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 | * | |
30 | * Copyright (c) 2011, 2012, Intel Corporation. | |
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 | ||
54 | #include <linux/libcfs/libcfs.h> | |
55 | /* class_put_type() */ | |
56 | #include <obd_class.h> | |
57 | #include <obd_support.h> | |
58 | #include <lustre_fid.h> | |
59 | #include <linux/list.h> | |
60 | #include <linux/libcfs/libcfs_hash.h> /* for cfs_hash stuff */ | |
61 | #include <cl_object.h> | |
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 | } | |
96 | RETURN(result); | |
97 | } | |
98 | EXPORT_SYMBOL(cl_object_header_init); | |
99 | ||
100 | /** | |
101 | * Finalize cl_object_header. | |
102 | */ | |
103 | void cl_object_header_fini(struct cl_object_header *h) | |
104 | { | |
105 | LASSERT(list_empty(&h->coh_locks)); | |
106 | lu_object_header_fini(&h->coh_lu); | |
107 | } | |
108 | EXPORT_SYMBOL(cl_object_header_fini); | |
109 | ||
110 | /** | |
111 | * Returns a cl_object with a given \a fid. | |
112 | * | |
113 | * Returns either cached or newly created object. Additional reference on the | |
114 | * returned object is acquired. | |
115 | * | |
116 | * \see lu_object_find(), cl_page_find(), cl_lock_find() | |
117 | */ | |
118 | struct cl_object *cl_object_find(const struct lu_env *env, | |
119 | struct cl_device *cd, const struct lu_fid *fid, | |
120 | const struct cl_object_conf *c) | |
121 | { | |
122 | might_sleep(); | |
123 | return lu2cl(lu_object_find_slice(env, cl2lu_dev(cd), fid, &c->coc_lu)); | |
124 | } | |
125 | EXPORT_SYMBOL(cl_object_find); | |
126 | ||
127 | /** | |
128 | * Releases a reference on \a o. | |
129 | * | |
130 | * When last reference is released object is returned to the cache, unless | |
131 | * lu_object_header_flags::LU_OBJECT_HEARD_BANSHEE bit is set in its header. | |
132 | * | |
133 | * \see cl_page_put(), cl_lock_put(). | |
134 | */ | |
135 | void cl_object_put(const struct lu_env *env, struct cl_object *o) | |
136 | { | |
137 | lu_object_put(env, &o->co_lu); | |
138 | } | |
139 | EXPORT_SYMBOL(cl_object_put); | |
140 | ||
141 | /** | |
142 | * Acquire an additional reference to the object \a o. | |
143 | * | |
144 | * This can only be used to acquire _additional_ reference, i.e., caller | |
145 | * already has to possess at least one reference to \a o before calling this. | |
146 | * | |
147 | * \see cl_page_get(), cl_lock_get(). | |
148 | */ | |
149 | void cl_object_get(struct cl_object *o) | |
150 | { | |
151 | lu_object_get(&o->co_lu); | |
152 | } | |
153 | EXPORT_SYMBOL(cl_object_get); | |
154 | ||
155 | /** | |
156 | * Returns the top-object for a given \a o. | |
157 | * | |
158 | * \see cl_page_top(), cl_io_top() | |
159 | */ | |
160 | struct cl_object *cl_object_top(struct cl_object *o) | |
161 | { | |
162 | struct cl_object_header *hdr = cl_object_header(o); | |
163 | struct cl_object *top; | |
164 | ||
165 | while (hdr->coh_parent != NULL) | |
166 | hdr = hdr->coh_parent; | |
167 | ||
168 | top = lu2cl(lu_object_top(&hdr->coh_lu)); | |
169 | CDEBUG(D_TRACE, "%p -> %p\n", o, top); | |
170 | return top; | |
171 | } | |
172 | EXPORT_SYMBOL(cl_object_top); | |
173 | ||
174 | /** | |
175 | * Returns pointer to the lock protecting data-attributes for the given object | |
176 | * \a o. | |
177 | * | |
178 | * Data-attributes are protected by the cl_object_header::coh_attr_guard | |
179 | * spin-lock in the top-object. | |
180 | * | |
181 | * \see cl_attr, cl_object_attr_lock(), cl_object_operations::coo_attr_get(). | |
182 | */ | |
183 | static spinlock_t *cl_object_attr_guard(struct cl_object *o) | |
184 | { | |
185 | return &cl_object_header(cl_object_top(o))->coh_attr_guard; | |
186 | } | |
187 | ||
188 | /** | |
189 | * Locks data-attributes. | |
190 | * | |
191 | * Prevents data-attributes from changing, until lock is released by | |
192 | * cl_object_attr_unlock(). This has to be called before calls to | |
193 | * cl_object_attr_get(), cl_object_attr_set(). | |
194 | */ | |
195 | void cl_object_attr_lock(struct cl_object *o) | |
196 | { | |
197 | spin_lock(cl_object_attr_guard(o)); | |
198 | } | |
199 | EXPORT_SYMBOL(cl_object_attr_lock); | |
200 | ||
201 | /** | |
202 | * Releases data-attributes lock, acquired by cl_object_attr_lock(). | |
203 | */ | |
204 | void cl_object_attr_unlock(struct cl_object *o) | |
205 | { | |
206 | spin_unlock(cl_object_attr_guard(o)); | |
