These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / fs / fscache / object.c
1 /* FS-Cache object state machine handler
2  *
3  * Copyright (C) 2007 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/object.txt for a description of the
12  * object state machine and the in-kernel representations.
13  */
14
15 #define FSCACHE_DEBUG_LEVEL COOKIE
16 #include <linux/module.h>
17 #include <linux/slab.h>
18 #include <linux/prefetch.h>
19 #include "internal.h"
20
21 static const struct fscache_state *fscache_abort_initialisation(struct fscache_object *, int);
22 static const struct fscache_state *fscache_kill_dependents(struct fscache_object *, int);
23 static const struct fscache_state *fscache_drop_object(struct fscache_object *, int);
24 static const struct fscache_state *fscache_initialise_object(struct fscache_object *, int);
25 static const struct fscache_state *fscache_invalidate_object(struct fscache_object *, int);
26 static const struct fscache_state *fscache_jumpstart_dependents(struct fscache_object *, int);
27 static const struct fscache_state *fscache_kill_object(struct fscache_object *, int);
28 static const struct fscache_state *fscache_lookup_failure(struct fscache_object *, int);
29 static const struct fscache_state *fscache_look_up_object(struct fscache_object *, int);
30 static const struct fscache_state *fscache_object_available(struct fscache_object *, int);
31 static const struct fscache_state *fscache_parent_ready(struct fscache_object *, int);
32 static const struct fscache_state *fscache_update_object(struct fscache_object *, int);
33
34 #define __STATE_NAME(n) fscache_osm_##n
35 #define STATE(n) (&__STATE_NAME(n))
36
37 /*
38  * Define a work state.  Work states are execution states.  No event processing
39  * is performed by them.  The function attached to a work state returns a
40  * pointer indicating the next state to which the state machine should
41  * transition.  Returning NO_TRANSIT repeats the current state, but goes back
42  * to the scheduler first.
43  */
44 #define WORK_STATE(n, sn, f) \
45         const struct fscache_state __STATE_NAME(n) = {                  \
46                 .name = #n,                                             \
47                 .short_name = sn,                                       \
48                 .work = f                                               \
49         }
50
51 /*
52  * Returns from work states.
53  */
54 #define transit_to(state) ({ prefetch(&STATE(state)->work); STATE(state); })
55
56 #define NO_TRANSIT ((struct fscache_state *)NULL)
57
58 /*
59  * Define a wait state.  Wait states are event processing states.  No execution
60  * is performed by them.  Wait states are just tables of "if event X occurs,
61  * clear it and transition to state Y".  The dispatcher returns to the
62  * scheduler if none of the events in which the wait state has an interest are
63  * currently pending.
64  */
65 #define WAIT_STATE(n, sn, ...) \
66         const struct fscache_state __STATE_NAME(n) = {                  \
67                 .name = #n,                                             \
68                 .short_name = sn,                                       \
69                 .work = NULL,                                           \
70                 .transitions = { __VA_ARGS__, { 0, NULL } }             \
71         }
72
73 #define TRANSIT_TO(state, emask) \
74         { .events = (emask), .transit_to = STATE(state) }
75
76 /*
77  * The object state machine.
78  */
79 static WORK_STATE(INIT_OBJECT,          "INIT", fscache_initialise_object);
80 static WORK_STATE(PARENT_READY,         "PRDY", fscache_parent_ready);
81 static WORK_STATE(ABORT_INIT,           "ABRT", fscache_abort_initialisation);
82 static WORK_STATE(LOOK_UP_OBJECT,       "LOOK", fscache_look_up_object);
83 static WORK_STATE(CREATE_OBJECT,        "CRTO", fscache_look_up_object);
84 static WORK_STATE(OBJECT_AVAILABLE,     "AVBL", fscache_object_available);
85 static WORK_STATE(JUMPSTART_DEPS,       "JUMP", fscache_jumpstart_dependents);
86
87 static WORK_STATE(INVALIDATE_OBJECT,    "INVL", fscache_invalidate_object);
88 static WORK_STATE(UPDATE_OBJECT,        "UPDT", fscache_update_object);
89
90 static WORK_STATE(LOOKUP_FAILURE,       "LCFL", fscache_lookup_failure);
91 static WORK_STATE(KILL_OBJECT,          "KILL", fscache_kill_object);
92 static WORK_STATE(KILL_DEPENDENTS,      "KDEP", fscache_kill_dependents);
93 static WORK_STATE(DROP_OBJECT,          "DROP", fscache_drop_object);
94 static WORK_STATE(OBJECT_DEAD,          "DEAD", (void*)2UL);
95
96 static WAIT_STATE(WAIT_FOR_INIT,        "?INI",
97                   TRANSIT_TO(INIT_OBJECT,       1 << FSCACHE_OBJECT_EV_NEW_CHILD));
98
99 static WAIT_STATE(WAIT_FOR_PARENT,      "?PRN",
100                   TRANSIT_TO(PARENT_READY,      1 << FSCACHE_OBJECT_EV_PARENT_READY));
101
102 static WAIT_STATE(WAIT_FOR_CMD,         "?CMD",
103                   TRANSIT_TO(INVALIDATE_OBJECT, 1 << FSCACHE_OBJECT_EV_INVALIDATE),
104                   TRANSIT_TO(UPDATE_OBJECT,     1 << FSCACHE_OBJECT_EV_UPDATE),
105                   TRANSIT_TO(JUMPSTART_DEPS,    1 << FSCACHE_OBJECT_EV_NEW_CHILD));
106
107 static WAIT_STATE(WAIT_FOR_CLEARANCE,   "?CLR",
108                   TRANSIT_TO(KILL_OBJECT,       1 << FSCACHE_OBJECT_EV_CLEARED));
109
110 /*
111  * Out-of-band event transition tables.  These are for handling unexpected
112  * events, such as an I/O error.  If an OOB event occurs, the state machine
113  * clears and disables the event and forces a transition to the nominated work
114  * state (acurrently executing work states will complete first).
