xref: /openbmc/linux/fs/nfs/dir.c (revision 9b9c2cd4)
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_inode *nfsi = NFS_I(dir);
73 	struct nfs_open_dir_context *ctx;
74 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
75 	if (ctx != NULL) {
76 		ctx->duped = 0;
77 		ctx->attr_gencount = nfsi->attr_gencount;
78 		ctx->dir_cookie = 0;
79 		ctx->dup_cookie = 0;
80 		ctx->cred = get_rpccred(cred);
81 		spin_lock(&dir->i_lock);
82 		list_add(&ctx->list, &nfsi->open_files);
83 		spin_unlock(&dir->i_lock);
84 		return ctx;
85 	}
86 	return  ERR_PTR(-ENOMEM);
87 }
88 
89 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
90 {
91 	spin_lock(&dir->i_lock);
92 	list_del(&ctx->list);
93 	spin_unlock(&dir->i_lock);
94 	put_rpccred(ctx->cred);
95 	kfree(ctx);
96 }
97 
98 /*
99  * Open file
100  */
101 static int
102 nfs_opendir(struct inode *inode, struct file *filp)
103 {
104 	int res = 0;
105 	struct nfs_open_dir_context *ctx;
106 	struct rpc_cred *cred;
107 
108 	dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
109 
110 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
111 
112 	cred = rpc_lookup_cred();
113 	if (IS_ERR(cred))
114 		return PTR_ERR(cred);
115 	ctx = alloc_nfs_open_dir_context(inode, cred);
116 	if (IS_ERR(ctx)) {
117 		res = PTR_ERR(ctx);
118 		goto out;
119 	}
120 	filp->private_data = ctx;
121 	if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
122 		/* This is a mountpoint, so d_revalidate will never
123 		 * have been called, so we need to refresh the
124 		 * inode (for close-open consistency) ourselves.
125 		 */
126 		__nfs_revalidate_inode(NFS_SERVER(inode), inode);
127 	}
128 out:
129 	put_rpccred(cred);
130 	return res;
131 }
132 
133 static int
134 nfs_closedir(struct inode *inode, struct file *filp)
135 {
136 	put_nfs_open_dir_context(file_inode(filp), filp->private_data);
137 	return 0;
138 }
139 
140 struct nfs_cache_array_entry {
141 	u64 cookie;
142 	u64 ino;
143 	struct qstr string;
144 	unsigned char d_type;
145 };
146 
147 struct nfs_cache_array {
148 	int size;
149 	int eof_index;
150 	u64 last_cookie;
151 	struct nfs_cache_array_entry array[0];
152 };
153 
154 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
155 typedef struct {
156 	struct file	*file;
157 	struct page	*page;
158 	struct dir_context *ctx;
159 	unsigned long	page_index;
160 	u64		*dir_cookie;
161 	u64		last_cookie;
162 	loff_t		current_index;
163 	decode_dirent_t	decode;
164 
165 	unsigned long	timestamp;
166 	unsigned long	gencount;
167 	unsigned int	cache_entry_index;
168 	unsigned int	plus:1;
169 	unsigned int	eof:1;
170 } nfs_readdir_descriptor_t;
171 
172 /*
173  * The caller is responsible for calling nfs_readdir_release_array(page)
174  */
175 static
176 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
177 {
178 	void *ptr;
179 	if (page == NULL)
180 		return ERR_PTR(-EIO);
181 	ptr = kmap(page);
182 	if (ptr == NULL)
183 		return ERR_PTR(-ENOMEM);
184 	return ptr;
185 }
186 
187 static
188 void nfs_readdir_release_array(struct page *page)
189 {
190 	kunmap(page);
191 }
192 
193 /*
194  * we are freeing strings created by nfs_add_to_readdir_array()
195  */
196 static
197 void nfs_readdir_clear_array(struct page *page)
198 {
199 	struct nfs_cache_array *array;
200 	int i;
201 
202 	array = kmap_atomic(page);
203 	for (i = 0; i < array->size; i++)
204 		kfree(array->array[i].string.name);
205 	kunmap_atomic(array);
206 }
207 
208 /*
209  * the caller is responsible for freeing qstr.name
210  * when called by nfs_readdir_add_to_array, the strings will be freed in
211  * nfs_clear_readdir_array()
212  */
213 static
214 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
215 {
216 	string->len = len;
217 	string->name = kmemdup(name, len, GFP_KERNEL);
218 	if (string->name == NULL)
219 		return -ENOMEM;
220 	/*
221 	 * Avoid a kmemleak false positive. The pointer to the name is stored
222 	 * in a page cache page which kmemleak does not scan.
223 	 */
224 	kmemleak_not_leak(string->name);
225 	string->hash = full_name_hash(name, len);
226 	return 0;
227 }
228 
229 static
230 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
231 {
232 	struct nfs_cache_array *array = nfs_readdir_get_array(page);
233 	struct nfs_cache_array_entry *cache_entry;
234 	int ret;
235 
236 	if (IS_ERR(array))
237 		return PTR_ERR(array);
238 
239 	cache_entry = &array->array[array->size];
240 
241 	/* Check that this entry lies within the page bounds */
242 	ret = -ENOSPC;
243 	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
244 		goto out;
245 
246 	cache_entry->cookie = entry->prev_cookie;
247 	cache_entry->ino = entry->ino;
248 	cache_entry->d_type = entry->d_type;
249 	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
250 	if (ret)
251 		goto out;
252 	array->last_cookie = entry->cookie;
253 	array->size++;
254 	if (entry->eof != 0)
255 		array->eof_index = array->size;
256 out:
257 	nfs_readdir_release_array(page);
258 	return ret;
259 }
260 
261 static
262 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
263 {
264 	loff_t diff = desc->ctx->pos - desc->current_index;
265 	unsigned int index;
266 
267 	if (diff < 0)
268 		goto out_eof;
269 	if (diff >= array->size) {
270 		if (array->eof_index >= 0)
271 			goto out_eof;
272 		return -EAGAIN;
273 	}
274 
275 	index = (unsigned int)diff;
276 	*desc->dir_cookie = array->array[index].cookie;
277 	desc->cache_entry_index = index;
278 	return 0;
279 out_eof:
280 	desc->eof = 1;
281 	return -EBADCOOKIE;
282 }
283 
284 static bool
285 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
286 {
287 	if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
288 		return false;
289 	smp_rmb();
290 	return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
291 }
292 
293 static
294 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
295 {
296 	int i;
297 	loff_t new_pos;
298 	int status = -EAGAIN;
299 
300 	for (i = 0; i < array->size; i++) {
301 		if (array->array[i].cookie == *desc->dir_cookie) {
302 			struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
303 			struct nfs_open_dir_context *ctx = desc->file->private_data;
304 
305 			new_pos = desc->current_index + i;
306 			if (ctx->attr_gencount != nfsi->attr_gencount ||
307 			    !nfs_readdir_inode_mapping_valid(nfsi)) {
308 				ctx->duped = 0;
309 				ctx->attr_gencount = nfsi->attr_gencount;
310 			} else if (new_pos < desc->ctx->pos) {
311 				if (ctx->duped > 0
312 				    && ctx->dup_cookie == *desc->dir_cookie) {
313 					if (printk_ratelimit()) {
314 						pr_notice("NFS: directory %pD2 contains a readdir loop."
315 								"Please contact your server vendor.  "
316 								"The file: %.*s has duplicate cookie %llu\n",
317 								desc->file, array->array[i].string.len,
318 								array->array[i].string.name, *desc->dir_cookie);
319 					}
320 					status = -ELOOP;
321 					goto out;
322 				}
323 				ctx->dup_cookie = *desc->dir_cookie;
324 				ctx->duped = -1;
325 			}
326 			desc->ctx->pos = new_pos;
327 			desc->cache_entry_index = i;
328 			return 0;
329 		}
330 	}
331 	if (array->eof_index >= 0) {
332 		status = -EBADCOOKIE;
333 		if (*desc->dir_cookie == array->last_cookie)
334 			desc->eof = 1;
335 	}
336 out:
337 	return status;
338 }
339 
340 static
341 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
342 {
343 	struct nfs_cache_array *array;
344 	int status;
345 
346 	array = nfs_readdir_get_array(desc->page);
347 	if (IS_ERR(array)) {
348 		status = PTR_ERR(array);
349 		goto out;
350 	}
351 
352 	if (*desc->dir_cookie == 0)
353 		status = nfs_readdir_search_for_pos(array, desc);
354 	else
355 		status = nfs_readdir_search_for_cookie(array, desc);
356 
357 	if (status == -EAGAIN) {
358 		desc->last_cookie = array->last_cookie;
359 		desc->current_index += array->size;
360 		desc->page_index++;
361 	}
362 	nfs_readdir_release_array(desc->page);
363 out:
364 	return status;
365 }
366 
367 /* Fill a page with xdr information before transferring to the cache page */
368 static
369 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
370 			struct nfs_entry *entry, struct file *file, struct inode *inode)
371 {
372 	struct nfs_open_dir_context *ctx = file->private_data;
373 	struct rpc_cred	*cred = ctx->cred;
374 	unsigned long	timestamp, gencount;
375 	int		error;
376 
377  again:
378 	timestamp = jiffies;
379 	gencount = nfs_inc_attr_generation_counter();
380 	error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
381 					  NFS_SERVER(inode)->dtsize, desc->plus);
382 	if (error < 0) {
383 		/* We requested READDIRPLUS, but the server doesn't grok it */
384 		if (error == -ENOTSUPP && desc->plus) {
385 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
386 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
387 			desc->plus = 0;
388 			goto again;
389 		}
390 		goto error;
391 	}
392 	desc->timestamp = timestamp;
393 	desc->gencount = gencount;
394 error:
395 	return error;
396 }
397 
398 static int xdr_decode(nfs_readdir_descriptor_t *desc,
399 		      struct nfs_entry *entry, struct xdr_stream *xdr)
400 {
401 	int error;
402 
403 	error = desc->decode(xdr, entry, desc->plus);
404 	if (error)
405 		return error;
406 	entry->fattr->time_start = desc->timestamp;
407 	entry->fattr->gencount = desc->gencount;
408 	return 0;
409 }
410 
411 /* Match file and dirent using either filehandle or fileid
412  * Note: caller is responsible for checking the fsid
413  */
414 static
415 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
416 {
417 	struct nfs_inode *nfsi;
418 
419 	if (d_really_is_negative(dentry))
420 		return 0;
421 
422 	nfsi = NFS_I(d_inode(dentry));
423 	if (entry->fattr->fileid == nfsi->fileid)
424 		return 1;
425 	if (nfs_compare_fh(entry->fh, &nfsi->fh) == 0)
426 		return 1;
427 	return 0;
428 }
429 
430 static
431 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
432 {
433 	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
434 		return false;
435 	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
436 		return true;
437 	if (ctx->pos == 0)
438 		return true;
439 	return false;
440 }
441 
442 /*
443  * This function is called by the lookup code to request the use of
444  * readdirplus to accelerate any future lookups in the same
445  * directory.