207 | } | |
208 | EXPORT_SYMBOL(cl_object_attr_unlock); | |
209 | ||
210 | /** | |
211 | * Returns data-attributes of an object \a obj. | |
212 | * | |
213 | * Every layer is asked (by calling cl_object_operations::coo_attr_get()) | |
214 | * top-to-bottom to fill in parts of \a attr that this layer is responsible | |
215 | * for. | |
216 | */ | |
217 | int cl_object_attr_get(const struct lu_env *env, struct cl_object *obj, | |
218 | struct cl_attr *attr) | |
219 | { | |
220 | struct lu_object_header *top; | |
221 | int result; | |
222 | ||
223 | LASSERT(spin_is_locked(cl_object_attr_guard(obj))); | |
d7e09d03 PT |
224 | |
225 | top = obj->co_lu.lo_header; | |
226 | result = 0; | |
227 | list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) { | |
228 | if (obj->co_ops->coo_attr_get != NULL) { | |
229 | result = obj->co_ops->coo_attr_get(env, obj, attr); | |
230 | if (result != 0) { | |
231 | if (result > 0) | |
232 | result = 0; | |
233 | break; | |
234 | } | |
235 | } | |
236 | } | |
237 | RETURN(result); | |
238 | } | |
239 | EXPORT_SYMBOL(cl_object_attr_get); | |
240 | ||
241 | /** | |
242 | * Updates data-attributes of an object \a obj. | |
243 | * | |
244 | * Only attributes, mentioned in a validness bit-mask \a v are | |
245 | * updated. Calls cl_object_operations::coo_attr_set() on every layer, bottom | |
246 | * to top. | |
247 | */ | |
248 | int cl_object_attr_set(const struct lu_env *env, struct cl_object *obj, | |
249 | const struct cl_attr *attr, unsigned v) | |
250 | { | |
251 | struct lu_object_header *top; | |
252 | int result; | |
253 | ||
254 | LASSERT(spin_is_locked(cl_object_attr_guard(obj))); | |
d7e09d03 PT |
255 | |
256 | top = obj->co_lu.lo_header; | |
257 | result = 0; | |
258 | list_for_each_entry_reverse(obj, &top->loh_layers, | |
259 | co_lu.lo_linkage) { | |
260 | if (obj->co_ops->coo_attr_set != NULL) { | |
261 | result = obj->co_ops->coo_attr_set(env, obj, attr, v); | |
262 | if (result != 0) { | |
263 | if (result > 0) | |
264 | result = 0; | |
265 | break; | |
266 | } | |
267 | } | |
268 | } | |
269 | RETURN(result); | |
270 | } | |
271 | EXPORT_SYMBOL(cl_object_attr_set); | |
272 | ||
273 | /** | |
274 | * Notifies layers (bottom-to-top) that glimpse AST was received. | |
275 | * | |
276 | * Layers have to fill \a lvb fields with information that will be shipped | |
277 | * back to glimpse issuer. | |
278 | * | |
279 | * \see cl_lock_operations::clo_glimpse() | |
280 | */ | |
281 | int cl_object_glimpse(const struct lu_env *env, struct cl_object *obj, | |
282 | struct ost_lvb *lvb) | |
283 | { | |
284 | struct lu_object_header *top; | |
285 | int result; | |
286 | ||
d7e09d03 PT |
287 | top = obj->co_lu.lo_header; |
288 | result = 0; | |
289 | list_for_each_entry_reverse(obj, &top->loh_layers, | |
290 | co_lu.lo_linkage) { | |
291 | if (obj->co_ops->coo_glimpse != NULL) { | |
292 | result = obj->co_ops->coo_glimpse(env, obj, lvb); | |
293 | if (result != 0) | |
294 | break; | |
295 | } | |
296 | } | |
297 | LU_OBJECT_HEADER(D_DLMTRACE, env, lu_object_top(top), | |
298 | "size: "LPU64" mtime: "LPU64" atime: "LPU64" " | |
299 | "ctime: "LPU64" blocks: "LPU64"\n", | |
300 | lvb->lvb_size, lvb->lvb_mtime, lvb->lvb_atime, | |
301 | lvb->lvb_ctime, lvb->lvb_blocks); | |
302 | RETURN(result); | |
303 | } | |
304 | EXPORT_SYMBOL(cl_object_glimpse); | |
305 | ||
306 | /** | |
307 | * Updates a configuration of an object \a obj. | |
308 | */ | |
309 | int cl_conf_set(const struct lu_env *env, struct cl_object *obj, | |
310 | const struct cl_object_conf *conf) | |
311 | { | |
312 | struct lu_object_header *top; | |
313 | int result; | |
314 | ||
d7e09d03 PT |
315 | top = obj->co_lu.lo_header; |
316 | result = 0; | |
317 | list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) { | |
318 | if (obj->co_ops->coo_conf_set != NULL) { | |
319 | result = obj->co_ops->coo_conf_set(env, obj, conf); | |
320 | if (result != 0) | |
321 | break; | |
322 | } | |
323 | } | |
324 | RETURN(result); | |
325 | } | |
326 | EXPORT_SYMBOL(cl_conf_set); | |
327 | ||
328 | /** | |
329 | * Helper function removing all object locks, and marking object for | |
330 | * deletion. All object pages must have been deleted at this point. | |
331 | * | |
332 | * This is called by cl_inode_fini() and lov_object_delete() to destroy top- | |
333 | * and sub- objects respectively. | |
334 | */ | |
