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> |
4480c27c AG |
36 | #include <linux/pid_namespace.h> |
37 | #include <linux/fdtable.h> | |
38 | #include <linux/file.h> | |
b3b94faa DT |
39 | |
40 | #include "gfs2.h" | |
5c676f6d | 41 | #include "incore.h" |
b3b94faa DT |
42 | #include "glock.h" |
43 | #include "glops.h" | |
44 | #include "inode.h" | |
b3b94faa DT |
45 | #include "lops.h" |
46 | #include "meta_io.h" | |
47 | #include "quota.h" | |
48 | #include "super.h" | |
5c676f6d | 49 | #include "util.h" |
813e0c46 | 50 | #include "bmap.h" |
63997775 SW |
51 | #define CREATE_TRACE_POINTS |
52 | #include "trace_gfs2.h" | |
b3b94faa | 53 | |
6802e340 | 54 | struct gfs2_glock_iter { |
ba1ddcb6 | 55 | struct gfs2_sbd *sdp; /* incore superblock */ |
88ffbf3e | 56 | struct rhashtable_iter hti; /* rhashtable iterator */ |
ba1ddcb6 SW |
57 | struct gfs2_glock *gl; /* current glock struct */ |
58 | loff_t last_pos; /* last position */ | |
7c52b166 RP |
59 | }; |
60 | ||
b3b94faa DT |
61 | typedef void (*glock_examiner) (struct gfs2_glock * gl); |
62 | ||
6802e340 | 63 | static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target); |
dc732906 | 64 | static void __gfs2_glock_dq(struct gfs2_holder *gh); |
86934198 BP |
65 | static void handle_callback(struct gfs2_glock *gl, unsigned int state, |
66 | unsigned long delay, bool remote); | |
c4f68a13 | 67 | |
7c52b166 | 68 | static struct dentry *gfs2_root; |
c4f68a13 | 69 | static struct workqueue_struct *glock_workqueue; |
97cc1025 SW |
70 | static LIST_HEAD(lru_list); |
71 | static atomic_t lru_count = ATOMIC_INIT(0); | |
eb8374e7 | 72 | static DEFINE_SPINLOCK(lru_lock); |
08bc2dbc | 73 | |
b6397893 | 74 | #define GFS2_GL_HASH_SHIFT 15 |
47a9a527 | 75 | #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT) |
087efdd3 | 76 | |
d296b15e | 77 | static const struct rhashtable_params ht_parms = { |
88ffbf3e | 78 | .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4, |
972b044e | 79 | .key_len = offsetofend(struct lm_lockname, ln_type), |
88ffbf3e BP |
80 | .key_offset = offsetof(struct gfs2_glock, gl_name), |
81 | .head_offset = offsetof(struct gfs2_glock, gl_node), | |
82 | }; | |
b3b94faa | 83 | |
88ffbf3e | 84 | static struct rhashtable gl_hash_table; |
b3b94faa | 85 | |
0515480a AG |
86 | #define GLOCK_WAIT_TABLE_BITS 12 |
87 | #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS) | |
88 | static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned; | |
89 | ||
90 | struct wait_glock_queue { | |
91 | struct lm_lockname *name; | |
92 | wait_queue_entry_t wait; | |
93 | }; | |
94 | ||
95 | static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode, | |
96 | int sync, void *key) | |
97 | { | |
98 | struct wait_glock_queue *wait_glock = | |
99 | container_of(wait, struct wait_glock_queue, wait); | |
100 | struct lm_lockname *wait_name = wait_glock->name; | |
101 | struct lm_lockname *wake_name = key; | |
102 | ||
103 | if (wake_name->ln_sbd != wait_name->ln_sbd || | |
104 | wake_name->ln_number != wait_name->ln_number || | |
105 | wake_name->ln_type != wait_name->ln_type) | |
106 | return 0; | |
107 | return autoremove_wake_function(wait, mode, sync, key); | |
108 | } | |
109 | ||
110 | static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name) | |
111 | { | |
605b0487 | 112 | u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0); |
0515480a AG |
113 | |
114 | return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS); | |
115 | } | |
116 | ||
0515480a AG |
117 | /** |
118 | * wake_up_glock - Wake up waiters on a glock | |
119 | * @gl: the glock | |
120 | */ | |
121 | static void wake_up_glock(struct gfs2_glock *gl) | |
122 | { | |
123 | wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name); | |
124 | ||
125 | if (waitqueue_active(wq)) | |
126 | __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name); | |
127 | } | |
128 | ||
961ae1d8 | 129 | static void gfs2_glock_dealloc(struct rcu_head *rcu) |
b3b94faa | 130 | { |
961ae1d8 | 131 | struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu); |
b3b94faa | 132 | |
15f2547b | 133 | kfree(gl->gl_lksb.sb_lvbptr); |
11d8b79e KC |
134 | if (gl->gl_ops->go_flags & GLOF_ASPACE) { |
135 | struct gfs2_glock_aspace *gla = | |
136 | container_of(gl, struct gfs2_glock_aspace, glock); | |
137 | kmem_cache_free(gfs2_glock_aspace_cachep, gla); | |
138 | } else | |
bc015cb8 | 139 | kmem_cache_free(gfs2_glock_cachep, gl); |
961ae1d8 AG |
140 | } |
141 | ||
a72d2401 BP |
142 | /** |
143 | * glock_blocked_by_withdraw - determine if we can still use a glock | |
144 | * @gl: the glock | |
145 | * | |
146 | * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted | |
147 | * when we're withdrawn. For example, to maintain metadata integrity, we should | |
af1abe11 AG |
148 | * disallow the use of inode and rgrp glocks when withdrawn. Other glocks like |
149 | * the iopen or freeze glock may be safely used because none of their | |
a72d2401 BP |
150 | * metadata goes through the journal. So in general, we should disallow all |
151 | * glocks that are journaled, and allow all the others. One exception is: | |
152 | * we need to allow our active journal to be promoted and demoted so others | |
153 | * may recover it and we can reacquire it when they're done. | |
154 | */ | |
155 | static bool glock_blocked_by_withdraw(struct gfs2_glock *gl) | |
156 | { | |
157 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
158 | ||
4d927b03 | 159 | if (!gfs2_withdrawing_or_withdrawn(sdp)) |
a72d2401 BP |
160 | return false; |
161 | if (gl->gl_ops->go_flags & GLOF_NONDISK) | |
162 | return false; | |
163 | if (!sdp->sd_jdesc || | |
164 | gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr) | |
165 | return false; | |
166 | return true; | |
167 | } | |
168 | ||
d98779e6 | 169 | static void __gfs2_glock_free(struct gfs2_glock *gl) |
961ae1d8 | 170 | { |
0515480a AG |
171 | rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms); |
172 | smp_mb(); | |
173 | wake_up_glock(gl); | |
961ae1d8 | 174 | call_rcu(&gl->gl_rcu, gfs2_glock_dealloc); |
d98779e6 AG |
175 | } |
176 | ||
177 | void gfs2_glock_free(struct gfs2_glock *gl) { | |
178 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
179 | ||
180 | __gfs2_glock_free(gl); | |
bc015cb8 | 181 | if (atomic_dec_and_test(&sdp->sd_glock_disposal)) |
3c69c437 | 182 | wake_up(&sdp->sd_kill_wait); |
b3b94faa DT |
183 | } |
184 | ||
d98779e6 AG |
185 | void gfs2_glock_free_later(struct gfs2_glock *gl) { |
186 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
187 | ||
188 | spin_lock(&lru_lock); | |
189 | list_add(&gl->gl_lru, &sdp->sd_dead_glocks); | |