115  *
116  * In such a situation, object->state remembers the state the machine should
117  * have been in/gone to and returning NO_TRANSIT returns to that.
118  */
119 static const struct fscache_transition fscache_osm_init_oob[] = {
120            TRANSIT_TO(ABORT_INIT,
121                       (1 << FSCACHE_OBJECT_EV_ERROR) |
122                       (1 << FSCACHE_OBJECT_EV_KILL)),
123            { 0, NULL }
124 };
125
126 static const struct fscache_transition fscache_osm_lookup_oob[] = {
127            TRANSIT_TO(LOOKUP_FAILURE,
128                       (1 << FSCACHE_OBJECT_EV_ERROR) |
129                       (1 << FSCACHE_OBJECT_EV_KILL)),
130            { 0, NULL }
131 };
132
133 static const struct fscache_transition fscache_osm_run_oob[] = {
134            TRANSIT_TO(KILL_OBJECT,
135                       (1 << FSCACHE_OBJECT_EV_ERROR) |
136                       (1 << FSCACHE_OBJECT_EV_KILL)),
137            { 0, NULL }
138 };
139
140 static int  fscache_get_object(struct fscache_object *);
141 static void fscache_put_object(struct fscache_object *);
142 static bool fscache_enqueue_dependents(struct fscache_object *, int);
143 static void fscache_dequeue_object(struct fscache_object *);
144
145 /*
146  * we need to notify the parent when an op completes that we had outstanding
147  * upon it
148  */
149 static inline void fscache_done_parent_op(struct fscache_object *object)
150 {
151         struct fscache_object *parent = object->parent;
152
153         _enter("OBJ%x {OBJ%x,%x}",
154                object->debug_id, parent->debug_id, parent->n_ops);
155
156         spin_lock_nested(&parent->lock, 1);
157         parent->n_obj_ops--;
158         parent->n_ops--;
159         if (parent->n_ops == 0)
160                 fscache_raise_event(parent, FSCACHE_OBJECT_EV_CLEARED);
161         spin_unlock(&parent->lock);
162 }
163
164 /*
165  * Object state machine dispatcher.
166  */
167 static void fscache_object_sm_dispatcher(struct fscache_object *object)
168 {
169         const struct fscache_transition *t;
170         const struct fscache_state *state, *new_state;
171         unsigned long events, event_mask;
172         int event = -1;
173
174         ASSERT(object != NULL);
175
176         _enter("{OBJ%x,%s,%lx}",
177                object->debug_id, object->state->name, object->events);
178
179         event_mask = object->event_mask;
180 restart:
181         object->event_mask = 0; /* Mask normal event handling */
182         state = object->state;
183 restart_masked:
184         events = object->events;
185
186         /* Handle any out-of-band events (typically an error) */
187         if (events & object->oob_event_mask) {
188                 _debug("{OBJ%x} oob %lx",
189                        object->debug_id, events & object->oob_event_mask);
190                 for (t = object->oob_table; t->events; t++) {
191                         if (events & t->events) {
192                                 state = t->transit_to;
193                                 ASSERT(state->work != NULL);
194                                 event = fls(events & t->events) - 1;
195                                 __clear_bit(event, &object->oob_event_mask);
196                                 clear_bit(event, &object->events);
197                                 goto execute_work_state;
198                         }
199                 }
200         }
201
202         /* Wait states are just transition tables */
203         if (!state->work) {
204                 if (events & event_mask) {
205                         for (t = state->transitions; t->events; t++) {
206                                 if (events & t->events) {
207                                         new_state = t->transit_to;
208                                         event = fls(events & t->events) - 1;
209                                         clear_bit(event, &object->events);
210                                         _debug("{OBJ%x} ev %d: %s -> %s",
211                                                object->debug_id, event,
212                                                state->name, new_state->name);
213                                         object->state = state = new_state;
214                                         goto execute_work_state;
215                                 }
216                         }
217
218                         /* The event mask didn't include all the tabled bits */
