xref: /openbmc/linux/fs/nfs/dir.c (revision a0b54add)
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/module.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/stat.h>
24 #include <linux/fcntl.h>
25 #include <linux/string.h>
26 #include <linux/kernel.h>
27 #include <linux/slab.h>
28 #include <linux/mm.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/pagemap.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/swap.h>
37 #include <linux/sched.h>
38 #include <linux/kmemleak.h>
39 #include <linux/xattr.h>
40 
41 #include "delegation.h"
42 #include "iostat.h"
43 #include "internal.h"
44 #include "fscache.h"
45 
46 #include "nfstrace.h"
47 
48 /* #define NFS_DEBUG_VERBOSE 1 */
49 
50 static int nfs_opendir(struct inode *, struct file *);
51 static int nfs_closedir(struct inode *, struct file *);
52 static int nfs_readdir(struct file *, struct dir_context *);
53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
54 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
55 static void nfs_readdir_clear_array(struct page*);
56 
57 const struct file_operations nfs_dir_operations = {
58 	.llseek		= nfs_llseek_dir,
59 	.read		= generic_read_dir,
60 	.iterate	= nfs_readdir,
61 	.open		= nfs_opendir,
62 	.release	= nfs_closedir,
63 	.fsync		= nfs_fsync_dir,
64 };
65 
66 const struct address_space_operations nfs_dir_aops = {
67 	.freepage = nfs_readdir_clear_array,
68 };
69 
70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
71 {
72 	struct nfs_open_dir_context *ctx;
73 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
74 	if (ctx != NULL) {
75 		ctx->duped = 0;
76 		ctx->attr_gencount = NFS_I(dir)->attr_gencount;
77 		ctx->dir_cookie = 0;
78 		ctx->dup_cookie = 0;
79 		ctx->cred = get_rpccred(cred);
80 		return ctx;
81 	}
82 	return  ERR_PTR(-ENOMEM);
83 }
84 
85 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
86 {
87 	put_rpccred(ctx->cred);
88 	kfree(ctx);
89 }
90 
91 /*
92  * Open file
93  */
94 static int
95 nfs_opendir(struct inode *inode, struct file *filp)
96 {
97 	int res = 0;
98 	struct nfs_open_dir_context *ctx;
99 	struct rpc_cred *cred;
100 
101 	dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
102 
103 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
104 
105 	cred = rpc_lookup_cred();
106 	if (IS_ERR(cred))
107 		return PTR_ERR(cred);
108 	ctx = alloc_nfs_open_dir_context(inode, cred);
109 	if (IS_ERR(ctx)) {
110 		res = PTR_ERR(ctx);
111 		goto out;
112 	}
113 	filp->private_data = ctx;
114 	if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
115 		/* This is a mountpoint, so d_revalidate will never
116 		 * have been called, so we need to refresh the
117 		 * inode (for close-open consistency) ourselves.
118 		 */
119 		__nfs_revalidate_inode(NFS_SERVER(inode), inode);
120 	}
121 out:
122 	put_rpccred(cred);
123 	return res;
124 }
125 
126 static int
127 nfs_closedir(struct inode *inode, struct file *filp)
128 {
129 	put_nfs_open_dir_context(filp->private_data);
130 	return 0;
131 }
132 
133 struct nfs_cache_array_entry {
134 	u64 cookie;
135 	u64 ino;
136 	struct qstr string;
137 	unsigned char d_type;
138 };
139 
140 struct nfs_cache_array {
141 	int size;
142 	int eof_index;
143 	u64 last_cookie;
144 	struct nfs_cache_array_entry array[0];
145 };
146 
147 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
148 typedef struct {
149 	struct file	*file;
150 	struct page	*page;
151 	struct dir_context *ctx;
152 	unsigned long	page_index;
153 	u64		*dir_cookie;
154 	u64		last_cookie;
155 	loff_t		current_index;
156 	decode_dirent_t	decode;
157 
158 	unsigned long	timestamp;
159 	unsigned long	gencount;
160 	unsigned int	cache_entry_index;
161 	unsigned int	plus:1;
162 	unsigned int	eof:1;
163 } nfs_readdir_descriptor_t;
164 
165 /*
166  * The caller is responsible for calling nfs_readdir_release_array(page)
167  */
168 static
169 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
170 {
171 	void *ptr;
172 	if (page == NULL)
173 		return ERR_PTR(-EIO);
174 	ptr = kmap(page);
175 	if (ptr == NULL)
176 		return ERR_PTR(-ENOMEM);
177 	return ptr;
178 }
179 
180 static
181 void nfs_readdir_release_array(struct page *page)
182 {
183 	kunmap(page);
184 }
185 
186 /*
187  * we are freeing strings created by nfs_add_to_readdir_array()
188  */
189 static
190 void nfs_readdir_clear_array(struct page *page)
191 {
192 	struct nfs_cache_array *array;
193 	int i;
194 
195 	array = kmap_atomic(page);
196 	for (i = 0; i < array->size; i++)
197 		kfree(array->array[i].string.name);
198 	kunmap_atomic(array);
199 }
200 
201 /*
202  * the caller is responsible for freeing qstr.name
203  * when called by nfs_readdir_add_to_array, the strings will be freed in
204  * nfs_clear_readdir_array()
205  */
206 static
207 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
208 {
209 	string->len = len;
210 	string->name = kmemdup(name, len, GFP_KERNEL);
211 	if (string->name == NULL)
212 		return -ENOMEM;
213 	/*
214 	 * Avoid a kmemleak false positive. The pointer to the name is stored
215 	 * in a page cache page which kmemleak does not scan.
216 	 */
217 	kmemleak_not_leak(string->name);
218 	string->hash = full_name_hash(name, len);
219 	return 0;
220 }
221 
222 static
223 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
224 {
225 	struct nfs_cache_array *array = nfs_readdir_get_array(page);
226 	struct nfs_cache_array_entry *cache_entry;
227 	int ret;
228 
229 	if (IS_ERR(array))
230 		return PTR_ERR(array);
231 
232 	cache_entry = &array->array[array->size];
233 
234 	/* Check that this entry lies within the page bounds */
235 	ret = -ENOSPC;
236 	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
237 		goto out;
238 
239 	cache_entry->cookie = entry->prev_cookie;
240 	cache_entry->ino = entry->ino;
241 	cache_entry->d_type = entry->d_type;
242 	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
243 	if (ret)
244 		goto out;
245 	array->last_cookie = entry->cookie;
246 	array->size++;
247 	if (entry->eof != 0)
248 		array->eof_index = array->size;
249 out:
250 	nfs_readdir_release_array(page);
251 	return ret;
252 }
253 
254 static
255 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
256 {
257 	loff_t diff = desc->ctx->pos - desc->current_index;
258 	unsigned int index;
259 
260 	if (diff < 0)
261 		goto out_eof;
262 	if (diff >= array->size) {
263 		if (array->eof_index >= 0)
264 			goto out_eof;
265 		return -EAGAIN;
266 	}
267 
268 	index = (unsigned int)diff;
269 	*desc->dir_cookie = array->array[index].cookie;
270 	desc->cache_entry_index = index;
271 	return 0;
272 out_eof:
273 	desc->eof = 1;
274 	return -EBADCOOKIE;
275 }
276 
277 static bool
278 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
279 {
280 	if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
281 		return false;
282 	smp_rmb();
283 	return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
284 }
285 
286 static
287 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
288 {
289 	int i;
290 	loff_t new_pos;
291 	int status = -EAGAIN;
292 
293 	for (i = 0; i < array->size; i++) {
294 		if (array->array[i].cookie == *desc->dir_cookie) {
295 			struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
296 			struct nfs_open_dir_context *ctx = desc->file->private_data;
297 
298 			new_pos = desc->current_index + i;
299 			if (ctx->attr_gencount != nfsi->attr_gencount ||
300 			    !nfs_readdir_inode_mapping_valid(nfsi)) {
301 				ctx->duped = 0;
302 				ctx->attr_gencount = nfsi->attr_gencount;
303 			} else if (new_pos < desc->ctx->pos) {
304 				if (ctx->duped > 0
305 				    && ctx->dup_cookie == *desc->dir_cookie) {
306 					if (printk_ratelimit()) {
307 						pr_notice("NFS: directory %pD2 contains a readdir loop."
