xref: /openbmc/linux/fs/nfs/fscache.c (revision b2765275)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* NFS filesystem cache interface
3  *
4  * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #include <linux/init.h>
9 #include <linux/kernel.h>
10 #include <linux/sched.h>
11 #include <linux/mm.h>
12 #include <linux/nfs_fs.h>
13 #include <linux/nfs_fs_sb.h>
14 #include <linux/in6.h>
15 #include <linux/seq_file.h>
16 #include <linux/slab.h>
17 #include <linux/iversion.h>
18 
19 #include "internal.h"
20 #include "iostat.h"
21 #include "fscache.h"
22 
23 #define NFSDBG_FACILITY		NFSDBG_FSCACHE
24 
25 static struct rb_root nfs_fscache_keys = RB_ROOT;
26 static DEFINE_SPINLOCK(nfs_fscache_keys_lock);
27 
28 /*
29  * Layout of the key for an NFS server cache object.
30  */
31 struct nfs_server_key {
32 	struct {
33 		uint16_t	nfsversion;		/* NFS protocol version */
34 		uint32_t	minorversion;		/* NFSv4 minor version */
35 		uint16_t	family;			/* address family */
36 		__be16		port;			/* IP port */
37 	} hdr;
38 	union {
39 		struct in_addr	ipv4_addr;	/* IPv4 address */
40 		struct in6_addr ipv6_addr;	/* IPv6 address */
41 	};
42 } __packed;
43 
44 /*
45  * Get the per-client index cookie for an NFS client if the appropriate mount
46  * flag was set
47  * - We always try and get an index cookie for the client, but get filehandle
48  *   cookies on a per-superblock basis, depending on the mount flags
49  */
50 void nfs_fscache_get_client_cookie(struct nfs_client *clp)
51 {
52 	const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) &clp->cl_addr;
53 	const struct sockaddr_in *sin = (struct sockaddr_in *) &clp->cl_addr;
54 	struct nfs_server_key key;
55 	uint16_t len = sizeof(key.hdr);
56 
57 	memset(&key, 0, sizeof(key));
58 	key.hdr.nfsversion = clp->rpc_ops->version;
59 	key.hdr.minorversion = clp->cl_minorversion;
60 	key.hdr.family = clp->cl_addr.ss_family;
61 
62 	switch (clp->cl_addr.ss_family) {
63 	case AF_INET:
64 		key.hdr.port = sin->sin_port;
65 		key.ipv4_addr = sin->sin_addr;
66 		len += sizeof(key.ipv4_addr);
67 		break;
68 
69 	case AF_INET6:
70 		key.hdr.port = sin6->sin6_port;
71 		key.ipv6_addr = sin6->sin6_addr;
72 		len += sizeof(key.ipv6_addr);
73 		break;
74 
75 	default:
76 		printk(KERN_WARNING "NFS: Unknown network family '%d'\n",
77 		       clp->cl_addr.ss_family);
78 		clp->fscache = NULL;
79 		return;
80 	}
81 
82 	/* create a cache index for looking up filehandles */
83 	clp->fscache = fscache_acquire_cookie(nfs_fscache_netfs.primary_index,
84 					      &nfs_fscache_server_index_def,
85 					      &key, len,
86 					      NULL, 0,
87 					      clp, 0, true);
88 	dfprintk(FSCACHE, "NFS: get client cookie (0x%p/0x%p)\n",
89 		 clp, clp->fscache);
90 }
91 
92 /*
93  * Dispose of a per-client cookie
94  */
95 void nfs_fscache_release_client_cookie(struct nfs_client *clp)
96 {
97 	dfprintk(FSCACHE, "NFS: releasing client cookie (0x%p/0x%p)\n",
98 		 clp, clp->fscache);
99 
100 	fscache_relinquish_cookie(clp->fscache, NULL, false);
101 	clp->fscache = NULL;
102 }
103 
104 /*
105  * Get the cache cookie for an NFS superblock.  We have to handle
106  * uniquification here because the cache doesn't do it for us.
107  *
108  * The default uniquifier is just an empty string, but it may be overridden
109  * either by the 'fsc=xxx' option to mount, or by inheriting it from the parent
110  * superblock across an automount point of some nature.
