]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
cf45b752 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
d77d1b58 JP |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
11 | ||
b3b94faa DT |
12 | #include <linux/sched.h> |
13 | #include <linux/slab.h> | |
14 | #include <linux/spinlock.h> | |
b3b94faa DT |
15 | #include <linux/buffer_head.h> |
16 | #include <linux/delay.h> | |
17 | #include <linux/sort.h> | |
0515480a | 18 | #include <linux/hash.h> |
b3b94faa | 19 | #include <linux/jhash.h> |
d0dc80db | 20 | #include <linux/kallsyms.h> |
5c676f6d | 21 | #include <linux/gfs2_ondisk.h> |
24264434 | 22 | #include <linux/list.h> |
fee852e3 | 23 | #include <linux/wait.h> |
95d97b7d | 24 | #include <linux/module.h> |
7c0f6ba6 | 25 | #include <linux/uaccess.h> |
7c52b166 RP |
26 | #include <linux/seq_file.h> |
27 | #include <linux/debugfs.h> | |
8fbbfd21 SW |
28 | #include <linux/kthread.h> |
29 | #include <linux/freezer.h> | |
c4f68a13 BM |
30 | #include <linux/workqueue.h> |
31 | #include <linux/jiffies.h> | |
bc015cb8 SW |
32 | #include <linux/rcupdate.h> |
33 | #include <linux/rculist_bl.h> | |
34 | #include <linux/bit_spinlock.h> | |
a245769f | 35 | #include <linux/percpu.h> |
4506a519 | 36 | #include <linux/list_sort.h> |
e66cf161 | 37 | #include <linux/lockref.h> |
88ffbf3e | 38 | #include <linux/rhashtable.h> |
b3b94faa DT |
39 | |
40 | #include "gfs2.h" | |
5c676f6d | 41 | #include "incore.h" |
b3b94faa DT |
42 | #include "glock.h" |
43 | #include "glops.h" | |
44 | #include "inode.h" | |
b3b94faa DT |
45 | #include "lops.h" |
46 | #include "meta_io.h" | |
47 | #include "quota.h" | |
48 | #include "super.h" | |
5c676f6d | 49 | #include "util.h" |
813e0c46 | 50 | #include "bmap.h" |
63997775 SW |
51 | #define CREATE_TRACE_POINTS |
52 | #include "trace_gfs2.h" | |
b3b94faa | 53 | |
6802e340 | 54 | struct gfs2_glock_iter { |
ba1ddcb6 | 55 | struct gfs2_sbd *sdp; /* incore superblock */ |
88ffbf3e | 56 | struct rhashtable_iter hti; /* rhashtable iterator */ |
ba1ddcb6 SW |
57 | struct gfs2_glock *gl; /* current glock struct */ |
58 | loff_t last_pos; /* last position */ | |
7c52b166 RP |
59 | }; |
60 | ||
b3b94faa DT |
61 | typedef void (*glock_examiner) (struct gfs2_glock * gl); |
62 | ||
6802e340 | 63 | static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target); |
c4f68a13 | 64 | |
7c52b166 | 65 | static struct dentry *gfs2_root; |
c4f68a13 | 66 | static struct workqueue_struct *glock_workqueue; |
b94a170e | 67 | struct workqueue_struct *gfs2_delete_workqueue; |
97cc1025 SW |
68 | static LIST_HEAD(lru_list); |
69 | static atomic_t lru_count = ATOMIC_INIT(0); | |
eb8374e7 | 70 | static DEFINE_SPINLOCK(lru_lock); |
08bc2dbc | 71 | |
b6397893 | 72 | #define GFS2_GL_HASH_SHIFT 15 |
47a9a527 | 73 | #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT) |
087efdd3 | 74 | |
d296b15e | 75 | static const struct rhashtable_params ht_parms = { |
88ffbf3e | 76 | .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4, |
972b044e | 77 | .key_len = offsetofend(struct lm_lockname, ln_type), |
88ffbf3e BP |
78 | .key_offset = offsetof(struct gfs2_glock, gl_name), |
79 | .head_offset = offsetof(struct gfs2_glock, gl_node), | |
80 | }; | |
b3b94faa | 81 | |
88ffbf3e | 82 | static struct rhashtable gl_hash_table; |
b3b94faa | 83 | |
0515480a AG |
84 | #define GLOCK_WAIT_TABLE_BITS 12 |
85 | #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS) | |
86 | static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned; | |
87 | ||
88 | struct wait_glock_queue { | |
89 | struct lm_lockname *name; | |
90 | wait_queue_entry_t wait; | |
91 | }; | |
92 | ||
93 | static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode, | |
94 | int sync, void *key) | |
95 | { | |
96 | struct wait_glock_queue *wait_glock = | |
97 | container_of(wait, struct wait_glock_queue, wait); | |
98 | struct lm_lockname *wait_name = wait_glock->name; | |
99 | struct lm_lockname *wake_name = key; | |
100 | ||
101 | if (wake_name->ln_sbd != wait_name->ln_sbd || | |
102 | wake_name->ln_number != wait_name->ln_number || | |
103 | wake_name->ln_type != wait_name->ln_type) | |
104 | return 0; | |
105 | return autoremove_wake_function(wait, mode, sync, key); | |
106 | } | |
107 | ||
108 | static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name) | |
109 | { | |
605b0487 | 110 | u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0); |
0515480a AG |
111 | |
112 | return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS); | |
113 | } | |
114 | ||
0515480a AG |
115 | /** |
116 | * wake_up_glock - Wake up waiters on a glock | |
117 | * @gl: the glock | |
118 | */ | |
119 | static void wake_up_glock(struct gfs2_glock *gl) | |
120 | { | |
121 | wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name); | |
122 | ||
123 | if (waitqueue_active(wq)) | |
124 | __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name); | |
125 | } | |
126 | ||
961ae1d8 | 127 | static void gfs2_glock_dealloc(struct rcu_head *rcu) |
b3b94faa | 128 | { |
961ae1d8 | 129 | struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu); |
b3b94faa | 130 | |
dba2d70c | 131 | if (gl->gl_ops->go_flags & GLOF_ASPACE) { |
bc015cb8 | 132 | kmem_cache_free(gfs2_glock_aspace_cachep, gl); |
dba2d70c | 133 | } else { |
4e2f8849 | 134 | kfree(gl->gl_lksb.sb_lvbptr); |
bc015cb8 | 135 | kmem_cache_free(gfs2_glock_cachep, gl); |
dba2d70c | 136 | } |
961ae1d8 AG |
137 | } |
138 | ||
139 | void gfs2_glock_free(struct gfs2_glock *gl) | |
140 | { | |
141 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
142 | ||
0515480a AG |
143 | rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms); |
144 | smp_mb(); | |
145 | wake_up_glock(gl); | |
961ae1d8 | 146 | call_rcu(&gl->gl_rcu, gfs2_glock_dealloc); |
bc015cb8 SW |
147 | if (atomic_dec_and_test(&sdp->sd_glock_disposal)) |
148 | wake_up(&sdp->sd_glock_wait); | |
b3b94faa DT |
149 | } |
150 | ||
151 | /** | |
152 | * gfs2_glock_hold() - increment reference count on glock | |
153 | * @gl: The glock to hold | |
154 | * | |
155 | */ | |
156 | ||
71c1b213 | 157 | void gfs2_glock_hold(struct gfs2_glock *gl) |
b3b94faa | 158 | { |
e66cf161 SW |
159 | GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref)); |
160 | lockref_get(&gl->gl_lockref); | |
b3b94faa DT |
161 | } |
162 | ||
8ff22a6f BM |
163 | /** |
164 | * demote_ok - Check to see if it's ok to unlock a glock | |
165 | * @gl: the glock | |
166 | * | |
167 | * Returns: 1 if it's ok | |
168 | */ | |
169 | ||
170 | static int demote_ok(const struct gfs2_glock *gl) | |
171 | { | |
172 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
173 | ||
174 | if (gl->gl_state == LM_ST_UNLOCKED) | |
175 | return 0; | |
f42ab085 | 176 | if (!list_empty(&gl->gl_holders)) |
8ff22a6f BM |
177 | return 0; |
178 | if (glops->go_demote_ok) | |
179 | return glops->go_demote_ok(gl); | |
180 | return 1; | |
181 | } | |
182 | ||
bc015cb8 | 183 | |
29687a2a SW |
184 | void gfs2_glock_add_to_lru(struct gfs2_glock *gl) |
185 | { | |
186 | spin_lock(&lru_lock); | |
187 | ||
188 | if (!list_empty(&gl->gl_lru)) | |
189 | list_del_init(&gl->gl_lru); | |
190 | else | |
191 | atomic_inc(&lru_count); | |
192 | ||
193 | list_add_tail(&gl->gl_lru, &lru_list); | |
627c10b7 | 194 | set_bit(GLF_LRU, &gl->gl_flags); |
29687a2a SW |
195 | spin_unlock(&lru_lock); |
196 | } | |
197 | ||
8f6cb409 | 198 | static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl) |
f42ab085 | 199 | { |
645ebd49 BP |
200 | if (!(gl->gl_ops->go_flags & GLOF_LRU)) |
201 | return; | |
202 | ||
8f6cb409 | 203 | spin_lock(&lru_lock); |
f42ab085 SW |
204 | if (!list_empty(&gl->gl_lru)) { |
205 | list_del_init(&gl->gl_lru); | |
206 | atomic_dec(&lru_count); | |
207 | clear_bit(GLF_LRU, &gl->gl_flags); | |
208 | } | |
209 | spin_unlock(&lru_lock); | |
210 | } | |
211 | ||
6b0c7440 AG |
212 | /* |
213 | * Enqueue the glock on the work queue. Passes one glock reference on to the | |
214 | * work queue. | |
b3b94faa | 215 | */ |
6b0c7440 AG |
216 | static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) { |
217 | if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) { | |
218 | /* | |
219 | * We are holding the lockref spinlock, and the work was still | |
220 | * queued above. The queued work (glock_work_func) takes that | |
221 | * spinlock before dropping its glock reference(s), so it | |
222 | * cannot have dropped them in the meantime. | |
223 | */ | |
224 | GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2); | |
225 | gl->gl_lockref.count--; | |
226 | } | |
227 | } | |
b3b94faa | 228 | |
6b0c7440 AG |
229 | static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) { |
230 | spin_lock(&gl->gl_lockref.lock); | |
231 | __gfs2_glock_queue_work(gl, delay); | |
232 | spin_unlock(&gl->gl_lockref.lock); | |
233 | } | |
234 | ||
235 | static void __gfs2_glock_put(struct gfs2_glock *gl) | |
b3b94faa | 236 | { |
15562c43 | 237 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
bc015cb8 | 238 | struct address_space *mapping = gfs2_glock2aspace(gl); |
b3b94faa | 239 | |
e66cf161 SW |
240 | lockref_mark_dead(&gl->gl_lockref); |
241 | ||
8f6cb409 | 242 | gfs2_glock_remove_from_lru(gl); |
e66cf161 | 243 | spin_unlock(&gl->gl_lockref.lock); |
e66cf161 SW |
244 | GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders)); |
245 | GLOCK_BUG_ON(gl, mapping && mapping->nrpages); | |
246 | trace_gfs2_glock_put(gl); | |
247 | sdp->sd_lockstruct.ls_ops->lm_put_lock(gl); | |
b3b94faa DT |
248 | } |
249 | ||
71c1b213 AG |
250 | /* |
251 | * Cause the glock to be put in work queue context. | |
252 | */ | |
253 | void gfs2_glock_queue_put(struct gfs2_glock *gl) | |
254 | { | |
255 | gfs2_glock_queue_work(gl, 0); | |
256 | } | |
257 | ||
6b0c7440 AG |
258 | /** |
259 | * gfs2_glock_put() - Decrement reference count on glock | |
260 | * @gl: The glock to put | |
261 | * | |
262 | */ | |
263 | ||
264 | void gfs2_glock_put(struct gfs2_glock *gl) | |
265 | { | |
266 | if (lockref_put_or_lock(&gl->gl_lockref)) | |
267 | return; | |
268 | ||
269 | __gfs2_glock_put(gl); | |
270 | } | |
271 | ||
6802e340 SW |
272 | /** |
273 | * may_grant - check if its ok to grant a new lock | |
274 | * @gl: The glock | |
275 | * @gh: The lock request which we wish to grant | |
276 | * | |
277 | * Returns: true if its ok to grant the lock | |
278 | */ | |
279 | ||
