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 */
nfs_fscache_get_client_cookie(struct nfs_client * clp)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 */
nfs_fscache_release_client_cookie(struct nfs_client * clp)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 */
nfs_fscache_get_super_cookie(struct super_block * sb,const char * uniq,int ulen)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 (!uniq) {
122 uniq = "";
123 ulen = 1;
124 }
125
126 key = kzalloc(sizeof(*key) + ulen, GFP_KERNEL);
127 if (!key)
128 return;
129
130 key->nfs_client = nfss->nfs_client;
131 key->key.super.s_flags = sb->s_flags & NFS_SB_MASK;
132 key->key.nfs_server.flags = nfss->flags;
133 key->key.nfs_server.rsize = nfss->rsize;
134 key->key.nfs_server.wsize = nfss->wsize;
135 key->key.nfs_server.acregmin = nfss->acregmin;
136 key->key.nfs_server.acregmax = nfss->acregmax;
137 key->key.nfs_server.acdirmin = nfss->acdirmin;
138 key->key.nfs_server.acdirmax = nfss->acdirmax;
139 key->key.nfs_server.fsid = nfss->fsid;
140 key->key.rpc_auth.au_flavor = nfss->client->cl_auth->au_flavor;
141
142 key->key.uniq_len = ulen;
143 memcpy(key->key.uniquifier, uniq, ulen);
144
145 spin_lock(&nfs_fscache_keys_lock);
146 p = &nfs_fscache_keys.rb_node;
147 parent = NULL;
148 while (*p) {
149 parent = *p;
150 xkey = rb_entry(parent, struct nfs_fscache_key, node);
151
152 if (key->nfs_client < xkey->nfs_client)
153 goto go_left;
154 if (key->nfs_client > xkey->nfs_client)
155 goto go_right;
156
157 diff = memcmp(&key->key, &xkey->key, sizeof(key->key));
158 if (diff < 0)
159 goto go_left;
160 if (diff > 0)
161 goto go_right;
162
163 if (key->key.uniq_len == 0)
164 goto non_unique;
165 diff = memcmp(key->key.uniquifier,
166 xkey->key.uniquifier,
167 key->key.uniq_len);
168 if (diff < 0)
169 goto go_left;
170 if (diff > 0)
171 goto go_right;
172 goto non_unique;
173
174 go_left:
175 p = &(*p)->rb_left;
176 continue;
177 go_right:
178 p = &(*p)->rb_right;
179 }
180
181 rb_link_node(&key->node, parent, p);
182 rb_insert_color(&key->node, &nfs_fscache_keys);
183 spin_unlock(&nfs_fscache_keys_lock);
184 nfss->fscache_key = key;
185
186 /* create a cache index for looking up filehandles */
187 nfss->fscache = fscache_acquire_cookie(nfss->nfs_client->fscache,
188 &nfs_fscache_super_index_def,
189 &key->key,
190 sizeof(key->key) + ulen,
191 NULL, 0,
192 nfss, 0, true);
193 dfprintk(FSCACHE, "NFS: get superblock cookie (0x%p/0x%p)\n",
194 nfss, nfss->fscache);
195 return;
196
197 non_unique:
198 spin_unlock(&nfs_fscache_keys_lock);
199 kfree(key);
200 nfss->fscache_key = NULL;
201 nfss->fscache = NULL;
202 printk(KERN_WARNING "NFS:"
203 " Cache request denied due to non-unique superblock keys\n");
204 }
205
206 /*
207 * release a per-superblock cookie
208 */
nfs_fscache_release_super_cookie(struct super_block * sb)209 void nfs_fscache_release_super_cookie(struct super_block *sb)
210 {
211 struct nfs_server *nfss = NFS_SB(sb);
212
213 dfprintk(FSCACHE, "NFS: releasing superblock cookie (0x%p/0x%p)\n",
214 nfss, nfss->fscache);
215
216 fscache_relinquish_cookie(nfss->fscache, NULL, false);
217 nfss->fscache = NULL;
218
219 if (nfss->fscache_key) {
220 spin_lock(&nfs_fscache_keys_lock);
221 rb_erase(&nfss->fscache_key->node, &nfs_fscache_keys);
222 spin_unlock(&nfs_fscache_keys_lock);
223 kfree(nfss->fscache_key);
224 nfss->fscache_key = NULL;
225 }
226 }
227
nfs_fscache_update_auxdata(struct nfs_fscache_inode_auxdata * auxdata,struct nfs_inode * nfsi)228 static void nfs_fscache_update_auxdata(struct nfs_fscache_inode_auxdata *auxdata,
229 struct nfs_inode *nfsi)
230 {
231 memset(auxdata, 0, sizeof(*auxdata));
232 auxdata->mtime_sec = nfsi->vfs_inode.i_mtime.tv_sec;
233 auxdata->mtime_nsec = nfsi->vfs_inode.i_mtime.tv_nsec;
234 auxdata->ctime_sec = nfsi->vfs_inode.i_ctime.tv_sec;
235 auxdata->ctime_nsec = nfsi->vfs_inode.i_ctime.tv_nsec;
236
237 if (NFS_SERVER(&nfsi->vfs_inode)->nfs_client->rpc_ops->version == 4)
238 auxdata->change_attr = inode_peek_iversion_raw(&nfsi->vfs_inode);
239 }
240
241 /*
242 * Initialise the per-inode cache cookie pointer for an NFS inode.