446  */
447 static
448 void nfs_advise_use_readdirplus(struct inode *dir)
449 {
450 	set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
451 }
452 
453 /*
454  * This function is mainly for use by nfs_getattr().
455  *
456  * If this is an 'ls -l', we want to force use of readdirplus.
457  * Do this by checking if there is an active file descriptor
458  * and calling nfs_advise_use_readdirplus, then forcing a
459  * cache flush.
460  */
461 void nfs_force_use_readdirplus(struct inode *dir)
462 {
463 	if (!list_empty(&NFS_I(dir)->open_files)) {
464 		nfs_advise_use_readdirplus(dir);
465 		nfs_zap_mapping(dir, dir->i_mapping);
466 	}
467 }
468 
469 static
470 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
471 {
472 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
473 	struct dentry *dentry;
474 	struct dentry *alias;
475 	struct inode *dir = d_inode(parent);
476 	struct inode *inode;
477 	int status;
478 
479 	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
480 		return;
481 	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
482 		return;
483 	if (filename.name[0] == '.') {
484 		if (filename.len == 1)
485 			return;
486 		if (filename.len == 2 && filename.name[1] == '.')
487 			return;
488 	}
489 	filename.hash = full_name_hash(filename.name, filename.len);
490 
491 	dentry = d_lookup(parent, &filename);
492 	if (dentry != NULL) {
493 		/* Is there a mountpoint here? If so, just exit */
494 		if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
495 					&entry->fattr->fsid))
496 			goto out;
497 		if (nfs_same_file(dentry, entry)) {
498 			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
499 			status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
500 			if (!status)
501 				nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
502 			goto out;
503 		} else {
504 			d_invalidate(dentry);
505 			dput(dentry);
506 		}
507 	}
508 
509 	dentry = d_alloc(parent, &filename);
510 	if (dentry == NULL)
511 		return;
512 
513 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
514 	if (IS_ERR(inode))
515 		goto out;
516 
517 	alias = d_splice_alias(inode, dentry);
518 	if (IS_ERR(alias))
519 		goto out;
520 	else if (alias) {
521 		nfs_set_verifier(alias, nfs_save_change_attribute(dir));
522 		dput(alias);
523 	} else
524 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
525 
526 out:
527 	dput(dentry);
528 }
529 
530 /* Perform conversion from xdr to cache array */
531 static
532 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
533 				struct page **xdr_pages, struct page *page, unsigned int buflen)
534 {
535 	struct xdr_stream stream;
536 	struct xdr_buf buf;
537 	struct page *scratch;
538 	struct nfs_cache_array *array;
539 	unsigned int count = 0;
540 	int status;
541 
542 	scratch = alloc_page(GFP_KERNEL);
543 	if (scratch == NULL)
544 		return -ENOMEM;
545 
546 	if (buflen == 0)
547 		goto out_nopages;
548 
549 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
550 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
551 
552 	do {
553 		status = xdr_decode(desc, entry, &stream);
554 		if (status != 0) {
555 			if (status == -EAGAIN)
556 				status = 0;
557 			break;
558 		}
559 
560 		count++;
561 
562 		if (desc->plus != 0)
563 			nfs_prime_dcache(desc->file->f_path.dentry, entry);
564 
565 		status = nfs_readdir_add_to_array(entry, page);
566 		if (status != 0)
567 			break;
568 	} while (!entry->eof);
569 
570 out_nopages:
571 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
572 		array = nfs_readdir_get_array(page);
573 		if (!IS_ERR(array)) {
574 			array->eof_index = array->size;
575 			status = 0;
576 			nfs_readdir_release_array(page);
577 		} else
578 			status = PTR_ERR(array);
579 	}
580 
581 	put_page(scratch);
582 	return status;
583 }
584 
585 static
586 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
587 {
588 	unsigned int i;
589 	for (i = 0; i < npages; i++)
590 		put_page(pages[i]);
591 }
592 
593 /*
594  * nfs_readdir_large_page will allocate pages that must be freed with a call
595  * to nfs_readdir_free_pagearray
596  */
597 static
598 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
599 {
600 	unsigned int i;
601 
602 	for (i = 0; i < npages; i++) {
603 		struct page *page = alloc_page(GFP_KERNEL);
604 		if (page == NULL)
605 			goto out_freepages;
606 		pages[i] = page;
607 	}
608 	return 0;
609 
610 out_freepages:
611 	nfs_readdir_free_pages(pages, i);
612 	return -ENOMEM;
613 }
614 
615 static
616 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
617 {
618 	struct page *pages[NFS_MAX_READDIR_PAGES];
619 	struct nfs_entry entry;
620 	struct file	*file = desc->file;
621 	struct nfs_cache_array *array;
622 	int status = -ENOMEM;
623 	unsigned int array_size = ARRAY_SIZE(pages);
624 
625 	entry.prev_cookie = 0;
626 	entry.cookie = desc->last_cookie;
627 	entry.eof = 0;
628 	entry.fh = nfs_alloc_fhandle();
629 	entry.fattr = nfs_alloc_fattr();
630 	entry.server = NFS_SERVER(inode);
631 	if (entry.fh == NULL || entry.fattr == NULL)
632 		goto out;
633 
634 	entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
635 	if (IS_ERR(entry.label)) {
636 		status = PTR_ERR(entry.label);
637 		goto out;
638 	}
639 
640 	array = nfs_readdir_get_array(page);
641 	if (IS_ERR(array)) {
642 		status = PTR_ERR(array);
643 		goto out_label_free;
644 	}
645 	memset(array, 0, sizeof(struct nfs_cache_array));
646 	array->eof_index = -1;
647 
648 	status = nfs_readdir_alloc_pages(pages, array_size);
649 	if (status < 0)
650 		goto out_release_array;
651 	do {
652 		unsigned int pglen;
653 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
654 
655 		if (status < 0)
656 			break;
657 		pglen = status;
658 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
659 		if (status < 0) {
660 			if (status == -ENOSPC)
661 				status = 0;
662 			break;
663 		}
664 	} while (array->eof_index < 0);
665 
666 	nfs_readdir_free_pages(pages, array_size);
667 out_release_array:
668 	nfs_readdir_release_array(page);
669 out_label_free:
670 	nfs4_label_free(entry.label);
671 out:
672 	nfs_free_fattr(entry.fattr);
673 	nfs_free_fhandle(entry.fh);
674 	return status;
675 }
676 
677 /*
678  * Now we cache directories properly, by converting xdr information
679  * to an array that can be used for lookups later.  This results in
680  * fewer cache pages, since we can store more information on each page.
681  * We only need to convert from xdr once so future lookups are much simpler
682  */
683 static
684 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
685 {
686 	struct inode	*inode = file_inode(desc->file);
687 	int ret;
688 
689 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
690 	if (ret < 0)
691 		goto error;
692 	SetPageUptodate(page);
693 
694 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
695 		/* Should never happen */
696 		nfs_zap_mapping(inode, inode->i_mapping);
697 	}
698 	unlock_page(page);
699 	return 0;
700  error:
701 	unlock_page(page);
702 	return ret;
703 }
704 
705 static
706 void cache_page_release(nfs_readdir_descriptor_t *desc)
707 {
708 	if (!desc->page->mapping)
709 		nfs_readdir_clear_array(desc->page);
710 	page_cache_release(desc->page);
711 	desc->page = NULL;
712 }
713 
714 static
715 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
716 {
717 	return read_cache_page(file_inode(desc->file)->i_mapping,
718 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
719 }
720 
721 /*
722  * Returns 0 if desc->dir_cookie was found on page desc->page_index
723  */
724 static
725 int find_cache_page(nfs_readdir_descriptor_t *desc)
726 {
727 	int res;
728 
729 	desc->page = get_cache_page(desc);
730 	if (IS_ERR(desc->page))
731 		return PTR_ERR(desc->page);
732 
733 	res = nfs_readdir_search_array(desc);
734 	if (res != 0)
735 		cache_page_release(desc);
736 	return res;
737 }
738 
739 /* Search for desc->dir_cookie from the beginning of the page cache */
740 static inline
741 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
742 {
743 	int res;
744 
745 	if (desc->page_index == 0) {
746 		desc->current_index = 0;
747 		desc->last_cookie = 0;
748 	}
749 	do {
750 		res = find_cache_page(desc);
751 	} while (res == -EAGAIN);
752 	return res;
753 }
754 
755 /*
756  * Once we've found the start of the dirent within a page: fill 'er up...
