Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * General Public License version 2 for more details (a copy is included | |
14 | * in the LICENSE file that accompanied this code). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf | |
19 | * | |
20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
21 | * CA 95054 USA or visit www.sun.com if you need additional information or | |
22 | * have any questions. | |
23 | * | |
24 | * GPL HEADER END | |
25 | */ | |
26 | /* | |
27 | * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. | |
28 | * Use is subject to license terms. | |
29 | * | |
1dc563a6 | 30 | * Copyright (c) 2011, 2015, Intel Corporation. |
d7e09d03 PT |
31 | */ |
32 | /* | |
33 | * This file is part of Lustre, http://www.lustre.org/ | |
34 | * Lustre is a trademark of Sun Microsystems, Inc. | |
35 | * | |
36 | * Client Lustre Object. | |
37 | * | |
38 | * Author: Nikita Danilov <nikita.danilov@sun.com> | |
39 | */ | |
40 | ||
41 | /* | |
42 | * Locking. | |
43 | * | |
44 | * i_mutex | |
45 | * PG_locked | |
46 | * ->coh_page_guard | |
47 | * ->coh_lock_guard | |
48 | * ->coh_attr_guard | |
49 | * ->ls_guard | |
50 | */ | |
51 | ||
52 | #define DEBUG_SUBSYSTEM S_CLASS | |
53 | ||
9fdaf8c0 | 54 | #include "../../include/linux/libcfs/libcfs.h" |
d7e09d03 | 55 | /* class_put_type() */ |
610f7377 GKH |
56 | #include "../include/obd_class.h" |
57 | #include "../include/obd_support.h" | |
58 | #include "../include/lustre_fid.h" | |
d7e09d03 | 59 | #include <linux/list.h> |
9fdaf8c0 | 60 | #include "../../include/linux/libcfs/libcfs_hash.h" /* for cfs_hash stuff */ |
610f7377 | 61 | #include "../include/cl_object.h" |
d7e09d03 PT |
62 | #include "cl_internal.h" |
63 | ||
64 | static struct kmem_cache *cl_env_kmem; | |
65 | ||
66 | /** Lock class of cl_object_header::coh_page_guard */ | |
67 | static struct lock_class_key cl_page_guard_class; | |
68 | /** Lock class of cl_object_header::coh_lock_guard */ | |
69 | static struct lock_class_key cl_lock_guard_class; | |
70 | /** Lock class of cl_object_header::coh_attr_guard */ | |
71 | static struct lock_class_key cl_attr_guard_class; | |
72 | ||
73 | extern __u32 lu_context_tags_default; | |
74 | extern __u32 lu_session_tags_default; | |
75 | /** | |
76 | * Initialize cl_object_header. | |
77 | */ | |
78 | int cl_object_header_init(struct cl_object_header *h) | |
79 | { | |
80 | int result; | |
81 | ||
d7e09d03 PT |
82 | result = lu_object_header_init(&h->coh_lu); |
83 | if (result == 0) { | |
84 | spin_lock_init(&h->coh_page_guard); | |
85 | spin_lock_init(&h->coh_lock_guard); | |
86 | spin_lock_init(&h->coh_attr_guard); | |
87 | lockdep_set_class(&h->coh_page_guard, &cl_page_guard_class); | |
88 | lockdep_set_class(&h->coh_lock_guard, &cl_lock_guard_class); | |
89 | lockdep_set_class(&h->coh_attr_guard, &cl_attr_guard_class); | |
90 | h->coh_pages = 0; | |
91 | /* XXX hard coded GFP_* mask. */ | |
92 | INIT_RADIX_TREE(&h->coh_tree, GFP_ATOMIC); | |
93 | INIT_LIST_HEAD(&h->coh_locks); | |
94 | h->coh_page_bufsize = ALIGN(sizeof(struct cl_page), 8); | |
95 | } | |
0a3bdb00 | 96 | return result; |
d7e09d03 PT |
97 | } |
98 | EXPORT_SYMBOL(cl_object_header_init); | |
99 | ||
d7e09d03 PT |
100 | /** |
101 | * Returns a cl_object with a given \a fid. | |
102 | * | |
103 | * Returns either cached or newly created object. Additional reference on the | |
104 | * returned object is acquired. | |
105 | * | |
106 | * \see lu_object_find(), cl_page_find(), cl_lock_find() | |
107 | */ | |
108 | struct cl_object *cl_object_find(const struct lu_env *env, | |
109 | struct cl_device *cd, const struct lu_fid *fid, | |
110 | const struct cl_object_conf *c) | |
111 | { | |
112 | might_sleep(); | |
113 | return lu2cl(lu_object_find_slice(env, cl2lu_dev(cd), fid, &c->coc_lu)); | |
114 | } | |
115 | EXPORT_SYMBOL(cl_object_find); | |
116 | ||
117 | /** | |
118 | * Releases a reference on \a o. | |
119 | * | |
120 | * When last reference is released object is returned to the cache, unless | |
121 | * lu_object_header_flags::LU_OBJECT_HEARD_BANSHEE bit is set in its header. | |
122 | * | |
123 | * \see cl_page_put(), cl_lock_put(). | |
124 | */ | |
125 | void cl_object_put(const struct lu_env *env, struct cl_object *o) | |
126 | { | |
127 | lu_object_put(env, &o->co_lu); | |
128 | } | |
129 | EXPORT_SYMBOL(cl_object_put); | |
130 | ||
131 | /** | |
132 | * Acquire an additional reference to the object \a o. | |
133 | * | |
134 | * This can only be used to acquire _additional_ reference, i.e., caller | |
135 | * already has to possess at least one reference to \a o before calling this. | |
136 | * | |
