operation.c 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491
  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. _enter("{OBJ%x OP%x},", object->debug_id, op->debug_id);
  80. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_INITIALISED);
  81. ASSERTCMP(atomic_read(&op->usage), >, 0);
  82. spin_lock(&object->lock);
  83. ASSERTCMP(object->n_ops, >=, object->n_in_progress);
  84. ASSERTCMP(object->n_ops, >=, object->n_exclusive);
  85. ASSERT(list_empty(&op->pend_link));
  86. op->state = FSCACHE_OP_ST_PENDING;
  87. if (fscache_object_is_active(object)) {
  88. op->object = object;
  89. object->n_ops++;
  90. object->n_exclusive++; /* reads and writes must wait */
  91. if (object->n_in_progress > 0) {
  92. atomic_inc(&op->usage);
  93. list_add_tail(&op->pend_link, &object->pending_ops);
  94. fscache_stat(&fscache_n_op_pend);
  95. } else if (!list_empty(&object->pending_ops)) {
  96. atomic_inc(&op->usage);
  97. list_add_tail(&op->pend_link, &object->pending_ops);
  98. fscache_stat(&fscache_n_op_pend);
  99. fscache_start_operations(object);
  100. } else {
  101. ASSERTCMP(object->n_in_progress, ==, 0);
  102. fscache_run_op(object, op);
  103. }
  104. /* need to issue a new write op after this */
  105. clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
  106. } else if (object->state == FSCACHE_OBJECT_CREATING) {
  107. op->object = object;
  108. object->n_ops++;
  109. object->n_exclusive++; /* reads and writes must wait */
  110. atomic_inc(&op->usage);
  111. list_add_tail(&op->pend_link, &object->pending_ops);
  112. fscache_stat(&fscache_n_op_pend);
  113. } else {
  114. /* not allowed to submit ops in any other state */
  115. BUG();
  116. }
  117. spin_unlock(&object->lock);
  118. return 0;
  119. }
  120. /*
  121. * report an unexpected submission
  122. */
  123. static void fscache_report_unexpected_submission(struct fscache_object *object,
  124. struct fscache_operation *op,
  125. unsigned long ostate)
  126. {
  127. static bool once_only;
  128. struct fscache_operation *p;
  129. unsigned n;
  130. if (once_only)
  131. return;
  132. once_only = true;
  133. kdebug("unexpected submission OP%x [OBJ%x %s]",
  134. op->debug_id, object->debug_id,
  135. fscache_object_states[object->state]);
  136. kdebug("objstate=%s [%s]",
  137. fscache_object_states[object->state],
  138. fscache_object_states[ostate]);
  139. kdebug("objflags=%lx", object->flags);
  140. kdebug("objevent=%lx [%lx]", object->events, object->event_mask);
  141. kdebug("ops=%u inp=%u exc=%u",
  142. object->n_ops, object->n_in_progress, object->n_exclusive);
  143. if (!list_empty(&object->pending_ops)) {
  144. n = 0;
  145. list_for_each_entry(p, &object->pending_ops, pend_link) {
  146. ASSERTCMP(p->object, ==, object);
  147. kdebug("%p %p", op->processor, op->release);
  148. n++;
  149. }
  150. kdebug("n=%u", n);
  151. }
  152. dump_stack();
  153. }
  154. /*
  155. * submit an operation for an object
  156. * - objects may be submitted only in the following states:
  157. * - during object creation (write ops may be submitted)
  158. * - whilst the object is active
  159. * - after an I/O error incurred in one of the two above states (op rejected)
  160. * - this gets any extra refs it needs on an op
  161. */
  162. int fscache_submit_op(struct fscache_object *object,
  163. struct fscache_operation *op)
  164. {
  165. unsigned long ostate;
  166. int ret;
  167. _enter("{OBJ%x OP%x},{%u}",
  168. object->debug_id, op->debug_id, atomic_read(&op->usage));
  169. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_INITIALISED);
  170. ASSERTCMP(atomic_read(&op->usage), >, 0);
  171. spin_lock(&object->lock);
  172. ASSERTCMP(object->n_ops, >=, object->n_in_progress);
  173. ASSERTCMP(object->n_ops, >=, object->n_exclusive);
  174. ASSERT(list_empty(&op->pend_link));
  175. ostate = object->state;
  176. smp_rmb();