757  */
758 static
759 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
760 {
761 	struct file	*file = desc->file;
762 	int i = 0;
763 	int res = 0;
764 	struct nfs_cache_array *array = NULL;
765 	struct nfs_open_dir_context *ctx = file->private_data;
766 
767 	array = nfs_readdir_get_array(desc->page);
768 	if (IS_ERR(array)) {
769 		res = PTR_ERR(array);
770 		goto out;
771 	}
772 
773 	for (i = desc->cache_entry_index; i < array->size; i++) {
774 		struct nfs_cache_array_entry *ent;
775 
776 		ent = &array->array[i];
777 		if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
778 		    nfs_compat_user_ino64(ent->ino), ent->d_type)) {
779 			desc->eof = 1;
780 			break;
781 		}
782 		desc->ctx->pos++;
783 		if (i < (array->size-1))
784 			*desc->dir_cookie = array->array[i+1].cookie;
785 		else
786 			*desc->dir_cookie = array->last_cookie;
787 		if (ctx->duped != 0)
788 			ctx->duped = 1;
789 	}
790 	if (array->eof_index >= 0)
791 		desc->eof = 1;
792 
793 	nfs_readdir_release_array(desc->page);
794 out:
795 	cache_page_release(desc);
796 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
797 			(unsigned long long)*desc->dir_cookie, res);
798 	return res;
799 }
800 
801 /*
802  * If we cannot find a cookie in our cache, we suspect that this is
803  * because it points to a deleted file, so we ask the server to return
804  * whatever it thinks is the next entry. We then feed this to filldir.
805  * If all goes well, we should then be able to find our way round the
806  * cache on the next call to readdir_search_pagecache();
807  *
808  * NOTE: we cannot add the anonymous page to the pagecache because
809  *	 the data it contains might not be page aligned. Besides,
810  *	 we should already have a complete representation of the
811  *	 directory in the page cache by the time we get here.
812  */
813 static inline
814 int uncached_readdir(nfs_readdir_descriptor_t *desc)
815 {
816 	struct page	*page = NULL;
817 	int		status;
818 	struct inode *inode = file_inode(desc->file);
819 	struct nfs_open_dir_context *ctx = desc->file->private_data;
820 
821 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
822 			(unsigned long long)*desc->dir_cookie);
823 
824 	page = alloc_page(GFP_HIGHUSER);
825 	if (!page) {
826 		status = -ENOMEM;
827 		goto out;
828 	}
829 
830 	desc->page_index = 0;
831 	desc->last_cookie = *desc->dir_cookie;
832 	desc->page = page;
833 	ctx->duped = 0;
834 
835 	status = nfs_readdir_xdr_to_array(desc, page, inode);
836 	if (status < 0)
837 		goto out_release;
838 
839 	status = nfs_do_filldir(desc);
840 
841  out:
842 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
843 			__func__, status);
844 	return status;
845  out_release:
846 	cache_page_release(desc);
847 	goto out;
848 }
849 
850 static bool nfs_dir_mapping_need_revalidate(struct inode *dir)
851 {
852 	struct nfs_inode *nfsi = NFS_I(dir);
853 
854 	if (nfs_attribute_cache_expired(dir))
855 		return true;
856 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
857 		return true;
858 	return false;
859 }
860 
861 /* The file offset position represents the dirent entry number.  A
862    last cookie cache takes care of the common case of reading the
863    whole directory.
864  */
865 static int nfs_readdir(struct file *file, struct dir_context *ctx)
866 {
867 	struct dentry	*dentry = file->f_path.dentry;
868 	struct inode	*inode = d_inode(dentry);
869 	nfs_readdir_descriptor_t my_desc,
870 			*desc = &my_desc;
871 	struct nfs_open_dir_context *dir_ctx = file->private_data;
872 	int res = 0;
873 
874 	dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
875 			file, (long long)ctx->pos);
876 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
877 
878 	/*
879 	 * ctx->pos points to the dirent entry number.
880 	 * *desc->dir_cookie has the cookie for the next entry. We have
881 	 * to either find the entry with the appropriate number or
882 	 * revalidate the cookie.
883 	 */
884 	memset(desc, 0, sizeof(*desc));
885 
886 	desc->file = file;
887 	desc->ctx = ctx;
888 	desc->dir_cookie = &dir_ctx->dir_cookie;
889 	desc->decode = NFS_PROTO(inode)->decode_dirent;
890 	desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
891 
892 	nfs_block_sillyrename(dentry);
893 	if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode))
894 		res = nfs_revalidate_mapping(inode, file->f_mapping);
895 	if (res < 0)
896 		goto out;
897 
898 	do {
899 		res = readdir_search_pagecache(desc);
900 
901 		if (res == -EBADCOOKIE) {
902 			res = 0;
903 			/* This means either end of directory */
904 			if (*desc->dir_cookie && desc->eof == 0) {
905 				/* Or that the server has 'lost' a cookie */
906 				res = uncached_readdir(desc);
907 				if (res == 0)
908 					continue;
909 			}
910 			break;
911 		}
912 		if (res == -ETOOSMALL && desc->plus) {
913 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
914 			nfs_zap_caches(inode);
915 			desc->page_index = 0;
916 			desc->plus = 0;
917 			desc->eof = 0;
918 			continue;
919 		}
920 		if (res < 0)
921 			break;
922 
923 		res = nfs_do_filldir(desc);
924 		if (res < 0)
925 			break;
926 	} while (!desc->eof);
927 out:
928 	nfs_unblock_sillyrename(dentry);
929 	if (res > 0)
930 		res = 0;
931 	dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
932 	return res;
933 }
934 
935 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
936 {
937 	struct inode *inode = file_inode(filp);
938 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
939 
940 	dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
941 			filp, offset, whence);
942 
943 	mutex_lock(&inode->i_mutex);
944 	switch (whence) {
945 		case 1:
946 			offset += filp->f_pos;
947 		case 0:
948 			if (offset >= 0)
949 				break;
950 		default:
951 			offset = -EINVAL;
952 			goto out;
953 	}
954 	if (offset != filp->f_pos) {
955 		filp->f_pos = offset;
956 		dir_ctx->dir_cookie = 0;
957 		dir_ctx->duped = 0;
958 	}
959 out:
960 	mutex_unlock(&inode->i_mutex);
961 	return offset;
962 }
963 
964 /*
965  * All directory operations under NFS are synchronous, so fsync()
966  * is a dummy operation.
967  */
968 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
969 			 int datasync)
970 {
971 	struct inode *inode = file_inode(filp);
972 
973 	dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
974 
975 	mutex_lock(&inode->i_mutex);
976 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
977 	mutex_unlock(&inode->i_mutex);
978 	return 0;
979 }
980 
981 /**
982  * nfs_force_lookup_revalidate - Mark the directory as having changed
983  * @dir - pointer to directory inode
984  *
985  * This forces the revalidation code in nfs_lookup_revalidate() to do a
986  * full lookup on all child dentries of 'dir' whenever a change occurs
987  * on the server that might have invalidated our dcache.
988  *
989  * The caller should be holding dir->i_lock
990  */
991 void nfs_force_lookup_revalidate(struct inode *dir)
992 {
993 	NFS_I(dir)->cache_change_attribute++;
994 }
995 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
996 
997 /*
998  * A check for whether or not the parent directory has changed.
999  * In the case it has, we assume that the dentries are untrustworthy
1000  * and may need to be looked up again.
1001  * If rcu_walk prevents us from performing a full check, return 0.
1002  */
1003 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1004 			      int rcu_walk)
1005 {
1006 	int ret;
1007 
1008 	if (IS_ROOT(dentry))
1009 		return 1;
1010 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1011 		return 0;
1012 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1013 		return 0;
1014 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1015 	if (rcu_walk)
1016 		ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir);
1017 	else
1018 		ret = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1019 	if (ret < 0)
1020 		return 0;
1021 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1022 		return 0;
1023 	return 1;
1024 }
1025 
1026 /*
1027  * Use intent information to check whether or not we're going to do
1028  * an O_EXCL create using this path component.
1029  */
1030 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1031 {
1032 	if (NFS_PROTO(dir)->version == 2)
1033 		return 0;
1034 	return flags & LOOKUP_EXCL;
1035 }
1036 
1037 /*
1038  * Inode and filehandle revalidation for lookups.
1039  *
1040  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1041  * or if the intent information indicates that we're about to open this
1042  * particular file and the "nocto" mount flag is not set.
