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
14#define FSCACHE_DEBUG_LEVEL OPERATION
15#include <linux/module.h>
16#include <linux/seq_file.h>
17#include <linux/slab.h>
18#include "internal.h"
19
20atomic_t fscache_op_debug_id;
21EXPORT_SYMBOL(fscache_op_debug_id);
22
23/**
24 * fscache_enqueue_operation - Enqueue an operation for processing
25 * @op: The operation to enqueue
26 *
27 * Enqueue an operation for processing by the FS-Cache thread pool.
28 *
29 * This will get its own ref on the object.
30 */
31void fscache_enqueue_operation(struct fscache_operation *op)
32{
33	_enter("{OBJ%x OP%x,%u}",
34	       op->object->debug_id, op->debug_id, atomic_read(&op->usage));
35
36	fscache_set_op_state(op, "EnQ");
37
38	ASSERT(list_empty(&op->pend_link));
39	ASSERT(op->processor != NULL);
40	ASSERTCMP(op->object->state, >=, FSCACHE_OBJECT_AVAILABLE);
41	ASSERTCMP(atomic_read(&op->usage), >, 0);
42
43	fscache_stat(&fscache_n_op_enqueue);
44	switch (op->flags & FSCACHE_OP_TYPE) {
45	case FSCACHE_OP_ASYNC:
46		_debug("queue async");
47		atomic_inc(&op->usage);
48		if (!queue_work(fscache_op_wq, &op->work))
49			fscache_put_operation(op);
50		break;
51	case FSCACHE_OP_MYTHREAD:
52		_debug("queue for caller's attention");
53		break;
54	default:
55		printk(KERN_ERR "FS-Cache: Unexpected op type %lx",
56		       op->flags);
57		BUG();
58		break;
59	}
60}
61EXPORT_SYMBOL(fscache_enqueue_operation);
62
63/*
64 * start an op running
65 */
66static void fscache_run_op(struct fscache_object *object,
67			   struct fscache_operation *op)
68{
69	fscache_set_op_state(op, "Run");
70
71	object->n_in_progress++;
72	if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
73		wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
74	if (op->processor)
75		fscache_enqueue_operation(op);
76	fscache_stat(&fscache_n_op_run);
77}
78
79/*
80 * submit an exclusive operation for an object
81 * - other ops are excluded from running simultaneously with this one
82 * - this gets any extra refs it needs on an op
83 */
84int fscache_submit_exclusive_op(struct fscache_object *object,
85				struct fscache_operation *op)
86{
87	int ret;
88
89	_enter("{OBJ%x OP%x},", object->debug_id, op->debug_id);
90
91	fscache_set_op_state(op, "SubmitX");
92
93	spin_lock(&object->lock);
94	ASSERTCMP(object->n_ops, >=, object->n_in_progress);
95	ASSERTCMP(object->n_ops, >=, object->n_exclusive);
96	ASSERT(list_empty(&op->pend_link));
97
98	ret = -ENOBUFS;
99	if (fscache_object_is_active(object)) {
100		op->object = object;
101		object->n_ops++;
102		object->n_exclusive++;	/* reads and writes must wait */
103
104		if (object->n_ops > 0) {
105			atomic_inc(&op->usage);
106			list_add_tail(&op->pend_link, &object->pending_ops);
107			fscache_stat(&fscache_n_op_pend);
108		} else if (!list_empty(&object->pending_ops)) {
109			atomic_inc(&op->usage);
110			list_add_tail(&op->pend_link, &object->pending_ops);
111			fscache_stat(&fscache_n_op_pend);
112			fscache_start_operations(object);
113		} else {
114			ASSERTCMP(object->n_in_progress, ==, 0);
115			fscache_run_op(object, op);
116		}
117
118		/* need to issue a new write op after this */
119		clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
120		ret = 0;
121	} else if (object->state == FSCACHE_OBJECT_CREATING) {
122		op->object = object;
123		object->n_ops++;
124		object->n_exclusive++;	/* reads and writes must wait */
125		atomic_inc(&op->usage);
126		list_add_tail(&op->pend_link, &object->pending_ops);
127		fscache_stat(&fscache_n_op_pend);
128		ret = 0;
129	} else {
130		/* not allowed to submit ops in any other state */
131		BUG();
132	}
133
134	spin_unlock(&object->lock);
135	return ret;
136}
137
138/*
139 * report an unexpected submission
140 */
141static void fscache_report_unexpected_submission(struct fscache_object *object,
142						 struct fscache_operation *op,
143						 unsigned long ostate)
144{
145	static bool once_only;
146	struct fscache_operation *p;
147	unsigned n;
148
149	if (once_only)
150		return;
151	once_only = true;
152
153	kdebug("unexpected submission OP%x [OBJ%x %s]",
154	       op->debug_id, object->debug_id,
155	       fscache_object_states[object->state]);
156	kdebug("objstate=%s [%s]",
157	       fscache_object_states[object->state],
158	       fscache_object_states[ostate]);
159	kdebug("objflags=%lx", object->flags);
160	kdebug("objevent=%lx [%lx]", object->events, object->event_mask);