111  */
112 void nfs_fscache_get_super_cookie(struct super_block *sb, const char *uniq, int ulen)
113 {
114 	struct nfs_fscache_key *key, *xkey;
115 	struct nfs_server *nfss = NFS_SB(sb);
116 	struct rb_node **p, *parent;
117 	int diff;
118 
119 	nfss->fscache_key = NULL;
120 	nfss->fscache = NULL;
121 	if (!(nfss->options & NFS_OPTION_FSCACHE))
122 		return;
123 	if (!uniq) {
124 		uniq = "";
125 		ulen = 1;
126 	}
127 
128 	key = kzalloc(sizeof(*key) + ulen, GFP_KERNEL);
129 	if (!key)
130 		return;
131 
132 	key->nfs_client = nfss->nfs_client;
133 	key->key.super.s_flags = sb->s_flags & NFS_SB_MASK;
134 	key->key.nfs_server.flags = nfss->flags;
135 	key->key.nfs_server.rsize = nfss->rsize;
136 	key->key.nfs_server.wsize = nfss->wsize;
137 	key->key.nfs_server.acregmin = nfss->acregmin;
138 	key->key.nfs_server.acregmax = nfss->acregmax;
139 	key->key.nfs_server.acdirmin = nfss->acdirmin;
140 	key->key.nfs_server.acdirmax = nfss->acdirmax;
141 	key->key.nfs_server.fsid = nfss->fsid;
142 	key->key.rpc_auth.au_flavor = nfss->client->cl_auth->au_flavor;
143 
144 	key->key.uniq_len = ulen;
145 	memcpy(key->key.uniquifier, uniq, ulen);
146 
147 	spin_lock(&nfs_fscache_keys_lock);
148 	p = &nfs_fscache_keys.rb_node;
149 	parent = NULL;
150 	while (*p) {
151 		parent = *p;
152 		xkey = rb_entry(parent, struct nfs_fscache_key, node);
153 
154 		if (key->nfs_client < xkey->nfs_client)
155 			goto go_left;
156 		if (key->nfs_client > xkey->nfs_client)
157 			goto go_right;
158 
159 		diff = memcmp(&key->key, &xkey->key, sizeof(key->key));
160 		if (diff < 0)
161 			goto go_left;
162 		if (diff > 0)
163 			goto go_right;
164 
165 		if (key->key.uniq_len == 0)
166 			goto non_unique;
167 		diff = memcmp(key->key.uniquifier,
168 			      xkey->key.uniquifier,
169 			      key->key.uniq_len);
170 		if (diff < 0)
171 			goto go_left;
172 		if (diff > 0)
173 			goto go_right;
174 		goto non_unique;
175 
176 	go_left:
177 		p = &(*p)->rb_left;
178 		continue;
179 	go_right:
180 		p = &(*p)->rb_right;
181 	}
182 
183 	rb_link_node(&key->node, parent, p);
184 	rb_insert_color(&key->node, &nfs_fscache_keys);
185 	spin_unlock(&nfs_fscache_keys_lock);
186 	nfss->fscache_key = key;
187 
188 	/* create a cache index for looking up filehandles */
189 	nfss->fscache = fscache_acquire_cookie(nfss->nfs_client->fscache,
190 					       &nfs_fscache_super_index_def,
191 					       key, sizeof(*key) + ulen,
192 					       NULL, 0,
193 					       nfss, 0, true);
194 	dfprintk(FSCACHE, "NFS: get superblock cookie (0x%p/0x%p)\n",
195 		 nfss, nfss->fscache);
196 	return;
197 
198 non_unique:
199 	spin_unlock(&nfs_fscache_keys_lock);
200 	kfree(key);
201 	nfss->fscache_key = NULL;
202 	nfss->fscache = NULL;
203 	printk(KERN_WARNING "NFS:"
204 	       " Cache request denied due to non-unique superblock keys\n");
205 }
206 
207 /*
208  * release a per-superblock cookie
209  */
210 void nfs_fscache_release_super_cookie(struct super_block *sb)
211 {
212 	struct nfs_server *nfss = NFS_SB(sb);
213 
214 	dfprintk(FSCACHE, "NFS: releasing superblock cookie (0x%p/0x%p)\n",
215 		 nfss, nfss->fscache);
216 
217 	fscache_relinquish_cookie(nfss->fscache, NULL, false);
218 	nfss->fscache = NULL;
219 
220 	if (nfss->fscache_key) {
221 		spin_lock(&nfs_fscache_keys_lock);
222 		rb_erase(&nfss->fscache_key->node, &nfs_fscache_keys);
223 		spin_unlock(&nfs_fscache_keys_lock);
224 		kfree(nfss->fscache_key);
225 		nfss->fscache_key = NULL;
226 	}
227 }
228 
229 /*
230  * Initialise the per-inode cache cookie pointer for an NFS inode.
