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 #include <linux/xarray.h>
19 #include <linux/fscache.h>
20 #include <linux/netfs.h>
21
22 #include "internal.h"
23 #include "iostat.h"
24 #include "fscache.h"
25 #include "nfstrace.h"
26
27 #define NFS_MAX_KEY_LEN 1000
28
29 static bool nfs_append_int(char *key, int *_len, unsigned long long x)
30 {
31 if (*_len > NFS_MAX_KEY_LEN)
32 return false;
33 if (x == 0)
34 key[(*_len)++] = ',';
35 else
36 *_len += sprintf(key + *_len, ",%llx", x);
37 return true;
38 }
39
40 /*
41 * Get the per-client index cookie for an NFS client if the appropriate mount
42 * flag was set
43 * - We always try and get an index cookie for the client, but get filehandle
44 * cookies on a per-superblock basis, depending on the mount flags
45 */
46 static bool nfs_fscache_get_client_key(struct nfs_client *clp,
47 char *key, int *_len)
48 {
49 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) &clp->cl_addr;
50 const struct sockaddr_in *sin = (struct sockaddr_in *) &clp->cl_addr;
51
52 *_len += snprintf(key + *_len, NFS_MAX_KEY_LEN - *_len,
53 ",%u.%u,%x",
54 clp->rpc_ops->version,
55 clp->cl_minorversion,
56 clp->cl_addr.ss_family);
57
58 switch (clp->cl_addr.ss_family) {
59 case AF_INET:
60 if (!nfs_append_int(key, _len, sin->sin_port) ||
61 !nfs_append_int(key, _len, sin->sin_addr.s_addr))
62 return false;
63 return true;
64
65 case AF_INET6:
66 if (!nfs_append_int(key, _len, sin6->sin6_port) ||
67 !nfs_append_int(key, _len, sin6->sin6_addr.s6_addr32[0]) ||
68 !nfs_append_int(key, _len, sin6->sin6_addr.s6_addr32[1]) ||
69 !nfs_append_int(key, _len, sin6->sin6_addr.s6_addr32[2]) ||
70 !nfs_append_int(key, _len, sin6->sin6_addr.s6_addr32[3]))
71 return false;
72 return true;
73
74 default:
75 printk(KERN_WARNING "NFS: Unknown network family '%d'\n",
76 clp->cl_addr.ss_family);
77 return false;
78 }
79 }
80
81 /*
82 * Get the cache cookie for an NFS superblock.
83 *
84 * The default uniquifier is just an empty string, but it may be overridden
85 * either by the 'fsc=xxx' option to mount, or by inheriting it from the parent
86 * superblock across an automount point of some nature.
87 */
88 int nfs_fscache_get_super_cookie(struct super_block *sb, const char *uniq, int ulen)
89 {
90 struct fscache_volume *vcookie;
91 struct nfs_server *nfss = NFS_SB(sb);
92 unsigned int len = 3;
93 char *key;
94
95 if (uniq) {
96 nfss->fscache_uniq = kmemdup_nul(uniq, ulen, GFP_KERNEL);
97 if (!nfss->fscache_uniq)
98 return -ENOMEM;
99 }
100
101 key = kmalloc(NFS_MAX_KEY_LEN + 24, GFP_KERNEL);
102 if (!key)
103 return -ENOMEM;
104
105 memcpy(key, "nfs", 3);
106 if (!nfs_fscache_get_client_key(nfss->nfs_client, key, &len) ||
107 !nfs_append_int(key, &len, nfss->fsid.major) ||
108 !nfs_append_int(key, &len, nfss->fsid.minor) ||
109 !nfs_append_int(key, &len, sb->s_flags & NFS_SB_MASK) ||
110 !nfs_append_int(key, &len, nfss->flags) ||
111 !nfs_append_int(key, &len, nfss->rsize) ||
112 !nfs_append_int(key, &len, nfss->wsize) ||
113 !nfs_append_int(key, &len, nfss->acregmin) ||
114 !nfs_append_int(key, &len, nfss->acregmax) ||
115 !nfs_append_int(key, &len, nfss->acdirmin) ||
116 !nfs_append_int(key, &len, nfss->acdirmax) ||
117 !nfs_append_int(key, &len, nfss->client->cl_auth->au_flavor))
118 goto out;
119
120 if (ulen > 0) {
121 if (ulen > NFS_MAX_KEY_LEN - len)
122 goto out;
123 key[len++] = ',';
124 memcpy(key + len, uniq, ulen);
125 len += ulen;
126 }
127 key[len] = 0;
128
129 /* create a cache index for looking up filehandles */
130 vcookie = fscache_acquire_volume(key,
131 NULL, /* preferred_cache */
132 NULL, 0 /* coherency_data */);
133 if (IS_ERR(vcookie)) {
134 if (vcookie != ERR_PTR(-EBUSY)) {
135 kfree(key);
136 return PTR_ERR(vcookie);
137 }
138 pr_err("NFS: Cache volume key already in use (%s)\n", key);
139 vcookie = NULL;
140 }
141 nfss->fscache = vcookie;
142
143 out:
144 kfree(key);
145 return 0;
146 }
147
148 /*
149 * release a per-superblock cookie
150 */
151 void nfs_fscache_release_super_cookie(struct super_block *sb)
152 {
153 struct nfs_server *nfss = NFS_SB(sb);
154
155 fscache_relinquish_volume(nfss->fscache, NULL, false);
156 nfss->fscache = NULL;
157 kfree(nfss->fscache_uniq);
158 }
159
160 /*
161 * Initialise the per-inode cache cookie pointer for an NFS inode.