190 | spin_unlock(&lru_lock); | |
191 | if (atomic_dec_and_test(&sdp->sd_glock_disposal)) | |
192 | wake_up(&sdp->sd_kill_wait); | |
193 | } | |
194 | ||
195 | static void gfs2_free_dead_glocks(struct gfs2_sbd *sdp) | |
196 | { | |
197 | struct list_head *list = &sdp->sd_dead_glocks; | |
198 | ||
199 | while(!list_empty(list)) { | |
200 | struct gfs2_glock *gl; | |
201 | ||
202 | gl = list_first_entry(list, struct gfs2_glock, gl_lru); | |
203 | list_del_init(&gl->gl_lru); | |
204 | __gfs2_glock_free(gl); | |
205 | } | |
206 | } | |
207 | ||
b3b94faa DT |
208 | /** |
209 | * gfs2_glock_hold() - increment reference count on glock | |
210 | * @gl: The glock to hold | |
211 | * | |
212 | */ | |
213 | ||
4ad02083 | 214 | struct gfs2_glock *gfs2_glock_hold(struct gfs2_glock *gl) |
b3b94faa | 215 | { |
e66cf161 SW |
216 | GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref)); |
217 | lockref_get(&gl->gl_lockref); | |
4ad02083 | 218 | return gl; |
b3b94faa DT |
219 | } |
220 | ||
8ff22a6f BM |
221 | /** |
222 | * demote_ok - Check to see if it's ok to unlock a glock | |
223 | * @gl: the glock | |
224 | * | |
225 | * Returns: 1 if it's ok | |
226 | */ | |
227 | ||
228 | static int demote_ok(const struct gfs2_glock *gl) | |
229 | { | |
230 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
231 | ||
232 | if (gl->gl_state == LM_ST_UNLOCKED) | |
233 | return 0; | |
f42ab085 | 234 | if (!list_empty(&gl->gl_holders)) |
8ff22a6f BM |
235 | return 0; |
236 | if (glops->go_demote_ok) | |
237 | return glops->go_demote_ok(gl); | |
238 | return 1; | |
239 | } | |
240 | ||
bc015cb8 | 241 | |
29687a2a SW |
242 | void gfs2_glock_add_to_lru(struct gfs2_glock *gl) |
243 | { | |
7881ef3f RL |
244 | if (!(gl->gl_ops->go_flags & GLOF_LRU)) |
245 | return; | |
246 | ||
29687a2a SW |
247 | spin_lock(&lru_lock); |
248 | ||
38a618db | 249 | list_move_tail(&gl->gl_lru, &lru_list); |
7881ef3f RL |
250 | |
251 | if (!test_bit(GLF_LRU, &gl->gl_flags)) { | |
252 | set_bit(GLF_LRU, &gl->gl_flags); | |
29687a2a | 253 | atomic_inc(&lru_count); |
7881ef3f | 254 | } |
29687a2a | 255 | |
29687a2a SW |
256 | spin_unlock(&lru_lock); |
257 | } | |
258 | ||
8f6cb409 | 259 | static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl) |
f42ab085 | 260 | { |
645ebd49 BP |
261 | if (!(gl->gl_ops->go_flags & GLOF_LRU)) |
262 | return; | |
263 | ||
8f6cb409 | 264 | spin_lock(&lru_lock); |
7881ef3f | 265 | if (test_bit(GLF_LRU, &gl->gl_flags)) { |
f42ab085 SW |
266 | list_del_init(&gl->gl_lru); |
267 | atomic_dec(&lru_count); | |
268 | clear_bit(GLF_LRU, &gl->gl_flags); | |
269 | } | |
270 | spin_unlock(&lru_lock); | |
271 | } | |
272 | ||
6b0c7440 AG |
273 | /* |
274 | * Enqueue the glock on the work queue. Passes one glock reference on to the | |
275 | * work queue. | |
b3b94faa | 276 | */ |
1e860444 | 277 | static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) { |
6b0c7440 AG |
278 | if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) { |
279 | /* | |
280 | * We are holding the lockref spinlock, and the work was still | |
281 | * queued above. The queued work (glock_work_func) takes that | |
282 | * spinlock before dropping its glock reference(s), so it | |
283 | * cannot have dropped them in the meantime. | |
284 | */ | |
285 | GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2); | |
286 | gl->gl_lockref.count--; | |
287 | } | |
288 | } | |
b3b94faa | 289 | |
6b0c7440 | 290 | static void __gfs2_glock_put(struct gfs2_glock *gl) |
b3b94faa | 291 | { |
15562c43 | 292 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
bc015cb8 | 293 | struct address_space *mapping = gfs2_glock2aspace(gl); |
b3b94faa | 294 | |
e66cf161 | 295 | lockref_mark_dead(&gl->gl_lockref); |
e66cf161 | 296 | spin_unlock(&gl->gl_lockref.lock); |
228804a3 | 297 | gfs2_glock_remove_from_lru(gl); |
e66cf161 | 298 | GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders)); |
ee1e2c77 BP |
299 | if (mapping) { |
300 | truncate_inode_pages_final(mapping); | |
4d927b03 | 301 | if (!gfs2_withdrawing_or_withdrawn(sdp)) |
7716506a | 302 | GLOCK_BUG_ON(gl, !mapping_empty(mapping)); |
ee1e2c77 | 303 | } |
e66cf161 SW |
304 | trace_gfs2_glock_put(gl); |
305 | sdp->sd_lockstruct.ls_ops->lm_put_lock(gl); | |
b3b94faa DT |
306 | } |
307 | ||
6b0c7440 AG |
308 | /** |
309 | * gfs2_glock_put() - Decrement reference count on glock | |
310 | * @gl: The glock to put | |
311 | * | |
312 | */ | |
313 | ||
314 | void gfs2_glock_put(struct gfs2_glock *gl) | |
315 | { | |
316 | if (lockref_put_or_lock(&gl->gl_lockref)) | |
317 | return; | |
318 | ||
319 | __gfs2_glock_put(gl); | |
320 | } | |
321 | ||
ee2be7d7 AG |
322 | /* |
323 | * gfs2_glock_put_async - Decrement reference count without sleeping | |
324 | * @gl: The glock to put | |
325 | * | |
326 | * Decrement the reference count on glock immediately unless it is the last | |
327 | * reference. Defer putting the last reference to work queue context. | |
328 | */ | |
329 | void gfs2_glock_put_async(struct gfs2_glock *gl) | |
330 | { | |
331 | if (lockref_put_or_lock(&gl->gl_lockref)) | |
332 | return; | |
333 | ||
1e860444 AG |
334 | GLOCK_BUG_ON(gl, gl->gl_lockref.count != 1); |
335 | gfs2_glock_queue_work(gl, 0); | |
ee2be7d7 AG |
336 | spin_unlock(&gl->gl_lockref.lock); |
337 | } | |
338 | ||
6802e340 | 339 | /** |
61444649 | 340 | * may_grant - check if it's ok to grant a new lock |
6802e340 | 341 | * @gl: The glock |
61444649 | 342 | * @current_gh: One of the current holders of @gl |
6802e340 SW |
343 | * @gh: The lock request which we wish to grant |
344 | * | |
61444649 AG |
345 | * With our current compatibility rules, if a glock has one or more active |
346 | * holders (HIF_HOLDER flag set), any of those holders can be passed in as | |
347 | * @current_gh; they are all the same as far as compatibility with the new @gh | |
348 | * goes. | |
349 | * | |
350 | * Returns true if it's ok to grant the lock. | |
6802e340 SW |
351 | */ |
352 | ||
61444649 AG |
353 | static inline bool may_grant(struct gfs2_glock *gl, |
354 | struct gfs2_holder *current_gh, | |
355 | struct gfs2_holder *gh) | |
6802e340 | 356 | { |
61444649 AG |
357 | if (current_gh) { |
358 | GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags)); | |
359 | ||
360 | switch(current_gh->gh_state) { | |
361 | case LM_ST_EXCLUSIVE: | |
362 | /* | |
363 | * Here we make a special exception to grant holders | |
364 | * who agree to share the EX lock with other holders | |
365 | * who also have the bit set. If the original holder | |
366 | * has the LM_FLAG_NODE_SCOPE bit set, we grant more | |
367 | * holders with the bit set. | |
368 | */ | |