231  */
232 void nfs_fscache_init_inode(struct inode *inode)
233 {
234 	struct nfs_fscache_inode_auxdata auxdata;
235 	struct nfs_server *nfss = NFS_SERVER(inode);
236 	struct nfs_inode *nfsi = NFS_I(inode);
237 
238 	nfsi->fscache = NULL;
239 	if (!(nfss->fscache && S_ISREG(inode->i_mode)))
240 		return;
241 
242 	memset(&auxdata, 0, sizeof(auxdata));
243 	auxdata.mtime_sec  = nfsi->vfs_inode.i_mtime.tv_sec;
244 	auxdata.mtime_nsec = nfsi->vfs_inode.i_mtime.tv_nsec;
245 	auxdata.ctime_sec  = nfsi->vfs_inode.i_ctime.tv_sec;
246 	auxdata.ctime_nsec = nfsi->vfs_inode.i_ctime.tv_nsec;
247 
248 	if (NFS_SERVER(&nfsi->vfs_inode)->nfs_client->rpc_ops->version == 4)
249 		auxdata.change_attr = inode_peek_iversion_raw(&nfsi->vfs_inode);
250 
251 	nfsi->fscache = fscache_acquire_cookie(NFS_SB(inode->i_sb)->fscache,
252 					       &nfs_fscache_inode_object_def,
253 					       nfsi->fh.data, nfsi->fh.size,
254 					       &auxdata, sizeof(auxdata),
255 					       nfsi, nfsi->vfs_inode.i_size, false);
256 }
257 
258 /*
259  * Release a per-inode cookie.
260  */
261 void nfs_fscache_clear_inode(struct inode *inode)
262 {
263 	struct nfs_fscache_inode_auxdata auxdata;
264 	struct nfs_inode *nfsi = NFS_I(inode);
265 	struct fscache_cookie *cookie = nfs_i_fscache(inode);
266 
267 	dfprintk(FSCACHE, "NFS: clear cookie (0x%p/0x%p)\n", nfsi, cookie);
268 
269 	memset(&auxdata, 0, sizeof(auxdata));
270 	auxdata.mtime_sec  = nfsi->vfs_inode.i_mtime.tv_sec;
271 	auxdata.mtime_nsec = nfsi->vfs_inode.i_mtime.tv_nsec;
272 	auxdata.ctime_sec  = nfsi->vfs_inode.i_ctime.tv_sec;
273 	auxdata.ctime_nsec = nfsi->vfs_inode.i_ctime.tv_nsec;
274 	fscache_relinquish_cookie(cookie, &auxdata, false);
275 	nfsi->fscache = NULL;
276 }
277 
278 static bool nfs_fscache_can_enable(void *data)
279 {
280 	struct inode *inode = data;
281 
282 	return !inode_is_open_for_write(inode);
283 }
284 
285 /*
286  * Enable or disable caching for a file that is being opened as appropriate.
287  * The cookie is allocated when the inode is initialised, but is not enabled at
288  * that time.  Enablement is deferred to file-open time to avoid stat() and
289  * access() thrashing the cache.
290  *
291  * For now, with NFS, only regular files that are open read-only will be able
292  * to use the cache.
293  *
294  * We enable the cache for an inode if we open it read-only and it isn't
295  * currently open for writing.  We disable the cache if the inode is open
296  * write-only.
297  *
298  * The caller uses the file struct to pin i_writecount on the inode before
299  * calling us when a file is opened for writing, so we can make use of that.
300  *
301  * Note that this may be invoked multiple times in parallel by parallel
302  * nfs_open() functions.
