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 
alloc_nfs_open_dir_context(struct inode * dir,struct rpc_cred * cred)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 
put_nfs_open_dir_context(struct inode * dir,struct nfs_open_dir_context * ctx)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
nfs_opendir(struct inode * inode,struct file * filp)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 out:
122 	put_rpccred(cred);
123 	return res;
124 }
125 
126 static int
nfs_closedir(struct inode * inode,struct file * filp)127 nfs_closedir(struct inode *inode, struct file *filp)
128 {
129 	put_nfs_open_dir_context(file_inode(filp), 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 *, bool);
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 	bool plus;
162 	bool eof;
163 } nfs_readdir_descriptor_t;
164 
165 static
nfs_readdir_init_array(struct page * page)166 void nfs_readdir_init_array(struct page *page)
167 {
168 	struct nfs_cache_array *array;
169 
170 	array = kmap_atomic(page);
171 	memset(array, 0, sizeof(struct nfs_cache_array));
172 	array->eof_index = -1;
173 	kunmap_atomic(array);
174 }
175 
176 /*
177  * we are freeing strings created by nfs_add_to_readdir_array()
178  */
179 static
nfs_readdir_clear_array(struct page * page)180 void nfs_readdir_clear_array(struct page *page)
181 {
182 	struct nfs_cache_array *array;
183 	int i;
184 
185 	array = kmap_atomic(page);
186 	for (i = 0; i < array->size; i++)
187 		kfree(array->array[i].string.name);
188 	array->size = 0;
189 	kunmap_atomic(array);
190 }
191 
192 /*
193  * the caller is responsible for freeing qstr.name
194  * when called by nfs_readdir_add_to_array, the strings will be freed in
195  * nfs_clear_readdir_array()
196  */
197 static
nfs_readdir_make_qstr(struct qstr * string,const char * name,unsigned int len)198 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
199 {
200 	string->len = len;
201 	string->name = kmemdup(name, len, GFP_KERNEL);
202 	if (string->name == NULL)
203 		return -ENOMEM;
204 	/*
205 	 * Avoid a kmemleak false positive. The pointer to the name is stored
206 	 * in a page cache page which kmemleak does not scan.
207 	 */
208 	kmemleak_not_leak(string->name);
209 	string->hash = full_name_hash(NULL, name, len);
210 	return 0;
211 }
212 
213 static
nfs_readdir_add_to_array(struct nfs_entry * entry,struct page * page)214 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
215 {
216 	struct nfs_cache_array *array = kmap(page);
217 	struct nfs_cache_array_entry *cache_entry;
218 	int ret;
219 
220 	cache_entry = &array->array[array->size];
221 
222 	/* Check that this entry lies within the page bounds */
223 	ret = -ENOSPC;
224 	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
225 		goto out;
226 
227 	cache_entry->cookie = entry->prev_cookie;
228 	cache_entry->ino = entry->ino;
229 	cache_entry->d_type = entry->d_type;
230 	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
231 	if (ret)
232 		goto out;
233 	array->last_cookie = entry->cookie;
234 	array->size++;
235 	if (entry->eof != 0)
236 		array->eof_index = array->size;
237 out:
238 	kunmap(page);
239 	return ret;
240 }
241 
242 static
nfs_readdir_search_for_pos(struct nfs_cache_array * array,nfs_readdir_descriptor_t * desc)243 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
244 {
245 	loff_t diff = desc->ctx->pos - desc->current_index;
246 	unsigned int index;
247 
248 	if (diff < 0)
249 		goto out_eof;
250 	if (diff >= array->size) {
251 		if (array->eof_index >= 0)
252 			goto out_eof;
253 		return -EAGAIN;
254 	}
255 
256 	index = (unsigned int)diff;
257 	*desc->dir_cookie = array->array[index].cookie;
258 	desc->cache_entry_index = index;
259 	return 0;
260 out_eof:
261 	desc->eof = true;
262 	return -EBADCOOKIE;
263 }
264 
265 static bool
nfs_readdir_inode_mapping_valid(struct nfs_inode * nfsi)266 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
267 {
268 	if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
269 		return false;
270 	smp_rmb();
271 	return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
272 }
273 
274 static
nfs_readdir_search_for_cookie(struct nfs_cache_array * array,nfs_readdir_descriptor_t * desc)275 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
276 {
277 	int i;
278 	loff_t new_pos;
279 	int status = -EAGAIN;
280 
281 	for (i = 0; i < array->size; i++) {
282 		if (array->array[i].cookie == *desc->dir_cookie) {
283 			struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
284 			struct nfs_open_dir_context *ctx = desc->file->private_data;
285 
286 			new_pos = desc->current_index + i;
287 			if (ctx->attr_gencount != nfsi->attr_gencount ||
288 			    !nfs_readdir_inode_mapping_valid(nfsi)) {
289 				ctx->duped = 0;
290 				ctx->attr_gencount = nfsi->attr_gencount;
291 			} else if (new_pos < desc->ctx->pos) {
292 				if (ctx->duped > 0
293 				    && ctx->dup_cookie == *desc->dir_cookie) {
294 					if (printk_ratelimit()) {
295 						pr_notice("NFS: directory %pD2 contains a readdir loop."
296 								"Please contact your server vendor.  "
297 								"The file: %.*s has duplicate cookie %llu\n",
298 								desc->file, array->array[i].string.len,
299 								array->array[i].string.name, *desc->dir_cookie);
300 					}
301 					status = -ELOOP;
302 					goto out;
303 				}
304 				ctx->dup_cookie = *desc->dir_cookie;
305 				ctx->duped = -1;
306 			}
307 			desc->ctx->pos = new_pos;
308 			desc->cache_entry_index = i;
309 			return 0;
310 		}
311 	}
312 	if (array->eof_index >= 0) {
313 		status = -EBADCOOKIE;
314 		if (*desc->dir_cookie == array->last_cookie)
315 			desc->eof = true;
316 	}
317 out:
318 	return status;
319 }
320 
321 static
nfs_readdir_search_array(nfs_readdir_descriptor_t * desc)322 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
323 {
324 	struct nfs_cache_array *array;
325 	int status;
326 
327 	array = kmap(desc->page);
328 
329 	if (*desc->dir_cookie == 0)
330 		status = nfs_readdir_search_for_pos(array, desc);
331 	else
332 		status = nfs_readdir_search_for_cookie(array, desc);
333 
334 	if (status == -EAGAIN) {
335 		desc->last_cookie = array->last_cookie;
336 		desc->current_index += array->size;
337 		desc->page_index++;
338 	}
339 	kunmap(desc->page);
340 	return status;
341 }
342 
343 /* Fill a page with xdr information before transferring to the cache page */
344 static
nfs_readdir_xdr_filler(struct page ** pages,nfs_readdir_descriptor_t * desc,struct nfs_entry * entry,struct file * file,struct inode * inode)345 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
346 			struct nfs_entry *entry, struct file *file, struct inode *inode)
347 {
348 	struct nfs_open_dir_context *ctx = file->private_data;
349 	struct rpc_cred	*cred = ctx->cred;
350 	unsigned long	timestamp, gencount;
351 	int		error;
352 
353  again:
354 	timestamp = jiffies;
355 	gencount = nfs_inc_attr_generation_counter();
356 	error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
357 					  NFS_SERVER(inode)->dtsize, desc->plus);
358 	if (error < 0) {
359 		/* We requested READDIRPLUS, but the server doesn't grok it */
360 		if (error == -ENOTSUPP && desc->plus) {
361 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
362 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
363 			desc->plus = false;
364 			goto again;
365 		}
366 		goto error;
367 	}
368 	desc->timestamp = timestamp;
369 	desc->gencount = gencount;
370 error:
371 	return error;
372 }
373 
xdr_decode(nfs_readdir_descriptor_t * desc,struct nfs_entry * entry,struct xdr_stream * xdr)374 static int xdr_decode(nfs_readdir_descriptor_t *desc,
375 		      struct nfs_entry *entry, struct xdr_stream *xdr)
376 {
377 	int error;
378 
379 	error = desc->decode(xdr, entry, desc->plus);
380 	if (error)
381 		return error;
382 	entry->fattr->time_start = desc->timestamp;
383 	entry->fattr->gencount = desc->gencount;
384 	return 0;
385 }
386 
387 /* Match file and dirent using either filehandle or fileid
388  * Note: caller is responsible for checking the fsid
389  */
390 static
nfs_same_file(struct dentry * dentry,struct nfs_entry * entry)391 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
392 {
393 	struct inode *inode;
394 	struct nfs_inode *nfsi;
395 
396 	if (d_really_is_negative(dentry))
397 		return 0;
398 
399 	inode = d_inode(dentry);
400 	if (is_bad_inode(inode) || NFS_STALE(inode))
401 		return 0;
402 
403 	nfsi = NFS_I(inode);
404 	if (entry->fattr->fileid != nfsi->fileid)
405 		return 0;
406 	if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
407 		return 0;
408 	return 1;
409 }
410 
411 static
nfs_use_readdirplus(struct inode * dir,struct dir_context * ctx)412 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
413 {
414 	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
415 		return false;
416 	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
417 		return true;
418 	if (ctx->pos == 0)
419 		return true;
420 	return false;
421 }
422 
423 /*
424  * This function is called by the lookup and getattr code to request the
425  * use of readdirplus to accelerate any future lookups in the same
426  * directory.
427  */
nfs_advise_use_readdirplus(struct inode * dir)428 void nfs_advise_use_readdirplus(struct inode *dir)
429 {
430 	struct nfs_inode *nfsi = NFS_I(dir);
431 
432 	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
433 	    !list_empty(&nfsi->open_files))
434 		set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
435 }
436 
437 /*
438  * This function is mainly for use by nfs_getattr().
439  *
440  * If this is an 'ls -l', we want to force use of readdirplus.
441  * Do this by checking if there is an active file descriptor
442  * and calling nfs_advise_use_readdirplus, then forcing a
443  * cache flush.