243 */
nfs_fscache_init_inode(struct inode * inode)244 void nfs_fscache_init_inode(struct inode *inode)
245 {
246 struct nfs_fscache_inode_auxdata auxdata;
247 struct nfs_server *nfss = NFS_SERVER(inode);
248 struct nfs_inode *nfsi = NFS_I(inode);
249
250 nfsi->fscache = NULL;
251 if (!(nfss->fscache && S_ISREG(inode->i_mode)))
252 return;
253
254 nfs_fscache_update_auxdata(&auxdata, nfsi);
255
256 nfsi->fscache = fscache_acquire_cookie(NFS_SB(inode->i_sb)->fscache,
257 &nfs_fscache_inode_object_def,
258 nfsi->fh.data, nfsi->fh.size,
259 &auxdata, sizeof(auxdata),
260 nfsi, nfsi->vfs_inode.i_size, false);
261 }
262
263 /*
264 * Release a per-inode cookie.
265 */
nfs_fscache_clear_inode(struct inode * inode)266 void nfs_fscache_clear_inode(struct inode *inode)
267 {
268 struct nfs_fscache_inode_auxdata auxdata;
269 struct nfs_inode *nfsi = NFS_I(inode);
270 struct fscache_cookie *cookie = nfs_i_fscache(inode);
271
272 dfprintk(FSCACHE, "NFS: clear cookie (0x%p/0x%p)\n", nfsi, cookie);
273
274 nfs_fscache_update_auxdata(&auxdata, nfsi);
275 fscache_relinquish_cookie(cookie, &auxdata, false);
276 nfsi->fscache = NULL;
277 }
278
nfs_fscache_can_enable(void * data)279 static bool nfs_fscache_can_enable(void *data)
280 {
281 struct inode *inode = data;
282
283 return !inode_is_open_for_write(inode);
284 }
285
286 /*
287 * Enable or disable caching for a file that is being opened as appropriate.
288 * The cookie is allocated when the inode is initialised, but is not enabled at
289 * that time. Enablement is deferred to file-open time to avoid stat() and
290 * access() thrashing the cache.
291 *
292 * For now, with NFS, only regular files that are open read-only will be able
293 * to use the cache.
294 *
295 * We enable the cache for an inode if we open it read-only and it isn't
296 * currently open for writing. We disable the cache if the inode is open
297 * write-only.
298 *
299 * The caller uses the file struct to pin i_writecount on the inode before
300 * calling us when a file is opened for writing, so we can make use of that.
301 *
302 * Note that this may be invoked multiple times in parallel by parallel
303 * nfs_open() functions.