280 | static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh) | |
281 | { | |
282 | const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list); | |
283 | if ((gh->gh_state == LM_ST_EXCLUSIVE || | |
284 | gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head) | |
285 | return 0; | |
286 | if (gl->gl_state == gh->gh_state) | |
287 | return 1; | |
288 | if (gh->gh_flags & GL_EXACT) | |
289 | return 0; | |
209806ab SW |
290 | if (gl->gl_state == LM_ST_EXCLUSIVE) { |
291 | if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED) | |
292 | return 1; | |
293 | if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED) | |
294 | return 1; | |
295 | } | |
6802e340 SW |
296 | if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY)) |
297 | return 1; | |
298 | return 0; | |
299 | } | |
300 | ||
301 | static void gfs2_holder_wake(struct gfs2_holder *gh) | |
302 | { | |
303 | clear_bit(HIF_WAIT, &gh->gh_iflags); | |
4e857c58 | 304 | smp_mb__after_atomic(); |
6802e340 SW |
305 | wake_up_bit(&gh->gh_iflags, HIF_WAIT); |
306 | } | |
307 | ||
d5341a92 SW |
308 | /** |
309 | * do_error - Something unexpected has happened during a lock request | |
310 | * | |
311 | */ | |
312 | ||
a527b38e | 313 | static void do_error(struct gfs2_glock *gl, const int ret) |
d5341a92 SW |
314 | { |
315 | struct gfs2_holder *gh, *tmp; | |
316 | ||
317 | list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) { | |
318 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
319 | continue; | |
320 | if (ret & LM_OUT_ERROR) | |
321 | gh->gh_error = -EIO; | |
322 | else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) | |
323 | gh->gh_error = GLR_TRYFAILED; | |
324 | else | |
325 | continue; | |
326 | list_del_init(&gh->gh_list); | |
327 | trace_gfs2_glock_queue(gh, 0); | |
328 | gfs2_holder_wake(gh); | |
329 | } | |
330 | } | |
331 | ||
6802e340 SW |
332 | /** |
333 | * do_promote - promote as many requests as possible on the current queue | |
334 | * @gl: The glock | |
335 | * | |
813e0c46 SW |
336 | * Returns: 1 if there is a blocked holder at the head of the list, or 2 |
337 | * if a type specific operation is underway. | |
6802e340 SW |
338 | */ |
339 | ||
340 | static int do_promote(struct gfs2_glock *gl) | |
f3dd1649 AG |
341 | __releases(&gl->gl_lockref.lock) |
342 | __acquires(&gl->gl_lockref.lock) | |
6802e340 SW |
343 | { |
344 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
345 | struct gfs2_holder *gh, *tmp; | |
346 | int ret; | |
347 | ||
348 | restart: | |
349 | list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) { | |
350 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
351 | continue; | |
352 | if (may_grant(gl, gh)) { | |
353 | if (gh->gh_list.prev == &gl->gl_holders && | |
354 | glops->go_lock) { | |
f3dd1649 | 355 | spin_unlock(&gl->gl_lockref.lock); |
6802e340 SW |
356 | /* FIXME: eliminate this eventually */ |
357 | ret = glops->go_lock(gh); | |
f3dd1649 | 358 | spin_lock(&gl->gl_lockref.lock); |
6802e340 | 359 | if (ret) { |
813e0c46 SW |
360 | if (ret == 1) |
361 | return 2; | |
6802e340 SW |
362 | gh->gh_error = ret; |
363 | list_del_init(&gh->gh_list); | |
63997775 | 364 | trace_gfs2_glock_queue(gh, 0); |
6802e340 SW |
365 | gfs2_holder_wake(gh); |
366 | goto restart; | |
367 | } | |
368 | set_bit(HIF_HOLDER, &gh->gh_iflags); | |
63997775 | 369 | trace_gfs2_promote(gh, 1); |
6802e340 SW |
370 | gfs2_holder_wake(gh); |
371 | goto restart; | |
372 | } | |
373 | set_bit(HIF_HOLDER, &gh->gh_iflags); | |
63997775 | 374 | trace_gfs2_promote(gh, 0); |
6802e340 SW |
375 | gfs2_holder_wake(gh); |
376 | continue; | |
377 | } | |
378 | if (gh->gh_list.prev == &gl->gl_holders) | |
379 | return 1; | |
d5341a92 | 380 | do_error(gl, 0); |
6802e340 SW |
381 | break; |
382 | } | |
383 | return 0; | |
384 | } | |
385 | ||
6802e340 SW |
386 | /** |
387 | * find_first_waiter - find the first gh that's waiting for the glock | |
388 | * @gl: the glock | |
389 | */ | |
390 | ||
391 | static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl) | |
392 | { | |
393 | struct gfs2_holder *gh; | |
394 | ||
395 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { | |
396 | if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
397 | return gh; | |
398 | } | |
399 | return NULL; | |
400 | } | |
401 | ||
402 | /** | |
403 | * state_change - record that the glock is now in a different state | |
404 | * @gl: the glock | |
405 | * @new_state the new state | |
406 | * | |
407 | */ | |
408 | ||
409 | static void state_change(struct gfs2_glock *gl, unsigned int new_state) | |
410 | { | |
411 | int held1, held2; | |
412 | ||
413 | held1 = (gl->gl_state != LM_ST_UNLOCKED); | |
414 | held2 = (new_state != LM_ST_UNLOCKED); | |
415 | ||
416 | if (held1 != held2) { | |
e66cf161 | 417 | GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref)); |
6802e340 | 418 | if (held2) |
e66cf161 | 419 | gl->gl_lockref.count++; |
6802e340 | 420 | else |
e66cf161 | 421 | gl->gl_lockref.count--; |
6802e340 | 422 | } |
7b5e3d5f SW |
423 | if (held1 && held2 && list_empty(&gl->gl_holders)) |
424 | clear_bit(GLF_QUEUED, &gl->gl_flags); | |
6802e340 | 425 | |
7cf8dcd3 BP |
426 | if (new_state != gl->gl_target) |
427 | /* shorten our minimum hold time */ | |
428 | gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR, | |
429 | GL_GLOCK_MIN_HOLD); | |
6802e340 SW |
430 | gl->gl_state = new_state; |
431 | gl->gl_tchange = jiffies; | |
432 | } | |
433 | ||
434 | static void gfs2_demote_wake(struct gfs2_glock *gl) | |
435 | { | |
436 | gl->gl_demote_state = LM_ST_EXCLUSIVE; | |
437 | clear_bit(GLF_DEMOTE, &gl->gl_flags); | |
4e857c58 | 438 | smp_mb__after_atomic(); |
6802e340 SW |
439 | wake_up_bit(&gl->gl_flags, GLF_DEMOTE); |
440 | } | |
441 | ||
442 | /** | |
443 | * finish_xmote - The DLM has replied to one of our lock requests | |
444 | * @gl: The glock | |
445 | * @ret: The status from the DLM | |
446 | * | |
447 | */ | |
448 | ||
449 | static void finish_xmote(struct gfs2_glock *gl, unsigned int ret) | |
450 | { | |
451 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
452 | struct gfs2_holder *gh; | |
453 | unsigned state = ret & LM_OUT_ST_MASK; | |
813e0c46 | 454 | int rv; |
6802e340 | 455 | |
f3dd1649 | 456 | spin_lock(&gl->gl_lockref.lock); |
63997775 | 457 | trace_gfs2_glock_state_change(gl, state); |
6802e340 SW |
458 | state_change(gl, state); |
459 | gh = find_first_waiter(gl); | |
460 | ||
461 | /* Demote to UN request arrived during demote to SH or DF */ | |
462 | if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) && | |
463 | state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED) | |
464 | gl->gl_target = LM_ST_UNLOCKED; | |
465 | ||
466 | /* Check for state != intended state */ | |
467 | if (unlikely(state != gl->gl_target)) { | |
468 | if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) { | |
469 | /* move to back of queue and try next entry */ | |
470 | if (ret & LM_OUT_CANCELED) { | |
471 | if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0) | |
472 | list_move_tail(&gh->gh_list, &gl->gl_holders); | |
473 | gh = find_first_waiter(gl); | |
474 | gl->gl_target = gh->gh_state; | |
475 | goto retry; | |
476 | } | |
477 | /* Some error or failed "try lock" - report it */ | |
478 | if ((ret & LM_OUT_ERROR) || | |
479 | (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) { | |
480 | gl->gl_target = gl->gl_state; | |
481 | do_error(gl, ret); | |
482 | goto out; | |
483 | } | |
484 | } | |
485 | switch(state) { | |
486 | /* Unlocked due to conversion deadlock, try again */ | |
487 | case LM_ST_UNLOCKED: | |
488 | retry: | |
489 | do_xmote(gl, gh, gl->gl_target); | |
490 | break; | |
491 | /* Conversion fails, unlock and try again */ | |
492 | case LM_ST_SHARED: | |
493 | case LM_ST_DEFERRED: | |
494 | do_xmote(gl, gh, LM_ST_UNLOCKED); | |
495 | break; | |
496 | default: /* Everything else */ | |
e54c78a2 BP |
497 | fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n", |
498 | gl->gl_target, state); | |
6802e340 SW |
499 | GLOCK_BUG_ON(gl, 1); |
500 | } | |
f3dd1649 | 501 | spin_unlock(&gl->gl_lockref.lock); |
6802e340 SW |
502 | return; |
503 | } | |
504 | ||
505 | /* Fast path - we got what we asked for */ | |
506 | if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) | |
507 | gfs2_demote_wake(gl); | |
508 | if (state != LM_ST_UNLOCKED) { | |
509 | if (glops->go_xmote_bh) { | |
f3dd1649 | 510 | spin_unlock(&gl->gl_lockref.lock); |
6802e340 | 511 | rv = glops->go_xmote_bh(gl, gh); |
f3dd1649 | 512 | spin_lock(&gl->gl_lockref.lock); |
6802e340 SW |
513 | if (rv) { |
514 | do_error(gl, rv); | |
515 | goto out; | |
516 | } | |
517 | } | |
813e0c46 SW |
518 | rv = do_promote(gl); |
519 | if (rv == 2) | |
520 | goto out_locked; | |
6802e340 SW |
521 | } |
522 | out: | |
523 | clear_bit(GLF_LOCK, &gl->gl_flags); | |
813e0c46 | 524 | out_locked: |
f3dd1649 | 525 | spin_unlock(&gl->gl_lockref.lock); |
6802e340 SW |
526 | } |
527 | ||
6802e340 SW |
528 | /** |
529 | * do_xmote - Calls the DLM to change the state of a lock | |
530 | * @gl: The lock state | |
531 | * @gh: The holder (only for promotes) | |
532 | * @target: The target lock state | |
533 | * | |
534 | */ | |
535 | ||
536 | static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target) | |
f3dd1649 AG |
537 | __releases(&gl->gl_lockref.lock) |
538 | __acquires(&gl->gl_lockref.lock) | |
6802e340 SW |
539 | { |
540 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
15562c43 | 541 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
b58bf407 | 542 | unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0); |
6802e340 SW |
543 | int ret; |
544 | ||
ed17545d BP |
545 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) && |
546 | target != LM_ST_UNLOCKED) | |
0d1c7ae9 | 547 | return; |
6802e340 SW |
548 | lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP | |
549 | LM_FLAG_PRIORITY); | |
921169ca SW |
550 | GLOCK_BUG_ON(gl, gl->gl_state == target); |
551 | GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target); | |
6802e340 SW |
552 | if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) && |
553 | glops->go_inval) { | |
554 | set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags); | |
555 | do_error(gl, 0); /* Fail queued try locks */ | |
556 | } | |
47a25380 | 557 | gl->gl_req = target; |
a245769f SW |
558 | set_bit(GLF_BLOCKING, &gl->gl_flags); |