444  */
nfs_force_use_readdirplus(struct inode * dir)445 void nfs_force_use_readdirplus(struct inode *dir)
446 {
447 	struct nfs_inode *nfsi = NFS_I(dir);
448 
449 	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
450 	    !list_empty(&nfsi->open_files)) {
451 		set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
452 		invalidate_mapping_pages(dir->i_mapping, 0, -1);
453 	}
454 }
455 
456 static
nfs_prime_dcache(struct dentry * parent,struct nfs_entry * entry)457 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
458 {
459 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
460 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
461 	struct dentry *dentry;
462 	struct dentry *alias;
463 	struct inode *dir = d_inode(parent);
464 	struct inode *inode;
465 	int status;
466 
467 	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
468 		return;
469 	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
470 		return;
471 	if (filename.len == 0)
472 		return;
473 	/* Validate that the name doesn't contain any illegal '\0' */
474 	if (strnlen(filename.name, filename.len) != filename.len)
475 		return;
476 	/* ...or '/' */
477 	if (strnchr(filename.name, filename.len, '/'))
478 		return;
479 	if (filename.name[0] == '.') {
480 		if (filename.len == 1)
481 			return;
482 		if (filename.len == 2 && filename.name[1] == '.')
483 			return;
484 	}
485 	filename.hash = full_name_hash(parent, filename.name, filename.len);
486 
487 	dentry = d_lookup(parent, &filename);
488 again:
489 	if (!dentry) {
490 		dentry = d_alloc_parallel(parent, &filename, &wq);
491 		if (IS_ERR(dentry))
492 			return;
493 	}
494 	if (!d_in_lookup(dentry)) {
495 		/* Is there a mountpoint here? If so, just exit */
496 		if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
497 					&entry->fattr->fsid))
498 			goto out;
499 		if (nfs_same_file(dentry, entry)) {
500 			if (!entry->fh->size)
501 				goto out;
502 			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
503 			status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
504 			if (!status)
505 				nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
506 			goto out;
507 		} else {
508 			d_invalidate(dentry);
509 			dput(dentry);
510 			dentry = NULL;
511 			goto again;
512 		}
513 	}
514 	if (!entry->fh->size) {
515 		d_lookup_done(dentry);
516 		goto out;
517 	}
518 
519 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
520 	alias = d_splice_alias(inode, dentry);
521 	d_lookup_done(dentry);
522 	if (alias) {
523 		if (IS_ERR(alias))
524 			goto out;
525 		dput(dentry);
526 		dentry = alias;
527 	}
528 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
529 out:
530 	dput(dentry);
531 }
532 
533 /* Perform conversion from xdr to cache array */
534 static
nfs_readdir_page_filler(nfs_readdir_descriptor_t * desc,struct nfs_entry * entry,struct page ** xdr_pages,struct page * page,unsigned int buflen)535 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
536 				struct page **xdr_pages, struct page *page, unsigned int buflen)
537 {
538 	struct xdr_stream stream;
539 	struct xdr_buf buf;
540 	struct page *scratch;
541 	struct nfs_cache_array *array;
542 	unsigned int count = 0;
543 	int status;
544 
545 	scratch = alloc_page(GFP_KERNEL);
546 	if (scratch == NULL)
547 		return -ENOMEM;
548 
549 	if (buflen == 0)
550 		goto out_nopages;
551 
552 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
553 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
554 
555 	do {
556 		if (entry->label)
557 			entry->label->len = NFS4_MAXLABELLEN;
558 
559 		status = xdr_decode(desc, entry, &stream);
560 		if (status != 0) {
561 			if (status == -EAGAIN)
562 				status = 0;
563 			break;
564 		}
565 
566 		count++;
567 
568 		if (desc->plus)
569 			nfs_prime_dcache(file_dentry(desc->file), entry);
570 
571 		status = nfs_readdir_add_to_array(entry, page);
572 		if (status != 0)
573 			break;
574 	} while (!entry->eof);
575 
576 out_nopages:
577 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
578 		array = kmap(page);
579 		array->eof_index = array->size;
580 		status = 0;
581 		kunmap(page);
582 	}
583 
584 	put_page(scratch);
585 	return status;
586 }
587 
588 static
nfs_readdir_free_pages(struct page ** pages,unsigned int npages)589 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
590 {
591 	unsigned int i;
592 	for (i = 0; i < npages; i++)
593 		put_page(pages[i]);
594 }
595 
596 /*
597  * nfs_readdir_large_page will allocate pages that must be freed with a call
598  * to nfs_readdir_free_pagearray
599  */
600 static
nfs_readdir_alloc_pages(struct page ** pages,unsigned int npages)601 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
602 {
603 	unsigned int i;
604 
605 	for (i = 0; i < npages; i++) {
606 		struct page *page = alloc_page(GFP_KERNEL);
607 		if (page == NULL)
608 			goto out_freepages;
609 		pages[i] = page;
610 	}
611 	return 0;
612 
613 out_freepages:
614 	nfs_readdir_free_pages(pages, i);
615 	return -ENOMEM;
616 }
617 
618 static
nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t * desc,struct page * page,struct inode * inode)619 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
620 {
621 	struct page *pages[NFS_MAX_READDIR_PAGES];
622 	struct nfs_entry entry;
623 	struct file	*file = desc->file;
624 	struct nfs_cache_array *array;
625 	int status = -ENOMEM;
626 	unsigned int array_size = ARRAY_SIZE(pages);
627 
628 	nfs_readdir_init_array(page);
629 
630 	entry.prev_cookie = 0;
631 	entry.cookie = desc->last_cookie;
632 	entry.eof = 0;
633 	entry.fh = nfs_alloc_fhandle();
634 	entry.fattr = nfs_alloc_fattr();
635 	entry.server = NFS_SERVER(inode);
636 	if (entry.fh == NULL || entry.fattr == NULL)
637 		goto out;
638 
639 	entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
640 	if (IS_ERR(entry.label)) {
641 		status = PTR_ERR(entry.label);
642 		goto out;
643 	}
644 
645 	array = kmap(page);
646 
647 	status = nfs_readdir_alloc_pages(pages, array_size);
648 	if (status < 0)
649 		goto out_release_array;
650 	do {
651 		unsigned int pglen;
652 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
653 
654 		if (status < 0)
655 			break;
656 		pglen = status;
657 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
658 		if (status < 0) {
659 			if (status == -ENOSPC)
660 				status = 0;
661 			break;
662 		}
663 	} while (array->eof_index < 0);
664 
665 	nfs_readdir_free_pages(pages, array_size);
666 out_release_array:
667 	kunmap(page);
668 	nfs4_label_free(entry.label);
669 out:
670 	nfs_free_fattr(entry.fattr);
671 	nfs_free_fhandle(entry.fh);
672 	return status;
673 }
674 
675 /*
676  * Now we cache directories properly, by converting xdr information
677  * to an array that can be used for lookups later.  This results in
678  * fewer cache pages, since we can store more information on each page.
679  * We only need to convert from xdr once so future lookups are much simpler
680  */
681 static
nfs_readdir_filler(nfs_readdir_descriptor_t * desc,struct page * page)682 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
683 {
684 	struct inode	*inode = file_inode(desc->file);
685 	int ret;
686 
687 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
688 	if (ret < 0)
689 		goto error;
690 	SetPageUptodate(page);
691 
692 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
693 		/* Should never happen */
694 		nfs_zap_mapping(inode, inode->i_mapping);
695 	}
696 	unlock_page(page);
697 	return 0;
698  error:
699 	nfs_readdir_clear_array(page);
700 	unlock_page(page);
701 	return ret;
702 }
703 
704 static
cache_page_release(nfs_readdir_descriptor_t * desc)705 void cache_page_release(nfs_readdir_descriptor_t *desc)
706 {
707 	put_page(desc->page);
708 	desc->page = NULL;
709 }
710 
711 static
get_cache_page(nfs_readdir_descriptor_t * desc)712 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
713 {
714 	return read_cache_page(desc->file->f_mapping,
715 			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
716 }
717 
718 /*
719  * Returns 0 if desc->dir_cookie was found on page desc->page_index
720  * and locks the page to prevent removal from the page cache.
721  */
722 static
find_and_lock_cache_page(nfs_readdir_descriptor_t * desc)723 int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc)
724 {
725 	int res;
726 
727 	desc->page = get_cache_page(desc);
728 	if (IS_ERR(desc->page))
729 		return PTR_ERR(desc->page);
730 	res = lock_page_killable(desc->page);
731 	if (res != 0)
732 		goto error;
733 	res = -EAGAIN;
734 	if (desc->page->mapping != NULL) {
735 		res = nfs_readdir_search_array(desc);
736 		if (res == 0)
737 			return 0;
738 	}
739 	unlock_page(desc->page);
740 error:
741 	cache_page_release(desc);
742 	return res;
743 }
744 
745 /* Search for desc->dir_cookie from the beginning of the page cache */
746 static inline
readdir_search_pagecache(nfs_readdir_descriptor_t * desc)747 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
748 {
749 	int res;
750 
751 	if (desc->page_index == 0) {
752 		desc->current_index = 0;
753 		desc->last_cookie = 0;
754 	}
755 	do {
756 		res = find_and_lock_cache_page(desc);
757 	} while (res == -EAGAIN);
758 	return res;
759 }
760 
761 /*
762  * Once we've found the start of the dirent within a page: fill 'er up...
763  */
764 static
nfs_do_filldir(nfs_readdir_descriptor_t * desc)765 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
766 {
767 	struct file	*file = desc->file;
768 	int i = 0;
769 	int res = 0;
770 	struct nfs_cache_array *array = NULL;
771 	struct nfs_open_dir_context *ctx = file->private_data;
772 
773 	array = kmap(desc->page);
774 	for (i = desc->cache_entry_index; i < array->size; i++) {
775 		struct nfs_cache_array_entry *ent;
776 
777 		ent = &array->array[i];
778 		if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
779 		    nfs_compat_user_ino64(ent->ino), ent->d_type)) {
780 			desc->eof = true;
781 			break;
782 		}
783 		desc->ctx->pos++;
784 		if (i < (array->size-1))
785 			*desc->dir_cookie = array->array[i+1].cookie;
786 		else
787 			*desc->dir_cookie = array->last_cookie;
788 		if (ctx->duped != 0)
789 			ctx->duped = 1;
790 	}
791 	if (array->eof_index >= 0)
792 		desc->eof = true;
793 
794 	kunmap(desc->page);
795 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
796 			(unsigned long long)*desc->dir_cookie, res);
797 	return res;
798 }
799 
800 /*
801  * If we cannot find a cookie in our cache, we suspect that this is
802  * because it points to a deleted file, so we ask the server to return
803  * whatever it thinks is the next entry. We then feed this to filldir.