369 | return gh->gh_state == LM_ST_EXCLUSIVE && | |
370 | (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) && | |
371 | (gh->gh_flags & LM_FLAG_NODE_SCOPE); | |
372 | ||
373 | case LM_ST_SHARED: | |
374 | case LM_ST_DEFERRED: | |
375 | return gh->gh_state == current_gh->gh_state; | |
376 | ||
377 | default: | |
378 | return false; | |
379 | } | |
06e908cd | 380 | } |
61444649 | 381 | |
6802e340 | 382 | if (gl->gl_state == gh->gh_state) |
61444649 | 383 | return true; |
6802e340 | 384 | if (gh->gh_flags & GL_EXACT) |
61444649 | 385 | return false; |
209806ab | 386 | if (gl->gl_state == LM_ST_EXCLUSIVE) { |
61444649 AG |
387 | return gh->gh_state == LM_ST_SHARED || |
388 | gh->gh_state == LM_ST_DEFERRED; | |
209806ab | 389 | } |
61444649 AG |
390 | if (gh->gh_flags & LM_FLAG_ANY) |
391 | return gl->gl_state != LM_ST_UNLOCKED; | |
392 | return false; | |
6802e340 SW |
393 | } |
394 | ||
395 | static void gfs2_holder_wake(struct gfs2_holder *gh) | |
396 | { | |
397 | clear_bit(HIF_WAIT, &gh->gh_iflags); | |
4e857c58 | 398 | smp_mb__after_atomic(); |
6802e340 | 399 | wake_up_bit(&gh->gh_iflags, HIF_WAIT); |
ad26967b BP |
400 | if (gh->gh_flags & GL_ASYNC) { |
401 | struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd; | |
402 | ||
403 | wake_up(&sdp->sd_async_glock_wait); | |
404 | } | |
6802e340 SW |
405 | } |
406 | ||
d5341a92 SW |
407 | /** |
408 | * do_error - Something unexpected has happened during a lock request | |
c551f66c LJ |
409 | * @gl: The glock |
410 | * @ret: The status from the DLM | |
d5341a92 SW |
411 | */ |
412 | ||
a527b38e | 413 | static void do_error(struct gfs2_glock *gl, const int ret) |
d5341a92 SW |
414 | { |
415 | struct gfs2_holder *gh, *tmp; | |
416 | ||
417 | list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) { | |
ba3e77a4 | 418 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) |
d5341a92 SW |
419 | continue; |
420 | if (ret & LM_OUT_ERROR) | |
421 | gh->gh_error = -EIO; | |
422 | else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) | |
423 | gh->gh_error = GLR_TRYFAILED; | |
424 | else | |
425 | continue; | |
426 | list_del_init(&gh->gh_list); | |
427 | trace_gfs2_glock_queue(gh, 0); | |
428 | gfs2_holder_wake(gh); | |
429 | } | |
430 | } | |
431 | ||
61444649 AG |
432 | /** |
433 | * find_first_holder - find the first "holder" gh | |
434 | * @gl: the glock | |
435 | */ | |
436 | ||
437 | static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl) | |
438 | { | |
439 | struct gfs2_holder *gh; | |
440 | ||
441 | if (!list_empty(&gl->gl_holders)) { | |
442 | gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, | |
443 | gh_list); | |
444 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
445 | return gh; | |
446 | } | |
447 | return NULL; | |
448 | } | |
449 | ||
e6f85600 BP |
450 | /* |
451 | * gfs2_instantiate - Call the glops instantiate function | |
3c5c67ec | 452 | * @gh: The glock holder |
e6f85600 | 453 | * |
53d69132 | 454 | * Returns: 0 if instantiate was successful, or error. |
e6f85600 | 455 | */ |
f2e70d8f | 456 | int gfs2_instantiate(struct gfs2_holder *gh) |
e6f85600 BP |
457 | { |
458 | struct gfs2_glock *gl = gh->gh_gl; | |
459 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
f2e70d8f BP |
460 | int ret; |
461 | ||
462 | again: | |
463 | if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags)) | |
86c30a01 | 464 | goto done; |
e6f85600 | 465 | |
f2e70d8f BP |
466 | /* |
467 | * Since we unlock the lockref lock, we set a flag to indicate | |
468 | * instantiate is in progress. | |
469 | */ | |
7a92deaa | 470 | if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) { |
f2e70d8f BP |
471 | wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG, |
472 | TASK_UNINTERRUPTIBLE); | |
473 | /* | |
474 | * Here we just waited for a different instantiate to finish. | |
475 | * But that may not have been successful, as when a process | |
476 | * locks an inode glock _before_ it has an actual inode to | |
477 | * instantiate into. So we check again. This process might | |
478 | * have an inode to instantiate, so might be successful. | |
479 | */ | |
480 | goto again; | |
481 | } | |
482 | ||
5f38a4d3 | 483 | ret = glops->go_instantiate(gl); |
f2e70d8f BP |
484 | if (!ret) |
485 | clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags); | |
7a92deaa | 486 | clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags); |
86c30a01 AG |
487 | if (ret) |
488 | return ret; | |
489 | ||
490 | done: | |
491 | if (glops->go_held) | |
492 | return glops->go_held(gh); | |
493 | return 0; | |
e6f85600 BP |
494 | } |
495 | ||
6802e340 SW |
496 | /** |
497 | * do_promote - promote as many requests as possible on the current queue | |
498 | * @gl: The glock | |
499 | * | |
de3e7f97 | 500 | * Returns true on success (i.e., progress was made or there are no waiters). |
6802e340 SW |
501 | */ |
502 | ||
de3e7f97 | 503 | static bool do_promote(struct gfs2_glock *gl) |
6802e340 | 504 | { |
6feaec81 | 505 | struct gfs2_holder *gh, *current_gh; |
6802e340 | 506 | |
ba3e77a4 | 507 | current_gh = find_first_holder(gl); |
6feaec81 | 508 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { |
60d8bae9 | 509 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) |
6802e340 | 510 | continue; |
8f0028fc | 511 | if (!may_grant(gl, current_gh, gh)) { |
17a6ecee | 512 | /* |
565f82b5 BP |
513 | * If we get here, it means we may not grant this |
514 | * holder for some reason. If this holder is at the | |
515 | * head of the list, it means we have a blocked holder | |
de3e7f97 | 516 | * at the head, so return false. |
17a6ecee | 517 | */ |
5a27a43e | 518 | if (list_is_first(&gh->gh_list, &gl->gl_holders)) |
de3e7f97 | 519 | return false; |
17a6ecee BP |
520 | do_error(gl, 0); |
521 | break; | |
522 | } | |
17a6ecee BP |
523 | set_bit(HIF_HOLDER, &gh->gh_iflags); |
524 | trace_gfs2_promote(gh); | |
525 | gfs2_holder_wake(gh); | |
ba3e77a4 | 526 | if (!current_gh) |
0befb851 | 527 | current_gh = gh; |
6802e340 | 528 | } |
de3e7f97 | 529 | return true; |
6802e340 SW |
530 | } |
531 | ||
6802e340 SW |
532 | /** |
533 | * find_first_waiter - find the first gh that's waiting for the glock | |
534 | * @gl: the glock | |
535 | */ | |
536 | ||
537 | static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl) | |
538 | { | |
539 | struct gfs2_holder *gh; | |
540 | ||
541 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { | |
542 | if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
543 | return gh; | |
544 | } | |
545 | return NULL; | |
546 | } | |
547 | ||
f9f229c1 AG |
548 | /** |
549 | * find_last_waiter - find the last gh that's waiting for the glock | |
