Commit | Line | Data |
---|---|---|
3f054211 | 1 | // SPDX-License-Identifier: GPL-2.0 |
65294c1f | 2 | /* |
3f054211 | 3 | * The NFSD open file cache. |
65294c1f JL |
4 | * |
5 | * (c) 2015 - Jeff Layton <jeff.layton@primarydata.com> | |
b3276c1f CL |
6 | * |
7 | * An nfsd_file object is a per-file collection of open state that binds | |
8 | * together: | |
9 | * - a struct file * | |
10 | * - a user credential | |
11 | * - a network namespace | |
12 | * - a read-ahead context | |
13 | * - monitoring for writeback errors | |
14 | * | |
15 | * nfsd_file objects are reference-counted. Consumers acquire a new | |
16 | * object via the nfsd_file_acquire API. They manage their interest in | |
17 | * the acquired object, and hence the object's reference count, via | |
18 | * nfsd_file_get and nfsd_file_put. There are two varieties of nfsd_file | |
19 | * object: | |
20 | * | |
21 | * * non-garbage-collected: When a consumer wants to precisely control | |
22 | * the lifetime of a file's open state, it acquires a non-garbage- | |
23 | * collected nfsd_file. The final nfsd_file_put releases the open | |
24 | * state immediately. | |
25 | * | |
26 | * * garbage-collected: When a consumer does not control the lifetime | |
27 | * of open state, it acquires a garbage-collected nfsd_file. The | |
28 | * final nfsd_file_put allows the open state to linger for a period | |
29 | * during which it may be re-used. | |
65294c1f JL |
30 | */ |
31 | ||
32 | #include <linux/hash.h> | |
33 | #include <linux/slab.h> | |
65294c1f | 34 | #include <linux/file.h> |
cbcc268b | 35 | #include <linux/pagemap.h> |
65294c1f JL |
36 | #include <linux/sched.h> |
37 | #include <linux/list_lru.h> | |
38 | #include <linux/fsnotify_backend.h> | |
39 | #include <linux/fsnotify.h> | |
40 | #include <linux/seq_file.h> | |
fc22945e | 41 | #include <linux/rhashtable.h> |
65294c1f JL |
42 | |
43 | #include "vfs.h" | |
44 | #include "nfsd.h" | |
45 | #include "nfsfh.h" | |
5e113224 | 46 | #include "netns.h" |
65294c1f JL |
47 | #include "filecache.h" |
48 | #include "trace.h" | |
49 | ||
65294c1f JL |
50 | #define NFSD_LAUNDRETTE_DELAY (2 * HZ) |
51 | ||
c7b824c3 | 52 | #define NFSD_FILE_CACHE_UP (0) |
65294c1f JL |
53 | |
54 | /* We only care about NFSD_MAY_READ/WRITE for this cache */ | |
55 | #define NFSD_FILE_MAY_MASK (NFSD_MAY_READ|NFSD_MAY_WRITE) | |
56 | ||
65294c1f | 57 | static DEFINE_PER_CPU(unsigned long, nfsd_file_cache_hits); |
29d4bdbb | 58 | static DEFINE_PER_CPU(unsigned long, nfsd_file_acquisitions); |
d6329327 | 59 | static DEFINE_PER_CPU(unsigned long, nfsd_file_releases); |
904940e9 | 60 | static DEFINE_PER_CPU(unsigned long, nfsd_file_total_age); |
94660cc1 | 61 | static DEFINE_PER_CPU(unsigned long, nfsd_file_evictions); |
65294c1f | 62 | |
9542e6a6 | 63 | struct nfsd_fcache_disposal { |
9542e6a6 | 64 | struct work_struct work; |
9542e6a6 TM |
65 | spinlock_t lock; |
66 | struct list_head freeme; | |
9542e6a6 TM |
67 | }; |
68 | ||
50d0def9 | 69 | static struct workqueue_struct *nfsd_filecache_wq __read_mostly; |
9542e6a6 | 70 | |
65294c1f JL |
71 | static struct kmem_cache *nfsd_file_slab; |
72 | static struct kmem_cache *nfsd_file_mark_slab; | |
65294c1f | 73 | static struct list_lru nfsd_file_lru; |
c7b824c3 | 74 | static unsigned long nfsd_file_flags; |
65294c1f | 75 | static struct fsnotify_group *nfsd_file_fsnotify_group; |
65294c1f | 76 | static struct delayed_work nfsd_filecache_laundrette; |
fc22945e CL |
77 | static struct rhashtable nfsd_file_rhash_tbl |
78 | ____cacheline_aligned_in_smp; | |
79 | ||
80 | enum nfsd_file_lookup_type { | |
81 | NFSD_FILE_KEY_INODE, | |
82 | NFSD_FILE_KEY_FULL, | |
83 | }; | |
84 | ||
85 | struct nfsd_file_lookup_key { | |
86 | struct inode *inode; | |
87 | struct net *net; | |
88 | const struct cred *cred; | |
89 | unsigned char need; | |
4d1ea845 | 90 | bool gc; |
fc22945e CL |
91 | enum nfsd_file_lookup_type type; |
92 | }; | |
93 | ||
94 | /* | |
95 | * The returned hash value is based solely on the address of an in-code | |
96 | * inode, a pointer to a slab-allocated object. The entropy in such a | |
97 | * pointer is concentrated in its middle bits. | |
98 | */ | |
99 | static u32 nfsd_file_inode_hash(const struct inode *inode, u32 seed) | |
100 | { | |
101 | unsigned long ptr = (unsigned long)inode; | |
102 | u32 k; | |
103 | ||
104 | k = ptr >> L1_CACHE_SHIFT; | |
105 | k &= 0x00ffffff; | |
106 | return jhash2(&k, 1, seed); | |
107 | } | |
108 | ||
109 | /** | |
110 | * nfsd_file_key_hashfn - Compute the hash value of a lookup key | |
111 | * @data: key on which to compute the hash value | |
112 | * @len: rhash table's key_len parameter (unused) | |
113 | * @seed: rhash table's random seed of the day | |
114 | * | |
115 | * Return value: | |
116 | * Computed 32-bit hash value | |
117 | */ | |
118 | static u32 nfsd_file_key_hashfn(const void *data, u32 len, u32 seed) | |
119 | { | |
120 | const struct nfsd_file_lookup_key *key = data; | |
121 | ||
122 | return nfsd_file_inode_hash(key->inode, seed); | |
123 | } | |
124 | ||
125 | /** | |
126 | * nfsd_file_obj_hashfn - Compute the hash value of an nfsd_file | |
127 | * @data: object on which to compute the hash value | |
128 | * @len: rhash table's key_len parameter (unused) | |
129 | * @seed: rhash table's random seed of the day | |
130 | * | |
131 | * Return value: | |
132 | * Computed 32-bit hash value | |
133 | */ | |
134 | static u32 nfsd_file_obj_hashfn(const void *data, u32 len, u32 seed) | |
135 | { | |
136 | const struct nfsd_file *nf = data; | |
137 | ||
138 | return nfsd_file_inode_hash(nf->nf_inode, seed); | |
139 | } | |
65294c1f | 140 | |
fc22945e CL |
141 | static bool |
142 | nfsd_match_cred(const struct cred *c1, const struct cred *c2) | |
143 | { | |
144 | int i; | |
145 | ||
146 | if (!uid_eq(c1->fsuid, c2->fsuid)) | |
147 | return false; | |
148 | if (!gid_eq(c1->fsgid, c2->fsgid)) | |
149 | return false; | |
150 | if (c1->group_info == NULL || c2->group_info == NULL) | |
151 | return c1->group_info == c2->group_info; | |
152 | if (c1->group_info->ngroups != c2->group_info->ngroups) | |
153 | return false; | |
154 | for (i = 0; i < c1->group_info->ngroups; i++) { | |
155 | if (!gid_eq(c1->group_info->gid[i], c2->group_info->gid[i])) | |
156 | return false; | |
157 | } | |
158 | return true; | |
159 | } | |
160 | ||