804  * If all goes well, we should then be able to find our way round the
805  * cache on the next call to readdir_search_pagecache();
806  *
807  * NOTE: we cannot add the anonymous page to the pagecache because
808  *	 the data it contains might not be page aligned. Besides,
809  *	 we should already have a complete representation of the
810  *	 directory in the page cache by the time we get here.
811  */
812 static inline
uncached_readdir(nfs_readdir_descriptor_t * desc)813 int uncached_readdir(nfs_readdir_descriptor_t *desc)
814 {
815 	struct page	*page = NULL;
816 	int		status;
817 	struct inode *inode = file_inode(desc->file);
818 	struct nfs_open_dir_context *ctx = desc->file->private_data;
819 
820 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
821 			(unsigned long long)*desc->dir_cookie);
822 
823 	page = alloc_page(GFP_HIGHUSER);
824 	if (!page) {
825 		status = -ENOMEM;
826 		goto out;
827 	}
828 
829 	desc->page_index = 0;
830 	desc->last_cookie = *desc->dir_cookie;
831 	desc->page = page;
832 	ctx->duped = 0;
833 
834 	status = nfs_readdir_xdr_to_array(desc, page, inode);
835 	if (status < 0)
836 		goto out_release;
837 
838 	status = nfs_do_filldir(desc);
839 
840  out_release:
841 	nfs_readdir_clear_array(desc->page);
842 	cache_page_release(desc);
843  out:
844 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
845 			__func__, status);
846 	return status;
847 }
848 
849 /* The file offset position represents the dirent entry number.  A
850    last cookie cache takes care of the common case of reading the
851    whole directory.
852  */
nfs_readdir(struct file * file,struct dir_context * ctx)853 static int nfs_readdir(struct file *file, struct dir_context *ctx)
854 {
855 	struct dentry	*dentry = file_dentry(file);
856 	struct inode	*inode = d_inode(dentry);
857 	nfs_readdir_descriptor_t my_desc,
858 			*desc = &my_desc;
859 	struct nfs_open_dir_context *dir_ctx = file->private_data;
860 	int res = 0;
861 
862 	dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
863 			file, (long long)ctx->pos);
864 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
865 
866 	/*
867 	 * ctx->pos points to the dirent entry number.
868 	 * *desc->dir_cookie has the cookie for the next entry. We have
869 	 * to either find the entry with the appropriate number or
870 	 * revalidate the cookie.
871 	 */
872 	memset(desc, 0, sizeof(*desc));
873 
874 	desc->file = file;
875 	desc->ctx = ctx;
876 	desc->dir_cookie = &dir_ctx->dir_cookie;
877 	desc->decode = NFS_PROTO(inode)->decode_dirent;
878 	desc->plus = nfs_use_readdirplus(inode, ctx);
879 
880 	if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
881 		res = nfs_revalidate_mapping(inode, file->f_mapping);
882 	if (res < 0)
883 		goto out;
884 
885 	do {
886 		res = readdir_search_pagecache(desc);
887 
888 		if (res == -EBADCOOKIE) {
889 			res = 0;
890 			/* This means either end of directory */
891 			if (*desc->dir_cookie && !desc->eof) {
892 				/* Or that the server has 'lost' a cookie */
893 				res = uncached_readdir(desc);
894 				if (res == 0)
895 					continue;
896 			}
897 			break;
898 		}
899 		if (res == -ETOOSMALL && desc->plus) {
900 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
901 			nfs_zap_caches(inode);
902 			desc->page_index = 0;
903 			desc->plus = false;
904 			desc->eof = false;
905 			continue;
906 		}
907 		if (res < 0)
908 			break;
909 
910 		res = nfs_do_filldir(desc);
911 		unlock_page(desc->page);
912 		cache_page_release(desc);
913 		if (res < 0)
914 			break;
915 	} while (!desc->eof);
916 out:
917 	if (res > 0)
918 		res = 0;
919 	dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
920 	return res;
921 }
922 
nfs_llseek_dir(struct file * filp,loff_t offset,int whence)923 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
924 {
925 	struct inode *inode = file_inode(filp);
926 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
927 
928 	dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
929 			filp, offset, whence);
930 
931 	switch (whence) {
932 	default:
933 		return -EINVAL;
934 	case SEEK_SET:
935 		if (offset < 0)
936 			return -EINVAL;
937 		inode_lock(inode);
938 		break;
939 	case SEEK_CUR:
940 		if (offset == 0)
941 			return filp->f_pos;
942 		inode_lock(inode);
943 		offset += filp->f_pos;
944 		if (offset < 0) {
945 			inode_unlock(inode);
946 			return -EINVAL;
947 		}
948 	}
949 	if (offset != filp->f_pos) {
950 		filp->f_pos = offset;
951 		dir_ctx->dir_cookie = 0;
952 		dir_ctx->duped = 0;
953 	}
954 	inode_unlock(inode);
955 	return offset;
956 }
957 
958 /*
959  * All directory operations under NFS are synchronous, so fsync()
960  * is a dummy operation.
961  */
nfs_fsync_dir(struct file * filp,loff_t start,loff_t end,int datasync)962 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
963 			 int datasync)
964 {
965 	struct inode *inode = file_inode(filp);
966 
967 	dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
968 
969 	inode_lock(inode);
970 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
971 	inode_unlock(inode);
972 	return 0;
973 }
974 
975 /**
976  * nfs_force_lookup_revalidate - Mark the directory as having changed
977  * @dir - pointer to directory inode
978  *
979  * This forces the revalidation code in nfs_lookup_revalidate() to do a
980  * full lookup on all child dentries of 'dir' whenever a change occurs
981  * on the server that might have invalidated our dcache.
982  *
983  * The caller should be holding dir->i_lock
984  */
nfs_force_lookup_revalidate(struct inode * dir)985 void nfs_force_lookup_revalidate(struct inode *dir)
986 {
987 	NFS_I(dir)->cache_change_attribute++;
988 }
989 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
990 
991 /*
992  * A check for whether or not the parent directory has changed.
993  * In the case it has, we assume that the dentries are untrustworthy
994  * and may need to be looked up again.
995  * If rcu_walk prevents us from performing a full check, return 0.
996  */
nfs_check_verifier(struct inode * dir,struct dentry * dentry,int rcu_walk)997 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
998 			      int rcu_walk)
999 {
1000 	if (IS_ROOT(dentry))
1001 		return 1;
1002 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1003 		return 0;
1004 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1005 		return 0;
1006 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1007 	if (nfs_mapping_need_revalidate_inode(dir)) {
1008 		if (rcu_walk)
1009 			return 0;
1010 		if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1011 			return 0;
1012 	}
1013 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1014 		return 0;
1015 	return 1;
1016 }
1017 
1018 /*
1019  * Use intent information to check whether or not we're going to do
1020  * an O_EXCL create using this path component.
1021  */
nfs_is_exclusive_create(struct inode * dir,unsigned int flags)1022 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1023 {
1024 	if (NFS_PROTO(dir)->version == 2)
1025 		return 0;
1026 	return flags & LOOKUP_EXCL;
1027 }
1028 
1029 /*
1030  * Inode and filehandle revalidation for lookups.
1031  *
1032  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1033  * or if the intent information indicates that we're about to open this
1034  * particular file and the "nocto" mount flag is not set.
1035  *
1036  */
1037 static
nfs_lookup_verify_inode(struct inode * inode,unsigned int flags)1038 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1039 {
1040 	struct nfs_server *server = NFS_SERVER(inode);
1041 	int ret;
1042 
1043 	if (IS_AUTOMOUNT(inode))
1044 		return 0;
1045 
1046 	if (flags & LOOKUP_OPEN) {
1047 		switch (inode->i_mode & S_IFMT) {
1048 		case S_IFREG:
1049 			/* A NFSv4 OPEN will revalidate later */
1050 			if (server->caps & NFS_CAP_ATOMIC_OPEN)
1051 				goto out;
1052 			/* Fallthrough */
1053 		case S_IFDIR:
1054 			if (server->flags & NFS_MOUNT_NOCTO)
1055 				break;
1056 			/* NFS close-to-open cache consistency validation */
1057 			goto out_force;
1058 		}
1059 	}
1060 
1061 	/* VFS wants an on-the-wire revalidation */
1062 	if (flags & LOOKUP_REVAL)
1063 		goto out_force;
1064 out:
1065 	return (inode->i_nlink == 0) ? -ESTALE : 0;
1066 out_force:
1067 	if (flags & LOOKUP_RCU)
1068 		return -ECHILD;
1069 	ret = __nfs_revalidate_inode(server, inode);
1070 	if (ret != 0)
1071 		return ret;
1072 	goto out;
1073 }
1074 
1075 /*
1076  * We judge how long we want to trust negative
1077  * dentries by looking at the parent inode mtime.
1078  *
1079  * If parent mtime has changed, we revalidate, else we wait for a
1080  * period corresponding to the parent's attribute cache timeout value.
1081  *
1082  * If LOOKUP_RCU prevents us from performing a full check, return 1
1083  * suggesting a reval is needed.
1084  *
1085  * Note that when creating a new file, or looking up a rename target,
1086  * then it shouldn't be necessary to revalidate a negative dentry.
1087  */
1088 static inline
nfs_neg_need_reval(struct inode * dir,struct dentry * dentry,unsigned int flags)1089 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1090 		       unsigned int flags)
1091 {
1092 	if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1093 		return 0;
1094 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1095 		return 1;
1096 	return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1097 }
1098 
1099 static int
nfs_lookup_revalidate_done(struct inode * dir,struct dentry * dentry,struct inode * inode,int error)1100 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1101 			   struct inode *inode, int error)
1102 {
1103 	switch (error) {
1104 	case 1:
1105 		dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1106 			__func__, dentry);
1107 		return 1;
1108 	case 0:
1109 		nfs_mark_for_revalidate(dir);
1110 		if (inode && S_ISDIR(inode->i_mode)) {
1111 			/* Purge readdir caches. */
1112 			nfs_zap_caches(inode);
1113 			/*
1114 			 * We can't d_drop the root of a disconnected tree:
1115 			 * its d_hash is on the s_anon list and d_drop() would hide
1116 			 * it from shrink_dcache_for_unmount(), leading to busy
1117 			 * inodes on unmount and further oopses.
