directory */
910 			if (*desc->dir_cookie && desc->eof == 0) {
911 				/* Or that the server has 'lost' a cookie */
912 				res = uncached_readdir(desc, dirent, filldir);
913 				if (res == 0)
914 					continue;
915 			}
916 			break;
917 		}
918 		if (res == -ETOOSMALL && desc->plus) {
919 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
920 			nfs_zap_caches(inode);
921 			desc->page_index = 0;
922 			desc->plus = 0;
923 			desc->eof = 0;
924 			continue;
925 		}
926 		if (res < 0)
927 			break;
928 
929 		res = nfs_do_filldir(desc, dirent, filldir);
930 		if (res < 0)
931 			break;
932 	} while (!desc->eof);
933 out:
934 	nfs_unblock_sillyrename(dentry);
935 	if (res > 0)
936 		res = 0;
937 	dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
938 			dentry->d_parent->d_name.name, dentry->d_name.name,
939 			res);
940 	return res;
941 }
942 
943 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
944 {
945 	struct dentry *dentry = filp->f_path.dentry;
946 	struct inode *inode = dentry->d_inode;
947 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
948 
949 	dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
950 			dentry->d_parent->d_name.name,
951 			dentry->d_name.name,
952 			offset, origin);
953 
954 	mutex_lock(&inode->i_mutex);
955 	switch (origin) {
956 		case 1:
957 			offset += filp->f_pos;
958 		case 0:
959 			if (offset >= 0)
960 				break;
961 		default:
962 			offset = -EINVAL;
963 			goto out;
964 	}
965 	if (offset != filp->f_pos) {
966 		filp->f_pos = offset;
967 		dir_ctx->dir_cookie = 0;
968 		dir_ctx->duped = 0;
969 	}
970 out:
971 	mutex_unlock(&inode->i_mutex);
972 	return offset;
973 }
974 
975 /*
976  * All directory operations under NFS are synchronous, so fsync()
977  * is a dummy operation.
978  */
979 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
980 			 int datasync)
981 {
982 	struct dentry *dentry = filp->f_path.dentry;
983 	struct inode *inode = dentry->d_inode;
984 
985 	dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
986 			dentry->d_parent->d_name.name, dentry->d_name.name,
987 			datasync);
988 
989 	mutex_lock(&inode->i_mutex);
990 	nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
991 	mutex_unlock(&inode->i_mutex);
992 	return 0;
993 }
994 
995 /**
996  * nfs_force_lookup_revalidate - Mark the directory as having changed
997  * @dir - pointer to directory inode
998  *
999  * This forces the revalidation code in nfs_lookup_revalidate() to do a
1000  * full lookup on all child dentries of 'dir' whenever a change occurs
1001  * on the server that might have invalidated our dcache.
1002  *
1003  * The caller should be holding dir->i_lock
1004  */
1005 void nfs_force_lookup_revalidate(struct inode *dir)
1006 {
1007 	NFS_I(dir)->cache_change_attribute++;
1008 }
1009 
1010 /*
1011  * A check for whether or not the parent directory has changed.
1012  * In the case it has, we assume that the dentries are untrustworthy
1013  * and may need to be looked up again.
1014  */
1015 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
1016 {
1017 	if (IS_ROOT(dentry))
1018 		return 1;
1019 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1020 		return 0;
1021 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1022 		return 0;
1023 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1024 	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1025 		return 0;
1026 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1027 		return 0;
1028 	return 1;
1029 }
1030 
1031 /*
1032  * Return the intent data that applies to this particular path component
1033  *
1034  * Note that the current set of intents only apply to the very last
1035  * component of the path and none of them is set before that last
1036  * component.
1037  */
1038 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1039 						unsigned int mask)
1040 {
1041 	return nd->flags & mask;
1042 }
1043 
1044 /*
1045  * Use intent information to check whether or not we're going to do
1046  * an O_EXCL create using this path component.
1047  */
1048 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1049 {
1050 	if (NFS_PROTO(dir)->version == 2)
1051 		return 0;
1052 	return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1053 }
1054 
1055 /*
1056  * Inode and filehandle revalidation for lookups.
1057  *
1058  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1059  * or if the intent information indicates that we're about to open this
1060  * particular file and the "nocto" mount flag is not set.