137 | * \see cl_page_get(), cl_lock_get(). | |
138 | */ | |
139 | void cl_object_get(struct cl_object *o) | |
140 | { | |
141 | lu_object_get(&o->co_lu); | |
142 | } | |
143 | EXPORT_SYMBOL(cl_object_get); | |
144 | ||
145 | /** | |
146 | * Returns the top-object for a given \a o. | |
147 | * | |
148 | * \see cl_page_top(), cl_io_top() | |
149 | */ | |
150 | struct cl_object *cl_object_top(struct cl_object *o) | |
151 | { | |
152 | struct cl_object_header *hdr = cl_object_header(o); | |
153 | struct cl_object *top; | |
154 | ||
155 | while (hdr->coh_parent != NULL) | |
156 | hdr = hdr->coh_parent; | |
157 | ||
158 | top = lu2cl(lu_object_top(&hdr->coh_lu)); | |
159 | CDEBUG(D_TRACE, "%p -> %p\n", o, top); | |
160 | return top; | |
161 | } | |
162 | EXPORT_SYMBOL(cl_object_top); | |
163 | ||
164 | /** | |
165 | * Returns pointer to the lock protecting data-attributes for the given object | |
166 | * \a o. | |
167 | * | |
168 | * Data-attributes are protected by the cl_object_header::coh_attr_guard | |
169 | * spin-lock in the top-object. | |
170 | * | |
171 | * \see cl_attr, cl_object_attr_lock(), cl_object_operations::coo_attr_get(). | |
172 | */ | |
173 | static spinlock_t *cl_object_attr_guard(struct cl_object *o) | |
174 | { | |
175 | return &cl_object_header(cl_object_top(o))->coh_attr_guard; | |
176 | } | |
177 | ||
178 | /** | |
179 | * Locks data-attributes. | |
180 | * | |
181 | * Prevents data-attributes from changing, until lock is released by | |
182 | * cl_object_attr_unlock(). This has to be called before calls to | |
183 | * cl_object_attr_get(), cl_object_attr_set(). | |
184 | */ | |
185 | void cl_object_attr_lock(struct cl_object *o) | |
f8bd34d2 | 186 | __acquires(cl_object_attr_guard(o)) |
d7e09d03 PT |
187 | { |
188 | spin_lock(cl_object_attr_guard(o)); | |
189 | } | |
190 | EXPORT_SYMBOL(cl_object_attr_lock); | |
191 | ||
192 | /** | |
193 | * Releases data-attributes lock, acquired by cl_object_attr_lock(). | |
194 | */ | |
195 | void cl_object_attr_unlock(struct cl_object *o) | |
f8bd34d2 | 196 | __releases(cl_object_attr_guard(o)) |
d7e09d03 PT |
197 | { |
198 | spin_unlock(cl_object_attr_guard(o)); | |
199 | } | |
200 | EXPORT_SYMBOL(cl_object_attr_unlock); | |
201 | ||
202 | /** | |
203 | * Returns data-attributes of an object \a obj. | |
204 | * | |
205 | * Every layer is asked (by calling cl_object_operations::coo_attr_get()) | |
206 | * top-to-bottom to fill in parts of \a attr that this layer is responsible | |
207 | * for. | |
208 | */ | |
209 | int cl_object_attr_get(const struct lu_env *env, struct cl_object *obj, | |
210 | struct cl_attr *attr) | |
211 | { | |
212 | struct lu_object_header *top; | |
213 | int result; | |
214 | ||
5e42bc9d | 215 | assert_spin_locked(cl_object_attr_guard(obj)); |
d7e09d03 PT |
216 | |
217 | top = obj->co_lu.lo_header; | |
218 | result = 0; | |
219 | list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) { | |
220 | if (obj->co_ops->coo_attr_get != NULL) { | |
221 | result = obj->co_ops->coo_attr_get(env, obj, attr); | |
222 | if (result != 0) { | |
223 | if (result > 0) | |
224 | result = 0; | |
225 | break; | |
226 | } | |
227 | } | |
228 | } | |
0a3bdb00 | 229 | return result; |
d7e09d03 PT |
230 | } |
231 | EXPORT_SYMBOL(cl_object_attr_get); | |
232 | ||
233 | /** | |
234 | * Updates data-attributes of an object \a obj. | |
235 | * | |
236 | * Only attributes, mentioned in a validness bit-mask \a v are | |
237 | * updated. Calls cl_object_operations::coo_attr_set() on every layer, bottom | |
238 | * to top. | |
239 | */ | |
240 | int cl_object_attr_set(const struct lu_env *env, struct cl_object *obj, | |
241 | const struct cl_attr *attr, unsigned v) | |
242 | { | |
243 | struct lu_object_header *top; | |
244 | int result; | |
245 | ||
5e42bc9d | 246 | assert_spin_locked(cl_object_attr_guard(obj)); |
d7e09d03 PT |
247 | |
248 | top = obj->co_lu.lo_header; | |
249 | result = 0; | |
250 | list_for_each_entry_reverse(obj, &top->loh_layers, | |
251 | co_lu.lo_linkage) { | |
252 | if (obj->co_ops->coo_attr_set != NULL) { | |
253 | result = obj->co_ops->coo_attr_set(env, obj, attr, v); | |
254 | if (result != 0) { | |
255 | if (result > 0) | |
256 | result = 0; | |
257 | break; | |
258 | } | |
259 | } | |
260 | } | |
0a3bdb00 | 261 | return result; |
d7e09d03 PT |
262 | } |
263 | EXPORT_SYMBOL(cl_object_attr_set); | |
264 | ||
265 | /** | |
266 | * Notifies layers (bottom-to-top) that glimpse AST was received. | |
267 | * | |
268 | * Layers have to fill \a lvb fields with information that will be shipped | |
269 | * back to glimpse issuer. | |
270 | * | |