335 | void cl_object_kill(const struct lu_env *env, struct cl_object *obj) | |
336 | { | |
337 | struct cl_object_header *hdr; | |
338 | ||
339 | hdr = cl_object_header(obj); | |
340 | LASSERT(hdr->coh_tree.rnode == NULL); | |
341 | LASSERT(hdr->coh_pages == 0); | |
342 | ||
343 | set_bit(LU_OBJECT_HEARD_BANSHEE, &hdr->coh_lu.loh_flags); | |
344 | /* | |
345 | * Destroy all locks. Object destruction (including cl_inode_fini()) | |
346 | * cannot cancel the locks, because in the case of a local client, | |
347 | * where client and server share the same thread running | |
348 | * prune_icache(), this can dead-lock with ldlm_cancel_handler() | |
349 | * waiting on __wait_on_freeing_inode(). | |
350 | */ | |
351 | cl_locks_prune(env, obj, 0); | |
352 | } | |
353 | EXPORT_SYMBOL(cl_object_kill); | |
354 | ||
355 | /** | |
356 | * Prunes caches of pages and locks for this object. | |
357 | */ | |
358 | void cl_object_prune(const struct lu_env *env, struct cl_object *obj) | |
359 | { | |
d7e09d03 PT |
360 | cl_pages_prune(env, obj); |
361 | cl_locks_prune(env, obj, 1); | |
362 | EXIT; | |
363 | } | |
364 | EXPORT_SYMBOL(cl_object_prune); | |
365 | ||
366 | /** | |
367 | * Check if the object has locks. | |
368 | */ | |
369 | int cl_object_has_locks(struct cl_object *obj) | |
370 | { | |
371 | struct cl_object_header *head = cl_object_header(obj); | |
372 | int has; | |
373 | ||
374 | spin_lock(&head->coh_lock_guard); | |
375 | has = list_empty(&head->coh_locks); | |
376 | spin_unlock(&head->coh_lock_guard); | |
377 | ||
378 | return (has == 0); | |
379 | } | |
380 | EXPORT_SYMBOL(cl_object_has_locks); | |
381 | ||
382 | void cache_stats_init(struct cache_stats *cs, const char *name) | |
383 | { | |
384 | int i; | |
385 | ||
386 | cs->cs_name = name; | |
387 | for (i = 0; i < CS_NR; i++) | |
388 | atomic_set(&cs->cs_stats[i], 0); | |
389 | } | |
390 | ||
73bb1da6 | 391 | int cache_stats_print(const struct cache_stats *cs, struct seq_file *m, int h) |
d7e09d03 | 392 | { |
d7e09d03 PT |
393 | int i; |
394 | /* | |
395 | * lookup hit total cached create | |
396 | * env: ...... ...... ...... ...... ...... | |
397 | */ | |
398 | if (h) { | |
399 | const char *names[CS_NR] = CS_NAMES; | |
400 | ||
73bb1da6 | 401 | seq_printf(m, "%6s", " "); |
d7e09d03 | 402 | for (i = 0; i < CS_NR; i++) |
73bb1da6 PT |
403 | seq_printf(m, "%8s", names[i]); |
404 | seq_printf(m, "\n"); | |
d7e09d03 PT |
405 | } |
406 | ||
73bb1da6 | 407 | seq_printf(m, "%5.5s:", cs->cs_name); |
d7e09d03 | 408 | for (i = 0; i < CS_NR; i++) |
73bb1da6 PT |
409 | seq_printf(m, "%8u", atomic_read(&cs->cs_stats[i])); |
410 | return 0; | |
d7e09d03 PT |
411 | } |
412 | ||
413 | /** | |
414 | * Initialize client site. | |
415 | * | |
416 | * Perform common initialization (lu_site_init()), and initialize statistical | |
417 | * counters. Also perform global initializations on the first call. | |
418 | */ | |
419 | int cl_site_init(struct cl_site *s, struct cl_device *d) | |
420 | { | |
421 | int i; | |
422 | int result; | |
423 | ||
424 | result = lu_site_init(&s->cs_lu, &d->cd_lu_dev); | |
425 | if (result == 0) { | |
426 | cache_stats_init(&s->cs_pages, "pages"); | |
427 | cache_stats_init(&s->cs_locks, "locks"); | |
428 | for (i = 0; i < ARRAY_SIZE(s->cs_pages_state); ++i) | |
429 | atomic_set(&s->cs_pages_state[0], 0); | |
430 | for (i = 0; i < ARRAY_SIZE(s->cs_locks_state); ++i) | |
431 | atomic_set(&s->cs_locks_state[i], 0); | |
432 | } | |
433 | return result; | |
434 | } | |
435 | EXPORT_SYMBOL(cl_site_init); | |
436 | ||
437 | /** | |
438 | * Finalize client site. Dual to cl_site_init(). | |
439 | */ | |
440 | void cl_site_fini(struct cl_site *s) | |
441 | { | |
442 | lu_site_fini(&s->cs_lu); | |
443 | } | |
444 | EXPORT_SYMBOL(cl_site_fini); | |
445 | ||
446 | static struct cache_stats cl_env_stats = { | |
447 | .cs_name = "envs", | |
448 | .cs_stats = { ATOMIC_INIT(0), } | |
449 | }; | |
450 | ||
451 | /** | |
452 | * Outputs client site statistical counters into a buffer. Suitable for | |
453 | * ll_rd_*()-style functions. | |
454 | */ | |
73bb1da6 | 455 | int cl_site_stats_print(const struct cl_site *site, struct seq_file *m) |
d7e09d03 | 456 | { |
d7e09d03 PT |
457 | int i; |
458 | static const char *pstate[] = { | |
459 | [CPS_CACHED] = "c", | |
460 | [CPS_OWNED] = "o", | |
461 | [CPS_PAGEOUT] = "w", | |
462 | [CPS_PAGEIN] = "r", | |