1061  *
1062  */
1063 static inline
1064 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1065 {
1066 	struct nfs_server *server = NFS_SERVER(inode);
1067 
1068 	if (IS_AUTOMOUNT(inode))
1069 		return 0;
1070 	if (nd != NULL) {
1071 		/* VFS wants an on-the-wire revalidation */
1072 		if (nd->flags & LOOKUP_REVAL)
1073 			goto out_force;
1074 		/* This is an open(2) */
1075 		if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1076 				!(server->flags & NFS_MOUNT_NOCTO) &&
1077 				(S_ISREG(inode->i_mode) ||
1078 				 S_ISDIR(inode->i_mode)))
1079 			goto out_force;
1080 		return 0;
1081 	}
1082 	return nfs_revalidate_inode(server, inode);
1083 out_force:
1084 	return __nfs_revalidate_inode(server, inode);
1085 }
1086 
1087 /*
1088  * We judge how long we want to trust negative
1089  * dentries by looking at the parent inode mtime.
1090  *
1091  * If parent mtime has changed, we revalidate, else we wait for a
1092  * period corresponding to the parent's attribute cache timeout value.
1093  */
1094 static inline
1095 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1096 		       struct nameidata *nd)
1097 {
1098 	/* Don't revalidate a negative dentry if we're creating a new file */
1099 	if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1100 		return 0;
1101 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1102 		return 1;
1103 	return !nfs_check_verifier(dir, dentry);
1104 }
1105 
1106 /*
1107  * This is called every time the dcache has a lookup hit,
1108  * and we should check whether we can really trust that
1109  * lookup.
1110  *
1111  * NOTE! The hit can be a negative hit too, don't assume
1112  * we have an inode!
1113  *
1114  * If the parent directory is seen to have changed, we throw out the
1115  * cached dentry and do a new lookup.
1116  */
1117 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1118 {
1119 	struct inode *dir;
1120 	struct inode *inode;
1121 	struct dentry *parent;
1122 	struct nfs_fh *fhandle = NULL;
1123 	struct nfs_fattr *fattr = NULL;
1124 	int error;
1125 
1126 	if (nd->flags & LOOKUP_RCU)
1127 		return -ECHILD;
1128 
1129 	parent = dget_parent(dentry);
1130 	dir = parent->d_inode;
1131 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1132 	inode = dentry->d_inode;
1133 
1134 	if (!inode) {
1135 		if (nfs_neg_need_reval(dir, dentry, nd))
1136 			goto out_bad;
1137 		goto out_valid_noent;
1138 	}
1139 
1140 	if (is_bad_inode(inode)) {
1141 		dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1142 				__func__, dentry->d_parent->d_name.name,
1143 				dentry->d_name.name);
1144 		goto out_bad;
1145 	}
1146 
1147 	if (nfs_have_delegation(inode, FMODE_READ))
1148 		goto out_set_verifier;
1149 
1150 	/* Force a full look up iff the parent directory has changed */
1151 	if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1152 		if (nfs_lookup_verify_inode(inode, nd))
1153 			goto out_zap_parent;
1154 		goto out_valid;
1155 	}
1156 
1157 	if (NFS_STALE(inode))
1158 		goto out_bad;
1159 
1160 	error = -ENOMEM;
1161 	fhandle = nfs_alloc_fhandle();
1162 	fattr = nfs_alloc_fattr();
1163 	if (fhandle == NULL || fattr == NULL)
1164 		goto out_error;
1165 
1166 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1167 	if (error)
1168 		goto out_bad;
1169 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1170 		goto out_bad;
1171 	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1172 		goto out_bad;
1173 
1174 	nfs_free_fattr(fattr);
1175 	nfs_free_fhandle(fhandle);
1176 out_set_verifier:
1177 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1178  out_valid:
1179 	/* Success: notify readdir to use READDIRPLUS */
1180 	nfs_advise_use_readdirplus(dir);
1181  out_valid_noent:
1182 	dput(parent);
1183 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1184 			__func__, dentry->d_parent->d_name.name,
1185 			dentry->d_name.name);
1186 	return 1;
1187 out_zap_parent:
1188 	nfs_zap_caches(dir);
1189  out_bad:
1190 	nfs_mark_for_revalidate(dir);
1191 	if (inode && S_ISDIR(inode->i_mode)) {
1192 		/* Purge readdir caches. */
1193 		nfs_zap_caches(inode);
1194 		/* If we have submounts, don't unhash ! */
1195 		if (have_submounts(dentry))
1196 			goto out_valid;
1197 		if (dentry->d_flags & DCACHE_DISCONNECTED)
1198 			goto out_valid;
1199 		shrink_dcache_parent(dentry);
1200 	}
1201 	d_drop(dentry);
1202 	nfs_free_fattr(fattr);
1203 	nfs_free_fhandle(fhandle);
1204 	dput(parent);
1205 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1206 			__func__, dentry->d_parent->d_name.name,
1207 			dentry->d_name.name);
1208 	return 0;
1209 out_error:
1210 	nfs_free_fattr(fattr);
1211 	nfs_free_fhandle(fhandle);
1212 	dput(parent);
1213 	dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1214 			__func__, dentry->d_parent->d_name.name,
1215 			dentry->d_name.name, error);
1216 	return error;
1217 }
1218 
1219 /*
1220  * This is called from dput() when d_count is going to 0.