1118 			 */
1119 			if (IS_ROOT(dentry))
1120 				return 1;
1121 		}
1122 		dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1123 				__func__, dentry);
1124 		return 0;
1125 	}
1126 	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1127 				__func__, dentry, error);
1128 	return error;
1129 }
1130 
1131 static int
nfs_lookup_revalidate_negative(struct inode * dir,struct dentry * dentry,unsigned int flags)1132 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1133 			       unsigned int flags)
1134 {
1135 	int ret = 1;
1136 	if (nfs_neg_need_reval(dir, dentry, flags)) {
1137 		if (flags & LOOKUP_RCU)
1138 			return -ECHILD;
1139 		ret = 0;
1140 	}
1141 	return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1142 }
1143 
1144 static int
nfs_lookup_revalidate_delegated(struct inode * dir,struct dentry * dentry,struct inode * inode)1145 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1146 				struct inode *inode)
1147 {
1148 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1149 	return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1150 }
1151 
1152 static int
nfs_lookup_revalidate_dentry(struct inode * dir,struct dentry * dentry,struct inode * inode)1153 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1154 			     struct inode *inode)
1155 {
1156 	struct nfs_fh *fhandle;
1157 	struct nfs_fattr *fattr;
1158 	struct nfs4_label *label;
1159 	int ret;
1160 
1161 	ret = -ENOMEM;
1162 	fhandle = nfs_alloc_fhandle();
1163 	fattr = nfs_alloc_fattr();
1164 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1165 	if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1166 		goto out;
1167 
1168 	ret = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1169 	if (ret < 0) {
1170 		if (ret == -ESTALE || ret == -ENOENT)
1171 			ret = 0;
1172 		goto out;
1173 	}
1174 	ret = 0;
1175 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1176 		goto out;
1177 	if (nfs_refresh_inode(inode, fattr) < 0)
1178 		goto out;
1179 
1180 	nfs_setsecurity(inode, fattr, label);
1181 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1182 
1183 	/* set a readdirplus hint that we had a cache miss */
1184 	nfs_force_use_readdirplus(dir);
1185 	ret = 1;
1186 out:
1187 	nfs_free_fattr(fattr);
1188 	nfs_free_fhandle(fhandle);
1189 	nfs4_label_free(label);
1190 	return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1191 }
1192 
1193 /*
1194  * This is called every time the dcache has a lookup hit,
1195  * and we should check whether we can really trust that
1196  * lookup.
1197  *
1198  * NOTE! The hit can be a negative hit too, don't assume
1199  * we have an inode!
1200  *
1201  * If the parent directory is seen to have changed, we throw out the
1202  * cached dentry and do a new lookup.
1203  */
1204 static int
nfs_do_lookup_revalidate(struct inode * dir,struct dentry * dentry,unsigned int flags)1205 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1206 			 unsigned int flags)
1207 {
1208 	struct inode *inode;
1209 	int error;
1210 
1211 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1212 	inode = d_inode(dentry);
1213 
1214 	if (!inode)
1215 		return nfs_lookup_revalidate_negative(dir, dentry, flags);
1216 
1217 	if (is_bad_inode(inode)) {
1218 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1219 				__func__, dentry);
1220 		goto out_bad;
1221 	}
1222 
1223 	if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1224 		return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1225 
1226 	/* Force a full look up iff the parent directory has changed */
1227 	if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1228 	    nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1229 		error = nfs_lookup_verify_inode(inode, flags);
1230 		if (error) {
1231 			if (error == -ESTALE)
1232 				nfs_zap_caches(dir);
1233 			goto out_bad;
1234 		}
1235 		nfs_advise_use_readdirplus(dir);
1236 		goto out_valid;
1237 	}
1238 
1239 	if (flags & LOOKUP_RCU)
1240 		return -ECHILD;
1241 
1242 	if (NFS_STALE(inode))
1243 		goto out_bad;
1244 
1245 	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1246 	error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1247 	trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1248 	return error;
1249 out_valid:
1250 	return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1251 out_bad:
1252 	if (flags & LOOKUP_RCU)
1253 		return -ECHILD;
1254 	return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1255 }
1256 
1257 static int
__nfs_lookup_revalidate(struct dentry * dentry,unsigned int flags,int (* reval)(struct inode *,struct dentry *,unsigned int))1258 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1259 			int (*reval)(struct inode *, struct dentry *, unsigned int))
1260 {
1261 	struct dentry *parent;
1262 	struct inode *dir;
1263 	int ret;
1264 
1265 	if (flags & LOOKUP_RCU) {
1266 		parent = READ_ONCE(dentry->d_parent);
1267 		dir = d_inode_rcu(parent);
1268 		if (!dir)
1269 			return -ECHILD;
1270 		ret = reval(dir, dentry, flags);
1271 		if (parent != READ_ONCE(dentry->d_parent))
1272 			return -ECHILD;
1273 	} else {
1274 		parent = dget_parent(dentry);
1275 		ret = reval(d_inode(parent), dentry, flags);
1276 		dput(parent);
1277 	}
1278 	return ret;
1279 }
1280 
nfs_lookup_revalidate(struct dentry * dentry,unsigned int flags)1281 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1282 {
1283 	return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1284 }
1285 
1286 /*
1287  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1288  * when we don't really care about the dentry name. This is called when a
1289  * pathwalk ends on a dentry that was not found via a normal lookup in the
1290  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1291  *
1292  * In this situation, we just want to verify that the inode itself is OK
1293  * since the dentry might have changed on the server.
1294  */
nfs_weak_revalidate(struct dentry * dentry,unsigned int flags)1295 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1296 {
1297 	struct inode *inode = d_inode(dentry);
1298 	int error = 0;
1299 
1300 	/*
1301 	 * I believe we can only get a negative dentry here in the case of a
1302 	 * procfs-style symlink. Just assume it's correct for now, but we may
1303 	 * eventually need to do something more here.
1304 	 */
1305 	if (!inode) {
1306 		dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1307 				__func__, dentry);
1308 		return 1;
1309 	}
1310 
1311 	if (is_bad_inode(inode)) {
1312 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1313 				__func__, dentry);
1314 		return 0;
1315 	}
1316 
1317 	error = nfs_lookup_verify_inode(inode, flags);
1318 	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1319 			__func__, inode->i_ino, error ? "invalid" : "valid");
1320 	return !error;
1321 }
1322 
1323 /*
1324  * This is called from dput() when d_count is going to 0.
1325  */
nfs_dentry_delete(const struct dentry * dentry)1326 static int nfs_dentry_delete(const struct dentry *dentry)
1327 {
1328 	dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1329 		dentry, dentry->d_flags);
1330 
1331 	/* Unhash any dentry with a stale inode */
1332 	if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1333 		return 1;
1334 
1335 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1336 		/* Unhash it, so that ->d_iput() would be called */
1337 		return 1;
1338 	}
1339 	if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1340 		/* Unhash it, so that ancestors of killed async unlink
1341 		 * files will be cleaned up during umount */
1342 		return 1;
1343 	}
1344 	return 0;
1345 
1346 }
1347 
1348 /* Ensure that we revalidate inode->i_nlink */
nfs_drop_nlink(struct inode * inode)1349 static void nfs_drop_nlink(struct inode *inode)
1350 {
1351 	spin_lock(&inode->i_lock);
1352 	/* drop the inode if we're reasonably sure this is the last link */
1353 	if (inode->i_nlink > 0)
1354 		drop_nlink(inode);
1355 	NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1356 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1357 		| NFS_INO_INVALID_CTIME
1358 		| NFS_INO_INVALID_OTHER
1359 		| NFS_INO_REVAL_FORCED;
1360 	spin_unlock(&inode->i_lock);
1361 }
1362 
1363 /*
1364  * Called when the dentry loses inode.
1365  * We use it to clean up silly-renamed files.
1366  */
nfs_dentry_iput(struct dentry * dentry,struct inode * inode)1367 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1368 {
1369 	if (S_ISDIR(inode->i_mode))
1370 		/* drop any readdir cache as it could easily be old */
1371 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1372 
1373 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1374 		nfs_complete_unlink(dentry, inode);
1375 		nfs_drop_nlink(inode);
1376 	}
1377 	iput(inode);
1378 }
1379 
nfs_d_release(struct dentry * dentry)1380 static void nfs_d_release(struct dentry *dentry)
1381 {
1382 	/* free cached devname value, if it survived that far */
1383 	if (unlikely(dentry->d_fsdata)) {
1384 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1385 			WARN_ON(1);
1386 		else
1387 			kfree(dentry->d_fsdata);
1388 	}
1389 }
1390 
1391 const struct dentry_operations nfs_dentry_operations = {
1392 	.d_revalidate	= nfs_lookup_revalidate,
1393 	.d_weak_revalidate	= nfs_weak_revalidate,
1394 	.d_delete	= nfs_dentry_delete,
1395 	.d_iput		= nfs_dentry_iput,
1396 	.d_automount	= nfs_d_automount,
1397 	.d_release	= nfs_d_release,
1398 };
1399 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1400 
nfs_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)1401 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1402 {
1403 	struct dentry *res;
1404 	struct inode *inode = NULL;
1405 	struct nfs_fh *fhandle = NULL;
1406 	struct nfs_fattr *fattr = NULL;
1407 	struct nfs4_label *label = NULL;
1408 	int error;
1409 
1410 	dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1411 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1412 
1413 	if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1414 		return ERR_PTR(-ENAMETOOLONG);
1415 
1416 	/*
1417 	 * If we're doing an exclusive create, optimize away the lookup
1418 	 * but don't hash the dentry.