308 								"Please contact your server vendor.  "
309 								"The file: %s has duplicate cookie %llu\n",
310 								desc->file,
311 								array->array[i].string.name,
312 								*desc->dir_cookie);
313 					}
314 					status = -ELOOP;
315 					goto out;
316 				}
317 				ctx->dup_cookie = *desc->dir_cookie;
318 				ctx->duped = -1;
319 			}
320 			desc->ctx->pos = new_pos;
321 			desc->cache_entry_index = i;
322 			return 0;
323 		}
324 	}
325 	if (array->eof_index >= 0) {
326 		status = -EBADCOOKIE;
327 		if (*desc->dir_cookie == array->last_cookie)
328 			desc->eof = 1;
329 	}
330 out:
331 	return status;
332 }
333 
334 static
335 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
336 {
337 	struct nfs_cache_array *array;
338 	int status;
339 
340 	array = nfs_readdir_get_array(desc->page);
341 	if (IS_ERR(array)) {
342 		status = PTR_ERR(array);
343 		goto out;
344 	}
345 
346 	if (*desc->dir_cookie == 0)
347 		status = nfs_readdir_search_for_pos(array, desc);
348 	else
349 		status = nfs_readdir_search_for_cookie(array, desc);
350 
351 	if (status == -EAGAIN) {
352 		desc->last_cookie = array->last_cookie;
353 		desc->current_index += array->size;
354 		desc->page_index++;
355 	}
356 	nfs_readdir_release_array(desc->page);
357 out:
358 	return status;
359 }
360 
361 /* Fill a page with xdr information before transferring to the cache page */
362 static
363 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
364 			struct nfs_entry *entry, struct file *file, struct inode *inode)
365 {
366 	struct nfs_open_dir_context *ctx = file->private_data;
367 	struct rpc_cred	*cred = ctx->cred;
368 	unsigned long	timestamp, gencount;
369 	int		error;
370 
371  again:
372 	timestamp = jiffies;
373 	gencount = nfs_inc_attr_generation_counter();
374 	error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
375 					  NFS_SERVER(inode)->dtsize, desc->plus);
376 	if (error < 0) {
377 		/* We requested READDIRPLUS, but the server doesn't grok it */
378 		if (error == -ENOTSUPP && desc->plus) {
379 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
380 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
381 			desc->plus = 0;
382 			goto again;
383 		}
384 		goto error;
385 	}
386 	desc->timestamp = timestamp;
387 	desc->gencount = gencount;
388 error:
389 	return error;
390 }
391 
392 static int xdr_decode(nfs_readdir_descriptor_t *desc,
393 		      struct nfs_entry *entry, struct xdr_stream *xdr)
394 {
395 	int error;
396 
397 	error = desc->decode(xdr, entry, desc->plus);
398 	if (error)
399 		return error;
400 	entry->fattr->time_start = desc->timestamp;
401 	entry->fattr->gencount = desc->gencount;
402 	return 0;
403 }
404 
405 static
406 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
407 {
408 	if (dentry->d_inode == NULL)
409 		goto different;
410 	if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
411 		goto different;
412 	return 1;
413 different:
414 	return 0;
415 }
416 
417 static
418 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
419 {
420 	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
421 		return false;
422 	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
423 		return true;
424 	if (ctx->pos == 0)
425 		return true;
426 	return false;
427 }
428 
429 /*
430  * This function is called by the lookup code to request the use of
431  * readdirplus to accelerate any future lookups in the same
432  * directory.
433  */
434 static
435 void nfs_advise_use_readdirplus(struct inode *dir)
436 {
437 	set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
438 }
439 
440 static
441 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
442 {
443 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
444 	struct dentry *dentry;
445 	struct dentry *alias;
446 	struct inode *dir = parent->d_inode;
447 	struct inode *inode;
448 	int status;
449 
450 	if (filename.name[0] == '.') {
451 		if (filename.len == 1)
452 			return;
453 		if (filename.len == 2 && filename.name[1] == '.')
454 			return;
455 	}
456 	filename.hash = full_name_hash(filename.name, filename.len);
457 
458 	dentry = d_lookup(parent, &filename);
459 	if (dentry != NULL) {
460 		if (nfs_same_file(dentry, entry)) {
461 			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
462 			status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
463 			if (!status)
464 				nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
465 			goto out;
466 		} else {
467 			if (d_invalidate(dentry) != 0)
468 				goto out;
469 			dput(dentry);
470 		}
471 	}
472 
473 	dentry = d_alloc(parent, &filename);
474 	if (dentry == NULL)
475 		return;
476 
477 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
478 	if (IS_ERR(inode))
479 		goto out;
480 
481 	alias = d_materialise_unique(dentry, inode);
482 	if (IS_ERR(alias))
483 		goto out;
484 	else if (alias) {
485 		nfs_set_verifier(alias, nfs_save_change_attribute(dir));
486 		dput(alias);
487 	} else
488 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
489 
490 out:
491 	dput(dentry);
492 }
493 
494 /* Perform conversion from xdr to cache array */
495 static
496 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
497 				struct page **xdr_pages, struct page *page, unsigned int buflen)
498 {
499 	struct xdr_stream stream;
500 	struct xdr_buf buf;
501 	struct page *scratch;
502 	struct nfs_cache_array *array;
503 	unsigned int count = 0;
504 	int status;
505 
506 	scratch = alloc_page(GFP_KERNEL);
507 	if (scratch == NULL)
508 		return -ENOMEM;
509 
510 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
511 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
512 
513 	do {
514 		status = xdr_decode(desc, entry, &stream);
515 		if (status != 0) {
516 			if (status == -EAGAIN)
517 				status = 0;
518 			break;
519 		}
520 
521 		count++;
522 
523 		if (desc->plus != 0)
524 			nfs_prime_dcache(desc->file->f_path.dentry, entry);
525 
526 		status = nfs_readdir_add_to_array(entry, page);
527 		if (status != 0)
528 			break;
529 	} while (!entry->eof);
530 
531 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
532 		array = nfs_readdir_get_array(page);
533 		if (!IS_ERR(array)) {
534 			array->eof_index = array->size;
535 			status = 0;
536 			nfs_readdir_release_array(page);
537 		} else
538 			status = PTR_ERR(array);
539 	}
540 
541 	put_page(scratch);
542 	return status;
543 }
544 
545 static
546 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
547 {
548 	unsigned int i;
549 	for (i = 0; i < npages; i++)
550 		put_page(pages[i]);
551 }
552 
553 static
554 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
555 		unsigned int npages)
556 {
557 	nfs_readdir_free_pagearray(pages, npages);
558 }
559 
560 /*
561  * nfs_readdir_large_page will allocate pages that must be freed with a call
562  * to nfs_readdir_free_large_page
563  */
564 static
565 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
566 {
567 	unsigned int i;
568 
569 	for (i = 0; i < npages; i++) {
570 		struct page *page = alloc_page(GFP_KERNEL);
571 		if (page == NULL)
572 			goto out_freepages;
573 		pages[i] = page;
574 	}
575 	return 0;
576 
577 out_freepages:
578 	nfs_readdir_free_pagearray(pages, i);
579 	return -ENOMEM;
580 }
581 
582 static
583 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
584 {
585 	struct page *pages[NFS_MAX_READDIR_PAGES];
586 	void *pages_ptr = NULL;
587 	struct nfs_entry entry;
588 	struct file	*file = desc->file;
589 	struct nfs_cache_array *array;
590 	int status = -ENOMEM;
591 	unsigned int array_size = ARRAY_SIZE(pages);
592 
593 	entry.prev_cookie = 0;
594 	entry.cookie = desc->last_cookie;
595 	entry.eof = 0;
596 	entry.fh = nfs_alloc_fhandle();
597 	entry.fattr = nfs_alloc_fattr();
598 	entry.server = NFS_SERVER(inode);
599 	if (entry.fh == NULL || entry.fattr == NULL)
600 		goto out;
601 
602 	entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
603 	if (IS_ERR(entry.label)) {
604 		status = PTR_ERR(entry.label);
605 		goto out;
606 	}
607 
608 	array = nfs_readdir_get_array(page);
609 	if (IS_ERR(array)) {
610 		status = PTR_ERR(array);
611 		goto out_label_free;
612 	}
613 	memset(array, 0, sizeof(struct nfs_cache_array));
614 	array->eof_index = -1;
615 
616 	status = nfs_readdir_large_page(pages, array_size);
617 	if (status < 0)
618 		goto out_release_array;
619 	do {
620 		unsigned int pglen;
621 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
622 
623 		if (status < 0)
624 			break;
625 		pglen = status;
626 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
627 		if (status < 0) {
628 			if (status == -ENOSPC)
629 				status = 0;
630 			break;
631 		}
632 	} while (array->eof_index < 0);
633 
634 	nfs_readdir_free_large_page(pages_ptr, pages, array_size);
635 out_release_array:
636 	nfs_readdir_release_array(page);
637 out_label_free:
638 	nfs4_label_free(entry.label);
639 out:
640 	nfs_free_fattr(entry.fattr);
641 	nfs_free_fhandle(entry.fh);
642 	return status;
643 }
644 
645 /*
646  * Now we cache directories properly, by converting xdr information
647  * to an array that can be used for lookups later.  This results in
648  * fewer cache pages, since we can store more information on each page.