271 | * \see cl_lock_operations::clo_glimpse() | |
272 | */ | |
273 | int cl_object_glimpse(const struct lu_env *env, struct cl_object *obj, | |
274 | struct ost_lvb *lvb) | |
275 | { | |
276 | struct lu_object_header *top; | |
277 | int result; | |
278 | ||
d7e09d03 PT |
279 | top = obj->co_lu.lo_header; |
280 | result = 0; | |
281 | list_for_each_entry_reverse(obj, &top->loh_layers, | |
282 | co_lu.lo_linkage) { | |
283 | if (obj->co_ops->coo_glimpse != NULL) { | |
284 | result = obj->co_ops->coo_glimpse(env, obj, lvb); | |
285 | if (result != 0) | |
286 | break; | |
287 | } | |
288 | } | |
289 | LU_OBJECT_HEADER(D_DLMTRACE, env, lu_object_top(top), | |
b0f5aad5 | 290 | "size: %llu mtime: %llu atime: %llu ctime: %llu blocks: %llu\n", |
d7e09d03 PT |
291 | lvb->lvb_size, lvb->lvb_mtime, lvb->lvb_atime, |
292 | lvb->lvb_ctime, lvb->lvb_blocks); | |
0a3bdb00 | 293 | return result; |
d7e09d03 PT |
294 | } |
295 | EXPORT_SYMBOL(cl_object_glimpse); | |
296 | ||
297 | /** | |
298 | * Updates a configuration of an object \a obj. | |
299 | */ | |
300 | int cl_conf_set(const struct lu_env *env, struct cl_object *obj, | |
301 | const struct cl_object_conf *conf) | |
302 | { | |
303 | struct lu_object_header *top; | |
304 | int result; | |
305 | ||
d7e09d03 PT |
306 | top = obj->co_lu.lo_header; |
307 | result = 0; | |
308 | list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) { | |
309 | if (obj->co_ops->coo_conf_set != NULL) { | |
310 | result = obj->co_ops->coo_conf_set(env, obj, conf); | |
311 | if (result != 0) | |
312 | break; | |
313 | } | |
314 | } | |
0a3bdb00 | 315 | return result; |
d7e09d03 PT |
316 | } |
317 | EXPORT_SYMBOL(cl_conf_set); | |
318 | ||
319 | /** | |
320 | * Helper function removing all object locks, and marking object for | |
321 | * deletion. All object pages must have been deleted at this point. | |
322 | * | |
323 | * This is called by cl_inode_fini() and lov_object_delete() to destroy top- | |
324 | * and sub- objects respectively. | |
325 | */ | |
326 | void cl_object_kill(const struct lu_env *env, struct cl_object *obj) | |
327 | { | |
328 | struct cl_object_header *hdr; | |
329 | ||
330 | hdr = cl_object_header(obj); | |
331 | LASSERT(hdr->coh_tree.rnode == NULL); | |
332 | LASSERT(hdr->coh_pages == 0); | |
333 | ||
334 | set_bit(LU_OBJECT_HEARD_BANSHEE, &hdr->coh_lu.loh_flags); | |
335 | /* | |
336 | * Destroy all locks. Object destruction (including cl_inode_fini()) | |
337 | * cannot cancel the locks, because in the case of a local client, | |
338 | * where client and server share the same thread running | |
339 | * prune_icache(), this can dead-lock with ldlm_cancel_handler() | |
340 | * waiting on __wait_on_freeing_inode(). | |
341 | */ | |
342 | cl_locks_prune(env, obj, 0); | |
343 | } | |
344 | EXPORT_SYMBOL(cl_object_kill); | |
345 | ||
346 | /** | |
347 | * Prunes caches of pages and locks for this object. | |
348 | */ | |
349 | void cl_object_prune(const struct lu_env *env, struct cl_object *obj) | |
350 | { | |
d7e09d03 PT |
351 | cl_pages_prune(env, obj); |
352 | cl_locks_prune(env, obj, 1); | |
d7e09d03 PT |
353 | } |
354 | EXPORT_SYMBOL(cl_object_prune); | |
355 | ||
d7e09d03 PT |
356 | void cache_stats_init(struct cache_stats *cs, const char *name) |
357 | { | |
358 | int i; | |
359 | ||
360 | cs->cs_name = name; | |
361 | for (i = 0; i < CS_NR; i++) | |
362 | atomic_set(&cs->cs_stats[i], 0); | |
363 | } | |
364 | ||
63986883 BG |
365 | static int cache_stats_print(const struct cache_stats *cs, |
366 | struct seq_file *m, int h) | |
d7e09d03 | 367 | { |
d7e09d03 PT |
368 | int i; |
369 | /* | |
370 | * lookup hit total cached create | |
371 | * env: ...... ...... ...... ...... ...... | |
372 | */ | |
373 | if (h) { | |
374 | const char *names[CS_NR] = CS_NAMES; | |
375 | ||
73bb1da6 | 376 | seq_printf(m, "%6s", " "); |
d7e09d03 | 377 | for (i = 0; i < CS_NR; i++) |
73bb1da6 PT |
378 | seq_printf(m, "%8s", names[i]); |
379 | seq_printf(m, "\n"); | |
d7e09d03 PT |
380 | } |
381 | ||
73bb1da6 | 382 | seq_printf(m, "%5.5s:", cs->cs_name); |
d7e09d03 | 383 | for (i = 0; i < CS_NR; i++) |
73bb1da6 PT |
384 | seq_printf(m, "%8u", atomic_read(&cs->cs_stats[i])); |
385 | return 0; | |
d7e09d03 PT |
386 | } |
387 | ||
388 | /** | |
389 | * Initialize client site. | |
390 | * | |
391 | * Perform common initialization (lu_site_init()), and initialize statistical | |
392 | * counters. Also perform global initializations on the first call. | |
393 | */ | |
394 | int cl_site_init(struct cl_site *s, struct cl_device *d) | |
395 | { | |
396 | int i; | |
397 | int result; | |
398 | ||
399 | result = lu_site_init(&s->cs_lu, &d->cd_lu_dev); | |