161 | /** | |
162 | * nfsd_file_obj_cmpfn - Match a cache item against search criteria | |
163 | * @arg: search criteria | |
164 | * @ptr: cache item to check | |
165 | * | |
166 | * Return values: | |
167 | * %0 - Item matches search criteria | |
168 | * %1 - Item does not match search criteria | |
169 | */ | |
170 | static int nfsd_file_obj_cmpfn(struct rhashtable_compare_arg *arg, | |
171 | const void *ptr) | |
172 | { | |
173 | const struct nfsd_file_lookup_key *key = arg->key; | |
174 | const struct nfsd_file *nf = ptr; | |
175 | ||
176 | switch (key->type) { | |
177 | case NFSD_FILE_KEY_INODE: | |
6c31e4c9 JL |
178 | if (test_bit(NFSD_FILE_GC, &nf->nf_flags) != key->gc) |
179 | return 1; | |
fc22945e CL |
180 | if (nf->nf_inode != key->inode) |
181 | return 1; | |
182 | break; | |
183 | case NFSD_FILE_KEY_FULL: | |
184 | if (nf->nf_inode != key->inode) | |
185 | return 1; | |
186 | if (nf->nf_may != key->need) | |
187 | return 1; | |
188 | if (nf->nf_net != key->net) | |
189 | return 1; | |
190 | if (!nfsd_match_cred(nf->nf_cred, key->cred)) | |
191 | return 1; | |
d69b8dbf | 192 | if (test_bit(NFSD_FILE_GC, &nf->nf_flags) != key->gc) |
4d1ea845 | 193 | return 1; |
fc22945e CL |
194 | if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags) == 0) |
195 | return 1; | |
196 | break; | |
197 | } | |
198 | return 0; | |
199 | } | |
200 | ||
201 | static const struct rhashtable_params nfsd_file_rhash_params = { | |
202 | .key_len = sizeof_field(struct nfsd_file, nf_inode), | |
203 | .key_offset = offsetof(struct nfsd_file, nf_inode), | |
204 | .head_offset = offsetof(struct nfsd_file, nf_rhash), | |
205 | .hashfn = nfsd_file_key_hashfn, | |
206 | .obj_hashfn = nfsd_file_obj_hashfn, | |
207 | .obj_cmpfn = nfsd_file_obj_cmpfn, | |
208 | /* Reduce resizing churn on light workloads */ | |
209 | .min_size = 512, /* buckets */ | |
210 | .automatic_shrinking = true, | |
211 | }; | |
65294c1f JL |
212 | |
213 | static void | |
9542e6a6 | 214 | nfsd_file_schedule_laundrette(void) |
65294c1f | 215 | { |
22ae4c11 JL |
216 | if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags)) |
217 | queue_delayed_work(system_wq, &nfsd_filecache_laundrette, | |
218 | NFSD_LAUNDRETTE_DELAY); | |
65294c1f JL |
219 | } |
220 | ||
221 | static void | |
222 | nfsd_file_slab_free(struct rcu_head *rcu) | |
223 | { | |
224 | struct nfsd_file *nf = container_of(rcu, struct nfsd_file, nf_rcu); | |
225 | ||
226 | put_cred(nf->nf_cred); | |
227 | kmem_cache_free(nfsd_file_slab, nf); | |
228 | } | |
229 | ||
230 | static void | |
231 | nfsd_file_mark_free(struct fsnotify_mark *mark) | |
232 | { | |
233 | struct nfsd_file_mark *nfm = container_of(mark, struct nfsd_file_mark, | |
234 | nfm_mark); | |
235 | ||
236 | kmem_cache_free(nfsd_file_mark_slab, nfm); | |
237 | } | |
238 | ||
239 | static struct nfsd_file_mark * | |
240 | nfsd_file_mark_get(struct nfsd_file_mark *nfm) | |
241 | { | |
689827cd | 242 | if (!refcount_inc_not_zero(&nfm->nfm_ref)) |
65294c1f JL |
243 | return NULL; |
244 | return nfm; | |
245 | } | |
246 | ||
247 | static void | |
248 | nfsd_file_mark_put(struct nfsd_file_mark *nfm) | |
249 | { | |
689827cd | 250 | if (refcount_dec_and_test(&nfm->nfm_ref)) { |
65294c1f JL |
251 | fsnotify_destroy_mark(&nfm->nfm_mark, nfsd_file_fsnotify_group); |
252 | fsnotify_put_mark(&nfm->nfm_mark); | |
253 | } | |
254 | } | |
255 | ||
256 | static struct nfsd_file_mark * | |
427f5f83 | 257 | nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode) |
65294c1f JL |
258 | { |
259 | int err; | |
260 | struct fsnotify_mark *mark; | |
261 | struct nfsd_file_mark *nfm = NULL, *new; | |
65294c1f JL |
262 | |
263 | do { | |
b8962a9d | 264 | fsnotify_group_lock(nfsd_file_fsnotify_group); |
65294c1f | 265 | mark = fsnotify_find_mark(&inode->i_fsnotify_marks, |
b8962a9d | 266 | nfsd_file_fsnotify_group); |
65294c1f JL |
267 | if (mark) { |
268 | nfm = nfsd_file_mark_get(container_of(mark, | |
269 | struct nfsd_file_mark, | |
270 | nfm_mark)); | |
b8962a9d | 271 | fsnotify_group_unlock(nfsd_file_fsnotify_group); |
90d2f1da TM |
272 | if (nfm) { |
273 | fsnotify_put_mark(mark); | |
65294c1f | 274 | break; |
90d2f1da TM |
275 | } |
276 | /* Avoid soft lockup race with nfsd_file_mark_put() */ | |
277 | fsnotify_destroy_mark(mark, nfsd_file_fsnotify_group); | |
278 | fsnotify_put_mark(mark); | |
b8962a9d AG |
279 | } else { |
280 | fsnotify_group_unlock(nfsd_file_fsnotify_group); | |
281 | } | |
65294c1f JL |
282 | |
283 | /* allocate a new nfm */ | |
284 | new = kmem_cache_alloc(nfsd_file_mark_slab, GFP_KERNEL); | |
285 | if (!new) | |
286 | return NULL; | |
287 | fsnotify_init_mark(&new->nfm_mark, nfsd_file_fsnotify_group); | |
288 | new->nfm_mark.mask = FS_ATTRIB|FS_DELETE_SELF; | |
689827cd | 289 | refcount_set(&new->nfm_ref, 1); |
65294c1f JL |
290 | |
291 | err = fsnotify_add_inode_mark(&new->nfm_mark, inode, 0); | |
292 | ||
293 | /* | |
294 | * If the add was successful, then return the object. | |
295 | * Otherwise, we need to put the reference we hold on the | |
296 | * nfm_mark. The fsnotify code will take a reference and put | |
297 | * it on failure, so we can't just free it directly. It's also | |
298 | * not safe to call fsnotify_destroy_mark on it as the | |
299 | * mark->group will be NULL. Thus, we can't let the nfm_ref | |
300 | * counter drive the destruction at this point. | |
301 | */ | |
302 | if (likely(!err)) | |
303 | nfm = new; | |
304 | else | |
305 | fsnotify_put_mark(&new->nfm_mark); | |
306 | } while (unlikely(err == -EEXIST)); | |
307 | ||
308 | return nfm; | |
309 | } | |
310 | ||
311 | static struct nfsd_file * | |
ce502f81 | 312 | nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may) |
65294c1f JL |
313 | { |
314 | struct nfsd_file *nf; | |
315 | ||
316 | nf = kmem_cache_alloc(nfsd_file_slab, GFP_KERNEL); | |
317 | if (nf) { | |
65294c1f | 318 | INIT_LIST_HEAD(&nf->nf_lru); |
904940e9 | 319 | nf->nf_birthtime = ktime_get(); |
65294c1f JL |
320 | nf->nf_file = NULL; |
321 | nf->nf_cred = get_current_cred(); | |
ce502f81 | 322 | nf->nf_net = key->net; |
65294c1f | 323 | nf->nf_flags = 0; |
ce502f81 CL |
324 | __set_bit(NFSD_FILE_HASHED, &nf->nf_flags); |
325 | __set_bit(NFSD_FILE_PENDING, &nf->nf_flags); | |
4d1ea845 CL |
326 | if (key->gc) |
327 | __set_bit(NFSD_FILE_GC, &nf->nf_flags); | |
ce502f81 | 328 | nf->nf_inode = key->inode; |