649  * We only need to convert from xdr once so future lookups are much simpler
650  */
651 static
652 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
653 {
654 	struct inode	*inode = file_inode(desc->file);
655 	int ret;
656 
657 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
658 	if (ret < 0)
659 		goto error;
660 	SetPageUptodate(page);
661 
662 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
663 		/* Should never happen */
664 		nfs_zap_mapping(inode, inode->i_mapping);
665 	}
666 	unlock_page(page);
667 	return 0;
668  error:
669 	unlock_page(page);
670 	return ret;
671 }
672 
673 static
674 void cache_page_release(nfs_readdir_descriptor_t *desc)
675 {
676 	if (!desc->page->mapping)
677 		nfs_readdir_clear_array(desc->page);
678 	page_cache_release(desc->page);
679 	desc->page = NULL;
680 }
681 
682 static
683 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
684 {
685 	return read_cache_page(file_inode(desc->file)->i_mapping,
686 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
687 }
688 
689 /*
690  * Returns 0 if desc->dir_cookie was found on page desc->page_index
691  */
692 static
693 int find_cache_page(nfs_readdir_descriptor_t *desc)
694 {
695 	int res;
696 
697 	desc->page = get_cache_page(desc);
698 	if (IS_ERR(desc->page))
699 		return PTR_ERR(desc->page);
700 
701 	res = nfs_readdir_search_array(desc);
702 	if (res != 0)
703 		cache_page_release(desc);
704 	return res;
705 }
706 
707 /* Search for desc->dir_cookie from the beginning of the page cache */
708 static inline
709 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
710 {
711 	int res;
712 
713 	if (desc->page_index == 0) {
714 		desc->current_index = 0;
715 		desc->last_cookie = 0;
716 	}
717 	do {
718 		res = find_cache_page(desc);
719 	} while (res == -EAGAIN);
720 	return res;
721 }
722 
723 /*
724  * Once we've found the start of the dirent within a page: fill 'er up...
725  */
726 static
727 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
728 {
729 	struct file	*file = desc->file;
730 	int i = 0;
731 	int res = 0;
732 	struct nfs_cache_array *array = NULL;
733 	struct nfs_open_dir_context *ctx = file->private_data;
734 
735 	array = nfs_readdir_get_array(desc->page);
736 	if (IS_ERR(array)) {
737 		res = PTR_ERR(array);
738 		goto out;
739 	}
740 
741 	for (i = desc->cache_entry_index; i < array->size; i++) {
742 		struct nfs_cache_array_entry *ent;
743 
744 		ent = &array->array[i];
745 		if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
746 		    nfs_compat_user_ino64(ent->ino), ent->d_type)) {
747 			desc->eof = 1;
748 			break;
749 		}
750 		desc->ctx->pos++;
751 		if (i < (array->size-1))
752 			*desc->dir_cookie = array->array[i+1].cookie;
753 		else
754 			*desc->dir_cookie = array->last_cookie;
755 		if (ctx->duped != 0)
756 			ctx->duped = 1;
757 	}
758 	if (array->eof_index >= 0)
759 		desc->eof = 1;
760 
761 	nfs_readdir_release_array(desc->page);
762 out:
763 	cache_page_release(desc);
764 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
765 			(unsigned long long)*desc->dir_cookie, res);
766 	return res;
767 }
768 
769 /*
770  * If we cannot find a cookie in our cache, we suspect that this is
771  * because it points to a deleted file, so we ask the server to return
772  * whatever it thinks is the next entry. We then feed this to filldir.
773  * If all goes well, we should then be able to find our way round the
774  * cache on the next call to readdir_search_pagecache();
775  *
776  * NOTE: we cannot add the anonymous page to the pagecache because
777  *	 the data it contains might not be page aligned. Besides,
778  *	 we should already have a complete representation of the
779  *	 directory in the page cache by the time we get here.
780  */
781 static inline
782 int uncached_readdir(nfs_readdir_descriptor_t *desc)
783 {
784 	struct page	*page = NULL;
785 	int		status;
786 	struct inode *inode = file_inode(desc->file);
787 	struct nfs_open_dir_context *ctx = desc->file->private_data;
788 
789 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
790 			(unsigned long long)*desc->dir_cookie);
791 
792 	page = alloc_page(GFP_HIGHUSER);
793 	if (!page) {
794 		status = -ENOMEM;
795 		goto out;
796 	}
797 
798 	desc->page_index = 0;
799 	desc->last_cookie = *desc->dir_cookie;
800 	desc->page = page;
801 	ctx->duped = 0;
802 
803 	status = nfs_readdir_xdr_to_array(desc, page, inode);
804 	if (status < 0)
805 		goto out_release;
806 
807 	status = nfs_do_filldir(desc);
808 
809  out:
810 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
811 			__func__, status);
812 	return status;
813  out_release:
814 	cache_page_release(desc);
815 	goto out;
816 }
817 
818 /* The file offset position represents the dirent entry number.  A
819    last cookie cache takes care of the common case of reading the
820    whole directory.
821  */
822 static int nfs_readdir(struct file *file, struct dir_context *ctx)
823 {
824 	struct dentry	*dentry = file->f_path.dentry;
825 	struct inode	*inode = dentry->d_inode;
826 	nfs_readdir_descriptor_t my_desc,
827 			*desc = &my_desc;
828 	struct nfs_open_dir_context *dir_ctx = file->private_data;
829 	int res = 0;
830 
831 	dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
832 			file, (long long)ctx->pos);
833 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
834 
835 	/*
836 	 * ctx->pos points to the dirent entry number.
837 	 * *desc->dir_cookie has the cookie for the next entry. We have
838 	 * to either find the entry with the appropriate number or
839 	 * revalidate the cookie.
840 	 */
841 	memset(desc, 0, sizeof(*desc));
842 
843 	desc->file = file;
844 	desc->ctx = ctx;
845 	desc->dir_cookie = &dir_ctx->dir_cookie;
846 	desc->decode = NFS_PROTO(inode)->decode_dirent;
847 	desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
848 
849 	nfs_block_sillyrename(dentry);
850 	if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
851 		res = nfs_revalidate_mapping(inode, file->f_mapping);
852 	if (res < 0)
853 		goto out;
854 
855 	do {
856 		res = readdir_search_pagecache(desc);
857 
858 		if (res == -EBADCOOKIE) {
859 			res = 0;
860 			/* This means either end of directory */
861 			if (*desc->dir_cookie && desc->eof == 0) {
862 				/* Or that the server has 'lost' a cookie */
863 				res = uncached_readdir(desc);
864 				if (res == 0)
865 					continue;
866 			}
867 			break;
868 		}
869 		if (res == -ETOOSMALL && desc->plus) {
870 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
871 			nfs_zap_caches(inode);
872 			desc->page_index = 0;
873 			desc->plus = 0;
874 			desc->eof = 0;
875 			continue;
876 		}
877 		if (res < 0)
878 			break;
879 
880 		res = nfs_do_filldir(desc);
881 		if (res < 0)
882 			break;
883 	} while (!desc->eof);
884 out:
885 	nfs_unblock_sillyrename(dentry);
886 	if (res > 0)
887 		res = 0;
888 	dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
889 	return res;
890 }
891 
892 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
893 {
894 	struct inode *inode = file_inode(filp);
895 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
896 
897 	dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
898 			filp, offset, whence);
899 
900 	mutex_lock(&inode->i_mutex);
901 	switch (whence) {
902 		case 1:
903 			offset += filp->f_pos;
904 		case 0:
905 			if (offset >= 0)
906 				break;
907 		default:
908 			offset = -EINVAL;
909 			goto out;
910 	}
911 	if (offset != filp->f_pos) {
912 		filp->f_pos = offset;
913 		dir_ctx->dir_cookie = 0;
914 		dir_ctx->duped = 0;
915 	}
916 out:
917 	mutex_unlock(&inode->i_mutex);
918 	return offset;
919 }
920 
921 /*
922  * All directory operations under NFS are synchronous, so fsync()
923  * is a dummy operation.
924  */
925 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
926 			 int datasync)
927 {
928 	struct inode *inode = file_inode(filp);
929 
930 	dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
931 
932 	mutex_lock(&inode->i_mutex);
933 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
934 	mutex_unlock(&inode->i_mutex);
935 	return 0;
936 }
937 
938 /**
939  * nfs_force_lookup_revalidate - Mark the directory as having changed
940  * @dir - pointer to directory inode
941  *
942  * This forces the revalidation code in nfs_lookup_revalidate() to do a
943  * full lookup on all child dentries of 'dir' whenever a change occurs
944  * on the server that might have invalidated our dcache.