463 | [CPS_FREEING] = "f" | |
464 | }; | |
465 | static const char *lstate[] = { | |
466 | [CLS_NEW] = "n", | |
467 | [CLS_QUEUING] = "q", | |
468 | [CLS_ENQUEUED] = "e", | |
469 | [CLS_HELD] = "h", | |
470 | [CLS_INTRANSIT] = "t", | |
471 | [CLS_CACHED] = "c", | |
472 | [CLS_FREEING] = "f" | |
473 | }; | |
474 | /* | |
475 | lookup hit total busy create | |
476 | pages: ...... ...... ...... ...... ...... [...... ...... ...... ......] | |
477 | locks: ...... ...... ...... ...... ...... [...... ...... ...... ...... ......] | |
478 | env: ...... ...... ...... ...... ...... | |
479 | */ | |
73bb1da6 PT |
480 | lu_site_stats_print(&site->cs_lu, m); |
481 | cache_stats_print(&site->cs_pages, m, 1); | |
482 | seq_printf(m, " ["); | |
d7e09d03 | 483 | for (i = 0; i < ARRAY_SIZE(site->cs_pages_state); ++i) |
73bb1da6 | 484 | seq_printf(m, "%s: %u ", pstate[i], |
d7e09d03 | 485 | atomic_read(&site->cs_pages_state[i])); |
73bb1da6 PT |
486 | seq_printf(m, "]\n"); |
487 | cache_stats_print(&site->cs_locks, m, 0); | |
488 | seq_printf(m, " ["); | |
d7e09d03 | 489 | for (i = 0; i < ARRAY_SIZE(site->cs_locks_state); ++i) |
73bb1da6 | 490 | seq_printf(m, "%s: %u ", lstate[i], |
d7e09d03 | 491 | atomic_read(&site->cs_locks_state[i])); |
73bb1da6 PT |
492 | seq_printf(m, "]\n"); |
493 | cache_stats_print(&cl_env_stats, m, 0); | |
494 | seq_printf(m, "\n"); | |
495 | return 0; | |
d7e09d03 PT |
496 | } |
497 | EXPORT_SYMBOL(cl_site_stats_print); | |
498 | ||
499 | /***************************************************************************** | |
500 | * | |
501 | * lu_env handling on client. | |
502 | * | |
503 | */ | |
504 | ||
505 | /** | |
506 | * The most efficient way is to store cl_env pointer in task specific | |
507 | * structures. On Linux, it wont' be easy to use task_struct->journal_info | |
508 | * because Lustre code may call into other fs which has certain assumptions | |
509 | * about journal_info. Currently following fields in task_struct are identified | |
510 | * can be used for this purpose: | |
511 | * - cl_env: for liblustre. | |
512 | * - tux_info: ony on RedHat kernel. | |
513 | * - ... | |
514 | * \note As long as we use task_struct to store cl_env, we assume that once | |
515 | * called into Lustre, we'll never call into the other part of the kernel | |
516 | * which will use those fields in task_struct without explicitly exiting | |
517 | * Lustre. | |
518 | * | |
519 | * If there's no space in task_struct is available, hash will be used. | |
520 | * bz20044, bz22683. | |
521 | */ | |
522 | ||
523 | struct cl_env { | |
524 | void *ce_magic; | |
525 | struct lu_env ce_lu; | |
526 | struct lu_context ce_ses; | |
527 | ||
528 | /** | |
529 | * This allows cl_env to be entered into cl_env_hash which implements | |
530 | * the current thread -> client environment lookup. | |
531 | */ | |
532 | struct hlist_node ce_node; | |
533 | /** | |
534 | * Owner for the current cl_env. | |
535 | * | |
536 | * If LL_TASK_CL_ENV is defined, this point to the owning current, | |
537 | * only for debugging purpose ; | |
538 | * Otherwise hash is used, and this is the key for cfs_hash. | |
539 | * Now current thread pid is stored. Note using thread pointer would | |
540 | * lead to unbalanced hash because of its specific allocation locality | |
541 | * and could be varied for different platforms and OSes, even different | |
542 | * OS versions. | |
543 | */ | |
544 | void *ce_owner; | |
545 | ||
546 | /* | |
547 | * Linkage into global list of all client environments. Used for | |
548 | * garbage collection. | |
549 | */ | |
550 | struct list_head ce_linkage; | |
551 | /* | |
552 | * | |
553 | */ | |
554 | int ce_ref; | |
555 | /* | |
556 | * Debugging field: address of the caller who made original | |
557 | * allocation. | |
558 | */ | |
559 | void *ce_debug; | |
560 | }; | |
561 | ||
562 | #define CL_ENV_INC(counter) | |
563 | #define CL_ENV_DEC(counter) | |
564 | ||
565 | static void cl_env_init0(struct cl_env *cle, void *debug) | |
566 | { | |
567 | LASSERT(cle->ce_ref == 0); | |
568 | LASSERT(cle->ce_magic == &cl_env_init0); | |
569 | LASSERT(cle->ce_debug == NULL && cle->ce_owner == NULL); | |
570 | ||
571 | cle->ce_ref = 1; | |
572 | cle->ce_debug = debug; | |
573 | CL_ENV_INC(busy); | |
574 | } | |
575 | ||
576 | ||
577 | /* | |
578 | * The implementation of using hash table to connect cl_env and thread | |
579 | */ | |