550 | * @gl: the glock | |
551 | * | |
552 | * This also is a fast way of finding out if there are any waiters. | |
553 | */ | |
554 | ||
555 | static inline struct gfs2_holder *find_last_waiter(const struct gfs2_glock *gl) | |
556 | { | |
557 | struct gfs2_holder *gh; | |
558 | ||
559 | if (list_empty(&gl->gl_holders)) | |
560 | return NULL; | |
561 | gh = list_last_entry(&gl->gl_holders, struct gfs2_holder, gh_list); | |
562 | return test_bit(HIF_HOLDER, &gh->gh_iflags) ? NULL : gh; | |
563 | } | |
564 | ||
6802e340 SW |
565 | /** |
566 | * state_change - record that the glock is now in a different state | |
567 | * @gl: the glock | |
c551f66c | 568 | * @new_state: the new state |
6802e340 SW |
569 | */ |
570 | ||
571 | static void state_change(struct gfs2_glock *gl, unsigned int new_state) | |
572 | { | |
573 | int held1, held2; | |
574 | ||
575 | held1 = (gl->gl_state != LM_ST_UNLOCKED); | |
576 | held2 = (new_state != LM_ST_UNLOCKED); | |
577 | ||
578 | if (held1 != held2) { | |
e66cf161 | 579 | GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref)); |
6802e340 | 580 | if (held2) |
e66cf161 | 581 | gl->gl_lockref.count++; |
6802e340 | 582 | else |
e66cf161 | 583 | gl->gl_lockref.count--; |
6802e340 | 584 | } |
7cf8dcd3 BP |
585 | if (new_state != gl->gl_target) |
586 | /* shorten our minimum hold time */ | |
587 | gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR, | |
588 | GL_GLOCK_MIN_HOLD); | |
6802e340 SW |
589 | gl->gl_state = new_state; |
590 | gl->gl_tchange = jiffies; | |
591 | } | |
592 | ||
35b6f8fb AG |
593 | static void gfs2_set_demote(struct gfs2_glock *gl) |
594 | { | |
595 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
596 | ||
597 | set_bit(GLF_DEMOTE, &gl->gl_flags); | |
598 | smp_mb(); | |
599 | wake_up(&sdp->sd_async_glock_wait); | |
600 | } | |
601 | ||
6802e340 SW |
602 | static void gfs2_demote_wake(struct gfs2_glock *gl) |
603 | { | |
604 | gl->gl_demote_state = LM_ST_EXCLUSIVE; | |
605 | clear_bit(GLF_DEMOTE, &gl->gl_flags); | |
4e857c58 | 606 | smp_mb__after_atomic(); |
6802e340 SW |
607 | wake_up_bit(&gl->gl_flags, GLF_DEMOTE); |
608 | } | |
609 | ||
610 | /** | |
611 | * finish_xmote - The DLM has replied to one of our lock requests | |
612 | * @gl: The glock | |
613 | * @ret: The status from the DLM | |
614 | * | |
615 | */ | |
616 | ||
617 | static void finish_xmote(struct gfs2_glock *gl, unsigned int ret) | |
618 | { | |
619 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
620 | struct gfs2_holder *gh; | |
621 | unsigned state = ret & LM_OUT_ST_MASK; | |
622 | ||
63997775 | 623 | trace_gfs2_glock_state_change(gl, state); |
6802e340 SW |
624 | state_change(gl, state); |
625 | gh = find_first_waiter(gl); | |
626 | ||
627 | /* Demote to UN request arrived during demote to SH or DF */ | |
628 | if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) && | |
629 | state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED) | |
630 | gl->gl_target = LM_ST_UNLOCKED; | |
631 | ||
632 | /* Check for state != intended state */ | |
633 | if (unlikely(state != gl->gl_target)) { | |
1fc05c8d AG |
634 | if (gh && (ret & LM_OUT_CANCELED)) |
635 | gfs2_holder_wake(gh); | |
6802e340 SW |
636 | if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) { |
637 | /* move to back of queue and try next entry */ | |
638 | if (ret & LM_OUT_CANCELED) { | |
0b93bac2 | 639 | list_move_tail(&gh->gh_list, &gl->gl_holders); |
6802e340 SW |
640 | gh = find_first_waiter(gl); |
641 | gl->gl_target = gh->gh_state; | |
66fa9912 BP |
642 | if (do_promote(gl)) |
643 | goto out; | |
6802e340 SW |
644 | goto retry; |
645 | } | |
646 | /* Some error or failed "try lock" - report it */ | |
647 | if ((ret & LM_OUT_ERROR) || | |
648 | (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) { | |
649 | gl->gl_target = gl->gl_state; | |
650 | do_error(gl, ret); | |
651 | goto out; | |
652 | } | |
653 | } | |
654 | switch(state) { | |
655 | /* Unlocked due to conversion deadlock, try again */ | |
656 | case LM_ST_UNLOCKED: | |
657 | retry: | |
658 | do_xmote(gl, gh, gl->gl_target); | |
659 | break; | |
660 | /* Conversion fails, unlock and try again */ | |
661 | case LM_ST_SHARED: | |
662 | case LM_ST_DEFERRED: | |
663 | do_xmote(gl, gh, LM_ST_UNLOCKED); | |
664 | break; | |
665 | default: /* Everything else */ | |
e54c78a2 BP |
666 | fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n", |
667 | gl->gl_target, state); | |
6802e340 SW |
668 | GLOCK_BUG_ON(gl, 1); |
669 | } | |
6802e340 SW |
670 | return; |
671 | } | |
672 | ||
673 | /* Fast path - we got what we asked for */ | |
674 | if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) | |
675 | gfs2_demote_wake(gl); | |
676 | if (state != LM_ST_UNLOCKED) { | |
677 | if (glops->go_xmote_bh) { | |
de3f906f AG |
678 | int rv; |
679 | ||
f3dd1649 | 680 | spin_unlock(&gl->gl_lockref.lock); |
f68effb3 | 681 | rv = glops->go_xmote_bh(gl); |
f3dd1649 | 682 | spin_lock(&gl->gl_lockref.lock); |
6802e340 SW |
683 | if (rv) { |
684 | do_error(gl, rv); | |
685 | goto out; | |
686 | } | |
687 | } | |
de3f906f | 688 | do_promote(gl); |
6802e340 SW |
689 | } |
690 | out: | |
691 | clear_bit(GLF_LOCK, &gl->gl_flags); | |
6802e340 SW |
692 | } |
693 | ||
865cc3e9 BP |
694 | static bool is_system_glock(struct gfs2_glock *gl) |
695 | { | |
696 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
697 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
698 | ||
699 | if (gl == m_ip->i_gl) | |
700 | return true; | |
701 | return false; | |
702 | } | |
703 | ||
6802e340 SW |
704 | /** |
705 | * do_xmote - Calls the DLM to change the state of a lock | |
706 | * @gl: The lock state | |
707 | * @gh: The holder (only for promotes) | |
708 | * @target: The target lock state | |
709 | * | |
710 | */ | |
711 | ||
86934198 BP |
712 | static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, |
713 | unsigned int target) | |
f3dd1649 AG |
714 | __releases(&gl->gl_lockref.lock) |
715 | __acquires(&gl->gl_lockref.lock) | |
6802e340 SW |
716 | { |
717 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
15562c43 | 718 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
9947a06d | 719 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
b58bf407 | 720 | unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0); |
6802e340 SW |
721 | int ret; |
722 | ||
601ef0d5 BP |
723 | if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) && |
724 | gh && !(gh->gh_flags & LM_FLAG_NOEXP)) | |
86934198 BP |
725 | goto skip_inval; |
726 | ||
0b93bac2 | 727 | lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP); |
921169ca SW |
728 | GLOCK_BUG_ON(gl, gl->gl_state == target); |