304 */
nfs_fscache_open_file(struct inode * inode,struct file * filp)305 void nfs_fscache_open_file(struct inode *inode, struct file *filp)
306 {
307 struct nfs_fscache_inode_auxdata auxdata;
308 struct nfs_inode *nfsi = NFS_I(inode);
309 struct fscache_cookie *cookie = nfs_i_fscache(inode);
310
311 if (!fscache_cookie_valid(cookie))
312 return;
313
314 nfs_fscache_update_auxdata(&auxdata, nfsi);
315
316 if (inode_is_open_for_write(inode)) {
317 dfprintk(FSCACHE, "NFS: nfsi 0x%p disabling cache\n", nfsi);
318 clear_bit(NFS_INO_FSCACHE, &nfsi->flags);
319 fscache_disable_cookie(cookie, &auxdata, true);
320 fscache_uncache_all_inode_pages(cookie, inode);
321 } else {
322 dfprintk(FSCACHE, "NFS: nfsi 0x%p enabling cache\n", nfsi);
323 fscache_enable_cookie(cookie, &auxdata, nfsi->vfs_inode.i_size,
324 nfs_fscache_can_enable, inode);
325 if (fscache_cookie_enabled(cookie))
326 set_bit(NFS_INO_FSCACHE, &NFS_I(inode)->flags);
327 }
328 }
329 EXPORT_SYMBOL_GPL(nfs_fscache_open_file);
330
331 /*
332 * Release the caching state associated with a page, if the page isn't busy
333 * interacting with the cache.
334 * - Returns true (can release page) or false (page busy).
335 */
nfs_fscache_release_page(struct page * page,gfp_t gfp)336 int nfs_fscache_release_page(struct page *page, gfp_t gfp)
337 {
338 if (PageFsCache(page)) {
339 struct fscache_cookie *cookie = nfs_i_fscache(page->mapping->host);
340
341 BUG_ON(!cookie);
342 dfprintk(FSCACHE, "NFS: fscache releasepage (0x%p/0x%p/0x%p)\n",
343 cookie, page, NFS_I(page->mapping->host));
344
345 if (!fscache_maybe_release_page(cookie, page, gfp))
346 return 0;
347
348 nfs_inc_fscache_stats(page->mapping->host,
349 NFSIOS_FSCACHE_PAGES_UNCACHED);
350 }
351
352 return 1;
353 }
354
355 /*
356 * Release the caching state associated with a page if undergoing complete page
357 * invalidation.
358 */
__nfs_fscache_invalidate_page(struct page * page,struct inode * inode)359 void __nfs_fscache_invalidate_page(struct page *page, struct inode *inode)
360 {
361 struct fscache_cookie *cookie = nfs_i_fscache(inode);
362
363 BUG_ON(!cookie);
364
365 dfprintk(FSCACHE, "NFS: fscache invalidatepage (0x%p/0x%p/0x%p)\n",
366 cookie, page, NFS_I(inode));
367
368 fscache_wait_on_page_write(cookie, page);
369
370 BUG_ON(!PageLocked(page));
371 fscache_uncache_page(cookie, page);
372 nfs_inc_fscache_stats(page->mapping->host,
373 NFSIOS_FSCACHE_PAGES_UNCACHED);
374 }
375
376 /*
377 * Handle completion of a page being read from the cache.
378 * - Called in process (keventd) context.
379 */
nfs_readpage_from_fscache_complete(struct page * page,void * context,int error)380 static void nfs_readpage_from_fscache_complete(struct page *page,
381 void *context,
382 int error)
383 {
384 dfprintk(FSCACHE,
385 "NFS: readpage_from_fscache_complete (0x%p/0x%p/%d)\n",
386 page, context, error);
387
388 /*
389 * If the read completes with an error, mark the page with PG_checked,
390 * unlock the page, and let the VM reissue the readpage.