559 | if ((gl->gl_req == LM_ST_UNLOCKED) || | |
560 | (gl->gl_state == LM_ST_EXCLUSIVE) || | |
561 | (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB))) | |
562 | clear_bit(GLF_BLOCKING, &gl->gl_flags); | |
f3dd1649 | 563 | spin_unlock(&gl->gl_lockref.lock); |
06dfc306 BP |
564 | if (glops->go_sync) |
565 | glops->go_sync(gl); | |
6802e340 SW |
566 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) |
567 | glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA); | |
568 | clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags); | |
569 | ||
570 | gfs2_glock_hold(gl); | |
921169ca SW |
571 | if (sdp->sd_lockstruct.ls_ops->lm_lock) { |
572 | /* lock_dlm */ | |
573 | ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags); | |
3e11e530 BM |
574 | if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED && |
575 | target == LM_ST_UNLOCKED && | |
576 | test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) { | |
577 | finish_xmote(gl, target); | |
6b0c7440 | 578 | gfs2_glock_queue_work(gl, 0); |
3e11e530 BM |
579 | } |
580 | else if (ret) { | |
e54c78a2 | 581 | fs_err(sdp, "lm_lock ret %d\n", ret); |
ed17545d BP |
582 | GLOCK_BUG_ON(gl, !test_bit(SDF_SHUTDOWN, |
583 | &sdp->sd_flags)); | |
dba2d70c | 584 | } |
921169ca SW |
585 | } else { /* lock_nolock */ |
586 | finish_xmote(gl, target); | |
6b0c7440 | 587 | gfs2_glock_queue_work(gl, 0); |
6802e340 | 588 | } |
921169ca | 589 | |
f3dd1649 | 590 | spin_lock(&gl->gl_lockref.lock); |
6802e340 SW |
591 | } |
592 | ||
593 | /** | |
594 | * find_first_holder - find the first "holder" gh | |
595 | * @gl: the glock | |
596 | */ | |
597 | ||
598 | static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl) | |
599 | { | |
600 | struct gfs2_holder *gh; | |
601 | ||
602 | if (!list_empty(&gl->gl_holders)) { | |
603 | gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list); | |
604 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
605 | return gh; | |
606 | } | |
607 | return NULL; | |
608 | } | |
609 | ||
610 | /** | |
611 | * run_queue - do all outstanding tasks related to a glock | |
612 | * @gl: The glock in question | |
613 | * @nonblock: True if we must not block in run_queue | |
614 | * | |
615 | */ | |
616 | ||
617 | static void run_queue(struct gfs2_glock *gl, const int nonblock) | |
f3dd1649 AG |
618 | __releases(&gl->gl_lockref.lock) |
619 | __acquires(&gl->gl_lockref.lock) | |
6802e340 SW |
620 | { |
621 | struct gfs2_holder *gh = NULL; | |
813e0c46 | 622 | int ret; |
6802e340 SW |
623 | |
624 | if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) | |
625 | return; | |
626 | ||
627 | GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)); | |
628 | ||
629 | if (test_bit(GLF_DEMOTE, &gl->gl_flags) && | |
630 | gl->gl_demote_state != gl->gl_state) { | |
631 | if (find_first_holder(gl)) | |
d8348de0 | 632 | goto out_unlock; |
6802e340 SW |
633 | if (nonblock) |
634 | goto out_sched; | |
635 | set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags); | |
265d529c | 636 | GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE); |
6802e340 SW |
637 | gl->gl_target = gl->gl_demote_state; |
638 | } else { | |
639 | if (test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
640 | gfs2_demote_wake(gl); | |
813e0c46 SW |
641 | ret = do_promote(gl); |
642 | if (ret == 0) | |
d8348de0 | 643 | goto out_unlock; |
813e0c46 | 644 | if (ret == 2) |
a228df63 | 645 | goto out; |
6802e340 SW |
646 | gh = find_first_waiter(gl); |
647 | gl->gl_target = gh->gh_state; | |
648 | if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) | |
649 | do_error(gl, 0); /* Fail queued try locks */ | |
650 | } | |
651 | do_xmote(gl, gh, gl->gl_target); | |
a228df63 | 652 | out: |
6802e340 SW |
653 | return; |
654 | ||
655 | out_sched: | |
7e71c55e | 656 | clear_bit(GLF_LOCK, &gl->gl_flags); |
4e857c58 | 657 | smp_mb__after_atomic(); |
e66cf161 | 658 | gl->gl_lockref.count++; |
6b0c7440 | 659 | __gfs2_glock_queue_work(gl, 0); |
7e71c55e SW |
660 | return; |
661 | ||
d8348de0 | 662 | out_unlock: |
6802e340 | 663 | clear_bit(GLF_LOCK, &gl->gl_flags); |
4e857c58 | 664 | smp_mb__after_atomic(); |
7e71c55e | 665 | return; |
6802e340 SW |
666 | } |
667 | ||
b94a170e BM |
668 | static void delete_work_func(struct work_struct *work) |
669 | { | |
670 | struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete); | |
15562c43 | 671 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
ec5ec66b | 672 | struct inode *inode; |
044b9414 SW |
673 | u64 no_addr = gl->gl_name.ln_number; |
674 | ||
a4923865 BP |
675 | /* If someone's using this glock to create a new dinode, the block must |
676 | have been freed by another node, then re-used, in which case our | |
677 | iopen callback is too late after the fact. Ignore it. */ | |
678 | if (test_bit(GLF_INODE_CREATING, &gl->gl_flags)) | |
679 | goto out; | |
680 | ||
ec5ec66b | 681 | inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED); |
044b9414 SW |
682 | if (inode && !IS_ERR(inode)) { |
683 | d_prune_aliases(inode); | |
684 | iput(inode); | |
b94a170e | 685 | } |
a4923865 | 686 | out: |
b94a170e BM |
687 | gfs2_glock_put(gl); |
688 | } | |
689 | ||
c4f68a13 BM |
690 | static void glock_work_func(struct work_struct *work) |
691 | { | |
6802e340 | 692 | unsigned long delay = 0; |
c4f68a13 | 693 | struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work); |
6b0c7440 | 694 | unsigned int drop_refs = 1; |
c4f68a13 | 695 | |
26bb7505 | 696 | if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) { |
6802e340 | 697 | finish_xmote(gl, gl->gl_reply); |
6b0c7440 | 698 | drop_refs++; |
26bb7505 | 699 | } |
f3dd1649 | 700 | spin_lock(&gl->gl_lockref.lock); |
f90e5b5b | 701 | if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) && |
265d529c SW |
702 | gl->gl_state != LM_ST_UNLOCKED && |
703 | gl->gl_demote_state != LM_ST_EXCLUSIVE) { | |
6802e340 | 704 | unsigned long holdtime, now = jiffies; |
f90e5b5b | 705 | |
7cf8dcd3 | 706 | holdtime = gl->gl_tchange + gl->gl_hold_time; |
6802e340 SW |
707 | if (time_before(now, holdtime)) |
708 | delay = holdtime - now; | |
f90e5b5b BP |
709 | |
710 | if (!delay) { | |
711 | clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags); | |
712 | set_bit(GLF_DEMOTE, &gl->gl_flags); | |
713 | } | |
6802e340 SW |
714 | } |
715 | run_queue(gl, 0); | |
6b0c7440 AG |
716 | if (delay) { |
717 | /* Keep one glock reference for the work we requeue. */ | |
718 | drop_refs--; | |
7cf8dcd3 BP |
719 | if (gl->gl_name.ln_type != LM_TYPE_INODE) |
720 | delay = 0; | |
6b0c7440 | 721 | __gfs2_glock_queue_work(gl, delay); |
7cf8dcd3 | 722 | } |
6b0c7440 AG |
723 | |
724 | /* | |
725 | * Drop the remaining glock references manually here. (Mind that | |
726 | * __gfs2_glock_queue_work depends on the lockref spinlock begin held | |
727 | * here as well.) | |
728 | */ | |
729 | gl->gl_lockref.count -= drop_refs; | |
730 | if (!gl->gl_lockref.count) { | |
731 | __gfs2_glock_put(gl); | |
732 | return; | |
733 | } | |
734 | spin_unlock(&gl->gl_lockref.lock); | |
c4f68a13 BM |
735 | } |
736 | ||
0515480a AG |
737 | static struct gfs2_glock *find_insert_glock(struct lm_lockname *name, |
738 | struct gfs2_glock *new) | |
739 | { | |
740 | struct wait_glock_queue wait; | |
a91323e2 | 741 | wait_queue_head_t *wq = glock_waitqueue(name); |
0515480a AG |
742 | struct gfs2_glock *gl; |
743 | ||
a91323e2 AG |
744 | wait.name = name; |
745 | init_wait(&wait.wait); | |
746 | wait.wait.func = glock_wake_function; | |
747 | ||
0515480a | 748 | again: |
a91323e2 | 749 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
0515480a AG |
750 | rcu_read_lock(); |
751 | if (new) { | |
752 | gl = rhashtable_lookup_get_insert_fast(&gl_hash_table, | |
753 | &new->gl_node, ht_parms); | |
754 | if (IS_ERR(gl)) | |
755 | goto out; | |
756 | } else { | |
757 | gl = rhashtable_lookup_fast(&gl_hash_table, | |
758 | name, ht_parms); | |
759 | } | |
760 | if (gl && !lockref_get_not_dead(&gl->gl_lockref)) { | |
761 | rcu_read_unlock(); | |
762 | schedule(); | |
763 | goto again; | |
764 | } | |
765 | out: | |
766 | rcu_read_unlock(); | |
a91323e2 | 767 | finish_wait(wq, &wait.wait); |
0515480a AG |
768 | return gl; |
769 | } | |
770 | ||
b3b94faa DT |
771 | /** |
772 | * gfs2_glock_get() - Get a glock, or create one if one doesn't exist | |
773 | * @sdp: The GFS2 superblock | |
774 | * @number: the lock number | |
775 | * @glops: The glock_operations to use | |
776 | * @create: If 0, don't create the glock if it doesn't exist | |
777 | * @glp: the glock is returned here | |
778 | * | |
779 | * This does not lock a glock, just finds/creates structures for one. | |
780 | * | |
781 | * Returns: errno | |
782 | */ | |
783 | ||
cd915493 | 784 | int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number, |
8fb4b536 | 785 | const struct gfs2_glock_operations *glops, int create, |
b3b94faa DT |
786 | struct gfs2_glock **glp) |
787 | { | |
009d8518 | 788 | struct super_block *s = sdp->sd_vfs; |
15562c43 BP |
789 | struct lm_lockname name = { .ln_number = number, |
790 | .ln_type = glops->go_type, | |
791 | .ln_sbd = sdp }; | |
0a52aba7 | 792 | struct gfs2_glock *gl, *tmp; |
009d8518 | 793 | struct address_space *mapping; |
bc015cb8 | 794 | struct kmem_cache *cachep; |
0a52aba7 | 795 | int ret = 0; |
b3b94faa | 796 | |
0515480a AG |
797 | gl = find_insert_glock(&name, NULL); |
798 | if (gl) { | |
799 | *glp = gl; | |
b3b94faa | 800 | return 0; |
0515480a | 801 | } |
64d576ba SW |
802 | if (!create) |
803 | return -ENOENT; | |
b3b94faa | 804 | |
009d8518 | 805 | if (glops->go_flags & GLOF_ASPACE) |
bc015cb8 | 806 | cachep = gfs2_glock_aspace_cachep; |
009d8518 | 807 | else |
bc015cb8 | 808 | cachep = gfs2_glock_cachep; |
fe0bbd29 | 809 | gl = kmem_cache_alloc(cachep, GFP_NOFS); |
b3b94faa DT |
810 | if (!gl) |
811 | return -ENOMEM; | |
812 | ||
dba2d70c | 813 | memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb)); |
dba2d70c DT |
814 | |
815 | if (glops->go_flags & GLOF_LVB) { | |
fe0bbd29 | 816 | gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS); |
4e2f8849 | 817 | if (!gl->gl_lksb.sb_lvbptr) { |
dba2d70c DT |
818 | kmem_cache_free(cachep, gl); |
819 | return -ENOMEM; | |
820 | } | |
dba2d70c DT |
821 | } |
822 | ||
8f05228e | 823 | atomic_inc(&sdp->sd_glock_disposal); |
88ffbf3e | 824 | gl->gl_node.next = NULL; |