1221  */
1222 static int nfs_dentry_delete(const struct dentry *dentry)
1223 {
1224 	dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1225 		dentry->d_parent->d_name.name, dentry->d_name.name,
1226 		dentry->d_flags);
1227 
1228 	/* Unhash any dentry with a stale inode */
1229 	if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1230 		return 1;
1231 
1232 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1233 		/* Unhash it, so that ->d_iput() would be called */
1234 		return 1;
1235 	}
1236 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1237 		/* Unhash it, so that ancestors of killed async unlink
1238 		 * files will be cleaned up during umount */
1239 		return 1;
1240 	}
1241 	return 0;
1242 
1243 }
1244 
1245 static void nfs_drop_nlink(struct inode *inode)
1246 {
1247 	spin_lock(&inode->i_lock);
1248 	if (inode->i_nlink > 0)
1249 		drop_nlink(inode);
1250 	spin_unlock(&inode->i_lock);
1251 }
1252 
1253 /*
1254  * Called when the dentry loses inode.
1255  * We use it to clean up silly-renamed files.
1256  */
1257 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1258 {
1259 	if (S_ISDIR(inode->i_mode))
1260 		/* drop any readdir cache as it could easily be old */
1261 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1262 
1263 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1264 		drop_nlink(inode);
1265 		nfs_complete_unlink(dentry, inode);
1266 	}
1267 	iput(inode);
1268 }
1269 
1270 static void nfs_d_release(struct dentry *dentry)
1271 {
1272 	/* free cached devname value, if it survived that far */
1273 	if (unlikely(dentry->d_fsdata)) {
1274 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1275 			WARN_ON(1);
1276 		else
1277 			kfree(dentry->d_fsdata);
1278 	}
1279 }
1280 
1281 const struct dentry_operations nfs_dentry_operations = {
1282 	.d_revalidate	= nfs_lookup_revalidate,
1283 	.d_delete	= nfs_dentry_delete,
1284 	.d_iput		= nfs_dentry_iput,
1285 	.d_automount	= nfs_d_automount,
1286 	.d_release	= nfs_d_release,
1287 };
1288 
1289 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1290 {
1291 	struct dentry *res;
1292 	struct dentry *parent;
1293 	struct inode *inode = NULL;
1294 	struct nfs_fh *fhandle = NULL;
1295 	struct nfs_fattr *fattr = NULL;
1296 	int error;
1297 
1298 	dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1299 		dentry->d_parent->d_name.name, dentry->d_name.name);
1300 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1301 
1302 	res = ERR_PTR(-ENAMETOOLONG);
1303 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1304 		goto out;
1305 
1306 	/*
1307 	 * If we're doing an exclusive create, optimize away the lookup
1308 	 * but don't hash the dentry.
1309 	 */
1310 	if (nfs_is_exclusive_create(dir, nd)) {
1311 		d_instantiate(dentry, NULL);
1312 		res = NULL;
1313 		goto out;
1314 	}
1315 
1316 	res = ERR_PTR(-ENOMEM);
1317 	fhandle = nfs_alloc_fhandle();
1318 	fattr = nfs_alloc_fattr();
1319 	if (fhandle == NULL || fattr == NULL)
1320 		goto out;
1321 
1322 	parent = dentry->d_parent;
1323 	/* Protect against concurrent sillydeletes */
1324 	nfs_block_sillyrename(parent);
1325 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1326 	if (error == -ENOENT)
1327 		goto no_entry;
1328 	if (error < 0) {
1329 		res = ERR_PTR(error);
1330 		goto out_unblock_sillyrename;
1331 	}
1332 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1333 	res = ERR_CAST(inode);
1334 	if (IS_ERR(res))
1335 		goto out_unblock_sillyrename;
1336 
1337 	/* Success: notify readdir to use READDIRPLUS */
1338 	nfs_advise_use_readdirplus(dir);
1339 
1340 no_entry:
1341 	res = d_materialise_unique(dentry, inode);
1342 	if (res != NULL) {
1343 		if (IS_ERR(res))
1344 			goto out_unblock_sillyrename;
1345 		dentry = res;