945  *
946  * The caller should be holding dir->i_lock
947  */
948 void nfs_force_lookup_revalidate(struct inode *dir)
949 {
950 	NFS_I(dir)->cache_change_attribute++;
951 }
952 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
953 
954 /*
955  * A check for whether or not the parent directory has changed.
956  * In the case it has, we assume that the dentries are untrustworthy
957  * and may need to be looked up again.
958  */
959 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
960 {
961 	if (IS_ROOT(dentry))
962 		return 1;
963 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
964 		return 0;
965 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
966 		return 0;
967 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
968 	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
969 		return 0;
970 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
971 		return 0;
972 	return 1;
973 }
974 
975 /*
976  * Use intent information to check whether or not we're going to do
977  * an O_EXCL create using this path component.
978  */
979 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
980 {
981 	if (NFS_PROTO(dir)->version == 2)
982 		return 0;
983 	return flags & LOOKUP_EXCL;
984 }
985 
986 /*
987  * Inode and filehandle revalidation for lookups.
988  *
989  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
990  * or if the intent information indicates that we're about to open this
991  * particular file and the "nocto" mount flag is not set.
992  *
993  */
994 static
995 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
996 {
997 	struct nfs_server *server = NFS_SERVER(inode);
998 	int ret;
999 
1000 	if (IS_AUTOMOUNT(inode))
1001 		return 0;
1002 	/* VFS wants an on-the-wire revalidation */
1003 	if (flags & LOOKUP_REVAL)
1004 		goto out_force;
1005 	/* This is an open(2) */
1006 	if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1007 	    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1008 		goto out_force;
1009 out:
1010 	return (inode->i_nlink == 0) ? -ENOENT : 0;
1011 out_force:
1012 	ret = __nfs_revalidate_inode(server, inode);
1013 	if (ret != 0)
1014 		return ret;
1015 	goto out;
1016 }
1017 
1018 /*
1019  * We judge how long we want to trust negative
1020  * dentries by looking at the parent inode mtime.
1021  *
1022  * If parent mtime has changed, we revalidate, else we wait for a
1023  * period corresponding to the parent's attribute cache timeout value.
1024  */
1025 static inline
1026 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1027 		       unsigned int flags)
1028 {
1029 	/* Don't revalidate a negative dentry if we're creating a new file */
1030 	if (flags & LOOKUP_CREATE)
1031 		return 0;
1032 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1033 		return 1;
1034 	return !nfs_check_verifier(dir, dentry);
1035 }
1036 
1037 /*
1038  * This is called every time the dcache has a lookup hit,
1039  * and we should check whether we can really trust that
1040  * lookup.
1041  *
1042  * NOTE! The hit can be a negative hit too, don't assume
1043  * we have an inode!
1044  *
1045  * If the parent directory is seen to have changed, we throw out the
1046  * cached dentry and do a new lookup.
1047  */
1048 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1049 {
1050 	struct inode *dir;
1051 	struct inode *inode;
1052 	struct dentry *parent;
1053 	struct nfs_fh *fhandle = NULL;
1054 	struct nfs_fattr *fattr = NULL;
1055 	struct nfs4_label *label = NULL;
1056 	int error;
1057 
1058 	if (flags & LOOKUP_RCU)
1059 		return -ECHILD;
1060 
1061 	parent = dget_parent(dentry);
1062 	dir = parent->d_inode;
1063 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1064 	inode = dentry->d_inode;
1065 
1066 	if (!inode) {
1067 		if (nfs_neg_need_reval(dir, dentry, flags))
1068 			goto out_bad;
1069 		goto out_valid_noent;
1070 	}
1071 
1072 	if (is_bad_inode(inode)) {
1073 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1074 				__func__, dentry);
1075 		goto out_bad;
1076 	}
1077 
1078 	if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1079 		goto out_set_verifier;
1080 
1081 	/* Force a full look up iff the parent directory has changed */
1082 	if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1083 		if (nfs_lookup_verify_inode(inode, flags))
1084 			goto out_zap_parent;
1085 		goto out_valid;
1086 	}
1087 
1088 	if (NFS_STALE(inode))
1089 		goto out_bad;
1090 
1091 	error = -ENOMEM;
1092 	fhandle = nfs_alloc_fhandle();
1093 	fattr = nfs_alloc_fattr();
1094 	if (fhandle == NULL || fattr == NULL)
1095 		goto out_error;
1096 
1097 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1098 	if (IS_ERR(label))
1099 		goto out_error;
1100 
1101 	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1102 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1103 	trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1104 	if (error)
1105 		goto out_bad;
1106 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1107 		goto out_bad;
1108 	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1109 		goto out_bad;
1110 
1111 	nfs_setsecurity(inode, fattr, label);
1112 
1113 	nfs_free_fattr(fattr);
1114 	nfs_free_fhandle(fhandle);
1115 	nfs4_label_free(label);
1116 
1117 out_set_verifier:
1118 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1119  out_valid:
1120 	/* Success: notify readdir to use READDIRPLUS */
1121 	nfs_advise_use_readdirplus(dir);
1122  out_valid_noent:
1123 	dput(parent);
1124 	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1125 			__func__, dentry);
1126 	return 1;
1127 out_zap_parent:
1128 	nfs_zap_caches(dir);
1129  out_bad:
1130 	nfs_free_fattr(fattr);
1131 	nfs_free_fhandle(fhandle);
1132 	nfs4_label_free(label);
1133 	nfs_mark_for_revalidate(dir);
1134 	if (inode && S_ISDIR(inode->i_mode)) {
1135 		/* Purge readdir caches. */
1136 		nfs_zap_caches(inode);
1137 		/*
1138 		 * We can't d_drop the root of a disconnected tree:
1139 		 * its d_hash is on the s_anon list and d_drop() would hide
1140 		 * it from shrink_dcache_for_unmount(), leading to busy
1141 		 * inodes on unmount and further oopses.
1142 		 */
1143 		if (IS_ROOT(dentry))
1144 			goto out_valid;
1145 	}
1146 	/* If we have submounts, don't unhash ! */
1147 	if (check_submounts_and_drop(dentry) != 0)
1148 		goto out_valid;
1149 
1150 	dput(parent);
1151 	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1152 			__func__, dentry);
1153 	return 0;
1154 out_error:
1155 	nfs_free_fattr(fattr);
1156 	nfs_free_fhandle(fhandle);
1157 	nfs4_label_free(label);
1158 	dput(parent);
1159 	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1160 			__func__, dentry, error);
1161 	return error;
1162 }
1163 
1164 /*
1165  * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1166  * when we don't really care about the dentry name. This is called when a
1167  * pathwalk ends on a dentry that was not found via a normal lookup in the
1168  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1169  *
1170  * In this situation, we just want to verify that the inode itself is OK
1171  * since the dentry might have changed on the server.
1172  */
1173 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1174 {
1175 	int error;
1176 	struct inode *inode = dentry->d_inode;
1177 
1178 	/*
1179 	 * I believe we can only get a negative dentry here in the case of a
1180 	 * procfs-style symlink. Just assume it's correct for now, but we may
1181 	 * eventually need to do something more here.
1182 	 */
1183 	if (!inode) {
1184 		dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1185 				__func__, dentry);
1186 		return 1;
1187 	}
1188 
1189 	if (is_bad_inode(inode)) {
1190 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1191 				__func__, dentry);
1192 		return 0;
1193 	}
1194 
1195 	error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1196 	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1197 			__func__, inode->i_ino, error ? "invalid" : "valid");
1198 	return !error;
1199 }
1200 
1201 /*
1202  * This is called from dput() when d_count is going to 0.
1203  */
1204 static int nfs_dentry_delete(const struct dentry *dentry)
1205 {
1206 	dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1207 		dentry, dentry->d_flags);
1208 
1209 	/* Unhash any dentry with a stale inode */
1210 	if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1211 		return 1;
1212 
1213 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1214 		/* Unhash it, so that ->d_iput() would be called */
1215 		return 1;
1216 	}
1217 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1218 		/* Unhash it, so that ancestors of killed async unlink
1219 		 * files will be cleaned up during umount */
1220 		return 1;
1221 	}
1222 	return 0;
1223 
1224 }
1225 
1226 /* Ensure that we revalidate inode->i_nlink */
1227 static void nfs_drop_nlink(struct inode *inode)
1228 {
1229 	spin_lock(&inode->i_lock);
1230 	/* drop the inode if we're reasonably sure this is the last link */
1231 	if (inode->i_nlink == 1)
1232 		clear_nlink(inode);
1233 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1234 	spin_unlock(&inode->i_lock);
1235 }
1236 
1237 /*
1238  * Called when the dentry loses inode.
1239  * We use it to clean up silly-renamed files.