729 | GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target); | |
6802e340 SW |
730 | if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) && |
731 | glops->go_inval) { | |
d99724c3 BP |
732 | /* |
733 | * If another process is already doing the invalidate, let that | |
734 | * finish first. The glock state machine will get back to this | |
735 | * holder again later. | |
736 | */ | |
737 | if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS, | |
738 | &gl->gl_flags)) | |
739 | return; | |
6802e340 SW |
740 | do_error(gl, 0); /* Fail queued try locks */ |
741 | } | |
47a25380 | 742 | gl->gl_req = target; |
a245769f SW |
743 | set_bit(GLF_BLOCKING, &gl->gl_flags); |
744 | if ((gl->gl_req == LM_ST_UNLOCKED) || | |
745 | (gl->gl_state == LM_ST_EXCLUSIVE) || | |
746 | (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB))) | |
747 | clear_bit(GLF_BLOCKING, &gl->gl_flags); | |
9947a06d AG |
748 | if (!glops->go_inval && !glops->go_sync) |
749 | goto skip_inval; | |
750 | ||
f3dd1649 | 751 | spin_unlock(&gl->gl_lockref.lock); |
1c634f94 BP |
752 | if (glops->go_sync) { |
753 | ret = glops->go_sync(gl); | |
754 | /* If we had a problem syncing (due to io errors or whatever, | |
755 | * we should not invalidate the metadata or tell dlm to | |
756 | * release the glock to other nodes. | |
757 | */ | |
758 | if (ret) { | |
759 | if (cmpxchg(&sdp->sd_log_error, 0, ret)) { | |
760 | fs_err(sdp, "Error %d syncing glock \n", ret); | |
761 | gfs2_dump_glock(NULL, gl, true); | |
762 | } | |
9947a06d | 763 | spin_lock(&gl->gl_lockref.lock); |
b11e1a84 | 764 | goto skip_inval; |
1c634f94 BP |
765 | } |
766 | } | |
33dbd1e4 BP |
767 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) { |
768 | /* | |
769 | * The call to go_sync should have cleared out the ail list. | |
770 | * If there are still items, we have a problem. We ought to | |
771 | * withdraw, but we can't because the withdraw code also uses | |
772 | * glocks. Warn about the error, dump the glock, then fall | |
773 | * through and wait for logd to do the withdraw for us. | |
774 | */ | |
775 | if ((atomic_read(&gl->gl_ail_count) != 0) && | |
776 | (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) { | |
ea4e61c7 BP |
777 | gfs2_glock_assert_warn(gl, |
778 | !atomic_read(&gl->gl_ail_count)); | |
33dbd1e4 BP |
779 | gfs2_dump_glock(NULL, gl, true); |
780 | } | |
6802e340 | 781 | glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA); |
33dbd1e4 BP |
782 | clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags); |
783 | } | |
9947a06d | 784 | spin_lock(&gl->gl_lockref.lock); |
6802e340 | 785 | |
b11e1a84 | 786 | skip_inval: |
9947a06d | 787 | gl->gl_lockref.count++; |
d93ae386 BP |
788 | /* |
789 | * Check for an error encountered since we called go_sync and go_inval. | |
790 | * If so, we can't withdraw from the glock code because the withdraw | |
791 | * code itself uses glocks (see function signal_our_withdraw) to | |
792 | * change the mount to read-only. Most importantly, we must not call | |
793 | * dlm to unlock the glock until the journal is in a known good state | |
794 | * (after journal replay) otherwise other nodes may use the object | |
795 | * (rgrp or dinode) and then later, journal replay will corrupt the | |
796 | * file system. The best we can do here is wait for the logd daemon | |
797 | * to see sd_log_error and withdraw, and in the meantime, requeue the | |
798 | * work for later. | |
799 | * | |
865cc3e9 BP |
800 | * We make a special exception for some system glocks, such as the |
801 | * system statfs inode glock, which needs to be granted before the | |
802 | * gfs2_quotad daemon can exit, and that exit needs to finish before | |
803 | * we can unmount the withdrawn file system. | |
804 | * | |
d93ae386 BP |
805 | * However, if we're just unlocking the lock (say, for unmount, when |
806 | * gfs2_gl_hash_clear calls clear_glock) and recovery is complete | |
807 | * then it's okay to tell dlm to unlock it. | |
808 | */ | |
4d927b03 | 809 | if (unlikely(sdp->sd_log_error) && !gfs2_withdrawing_or_withdrawn(sdp)) |
d93ae386 | 810 | gfs2_withdraw_delayed(sdp); |
865cc3e9 BP |
811 | if (glock_blocked_by_withdraw(gl) && |
812 | (target != LM_ST_UNLOCKED || | |
813 | test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) { | |
814 | if (!is_system_glock(gl)) { | |
86934198 BP |
815 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); /* sets demote */ |
816 | /* | |
817 | * Ordinarily, we would call dlm and its callback would call | |
818 | * finish_xmote, which would call state_change() to the new state. | |
819 | * Since we withdrew, we won't call dlm, so call state_change | |
820 | * manually, but to the UNLOCKED state we desire. | |
821 | */ | |
822 | state_change(gl, LM_ST_UNLOCKED); | |
823 | /* | |
824 | * We skip telling dlm to do the locking, so we won't get a | |
825 | * reply that would otherwise clear GLF_LOCK. So we clear it here. | |
826 | */ | |
827 | clear_bit(GLF_LOCK, &gl->gl_flags); | |
828 | clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags); | |
1e860444 | 829 | gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD); |
9947a06d | 830 | return; |
865cc3e9 BP |
831 | } else { |
832 | clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags); | |
d93ae386 BP |
833 | } |
834 | } | |
835 | ||
9947a06d AG |
836 | if (ls->ls_ops->lm_lock) { |
837 | spin_unlock(&gl->gl_lockref.lock); | |
838 | ret = ls->ls_ops->lm_lock(gl, target, lck_flags); | |
839 | spin_lock(&gl->gl_lockref.lock); | |
5d923111 | 840 | |
3e11e530 BM |
841 | if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED && |
842 | target == LM_ST_UNLOCKED && | |
5d923111 | 843 | test_bit(DFL_UNMOUNT, &ls->ls_recover_flags)) { |
9947a06d AG |
844 | /* |
845 | * The lockspace has been released and the lock has | |
846 | * been unlocked implicitly. | |
847 | */ | |
d93ae386 | 848 | } else if (ret) { |
e54c78a2 | 849 | fs_err(sdp, "lm_lock ret %d\n", ret); |
9947a06d AG |
850 | target = gl->gl_state | LM_OUT_ERROR; |
851 | } else { | |
852 | /* The operation will be completed asynchronously. */ | |
853 | return; | |
dba2d70c | 854 | } |
6802e340 | 855 | } |
9947a06d AG |
856 | |
857 | /* Complete the operation now. */ | |
858 | finish_xmote(gl, target); | |
1e860444 | 859 | gfs2_glock_queue_work(gl, 0); |
6802e340 SW |
860 | } |
861 | ||
6802e340 SW |
862 | /** |
863 | * run_queue - do all outstanding tasks related to a glock | |
864 | * @gl: The glock in question | |
865 | * @nonblock: True if we must not block in run_queue | |
866 | * | |
867 | */ | |
868 | ||
869 | static void run_queue(struct gfs2_glock *gl, const int nonblock) | |