ec45d9f5 | 825 | gl->gl_flags = 0; |
b3b94faa | 826 | gl->gl_name = name; |
e66cf161 | 827 | gl->gl_lockref.count = 1; |
b3b94faa | 828 | gl->gl_state = LM_ST_UNLOCKED; |
6802e340 | 829 | gl->gl_target = LM_ST_UNLOCKED; |
c4f68a13 | 830 | gl->gl_demote_state = LM_ST_EXCLUSIVE; |
b3b94faa | 831 | gl->gl_ops = glops; |
8b0e1953 | 832 | gl->gl_dstamp = 0; |
a245769f SW |
833 | preempt_disable(); |
834 | /* We use the global stats to estimate the initial per-glock stats */ | |
835 | gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type]; | |
836 | preempt_enable(); | |
837 | gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0; | |
838 | gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0; | |
c4f68a13 | 839 | gl->gl_tchange = jiffies; |
ec45d9f5 | 840 | gl->gl_object = NULL; |
7cf8dcd3 | 841 | gl->gl_hold_time = GL_GLOCK_DFT_HOLD; |
c4f68a13 | 842 | INIT_DELAYED_WORK(&gl->gl_work, glock_work_func); |
b94a170e | 843 | INIT_WORK(&gl->gl_delete, delete_work_func); |
b3b94faa | 844 | |
009d8518 SW |
845 | mapping = gfs2_glock2aspace(gl); |
846 | if (mapping) { | |
847 | mapping->a_ops = &gfs2_meta_aops; | |
848 | mapping->host = s->s_bdev->bd_inode; | |
849 | mapping->flags = 0; | |
850 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
252aa6f5 | 851 | mapping->private_data = NULL; |
009d8518 | 852 | mapping->writeback_index = 0; |
b3b94faa DT |
853 | } |
854 | ||
0515480a | 855 | tmp = find_insert_glock(&name, gl); |
0a52aba7 | 856 | if (!tmp) { |
88ffbf3e | 857 | *glp = gl; |
0a52aba7 | 858 | goto out; |
b3b94faa | 859 | } |
0a52aba7 AG |
860 | if (IS_ERR(tmp)) { |
861 | ret = PTR_ERR(tmp); | |
862 | goto out_free; | |
863 | } | |
0515480a | 864 | *glp = tmp; |
0a52aba7 AG |
865 | |
866 | out_free: | |
88ffbf3e BP |
867 | kfree(gl->gl_lksb.sb_lvbptr); |
868 | kmem_cache_free(cachep, gl); | |
869 | atomic_dec(&sdp->sd_glock_disposal); | |
b3b94faa | 870 | |
0a52aba7 | 871 | out: |
88ffbf3e | 872 | return ret; |
b3b94faa DT |
873 | } |
874 | ||
875 | /** | |
876 | * gfs2_holder_init - initialize a struct gfs2_holder in the default way | |
877 | * @gl: the glock | |
878 | * @state: the state we're requesting | |
879 | * @flags: the modifier flags | |
880 | * @gh: the holder structure | |
881 | * | |
882 | */ | |
883 | ||
b58bf407 | 884 | void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags, |
b3b94faa DT |
885 | struct gfs2_holder *gh) |
886 | { | |
887 | INIT_LIST_HEAD(&gh->gh_list); | |
888 | gh->gh_gl = gl; | |
d29c0afe | 889 | gh->gh_ip = _RET_IP_; |
b1e058da | 890 | gh->gh_owner_pid = get_pid(task_pid(current)); |
b3b94faa DT |
891 | gh->gh_state = state; |
892 | gh->gh_flags = flags; | |
893 | gh->gh_error = 0; | |
894 | gh->gh_iflags = 0; | |
b3b94faa DT |
895 | gfs2_glock_hold(gl); |
896 | } | |
897 | ||
898 | /** | |
899 | * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it | |
900 | * @state: the state we're requesting | |
901 | * @flags: the modifier flags | |
902 | * @gh: the holder structure | |
903 | * | |
904 | * Don't mess with the glock. | |
905 | * | |
906 | */ | |
907 | ||
b58bf407 | 908 | void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh) |
b3b94faa DT |
909 | { |
910 | gh->gh_state = state; | |
579b78a4 | 911 | gh->gh_flags = flags; |
3b8249f6 | 912 | gh->gh_iflags = 0; |
d29c0afe | 913 | gh->gh_ip = _RET_IP_; |
30badc95 | 914 | put_pid(gh->gh_owner_pid); |
1a0eae88 | 915 | gh->gh_owner_pid = get_pid(task_pid(current)); |
b3b94faa DT |
916 | } |
917 | ||
918 | /** | |
919 | * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference) | |
920 | * @gh: the holder structure | |
921 | * | |
922 | */ | |
923 | ||
924 | void gfs2_holder_uninit(struct gfs2_holder *gh) | |
925 | { | |
b1e058da | 926 | put_pid(gh->gh_owner_pid); |
b3b94faa | 927 | gfs2_glock_put(gh->gh_gl); |
6df9f9a2 | 928 | gfs2_holder_mark_uninitialized(gh); |
d0dc80db | 929 | gh->gh_ip = 0; |
b3b94faa DT |
930 | } |
931 | ||
07a79049 BP |
932 | /** |
933 | * gfs2_glock_wait - wait on a glock acquisition | |
934 | * @gh: the glock holder | |
935 | * | |
936 | * Returns: 0 on success | |
937 | */ | |
938 | ||
939 | int gfs2_glock_wait(struct gfs2_holder *gh) | |
da755fdb | 940 | { |
7cf8dcd3 BP |
941 | unsigned long time1 = jiffies; |
942 | ||
6802e340 | 943 | might_sleep(); |
74316201 | 944 | wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE); |
7cf8dcd3 BP |
945 | if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */ |
946 | /* Lengthen the minimum hold time. */ | |
947 | gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time + | |
948 | GL_GLOCK_HOLD_INCR, | |
949 | GL_GLOCK_MAX_HOLD); | |
07a79049 | 950 | return gh->gh_error; |
da755fdb SW |
951 | } |
952 | ||
b3b94faa | 953 | /** |
6802e340 SW |
954 | * handle_callback - process a demote request |
955 | * @gl: the glock | |
956 | * @state: the state the caller wants us to change to | |
b3b94faa | 957 | * |
6802e340 SW |
958 | * There are only two requests that we are going to see in actual |
959 | * practise: LM_ST_SHARED and LM_ST_UNLOCKED | |
b3b94faa DT |
960 | */ |
961 | ||
6802e340 | 962 | static void handle_callback(struct gfs2_glock *gl, unsigned int state, |
81ffbf65 | 963 | unsigned long delay, bool remote) |
b3b94faa | 964 | { |
6802e340 | 965 | int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE; |
b3b94faa | 966 | |
6802e340 SW |
967 | set_bit(bit, &gl->gl_flags); |
968 | if (gl->gl_demote_state == LM_ST_EXCLUSIVE) { | |
969 | gl->gl_demote_state = state; | |
970 | gl->gl_demote_time = jiffies; | |
6802e340 SW |
971 | } else if (gl->gl_demote_state != LM_ST_UNLOCKED && |
972 | gl->gl_demote_state != state) { | |
973 | gl->gl_demote_state = LM_ST_UNLOCKED; | |
b3b94faa | 974 | } |
b94a170e | 975 | if (gl->gl_ops->go_callback) |
81ffbf65 | 976 | gl->gl_ops->go_callback(gl, remote); |
7bd8b2eb | 977 | trace_gfs2_demote_rq(gl, remote); |
b3b94faa DT |
978 | } |
979 | ||
6802e340 | 980 | void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...) |
7c52b166 | 981 | { |
5e69069c | 982 | struct va_format vaf; |
7c52b166 RP |
983 | va_list args; |
984 | ||
985 | va_start(args, fmt); | |
5e69069c | 986 | |
6802e340 | 987 | if (seq) { |
1bb49303 | 988 | seq_vprintf(seq, fmt, args); |
6802e340 | 989 | } else { |
5e69069c JP |
990 | vaf.fmt = fmt; |
991 | vaf.va = &args; | |
992 | ||
d77d1b58 | 993 | pr_err("%pV", &vaf); |
6802e340 | 994 | } |
5e69069c | 995 | |
7c52b166 RP |
996 | va_end(args); |
997 | } | |
998 | ||
b3b94faa DT |
999 | /** |
1000 | * add_to_queue - Add a holder to the wait queue (but look for recursion) | |
1001 | * @gh: the holder structure to add | |
1002 | * | |
6802e340 SW |
1003 | * Eventually we should move the recursive locking trap to a |
1004 | * debugging option or something like that. This is the fast | |
1005 | * path and needs to have the minimum number of distractions. | |
1006 | * | |
b3b94faa DT |
1007 | */ |
1008 | ||
6802e340 | 1009 | static inline void add_to_queue(struct gfs2_holder *gh) |
f3dd1649 AG |
1010 | __releases(&gl->gl_lockref.lock) |
1011 | __acquires(&gl->gl_lockref.lock) | |
b3b94faa DT |
1012 | { |
1013 | struct gfs2_glock *gl = gh->gh_gl; | |
15562c43 | 1014 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
6802e340 SW |
1015 | struct list_head *insert_pt = NULL; |
1016 | struct gfs2_holder *gh2; | |
e5dc76b9 | 1017 | int try_futile = 0; |
b3b94faa | 1018 | |
b1e058da | 1019 | BUG_ON(gh->gh_owner_pid == NULL); |
fee852e3 SW |
1020 | if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags)) |
1021 | BUG(); | |
190562bd | 1022 | |
6802e340 SW |
1023 | if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) { |
1024 | if (test_bit(GLF_LOCK, &gl->gl_flags)) | |
e5dc76b9 | 1025 | try_futile = !may_grant(gl, gh); |
6802e340 SW |
1026 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) |
1027 | goto fail; | |
1028 | } | |
1029 | ||
1030 | list_for_each_entry(gh2, &gl->gl_holders, gh_list) { | |
1031 | if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid && | |
1032 | (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK))) | |
1033 | goto trap_recursive; | |
e5dc76b9 BP |
1034 | if (try_futile && |
1035 | !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) { | |
6802e340 SW |
1036 | fail: |
1037 | gh->gh_error = GLR_TRYFAILED; | |
1038 | gfs2_holder_wake(gh); | |
1039 | return; | |
b4c20166 | 1040 | } |
6802e340 SW |
1041 | if (test_bit(HIF_HOLDER, &gh2->gh_iflags)) |
1042 | continue; | |
1043 | if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt)) | |
1044 | insert_pt = &gh2->gh_list; | |
1045 | } | |
7b5e3d5f | 1046 | set_bit(GLF_QUEUED, &gl->gl_flags); |
edae38a6 | 1047 | trace_gfs2_glock_queue(gh, 1); |
a245769f SW |
1048 | gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT); |
1049 | gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT); | |
6802e340 SW |
1050 | if (likely(insert_pt == NULL)) { |
1051 | list_add_tail(&gh->gh_list, &gl->gl_holders); | |
1052 | if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY)) | |
1053 | goto do_cancel; | |
1054 | return; | |
1055 | } | |
1056 | list_add_tail(&gh->gh_list, insert_pt); | |
1057 | do_cancel: | |
1058 | gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list); | |
1059 | if (!(gh->gh_flags & LM_FLAG_PRIORITY)) { | |
f3dd1649 | 1060 | spin_unlock(&gl->gl_lockref.lock); |
048bca22 | 1061 | if (sdp->sd_lockstruct.ls_ops->lm_cancel) |
f057f6cd | 1062 | sdp->sd_lockstruct.ls_ops->lm_cancel(gl); |
f3dd1649 | 1063 | spin_lock(&gl->gl_lockref.lock); |
b3b94faa | 1064 | } |
6802e340 | 1065 | return; |
b3b94faa | 1066 | |
6802e340 | 1067 | trap_recursive: |
e54c78a2 BP |
1068 | fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip); |
1069 | fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid)); | |
1070 | fs_err(sdp, "lock type: %d req lock state : %d\n", | |
6802e340 | 1071 | gh2->gh_gl->gl_name.ln_type, gh2->gh_state); |
e54c78a2 BP |
1072 | fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip); |
1073 | fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid)); | |
1074 | fs_err(sdp, "lock type: %d req lock state : %d\n", | |
6802e340 | 1075 | gh->gh_gl->gl_name.ln_type, gh->gh_state); |
8eae1ca0 | 1076 | gfs2_dump_glock(NULL, gl); |
6802e340 | 1077 | BUG(); |
b3b94faa DT |
1078 | } |
1079 | ||
1080 | /** | |
1081 | * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock) | |