ac3a2585 | 329 | refcount_set(&nf->nf_ref, 1); |
ce502f81 | 330 | nf->nf_may = key->need; |
65294c1f | 331 | nf->nf_mark = NULL; |
65294c1f JL |
332 | } |
333 | return nf; | |
334 | } | |
335 | ||
4c475eee JL |
336 | /** |
337 | * nfsd_file_check_write_error - check for writeback errors on a file | |
338 | * @nf: nfsd_file to check for writeback errors | |
339 | * | |
340 | * Check whether a nfsd_file has an unseen error. Reset the write | |
341 | * verifier if so. | |
342 | */ | |
82141185 | 343 | static void |
82141185 JL |
344 | nfsd_file_check_write_error(struct nfsd_file *nf) |
345 | { | |
346 | struct file *file = nf->nf_file; | |
347 | ||
4c475eee JL |
348 | if ((file->f_mode & FMODE_WRITE) && |
349 | filemap_check_wb_err(file->f_mapping, READ_ONCE(file->f_wb_err))) | |
350 | nfsd_reset_write_verifier(net_generic(nf->nf_net, nfsd_net_id)); | |
82141185 JL |
351 | } |
352 | ||
353 | static void | |
354 | nfsd_file_hash_remove(struct nfsd_file *nf) | |
355 | { | |
356 | trace_nfsd_file_unhash(nf); | |
82141185 JL |
357 | rhashtable_remove_fast(&nfsd_file_rhash_tbl, &nf->nf_rhash, |
358 | nfsd_file_rhash_params); | |
359 | } | |
360 | ||
361 | static bool | |
362 | nfsd_file_unhash(struct nfsd_file *nf) | |
363 | { | |
364 | if (test_and_clear_bit(NFSD_FILE_HASHED, &nf->nf_flags)) { | |
365 | nfsd_file_hash_remove(nf); | |
366 | return true; | |
367 | } | |
368 | return false; | |
369 | } | |
370 | ||
ac3a2585 | 371 | static void |
65294c1f JL |
372 | nfsd_file_free(struct nfsd_file *nf) |
373 | { | |
904940e9 | 374 | s64 age = ktime_to_ms(ktime_sub(ktime_get(), nf->nf_birthtime)); |
65294c1f | 375 | |
82141185 JL |
376 | trace_nfsd_file_free(nf); |
377 | ||
d6329327 | 378 | this_cpu_inc(nfsd_file_releases); |
904940e9 | 379 | this_cpu_add(nfsd_file_total_age, age); |
d6329327 | 380 | |
ac3a2585 | 381 | nfsd_file_unhash(nf); |
65294c1f JL |
382 | if (nf->nf_mark) |
383 | nfsd_file_mark_put(nf->nf_mark); | |
384 | if (nf->nf_file) { | |
4c475eee | 385 | nfsd_file_check_write_error(nf); |
b2ff1bd7 | 386 | filp_close(nf->nf_file, NULL); |
65294c1f | 387 | } |
668ed92e CL |
388 | |
389 | /* | |
390 | * If this item is still linked via nf_lru, that's a bug. | |
391 | * WARN and leak it to preserve system stability. | |
392 | */ | |
393 | if (WARN_ON_ONCE(!list_empty(&nf->nf_lru))) | |
ac3a2585 | 394 | return; |
668ed92e | 395 | |
65294c1f | 396 | call_rcu(&nf->nf_rcu, nfsd_file_slab_free); |
65294c1f JL |
397 | } |
398 | ||
055b24a8 TM |
399 | static bool |
400 | nfsd_file_check_writeback(struct nfsd_file *nf) | |
401 | { | |
402 | struct file *file = nf->nf_file; | |
403 | struct address_space *mapping; | |
404 | ||
405 | if (!file || !(file->f_mode & FMODE_WRITE)) | |
406 | return false; | |
407 | mapping = file->f_mapping; | |
408 | return mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) || | |
409 | mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK); | |
410 | } | |
411 | ||
ac3a2585 | 412 | static bool nfsd_file_lru_add(struct nfsd_file *nf) |
65294c1f | 413 | { |
4a0e73e6 | 414 | set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags); |
ac3a2585 | 415 | if (list_lru_add(&nfsd_file_lru, &nf->nf_lru)) { |
c46203ac | 416 | trace_nfsd_file_lru_add(nf); |
ac3a2585 JL |
417 | return true; |
418 | } | |
419 | return false; | |
c46203ac | 420 | } |
65294c1f | 421 | |
ac3a2585 | 422 | static bool nfsd_file_lru_remove(struct nfsd_file *nf) |
c46203ac | 423 | { |
ac3a2585 | 424 | if (list_lru_del(&nfsd_file_lru, &nf->nf_lru)) { |
c46203ac | 425 | trace_nfsd_file_lru_del(nf); |
ac3a2585 JL |
426 | return true; |
427 | } | |
428 | return false; | |
c46203ac CL |
429 | } |
430 | ||
82141185 JL |
431 | struct nfsd_file * |
432 | nfsd_file_get(struct nfsd_file *nf) | |
65294c1f | 433 | { |
70f62231 | 434 | if (nf && refcount_inc_not_zero(&nf->nf_ref)) |
82141185 JL |
435 | return nf; |
436 | return NULL; | |
65294c1f JL |
437 | } |
438 | ||
ac3a2585 JL |
439 | /** |
440 | * nfsd_file_put - put the reference to a nfsd_file | |
441 | * @nf: nfsd_file of which to put the reference | |
442 | * | |
443 | * Put a reference to a nfsd_file. In the non-GC case, we just put the | |
444 | * reference immediately. In the GC case, if the reference would be | |
445 | * the last one, the put it on the LRU instead to be cleaned up later. | |
446 | */ | |
65294c1f JL |
447 | void |
448 | nfsd_file_put(struct nfsd_file *nf) | |
449 | { | |
08af54b3 | 450 | might_sleep(); |
ac3a2585 | 451 | trace_nfsd_file_put(nf); |
08af54b3 | 452 | |
ac3a2585 JL |
453 | if (test_bit(NFSD_FILE_GC, &nf->nf_flags) && |
454 | test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) { | |
455 | /* | |
456 | * If this is the last reference (nf_ref == 1), then try to | |
457 | * transfer it to the LRU. | |
458 | */ | |
459 | if (refcount_dec_not_one(&nf->nf_ref)) | |
460 | return; | |
461 | ||
462 | /* Try to add it to the LRU. If that fails, decrement. */ | |
463 | if (nfsd_file_lru_add(nf)) { | |
464 | /* If it's still hashed, we're done */ | |
465 | if (test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) { | |
466 | nfsd_file_schedule_laundrette(); | |
467 | return; | |
468 | } | |
65294c1f | 469 | |
ac3a2585 JL |
470 | /* |
471 | * We're racing with unhashing, so try to remove it from | |
472 | * the LRU. If removal fails, then someone else already | |
473 | * has our reference. | |
474 | */ | |
475 | if (!nfsd_file_lru_remove(nf)) | |
476 | return; | |
477 | } | |
65294c1f | 478 | } |
ac3a2585 JL |
479 | if (refcount_dec_and_test(&nf->nf_ref)) |
480 | nfsd_file_free(nf); | |
65294c1f JL |
481 | } |
482 | ||
483 | static void | |
ac3a2585 | 484 | nfsd_file_dispose_list(struct list_head *dispose) |
65294c1f | 485 | { |
65294c1f JL |
486 | struct nfsd_file *nf; |
487 | ||
ac3a2585 | 488 | while (!list_empty(dispose)) { |
65294c1f | 489 | nf = list_first_entry(dispose, struct nfsd_file, nf_lru); |
668ed92e | 490 | list_del_init(&nf->nf_lru); |
ac3a2585 | 491 | nfsd_file_free(nf); |
65294c1f | 492 | } |
65294c1f JL |
493 | } |
494 | ||
9542e6a6 TM |
495 | static void |
496 | nfsd_file_list_remove_disposal(struct list_head *dst, | |
497 | struct nfsd_fcache_disposal *l) | |
498 | { | |
499 | spin_lock(&l->lock); | |
500 | list_splice_init(&l->freeme, dst); | |
501 | spin_unlock(&l->lock); | |
502 | } | |
503 | ||