161	kdebug("ops=%u inp=%u exc=%u",
162	       object->n_ops, object->n_in_progress, object->n_exclusive);
163
164	if (!list_empty(&object->pending_ops)) {
165		n = 0;
166		list_for_each_entry(p, &object->pending_ops, pend_link) {
167			ASSERTCMP(p->object, ==, object);
168			kdebug("%p %p", op->processor, op->release);
169			n++;
170		}
171
172		kdebug("n=%u", n);
173	}
174
175	dump_stack();
176}
177
178/*
179 * submit an operation for an object
180 * - objects may be submitted only in the following states:
181 *   - during object creation (write ops may be submitted)
182 *   - whilst the object is active
183 *   - after an I/O error incurred in one of the two above states (op rejected)
184 * - this gets any extra refs it needs on an op
185 */
186int fscache_submit_op(struct fscache_object *object,
187		      struct fscache_operation *op)
188{
189	unsigned long ostate;
190	int ret;
191
192	_enter("{OBJ%x OP%x},{%u}",
193	       object->debug_id, op->debug_id, atomic_read(&op->usage));
194
195	ASSERTCMP(atomic_read(&op->usage), >, 0);
196
197	fscache_set_op_state(op, "Submit");
198
199	spin_lock(&object->lock);
200	ASSERTCMP(object->n_ops, >=, object->n_in_progress);
201	ASSERTCMP(object->n_ops, >=, object->n_exclusive);
202	ASSERT(list_empty(&op->pend_link));
203
204	ostate = object->state;
205	smp_rmb();
206
207	if (fscache_object_is_active(object)) {
208		op->object = object;
209		object->n_ops++;
210
211		if (object->n_exclusive > 0) {
212			atomic_inc(&op->usage);
213			list_add_tail(&op->pend_link, &object->pending_ops);
214			fscache_stat(&fscache_n_op_pend);
215		} else if (!list_empty(&object->pending_ops)) {
216			atomic_inc(&op->usage);
217			list_add_tail(&op->pend_link, &object->pending_ops);
218			fscache_stat(&fscache_n_op_pend);
219			fscache_start_operations(object);
220		} else {
221			ASSERTCMP(object->n_exclusive, ==, 0);
222			fscache_run_op(object, op);
223		}
224		ret = 0;
225	} else if (object->state == FSCACHE_OBJECT_CREATING) {
226		op->object = object;
227		object->n_ops++;
228		atomic_inc(&op->usage);
229		list_add_tail(&op->pend_link, &object->pending_ops);
230		fscache_stat(&fscache_n_op_pend);
231		ret = 0;
232	} else if (object->state == FSCACHE_OBJECT_DYING ||
233		   object->state == FSCACHE_OBJECT_LC_DYING ||
234		   object->state == FSCACHE_OBJECT_WITHDRAWING) {
235		fscache_stat(&fscache_n_op_rejected);
236		ret = -ENOBUFS;
237	} else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
238		fscache_report_unexpected_submission(object, op, ostate);
239		ASSERT(!fscache_object_is_active(object));
240		ret = -ENOBUFS;
241	} else {
242		ret = -ENOBUFS;
243	}
244
245	spin_unlock(&object->lock);
246	return ret;
247}
248
249/*
250 * queue an object for withdrawal on error, aborting all following asynchronous
251 * operations
252 */
253void fscache_abort_object(struct fscache_object *object)
254{
255	_enter("{OBJ%x}", object->debug_id);
256
257	fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
258}
259
260/*
261 * jump start the operation processing on an object
262 * - caller must hold object->lock
263 */
264void fscache_start_operations(struct fscache_object *object)
265{
266	struct fscache_operation *op;
267	bool stop = false;
268
269	while (!list_empty(&object->pending_ops) && !stop) {
270		op = list_entry(object->pending_ops.next,
271				struct fscache_operation, pend_link);
272
273		if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
274			if (object->n_in_progress > 0)
275				break;
276			stop = true;
277		}
278		list_del_init(&op->pend_link);
279		fscache_run_op(object, op);
280
281		/* the pending queue was holding a ref on the object */
282		fscache_put_operation(op);
283	}
284
285	ASSERTCMP(object->n_in_progress, <=, object->n_ops);
286
287	_debug("woke %d ops on OBJ%x",
288	       object->n_in_progress, object->debug_id);
289}
290
291/*
292 * cancel an operation that's pending on an object
293 */
294int fscache_cancel_op(struct fscache_operation *op)
295{
296	struct fscache_object *object = op->object;
297	int ret;
298
299	_enter("OBJ%x OP%x}", op->object->debug_id, op->debug_id);
300
301	spin_lock(&object->lock);
302
303	ret = -EBUSY;
304	if (!list_empty(&op->pend_link)) {
305		fscache_stat(&fscache_n_op_cancelled);
306		list_del_init(&op->pend_link);
307		object->n_ops--;
308		if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
309			object->n_exclusive--;