219                         BUG();
220                 }
221                 /* Randomly woke up */
222                 goto unmask_events;
223         }
224
225 execute_work_state:
226         _debug("{OBJ%x} exec %s", object->debug_id, state->name);
227
228         new_state = state->work(object, event);
229         event = -1;
230         if (new_state == NO_TRANSIT) {
231                 _debug("{OBJ%x} %s notrans", object->debug_id, state->name);
232                 fscache_enqueue_object(object);
233                 event_mask = object->oob_event_mask;
234                 goto unmask_events;
235         }
236
237         _debug("{OBJ%x} %s -> %s",
238                object->debug_id, state->name, new_state->name);
239         object->state = state = new_state;
240
241         if (state->work) {
242                 if (unlikely(state->work == ((void *)2UL))) {
243                         _leave(" [dead]");
244                         return;
245                 }
246                 goto restart_masked;
247         }
248
249         /* Transited to wait state */
250         event_mask = object->oob_event_mask;
251         for (t = state->transitions; t->events; t++)
252                 event_mask |= t->events;
253
254 unmask_events:
255         object->event_mask = event_mask;
256         smp_mb();
257         events = object->events;
258         if (events & event_mask)
259                 goto restart;
260         _leave(" [msk %lx]", event_mask);
261 }
262
263 /*
264  * execute an object
265  */
266 static void fscache_object_work_func(struct work_struct *work)
267 {
268         struct fscache_object *object =
269                 container_of(work, struct fscache_object, work);
270         unsigned long start;
271
272         _enter("{OBJ%x}", object->debug_id);
273
274         start = jiffies;
275         fscache_object_sm_dispatcher(object);
276         fscache_hist(fscache_objs_histogram, start);
277         fscache_put_object(object);
278 }
279
280 /**
281  * fscache_object_init - Initialise a cache object description
282  * @object: Object description
283  * @cookie: Cookie object will be attached to
284  * @cache: Cache in which backing object will be found
285  *
286  * Initialise a cache object description to its basic values.
287  *
288  * See Documentation/filesystems/caching/backend-api.txt for a complete
289  * description.
290  */
291 void fscache_object_init(struct fscache_object *object,
292                          struct fscache_cookie *cookie,
293                          struct fscache_cache *cache)
294 {
295         const struct fscache_transition *t;
296
297         atomic_inc(&cache->object_count);
298
299         object->state = STATE(WAIT_FOR_INIT);
300         object->oob_table = fscache_osm_init_oob;
301         object->flags = 1 << FSCACHE_OBJECT_IS_LIVE;
302         spin_lock_init(&object->lock);
303         INIT_LIST_HEAD(&object->cache_link);
304         INIT_HLIST_NODE(&object->cookie_link);
305         INIT_WORK(&object->work, fscache_object_work_func);
306         INIT_LIST_HEAD(&object->dependents);
307         INIT_LIST_HEAD(&object->dep_link);
308         INIT_LIST_HEAD(&object->pending_ops);
309         object->n_children = 0;
310         object->n_ops = object->n_in_progress = object->n_exclusive = 0;
311         object->events = 0;
312         object->store_limit = 0;
313         object->store_limit_l = 0;
314         object->cache = cache;
315         object->cookie = cookie;
316         object->parent = NULL;
317 #ifdef CONFIG_FSCACHE_OBJECT_LIST
318         RB_CLEAR_NODE(&object->objlist_link);
319 #endif
320
321         object->oob_event_mask = 0;
322         for (t = object->oob_table; t->events; t++)
323                 object->oob_event_mask |= t->events;
324         object->event_mask = object->oob_event_mask;
325         for (t = object->state->transitions; t->events; t++)
326                 object->event_mask |= t->events;
327 }
328 EXPORT_SYMBOL(fscache_object_init);
329
330 /*
331  * Mark the object as no longer being live, making sure that we synchronise
332  * against op submission.
333  */
334 static inline void fscache_mark_object_dead(struct fscache_object *object)
335 {
336         spin_lock(&object->lock);
337         clear_bit(FSCACHE_OBJECT_IS_LIVE, &object->flags);
338         spin_unlock(&object->lock);
339 }
340
341 /*
342  * Abort object initialisation before we start it.