303  */
304 void nfs_fscache_open_file(struct inode *inode, struct file *filp)
305 {
306 	struct nfs_fscache_inode_auxdata auxdata;
307 	struct nfs_inode *nfsi = NFS_I(inode);
308 	struct fscache_cookie *cookie = nfs_i_fscache(inode);
309 
310 	if (!fscache_cookie_valid(cookie))
311 		return;
312 
313 	memset(&auxdata, 0, sizeof(auxdata));
314 	auxdata.mtime_sec  = nfsi->vfs_inode.i_mtime.tv_sec;
315 	auxdata.mtime_nsec = nfsi->vfs_inode.i_mtime.tv_nsec;
316 	auxdata.ctime_sec  = nfsi->vfs_inode.i_ctime.tv_sec;
317 	auxdata.ctime_nsec = nfsi->vfs_inode.i_ctime.tv_nsec;
318 
319 	if (inode_is_open_for_write(inode)) {
320 		dfprintk(FSCACHE, "NFS: nfsi 0x%p disabling cache\n", nfsi);
321 		clear_bit(NFS_INO_FSCACHE, &nfsi->flags);
322 		fscache_disable_cookie(cookie, &auxdata, true);
323 		fscache_uncache_all_inode_pages(cookie, inode);
324 	} else {
325 		dfprintk(FSCACHE, "NFS: nfsi 0x%p enabling cache\n", nfsi);
326 		fscache_enable_cookie(cookie, &auxdata, nfsi->vfs_inode.i_size,
327 				      nfs_fscache_can_enable, inode);
328 		if (fscache_cookie_enabled(cookie))
329 			set_bit(NFS_INO_FSCACHE, &NFS_I(inode)->flags);
330 	}
331 }
332 EXPORT_SYMBOL_GPL(nfs_fscache_open_file);
333 
334 /*
335  * Release the caching state associated with a page, if the page isn't busy
336  * interacting with the cache.
337  * - Returns true (can release page) or false (page busy).
338  */
339 int nfs_fscache_release_page(struct page *page, gfp_t gfp)
340 {
341 	if (PageFsCache(page)) {
342 		struct fscache_cookie *cookie = nfs_i_fscache(page->mapping->host);
343 
344 		BUG_ON(!cookie);
345 		dfprintk(FSCACHE, "NFS: fscache releasepage (0x%p/0x%p/0x%p)\n",
346 			 cookie, page, NFS_I(page->mapping->host));
347 
348 		if (!fscache_maybe_release_page(cookie, page, gfp))
349 			return 0;
350 
351 		nfs_inc_fscache_stats(page->mapping->host,
352 				      NFSIOS_FSCACHE_PAGES_UNCACHED);
353 	}
354 
355 	return 1;
356 }
357 
358 /*
359  * Release the caching state associated with a page if undergoing complete page
360  * invalidation.
361  */
362 void __nfs_fscache_invalidate_page(struct page *page, struct inode *inode)
363 {
364 	struct fscache_cookie *cookie = nfs_i_fscache(inode);
365 
366 	BUG_ON(!cookie);
367 
368 	dfprintk(FSCACHE, "NFS: fscache invalidatepage (0x%p/0x%p/0x%p)\n",
369 		 cookie, page, NFS_I(inode));
370 
371 	fscache_wait_on_page_write(cookie, page);
372 
373 	BUG_ON(!PageLocked(page));
374 	fscache_uncache_page(cookie, page);
375 	nfs_inc_fscache_stats(page->mapping->host,
376 			      NFSIOS_FSCACHE_PAGES_UNCACHED);
377 }
378 
379 /*
380  * Handle completion of a page being read from the cache.
381  * - Called in process (keventd) context.