504 | static void | |
505 | nfsd_file_list_add_disposal(struct list_head *files, struct net *net) | |
506 | { | |
1463b38e N |
507 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
508 | struct nfsd_fcache_disposal *l = nn->fcache_disposal; | |
9542e6a6 | 509 | |
1463b38e N |
510 | spin_lock(&l->lock); |
511 | list_splice_tail_init(files, &l->freeme); | |
512 | spin_unlock(&l->lock); | |
513 | queue_work(nfsd_filecache_wq, &l->work); | |
9542e6a6 TM |
514 | } |
515 | ||
516 | static void | |
517 | nfsd_file_list_add_pernet(struct list_head *dst, struct list_head *src, | |
518 | struct net *net) | |
519 | { | |
520 | struct nfsd_file *nf, *tmp; | |
521 | ||
522 | list_for_each_entry_safe(nf, tmp, src, nf_lru) { | |
523 | if (nf->nf_net == net) | |
524 | list_move_tail(&nf->nf_lru, dst); | |
525 | } | |
526 | } | |
527 | ||
528 | static void | |
529 | nfsd_file_dispose_list_delayed(struct list_head *dispose) | |
530 | { | |
531 | LIST_HEAD(list); | |
532 | struct nfsd_file *nf; | |
533 | ||
534 | while(!list_empty(dispose)) { | |
535 | nf = list_first_entry(dispose, struct nfsd_file, nf_lru); | |
536 | nfsd_file_list_add_pernet(&list, dispose, nf->nf_net); | |
537 | nfsd_file_list_add_disposal(&list, nf->nf_net); | |
538 | } | |
539 | } | |
540 | ||
4a0e73e6 CL |
541 | /** |
542 | * nfsd_file_lru_cb - Examine an entry on the LRU list | |
543 | * @item: LRU entry to examine | |
544 | * @lru: controlling LRU | |
545 | * @lock: LRU list lock (unused) | |
546 | * @arg: dispose list | |
547 | * | |
4a0e73e6 CL |
548 | * Return values: |
549 | * %LRU_REMOVED: @item was removed from the LRU | |
edead3a5 | 550 | * %LRU_ROTATE: @item is to be moved to the LRU tail |
4a0e73e6 | 551 | * %LRU_SKIP: @item cannot be evicted |
65294c1f JL |
552 | */ |
553 | static enum lru_status | |
554 | nfsd_file_lru_cb(struct list_head *item, struct list_lru_one *lru, | |
555 | spinlock_t *lock, void *arg) | |
556 | __releases(lock) | |
557 | __acquires(lock) | |
558 | { | |
559 | struct list_head *head = arg; | |
560 | struct nfsd_file *nf = list_entry(item, struct nfsd_file, nf_lru); | |
561 | ||
ac3a2585 JL |
562 | /* We should only be dealing with GC entries here */ |
563 | WARN_ON_ONCE(!test_bit(NFSD_FILE_GC, &nf->nf_flags)); | |
055b24a8 TM |
564 | |
565 | /* | |
566 | * Don't throw out files that are still undergoing I/O or | |
567 | * that have uncleared errors pending. | |
568 | */ | |
c46203ac CL |
569 | if (nfsd_file_check_writeback(nf)) { |
570 | trace_nfsd_file_gc_writeback(nf); | |
571 | return LRU_SKIP; | |
572 | } | |
055b24a8 | 573 | |
ac3a2585 | 574 | /* If it was recently added to the list, skip it */ |
c46203ac CL |
575 | if (test_and_clear_bit(NFSD_FILE_REFERENCED, &nf->nf_flags)) { |
576 | trace_nfsd_file_gc_referenced(nf); | |
edead3a5 | 577 | return LRU_ROTATE; |
c46203ac | 578 | } |
65294c1f | 579 | |
ac3a2585 JL |
580 | /* |
581 | * Put the reference held on behalf of the LRU. If it wasn't the last | |
582 | * one, then just remove it from the LRU and ignore it. | |
583 | */ | |
584 | if (!refcount_dec_and_test(&nf->nf_ref)) { | |
585 | trace_nfsd_file_gc_in_use(nf); | |
586 | list_lru_isolate(lru, &nf->nf_lru); | |
587 | return LRU_REMOVED; | |
c46203ac | 588 | } |
65294c1f | 589 | |
ac3a2585 JL |
590 | /* Refcount went to zero. Unhash it and queue it to the dispose list */ |
591 | nfsd_file_unhash(nf); | |
65294c1f | 592 | list_lru_isolate_move(lru, &nf->nf_lru, head); |
94660cc1 | 593 | this_cpu_inc(nfsd_file_evictions); |
c46203ac | 594 | trace_nfsd_file_gc_disposed(nf); |
65294c1f | 595 | return LRU_REMOVED; |
65294c1f JL |
596 | } |
597 | ||
9542e6a6 TM |
598 | static void |
599 | nfsd_file_gc(void) | |
600 | { | |
3bc6d347 | 601 | LIST_HEAD(dispose); |
94660cc1 | 602 | unsigned long ret; |
3bc6d347 | 603 | |
94660cc1 | 604 | ret = list_lru_walk(&nfsd_file_lru, nfsd_file_lru_cb, |
edead3a5 | 605 | &dispose, list_lru_count(&nfsd_file_lru)); |
94660cc1 | 606 | trace_nfsd_file_gc_removed(ret, list_lru_count(&nfsd_file_lru)); |
ac3a2585 | 607 | nfsd_file_dispose_list_delayed(&dispose); |
9542e6a6 TM |
608 | } |
609 | ||
610 | static void | |
611 | nfsd_file_gc_worker(struct work_struct *work) | |
612 | { | |
613 | nfsd_file_gc(); | |
22ae4c11 JL |
614 | if (list_lru_count(&nfsd_file_lru)) |
615 | nfsd_file_schedule_laundrette(); | |
65294c1f JL |
616 | } |
617 | ||
618 | static unsigned long | |
619 | nfsd_file_lru_count(struct shrinker *s, struct shrink_control *sc) | |
620 | { | |
621 | return list_lru_count(&nfsd_file_lru); | |
622 | } | |
623 | ||
624 | static unsigned long | |
625 | nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc) | |
626 | { | |
39f1d1ff CL |
627 | LIST_HEAD(dispose); |
628 | unsigned long ret; | |
629 | ||
630 | ret = list_lru_shrink_walk(&nfsd_file_lru, sc, | |
631 | nfsd_file_lru_cb, &dispose); | |
94660cc1 | 632 | trace_nfsd_file_shrinker_removed(ret, list_lru_count(&nfsd_file_lru)); |
ac3a2585 | 633 | nfsd_file_dispose_list_delayed(&dispose); |
39f1d1ff | 634 | return ret; |
65294c1f JL |
635 | } |
636 | ||
637 | static struct shrinker nfsd_file_shrinker = { | |
638 | .scan_objects = nfsd_file_lru_scan, | |
639 | .count_objects = nfsd_file_lru_count, | |
640 | .seeks = 1, | |
641 | }; | |
642 | ||
4bdbba54 JL |
643 | /** |
644 | * nfsd_file_cond_queue - conditionally unhash and queue a nfsd_file | |
645 | * @nf: nfsd_file to attempt to queue | |
646 | * @dispose: private list to queue successfully-put objects | |
647 | * | |
648 | * Unhash an nfsd_file, try to get a reference to it, and then put that | |
649 | * reference. If it's the last reference, queue it to the dispose list. | |
650 | */ | |
651 | static void | |
652 | nfsd_file_cond_queue(struct nfsd_file *nf, struct list_head *dispose) | |
653 | __must_hold(RCU) | |
654 | { | |
655 | int decrement = 1; | |
656 | ||
657 | /* If we raced with someone else unhashing, ignore it */ | |
658 | if (!nfsd_file_unhash(nf)) | |
659 | return; | |
660 | ||
661 | /* If we can't get a reference, ignore it */ | |
662 | if (!nfsd_file_get(nf)) | |
663 | return; | |
664 | ||
665 | /* Extra decrement if we remove from the LRU */ | |
666 | if (nfsd_file_lru_remove(nf)) | |
667 | ++decrement; | |
668 | ||
669 | /* If refcount goes to 0, then put on the dispose list */ | |
670 | if (refcount_sub_and_test(decrement, &nf->nf_ref)) { | |
