]>
Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
cf45b752 | 4 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
d77d1b58 JP |
7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
8 | ||
b3b94faa DT |
9 | #include <linux/sched.h> |
10 | #include <linux/slab.h> | |
11 | #include <linux/spinlock.h> | |
b3b94faa DT |
12 | #include <linux/buffer_head.h> |
13 | #include <linux/delay.h> | |
14 | #include <linux/sort.h> | |
0515480a | 15 | #include <linux/hash.h> |
b3b94faa | 16 | #include <linux/jhash.h> |
d0dc80db | 17 | #include <linux/kallsyms.h> |
5c676f6d | 18 | #include <linux/gfs2_ondisk.h> |
24264434 | 19 | #include <linux/list.h> |
fee852e3 | 20 | #include <linux/wait.h> |
95d97b7d | 21 | #include <linux/module.h> |
7c0f6ba6 | 22 | #include <linux/uaccess.h> |
7c52b166 RP |
23 | #include <linux/seq_file.h> |
24 | #include <linux/debugfs.h> | |
8fbbfd21 SW |
25 | #include <linux/kthread.h> |
26 | #include <linux/freezer.h> | |
c4f68a13 BM |
27 | #include <linux/workqueue.h> |
28 | #include <linux/jiffies.h> | |
bc015cb8 SW |
29 | #include <linux/rcupdate.h> |
30 | #include <linux/rculist_bl.h> | |
31 | #include <linux/bit_spinlock.h> | |
a245769f | 32 | #include <linux/percpu.h> |
4506a519 | 33 | #include <linux/list_sort.h> |
e66cf161 | 34 | #include <linux/lockref.h> |
88ffbf3e | 35 | #include <linux/rhashtable.h> |
b3b94faa DT |
36 | |
37 | #include "gfs2.h" | |
5c676f6d | 38 | #include "incore.h" |
b3b94faa DT |
39 | #include "glock.h" |
40 | #include "glops.h" | |
41 | #include "inode.h" | |
b3b94faa DT |
42 | #include "lops.h" |
43 | #include "meta_io.h" | |
44 | #include "quota.h" | |
45 | #include "super.h" | |
5c676f6d | 46 | #include "util.h" |
813e0c46 | 47 | #include "bmap.h" |
63997775 SW |
48 | #define CREATE_TRACE_POINTS |
49 | #include "trace_gfs2.h" | |
b3b94faa | 50 | |
6802e340 | 51 | struct gfs2_glock_iter { |
ba1ddcb6 | 52 | struct gfs2_sbd *sdp; /* incore superblock */ |
88ffbf3e | 53 | struct rhashtable_iter hti; /* rhashtable iterator */ |
ba1ddcb6 SW |
54 | struct gfs2_glock *gl; /* current glock struct */ |
55 | loff_t last_pos; /* last position */ | |
7c52b166 RP |
56 | }; |
57 | ||
b3b94faa DT |
58 | typedef void (*glock_examiner) (struct gfs2_glock * gl); |
59 | ||
6802e340 | 60 | static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target); |
dc732906 | 61 | static void __gfs2_glock_dq(struct gfs2_holder *gh); |
c4f68a13 | 62 | |
7c52b166 | 63 | static struct dentry *gfs2_root; |
c4f68a13 | 64 | static struct workqueue_struct *glock_workqueue; |
b94a170e | 65 | struct workqueue_struct *gfs2_delete_workqueue; |
97cc1025 SW |
66 | static LIST_HEAD(lru_list); |
67 | static atomic_t lru_count = ATOMIC_INIT(0); | |
eb8374e7 | 68 | static DEFINE_SPINLOCK(lru_lock); |
08bc2dbc | 69 | |
b6397893 | 70 | #define GFS2_GL_HASH_SHIFT 15 |
47a9a527 | 71 | #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT) |
087efdd3 | 72 | |
d296b15e | 73 | static const struct rhashtable_params ht_parms = { |
88ffbf3e | 74 | .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4, |
972b044e | 75 | .key_len = offsetofend(struct lm_lockname, ln_type), |
88ffbf3e BP |
76 | .key_offset = offsetof(struct gfs2_glock, gl_name), |
77 | .head_offset = offsetof(struct gfs2_glock, gl_node), | |
78 | }; | |
b3b94faa | 79 | |
88ffbf3e | 80 | static struct rhashtable gl_hash_table; |
b3b94faa | 81 | |
0515480a AG |
82 | #define GLOCK_WAIT_TABLE_BITS 12 |
83 | #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS) | |
84 | static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned; | |
85 | ||
86 | struct wait_glock_queue { | |
87 | struct lm_lockname *name; | |
88 | wait_queue_entry_t wait; | |
89 | }; | |
90 | ||
91 | static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode, | |
92 | int sync, void *key) | |
93 | { | |
94 | struct wait_glock_queue *wait_glock = | |
95 | container_of(wait, struct wait_glock_queue, wait); | |
96 | struct lm_lockname *wait_name = wait_glock->name; | |
97 | struct lm_lockname *wake_name = key; | |
98 | ||
99 | if (wake_name->ln_sbd != wait_name->ln_sbd || | |
100 | wake_name->ln_number != wait_name->ln_number || | |
101 | wake_name->ln_type != wait_name->ln_type) | |
102 | return 0; | |
103 | return autoremove_wake_function(wait, mode, sync, key); | |
104 | } | |
105 | ||
106 | static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name) | |
107 | { | |
605b0487 | 108 | u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0); |
0515480a AG |
109 | |
110 | return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS); | |
111 | } | |
112 | ||
0515480a AG |
113 | /** |
114 | * wake_up_glock - Wake up waiters on a glock | |
115 | * @gl: the glock | |
116 | */ | |
117 | static void wake_up_glock(struct gfs2_glock *gl) | |
118 | { | |
119 | wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name); | |
120 | ||
121 | if (waitqueue_active(wq)) | |
122 | __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name); | |
123 | } | |
124 | ||
961ae1d8 | 125 | static void gfs2_glock_dealloc(struct rcu_head *rcu) |
b3b94faa | 126 | { |
961ae1d8 | 127 | struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu); |
b3b94faa | 128 | |
15f2547b | 129 | kfree(gl->gl_lksb.sb_lvbptr); |
11d8b79e KC |
130 | if (gl->gl_ops->go_flags & GLOF_ASPACE) { |
131 | struct gfs2_glock_aspace *gla = | |
132 | container_of(gl, struct gfs2_glock_aspace, glock); | |
133 | kmem_cache_free(gfs2_glock_aspace_cachep, gla); | |
134 | } else | |
bc015cb8 | 135 | kmem_cache_free(gfs2_glock_cachep, gl); |
961ae1d8 AG |
136 | } |
137 | ||
a72d2401 BP |
138 | /** |
139 | * glock_blocked_by_withdraw - determine if we can still use a glock | |
140 | * @gl: the glock | |
141 | * | |
142 | * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted | |
143 | * when we're withdrawn. For example, to maintain metadata integrity, we should | |
144 | * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like | |
145 | * iopen or the transaction glocks may be safely used because none of their | |
146 | * metadata goes through the journal. So in general, we should disallow all | |
147 | * glocks that are journaled, and allow all the others. One exception is: | |
148 | * we need to allow our active journal to be promoted and demoted so others | |
149 | * may recover it and we can reacquire it when they're done. | |
150 | */ | |
151 | static bool glock_blocked_by_withdraw(struct gfs2_glock *gl) | |
152 | { | |
153 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
154 | ||
155 | if (likely(!gfs2_withdrawn(sdp))) | |
156 | return false; | |
157 | if (gl->gl_ops->go_flags & GLOF_NONDISK) | |
158 | return false; | |
159 | if (!sdp->sd_jdesc || | |
160 | gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr) | |
161 | return false; | |
162 | return true; | |
163 | } | |
164 | ||
961ae1d8 AG |
165 | void gfs2_glock_free(struct gfs2_glock *gl) |
166 | { | |
167 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
168 | ||
ea4e61c7 | 169 | gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0); |
0515480a AG |
170 | rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms); |
171 | smp_mb(); | |
172 | wake_up_glock(gl); | |
961ae1d8 | 173 | call_rcu(&gl->gl_rcu, gfs2_glock_dealloc); |
bc015cb8 SW |
174 | if (atomic_dec_and_test(&sdp->sd_glock_disposal)) |
175 | wake_up(&sdp->sd_glock_wait); | |
b3b94faa DT |
176 | } |
177 | ||
178 | /** | |
179 | * gfs2_glock_hold() - increment reference count on glock | |
180 | * @gl: The glock to hold | |
181 | * | |
182 | */ | |
183 | ||
71c1b213 | 184 | void gfs2_glock_hold(struct gfs2_glock *gl) |
b3b94faa | 185 | { |
e66cf161 SW |
186 | GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref)); |
187 | lockref_get(&gl->gl_lockref); | |
b3b94faa DT |
188 | } |
189 | ||
8ff22a6f BM |
190 | /** |
191 | * demote_ok - Check to see if it's ok to unlock a glock | |
192 | * @gl: the glock | |
193 | * | |
194 | * Returns: 1 if it's ok | |
195 | */ | |
196 | ||
197 | static int demote_ok(const struct gfs2_glock *gl) | |
198 | { | |
199 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
200 | ||
201 | if (gl->gl_state == LM_ST_UNLOCKED) | |
202 | return 0; | |
dc732906 BP |
203 | /* |
204 | * Note that demote_ok is used for the lru process of disposing of | |
205 | * glocks. For this purpose, we don't care if the glock's holders | |
206 | * have the HIF_MAY_DEMOTE flag set or not. If someone is using | |
207 | * them, don't demote. | |
208 | */ | |
f42ab085 | 209 | if (!list_empty(&gl->gl_holders)) |
8ff22a6f BM |
210 | return 0; |
211 | if (glops->go_demote_ok) | |
212 | return glops->go_demote_ok(gl); | |
213 | return 1; | |
214 | } | |
215 | ||
bc015cb8 | 216 | |
29687a2a SW |
217 | void gfs2_glock_add_to_lru(struct gfs2_glock *gl) |
218 | { | |
7881ef3f RL |
219 | if (!(gl->gl_ops->go_flags & GLOF_LRU)) |
220 | return; | |
221 | ||
29687a2a SW |
222 | spin_lock(&lru_lock); |
223 | ||
38a618db | 224 | list_move_tail(&gl->gl_lru, &lru_list); |
7881ef3f RL |
225 | |
226 | if (!test_bit(GLF_LRU, &gl->gl_flags)) { | |
227 | set_bit(GLF_LRU, &gl->gl_flags); | |
29687a2a | 228 | atomic_inc(&lru_count); |
7881ef3f | 229 | } |
29687a2a | 230 | |
29687a2a SW |
231 | spin_unlock(&lru_lock); |
232 | } | |
233 | ||
8f6cb409 | 234 | static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl) |
f42ab085 | 235 | { |
645ebd49 BP |
236 | if (!(gl->gl_ops->go_flags & GLOF_LRU)) |
237 | return; | |
238 | ||
8f6cb409 | 239 | spin_lock(&lru_lock); |
7881ef3f | 240 | if (test_bit(GLF_LRU, &gl->gl_flags)) { |
f42ab085 SW |
241 | list_del_init(&gl->gl_lru); |
242 | atomic_dec(&lru_count); | |
243 | clear_bit(GLF_LRU, &gl->gl_flags); | |
244 | } | |
245 | spin_unlock(&lru_lock); | |
246 | } | |
247 | ||
6b0c7440 AG |
248 | /* |
249 | * Enqueue the glock on the work queue. Passes one glock reference on to the | |
250 | * work queue. | |
b3b94faa | 251 | */ |
6b0c7440 AG |
252 | static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) { |
253 | if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) { | |
254 | /* | |
255 | * We are holding the lockref spinlock, and the work was still | |
256 | * queued above. The queued work (glock_work_func) takes that | |
257 | * spinlock before dropping its glock reference(s), so it | |
258 | * cannot have dropped them in the meantime. | |
259 | */ | |
260 | GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2); | |
261 | gl->gl_lockref.count--; | |
262 | } | |
263 | } | |
b3b94faa | 264 | |
6b0c7440 AG |
265 | static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) { |
266 | spin_lock(&gl->gl_lockref.lock); | |
267 | __gfs2_glock_queue_work(gl, delay); | |
268 | spin_unlock(&gl->gl_lockref.lock); | |
269 | } | |
270 | ||
271 | static void __gfs2_glock_put(struct gfs2_glock *gl) | |
b3b94faa | 272 | { |
15562c43 | 273 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
bc015cb8 | 274 | struct address_space *mapping = gfs2_glock2aspace(gl); |
b3b94faa | 275 | |
e66cf161 SW |
276 | lockref_mark_dead(&gl->gl_lockref); |
277 | ||
8f6cb409 | 278 | gfs2_glock_remove_from_lru(gl); |
e66cf161 | 279 | spin_unlock(&gl->gl_lockref.lock); |
e66cf161 | 280 | GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders)); |
ee1e2c77 BP |
281 | if (mapping) { |
282 | truncate_inode_pages_final(mapping); | |
283 | if (!gfs2_withdrawn(sdp)) | |
7716506a | 284 | GLOCK_BUG_ON(gl, !mapping_empty(mapping)); |
ee1e2c77 | 285 | } |
e66cf161 SW |
286 | trace_gfs2_glock_put(gl); |
287 | sdp->sd_lockstruct.ls_ops->lm_put_lock(gl); | |
b3b94faa DT |
288 | } |
289 | ||
71c1b213 AG |
290 | /* |
291 | * Cause the glock to be put in work queue context. | |
292 | */ | |
293 | void gfs2_glock_queue_put(struct gfs2_glock *gl) | |
294 | { | |
295 | gfs2_glock_queue_work(gl, 0); | |
296 | } | |
297 | ||
6b0c7440 AG |
298 | /** |
299 | * gfs2_glock_put() - Decrement reference count on glock | |
300 | * @gl: The glock to put | |
301 | * | |
302 | */ | |
303 | ||
304 | void gfs2_glock_put(struct gfs2_glock *gl) | |
305 | { | |
306 | if (lockref_put_or_lock(&gl->gl_lockref)) | |
307 | return; | |
308 | ||
309 | __gfs2_glock_put(gl); | |
310 | } | |
311 | ||
6802e340 | 312 | /** |
61444649 | 313 | * may_grant - check if it's ok to grant a new lock |
6802e340 | 314 | * @gl: The glock |
61444649 | 315 | * @current_gh: One of the current holders of @gl |
6802e340 SW |
316 | * @gh: The lock request which we wish to grant |
317 | * | |
61444649 AG |
318 | * With our current compatibility rules, if a glock has one or more active |
319 | * holders (HIF_HOLDER flag set), any of those holders can be passed in as | |
320 | * @current_gh; they are all the same as far as compatibility with the new @gh | |
321 | * goes. | |
322 | * | |
323 | * Returns true if it's ok to grant the lock. | |
6802e340 SW |
324 | */ |
325 | ||
61444649 AG |
326 | static inline bool may_grant(struct gfs2_glock *gl, |
327 | struct gfs2_holder *current_gh, | |
328 | struct gfs2_holder *gh) | |
6802e340 | 329 | { |
61444649 AG |
330 | if (current_gh) { |
331 | GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags)); | |
332 | ||
333 | switch(current_gh->gh_state) { | |
334 | case LM_ST_EXCLUSIVE: | |
335 | /* | |
336 | * Here we make a special exception to grant holders | |
337 | * who agree to share the EX lock with other holders | |
338 | * who also have the bit set. If the original holder | |
339 | * has the LM_FLAG_NODE_SCOPE bit set, we grant more | |
340 | * holders with the bit set. | |
341 | */ | |