343  */
344 static const struct fscache_state *fscache_abort_initialisation(struct fscache_object *object,
345                                                                 int event)
346 {
347         _enter("{OBJ%x},%d", object->debug_id, event);
348
349         object->oob_event_mask = 0;
350         fscache_dequeue_object(object);
351         return transit_to(KILL_OBJECT);
352 }
353
354 /*
355  * initialise an object
356  * - check the specified object's parent to see if we can make use of it
357  *   immediately to do a creation
358  * - we may need to start the process of creating a parent and we need to wait
359  *   for the parent's lookup and creation to complete if it's not there yet
360  */
361 static const struct fscache_state *fscache_initialise_object(struct fscache_object *object,
362                                                              int event)
363 {
364         struct fscache_object *parent;
365         bool success;
366
367         _enter("{OBJ%x},%d", object->debug_id, event);
368
369         ASSERT(list_empty(&object->dep_link));
370
371         parent = object->parent;
372         if (!parent) {
373                 _leave(" [no parent]");
374                 return transit_to(DROP_OBJECT);
375         }
376
377         _debug("parent: %s of:%lx", parent->state->name, parent->flags);
378
379         if (fscache_object_is_dying(parent)) {
380                 _leave(" [bad parent]");
381                 return transit_to(DROP_OBJECT);
382         }
383
384         if (fscache_object_is_available(parent)) {
385                 _leave(" [ready]");
386                 return transit_to(PARENT_READY);
387         }
388
389         _debug("wait");
390
391         spin_lock(&parent->lock);
392         fscache_stat(&fscache_n_cop_grab_object);
393         success = false;
394         if (fscache_object_is_live(parent) &&
395             object->cache->ops->grab_object(object)) {
396                 list_add(&object->dep_link, &parent->dependents);
397                 success = true;
398         }
399         fscache_stat_d(&fscache_n_cop_grab_object);
400         spin_unlock(&parent->lock);
401         if (!success) {
402                 _leave(" [grab failed]");
403                 return transit_to(DROP_OBJECT);
404         }
405
406         /* fscache_acquire_non_index_cookie() uses this
407          * to wake the chain up */
408         fscache_raise_event(parent, FSCACHE_OBJECT_EV_NEW_CHILD);
409         _leave(" [wait]");
410         return transit_to(WAIT_FOR_PARENT);
411 }
412
413 /*
414  * Once the parent object is ready, we should kick off our lookup op.
415  */
416 static const struct fscache_state *fscache_parent_ready(struct fscache_object *object,
417                                                         int event)
418 {
419         struct fscache_object *parent = object->parent;
420
421         _enter("{OBJ%x},%d", object->debug_id, event);
422
423         ASSERT(parent != NULL);
424
425         spin_lock(&parent->lock);
426         parent->n_ops++;
427         parent->n_obj_ops++;
428         object->lookup_jif = jiffies;
429         spin_unlock(&parent->lock);
430
431         _leave("");
432         return transit_to(LOOK_UP_OBJECT);
433 }
434
435 /*
436  * look an object up in the cache from which it was allocated
437  * - we hold an "access lock" on the parent object, so the parent object cannot
438  *   be withdrawn by either party till we've finished
439  */
440 static const struct fscache_state *fscache_look_up_object(struct fscache_object *object,
441                                                           int event)
442 {
443         struct fscache_cookie *cookie = object->cookie;
444         struct fscache_object *parent = object->parent;
445         int ret;
446
447         _enter("{OBJ%x},%d", object->debug_id, event);
448
449         object->oob_table = fscache_osm_lookup_oob;
450
451         ASSERT(parent != NULL);
452         ASSERTCMP(parent->n_ops, >, 0);
453         ASSERTCMP(parent->n_obj_ops, >, 0);
454
455         /* make sure the parent is still available */
456         ASSERT(fscache_object_is_available(parent));
457
458         if (fscache_object_is_dying(parent) ||
459             test_bit(FSCACHE_IOERROR, &object->cache->flags) ||
460             !fscache_use_cookie(object)) {
461                 _leave(" [unavailable]");
462                 return transit_to(LOOKUP_FAILURE);
463         }
464
465         _debug("LOOKUP \"%s\" in \"%s\"",
466                cookie->def->name, object->cache->tag->name);
467
468         fscache_stat(&fscache_n_object_lookups);
469         fscache_stat(&fscache_n_cop_lookup_object);
470         ret = object->cache->ops->lookup_object(object);
471         fscache_stat_d(&fscache_n_cop_lookup_object);
472
473         fscache_unuse_cookie(object);
474
475         if (ret == -ETIMEDOUT) {
476                 /* probably stuck behind another object, so move this one to
477                  * the back of the queue */
478                 fscache_stat(&fscache_n_object_lookups_timed_out);
479                 _leave(" [timeout]");
480                 return NO_TRANSIT;
481         }
482
483         if (ret < 0) {
484                 _leave(" [error]");
485                 return transit_to(LOOKUP_FAILURE);
486         }
487
488         _leave(" [ok]");
489         return transit_to(OBJECT_AVAILABLE);
490 }
491
492 /**
493  * fscache_object_lookup_negative - Note negative cookie lookup
494  * @object: Object pointing to cookie to mark
495  *
496  * Note negative lookup, permitting those waiting to read data from an already
497  * existing backing object to continue as there's no data for them to read.
498  */
499 void fscache_object_lookup_negative(struct fscache_object *object)
500 {
501         struct fscache_cookie *cookie = object->cookie;
502
503         _enter("{OBJ%x,%s}", object->debug_id, object->state->name);
504
505         if (!test_and_set_bit(FSCACHE_OBJECT_IS_LOOKED_UP, &object->flags)) {
506                 fscache_stat(&fscache_n_object_lookups_negative);
507
508                 /* Allow write requests to begin stacking up and read requests to begin
509                  * returning ENODATA.