1346 	}
1347 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1348 out_unblock_sillyrename:
1349 	nfs_unblock_sillyrename(parent);
1350 out:
1351 	nfs_free_fattr(fattr);
1352 	nfs_free_fhandle(fhandle);
1353 	return res;
1354 }
1355 
1356 #ifdef CONFIG_NFS_V4
1357 static int nfs4_lookup_revalidate(struct dentry *, struct nameidata *);
1358 
1359 const struct dentry_operations nfs4_dentry_operations = {
1360 	.d_revalidate	= nfs4_lookup_revalidate,
1361 	.d_delete	= nfs_dentry_delete,
1362 	.d_iput		= nfs_dentry_iput,
1363 	.d_automount	= nfs_d_automount,
1364 	.d_release	= nfs_d_release,
1365 };
1366 
1367 /*
1368  * Use intent information to determine whether we need to substitute
1369  * the NFSv4-style stateful OPEN for the LOOKUP call
1370  */
1371 static int is_atomic_open(struct nameidata *nd)
1372 {
1373 	if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1374 		return 0;
1375 	/* NFS does not (yet) have a stateful open for directories */
1376 	if (nd->flags & LOOKUP_DIRECTORY)
1377 		return 0;
1378 	/* Are we trying to write to a read only partition? */
1379 	if (__mnt_is_readonly(nd->path.mnt) &&
1380 	    (nd->intent.open.flags & (O_CREAT|O_TRUNC|O_ACCMODE)))
1381 		return 0;
1382 	return 1;
1383 }
1384 
1385 static fmode_t flags_to_mode(int flags)
1386 {
1387 	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1388 	if ((flags & O_ACCMODE) != O_WRONLY)
1389 		res |= FMODE_READ;
1390 	if ((flags & O_ACCMODE) != O_RDONLY)
1391 		res |= FMODE_WRITE;
1392 	return res;
1393 }
1394 
1395 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1396 {
1397 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1398 }
1399 
1400 static int do_open(struct inode *inode, struct file *filp)
1401 {
1402 	nfs_fscache_set_inode_cookie(inode, filp);
1403 	return 0;
1404 }
1405 
1406 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1407 {
1408 	struct file *filp;
1409 	int ret = 0;
1410 
1411 	/* If the open_intent is for execute, we have an extra check to make */
1412 	if (ctx->mode & FMODE_EXEC) {
1413 		ret = nfs_may_open(ctx->dentry->d_inode,
1414 				ctx->cred,
1415 				nd->intent.open.flags);
1416 		if (ret < 0)
1417 			goto out;
1418 	}
1419 	filp = lookup_instantiate_filp(nd, ctx->dentry, do_open);
1420 	if (IS_ERR(filp))
1421 		ret = PTR_ERR(filp);
1422 	else
1423 		nfs_file_set_open_context(filp, ctx);
1424 out:
1425 	put_nfs_open_context(ctx);
1426 	return ret;
1427 }
1428 
1429 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1430 {
1431 	struct nfs_open_context *ctx;
1432 	struct iattr attr;
1433 	struct dentry *res = NULL;
1434 	struct inode *inode;
1435 	int open_flags;
1436 	int err;
1437 
1438 	dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1439 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1440 
1441 	/* Check that we are indeed trying to open this file */
1442 	if (!is_atomic_open(nd))
1443 		goto no_open;
1444 
1445 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1446 		res = ERR_PTR(-ENAMETOOLONG);
1447 		goto out;
1448 	}
1449 
1450 	/* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1451 	 * the dentry. */
1452 	if (nd->flags & LOOKUP_EXCL) {
1453 		d_instantiate(dentry, NULL);
1454 		goto out;
1455 	}
1456 
1457 	open_flags = nd->intent.open.flags;
1458 	attr.ia_valid = ATTR_OPEN;
1459 
1460 	ctx = create_nfs_open_context(dentry, open_flags);
1461 	res = ERR_CAST(ctx);
1462 	if (IS_ERR(ctx))
1463 		goto out;
1464 
1465 	if (nd->flags & LOOKUP_CREATE) {
1466 		attr.ia_mode = nd->intent.open.create_mode;
1467 		attr.ia_valid |= ATTR_MODE;
1468 		attr.ia_mode &= ~current_umask();
1469 	} else
1470 		open_flags &= ~(O_EXCL | O_CREAT);
1471 
1472 	if (open_flags & O_TRUNC) {
1473 		attr.ia_valid |= ATTR_SIZE;