1082 | * @gh: the holder structure | |
1083 | * | |
1084 | * if (gh->gh_flags & GL_ASYNC), this never returns an error | |
1085 | * | |
1086 | * Returns: 0, GLR_TRYFAILED, or errno on failure | |
1087 | */ | |
1088 | ||
1089 | int gfs2_glock_nq(struct gfs2_holder *gh) | |
1090 | { | |
1091 | struct gfs2_glock *gl = gh->gh_gl; | |
15562c43 | 1092 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
b3b94faa DT |
1093 | int error = 0; |
1094 | ||
6802e340 | 1095 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
b3b94faa | 1096 | return -EIO; |
b3b94faa | 1097 | |
f42ab085 SW |
1098 | if (test_bit(GLF_LRU, &gl->gl_flags)) |
1099 | gfs2_glock_remove_from_lru(gl); | |
1100 | ||
f3dd1649 | 1101 | spin_lock(&gl->gl_lockref.lock); |
b3b94faa | 1102 | add_to_queue(gh); |
01b172b7 BP |
1103 | if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) && |
1104 | test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) { | |
0809f6ec | 1105 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
01b172b7 | 1106 | gl->gl_lockref.count++; |
6b0c7440 | 1107 | __gfs2_glock_queue_work(gl, 0); |
01b172b7 | 1108 | } |
6802e340 | 1109 | run_queue(gl, 1); |
f3dd1649 | 1110 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa | 1111 | |
6802e340 SW |
1112 | if (!(gh->gh_flags & GL_ASYNC)) |
1113 | error = gfs2_glock_wait(gh); | |
b3b94faa | 1114 | |
b3b94faa DT |
1115 | return error; |
1116 | } | |
1117 | ||
1118 | /** | |
1119 | * gfs2_glock_poll - poll to see if an async request has been completed | |
1120 | * @gh: the holder | |
1121 | * | |
1122 | * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on | |
1123 | */ | |
1124 | ||
1125 | int gfs2_glock_poll(struct gfs2_holder *gh) | |
1126 | { | |
6802e340 | 1127 | return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1; |
b3b94faa DT |
1128 | } |
1129 | ||
1130 | /** | |
1131 | * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock) | |
1132 | * @gh: the glock holder | |
1133 | * | |
1134 | */ | |
1135 | ||
1136 | void gfs2_glock_dq(struct gfs2_holder *gh) | |
1137 | { | |
1138 | struct gfs2_glock *gl = gh->gh_gl; | |
8fb4b536 | 1139 | const struct gfs2_glock_operations *glops = gl->gl_ops; |
c4f68a13 | 1140 | unsigned delay = 0; |
6802e340 | 1141 | int fast_path = 0; |
b3b94faa | 1142 | |
f3dd1649 | 1143 | spin_lock(&gl->gl_lockref.lock); |
b3b94faa | 1144 | if (gh->gh_flags & GL_NOCACHE) |
81ffbf65 | 1145 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); |
b3b94faa | 1146 | |
b3b94faa | 1147 | list_del_init(&gh->gh_list); |
7508abc4 | 1148 | clear_bit(HIF_HOLDER, &gh->gh_iflags); |
6802e340 | 1149 | if (find_first_holder(gl) == NULL) { |
3042a2cc | 1150 | if (glops->go_unlock) { |
6802e340 | 1151 | GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags)); |
f3dd1649 | 1152 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa | 1153 | glops->go_unlock(gh); |
f3dd1649 | 1154 | spin_lock(&gl->gl_lockref.lock); |
6802e340 | 1155 | clear_bit(GLF_LOCK, &gl->gl_flags); |
3042a2cc | 1156 | } |
6802e340 SW |
1157 | if (list_empty(&gl->gl_holders) && |
1158 | !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) && | |
1159 | !test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
1160 | fast_path = 1; | |
b3b94faa | 1161 | } |
e7ccaf5f BP |
1162 | if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) && |
1163 | (glops->go_flags & GLOF_LRU)) | |
4abb6ad9 BP |
1164 | gfs2_glock_add_to_lru(gl); |
1165 | ||
63997775 | 1166 | trace_gfs2_glock_queue(gh, 0); |
6b0c7440 AG |
1167 | if (unlikely(!fast_path)) { |
1168 | gl->gl_lockref.count++; | |
1169 | if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) && | |
1170 | !test_bit(GLF_DEMOTE, &gl->gl_flags) && | |
1171 | gl->gl_name.ln_type == LM_TYPE_INODE) | |
1172 | delay = gl->gl_hold_time; | |
1173 | __gfs2_glock_queue_work(gl, delay); | |
1174 | } | |
f3dd1649 | 1175 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1176 | } |
1177 | ||
d93cfa98 AD |
1178 | void gfs2_glock_dq_wait(struct gfs2_holder *gh) |
1179 | { | |
1180 | struct gfs2_glock *gl = gh->gh_gl; | |
1181 | gfs2_glock_dq(gh); | |
81e1d450 | 1182 | might_sleep(); |
74316201 | 1183 | wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE); |
d93cfa98 AD |
1184 | } |
1185 | ||
b3b94faa DT |
1186 | /** |
1187 | * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it | |
1188 | * @gh: the holder structure | |
1189 | * | |
1190 | */ | |
1191 | ||
1192 | void gfs2_glock_dq_uninit(struct gfs2_holder *gh) | |
1193 | { | |
1194 | gfs2_glock_dq(gh); | |
1195 | gfs2_holder_uninit(gh); | |
1196 | } | |
1197 | ||
1198 | /** | |
1199 | * gfs2_glock_nq_num - acquire a glock based on lock number | |
1200 | * @sdp: the filesystem | |
1201 | * @number: the lock number | |
1202 | * @glops: the glock operations for the type of glock | |
1203 | * @state: the state to acquire the glock in | |
25985edc | 1204 | * @flags: modifier flags for the acquisition |
b3b94faa DT |
1205 | * @gh: the struct gfs2_holder |
1206 | * | |
1207 | * Returns: errno | |
1208 | */ | |
1209 | ||
cd915493 | 1210 | int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number, |
8fb4b536 | 1211 | const struct gfs2_glock_operations *glops, |
b58bf407 | 1212 | unsigned int state, u16 flags, struct gfs2_holder *gh) |
b3b94faa DT |
1213 | { |
1214 | struct gfs2_glock *gl; | |
1215 | int error; | |
1216 | ||
1217 | error = gfs2_glock_get(sdp, number, glops, CREATE, &gl); | |
1218 | if (!error) { | |
1219 | error = gfs2_glock_nq_init(gl, state, flags, gh); | |
1220 | gfs2_glock_put(gl); | |
1221 | } | |
1222 | ||
1223 | return error; | |
1224 | } | |
1225 | ||
1226 | /** | |
1227 | * glock_compare - Compare two struct gfs2_glock structures for sorting | |
1228 | * @arg_a: the first structure | |
1229 | * @arg_b: the second structure | |
1230 | * | |
1231 | */ | |
1232 | ||
1233 | static int glock_compare(const void *arg_a, const void *arg_b) | |
1234 | { | |
a5e08a9e SW |
1235 | const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a; |
1236 | const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b; | |
1237 | const struct lm_lockname *a = &gh_a->gh_gl->gl_name; | |
1238 | const struct lm_lockname *b = &gh_b->gh_gl->gl_name; | |
b3b94faa DT |
1239 | |
1240 | if (a->ln_number > b->ln_number) | |
a5e08a9e SW |
1241 | return 1; |
1242 | if (a->ln_number < b->ln_number) | |
1243 | return -1; | |
1c0f4872 | 1244 | BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type); |
a5e08a9e | 1245 | return 0; |
b3b94faa DT |
1246 | } |
1247 | ||
1248 | /** | |
1249 | * nq_m_sync - synchonously acquire more than one glock in deadlock free order | |
1250 | * @num_gh: the number of structures | |
1251 | * @ghs: an array of struct gfs2_holder structures | |
1252 | * | |
1253 | * Returns: 0 on success (all glocks acquired), | |
1254 | * errno on failure (no glocks acquired) | |
1255 | */ | |
1256 | ||
1257 | static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs, | |
1258 | struct gfs2_holder **p) | |
1259 | { | |
1260 | unsigned int x; | |
1261 | int error = 0; | |
1262 | ||
1263 | for (x = 0; x < num_gh; x++) | |
1264 | p[x] = &ghs[x]; | |
1265 | ||
1266 | sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL); | |
1267 | ||
1268 | for (x = 0; x < num_gh; x++) { | |
1269 | p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); | |
1270 | ||
1271 | error = gfs2_glock_nq(p[x]); | |
1272 | if (error) { | |
1273 | while (x--) | |
1274 | gfs2_glock_dq(p[x]); | |
1275 | break; | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | return error; | |
1280 | } | |
1281 | ||
1282 | /** | |
1283 | * gfs2_glock_nq_m - acquire multiple glocks | |
1284 | * @num_gh: the number of structures | |
1285 | * @ghs: an array of struct gfs2_holder structures | |
1286 | * | |
b3b94faa DT |
1287 | * |
1288 | * Returns: 0 on success (all glocks acquired), | |
1289 | * errno on failure (no glocks acquired) | |
1290 | */ | |
1291 | ||
1292 | int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs) | |
1293 | { | |
eaf5bd3c SW |
1294 | struct gfs2_holder *tmp[4]; |
1295 | struct gfs2_holder **pph = tmp; | |
b3b94faa DT |
1296 | int error = 0; |
1297 | ||
eaf5bd3c SW |
1298 | switch(num_gh) { |
1299 | case 0: | |
b3b94faa | 1300 | return 0; |
eaf5bd3c | 1301 | case 1: |
b3b94faa DT |
1302 | ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); |
1303 | return gfs2_glock_nq(ghs); | |
eaf5bd3c SW |
1304 | default: |
1305 | if (num_gh <= 4) | |
b3b94faa | 1306 | break; |
6da2ec56 KC |
1307 | pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *), |
1308 | GFP_NOFS); | |
eaf5bd3c SW |
1309 | if (!pph) |
1310 | return -ENOMEM; | |
b3b94faa DT |
1311 | } |
1312 | ||
eaf5bd3c | 1313 | error = nq_m_sync(num_gh, ghs, pph); |
b3b94faa | 1314 | |
eaf5bd3c SW |
1315 | if (pph != tmp) |
1316 | kfree(pph); | |
b3b94faa DT |
1317 | |
1318 | return error; | |
1319 | } | |
1320 | ||
1321 | /** | |
1322 | * gfs2_glock_dq_m - release multiple glocks | |
1323 | * @num_gh: the number of structures | |
1324 | * @ghs: an array of struct gfs2_holder structures | |
1325 | * | |
1326 | */ | |
1327 | ||
1328 | void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs) | |
1329 | { | |
fa1bbdea BP |
1330 | while (num_gh--) |
1331 | gfs2_glock_dq(&ghs[num_gh]); | |
b3b94faa DT |
1332 | } |
1333 | ||
f057f6cd | 1334 | void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state) |
da755fdb | 1335 | { |
c4f68a13 BM |
1336 | unsigned long delay = 0; |
1337 | unsigned long holdtime; | |
1338 | unsigned long now = jiffies; | |
b3b94faa | 1339 | |
f057f6cd | 1340 | gfs2_glock_hold(gl); |
7cf8dcd3 BP |
1341 | holdtime = gl->gl_tchange + gl->gl_hold_time; |
1342 | if (test_bit(GLF_QUEUED, &gl->gl_flags) && | |
1343 | gl->gl_name.ln_type == LM_TYPE_INODE) { | |
7b5e3d5f SW |
1344 | if (time_before(now, holdtime)) |
1345 | delay = holdtime - now; | |
1346 | if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags)) | |
7cf8dcd3 | 1347 | delay = gl->gl_hold_time; |
7b5e3d5f | 1348 | } |
b3b94faa | 1349 | |
f3dd1649 | 1350 | spin_lock(&gl->gl_lockref.lock); |
81ffbf65 | 1351 | handle_callback(gl, state, delay, true); |
6b0c7440 | 1352 | __gfs2_glock_queue_work(gl, delay); |
f3dd1649 | 1353 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1354 | } |
1355 | ||
0809f6ec SW |
1356 | /** |
1357 | * gfs2_should_freeze - Figure out if glock should be frozen | |