342 | return gh->gh_state == LM_ST_EXCLUSIVE && | |
343 | (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) && | |
344 | (gh->gh_flags & LM_FLAG_NODE_SCOPE); | |
345 | ||
346 | case LM_ST_SHARED: | |
347 | case LM_ST_DEFERRED: | |
348 | return gh->gh_state == current_gh->gh_state; | |
349 | ||
350 | default: | |
351 | return false; | |
352 | } | |
06e908cd | 353 | } |
61444649 | 354 | |
6802e340 | 355 | if (gl->gl_state == gh->gh_state) |
61444649 | 356 | return true; |
6802e340 | 357 | if (gh->gh_flags & GL_EXACT) |
61444649 | 358 | return false; |
209806ab | 359 | if (gl->gl_state == LM_ST_EXCLUSIVE) { |
61444649 AG |
360 | return gh->gh_state == LM_ST_SHARED || |
361 | gh->gh_state == LM_ST_DEFERRED; | |
209806ab | 362 | } |
61444649 AG |
363 | if (gh->gh_flags & LM_FLAG_ANY) |
364 | return gl->gl_state != LM_ST_UNLOCKED; | |
365 | return false; | |
6802e340 SW |
366 | } |
367 | ||
368 | static void gfs2_holder_wake(struct gfs2_holder *gh) | |
369 | { | |
370 | clear_bit(HIF_WAIT, &gh->gh_iflags); | |
4e857c58 | 371 | smp_mb__after_atomic(); |
6802e340 | 372 | wake_up_bit(&gh->gh_iflags, HIF_WAIT); |
ad26967b BP |
373 | if (gh->gh_flags & GL_ASYNC) { |
374 | struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd; | |
375 | ||
376 | wake_up(&sdp->sd_async_glock_wait); | |
377 | } | |
6802e340 SW |
378 | } |
379 | ||
d5341a92 SW |
380 | /** |
381 | * do_error - Something unexpected has happened during a lock request | |
c551f66c LJ |
382 | * @gl: The glock |
383 | * @ret: The status from the DLM | |
d5341a92 SW |
384 | */ |
385 | ||
a527b38e | 386 | static void do_error(struct gfs2_glock *gl, const int ret) |
d5341a92 SW |
387 | { |
388 | struct gfs2_holder *gh, *tmp; | |
389 | ||
390 | list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) { | |
dc732906 | 391 | if (!test_bit(HIF_WAIT, &gh->gh_iflags)) |
d5341a92 SW |
392 | continue; |
393 | if (ret & LM_OUT_ERROR) | |
394 | gh->gh_error = -EIO; | |
395 | else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) | |
396 | gh->gh_error = GLR_TRYFAILED; | |
397 | else | |
398 | continue; | |
399 | list_del_init(&gh->gh_list); | |
400 | trace_gfs2_glock_queue(gh, 0); | |
401 | gfs2_holder_wake(gh); | |
402 | } | |
403 | } | |
404 | ||
dc732906 BP |
405 | /** |
406 | * demote_incompat_holders - demote incompatible demoteable holders | |
407 | * @gl: the glock we want to promote | |
0befb851 BP |
408 | * @current_gh: the newly promoted holder |
409 | * | |
410 | * We're passing the newly promoted holder in @current_gh, but actually, any of | |
411 | * the strong holders would do. | |
dc732906 BP |
412 | */ |
413 | static void demote_incompat_holders(struct gfs2_glock *gl, | |
0befb851 | 414 | struct gfs2_holder *current_gh) |
dc732906 | 415 | { |
a7ac203d | 416 | struct gfs2_holder *gh, *tmp; |
dc732906 BP |
417 | |
418 | /* | |
419 | * Demote incompatible holders before we make ourselves eligible. | |
420 | * (This holder may or may not allow auto-demoting, but we don't want | |
421 | * to demote the new holder before it's even granted.) | |
422 | */ | |
a7ac203d | 423 | list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) { |
dc732906 BP |
424 | /* |
425 | * Since holders are at the front of the list, we stop when we | |
426 | * find the first non-holder. | |
427 | */ | |
428 | if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
429 | return; | |
0befb851 BP |
430 | if (gh == current_gh) |
431 | continue; | |
dc732906 | 432 | if (test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags) && |
0befb851 | 433 | !may_grant(gl, current_gh, gh)) { |
dc732906 BP |
434 | /* |
435 | * We should not recurse into do_promote because | |
436 | * __gfs2_glock_dq only calls handle_callback, | |
437 | * gfs2_glock_add_to_lru and __gfs2_glock_queue_work. | |
438 | */ | |
439 | __gfs2_glock_dq(gh); | |
440 | } | |
441 | } | |
442 | } | |
443 | ||
61444649 AG |
444 | /** |
445 | * find_first_holder - find the first "holder" gh | |
446 | * @gl: the glock | |
447 | */ | |
448 | ||
449 | static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl) | |
450 | { | |
451 | struct gfs2_holder *gh; | |
452 | ||
453 | if (!list_empty(&gl->gl_holders)) { | |
454 | gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, | |
455 | gh_list); | |
456 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
457 | return gh; | |
458 | } | |
459 | return NULL; | |
460 | } | |
461 | ||
dc732906 BP |
462 | /** |
463 | * find_first_strong_holder - find the first non-demoteable holder | |
464 | * @gl: the glock | |
465 | * | |
466 | * Find the first holder that doesn't have the HIF_MAY_DEMOTE flag set. | |
467 | */ | |
468 | static inline struct gfs2_holder * | |
469 | find_first_strong_holder(struct gfs2_glock *gl) | |
470 | { | |
471 | struct gfs2_holder *gh; | |
472 | ||
473 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { | |
474 | if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
475 | return NULL; | |
476 | if (!test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags)) | |
477 | return gh; | |
478 | } | |
479 | return NULL; | |
480 | } | |
481 | ||
e6f85600 BP |
482 | /* |
483 | * gfs2_instantiate - Call the glops instantiate function | |
3c5c67ec | 484 | * @gh: The glock holder |
e6f85600 | 485 | * |
53d69132 | 486 | * Returns: 0 if instantiate was successful, or error. |
e6f85600 | 487 | */ |
f2e70d8f | 488 | int gfs2_instantiate(struct gfs2_holder *gh) |
e6f85600 BP |
489 | { |
490 | struct gfs2_glock *gl = gh->gh_gl; | |
491 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
f2e70d8f BP |
492 | int ret; |
493 | ||
494 | again: | |
495 | if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags)) | |
86c30a01 | 496 | goto done; |
e6f85600 | 497 | |
f2e70d8f BP |
498 | /* |
499 | * Since we unlock the lockref lock, we set a flag to indicate | |
500 | * instantiate is in progress. | |
501 | */ | |
7a92deaa | 502 | if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) { |
f2e70d8f BP |
503 | wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG, |
504 | TASK_UNINTERRUPTIBLE); | |
505 | /* | |
506 | * Here we just waited for a different instantiate to finish. | |
507 | * But that may not have been successful, as when a process | |
508 | * locks an inode glock _before_ it has an actual inode to | |
509 | * instantiate into. So we check again. This process might | |
510 | * have an inode to instantiate, so might be successful. | |
511 | */ | |
512 | goto again; | |
513 | } | |
514 | ||
5f38a4d3 | 515 | ret = glops->go_instantiate(gl); |
f2e70d8f BP |
516 | if (!ret) |
517 | clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags); | |
7a92deaa | 518 | clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags); |
86c30a01 AG |
519 | if (ret) |
520 | return ret; | |
521 | ||
522 | done: | |
523 | if (glops->go_held) | |
524 | return glops->go_held(gh); | |
525 | return 0; | |
e6f85600 BP |
526 | } |
527 | ||
6802e340 SW |
528 | /** |
529 | * do_promote - promote as many requests as possible on the current queue | |
530 | * @gl: The glock | |
531 | * | |
de3f906f | 532 | * Returns: 1 if there is a blocked holder at the head of the list |
6802e340 SW |
533 | */ |
534 | ||
535 | static int do_promote(struct gfs2_glock *gl) | |
536 | { | |
6feaec81 | 537 | struct gfs2_holder *gh, *current_gh; |
dc732906 | 538 | bool incompat_holders_demoted = false; |
6802e340 | 539 | |
8f0028fc | 540 | current_gh = find_first_strong_holder(gl); |
6feaec81 | 541 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { |
60d8bae9 | 542 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) |
6802e340 | 543 | continue; |
8f0028fc | 544 | if (!may_grant(gl, current_gh, gh)) { |
17a6ecee | 545 | /* |
565f82b5 BP |
546 | * If we get here, it means we may not grant this |
547 | * holder for some reason. If this holder is at the | |
548 | * head of the list, it means we have a blocked holder | |
549 | * at the head, so return 1. | |
17a6ecee | 550 | */ |
5a27a43e | 551 | if (list_is_first(&gh->gh_list, &gl->gl_holders)) |
17a6ecee BP |
552 | return 1; |
553 | do_error(gl, 0); | |
554 | break; | |
555 | } | |
17a6ecee BP |
556 | set_bit(HIF_HOLDER, &gh->gh_iflags); |
557 | trace_gfs2_promote(gh); | |
558 | gfs2_holder_wake(gh); | |
17a6ecee | 559 | if (!incompat_holders_demoted) { |
0befb851 | 560 | current_gh = gh; |
8f0028fc | 561 | demote_incompat_holders(gl, current_gh); |
17a6ecee | 562 | incompat_holders_demoted = true; |
17a6ecee | 563 | } |
6802e340 SW |
564 | } |
565 | return 0; | |
566 | } | |
567 | ||
6802e340 SW |
568 | /** |
569 | * find_first_waiter - find the first gh that's waiting for the glock | |
570 | * @gl: the glock | |
571 | */ | |
572 | ||
573 | static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl) | |
574 | { | |
575 | struct gfs2_holder *gh; | |
576 | ||
577 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { | |
578 | if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
579 | return gh; | |
580 | } | |
581 | return NULL; | |
582 | } | |
583 | ||
584 | /** | |
585 | * state_change - record that the glock is now in a different state | |
586 | * @gl: the glock | |
c551f66c | 587 | * @new_state: the new state |
6802e340 SW |
588 | */ |
589 | ||
590 | static void state_change(struct gfs2_glock *gl, unsigned int new_state) | |
591 | { | |
592 | int held1, held2; | |
593 | ||
594 | held1 = (gl->gl_state != LM_ST_UNLOCKED); | |
595 | held2 = (new_state != LM_ST_UNLOCKED); | |
596 | ||
597 | if (held1 != held2) { | |
e66cf161 | 598 | GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref)); |
6802e340 | 599 | if (held2) |
e66cf161 | 600 | gl->gl_lockref.count++; |
6802e340 | 601 | else |
e66cf161 | 602 | gl->gl_lockref.count--; |
6802e340 | 603 | } |
7cf8dcd3 BP |
604 | if (new_state != gl->gl_target) |
605 | /* shorten our minimum hold time */ | |
606 | gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR, | |
607 | GL_GLOCK_MIN_HOLD); | |
6802e340 SW |
608 | gl->gl_state = new_state; |
609 | gl->gl_tchange = jiffies; | |
610 | } | |
611 | ||
35b6f8fb AG |
612 | static void gfs2_set_demote(struct gfs2_glock *gl) |
613 | { | |
614 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
615 | ||
616 | set_bit(GLF_DEMOTE, &gl->gl_flags); | |
617 | smp_mb(); | |
618 | wake_up(&sdp->sd_async_glock_wait); | |
619 | } | |
620 | ||
6802e340 SW |
621 | static void gfs2_demote_wake(struct gfs2_glock *gl) |
622 | { | |
623 | gl->gl_demote_state = LM_ST_EXCLUSIVE; | |
624 | clear_bit(GLF_DEMOTE, &gl->gl_flags); | |
4e857c58 | 625 | smp_mb__after_atomic(); |
6802e340 SW |
626 | wake_up_bit(&gl->gl_flags, GLF_DEMOTE); |
627 | } | |
628 | ||
629 | /** | |
630 | * finish_xmote - The DLM has replied to one of our lock requests | |
631 | * @gl: The glock | |
632 | * @ret: The status from the DLM | |
633 | * | |
634 | */ | |
635 | ||
636 | static void finish_xmote(struct gfs2_glock *gl, unsigned int ret) | |
637 | { | |
638 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
639 | struct gfs2_holder *gh; | |
640 | unsigned state = ret & LM_OUT_ST_MASK; | |
641 | ||
f3dd1649 | 642 | spin_lock(&gl->gl_lockref.lock); |
63997775 | 643 | trace_gfs2_glock_state_change(gl, state); |
6802e340 SW |
644 | state_change(gl, state); |
645 | gh = find_first_waiter(gl); | |
646 | ||
647 | /* Demote to UN request arrived during demote to SH or DF */ | |
648 | if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) && | |
649 | state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED) | |
650 | gl->gl_target = LM_ST_UNLOCKED; | |
651 | ||
652 | /* Check for state != intended state */ | |
653 | if (unlikely(state != gl->gl_target)) { | |
1fc05c8d AG |
654 | if (gh && (ret & LM_OUT_CANCELED)) |
655 | gfs2_holder_wake(gh); | |
6802e340 SW |
656 | if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) { |
657 | /* move to back of queue and try next entry */ | |
658 | if (ret & LM_OUT_CANCELED) { | |
659 | if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0) | |
660 | list_move_tail(&gh->gh_list, &gl->gl_holders); | |
661 | gh = find_first_waiter(gl); | |
662 | gl->gl_target = gh->gh_state; | |
663 | goto retry; | |
664 | } | |
665 | /* Some error or failed "try lock" - report it */ | |
666 | if ((ret & LM_OUT_ERROR) || | |
667 | (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) { | |
668 | gl->gl_target = gl->gl_state; | |
669 | do_error(gl, ret); | |
670 | goto out; | |
671 | } | |
672 | } | |
673 | switch(state) { | |
674 | /* Unlocked due to conversion deadlock, try again */ | |
675 | case LM_ST_UNLOCKED: | |
676 | retry: | |
677 | do_xmote(gl, gh, gl->gl_target); | |
678 | break; | |
679 | /* Conversion fails, unlock and try again */ | |
680 | case LM_ST_SHARED: | |
681 | case LM_ST_DEFERRED: | |
682 | do_xmote(gl, gh, LM_ST_UNLOCKED); | |
683 | break; | |
684 | default: /* Everything else */ | |
e54c78a2 BP |
685 | fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n", |
686 | gl->gl_target, state); | |
6802e340 SW |
687 | GLOCK_BUG_ON(gl, 1); |
688 | } | |
f3dd1649 | 689 | spin_unlock(&gl->gl_lockref.lock); |
6802e340 SW |
690 | return; |
691 | } | |
692 | ||
693 | /* Fast path - we got what we asked for */ | |
694 | if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) | |
695 | gfs2_demote_wake(gl); | |
696 | if (state != LM_ST_UNLOCKED) { | |
697 | if (glops->go_xmote_bh) { | |
de3f906f AG |
698 | int rv; |
699 | ||
f3dd1649 | 700 | spin_unlock(&gl->gl_lockref.lock); |
f68effb3 | 701 | rv = glops->go_xmote_bh(gl); |