400 | if (result == 0) { | |
401 | cache_stats_init(&s->cs_pages, "pages"); | |
402 | cache_stats_init(&s->cs_locks, "locks"); | |
403 | for (i = 0; i < ARRAY_SIZE(s->cs_pages_state); ++i) | |
404 | atomic_set(&s->cs_pages_state[0], 0); | |
405 | for (i = 0; i < ARRAY_SIZE(s->cs_locks_state); ++i) | |
406 | atomic_set(&s->cs_locks_state[i], 0); | |
407 | } | |
408 | return result; | |
409 | } | |
410 | EXPORT_SYMBOL(cl_site_init); | |
411 | ||
412 | /** | |
413 | * Finalize client site. Dual to cl_site_init(). | |
414 | */ | |
415 | void cl_site_fini(struct cl_site *s) | |
416 | { | |
417 | lu_site_fini(&s->cs_lu); | |
418 | } | |
419 | EXPORT_SYMBOL(cl_site_fini); | |
420 | ||
421 | static struct cache_stats cl_env_stats = { | |
422 | .cs_name = "envs", | |
423 | .cs_stats = { ATOMIC_INIT(0), } | |
424 | }; | |
425 | ||
426 | /** | |
427 | * Outputs client site statistical counters into a buffer. Suitable for | |
428 | * ll_rd_*()-style functions. | |
429 | */ | |
73bb1da6 | 430 | int cl_site_stats_print(const struct cl_site *site, struct seq_file *m) |
d7e09d03 | 431 | { |
d7e09d03 PT |
432 | int i; |
433 | static const char *pstate[] = { | |
434 | [CPS_CACHED] = "c", | |
435 | [CPS_OWNED] = "o", | |
436 | [CPS_PAGEOUT] = "w", | |
437 | [CPS_PAGEIN] = "r", | |
438 | [CPS_FREEING] = "f" | |
439 | }; | |
440 | static const char *lstate[] = { | |
441 | [CLS_NEW] = "n", | |
442 | [CLS_QUEUING] = "q", | |
443 | [CLS_ENQUEUED] = "e", | |
444 | [CLS_HELD] = "h", | |
445 | [CLS_INTRANSIT] = "t", | |
446 | [CLS_CACHED] = "c", | |
447 | [CLS_FREEING] = "f" | |
448 | }; | |
449 | /* | |
450 | lookup hit total busy create | |
451 | pages: ...... ...... ...... ...... ...... [...... ...... ...... ......] | |
452 | locks: ...... ...... ...... ...... ...... [...... ...... ...... ...... ......] | |
453 | env: ...... ...... ...... ...... ...... | |
454 | */ | |
73bb1da6 PT |
455 | lu_site_stats_print(&site->cs_lu, m); |
456 | cache_stats_print(&site->cs_pages, m, 1); | |
457 | seq_printf(m, " ["); | |
d7e09d03 | 458 | for (i = 0; i < ARRAY_SIZE(site->cs_pages_state); ++i) |
73bb1da6 | 459 | seq_printf(m, "%s: %u ", pstate[i], |
d7e09d03 | 460 | atomic_read(&site->cs_pages_state[i])); |
73bb1da6 PT |
461 | seq_printf(m, "]\n"); |
462 | cache_stats_print(&site->cs_locks, m, 0); | |
463 | seq_printf(m, " ["); | |
d7e09d03 | 464 | for (i = 0; i < ARRAY_SIZE(site->cs_locks_state); ++i) |
73bb1da6 | 465 | seq_printf(m, "%s: %u ", lstate[i], |
d7e09d03 | 466 | atomic_read(&site->cs_locks_state[i])); |
73bb1da6 PT |
467 | seq_printf(m, "]\n"); |
468 | cache_stats_print(&cl_env_stats, m, 0); | |
469 | seq_printf(m, "\n"); | |
470 | return 0; | |
d7e09d03 PT |
471 | } |
472 | EXPORT_SYMBOL(cl_site_stats_print); | |
473 | ||
474 | /***************************************************************************** | |
475 | * | |
476 | * lu_env handling on client. | |
477 | * | |
478 | */ | |
479 | ||
480 | /** | |
481 | * The most efficient way is to store cl_env pointer in task specific | |
482 | * structures. On Linux, it wont' be easy to use task_struct->journal_info | |
483 | * because Lustre code may call into other fs which has certain assumptions | |
484 | * about journal_info. Currently following fields in task_struct are identified | |
485 | * can be used for this purpose: | |
486 | * - cl_env: for liblustre. | |
52cf6acf | 487 | * - tux_info: only on RedHat kernel. |
d7e09d03 PT |
488 | * - ... |
489 | * \note As long as we use task_struct to store cl_env, we assume that once | |
490 | * called into Lustre, we'll never call into the other part of the kernel | |
491 | * which will use those fields in task_struct without explicitly exiting | |
492 | * Lustre. | |
493 | * | |
494 | * If there's no space in task_struct is available, hash will be used. | |
495 | * bz20044, bz22683. | |
496 | */ | |
497 | ||
498 | struct cl_env { | |
499 | void *ce_magic; | |
500 | struct lu_env ce_lu; | |
501 | struct lu_context ce_ses; | |
502 | ||
503 | /** | |
504 | * This allows cl_env to be entered into cl_env_hash which implements | |
505 | * the current thread -> client environment lookup. | |
506 | */ | |
507 | struct hlist_node ce_node; | |
508 | /** | |
509 | * Owner for the current cl_env. | |
510 | * | |
511 | * If LL_TASK_CL_ENV is defined, this point to the owning current, | |
512 | * only for debugging purpose ; | |
513 | * Otherwise hash is used, and this is the key for cfs_hash. | |
514 | * Now current thread pid is stored. Note using thread pointer would | |
515 | * lead to unbalanced hash because of its specific allocation locality | |