162 */
163 void nfs_fscache_init_inode(struct inode *inode)
164 {
165 struct nfs_fscache_inode_auxdata auxdata;
166 struct nfs_server *nfss = NFS_SERVER(inode);
167 struct nfs_inode *nfsi = NFS_I(inode);
168
169 netfs_inode(inode)->cache = NULL;
170 if (!(nfss->fscache && S_ISREG(inode->i_mode)))
171 return;
172
173 nfs_fscache_update_auxdata(&auxdata, inode);
174
175 netfs_inode(inode)->cache = fscache_acquire_cookie(
176 nfss->fscache,
177 0,
178 nfsi->fh.data, /* index_key */
179 nfsi->fh.size,
180 &auxdata, /* aux_data */
181 sizeof(auxdata),
182 i_size_read(inode));
183
184 if (netfs_inode(inode)->cache)
185 mapping_set_release_always(inode->i_mapping);
186 }
187
188 /*
189 * Release a per-inode cookie.
190 */
191 void nfs_fscache_clear_inode(struct inode *inode)
192 {
193 fscache_relinquish_cookie(netfs_i_cookie(netfs_inode(inode)), false);
194 netfs_inode(inode)->cache = NULL;
195 }
196
197 /*
198 * Enable or disable caching for a file that is being opened as appropriate.
199 * The cookie is allocated when the inode is initialised, but is not enabled at
200 * that time. Enablement is deferred to file-open time to avoid stat() and
201 * access() thrashing the cache.
202 *
203 * For now, with NFS, only regular files that are open read-only will be able
204 * to use the cache.
205 *
206 * We enable the cache for an inode if we open it read-only and it isn't
207 * currently open for writing. We disable the cache if the inode is open
208 * write-only.
209 *
210 * The caller uses the file struct to pin i_writecount on the inode before
211 * calling us when a file is opened for writing, so we can make use of that.
212 *
213 * Note that this may be invoked multiple times in parallel by parallel
214 * nfs_open() functions.
215 */
216 void nfs_fscache_open_file(struct inode *inode, struct file *filp)
217 {
218 struct nfs_fscache_inode_auxdata auxdata;
219 struct fscache_cookie *cookie = netfs_i_cookie(netfs_inode(inode));
220 bool open_for_write = inode_is_open_for_write(inode);
221
222 if (!fscache_cookie_valid(cookie))
223 return;
224
225 fscache_use_cookie(cookie, open_for_write);
226 if (open_for_write) {
227 nfs_fscache_update_auxdata(&auxdata, inode);
228 fscache_invalidate(cookie, &auxdata, i_size_read(inode),
229 FSCACHE_INVAL_DIO_WRITE);
230 }
231 }
232 EXPORT_SYMBOL_GPL(nfs_fscache_open_file);
233
234 void nfs_fscache_release_file(struct inode *inode, struct file *filp)
235 {
236 struct nfs_fscache_inode_auxdata auxdata;
237 struct fscache_cookie *cookie = netfs_i_cookie(netfs_inode(inode));
238 loff_t i_size = i_size_read(inode);
239
240 nfs_fscache_update_auxdata(&auxdata, inode);
241 fscache_unuse_cookie(cookie, &auxdata, &i_size);
242 }
243
244 int nfs_netfs_read_folio(struct file *file, struct folio *folio)
245 {
246 if (!netfs_inode(folio_inode(folio))->cache)
247 return -ENOBUFS;
248
249 return netfs_read_folio(file, folio);
250 }
251
252 int nfs_netfs_readahead(struct readahead_control *ractl)
253 {
254 struct inode *inode = ractl->mapping->host;
255
256 if (!netfs_inode(inode)->cache)
257 return -ENOBUFS;
258
259 netfs_readahead(ractl);
260 return 0;
261 }
262
263 static atomic_t nfs_netfs_debug_id;
264 static int nfs_netfs_init_request(struct netfs_io_request *rreq, struct file *file)
265 {