671 | list_add(&nf->nf_lru, dispose); | |
672 | trace_nfsd_file_closing(nf); | |
673 | } | |
674 | } | |
675 | ||
ac3a2585 JL |
676 | /** |
677 | * nfsd_file_queue_for_close: try to close out any open nfsd_files for an inode | |
678 | * @inode: inode on which to close out nfsd_files | |
679 | * @dispose: list on which to gather nfsd_files to close out | |
680 | * | |
681 | * An nfsd_file represents a struct file being held open on behalf of nfsd. An | |
682 | * open file however can block other activity (such as leases), or cause | |
683 | * undesirable behavior (e.g. spurious silly-renames when reexporting NFS). | |
684 | * | |
685 | * This function is intended to find open nfsd_files when this sort of | |
686 | * conflicting access occurs and then attempt to close those files out. | |
687 | * | |
688 | * Populates the dispose list with entries that have already had their | |
689 | * refcounts go to zero. The actual free of an nfsd_file can be expensive, | |
690 | * so we leave it up to the caller whether it wants to wait or not. | |
a8455110 | 691 | */ |
ac3a2585 JL |
692 | static void |
693 | nfsd_file_queue_for_close(struct inode *inode, struct list_head *dispose) | |
65294c1f | 694 | { |
ce502f81 CL |
695 | struct nfsd_file_lookup_key key = { |
696 | .type = NFSD_FILE_KEY_INODE, | |
697 | .inode = inode, | |
6c31e4c9 | 698 | .gc = true, |
ce502f81 | 699 | }; |
ce502f81 | 700 | struct nfsd_file *nf; |
65294c1f | 701 | |
ce502f81 CL |
702 | rcu_read_lock(); |
703 | do { | |
704 | nf = rhashtable_lookup(&nfsd_file_rhash_tbl, &key, | |
705 | nfsd_file_rhash_params); | |
706 | if (!nf) | |
707 | break; | |
4bdbba54 | 708 | nfsd_file_cond_queue(nf, dispose); |
ce502f81 CL |
709 | } while (1); |
710 | rcu_read_unlock(); | |
65294c1f JL |
711 | } |
712 | ||
713 | /** | |
ac3a2585 | 714 | * nfsd_file_close_inode - attempt a delayed close of a nfsd_file |
65294c1f JL |
715 | * @inode: inode of the file to attempt to remove |
716 | * | |
ac3a2585 JL |
717 | * Close out any open nfsd_files that can be reaped for @inode. The |
718 | * actual freeing is deferred to the dispose_list_delayed infrastructure. | |
719 | * | |
720 | * This is used by the fsnotify callbacks and setlease notifier. | |
65294c1f | 721 | */ |
ac3a2585 JL |
722 | static void |
723 | nfsd_file_close_inode(struct inode *inode) | |
65294c1f | 724 | { |
65294c1f JL |
725 | LIST_HEAD(dispose); |
726 | ||
ac3a2585 JL |
727 | nfsd_file_queue_for_close(inode, &dispose); |
728 | nfsd_file_dispose_list_delayed(&dispose); | |
65294c1f JL |
729 | } |
730 | ||
731 | /** | |
ac3a2585 | 732 | * nfsd_file_close_inode_sync - attempt to forcibly close a nfsd_file |
65294c1f JL |
733 | * @inode: inode of the file to attempt to remove |
734 | * | |
ac3a2585 JL |
735 | * Close out any open nfsd_files that can be reaped for @inode. The |
736 | * nfsd_files are closed out synchronously. | |
737 | * | |
738 | * This is called from nfsd_rename and nfsd_unlink to avoid silly-renames | |
739 | * when reexporting NFS. | |
65294c1f | 740 | */ |
ac3a2585 JL |
741 | void |
742 | nfsd_file_close_inode_sync(struct inode *inode) | |
65294c1f | 743 | { |
ac3a2585 | 744 | struct nfsd_file *nf; |
65294c1f JL |
745 | LIST_HEAD(dispose); |
746 | ||
ac3a2585 JL |
747 | trace_nfsd_file_close(inode); |
748 | ||
749 | nfsd_file_queue_for_close(inode, &dispose); | |
750 | while (!list_empty(&dispose)) { | |
751 | nf = list_first_entry(&dispose, struct nfsd_file, nf_lru); | |
752 | list_del_init(&nf->nf_lru); | |
753 | nfsd_file_free(nf); | |
754 | } | |
755 | flush_delayed_fput(); | |
65294c1f JL |
756 | } |
757 | ||
758 | /** | |
759 | * nfsd_file_delayed_close - close unused nfsd_files | |
760 | * @work: dummy | |
761 | * | |
ac3a2585 | 762 | * Walk the LRU list and destroy any entries that have not been used since |
65294c1f | 763 | * the last scan. |
65294c1f JL |
764 | */ |
765 | static void | |
766 | nfsd_file_delayed_close(struct work_struct *work) | |
767 | { | |
768 | LIST_HEAD(head); | |
9542e6a6 TM |
769 | struct nfsd_fcache_disposal *l = container_of(work, |
770 | struct nfsd_fcache_disposal, work); | |
65294c1f | 771 | |
9542e6a6 TM |
772 | nfsd_file_list_remove_disposal(&head, l); |
773 | nfsd_file_dispose_list(&head); | |
65294c1f JL |
774 | } |
775 | ||
776 | static int | |
777 | nfsd_file_lease_notifier_call(struct notifier_block *nb, unsigned long arg, | |
778 | void *data) | |
779 | { | |
780 | struct file_lock *fl = data; | |
781 | ||
782 | /* Only close files for F_SETLEASE leases */ | |
783 | if (fl->fl_flags & FL_LEASE) | |
ac3a2585 | 784 | nfsd_file_close_inode(file_inode(fl->fl_file)); |
65294c1f JL |
785 | return 0; |
786 | } | |
787 | ||
788 | static struct notifier_block nfsd_file_lease_notifier = { | |
789 | .notifier_call = nfsd_file_lease_notifier_call, | |
790 | }; | |
791 | ||
792 | static int | |
b9a1b977 AG |
793 | nfsd_file_fsnotify_handle_event(struct fsnotify_mark *mark, u32 mask, |
794 | struct inode *inode, struct inode *dir, | |
950cc0d2 | 795 | const struct qstr *name, u32 cookie) |
65294c1f | 796 | { |
24dca905 GKB |
797 | if (WARN_ON_ONCE(!inode)) |
798 | return 0; | |
799 | ||
65294c1f JL |
800 | trace_nfsd_file_fsnotify_handle_event(inode, mask); |
801 | ||
802 | /* Should be no marks on non-regular files */ | |
803 | if (!S_ISREG(inode->i_mode)) { | |
804 | WARN_ON_ONCE(1); | |
805 | return 0; | |
806 | } | |
807 | ||
808 | /* don't close files if this was not the last link */ | |
809 | if (mask & FS_ATTRIB) { | |
810 | if (inode->i_nlink) | |
811 | return 0; | |
812 | } | |
813 | ||
814 | nfsd_file_close_inode(inode); | |
815 | return 0; | |
816 | } | |
817 | ||
818 | ||
819 | static const struct fsnotify_ops nfsd_file_fsnotify_ops = { | |
b9a1b977 | 820 | .handle_inode_event = nfsd_file_fsnotify_handle_event, |
65294c1f JL |
821 | .free_mark = nfsd_file_mark_free, |
822 | }; | |
823 | ||
824 | int | |
825 | nfsd_file_cache_init(void) | |
826 | { | |
0ec8e9d1 | 827 | int ret; |
65294c1f | 828 | |
c7b824c3 CL |
829 | lockdep_assert_held(&nfsd_mutex); |
830 | if (test_and_set_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1) | |
65294c1f JL |
831 | return 0; |
832 | ||
fc22945e CL |
833 | ret = rhashtable_init(&nfsd_file_rhash_tbl, &nfsd_file_rhash_params); |
834 | if (ret) | |
835 | return ret; | |
836 | ||
837 | ret = -ENOMEM; | |
9542e6a6 TM |