1043  *
1044  */
1045 static
1046 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1047 {
1048 	struct nfs_server *server = NFS_SERVER(inode);
1049 	int ret;
1050 
1051 	if (IS_AUTOMOUNT(inode))
1052 		return 0;
1053 	/* VFS wants an on-the-wire revalidation */
1054 	if (flags & LOOKUP_REVAL)
1055 		goto out_force;
1056 	/* This is an open(2) */
1057 	if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1058 	    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1059 		goto out_force;
1060 out:
1061 	return (inode->i_nlink == 0) ? -ENOENT : 0;
1062 out_force:
1063 	if (flags & LOOKUP_RCU)
1064 		return -ECHILD;
1065 	ret = __nfs_revalidate_inode(server, inode);
1066 	if (ret != 0)
1067 		return ret;
1068 	goto out;
1069 }
1070 
1071 /*
1072  * We judge how long we want to trust negative
1073  * dentries by looking at the parent inode mtime.
1074  *
1075  * If parent mtime has changed, we revalidate, else we wait for a
1076  * period corresponding to the parent's attribute cache timeout value.
1077  *
1078  * If LOOKUP_RCU prevents us from performing a full check, return 1
1079  * suggesting a reval is needed.
1080  */
1081 static inline
1082 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1083 		       unsigned int flags)
1084 {
1085 	/* Don't revalidate a negative dentry if we're creating a new file */
1086 	if (flags & LOOKUP_CREATE)
1087 		return 0;
1088 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1089 		return 1;
1090 	return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1091 }
1092 
1093 /*
1094  * This is called every time the dcache has a lookup hit,
1095  * and we should check whether we can really trust that
1096  * lookup.
1097  *
1098  * NOTE! The hit can be a negative hit too, don't assume
1099  * we have an inode!
1100  *
1101  * If the parent directory is seen to have changed, we throw out the
1102  * cached dentry and do a new lookup.
1103  */
1104 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1105 {
1106 	struct inode *dir;
1107 	struct inode *inode;
1108 	struct dentry *parent;
1109 	struct nfs_fh *fhandle = NULL;
1110 	struct nfs_fattr *fattr = NULL;
1111 	struct nfs4_label *label = NULL;
1112 	int error;
1113 
1114 	if (flags & LOOKUP_RCU) {
1115 		parent = ACCESS_ONCE(dentry->d_parent);
1116 		dir = d_inode_rcu(parent);
1117 		if (!dir)
1118 			return -ECHILD;
1119 	} else {
1120 		parent = dget_parent(dentry);
1121 		dir = d_inode(parent);
1122 	}
1123 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1124 	inode = d_inode(dentry);
1125 
1126 	if (!inode) {
1127 		if (nfs_neg_need_reval(dir, dentry, flags)) {
1128 			if (flags & LOOKUP_RCU)
1129 				return -ECHILD;
1130 			goto out_bad;
1131 		}
1132 		goto out_valid_noent;
1133 	}
1134 
1135 	if (is_bad_inode(inode)) {
1136 		if (flags & LOOKUP_RCU)
1137 			return -ECHILD;
1138 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1139 				__func__, dentry);
1140 		goto out_bad;
1141 	}
1142 
1143 	if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1144 		goto out_set_verifier;
1145 
1146 	/* Force a full look up iff the parent directory has changed */
1147 	if (!nfs_is_exclusive_create(dir, flags) &&
1148 	    nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1149 
1150 		if (nfs_lookup_verify_inode(inode, flags)) {
1151 			if (flags & LOOKUP_RCU)
1152 				return -ECHILD;
1153 			goto out_zap_parent;
1154 		}
1155 		goto out_valid;
1156 	}
1157 
1158 	if (flags & LOOKUP_RCU)
1159 		return -ECHILD;
1160 
1161 	if (NFS_STALE(inode))
1162 		goto out_bad;
1163 
1164 	error = -ENOMEM;
1165 	fhandle = nfs_alloc_fhandle();
1166 	fattr = nfs_alloc_fattr();
1167 	if (fhandle == NULL || fattr == NULL)
1168 		goto out_error;
1169 
1170 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1171 	if (IS_ERR(label))
1172 		goto out_error;
1173 
1174 	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1175 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1176 	trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1177 	if (error)
1178 		goto out_bad;
1179 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1180 		goto out_bad;
1181 	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1182 		goto out_bad;
1183 
1184 	nfs_setsecurity(inode, fattr, label);
1185 
1186 	nfs_free_fattr(fattr);
1187 	nfs_free_fhandle(fhandle);
1188 	nfs4_label_free(label);
1189 
1190 out_set_verifier:
1191 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1192  out_valid:
1193 	/* Success: notify readdir to use READDIRPLUS */
1194 	nfs_advise_use_readdirplus(dir);
1195  out_valid_noent:
1196 	if (flags & LOOKUP_RCU) {
1197 		if (parent != ACCESS_ONCE(dentry->d_parent))
1198 			return -ECHILD;
1199 	} else
1200 		dput(parent);
1201 	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1202 			__func__, dentry);
1203 	return 1;
1204 out_zap_parent:
1205 	nfs_zap_caches(dir);
1206  out_bad:
1207 	WARN_ON(flags & LOOKUP_RCU);
1208 	nfs_free_fattr(fattr);
1209 	nfs_free_fhandle(fhandle);
1210 	nfs4_label_free(label);
1211 	nfs_mark_for_revalidate(dir);
1212 	if (inode && S_ISDIR(inode->i_mode)) {
1213 		/* Purge readdir caches. */
1214 		nfs_zap_caches(inode);
1215 		/*
1216 		 * We can't d_drop the root of a disconnected tree:
1217 		 * its d_hash is on the s_anon list and d_drop() would hide
1218 		 * it from shrink_dcache_for_unmount(), leading to busy
1219 		 * inodes on unmount and further oopses.
1220 		 */
1221 		if (IS_ROOT(dentry))
1222 			goto out_valid;
1223 	}
1224 	dput(parent);
1225 	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1226 			__func__, dentry);
1227 	return 0;
1228 out_error:
1229 	WARN_ON(flags & LOOKUP_RCU);
1230 	nfs_free_fattr(fattr);
1231 	nfs_free_fhandle(fhandle);
1232 	nfs4_label_free(label);
1233 	dput(parent);
1234 	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1235 			__func__, dentry, error);
1236 	return error;
1237 }
1238 
1239 /*
1240  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1241  * when we don't really care about the dentry name. This is called when a
1242  * pathwalk ends on a dentry that was not found via a normal lookup in the
1243  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1244  *
1245  * In this situation, we just want to verify that the inode itself is OK
1246  * since the dentry might have changed on the server.
1247  */
1248 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1249 {
1250 	int error;
1251 	struct inode *inode = d_inode(dentry);
1252 
1253 	/*
1254 	 * I believe we can only get a negative dentry here in the case of a
1255 	 * procfs-style symlink. Just assume it's correct for now, but we may
1256 	 * eventually need to do something more here.
1257 	 */
1258 	if (!inode) {
1259 		dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1260 				__func__, dentry);
1261 		return 1;
1262 	}
1263 
1264 	if (is_bad_inode(inode)) {
1265 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1266 				__func__, dentry);
1267 		return 0;
1268 	}
1269 
1270 	error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1271 	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1272 			__func__, inode->i_ino, error ? "invalid" : "valid");
1273 	return !error;
1274 }
1275 
1276 /*
1277  * This is called from dput() when d_count is going to 0.
1278  */
1279 static int nfs_dentry_delete(const struct dentry *dentry)
1280 {
1281 	dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1282 		dentry, dentry->d_flags);
1283 
1284 	/* Unhash any dentry with a stale inode */
1285 	if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1286 		return 1;
1287 
1288 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1289 		/* Unhash it, so that ->d_iput() would be called */
1290 		return 1;
1291 	}
1292 	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1293 		/* Unhash it, so that ancestors of killed async unlink
1294 		 * files will be cleaned up during umount */
1295 		return 1;
1296 	}
1297 	return 0;
1298 
1299 }
1300 
1301 /* Ensure that we revalidate inode->i_nlink */
1302 static void nfs_drop_nlink(struct inode *inode)
1303 {
1304 	spin_lock(&inode->i_lock);
1305 	/* drop the inode if we're reasonably sure this is the last link */
1306 	if (inode->i_nlink == 1)
1307 		clear_nlink(inode);
1308 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1309 	spin_unlock(&inode->i_lock);
1310 }
1311 
1312 /*
1313  * Called when the dentry loses inode.
1314  * We use it to clean up silly-renamed files.