f3dd1649 AG |
870 | __releases(&gl->gl_lockref.lock) |
871 | __acquires(&gl->gl_lockref.lock) | |
6802e340 SW |
872 | { |
873 | struct gfs2_holder *gh = NULL; | |
874 | ||
262ee3a0 | 875 | if (test_bit(GLF_LOCK, &gl->gl_flags)) |
6802e340 | 876 | return; |
262ee3a0 | 877 | set_bit(GLF_LOCK, &gl->gl_flags); |
6802e340 SW |
878 | |
879 | GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)); | |
880 | ||
881 | if (test_bit(GLF_DEMOTE, &gl->gl_flags) && | |
882 | gl->gl_demote_state != gl->gl_state) { | |
883 | if (find_first_holder(gl)) | |
d8348de0 | 884 | goto out_unlock; |
6802e340 SW |
885 | if (nonblock) |
886 | goto out_sched; | |
887 | set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags); | |
265d529c | 888 | GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE); |
6802e340 SW |
889 | gl->gl_target = gl->gl_demote_state; |
890 | } else { | |
891 | if (test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
892 | gfs2_demote_wake(gl); | |
de3e7f97 | 893 | if (do_promote(gl)) |
d8348de0 | 894 | goto out_unlock; |
6802e340 SW |
895 | gh = find_first_waiter(gl); |
896 | gl->gl_target = gh->gh_state; | |
897 | if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) | |
898 | do_error(gl, 0); /* Fail queued try locks */ | |
899 | } | |
900 | do_xmote(gl, gh, gl->gl_target); | |
901 | return; | |
902 | ||
903 | out_sched: | |
7e71c55e | 904 | clear_bit(GLF_LOCK, &gl->gl_flags); |
4e857c58 | 905 | smp_mb__after_atomic(); |
e66cf161 | 906 | gl->gl_lockref.count++; |
1e860444 | 907 | gfs2_glock_queue_work(gl, 0); |
7e71c55e SW |
908 | return; |
909 | ||
d8348de0 | 910 | out_unlock: |
6802e340 | 911 | clear_bit(GLF_LOCK, &gl->gl_flags); |
4e857c58 | 912 | smp_mb__after_atomic(); |
7e71c55e | 913 | return; |
6802e340 SW |
914 | } |
915 | ||
3781ec9e AG |
916 | /** |
917 | * glock_set_object - set the gl_object field of a glock | |
918 | * @gl: the glock | |
919 | * @object: the object | |
920 | */ | |
921 | void glock_set_object(struct gfs2_glock *gl, void *object) | |
922 | { | |
923 | void *prev_object; | |
924 | ||
925 | spin_lock(&gl->gl_lockref.lock); | |
926 | prev_object = gl->gl_object; | |
927 | gl->gl_object = object; | |
928 | spin_unlock(&gl->gl_lockref.lock); | |
929 | if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == NULL)) { | |
930 | pr_warn("glock=%u/%llx\n", | |
931 | gl->gl_name.ln_type, | |
932 | (unsigned long long)gl->gl_name.ln_number); | |
933 | gfs2_dump_glock(NULL, gl, true); | |
934 | } | |
935 | } | |
936 | ||
937 | /** | |
938 | * glock_clear_object - clear the gl_object field of a glock | |
939 | * @gl: the glock | |
9ffa1888 | 940 | * @object: object the glock currently points at |
3781ec9e AG |
941 | */ |
942 | void glock_clear_object(struct gfs2_glock *gl, void *object) | |
943 | { | |
944 | void *prev_object; | |
945 | ||
946 | spin_lock(&gl->gl_lockref.lock); | |
947 | prev_object = gl->gl_object; | |
948 | gl->gl_object = NULL; | |
949 | spin_unlock(&gl->gl_lockref.lock); | |
9ffa1888 | 950 | if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == object)) { |
3781ec9e AG |
951 | pr_warn("glock=%u/%llx\n", |
952 | gl->gl_name.ln_type, | |
953 | (unsigned long long)gl->gl_name.ln_number); | |
954 | gfs2_dump_glock(NULL, gl, true); | |
955 | } | |
956 | } | |
957 | ||
f286d627 AG |
958 | void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation) |
959 | { | |
960 | struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr; | |
961 | ||
962 | if (ri->ri_magic == 0) | |
963 | ri->ri_magic = cpu_to_be32(GFS2_MAGIC); | |
964 | if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC)) | |
965 | ri->ri_generation_deleted = cpu_to_be64(generation); | |
966 | } | |
967 | ||
968 | bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation) | |
969 | { | |
970 | struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr; | |
971 | ||
972 | if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC)) | |
973 | return false; | |
974 | return generation <= be64_to_cpu(ri->ri_generation_deleted); | |
975 | } | |
976 | ||
9e8990de AG |
977 | static void gfs2_glock_poke(struct gfs2_glock *gl) |
978 | { | |
979 | int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP; | |
980 | struct gfs2_holder gh; | |
981 | int error; | |
982 | ||
b016d9a8 | 983 | __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_); |
c07bfb4d | 984 | error = gfs2_glock_nq(&gh); |
9e8990de AG |
985 | if (!error) |
986 | gfs2_glock_dq(&gh); | |
c07bfb4d | 987 | gfs2_holder_uninit(&gh); |
9e8990de AG |
988 | } |
989 | ||
8c7b9262 AG |
990 | static bool gfs2_try_evict(struct gfs2_glock *gl) |
991 | { | |
992 | struct gfs2_inode *ip; | |
993 | bool evicted = false; | |
994 | ||
995 | /* | |
996 | * If there is contention on the iopen glock and we have an inode, try | |
997 | * to grab and release the inode so that it can be evicted. This will | |
998 | * allow the remote node to go ahead and delete the inode without us | |
999 | * having to do it, which will avoid rgrp glock thrashing. | |
1000 | * | |
1001 | * The remote node is likely still holding the corresponding inode | |
1002 | * glock, so it will run before we get to verify that the delete has | |
1003 | * happened below. | |
1004 | */ | |
1005 | spin_lock(&gl->gl_lockref.lock); | |
1006 | ip = gl->gl_object; | |
1007 | if (ip && !igrab(&ip->i_inode)) | |
1008 | ip = NULL; | |
1009 | spin_unlock(&gl->gl_lockref.lock); | |
1010 | if (ip) { | |
b0dcffd8 | 1011 | gl->gl_no_formal_ino = ip->i_no_formal_ino; |
8c7b9262 AG |
1012 | set_bit(GIF_DEFERRED_DELETE, &ip->i_flags); |
1013 | d_prune_aliases(&ip->i_inode); | |
1014 | iput(&ip->i_inode); | |
1015 | ||
1016 | /* If the inode was evicted, gl->gl_object will now be NULL. */ | |
1017 | spin_lock(&gl->gl_lockref.lock); | |
1018 | ip = gl->gl_object; | |
9e8990de | 1019 | if (ip) { |
8c7b9262 | 1020 | clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags); |
f0c0ade8 AG |
1021 | if (!igrab(&ip->i_inode)) |
1022 | ip = NULL; | |
9e8990de | 1023 | } |
8c7b9262 | 1024 | spin_unlock(&gl->gl_lockref.lock); |
f0c0ade8 AG |
1025 | if (ip) { |
1026 | gfs2_glock_poke(ip->i_gl); | |
1027 | iput(&ip->i_inode); | |
9e8990de | 1028 | } |
8c7b9262 AG |
1029 | evicted = !ip; |
1030 | } | |
1031 | return evicted; | |
1032 | } | |
1033 | ||
f0e56edc AG |
1034 | bool gfs2_queue_try_to_evict(struct gfs2_glock *gl) |
1035 | { | |
1036 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
1037 | ||
1038 | if (test_and_set_bit(GLF_TRY_TO_EVICT, &gl->gl_flags)) | |
1039 | return false; | |
1040 | return queue_delayed_work(sdp->sd_delete_wq, | |
1041 | &gl->gl_delete, 0); | |