580 | ||
581 | static cfs_hash_t *cl_env_hash; | |
582 | ||
583 | static unsigned cl_env_hops_hash(cfs_hash_t *lh, | |
584 | const void *key, unsigned mask) | |
585 | { | |
586 | #if BITS_PER_LONG == 64 | |
587 | return cfs_hash_u64_hash((__u64)key, mask); | |
588 | #else | |
589 | return cfs_hash_u32_hash((__u32)key, mask); | |
590 | #endif | |
591 | } | |
592 | ||
593 | static void *cl_env_hops_obj(struct hlist_node *hn) | |
594 | { | |
595 | struct cl_env *cle = hlist_entry(hn, struct cl_env, ce_node); | |
596 | LASSERT(cle->ce_magic == &cl_env_init0); | |
597 | return (void *)cle; | |
598 | } | |
599 | ||
600 | static int cl_env_hops_keycmp(const void *key, struct hlist_node *hn) | |
601 | { | |
602 | struct cl_env *cle = cl_env_hops_obj(hn); | |
603 | ||
604 | LASSERT(cle->ce_owner != NULL); | |
605 | return (key == cle->ce_owner); | |
606 | } | |
607 | ||
608 | static void cl_env_hops_noop(cfs_hash_t *hs, struct hlist_node *hn) | |
609 | { | |
610 | struct cl_env *cle = hlist_entry(hn, struct cl_env, ce_node); | |
611 | LASSERT(cle->ce_magic == &cl_env_init0); | |
612 | } | |
613 | ||
614 | static cfs_hash_ops_t cl_env_hops = { | |
615 | .hs_hash = cl_env_hops_hash, | |
616 | .hs_key = cl_env_hops_obj, | |
617 | .hs_keycmp = cl_env_hops_keycmp, | |
618 | .hs_object = cl_env_hops_obj, | |
619 | .hs_get = cl_env_hops_noop, | |
620 | .hs_put_locked = cl_env_hops_noop, | |
621 | }; | |
622 | ||
623 | static inline struct cl_env *cl_env_fetch(void) | |
624 | { | |
625 | struct cl_env *cle; | |
626 | ||
627 | cle = cfs_hash_lookup(cl_env_hash, (void *) (long) current->pid); | |
628 | LASSERT(ergo(cle, cle->ce_magic == &cl_env_init0)); | |
629 | return cle; | |
630 | } | |
631 | ||
632 | static inline void cl_env_attach(struct cl_env *cle) | |
633 | { | |
634 | if (cle) { | |
635 | int rc; | |
636 | ||
637 | LASSERT(cle->ce_owner == NULL); | |
638 | cle->ce_owner = (void *) (long) current->pid; | |
639 | rc = cfs_hash_add_unique(cl_env_hash, cle->ce_owner, | |
640 | &cle->ce_node); | |
641 | LASSERT(rc == 0); | |
642 | } | |
643 | } | |
644 | ||
645 | static inline void cl_env_do_detach(struct cl_env *cle) | |
646 | { | |
647 | void *cookie; | |
648 | ||
649 | LASSERT(cle->ce_owner == (void *) (long) current->pid); | |
650 | cookie = cfs_hash_del(cl_env_hash, cle->ce_owner, | |
651 | &cle->ce_node); | |
652 | LASSERT(cookie == cle); | |
653 | cle->ce_owner = NULL; | |
654 | } | |
655 | ||
656 | static int cl_env_store_init(void) { | |
657 | cl_env_hash = cfs_hash_create("cl_env", | |
658 | HASH_CL_ENV_BITS, HASH_CL_ENV_BITS, | |
659 | HASH_CL_ENV_BKT_BITS, 0, | |
660 | CFS_HASH_MIN_THETA, | |
661 | CFS_HASH_MAX_THETA, | |
662 | &cl_env_hops, | |
663 | CFS_HASH_RW_BKTLOCK); | |
664 | return cl_env_hash != NULL ? 0 :-ENOMEM; | |
665 | } | |
666 | ||
667 | static void cl_env_store_fini(void) { | |
668 | cfs_hash_putref(cl_env_hash); | |
669 | } | |
670 | ||
671 | ||
672 | static inline struct cl_env *cl_env_detach(struct cl_env *cle) | |
673 | { | |
674 | if (cle == NULL) | |
675 | cle = cl_env_fetch(); | |
676 | ||
677 | if (cle && cle->ce_owner) | |
678 | cl_env_do_detach(cle); | |
679 | ||
680 | return cle; | |
681 | } | |
682 | ||
683 | static struct lu_env *cl_env_new(__u32 ctx_tags, __u32 ses_tags, void *debug) | |
684 | { | |
685 | struct lu_env *env; | |
686 | struct cl_env *cle; | |
687 | ||
688 | OBD_SLAB_ALLOC_PTR_GFP(cle, cl_env_kmem, __GFP_IO); | |
689 | if (cle != NULL) { | |
690 | int rc; | |
691 | ||
692 | INIT_LIST_HEAD(&cle->ce_linkage); | |
693 | cle->ce_magic = &cl_env_init0; | |
694 | env = &cle->ce_lu; | |
695 | rc = lu_env_init(env, LCT_CL_THREAD|ctx_tags); | |
696 | if (rc == 0) { | |
697 | rc = lu_context_init(&cle->ce_ses, | |
698 | LCT_SESSION | ses_tags); | |
699 | if (rc == 0) { | |
700 | lu_context_enter(&cle->ce_ses); | |
701 | env->le_ses = &cle->ce_ses; | |
702 | cl_env_init0(cle, debug); | |
703 | } else | |
704 | lu_env_fini(env); | |
705 | } | |
706 | if (rc != 0) { | |
707 | OBD_SLAB_FREE_PTR(cle, cl_env_kmem); | |
708 | env = ERR_PTR(rc); | |
709 | } else { | |
710 | CL_ENV_INC(create); | |
711 | CL_ENV_INC(total); | |
712 | } | |
713 | } else | |
714 | env = ERR_PTR(-ENOMEM); | |
715 | return env; | |
716 | } | |
717 | ||
718 | static void cl_env_fini(struct cl_env *cle) | |
719 | { | |
720 | CL_ENV_DEC(total); | |
721 | lu_context_fini(&cle->ce_lu.le_ctx); | |