  177. op->state = FSCACHE_OP_ST_PENDING;
  178. if (fscache_object_is_active(object)) {
  179. op->object = object;
  180. object->n_ops++;
  181. if (object->n_exclusive > 0) {
  182. atomic_inc(&op->usage);
  183. list_add_tail(&op->pend_link, &object->pending_ops);
  184. fscache_stat(&fscache_n_op_pend);
  185. } else if (!list_empty(&object->pending_ops)) {
  186. atomic_inc(&op->usage);
  187. list_add_tail(&op->pend_link, &object->pending_ops);
  188. fscache_stat(&fscache_n_op_pend);
  189. fscache_start_operations(object);
  190. } else {
  191. ASSERTCMP(object->n_exclusive, ==, 0);
  192. fscache_run_op(object, op);
  193. }
  194. ret = 0;
  195. } else if (object->state == FSCACHE_OBJECT_CREATING) {
  196. op->object = object;
  197. object->n_ops++;
  198. atomic_inc(&op->usage);
  199. list_add_tail(&op->pend_link, &object->pending_ops);
  200. fscache_stat(&fscache_n_op_pend);
  201. ret = 0;
  202. } else if (object->state == FSCACHE_OBJECT_DYING ||
  203. object->state == FSCACHE_OBJECT_LC_DYING ||
  204. object->state == FSCACHE_OBJECT_WITHDRAWING) {
  205. fscache_stat(&fscache_n_op_rejected);
  206. op->state = FSCACHE_OP_ST_CANCELLED;
  207. ret = -ENOBUFS;
  208. } else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
  209. fscache_report_unexpected_submission(object, op, ostate);
  210. ASSERT(!fscache_object_is_active(object));
  211. op->state = FSCACHE_OP_ST_CANCELLED;
  212. ret = -ENOBUFS;
  213. } else {
  214. op->state = FSCACHE_OP_ST_CANCELLED;
  215. ret = -ENOBUFS;
  216. }
  217. spin_unlock(&object->lock);
  218. return ret;
  219. }
  220. /*
  221. * queue an object for withdrawal on error, aborting all following asynchronous
  222. * operations
  223. */
  224. void fscache_abort_object(struct fscache_object *object)
  225. {
  226. _enter("{OBJ%x}", object->debug_id);
  227. fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
  228. }
  229. /*
  230. * jump start the operation processing on an object
  231. * - caller must hold object->lock
  232. */
  233. void fscache_start_operations(struct fscache_object *object)
  234. {
  235. struct fscache_operation *op;
  236. bool stop = false;
  237. while (!list_empty(&object->pending_ops) && !stop) {
  238. op = list_entry(object->pending_ops.next,
  239. struct fscache_operation, pend_link);
  240. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
  241. if (object->n_in_progress > 0)
  242. break;
  243. stop = true;
  244. }
  245. list_del_init(&op->pend_link);
  246. fscache_run_op(object, op);
  247. /* the pending queue was holding a ref on the object */
  248. fscache_put_operation(op);
  249. }
  250. ASSERTCMP(object->n_in_progress, <=, object->n_ops);
  251. _debug("woke %d ops on OBJ%x",
  252. object->n_in_progress, object->debug_id);
  253. }
  254. /*
  255. * cancel an operation that's pending on an object
  256. */
  257. int fscache_cancel_op(struct fscache_operation *op)
  258. {
  259. struct fscache_object *object = op->object;
  260. int ret;
  261. _enter("OBJ%x OP%x}", op->object->debug_id, op->debug_id);
  262. ASSERTCMP(op->state, >=, FSCACHE_OP_ST_PENDING);
  263. ASSERTCMP(op->state, !=, FSCACHE_OP_ST_CANCELLED);
  264. ASSERTCMP(atomic_read(&op->usage), >, 0);
  265. spin_lock(&object->lock);
  266. ret = -EBUSY;
  267. if (op->state == FSCACHE_OP_ST_PENDING) {
  268. ASSERT(!list_empty(&op->pend_link));
  269. fscache_stat(&fscache_n_op_cancelled);
  270. list_del_init(&op->pend_link);
  271. op->state = FSCACHE_OP_ST_CANCELLED;
  272. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
  273. object->n_exclusive--;
  274. if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
  275. wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
  276. fscache_put_operation(op);
  277. ret = 0;
  278. }
  279. spin_unlock(&object->lock);
  280. _leave(" = %d", ret);
  281. return ret;
  282. }
  283. /*
  284. * Record the completion of an in-progress operation.