f3dd1649 | 702 | spin_lock(&gl->gl_lockref.lock); |
6802e340 SW |
703 | if (rv) { |
704 | do_error(gl, rv); | |
705 | goto out; | |
706 | } | |
707 | } | |
de3f906f | 708 | do_promote(gl); |
6802e340 SW |
709 | } |
710 | out: | |
711 | clear_bit(GLF_LOCK, &gl->gl_flags); | |
f3dd1649 | 712 | spin_unlock(&gl->gl_lockref.lock); |
6802e340 SW |
713 | } |
714 | ||
865cc3e9 BP |
715 | static bool is_system_glock(struct gfs2_glock *gl) |
716 | { | |
717 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
718 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
719 | ||
720 | if (gl == m_ip->i_gl) | |
721 | return true; | |
722 | return false; | |
723 | } | |
724 | ||
6802e340 SW |
725 | /** |
726 | * do_xmote - Calls the DLM to change the state of a lock | |
727 | * @gl: The lock state | |
728 | * @gh: The holder (only for promotes) | |
729 | * @target: The target lock state | |
730 | * | |
731 | */ | |
732 | ||
733 | static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target) | |
f3dd1649 AG |
734 | __releases(&gl->gl_lockref.lock) |
735 | __acquires(&gl->gl_lockref.lock) | |
6802e340 SW |
736 | { |
737 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
15562c43 | 738 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
b58bf407 | 739 | unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0); |
6802e340 SW |
740 | int ret; |
741 | ||
601ef0d5 BP |
742 | if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) && |
743 | gh && !(gh->gh_flags & LM_FLAG_NOEXP)) | |
0d1c7ae9 | 744 | return; |
6802e340 SW |
745 | lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP | |
746 | LM_FLAG_PRIORITY); | |
921169ca SW |
747 | GLOCK_BUG_ON(gl, gl->gl_state == target); |
748 | GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target); | |
6802e340 SW |
749 | if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) && |
750 | glops->go_inval) { | |
d99724c3 BP |
751 | /* |
752 | * If another process is already doing the invalidate, let that | |
753 | * finish first. The glock state machine will get back to this | |
754 | * holder again later. | |
755 | */ | |
756 | if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS, | |
757 | &gl->gl_flags)) | |
758 | return; | |
6802e340 SW |
759 | do_error(gl, 0); /* Fail queued try locks */ |
760 | } | |
47a25380 | 761 | gl->gl_req = target; |
a245769f SW |
762 | set_bit(GLF_BLOCKING, &gl->gl_flags); |
763 | if ((gl->gl_req == LM_ST_UNLOCKED) || | |
764 | (gl->gl_state == LM_ST_EXCLUSIVE) || | |
765 | (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB))) | |
766 | clear_bit(GLF_BLOCKING, &gl->gl_flags); | |
f3dd1649 | 767 | spin_unlock(&gl->gl_lockref.lock); |
1c634f94 BP |
768 | if (glops->go_sync) { |
769 | ret = glops->go_sync(gl); | |
770 | /* If we had a problem syncing (due to io errors or whatever, | |
771 | * we should not invalidate the metadata or tell dlm to | |
772 | * release the glock to other nodes. | |
773 | */ | |
774 | if (ret) { | |
775 | if (cmpxchg(&sdp->sd_log_error, 0, ret)) { | |
776 | fs_err(sdp, "Error %d syncing glock \n", ret); | |
777 | gfs2_dump_glock(NULL, gl, true); | |
778 | } | |
b11e1a84 | 779 | goto skip_inval; |
1c634f94 BP |
780 | } |
781 | } | |
33dbd1e4 BP |
782 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) { |
783 | /* | |
784 | * The call to go_sync should have cleared out the ail list. | |
785 | * If there are still items, we have a problem. We ought to | |
786 | * withdraw, but we can't because the withdraw code also uses | |
787 | * glocks. Warn about the error, dump the glock, then fall | |
788 | * through and wait for logd to do the withdraw for us. | |
789 | */ | |
790 | if ((atomic_read(&gl->gl_ail_count) != 0) && | |
791 | (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) { | |
ea4e61c7 BP |
792 | gfs2_glock_assert_warn(gl, |
793 | !atomic_read(&gl->gl_ail_count)); | |
33dbd1e4 BP |
794 | gfs2_dump_glock(NULL, gl, true); |
795 | } | |
6802e340 | 796 | glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA); |
33dbd1e4 BP |
797 | clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags); |
798 | } | |
6802e340 | 799 | |
b11e1a84 | 800 | skip_inval: |
6802e340 | 801 | gfs2_glock_hold(gl); |
d93ae386 BP |
802 | /* |
803 | * Check for an error encountered since we called go_sync and go_inval. | |
804 | * If so, we can't withdraw from the glock code because the withdraw | |
805 | * code itself uses glocks (see function signal_our_withdraw) to | |
806 | * change the mount to read-only. Most importantly, we must not call | |
807 | * dlm to unlock the glock until the journal is in a known good state | |
808 | * (after journal replay) otherwise other nodes may use the object | |
809 | * (rgrp or dinode) and then later, journal replay will corrupt the | |
810 | * file system. The best we can do here is wait for the logd daemon | |
811 | * to see sd_log_error and withdraw, and in the meantime, requeue the | |
812 | * work for later. | |
813 | * | |
865cc3e9 BP |
814 | * We make a special exception for some system glocks, such as the |
815 | * system statfs inode glock, which needs to be granted before the | |
816 | * gfs2_quotad daemon can exit, and that exit needs to finish before | |
817 | * we can unmount the withdrawn file system. | |
818 | * | |
d93ae386 BP |
819 | * However, if we're just unlocking the lock (say, for unmount, when |
820 | * gfs2_gl_hash_clear calls clear_glock) and recovery is complete | |
821 | * then it's okay to tell dlm to unlock it. | |
822 | */ | |
823 | if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp))) | |
824 | gfs2_withdraw_delayed(sdp); | |
865cc3e9 BP |
825 | if (glock_blocked_by_withdraw(gl) && |
826 | (target != LM_ST_UNLOCKED || | |
827 | test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) { | |
828 | if (!is_system_glock(gl)) { | |
d93ae386 BP |
829 | gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD); |
830 | goto out; | |
865cc3e9 BP |
831 | } else { |
832 | clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags); | |
d93ae386 BP |
833 | } |
834 | } | |
835 | ||
921169ca SW |
836 | if (sdp->sd_lockstruct.ls_ops->lm_lock) { |
837 | /* lock_dlm */ | |
838 | ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags); | |
3e11e530 BM |
839 | if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED && |
840 | target == LM_ST_UNLOCKED && | |
841 | test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) { | |
842 | finish_xmote(gl, target); | |
6b0c7440 | 843 | gfs2_glock_queue_work(gl, 0); |
d93ae386 | 844 | } else if (ret) { |
e54c78a2 | 845 | fs_err(sdp, "lm_lock ret %d\n", ret); |
eb43e660 | 846 | GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp)); |
dba2d70c | 847 | } |
921169ca SW |
848 | } else { /* lock_nolock */ |
849 | finish_xmote(gl, target); | |
6b0c7440 | 850 | gfs2_glock_queue_work(gl, 0); |
6802e340 | 851 | } |
d93ae386 | 852 | out: |
f3dd1649 | 853 | spin_lock(&gl->gl_lockref.lock); |
6802e340 SW |
854 | } |
855 | ||
6802e340 SW |
856 | /** |
857 | * run_queue - do all outstanding tasks related to a glock | |
858 | * @gl: The glock in question | |
859 | * @nonblock: True if we must not block in run_queue | |
860 | * | |
861 | */ | |
862 | ||
863 | static void run_queue(struct gfs2_glock *gl, const int nonblock) | |
f3dd1649 AG |
864 | __releases(&gl->gl_lockref.lock) |
865 | __acquires(&gl->gl_lockref.lock) | |
6802e340 SW |
866 | { |
867 | struct gfs2_holder *gh = NULL; | |
868 | ||
869 | if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) | |
870 | return; | |
871 | ||
872 | GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)); | |
873 | ||
874 | if (test_bit(GLF_DEMOTE, &gl->gl_flags) && | |
875 | gl->gl_demote_state != gl->gl_state) { | |
876 | if (find_first_holder(gl)) | |
d8348de0 | 877 | goto out_unlock; |
6802e340 SW |
878 | if (nonblock) |
879 | goto out_sched; | |
880 | set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags); | |
265d529c | 881 | GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE); |
6802e340 SW |
882 | gl->gl_target = gl->gl_demote_state; |
883 | } else { | |
884 | if (test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
885 | gfs2_demote_wake(gl); | |
de3f906f | 886 | if (do_promote(gl) == 0) |
d8348de0 | 887 | goto out_unlock; |
6802e340 SW |
888 | gh = find_first_waiter(gl); |
889 | gl->gl_target = gh->gh_state; | |
890 | if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) | |
891 | do_error(gl, 0); /* Fail queued try locks */ | |
892 | } | |
893 | do_xmote(gl, gh, gl->gl_target); | |
894 | return; | |
895 | ||
896 | out_sched: | |
7e71c55e | 897 | clear_bit(GLF_LOCK, &gl->gl_flags); |
4e857c58 | 898 | smp_mb__after_atomic(); |
e66cf161 | 899 | gl->gl_lockref.count++; |
6b0c7440 | 900 | __gfs2_glock_queue_work(gl, 0); |
7e71c55e SW |
901 | return; |
902 | ||
d8348de0 | 903 | out_unlock: |
6802e340 | 904 | clear_bit(GLF_LOCK, &gl->gl_flags); |
4e857c58 | 905 | smp_mb__after_atomic(); |
7e71c55e | 906 | return; |
6802e340 SW |
907 | } |
908 | ||
f286d627 AG |
909 | void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation) |
910 | { | |
911 | struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr; | |
912 | ||
913 | if (ri->ri_magic == 0) | |
914 | ri->ri_magic = cpu_to_be32(GFS2_MAGIC); | |
915 | if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC)) | |
916 | ri->ri_generation_deleted = cpu_to_be64(generation); | |
917 | } | |
918 | ||
919 | bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation) | |
920 | { | |
921 | struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr; | |
922 | ||
923 | if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC)) | |
924 | return false; | |
925 | return generation <= be64_to_cpu(ri->ri_generation_deleted); | |
926 | } | |
927 | ||
9e8990de AG |
928 | static void gfs2_glock_poke(struct gfs2_glock *gl) |
929 | { | |
930 | int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP; | |
931 | struct gfs2_holder gh; | |
932 | int error; | |
933 | ||
b016d9a8 | 934 | __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_); |
c07bfb4d | 935 | error = gfs2_glock_nq(&gh); |
9e8990de AG |
936 | if (!error) |
937 | gfs2_glock_dq(&gh); | |
c07bfb4d | 938 | gfs2_holder_uninit(&gh); |
9e8990de AG |
939 | } |
940 | ||
8c7b9262 AG |
941 | static bool gfs2_try_evict(struct gfs2_glock *gl) |
942 | { | |
943 | struct gfs2_inode *ip; | |
944 | bool evicted = false; | |
945 | ||
946 | /* | |
947 | * If there is contention on the iopen glock and we have an inode, try | |
948 | * to grab and release the inode so that it can be evicted. This will | |
949 | * allow the remote node to go ahead and delete the inode without us | |
950 | * having to do it, which will avoid rgrp glock thrashing. | |
951 | * | |
952 | * The remote node is likely still holding the corresponding inode | |
953 | * glock, so it will run before we get to verify that the delete has | |
954 | * happened below. | |
955 | */ | |
956 | spin_lock(&gl->gl_lockref.lock); | |
957 | ip = gl->gl_object; | |
958 | if (ip && !igrab(&ip->i_inode)) | |
959 | ip = NULL; | |
960 | spin_unlock(&gl->gl_lockref.lock); | |
961 | if (ip) { | |
9e8990de AG |
962 | struct gfs2_glock *inode_gl = NULL; |
963 | ||
b0dcffd8 | 964 | gl->gl_no_formal_ino = ip->i_no_formal_ino; |
8c7b9262 AG |
965 | set_bit(GIF_DEFERRED_DELETE, &ip->i_flags); |
966 | d_prune_aliases(&ip->i_inode); | |
967 | iput(&ip->i_inode); | |
968 | ||
969 | /* If the inode was evicted, gl->gl_object will now be NULL. */ | |
970 | spin_lock(&gl->gl_lockref.lock); | |
971 | ip = gl->gl_object; | |
9e8990de AG |
972 | if (ip) { |
973 | inode_gl = ip->i_gl; | |
974 | lockref_get(&inode_gl->gl_lockref); | |
8c7b9262 | 975 | clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags); |
9e8990de | 976 | } |
8c7b9262 | 977 | spin_unlock(&gl->gl_lockref.lock); |
9e8990de AG |
978 | if (inode_gl) { |
979 | gfs2_glock_poke(inode_gl); | |
980 | gfs2_glock_put(inode_gl); | |
981 | } | |
8c7b9262 AG |
982 | evicted = !ip; |
983 | } | |
984 | return evicted; | |
985 | } | |
986 | ||
b94a170e BM |
987 | static void delete_work_func(struct work_struct *work) |
988 | { | |
a0e3cc65 AG |
989 | struct delayed_work *dwork = to_delayed_work(work); |
990 | struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete); | |
15562c43 | 991 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
ec5ec66b | 992 | struct inode *inode; |
044b9414 SW |
993 | u64 no_addr = gl->gl_name.ln_number; |
994 | ||
a0e3cc65 AG |
995 | spin_lock(&gl->gl_lockref.lock); |
996 | clear_bit(GLF_PENDING_DELETE, &gl->gl_flags); | |
997 | spin_unlock(&gl->gl_lockref.lock); | |
998 | ||
8c7b9262 AG |
999 | if (test_bit(GLF_DEMOTE, &gl->gl_flags)) { |
1000 | /* | |
1001 | * If we can evict the inode, give the remote node trying to | |
1002 | * delete the inode some time before verifying that the delete | |
1003 | * has happened. Otherwise, if we cause contention on the inode glock | |
1004 | * immediately, the remote node will think that we still have | |
1005 | * the inode in use, and so it will give up waiting. | |
9e8990de AG |
1006 | * |
1007 | * If we can't evict the inode, signal to the remote node that | |
1008 | * the inode is still in use. We'll later try to delete the | |
1009 | * inode locally in gfs2_evict_inode. | |
1010 | * | |
1011 | * FIXME: We only need to verify that the remote node has | |
1012 | * deleted the inode because nodes before this remote delete | |
1013 | * rework won't cooperate. At a later time, when we no longer | |