722 | lu_context_fini(&cle->ce_ses); | |
723 | OBD_SLAB_FREE_PTR(cle, cl_env_kmem); | |
724 | } | |
725 | ||
726 | static inline struct cl_env *cl_env_container(struct lu_env *env) | |
727 | { | |
728 | return container_of(env, struct cl_env, ce_lu); | |
729 | } | |
730 | ||
731 | struct lu_env *cl_env_peek(int *refcheck) | |
732 | { | |
733 | struct lu_env *env; | |
734 | struct cl_env *cle; | |
735 | ||
736 | CL_ENV_INC(lookup); | |
737 | ||
738 | /* check that we don't go far from untrusted pointer */ | |
739 | CLASSERT(offsetof(struct cl_env, ce_magic) == 0); | |
740 | ||
741 | env = NULL; | |
742 | cle = cl_env_fetch(); | |
743 | if (cle != NULL) { | |
744 | CL_ENV_INC(hit); | |
745 | env = &cle->ce_lu; | |
746 | *refcheck = ++cle->ce_ref; | |
747 | } | |
748 | CDEBUG(D_OTHER, "%d@%p\n", cle ? cle->ce_ref : 0, cle); | |
749 | return env; | |
750 | } | |
751 | EXPORT_SYMBOL(cl_env_peek); | |
752 | ||
753 | /** | |
754 | * Returns lu_env: if there already is an environment associated with the | |
755 | * current thread, it is returned, otherwise, new environment is allocated. | |
756 | * | |
757 | * \param refcheck pointer to a counter used to detect environment leaks. In | |
758 | * the usual case cl_env_get() and cl_env_put() are called in the same lexical | |
759 | * scope and pointer to the same integer is passed as \a refcheck. This is | |
760 | * used to detect missed cl_env_put(). | |
761 | * | |
762 | * \see cl_env_put() | |
763 | */ | |
764 | struct lu_env *cl_env_get(int *refcheck) | |
765 | { | |
766 | struct lu_env *env; | |
767 | ||
768 | env = cl_env_peek(refcheck); | |
769 | if (env == NULL) { | |
770 | env = cl_env_new(lu_context_tags_default, | |
771 | lu_session_tags_default, | |
772 | __builtin_return_address(0)); | |
773 | ||
774 | if (!IS_ERR(env)) { | |
775 | struct cl_env *cle; | |
776 | ||
777 | cle = cl_env_container(env); | |
778 | cl_env_attach(cle); | |
779 | *refcheck = cle->ce_ref; | |
780 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
781 | } | |
782 | } | |
783 | return env; | |
784 | } | |
785 | EXPORT_SYMBOL(cl_env_get); | |
786 | ||
787 | /** | |
788 | * Forces an allocation of a fresh environment with given tags. | |
789 | * | |
790 | * \see cl_env_get() | |
791 | */ | |
792 | struct lu_env *cl_env_alloc(int *refcheck, __u32 tags) | |
793 | { | |
794 | struct lu_env *env; | |
795 | ||
796 | LASSERT(cl_env_peek(refcheck) == NULL); | |
797 | env = cl_env_new(tags, tags, __builtin_return_address(0)); | |
798 | if (!IS_ERR(env)) { | |
799 | struct cl_env *cle; | |
800 | ||
801 | cle = cl_env_container(env); | |
802 | *refcheck = cle->ce_ref; | |
803 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
804 | } | |
805 | return env; | |
806 | } | |
807 | EXPORT_SYMBOL(cl_env_alloc); | |
808 | ||
809 | static void cl_env_exit(struct cl_env *cle) | |
810 | { | |
811 | LASSERT(cle->ce_owner == NULL); | |
812 | lu_context_exit(&cle->ce_lu.le_ctx); | |
813 | lu_context_exit(&cle->ce_ses); | |
814 | } | |
815 | ||
816 | /** | |
817 | * Release an environment. | |
818 | * | |
819 | * Decrement \a env reference counter. When counter drops to 0, nothing in | |
820 | * this thread is using environment and it is returned to the allocation | |
821 | * cache, or freed straight away, if cache is large enough. | |
822 | */ | |
823 | void cl_env_put(struct lu_env *env, int *refcheck) | |
824 | { | |
825 | struct cl_env *cle; | |
826 | ||
827 | cle = cl_env_container(env); | |
828 | ||
829 | LASSERT(cle->ce_ref > 0); | |
830 | LASSERT(ergo(refcheck != NULL, cle->ce_ref == *refcheck)); | |
831 | ||
832 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
833 | if (--cle->ce_ref == 0) { | |
834 | CL_ENV_DEC(busy); | |
835 | cl_env_detach(cle); | |
836 | cle->ce_debug = NULL; | |
837 | cl_env_exit(cle); | |
838 | cl_env_fini(cle); | |
839 | } | |
840 | } | |
841 | EXPORT_SYMBOL(cl_env_put); | |
842 | ||
843 | /** | |
844 | * Declares a point of re-entrancy. | |
845 | * | |
846 | * \see cl_env_reexit() | |
847 | */ | |
848 | void *cl_env_reenter(void) | |
849 | { | |
850 | return cl_env_detach(NULL); | |
851 | } | |
852 | EXPORT_SYMBOL(cl_env_reenter); | |
853 | ||
854 | /** | |
855 | * Exits re-entrancy. | |
856 | */ | |
857 | void cl_env_reexit(void *cookie) | |
858 | { | |
859 | cl_env_detach(NULL); | |
860 | cl_env_attach(cookie); | |
861 | } | |
862 | EXPORT_SYMBOL(cl_env_reexit); | |