1474 		attr.ia_size = 0;
1475 	}
1476 
1477 	/* Open the file on the server */
1478 	nfs_block_sillyrename(dentry->d_parent);
1479 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1480 	if (IS_ERR(inode)) {
1481 		nfs_unblock_sillyrename(dentry->d_parent);
1482 		put_nfs_open_context(ctx);
1483 		switch (PTR_ERR(inode)) {
1484 			/* Make a negative dentry */
1485 			case -ENOENT:
1486 				d_add(dentry, NULL);
1487 				res = NULL;
1488 				goto out;
1489 			/* This turned out not to be a regular file */
1490 			case -EISDIR:
1491 			case -ENOTDIR:
1492 				goto no_open;
1493 			case -ELOOP:
1494 				if (!(nd->intent.open.flags & O_NOFOLLOW))
1495 					goto no_open;
1496 			/* case -EINVAL: */
1497 			default:
1498 				res = ERR_CAST(inode);
1499 				goto out;
1500 		}
1501 	}
1502 	res = d_add_unique(dentry, inode);
1503 	nfs_unblock_sillyrename(dentry->d_parent);
1504 	if (res != NULL) {
1505 		dput(ctx->dentry);
1506 		ctx->dentry = dget(res);
1507 		dentry = res;
1508 	}
1509 	err = nfs_intent_set_file(nd, ctx);
1510 	if (err < 0) {
1511 		if (res != NULL)
1512 			dput(res);
1513 		return ERR_PTR(err);
1514 	}
1515 out:
1516 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1517 	return res;
1518 no_open:
1519 	return nfs_lookup(dir, dentry, nd);
1520 }
1521 
1522 static int nfs4_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1523 {
1524 	struct dentry *parent = NULL;
1525 	struct inode *inode;
1526 	struct inode *dir;
1527 	int openflags, ret = 0;
1528 
1529 	if (nd->flags & LOOKUP_RCU)
1530 		return -ECHILD;
1531 
1532 	inode = dentry->d_inode;
1533 	if (!is_atomic_open(nd) || d_mountpoint(dentry))
1534 		goto no_open;
1535 
1536 	parent = dget_parent(dentry);
1537 	dir = parent->d_inode;
1538 
1539 	/* We can't create new files in nfs_open_revalidate(), so we
1540 	 * optimize away revalidation of negative dentries.
1541 	 */
1542 	if (inode == NULL) {
1543 		if (!nfs_neg_need_reval(dir, dentry, nd))
1544 			ret = 1;
1545 		goto out;
1546 	}
1547 
1548 	/* NFS only supports OPEN on regular files */
1549 	if (!S_ISREG(inode->i_mode))
1550 		goto no_open_dput;
1551 	openflags = nd->intent.open.flags;
1552 	/* We cannot do exclusive creation on a positive dentry */
1553 	if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1554 		goto no_open_dput;
1555 
1556 	/* Let f_op->open() actually open (and revalidate) the file */
1557 	ret = 1;
1558 
1559 out:
1560 	dput(parent);
1561 	return ret;
1562 
1563 no_open_dput:
1564 	dput(parent);
1565 no_open:
1566 	return nfs_lookup_revalidate(dentry, nd);
1567 }
1568 
1569 static int nfs_open_create(struct inode *dir, struct dentry *dentry,
1570 		umode_t mode, struct nameidata *nd)
1571 {
1572 	struct nfs_open_context *ctx = NULL;
1573 	struct iattr attr;
1574 	int error;
1575 	int open_flags = O_CREAT|O_EXCL;
1576 
1577 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1578 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1579 
1580 	attr.ia_mode = mode;
1581 	attr.ia_valid = ATTR_MODE;
1582 
1583 	if (nd)
1584 		open_flags = nd->intent.open.flags;
1585 
1586 	ctx = create_nfs_open_context(dentry, open_flags);
1587 	error = PTR_ERR(ctx);
1588 	if (IS_ERR(ctx))
1589 		goto out_err_drop;
1590 
1591 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1592 	if (error != 0)
1593 		goto out_put_ctx;
1594 	if (nd) {
1595 		error = nfs_intent_set_file(nd, ctx);
1596 		if (error < 0)
1597 			goto out_err;
1598 	} else {
1599 		put_nfs_open_context(ctx);
1600 	}
1601 	return 0;
1602 out_put_ctx:
1603 	put_nfs_open_context(ctx);
1604 out_err_drop:
1605 	d_drop(dentry);
1606 out_err:
1607 	return error;
1608 }
1609 
1610 #endif /* CONFIG_NFSV4 */
1611 
1612 /*
1613  * Code common to create, mkdir, and mknod.