382  */
383 static void nfs_readpage_from_fscache_complete(struct page *page,
384 					       void *context,
385 					       int error)
386 {
387 	dfprintk(FSCACHE,
388 		 "NFS: readpage_from_fscache_complete (0x%p/0x%p/%d)\n",
389 		 page, context, error);
390 
391 	/* if the read completes with an error, we just unlock the page and let
392 	 * the VM reissue the readpage */
393 	if (!error) {
394 		SetPageUptodate(page);
395 		unlock_page(page);
396 	} else {
397 		error = nfs_readpage_async(context, page->mapping->host, page);
398 		if (error)
399 			unlock_page(page);
400 	}
401 }
402 
403 /*
404  * Retrieve a page from fscache
405  */
406 int __nfs_readpage_from_fscache(struct nfs_open_context *ctx,
407 				struct inode *inode, struct page *page)
408 {
409 	int ret;
410 
411 	dfprintk(FSCACHE,
412 		 "NFS: readpage_from_fscache(fsc:%p/p:%p(i:%lx f:%lx)/0x%p)\n",
413 		 nfs_i_fscache(inode), page, page->index, page->flags, inode);
414 
415 	ret = fscache_read_or_alloc_page(nfs_i_fscache(inode),
416 					 page,
417 					 nfs_readpage_from_fscache_complete,
418 					 ctx,
419 					 GFP_KERNEL);
420 
421 	switch (ret) {
422 	case 0: /* read BIO submitted (page in fscache) */
423 		dfprintk(FSCACHE,
424 			 "NFS:    readpage_from_fscache: BIO submitted\n");
425 		nfs_inc_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_OK);
426 		return ret;
427 
428 	case -ENOBUFS: /* inode not in cache */
429 	case -ENODATA: /* page not in cache */
430 		nfs_inc_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_FAIL);
431 		dfprintk(FSCACHE,
432 			 "NFS:    readpage_from_fscache %d\n", ret);
433 		return 1;
434 
435 	default:
436 		dfprintk(FSCACHE, "NFS:    readpage_from_fscache %d\n", ret);
437 		nfs_inc_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_FAIL);
438 	}
439 	return ret;
440 }
441 
442 /*
443  * Retrieve a set of pages from fscache
444  */
445 int __nfs_readpages_from_fscache(struct nfs_open_context *ctx,
446 				 struct inode *inode,
447 				 struct address_space *mapping,
448 				 struct list_head *pages,
449 				 unsigned *nr_pages)
450 {
451 	unsigned npages = *nr_pages;
452 	int ret;
453 
454 	dfprintk(FSCACHE, "NFS: nfs_getpages_from_fscache (0x%p/%u/0x%p)\n",
455 		 nfs_i_fscache(inode), npages, inode);
456 
457 	ret = fscache_read_or_alloc_pages(nfs_i_fscache(inode),
458 					  mapping, pages, nr_pages,
459 					  nfs_readpage_from_fscache_complete,
460 					  ctx,
461 					  mapping_gfp_mask(mapping));
462 	if (*nr_pages < npages)
463 		nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_OK,
464 				      npages);
465 	if (*nr_pages > 0)
466 		nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_FAIL,
467 				      *nr_pages);
468 
469 	switch (ret) {
470 	case 0: /* read submitted to the cache for all pages */
471 		BUG_ON(!list_empty(pages));
472 		BUG_ON(*nr_pages != 0);
473 		dfprintk(FSCACHE,
474 			 "NFS: nfs_getpages_from_fscache: submitted\n");
475 
476 		return ret;
477 
478 	case -ENOBUFS: /* some pages aren't cached and can't be */
479 	case -ENODATA: /* some pages aren't cached */
480 		dfprintk(FSCACHE,
481 			 "NFS: nfs_getpages_from_fscache: no page: %d\n", ret);
482 		return 1;
483 
484 	default:
485 		dfprintk(FSCACHE,
486 			 "NFS: nfs_getpages_from_fscache: ret  %d\n", ret);
487 	}
488 
489 	return ret;
490 }
491 
492 /*
493  * Store a newly fetched page in fscache
494  * - PG_fscache must be set on the page
495  */
496 void __nfs_readpage_to_fscache(struct inode *inode, struct page *page, int sync)
497 {
498 	int ret;
499 
500 	dfprintk(FSCACHE,
501 		 "NFS: readpage_to_fscache(fsc:%p/p:%p(i:%lx f:%lx)/%d)\n",
502 		 nfs_i_fscache(inode), page, page->index, page->flags, sync);
503 
504 	ret = fscache_write_page(nfs_i_fscache(inode), page,
505 				 inode->i_size, GFP_KERNEL);
506 	dfprintk(FSCACHE,
507 		 "NFS:     readpage_to_fscache: p:%p(i:%lu f:%lx) ret %d\n",
508 		 page, page->index, page->flags, ret);
509 
510 	if (ret != 0) {
511 		fscache_uncache_page(nfs_i_fscache(inode), page);
512 		nfs_inc_fscache_stats(inode,
513 				      NFSIOS_FSCACHE_PAGES_WRITTEN_FAIL);
514 		nfs_inc_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_UNCACHED);
515 	} else {
516 		nfs_inc_fscache_stats(inode,
517 				      NFSIOS_FSCACHE_PAGES_WRITTEN_OK);
518 	}
519 }
520