863 | ||
864 | /** | |
865 | * Setup user-supplied \a env as a current environment. This is to be used to | |
866 | * guaranteed that environment exists even when cl_env_get() fails. It is up | |
867 | * to user to ensure proper concurrency control. | |
868 | * | |
869 | * \see cl_env_unplant() | |
870 | */ | |
871 | void cl_env_implant(struct lu_env *env, int *refcheck) | |
872 | { | |
873 | struct cl_env *cle = cl_env_container(env); | |
874 | ||
875 | LASSERT(cle->ce_ref > 0); | |
876 | ||
877 | cl_env_attach(cle); | |
878 | cl_env_get(refcheck); | |
879 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
880 | } | |
881 | EXPORT_SYMBOL(cl_env_implant); | |
882 | ||
883 | /** | |
884 | * Detach environment installed earlier by cl_env_implant(). | |
885 | */ | |
886 | void cl_env_unplant(struct lu_env *env, int *refcheck) | |
887 | { | |
888 | struct cl_env *cle = cl_env_container(env); | |
889 | ||
890 | LASSERT(cle->ce_ref > 1); | |
891 | ||
892 | CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle); | |
893 | ||
894 | cl_env_detach(cle); | |
895 | cl_env_put(env, refcheck); | |
896 | } | |
897 | EXPORT_SYMBOL(cl_env_unplant); | |
898 | ||
899 | struct lu_env *cl_env_nested_get(struct cl_env_nest *nest) | |
900 | { | |
901 | struct lu_env *env; | |
902 | ||
903 | nest->cen_cookie = NULL; | |
904 | env = cl_env_peek(&nest->cen_refcheck); | |
905 | if (env != NULL) { | |
906 | if (!cl_io_is_going(env)) | |
907 | return env; | |
908 | else { | |
909 | cl_env_put(env, &nest->cen_refcheck); | |
910 | nest->cen_cookie = cl_env_reenter(); | |
911 | } | |
912 | } | |
913 | env = cl_env_get(&nest->cen_refcheck); | |
914 | if (IS_ERR(env)) { | |
915 | cl_env_reexit(nest->cen_cookie); | |
916 | return env; | |
917 | } | |
918 | ||
919 | LASSERT(!cl_io_is_going(env)); | |
920 | return env; | |
921 | } | |
922 | EXPORT_SYMBOL(cl_env_nested_get); | |
923 | ||
924 | void cl_env_nested_put(struct cl_env_nest *nest, struct lu_env *env) | |
925 | { | |
926 | cl_env_put(env, &nest->cen_refcheck); | |
927 | cl_env_reexit(nest->cen_cookie); | |
928 | } | |
929 | EXPORT_SYMBOL(cl_env_nested_put); | |
930 | ||
931 | /** | |
932 | * Converts struct cl_attr to struct ost_lvb. | |
933 | * | |
934 | * \see cl_lvb2attr | |
935 | */ | |
936 | void cl_attr2lvb(struct ost_lvb *lvb, const struct cl_attr *attr) | |
937 | { | |
d7e09d03 PT |
938 | lvb->lvb_size = attr->cat_size; |
939 | lvb->lvb_mtime = attr->cat_mtime; | |
940 | lvb->lvb_atime = attr->cat_atime; | |
941 | lvb->lvb_ctime = attr->cat_ctime; | |
942 | lvb->lvb_blocks = attr->cat_blocks; | |
943 | EXIT; | |
944 | } | |
945 | EXPORT_SYMBOL(cl_attr2lvb); | |
946 | ||
947 | /** | |
948 | * Converts struct ost_lvb to struct cl_attr. | |
949 | * | |
950 | * \see cl_attr2lvb | |
951 | */ | |
952 | void cl_lvb2attr(struct cl_attr *attr, const struct ost_lvb *lvb) | |
953 | { | |
d7e09d03 PT |
954 | attr->cat_size = lvb->lvb_size; |
955 | attr->cat_mtime = lvb->lvb_mtime; | |
956 | attr->cat_atime = lvb->lvb_atime; | |
957 | attr->cat_ctime = lvb->lvb_ctime; | |
958 | attr->cat_blocks = lvb->lvb_blocks; | |
959 | EXIT; | |
960 | } | |
961 | EXPORT_SYMBOL(cl_lvb2attr); | |
962 | ||
963 | /***************************************************************************** | |
964 | * | |
965 | * Temporary prototype thing: mirror obd-devices into cl devices. | |
966 | * | |
967 | */ | |
968 | ||
969 | struct cl_device *cl_type_setup(const struct lu_env *env, struct lu_site *site, | |
970 | struct lu_device_type *ldt, | |
971 | struct lu_device *next) | |
972 | { | |
973 | const char *typename; | |
974 | struct lu_device *d; | |
975 | ||
976 | LASSERT(ldt != NULL); | |
977 | ||
978 | typename = ldt->ldt_name; | |
979 | d = ldt->ldt_ops->ldto_device_alloc(env, ldt, NULL); | |
980 | if (!IS_ERR(d)) { | |
981 | int rc; | |
982 | ||
983 | if (site != NULL) | |
984 | d->ld_site = site; | |
985 | rc = ldt->ldt_ops->ldto_device_init(env, d, typename, next); | |
986 | if (rc == 0) { | |
987 | lu_device_get(d); | |
988 | lu_ref_add(&d->ld_reference, | |
989 | "lu-stack", &lu_site_init); | |
990 | } else { | |
991 | ldt->ldt_ops->ldto_device_free(env, d); | |
992 | CERROR("can't init device '%s', %d\n", typename, rc); | |
993 | d = ERR_PTR(rc); | |
994 | } | |
995 | } else | |
996 | CERROR("Cannot allocate device: '%s'\n", typename); | |
997 | return lu2cl_dev(d); | |
998 | } | |