1240  */
1241 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1242 {
1243 	if (S_ISDIR(inode->i_mode))
1244 		/* drop any readdir cache as it could easily be old */
1245 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1246 
1247 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1248 		nfs_complete_unlink(dentry, inode);
1249 		nfs_drop_nlink(inode);
1250 	}
1251 	iput(inode);
1252 }
1253 
1254 static void nfs_d_release(struct dentry *dentry)
1255 {
1256 	/* free cached devname value, if it survived that far */
1257 	if (unlikely(dentry->d_fsdata)) {
1258 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1259 			WARN_ON(1);
1260 		else
1261 			kfree(dentry->d_fsdata);
1262 	}
1263 }
1264 
1265 const struct dentry_operations nfs_dentry_operations = {
1266 	.d_revalidate	= nfs_lookup_revalidate,
1267 	.d_weak_revalidate	= nfs_weak_revalidate,
1268 	.d_delete	= nfs_dentry_delete,
1269 	.d_iput		= nfs_dentry_iput,
1270 	.d_automount	= nfs_d_automount,
1271 	.d_release	= nfs_d_release,
1272 };
1273 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1274 
1275 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1276 {
1277 	struct dentry *res;
1278 	struct dentry *parent;
1279 	struct inode *inode = NULL;
1280 	struct nfs_fh *fhandle = NULL;
1281 	struct nfs_fattr *fattr = NULL;
1282 	struct nfs4_label *label = NULL;
1283 	int error;
1284 
1285 	dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1286 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1287 
1288 	res = ERR_PTR(-ENAMETOOLONG);
1289 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1290 		goto out;
1291 
1292 	/*
1293 	 * If we're doing an exclusive create, optimize away the lookup
1294 	 * but don't hash the dentry.
1295 	 */
1296 	if (nfs_is_exclusive_create(dir, flags)) {
1297 		d_instantiate(dentry, NULL);
1298 		res = NULL;
1299 		goto out;
1300 	}
1301 
1302 	res = ERR_PTR(-ENOMEM);
1303 	fhandle = nfs_alloc_fhandle();
1304 	fattr = nfs_alloc_fattr();
1305 	if (fhandle == NULL || fattr == NULL)
1306 		goto out;
1307 
1308 	label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1309 	if (IS_ERR(label))
1310 		goto out;
1311 
1312 	parent = dentry->d_parent;
1313 	/* Protect against concurrent sillydeletes */
1314 	trace_nfs_lookup_enter(dir, dentry, flags);
1315 	nfs_block_sillyrename(parent);
1316 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1317 	if (error == -ENOENT)
1318 		goto no_entry;
1319 	if (error < 0) {
1320 		res = ERR_PTR(error);
1321 		goto out_unblock_sillyrename;
1322 	}
1323 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1324 	res = ERR_CAST(inode);
1325 	if (IS_ERR(res))
1326 		goto out_unblock_sillyrename;
1327 
1328 	/* Success: notify readdir to use READDIRPLUS */
1329 	nfs_advise_use_readdirplus(dir);
1330 
1331 no_entry:
1332 	res = d_materialise_unique(dentry, inode);
1333 	if (res != NULL) {
1334 		if (IS_ERR(res))
1335 			goto out_unblock_sillyrename;
1336 		dentry = res;
1337 	}
1338 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1339 out_unblock_sillyrename:
1340 	nfs_unblock_sillyrename(parent);
1341 	trace_nfs_lookup_exit(dir, dentry, flags, error);
1342 	nfs4_label_free(label);
1343 out:
1344 	nfs_free_fattr(fattr);
1345 	nfs_free_fhandle(fhandle);
1346 	return res;
1347 }
1348 EXPORT_SYMBOL_GPL(nfs_lookup);
1349 
1350 #if IS_ENABLED(CONFIG_NFS_V4)
1351 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1352 
1353 const struct dentry_operations nfs4_dentry_operations = {
1354 	.d_revalidate	= nfs4_lookup_revalidate,
1355 	.d_delete	= nfs_dentry_delete,
1356 	.d_iput		= nfs_dentry_iput,
1357 	.d_automount	= nfs_d_automount,
1358 	.d_release	= nfs_d_release,
1359 };
1360 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1361 
1362 static fmode_t flags_to_mode(int flags)
1363 {
1364 	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1365 	if ((flags & O_ACCMODE) != O_WRONLY)
1366 		res |= FMODE_READ;
1367 	if ((flags & O_ACCMODE) != O_RDONLY)
1368 		res |= FMODE_WRITE;
1369 	return res;
1370 }
1371 
1372 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1373 {
1374 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1375 }
1376 
1377 static int do_open(struct inode *inode, struct file *filp)
1378 {
1379 	nfs_fscache_open_file(inode, filp);
1380 	return 0;
1381 }
1382 
1383 static int nfs_finish_open(struct nfs_open_context *ctx,
1384 			   struct dentry *dentry,
1385 			   struct file *file, unsigned open_flags,
1386 			   int *opened)
1387 {
1388 	int err;
1389 
1390 	if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
1391 		*opened |= FILE_CREATED;
1392 
1393 	err = finish_open(file, dentry, do_open, opened);
1394 	if (err)
1395 		goto out;
1396 	nfs_file_set_open_context(file, ctx);
1397 
1398 out:
1399 	return err;
1400 }
1401 
1402 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1403 		    struct file *file, unsigned open_flags,
1404 		    umode_t mode, int *opened)
1405 {
1406 	struct nfs_open_context *ctx;
1407 	struct dentry *res;
1408 	struct iattr attr = { .ia_valid = ATTR_OPEN };
1409 	struct inode *inode;
1410 	unsigned int lookup_flags = 0;
1411 	int err;
1412 
1413 	/* Expect a negative dentry */
1414 	BUG_ON(dentry->d_inode);
1415 
1416 	dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1417 			dir->i_sb->s_id, dir->i_ino, dentry);
1418 
1419 	err = nfs_check_flags(open_flags);
1420 	if (err)
1421 		return err;
1422 
1423 	/* NFS only supports OPEN on regular files */
1424 	if ((open_flags & O_DIRECTORY)) {
1425 		if (!d_unhashed(dentry)) {
1426 			/*
1427 			 * Hashed negative dentry with O_DIRECTORY: dentry was
1428 			 * revalidated and is fine, no need to perform lookup
1429 			 * again
1430 			 */
1431 			return -ENOENT;
1432 		}
1433 		lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1434 		goto no_open;
1435 	}
1436 
1437 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1438 		return -ENAMETOOLONG;
1439 
1440 	if (open_flags & O_CREAT) {
1441 		attr.ia_valid |= ATTR_MODE;
1442 		attr.ia_mode = mode & ~current_umask();
1443 	}
1444 	if (open_flags & O_TRUNC) {
1445 		attr.ia_valid |= ATTR_SIZE;
1446 		attr.ia_size = 0;
1447 	}
1448 
1449 	ctx = create_nfs_open_context(dentry, open_flags);
1450 	err = PTR_ERR(ctx);
1451 	if (IS_ERR(ctx))
1452 		goto out;
1453 
1454 	trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1455 	nfs_block_sillyrename(dentry->d_parent);
1456 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1457 	nfs_unblock_sillyrename(dentry->d_parent);
1458 	if (IS_ERR(inode)) {
1459 		err = PTR_ERR(inode);
1460 		trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1461 		put_nfs_open_context(ctx);
1462 		switch (err) {
1463 		case -ENOENT:
1464 			d_drop(dentry);
1465 			d_add(dentry, NULL);
1466 			break;
1467 		case -EISDIR:
1468 		case -ENOTDIR:
1469 			goto no_open;
1470 		case -ELOOP:
1471 			if (!(open_flags & O_NOFOLLOW))
1472 				goto no_open;
1473 			break;
1474 			/* case -EINVAL: */
1475 		default:
1476 			break;
1477 		}
1478 		goto out;
1479 	}
1480 
1481 	err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1482 	trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1483 	put_nfs_open_context(ctx);
1484 out:
1485 	return err;
1486 
1487 no_open:
1488 	res = nfs_lookup(dir, dentry, lookup_flags);
1489 	err = PTR_ERR(res);
1490 	if (IS_ERR(res))
1491 		goto out;
1492 
1493 	return finish_no_open(file, res);
1494 }
1495 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1496 
1497 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1498 {
1499 	struct dentry *parent = NULL;
1500 	struct inode *inode;
1501 	struct inode *dir;
1502 	int ret = 0;
1503 
1504 	if (flags & LOOKUP_RCU)
1505 		return -ECHILD;
1506 
1507 	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1508 		goto no_open;
1509 	if (d_mountpoint(dentry))
1510 		goto no_open;
1511 	if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1512 		goto no_open;
1513 
1514 	inode = dentry->d_inode;
1515 	parent = dget_parent(dentry);
1516 	dir = parent->d_inode;
1517 
1518 	/* We can't create new files in nfs_open_revalidate(), so we
1519 	 * optimize away revalidation of negative dentries.
