operation.c 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535
  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. ASSERTCMP(op->object->state, >=, FSCACHE_OBJECT_AVAILABLE);
  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 (object->state == FSCACHE_OBJECT_CREATING) {
  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. unsigned long 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,
  141. fscache_object_states[object->state]);
  142. kdebug("objstate=%s [%s]",
  143. fscache_object_states[object->state],
  144. fscache_object_states[ostate]);
  145. kdebug("objflags=%lx", object->flags);
  146. kdebug("objevent=%lx [%lx]", object->events, object->event_mask);
  147. kdebug("ops=%u inp=%u exc=%u",
  148. object->n_ops, object->n_in_progress, object->n_exclusive);
  149. if (!list_empty(&object->pending_ops)) {
  150. n = 0;
  151. list_for_each_entry(p, &object->pending_ops, pend_link) {
  152. ASSERTCMP(p->object, ==, object);
  153. kdebug("%p %p", op->processor, op->release);
  154. n++;
  155. }
  156. kdebug("n=%u", n);
  157. }
  158. dump_stack();
  159. }
  160. /*
  161. * submit an operation for an object
  162. * - objects may be submitted only in the following states:
  163. * - during object creation (write ops may be submitted)
  164. * - whilst the object is active
  165. * - after an I/O error incurred in one of the two above states (op rejected)
  166. * - this gets any extra refs it needs on an op
  167. */
  168. int fscache_submit_op(struct fscache_object *object,
  169. struct fscache_operation *op)
  170. {
  171. unsigned long ostate;
  172. int ret;
  173. _enter("{OBJ%x OP%x},{%u}",
  174. object->debug_id, op->debug_id, atomic_read(&op->usage));
  175. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_INITIALISED);
  176. ASSERTCMP(atomic_read(&op->usage), >, 0);
  177. spin_lock(&object->lock);
  178. ASSERTCMP(object->n_ops, >=, object->n_in_progress);
  179. ASSERTCMP(object->n_ops, >=, object->n_exclusive);
  180. ASSERT(list_empty(&op->pend_link));
  181. ostate = object->state;
  182. smp_rmb();
  183. op->state = FSCACHE_OP_ST_PENDING;
  184. if (fscache_object_is_active(object)) {
  185. op->object = object;
  186. object->n_ops++;
  187. if (object->n_exclusive > 0) {
  188. atomic_inc(&op->usage);
  189. list_add_tail(&op->pend_link, &object->pending_ops);
  190. fscache_stat(&fscache_n_op_pend);
  191. } else if (!list_empty(&object->pending_ops)) {
  192. atomic_inc(&op->usage);
  193. list_add_tail(&op->pend_link, &object->pending_ops);
  194. fscache_stat(&fscache_n_op_pend);
  195. fscache_start_operations(object);
  196. } else {
  197. ASSERTCMP(object->n_exclusive, ==, 0);
  198. fscache_run_op(object, op);
  199. }
  200. ret = 0;
  201. } else if (object->state == FSCACHE_OBJECT_CREATING) {
  202. op->object = object;
  203. object->n_ops++;
  204. atomic_inc(&op->usage);
  205. list_add_tail(&op->pend_link, &object->pending_ops);
  206. fscache_stat(&fscache_n_op_pend);
  207. ret = 0;
  208. } else if (object->state == FSCACHE_OBJECT_DYING ||
  209. object->state == FSCACHE_OBJECT_LC_DYING ||
  210. object->state == FSCACHE_OBJECT_WITHDRAWING) {
  211. fscache_stat(&fscache_n_op_rejected);
  212. op->state = FSCACHE_OP_ST_CANCELLED;
  213. ret = -ENOBUFS;
  214. } else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
  215. fscache_report_unexpected_submission(object, op, ostate);
  216. ASSERT(!fscache_object_is_active(object));
  217. op->state = FSCACHE_OP_ST_CANCELLED;