516 | * and could be varied for different platforms and OSes, even different | |
517 | * OS versions. | |
518 | */ | |
519 | void *ce_owner; | |
520 | ||
521 | /* | |
522 | * Linkage into global list of all client environments. Used for | |
523 | * garbage collection. | |
524 | */ | |
525 | struct list_head ce_linkage; | |
526 | /* | |
527 | * | |
528 | */ | |
529 | int ce_ref; | |
530 | /* | |
531 | * Debugging field: address of the caller who made original | |
532 | * allocation. | |
533 | */ | |
534 | void *ce_debug; | |
535 | }; | |
536 | ||
537 | #define CL_ENV_INC(counter) | |
538 | #define CL_ENV_DEC(counter) | |
539 | ||
540 | static void cl_env_init0(struct cl_env *cle, void *debug) | |
541 | { | |
542 | LASSERT(cle->ce_ref == 0); | |
543 | LASSERT(cle->ce_magic == &cl_env_init0); | |
544 | LASSERT(cle->ce_debug == NULL && cle->ce_owner == NULL); | |
545 | ||
546 | cle->ce_ref = 1; | |
547 | cle->ce_debug = debug; | |
548 | CL_ENV_INC(busy); | |
549 | } | |
550 | ||
d7e09d03 PT |
551 | /* |
552 | * The implementation of using hash table to connect cl_env and thread | |
553 | */ | |
554 | ||
6da6eabe | 555 | static struct cfs_hash *cl_env_hash; |
d7e09d03 | 556 | |
6da6eabe | 557 | static unsigned cl_env_hops_hash(struct cfs_hash *lh, |
d7e09d03 PT |
558 | const void *key, unsigned mask) |
559 | { | |
560 | #if BITS_PER_LONG == 64 | |
561 | return cfs_hash_u64_hash((__u64)key, mask); | |
562 | #else | |
563 | return cfs_hash_u32_hash((__u32)key, mask); | |
564 | #endif | |
565 | } | |
566 | ||
567 | static void *cl_env_hops_obj(struct hlist_node *hn) | |
568 | { | |
569 | struct cl_env *cle = hlist_entry(hn, struct cl_env, ce_node); | |
50ffcb7e | 570 | |
d7e09d03 PT |
571 | LASSERT(cle->ce_magic == &cl_env_init0); |
572 | return (void *)cle; | |
573 | } | |
574 | ||
575 | static int cl_env_hops_keycmp(const void *key, struct hlist_node *hn) | |
576 | { | |
577 | struct cl_env *cle = cl_env_hops_obj(hn); | |
578 | ||
579 | LASSERT(cle->ce_owner != NULL); | |
580 | return (key == cle->ce_owner); | |
581 | } | |
582 | ||
6da6eabe | 583 | static void cl_env_hops_noop(struct cfs_hash *hs, struct hlist_node *hn) |
d7e09d03 PT |
584 | { |
585 | struct cl_env *cle = hlist_entry(hn, struct cl_env, ce_node); | |
50ffcb7e | 586 | |
d7e09d03 PT |
587 | LASSERT(cle->ce_magic == &cl_env_init0); |
588 | } | |
589 | ||
db9fc06b | 590 | static struct cfs_hash_ops cl_env_hops = { |
d7e09d03 | 591 | .hs_hash = cl_env_hops_hash, |
db9fc06b | 592 | .hs_key = cl_env_hops_obj, |
d7e09d03 PT |
593 | .hs_keycmp = cl_env_hops_keycmp, |
594 | .hs_object = cl_env_hops_obj, | |
db9fc06b | 595 | .hs_get = cl_env_hops_noop, |
d7e09d03 PT |
596 | .hs_put_locked = cl_env_hops_noop, |
597 | }; | |
598 | ||
599 | static inline struct cl_env *cl_env_fetch(void) | |
600 | { | |
601 | struct cl_env *cle; | |
602 | ||
603 | cle = cfs_hash_lookup(cl_env_hash, (void *) (long) current->pid); | |
604 | LASSERT(ergo(cle, cle->ce_magic == &cl_env_init0)); | |
605 | return cle; | |
606 | } | |
607 | ||
608 | static inline void cl_env_attach(struct cl_env *cle) | |
609 | { | |
610 | if (cle) { | |
611 | int rc; | |
612 | ||
613 | LASSERT(cle->ce_owner == NULL); | |
614 | cle->ce_owner = (void *) (long) current->pid; | |
615 | rc = cfs_hash_add_unique(cl_env_hash, cle->ce_owner, | |
616 | &cle->ce_node); | |
617 | LASSERT(rc == 0); | |
618 | } | |
619 | } | |
620 | ||
621 | static inline void cl_env_do_detach(struct cl_env *cle) | |
622 | { | |
623 | void *cookie; | |
624 | ||
625 | LASSERT(cle->ce_owner == (void *) (long) current->pid); | |
626 | cookie = cfs_hash_del(cl_env_hash, cle->ce_owner, | |
627 | &cle->ce_node); | |
628 | LASSERT(cookie == cle); | |
629 | cle->ce_owner = NULL; | |
630 | } | |
631 | ||
60753e90 MR |
632 | static int cl_env_store_init(void) |
633 | { | |
d7e09d03 PT |
634 | cl_env_hash = cfs_hash_create("cl_env", |
635 | HASH_CL_ENV_BITS, HASH_CL_ENV_BITS, | |
636 | HASH_CL_ENV_BKT_BITS, 0, | |
637 | CFS_HASH_MIN_THETA, | |
638 | CFS_HASH_MAX_THETA, | |
639 | &cl_env_hops, | |
640 | CFS_HASH_RW_BKTLOCK); | |
b2952d62 | 641 | return cl_env_hash != NULL ? 0 : -ENOMEM; |
d7e09d03 PT |
642 | } |
643 | ||
4d4e1eef DA |
644 | static void cl_env_store_fini(void) |
645 | { | |
d7e09d03 PT |
646 | cfs_hash_putref(cl_env_hash); |
647 | } | |
648 | ||
d7e09d03 PT |
649 | static inline struct cl_env *cl_env_detach(struct cl_env *cle) |
650 | { | |
651 | if (cle == NULL) | |
652 | cle = cl_env_fetch(); | |
653 | ||
654 | if (cle && cle->ce_owner) | |
655 | cl_env_do_detach(cle); | |
656 | ||
657 | return cle; | |
658 | } | |
659 | ||