1520 	 */
1521 	if (inode == NULL) {
1522 		if (!nfs_neg_need_reval(dir, dentry, flags))
1523 			ret = 1;
1524 		goto out;
1525 	}
1526 
1527 	/* NFS only supports OPEN on regular files */
1528 	if (!S_ISREG(inode->i_mode))
1529 		goto no_open_dput;
1530 	/* We cannot do exclusive creation on a positive dentry */
1531 	if (flags & LOOKUP_EXCL)
1532 		goto no_open_dput;
1533 
1534 	/* Let f_op->open() actually open (and revalidate) the file */
1535 	ret = 1;
1536 
1537 out:
1538 	dput(parent);
1539 	return ret;
1540 
1541 no_open_dput:
1542 	dput(parent);
1543 no_open:
1544 	return nfs_lookup_revalidate(dentry, flags);
1545 }
1546 
1547 #endif /* CONFIG_NFSV4 */
1548 
1549 /*
1550  * Code common to create, mkdir, and mknod.
1551  */
1552 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1553 				struct nfs_fattr *fattr,
1554 				struct nfs4_label *label)
1555 {
1556 	struct dentry *parent = dget_parent(dentry);
1557 	struct inode *dir = parent->d_inode;
1558 	struct inode *inode;
1559 	int error = -EACCES;
1560 
1561 	d_drop(dentry);
1562 
1563 	/* We may have been initialized further down */
1564 	if (dentry->d_inode)
1565 		goto out;
1566 	if (fhandle->size == 0) {
1567 		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1568 		if (error)
1569 			goto out_error;
1570 	}
1571 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1572 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1573 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1574 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1575 		if (error < 0)
1576 			goto out_error;
1577 	}
1578 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1579 	error = PTR_ERR(inode);
1580 	if (IS_ERR(inode))
1581 		goto out_error;
1582 	d_add(dentry, inode);
1583 out:
1584 	dput(parent);
1585 	return 0;
1586 out_error:
1587 	nfs_mark_for_revalidate(dir);
1588 	dput(parent);
1589 	return error;
1590 }
1591 EXPORT_SYMBOL_GPL(nfs_instantiate);
1592 
1593 /*
1594  * Following a failed create operation, we drop the dentry rather
1595  * than retain a negative dentry. This avoids a problem in the event
1596  * that the operation succeeded on the server, but an error in the
1597  * reply path made it appear to have failed.
1598  */
1599 int nfs_create(struct inode *dir, struct dentry *dentry,
1600 		umode_t mode, bool excl)
1601 {
1602 	struct iattr attr;
1603 	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1604 	int error;
1605 
1606 	dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1607 			dir->i_sb->s_id, dir->i_ino, dentry);
1608 
1609 	attr.ia_mode = mode;
1610 	attr.ia_valid = ATTR_MODE;
1611 
1612 	trace_nfs_create_enter(dir, dentry, open_flags);
1613 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1614 	trace_nfs_create_exit(dir, dentry, open_flags, error);
1615 	if (error != 0)
1616 		goto out_err;
1617 	return 0;
1618 out_err:
1619 	d_drop(dentry);
1620 	return error;
1621 }
1622 EXPORT_SYMBOL_GPL(nfs_create);
1623 
1624 /*
1625  * See comments for nfs_proc_create regarding failed operations.
1626  */
1627 int
1628 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1629 {
1630 	struct iattr attr;
1631 	int status;
1632 
1633 	dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1634 			dir->i_sb->s_id, dir->i_ino, dentry);
1635 
1636 	if (!new_valid_dev(rdev))
1637 		return -EINVAL;
1638 
1639 	attr.ia_mode = mode;
1640 	attr.ia_valid = ATTR_MODE;
1641 
1642 	trace_nfs_mknod_enter(dir, dentry);
1643 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1644 	trace_nfs_mknod_exit(dir, dentry, status);
1645 	if (status != 0)
1646 		goto out_err;
1647 	return 0;
1648 out_err:
1649 	d_drop(dentry);
1650 	return status;
1651 }
1652 EXPORT_SYMBOL_GPL(nfs_mknod);
1653 
1654 /*
1655  * See comments for nfs_proc_create regarding failed operations.
1656  */
1657 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1658 {
1659 	struct iattr attr;
1660 	int error;
1661 
1662 	dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1663 			dir->i_sb->s_id, dir->i_ino, dentry);
1664 
1665 	attr.ia_valid = ATTR_MODE;
1666 	attr.ia_mode = mode | S_IFDIR;
1667 
1668 	trace_nfs_mkdir_enter(dir, dentry);
1669 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1670 	trace_nfs_mkdir_exit(dir, dentry, error);
1671 	if (error != 0)
1672 		goto out_err;
1673 	return 0;
1674 out_err:
1675 	d_drop(dentry);
1676 	return error;
1677 }
1678 EXPORT_SYMBOL_GPL(nfs_mkdir);
1679 
1680 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1681 {
1682 	if (dentry->d_inode != NULL && !d_unhashed(dentry))
1683 		d_delete(dentry);
1684 }
1685 
1686 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1687 {
1688 	int error;
1689 
1690 	dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1691 			dir->i_sb->s_id, dir->i_ino, dentry);
1692 
1693 	trace_nfs_rmdir_enter(dir, dentry);
1694 	if (dentry->d_inode) {
1695 		nfs_wait_on_sillyrename(dentry);
1696 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1697 		/* Ensure the VFS deletes this inode */
1698 		switch (error) {
1699 		case 0:
1700 			clear_nlink(dentry->d_inode);
1701 			break;
1702 		case -ENOENT:
1703 			nfs_dentry_handle_enoent(dentry);
1704 		}
1705 	} else
1706 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1707 	trace_nfs_rmdir_exit(dir, dentry, error);
1708 
1709 	return error;
1710 }
1711 EXPORT_SYMBOL_GPL(nfs_rmdir);
1712 
1713 /*
1714  * Remove a file after making sure there are no pending writes,
1715  * and after checking that the file has only one user.
1716  *
1717  * We invalidate the attribute cache and free the inode prior to the operation
1718  * to avoid possible races if the server reuses the inode.
1719  */
1720 static int nfs_safe_remove(struct dentry *dentry)
1721 {
1722 	struct inode *dir = dentry->d_parent->d_inode;
1723 	struct inode *inode = dentry->d_inode;
1724 	int error = -EBUSY;
1725 
1726 	dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1727 
1728 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1729 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1730 		error = 0;
1731 		goto out;
1732 	}
1733 
1734 	trace_nfs_remove_enter(dir, dentry);
1735 	if (inode != NULL) {
1736 		NFS_PROTO(inode)->return_delegation(inode);
1737 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1738 		if (error == 0)
1739 			nfs_drop_nlink(inode);
1740 	} else
1741 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1742 	if (error == -ENOENT)
1743 		nfs_dentry_handle_enoent(dentry);
1744 	trace_nfs_remove_exit(dir, dentry, error);
1745 out:
1746 	return error;
1747 }
1748 
1749 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1750  *  belongs to an active ".nfs..." file and we return -EBUSY.
1751  *
1752  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1753  */
1754 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1755 {
1756 	int error;
1757 	int need_rehash = 0;
1758 
1759 	dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1760 		dir->i_ino, dentry);
1761 
1762 	trace_nfs_unlink_enter(dir, dentry);
1763 	spin_lock(&dentry->d_lock);
1764 	if (d_count(dentry) > 1) {
1765 		spin_unlock(&dentry->d_lock);
1766 		/* Start asynchronous writeout of the inode */
1767 		write_inode_now(dentry->d_inode, 0);
1768 		error = nfs_sillyrename(dir, dentry);
1769 		goto out;
1770 	}
1771 	if (!d_unhashed(dentry)) {
1772 		__d_drop(dentry);
1773 		need_rehash = 1;
1774 	}
1775 	spin_unlock(&dentry->d_lock);
1776 	error = nfs_safe_remove(dentry);
1777 	if (!error || error == -ENOENT) {
1778 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1779 	} else if (need_rehash)
1780 		d_rehash(dentry);
1781 out:
1782 	trace_nfs_unlink_exit(dir, dentry, error);
1783 	return error;
1784 }
1785 EXPORT_SYMBOL_GPL(nfs_unlink);
1786 
1787 /*
1788  * To create a symbolic link, most file systems instantiate a new inode,
1789  * add a page to it containing the path, then write it out to the disk
1790  * using prepare_write/commit_write.