1358 | * @gl: The glock in question | |
1359 | * | |
1360 | * Glocks are not frozen if (a) the result of the dlm operation is | |
1361 | * an error, (b) the locking operation was an unlock operation or | |
1362 | * (c) if there is a "noexp" flagged request anywhere in the queue | |
1363 | * | |
1364 | * Returns: 1 if freezing should occur, 0 otherwise | |
1365 | */ | |
1366 | ||
1367 | static int gfs2_should_freeze(const struct gfs2_glock *gl) | |
1368 | { | |
1369 | const struct gfs2_holder *gh; | |
1370 | ||
1371 | if (gl->gl_reply & ~LM_OUT_ST_MASK) | |
1372 | return 0; | |
1373 | if (gl->gl_target == LM_ST_UNLOCKED) | |
1374 | return 0; | |
1375 | ||
1376 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { | |
1377 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
1378 | continue; | |
1379 | if (LM_FLAG_NOEXP & gh->gh_flags) | |
1380 | return 0; | |
1381 | } | |
1382 | ||
1383 | return 1; | |
1384 | } | |
1385 | ||
b3b94faa | 1386 | /** |
f057f6cd SW |
1387 | * gfs2_glock_complete - Callback used by locking |
1388 | * @gl: Pointer to the glock | |
1389 | * @ret: The return value from the dlm | |
b3b94faa | 1390 | * |
f3dd1649 | 1391 | * The gl_reply field is under the gl_lockref.lock lock so that it is ok |
47a25380 | 1392 | * to use a bitfield shared with other glock state fields. |
b3b94faa DT |
1393 | */ |
1394 | ||
f057f6cd | 1395 | void gfs2_glock_complete(struct gfs2_glock *gl, int ret) |
b3b94faa | 1396 | { |
15562c43 | 1397 | struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct; |
0809f6ec | 1398 | |
f3dd1649 | 1399 | spin_lock(&gl->gl_lockref.lock); |
f057f6cd | 1400 | gl->gl_reply = ret; |
0809f6ec | 1401 | |
e0c2a9aa | 1402 | if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) { |
0809f6ec | 1403 | if (gfs2_should_freeze(gl)) { |
f057f6cd | 1404 | set_bit(GLF_FROZEN, &gl->gl_flags); |
f3dd1649 | 1405 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa | 1406 | return; |
0809f6ec | 1407 | } |
b3b94faa | 1408 | } |
47a25380 | 1409 | |
e66cf161 | 1410 | gl->gl_lockref.count++; |
f057f6cd | 1411 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
6b0c7440 | 1412 | __gfs2_glock_queue_work(gl, 0); |
f3dd1649 | 1413 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1414 | } |
1415 | ||
4506a519 SW |
1416 | static int glock_cmp(void *priv, struct list_head *a, struct list_head *b) |
1417 | { | |
1418 | struct gfs2_glock *gla, *glb; | |
1419 | ||
1420 | gla = list_entry(a, struct gfs2_glock, gl_lru); | |
1421 | glb = list_entry(b, struct gfs2_glock, gl_lru); | |
1422 | ||
1423 | if (gla->gl_name.ln_number > glb->gl_name.ln_number) | |
1424 | return 1; | |
1425 | if (gla->gl_name.ln_number < glb->gl_name.ln_number) | |
1426 | return -1; | |
1427 | ||
1428 | return 0; | |
1429 | } | |
1430 | ||
1431 | /** | |
1432 | * gfs2_dispose_glock_lru - Demote a list of glocks | |
1433 | * @list: The list to dispose of | |
1434 | * | |
1435 | * Disposing of glocks may involve disk accesses, so that here we sort | |
1436 | * the glocks by number (i.e. disk location of the inodes) so that if | |
1437 | * there are any such accesses, they'll be sent in order (mostly). | |
1438 | * | |
1439 | * Must be called under the lru_lock, but may drop and retake this | |
1440 | * lock. While the lru_lock is dropped, entries may vanish from the | |
1441 | * list, but no new entries will appear on the list (since it is | |
1442 | * private) | |
1443 | */ | |
1444 | ||
1445 | static void gfs2_dispose_glock_lru(struct list_head *list) | |
1446 | __releases(&lru_lock) | |
1447 | __acquires(&lru_lock) | |
1448 | { | |
1449 | struct gfs2_glock *gl; | |
1450 | ||
1451 | list_sort(NULL, list, glock_cmp); | |
1452 | ||
1453 | while(!list_empty(list)) { | |
1454 | gl = list_entry(list->next, struct gfs2_glock, gl_lru); | |
1455 | list_del_init(&gl->gl_lru); | |
f3dd1649 | 1456 | if (!spin_trylock(&gl->gl_lockref.lock)) { |
94a09a39 | 1457 | add_back_to_lru: |
e66cf161 SW |
1458 | list_add(&gl->gl_lru, &lru_list); |
1459 | atomic_inc(&lru_count); | |
1460 | continue; | |
1461 | } | |
94a09a39 | 1462 | if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) { |
f3dd1649 | 1463 | spin_unlock(&gl->gl_lockref.lock); |
94a09a39 SW |
1464 | goto add_back_to_lru; |
1465 | } | |
4506a519 | 1466 | clear_bit(GLF_LRU, &gl->gl_flags); |
e66cf161 | 1467 | gl->gl_lockref.count++; |
4506a519 | 1468 | if (demote_ok(gl)) |
81ffbf65 | 1469 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); |
4506a519 | 1470 | WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags)); |
6b0c7440 | 1471 | __gfs2_glock_queue_work(gl, 0); |
f3dd1649 | 1472 | spin_unlock(&gl->gl_lockref.lock); |
94a09a39 | 1473 | cond_resched_lock(&lru_lock); |
4506a519 SW |
1474 | } |
1475 | } | |
1476 | ||
2a005855 SW |
1477 | /** |
1478 | * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote | |
1479 | * @nr: The number of entries to scan | |
1480 | * | |
4506a519 SW |
1481 | * This function selects the entries on the LRU which are able to |
1482 | * be demoted, and then kicks off the process by calling | |
1483 | * gfs2_dispose_glock_lru() above. | |
2a005855 | 1484 | */ |
b3b94faa | 1485 | |
1ab6c499 | 1486 | static long gfs2_scan_glock_lru(int nr) |
b3b94faa DT |
1487 | { |
1488 | struct gfs2_glock *gl; | |
97cc1025 | 1489 | LIST_HEAD(skipped); |
4506a519 | 1490 | LIST_HEAD(dispose); |
1ab6c499 | 1491 | long freed = 0; |
b3b94faa | 1492 | |
97cc1025 | 1493 | spin_lock(&lru_lock); |
1ab6c499 | 1494 | while ((nr-- >= 0) && !list_empty(&lru_list)) { |
97cc1025 | 1495 | gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru); |
97cc1025 SW |
1496 | |
1497 | /* Test for being demotable */ | |
94a09a39 | 1498 | if (!test_bit(GLF_LOCK, &gl->gl_flags)) { |
4506a519 SW |
1499 | list_move(&gl->gl_lru, &dispose); |
1500 | atomic_dec(&lru_count); | |
1ab6c499 | 1501 | freed++; |
2163b1e6 | 1502 | continue; |
97cc1025 | 1503 | } |
4506a519 SW |
1504 | |
1505 | list_move(&gl->gl_lru, &skipped); | |
b3b94faa | 1506 | } |
97cc1025 | 1507 | list_splice(&skipped, &lru_list); |
4506a519 SW |
1508 | if (!list_empty(&dispose)) |
1509 | gfs2_dispose_glock_lru(&dispose); | |
97cc1025 | 1510 | spin_unlock(&lru_lock); |
1ab6c499 DC |
1511 | |
1512 | return freed; | |
2a005855 SW |
1513 | } |
1514 | ||
1ab6c499 DC |
1515 | static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink, |
1516 | struct shrink_control *sc) | |
2a005855 | 1517 | { |
1ab6c499 DC |
1518 | if (!(sc->gfp_mask & __GFP_FS)) |
1519 | return SHRINK_STOP; | |
1520 | return gfs2_scan_glock_lru(sc->nr_to_scan); | |
1521 | } | |
2a005855 | 1522 | |
1ab6c499 DC |
1523 | static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink, |
1524 | struct shrink_control *sc) | |
1525 | { | |
55f841ce | 1526 | return vfs_pressure_ratio(atomic_read(&lru_count)); |
b3b94faa DT |
1527 | } |
1528 | ||
97cc1025 | 1529 | static struct shrinker glock_shrinker = { |
97cc1025 | 1530 | .seeks = DEFAULT_SEEKS, |
1ab6c499 DC |
1531 | .count_objects = gfs2_glock_shrink_count, |
1532 | .scan_objects = gfs2_glock_shrink_scan, | |
97cc1025 SW |
1533 | }; |
1534 | ||
b3b94faa DT |
1535 | /** |
1536 | * examine_bucket - Call a function for glock in a hash bucket | |
1537 | * @examiner: the function | |
1538 | * @sdp: the filesystem | |
1539 | * @bucket: the bucket | |
1540 | * | |
98687f42 HX |
1541 | * Note that the function can be called multiple times on the same |
1542 | * object. So the user must ensure that the function can cope with | |
1543 | * that. | |
b3b94faa DT |
1544 | */ |
1545 | ||
88ffbf3e | 1546 | static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp) |
b3b94faa | 1547 | { |
bc015cb8 | 1548 | struct gfs2_glock *gl; |
98687f42 HX |
1549 | struct rhashtable_iter iter; |
1550 | ||
1551 | rhashtable_walk_enter(&gl_hash_table, &iter); | |
1552 | ||
1553 | do { | |
97a6ec4a | 1554 | rhashtable_walk_start(&iter); |
98687f42 HX |
1555 | |
1556 | while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) | |
27c3b415 | 1557 | if (gl->gl_name.ln_sbd == sdp && |
88ffbf3e BP |
1558 | lockref_get_not_dead(&gl->gl_lockref)) |
1559 | examiner(gl); | |
98687f42 HX |
1560 | |
1561 | rhashtable_walk_stop(&iter); | |
1562 | } while (cond_resched(), gl == ERR_PTR(-EAGAIN)); | |
1563 | ||
1564 | rhashtable_walk_exit(&iter); | |
bc015cb8 SW |
1565 | } |
1566 | ||
f057f6cd SW |
1567 | /** |
1568 | * thaw_glock - thaw out a glock which has an unprocessed reply waiting | |
1569 | * @gl: The glock to thaw | |
1570 | * | |
f057f6cd SW |
1571 | */ |
1572 | ||
1573 | static void thaw_glock(struct gfs2_glock *gl) | |
1574 | { | |
6b0c7440 | 1575 | if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) { |
f057f6cd | 1576 | gfs2_glock_put(gl); |
6b0c7440 | 1577 | return; |
7286b31e | 1578 | } |
6b0c7440 AG |
1579 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
1580 | gfs2_glock_queue_work(gl, 0); | |
f057f6cd SW |
1581 | } |
1582 | ||
b3b94faa DT |
1583 | /** |
1584 | * clear_glock - look at a glock and see if we can free it from glock cache | |
1585 | * @gl: the glock to look at | |
1586 | * | |
1587 | */ | |
1588 | ||
1589 | static void clear_glock(struct gfs2_glock *gl) | |
1590 | { | |
f42ab085 | 1591 | gfs2_glock_remove_from_lru(gl); |
b3b94faa | 1592 | |
f3dd1649 | 1593 | spin_lock(&gl->gl_lockref.lock); |
c741c455 | 1594 | if (gl->gl_state != LM_ST_UNLOCKED) |
81ffbf65 | 1595 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); |
6b0c7440 | 1596 | __gfs2_glock_queue_work(gl, 0); |
f3dd1649 | 1597 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1598 | } |
1599 | ||
f057f6cd SW |
1600 | /** |
1601 | * gfs2_glock_thaw - Thaw any frozen glocks | |
1602 | * @sdp: The super block | |
1603 | * | |
1604 | */ | |
1605 | ||
1606 | void gfs2_glock_thaw(struct gfs2_sbd *sdp) | |
1607 | { | |
bc015cb8 SW |
1608 | glock_hash_walk(thaw_glock, sdp); |
1609 | } | |
f057f6cd | 1610 | |
ac3beb6a | 1611 | static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl) |
bc015cb8 | 1612 | { |
f3dd1649 | 1613 | spin_lock(&gl->gl_lockref.lock); |
ac3beb6a | 1614 | gfs2_dump_glock(seq, gl); |
f3dd1649 | 1615 | spin_unlock(&gl->gl_lockref.lock); |
bc015cb8 SW |
1616 | } |
1617 | ||
1618 | static void dump_glock_func(struct gfs2_glock *gl) | |
1619 | { | |
1620 | dump_glock(NULL, gl); | |
f057f6cd SW |
1621 | } |
1622 | ||
b3b94faa DT |
1623 | /** |