  218. ret = -ENOBUFS;
  219. } else {
  220. op->state = FSCACHE_OP_ST_CANCELLED;
  221. ret = -ENOBUFS;
  222. }
  223. spin_unlock(&object->lock);
  224. return ret;
  225. }
  226. /*
  227. * queue an object for withdrawal on error, aborting all following asynchronous
  228. * operations
  229. */
  230. void fscache_abort_object(struct fscache_object *object)
  231. {
  232. _enter("{OBJ%x}", object->debug_id);
  233. fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
  234. }
  235. /*
  236. * jump start the operation processing on an object
  237. * - caller must hold object->lock
  238. */
  239. void fscache_start_operations(struct fscache_object *object)
  240. {
  241. struct fscache_operation *op;
  242. bool stop = false;
  243. while (!list_empty(&object->pending_ops) && !stop) {
  244. op = list_entry(object->pending_ops.next,
  245. struct fscache_operation, pend_link);
  246. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
  247. if (object->n_in_progress > 0)
  248. break;
  249. stop = true;
  250. }
  251. list_del_init(&op->pend_link);
  252. fscache_run_op(object, op);
  253. /* the pending queue was holding a ref on the object */
  254. fscache_put_operation(op);
  255. }
  256. ASSERTCMP(object->n_in_progress, <=, object->n_ops);
  257. _debug("woke %d ops on OBJ%x",
  258. object->n_in_progress, object->debug_id);
  259. }
  260. /*
  261. * cancel an operation that's pending on an object
  262. */
  263. int fscache_cancel_op(struct fscache_operation *op)
  264. {
  265. struct fscache_object *object = op->object;
  266. int ret;
  267. _enter("OBJ%x OP%x}", op->object->debug_id, op->debug_id);
  268. ASSERTCMP(op->state, >=, FSCACHE_OP_ST_PENDING);
  269. ASSERTCMP(op->state, !=, FSCACHE_OP_ST_CANCELLED);
  270. ASSERTCMP(atomic_read(&op->usage), >, 0);
  271. spin_lock(&object->lock);
  272. ret = -EBUSY;
  273. if (op->state == FSCACHE_OP_ST_PENDING) {
  274. ASSERT(!list_empty(&op->pend_link));
  275. fscache_stat(&fscache_n_op_cancelled);
  276. list_del_init(&op->pend_link);
  277. op->state = FSCACHE_OP_ST_CANCELLED;
  278. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
  279. object->n_exclusive--;
  280. if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
  281. wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
  282. fscache_put_operation(op);
  283. ret = 0;
  284. }
  285. spin_unlock(&object->lock);
  286. _leave(" = %d", ret);
  287. return ret;
  288. }
  289. /*
  290. * Cancel all pending operations on an object
  291. */
  292. void fscache_cancel_all_ops(struct fscache_object *object)
  293. {
  294. struct fscache_operation *op;
  295. _enter("OBJ%x", object->debug_id);
  296. spin_lock(&object->lock);
  297. while (!list_empty(&object->pending_ops)) {
  298. op = list_entry(object->pending_ops.next,
  299. struct fscache_operation, pend_link);
  300. fscache_stat(&fscache_n_op_cancelled);
  301. list_del_init(&op->pend_link);
  302. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_PENDING);
  303. op->state = FSCACHE_OP_ST_CANCELLED;
  304. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
  305. object->n_exclusive--;
  306. if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
  307. wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
  308. fscache_put_operation(op);
  309. cond_resched_lock(&object->lock);
  310. }
  311. spin_unlock(&object->lock);
  312. _leave("");
  313. }
  314. /*
  315. * Record the completion of an in-progress operation.