1419 	 */
1420 	if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1421 		return NULL;
1422 
1423 	res = ERR_PTR(-ENOMEM);
1424 	fhandle = nfs_alloc_fhandle();
1425 	fattr = nfs_alloc_fattr();
1426 	if (fhandle == NULL || fattr == NULL)
1427 		goto out;
1428 
1429 	label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1430 	if (IS_ERR(label))
1431 		goto out;
1432 
1433 	trace_nfs_lookup_enter(dir, dentry, flags);
1434 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1435 	if (error == -ENOENT)
1436 		goto no_entry;
1437 	if (error < 0) {
1438 		res = ERR_PTR(error);
1439 		goto out_label;
1440 	}
1441 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1442 	res = ERR_CAST(inode);
1443 	if (IS_ERR(res))
1444 		goto out_label;
1445 
1446 	/* Notify readdir to use READDIRPLUS */
1447 	nfs_force_use_readdirplus(dir);
1448 
1449 no_entry:
1450 	res = d_splice_alias(inode, dentry);
1451 	if (res != NULL) {
1452 		if (IS_ERR(res))
1453 			goto out_label;
1454 		dentry = res;
1455 	}
1456 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1457 out_label:
1458 	trace_nfs_lookup_exit(dir, dentry, flags, error);
1459 	nfs4_label_free(label);
1460 out:
1461 	nfs_free_fattr(fattr);
1462 	nfs_free_fhandle(fhandle);
1463 	return res;
1464 }
1465 EXPORT_SYMBOL_GPL(nfs_lookup);
1466 
1467 #if IS_ENABLED(CONFIG_NFS_V4)
1468 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1469 
1470 const struct dentry_operations nfs4_dentry_operations = {
1471 	.d_revalidate	= nfs4_lookup_revalidate,
1472 	.d_weak_revalidate	= nfs_weak_revalidate,
1473 	.d_delete	= nfs_dentry_delete,
1474 	.d_iput		= nfs_dentry_iput,
1475 	.d_automount	= nfs_d_automount,
1476 	.d_release	= nfs_d_release,
1477 };
1478 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1479 
flags_to_mode(int flags)1480 static fmode_t flags_to_mode(int flags)
1481 {
1482 	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1483 	if ((flags & O_ACCMODE) != O_WRONLY)
1484 		res |= FMODE_READ;
1485 	if ((flags & O_ACCMODE) != O_RDONLY)
1486 		res |= FMODE_WRITE;
1487 	return res;
1488 }
1489 
create_nfs_open_context(struct dentry * dentry,int open_flags,struct file * filp)1490 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1491 {
1492 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1493 }
1494 
do_open(struct inode * inode,struct file * filp)1495 static int do_open(struct inode *inode, struct file *filp)
1496 {
1497 	nfs_fscache_open_file(inode, filp);
1498 	return 0;
1499 }
1500 
nfs_finish_open(struct nfs_open_context * ctx,struct dentry * dentry,struct file * file,unsigned open_flags)1501 static int nfs_finish_open(struct nfs_open_context *ctx,
1502 			   struct dentry *dentry,
1503 			   struct file *file, unsigned open_flags)
1504 {
1505 	int err;
1506 
1507 	err = finish_open(file, dentry, do_open);
1508 	if (err)
1509 		goto out;
1510 	if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1511 		nfs_file_set_open_context(file, ctx);
1512 	else
1513 		err = -EOPENSTALE;
1514 out:
1515 	return err;
1516 }
1517 
nfs_atomic_open(struct inode * dir,struct dentry * dentry,struct file * file,unsigned open_flags,umode_t mode)1518 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1519 		    struct file *file, unsigned open_flags,
1520 		    umode_t mode)
1521 {
1522 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1523 	struct nfs_open_context *ctx;
1524 	struct dentry *res;
1525 	struct iattr attr = { .ia_valid = ATTR_OPEN };
1526 	struct inode *inode;
1527 	unsigned int lookup_flags = 0;
1528 	bool switched = false;
1529 	int created = 0;
1530 	int err;
1531 
1532 	/* Expect a negative dentry */
1533 	BUG_ON(d_inode(dentry));
1534 
1535 	dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1536 			dir->i_sb->s_id, dir->i_ino, dentry);
1537 
1538 	err = nfs_check_flags(open_flags);
1539 	if (err)
1540 		return err;
1541 
1542 	/* NFS only supports OPEN on regular files */
1543 	if ((open_flags & O_DIRECTORY)) {
1544 		if (!d_in_lookup(dentry)) {
1545 			/*
1546 			 * Hashed negative dentry with O_DIRECTORY: dentry was
1547 			 * revalidated and is fine, no need to perform lookup
1548 			 * again
1549 			 */
1550 			return -ENOENT;
1551 		}
1552 		lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1553 		goto no_open;
1554 	}
1555 
1556 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1557 		return -ENAMETOOLONG;
1558 
1559 	if (open_flags & O_CREAT) {
1560 		struct nfs_server *server = NFS_SERVER(dir);
1561 
1562 		if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1563 			mode &= ~current_umask();
1564 
1565 		attr.ia_valid |= ATTR_MODE;
1566 		attr.ia_mode = mode;
1567 	}
1568 	if (open_flags & O_TRUNC) {
1569 		attr.ia_valid |= ATTR_SIZE;
1570 		attr.ia_size = 0;
1571 	}
1572 
1573 	if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1574 		d_drop(dentry);
1575 		switched = true;
1576 		dentry = d_alloc_parallel(dentry->d_parent,
1577 					  &dentry->d_name, &wq);
1578 		if (IS_ERR(dentry))
1579 			return PTR_ERR(dentry);
1580 		if (unlikely(!d_in_lookup(dentry)))
1581 			return finish_no_open(file, dentry);
1582 	}
1583 
1584 	ctx = create_nfs_open_context(dentry, open_flags, file);
1585 	err = PTR_ERR(ctx);
1586 	if (IS_ERR(ctx))
1587 		goto out;
1588 
1589 	trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1590 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1591 	if (created)
1592 		file->f_mode |= FMODE_CREATED;
1593 	if (IS_ERR(inode)) {
1594 		err = PTR_ERR(inode);
1595 		trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1596 		put_nfs_open_context(ctx);
1597 		d_drop(dentry);
1598 		switch (err) {
1599 		case -ENOENT:
1600 			d_splice_alias(NULL, dentry);
1601 			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1602 			break;
1603 		case -EISDIR:
1604 		case -ENOTDIR:
1605 			goto no_open;
1606 		case -ELOOP:
1607 			if (!(open_flags & O_NOFOLLOW))
1608 				goto no_open;
1609 			break;
1610 			/* case -EINVAL: */
1611 		default:
1612 			break;
1613 		}
1614 		goto out;
1615 	}
1616 
1617 	err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1618 	trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1619 	put_nfs_open_context(ctx);
1620 out:
1621 	if (unlikely(switched)) {
1622 		d_lookup_done(dentry);
1623 		dput(dentry);
1624 	}
1625 	return err;
1626 
1627 no_open:
1628 	res = nfs_lookup(dir, dentry, lookup_flags);
1629 	if (!res) {
1630 		inode = d_inode(dentry);
1631 		if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
1632 		    !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
1633 			res = ERR_PTR(-ENOTDIR);
1634 		else if (inode && S_ISREG(inode->i_mode))
1635 			res = ERR_PTR(-EOPENSTALE);
1636 	} else if (!IS_ERR(res)) {
1637 		inode = d_inode(res);
1638 		if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
1639 		    !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
1640 			dput(res);
1641 			res = ERR_PTR(-ENOTDIR);
1642 		} else if (inode && S_ISREG(inode->i_mode)) {
1643 			dput(res);
1644 			res = ERR_PTR(-EOPENSTALE);
1645 		}
1646 	}
1647 	if (switched) {
1648 		d_lookup_done(dentry);
1649 		if (!res)
1650 			res = dentry;
1651 		else
1652 			dput(dentry);
1653 	}
1654 	if (IS_ERR(res))
1655 		return PTR_ERR(res);
1656 	return finish_no_open(file, res);
1657 }
1658 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1659 
1660 static int
nfs4_do_lookup_revalidate(struct inode * dir,struct dentry * dentry,unsigned int flags)1661 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1662 			  unsigned int flags)
1663 {
1664 	struct inode *inode;
1665 
1666 	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1667 		goto full_reval;
1668 	if (d_mountpoint(dentry))
1669 		goto full_reval;
1670 
1671 	inode = d_inode(dentry);
1672 
1673 	/* We can't create new files in nfs_open_revalidate(), so we
1674 	 * optimize away revalidation of negative dentries.
1675 	 */
1676 	if (inode == NULL)
1677 		goto full_reval;
1678 
1679 	if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1680 		return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1681 
1682 	/* NFS only supports OPEN on regular files */
1683 	if (!S_ISREG(inode->i_mode))
1684 		goto full_reval;
1685 
1686 	/* We cannot do exclusive creation on a positive dentry */
1687 	if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
1688 		goto reval_dentry;
1689 
1690 	/* Check if the directory changed */
1691 	if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
1692 		goto reval_dentry;
1693 
1694 	/* Let f_op->open() actually open (and revalidate) the file */
1695 	return 1;
1696 reval_dentry:
1697 	if (flags & LOOKUP_RCU)
1698 		return -ECHILD;
1699 	return nfs_lookup_revalidate_dentry(dir, dentry, inode);;
1700 
1701 full_reval:
1702 	return nfs_do_lookup_revalidate(dir, dentry, flags);
1703 }
1704 
nfs4_lookup_revalidate(struct dentry * dentry,unsigned int flags)1705 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1706 {
1707 	return __nfs_lookup_revalidate(dentry, flags,
1708 			nfs4_do_lookup_revalidate);
1709 }
1710 
1711 #endif /* CONFIG_NFSV4 */
1712 
1713 /*
1714  * Code common to create, mkdir, and mknod.
1715  */
nfs_instantiate(struct dentry * dentry,struct nfs_fh * fhandle,struct nfs_fattr * fattr,struct nfs4_label * label)1716 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1717 				struct nfs_fattr *fattr,
1718 				struct nfs4_label *label)
1719 {
1720 	struct dentry *parent = dget_parent(dentry);
1721 	struct inode *dir = d_inode(parent);
1722 	struct inode *inode;
1723 	struct dentry *d;
1724 	int error = -EACCES;
1725 
1726 	d_drop(dentry);
1727 
1728 	/* We may have been initialized further down */
1729 	if (d_really_is_positive(dentry))
1730 		goto out;
1731 	if (fhandle->size == 0) {
1732 		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1733 		if (error)
1734 			goto out_error;
1735 	}
1736 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1737 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1738 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1739 		error = server->nfs_client->rpc_ops->getattr(server, fhandle,
1740 				fattr, NULL, NULL);
1741 		if (error < 0)
1742 			goto out_error;
1743 	}
1744 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1745 	d = d_splice_alias(inode, dentry);
1746 	if (IS_ERR(d)) {
1747 		error = PTR_ERR(d);
1748 		goto out_error;
1749 	}
1750 	dput(d);
1751 out:
1752 	dput(parent);
1753 	return 0;
1754 out_error:
1755 	nfs_mark_for_revalidate(dir);
1756 	dput(parent);
1757 	return error;
1758 }
1759 EXPORT_SYMBOL_GPL(nfs_instantiate);
1760 
1761 /*
1762  * Following a failed create operation, we drop the dentry rather
1763  * than retain a negative dentry. This avoids a problem in the event
1764  * that the operation succeeded on the server, but an error in the
1765  * reply path made it appear to have failed.