999 | EXPORT_SYMBOL(cl_type_setup); | |
1000 | ||
1001 | /** | |
1002 | * Finalize device stack by calling lu_stack_fini(). | |
1003 | */ | |
1004 | void cl_stack_fini(const struct lu_env *env, struct cl_device *cl) | |
1005 | { | |
1006 | lu_stack_fini(env, cl2lu_dev(cl)); | |
1007 | } | |
1008 | EXPORT_SYMBOL(cl_stack_fini); | |
1009 | ||
1010 | int cl_lock_init(void); | |
1011 | void cl_lock_fini(void); | |
1012 | ||
1013 | int cl_page_init(void); | |
1014 | void cl_page_fini(void); | |
1015 | ||
1016 | static struct lu_context_key cl_key; | |
1017 | ||
1018 | struct cl_thread_info *cl_env_info(const struct lu_env *env) | |
1019 | { | |
1020 | return lu_context_key_get(&env->le_ctx, &cl_key); | |
1021 | } | |
1022 | ||
1023 | /* defines cl0_key_{init,fini}() */ | |
1024 | LU_KEY_INIT_FINI(cl0, struct cl_thread_info); | |
1025 | ||
1026 | static void *cl_key_init(const struct lu_context *ctx, | |
1027 | struct lu_context_key *key) | |
1028 | { | |
1029 | struct cl_thread_info *info; | |
1030 | ||
1031 | info = cl0_key_init(ctx, key); | |
1032 | if (!IS_ERR(info)) { | |
1033 | int i; | |
1034 | ||
1035 | for (i = 0; i < ARRAY_SIZE(info->clt_counters); ++i) | |
1036 | lu_ref_init(&info->clt_counters[i].ctc_locks_locked); | |
1037 | } | |
1038 | return info; | |
1039 | } | |
1040 | ||
1041 | static void cl_key_fini(const struct lu_context *ctx, | |
1042 | struct lu_context_key *key, void *data) | |
1043 | { | |
1044 | struct cl_thread_info *info; | |
1045 | int i; | |
1046 | ||
1047 | info = data; | |
1048 | for (i = 0; i < ARRAY_SIZE(info->clt_counters); ++i) | |
1049 | lu_ref_fini(&info->clt_counters[i].ctc_locks_locked); | |
1050 | cl0_key_fini(ctx, key, data); | |
1051 | } | |
1052 | ||
1053 | static void cl_key_exit(const struct lu_context *ctx, | |
1054 | struct lu_context_key *key, void *data) | |
1055 | { | |
1056 | struct cl_thread_info *info = data; | |
1057 | int i; | |
1058 | ||
1059 | for (i = 0; i < ARRAY_SIZE(info->clt_counters); ++i) { | |
1060 | LASSERT(info->clt_counters[i].ctc_nr_held == 0); | |
1061 | LASSERT(info->clt_counters[i].ctc_nr_used == 0); | |
1062 | LASSERT(info->clt_counters[i].ctc_nr_locks_acquired == 0); | |
1063 | LASSERT(info->clt_counters[i].ctc_nr_locks_locked == 0); | |
1064 | lu_ref_fini(&info->clt_counters[i].ctc_locks_locked); | |
1065 | lu_ref_init(&info->clt_counters[i].ctc_locks_locked); | |
1066 | } | |
1067 | } | |
1068 | ||
1069 | static struct lu_context_key cl_key = { | |
1070 | .lct_tags = LCT_CL_THREAD, | |
1071 | .lct_init = cl_key_init, | |
1072 | .lct_fini = cl_key_fini, | |
1073 | .lct_exit = cl_key_exit | |
1074 | }; | |
1075 | ||
1076 | static struct lu_kmem_descr cl_object_caches[] = { | |
1077 | { | |
1078 | .ckd_cache = &cl_env_kmem, | |
1079 | .ckd_name = "cl_env_kmem", | |
1080 | .ckd_size = sizeof (struct cl_env) | |
1081 | }, | |
1082 | { | |
1083 | .ckd_cache = NULL | |
1084 | } | |
1085 | }; | |
1086 | ||
1087 | /** | |
1088 | * Global initialization of cl-data. Create kmem caches, register | |
1089 | * lu_context_key's, etc. | |
1090 | * | |
1091 | * \see cl_global_fini() | |
1092 | */ | |
1093 | int cl_global_init(void) | |
1094 | { | |
1095 | int result; | |
1096 | ||
1097 | result = cl_env_store_init(); | |
1098 | if (result) | |
1099 | return result; | |
1100 | ||
1101 | result = lu_kmem_init(cl_object_caches); | |
1102 | if (result) | |
1103 | goto out_store; | |
1104 | ||
1105 | LU_CONTEXT_KEY_INIT(&cl_key); | |
1106 | result = lu_context_key_register(&cl_key); | |
1107 | if (result) | |
1108 | goto out_kmem; | |
1109 | ||
1110 | result = cl_lock_init(); | |
1111 | if (result) | |
1112 | goto out_context; | |
1113 | ||
1114 | result = cl_page_init(); | |
1115 | if (result) | |
1116 | goto out_lock; | |
1117 | ||
1118 | return 0; | |
1119 | out_lock: | |
1120 | cl_lock_fini(); | |
1121 | out_context: | |
1122 | lu_context_key_degister(&cl_key); | |
1123 | out_kmem: | |
1124 | lu_kmem_fini(cl_object_caches); | |
1125 | out_store: | |
1126 | cl_env_store_fini(); | |
1127 | return result; | |
1128 | } | |
1129 | ||
1130 | /** | |
1131 | * Finalization of global cl-data. Dual to cl_global_init(). | |
1132 | */ | |
1133 | void cl_global_fini(void) | |
1134 | { | |
1135 | cl_lock_fini(); | |
1136 | cl_page_fini(); | |
1137 | lu_context_key_degister(&cl_key); | |
1138 | lu_kmem_fini(cl_object_caches); | |
1139 | cl_env_store_fini(); | |
1140 | } |