1624 | * gfs2_gl_hash_clear - Empty out the glock hash table | |
1625 | * @sdp: the filesystem | |
1626 | * @wait: wait until it's all gone | |
1627 | * | |
1bdad606 | 1628 | * Called when unmounting the filesystem. |
b3b94faa DT |
1629 | */ |
1630 | ||
fefc03bf | 1631 | void gfs2_gl_hash_clear(struct gfs2_sbd *sdp) |
b3b94faa | 1632 | { |
fb6791d1 | 1633 | set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags); |
222cb538 | 1634 | flush_workqueue(glock_workqueue); |
bc015cb8 | 1635 | glock_hash_walk(clear_glock, sdp); |
8f05228e | 1636 | flush_workqueue(glock_workqueue); |
2aba1b5b BP |
1637 | wait_event_timeout(sdp->sd_glock_wait, |
1638 | atomic_read(&sdp->sd_glock_disposal) == 0, | |
1639 | HZ * 600); | |
bc015cb8 | 1640 | glock_hash_walk(dump_glock_func, sdp); |
b3b94faa DT |
1641 | } |
1642 | ||
813e0c46 SW |
1643 | void gfs2_glock_finish_truncate(struct gfs2_inode *ip) |
1644 | { | |
1645 | struct gfs2_glock *gl = ip->i_gl; | |
1646 | int ret; | |
1647 | ||
1648 | ret = gfs2_truncatei_resume(ip); | |
15562c43 | 1649 | gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0); |
813e0c46 | 1650 | |
f3dd1649 | 1651 | spin_lock(&gl->gl_lockref.lock); |
813e0c46 SW |
1652 | clear_bit(GLF_LOCK, &gl->gl_flags); |
1653 | run_queue(gl, 1); | |
f3dd1649 | 1654 | spin_unlock(&gl->gl_lockref.lock); |
813e0c46 SW |
1655 | } |
1656 | ||
6802e340 | 1657 | static const char *state2str(unsigned state) |
04b933f2 | 1658 | { |
6802e340 SW |
1659 | switch(state) { |
1660 | case LM_ST_UNLOCKED: | |
1661 | return "UN"; | |
1662 | case LM_ST_SHARED: | |
1663 | return "SH"; | |
1664 | case LM_ST_DEFERRED: | |
1665 | return "DF"; | |
1666 | case LM_ST_EXCLUSIVE: | |
1667 | return "EX"; | |
1668 | } | |
1669 | return "??"; | |
1670 | } | |
1671 | ||
b58bf407 | 1672 | static const char *hflags2str(char *buf, u16 flags, unsigned long iflags) |
6802e340 SW |
1673 | { |
1674 | char *p = buf; | |
1675 | if (flags & LM_FLAG_TRY) | |
1676 | *p++ = 't'; | |
1677 | if (flags & LM_FLAG_TRY_1CB) | |
1678 | *p++ = 'T'; | |
1679 | if (flags & LM_FLAG_NOEXP) | |
1680 | *p++ = 'e'; | |
1681 | if (flags & LM_FLAG_ANY) | |
f057f6cd | 1682 | *p++ = 'A'; |
6802e340 SW |
1683 | if (flags & LM_FLAG_PRIORITY) |
1684 | *p++ = 'p'; | |
1685 | if (flags & GL_ASYNC) | |
1686 | *p++ = 'a'; | |
1687 | if (flags & GL_EXACT) | |
1688 | *p++ = 'E'; | |
6802e340 SW |
1689 | if (flags & GL_NOCACHE) |
1690 | *p++ = 'c'; | |
1691 | if (test_bit(HIF_HOLDER, &iflags)) | |
1692 | *p++ = 'H'; | |
1693 | if (test_bit(HIF_WAIT, &iflags)) | |
1694 | *p++ = 'W'; | |
1695 | if (test_bit(HIF_FIRST, &iflags)) | |
1696 | *p++ = 'F'; | |
1697 | *p = 0; | |
1698 | return buf; | |
04b933f2 RP |
1699 | } |
1700 | ||
b3b94faa DT |
1701 | /** |
1702 | * dump_holder - print information about a glock holder | |
6802e340 | 1703 | * @seq: the seq_file struct |
b3b94faa DT |
1704 | * @gh: the glock holder |
1705 | * | |
b3b94faa DT |
1706 | */ |
1707 | ||
ac3beb6a | 1708 | static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh) |
b3b94faa | 1709 | { |
6802e340 | 1710 | struct task_struct *gh_owner = NULL; |
6802e340 | 1711 | char flags_buf[32]; |
b3b94faa | 1712 | |
0b3a2c99 | 1713 | rcu_read_lock(); |
6802e340 | 1714 | if (gh->gh_owner_pid) |
b1e058da | 1715 | gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID); |
cc18152e JP |
1716 | gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n", |
1717 | state2str(gh->gh_state), | |
1718 | hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags), | |
1719 | gh->gh_error, | |
1720 | gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1, | |
1721 | gh_owner ? gh_owner->comm : "(ended)", | |
1722 | (void *)gh->gh_ip); | |
0b3a2c99 | 1723 | rcu_read_unlock(); |
b3b94faa DT |
1724 | } |
1725 | ||
627c10b7 | 1726 | static const char *gflags2str(char *buf, const struct gfs2_glock *gl) |
6802e340 | 1727 | { |
627c10b7 | 1728 | const unsigned long *gflags = &gl->gl_flags; |
6802e340 | 1729 | char *p = buf; |
627c10b7 | 1730 | |
6802e340 SW |
1731 | if (test_bit(GLF_LOCK, gflags)) |
1732 | *p++ = 'l'; | |
6802e340 SW |
1733 | if (test_bit(GLF_DEMOTE, gflags)) |
1734 | *p++ = 'D'; | |
1735 | if (test_bit(GLF_PENDING_DEMOTE, gflags)) | |
1736 | *p++ = 'd'; | |
1737 | if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags)) | |
1738 | *p++ = 'p'; | |
1739 | if (test_bit(GLF_DIRTY, gflags)) | |
1740 | *p++ = 'y'; | |
1741 | if (test_bit(GLF_LFLUSH, gflags)) | |
1742 | *p++ = 'f'; | |
1743 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags)) | |
1744 | *p++ = 'i'; | |
1745 | if (test_bit(GLF_REPLY_PENDING, gflags)) | |
1746 | *p++ = 'r'; | |
f057f6cd | 1747 | if (test_bit(GLF_INITIAL, gflags)) |
d8348de0 | 1748 | *p++ = 'I'; |
f057f6cd SW |
1749 | if (test_bit(GLF_FROZEN, gflags)) |
1750 | *p++ = 'F'; | |
7b5e3d5f SW |
1751 | if (test_bit(GLF_QUEUED, gflags)) |
1752 | *p++ = 'q'; | |
627c10b7 SW |
1753 | if (test_bit(GLF_LRU, gflags)) |
1754 | *p++ = 'L'; | |
1755 | if (gl->gl_object) | |
1756 | *p++ = 'o'; | |
a245769f SW |
1757 | if (test_bit(GLF_BLOCKING, gflags)) |
1758 | *p++ = 'b'; | |
6802e340 SW |
1759 | *p = 0; |
1760 | return buf; | |
b3b94faa DT |
1761 | } |
1762 | ||
1763 | /** | |
8eae1ca0 | 1764 | * gfs2_dump_glock - print information about a glock |
6802e340 | 1765 | * @seq: The seq_file struct |
b3b94faa | 1766 | * @gl: the glock |
6802e340 SW |
1767 | * |
1768 | * The file format is as follows: | |
1769 | * One line per object, capital letters are used to indicate objects | |
1770 | * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented, | |
1771 | * other objects are indented by a single space and follow the glock to | |
1772 | * which they are related. Fields are indicated by lower case letters | |
1773 | * followed by a colon and the field value, except for strings which are in | |
1774 | * [] so that its possible to see if they are composed of spaces for | |
1775 | * example. The field's are n = number (id of the object), f = flags, | |
1776 | * t = type, s = state, r = refcount, e = error, p = pid. | |
b3b94faa | 1777 | * |
b3b94faa DT |
1778 | */ |
1779 | ||
27a2660f | 1780 | void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl) |
b3b94faa | 1781 | { |
6802e340 SW |
1782 | const struct gfs2_glock_operations *glops = gl->gl_ops; |
1783 | unsigned long long dtime; | |
1784 | const struct gfs2_holder *gh; | |
1785 | char gflags_buf[32]; | |
b3b94faa | 1786 | |
6802e340 SW |
1787 | dtime = jiffies - gl->gl_demote_time; |
1788 | dtime *= 1000000/HZ; /* demote time in uSec */ | |
1789 | if (!test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
1790 | dtime = 0; | |
7cf8dcd3 | 1791 | gfs2_print_dbg(seq, "G: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d m:%ld\n", |
6802e340 SW |
1792 | state2str(gl->gl_state), |
1793 | gl->gl_name.ln_type, | |
1794 | (unsigned long long)gl->gl_name.ln_number, | |
627c10b7 | 1795 | gflags2str(gflags_buf, gl), |
6802e340 SW |
1796 | state2str(gl->gl_target), |
1797 | state2str(gl->gl_demote_state), dtime, | |
6802e340 | 1798 | atomic_read(&gl->gl_ail_count), |
f42ab085 | 1799 | atomic_read(&gl->gl_revokes), |
e66cf161 | 1800 | (int)gl->gl_lockref.count, gl->gl_hold_time); |
b3b94faa | 1801 | |
ac3beb6a SW |
1802 | list_for_each_entry(gh, &gl->gl_holders, gh_list) |
1803 | dump_holder(seq, gh); | |
1804 | ||
6802e340 | 1805 | if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump) |
ac3beb6a | 1806 | glops->go_dump(seq, gl); |
b3b94faa DT |
1807 | } |
1808 | ||
a245769f SW |
1809 | static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr) |
1810 | { | |
1811 | struct gfs2_glock *gl = iter_ptr; | |
1812 | ||
4d207133 | 1813 | seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n", |
a245769f SW |
1814 | gl->gl_name.ln_type, |
1815 | (unsigned long long)gl->gl_name.ln_number, | |
4d207133 BH |
1816 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT], |
1817 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR], | |
1818 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB], | |
1819 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB], | |
1820 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT], | |
1821 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR], | |
1822 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT], | |
1823 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]); | |
a245769f SW |
1824 | return 0; |
1825 | } | |
1826 | ||
1827 | static const char *gfs2_gltype[] = { | |
1828 | "type", | |
1829 | "reserved", | |
1830 | "nondisk", | |
1831 | "inode", | |
1832 | "rgrp", | |
1833 | "meta", | |
1834 | "iopen", | |
1835 | "flock", | |
1836 | "plock", | |
1837 | "quota", | |
1838 | "journal", | |
1839 | }; | |
1840 | ||
1841 | static const char *gfs2_stype[] = { | |
1842 | [GFS2_LKS_SRTT] = "srtt", | |
1843 | [GFS2_LKS_SRTTVAR] = "srttvar", | |
1844 | [GFS2_LKS_SRTTB] = "srttb", | |
1845 | [GFS2_LKS_SRTTVARB] = "srttvarb", | |
1846 | [GFS2_LKS_SIRT] = "sirt", | |
1847 | [GFS2_LKS_SIRTVAR] = "sirtvar", | |
1848 | [GFS2_LKS_DCOUNT] = "dlm", | |
1849 | [GFS2_LKS_QCOUNT] = "queue", | |
1850 | }; | |
1851 | ||
1852 | #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype)) | |
1853 | ||
1854 | static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr) | |
1855 | { | |
81648d04 AG |
1856 | struct gfs2_sbd *sdp = seq->private; |
1857 | loff_t pos = *(loff_t *)iter_ptr; | |
1858 | unsigned index = pos >> 3; | |
1859 | unsigned subindex = pos & 0x07; | |
a245769f SW |
1860 | int i; |
1861 | ||
1862 | if (index == 0 && subindex != 0) | |
1863 | return 0; | |
6802e340 | 1864 | |
a245769f SW |
1865 | seq_printf(seq, "%-10s %8s:", gfs2_gltype[index], |
1866 | (index == 0) ? "cpu": gfs2_stype[subindex]); | |
b3b94faa | 1867 | |
a245769f SW |
1868 | for_each_possible_cpu(i) { |
1869 | const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i); | |
8f7e0a80 AG |
1870 | |
1871 | if (index == 0) | |
1872 | seq_printf(seq, " %15u", i); | |
1873 | else | |
1874 | seq_printf(seq, " %15llu", (unsigned long long)lkstats-> | |
1875 | lkstats[index - 1].stats[subindex]); | |
a245769f SW |
1876 | } |
1877 | seq_putc(seq, '\n'); | |
1878 | return 0; | |
1879 | } | |
8fbbfd21 | 1880 | |
85d1da67 SW |
1881 | int __init gfs2_glock_init(void) |
1882 | { | |
0515480a | 1883 | int i, ret; |
88ffbf3e BP |
1884 | |
1885 | ret = rhashtable_init(&gl_hash_table, &ht_parms); | |