510                  */
511                 set_bit(FSCACHE_COOKIE_NO_DATA_YET, &cookie->flags);
512                 clear_bit(FSCACHE_COOKIE_UNAVAILABLE, &cookie->flags);
513
514                 _debug("wake up lookup %p", &cookie->flags);
515                 clear_bit_unlock(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags);
516                 wake_up_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP);
517         }
518         _leave("");
519 }
520 EXPORT_SYMBOL(fscache_object_lookup_negative);
521
522 /**
523  * fscache_obtained_object - Note successful object lookup or creation
524  * @object: Object pointing to cookie to mark
525  *
526  * Note successful lookup and/or creation, permitting those waiting to write
527  * data to a backing object to continue.
528  *
529  * Note that after calling this, an object's cookie may be relinquished by the
530  * netfs, and so must be accessed with object lock held.
531  */
532 void fscache_obtained_object(struct fscache_object *object)
533 {
534         struct fscache_cookie *cookie = object->cookie;
535
536         _enter("{OBJ%x,%s}", object->debug_id, object->state->name);
537
538         /* if we were still looking up, then we must have a positive lookup
539          * result, in which case there may be data available */
540         if (!test_and_set_bit(FSCACHE_OBJECT_IS_LOOKED_UP, &object->flags)) {
541                 fscache_stat(&fscache_n_object_lookups_positive);
542
543                 /* We do (presumably) have data */
544                 clear_bit_unlock(FSCACHE_COOKIE_NO_DATA_YET, &cookie->flags);
545                 clear_bit(FSCACHE_COOKIE_UNAVAILABLE, &cookie->flags);
546
547                 /* Allow write requests to begin stacking up and read requests
548                  * to begin shovelling data.
549                  */
550                 clear_bit_unlock(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags);
551                 wake_up_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP);
552         } else {
553                 fscache_stat(&fscache_n_object_created);
554         }
555
556         set_bit(FSCACHE_OBJECT_IS_AVAILABLE, &object->flags);
557         _leave("");
558 }
559 EXPORT_SYMBOL(fscache_obtained_object);
560
561 /*
562  * handle an object that has just become available
563  */
564 static const struct fscache_state *fscache_object_available(struct fscache_object *object,
565                                                             int event)
566 {
567         _enter("{OBJ%x},%d", object->debug_id, event);
568
569         object->oob_table = fscache_osm_run_oob;
570
571         spin_lock(&object->lock);
572
573         fscache_done_parent_op(object);
574         if (object->n_in_progress == 0) {
575                 if (object->n_ops > 0) {
576                         ASSERTCMP(object->n_ops, >=, object->n_obj_ops);
577                         fscache_start_operations(object);
578                 } else {
579                         ASSERT(list_empty(&object->pending_ops));
580                 }
581         }
582         spin_unlock(&object->lock);
583
584         fscache_stat(&fscache_n_cop_lookup_complete);
585         object->cache->ops->lookup_complete(object);
586         fscache_stat_d(&fscache_n_cop_lookup_complete);
587
588         fscache_hist(fscache_obj_instantiate_histogram, object->lookup_jif);
589         fscache_stat(&fscache_n_object_avail);
590
591         _leave("");
592         return transit_to(JUMPSTART_DEPS);
593 }
594
595 /*
596  * Wake up this object's dependent objects now that we've become available.
597  */
598 static const struct fscache_state *fscache_jumpstart_dependents(struct fscache_object *object,
599                                                                 int event)
600 {
601         _enter("{OBJ%x},%d", object->debug_id, event);
602
603         if (!fscache_enqueue_dependents(object, FSCACHE_OBJECT_EV_PARENT_READY))
604                 return NO_TRANSIT; /* Not finished; requeue */
605         return transit_to(WAIT_FOR_CMD);
606 }
607
608 /*
609  * Handle lookup or creation failute.
610  */
611 static const struct fscache_state *fscache_lookup_failure(struct fscache_object *object,
612                                                           int event)
613 {
614         struct fscache_cookie *cookie;
615
616         _enter("{OBJ%x},%d", object->debug_id, event);
617
618         object->oob_event_mask = 0;
619
620         fscache_stat(&fscache_n_cop_lookup_complete);
621         object->cache->ops->lookup_complete(object);
622         fscache_stat_d(&fscache_n_cop_lookup_complete);
623
624         set_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->flags);
625
626         cookie = object->cookie;
627         set_bit(FSCACHE_COOKIE_UNAVAILABLE, &cookie->flags);
628         if (test_and_clear_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags))
629                 wake_up_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP);
630
631         fscache_done_parent_op(object);
632         return transit_to(KILL_OBJECT);
633 }
634
635 /*
636  * Wait for completion of all active operations on this object and the death of
637  * all child objects of this object.
638  */
639 static const struct fscache_state *fscache_kill_object(struct fscache_object *object,
640                                                        int event)
641 {
642         _enter("{OBJ%x,%d,%d},%d",
643                object->debug_id, object->n_ops, object->n_children, event);
644
645         fscache_mark_object_dead(object);
646         object->oob_event_mask = 0;
647
648         if (list_empty(&object->dependents) &&
649             object->n_ops == 0 &&
650             object->n_children == 0)
651                 return transit_to(DROP_OBJECT);
652
653         if (object->n_in_progress == 0) {
654                 spin_lock(&object->lock);
655                 if (object->n_ops > 0 && object->n_in_progress == 0)
656                         fscache_start_operations(object);
657                 spin_unlock(&object->lock);
658         }
659
660         if (!list_empty(&object->dependents))
661                 return transit_to(KILL_DEPENDENTS);
662
663         return transit_to(WAIT_FOR_CLEARANCE);
664 }
665
666 /*
667  * Kill dependent objects.