1614  */
1615 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1616 				struct nfs_fattr *fattr)
1617 {
1618 	struct dentry *parent = dget_parent(dentry);
1619 	struct inode *dir = parent->d_inode;
1620 	struct inode *inode;
1621 	int error = -EACCES;
1622 
1623 	d_drop(dentry);
1624 
1625 	/* We may have been initialized further down */
1626 	if (dentry->d_inode)
1627 		goto out;
1628 	if (fhandle->size == 0) {
1629 		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1630 		if (error)
1631 			goto out_error;
1632 	}
1633 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1634 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1635 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1636 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1637 		if (error < 0)
1638 			goto out_error;
1639 	}
1640 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1641 	error = PTR_ERR(inode);
1642 	if (IS_ERR(inode))
1643 		goto out_error;
1644 	d_add(dentry, inode);
1645 out:
1646 	dput(parent);
1647 	return 0;
1648 out_error:
1649 	nfs_mark_for_revalidate(dir);
1650 	dput(parent);
1651 	return error;
1652 }
1653 
1654 /*
1655  * Following a failed create operation, we drop the dentry rather
1656  * than retain a negative dentry. This avoids a problem in the event
1657  * that the operation succeeded on the server, but an error in the
1658  * reply path made it appear to have failed.
1659  */
1660 static int nfs_create(struct inode *dir, struct dentry *dentry,
1661 		umode_t mode, struct nameidata *nd)
1662 {
1663 	struct iattr attr;
1664 	int error;
1665 	int open_flags = O_CREAT|O_EXCL;
1666 
1667 	dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1668 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1669 
1670 	attr.ia_mode = mode;
1671 	attr.ia_valid = ATTR_MODE;
1672 
1673 	if (nd)
1674 		open_flags = nd->intent.open.flags;
1675 
1676 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1677 	if (error != 0)
1678 		goto out_err;
1679 	return 0;
1680 out_err:
1681 	d_drop(dentry);
1682 	return error;
1683 }
1684 
1685 /*
1686  * See comments for nfs_proc_create regarding failed operations.
1687  */
1688 static int
1689 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1690 {
1691 	struct iattr attr;
1692 	int status;
1693 
1694 	dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1695 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1696 
1697 	if (!new_valid_dev(rdev))
1698 		return -EINVAL;
1699 
1700 	attr.ia_mode = mode;
1701 	attr.ia_valid = ATTR_MODE;
1702 
1703 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1704 	if (status != 0)
1705 		goto out_err;
1706 	return 0;
1707 out_err:
1708 	d_drop(dentry);
1709 	return status;
1710 }
1711 
1712 /*
1713  * See comments for nfs_proc_create regarding failed operations.
1714  */
1715 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1716 {
1717 	struct iattr attr;
1718 	int error;
1719 
1720 	dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1721 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1722 
1723 	attr.ia_valid = ATTR_MODE;
1724 	attr.ia_mode = mode | S_IFDIR;
1725 
1726 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1727 	if (error != 0)
1728 		goto out_err;
1729 	return 0;
1730 out_err:
1731 	d_drop(dentry);
1732 	return error;
1733 }
1734 
1735 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1736 {
1737 	if (dentry->d_inode != NULL && !d_unhashed(dentry))
1738 		d_delete(dentry);
1739 }
1740 
1741 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1742 {
1743 	int error;
1744 
1745 	dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1746 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1747 
1748 	error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1749 	/* Ensure the VFS deletes this inode */
1750 	if (error == 0 && dentry->d_inode != NULL)
1751 		clear_nlink(dentry->d_inode);
1752 	else if (error == -ENOENT)
1753 		nfs_dentry_handle_enoent(dentry);
1754 
1755 	return error;
1756 }
1757 
1758 /*
1759  * Remove a file after making sure there are no pending writes,
1760  * and after checking that the file has only one user.
1761  *
1762  * We invalidate the attribute cache and free the inode prior to the operation
1763  * to avoid possible races if the server reuses the inode.
1764  */
1765 static int nfs_safe_remove(struct dentry *dentry)
1766 {
1767 	struct inode *dir = dentry->d_parent->d_inode;
1768 	struct inode *inode = dentry->d_inode;
1769 	int error = -EBUSY;
1770 
1771 	dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1772 		dentry->d_parent->d_name.name, dentry->d_name.name);
1773 
1774 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1775 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1776 		error = 0;
1777 		goto out;
1778 	}
1779 
1780 	if (inode != NULL) {
1781 		nfs_inode_return_delegation(inode);
1782 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1783 		/* The VFS may want to delete this inode */
1784 		if (error == 0)
1785 			nfs_drop_nlink(inode);
1786 		nfs_mark_for_revalidate(inode);
1787 	} else
1788 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1789 	if (error == -ENOENT)
1790 		nfs_dentry_handle_enoent(dentry);
1791 out:
1792 	return error;
1793 }
1794 
1795 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1796  *  belongs to an active ".nfs..." file and we return -EBUSY.