1791  *
1792  * Unfortunately the NFS client can't create the in-core inode first
1793  * because it needs a file handle to create an in-core inode (see
1794  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1795  * symlink request has completed on the server.
1796  *
1797  * So instead we allocate a raw page, copy the symname into it, then do
1798  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1799  * now have a new file handle and can instantiate an in-core NFS inode
1800  * and move the raw page into its mapping.
1801  */
1802 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1803 {
1804 	struct page *page;
1805 	char *kaddr;
1806 	struct iattr attr;
1807 	unsigned int pathlen = strlen(symname);
1808 	int error;
1809 
1810 	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1811 		dir->i_ino, dentry, symname);
1812 
1813 	if (pathlen > PAGE_SIZE)
1814 		return -ENAMETOOLONG;
1815 
1816 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1817 	attr.ia_valid = ATTR_MODE;
1818 
1819 	page = alloc_page(GFP_HIGHUSER);
1820 	if (!page)
1821 		return -ENOMEM;
1822 
1823 	kaddr = kmap_atomic(page);
1824 	memcpy(kaddr, symname, pathlen);
1825 	if (pathlen < PAGE_SIZE)
1826 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1827 	kunmap_atomic(kaddr);
1828 
1829 	trace_nfs_symlink_enter(dir, dentry);
1830 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1831 	trace_nfs_symlink_exit(dir, dentry, error);
1832 	if (error != 0) {
1833 		dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1834 			dir->i_sb->s_id, dir->i_ino,
1835 			dentry, symname, error);
1836 		d_drop(dentry);
1837 		__free_page(page);
1838 		return error;
1839 	}
1840 
1841 	/*
1842 	 * No big deal if we can't add this page to the page cache here.
1843 	 * READLINK will get the missing page from the server if needed.
1844 	 */
1845 	if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
1846 							GFP_KERNEL)) {
1847 		SetPageUptodate(page);
1848 		unlock_page(page);
1849 		/*
1850 		 * add_to_page_cache_lru() grabs an extra page refcount.
1851 		 * Drop it here to avoid leaking this page later.
1852 		 */
1853 		page_cache_release(page);
1854 	} else
1855 		__free_page(page);
1856 
1857 	return 0;
1858 }
1859 EXPORT_SYMBOL_GPL(nfs_symlink);
1860 
1861 int
1862 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1863 {
1864 	struct inode *inode = old_dentry->d_inode;
1865 	int error;
1866 
1867 	dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1868 		old_dentry, dentry);
1869 
1870 	trace_nfs_link_enter(inode, dir, dentry);
1871 	NFS_PROTO(inode)->return_delegation(inode);
1872 
1873 	d_drop(dentry);
1874 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1875 	if (error == 0) {
1876 		ihold(inode);
1877 		d_add(dentry, inode);
1878 	}
1879 	trace_nfs_link_exit(inode, dir, dentry, error);
1880 	return error;
1881 }
1882 EXPORT_SYMBOL_GPL(nfs_link);
1883 
1884 /*
1885  * RENAME
1886  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1887  * different file handle for the same inode after a rename (e.g. when
1888  * moving to a different directory). A fail-safe method to do so would
1889  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1890  * rename the old file using the sillyrename stuff. This way, the original
1891  * file in old_dir will go away when the last process iput()s the inode.
1892  *
1893  * FIXED.
1894  *
1895  * It actually works quite well. One needs to have the possibility for
1896  * at least one ".nfs..." file in each directory the file ever gets
1897  * moved or linked to which happens automagically with the new
1898  * implementation that only depends on the dcache stuff instead of
1899  * using the inode layer
1900  *
1901  * Unfortunately, things are a little more complicated than indicated
1902  * above. For a cross-directory move, we want to make sure we can get
1903  * rid of the old inode after the operation.  This means there must be
1904  * no pending writes (if it's a file), and the use count must be 1.
1905  * If these conditions are met, we can drop the dentries before doing
1906  * the rename.
1907  */
1908 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1909 		      struct inode *new_dir, struct dentry *new_dentry)
1910 {
1911 	struct inode *old_inode = old_dentry->d_inode;
1912 	struct inode *new_inode = new_dentry->d_inode;
1913 	struct dentry *dentry = NULL, *rehash = NULL;
1914 	int error = -EBUSY;
1915 
1916 	dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1917 		 old_dentry, new_dentry,
1918 		 d_count(new_dentry));
1919 
1920 	trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1921 	/*
1922 	 * For non-directories, check whether the target is busy and if so,
1923 	 * make a copy of the dentry and then do a silly-rename. If the
1924 	 * silly-rename succeeds, the copied dentry is hashed and becomes
1925 	 * the new target.
1926 	 */
1927 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1928 		/*
1929 		 * To prevent any new references to the target during the
1930 		 * rename, we unhash the dentry in advance.
1931 		 */
1932 		if (!d_unhashed(new_dentry)) {
1933 			d_drop(new_dentry);
1934 			rehash = new_dentry;
1935 		}
1936 
1937 		if (d_count(new_dentry) > 2) {
1938 			int err;
1939 
1940 			/* copy the target dentry's name */
1941 			dentry = d_alloc(new_dentry->d_parent,
1942 					 &new_dentry->d_name);
1943 			if (!dentry)
1944 				goto out;
1945 
1946 			/* silly-rename the existing target ... */
1947 			err = nfs_sillyrename(new_dir, new_dentry);
1948 			if (err)
1949 				goto out;
1950 
1951 			new_dentry = dentry;
1952 			rehash = NULL;
1953 			new_inode = NULL;
1954 		}
1955 	}
1956 
1957 	NFS_PROTO(old_inode)->return_delegation(old_inode);
1958 	if (new_inode != NULL)
1959 		NFS_PROTO(new_inode)->return_delegation(new_inode);
1960 
1961 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1962 					   new_dir, &new_dentry->d_name);
1963 	nfs_mark_for_revalidate(old_inode);
1964 out:
1965 	if (rehash)
1966 		d_rehash(rehash);
1967 	trace_nfs_rename_exit(old_dir, old_dentry,
1968 			new_dir, new_dentry, error);
1969 	if (!error) {
1970 		if (new_inode != NULL)
1971 			nfs_drop_nlink(new_inode);
1972 		d_move(old_dentry, new_dentry);
1973 		nfs_set_verifier(new_dentry,
1974 					nfs_save_change_attribute(new_dir));
1975 	} else if (error == -ENOENT)
1976 		nfs_dentry_handle_enoent(old_dentry);
1977 
1978 	/* new dentry created? */
1979 	if (dentry)
1980 		dput(dentry);
1981 	return error;
1982 }
1983 EXPORT_SYMBOL_GPL(nfs_rename);
1984 
1985 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1986 static LIST_HEAD(nfs_access_lru_list);
1987 static atomic_long_t nfs_access_nr_entries;
1988 
1989 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1990 {
1991 	put_rpccred(entry->cred);
1992 	kfree(entry);
1993 	smp_mb__before_atomic_dec();
1994 	atomic_long_dec(&nfs_access_nr_entries);
1995 	smp_mb__after_atomic_dec();
1996 }
1997 
1998 static void nfs_access_free_list(struct list_head *head)
1999 {
2000 	struct nfs_access_entry *cache;
2001 
2002 	while (!list_empty(head)) {
2003 		cache = list_entry(head->next, struct nfs_access_entry, lru);
2004 		list_del(&cache->lru);
2005 		nfs_access_free_entry(cache);
2006 	}
2007 }
2008 
2009 unsigned long
2010 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2011 {
2012 	LIST_HEAD(head);
2013 	struct nfs_inode *nfsi, *next;
2014 	struct nfs_access_entry *cache;
2015 	int nr_to_scan = sc->nr_to_scan;
2016 	gfp_t gfp_mask = sc->gfp_mask;
2017 	long freed = 0;
2018 
2019 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2020 		return SHRINK_STOP;
2021 
2022 	spin_lock(&nfs_access_lru_lock);
2023 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2024 		struct inode *inode;
2025 
2026 		if (nr_to_scan-- == 0)
2027 			break;
2028 		inode = &nfsi->vfs_inode;
2029 		spin_lock(&inode->i_lock);
2030 		if (list_empty(&nfsi->access_cache_entry_lru))
2031 			goto remove_lru_entry;
2032 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2033 				struct nfs_access_entry, lru);
2034 		list_move(&cache->lru, &head);
2035 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2036 		freed++;
2037 		if (!list_empty(&nfsi->access_cache_entry_lru))
2038 			list_move_tail(&nfsi->access_cache_inode_lru,
2039 					&nfs_access_lru_list);
2040 		else {
2041 remove_lru_entry:
2042 			list_del_init(&nfsi->access_cache_inode_lru);
2043 			smp_mb__before_clear_bit();
2044 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2045 			smp_mb__after_clear_bit();
2046 		}
