operation.c 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530
  1. /* FS-Cache worker operation management routines
  2. *
  3. * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
  4. * Written by David Howells (dhowells@redhat.com)
  5. *
  6. * This program is free software; you can redistribute it and/or
  7. * modify it under the terms of the GNU General Public License
  8. * as published by the Free Software Foundation; either version
  9. * 2 of the License, or (at your option) any later version.
  10. *
  11. * See Documentation/filesystems/caching/operations.txt
  12. */
  13. #define FSCACHE_DEBUG_LEVEL OPERATION
  14. #include <linux/module.h>
  15. #include <linux/seq_file.h>
  16. #include <linux/slab.h>
  17. #include "internal.h"
  18. atomic_t fscache_op_debug_id;
  19. EXPORT_SYMBOL(fscache_op_debug_id);
  20. /**
  21. * fscache_enqueue_operation - Enqueue an operation for processing
  22. * @op: The operation to enqueue
  23. *
  24. * Enqueue an operation for processing by the FS-Cache thread pool.
  25. *
  26. * This will get its own ref on the object.
  27. */
  28. void fscache_enqueue_operation(struct fscache_operation *op)
  29. {
  30. _enter("{OBJ%x OP%x,%u}",
  31. op->object->debug_id, op->debug_id, atomic_read(&op->usage));
  32. ASSERT(list_empty(&op->pend_link));
  33. ASSERT(op->processor != NULL);
  34. ASSERT(fscache_object_is_available(op->object));
  35. ASSERTCMP(atomic_read(&op->usage), >, 0);
  36. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_IN_PROGRESS);
  37. fscache_stat(&fscache_n_op_enqueue);
  38. switch (op->flags & FSCACHE_OP_TYPE) {
  39. case FSCACHE_OP_ASYNC:
  40. _debug("queue async");
  41. atomic_inc(&op->usage);
  42. if (!queue_work(fscache_op_wq, &op->work))
  43. fscache_put_operation(op);
  44. break;
  45. case FSCACHE_OP_MYTHREAD:
  46. _debug("queue for caller's attention");
  47. break;
  48. default:
  49. printk(KERN_ERR "FS-Cache: Unexpected op type %lx",
  50. op->flags);
  51. BUG();
  52. break;
  53. }
  54. }
  55. EXPORT_SYMBOL(fscache_enqueue_operation);
  56. /*
  57. * start an op running
  58. */
  59. static void fscache_run_op(struct fscache_object *object,
  60. struct fscache_operation *op)
  61. {
  62. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_PENDING);
  63. op->state = FSCACHE_OP_ST_IN_PROGRESS;
  64. object->n_in_progress++;
  65. if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
  66. wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
  67. if (op->processor)
  68. fscache_enqueue_operation(op);
  69. fscache_stat(&fscache_n_op_run);
  70. }
  71. /*
  72. * submit an exclusive operation for an object
  73. * - other ops are excluded from running simultaneously with this one
  74. * - this gets any extra refs it needs on an op
  75. */
  76. int fscache_submit_exclusive_op(struct fscache_object *object,
  77. struct fscache_operation *op)
  78. {
  79. int ret;
  80. _enter("{OBJ%x OP%x},", object->debug_id, op->debug_id);
  81. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_INITIALISED);
  82. ASSERTCMP(atomic_read(&op->usage), >, 0);
  83. spin_lock(&object->lock);
  84. ASSERTCMP(object->n_ops, >=, object->n_in_progress);
  85. ASSERTCMP(object->n_ops, >=, object->n_exclusive);
  86. ASSERT(list_empty(&op->pend_link));
  87. op->state = FSCACHE_OP_ST_PENDING;
  88. if (fscache_object_is_active(object)) {
  89. op->object = object;
  90. object->n_ops++;
  91. object->n_exclusive++; /* reads and writes must wait */
  92. if (object->n_in_progress > 0) {
  93. atomic_inc(&op->usage);
  94. list_add_tail(&op->pend_link, &object->pending_ops);
  95. fscache_stat(&fscache_n_op_pend);
  96. } else if (!list_empty(&object->pending_ops)) {
  97. atomic_inc(&op->usage);
  98. list_add_tail(&op->pend_link, &object->pending_ops);
  99. fscache_stat(&fscache_n_op_pend);
  100. fscache_start_operations(object);
  101. } else {
  102. ASSERTCMP(object->n_in_progress, ==, 0);
  103. fscache_run_op(object, op);
  104. }
  105. /* need to issue a new write op after this */
  106. clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
  107. ret = 0;
  108. } else if (test_bit(FSCACHE_OBJECT_IS_LOOKED_UP, &object->flags)) {
  109. op->object = object;
  110. object->n_ops++;
  111. object->n_exclusive++; /* reads and writes must wait */
  112. atomic_inc(&op->usage);
  113. list_add_tail(&op->pend_link, &object->pending_ops);
  114. fscache_stat(&fscache_n_op_pend);
  115. ret = 0;
  116. } else {
  117. /* If we're in any other state, there must have been an I/O
  118. * error of some nature.