1014 | * care about compatibility with such nodes, we can skip this | |
1015 | * step entirely. | |
8c7b9262 AG |
1016 | */ |
1017 | if (gfs2_try_evict(gl)) { | |
1018 | if (gfs2_queue_delete_work(gl, 5 * HZ)) | |
1019 | return; | |
8c7b9262 | 1020 | } |
9e8990de | 1021 | goto out; |
8c7b9262 AG |
1022 | } |
1023 | ||
b0dcffd8 AG |
1024 | inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino, |
1025 | GFS2_BLKST_UNLINKED); | |
15a798f7 | 1026 | if (!IS_ERR_OR_NULL(inode)) { |
044b9414 SW |
1027 | d_prune_aliases(inode); |
1028 | iput(inode); | |
b94a170e | 1029 | } |
a4923865 | 1030 | out: |
b94a170e BM |
1031 | gfs2_glock_put(gl); |
1032 | } | |
1033 | ||
c4f68a13 BM |
1034 | static void glock_work_func(struct work_struct *work) |
1035 | { | |
6802e340 | 1036 | unsigned long delay = 0; |
c4f68a13 | 1037 | struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work); |
6b0c7440 | 1038 | unsigned int drop_refs = 1; |
c4f68a13 | 1039 | |
26bb7505 | 1040 | if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) { |
6802e340 | 1041 | finish_xmote(gl, gl->gl_reply); |
6b0c7440 | 1042 | drop_refs++; |
26bb7505 | 1043 | } |
f3dd1649 | 1044 | spin_lock(&gl->gl_lockref.lock); |
f90e5b5b | 1045 | if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) && |
265d529c SW |
1046 | gl->gl_state != LM_ST_UNLOCKED && |
1047 | gl->gl_demote_state != LM_ST_EXCLUSIVE) { | |
6802e340 | 1048 | unsigned long holdtime, now = jiffies; |
f90e5b5b | 1049 | |
7cf8dcd3 | 1050 | holdtime = gl->gl_tchange + gl->gl_hold_time; |
6802e340 SW |
1051 | if (time_before(now, holdtime)) |
1052 | delay = holdtime - now; | |
f90e5b5b BP |
1053 | |
1054 | if (!delay) { | |
1055 | clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags); | |
35b6f8fb | 1056 | gfs2_set_demote(gl); |
f90e5b5b | 1057 | } |
6802e340 SW |
1058 | } |
1059 | run_queue(gl, 0); | |
6b0c7440 AG |
1060 | if (delay) { |
1061 | /* Keep one glock reference for the work we requeue. */ | |
1062 | drop_refs--; | |
7cf8dcd3 BP |
1063 | if (gl->gl_name.ln_type != LM_TYPE_INODE) |
1064 | delay = 0; | |
6b0c7440 | 1065 | __gfs2_glock_queue_work(gl, delay); |
7cf8dcd3 | 1066 | } |
6b0c7440 AG |
1067 | |
1068 | /* | |
1069 | * Drop the remaining glock references manually here. (Mind that | |
1070 | * __gfs2_glock_queue_work depends on the lockref spinlock begin held | |
1071 | * here as well.) | |
1072 | */ | |
1073 | gl->gl_lockref.count -= drop_refs; | |
1074 | if (!gl->gl_lockref.count) { | |
1075 | __gfs2_glock_put(gl); | |
1076 | return; | |
1077 | } | |
1078 | spin_unlock(&gl->gl_lockref.lock); | |
c4f68a13 BM |
1079 | } |
1080 | ||
0515480a AG |
1081 | static struct gfs2_glock *find_insert_glock(struct lm_lockname *name, |
1082 | struct gfs2_glock *new) | |
1083 | { | |
1084 | struct wait_glock_queue wait; | |
a91323e2 | 1085 | wait_queue_head_t *wq = glock_waitqueue(name); |
0515480a AG |
1086 | struct gfs2_glock *gl; |
1087 | ||
a91323e2 AG |
1088 | wait.name = name; |
1089 | init_wait(&wait.wait); | |
1090 | wait.wait.func = glock_wake_function; | |
1091 | ||
0515480a | 1092 | again: |
a91323e2 | 1093 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
0515480a AG |
1094 | rcu_read_lock(); |
1095 | if (new) { | |
1096 | gl = rhashtable_lookup_get_insert_fast(&gl_hash_table, | |
1097 | &new->gl_node, ht_parms); | |
1098 | if (IS_ERR(gl)) | |
1099 | goto out; | |
1100 | } else { | |
1101 | gl = rhashtable_lookup_fast(&gl_hash_table, | |
1102 | name, ht_parms); | |
1103 | } | |
1104 | if (gl && !lockref_get_not_dead(&gl->gl_lockref)) { | |
1105 | rcu_read_unlock(); | |
1106 | schedule(); | |
1107 | goto again; | |
1108 | } | |
1109 | out: | |
1110 | rcu_read_unlock(); | |
a91323e2 | 1111 | finish_wait(wq, &wait.wait); |
0515480a AG |
1112 | return gl; |
1113 | } | |
1114 | ||
b3b94faa DT |
1115 | /** |
1116 | * gfs2_glock_get() - Get a glock, or create one if one doesn't exist | |
1117 | * @sdp: The GFS2 superblock | |
1118 | * @number: the lock number | |
1119 | * @glops: The glock_operations to use | |
1120 | * @create: If 0, don't create the glock if it doesn't exist | |
1121 | * @glp: the glock is returned here | |
1122 | * | |
1123 | * This does not lock a glock, just finds/creates structures for one. | |
1124 | * | |
1125 | * Returns: errno | |
1126 | */ | |
1127 | ||
cd915493 | 1128 | int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number, |
8fb4b536 | 1129 | const struct gfs2_glock_operations *glops, int create, |
b3b94faa DT |
1130 | struct gfs2_glock **glp) |
1131 | { | |
009d8518 | 1132 | struct super_block *s = sdp->sd_vfs; |
15562c43 BP |
1133 | struct lm_lockname name = { .ln_number = number, |
1134 | .ln_type = glops->go_type, | |
1135 | .ln_sbd = sdp }; | |
0a52aba7 | 1136 | struct gfs2_glock *gl, *tmp; |
009d8518 | 1137 | struct address_space *mapping; |
0a52aba7 | 1138 | int ret = 0; |
b3b94faa | 1139 | |
0515480a AG |
1140 | gl = find_insert_glock(&name, NULL); |
1141 | if (gl) { | |
1142 | *glp = gl; | |
b3b94faa | 1143 | return 0; |
0515480a | 1144 | } |
64d576ba SW |
1145 | if (!create) |
1146 | return -ENOENT; | |
b3b94faa | 1147 | |
11d8b79e KC |
1148 | if (glops->go_flags & GLOF_ASPACE) { |
1149 | struct gfs2_glock_aspace *gla = | |
1150 | kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS); | |
1151 | if (!gla) | |
1152 | return -ENOMEM; | |
1153 | gl = &gla->glock; | |
1154 | } else { | |
1155 | gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS); | |
1156 | if (!gl) | |
1157 | return -ENOMEM; | |
1158 | } | |
dba2d70c | 1159 | memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb)); |
11d8b79e | 1160 | gl->gl_ops = glops; |
dba2d70c DT |
1161 | |
1162 | if (glops->go_flags & GLOF_LVB) { | |
f7be987b | 1163 | gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS); |
4e2f8849 | 1164 | if (!gl->gl_lksb.sb_lvbptr) { |
11d8b79e | 1165 | gfs2_glock_dealloc(&gl->gl_rcu); |
dba2d70c DT |
1166 | return -ENOMEM; |
1167 | } | |
dba2d70c DT |
1168 | } |
1169 | ||
8f05228e | 1170 | atomic_inc(&sdp->sd_glock_disposal); |
88ffbf3e | 1171 | gl->gl_node.next = NULL; |
f2e70d8f | 1172 | gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0; |
b3b94faa | 1173 | gl->gl_name = name; |
515b269d | 1174 | lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass); |
e66cf161 | 1175 | gl->gl_lockref.count = 1; |
b3b94faa | 1176 | gl->gl_state = LM_ST_UNLOCKED; |
6802e340 | 1177 | gl->gl_target = LM_ST_UNLOCKED; |
c4f68a13 | 1178 | gl->gl_demote_state = LM_ST_EXCLUSIVE; |
8b0e1953 | 1179 | gl->gl_dstamp = 0; |
a245769f SW |
1180 | preempt_disable(); |
1181 | /* We use the global stats to estimate the initial per-glock stats */ | |
1182 | gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type]; | |
1183 | preempt_enable(); | |
1184 | gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0; | |
1185 | gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0; | |
c4f68a13 | 1186 | gl->gl_tchange = jiffies; |
ec45d9f5 | 1187 | gl->gl_object = NULL; |
7cf8dcd3 | 1188 | gl->gl_hold_time = GL_GLOCK_DFT_HOLD; |
c4f68a13 | 1189 | INIT_DELAYED_WORK(&gl->gl_work, glock_work_func); |
2ffed529 BP |
1190 | if (gl->gl_name.ln_type == LM_TYPE_IOPEN) |
1191 | INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func); | |
b3b94faa | 1192 | |
009d8518 SW |
1193 | mapping = gfs2_glock2aspace(gl); |
1194 | if (mapping) { | |
1195 | mapping->a_ops = &gfs2_meta_aops; | |
1196 | mapping->host = s->s_bdev->bd_inode; | |
1197 | mapping->flags = 0; | |
1198 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
252aa6f5 | 1199 | mapping->private_data = NULL; |
009d8518 | 1200 | mapping->writeback_index = 0; |
b3b94faa DT |
1201 | } |
1202 | ||
0515480a | 1203 | tmp = find_insert_glock(&name, gl); |
0a52aba7 | 1204 | if (!tmp) { |
88ffbf3e | 1205 | *glp = gl; |
0a52aba7 | 1206 | goto out; |
b3b94faa | 1207 | } |
0a52aba7 AG |
1208 | if (IS_ERR(tmp)) { |
1209 | ret = PTR_ERR(tmp); | |
1210 | goto out_free; | |
1211 | } | |
0515480a | 1212 | *glp = tmp; |
0a52aba7 AG |
1213 | |
1214 | out_free: | |
11d8b79e | 1215 | gfs2_glock_dealloc(&gl->gl_rcu); |
da7d554f AA |
1216 | if (atomic_dec_and_test(&sdp->sd_glock_disposal)) |
1217 | wake_up(&sdp->sd_glock_wait); | |
b3b94faa | 1218 | |
0a52aba7 | 1219 | out: |
88ffbf3e | 1220 | return ret; |
b3b94faa DT |
1221 | } |
1222 | ||
1223 | /** | |
ffd0cd3c | 1224 | * __gfs2_holder_init - initialize a struct gfs2_holder in the default way |
b3b94faa DT |
1225 | * @gl: the glock |
1226 | * @state: the state we're requesting | |
1227 | * @flags: the modifier flags | |
1228 | * @gh: the holder structure | |
1229 | * | |
1230 | */ | |
1231 | ||
b016d9a8 AG |
1232 | void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags, |
1233 | struct gfs2_holder *gh, unsigned long ip) | |
b3b94faa DT |
1234 | { |
1235 | INIT_LIST_HEAD(&gh->gh_list); | |
1236 | gh->gh_gl = gl; | |
b016d9a8 | 1237 | gh->gh_ip = ip; |
b1e058da | 1238 | gh->gh_owner_pid = get_pid(task_pid(current)); |
b3b94faa DT |
1239 | gh->gh_state = state; |
1240 | gh->gh_flags = flags; | |
b3b94faa | 1241 | gh->gh_iflags = 0; |
b3b94faa DT |
1242 | gfs2_glock_hold(gl); |
1243 | } | |
1244 | ||
1245 | /** | |
1246 | * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it | |
1247 | * @state: the state we're requesting | |
1248 | * @flags: the modifier flags | |
1249 | * @gh: the holder structure | |
1250 | * | |
1251 | * Don't mess with the glock. | |
1252 | * | |
1253 | */ | |
1254 | ||
b58bf407 | 1255 | void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh) |
b3b94faa DT |
1256 | { |
1257 | gh->gh_state = state; | |
579b78a4 | 1258 | gh->gh_flags = flags; |
3b8249f6 | 1259 | gh->gh_iflags = 0; |
d29c0afe | 1260 | gh->gh_ip = _RET_IP_; |
30badc95 | 1261 | put_pid(gh->gh_owner_pid); |
1a0eae88 | 1262 | gh->gh_owner_pid = get_pid(task_pid(current)); |
b3b94faa DT |
1263 | } |
1264 | ||
1265 | /** | |
1266 | * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference) | |
1267 | * @gh: the holder structure | |
1268 | * | |
1269 | */ | |
1270 | ||
1271 | void gfs2_holder_uninit(struct gfs2_holder *gh) | |
1272 | { | |
b1e058da | 1273 | put_pid(gh->gh_owner_pid); |
b3b94faa | 1274 | gfs2_glock_put(gh->gh_gl); |
6df9f9a2 | 1275 | gfs2_holder_mark_uninitialized(gh); |
d0dc80db | 1276 | gh->gh_ip = 0; |
b3b94faa DT |
1277 | } |
1278 | ||
01123cf1 AG |
1279 | static void gfs2_glock_update_hold_time(struct gfs2_glock *gl, |
1280 | unsigned long start_time) | |
1281 | { | |
1282 | /* Have we waited longer that a second? */ | |
1283 | if (time_after(jiffies, start_time + HZ)) { | |
1284 | /* Lengthen the minimum hold time. */ | |
1285 | gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR, | |
1286 | GL_GLOCK_MAX_HOLD); | |
1287 | } | |
1288 | } | |
1289 | ||
53d69132 AG |
1290 | /** |
1291 | * gfs2_glock_holder_ready - holder is ready and its error code can be collected | |
1292 | * @gh: the glock holder | |
1293 | * | |
1294 | * Called when a glock holder no longer needs to be waited for because it is | |
1295 | * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has | |
1296 | * failed (gh_error != 0). | |
1297 | */ | |
1298 | ||
1299 | int gfs2_glock_holder_ready(struct gfs2_holder *gh) | |
1300 | { | |
1301 | if (gh->gh_error || (gh->gh_flags & GL_SKIP)) | |
1302 | return gh->gh_error; | |
1303 | gh->gh_error = gfs2_instantiate(gh); | |
1304 | if (gh->gh_error) | |
1305 | gfs2_glock_dq(gh); | |
1306 | return gh->gh_error; | |
1307 | } | |
1308 | ||
07a79049 BP |
1309 | /** |
1310 | * gfs2_glock_wait - wait on a glock acquisition | |
1311 | * @gh: the glock holder | |
1312 | * | |
1313 | * Returns: 0 on success | |
1314 | */ | |
1315 | ||
1316 | int gfs2_glock_wait(struct gfs2_holder *gh) | |
da755fdb | 1317 | { |
01123cf1 | 1318 | unsigned long start_time = jiffies; |
7cf8dcd3 | 1319 | |
6802e340 | 1320 | might_sleep(); |
74316201 | 1321 | wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE); |
01123cf1 | 1322 | gfs2_glock_update_hold_time(gh->gh_gl, start_time); |
53d69132 | 1323 | return gfs2_glock_holder_ready(gh); |
da755fdb SW |
1324 | } |
1325 | ||
ad26967b BP |
1326 | static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs) |
1327 | { | |
1328 | int i; | |
1329 | ||
1330 | for (i = 0; i < num_gh; i++) | |
1331 | if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) | |
1332 | return 1; | |
1333 | return 0; | |
1334 | } | |
1335 | ||
1336 | /** | |
1337 | * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions | |
1338 | * @num_gh: the number of holders in the array | |
1339 | * @ghs: the glock holder array | |
1340 | * | |
1341 | * Returns: 0 on success, meaning all glocks have been granted and are held. | |
1342 | * -ESTALE if the request timed out, meaning all glocks were released, | |
1343 | * and the caller should retry the operation. | |
1344 | */ | |
1345 | ||
1346 | int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs) | |
1347 | { | |
1348 | struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd; | |
1349 | int i, ret = 0, timeout = 0; | |
1350 | unsigned long start_time = jiffies; | |
ad26967b BP |
1351 | |
1352 | might_sleep(); | |
1353 | /* | |
1354 | * Total up the (minimum hold time * 2) of all glocks and use that to | |
1355 | * determine the max amount of time we should wait. | |
1356 | */ | |
1357 | for (i = 0; i < num_gh; i++) | |
1358 | timeout += ghs[i].gh_gl->gl_hold_time << 1; | |
1359 | ||
ad26967b | 1360 | if (!wait_event_timeout(sdp->sd_async_glock_wait, |
bdff777c | 1361 | !glocks_pending(num_gh, ghs), timeout)) { |