2047 		spin_unlock(&inode->i_lock);
2048 	}
2049 	spin_unlock(&nfs_access_lru_lock);
2050 	nfs_access_free_list(&head);
2051 	return freed;
2052 }
2053 
2054 unsigned long
2055 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2056 {
2057 	return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2058 }
2059 
2060 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2061 {
2062 	struct rb_root *root_node = &nfsi->access_cache;
2063 	struct rb_node *n;
2064 	struct nfs_access_entry *entry;
2065 
2066 	/* Unhook entries from the cache */
2067 	while ((n = rb_first(root_node)) != NULL) {
2068 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2069 		rb_erase(n, root_node);
2070 		list_move(&entry->lru, head);
2071 	}
2072 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2073 }
2074 
2075 void nfs_access_zap_cache(struct inode *inode)
2076 {
2077 	LIST_HEAD(head);
2078 
2079 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2080 		return;
2081 	/* Remove from global LRU init */
2082 	spin_lock(&nfs_access_lru_lock);
2083 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2084 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2085 
2086 	spin_lock(&inode->i_lock);
2087 	__nfs_access_zap_cache(NFS_I(inode), &head);
2088 	spin_unlock(&inode->i_lock);
2089 	spin_unlock(&nfs_access_lru_lock);
2090 	nfs_access_free_list(&head);
2091 }
2092 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2093 
2094 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2095 {
2096 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2097 	struct nfs_access_entry *entry;
2098 
2099 	while (n != NULL) {
2100 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2101 
2102 		if (cred < entry->cred)
2103 			n = n->rb_left;
2104 		else if (cred > entry->cred)
2105 			n = n->rb_right;
2106 		else
2107 			return entry;
2108 	}
2109 	return NULL;
2110 }
2111 
2112 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2113 {
2114 	struct nfs_inode *nfsi = NFS_I(inode);
2115 	struct nfs_access_entry *cache;
2116 	int err = -ENOENT;
2117 
2118 	spin_lock(&inode->i_lock);
2119 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2120 		goto out_zap;
2121 	cache = nfs_access_search_rbtree(inode, cred);
2122 	if (cache == NULL)
2123 		goto out;
2124 	if (!nfs_have_delegated_attributes(inode) &&
2125 	    !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2126 		goto out_stale;
2127 	res->jiffies = cache->jiffies;
2128 	res->cred = cache->cred;
2129 	res->mask = cache->mask;
2130 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2131 	err = 0;
2132 out:
2133 	spin_unlock(&inode->i_lock);
2134 	return err;
2135 out_stale:
2136 	rb_erase(&cache->rb_node, &nfsi->access_cache);
2137 	list_del(&cache->lru);
2138 	spin_unlock(&inode->i_lock);
2139 	nfs_access_free_entry(cache);
2140 	return -ENOENT;
2141 out_zap:
2142 	spin_unlock(&inode->i_lock);
2143 	nfs_access_zap_cache(inode);
2144 	return -ENOENT;
2145 }
2146 
2147 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2148 {
2149 	struct nfs_inode *nfsi = NFS_I(inode);
2150 	struct rb_root *root_node = &nfsi->access_cache;
2151 	struct rb_node **p = &root_node->rb_node;
2152 	struct rb_node *parent = NULL;
2153 	struct nfs_access_entry *entry;
2154 
2155 	spin_lock(&inode->i_lock);
2156 	while (*p != NULL) {
2157 		parent = *p;
2158 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2159 
2160 		if (set->cred < entry->cred)
2161 			p = &parent->rb_left;
2162 		else if (set->cred > entry->cred)
2163 			p = &parent->rb_right;
2164 		else
2165 			goto found;
2166 	}
2167 	rb_link_node(&set->rb_node, parent, p);
2168 	rb_insert_color(&set->rb_node, root_node);
2169 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2170 	spin_unlock(&inode->i_lock);
2171 	return;
2172 found:
2173 	rb_replace_node(parent, &set->rb_node, root_node);
2174 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2175 	list_del(&entry->lru);
2176 	spin_unlock(&inode->i_lock);
2177 	nfs_access_free_entry(entry);
2178 }
2179 
2180 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2181 {
2182 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2183 	if (cache == NULL)
2184 		return;
2185 	RB_CLEAR_NODE(&cache->rb_node);
2186 	cache->jiffies = set->jiffies;
2187 	cache->cred = get_rpccred(set->cred);
2188 	cache->mask = set->mask;
2189 
2190 	nfs_access_add_rbtree(inode, cache);
2191 
2192 	/* Update accounting */
2193 	smp_mb__before_atomic_inc();
2194 	atomic_long_inc(&nfs_access_nr_entries);
2195 	smp_mb__after_atomic_inc();
2196 
2197 	/* Add inode to global LRU list */
2198 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2199 		spin_lock(&nfs_access_lru_lock);
2200 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2201 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2202 					&nfs_access_lru_list);
2203 		spin_unlock(&nfs_access_lru_lock);
2204 	}
2205 }
2206 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2207 
2208 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2209 {
2210 	entry->mask = 0;
2211 	if (access_result & NFS4_ACCESS_READ)
2212 		entry->mask |= MAY_READ;
2213 	if (access_result &
2214 	    (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2215 		entry->mask |= MAY_WRITE;
2216 	if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2217 		entry->mask |= MAY_EXEC;
2218 }
2219 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2220 
2221 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2222 {
2223 	struct nfs_access_entry cache;
2224 	int status;
2225 
2226 	trace_nfs_access_enter(inode);
2227 
2228 	status = nfs_access_get_cached(inode, cred, &cache);
2229 	if (status == 0)
2230 		goto out_cached;
2231 
2232 	/* Be clever: ask server to check for all possible rights */
2233 	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2234 	cache.cred = cred;
2235 	cache.jiffies = jiffies;
2236 	status = NFS_PROTO(inode)->access(inode, &cache);
2237 	if (status != 0) {
2238 		if (status == -ESTALE) {
2239 			nfs_zap_caches(inode);
2240 			if (!S_ISDIR(inode->i_mode))
2241 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2242 		}
2243 		goto out;
2244 	}
2245 	nfs_access_add_cache(inode, &cache);
2246 out_cached:
2247 	if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2248 		status = -EACCES;
2249 out:
2250 	trace_nfs_access_exit(inode, status);
2251 	return status;
2252 }
2253 
2254 static int nfs_open_permission_mask(int openflags)
2255 {
2256 	int mask = 0;
2257 
2258 	if (openflags & __FMODE_EXEC) {
2259 		/* ONLY check exec rights */
2260 		mask = MAY_EXEC;
2261 	} else {
2262 		if ((openflags & O_ACCMODE) != O_WRONLY)
2263 			mask |= MAY_READ;
2264 		if ((openflags & O_ACCMODE) != O_RDONLY)
2265 			mask |= MAY_WRITE;
2266 	}
2267 
2268 	return mask;
2269 }
2270 
2271 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2272 {
2273 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2274 }
2275 EXPORT_SYMBOL_GPL(nfs_may_open);
2276 
2277 int nfs_permission(struct inode *inode, int mask)
2278 {
2279 	struct rpc_cred *cred;
2280 	int res = 0;
2281 
2282 	if (mask & MAY_NOT_BLOCK)
2283 		return -ECHILD;
2284 
2285 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2286 
2287 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2288 		goto out;
2289 	/* Is this sys_access() ? */
2290 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2291 		goto force_lookup;
2292 
2293 	switch (inode->i_mode & S_IFMT) {
2294 		case S_IFLNK:
2295 			goto out;
2296 		case S_IFREG:
2297 			break;
2298 		case S_IFDIR:
2299 			/*
2300 			 * Optimize away all write operations, since the server
2301 			 * will check permissions when we perform the op.
2302 			 */
2303 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2304 				goto out;
2305 	}
2306 
2307 force_lookup:
2308 	if (!NFS_PROTO(inode)->access)
2309 		goto out_notsup;
2310 
2311 	cred = rpc_lookup_cred();
2312 	if (!IS_ERR(cred)) {
2313 		res = nfs_do_access(inode, cred, mask);
2314 		put_rpccred(cred);
2315 	} else
2316 		res = PTR_ERR(cred);
2317 out:
2318 	if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2319 		res = -EACCES;
2320 
2321 	dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2322 		inode->i_sb->s_id, inode->i_ino, mask, res);
2323 	return res;
2324 out_notsup:
2325 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2326 	if (res == 0)
2327 		res = generic_permission(inode, mask);
2328 	goto out;
2329 }
2330 EXPORT_SYMBOL_GPL(nfs_permission);
2331 
2332 /*
2333  * Local variables:
2334  *  version-control: t
2335  *  kept-new-versions: 5
2336  * End:
2337  */
2338