660 | static struct lu_env *cl_env_new(__u32 ctx_tags, __u32 ses_tags, void *debug) | |
661 | { | |
662 | struct lu_env *env; | |
663 | struct cl_env *cle; | |
664 | ||
ccaabce1 | 665 | cle = kmem_cache_alloc(cl_env_kmem, GFP_NOFS | __GFP_ZERO); |
d7e09d03 PT |
666 | if (cle != NULL) { |
667 | int rc; | |
668 | ||
669 | INIT_LIST_HEAD(&cle->ce_linkage); | |
670 | cle->ce_magic = &cl_env_init0; | |
671 | env = &cle->ce_lu; | |
9c5d9fa0 | 672 | rc = lu_env_init(env, ctx_tags | LCT_CL_THREAD); |
d7e09d03 PT |
673 | if (rc == 0) { |
674 | rc = lu_context_init(&cle->ce_ses, | |
9c5d9fa0 | 675 | ses_tags | LCT_SESSION); |
d7e09d03 PT |
676 | if (rc == 0) { |
677 | lu_context_enter(&cle->ce_ses); | |
678 | env->le_ses = &cle->ce_ses; | |
679 | cl_env_init0(cle, debug); | |
680 | } else | |
681 | lu_env_fini(env); | |
682 | } | |
683 | if (rc != 0) { | |
50d30362 | 684 | kmem_cache_free(cl_env_kmem, cle); |
d7e09d03 PT |
685 | env = ERR_PTR(rc); |
686 | } else { | |
687 | CL_ENV_INC(create); | |
688 | CL_ENV_INC(total); | |
689 | } | |
690 | } else | |
691 | env = ERR_PTR(-ENOMEM); | |
692 | return env; | |
693 | } | |
694 | ||
695 | static void cl_env_fini(struct cl_env *cle) | |
696 | { | |
697 | CL_ENV_DEC(total); | |
698 | lu_context_fini(&cle->ce_lu.le_ctx); | |
699 | lu_context_fini(&cle->ce_ses); | |
50d30362 | 700 | kmem_cache_free(cl_env_kmem, cle); |
d7e09d03 PT |
701 | } |
702 | ||
703 | static inline struct cl_env *cl_env_container(struct lu_env *env) | |
704 | { | |
705 | return container_of(env, struct cl_env, ce_lu); | |
706 | } | |
707 | ||
9569ea54 | 708 | static struct lu_env *cl_env_peek(int *refcheck) |
d7e09d03 PT |
709 | { |
710 | struct lu_env *env; | |
711 | struct cl_env *cle; | |
712 | ||
713 | CL_ENV_INC(lookup); | |
714 | ||
715 | /* check that we don't go far from untrusted pointer */ | |
716 | CLASSERT(offsetof(struct cl_env, ce_magic) == 0); | |
717 | ||
718 | env = NULL; | |
719 | cle = cl_env_fetch(); | |
720 | if (cle != NULL) { | |
721 | CL_ENV_INC(hit); | |
722 | env = &cle->ce_lu; | |
723 | *refcheck = ++cle->ce_ref; | |
724 | } | |
725 | CDEBUG(D_OTHER, "%d@%p\n", cle ? cle->ce_ref : 0, cle); | |
726 | return env; | |
727 | } | |
d7e09d03 PT |
728 | |
729 | /** | |
730 | * Returns lu_env: if there already is an environment associated with the | |
731 | * current thread, it is returned, otherwise, new environment is allocated. | |
732 | * | |
733 | * \param refcheck pointer to a counter used to detect environment leaks. In | |
734 | * the usual case cl_env_get() and cl_env_put() are called in the same lexical | |
735 | * scope and pointer to the same integer is passed as \a refcheck. This is | |
736 | * used to detect missed cl_env_put(). | |
737 | * | |
738 | * \see cl_env_put() | |
739 | */ | |
740 | struct lu_env *cl_env_get(int *refcheck) | |
741 | { | |
742 | struct lu_env *env; | |
743 | ||
744 | env = cl_env_peek(refcheck); | |
745 | if (env == NULL) { | |
746 | env = cl_env_new(lu_context_tags_default, | |
747 | lu_session_tags_default, | |
748 | __builtin_return_address(0)); | |
749 | ||
750 | if (!IS_ERR(env)) { | |
751 | struct cl_env *cle; | |
752 | ||
753 | cle = cl_env_container(env); | |
754 | cl_env_attach(cle); | |
755 | *refcheck = cle->ce_ref; | |
756 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
757 | } | |
758 | } | |
759 | return env; | |
760 | } | |
761 | EXPORT_SYMBOL(cl_env_get); | |
762 | ||
763 | /** | |
764 | * Forces an allocation of a fresh environment with given tags. | |
765 | * | |
766 | * \see cl_env_get() | |
767 | */ | |
768 | struct lu_env *cl_env_alloc(int *refcheck, __u32 tags) | |
769 | { | |
770 | struct lu_env *env; | |
771 | ||
772 | LASSERT(cl_env_peek(refcheck) == NULL); | |
773 | env = cl_env_new(tags, tags, __builtin_return_address(0)); | |
774 | if (!IS_ERR(env)) { | |
775 | struct cl_env *cle; | |
776 | ||
777 | cle = cl_env_container(env); | |
778 | *refcheck = cle->ce_ref; | |
779 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
780 | } | |
781 | return env; | |
782 | } | |
783 | EXPORT_SYMBOL(cl_env_alloc); | |
784 | ||
785 | static void cl_env_exit(struct cl_env *cle) | |
786 | { | |
787 | LASSERT(cle->ce_owner == NULL); | |
788 | lu_context_exit(&cle->ce_lu.le_ctx); | |
789 | lu_context_exit(&cle->ce_ses); | |
790 | } | |
791 | ||
792 | /** | |
793 | * Release an environment. | |
794 | * | |
795 | * Decrement \a env reference counter. When counter drops to 0, nothing in | |
796 | * this thread is using environment and it is returned to the allocation | |
797 | * cache, or freed straight away, if cache is large enough. | |
798 | */ | |
799 | void cl_env_put(struct lu_env *env, int *refcheck) | |