1766  */
nfs_create(struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)1767 int nfs_create(struct inode *dir, struct dentry *dentry,
1768 		umode_t mode, bool excl)
1769 {
1770 	struct iattr attr;
1771 	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1772 	int error;
1773 
1774 	dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1775 			dir->i_sb->s_id, dir->i_ino, dentry);
1776 
1777 	attr.ia_mode = mode;
1778 	attr.ia_valid = ATTR_MODE;
1779 
1780 	trace_nfs_create_enter(dir, dentry, open_flags);
1781 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1782 	trace_nfs_create_exit(dir, dentry, open_flags, error);
1783 	if (error != 0)
1784 		goto out_err;
1785 	return 0;
1786 out_err:
1787 	d_drop(dentry);
1788 	return error;
1789 }
1790 EXPORT_SYMBOL_GPL(nfs_create);
1791 
1792 /*
1793  * See comments for nfs_proc_create regarding failed operations.
1794  */
1795 int
nfs_mknod(struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)1796 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1797 {
1798 	struct iattr attr;
1799 	int status;
1800 
1801 	dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1802 			dir->i_sb->s_id, dir->i_ino, dentry);
1803 
1804 	attr.ia_mode = mode;
1805 	attr.ia_valid = ATTR_MODE;
1806 
1807 	trace_nfs_mknod_enter(dir, dentry);
1808 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1809 	trace_nfs_mknod_exit(dir, dentry, status);
1810 	if (status != 0)
1811 		goto out_err;
1812 	return 0;
1813 out_err:
1814 	d_drop(dentry);
1815 	return status;
1816 }
1817 EXPORT_SYMBOL_GPL(nfs_mknod);
1818 
1819 /*
1820  * See comments for nfs_proc_create regarding failed operations.
1821  */
nfs_mkdir(struct inode * dir,struct dentry * dentry,umode_t mode)1822 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1823 {
1824 	struct iattr attr;
1825 	int error;
1826 
1827 	dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1828 			dir->i_sb->s_id, dir->i_ino, dentry);
1829 
1830 	attr.ia_valid = ATTR_MODE;
1831 	attr.ia_mode = mode | S_IFDIR;
1832 
1833 	trace_nfs_mkdir_enter(dir, dentry);
1834 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1835 	trace_nfs_mkdir_exit(dir, dentry, error);
1836 	if (error != 0)
1837 		goto out_err;
1838 	return 0;
1839 out_err:
1840 	d_drop(dentry);
1841 	return error;
1842 }
1843 EXPORT_SYMBOL_GPL(nfs_mkdir);
1844 
nfs_dentry_handle_enoent(struct dentry * dentry)1845 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1846 {
1847 	if (simple_positive(dentry))
1848 		d_delete(dentry);
1849 }
1850 
nfs_rmdir(struct inode * dir,struct dentry * dentry)1851 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1852 {
1853 	int error;
1854 
1855 	dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1856 			dir->i_sb->s_id, dir->i_ino, dentry);
1857 
1858 	trace_nfs_rmdir_enter(dir, dentry);
1859 	if (d_really_is_positive(dentry)) {
1860 		down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1861 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1862 		/* Ensure the VFS deletes this inode */
1863 		switch (error) {
1864 		case 0:
1865 			clear_nlink(d_inode(dentry));
1866 			break;
1867 		case -ENOENT:
1868 			nfs_dentry_handle_enoent(dentry);
1869 		}
1870 		up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1871 	} else
1872 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1873 	trace_nfs_rmdir_exit(dir, dentry, error);
1874 
1875 	return error;
1876 }
1877 EXPORT_SYMBOL_GPL(nfs_rmdir);
1878 
1879 /*
1880  * Remove a file after making sure there are no pending writes,
1881  * and after checking that the file has only one user.
1882  *
1883  * We invalidate the attribute cache and free the inode prior to the operation
1884  * to avoid possible races if the server reuses the inode.
1885  */
nfs_safe_remove(struct dentry * dentry)1886 static int nfs_safe_remove(struct dentry *dentry)
1887 {
1888 	struct inode *dir = d_inode(dentry->d_parent);
1889 	struct inode *inode = d_inode(dentry);
1890 	int error = -EBUSY;
1891 
1892 	dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1893 
1894 	/* If the dentry was sillyrenamed, we simply call d_delete() */
1895 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1896 		error = 0;
1897 		goto out;
1898 	}
1899 
1900 	trace_nfs_remove_enter(dir, dentry);
1901 	if (inode != NULL) {
1902 		error = NFS_PROTO(dir)->remove(dir, dentry);
1903 		if (error == 0)
1904 			nfs_drop_nlink(inode);
1905 	} else
1906 		error = NFS_PROTO(dir)->remove(dir, dentry);
1907 	if (error == -ENOENT)
1908 		nfs_dentry_handle_enoent(dentry);
1909 	trace_nfs_remove_exit(dir, dentry, error);
1910 out:
1911 	return error;
1912 }
1913 
1914 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1915  *  belongs to an active ".nfs..." file and we return -EBUSY.
1916  *
1917  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1918  */
nfs_unlink(struct inode * dir,struct dentry * dentry)1919 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1920 {
1921 	int error;
1922 	int need_rehash = 0;
1923 
1924 	dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1925 		dir->i_ino, dentry);
1926 
1927 	trace_nfs_unlink_enter(dir, dentry);
1928 	spin_lock(&dentry->d_lock);
1929 	if (d_count(dentry) > 1) {
1930 		spin_unlock(&dentry->d_lock);
1931 		/* Start asynchronous writeout of the inode */
1932 		write_inode_now(d_inode(dentry), 0);
1933 		error = nfs_sillyrename(dir, dentry);
1934 		goto out;
1935 	}
1936 	if (!d_unhashed(dentry)) {
1937 		__d_drop(dentry);
1938 		need_rehash = 1;
1939 	}
1940 	spin_unlock(&dentry->d_lock);
1941 	error = nfs_safe_remove(dentry);
1942 	if (!error || error == -ENOENT) {
1943 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1944 	} else if (need_rehash)
1945 		d_rehash(dentry);
1946 out:
1947 	trace_nfs_unlink_exit(dir, dentry, error);
1948 	return error;
1949 }
1950 EXPORT_SYMBOL_GPL(nfs_unlink);
1951 
1952 /*
1953  * To create a symbolic link, most file systems instantiate a new inode,
1954  * add a page to it containing the path, then write it out to the disk
1955  * using prepare_write/commit_write.
1956  *
1957  * Unfortunately the NFS client can't create the in-core inode first
1958  * because it needs a file handle to create an in-core inode (see
1959  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1960  * symlink request has completed on the server.
1961  *
1962  * So instead we allocate a raw page, copy the symname into it, then do
1963  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1964  * now have a new file handle and can instantiate an in-core NFS inode
1965  * and move the raw page into its mapping.
1966  */
nfs_symlink(struct inode * dir,struct dentry * dentry,const char * symname)1967 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1968 {
1969 	struct page *page;
1970 	char *kaddr;
1971 	struct iattr attr;
1972 	unsigned int pathlen = strlen(symname);
1973 	int error;
1974 
1975 	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1976 		dir->i_ino, dentry, symname);
1977 
1978 	if (pathlen > PAGE_SIZE)
1979 		return -ENAMETOOLONG;
1980 
1981 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1982 	attr.ia_valid = ATTR_MODE;
1983 
1984 	page = alloc_page(GFP_USER);
1985 	if (!page)
1986 		return -ENOMEM;
1987 
1988 	kaddr = page_address(page);
1989 	memcpy(kaddr, symname, pathlen);
1990 	if (pathlen < PAGE_SIZE)
1991 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1992 
1993 	trace_nfs_symlink_enter(dir, dentry);
1994 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1995 	trace_nfs_symlink_exit(dir, dentry, error);
1996 	if (error != 0) {
1997 		dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1998 			dir->i_sb->s_id, dir->i_ino,
1999 			dentry, symname, error);
2000 		d_drop(dentry);
2001 		__free_page(page);
2002 		return error;
2003 	}
2004 
2005 	/*
2006 	 * No big deal if we can't add this page to the page cache here.
2007 	 * READLINK will get the missing page from the server if needed.
2008 	 */
2009 	if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2010 							GFP_KERNEL)) {
2011 		SetPageUptodate(page);
2012 		unlock_page(page);
2013 		/*
2014 		 * add_to_page_cache_lru() grabs an extra page refcount.
2015 		 * Drop it here to avoid leaking this page later.
2016 		 */
2017 		put_page(page);
2018 	} else
2019 		__free_page(page);
2020 
2021 	return 0;
2022 }
2023 EXPORT_SYMBOL_GPL(nfs_symlink);
2024 
2025 int
nfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)2026 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2027 {
2028 	struct inode *inode = d_inode(old_dentry);
2029 	int error;
2030 
2031 	dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2032 		old_dentry, dentry);
2033 
2034 	trace_nfs_link_enter(inode, dir, dentry);
2035 	d_drop(dentry);
2036 	if (S_ISREG(inode->i_mode))
2037 		nfs_sync_inode(inode);
2038 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2039 	if (error == 0) {
2040 		ihold(inode);
2041 		d_add(dentry, inode);
2042 	}
2043 	trace_nfs_link_exit(inode, dir, dentry, error);
2044 	return error;
2045 }
2046 EXPORT_SYMBOL_GPL(nfs_link);
2047 
2048 /*
2049  * RENAME
2050  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2051  * different file handle for the same inode after a rename (e.g. when
2052  * moving to a different directory). A fail-safe method to do so would
2053  * be to look up old_dir/old_name, create a link to new_dir/new_name and
2054  * rename the old file using the sillyrename stuff. This way, the original
2055  * file in old_dir will go away when the last process iput()s the inode.
2056  *
2057  * FIXED.
