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