  119. */
  120. ASSERT(test_bit(FSCACHE_IOERROR, &object->cache->flags));
  121. ret = -EIO;
  122. }
  123. spin_unlock(&object->lock);
  124. return ret;
  125. }
  126. /*
  127. * report an unexpected submission
  128. */
  129. static void fscache_report_unexpected_submission(struct fscache_object *object,
  130. struct fscache_operation *op,
  131. const struct fscache_state *ostate)
  132. {
  133. static bool once_only;
  134. struct fscache_operation *p;
  135. unsigned n;
  136. if (once_only)
  137. return;
  138. once_only = true;
  139. kdebug("unexpected submission OP%x [OBJ%x %s]",
  140. op->debug_id, object->debug_id, object->state->name);
  141. kdebug("objstate=%s [%s]", object->state->name, ostate->name);
  142. kdebug("objflags=%lx", object->flags);
  143. kdebug("objevent=%lx [%lx]", object->events, object->event_mask);
  144. kdebug("ops=%u inp=%u exc=%u",
  145. object->n_ops, object->n_in_progress, object->n_exclusive);
  146. if (!list_empty(&object->pending_ops)) {
  147. n = 0;
  148. list_for_each_entry(p, &object->pending_ops, pend_link) {
  149. ASSERTCMP(p->object, ==, object);
  150. kdebug("%p %p", op->processor, op->release);
  151. n++;
  152. }
  153. kdebug("n=%u", n);
  154. }
  155. dump_stack();
  156. }
  157. /*
  158. * submit an operation for an object
  159. * - objects may be submitted only in the following states:
  160. * - during object creation (write ops may be submitted)
  161. * - whilst the object is active
  162. * - after an I/O error incurred in one of the two above states (op rejected)
  163. * - this gets any extra refs it needs on an op
  164. */
  165. int fscache_submit_op(struct fscache_object *object,
  166. struct fscache_operation *op)
  167. {
  168. const struct fscache_state *ostate;
  169. int ret;
  170. _enter("{OBJ%x OP%x},{%u}",
  171. object->debug_id, op->debug_id, atomic_read(&op->usage));
  172. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_INITIALISED);
  173. ASSERTCMP(atomic_read(&op->usage), >, 0);
  174. spin_lock(&object->lock);
  175. ASSERTCMP(object->n_ops, >=, object->n_in_progress);
  176. ASSERTCMP(object->n_ops, >=, object->n_exclusive);
  177. ASSERT(list_empty(&op->pend_link));
  178. ostate = object->state;
  179. smp_rmb();
  180. op->state = FSCACHE_OP_ST_PENDING;
  181. if (fscache_object_is_active(object)) {
  182. op->object = object;
  183. object->n_ops++;
  184. if (object->n_exclusive > 0) {
  185. atomic_inc(&op->usage);
  186. list_add_tail(&op->pend_link, &object->pending_ops);
  187. fscache_stat(&fscache_n_op_pend);
  188. } else if (!list_empty(&object->pending_ops)) {
  189. atomic_inc(&op->usage);
  190. list_add_tail(&op->pend_link, &object->pending_ops);
  191. fscache_stat(&fscache_n_op_pend);
  192. fscache_start_operations(object);
  193. } else {
  194. ASSERTCMP(object->n_exclusive, ==, 0);
  195. fscache_run_op(object, op);
  196. }
  197. ret = 0;
  198. } else if (test_bit(FSCACHE_OBJECT_IS_LOOKED_UP, &object->flags)) {
  199. op->object = object;
  200. object->n_ops++;
  201. atomic_inc(&op->usage);
  202. list_add_tail(&op->pend_link, &object->pending_ops);
  203. fscache_stat(&fscache_n_op_pend);
  204. ret = 0;
  205. } else if (fscache_object_is_dying(object)) {
  206. fscache_stat(&fscache_n_op_rejected);
  207. op->state = FSCACHE_OP_ST_CANCELLED;
  208. ret = -ENOBUFS;
  209. } else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
  210. fscache_report_unexpected_submission(object, op, ostate);
  211. ASSERT(!fscache_object_is_active(object));
  212. op->state = FSCACHE_OP_ST_CANCELLED;
  213. ret = -ENOBUFS;
  214. } else {
  215. op->state = FSCACHE_OP_ST_CANCELLED;
  216. ret = -ENOBUFS;
  217. }
  218. spin_unlock(&object->lock);
  219. return ret;
  220. }
  221. /*
  222. * queue an object for withdrawal on error, aborting all following asynchronous
  223. * operations
  224. */
  225. void fscache_abort_object(struct fscache_object *object)
  226. {
  227. _enter("{OBJ%x}", object->debug_id);
  228. fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
  229. }
  230. /*
  231. * Jump start the operation processing on an object. The caller must hold
  232. * object->lock.