  316. */
  317. void fscache_op_complete(struct fscache_operation *op)
  318. {
  319. struct fscache_object *object = op->object;
  320. _enter("OBJ%x", object->debug_id);
  321. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_IN_PROGRESS);
  322. ASSERTCMP(object->n_in_progress, >, 0);
  323. ASSERTIFCMP(test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags),
  324. object->n_exclusive, >, 0);
  325. ASSERTIFCMP(test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags),
  326. object->n_in_progress, ==, 1);
  327. spin_lock(&object->lock);
  328. op->state = FSCACHE_OP_ST_COMPLETE;
  329. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
  330. object->n_exclusive--;
  331. object->n_in_progress--;
  332. if (object->n_in_progress == 0)
  333. fscache_start_operations(object);
  334. spin_unlock(&object->lock);
  335. _leave("");
  336. }
  337. EXPORT_SYMBOL(fscache_op_complete);
  338. /*
  339. * release an operation
  340. * - queues pending ops if this is the last in-progress op
  341. */
  342. void fscache_put_operation(struct fscache_operation *op)
  343. {
  344. struct fscache_object *object;
  345. struct fscache_cache *cache;
  346. _enter("{OBJ%x OP%x,%d}",
  347. op->object->debug_id, op->debug_id, atomic_read(&op->usage));
  348. ASSERTCMP(atomic_read(&op->usage), >, 0);
  349. if (!atomic_dec_and_test(&op->usage))
  350. return;
  351. _debug("PUT OP");
  352. ASSERTIFCMP(op->state != FSCACHE_OP_ST_COMPLETE,
  353. op->state, ==, FSCACHE_OP_ST_CANCELLED);
  354. op->state = FSCACHE_OP_ST_DEAD;
  355. fscache_stat(&fscache_n_op_release);
  356. if (op->release) {
  357. op->release(op);
  358. op->release = NULL;
  359. }
  360. object = op->object;
  361. if (test_bit(FSCACHE_OP_DEC_READ_CNT, &op->flags)) {
  362. if (atomic_dec_and_test(&object->n_reads)) {
  363. clear_bit(FSCACHE_COOKIE_WAITING_ON_READS,
  364. &object->cookie->flags);
  365. wake_up_bit(&object->cookie->flags,
  366. FSCACHE_COOKIE_WAITING_ON_READS);
  367. }
  368. }
  369. /* now... we may get called with the object spinlock held, so we
  370. * complete the cleanup here only if we can immediately acquire the
  371. * lock, and defer it otherwise */
  372. if (!spin_trylock(&object->lock)) {
  373. _debug("defer put");
  374. fscache_stat(&fscache_n_op_deferred_release);
  375. cache = object->cache;
  376. spin_lock(&cache->op_gc_list_lock);
  377. list_add_tail(&op->pend_link, &cache->op_gc_list);
  378. spin_unlock(&cache->op_gc_list_lock);
  379. schedule_work(&cache->op_gc);
  380. _leave(" [defer]");
  381. return;
  382. }
  383. ASSERTCMP(object->n_ops, >, 0);
  384. object->n_ops--;
  385. if (object->n_ops == 0)
  386. fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
  387. spin_unlock(&object->lock);
  388. kfree(op);
  389. _leave(" [done]");
  390. }
  391. EXPORT_SYMBOL(fscache_put_operation);
  392. /*
  393. * garbage collect operations that have had their release deferred
  394. */
  395. void fscache_operation_gc(struct work_struct *work)
  396. {
  397. struct fscache_operation *op;
  398. struct fscache_object *object;
  399. struct fscache_cache *cache =
  400. container_of(work, struct fscache_cache, op_gc);
  401. int count = 0;
  402. _enter("");
  403. do {
  404. spin_lock(&cache->op_gc_list_lock);
  405. if (list_empty(&cache->op_gc_list)) {
  406. spin_unlock(&cache->op_gc_list_lock);
  407. break;
  408. }
  409. op = list_entry(cache->op_gc_list.next,
  410. struct fscache_operation, pend_link);
  411. list_del(&op->pend_link);
  412. spin_unlock(&cache->op_gc_list_lock);
  413. object = op->object;
  414. spin_lock(&object->lock);
  415. _debug("GC DEFERRED REL OBJ%x OP%x",
  416. object->debug_id, op->debug_id);
  417. fscache_stat(&fscache_n_op_gc);
  418. ASSERTCMP(atomic_read(&op->usage), ==, 0);
  419. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_DEAD);
  420. ASSERTCMP(object->n_ops, >, 0);
  421. object->n_ops--;
  422. if (object->n_ops == 0)
  423. fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
  424. spin_unlock(&object->lock);
  425. kfree(op);
  426. } while (count++ < 20);
  427. if (!list_empty(&cache->op_gc_list))
  428. schedule_work(&cache->op_gc);
  429. _leave("");
  430. }
  431. /*
  432. * execute an operation using fs_op_wq to provide processing context -
  433. * the caller holds a ref to this object, so we don't need to hold one
  434. */
  435. void fscache_op_work_func(struct work_struct *work)
  436. {
  437. struct fscache_operation *op =
  438. container_of(work, struct fscache_operation, work);
  439. unsigned long start;
  440. _enter("{OBJ%x OP%x,%d}",
  441. op->object->debug_id, op->debug_id, atomic_read(&op->usage));
  442. ASSERT(op->processor != NULL);
  443. start = jiffies;
  444. op->processor(op);
  445. fscache_hist(fscache_ops_histogram, start);
  446. fscache_put_operation(op);
  447. _leave("");
  448. }