838 | nfsd_filecache_wq = alloc_workqueue("nfsd_filecache", 0, 0); |
839 | if (!nfsd_filecache_wq) | |
840 | goto out; | |
841 | ||
65294c1f JL |
842 | nfsd_file_slab = kmem_cache_create("nfsd_file", |
843 | sizeof(struct nfsd_file), 0, 0, NULL); | |
844 | if (!nfsd_file_slab) { | |
845 | pr_err("nfsd: unable to create nfsd_file_slab\n"); | |
846 | goto out_err; | |
847 | } | |
848 | ||
849 | nfsd_file_mark_slab = kmem_cache_create("nfsd_file_mark", | |
850 | sizeof(struct nfsd_file_mark), 0, 0, NULL); | |
851 | if (!nfsd_file_mark_slab) { | |
852 | pr_err("nfsd: unable to create nfsd_file_mark_slab\n"); | |
853 | goto out_err; | |
854 | } | |
855 | ||
856 | ||
857 | ret = list_lru_init(&nfsd_file_lru); | |
858 | if (ret) { | |
859 | pr_err("nfsd: failed to init nfsd_file_lru: %d\n", ret); | |
860 | goto out_err; | |
861 | } | |
862 | ||
e33c267a | 863 | ret = register_shrinker(&nfsd_file_shrinker, "nfsd-filecache"); |
65294c1f JL |
864 | if (ret) { |
865 | pr_err("nfsd: failed to register nfsd_file_shrinker: %d\n", ret); | |
866 | goto out_lru; | |
867 | } | |
868 | ||
869 | ret = lease_register_notifier(&nfsd_file_lease_notifier); | |
870 | if (ret) { | |
871 | pr_err("nfsd: unable to register lease notifier: %d\n", ret); | |
872 | goto out_shrinker; | |
873 | } | |
874 | ||
867a448d | 875 | nfsd_file_fsnotify_group = fsnotify_alloc_group(&nfsd_file_fsnotify_ops, |
b8962a9d | 876 | FSNOTIFY_GROUP_NOFS); |
65294c1f JL |
877 | if (IS_ERR(nfsd_file_fsnotify_group)) { |
878 | pr_err("nfsd: unable to create fsnotify group: %ld\n", | |
879 | PTR_ERR(nfsd_file_fsnotify_group)); | |
231307df | 880 | ret = PTR_ERR(nfsd_file_fsnotify_group); |
65294c1f JL |
881 | nfsd_file_fsnotify_group = NULL; |
882 | goto out_notifier; | |
883 | } | |
884 | ||
9542e6a6 | 885 | INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker); |
65294c1f JL |
886 | out: |
887 | return ret; | |
888 | out_notifier: | |
889 | lease_unregister_notifier(&nfsd_file_lease_notifier); | |
890 | out_shrinker: | |
891 | unregister_shrinker(&nfsd_file_shrinker); | |
892 | out_lru: | |
893 | list_lru_destroy(&nfsd_file_lru); | |
894 | out_err: | |
895 | kmem_cache_destroy(nfsd_file_slab); | |
896 | nfsd_file_slab = NULL; | |
897 | kmem_cache_destroy(nfsd_file_mark_slab); | |
898 | nfsd_file_mark_slab = NULL; | |
9542e6a6 TM |
899 | destroy_workqueue(nfsd_filecache_wq); |
900 | nfsd_filecache_wq = NULL; | |
fc22945e | 901 | rhashtable_destroy(&nfsd_file_rhash_tbl); |
65294c1f JL |
902 | goto out; |
903 | } | |
904 | ||
ac3a2585 JL |
905 | /** |
906 | * __nfsd_file_cache_purge: clean out the cache for shutdown | |
907 | * @net: net-namespace to shut down the cache (may be NULL) | |
908 | * | |
909 | * Walk the nfsd_file cache and close out any that match @net. If @net is NULL, | |
972cc0e0 JL |
910 | * then close out everything. Called when an nfsd instance is being shut down, |
911 | * and when the exports table is flushed. | |
ac3a2585 | 912 | */ |
c7b824c3 CL |
913 | static void |
914 | __nfsd_file_cache_purge(struct net *net) | |
65294c1f | 915 | { |
ce502f81 CL |
916 | struct rhashtable_iter iter; |
917 | struct nfsd_file *nf; | |
65294c1f | 918 | LIST_HEAD(dispose); |
65294c1f | 919 | |
ce502f81 CL |
920 | rhashtable_walk_enter(&nfsd_file_rhash_tbl, &iter); |
921 | do { | |
922 | rhashtable_walk_start(&iter); | |
5e113224 | 923 | |
ce502f81 CL |
924 | nf = rhashtable_walk_next(&iter); |
925 | while (!IS_ERR_OR_NULL(nf)) { | |
4bdbba54 JL |
926 | if (!net || nf->nf_net == net) |
927 | nfsd_file_cond_queue(nf, &dispose); | |
ce502f81 | 928 | nf = rhashtable_walk_next(&iter); |
65294c1f | 929 | } |
ce502f81 CL |
930 | |
931 | rhashtable_walk_stop(&iter); | |
932 | } while (nf == ERR_PTR(-EAGAIN)); | |
933 | rhashtable_walk_exit(&iter); | |
934 | ||
935 | nfsd_file_dispose_list(&dispose); | |
65294c1f JL |
936 | } |
937 | ||
9542e6a6 | 938 | static struct nfsd_fcache_disposal * |
1463b38e | 939 | nfsd_alloc_fcache_disposal(void) |
9542e6a6 TM |
940 | { |
941 | struct nfsd_fcache_disposal *l; | |
942 | ||
943 | l = kmalloc(sizeof(*l), GFP_KERNEL); | |
944 | if (!l) | |
945 | return NULL; | |
946 | INIT_WORK(&l->work, nfsd_file_delayed_close); | |
9542e6a6 TM |
947 | spin_lock_init(&l->lock); |
948 | INIT_LIST_HEAD(&l->freeme); | |
949 | return l; | |
950 | } | |
951 | ||
952 | static void | |
953 | nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l) | |
954 | { | |
9542e6a6 TM |
955 | cancel_work_sync(&l->work); |
956 | nfsd_file_dispose_list(&l->freeme); | |
1463b38e | 957 | kfree(l); |
9542e6a6 TM |
958 | } |
959 | ||
960 | static void | |
961 | nfsd_free_fcache_disposal_net(struct net *net) | |
962 | { | |
1463b38e N |
963 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
964 | struct nfsd_fcache_disposal *l = nn->fcache_disposal; | |
9542e6a6 | 965 | |
1463b38e | 966 | nfsd_free_fcache_disposal(l); |
9542e6a6 TM |
967 | } |
968 | ||
969 | int | |
970 | nfsd_file_cache_start_net(struct net *net) | |
971 | { | |
1463b38e N |
972 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
973 | ||
974 | nn->fcache_disposal = nfsd_alloc_fcache_disposal(); | |
975 | return nn->fcache_disposal ? 0 : -ENOMEM; | |
9542e6a6 TM |
976 | } |
977 | ||
c7b824c3 CL |
978 | /** |
979 | * nfsd_file_cache_purge - Remove all cache items associated with @net | |
980 | * @net: target net namespace | |
981 | * | |
982 | */ | |
983 | void | |
984 | nfsd_file_cache_purge(struct net *net) | |
985 | { | |
986 | lockdep_assert_held(&nfsd_mutex); | |
987 | if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1) | |
988 | __nfsd_file_cache_purge(net); | |
989 | } | |
990 | ||
9542e6a6 TM |
991 | void |
992 | nfsd_file_cache_shutdown_net(struct net *net) | |
993 | { | |
994 | nfsd_file_cache_purge(net); | |
995 | nfsd_free_fcache_disposal_net(net); | |
996 | } | |
997 | ||
65294c1f JL |
998 | void |
999 | nfsd_file_cache_shutdown(void) | |
1000 | { | |
8b330f78 CL |
1001 | int i; |
1002 | ||
c7b824c3 CL |
1003 | lockdep_assert_held(&nfsd_mutex); |
1004 | if (test_and_clear_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 0) | |
1005 | return; | |
65294c1f JL |
1006 | |
1007 | lease_unregister_notifier(&nfsd_file_lease_notifier); | |
1008 | unregister_shrinker(&nfsd_file_shrinker); | |
1009 | /* | |
1010 | * make sure all callers of nfsd_file_lru_cb are done before | |