  233. */
  234. void fscache_start_operations(struct fscache_object *object)
  235. {
  236. struct fscache_operation *op;
  237. bool stop = false;
  238. while (!list_empty(&object->pending_ops) && !stop) {
  239. op = list_entry(object->pending_ops.next,
  240. struct fscache_operation, pend_link);
  241. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
  242. if (object->n_in_progress > 0)
  243. break;
  244. stop = true;
  245. }
  246. list_del_init(&op->pend_link);
  247. fscache_run_op(object, op);
  248. /* the pending queue was holding a ref on the object */
  249. fscache_put_operation(op);
  250. }
  251. ASSERTCMP(object->n_in_progress, <=, object->n_ops);
  252. _debug("woke %d ops on OBJ%x",
  253. object->n_in_progress, object->debug_id);
  254. }
  255. /*
  256. * cancel an operation that's pending on an object
  257. */
  258. int fscache_cancel_op(struct fscache_operation *op,
  259. void (*do_cancel)(struct fscache_operation *))
  260. {
  261. struct fscache_object *object = op->object;
  262. int ret;
  263. _enter("OBJ%x OP%x}", op->object->debug_id, op->debug_id);
  264. ASSERTCMP(op->state, >=, FSCACHE_OP_ST_PENDING);
  265. ASSERTCMP(op->state, !=, FSCACHE_OP_ST_CANCELLED);
  266. ASSERTCMP(atomic_read(&op->usage), >, 0);
  267. spin_lock(&object->lock);
  268. ret = -EBUSY;
  269. if (op->state == FSCACHE_OP_ST_PENDING) {
  270. ASSERT(!list_empty(&op->pend_link));
  271. fscache_stat(&fscache_n_op_cancelled);
  272. list_del_init(&op->pend_link);
  273. if (do_cancel)
  274. do_cancel(op);
  275. op->state = FSCACHE_OP_ST_CANCELLED;
  276. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
  277. object->n_exclusive--;
  278. if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
  279. wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
  280. fscache_put_operation(op);
  281. ret = 0;
  282. }
  283. spin_unlock(&object->lock);
  284. _leave(" = %d", ret);
  285. return ret;
  286. }
  287. /*
  288. * Cancel all pending operations on an object
  289. */
  290. void fscache_cancel_all_ops(struct fscache_object *object)
  291. {
  292. struct fscache_operation *op;
  293. _enter("OBJ%x", object->debug_id);
  294. spin_lock(&object->lock);
  295. while (!list_empty(&object->pending_ops)) {
  296. op = list_entry(object->pending_ops.next,
  297. struct fscache_operation, pend_link);
  298. fscache_stat(&fscache_n_op_cancelled);
  299. list_del_init(&op->pend_link);
  300. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_PENDING);
  301. op->state = FSCACHE_OP_ST_CANCELLED;
  302. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
  303. object->n_exclusive--;
  304. if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
  305. wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
  306. fscache_put_operation(op);
  307. cond_resched_lock(&object->lock);
  308. }
  309. spin_unlock(&object->lock);
  310. _leave("");
  311. }
  312. /*
  313. * Record the completion or cancellation of an in-progress operation.
  314. */
  315. void fscache_op_complete(struct fscache_operation *op, bool cancelled)
  316. {
  317. struct fscache_object *object = op->object;
  318. _enter("OBJ%x", object->debug_id);
  319. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_IN_PROGRESS);
  320. ASSERTCMP(object->n_in_progress, >, 0);
  321. ASSERTIFCMP(test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags),
  322. object->n_exclusive, >, 0);
  323. ASSERTIFCMP(test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags),
  324. object->n_in_progress, ==, 1);
  325. spin_lock(&object->lock);
  326. op->state = cancelled ?