ad26967b | 1362 | ret = -ESTALE; /* request timed out. */ |
bdff777c AG |
1363 | goto out; |
1364 | } | |
ad26967b | 1365 | |
ad26967b | 1366 | for (i = 0; i < num_gh; i++) { |
bdff777c | 1367 | struct gfs2_holder *gh = &ghs[i]; |
53d69132 | 1368 | int ret2; |
ad26967b | 1369 | |
bdff777c AG |
1370 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) { |
1371 | gfs2_glock_update_hold_time(gh->gh_gl, | |
1372 | start_time); | |
ad26967b | 1373 | } |
53d69132 | 1374 | ret2 = gfs2_glock_holder_ready(gh); |
ad26967b | 1375 | if (!ret) |
53d69132 | 1376 | ret = ret2; |
ad26967b BP |
1377 | } |
1378 | ||
bdff777c AG |
1379 | out: |
1380 | if (ret) { | |
1381 | for (i = 0; i < num_gh; i++) { | |
1382 | struct gfs2_holder *gh = &ghs[i]; | |
ad26967b | 1383 | |
bdff777c AG |
1384 | gfs2_glock_dq(gh); |
1385 | } | |
1386 | } | |
ad26967b BP |
1387 | return ret; |
1388 | } | |
1389 | ||
b3b94faa | 1390 | /** |
6802e340 SW |
1391 | * handle_callback - process a demote request |
1392 | * @gl: the glock | |
1393 | * @state: the state the caller wants us to change to | |
c551f66c LJ |
1394 | * @delay: zero to demote immediately; otherwise pending demote |
1395 | * @remote: true if this came from a different cluster node | |
b3b94faa | 1396 | * |
6802e340 SW |
1397 | * There are only two requests that we are going to see in actual |
1398 | * practise: LM_ST_SHARED and LM_ST_UNLOCKED | |
b3b94faa DT |
1399 | */ |
1400 | ||
6802e340 | 1401 | static void handle_callback(struct gfs2_glock *gl, unsigned int state, |
81ffbf65 | 1402 | unsigned long delay, bool remote) |
b3b94faa | 1403 | { |
35b6f8fb AG |
1404 | if (delay) |
1405 | set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags); | |
1406 | else | |
1407 | gfs2_set_demote(gl); | |
6802e340 SW |
1408 | if (gl->gl_demote_state == LM_ST_EXCLUSIVE) { |
1409 | gl->gl_demote_state = state; | |
1410 | gl->gl_demote_time = jiffies; | |
6802e340 SW |
1411 | } else if (gl->gl_demote_state != LM_ST_UNLOCKED && |
1412 | gl->gl_demote_state != state) { | |
1413 | gl->gl_demote_state = LM_ST_UNLOCKED; | |
b3b94faa | 1414 | } |
b94a170e | 1415 | if (gl->gl_ops->go_callback) |
81ffbf65 | 1416 | gl->gl_ops->go_callback(gl, remote); |
7bd8b2eb | 1417 | trace_gfs2_demote_rq(gl, remote); |
b3b94faa DT |
1418 | } |
1419 | ||
6802e340 | 1420 | void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...) |
7c52b166 | 1421 | { |
5e69069c | 1422 | struct va_format vaf; |
7c52b166 RP |
1423 | va_list args; |
1424 | ||
1425 | va_start(args, fmt); | |
5e69069c | 1426 | |
6802e340 | 1427 | if (seq) { |
1bb49303 | 1428 | seq_vprintf(seq, fmt, args); |
6802e340 | 1429 | } else { |
5e69069c JP |
1430 | vaf.fmt = fmt; |
1431 | vaf.va = &args; | |
1432 | ||
d77d1b58 | 1433 | pr_err("%pV", &vaf); |
6802e340 | 1434 | } |
5e69069c | 1435 | |
7c52b166 RP |
1436 | va_end(args); |
1437 | } | |
1438 | ||
b3b94faa DT |
1439 | /** |
1440 | * add_to_queue - Add a holder to the wait queue (but look for recursion) | |
1441 | * @gh: the holder structure to add | |
1442 | * | |
6802e340 SW |
1443 | * Eventually we should move the recursive locking trap to a |
1444 | * debugging option or something like that. This is the fast | |
1445 | * path and needs to have the minimum number of distractions. | |
1446 | * | |
b3b94faa DT |
1447 | */ |
1448 | ||
6802e340 | 1449 | static inline void add_to_queue(struct gfs2_holder *gh) |
f3dd1649 AG |
1450 | __releases(&gl->gl_lockref.lock) |
1451 | __acquires(&gl->gl_lockref.lock) | |
b3b94faa DT |
1452 | { |
1453 | struct gfs2_glock *gl = gh->gh_gl; | |
15562c43 | 1454 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
6802e340 SW |
1455 | struct list_head *insert_pt = NULL; |
1456 | struct gfs2_holder *gh2; | |
e5dc76b9 | 1457 | int try_futile = 0; |
b3b94faa | 1458 | |
ad26967b | 1459 | GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL); |
fee852e3 | 1460 | if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags)) |
ad26967b | 1461 | GLOCK_BUG_ON(gl, true); |
190562bd | 1462 | |
6802e340 | 1463 | if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) { |
61444649 | 1464 | if (test_bit(GLF_LOCK, &gl->gl_flags)) { |
8f0028fc | 1465 | struct gfs2_holder *current_gh; |
61444649 | 1466 | |
8f0028fc AG |
1467 | current_gh = find_first_strong_holder(gl); |
1468 | try_futile = !may_grant(gl, current_gh, gh); | |
61444649 | 1469 | } |
6802e340 SW |
1470 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) |
1471 | goto fail; | |
1472 | } | |
1473 | ||
1474 | list_for_each_entry(gh2, &gl->gl_holders, gh_list) { | |
1475 | if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid && | |
dc732906 BP |
1476 | (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) && |
1477 | !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags))) | |
6802e340 | 1478 | goto trap_recursive; |
e5dc76b9 BP |
1479 | if (try_futile && |
1480 | !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) { | |
6802e340 SW |
1481 | fail: |
1482 | gh->gh_error = GLR_TRYFAILED; | |
1483 | gfs2_holder_wake(gh); | |
1484 | return; | |
b4c20166 | 1485 | } |
6802e340 SW |
1486 | if (test_bit(HIF_HOLDER, &gh2->gh_iflags)) |
1487 | continue; | |
1488 | if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt)) | |
1489 | insert_pt = &gh2->gh_list; | |
1490 | } | |
edae38a6 | 1491 | trace_gfs2_glock_queue(gh, 1); |
a245769f SW |
1492 | gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT); |
1493 | gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT); | |
6802e340 SW |
1494 | if (likely(insert_pt == NULL)) { |
1495 | list_add_tail(&gh->gh_list, &gl->gl_holders); | |
1496 | if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY)) | |
1497 | goto do_cancel; | |
1498 | return; | |
1499 | } | |
1500 | list_add_tail(&gh->gh_list, insert_pt); | |
1501 | do_cancel: | |
969183bc | 1502 | gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list); |
6802e340 | 1503 | if (!(gh->gh_flags & LM_FLAG_PRIORITY)) { |
f3dd1649 | 1504 | spin_unlock(&gl->gl_lockref.lock); |
048bca22 | 1505 | if (sdp->sd_lockstruct.ls_ops->lm_cancel) |
f057f6cd | 1506 | sdp->sd_lockstruct.ls_ops->lm_cancel(gl); |
f3dd1649 | 1507 | spin_lock(&gl->gl_lockref.lock); |
b3b94faa | 1508 | } |
6802e340 | 1509 | return; |
b3b94faa | 1510 | |
6802e340 | 1511 | trap_recursive: |
e54c78a2 BP |
1512 | fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip); |
1513 | fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid)); | |
1514 | fs_err(sdp, "lock type: %d req lock state : %d\n", | |
6802e340 | 1515 | gh2->gh_gl->gl_name.ln_type, gh2->gh_state); |
e54c78a2 BP |
1516 | fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip); |
1517 | fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid)); | |
1518 | fs_err(sdp, "lock type: %d req lock state : %d\n", | |
6802e340 | 1519 | gh->gh_gl->gl_name.ln_type, gh->gh_state); |
3792ce97 | 1520 | gfs2_dump_glock(NULL, gl, true); |
6802e340 | 1521 | BUG(); |
b3b94faa DT |
1522 | } |
1523 | ||
1524 | /** | |
1525 | * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock) | |
1526 | * @gh: the holder structure | |
1527 | * | |
1528 | * if (gh->gh_flags & GL_ASYNC), this never returns an error | |
1529 | * | |
1530 | * Returns: 0, GLR_TRYFAILED, or errno on failure | |
1531 | */ | |
1532 | ||
1533 | int gfs2_glock_nq(struct gfs2_holder *gh) | |
1534 | { | |
1535 | struct gfs2_glock *gl = gh->gh_gl; | |
b3b94faa DT |
1536 | int error = 0; |
1537 | ||
601ef0d5 | 1538 | if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP)) |
b3b94faa | 1539 | return -EIO; |
b3b94faa | 1540 | |
f42ab085 SW |
1541 | if (test_bit(GLF_LRU, &gl->gl_flags)) |
1542 | gfs2_glock_remove_from_lru(gl); | |
1543 | ||
a4e8145e | 1544 | gh->gh_error = 0; |
f3dd1649 | 1545 | spin_lock(&gl->gl_lockref.lock); |
b3b94faa | 1546 | add_to_queue(gh); |
01b172b7 BP |
1547 | if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) && |
1548 | test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) { | |
0809f6ec | 1549 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
01b172b7 | 1550 | gl->gl_lockref.count++; |
6b0c7440 | 1551 | __gfs2_glock_queue_work(gl, 0); |
01b172b7 | 1552 | } |
6802e340 | 1553 | run_queue(gl, 1); |
f3dd1649 | 1554 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa | 1555 | |
6802e340 SW |
1556 | if (!(gh->gh_flags & GL_ASYNC)) |
1557 | error = gfs2_glock_wait(gh); | |
b3b94faa | 1558 | |
b3b94faa DT |
1559 | return error; |
1560 | } | |
1561 | ||
1562 | /** | |
1563 | * gfs2_glock_poll - poll to see if an async request has been completed | |
1564 | * @gh: the holder | |
1565 | * | |
1566 | * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on | |
1567 | */ | |
1568 | ||
1569 | int gfs2_glock_poll(struct gfs2_holder *gh) | |
1570 | { | |
6802e340 | 1571 | return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1; |
b3b94faa DT |
1572 | } |
1573 | ||
dc732906 BP |
1574 | static inline bool needs_demote(struct gfs2_glock *gl) |
1575 | { | |
1576 | return (test_bit(GLF_DEMOTE, &gl->gl_flags) || | |
1577 | test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags)); | |
1578 | } | |
b3b94faa | 1579 | |
dc732906 | 1580 | static void __gfs2_glock_dq(struct gfs2_holder *gh) |
b3b94faa DT |
1581 | { |
1582 | struct gfs2_glock *gl = gh->gh_gl; | |
601ef0d5 | 1583 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
c4f68a13 | 1584 | unsigned delay = 0; |
6802e340 | 1585 | int fast_path = 0; |
b3b94faa | 1586 | |
601ef0d5 | 1587 | /* |
dc732906 BP |
1588 | * This while loop is similar to function demote_incompat_holders: |
1589 | * If the glock is due to be demoted (which may be from another node | |
1590 | * or even if this holder is GL_NOCACHE), the weak holders are | |
1591 | * demoted as well, allowing the glock to be demoted. | |
601ef0d5 | 1592 | */ |
dc732906 BP |
1593 | while (gh) { |
1594 | /* | |
1595 | * If we're in the process of file system withdraw, we cannot | |
1596 | * just dequeue any glocks until our journal is recovered, lest | |
1597 | * we introduce file system corruption. We need two exceptions | |
1598 | * to this rule: We need to allow unlocking of nondisk glocks | |
1599 | * and the glock for our own journal that needs recovery. | |
1600 | */ | |
1601 | if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) && | |
1602 | glock_blocked_by_withdraw(gl) && | |
1603 | gh->gh_gl != sdp->sd_jinode_gl) { | |
1604 | sdp->sd_glock_dqs_held++; | |
1605 | spin_unlock(&gl->gl_lockref.lock); | |
1606 | might_sleep(); | |
1607 | wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY, | |
1608 | TASK_UNINTERRUPTIBLE); | |
1609 | spin_lock(&gl->gl_lockref.lock); | |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * This holder should not be cached, so mark it for demote. | |
1614 | * Note: this should be done before the check for needs_demote | |
1615 | * below. | |
1616 | */ | |
1617 | if (gh->gh_flags & GL_NOCACHE) | |
1618 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); | |
1619 | ||
1620 | list_del_init(&gh->gh_list); | |
1621 | clear_bit(HIF_HOLDER, &gh->gh_iflags); | |
1622 | trace_gfs2_glock_queue(gh, 0); | |
1623 | ||
1624 | /* | |
1625 | * If there hasn't been a demote request we are done. | |
1626 | * (Let the remaining holders, if any, keep holding it.) | |
1627 | */ | |
1628 | if (!needs_demote(gl)) { | |
1629 | if (list_empty(&gl->gl_holders)) | |
1630 | fast_path = 1; | |
1631 | break; | |
1632 | } | |
1633 | /* | |
1634 | * If we have another strong holder (we cannot auto-demote) | |
1635 | * we are done. It keeps holding it until it is done. | |
1636 | */ | |
1637 | if (find_first_strong_holder(gl)) | |
1638 | break; | |
b3b94faa | 1639 | |
dc732906 BP |
1640 | /* |
1641 | * If we have a weak holder at the head of the list, it | |
1642 | * (and all others like it) must be auto-demoted. If there | |
1643 | * are no more weak holders, we exit the while loop. | |
1644 | */ | |
1645 | gh = find_first_holder(gl); | |
1646 | } | |
08d73666 | 1647 | |
7881ef3f | 1648 | if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl)) |
4abb6ad9 BP |
1649 | gfs2_glock_add_to_lru(gl); |
1650 | ||
6b0c7440 AG |
1651 | if (unlikely(!fast_path)) { |
1652 | gl->gl_lockref.count++; | |
1653 | if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) && | |
1654 | !test_bit(GLF_DEMOTE, &gl->gl_flags) && | |
1655 | gl->gl_name.ln_type == LM_TYPE_INODE) | |
1656 | delay = gl->gl_hold_time; | |
1657 | __gfs2_glock_queue_work(gl, delay); | |
1658 | } | |
dc732906 BP |
1659 | } |
1660 | ||
1661 | /** | |
1662 | * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock) | |
1663 | * @gh: the glock holder | |
1664 | * | |
1665 | */ | |
1666 | void gfs2_glock_dq(struct gfs2_holder *gh) | |
1667 | { | |
1668 | struct gfs2_glock *gl = gh->gh_gl; | |
1669 | ||
1670 | spin_lock(&gl->gl_lockref.lock); | |
1fc05c8d AG |
1671 | if (list_is_first(&gh->gh_list, &gl->gl_holders) && |
1672 | !test_bit(HIF_HOLDER, &gh->gh_iflags)) { | |
1673 | spin_unlock(&gl->gl_lockref.lock); | |
1674 | gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl); | |
1675 | wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE); | |
1676 | spin_lock(&gl->gl_lockref.lock); | |
1677 | } | |
1678 | ||
dc732906 | 1679 | __gfs2_glock_dq(gh); |
f3dd1649 | 1680 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1681 | } |
1682 | ||
d93cfa98 AD |
1683 | void gfs2_glock_dq_wait(struct gfs2_holder *gh) |
1684 | { | |
1685 | struct gfs2_glock *gl = gh->gh_gl; | |
1686 | gfs2_glock_dq(gh); | |
81e1d450 | 1687 | might_sleep(); |