668  */
669 static const struct fscache_state *fscache_kill_dependents(struct fscache_object *object,
670                                                            int event)
671 {
672         _enter("{OBJ%x},%d", object->debug_id, event);
673
674         if (!fscache_enqueue_dependents(object, FSCACHE_OBJECT_EV_KILL))
675                 return NO_TRANSIT; /* Not finished */
676         return transit_to(WAIT_FOR_CLEARANCE);
677 }
678
679 /*
680  * Drop an object's attachments
681  */
682 static const struct fscache_state *fscache_drop_object(struct fscache_object *object,
683                                                        int event)
684 {
685         struct fscache_object *parent = object->parent;
686         struct fscache_cookie *cookie = object->cookie;
687         struct fscache_cache *cache = object->cache;
688         bool awaken = false;
689
690         _enter("{OBJ%x,%d},%d", object->debug_id, object->n_children, event);
691
692         ASSERT(cookie != NULL);
693         ASSERT(!hlist_unhashed(&object->cookie_link));
694
695         /* Make sure the cookie no longer points here and that the netfs isn't
696          * waiting for us.
697          */
698         spin_lock(&cookie->lock);
699         hlist_del_init(&object->cookie_link);
700         if (hlist_empty(&cookie->backing_objects) &&
701             test_and_clear_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags))
702                 awaken = true;
703         spin_unlock(&cookie->lock);
704
705         if (awaken)
706                 wake_up_bit(&cookie->flags, FSCACHE_COOKIE_INVALIDATING);
707
708         /* Prevent a race with our last child, which has to signal EV_CLEARED
709          * before dropping our spinlock.
710          */
711         spin_lock(&object->lock);
712         spin_unlock(&object->lock);
713
714         /* Discard from the cache's collection of objects */
715         spin_lock(&cache->object_list_lock);
716         list_del_init(&object->cache_link);
717         spin_unlock(&cache->object_list_lock);
718
719         fscache_stat(&fscache_n_cop_drop_object);
720         cache->ops->drop_object(object);
721         fscache_stat_d(&fscache_n_cop_drop_object);
722
723         /* The parent object wants to know when all it dependents have gone */
724         if (parent) {
725                 _debug("release parent OBJ%x {%d}",
726                        parent->debug_id, parent->n_children);
727
728                 spin_lock(&parent->lock);
729                 parent->n_children--;
730                 if (parent->n_children == 0)
731                         fscache_raise_event(parent, FSCACHE_OBJECT_EV_CLEARED);
732                 spin_unlock(&parent->lock);
733                 object->parent = NULL;
734         }
735
736         /* this just shifts the object release to the work processor */
737         fscache_put_object(object);
738         fscache_stat(&fscache_n_object_dead);
739
740         _leave("");
741         return transit_to(OBJECT_DEAD);
742 }
743
744 /*
745  * get a ref on an object
746  */
747 static int fscache_get_object(struct fscache_object *object)
748 {
749         int ret;
750
751         fscache_stat(&fscache_n_cop_grab_object);
752         ret = object->cache->ops->grab_object(object) ? 0 : -EAGAIN;
753         fscache_stat_d(&fscache_n_cop_grab_object);
754         return ret;
755 }
756
757 /*
758  * Discard a ref on an object
759  */
760 static void fscache_put_object(struct fscache_object *object)
761 {
762         fscache_stat(&fscache_n_cop_put_object);
763         object->cache->ops->put_object(object);
764         fscache_stat_d(&fscache_n_cop_put_object);
765 }
766
767 /**
768  * fscache_object_destroy - Note that a cache object is about to be destroyed
769  * @object: The object to be destroyed
770  *
771  * Note the imminent destruction and deallocation of a cache object record.
772  */
773 void fscache_object_destroy(struct fscache_object *object)
774 {
775         fscache_objlist_remove(object);
776
777         /* We can get rid of the cookie now */
778         fscache_cookie_put(object->cookie);
779         object->cookie = NULL;
780 }
781 EXPORT_SYMBOL(fscache_object_destroy);
782
783 /*
784  * enqueue an object for metadata-type processing
785  */
786 void fscache_enqueue_object(struct fscache_object *object)
787 {
788         _enter("{OBJ%x}", object->debug_id);
789
790         if (fscache_get_object(object) >= 0) {
791                 wait_queue_head_t *cong_wq =
792                         &get_cpu_var(fscache_object_cong_wait);
793
794                 if (queue_work(fscache_object_wq, &object->work)) {
795                         if (fscache_object_congested())
796                                 wake_up(cong_wq);
797                 } else
798                         fscache_put_object(object);
799
800                 put_cpu_var(fscache_object_cong_wait);
801         }
802 }
803
804 /**
805  * fscache_object_sleep_till_congested - Sleep until object wq is congested
806  * @timeoutp: Scheduler sleep timeout
807  *
808  * Allow an object handler to sleep until the object workqueue is congested.