  285. */
  286. void fscache_op_complete(struct fscache_operation *op)
  287. {
  288. struct fscache_object *object = op->object;
  289. _enter("OBJ%x", object->debug_id);
  290. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_IN_PROGRESS);
  291. ASSERTCMP(object->n_in_progress, >, 0);
  292. ASSERTIFCMP(test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags),
  293. object->n_exclusive, >, 0);
  294. ASSERTIFCMP(test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags),
  295. object->n_in_progress, ==, 1);
  296. spin_lock(&object->lock);
  297. op->state = FSCACHE_OP_ST_COMPLETE;
  298. if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
  299. object->n_exclusive--;
  300. object->n_in_progress--;
  301. if (object->n_in_progress == 0)
  302. fscache_start_operations(object);
  303. spin_unlock(&object->lock);
  304. _leave("");
  305. }
  306. EXPORT_SYMBOL(fscache_op_complete);
  307. /*
  308. * release an operation
  309. * - queues pending ops if this is the last in-progress op
  310. */
  311. void fscache_put_operation(struct fscache_operation *op)
  312. {
  313. struct fscache_object *object;
  314. struct fscache_cache *cache;
  315. _enter("{OBJ%x OP%x,%d}",
  316. op->object->debug_id, op->debug_id, atomic_read(&op->usage));
  317. ASSERTCMP(atomic_read(&op->usage), >, 0);
  318. if (!atomic_dec_and_test(&op->usage))
  319. return;
  320. _debug("PUT OP");
  321. ASSERTIFCMP(op->state != FSCACHE_OP_ST_COMPLETE,
  322. op->state, ==, FSCACHE_OP_ST_CANCELLED);
  323. op->state = FSCACHE_OP_ST_DEAD;
  324. fscache_stat(&fscache_n_op_release);
  325. if (op->release) {
  326. op->release(op);
  327. op->release = NULL;
  328. }
  329. object = op->object;
  330. if (test_bit(FSCACHE_OP_DEC_READ_CNT, &op->flags))
  331. atomic_dec(&object->n_reads);
  332. /* now... we may get called with the object spinlock held, so we
  333. * complete the cleanup here only if we can immediately acquire the
  334. * lock, and defer it otherwise */
  335. if (!spin_trylock(&object->lock)) {
  336. _debug("defer put");
  337. fscache_stat(&fscache_n_op_deferred_release);
  338. cache = object->cache;
  339. spin_lock(&cache->op_gc_list_lock);
  340. list_add_tail(&op->pend_link, &cache->op_gc_list);
  341. spin_unlock(&cache->op_gc_list_lock);
  342. schedule_work(&cache->op_gc);
  343. _leave(" [defer]");
  344. return;
  345. }
  346. ASSERTCMP(object->n_ops, >, 0);
  347. object->n_ops--;
  348. if (object->n_ops == 0)
  349. fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
  350. spin_unlock(&object->lock);
  351. kfree(op);
  352. _leave(" [done]");
  353. }
  354. EXPORT_SYMBOL(fscache_put_operation);
  355. /*
  356. * garbage collect operations that have had their release deferred
  357. */
  358. void fscache_operation_gc(struct work_struct *work)
  359. {
  360. struct fscache_operation *op;
  361. struct fscache_object *object;
  362. struct fscache_cache *cache =
  363. container_of(work, struct fscache_cache, op_gc);
  364. int count = 0;
  365. _enter("");
  366. do {
  367. spin_lock(&cache->op_gc_list_lock);
  368. if (list_empty(&cache->op_gc_list)) {
  369. spin_unlock(&cache->op_gc_list_lock);
  370. break;
  371. }
  372. op = list_entry(cache->op_gc_list.next,
  373. struct fscache_operation, pend_link);
  374. list_del(&op->pend_link);
  375. spin_unlock(&cache->op_gc_list_lock);
  376. object = op->object;
  377. spin_lock(&object->lock);
  378. _debug("GC DEFERRED REL OBJ%x OP%x",
  379. object->debug_id, op->debug_id);
  380. fscache_stat(&fscache_n_op_gc);
  381. ASSERTCMP(atomic_read(&op->usage), ==, 0);
  382. ASSERTCMP(op->state, ==, FSCACHE_OP_ST_DEAD);
  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. } while (count++ < 20);
  390. if (!list_empty(&cache->op_gc_list))
  391. schedule_work(&cache->op_gc);
  392. _leave("");
  393. }
  394. /*
  395. * execute an operation using fs_op_wq to provide processing context -
  396. * the caller holds a ref to this object, so we don't need to hold one
  397. */
  398. void fscache_op_work_func(struct work_struct *work)
  399. {
  400. struct fscache_operation *op =
  401. container_of(work, struct fscache_operation, work);
  402. unsigned long start;
  403. _enter("{OBJ%x OP%x,%d}",
  404. op->object->debug_id, op->debug_id, atomic_read(&op->usage));
  405. ASSERT(op->processor != NULL);
  406. start = jiffies;
  407. op->processor(op);
  408. fscache_hist(fscache_ops_histogram, start);
  409. fscache_put_operation(op);
  410. _leave("");
  411. }