310		if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
311			wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
312		fscache_put_operation(op);
313		ret = 0;
314	}
315
316	spin_unlock(&object->lock);
317	_leave(" = %d", ret);
318	return ret;
319}
320
321/*
322 * release an operation
323 * - queues pending ops if this is the last in-progress op
324 */
325void fscache_put_operation(struct fscache_operation *op)
326{
327	struct fscache_object *object;
328	struct fscache_cache *cache;
329
330	_enter("{OBJ%x OP%x,%d}",
331	       op->object->debug_id, op->debug_id, atomic_read(&op->usage));
332
333	ASSERTCMP(atomic_read(&op->usage), >, 0);
334
335	if (!atomic_dec_and_test(&op->usage))
336		return;
337
338	fscache_set_op_state(op, "Put");
339
340	_debug("PUT OP");
341	if (test_and_set_bit(FSCACHE_OP_DEAD, &op->flags))
342		BUG();
343
344	fscache_stat(&fscache_n_op_release);
345
346	if (op->release) {
347		op->release(op);
348		op->release = NULL;
349	}
350
351	object = op->object;
352
353	if (test_bit(FSCACHE_OP_DEC_READ_CNT, &op->flags))
354		atomic_dec(&object->n_reads);
355
356	/* now... we may get called with the object spinlock held, so we
357	 * complete the cleanup here only if we can immediately acquire the
358	 * lock, and defer it otherwise */
359	if (!spin_trylock(&object->lock)) {
360		_debug("defer put");
361		fscache_stat(&fscache_n_op_deferred_release);
362
363		cache = object->cache;
364		spin_lock(&cache->op_gc_list_lock);
365		list_add_tail(&op->pend_link, &cache->op_gc_list);
366		spin_unlock(&cache->op_gc_list_lock);
367		schedule_work(&cache->op_gc);
368		_leave(" [defer]");
369		return;
370	}
371
372	if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
373		ASSERTCMP(object->n_exclusive, >, 0);
374		object->n_exclusive--;
375	}
376
377	ASSERTCMP(object->n_in_progress, >, 0);
378	object->n_in_progress--;
379	if (object->n_in_progress == 0)
380		fscache_start_operations(object);
381
382	ASSERTCMP(object->n_ops, >, 0);
383	object->n_ops--;
384	if (object->n_ops == 0)
385		fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
386
387	spin_unlock(&object->lock);
388
389	kfree(op);
390	_leave(" [done]");
391}
392EXPORT_SYMBOL(fscache_put_operation);
393
394/*
395 * garbage collect operations that have had their release deferred
396 */
397void fscache_operation_gc(struct work_struct *work)
398{
399	struct fscache_operation *op;
400	struct fscache_object *object;
401	struct fscache_cache *cache =
402		container_of(work, struct fscache_cache, op_gc);
403	int count = 0;
404
405	_enter("");
406
407	do {
408		spin_lock(&cache->op_gc_list_lock);
409		if (list_empty(&cache->op_gc_list)) {
410			spin_unlock(&cache->op_gc_list_lock);
411			break;
412		}
413
414		op = list_entry(cache->op_gc_list.next,
415				struct fscache_operation, pend_link);
416		list_del(&op->pend_link);
417		spin_unlock(&cache->op_gc_list_lock);
418
419		object = op->object;
420
421		_debug("GC DEFERRED REL OBJ%x OP%x",
422		       object->debug_id, op->debug_id);
423		fscache_stat(&fscache_n_op_gc);
424
425		ASSERTCMP(atomic_read(&op->usage), ==, 0);
426
427		spin_lock(&object->lock);
428		if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
429			ASSERTCMP(object->n_exclusive, >, 0);
430			object->n_exclusive--;
431		}
432
433		ASSERTCMP(object->n_in_progress, >, 0);
434		object->n_in_progress--;
435		if (object->n_in_progress == 0)
436			fscache_start_operations(object);
437
438		ASSERTCMP(object->n_ops, >, 0);
439		object->n_ops--;
440		if (object->n_ops == 0)
441			fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
442
443		spin_unlock(&object->lock);
444
445	} while (count++ < 20);
446
447	if (!list_empty(&cache->op_gc_list))
448		schedule_work(&cache->op_gc);
449
450	_leave("");
451}
452
453/*
454 * execute an operation using fs_op_wq to provide processing context -
455 * the caller holds a ref to this object, so we don't need to hold one
456 */
457void fscache_op_work_func(struct work_struct *work)
458{
459	struct fscache_operation *op =
460		container_of(work, struct fscache_operation, work);
461	unsigned long start;
462
463	_enter("{OBJ%x OP%x,%d}",
464	       op->object->debug_id, op->debug_id, atomic_read(&op->usage));
465
466	ASSERT(op->processor != NULL);
467	start = jiffies;
468	op->processor(op);
469	fscache_hist(fscache_ops_histogram, start);
470	fscache_put_operation(op);
471
472	_leave("");
473}
474