1797  *
1798  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1799  */
1800 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1801 {
1802 	int error;
1803 	int need_rehash = 0;
1804 
1805 	dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1806 		dir->i_ino, dentry->d_name.name);
1807 
1808 	spin_lock(&dentry->d_lock);
1809 	if (dentry->d_count > 1) {
1810 		spin_unlock(&dentry->d_lock);
1811 		/* Start asynchronous writeout of the inode */
1812 		write_inode_now(dentry->d_inode, 0);
1813 		error = nfs_sillyrename(dir, dentry);
1814 		return error;
1815 	}
1816 	if (!d_unhashed(dentry)) {
1817 		__d_drop(dentry);
1818 		need_rehash = 1;
1819 	}
1820 	spin_unlock(&dentry->d_lock);
1821 	error = nfs_safe_remove(dentry);
1822 	if (!error || error == -ENOENT) {
1823 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1824 	} else if (need_rehash)
1825 		d_rehash(dentry);
1826 	return error;
1827 }
1828 
1829 /*
1830  * To create a symbolic link, most file systems instantiate a new inode,
1831  * add a page to it containing the path, then write it out to the disk
1832  * using prepare_write/commit_write.
1833  *
1834  * Unfortunately the NFS client can't create the in-core inode first
1835  * because it needs a file handle to create an in-core inode (see
1836  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1837  * symlink request has completed on the server.
1838  *
1839  * So instead we allocate a raw page, copy the symname into it, then do
1840  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1841  * now have a new file handle and can instantiate an in-core NFS inode
1842  * and move the raw page into its mapping.
1843  */
1844 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1845 {
1846 	struct pagevec lru_pvec;
1847 	struct page *page;
1848 	char *kaddr;
1849 	struct iattr attr;
1850 	unsigned int pathlen = strlen(symname);
1851 	int error;
1852 
1853 	dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1854 		dir->i_ino, dentry->d_name.name, symname);
1855 
1856 	if (pathlen > PAGE_SIZE)
1857 		return -ENAMETOOLONG;
1858 
1859 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1860 	attr.ia_valid = ATTR_MODE;
1861 
1862 	page = alloc_page(GFP_HIGHUSER);
1863 	if (!page)
1864 		return -ENOMEM;
1865 
1866 	kaddr = kmap_atomic(page);
1867 	memcpy(kaddr, symname, pathlen);
1868 	if (pathlen < PAGE_SIZE)
1869 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1870 	kunmap_atomic(kaddr);
1871 
1872 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1873 	if (error != 0) {
1874 		dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1875 			dir->i_sb->s_id, dir->i_ino,
1876 			dentry->d_name.name, symname, error);
1877 		d_drop(dentry);
1878 		__free_page(page);
1879 		return error;
1880 	}
1881 
1882 	/*
1883 	 * No big deal if we can't add this page to the page cache here.
1884 	 * READLINK will get the missing page from the server if needed.
1885 	 */
1886 	pagevec_init(&lru_pvec, 0);
1887 	if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1888 							GFP_KERNEL)) {
1889 		pagevec_add(&lru_pvec, page);
1890 		pagevec_lru_add_file(&lru_pvec);
1891 		SetPageUptodate(page);
1892 		unlock_page(page);
1893 	} else
1894 		__free_page(page);
1895 
1896 	return 0;
1897 }
1898 
1899 static int
1900 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1901 {
1902 	struct inode *inode = old_dentry->d_inode;
1903 	int error;
1904 
1905 	dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1906 		old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1907 		dentry->d_parent->d_name.name, dentry->d_name.name);
1908 
1909 	nfs_inode_return_delegation(inode);
1910 
1911 	d_drop(dentry);
1912 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1913 	if (error == 0) {
1914 		ihold(inode);
1915 		d_add(dentry, inode);
1916 	}
1917 	return error;
1918 }
1919 
1920 /*
1921  * RENAME
1922  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1923  * different file handle for the same inode after a rename (e.g. when
1924  * moving to a different directory). A fail-safe method to do so would
1925  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1926  * rename the old file using the sillyrename stuff. This way, the original
1927  * file in old_dir will go away when the last process iput()s the inode.
1928  *
1929  * FIXED.
1930  *
1931  * It actually works quite well. One needs to have the possibility for
1932  * at least one ".nfs..." file in each directory the file ever gets
1933  * moved or linked to which happens automagically with the new
1934  * implementation that only depends on the dcache stuff instead of
1935  * using the inode layer
1936  *
1937  * Unfortunately, things are a little more complicated than indicated
1938  * above. For a cross-directory move, we want to make sure we can get
1939  * rid of the old inode after the operation.  This means there must be
1940  * no pending writes (if it's a file), and the use count must be 1.