1042 | } | |
1043 | ||
1044 | static bool gfs2_queue_verify_evict(struct gfs2_glock *gl) | |
1045 | { | |
1046 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
1047 | ||
1048 | if (test_and_set_bit(GLF_VERIFY_EVICT, &gl->gl_flags)) | |
1049 | return false; | |
1050 | return queue_delayed_work(sdp->sd_delete_wq, | |
1051 | &gl->gl_delete, 5 * HZ); | |
1052 | } | |
1053 | ||
b94a170e BM |
1054 | static void delete_work_func(struct work_struct *work) |
1055 | { | |
a0e3cc65 AG |
1056 | struct delayed_work *dwork = to_delayed_work(work); |
1057 | struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete); | |
15562c43 | 1058 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
ec5ec66b | 1059 | struct inode *inode; |
044b9414 SW |
1060 | u64 no_addr = gl->gl_name.ln_number; |
1061 | ||
f0e56edc | 1062 | if (test_and_clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags)) { |
8c7b9262 AG |
1063 | /* |
1064 | * If we can evict the inode, give the remote node trying to | |
1065 | * delete the inode some time before verifying that the delete | |
1066 | * has happened. Otherwise, if we cause contention on the inode glock | |
1067 | * immediately, the remote node will think that we still have | |
1068 | * the inode in use, and so it will give up waiting. | |
9e8990de AG |
1069 | * |
1070 | * If we can't evict the inode, signal to the remote node that | |
1071 | * the inode is still in use. We'll later try to delete the | |
1072 | * inode locally in gfs2_evict_inode. | |
1073 | * | |
1074 | * FIXME: We only need to verify that the remote node has | |
1075 | * deleted the inode because nodes before this remote delete | |
1076 | * rework won't cooperate. At a later time, when we no longer | |
1077 | * care about compatibility with such nodes, we can skip this | |
1078 | * step entirely. | |
8c7b9262 AG |
1079 | */ |
1080 | if (gfs2_try_evict(gl)) { | |
e7beb8b6 | 1081 | if (test_bit(SDF_KILL, &sdp->sd_flags)) |
6c0246a9 | 1082 | goto out; |
f0e56edc | 1083 | if (gfs2_queue_verify_evict(gl)) |
8c7b9262 | 1084 | return; |
8c7b9262 | 1085 | } |
88f4a9f8 | 1086 | goto out; |
8c7b9262 AG |
1087 | } |
1088 | ||
f0e56edc AG |
1089 | if (test_and_clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags)) { |
1090 | inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino, | |
1091 | GFS2_BLKST_UNLINKED); | |
1092 | if (IS_ERR(inode)) { | |
1093 | if (PTR_ERR(inode) == -EAGAIN && | |
e7beb8b6 | 1094 | !test_bit(SDF_KILL, &sdp->sd_flags) && |
f0e56edc | 1095 | gfs2_queue_verify_evict(gl)) |
c412a97c | 1096 | return; |
f0e56edc AG |
1097 | } else { |
1098 | d_prune_aliases(inode); | |
1099 | iput(inode); | |
1100 | } | |
b94a170e | 1101 | } |
f0e56edc | 1102 | |
88f4a9f8 | 1103 | out: |
b94a170e BM |
1104 | gfs2_glock_put(gl); |
1105 | } | |
1106 | ||
c4f68a13 BM |
1107 | static void glock_work_func(struct work_struct *work) |
1108 | { | |
6802e340 | 1109 | unsigned long delay = 0; |
c4f68a13 | 1110 | struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work); |
6b0c7440 | 1111 | unsigned int drop_refs = 1; |
c4f68a13 | 1112 | |
1cd28e15 AG |
1113 | spin_lock(&gl->gl_lockref.lock); |
1114 | if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags)) { | |
1115 | clear_bit(GLF_REPLY_PENDING, &gl->gl_flags); | |
6802e340 | 1116 | finish_xmote(gl, gl->gl_reply); |
6b0c7440 | 1117 | drop_refs++; |
26bb7505 | 1118 | } |
f90e5b5b | 1119 | if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) && |
265d529c SW |
1120 | gl->gl_state != LM_ST_UNLOCKED && |
1121 | gl->gl_demote_state != LM_ST_EXCLUSIVE) { | |
6802e340 | 1122 | unsigned long holdtime, now = jiffies; |
f90e5b5b | 1123 | |
7cf8dcd3 | 1124 | holdtime = gl->gl_tchange + gl->gl_hold_time; |
6802e340 SW |
1125 | if (time_before(now, holdtime)) |
1126 | delay = holdtime - now; | |
f90e5b5b BP |
1127 | |
1128 | if (!delay) { | |
1129 | clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags); | |
35b6f8fb | 1130 | gfs2_set_demote(gl); |
f90e5b5b | 1131 | } |
6802e340 SW |
1132 | } |
1133 | run_queue(gl, 0); | |
6b0c7440 AG |
1134 | if (delay) { |
1135 | /* Keep one glock reference for the work we requeue. */ | |
1136 | drop_refs--; | |
7cf8dcd3 BP |
1137 | if (gl->gl_name.ln_type != LM_TYPE_INODE) |
1138 | delay = 0; | |
1e860444 | 1139 | gfs2_glock_queue_work(gl, delay); |
7cf8dcd3 | 1140 | } |
6b0c7440 AG |
1141 | |
1142 | /* | |
1143 | * Drop the remaining glock references manually here. (Mind that | |
1e860444 | 1144 | * gfs2_glock_queue_work depends on the lockref spinlock begin held |
6b0c7440 AG |
1145 | * here as well.) |
1146 | */ | |
1147 | gl->gl_lockref.count -= drop_refs; | |
1148 | if (!gl->gl_lockref.count) { | |
1149 | __gfs2_glock_put(gl); | |
1150 | return; | |
1151 | } | |
1152 | spin_unlock(&gl->gl_lockref.lock); | |
c4f68a13 BM |
1153 | } |
1154 | ||
0515480a AG |
1155 | static struct gfs2_glock *find_insert_glock(struct lm_lockname *name, |
1156 | struct gfs2_glock *new) | |
1157 | { | |
1158 | struct wait_glock_queue wait; | |
a91323e2 | 1159 | wait_queue_head_t *wq = glock_waitqueue(name); |
0515480a AG |
1160 | struct gfs2_glock *gl; |
1161 | ||
a91323e2 AG |
1162 | wait.name = name; |
1163 | init_wait(&wait.wait); | |
1164 | wait.wait.func = glock_wake_function; | |
1165 | ||
0515480a | 1166 | again: |
a91323e2 | 1167 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
0515480a AG |
1168 | rcu_read_lock(); |
1169 | if (new) { | |
1170 | gl = rhashtable_lookup_get_insert_fast(&gl_hash_table, | |
1171 | &new->gl_node, ht_parms); | |
1172 | if (IS_ERR(gl)) | |
1173 | goto out; | |
1174 | } else { | |
1175 | gl = rhashtable_lookup_fast(&gl_hash_table, | |
1176 | name, ht_parms); | |
1177 | } | |
1178 | if (gl && !lockref_get_not_dead(&gl->gl_lockref)) { | |
1179 | rcu_read_unlock(); | |
1180 | schedule(); | |
1181 | goto again; | |
1182 | } | |
1183 | out: | |
1184 | rcu_read_unlock(); | |
a91323e2 | 1185 | finish_wait(wq, &wait.wait); |
0515480a AG |
1186 | return gl; |
1187 | } | |
1188 | ||
b3b94faa DT |
1189 | /** |
1190 | * gfs2_glock_get() - Get a glock, or create one if one doesn't exist | |
1191 | * @sdp: The GFS2 superblock | |
1192 | * @number: the lock number | |
1193 | * @glops: The glock_operations to use | |
1194 | * @create: If 0, don't create the glock if it doesn't exist | |
1195 | * @glp: the glock is returned here | |
1196 | * | |
1197 | * This does not lock a glock, just finds/creates structures for one. | |
1198 | * | |
1199 | * Returns: errno | |
1200 | */ | |
1201 | ||
cd915493 | 1202 | int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number, |
8fb4b536 | 1203 | const struct gfs2_glock_operations *glops, int create, |