1315  */
1316 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1317 {
1318 	if (S_ISDIR(inode->i_mode))
1319 		/* drop any readdir cache as it could easily be old */
1320 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1321 
1322 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1323 		nfs_complete_unlink(dentry, inode);
1324 		nfs_drop_nlink(inode);
1325 	}
1326 	iput(inode);
1327 }
1328 
1329 static void nfs_d_release(struct dentry *dentry)
1330 {
1331 	/* free cached devname value, if it survived that far */
1332 	if (unlikely(dentry->d_fsdata)) {
1333 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1334 			WARN_ON(1);
1335 		else
1336 			kfree(dentry->d_fsdata);
1337 	}
1338 }
1339 
1340 const struct dentry_operations nfs_dentry_operations = {
1341 	.d_revalidate	= nfs_lookup_revalidate,
1342 	.d_weak_revalidate	= nfs_weak_revalidate,
1343 	.d_delete	= nfs_dentry_delete,
1344 	.d_iput		= nfs_dentry_iput,
1345 	.d_automount	= nfs_d_automount,
1346 	.d_release	= nfs_d_release,
1347 };
1348 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1349 
1350 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1351 {
1352 	struct dentry *res;
1353 	struct dentry *parent;
1354 	struct inode *inode = NULL;
1355 	struct nfs_fh *fhandle = NULL;
1356 	struct nfs_fattr *fattr = NULL;
1357 	struct nfs4_label *label = NULL;
1358 	int error;
1359 
1360 	dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1361 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1362 
1363 	res = ERR_PTR(-ENAMETOOLONG);
1364 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1365 		goto out;
1366 
1367 	/*
1368 	 * If we're doing an exclusive create, optimize away the lookup
1369 	 * but don't hash the dentry.
1370 	 */
1371 	if (nfs_is_exclusive_create(dir, flags)) {
1372 		d_instantiate(dentry, NULL);
1373 		res = NULL;
1374 		goto out;
1375 	}
1376 
1377 	res = ERR_PTR(-ENOMEM);
1378 	fhandle = nfs_alloc_fhandle();
1379 	fattr = nfs_alloc_fattr();
1380 	if (fhandle == NULL || fattr == NULL)
1381 		goto out;
1382 
1383 	label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1384 	if (IS_ERR(label))
1385 		goto out;
1386 
1387 	parent = dentry->d_parent;
1388 	/* Protect against concurrent sillydeletes */
1389 	trace_nfs_lookup_enter(dir, dentry, flags);
1390 	nfs_block_sillyrename(parent);
1391 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1392 	if (error == -ENOENT)
1393 		goto no_entry;
1394 	if (error < 0) {
1395 		res = ERR_PTR(error);
1396 		goto out_unblock_sillyrename;
1397 	}
1398 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1399 	res = ERR_CAST(inode);
1400 	if (IS_ERR(res))
1401 		goto out_unblock_sillyrename;
1402 
1403 	/* Success: notify readdir to use READDIRPLUS */
1404 	nfs_advise_use_readdirplus(dir);
1405 
1406 no_entry:
1407 	res = d_splice_alias(inode, dentry);
1408 	if (res != NULL) {
1409 		if (IS_ERR(res))
1410 			goto out_unblock_sillyrename;
1411 		dentry = res;
1412 	}
1413 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1414 out_unblock_sillyrename:
1415 	nfs_unblock_sillyrename(parent);
1416 	trace_nfs_lookup_exit(dir, dentry, flags, error);
1417 	nfs4_label_free(label);
1418 out:
1419 	nfs_free_fattr(fattr);
1420 	nfs_free_fhandle(fhandle);
1421 	return res;
1422 }
1423 EXPORT_SYMBOL_GPL(nfs_lookup);
1424 
1425 #if IS_ENABLED(CONFIG_NFS_V4)
1426 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1427 
1428 const struct dentry_operations nfs4_dentry_operations = {
1429 	.d_revalidate	= nfs4_lookup_revalidate,
1430 	.d_delete	= nfs_dentry_delete,
1431 	.d_iput		= nfs_dentry_iput,
1432 	.d_automount	= nfs_d_automount,
1433 	.d_release	= nfs_d_release,
1434 };
1435 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1436 
1437 static fmode_t flags_to_mode(int flags)
1438 {
1439 	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1440 	if ((flags & O_ACCMODE) != O_WRONLY)
1441 		res |= FMODE_READ;
1442 	if ((flags & O_ACCMODE) != O_RDONLY)
1443 		res |= FMODE_WRITE;
1444 	return res;
1445 }
1446 
1447 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1448 {
1449 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1450 }
1451 
1452 static int do_open(struct inode *inode, struct file *filp)
1453 {
1454 	nfs_fscache_open_file(inode, filp);
1455 	return 0;
1456 }
1457 
1458 static int nfs_finish_open(struct nfs_open_context *ctx,
1459 			   struct dentry *dentry,
1460 			   struct file *file, unsigned open_flags,
1461 			   int *opened)
1462 {
1463 	int err;
1464 
1465 	err = finish_open(file, dentry, do_open, opened);
1466 	if (err)
1467 		goto out;
1468 	nfs_file_set_open_context(file, ctx);
1469 
1470 out:
1471 	return err;
1472 }
1473 
1474 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1475 		    struct file *file, unsigned open_flags,
1476 		    umode_t mode, int *opened)
1477 {
1478 	struct nfs_open_context *ctx;
1479 	struct dentry *res;
1480 	struct iattr attr = { .ia_valid = ATTR_OPEN };
1481 	struct inode *inode;
1482 	unsigned int lookup_flags = 0;
1483 	int err;
1484 
1485 	/* Expect a negative dentry */
1486 	BUG_ON(d_inode(dentry));
1487 
1488 	dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1489 			dir->i_sb->s_id, dir->i_ino, dentry);
1490 
1491 	err = nfs_check_flags(open_flags);
1492 	if (err)
1493 		return err;
1494 
1495 	/* NFS only supports OPEN on regular files */
1496 	if ((open_flags & O_DIRECTORY)) {
1497 		if (!d_unhashed(dentry)) {
1498 			/*
1499 			 * Hashed negative dentry with O_DIRECTORY: dentry was
1500 			 * revalidated and is fine, no need to perform lookup
1501 			 * again
1502 			 */
1503 			return -ENOENT;
1504 		}
1505 		lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1506 		goto no_open;
1507 	}
1508 
1509 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1510 		return -ENAMETOOLONG;
1511 
1512 	if (open_flags & O_CREAT) {
1513 		attr.ia_valid |= ATTR_MODE;
1514 		attr.ia_mode = mode & ~current_umask();
1515 	}
1516 	if (open_flags & O_TRUNC) {
1517 		attr.ia_valid |= ATTR_SIZE;
1518 		attr.ia_size = 0;
1519 	}
1520 
1521 	ctx = create_nfs_open_context(dentry, open_flags);
1522 	err = PTR_ERR(ctx);
1523 	if (IS_ERR(ctx))
1524 		goto out;
1525 
1526 	trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1527 	nfs_block_sillyrename(dentry->d_parent);
1528 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1529 	nfs_unblock_sillyrename(dentry->d_parent);
1530 	if (IS_ERR(inode)) {
1531 		err = PTR_ERR(inode);
1532 		trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1533 		put_nfs_open_context(ctx);
1534 		switch (err) {
1535 		case -ENOENT:
1536 			d_drop(dentry);
1537 			d_add(dentry, NULL);
1538 			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1539 			break;
1540 		case -EISDIR:
1541 		case -ENOTDIR:
1542 			goto no_open;
1543 		case -ELOOP:
1544 			if (!(open_flags & O_NOFOLLOW))
1545 				goto no_open;
1546 			break;
1547 			/* case -EINVAL: */
1548 		default:
1549 			break;
1550 		}
1551 		goto out;
1552 	}
1553 
1554 	err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1555 	trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1556 	put_nfs_open_context(ctx);
1557 out:
1558 	return err;
1559 
1560 no_open:
1561 	res = nfs_lookup(dir, dentry, lookup_flags);
1562 	err = PTR_ERR(res);
1563 	if (IS_ERR(res))
1564 		goto out;
1565 
1566 	return finish_no_open(file, res);
1567 }
1568 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1569 
1570 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1571 {
1572 	struct inode *inode;
1573 	int ret = 0;
1574 
1575 	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1576 		goto no_open;
1577 	if (d_mountpoint(dentry))
1578 		goto no_open;
1579 	if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1580 		goto no_open;
1581 
1582 	inode = d_inode(dentry);
1583 
1584 	/* We can't create new files in nfs_open_revalidate(), so we
1585 	 * optimize away revalidation of negative dentries.
1586 	 */
1587 	if (inode == NULL) {
1588 		struct dentry *parent;
1589 		struct inode *dir;
1590 
1591 		if (flags & LOOKUP_RCU) {
1592 			parent = ACCESS_ONCE(dentry->d_parent);
1593 			dir = d_inode_rcu(parent);
1594 			if (!dir)
1595 				return -ECHILD;
1596 		} else {
1597 			parent = dget_parent(dentry);
1598 			dir = d_inode(parent);
1599 		}
1600 		if (!nfs_neg_need_reval(dir, dentry, flags))
1601 			ret = 1;
1602 		else if (flags & LOOKUP_RCU)
1603 			ret = -ECHILD;
1604 		if (!(flags & LOOKUP_RCU))
1605 			dput(parent);
1606 		else if (parent != ACCESS_ONCE(dentry->d_parent))
1607 			return -ECHILD;
1608 		goto out;
1609 	}
1610 
1611 	/* NFS only supports OPEN on regular files */
1612 	if (!S_ISREG(inode->i_mode))
1613 		goto no_open;
1614 	/* We cannot do exclusive creation on a positive dentry */
1615 	if (flags & LOOKUP_EXCL)
1616 		goto no_open;
1617 
1618 	/* Let f_op->open() actually open (and revalidate) the file */
1619 	ret = 1;
1620 
1621 out:
1622 	return ret;
1623 
1624 no_open:
1625 	return nfs_lookup_revalidate(dentry, flags);
1626 }
1627 
1628 #endif /* CONFIG_NFSV4 */
1629 
1630 /*
1631  * Code common to create, mkdir, and mknod.