391 */
392 if (!error)
393 SetPageUptodate(page);
394 else
395 SetPageChecked(page);
396 unlock_page(page);
397 }
398
399 /*
400 * Retrieve a page from fscache
401 */
__nfs_readpage_from_fscache(struct nfs_open_context * ctx,struct inode * inode,struct page * page)402 int __nfs_readpage_from_fscache(struct nfs_open_context *ctx,
403 struct inode *inode, struct page *page)
404 {
405 int ret;
406
407 dfprintk(FSCACHE,
408 "NFS: readpage_from_fscache(fsc:%p/p:%p(i:%lx f:%lx)/0x%p)\n",
409 nfs_i_fscache(inode), page, page->index, page->flags, inode);
410
411 if (PageChecked(page)) {
412 ClearPageChecked(page);
413 return 1;
414 }
415
416 ret = fscache_read_or_alloc_page(nfs_i_fscache(inode),
417 page,
418 nfs_readpage_from_fscache_complete,
419 ctx,
420 GFP_KERNEL);
421
422 switch (ret) {
423 case 0: /* read BIO submitted (page in fscache) */
424 dfprintk(FSCACHE,
425 "NFS: readpage_from_fscache: BIO submitted\n");
426 nfs_inc_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_OK);
427 return ret;
428
429 case -ENOBUFS: /* inode not in cache */
430 case -ENODATA: /* page not in cache */
431 nfs_inc_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_FAIL);
432 dfprintk(FSCACHE,
433 "NFS: readpage_from_fscache %d\n", ret);
434 return 1;
435
436 default:
437 dfprintk(FSCACHE, "NFS: readpage_from_fscache %d\n", ret);
438 nfs_inc_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_FAIL);
439 }
440 return ret;
441 }
442
443 /*
444 * Retrieve a set of pages from fscache
445 */
__nfs_readpages_from_fscache(struct nfs_open_context * ctx,struct inode * inode,struct address_space * mapping,struct list_head * pages,unsigned * nr_pages)446 int __nfs_readpages_from_fscache(struct nfs_open_context *ctx,
447 struct inode *inode,
448 struct address_space *mapping,
449 struct list_head *pages,
450 unsigned *nr_pages)
451 {
452 unsigned npages = *nr_pages;
453 int ret;
454
455 dfprintk(FSCACHE, "NFS: nfs_getpages_from_fscache (0x%p/%u/0x%p)\n",
456 nfs_i_fscache(inode), npages, inode);
457
458 ret = fscache_read_or_alloc_pages(nfs_i_fscache(inode),
459 mapping, pages, nr_pages,
460 nfs_readpage_from_fscache_complete,
461 ctx,
462 mapping_gfp_mask(mapping));
463 if (*nr_pages < npages)
464 nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_OK,
465 npages);
466 if (*nr_pages > 0)
467 nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_FAIL,
468 *nr_pages);
469
470 switch (ret) {
471 case 0: /* read submitted to the cache for all pages */
472 BUG_ON(!list_empty(pages));
473 BUG_ON(*nr_pages != 0);
474 dfprintk(FSCACHE,
475 "NFS: nfs_getpages_from_fscache: submitted\n");
476
477 return ret;
478
479 case -ENOBUFS: /* some pages aren't cached and can't be */
480 case -ENODATA: /* some pages aren't cached */
481 dfprintk(FSCACHE,
482 "NFS: nfs_getpages_from_fscache: no page: %d\n", ret);
483 return 1;
484
485 default:
486 dfprintk(FSCACHE,
487 "NFS: nfs_getpages_from_fscache: ret %d\n", ret);
488 }
489
490 return ret;
491 }
492
493 /*
494 * Store a newly fetched page in fscache
495 * - PG_fscache must be set on the page
496 */
__nfs_readpage_to_fscache(struct inode * inode,struct page * page,int sync)497 void __nfs_readpage_to_fscache(struct inode *inode, struct page *page, int sync)
498 {
499 int ret;
500
501 dfprintk(FSCACHE,
502 "NFS: readpage_to_fscache(fsc:%p/p:%p(i:%lx f:%lx)/%d)\n",
503 nfs_i_fscache(inode), page, page->index, page->flags, sync);
504
505 ret = fscache_write_page(nfs_i_fscache(inode), page,
506 inode->i_size, GFP_KERNEL);
507 dfprintk(FSCACHE,
508 "NFS: readpage_to_fscache: p:%p(i:%lu f:%lx) ret %d\n",
509 page, page->index, page->flags, ret);
510
511 if (ret != 0) {
512 fscache_uncache_page(nfs_i_fscache(inode), page);
513 nfs_inc_fscache_stats(inode,
514 NFSIOS_FSCACHE_PAGES_WRITTEN_FAIL);
515 nfs_inc_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_UNCACHED);
516 } else {
517 nfs_inc_fscache_stats(inode,
518 NFSIOS_FSCACHE_PAGES_WRITTEN_OK);
519 }
520 }
521