74316201 | 1688 | wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE); |
d93cfa98 AD |
1689 | } |
1690 | ||
b3b94faa DT |
1691 | /** |
1692 | * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it | |
1693 | * @gh: the holder structure | |
1694 | * | |
1695 | */ | |
1696 | ||
1697 | void gfs2_glock_dq_uninit(struct gfs2_holder *gh) | |
1698 | { | |
1699 | gfs2_glock_dq(gh); | |
1700 | gfs2_holder_uninit(gh); | |
1701 | } | |
1702 | ||
1703 | /** | |
1704 | * gfs2_glock_nq_num - acquire a glock based on lock number | |
1705 | * @sdp: the filesystem | |
1706 | * @number: the lock number | |
1707 | * @glops: the glock operations for the type of glock | |
1708 | * @state: the state to acquire the glock in | |
25985edc | 1709 | * @flags: modifier flags for the acquisition |
b3b94faa DT |
1710 | * @gh: the struct gfs2_holder |
1711 | * | |
1712 | * Returns: errno | |
1713 | */ | |
1714 | ||
cd915493 | 1715 | int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number, |
8fb4b536 | 1716 | const struct gfs2_glock_operations *glops, |
b58bf407 | 1717 | unsigned int state, u16 flags, struct gfs2_holder *gh) |
b3b94faa DT |
1718 | { |
1719 | struct gfs2_glock *gl; | |
1720 | int error; | |
1721 | ||
1722 | error = gfs2_glock_get(sdp, number, glops, CREATE, &gl); | |
1723 | if (!error) { | |
1724 | error = gfs2_glock_nq_init(gl, state, flags, gh); | |
1725 | gfs2_glock_put(gl); | |
1726 | } | |
1727 | ||
1728 | return error; | |
1729 | } | |
1730 | ||
1731 | /** | |
1732 | * glock_compare - Compare two struct gfs2_glock structures for sorting | |
1733 | * @arg_a: the first structure | |
1734 | * @arg_b: the second structure | |
1735 | * | |
1736 | */ | |
1737 | ||
1738 | static int glock_compare(const void *arg_a, const void *arg_b) | |
1739 | { | |
a5e08a9e SW |
1740 | const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a; |
1741 | const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b; | |
1742 | const struct lm_lockname *a = &gh_a->gh_gl->gl_name; | |
1743 | const struct lm_lockname *b = &gh_b->gh_gl->gl_name; | |
b3b94faa DT |
1744 | |
1745 | if (a->ln_number > b->ln_number) | |
a5e08a9e SW |
1746 | return 1; |
1747 | if (a->ln_number < b->ln_number) | |
1748 | return -1; | |
1c0f4872 | 1749 | BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type); |
a5e08a9e | 1750 | return 0; |
b3b94faa DT |
1751 | } |
1752 | ||
1753 | /** | |
44dab005 | 1754 | * nq_m_sync - synchronously acquire more than one glock in deadlock free order |
b3b94faa DT |
1755 | * @num_gh: the number of structures |
1756 | * @ghs: an array of struct gfs2_holder structures | |
c551f66c | 1757 | * @p: placeholder for the holder structure to pass back |
b3b94faa DT |
1758 | * |
1759 | * Returns: 0 on success (all glocks acquired), | |
1760 | * errno on failure (no glocks acquired) | |
1761 | */ | |
1762 | ||
1763 | static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs, | |
1764 | struct gfs2_holder **p) | |
1765 | { | |
1766 | unsigned int x; | |
1767 | int error = 0; | |
1768 | ||
1769 | for (x = 0; x < num_gh; x++) | |
1770 | p[x] = &ghs[x]; | |
1771 | ||
1772 | sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL); | |
1773 | ||
1774 | for (x = 0; x < num_gh; x++) { | |
b3b94faa DT |
1775 | error = gfs2_glock_nq(p[x]); |
1776 | if (error) { | |
1777 | while (x--) | |
1778 | gfs2_glock_dq(p[x]); | |
1779 | break; | |
1780 | } | |
1781 | } | |
1782 | ||
1783 | return error; | |
1784 | } | |
1785 | ||
1786 | /** | |
1787 | * gfs2_glock_nq_m - acquire multiple glocks | |
1788 | * @num_gh: the number of structures | |
1789 | * @ghs: an array of struct gfs2_holder structures | |
1790 | * | |
b3b94faa DT |
1791 | * Returns: 0 on success (all glocks acquired), |
1792 | * errno on failure (no glocks acquired) | |
1793 | */ | |
1794 | ||
1795 | int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs) | |
1796 | { | |
eaf5bd3c SW |
1797 | struct gfs2_holder *tmp[4]; |
1798 | struct gfs2_holder **pph = tmp; | |
b3b94faa DT |
1799 | int error = 0; |
1800 | ||
eaf5bd3c SW |
1801 | switch(num_gh) { |
1802 | case 0: | |
b3b94faa | 1803 | return 0; |
eaf5bd3c | 1804 | case 1: |
b3b94faa | 1805 | return gfs2_glock_nq(ghs); |
eaf5bd3c SW |
1806 | default: |
1807 | if (num_gh <= 4) | |
b3b94faa | 1808 | break; |
6da2ec56 KC |
1809 | pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *), |
1810 | GFP_NOFS); | |
eaf5bd3c SW |
1811 | if (!pph) |
1812 | return -ENOMEM; | |
b3b94faa DT |
1813 | } |
1814 | ||
eaf5bd3c | 1815 | error = nq_m_sync(num_gh, ghs, pph); |
b3b94faa | 1816 | |
eaf5bd3c SW |
1817 | if (pph != tmp) |
1818 | kfree(pph); | |
b3b94faa DT |
1819 | |
1820 | return error; | |
1821 | } | |
1822 | ||
1823 | /** | |
1824 | * gfs2_glock_dq_m - release multiple glocks | |
1825 | * @num_gh: the number of structures | |
1826 | * @ghs: an array of struct gfs2_holder structures | |
1827 | * | |
1828 | */ | |
1829 | ||
1830 | void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs) | |
1831 | { | |
fa1bbdea BP |
1832 | while (num_gh--) |
1833 | gfs2_glock_dq(&ghs[num_gh]); | |
b3b94faa DT |
1834 | } |
1835 | ||
f057f6cd | 1836 | void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state) |
da755fdb | 1837 | { |
c4f68a13 BM |
1838 | unsigned long delay = 0; |
1839 | unsigned long holdtime; | |
1840 | unsigned long now = jiffies; | |
b3b94faa | 1841 | |
f057f6cd | 1842 | gfs2_glock_hold(gl); |
e2c6c8a7 | 1843 | spin_lock(&gl->gl_lockref.lock); |
7cf8dcd3 | 1844 | holdtime = gl->gl_tchange + gl->gl_hold_time; |
e2c6c8a7 | 1845 | if (!list_empty(&gl->gl_holders) && |
7cf8dcd3 | 1846 | gl->gl_name.ln_type == LM_TYPE_INODE) { |
7b5e3d5f SW |
1847 | if (time_before(now, holdtime)) |
1848 | delay = holdtime - now; | |
1849 | if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags)) | |
7cf8dcd3 | 1850 | delay = gl->gl_hold_time; |
7b5e3d5f | 1851 | } |
dc732906 BP |
1852 | /* |
1853 | * Note 1: We cannot call demote_incompat_holders from handle_callback | |
1854 | * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq -> | |
1855 | * handle_callback -> demote_incompat_holders -> gfs2_glock_dq | |
1856 | * Plus, we only want to demote the holders if the request comes from | |
1857 | * a remote cluster node because local holder conflicts are resolved | |
1858 | * elsewhere. | |
1859 | * | |
1860 | * Note 2: if a remote node wants this glock in EX mode, lock_dlm will | |
1861 | * request that we set our state to UNLOCKED. Here we mock up a holder | |
1862 | * to make it look like someone wants the lock EX locally. Any SH | |
1863 | * and DF requests should be able to share the lock without demoting. | |
1864 | * | |
1865 | * Note 3: We only want to demote the demoteable holders when there | |
1866 | * are no more strong holders. The demoteable holders might as well | |
1867 | * keep the glock until the last strong holder is done with it. | |
1868 | */ | |
1869 | if (!find_first_strong_holder(gl)) { | |
e11b02df AG |
1870 | struct gfs2_holder mock_gh = { |
1871 | .gh_gl = gl, | |
1872 | .gh_state = (state == LM_ST_UNLOCKED) ? | |
1873 | LM_ST_EXCLUSIVE : state, | |
1874 | .gh_iflags = BIT(HIF_HOLDER) | |
1875 | }; | |
1876 | ||
dc732906 BP |
1877 | demote_incompat_holders(gl, &mock_gh); |
1878 | } | |
81ffbf65 | 1879 | handle_callback(gl, state, delay, true); |
6b0c7440 | 1880 | __gfs2_glock_queue_work(gl, delay); |
f3dd1649 | 1881 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1882 | } |
1883 | ||
0809f6ec SW |
1884 | /** |
1885 | * gfs2_should_freeze - Figure out if glock should be frozen | |
1886 | * @gl: The glock in question | |
1887 | * | |
1888 | * Glocks are not frozen if (a) the result of the dlm operation is | |
1889 | * an error, (b) the locking operation was an unlock operation or | |
1890 | * (c) if there is a "noexp" flagged request anywhere in the queue | |
1891 | * | |
1892 | * Returns: 1 if freezing should occur, 0 otherwise | |
1893 | */ | |
1894 | ||
1895 | static int gfs2_should_freeze(const struct gfs2_glock *gl) | |
1896 | { | |
1897 | const struct gfs2_holder *gh; | |
1898 | ||
1899 | if (gl->gl_reply & ~LM_OUT_ST_MASK) | |
1900 | return 0; | |
1901 | if (gl->gl_target == LM_ST_UNLOCKED) | |
1902 | return 0; | |
1903 | ||
1904 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { | |
1905 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
1906 | continue; | |
1907 | if (LM_FLAG_NOEXP & gh->gh_flags) | |
1908 | return 0; | |
1909 | } | |
1910 | ||
1911 | return 1; | |
1912 | } | |
1913 | ||
b3b94faa | 1914 | /** |
f057f6cd SW |
1915 | * gfs2_glock_complete - Callback used by locking |
1916 | * @gl: Pointer to the glock | |
1917 | * @ret: The return value from the dlm | |
b3b94faa | 1918 | * |
f3dd1649 | 1919 | * The gl_reply field is under the gl_lockref.lock lock so that it is ok |
47a25380 | 1920 | * to use a bitfield shared with other glock state fields. |
b3b94faa DT |
1921 | */ |
1922 | ||
f057f6cd | 1923 | void gfs2_glock_complete(struct gfs2_glock *gl, int ret) |
b3b94faa | 1924 | { |
15562c43 | 1925 | struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct; |
0809f6ec | 1926 | |
f3dd1649 | 1927 | spin_lock(&gl->gl_lockref.lock); |
f057f6cd | 1928 | gl->gl_reply = ret; |
0809f6ec | 1929 | |
e0c2a9aa | 1930 | if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) { |
0809f6ec | 1931 | if (gfs2_should_freeze(gl)) { |
f057f6cd | 1932 | set_bit(GLF_FROZEN, &gl->gl_flags); |
f3dd1649 | 1933 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa | 1934 | return; |
0809f6ec | 1935 | } |
b3b94faa | 1936 | } |
47a25380 | 1937 | |
e66cf161 | 1938 | gl->gl_lockref.count++; |
f057f6cd | 1939 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
6b0c7440 | 1940 | __gfs2_glock_queue_work(gl, 0); |
f3dd1649 | 1941 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1942 | } |
1943 | ||
4f0f586b ST |
1944 | static int glock_cmp(void *priv, const struct list_head *a, |
1945 | const struct list_head *b) | |
4506a519 SW |
1946 | { |
1947 | struct gfs2_glock *gla, *glb; | |
1948 | ||
1949 | gla = list_entry(a, struct gfs2_glock, gl_lru); | |
1950 | glb = list_entry(b, struct gfs2_glock, gl_lru); | |
1951 | ||
1952 | if (gla->gl_name.ln_number > glb->gl_name.ln_number) | |
1953 | return 1; | |
1954 | if (gla->gl_name.ln_number < glb->gl_name.ln_number) | |
1955 | return -1; | |
1956 | ||
1957 | return 0; | |
1958 | } | |
1959 | ||
1960 | /** | |
1961 | * gfs2_dispose_glock_lru - Demote a list of glocks | |
1962 | * @list: The list to dispose of | |
1963 | * | |
1964 | * Disposing of glocks may involve disk accesses, so that here we sort | |
1965 | * the glocks by number (i.e. disk location of the inodes) so that if | |
1966 | * there are any such accesses, they'll be sent in order (mostly). | |
1967 | * | |
1968 | * Must be called under the lru_lock, but may drop and retake this | |
1969 | * lock. While the lru_lock is dropped, entries may vanish from the | |
1970 | * list, but no new entries will appear on the list (since it is | |
1971 | * private) | |
1972 | */ | |
1973 | ||
1974 | static void gfs2_dispose_glock_lru(struct list_head *list) | |
1975 | __releases(&lru_lock) | |
1976 | __acquires(&lru_lock) | |
1977 | { | |
1978 | struct gfs2_glock *gl; | |
1979 | ||
1980 | list_sort(NULL, list, glock_cmp); | |
1981 | ||
1982 | while(!list_empty(list)) { | |
969183bc | 1983 | gl = list_first_entry(list, struct gfs2_glock, gl_lru); |
4506a519 | 1984 | list_del_init(&gl->gl_lru); |
1ab19c5d | 1985 | clear_bit(GLF_LRU, &gl->gl_flags); |
f3dd1649 | 1986 | if (!spin_trylock(&gl->gl_lockref.lock)) { |
94a09a39 | 1987 | add_back_to_lru: |
e66cf161 | 1988 | list_add(&gl->gl_lru, &lru_list); |
7881ef3f | 1989 | set_bit(GLF_LRU, &gl->gl_flags); |
e66cf161 SW |
1990 | atomic_inc(&lru_count); |
1991 | continue; | |
1992 | } | |
94a09a39 | 1993 | if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) { |
f3dd1649 | 1994 | spin_unlock(&gl->gl_lockref.lock); |
94a09a39 SW |
1995 | goto add_back_to_lru; |
1996 | } | |
e66cf161 | 1997 | gl->gl_lockref.count++; |
4506a519 | 1998 | if (demote_ok(gl)) |
81ffbf65 | 1999 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); |
4506a519 | 2000 | WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags)); |
6b0c7440 | 2001 | __gfs2_glock_queue_work(gl, 0); |
f3dd1649 | 2002 | spin_unlock(&gl->gl_lockref.lock); |
94a09a39 | 2003 | cond_resched_lock(&lru_lock); |
4506a519 SW |
2004 | } |
2005 | } | |
2006 | ||
2a005855 SW |
2007 | /** |
2008 | * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote | |
2009 | * @nr: The number of entries to scan | |
2010 | * | |
4506a519 SW |
2011 | * This function selects the entries on the LRU which are able to |
2012 | * be demoted, and then kicks off the process by calling | |
2013 | * gfs2_dispose_glock_lru() above. | |
2a005855 | 2014 | */ |
b3b94faa | 2015 | |
1ab6c499 | 2016 | static long gfs2_scan_glock_lru(int nr) |
b3b94faa DT |
2017 | { |
2018 | struct gfs2_glock *gl; | |
97cc1025 | 2019 | LIST_HEAD(skipped); |
4506a519 | 2020 | LIST_HEAD(dispose); |
1ab6c499 | 2021 | long freed = 0; |
b3b94faa | 2022 | |
97cc1025 | 2023 | spin_lock(&lru_lock); |
1ab6c499 | 2024 | while ((nr-- >= 0) && !list_empty(&lru_list)) { |
969183bc | 2025 | gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru); |
97cc1025 SW |
2026 | |
2027 | /* Test for being demotable */ | |
94a09a39 | 2028 | if (!test_bit(GLF_LOCK, &gl->gl_flags)) { |
4506a519 SW |
2029 | list_move(&gl->gl_lru, &dispose); |
2030 | atomic_dec(&lru_count); | |