1011 | * calling nfsd_file_cache_purge | |
1012 | */ | |
1013 | cancel_delayed_work_sync(&nfsd_filecache_laundrette); | |
c7b824c3 | 1014 | __nfsd_file_cache_purge(NULL); |
65294c1f JL |
1015 | list_lru_destroy(&nfsd_file_lru); |
1016 | rcu_barrier(); | |
1017 | fsnotify_put_group(nfsd_file_fsnotify_group); | |
1018 | nfsd_file_fsnotify_group = NULL; | |
1019 | kmem_cache_destroy(nfsd_file_slab); | |
1020 | nfsd_file_slab = NULL; | |
1021 | fsnotify_wait_marks_destroyed(); | |
1022 | kmem_cache_destroy(nfsd_file_mark_slab); | |
1023 | nfsd_file_mark_slab = NULL; | |
9542e6a6 TM |
1024 | destroy_workqueue(nfsd_filecache_wq); |
1025 | nfsd_filecache_wq = NULL; | |
fc22945e | 1026 | rhashtable_destroy(&nfsd_file_rhash_tbl); |
8b330f78 CL |
1027 | |
1028 | for_each_possible_cpu(i) { | |
1029 | per_cpu(nfsd_file_cache_hits, i) = 0; | |
1030 | per_cpu(nfsd_file_acquisitions, i) = 0; | |
1031 | per_cpu(nfsd_file_releases, i) = 0; | |
1032 | per_cpu(nfsd_file_total_age, i) = 0; | |
8b330f78 | 1033 | per_cpu(nfsd_file_evictions, i) = 0; |
65294c1f | 1034 | } |
65294c1f JL |
1035 | } |
1036 | ||
1037 | /** | |
ce502f81 CL |
1038 | * nfsd_file_is_cached - are there any cached open files for this inode? |
1039 | * @inode: inode to check | |
1040 | * | |
1041 | * The lookup matches inodes in all net namespaces and is atomic wrt | |
1042 | * nfsd_file_acquire(). | |
65294c1f | 1043 | * |
ce502f81 CL |
1044 | * Return values: |
1045 | * %true: filecache contains at least one file matching this inode | |
1046 | * %false: filecache contains no files matching this inode | |
65294c1f JL |
1047 | */ |
1048 | bool | |
1049 | nfsd_file_is_cached(struct inode *inode) | |
1050 | { | |
ce502f81 CL |
1051 | struct nfsd_file_lookup_key key = { |
1052 | .type = NFSD_FILE_KEY_INODE, | |
1053 | .inode = inode, | |
6c31e4c9 | 1054 | .gc = true, |
ce502f81 CL |
1055 | }; |
1056 | bool ret = false; | |
1057 | ||
1058 | if (rhashtable_lookup_fast(&nfsd_file_rhash_tbl, &key, | |
1059 | nfsd_file_rhash_params) != NULL) | |
1060 | ret = true; | |
54f7df70 | 1061 | trace_nfsd_file_is_cached(inode, (int)ret); |
65294c1f JL |
1062 | return ret; |
1063 | } | |
1064 | ||
fb70bf12 | 1065 | static __be32 |
be023006 | 1066 | nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp, |
0b3a551f JL |
1067 | unsigned int may_flags, struct file *file, |
1068 | struct nfsd_file **pnf, bool want_gc) | |
65294c1f | 1069 | { |
ce502f81 CL |
1070 | struct nfsd_file_lookup_key key = { |
1071 | .type = NFSD_FILE_KEY_FULL, | |
1072 | .need = may_flags & NFSD_FILE_MAY_MASK, | |
1073 | .net = SVC_NET(rqstp), | |
4d1ea845 | 1074 | .gc = want_gc, |
ce502f81 | 1075 | }; |
243a5263 JL |
1076 | bool open_retry = true; |
1077 | struct nfsd_file *nf; | |
ce502f81 | 1078 | __be32 status; |
243a5263 | 1079 | int ret; |
65294c1f | 1080 | |
65294c1f JL |
1081 | status = fh_verify(rqstp, fhp, S_IFREG, |
1082 | may_flags|NFSD_MAY_OWNER_OVERRIDE); | |
1083 | if (status != nfs_ok) | |
1084 | return status; | |
ce502f81 CL |
1085 | key.inode = d_inode(fhp->fh_dentry); |
1086 | key.cred = get_current_cred(); | |
65294c1f | 1087 | |
65294c1f | 1088 | retry: |
243a5263 JL |
1089 | rcu_read_lock(); |
1090 | nf = rhashtable_lookup(&nfsd_file_rhash_tbl, &key, | |
1091 | nfsd_file_rhash_params); | |
70f62231 | 1092 | nf = nfsd_file_get(nf); |
243a5263 | 1093 | rcu_read_unlock(); |
ac3a2585 JL |
1094 | |
1095 | if (nf) { | |
b680cb9b JL |
1096 | /* |
1097 | * If the nf is on the LRU then it holds an extra reference | |
1098 | * that must be put if it's removed. It had better not be | |
1099 | * the last one however, since we should hold another. | |
1100 | */ | |
ac3a2585 JL |
1101 | if (nfsd_file_lru_remove(nf)) |
1102 | WARN_ON_ONCE(refcount_dec_and_test(&nf->nf_ref)); | |
65294c1f | 1103 | goto wait_for_construction; |
ac3a2585 | 1104 | } |
65294c1f | 1105 | |
243a5263 JL |
1106 | nf = nfsd_file_alloc(&key, may_flags); |
1107 | if (!nf) { | |
54f7df70 | 1108 | status = nfserr_jukebox; |
c6593366 | 1109 | goto out; |
65294c1f JL |
1110 | } |
1111 | ||
243a5263 JL |
1112 | ret = rhashtable_lookup_insert_key(&nfsd_file_rhash_tbl, |
1113 | &key, &nf->nf_rhash, | |
1114 | nfsd_file_rhash_params); | |
1115 | if (likely(ret == 0)) | |
65294c1f | 1116 | goto open_file; |
243a5263 | 1117 | |
243a5263 JL |
1118 | if (ret == -EEXIST) |
1119 | goto retry; | |
1120 | trace_nfsd_file_insert_err(rqstp, key.inode, may_flags, ret); | |
1121 | status = nfserr_jukebox; | |
c6593366 | 1122 | goto construction_err; |
65294c1f JL |
1123 | |
1124 | wait_for_construction: | |
1125 | wait_on_bit(&nf->nf_flags, NFSD_FILE_PENDING, TASK_UNINTERRUPTIBLE); | |
1126 | ||
1127 | /* Did construction of this file fail? */ | |
1128 | if (!test_bit(NFSD_FILE_HASHED, &nf->nf_flags)) { | |
ce502f81 | 1129 | trace_nfsd_file_cons_err(rqstp, key.inode, may_flags, nf); |
243a5263 | 1130 | if (!open_retry) { |
28c7d86b | 1131 | status = nfserr_jukebox; |
c6593366 | 1132 | goto construction_err; |
28c7d86b | 1133 | } |
243a5263 | 1134 | open_retry = false; |
65294c1f JL |
1135 | goto retry; |
1136 | } | |
65294c1f JL |
1137 | this_cpu_inc(nfsd_file_cache_hits); |
1138 | ||
23ba98de | 1139 | status = nfserrno(nfsd_open_break_lease(file_inode(nf->nf_file), may_flags)); |
c6593366 JL |
1140 | if (status != nfs_ok) { |
1141 | nfsd_file_put(nf); | |
1142 | nf = NULL; | |
1143 | } | |
1144 | ||
65294c1f JL |
1145 | out: |
1146 | if (status == nfs_ok) { | |
0b3a551f | 1147 | this_cpu_inc(nfsd_file_acquisitions); |
4c475eee | 1148 | nfsd_file_check_write_error(nf); |
65294c1f | 1149 | *pnf = nf; |
65294c1f | 1150 | } |
ce502f81 | 1151 | put_cred(key.cred); |
0b3a551f | 1152 | trace_nfsd_file_acquire(rqstp, key.inode, may_flags, nf, status); |
65294c1f | 1153 | return status; |
54f7df70 | 1154 | |
65294c1f | 1155 | open_file: |
b40a2839 | 1156 | trace_nfsd_file_alloc(nf); |
427f5f83 | 1157 | nf->nf_mark = nfsd_file_mark_find_or_create(nf, key.inode); |
fb70bf12 | 1158 | if (nf->nf_mark) { |
0b3a551f JL |
1159 | if (file) { |
1160 | get_file(file); | |
1161 | nf->nf_file = file; | |
1162 | status = nfs_ok; | |
1163 | trace_nfsd_file_opened(nf, status); | |
1164 | } else { | |
fb70bf12 CL |
1165 | status = nfsd_open_verified(rqstp, fhp, may_flags, |
1166 | &nf->nf_file); | |
0122e882 | 1167 | trace_nfsd_file_open(nf, status); |