800 | { | |
801 | struct cl_env *cle; | |
802 | ||
803 | cle = cl_env_container(env); | |
804 | ||
805 | LASSERT(cle->ce_ref > 0); | |
806 | LASSERT(ergo(refcheck != NULL, cle->ce_ref == *refcheck)); | |
807 | ||
808 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
809 | if (--cle->ce_ref == 0) { | |
810 | CL_ENV_DEC(busy); | |
811 | cl_env_detach(cle); | |
812 | cle->ce_debug = NULL; | |
813 | cl_env_exit(cle); | |
814 | cl_env_fini(cle); | |
815 | } | |
816 | } | |
817 | EXPORT_SYMBOL(cl_env_put); | |
818 | ||
819 | /** | |
820 | * Declares a point of re-entrancy. | |
821 | * | |
822 | * \see cl_env_reexit() | |
823 | */ | |
824 | void *cl_env_reenter(void) | |
825 | { | |
826 | return cl_env_detach(NULL); | |
827 | } | |
828 | EXPORT_SYMBOL(cl_env_reenter); | |
829 | ||
830 | /** | |
831 | * Exits re-entrancy. | |
832 | */ | |
833 | void cl_env_reexit(void *cookie) | |
834 | { | |
835 | cl_env_detach(NULL); | |
836 | cl_env_attach(cookie); | |
837 | } | |
838 | EXPORT_SYMBOL(cl_env_reexit); | |
839 | ||
840 | /** | |
841 | * Setup user-supplied \a env as a current environment. This is to be used to | |
842 | * guaranteed that environment exists even when cl_env_get() fails. It is up | |
843 | * to user to ensure proper concurrency control. | |
844 | * | |
845 | * \see cl_env_unplant() | |
846 | */ | |
847 | void cl_env_implant(struct lu_env *env, int *refcheck) | |
848 | { | |
849 | struct cl_env *cle = cl_env_container(env); | |
850 | ||
851 | LASSERT(cle->ce_ref > 0); | |
852 | ||
853 | cl_env_attach(cle); | |
854 | cl_env_get(refcheck); | |
855 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
856 | } | |
857 | EXPORT_SYMBOL(cl_env_implant); | |
858 | ||
859 | /** | |
860 | * Detach environment installed earlier by cl_env_implant(). | |
861 | */ | |
862 | void cl_env_unplant(struct lu_env *env, int *refcheck) | |
863 | { | |
864 | struct cl_env *cle = cl_env_container(env); | |
865 | ||
866 | LASSERT(cle->ce_ref > 1); | |
867 | ||
868 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
869 | ||
870 | cl_env_detach(cle); | |
871 | cl_env_put(env, refcheck); | |
872 | } | |
873 | EXPORT_SYMBOL(cl_env_unplant); | |
874 | ||
875 | struct lu_env *cl_env_nested_get(struct cl_env_nest *nest) | |
876 | { | |
877 | struct lu_env *env; | |
878 | ||
879 | nest->cen_cookie = NULL; | |
880 | env = cl_env_peek(&nest->cen_refcheck); | |
881 | if (env != NULL) { | |
882 | if (!cl_io_is_going(env)) | |
883 | return env; | |
71e8dd9a AM |
884 | cl_env_put(env, &nest->cen_refcheck); |
885 | nest->cen_cookie = cl_env_reenter(); | |
d7e09d03 PT |
886 | } |
887 | env = cl_env_get(&nest->cen_refcheck); | |
888 | if (IS_ERR(env)) { | |
889 | cl_env_reexit(nest->cen_cookie); | |
890 | return env; | |
891 | } | |
892 | ||
893 | LASSERT(!cl_io_is_going(env)); | |
894 | return env; | |
895 | } | |
896 | EXPORT_SYMBOL(cl_env_nested_get); | |
897 | ||
898 | void cl_env_nested_put(struct cl_env_nest *nest, struct lu_env *env) | |
899 | { | |
900 | cl_env_put(env, &nest->cen_refcheck); | |
901 | cl_env_reexit(nest->cen_cookie); | |
902 | } | |
903 | EXPORT_SYMBOL(cl_env_nested_put); | |
904 | ||
d7e09d03 PT |
905 | /** |
906 | * Converts struct ost_lvb to struct cl_attr. | |
907 | * | |
908 | * \see cl_attr2lvb | |
909 | */ | |
910 | void cl_lvb2attr(struct cl_attr *attr, const struct ost_lvb *lvb) | |
911 | { | |
d7e09d03 PT |
912 | attr->cat_size = lvb->lvb_size; |
913 | attr->cat_mtime = lvb->lvb_mtime; | |
914 | attr->cat_atime = lvb->lvb_atime; | |
915 | attr->cat_ctime = lvb->lvb_ctime; | |
916 | attr->cat_blocks = lvb->lvb_blocks; | |
d7e09d03 PT |
917 | } |
918 | EXPORT_SYMBOL(cl_lvb2attr); | |
919 | ||
920 | /***************************************************************************** | |
921 | * | |
922 | * Temporary prototype thing: mirror obd-devices into cl devices. | |
923 | * | |
924 | */ | |
925 | ||
926 | struct cl_device *cl_type_setup(const struct lu_env *env, struct lu_site *site, | |
927 | struct lu_device_type *ldt, | |
928 | struct lu_device *next) | |
929 | { | |
930 | const char *typename; | |
931 | struct lu_device *d; | |
932 | ||
933 | LASSERT(ldt != NULL); | |
934 | ||
935 | typename = ldt->ldt_name; | |
936 | d = ldt->ldt_ops->ldto_device_alloc(env, ldt, NULL); | |
937 | if (!IS_ERR(d)) { | |
938 | int rc; | |
939 | ||
940 | if (site != NULL) | |
941 | d->ld_site = site; | |
942 | rc = ldt->ldt_ops->ldto_device_init(env, d, typename, next); | |
943 | if (rc == 0) { | |
944 | lu_device_get(d); | |
945 | lu_ref_add(&d->ld_reference, | |
946 | "lu-stack", &lu_site_init); | |
947 | } else { | |