2058  *
2059  * It actually works quite well. One needs to have the possibility for
2060  * at least one ".nfs..." file in each directory the file ever gets
2061  * moved or linked to which happens automagically with the new
2062  * implementation that only depends on the dcache stuff instead of
2063  * using the inode layer
2064  *
2065  * Unfortunately, things are a little more complicated than indicated
2066  * above. For a cross-directory move, we want to make sure we can get
2067  * rid of the old inode after the operation.  This means there must be
2068  * no pending writes (if it's a file), and the use count must be 1.
2069  * If these conditions are met, we can drop the dentries before doing
2070  * the rename.
2071  */
nfs_rename(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)2072 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2073 	       struct inode *new_dir, struct dentry *new_dentry,
2074 	       unsigned int flags)
2075 {
2076 	struct inode *old_inode = d_inode(old_dentry);
2077 	struct inode *new_inode = d_inode(new_dentry);
2078 	struct dentry *dentry = NULL, *rehash = NULL;
2079 	struct rpc_task *task;
2080 	int error = -EBUSY;
2081 
2082 	if (flags)
2083 		return -EINVAL;
2084 
2085 	dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2086 		 old_dentry, new_dentry,
2087 		 d_count(new_dentry));
2088 
2089 	trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2090 	/*
2091 	 * For non-directories, check whether the target is busy and if so,
2092 	 * make a copy of the dentry and then do a silly-rename. If the
2093 	 * silly-rename succeeds, the copied dentry is hashed and becomes
2094 	 * the new target.
2095 	 */
2096 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2097 		/*
2098 		 * To prevent any new references to the target during the
2099 		 * rename, we unhash the dentry in advance.
2100 		 */
2101 		if (!d_unhashed(new_dentry)) {
2102 			d_drop(new_dentry);
2103 			rehash = new_dentry;
2104 		}
2105 
2106 		if (d_count(new_dentry) > 2) {
2107 			int err;
2108 
2109 			/* copy the target dentry's name */
2110 			dentry = d_alloc(new_dentry->d_parent,
2111 					 &new_dentry->d_name);
2112 			if (!dentry)
2113 				goto out;
2114 
2115 			/* silly-rename the existing target ... */
2116 			err = nfs_sillyrename(new_dir, new_dentry);
2117 			if (err)
2118 				goto out;
2119 
2120 			new_dentry = dentry;
2121 			rehash = NULL;
2122 			new_inode = NULL;
2123 		}
2124 	}
2125 
2126 	if (S_ISREG(old_inode->i_mode))
2127 		nfs_sync_inode(old_inode);
2128 	task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2129 	if (IS_ERR(task)) {
2130 		error = PTR_ERR(task);
2131 		goto out;
2132 	}
2133 
2134 	error = rpc_wait_for_completion_task(task);
2135 	if (error != 0) {
2136 		((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2137 		/* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2138 		smp_wmb();
2139 	} else
2140 		error = task->tk_status;
2141 	rpc_put_task(task);
2142 	/* Ensure the inode attributes are revalidated */
2143 	if (error == 0) {
2144 		spin_lock(&old_inode->i_lock);
2145 		NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2146 		NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2147 			| NFS_INO_INVALID_CTIME
2148 			| NFS_INO_REVAL_FORCED;
2149 		spin_unlock(&old_inode->i_lock);
2150 	}
2151 out:
2152 	if (rehash)
2153 		d_rehash(rehash);
2154 	trace_nfs_rename_exit(old_dir, old_dentry,
2155 			new_dir, new_dentry, error);
2156 	if (!error) {
2157 		if (new_inode != NULL)
2158 			nfs_drop_nlink(new_inode);
2159 		/*
2160 		 * The d_move() should be here instead of in an async RPC completion
2161 		 * handler because we need the proper locks to move the dentry.  If
2162 		 * we're interrupted by a signal, the async RPC completion handler
2163 		 * should mark the directories for revalidation.
2164 		 */
2165 		d_move(old_dentry, new_dentry);
2166 		nfs_set_verifier(old_dentry,
2167 					nfs_save_change_attribute(new_dir));
2168 	} else if (error == -ENOENT)
2169 		nfs_dentry_handle_enoent(old_dentry);
2170 
2171 	/* new dentry created? */
2172 	if (dentry)
2173 		dput(dentry);
2174 	return error;
2175 }
2176 EXPORT_SYMBOL_GPL(nfs_rename);
2177 
2178 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2179 static LIST_HEAD(nfs_access_lru_list);
2180 static atomic_long_t nfs_access_nr_entries;
2181 
2182 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2183 module_param(nfs_access_max_cachesize, ulong, 0644);
2184 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2185 
nfs_access_free_entry(struct nfs_access_entry * entry)2186 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2187 {
2188 	put_rpccred(entry->cred);
2189 	kfree_rcu(entry, rcu_head);
2190 	smp_mb__before_atomic();
2191 	atomic_long_dec(&nfs_access_nr_entries);
2192 	smp_mb__after_atomic();
2193 }
2194 
nfs_access_free_list(struct list_head * head)2195 static void nfs_access_free_list(struct list_head *head)
2196 {
2197 	struct nfs_access_entry *cache;
2198 
2199 	while (!list_empty(head)) {
2200 		cache = list_entry(head->next, struct nfs_access_entry, lru);
2201 		list_del(&cache->lru);
2202 		nfs_access_free_entry(cache);
2203 	}
2204 }
2205 
2206 static unsigned long
nfs_do_access_cache_scan(unsigned int nr_to_scan)2207 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2208 {
2209 	LIST_HEAD(head);
2210 	struct nfs_inode *nfsi, *next;
2211 	struct nfs_access_entry *cache;
2212 	long freed = 0;
2213 
2214 	spin_lock(&nfs_access_lru_lock);
2215 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2216 		struct inode *inode;
2217 
2218 		if (nr_to_scan-- == 0)
2219 			break;
2220 		inode = &nfsi->vfs_inode;
2221 		spin_lock(&inode->i_lock);
2222 		if (list_empty(&nfsi->access_cache_entry_lru))
2223 			goto remove_lru_entry;
2224 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2225 				struct nfs_access_entry, lru);
2226 		list_move(&cache->lru, &head);
2227 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2228 		freed++;
2229 		if (!list_empty(&nfsi->access_cache_entry_lru))
2230 			list_move_tail(&nfsi->access_cache_inode_lru,
2231 					&nfs_access_lru_list);
2232 		else {
2233 remove_lru_entry:
2234 			list_del_init(&nfsi->access_cache_inode_lru);
2235 			smp_mb__before_atomic();
2236 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2237 			smp_mb__after_atomic();
2238 		}
2239 		spin_unlock(&inode->i_lock);
2240 	}
2241 	spin_unlock(&nfs_access_lru_lock);
2242 	nfs_access_free_list(&head);
2243 	return freed;
2244 }
2245 
2246 unsigned long
nfs_access_cache_scan(struct shrinker * shrink,struct shrink_control * sc)2247 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2248 {
2249 	int nr_to_scan = sc->nr_to_scan;
2250 	gfp_t gfp_mask = sc->gfp_mask;
2251 
2252 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2253 		return SHRINK_STOP;
2254 	return nfs_do_access_cache_scan(nr_to_scan);
2255 }
2256 
2257 
2258 unsigned long
nfs_access_cache_count(struct shrinker * shrink,struct shrink_control * sc)2259 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2260 {
2261 	return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2262 }
2263 
2264 static void
nfs_access_cache_enforce_limit(void)2265 nfs_access_cache_enforce_limit(void)
2266 {
2267 	long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2268 	unsigned long diff;
2269 	unsigned int nr_to_scan;
2270 
2271 	if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2272 		return;
2273 	nr_to_scan = 100;
2274 	diff = nr_entries - nfs_access_max_cachesize;
2275 	if (diff < nr_to_scan)
2276 		nr_to_scan = diff;
2277 	nfs_do_access_cache_scan(nr_to_scan);
2278 }
2279 
__nfs_access_zap_cache(struct nfs_inode * nfsi,struct list_head * head)2280 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2281 {
2282 	struct rb_root *root_node = &nfsi->access_cache;
2283 	struct rb_node *n;
2284 	struct nfs_access_entry *entry;
2285 
2286 	/* Unhook entries from the cache */
2287 	while ((n = rb_first(root_node)) != NULL) {
2288 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2289 		rb_erase(n, root_node);
2290 		list_move(&entry->lru, head);
2291 	}
2292 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2293 }
2294 
nfs_access_zap_cache(struct inode * inode)2295 void nfs_access_zap_cache(struct inode *inode)
2296 {
2297 	LIST_HEAD(head);
2298 
2299 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2300 		return;
2301 	/* Remove from global LRU init */
2302 	spin_lock(&nfs_access_lru_lock);
2303 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2304 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2305 
2306 	spin_lock(&inode->i_lock);
2307 	__nfs_access_zap_cache(NFS_I(inode), &head);
2308 	spin_unlock(&inode->i_lock);
2309 	spin_unlock(&nfs_access_lru_lock);
2310 	nfs_access_free_list(&head);
2311 }
2312 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2313 
nfs_access_search_rbtree(struct inode * inode,struct rpc_cred * cred)2314 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2315 {
2316 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2317 	struct nfs_access_entry *entry;
2318 
2319 	while (n != NULL) {
2320 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2321 
2322 		if (cred < entry->cred)
2323 			n = n->rb_left;
2324 		else if (cred > entry->cred)
2325 			n = n->rb_right;
2326 		else
2327 			return entry;
2328 	}
2329 	return NULL;
2330 }
2331 
nfs_access_get_cached(struct inode * inode,struct rpc_cred * cred,struct nfs_access_entry * res,bool may_block)2332 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block)
2333 {
2334 	struct nfs_inode *nfsi = NFS_I(inode);
2335 	struct nfs_access_entry *cache;
2336 	bool retry = true;
2337 	int err;
2338 
2339 	spin_lock(&inode->i_lock);
2340 	for(;;) {
2341 		if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2342 			goto out_zap;
2343 		cache = nfs_access_search_rbtree(inode, cred);
2344 		err = -ENOENT;
2345 		if (cache == NULL)
2346 			goto out;
2347 		/* Found an entry, is our attribute cache valid? */
2348 		if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2349 			break;
2350 		err = -ECHILD;
2351 		if (!may_block)
2352 			goto out;
2353 		if (!retry)
2354 			goto out_zap;
2355 		spin_unlock(&inode->i_lock);
2356 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2357 		if (err)
2358 			return err;
2359 		spin_lock(&inode->i_lock);
2360 		retry = false;
2361 	}
2362 	res->cred = cache->cred;
2363 	res->mask = cache->mask;
2364 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2365 	err = 0;
2366 out:
2367 	spin_unlock(&inode->i_lock);
2368 	return err;
2369 out_zap:
2370 	spin_unlock(&inode->i_lock);
2371 	nfs_access_zap_cache(inode);
2372 	return -ENOENT;
2373 }
2374 
nfs_access_get_cached_rcu(struct inode * inode,struct rpc_cred * cred,struct nfs_access_entry * res)2375 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2376 {
2377 	/* Only check the most recently returned cache entry,
2378 	 * but do it without locking.