1632  */
1633 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1634 				struct nfs_fattr *fattr,
1635 				struct nfs4_label *label)
1636 {
1637 	struct dentry *parent = dget_parent(dentry);
1638 	struct inode *dir = d_inode(parent);
1639 	struct inode *inode;
1640 	int error = -EACCES;
1641 
1642 	d_drop(dentry);
1643 
1644 	/* We may have been initialized further down */
1645 	if (d_really_is_positive(dentry))
1646 		goto out;
1647 	if (fhandle->size == 0) {
1648 		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1649 		if (error)
1650 			goto out_error;
1651 	}
1652 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1653 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1654 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1655 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1656 		if (error < 0)
1657 			goto out_error;
1658 	}
1659 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1660 	error = PTR_ERR(inode);
1661 	if (IS_ERR(inode))
1662 		goto out_error;
1663 	d_add(dentry, inode);
1664 out:
1665 	dput(parent);
1666 	return 0;
1667 out_error:
1668 	nfs_mark_for_revalidate(dir);
1669 	dput(parent);
1670 	return error;
1671 }
1672 EXPORT_SYMBOL_GPL(nfs_instantiate);
1673 
1674 /*
1675  * Following a failed create operation, we drop the dentry rather
1676  * than retain a negative dentry. This avoids a problem in the event
1677  * that the operation succeeded on the server, but an error in the
1678  * reply path made it appear to have failed.
1679  */
1680 int nfs_create(struct inode *dir, struct dentry *dentry,
1681 		umode_t mode, bool excl)
1682 {
1683 	struct iattr attr;
1684 	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1685 	int error;
1686 
1687 	dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1688 			dir->i_sb->s_id, dir->i_ino, dentry);
1689 
1690 	attr.ia_mode = mode;
1691 	attr.ia_valid = ATTR_MODE;
1692 
1693 	trace_nfs_create_enter(dir, dentry, open_flags);
1694 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1695 	trace_nfs_create_exit(dir, dentry, open_flags, error);
1696 	if (error != 0)
1697 		goto out_err;
1698 	return 0;
1699 out_err:
1700 	d_drop(dentry);
1701 	return error;
1702 }
1703 EXPORT_SYMBOL_GPL(nfs_create);
1704 
1705 /*
1706  * See comments for nfs_proc_create regarding failed operations.
1707  */
1708 int
1709 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1710 {
1711 	struct iattr attr;
1712 	int status;
1713 
1714 	dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1715 			dir->i_sb->s_id, dir->i_ino, dentry);
1716 
1717 	attr.ia_mode = mode;
1718 	attr.ia_valid = ATTR_MODE;
1719 
1720 	trace_nfs_mknod_enter(dir, dentry);
1721 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1722 	trace_nfs_mknod_exit(dir, dentry, status);
1723 	if (status != 0)
1724 		goto out_err;
1725 	return 0;
1726 out_err:
1727 	d_drop(dentry);
1728 	return status;
1729 }
1730 EXPORT_SYMBOL_GPL(nfs_mknod);
1731 
1732 /*
1733  * See comments for nfs_proc_create regarding failed operations.
1734  */
1735 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1736 {
1737 	struct iattr attr;
1738 	int error;
1739 
1740 	dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1741 			dir->i_sb->s_id, dir->i_ino, dentry);
1742 
1743 	attr.ia_valid = ATTR_MODE;
1744 	attr.ia_mode = mode | S_IFDIR;
1745 
1746 	trace_nfs_mkdir_enter(dir, dentry);
1747 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1748 	trace_nfs_mkdir_exit(dir, dentry, error);
1749 	if (error != 0)
1750 		goto out_err;
1751 	return 0;
1752 out_err:
1753 	d_drop(dentry);
1754 	return error;
1755 }
1756 EXPORT_SYMBOL_GPL(nfs_mkdir);
1757 
1758 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1759 {
1760 	if (simple_positive(dentry))
1761 		d_delete(dentry);
1762 }
1763 
1764 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1765 {
1766 	int error;
1767 
1768 	dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1769 			dir->i_sb->s_id, dir->i_ino, dentry);
1770 
1771 	trace_nfs_rmdir_enter(dir, dentry);
1772 	if (d_really_is_positive(dentry)) {
1773 		nfs_wait_on_sillyrename(dentry);
1774 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1775 		/* Ensure the VFS deletes this inode */
1776 		switch (error) {
1777 		case 0:
1778 			clear_nlink(d_inode(dentry));
1779 			break;
1780 		case -ENOENT:
1781 			nfs_dentry_handle_enoent(dentry);
1782 		}
1783 	} else
1784 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1785 	trace_nfs_rmdir_exit(dir, dentry, error);
1786 
1787 	return error;
1788 }
1789 EXPORT_SYMBOL_GPL(nfs_rmdir);
1790 
1791 /*
1792  * Remove a file after making sure there are no pending writes,
1793  * and after checking that the file has only one user.
1794  *
1795  * We invalidate the attribute cache and free the inode prior to the operation
1796  * to avoid possible races if the server reuses the inode.
1797  */
1798 static int nfs_safe_remove(struct dentry *dentry)
1799 {
1800 	struct inode *dir = d_inode(dentry->d_parent);
1801 	struct inode *inode = d_inode(dentry);
1802 	int error = -EBUSY;
1803 
1804 	dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1805 
1806 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1807 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1808 		error = 0;
1809 		goto out;
1810 	}
1811 
1812 	trace_nfs_remove_enter(dir, dentry);
1813 	if (inode != NULL) {
1814 		NFS_PROTO(inode)->return_delegation(inode);
1815 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1816 		if (error == 0)
1817 			nfs_drop_nlink(inode);
1818 	} else
1819 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1820 	if (error == -ENOENT)
1821 		nfs_dentry_handle_enoent(dentry);
1822 	trace_nfs_remove_exit(dir, dentry, error);
1823 out:
1824 	return error;
1825 }
1826 
1827 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1828  *  belongs to an active ".nfs..." file and we return -EBUSY.
1829  *
1830  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1831  */
1832 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1833 {
1834 	int error;
1835 	int need_rehash = 0;
1836 
1837 	dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1838 		dir->i_ino, dentry);
1839 
1840 	trace_nfs_unlink_enter(dir, dentry);
1841 	spin_lock(&dentry->d_lock);
1842 	if (d_count(dentry) > 1) {
1843 		spin_unlock(&dentry->d_lock);
1844 		/* Start asynchronous writeout of the inode */
1845 		write_inode_now(d_inode(dentry), 0);
1846 		error = nfs_sillyrename(dir, dentry);
1847 		goto out;
1848 	}
1849 	if (!d_unhashed(dentry)) {
1850 		__d_drop(dentry);
1851 		need_rehash = 1;
1852 	}
1853 	spin_unlock(&dentry->d_lock);
1854 	error = nfs_safe_remove(dentry);
1855 	if (!error || error == -ENOENT) {
1856 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1857 	} else if (need_rehash)
1858 		d_rehash(dentry);
1859 out:
1860 	trace_nfs_unlink_exit(dir, dentry, error);
1861 	return error;
1862 }
1863 EXPORT_SYMBOL_GPL(nfs_unlink);
1864 
1865 /*
1866  * To create a symbolic link, most file systems instantiate a new inode,
1867  * add a page to it containing the path, then write it out to the disk
1868  * using prepare_write/commit_write.
1869  *
1870  * Unfortunately the NFS client can't create the in-core inode first
1871  * because it needs a file handle to create an in-core inode (see
1872  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1873  * symlink request has completed on the server.
1874  *
1875  * So instead we allocate a raw page, copy the symname into it, then do
1876  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1877  * now have a new file handle and can instantiate an in-core NFS inode
1878  * and move the raw page into its mapping.
1879  */
1880 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1881 {
1882 	struct page *page;
1883 	char *kaddr;
1884 	struct iattr attr;
1885 	unsigned int pathlen = strlen(symname);
1886 	int error;
1887 
1888 	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1889 		dir->i_ino, dentry, symname);
1890 
1891 	if (pathlen > PAGE_SIZE)
1892 		return -ENAMETOOLONG;
1893 
1894 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1895 	attr.ia_valid = ATTR_MODE;
1896 
1897 	page = alloc_page(GFP_HIGHUSER);
1898 	if (!page)
1899 		return -ENOMEM;
1900 
1901 	kaddr = kmap_atomic(page);
1902 	memcpy(kaddr, symname, pathlen);
1903 	if (pathlen < PAGE_SIZE)
1904 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1905 	kunmap_atomic(kaddr);
1906 
1907 	trace_nfs_symlink_enter(dir, dentry);
1908 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1909 	trace_nfs_symlink_exit(dir, dentry, error);
1910 	if (error != 0) {
1911 		dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1912 			dir->i_sb->s_id, dir->i_ino,
1913 			dentry, symname, error);
1914 		d_drop(dentry);
1915 		__free_page(page);
1916 		return error;
1917 	}
1918 
1919 	/*
1920 	 * No big deal if we can't add this page to the page cache here.
1921 	 * READLINK will get the missing page from the server if needed.