0b3a551f | 1168 | } |
fb70bf12 | 1169 | } else |
65294c1f JL |
1170 | status = nfserr_jukebox; |
1171 | /* | |
1172 | * If construction failed, or we raced with a call to unlink() | |
1173 | * then unhash. | |
1174 | */ | |
ac3a2585 JL |
1175 | if (status == nfs_ok && key.inode->i_nlink == 0) |
1176 | status = nfserr_jukebox; | |
1177 | if (status != nfs_ok) | |
1178 | nfsd_file_unhash(nf); | |
b8bea9f6 | 1179 | clear_and_wake_up_bit(NFSD_FILE_PENDING, &nf->nf_flags); |
c6593366 JL |
1180 | if (status == nfs_ok) |
1181 | goto out; | |
1182 | ||
1183 | construction_err: | |
1184 | if (refcount_dec_and_test(&nf->nf_ref)) | |
1185 | nfsd_file_free(nf); | |
1186 | nf = NULL; | |
65294c1f JL |
1187 | goto out; |
1188 | } | |
1189 | ||
4d1ea845 CL |
1190 | /** |
1191 | * nfsd_file_acquire_gc - Get a struct nfsd_file with an open file | |
1192 | * @rqstp: the RPC transaction being executed | |
1193 | * @fhp: the NFS filehandle of the file to be opened | |
1194 | * @may_flags: NFSD_MAY_ settings for the file | |
1195 | * @pnf: OUT: new or found "struct nfsd_file" object | |
1196 | * | |
1197 | * The nfsd_file object returned by this API is reference-counted | |
1198 | * and garbage-collected. The object is retained for a few | |
1199 | * seconds after the final nfsd_file_put() in case the caller | |
1200 | * wants to re-use it. | |
1201 | * | |
1202 | * Returns nfs_ok and sets @pnf on success; otherwise an nfsstat in | |
1203 | * network byte order is returned. | |
1204 | */ | |
1205 | __be32 | |
1206 | nfsd_file_acquire_gc(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1207 | unsigned int may_flags, struct nfsd_file **pnf) | |
1208 | { | |
0b3a551f | 1209 | return nfsd_file_do_acquire(rqstp, fhp, may_flags, NULL, pnf, true); |
4d1ea845 CL |
1210 | } |
1211 | ||
fb70bf12 CL |
1212 | /** |
1213 | * nfsd_file_acquire - Get a struct nfsd_file with an open file | |
1214 | * @rqstp: the RPC transaction being executed | |
1215 | * @fhp: the NFS filehandle of the file to be opened | |
1216 | * @may_flags: NFSD_MAY_ settings for the file | |
1217 | * @pnf: OUT: new or found "struct nfsd_file" object | |
1218 | * | |
4d1ea845 CL |
1219 | * The nfsd_file_object returned by this API is reference-counted |
1220 | * but not garbage-collected. The object is unhashed after the | |
1221 | * final nfsd_file_put(). | |
1222 | * | |
fb70bf12 CL |
1223 | * Returns nfs_ok and sets @pnf on success; otherwise an nfsstat in |
1224 | * network byte order is returned. | |
1225 | */ | |
1226 | __be32 | |
1227 | nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp, | |
1228 | unsigned int may_flags, struct nfsd_file **pnf) | |
1229 | { | |
0b3a551f | 1230 | return nfsd_file_do_acquire(rqstp, fhp, may_flags, NULL, pnf, false); |
fb70bf12 CL |
1231 | } |
1232 | ||
1233 | /** | |
0b3a551f | 1234 | * nfsd_file_acquire_opened - Get a struct nfsd_file using existing open file |
fb70bf12 CL |
1235 | * @rqstp: the RPC transaction being executed |
1236 | * @fhp: the NFS filehandle of the file just created | |
1237 | * @may_flags: NFSD_MAY_ settings for the file | |
0b3a551f | 1238 | * @file: cached, already-open file (may be NULL) |
fb70bf12 CL |
1239 | * @pnf: OUT: new or found "struct nfsd_file" object |
1240 | * | |
0b3a551f JL |
1241 | * Acquire a nfsd_file object that is not GC'ed. If one doesn't already exist, |
1242 | * and @file is non-NULL, use it to instantiate a new nfsd_file instead of | |
1243 | * opening a new one. | |
4d1ea845 | 1244 | * |
fb70bf12 CL |
1245 | * Returns nfs_ok and sets @pnf on success; otherwise an nfsstat in |
1246 | * network byte order is returned. | |
1247 | */ | |
1248 | __be32 | |
0b3a551f JL |
1249 | nfsd_file_acquire_opened(struct svc_rqst *rqstp, struct svc_fh *fhp, |
1250 | unsigned int may_flags, struct file *file, | |
1251 | struct nfsd_file **pnf) | |
fb70bf12 | 1252 | { |
0b3a551f | 1253 | return nfsd_file_do_acquire(rqstp, fhp, may_flags, file, pnf, false); |
fb70bf12 CL |
1254 | } |
1255 | ||
65294c1f JL |
1256 | /* |
1257 | * Note that fields may be added, removed or reordered in the future. Programs | |
1258 | * scraping this file for info should test the labels to ensure they're | |
1259 | * getting the correct field. | |
1260 | */ | |
1342f9dd | 1261 | int nfsd_file_cache_stats_show(struct seq_file *m, void *v) |
65294c1f | 1262 | { |
1f696e23 | 1263 | unsigned long releases = 0, evictions = 0; |
df2aff52 | 1264 | unsigned long hits = 0, acquisitions = 0; |
ce502f81 | 1265 | unsigned int i, count = 0, buckets = 0; |
904940e9 | 1266 | unsigned long lru = 0, total_age = 0; |
65294c1f | 1267 | |
ce502f81 | 1268 | /* Serialize with server shutdown */ |
65294c1f | 1269 | mutex_lock(&nfsd_mutex); |
c7b824c3 | 1270 | if (test_bit(NFSD_FILE_CACHE_UP, &nfsd_file_flags) == 1) { |
ce502f81 CL |
1271 | struct bucket_table *tbl; |
1272 | struct rhashtable *ht; | |
1273 | ||
0fd244c1 | 1274 | lru = list_lru_count(&nfsd_file_lru); |
ce502f81 CL |
1275 | |
1276 | rcu_read_lock(); | |
1277 | ht = &nfsd_file_rhash_tbl; | |
1278 | count = atomic_read(&ht->nelems); | |
1279 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
1280 | buckets = tbl->size; | |
1281 | rcu_read_unlock(); | |
65294c1f JL |
1282 | } |
1283 | mutex_unlock(&nfsd_mutex); | |
1284 | ||
29d4bdbb | 1285 | for_each_possible_cpu(i) { |
65294c1f | 1286 | hits += per_cpu(nfsd_file_cache_hits, i); |
29d4bdbb | 1287 | acquisitions += per_cpu(nfsd_file_acquisitions, i); |
d6329327 | 1288 | releases += per_cpu(nfsd_file_releases, i); |
904940e9 | 1289 | total_age += per_cpu(nfsd_file_total_age, i); |
94660cc1 | 1290 | evictions += per_cpu(nfsd_file_evictions, i); |
29d4bdbb | 1291 | } |
65294c1f JL |
1292 | |
1293 | seq_printf(m, "total entries: %u\n", count); | |
ce502f81 | 1294 | seq_printf(m, "hash buckets: %u\n", buckets); |
0fd244c1 | 1295 | seq_printf(m, "lru entries: %lu\n", lru); |
65294c1f | 1296 | seq_printf(m, "cache hits: %lu\n", hits); |
29d4bdbb | 1297 | seq_printf(m, "acquisitions: %lu\n", acquisitions); |
d6329327 | 1298 | seq_printf(m, "releases: %lu\n", releases); |
94660cc1 | 1299 | seq_printf(m, "evictions: %lu\n", evictions); |
904940e9 CL |
1300 | if (releases) |
1301 | seq_printf(m, "mean age (ms): %ld\n", total_age / releases); | |
1302 | else | |
1303 | seq_printf(m, "mean age (ms): -\n"); | |
65294c1f JL |
1304 | return 0; |
1305 | } |