948 | ldt->ldt_ops->ldto_device_free(env, d); | |
949 | CERROR("can't init device '%s', %d\n", typename, rc); | |
950 | d = ERR_PTR(rc); | |
951 | } | |
952 | } else | |
953 | CERROR("Cannot allocate device: '%s'\n", typename); | |
954 | return lu2cl_dev(d); | |
955 | } | |
956 | EXPORT_SYMBOL(cl_type_setup); | |
957 | ||
958 | /** | |
959 | * Finalize device stack by calling lu_stack_fini(). | |
960 | */ | |
961 | void cl_stack_fini(const struct lu_env *env, struct cl_device *cl) | |
962 | { | |
963 | lu_stack_fini(env, cl2lu_dev(cl)); | |
964 | } | |
965 | EXPORT_SYMBOL(cl_stack_fini); | |
966 | ||
967 | int cl_lock_init(void); | |
968 | void cl_lock_fini(void); | |
969 | ||
970 | int cl_page_init(void); | |
971 | void cl_page_fini(void); | |
972 | ||
973 | static struct lu_context_key cl_key; | |
974 | ||
975 | struct cl_thread_info *cl_env_info(const struct lu_env *env) | |
976 | { | |
977 | return lu_context_key_get(&env->le_ctx, &cl_key); | |
978 | } | |
979 | ||
980 | /* defines cl0_key_{init,fini}() */ | |
981 | LU_KEY_INIT_FINI(cl0, struct cl_thread_info); | |
982 | ||
983 | static void *cl_key_init(const struct lu_context *ctx, | |
984 | struct lu_context_key *key) | |
985 | { | |
986 | struct cl_thread_info *info; | |
987 | ||
988 | info = cl0_key_init(ctx, key); | |
989 | if (!IS_ERR(info)) { | |
990 | int i; | |
991 | ||
992 | for (i = 0; i < ARRAY_SIZE(info->clt_counters); ++i) | |
993 | lu_ref_init(&info->clt_counters[i].ctc_locks_locked); | |
994 | } | |
995 | return info; | |
996 | } | |
997 | ||
998 | static void cl_key_fini(const struct lu_context *ctx, | |
999 | struct lu_context_key *key, void *data) | |
1000 | { | |
1001 | struct cl_thread_info *info; | |
1002 | int i; | |
1003 | ||
1004 | info = data; | |
1005 | for (i = 0; i < ARRAY_SIZE(info->clt_counters); ++i) | |
1006 | lu_ref_fini(&info->clt_counters[i].ctc_locks_locked); | |
1007 | cl0_key_fini(ctx, key, data); | |
1008 | } | |
1009 | ||
1010 | static void cl_key_exit(const struct lu_context *ctx, | |
1011 | struct lu_context_key *key, void *data) | |
1012 | { | |
1013 | struct cl_thread_info *info = data; | |
1014 | int i; | |
1015 | ||
1016 | for (i = 0; i < ARRAY_SIZE(info->clt_counters); ++i) { | |
1017 | LASSERT(info->clt_counters[i].ctc_nr_held == 0); | |
1018 | LASSERT(info->clt_counters[i].ctc_nr_used == 0); | |
1019 | LASSERT(info->clt_counters[i].ctc_nr_locks_acquired == 0); | |
1020 | LASSERT(info->clt_counters[i].ctc_nr_locks_locked == 0); | |
1021 | lu_ref_fini(&info->clt_counters[i].ctc_locks_locked); | |
1022 | lu_ref_init(&info->clt_counters[i].ctc_locks_locked); | |
1023 | } | |
1024 | } | |
1025 | ||
1026 | static struct lu_context_key cl_key = { | |
1027 | .lct_tags = LCT_CL_THREAD, | |
1028 | .lct_init = cl_key_init, | |
1029 | .lct_fini = cl_key_fini, | |
1030 | .lct_exit = cl_key_exit | |
1031 | }; | |
1032 | ||
1033 | static struct lu_kmem_descr cl_object_caches[] = { | |
1034 | { | |
1035 | .ckd_cache = &cl_env_kmem, | |
1036 | .ckd_name = "cl_env_kmem", | |
60aff4ff | 1037 | .ckd_size = sizeof(struct cl_env) |
d7e09d03 PT |
1038 | }, |
1039 | { | |
1040 | .ckd_cache = NULL | |
1041 | } | |
1042 | }; | |
1043 | ||
1044 | /** | |
1045 | * Global initialization of cl-data. Create kmem caches, register | |
1046 | * lu_context_key's, etc. | |
1047 | * | |
1048 | * \see cl_global_fini() | |
1049 | */ | |
1050 | int cl_global_init(void) | |
1051 | { | |
1052 | int result; | |
1053 | ||
1054 | result = cl_env_store_init(); | |
1055 | if (result) | |
1056 | return result; | |
1057 | ||
1058 | result = lu_kmem_init(cl_object_caches); | |
1059 | if (result) | |
1060 | goto out_store; | |
1061 | ||
1062 | LU_CONTEXT_KEY_INIT(&cl_key); | |
1063 | result = lu_context_key_register(&cl_key); | |
1064 | if (result) | |
1065 | goto out_kmem; | |
1066 | ||
1067 | result = cl_lock_init(); | |
1068 | if (result) | |
1069 | goto out_context; | |
1070 | ||
1071 | result = cl_page_init(); | |
1072 | if (result) | |
1073 | goto out_lock; | |
1074 | ||
1075 | return 0; | |
1076 | out_lock: | |
1077 | cl_lock_fini(); | |
1078 | out_context: | |
1079 | lu_context_key_degister(&cl_key); | |
1080 | out_kmem: | |
1081 | lu_kmem_fini(cl_object_caches); | |
1082 | out_store: | |
1083 | cl_env_store_fini(); | |
1084 | return result; | |
1085 | } | |
1086 | ||
1087 | /** | |
1088 | * Finalization of global cl-data. Dual to cl_global_init(). | |
1089 | */ | |
1090 | void cl_global_fini(void) | |
1091 | { | |
1092 | cl_lock_fini(); | |
1093 | cl_page_fini(); | |
1094 | lu_context_key_degister(&cl_key); | |
1095 | lu_kmem_fini(cl_object_caches); | |
1096 | cl_env_store_fini(); | |
1097 | } |