1922 	 */
1923 	if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
1924 							GFP_KERNEL)) {
1925 		SetPageUptodate(page);
1926 		unlock_page(page);
1927 		/*
1928 		 * add_to_page_cache_lru() grabs an extra page refcount.
1929 		 * Drop it here to avoid leaking this page later.
1930 		 */
1931 		page_cache_release(page);
1932 	} else
1933 		__free_page(page);
1934 
1935 	return 0;
1936 }
1937 EXPORT_SYMBOL_GPL(nfs_symlink);
1938 
1939 int
1940 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1941 {
1942 	struct inode *inode = d_inode(old_dentry);
1943 	int error;
1944 
1945 	dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1946 		old_dentry, dentry);
1947 
1948 	trace_nfs_link_enter(inode, dir, dentry);
1949 	NFS_PROTO(inode)->return_delegation(inode);
1950 
1951 	d_drop(dentry);
1952 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1953 	if (error == 0) {
1954 		ihold(inode);
1955 		d_add(dentry, inode);
1956 	}
1957 	trace_nfs_link_exit(inode, dir, dentry, error);
1958 	return error;
1959 }
1960 EXPORT_SYMBOL_GPL(nfs_link);
1961 
1962 /*
1963  * RENAME
1964  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1965  * different file handle for the same inode after a rename (e.g. when
1966  * moving to a different directory). A fail-safe method to do so would
1967  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1968  * rename the old file using the sillyrename stuff. This way, the original
1969  * file in old_dir will go away when the last process iput()s the inode.
1970  *
1971  * FIXED.
1972  *
1973  * It actually works quite well. One needs to have the possibility for
1974  * at least one ".nfs..." file in each directory the file ever gets
1975  * moved or linked to which happens automagically with the new
1976  * implementation that only depends on the dcache stuff instead of
1977  * using the inode layer
1978  *
1979  * Unfortunately, things are a little more complicated than indicated
1980  * above. For a cross-directory move, we want to make sure we can get
1981  * rid of the old inode after the operation.  This means there must be
1982  * no pending writes (if it's a file), and the use count must be 1.
1983  * If these conditions are met, we can drop the dentries before doing
1984  * the rename.
1985  */
1986 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1987 		      struct inode *new_dir, struct dentry *new_dentry)
1988 {
1989 	struct inode *old_inode = d_inode(old_dentry);
1990 	struct inode *new_inode = d_inode(new_dentry);
1991 	struct dentry *dentry = NULL, *rehash = NULL;
1992 	struct rpc_task *task;
1993 	int error = -EBUSY;
1994 
1995 	dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1996 		 old_dentry, new_dentry,
1997 		 d_count(new_dentry));
1998 
1999 	trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2000 	/*
2001 	 * For non-directories, check whether the target is busy and if so,
2002 	 * make a copy of the dentry and then do a silly-rename. If the
2003 	 * silly-rename succeeds, the copied dentry is hashed and becomes
2004 	 * the new target.
2005 	 */
2006 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2007 		/*
2008 		 * To prevent any new references to the target during the
2009 		 * rename, we unhash the dentry in advance.
2010 		 */
2011 		if (!d_unhashed(new_dentry)) {
2012 			d_drop(new_dentry);
2013 			rehash = new_dentry;
2014 		}
2015 
2016 		if (d_count(new_dentry) > 2) {
2017 			int err;
2018 
2019 			/* copy the target dentry's name */
2020 			dentry = d_alloc(new_dentry->d_parent,
2021 					 &new_dentry->d_name);
2022 			if (!dentry)
2023 				goto out;
2024 
2025 			/* silly-rename the existing target ... */
2026 			err = nfs_sillyrename(new_dir, new_dentry);
2027 			if (err)
2028 				goto out;
2029 
2030 			new_dentry = dentry;
2031 			rehash = NULL;
2032 			new_inode = NULL;
2033 		}
2034 	}
2035 
2036 	NFS_PROTO(old_inode)->return_delegation(old_inode);
2037 	if (new_inode != NULL)
2038 		NFS_PROTO(new_inode)->return_delegation(new_inode);
2039 
2040 	task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2041 	if (IS_ERR(task)) {
2042 		error = PTR_ERR(task);
2043 		goto out;
2044 	}
2045 
2046 	error = rpc_wait_for_completion_task(task);
2047 	if (error == 0)
2048 		error = task->tk_status;
2049 	rpc_put_task(task);
2050 	nfs_mark_for_revalidate(old_inode);
2051 out:
2052 	if (rehash)
2053 		d_rehash(rehash);
2054 	trace_nfs_rename_exit(old_dir, old_dentry,
2055 			new_dir, new_dentry, error);
2056 	if (!error) {
2057 		if (new_inode != NULL)
2058 			nfs_drop_nlink(new_inode);
2059 		d_move(old_dentry, new_dentry);
2060 		nfs_set_verifier(new_dentry,
2061 					nfs_save_change_attribute(new_dir));
2062 	} else if (error == -ENOENT)
2063 		nfs_dentry_handle_enoent(old_dentry);
2064 
2065 	/* new dentry created? */
2066 	if (dentry)
2067 		dput(dentry);
2068 	return error;
2069 }
2070 EXPORT_SYMBOL_GPL(nfs_rename);
2071 
2072 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2073 static LIST_HEAD(nfs_access_lru_list);
2074 static atomic_long_t nfs_access_nr_entries;
2075 
2076 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2077 module_param(nfs_access_max_cachesize, ulong, 0644);
2078 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2079 
2080 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2081 {
2082 	put_rpccred(entry->cred);
2083 	kfree_rcu(entry, rcu_head);
2084 	smp_mb__before_atomic();
2085 	atomic_long_dec(&nfs_access_nr_entries);
2086 	smp_mb__after_atomic();
2087 }
2088 
2089 static void nfs_access_free_list(struct list_head *head)
2090 {
2091 	struct nfs_access_entry *cache;
2092 
2093 	while (!list_empty(head)) {
2094 		cache = list_entry(head->next, struct nfs_access_entry, lru);
2095 		list_del(&cache->lru);
2096 		nfs_access_free_entry(cache);
2097 	}
2098 }
2099 
2100 static unsigned long
2101 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2102 {
2103 	LIST_HEAD(head);
2104 	struct nfs_inode *nfsi, *next;
2105 	struct nfs_access_entry *cache;
2106 	long freed = 0;
2107 
2108 	spin_lock(&nfs_access_lru_lock);
2109 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2110 		struct inode *inode;
2111 
2112 		if (nr_to_scan-- == 0)
2113 			break;
2114 		inode = &nfsi->vfs_inode;
2115 		spin_lock(&inode->i_lock);
2116 		if (list_empty(&nfsi->access_cache_entry_lru))
2117 			goto remove_lru_entry;
2118 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2119 				struct nfs_access_entry, lru);
2120 		list_move(&cache->lru, &head);
2121 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2122 		freed++;
2123 		if (!list_empty(&nfsi->access_cache_entry_lru))
2124 			list_move_tail(&nfsi->access_cache_inode_lru,
2125 					&nfs_access_lru_list);
2126 		else {
2127 remove_lru_entry:
2128 			list_del_init(&nfsi->access_cache_inode_lru);
2129 			smp_mb__before_atomic();
2130 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2131 			smp_mb__after_atomic();
2132 		}
2133 		spin_unlock(&inode->i_lock);
2134 	}
2135 	spin_unlock(&nfs_access_lru_lock);
2136 	nfs_access_free_list(&head);
2137 	return freed;
2138 }
2139 
2140 unsigned long
2141 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2142 {
2143 	int nr_to_scan = sc->nr_to_scan;
2144 	gfp_t gfp_mask = sc->gfp_mask;
2145 
2146 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2147 		return SHRINK_STOP;
2148 	return nfs_do_access_cache_scan(nr_to_scan);
2149 }
2150 
2151 
2152 unsigned long
2153 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2154 {
2155 	return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2156 }
2157 
2158 static void
2159 nfs_access_cache_enforce_limit(void)
2160 {
2161 	long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2162 	unsigned long diff;
2163 	unsigned int nr_to_scan;
2164 
2165 	if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2166 		return;
2167 	nr_to_scan = 100;
2168 	diff = nr_entries - nfs_access_max_cachesize;
2169 	if (diff < nr_to_scan)
2170 		nr_to_scan = diff;
2171 	nfs_do_access_cache_scan(nr_to_scan);
2172 }
2173 
2174 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2175 {
2176 	struct rb_root *root_node = &nfsi->access_cache;
2177 	struct rb_node *n;
2178 	struct nfs_access_entry *entry;
2179 
2180 	/* Unhook entries from the cache */
2181 	while ((n = rb_first(root_node)) != NULL) {
2182 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2183 		rb_erase(n, root_node);
2184 		list_move(&entry->lru, head);
2185 	}
2186 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2187 }
2188 
2189 void nfs_access_zap_cache(struct inode *inode)
2190 {
2191 	LIST_HEAD(head);
2192 
2193 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2194 		return;
2195 	/* Remove from global LRU init */
2196 	spin_lock(&nfs_access_lru_lock);
2197 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2198 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2199 
2200 	spin_lock(&inode->i_lock);
2201 	__nfs_access_zap_cache(NFS_I(inode), &head);
2202 	spin_unlock(&inode->i_lock);
2203 	spin_unlock(&nfs_access_lru_lock);
2204 	nfs_access_free_list(&head);
2205 }
2206 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2207 
2208 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2209 {
2210 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2211 	struct nfs_access_entry *entry;
2212 
2213 	while (n != NULL) {
2214 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2215 
2216 		if (cred < entry->cred)
2217 			n = n->rb_left;
2218 		else if (cred > entry->cred)
2219 			n = n->rb_right;
2220 		else
2221 			return entry;
2222 	}
2223 	return NULL;
2224 }
2225 
2226 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2227 {
2228 	struct nfs_inode *nfsi = NFS_I(inode);
2229 	struct nfs_access_entry *cache;
2230 	int err = -ENOENT;
2231 
2232 	spin_lock(&inode->i_lock);
2233 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2234 		goto out_zap;
2235 	cache = nfs_access_search_rbtree(inode, cred);
2236 	if (cache == NULL)
2237 		goto out;
2238 	if (!nfs_have_delegated_attributes(inode) &&
2239 	    !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2240 		goto out_stale;
2241 	res->jiffies = cache->jiffies;
2242 	res->cred = cache->cred;
2243 	res->mask = cache->mask;
2244 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2245 	err = 0;
2246 out:
2247 	spin_unlock(&inode->i_lock);
2248 	return err;
2249 out_stale:
2250 	rb_erase(&cache->rb_node, &nfsi->access_cache);
2251 	list_del(&cache->lru);
2252 	spin_unlock(&inode->i_lock);
2253 	nfs_access_free_entry(cache);
2254 	return -ENOENT;
2255 out_zap:
2256 	spin_unlock(&inode->i_lock);
2257 	nfs_access_zap_cache(inode);
2258 	return -ENOENT;
2259 }
2260 
2261 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2262 {
2263 	/* Only check the most recently returned cache entry,
2264 	 * but do it without locking.