266 rreq->netfs_priv = get_nfs_open_context(nfs_file_open_context(file));
267 rreq->debug_id = atomic_inc_return(&nfs_netfs_debug_id);
268 /* [DEPRECATED] Use PG_private_2 to mark folio being written to the cache. */
269 __set_bit(NETFS_RREQ_USE_PGPRIV2, &rreq->flags);
270
271 return 0;
272 }
273
274 static void nfs_netfs_free_request(struct netfs_io_request *rreq)
275 {
276 put_nfs_open_context(rreq->netfs_priv);
277 }
278
279 static struct nfs_netfs_io_data *nfs_netfs_alloc(struct netfs_io_subrequest *sreq)
280 {
281 struct nfs_netfs_io_data *netfs;
282
283 netfs = kzalloc(sizeof(*netfs), GFP_KERNEL_ACCOUNT);
284 if (!netfs)
285 return NULL;
286 netfs->sreq = sreq;
287 refcount_set(&netfs->refcount, 1);
288 return netfs;
289 }
290
291 static bool nfs_netfs_clamp_length(struct netfs_io_subrequest *sreq)
292 {
293 size_t rsize = NFS_SB(sreq->rreq->inode->i_sb)->rsize;
294
295 sreq->len = min(sreq->len, rsize);
296 return true;
297 }
298
299 static void nfs_netfs_issue_read(struct netfs_io_subrequest *sreq)
300 {
301 struct nfs_netfs_io_data *netfs;
302 struct nfs_pageio_descriptor pgio;
303 struct inode *inode = sreq->rreq->inode;
304 struct nfs_open_context *ctx = sreq->rreq->netfs_priv;
305 struct page *page;
306 unsigned long idx;
307 int err;
308 pgoff_t start = (sreq->start + sreq->transferred) >> PAGE_SHIFT;
309 pgoff_t last = ((sreq->start + sreq->len -
310 sreq->transferred - 1) >> PAGE_SHIFT);
311
312 nfs_pageio_init_read(&pgio, inode, false,
313 &nfs_async_read_completion_ops);
314
315 netfs = nfs_netfs_alloc(sreq);
316 if (!netfs)
317 return netfs_subreq_terminated(sreq, -ENOMEM, false);
318
319 pgio.pg_netfs = netfs; /* used in completion */
320
321 xa_for_each_range(&sreq->rreq->mapping->i_pages, idx, page, start, last) {
322 /* nfs_read_add_folio() may schedule() due to pNFS layout and other RPCs */
323 err = nfs_read_add_folio(&pgio, ctx, page_folio(page));
324 if (err < 0) {
325 netfs->error = err;
326 goto out;
327 }
328 }
329 out:
330 nfs_pageio_complete_read(&pgio);
331 nfs_netfs_put(netfs);
332 }
333
334 void nfs_netfs_initiate_read(struct nfs_pgio_header *hdr)
335 {
336 struct nfs_netfs_io_data *netfs = hdr->netfs;
337
338 if (!netfs)
339 return;
340
341 nfs_netfs_get(netfs);
342 }
343
344 int nfs_netfs_folio_unlock(struct folio *folio)
345 {
346 struct inode *inode = folio->mapping->host;
347
348 /*
349 * If fscache is enabled, netfs will unlock pages.
350 */
351 if (netfs_inode(inode)->cache)
352 return 0;
353
354 return 1;
355 }
356
357 void nfs_netfs_read_completion(struct nfs_pgio_header *hdr)
358 {
359 struct nfs_netfs_io_data *netfs = hdr->netfs;
360 struct netfs_io_subrequest *sreq;
361
362 if (!netfs)
363 return;
364
365 sreq = netfs->sreq;
366 if (test_bit(NFS_IOHDR_EOF, &hdr->flags) &&
367 sreq->rreq->origin != NETFS_DIO_READ)
368 __set_bit(NETFS_SREQ_CLEAR_TAIL, &sreq->flags);
369
370 if (hdr->error)
371 netfs->error = hdr->error;
372 else
373 atomic64_add(hdr->res.count, &netfs->transferred);
374
375 nfs_netfs_put(netfs);
376 hdr->netfs = NULL;
377 }
378
379 const struct netfs_request_ops nfs_netfs_ops = {
380 .init_request = nfs_netfs_init_request,
381 .free_request = nfs_netfs_free_request,
382 .issue_read = nfs_netfs_issue_read,
383 .clamp_length = nfs_netfs_clamp_length
384 };
385
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.