1886 | if (ret < 0) | |
1887 | return ret; | |
8fbbfd21 | 1888 | |
d2115778 | 1889 | glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM | |
58a69cb4 | 1890 | WQ_HIGHPRI | WQ_FREEZABLE, 0); |
88ffbf3e BP |
1891 | if (!glock_workqueue) { |
1892 | rhashtable_destroy(&gl_hash_table); | |
dfc4616d | 1893 | return -ENOMEM; |
88ffbf3e | 1894 | } |
d2115778 | 1895 | gfs2_delete_workqueue = alloc_workqueue("delete_workqueue", |
58a69cb4 | 1896 | WQ_MEM_RECLAIM | WQ_FREEZABLE, |
d2115778 | 1897 | 0); |
dfc4616d | 1898 | if (!gfs2_delete_workqueue) { |
b94a170e | 1899 | destroy_workqueue(glock_workqueue); |
88ffbf3e | 1900 | rhashtable_destroy(&gl_hash_table); |
dfc4616d | 1901 | return -ENOMEM; |
b94a170e | 1902 | } |
97cc1025 | 1903 | |
e0d735c1 CY |
1904 | ret = register_shrinker(&glock_shrinker); |
1905 | if (ret) { | |
1906 | destroy_workqueue(gfs2_delete_workqueue); | |
1907 | destroy_workqueue(glock_workqueue); | |
1908 | rhashtable_destroy(&gl_hash_table); | |
1909 | return ret; | |
1910 | } | |
c4f68a13 | 1911 | |
0515480a AG |
1912 | for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++) |
1913 | init_waitqueue_head(glock_wait_table + i); | |
1914 | ||
85d1da67 SW |
1915 | return 0; |
1916 | } | |
1917 | ||
8fbbfd21 SW |
1918 | void gfs2_glock_exit(void) |
1919 | { | |
97cc1025 | 1920 | unregister_shrinker(&glock_shrinker); |
88ffbf3e | 1921 | rhashtable_destroy(&gl_hash_table); |
c4f68a13 | 1922 | destroy_workqueue(glock_workqueue); |
b94a170e | 1923 | destroy_workqueue(gfs2_delete_workqueue); |
8fbbfd21 SW |
1924 | } |
1925 | ||
7ac07fda | 1926 | static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n) |
bc015cb8 | 1927 | { |
3fd5d3ad AG |
1928 | struct gfs2_glock *gl = gi->gl; |
1929 | ||
1930 | if (gl) { | |
1931 | if (n == 0) | |
1932 | return; | |
1933 | if (!lockref_put_not_zero(&gl->gl_lockref)) | |
1934 | gfs2_glock_queue_put(gl); | |
7ac07fda AG |
1935 | } |
1936 | for (;;) { | |
3fd5d3ad AG |
1937 | gl = rhashtable_walk_next(&gi->hti); |
1938 | if (IS_ERR_OR_NULL(gl)) { | |
1939 | if (gl == ERR_PTR(-EAGAIN)) { | |
1940 | n = 1; | |
1941 | continue; | |
7ac07fda | 1942 | } |
3fd5d3ad AG |
1943 | gl = NULL; |
1944 | break; | |
1945 | } | |
1946 | if (gl->gl_name.ln_sbd != gi->sdp) | |
1947 | continue; | |
1948 | if (n <= 1) { | |
1949 | if (!lockref_get_not_dead(&gl->gl_lockref)) | |
1950 | continue; | |
1951 | break; | |
1952 | } else { | |
1953 | if (__lockref_is_dead(&gl->gl_lockref)) | |
1954 | continue; | |
1955 | n--; | |
bc015cb8 | 1956 | } |
14d37564 | 1957 | } |
3fd5d3ad | 1958 | gi->gl = gl; |
7c52b166 RP |
1959 | } |
1960 | ||
6802e340 | 1961 | static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos) |
27c3b415 | 1962 | __acquires(RCU) |
7c52b166 | 1963 | { |
6802e340 | 1964 | struct gfs2_glock_iter *gi = seq->private; |
7ac07fda | 1965 | loff_t n; |
ba1ddcb6 | 1966 | |
7ac07fda AG |
1967 | /* |
1968 | * We can either stay where we are, skip to the next hash table | |
1969 | * entry, or start from the beginning. | |
1970 | */ | |
1971 | if (*pos < gi->last_pos) { | |
1972 | rhashtable_walk_exit(&gi->hti); | |
1973 | rhashtable_walk_enter(&gl_hash_table, &gi->hti); | |
1974 | n = *pos + 1; | |
1975 | } else { | |
1976 | n = *pos - gi->last_pos; | |
1977 | } | |
7c52b166 | 1978 | |
7ac07fda | 1979 | rhashtable_walk_start(&gi->hti); |
7c52b166 | 1980 | |
7ac07fda | 1981 | gfs2_glock_iter_next(gi, n); |
ba1ddcb6 | 1982 | gi->last_pos = *pos; |
6802e340 | 1983 | return gi->gl; |
7c52b166 RP |
1984 | } |
1985 | ||
6802e340 | 1986 | static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr, |
7c52b166 RP |
1987 | loff_t *pos) |
1988 | { | |
6802e340 | 1989 | struct gfs2_glock_iter *gi = seq->private; |
7c52b166 RP |
1990 | |
1991 | (*pos)++; | |
ba1ddcb6 | 1992 | gi->last_pos = *pos; |
7ac07fda | 1993 | gfs2_glock_iter_next(gi, 1); |
6802e340 | 1994 | return gi->gl; |
7c52b166 RP |
1995 | } |
1996 | ||
6802e340 | 1997 | static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr) |
27c3b415 | 1998 | __releases(RCU) |
7c52b166 | 1999 | { |
6802e340 | 2000 | struct gfs2_glock_iter *gi = seq->private; |
bc015cb8 | 2001 | |
88ffbf3e | 2002 | rhashtable_walk_stop(&gi->hti); |
7c52b166 RP |
2003 | } |
2004 | ||
6802e340 | 2005 | static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr) |
7c52b166 | 2006 | { |
ac3beb6a SW |
2007 | dump_glock(seq, iter_ptr); |
2008 | return 0; | |
7c52b166 RP |
2009 | } |
2010 | ||
a245769f SW |
2011 | static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos) |
2012 | { | |
81648d04 | 2013 | preempt_disable(); |
a245769f SW |
2014 | if (*pos >= GFS2_NR_SBSTATS) |
2015 | return NULL; | |
81648d04 | 2016 | return pos; |
a245769f SW |
2017 | } |
2018 | ||
2019 | static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr, | |
2020 | loff_t *pos) | |
2021 | { | |
a245769f | 2022 | (*pos)++; |
81648d04 | 2023 | if (*pos >= GFS2_NR_SBSTATS) |
a245769f | 2024 | return NULL; |
81648d04 | 2025 | return pos; |
a245769f SW |
2026 | } |
2027 | ||
2028 | static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr) | |
2029 | { | |
2030 | preempt_enable(); | |
2031 | } | |
2032 | ||
4ef29002 | 2033 | static const struct seq_operations gfs2_glock_seq_ops = { |
7c52b166 RP |
2034 | .start = gfs2_glock_seq_start, |
2035 | .next = gfs2_glock_seq_next, | |
2036 | .stop = gfs2_glock_seq_stop, | |
2037 | .show = gfs2_glock_seq_show, | |
2038 | }; | |
2039 | ||
a245769f SW |
2040 | static const struct seq_operations gfs2_glstats_seq_ops = { |
2041 | .start = gfs2_glock_seq_start, | |
2042 | .next = gfs2_glock_seq_next, | |
2043 | .stop = gfs2_glock_seq_stop, | |
2044 | .show = gfs2_glstats_seq_show, | |
2045 | }; | |
2046 | ||
2047 | static const struct seq_operations gfs2_sbstats_seq_ops = { | |
2048 | .start = gfs2_sbstats_seq_start, | |
2049 | .next = gfs2_sbstats_seq_next, | |
2050 | .stop = gfs2_sbstats_seq_stop, | |
2051 | .show = gfs2_sbstats_seq_show, | |
2052 | }; | |
2053 | ||
0fe2f1e9 SW |
2054 | #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL) |
2055 | ||
92ecd73a AG |
2056 | static int __gfs2_glocks_open(struct inode *inode, struct file *file, |
2057 | const struct seq_operations *ops) | |
7c52b166 | 2058 | { |
92ecd73a | 2059 | int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter)); |
6802e340 SW |
2060 | if (ret == 0) { |
2061 | struct seq_file *seq = file->private_data; | |
2062 | struct gfs2_glock_iter *gi = seq->private; | |
88ffbf3e | 2063 | |
6802e340 | 2064 | gi->sdp = inode->i_private; |
0fe2f1e9 | 2065 | seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN); |
df5d2f55 | 2066 | if (seq->buf) |
0fe2f1e9 | 2067 | seq->size = GFS2_SEQ_GOODSIZE; |
7ac07fda AG |
2068 | /* |
2069 | * Initially, we are "before" the first hash table entry; the | |
2070 | * first call to rhashtable_walk_next gets us the first entry. | |
2071 | */ | |
2072 | gi->last_pos = -1; | |
88ffbf3e | 2073 | gi->gl = NULL; |
7ac07fda | 2074 | rhashtable_walk_enter(&gl_hash_table, &gi->hti); |
6802e340 SW |
2075 | } |
2076 | return ret; | |
7c52b166 RP |
2077 | } |
2078 | ||
92ecd73a AG |
2079 | static int gfs2_glocks_open(struct inode *inode, struct file *file) |
2080 | { | |
2081 | return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops); | |
2082 | } | |
2083 | ||
88ffbf3e BP |
2084 | static int gfs2_glocks_release(struct inode *inode, struct file *file) |
2085 | { | |
2086 | struct seq_file *seq = file->private_data; | |
2087 | struct gfs2_glock_iter *gi = seq->private; | |
2088 | ||
3fd5d3ad AG |
2089 | if (gi->gl) |
2090 | gfs2_glock_put(gi->gl); | |
7ac07fda | 2091 | rhashtable_walk_exit(&gi->hti); |
88ffbf3e BP |
2092 | return seq_release_private(inode, file); |
2093 | } | |
2094 | ||
a245769f SW |
2095 | static int gfs2_glstats_open(struct inode *inode, struct file *file) |
2096 | { | |
92ecd73a | 2097 | return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops); |
a245769f SW |
2098 | } |
2099 | ||
2100 | static int gfs2_sbstats_open(struct inode *inode, struct file *file) | |
2101 | { | |
81648d04 | 2102 | int ret = seq_open(file, &gfs2_sbstats_seq_ops); |
a245769f SW |
2103 | if (ret == 0) { |
2104 | struct seq_file *seq = file->private_data; | |
81648d04 | 2105 | seq->private = inode->i_private; /* sdp */ |
a245769f SW |
2106 | } |
2107 | return ret; | |
2108 | } | |
2109 | ||
2110 | static const struct file_operations gfs2_glocks_fops = { | |
2111 | .owner = THIS_MODULE, | |
2112 | .open = gfs2_glocks_open, | |
2113 | .read = seq_read, | |
2114 | .llseek = seq_lseek, | |
88ffbf3e | 2115 | .release = gfs2_glocks_release, |
a245769f SW |
2116 | }; |
2117 | ||
2118 | static const struct file_operations gfs2_glstats_fops = { | |
7c52b166 | 2119 | .owner = THIS_MODULE, |
a245769f SW |
2120 | .open = gfs2_glstats_open, |
2121 | .read = seq_read, | |
2122 | .llseek = seq_lseek, | |
88ffbf3e | 2123 | .release = gfs2_glocks_release, |
a245769f SW |
2124 | }; |
2125 | ||
2126 | static const struct file_operations gfs2_sbstats_fops = { | |
2127 | .owner = THIS_MODULE, | |
2128 | .open = gfs2_sbstats_open, | |
7c52b166 RP |
2129 | .read = seq_read, |
2130 | .llseek = seq_lseek, | |
81648d04 | 2131 | .release = seq_release, |
7c52b166 RP |
2132 | }; |
2133 | ||
2abbf9a4 GKH |
2134 | void gfs2_create_debugfs_file(struct gfs2_sbd *sdp) |
2135 | { | |
2136 | sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root); | |
7c52b166 | 2137 | |
2abbf9a4 GKH |
2138 | debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, |
2139 | &gfs2_glocks_fops); | |
2140 | ||
2141 | debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, | |
2142 | &gfs2_glstats_fops); | |
2143 | ||
2144 | debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, | |
2145 | &gfs2_sbstats_fops); | |
7c52b166 RP |
2146 | } |
2147 | ||
2148 | void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp) | |
2149 | { | |
2abbf9a4 GKH |
2150 | debugfs_remove_recursive(sdp->debugfs_dir); |
2151 | sdp->debugfs_dir = NULL; | |
7c52b166 RP |
2152 | } |
2153 | ||
2abbf9a4 | 2154 | void gfs2_register_debugfs(void) |
7c52b166 RP |
2155 | { |
2156 | gfs2_root = debugfs_create_dir("gfs2", NULL); | |
7c52b166 RP |
2157 | } |
2158 | ||
2159 | void gfs2_unregister_debugfs(void) | |
2160 | { | |
2161 | debugfs_remove(gfs2_root); | |
5f882096 | 2162 | gfs2_root = NULL; |
7c52b166 | 2163 | } |