2265 	 */
2266 	struct nfs_inode *nfsi = NFS_I(inode);
2267 	struct nfs_access_entry *cache;
2268 	int err = -ECHILD;
2269 	struct list_head *lh;
2270 
2271 	rcu_read_lock();
2272 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2273 		goto out;
2274 	lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2275 	cache = list_entry(lh, struct nfs_access_entry, lru);
2276 	if (lh == &nfsi->access_cache_entry_lru ||
2277 	    cred != cache->cred)
2278 		cache = NULL;
2279 	if (cache == NULL)
2280 		goto out;
2281 	if (!nfs_have_delegated_attributes(inode) &&
2282 	    !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2283 		goto out;
2284 	res->jiffies = cache->jiffies;
2285 	res->cred = cache->cred;
2286 	res->mask = cache->mask;
2287 	err = 0;
2288 out:
2289 	rcu_read_unlock();
2290 	return err;
2291 }
2292 
2293 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2294 {
2295 	struct nfs_inode *nfsi = NFS_I(inode);
2296 	struct rb_root *root_node = &nfsi->access_cache;
2297 	struct rb_node **p = &root_node->rb_node;
2298 	struct rb_node *parent = NULL;
2299 	struct nfs_access_entry *entry;
2300 
2301 	spin_lock(&inode->i_lock);
2302 	while (*p != NULL) {
2303 		parent = *p;
2304 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2305 
2306 		if (set->cred < entry->cred)
2307 			p = &parent->rb_left;
2308 		else if (set->cred > entry->cred)
2309 			p = &parent->rb_right;
2310 		else
2311 			goto found;
2312 	}
2313 	rb_link_node(&set->rb_node, parent, p);
2314 	rb_insert_color(&set->rb_node, root_node);
2315 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2316 	spin_unlock(&inode->i_lock);
2317 	return;
2318 found:
2319 	rb_replace_node(parent, &set->rb_node, root_node);
2320 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2321 	list_del(&entry->lru);
2322 	spin_unlock(&inode->i_lock);
2323 	nfs_access_free_entry(entry);
2324 }
2325 
2326 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2327 {
2328 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2329 	if (cache == NULL)
2330 		return;
2331 	RB_CLEAR_NODE(&cache->rb_node);
2332 	cache->jiffies = set->jiffies;
2333 	cache->cred = get_rpccred(set->cred);
2334 	cache->mask = set->mask;
2335 
2336 	/* The above field assignments must be visible
2337 	 * before this item appears on the lru.  We cannot easily
2338 	 * use rcu_assign_pointer, so just force the memory barrier.
2339 	 */
2340 	smp_wmb();
2341 	nfs_access_add_rbtree(inode, cache);
2342 
2343 	/* Update accounting */
2344 	smp_mb__before_atomic();
2345 	atomic_long_inc(&nfs_access_nr_entries);
2346 	smp_mb__after_atomic();
2347 
2348 	/* Add inode to global LRU list */
2349 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2350 		spin_lock(&nfs_access_lru_lock);
2351 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2352 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2353 					&nfs_access_lru_list);
2354 		spin_unlock(&nfs_access_lru_lock);
2355 	}
2356 	nfs_access_cache_enforce_limit();
2357 }
2358 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2359 
2360 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2361 {
2362 	entry->mask = 0;
2363 	if (access_result & NFS4_ACCESS_READ)
2364 		entry->mask |= MAY_READ;
2365 	if (access_result &
2366 	    (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2367 		entry->mask |= MAY_WRITE;
2368 	if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2369 		entry->mask |= MAY_EXEC;
2370 }
2371 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2372 
2373 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2374 {
2375 	struct nfs_access_entry cache;
2376 	int status;
2377 
2378 	trace_nfs_access_enter(inode);
2379 
2380 	status = nfs_access_get_cached_rcu(inode, cred, &cache);
2381 	if (status != 0)
2382 		status = nfs_access_get_cached(inode, cred, &cache);
2383 	if (status == 0)
2384 		goto out_cached;
2385 
2386 	status = -ECHILD;
2387 	if (mask & MAY_NOT_BLOCK)
2388 		goto out;
2389 
2390 	/* Be clever: ask server to check for all possible rights */
2391 	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2392 	cache.cred = cred;
2393 	cache.jiffies = jiffies;
2394 	status = NFS_PROTO(inode)->access(inode, &cache);
2395 	if (status != 0) {
2396 		if (status == -ESTALE) {
2397 			nfs_zap_caches(inode);
2398 			if (!S_ISDIR(inode->i_mode))
2399 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2400 		}
2401 		goto out;
2402 	}
2403 	nfs_access_add_cache(inode, &cache);
2404 out_cached:
2405 	if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2406 		status = -EACCES;
2407 out:
2408 	trace_nfs_access_exit(inode, status);
2409 	return status;
2410 }
2411 
2412 static int nfs_open_permission_mask(int openflags)
2413 {
2414 	int mask = 0;
2415 
2416 	if (openflags & __FMODE_EXEC) {
2417 		/* ONLY check exec rights */
2418 		mask = MAY_EXEC;
2419 	} else {
2420 		if ((openflags & O_ACCMODE) != O_WRONLY)
2421 			mask |= MAY_READ;
2422 		if ((openflags & O_ACCMODE) != O_RDONLY)
2423 			mask |= MAY_WRITE;
2424 	}
2425 
2426 	return mask;
2427 }
2428 
2429 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2430 {
2431 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2432 }
2433 EXPORT_SYMBOL_GPL(nfs_may_open);
2434 
2435 int nfs_permission(struct inode *inode, int mask)
2436 {
2437 	struct rpc_cred *cred;
2438 	int res = 0;
2439 
2440 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2441 
2442 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2443 		goto out;
2444 	/* Is this sys_access() ? */
2445 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2446 		goto force_lookup;
2447 
2448 	switch (inode->i_mode & S_IFMT) {
2449 		case S_IFLNK:
2450 			goto out;
2451 		case S_IFREG:
2452 			break;
2453 		case S_IFDIR:
2454 			/*
2455 			 * Optimize away all write operations, since the server
2456 			 * will check permissions when we perform the op.
2457 			 */
2458 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2459 				goto out;
2460 	}
2461 
2462 force_lookup:
2463 	if (!NFS_PROTO(inode)->access)
2464 		goto out_notsup;
2465 
2466 	/* Always try fast lookups first */
2467 	rcu_read_lock();
2468 	cred = rpc_lookup_cred_nonblock();
2469 	if (!IS_ERR(cred))
2470 		res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2471 	else
2472 		res = PTR_ERR(cred);
2473 	rcu_read_unlock();
2474 	if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2475 		/* Fast lookup failed, try the slow way */
2476 		cred = rpc_lookup_cred();
2477 		if (!IS_ERR(cred)) {
2478 			res = nfs_do_access(inode, cred, mask);
2479 			put_rpccred(cred);
2480 		} else
2481 			res = PTR_ERR(cred);
2482 	}
2483 out:
2484 	if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2485 		res = -EACCES;
2486 
2487 	dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2488 		inode->i_sb->s_id, inode->i_ino, mask, res);
2489 	return res;
2490 out_notsup:
2491 	if (mask & MAY_NOT_BLOCK)
2492 		return -ECHILD;
2493 
2494 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2495 	if (res == 0)
2496 		res = generic_permission(inode, mask);
2497 	goto out;
2498 }
2499 EXPORT_SYMBOL_GPL(nfs_permission);
2500 
2501 /*
2502  * Local variables:
2503  *  version-control: t
2504  *  kept-new-versions: 5
2505  * End:
2506  */
2507