1ab6c499 | 2031 | freed++; |
2163b1e6 | 2032 | continue; |
97cc1025 | 2033 | } |
4506a519 SW |
2034 | |
2035 | list_move(&gl->gl_lru, &skipped); | |
b3b94faa | 2036 | } |
97cc1025 | 2037 | list_splice(&skipped, &lru_list); |
4506a519 SW |
2038 | if (!list_empty(&dispose)) |
2039 | gfs2_dispose_glock_lru(&dispose); | |
97cc1025 | 2040 | spin_unlock(&lru_lock); |
1ab6c499 DC |
2041 | |
2042 | return freed; | |
2a005855 SW |
2043 | } |
2044 | ||
1ab6c499 DC |
2045 | static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink, |
2046 | struct shrink_control *sc) | |
2a005855 | 2047 | { |
1ab6c499 DC |
2048 | if (!(sc->gfp_mask & __GFP_FS)) |
2049 | return SHRINK_STOP; | |
2050 | return gfs2_scan_glock_lru(sc->nr_to_scan); | |
2051 | } | |
2a005855 | 2052 | |
1ab6c499 DC |
2053 | static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink, |
2054 | struct shrink_control *sc) | |
2055 | { | |
55f841ce | 2056 | return vfs_pressure_ratio(atomic_read(&lru_count)); |
b3b94faa DT |
2057 | } |
2058 | ||
97cc1025 | 2059 | static struct shrinker glock_shrinker = { |
97cc1025 | 2060 | .seeks = DEFAULT_SEEKS, |
1ab6c499 DC |
2061 | .count_objects = gfs2_glock_shrink_count, |
2062 | .scan_objects = gfs2_glock_shrink_scan, | |
97cc1025 SW |
2063 | }; |
2064 | ||
b3b94faa | 2065 | /** |
dbffb29d | 2066 | * glock_hash_walk - Call a function for glock in a hash bucket |
b3b94faa DT |
2067 | * @examiner: the function |
2068 | * @sdp: the filesystem | |
b3b94faa | 2069 | * |
98687f42 HX |
2070 | * Note that the function can be called multiple times on the same |
2071 | * object. So the user must ensure that the function can cope with | |
2072 | * that. | |
b3b94faa DT |
2073 | */ |
2074 | ||
88ffbf3e | 2075 | static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp) |
b3b94faa | 2076 | { |
bc015cb8 | 2077 | struct gfs2_glock *gl; |
98687f42 HX |
2078 | struct rhashtable_iter iter; |
2079 | ||
2080 | rhashtable_walk_enter(&gl_hash_table, &iter); | |
2081 | ||
2082 | do { | |
97a6ec4a | 2083 | rhashtable_walk_start(&iter); |
98687f42 | 2084 | |
7427f3bb AG |
2085 | while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) { |
2086 | if (gl->gl_name.ln_sbd == sdp) | |
88ffbf3e | 2087 | examiner(gl); |
7427f3bb | 2088 | } |
98687f42 HX |
2089 | |
2090 | rhashtable_walk_stop(&iter); | |
2091 | } while (cond_resched(), gl == ERR_PTR(-EAGAIN)); | |
2092 | ||
2093 | rhashtable_walk_exit(&iter); | |
bc015cb8 SW |
2094 | } |
2095 | ||
a0e3cc65 AG |
2096 | bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay) |
2097 | { | |
2098 | bool queued; | |
2099 | ||
2100 | spin_lock(&gl->gl_lockref.lock); | |
2101 | queued = queue_delayed_work(gfs2_delete_workqueue, | |
2102 | &gl->gl_delete, delay); | |
2103 | if (queued) | |
2104 | set_bit(GLF_PENDING_DELETE, &gl->gl_flags); | |
2105 | spin_unlock(&gl->gl_lockref.lock); | |
2106 | return queued; | |
2107 | } | |
2108 | ||
2109 | void gfs2_cancel_delete_work(struct gfs2_glock *gl) | |
2110 | { | |
486408d6 | 2111 | if (cancel_delayed_work(&gl->gl_delete)) { |
a0e3cc65 AG |
2112 | clear_bit(GLF_PENDING_DELETE, &gl->gl_flags); |
2113 | gfs2_glock_put(gl); | |
2114 | } | |
2115 | } | |
2116 | ||
2117 | bool gfs2_delete_work_queued(const struct gfs2_glock *gl) | |
2118 | { | |
2119 | return test_bit(GLF_PENDING_DELETE, &gl->gl_flags); | |
2120 | } | |
2121 | ||
2122 | static void flush_delete_work(struct gfs2_glock *gl) | |
2123 | { | |
2ffed529 BP |
2124 | if (gl->gl_name.ln_type == LM_TYPE_IOPEN) { |
2125 | if (cancel_delayed_work(&gl->gl_delete)) { | |
2126 | queue_delayed_work(gfs2_delete_workqueue, | |
2127 | &gl->gl_delete, 0); | |
2128 | } | |
34244d71 | 2129 | } |
a0e3cc65 AG |
2130 | } |
2131 | ||
2132 | void gfs2_flush_delete_work(struct gfs2_sbd *sdp) | |
2133 | { | |
2134 | glock_hash_walk(flush_delete_work, sdp); | |
2135 | flush_workqueue(gfs2_delete_workqueue); | |
2136 | } | |
2137 | ||
f057f6cd SW |
2138 | /** |
2139 | * thaw_glock - thaw out a glock which has an unprocessed reply waiting | |
2140 | * @gl: The glock to thaw | |
2141 | * | |
f057f6cd SW |
2142 | */ |
2143 | ||
2144 | static void thaw_glock(struct gfs2_glock *gl) | |
2145 | { | |
7427f3bb AG |
2146 | if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) |
2147 | return; | |
2148 | if (!lockref_get_not_dead(&gl->gl_lockref)) | |
6b0c7440 | 2149 | return; |
6b0c7440 AG |
2150 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
2151 | gfs2_glock_queue_work(gl, 0); | |
f057f6cd SW |
2152 | } |
2153 | ||
b3b94faa DT |
2154 | /** |
2155 | * clear_glock - look at a glock and see if we can free it from glock cache | |
2156 | * @gl: the glock to look at | |
2157 | * | |
2158 | */ | |
2159 | ||
2160 | static void clear_glock(struct gfs2_glock *gl) | |
2161 | { | |
f42ab085 | 2162 | gfs2_glock_remove_from_lru(gl); |
b3b94faa | 2163 | |
f3dd1649 | 2164 | spin_lock(&gl->gl_lockref.lock); |
7427f3bb AG |
2165 | if (!__lockref_is_dead(&gl->gl_lockref)) { |
2166 | gl->gl_lockref.count++; | |
2167 | if (gl->gl_state != LM_ST_UNLOCKED) | |
2168 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); | |
2169 | __gfs2_glock_queue_work(gl, 0); | |
2170 | } | |
f3dd1649 | 2171 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
2172 | } |
2173 | ||
f057f6cd SW |
2174 | /** |
2175 | * gfs2_glock_thaw - Thaw any frozen glocks | |
2176 | * @sdp: The super block | |
2177 | * | |
2178 | */ | |
2179 | ||
2180 | void gfs2_glock_thaw(struct gfs2_sbd *sdp) | |
2181 | { | |
bc015cb8 SW |
2182 | glock_hash_walk(thaw_glock, sdp); |
2183 | } | |
f057f6cd | 2184 | |
3792ce97 | 2185 | static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid) |
bc015cb8 | 2186 | { |
f3dd1649 | 2187 | spin_lock(&gl->gl_lockref.lock); |
3792ce97 | 2188 | gfs2_dump_glock(seq, gl, fsid); |
f3dd1649 | 2189 | spin_unlock(&gl->gl_lockref.lock); |
bc015cb8 SW |
2190 | } |
2191 | ||
2192 | static void dump_glock_func(struct gfs2_glock *gl) | |
2193 | { | |
3792ce97 | 2194 | dump_glock(NULL, gl, true); |
f057f6cd SW |
2195 | } |
2196 | ||
b3b94faa DT |
2197 | /** |
2198 | * gfs2_gl_hash_clear - Empty out the glock hash table | |
2199 | * @sdp: the filesystem | |
b3b94faa | 2200 | * |
1bdad606 | 2201 | * Called when unmounting the filesystem. |
b3b94faa DT |
2202 | */ |
2203 | ||
fefc03bf | 2204 | void gfs2_gl_hash_clear(struct gfs2_sbd *sdp) |
b3b94faa | 2205 | { |
fb6791d1 | 2206 | set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags); |
222cb538 | 2207 | flush_workqueue(glock_workqueue); |
bc015cb8 | 2208 | glock_hash_walk(clear_glock, sdp); |
8f05228e | 2209 | flush_workqueue(glock_workqueue); |
2aba1b5b BP |
2210 | wait_event_timeout(sdp->sd_glock_wait, |
2211 | atomic_read(&sdp->sd_glock_disposal) == 0, | |
2212 | HZ * 600); | |
bc015cb8 | 2213 | glock_hash_walk(dump_glock_func, sdp); |
b3b94faa DT |
2214 | } |
2215 | ||
6802e340 | 2216 | static const char *state2str(unsigned state) |
04b933f2 | 2217 | { |
6802e340 SW |
2218 | switch(state) { |
2219 | case LM_ST_UNLOCKED: | |
2220 | return "UN"; | |
2221 | case LM_ST_SHARED: | |
2222 | return "SH"; | |
2223 | case LM_ST_DEFERRED: | |
2224 | return "DF"; | |
2225 | case LM_ST_EXCLUSIVE: | |
2226 | return "EX"; | |
2227 | } | |
2228 | return "??"; | |
2229 | } | |
2230 | ||
b58bf407 | 2231 | static const char *hflags2str(char *buf, u16 flags, unsigned long iflags) |
6802e340 SW |
2232 | { |
2233 | char *p = buf; | |
2234 | if (flags & LM_FLAG_TRY) | |
2235 | *p++ = 't'; | |
2236 | if (flags & LM_FLAG_TRY_1CB) | |
2237 | *p++ = 'T'; | |
2238 | if (flags & LM_FLAG_NOEXP) | |
2239 | *p++ = 'e'; | |
2240 | if (flags & LM_FLAG_ANY) | |
f057f6cd | 2241 | *p++ = 'A'; |
6802e340 SW |
2242 | if (flags & LM_FLAG_PRIORITY) |
2243 | *p++ = 'p'; | |
06e908cd BP |
2244 | if (flags & LM_FLAG_NODE_SCOPE) |
2245 | *p++ = 'n'; | |
6802e340 SW |
2246 | if (flags & GL_ASYNC) |
2247 | *p++ = 'a'; | |
2248 | if (flags & GL_EXACT) | |
2249 | *p++ = 'E'; | |
6802e340 SW |
2250 | if (flags & GL_NOCACHE) |
2251 | *p++ = 'c'; | |
2252 | if (test_bit(HIF_HOLDER, &iflags)) | |
2253 | *p++ = 'H'; | |
2254 | if (test_bit(HIF_WAIT, &iflags)) | |
2255 | *p++ = 'W'; | |
dc732906 BP |
2256 | if (test_bit(HIF_MAY_DEMOTE, &iflags)) |
2257 | *p++ = 'D'; | |
4c69038d BP |
2258 | if (flags & GL_SKIP) |
2259 | *p++ = 's'; | |
6802e340 SW |
2260 | *p = 0; |
2261 | return buf; | |
04b933f2 RP |
2262 | } |
2263 | ||
b3b94faa DT |
2264 | /** |
2265 | * dump_holder - print information about a glock holder | |
6802e340 | 2266 | * @seq: the seq_file struct |
b3b94faa | 2267 | * @gh: the glock holder |
3792ce97 | 2268 | * @fs_id_buf: pointer to file system id (if requested) |
b3b94faa | 2269 | * |
b3b94faa DT |
2270 | */ |
2271 | ||
3792ce97 BP |
2272 | static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh, |
2273 | const char *fs_id_buf) | |
b3b94faa | 2274 | { |
6802e340 | 2275 | struct task_struct *gh_owner = NULL; |
6802e340 | 2276 | char flags_buf[32]; |
b3b94faa | 2277 | |
0b3a2c99 | 2278 | rcu_read_lock(); |
6802e340 | 2279 | if (gh->gh_owner_pid) |
b1e058da | 2280 | gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID); |
3792ce97 BP |
2281 | gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n", |
2282 | fs_id_buf, state2str(gh->gh_state), | |
cc18152e JP |
2283 | hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags), |
2284 | gh->gh_error, | |
2285 | gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1, | |
2286 | gh_owner ? gh_owner->comm : "(ended)", | |
2287 | (void *)gh->gh_ip); | |
0b3a2c99 | 2288 | rcu_read_unlock(); |
b3b94faa DT |
2289 | } |
2290 | ||
627c10b7 | 2291 | static const char *gflags2str(char *buf, const struct gfs2_glock *gl) |
6802e340 | 2292 | { |
627c10b7 | 2293 | const unsigned long *gflags = &gl->gl_flags; |
6802e340 | 2294 | char *p = buf; |
627c10b7 | 2295 | |
6802e340 SW |
2296 | if (test_bit(GLF_LOCK, gflags)) |
2297 | *p++ = 'l'; | |
6802e340 SW |
2298 | if (test_bit(GLF_DEMOTE, gflags)) |
2299 | *p++ = 'D'; | |
2300 | if (test_bit(GLF_PENDING_DEMOTE, gflags)) | |
2301 | *p++ = 'd'; | |
2302 | if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags)) | |
2303 | *p++ = 'p'; | |
2304 | if (test_bit(GLF_DIRTY, gflags)) | |
2305 | *p++ = 'y'; | |
2306 | if (test_bit(GLF_LFLUSH, gflags)) | |
2307 | *p++ = 'f'; | |
2308 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags)) | |
2309 | *p++ = 'i'; | |
2310 | if (test_bit(GLF_REPLY_PENDING, gflags)) | |
2311 | *p++ = 'r'; | |
f057f6cd | 2312 | if (test_bit(GLF_INITIAL, gflags)) |
d8348de0 | 2313 | *p++ = 'I'; |
f057f6cd SW |
2314 | if (test_bit(GLF_FROZEN, gflags)) |
2315 | *p++ = 'F'; | |
e2c6c8a7 | 2316 | if (!list_empty(&gl->gl_holders)) |
7b5e3d5f | 2317 | *p++ = 'q'; |
627c10b7 SW |
2318 | if (test_bit(GLF_LRU, gflags)) |
2319 | *p++ = 'L'; | |
2320 | if (gl->gl_object) | |
2321 | *p++ = 'o'; | |
a245769f SW |
2322 | if (test_bit(GLF_BLOCKING, gflags)) |
2323 | *p++ = 'b'; | |
5deaf1f6 BP |
2324 | if (test_bit(GLF_PENDING_DELETE, gflags)) |
2325 | *p++ = 'P'; | |
2326 | if (test_bit(GLF_FREEING, gflags)) | |
2327 | *p++ = 'x'; | |
f2e70d8f BP |
2328 | if (test_bit(GLF_INSTANTIATE_NEEDED, gflags)) |
2329 | *p++ = 'n'; | |
2330 | if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags)) | |
2331 | *p++ = 'N'; | |
6802e340 SW |
2332 | *p = 0; |
2333 | return buf; | |
b3b94faa DT |
2334 | } |
2335 | ||
2336 | /** | |
8eae1ca0 | 2337 | * gfs2_dump_glock - print information about a glock |
6802e340 | 2338 | * @seq: The seq_file struct |
b3b94faa | 2339 | * @gl: the glock |
3792ce97 | 2340 | * @fsid: If true, also dump the file system id |
6802e340 SW |
2341 | * |
2342 | * The file format is as follows: | |
2343 | * One line per object, capital letters are used to indicate objects | |
2344 | * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented, | |
2345 | * other objects are indented by a single space and follow the glock to | |
2346 | * which they are related. Fields are indicated by lower case letters | |
2347 | * followed by a colon and the field value, except for strings which are in | |
2348 | * [] so that its possible to see if they are composed of spaces for | |
2349 | * example. The field's are n = number (id of the object), f = flags, | |
2350 | * t = type, s = state, r = refcount, e = error, p = pid. | |
b3b94faa | 2351 | * |
b3b94faa DT |
2352 | */ |
2353 | ||
3792ce97 | 2354 | void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid) |
b3b94faa | 2355 | { |
6802e340 SW |
2356 | const struct gfs2_glock_operations *glops = gl->gl_ops; |
2357 | unsigned long long dtime; | |
2358 | const struct gfs2_holder *gh; | |
2359 | char gflags_buf[32]; | |
3792ce97 | 2360 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
98fb0574 | 2361 | char fs_id_buf[sizeof(sdp->sd_fsname) + 7]; |
7e901d6e | 2362 | unsigned long nrpages = 0; |
b3b94faa | 2363 | |
7e901d6e BP |
2364 | if (gl->gl_ops->go_flags & GLOF_ASPACE) { |
2365 | struct address_space *mapping = gfs2_glock2aspace(gl); | |
2366 | ||
2367 | nrpages = mapping->nrpages; | |
2368 | } | |
3792ce97 BP |
2369 | memset(fs_id_buf, 0, sizeof(fs_id_buf)); |
2370 | if (fsid && sdp) /* safety precaution */ | |
2371 | sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname); | |
6802e340 SW |
2372 | dtime = jiffies - gl->gl_demote_time; |
2373 | dtime *= 1000000/HZ; /* demote time in uSec */ | |
2374 | if (!test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
2375 | dtime = 0; | |
3792ce97 | 2376 | gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d " |
7e901d6e BP |