  327. FSCACHE_OP_ST_CANCELLED : FSCACHE_OP_ST_COMPLETE;
  328. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
  329. object->n_exclusive--;
  330. object->n_in_progress--;
  331. if (object->n_in_progress == 0)
  332. fscache_start_operations(object);
  333. spin_unlock(&object->lock);
  334. _leave("");
  335. }
  336. EXPORT_SYMBOL(fscache_op_complete);
  337. /*
  338. * release an operation
  339. * - queues pending ops if this is the last in-progress op
  340. */
  341. void fscache_put_operation(struct fscache_operation *op)
  342. {
  343. struct fscache_object *object;
  344. struct fscache_cache *cache;
  345. _enter("{OBJ%x OP%x,%d}",
  346. op->object->debug_id, op->debug_id, atomic_read(&op->usage));
  347. ASSERTCMP(atomic_read(&op->usage), >, 0);
  348. if (!atomic_dec_and_test(&op->usage))
  349. return;
  350. _debug("PUT OP");
  351. ASSERTIFCMP(op->state != FSCACHE_OP_ST_COMPLETE,
  352. op->state, ==, FSCACHE_OP_ST_CANCELLED);
  353. op->state = FSCACHE_OP_ST_DEAD;
  354. fscache_stat(&fscache_n_op_release);
  355. if (op->release) {
  356. op->release(op);
  357. op->release = NULL;
  358. }
  359. object = op->object;
  360. if (test_bit(FSCACHE_OP_DEC_READ_CNT, &op->flags))
  361. atomic_dec(&object->n_reads);
  362. if (test_bit(FSCACHE_OP_UNUSE_COOKIE, &op->flags))
  363. fscache_unuse_cookie(object);
  364. /* now... we may get called with the object spinlock held, so we
  365. * complete the cleanup here only if we can immediately acquire the
  366. * lock, and defer it otherwise */
  367. if (!spin_trylock(&object->lock)) {
  368. _debug("defer put");
  369. fscache_stat(&fscache_n_op_deferred_release);
  370. cache = object->cache;
  371. spin_lock(&cache->op_gc_list_lock);
  372. list_add_tail(&op->pend_link, &cache->op_gc_list);
  373. spin_unlock(&cache->op_gc_list_lock);
  374. schedule_work(&cache->op_gc);
  375. _leave(" [defer]");
  376. return;
  377. }
  378. ASSERTCMP(object->n_ops, >, 0);
  379. object->n_ops--;
  380. if (object->n_ops == 0)
  381. fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
  382. spin_unlock(&object->lock);
  383. kfree(op);
  384. _leave(" [done]");
  385. }
  386. EXPORT_SYMBOL(fscache_put_operation);
  387. /*
  388. * garbage collect operations that have had their release deferred
  389. */
  390. void fscache_operation_gc(struct work_struct *work)
  391. {
  392. struct fscache_operation *op;
  393. struct fscache_object *object;
  394. struct fscache_cache *cache =
  395. container_of(work, struct fscache_cache, op_gc);
  396. int count = 0;
  397. _enter("");
  398. do {
  399. spin_lock(&cache->op_gc_list_lock);
  400. if (list_empty(&cache->op_gc_list)) {
  401. spin_unlock(&cache->op_gc_list_lock);
  402. break;
  403. }
  404. op = list_entry(cache->op_gc_list.next,
  405. struct fscache_operation, pend_link);
  406. list_del(&op->pend_link);
  407. spin_unlock(&cache->op_gc_list_lock);
  408. object = op->object;
  409. spin_lock(&object->lock);
  410. _debug("GC DEFERRED REL OBJ%x OP%x",
  411. object->debug_id, op->debug_id);
  412. fscache_stat(&fscache_n_op_gc);
  413. ASSERTCMP(atomic_read(&op->usage), ==, 0);
  414. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_DEAD);
  415. ASSERTCMP(object->n_ops, >, 0);
  416. object->n_ops--;
  417. if (object->n_ops == 0)
  418. fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
  419. spin_unlock(&object->lock);
  420. kfree(op);
  421. } while (count++ < 20);
  422. if (!list_empty(&cache->op_gc_list))
  423. schedule_work(&cache->op_gc);
  424. _leave("");
  425. }
  426. /*
  427. * execute an operation using fs_op_wq to provide processing context -
  428. * the caller holds a ref to this object, so we don't need to hold one
  429. */
  430. void fscache_op_work_func(struct work_struct *work)
  431. {
  432. struct fscache_operation *op =
  433. container_of(work, struct fscache_operation, work);
  434. unsigned long start;
  435. _enter("{OBJ%x OP%x,%d}",
  436. op->object->debug_id, op->debug_id, atomic_read(&op->usage));
  437. ASSERT(op->processor != NULL);
  438. start = jiffies;
  439. op->processor(op);
  440. fscache_hist(fscache_ops_histogram, start);
  441. fscache_put_operation(op);
  442. _leave("");
  443. }