809  *
810  * The caller must set up a wake up event before calling this and must have set
811  * the appropriate sleep mode (such as TASK_UNINTERRUPTIBLE) and tested its own
812  * condition before calling this function as no test is made here.
813  *
814  * %true is returned if the object wq is congested, %false otherwise.
815  */
816 bool fscache_object_sleep_till_congested(signed long *timeoutp)
817 {
818         wait_queue_head_t *cong_wq = this_cpu_ptr(&fscache_object_cong_wait);
819         DEFINE_WAIT(wait);
820
821         if (fscache_object_congested())
822                 return true;
823
824         add_wait_queue_exclusive(cong_wq, &wait);
825         if (!fscache_object_congested())
826                 *timeoutp = schedule_timeout(*timeoutp);
827         finish_wait(cong_wq, &wait);
828
829         return fscache_object_congested();
830 }
831 EXPORT_SYMBOL_GPL(fscache_object_sleep_till_congested);
832
833 /*
834  * Enqueue the dependents of an object for metadata-type processing.
835  *
836  * If we don't manage to finish the list before the scheduler wants to run
837  * again then return false immediately.  We return true if the list was
838  * cleared.
839  */
840 static bool fscache_enqueue_dependents(struct fscache_object *object, int event)
841 {
842         struct fscache_object *dep;
843         bool ret = true;
844
845         _enter("{OBJ%x}", object->debug_id);
846
847         if (list_empty(&object->dependents))
848                 return true;
849
850         spin_lock(&object->lock);
851
852         while (!list_empty(&object->dependents)) {
853                 dep = list_entry(object->dependents.next,
854                                  struct fscache_object, dep_link);
855                 list_del_init(&dep->dep_link);
856
857                 fscache_raise_event(dep, event);
858                 fscache_put_object(dep);
859
860                 if (!list_empty(&object->dependents) && need_resched()) {
861                         ret = false;
862                         break;
863                 }
864         }
865
866         spin_unlock(&object->lock);
867         return ret;
868 }
869
870 /*
871  * remove an object from whatever queue it's waiting on
872  */
873 static void fscache_dequeue_object(struct fscache_object *object)
874 {
875         _enter("{OBJ%x}", object->debug_id);
876
877         if (!list_empty(&object->dep_link)) {
878                 spin_lock(&object->parent->lock);
879                 list_del_init(&object->dep_link);
880                 spin_unlock(&object->parent->lock);
881         }
882
883         _leave("");
884 }
885
886 /**
887  * fscache_check_aux - Ask the netfs whether an object on disk is still valid
888  * @object: The object to ask about
889  * @data: The auxiliary data for the object
890  * @datalen: The size of the auxiliary data
891  *
892  * This function consults the netfs about the coherency state of an object.
893  * The caller must be holding a ref on cookie->n_active (held by
894  * fscache_look_up_object() on behalf of the cache backend during object lookup
895  * and creation).
896  */
897 enum fscache_checkaux fscache_check_aux(struct fscache_object *object,
898                                         const void *data, uint16_t datalen)
899 {
900         enum fscache_checkaux result;
901
902         if (!object->cookie->def->check_aux) {
903                 fscache_stat(&fscache_n_checkaux_none);
904                 return FSCACHE_CHECKAUX_OKAY;
905         }
906
907         result = object->cookie->def->check_aux(object->cookie->netfs_data,
908                                                 data, datalen);
909         switch (result) {
910                 /* entry okay as is */
911         case FSCACHE_CHECKAUX_OKAY:
912                 fscache_stat(&fscache_n_checkaux_okay);
913                 break;
914
915                 /* entry requires update */
916         case FSCACHE_CHECKAUX_NEEDS_UPDATE:
917                 fscache_stat(&fscache_n_checkaux_update);
918                 break;
919
920                 /* entry requires deletion */
921         case FSCACHE_CHECKAUX_OBSOLETE:
922                 fscache_stat(&fscache_n_checkaux_obsolete);
923                 break;
924
925         default:
926                 BUG();
927         }
928
929         return result;
930 }
931 EXPORT_SYMBOL(fscache_check_aux);
932
933 /*
934  * Asynchronously invalidate an object.
935  */
936 static const struct fscache_state *_fscache_invalidate_object(struct fscache_object *object,
937                                                               int event)
938 {
939         struct fscache_operation *op;
940         struct fscache_cookie *cookie = object->cookie;
941
942         _enter("{OBJ%x},%d", object->debug_id, event);
943
944         /* We're going to need the cookie.  If the cookie is not available then
945          * retire the object instead.