2377 | "v:%d r:%d m:%ld p:%lu\n", |
2378 | fs_id_buf, state2str(gl->gl_state), | |
2379 | gl->gl_name.ln_type, | |
2380 | (unsigned long long)gl->gl_name.ln_number, | |
2381 | gflags2str(gflags_buf, gl), | |
2382 | state2str(gl->gl_target), | |
2383 | state2str(gl->gl_demote_state), dtime, | |
2384 | atomic_read(&gl->gl_ail_count), | |
2385 | atomic_read(&gl->gl_revokes), | |
2386 | (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages); | |
b3b94faa | 2387 | |
ac3beb6a | 2388 | list_for_each_entry(gh, &gl->gl_holders, gh_list) |
3792ce97 | 2389 | dump_holder(seq, gh, fs_id_buf); |
ac3beb6a | 2390 | |
6802e340 | 2391 | if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump) |
3792ce97 | 2392 | glops->go_dump(seq, gl, fs_id_buf); |
b3b94faa DT |
2393 | } |
2394 | ||
a245769f SW |
2395 | static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr) |
2396 | { | |
2397 | struct gfs2_glock *gl = iter_ptr; | |
2398 | ||
4d207133 | 2399 | seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n", |
a245769f SW |
2400 | gl->gl_name.ln_type, |
2401 | (unsigned long long)gl->gl_name.ln_number, | |
4d207133 BH |
2402 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT], |
2403 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR], | |
2404 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB], | |
2405 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB], | |
2406 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT], | |
2407 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR], | |
2408 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT], | |
2409 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]); | |
a245769f SW |
2410 | return 0; |
2411 | } | |
2412 | ||
2413 | static const char *gfs2_gltype[] = { | |
2414 | "type", | |
2415 | "reserved", | |
2416 | "nondisk", | |
2417 | "inode", | |
2418 | "rgrp", | |
2419 | "meta", | |
2420 | "iopen", | |
2421 | "flock", | |
2422 | "plock", | |
2423 | "quota", | |
2424 | "journal", | |
2425 | }; | |
2426 | ||
2427 | static const char *gfs2_stype[] = { | |
2428 | [GFS2_LKS_SRTT] = "srtt", | |
2429 | [GFS2_LKS_SRTTVAR] = "srttvar", | |
2430 | [GFS2_LKS_SRTTB] = "srttb", | |
2431 | [GFS2_LKS_SRTTVARB] = "srttvarb", | |
2432 | [GFS2_LKS_SIRT] = "sirt", | |
2433 | [GFS2_LKS_SIRTVAR] = "sirtvar", | |
2434 | [GFS2_LKS_DCOUNT] = "dlm", | |
2435 | [GFS2_LKS_QCOUNT] = "queue", | |
2436 | }; | |
2437 | ||
2438 | #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype)) | |
2439 | ||
2440 | static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr) | |
2441 | { | |
81648d04 AG |
2442 | struct gfs2_sbd *sdp = seq->private; |
2443 | loff_t pos = *(loff_t *)iter_ptr; | |
2444 | unsigned index = pos >> 3; | |
2445 | unsigned subindex = pos & 0x07; | |
a245769f SW |
2446 | int i; |
2447 | ||
2448 | if (index == 0 && subindex != 0) | |
2449 | return 0; | |
6802e340 | 2450 | |
a245769f SW |
2451 | seq_printf(seq, "%-10s %8s:", gfs2_gltype[index], |
2452 | (index == 0) ? "cpu": gfs2_stype[subindex]); | |
b3b94faa | 2453 | |
a245769f SW |
2454 | for_each_possible_cpu(i) { |
2455 | const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i); | |
8f7e0a80 AG |
2456 | |
2457 | if (index == 0) | |
2458 | seq_printf(seq, " %15u", i); | |
2459 | else | |
2460 | seq_printf(seq, " %15llu", (unsigned long long)lkstats-> | |
2461 | lkstats[index - 1].stats[subindex]); | |
a245769f SW |
2462 | } |
2463 | seq_putc(seq, '\n'); | |
2464 | return 0; | |
2465 | } | |
8fbbfd21 | 2466 | |
85d1da67 SW |
2467 | int __init gfs2_glock_init(void) |
2468 | { | |
0515480a | 2469 | int i, ret; |
88ffbf3e BP |
2470 | |
2471 | ret = rhashtable_init(&gl_hash_table, &ht_parms); | |
2472 | if (ret < 0) | |
2473 | return ret; | |
8fbbfd21 | 2474 | |
d2115778 | 2475 | glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM | |
58a69cb4 | 2476 | WQ_HIGHPRI | WQ_FREEZABLE, 0); |
88ffbf3e BP |
2477 | if (!glock_workqueue) { |
2478 | rhashtable_destroy(&gl_hash_table); | |
dfc4616d | 2479 | return -ENOMEM; |
88ffbf3e | 2480 | } |
d2115778 | 2481 | gfs2_delete_workqueue = alloc_workqueue("delete_workqueue", |
58a69cb4 | 2482 | WQ_MEM_RECLAIM | WQ_FREEZABLE, |
d2115778 | 2483 | 0); |
dfc4616d | 2484 | if (!gfs2_delete_workqueue) { |
b94a170e | 2485 | destroy_workqueue(glock_workqueue); |
88ffbf3e | 2486 | rhashtable_destroy(&gl_hash_table); |
dfc4616d | 2487 | return -ENOMEM; |
b94a170e | 2488 | } |
97cc1025 | 2489 | |
e33c267a | 2490 | ret = register_shrinker(&glock_shrinker, "gfs2-glock"); |
e0d735c1 CY |
2491 | if (ret) { |
2492 | destroy_workqueue(gfs2_delete_workqueue); | |
2493 | destroy_workqueue(glock_workqueue); | |
2494 | rhashtable_destroy(&gl_hash_table); | |
2495 | return ret; | |
2496 | } | |
c4f68a13 | 2497 | |
0515480a AG |
2498 | for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++) |
2499 | init_waitqueue_head(glock_wait_table + i); | |
2500 | ||
85d1da67 SW |
2501 | return 0; |
2502 | } | |
2503 | ||
8fbbfd21 SW |
2504 | void gfs2_glock_exit(void) |
2505 | { | |
97cc1025 | 2506 | unregister_shrinker(&glock_shrinker); |
88ffbf3e | 2507 | rhashtable_destroy(&gl_hash_table); |
c4f68a13 | 2508 | destroy_workqueue(glock_workqueue); |
b94a170e | 2509 | destroy_workqueue(gfs2_delete_workqueue); |
8fbbfd21 SW |
2510 | } |
2511 | ||
7ac07fda | 2512 | static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n) |
bc015cb8 | 2513 | { |
3fd5d3ad AG |
2514 | struct gfs2_glock *gl = gi->gl; |
2515 | ||
2516 | if (gl) { | |
2517 | if (n == 0) | |
2518 | return; | |
2519 | if (!lockref_put_not_zero(&gl->gl_lockref)) | |
2520 | gfs2_glock_queue_put(gl); | |
7ac07fda AG |
2521 | } |
2522 | for (;;) { | |
3fd5d3ad AG |
2523 | gl = rhashtable_walk_next(&gi->hti); |
2524 | if (IS_ERR_OR_NULL(gl)) { | |
2525 | if (gl == ERR_PTR(-EAGAIN)) { | |
2526 | n = 1; | |
2527 | continue; | |
7ac07fda | 2528 | } |
3fd5d3ad AG |
2529 | gl = NULL; |
2530 | break; | |
2531 | } | |
2532 | if (gl->gl_name.ln_sbd != gi->sdp) | |
2533 | continue; | |
2534 | if (n <= 1) { | |
2535 | if (!lockref_get_not_dead(&gl->gl_lockref)) | |
2536 | continue; | |
2537 | break; | |
2538 | } else { | |
2539 | if (__lockref_is_dead(&gl->gl_lockref)) | |
2540 | continue; | |
2541 | n--; | |
bc015cb8 | 2542 | } |
14d37564 | 2543 | } |
3fd5d3ad | 2544 | gi->gl = gl; |
7c52b166 RP |
2545 | } |
2546 | ||
6802e340 | 2547 | static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos) |
27c3b415 | 2548 | __acquires(RCU) |
7c52b166 | 2549 | { |
6802e340 | 2550 | struct gfs2_glock_iter *gi = seq->private; |
7ac07fda | 2551 | loff_t n; |
ba1ddcb6 | 2552 | |
7ac07fda AG |
2553 | /* |
2554 | * We can either stay where we are, skip to the next hash table | |
2555 | * entry, or start from the beginning. | |
2556 | */ | |
2557 | if (*pos < gi->last_pos) { | |
2558 | rhashtable_walk_exit(&gi->hti); | |
2559 | rhashtable_walk_enter(&gl_hash_table, &gi->hti); | |
2560 | n = *pos + 1; | |
2561 | } else { | |
2562 | n = *pos - gi->last_pos; | |
2563 | } | |
7c52b166 | 2564 | |
7ac07fda | 2565 | rhashtable_walk_start(&gi->hti); |
7c52b166 | 2566 | |
7ac07fda | 2567 | gfs2_glock_iter_next(gi, n); |
ba1ddcb6 | 2568 | gi->last_pos = *pos; |
6802e340 | 2569 | return gi->gl; |
7c52b166 RP |
2570 | } |
2571 | ||
6802e340 | 2572 | static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr, |
7c52b166 RP |
2573 | loff_t *pos) |
2574 | { | |
6802e340 | 2575 | struct gfs2_glock_iter *gi = seq->private; |
7c52b166 RP |
2576 | |
2577 | (*pos)++; | |
ba1ddcb6 | 2578 | gi->last_pos = *pos; |
7ac07fda | 2579 | gfs2_glock_iter_next(gi, 1); |
6802e340 | 2580 | return gi->gl; |
7c52b166 RP |
2581 | } |
2582 | ||
6802e340 | 2583 | static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr) |
27c3b415 | 2584 | __releases(RCU) |
7c52b166 | 2585 | { |
6802e340 | 2586 | struct gfs2_glock_iter *gi = seq->private; |
bc015cb8 | 2587 | |
88ffbf3e | 2588 | rhashtable_walk_stop(&gi->hti); |
7c52b166 RP |
2589 | } |
2590 | ||
6802e340 | 2591 | static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr) |
7c52b166 | 2592 | { |
3792ce97 | 2593 | dump_glock(seq, iter_ptr, false); |
ac3beb6a | 2594 | return 0; |
7c52b166 RP |
2595 | } |
2596 | ||
a245769f SW |
2597 | static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos) |
2598 | { | |
81648d04 | 2599 | preempt_disable(); |
a245769f SW |
2600 | if (*pos >= GFS2_NR_SBSTATS) |
2601 | return NULL; | |
81648d04 | 2602 | return pos; |
a245769f SW |
2603 | } |
2604 | ||
2605 | static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr, | |
2606 | loff_t *pos) | |
2607 | { | |
a245769f | 2608 | (*pos)++; |
81648d04 | 2609 | if (*pos >= GFS2_NR_SBSTATS) |
a245769f | 2610 | return NULL; |
81648d04 | 2611 | return pos; |
a245769f SW |
2612 | } |
2613 | ||
2614 | static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr) | |
2615 | { | |
2616 | preempt_enable(); | |
2617 | } | |
2618 | ||
4ef29002 | 2619 | static const struct seq_operations gfs2_glock_seq_ops = { |
7c52b166 RP |
2620 | .start = gfs2_glock_seq_start, |
2621 | .next = gfs2_glock_seq_next, | |
2622 | .stop = gfs2_glock_seq_stop, | |
2623 | .show = gfs2_glock_seq_show, | |
2624 | }; | |
2625 | ||
a245769f SW |
2626 | static const struct seq_operations gfs2_glstats_seq_ops = { |
2627 | .start = gfs2_glock_seq_start, | |
2628 | .next = gfs2_glock_seq_next, | |
2629 | .stop = gfs2_glock_seq_stop, | |
2630 | .show = gfs2_glstats_seq_show, | |
2631 | }; | |
2632 | ||
e8a8023e | 2633 | static const struct seq_operations gfs2_sbstats_sops = { |
a245769f SW |
2634 | .start = gfs2_sbstats_seq_start, |
2635 | .next = gfs2_sbstats_seq_next, | |
2636 | .stop = gfs2_sbstats_seq_stop, | |
2637 | .show = gfs2_sbstats_seq_show, | |
2638 | }; | |
2639 | ||
0fe2f1e9 SW |
2640 | #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL) |
2641 | ||
92ecd73a AG |
2642 | static int __gfs2_glocks_open(struct inode *inode, struct file *file, |
2643 | const struct seq_operations *ops) | |
7c52b166 | 2644 | { |
92ecd73a | 2645 | int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter)); |
6802e340 SW |
2646 | if (ret == 0) { |
2647 | struct seq_file *seq = file->private_data; | |
2648 | struct gfs2_glock_iter *gi = seq->private; | |
88ffbf3e | 2649 | |
6802e340 | 2650 | gi->sdp = inode->i_private; |
0fe2f1e9 | 2651 | seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN); |
df5d2f55 | 2652 | if (seq->buf) |
0fe2f1e9 | 2653 | seq->size = GFS2_SEQ_GOODSIZE; |
7ac07fda AG |
2654 | /* |
2655 | * Initially, we are "before" the first hash table entry; the | |
2656 | * first call to rhashtable_walk_next gets us the first entry. | |
2657 | */ | |
2658 | gi->last_pos = -1; | |
88ffbf3e | 2659 | gi->gl = NULL; |
7ac07fda | 2660 | rhashtable_walk_enter(&gl_hash_table, &gi->hti); |
6802e340 SW |
2661 | } |
2662 | return ret; | |
7c52b166 RP |
2663 | } |
2664 | ||
92ecd73a AG |
2665 | static int gfs2_glocks_open(struct inode *inode, struct file *file) |
2666 | { | |
2667 | return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops); | |
2668 | } | |
2669 | ||
88ffbf3e BP |
2670 | static int gfs2_glocks_release(struct inode *inode, struct file *file) |
2671 | { | |
2672 | struct seq_file *seq = file->private_data; | |
2673 | struct gfs2_glock_iter *gi = seq->private; | |
2674 | ||
3fd5d3ad AG |
2675 | if (gi->gl) |
2676 | gfs2_glock_put(gi->gl); | |
7ac07fda | 2677 | rhashtable_walk_exit(&gi->hti); |
88ffbf3e BP |
2678 | return seq_release_private(inode, file); |
2679 | } | |
2680 | ||
a245769f SW |
2681 | static int gfs2_glstats_open(struct inode *inode, struct file *file) |
2682 | { | |
92ecd73a | 2683 | return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops); |
a245769f SW |
2684 | } |
2685 | ||
a245769f SW |
2686 | static const struct file_operations gfs2_glocks_fops = { |
2687 | .owner = THIS_MODULE, | |
2688 | .open = gfs2_glocks_open, | |
2689 | .read = seq_read, | |
2690 | .llseek = seq_lseek, | |
88ffbf3e | 2691 | .release = gfs2_glocks_release, |
a245769f SW |
2692 | }; |
2693 | ||
2694 | static const struct file_operations gfs2_glstats_fops = { | |
7c52b166 | 2695 | .owner = THIS_MODULE, |
a245769f SW |
2696 | .open = gfs2_glstats_open, |
2697 | .read = seq_read, | |
2698 | .llseek = seq_lseek, | |
88ffbf3e | 2699 | .release = gfs2_glocks_release, |
a245769f SW |
2700 | }; |
2701 | ||
e8a8023e | 2702 | DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats); |
7c52b166 | 2703 | |
2abbf9a4 GKH |
2704 | void gfs2_create_debugfs_file(struct gfs2_sbd *sdp) |
2705 | { | |
2706 | sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root); | |
7c52b166 | 2707 | |
2abbf9a4 GKH |
2708 | debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, |
2709 | &gfs2_glocks_fops); | |
2710 | ||
2711 | debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, | |
2712 | &gfs2_glstats_fops); | |
2713 | ||
2714 | debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, | |
2715 | &gfs2_sbstats_fops); | |
7c52b166 RP |
2716 | } |
2717 | ||
2718 | void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp) | |
2719 | { | |
2abbf9a4 GKH |
2720 | debugfs_remove_recursive(sdp->debugfs_dir); |
2721 | sdp->debugfs_dir = NULL; | |
7c52b166 RP |
2722 | } |
2723 | ||
2abbf9a4 | 2724 | void gfs2_register_debugfs(void) |
7c52b166 RP |
2725 | { |
2726 | gfs2_root = debugfs_create_dir("gfs2", NULL); | |
7c52b166 RP |
2727 | } |
2728 | ||
2729 | void gfs2_unregister_debugfs(void) | |
2730 | { | |
2731 | debugfs_remove(gfs2_root); | |
5f882096 | 2732 | gfs2_root = NULL; |
7c52b166 | 2733 | } |