1941  * If these conditions are met, we can drop the dentries before doing
1942  * the rename.
1943  */
1944 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1945 		      struct inode *new_dir, struct dentry *new_dentry)
1946 {
1947 	struct inode *old_inode = old_dentry->d_inode;
1948 	struct inode *new_inode = new_dentry->d_inode;
1949 	struct dentry *dentry = NULL, *rehash = NULL;
1950 	int error = -EBUSY;
1951 
1952 	dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1953 		 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1954 		 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1955 		 new_dentry->d_count);
1956 
1957 	/*
1958 	 * For non-directories, check whether the target is busy and if so,
1959 	 * make a copy of the dentry and then do a silly-rename. If the
1960 	 * silly-rename succeeds, the copied dentry is hashed and becomes
1961 	 * the new target.
1962 	 */
1963 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1964 		/*
1965 		 * To prevent any new references to the target during the
1966 		 * rename, we unhash the dentry in advance.
1967 		 */
1968 		if (!d_unhashed(new_dentry)) {
1969 			d_drop(new_dentry);
1970 			rehash = new_dentry;
1971 		}
1972 
1973 		if (new_dentry->d_count > 2) {
1974 			int err;
1975 
1976 			/* copy the target dentry's name */
1977 			dentry = d_alloc(new_dentry->d_parent,
1978 					 &new_dentry->d_name);
1979 			if (!dentry)
1980 				goto out;
1981 
1982 			/* silly-rename the existing target ... */
1983 			err = nfs_sillyrename(new_dir, new_dentry);
1984 			if (err)
1985 				goto out;
1986 
1987 			new_dentry = dentry;
1988 			rehash = NULL;
1989 			new_inode = NULL;
1990 		}
1991 	}
1992 
1993 	nfs_inode_return_delegation(old_inode);
1994 	if (new_inode != NULL)
1995 		nfs_inode_return_delegation(new_inode);
1996 
1997 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1998 					   new_dir, &new_dentry->d_name);
1999 	nfs_mark_for_revalidate(old_inode);
2000 out:
2001 	if (rehash)
2002 		d_rehash(rehash);
2003 	if (!error) {
2004 		if (new_inode != NULL)
2005 			nfs_drop_nlink(new_inode);
2006 		d_move(old_dentry, new_dentry);
2007 		nfs_set_verifier(new_dentry,
2008 					nfs_save_change_attribute(new_dir));
2009 	} else if (error == -ENOENT)
2010 		nfs_dentry_handle_enoent(old_dentry);
2011 
2012 	/* new dentry created? */
2013 	if (dentry)
2014 		dput(dentry);
2015 	return error;
2016 }
2017 
2018 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2019 static LIST_HEAD(nfs_access_lru_list);
2020 static atomic_long_t nfs_access_nr_entries;
2021 
2022 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2023 {
2024 	put_rpccred(entry->cred);
2025 	kfree(entry);
2026 	smp_mb__before_atomic_dec();
2027 	atomic_long_dec(&nfs_access_nr_entries);
2028 	smp_mb__after_atomic_dec();
2029 }
2030 
2031 static void nfs_access_free_list(struct list_head *head)
2032 {
2033 	struct nfs_access_entry *cache;
2034 
2035 	while (!list_empty(head)) {
2036 		cache = list_entry(head->next, struct nfs_access_entry, lru);
2037 		list_del(&cache->lru);
2038 		nfs_access_free_entry(cache);
2039 	}
2040 }
2041 
2042 int nfs_access_cache_shrinker(struct shrinker *shrink,
2043 			      struct shrink_control *sc)
2044 {
2045 	LIST_HEAD(head);
2046 	struct nfs_inode *nfsi, *next;
2047 	struct nfs_access_entry *cache;
2048 	int nr_to_scan = sc->nr_to_scan;
2049 	gfp_t gfp_mask = sc->gfp_mask;
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 & O_ACCMODE) != O_WRONLY)
2265 		mask |= MAY_READ;
2266 	if ((openflags & O_ACCMODE) != O_RDONLY)
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)
2279 {
2280 	struct rpc_cred *cred;
2281 	int res = 0;
2282 
2283 	if (mask & MAY_NOT_BLOCK)
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);
2334 	goto out;
2335 }
2336 
2337 /*
2338  * Local variables:
2339  *  version-control: t
2340  *  kept-new-versions: 5
2341  * End:
2342  */
2343