946          */
947         if (!fscache_use_cookie(object)) {
948                 ASSERT(object->cookie->stores.rnode == NULL);
949                 set_bit(FSCACHE_OBJECT_RETIRED, &object->flags);
950                 _leave(" [no cookie]");
951                 return transit_to(KILL_OBJECT);
952         }
953
954         /* Reject any new read/write ops and abort any that are pending. */
955         fscache_invalidate_writes(cookie);
956         clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
957         fscache_cancel_all_ops(object);
958
959         /* Now we have to wait for in-progress reads and writes */
960         op = kzalloc(sizeof(*op), GFP_KERNEL);
961         if (!op)
962                 goto nomem;
963
964         fscache_operation_init(op, object->cache->ops->invalidate_object,
965                                NULL, NULL);
966         op->flags = FSCACHE_OP_ASYNC |
967                 (1 << FSCACHE_OP_EXCLUSIVE) |
968                 (1 << FSCACHE_OP_UNUSE_COOKIE);
969
970         spin_lock(&cookie->lock);
971         if (fscache_submit_exclusive_op(object, op) < 0)
972                 goto submit_op_failed;
973         spin_unlock(&cookie->lock);
974         fscache_put_operation(op);
975
976         /* Once we've completed the invalidation, we know there will be no data
977          * stored in the cache and thus we can reinstate the data-check-skip
978          * optimisation.
979          */
980         set_bit(FSCACHE_COOKIE_NO_DATA_YET, &cookie->flags);
981
982         /* We can allow read and write requests to come in once again.  They'll
983          * queue up behind our exclusive invalidation operation.
984          */
985         if (test_and_clear_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags))
986                 wake_up_bit(&cookie->flags, FSCACHE_COOKIE_INVALIDATING);
987         _leave(" [ok]");
988         return transit_to(UPDATE_OBJECT);
989
990 nomem:
991         fscache_mark_object_dead(object);
992         fscache_unuse_cookie(object);
993         _leave(" [ENOMEM]");
994         return transit_to(KILL_OBJECT);
995
996 submit_op_failed:
997         fscache_mark_object_dead(object);
998         spin_unlock(&cookie->lock);
999         fscache_unuse_cookie(object);
1000         kfree(op);
1001         _leave(" [EIO]");
1002         return transit_to(KILL_OBJECT);
1003 }
1004
1005 static const struct fscache_state *fscache_invalidate_object(struct fscache_object *object,
1006                                                              int event)
1007 {
1008         const struct fscache_state *s;
1009
1010         fscache_stat(&fscache_n_invalidates_run);
1011         fscache_stat(&fscache_n_cop_invalidate_object);
1012         s = _fscache_invalidate_object(object, event);
1013         fscache_stat_d(&fscache_n_cop_invalidate_object);
1014         return s;
1015 }
1016
1017 /*
1018  * Asynchronously update an object.
1019  */
1020 static const struct fscache_state *fscache_update_object(struct fscache_object *object,
1021                                                          int event)
1022 {
1023         _enter("{OBJ%x},%d", object->debug_id, event);
1024
1025         fscache_stat(&fscache_n_updates_run);
1026         fscache_stat(&fscache_n_cop_update_object);
1027         object->cache->ops->update_object(object);
1028         fscache_stat_d(&fscache_n_cop_update_object);
1029
1030         _leave("");
1031         return transit_to(WAIT_FOR_CMD);
1032 }
1033
1034 /**
1035  * fscache_object_retrying_stale - Note retrying stale object
1036  * @object: The object that will be retried
1037  *
1038  * Note that an object lookup found an on-disk object that was adjudged to be
1039  * stale and has been deleted.  The lookup will be retried.
1040  */
1041 void fscache_object_retrying_stale(struct fscache_object *object)
1042 {
1043         fscache_stat(&fscache_n_cache_no_space_reject);
1044 }
1045 EXPORT_SYMBOL(fscache_object_retrying_stale);
1046
1047 /**
1048  * fscache_object_mark_killed - Note that an object was killed
1049  * @object: The object that was culled
1050  * @why: The reason the object was killed.
1051  *
1052  * Note that an object was killed.  Returns true if the object was
1053  * already marked killed, false if it wasn't.
1054  */
1055 void fscache_object_mark_killed(struct fscache_object *object,
1056                                 enum fscache_why_object_killed why)
1057 {
1058         if (test_and_set_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->flags)) {
1059                 pr_err("Error: Object already killed by cache [%s]\n",
1060                        object->cache->identifier);
1061                 return;
1062         }
1063
1064         switch (why) {
1065         case FSCACHE_OBJECT_NO_SPACE:
1066                 fscache_stat(&fscache_n_cache_no_space_reject);
1067                 break;
1068         case FSCACHE_OBJECT_IS_STALE:
1069                 fscache_stat(&fscache_n_cache_stale_objects);
1070                 break;
1071         case FSCACHE_OBJECT_WAS_RETIRED:
1072                 fscache_stat(&fscache_n_cache_retired_objects);
1073                 break;
1074         case FSCACHE_OBJECT_WAS_CULLED:
1075                 fscache_stat(&fscache_n_cache_culled_objects);
1076                 break;
1077         }
1078 }
1079 EXPORT_SYMBOL(fscache_object_mark_killed);