2379 	 */
2380 	struct nfs_inode *nfsi = NFS_I(inode);
2381 	struct nfs_access_entry *cache;
2382 	int err = -ECHILD;
2383 	struct list_head *lh;
2384 
2385 	rcu_read_lock();
2386 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2387 		goto out;
2388 	lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2389 	cache = list_entry(lh, struct nfs_access_entry, lru);
2390 	if (lh == &nfsi->access_cache_entry_lru ||
2391 	    cred != cache->cred)
2392 		cache = NULL;
2393 	if (cache == NULL)
2394 		goto out;
2395 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2396 		goto out;
2397 	res->cred = cache->cred;
2398 	res->mask = cache->mask;
2399 	err = 0;
2400 out:
2401 	rcu_read_unlock();
2402 	return err;
2403 }
2404 
nfs_access_add_rbtree(struct inode * inode,struct nfs_access_entry * set)2405 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2406 {
2407 	struct nfs_inode *nfsi = NFS_I(inode);
2408 	struct rb_root *root_node = &nfsi->access_cache;
2409 	struct rb_node **p = &root_node->rb_node;
2410 	struct rb_node *parent = NULL;
2411 	struct nfs_access_entry *entry;
2412 
2413 	spin_lock(&inode->i_lock);
2414 	while (*p != NULL) {
2415 		parent = *p;
2416 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2417 
2418 		if (set->cred < entry->cred)
2419 			p = &parent->rb_left;
2420 		else if (set->cred > entry->cred)
2421 			p = &parent->rb_right;
2422 		else
2423 			goto found;
2424 	}
2425 	rb_link_node(&set->rb_node, parent, p);
2426 	rb_insert_color(&set->rb_node, root_node);
2427 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2428 	spin_unlock(&inode->i_lock);
2429 	return;
2430 found:
2431 	rb_replace_node(parent, &set->rb_node, root_node);
2432 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2433 	list_del(&entry->lru);
2434 	spin_unlock(&inode->i_lock);
2435 	nfs_access_free_entry(entry);
2436 }
2437 
nfs_access_add_cache(struct inode * inode,struct nfs_access_entry * set)2438 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2439 {
2440 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2441 	if (cache == NULL)
2442 		return;
2443 	RB_CLEAR_NODE(&cache->rb_node);
2444 	cache->cred = get_rpccred(set->cred);
2445 	cache->mask = set->mask;
2446 
2447 	/* The above field assignments must be visible
2448 	 * before this item appears on the lru.  We cannot easily
2449 	 * use rcu_assign_pointer, so just force the memory barrier.
2450 	 */
2451 	smp_wmb();
2452 	nfs_access_add_rbtree(inode, cache);
2453 
2454 	/* Update accounting */
2455 	smp_mb__before_atomic();
2456 	atomic_long_inc(&nfs_access_nr_entries);
2457 	smp_mb__after_atomic();
2458 
2459 	/* Add inode to global LRU list */
2460 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2461 		spin_lock(&nfs_access_lru_lock);
2462 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2463 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2464 					&nfs_access_lru_list);
2465 		spin_unlock(&nfs_access_lru_lock);
2466 	}
2467 	nfs_access_cache_enforce_limit();
2468 }
2469 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2470 
2471 #define NFS_MAY_READ (NFS_ACCESS_READ)
2472 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2473 		NFS_ACCESS_EXTEND | \
2474 		NFS_ACCESS_DELETE)
2475 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2476 		NFS_ACCESS_EXTEND)
2477 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2478 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2479 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2480 static int
nfs_access_calc_mask(u32 access_result,umode_t umode)2481 nfs_access_calc_mask(u32 access_result, umode_t umode)
2482 {
2483 	int mask = 0;
2484 
2485 	if (access_result & NFS_MAY_READ)
2486 		mask |= MAY_READ;
2487 	if (S_ISDIR(umode)) {
2488 		if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2489 			mask |= MAY_WRITE;
2490 		if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2491 			mask |= MAY_EXEC;
2492 	} else if (S_ISREG(umode)) {
2493 		if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2494 			mask |= MAY_WRITE;
2495 		if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2496 			mask |= MAY_EXEC;
2497 	} else if (access_result & NFS_MAY_WRITE)
2498 			mask |= MAY_WRITE;
2499 	return mask;
2500 }
2501 
nfs_access_set_mask(struct nfs_access_entry * entry,u32 access_result)2502 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2503 {
2504 	entry->mask = access_result;
2505 }
2506 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2507 
nfs_do_access(struct inode * inode,struct rpc_cred * cred,int mask)2508 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2509 {
2510 	struct nfs_access_entry cache;
2511 	bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2512 	int cache_mask;
2513 	int status;
2514 
2515 	trace_nfs_access_enter(inode);
2516 
2517 	status = nfs_access_get_cached_rcu(inode, cred, &cache);
2518 	if (status != 0)
2519 		status = nfs_access_get_cached(inode, cred, &cache, may_block);
2520 	if (status == 0)
2521 		goto out_cached;
2522 
2523 	status = -ECHILD;
2524 	if (!may_block)
2525 		goto out;
2526 
2527 	/*
2528 	 * Determine which access bits we want to ask for...
2529 	 */
2530 	cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2531 	if (S_ISDIR(inode->i_mode))
2532 		cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2533 	else
2534 		cache.mask |= NFS_ACCESS_EXECUTE;
2535 	cache.cred = cred;
2536 	status = NFS_PROTO(inode)->access(inode, &cache);
2537 	if (status != 0) {
2538 		if (status == -ESTALE) {
2539 			nfs_zap_caches(inode);
2540 			if (!S_ISDIR(inode->i_mode))
2541 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2542 		}
2543 		goto out;
2544 	}
2545 	nfs_access_add_cache(inode, &cache);
2546 out_cached:
2547 	cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2548 	if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2549 		status = -EACCES;
2550 out:
2551 	trace_nfs_access_exit(inode, status);
2552 	return status;
2553 }
2554 
nfs_open_permission_mask(int openflags)2555 static int nfs_open_permission_mask(int openflags)
2556 {
2557 	int mask = 0;
2558 
2559 	if (openflags & __FMODE_EXEC) {
2560 		/* ONLY check exec rights */
2561 		mask = MAY_EXEC;
2562 	} else {
2563 		if ((openflags & O_ACCMODE) != O_WRONLY)
2564 			mask |= MAY_READ;
2565 		if ((openflags & O_ACCMODE) != O_RDONLY)
2566 			mask |= MAY_WRITE;
2567 	}
2568 
2569 	return mask;
2570 }
2571 
nfs_may_open(struct inode * inode,struct rpc_cred * cred,int openflags)2572 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2573 {
2574 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2575 }
2576 EXPORT_SYMBOL_GPL(nfs_may_open);
2577 
nfs_execute_ok(struct inode * inode,int mask)2578 static int nfs_execute_ok(struct inode *inode, int mask)
2579 {
2580 	struct nfs_server *server = NFS_SERVER(inode);
2581 	int ret = 0;
2582 
2583 	if (S_ISDIR(inode->i_mode))
2584 		return 0;
2585 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2586 		if (mask & MAY_NOT_BLOCK)
2587 			return -ECHILD;
2588 		ret = __nfs_revalidate_inode(server, inode);
2589 	}
2590 	if (ret == 0 && !execute_ok(inode))
2591 		ret = -EACCES;
2592 	return ret;
2593 }
2594 
nfs_permission(struct inode * inode,int mask)2595 int nfs_permission(struct inode *inode, int mask)
2596 {
2597 	struct rpc_cred *cred;
2598 	int res = 0;
2599 
2600 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2601 
2602 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2603 		goto out;
2604 	/* Is this sys_access() ? */
2605 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2606 		goto force_lookup;
2607 
2608 	switch (inode->i_mode & S_IFMT) {
2609 		case S_IFLNK:
2610 			goto out;
2611 		case S_IFREG:
2612 			if ((mask & MAY_OPEN) &&
2613 			   nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2614 				return 0;
2615 			break;
2616 		case S_IFDIR:
2617 			/*
2618 			 * Optimize away all write operations, since the server
2619 			 * will check permissions when we perform the op.
2620 			 */
2621 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2622 				goto out;
2623 	}
2624 
2625 force_lookup:
2626 	if (!NFS_PROTO(inode)->access)
2627 		goto out_notsup;
2628 
2629 	/* Always try fast lookups first */
2630 	rcu_read_lock();
2631 	cred = rpc_lookup_cred_nonblock();
2632 	if (!IS_ERR(cred))
2633 		res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2634 	else
2635 		res = PTR_ERR(cred);
2636 	rcu_read_unlock();
2637 	if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2638 		/* Fast lookup failed, try the slow way */
2639 		cred = rpc_lookup_cred();
2640 		if (!IS_ERR(cred)) {
2641 			res = nfs_do_access(inode, cred, mask);
2642 			put_rpccred(cred);
2643 		} else
2644 			res = PTR_ERR(cred);
2645 	}
2646 out:
2647 	if (!res && (mask & MAY_EXEC))
2648 		res = nfs_execute_ok(inode, mask);
2649 
2650 	dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2651 		inode->i_sb->s_id, inode->i_ino, mask, res);
2652 	return res;
2653 out_notsup:
2654 	if (mask & MAY_NOT_BLOCK)
2655 		return -ECHILD;
2656 
2657 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2658 	if (res == 0)
2659 		res = generic_permission(inode, mask);
2660 	goto out;
2661 }
2662 EXPORT_SYMBOL_GPL(nfs_permission);
2663 
2664 /*
2665  * Local variables:
2666  *  version-control: t
2667  *  kept-new-versions: 5
2668  * End:
2669  */
2670