b3b94faa DT |
1204 | struct gfs2_glock **glp) |
1205 | { | |
009d8518 | 1206 | struct super_block *s = sdp->sd_vfs; |
15562c43 BP |
1207 | struct lm_lockname name = { .ln_number = number, |
1208 | .ln_type = glops->go_type, | |
1209 | .ln_sbd = sdp }; | |
0a52aba7 | 1210 | struct gfs2_glock *gl, *tmp; |
009d8518 | 1211 | struct address_space *mapping; |
0a52aba7 | 1212 | int ret = 0; |
b3b94faa | 1213 | |
0515480a AG |
1214 | gl = find_insert_glock(&name, NULL); |
1215 | if (gl) { | |
1216 | *glp = gl; | |
b3b94faa | 1217 | return 0; |
0515480a | 1218 | } |
64d576ba SW |
1219 | if (!create) |
1220 | return -ENOENT; | |
b3b94faa | 1221 | |
11d8b79e KC |
1222 | if (glops->go_flags & GLOF_ASPACE) { |
1223 | struct gfs2_glock_aspace *gla = | |
1224 | kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS); | |
1225 | if (!gla) | |
1226 | return -ENOMEM; | |
1227 | gl = &gla->glock; | |
1228 | } else { | |
1229 | gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS); | |
1230 | if (!gl) | |
1231 | return -ENOMEM; | |
1232 | } | |
dba2d70c | 1233 | memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb)); |
11d8b79e | 1234 | gl->gl_ops = glops; |
dba2d70c DT |
1235 | |
1236 | if (glops->go_flags & GLOF_LVB) { | |
f7be987b | 1237 | gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS); |
4e2f8849 | 1238 | if (!gl->gl_lksb.sb_lvbptr) { |
11d8b79e | 1239 | gfs2_glock_dealloc(&gl->gl_rcu); |
dba2d70c DT |
1240 | return -ENOMEM; |
1241 | } | |
dba2d70c DT |
1242 | } |
1243 | ||
8f05228e | 1244 | atomic_inc(&sdp->sd_glock_disposal); |
88ffbf3e | 1245 | gl->gl_node.next = NULL; |
f2e70d8f | 1246 | gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0; |
b3b94faa | 1247 | gl->gl_name = name; |
515b269d | 1248 | lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass); |
e66cf161 | 1249 | gl->gl_lockref.count = 1; |
b3b94faa | 1250 | gl->gl_state = LM_ST_UNLOCKED; |
6802e340 | 1251 | gl->gl_target = LM_ST_UNLOCKED; |
c4f68a13 | 1252 | gl->gl_demote_state = LM_ST_EXCLUSIVE; |
8b0e1953 | 1253 | gl->gl_dstamp = 0; |
a245769f SW |
1254 | preempt_disable(); |
1255 | /* We use the global stats to estimate the initial per-glock stats */ | |
1256 | gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type]; | |
1257 | preempt_enable(); | |
1258 | gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0; | |
1259 | gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0; | |
c4f68a13 | 1260 | gl->gl_tchange = jiffies; |
ec45d9f5 | 1261 | gl->gl_object = NULL; |
7cf8dcd3 | 1262 | gl->gl_hold_time = GL_GLOCK_DFT_HOLD; |
c4f68a13 | 1263 | INIT_DELAYED_WORK(&gl->gl_work, glock_work_func); |
2ffed529 BP |
1264 | if (gl->gl_name.ln_type == LM_TYPE_IOPEN) |
1265 | INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func); | |
b3b94faa | 1266 | |
009d8518 SW |
1267 | mapping = gfs2_glock2aspace(gl); |
1268 | if (mapping) { | |
1269 | mapping->a_ops = &gfs2_meta_aops; | |
1270 | mapping->host = s->s_bdev->bd_inode; | |
1271 | mapping->flags = 0; | |
1272 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
600f111e | 1273 | mapping->i_private_data = NULL; |
009d8518 | 1274 | mapping->writeback_index = 0; |
b3b94faa DT |
1275 | } |
1276 | ||
0515480a | 1277 | tmp = find_insert_glock(&name, gl); |
0a52aba7 | 1278 | if (!tmp) { |
88ffbf3e | 1279 | *glp = gl; |
0a52aba7 | 1280 | goto out; |
b3b94faa | 1281 | } |
0a52aba7 AG |
1282 | if (IS_ERR(tmp)) { |
1283 | ret = PTR_ERR(tmp); | |
1284 | goto out_free; | |
1285 | } | |
0515480a | 1286 | *glp = tmp; |
0a52aba7 AG |
1287 | |
1288 | out_free: | |
11d8b79e | 1289 | gfs2_glock_dealloc(&gl->gl_rcu); |
da7d554f | 1290 | if (atomic_dec_and_test(&sdp->sd_glock_disposal)) |
3c69c437 | 1291 | wake_up(&sdp->sd_kill_wait); |
b3b94faa | 1292 | |
0a52aba7 | 1293 | out: |
88ffbf3e | 1294 | return ret; |
b3b94faa DT |
1295 | } |
1296 | ||
1297 | /** | |
ffd0cd3c | 1298 | * __gfs2_holder_init - initialize a struct gfs2_holder in the default way |
b3b94faa DT |
1299 | * @gl: the glock |
1300 | * @state: the state we're requesting | |
1301 | * @flags: the modifier flags | |
1302 | * @gh: the holder structure | |
1303 | * | |
1304 | */ | |
1305 | ||
b016d9a8 AG |
1306 | void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags, |
1307 | struct gfs2_holder *gh, unsigned long ip) | |
b3b94faa DT |
1308 | { |
1309 | INIT_LIST_HEAD(&gh->gh_list); | |
4ad02083 | 1310 | gh->gh_gl = gfs2_glock_hold(gl); |
b016d9a8 | 1311 | gh->gh_ip = ip; |
b1e058da | 1312 | gh->gh_owner_pid = get_pid(task_pid(current)); |
b3b94faa DT |
1313 | gh->gh_state = state; |
1314 | gh->gh_flags = flags; | |
b3b94faa | 1315 | gh->gh_iflags = 0; |
b3b94faa DT |
1316 | } |
1317 | ||
1318 | /** | |
1319 | * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it | |
1320 | * @state: the state we're requesting | |
1321 | * @flags: the modifier flags | |
1322 | * @gh: the holder structure | |
1323 | * | |
1324 | * Don't mess with the glock. | |
1325 | * | |
1326 | */ | |
1327 | ||
b58bf407 | 1328 | void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh) |
b3b94faa DT |
1329 | { |
1330 | gh->gh_state = state; | |
579b78a4 | 1331 | gh->gh_flags = flags; |
3b8249f6 | 1332 | gh->gh_iflags = 0; |
d29c0afe | 1333 | gh->gh_ip = _RET_IP_; |
30badc95 | 1334 | put_pid(gh->gh_owner_pid); |
1a0eae88 | 1335 | gh->gh_owner_pid = get_pid(task_pid(current)); |
b3b94faa DT |
1336 | } |
1337 | ||
1338 | /** | |
1339 | * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference) | |
1340 | * @gh: the holder structure | |
1341 | * | |
1342 | */ | |
1343 | ||
1344 | void gfs2_holder_uninit(struct gfs2_holder *gh) | |
1345 | { | |
b1e058da | 1346 | put_pid(gh->gh_owner_pid); |
b3b94faa | 1347 | gfs2_glock_put(gh->gh_gl); |
6df9f9a2 | 1348 | gfs2_holder_mark_uninitialized(gh); |
d0dc80db | 1349 | gh->gh_ip = 0; |
b3b94faa DT |
1350 | } |
1351 | ||
01123cf1 AG |
1352 | static void gfs2_glock_update_hold_time(struct gfs2_glock *gl, |
1353 | unsigned long start_time) | |
1354 | { | |
1355 | /* Have we waited longer that a second? */ | |
1356 | if (time_after(jiffies, start_time + HZ)) { | |
1357 | /* Lengthen the minimum hold time. */ | |
1358 | gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR, | |
1359 | GL_GLOCK_MAX_HOLD); | |
1360 | } | |
1361 | } | |
1362 | ||
53d69132 AG |
1363 | /** |
1364 | * gfs2_glock_holder_ready - holder is ready and its error code can be collected | |
1365 | * @gh: the glock holder | |
1366 | * | |
1367 | * Called when a glock holder no longer needs to be waited for because it is | |
1368 | * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has | |
1369 | * failed (gh_error != 0). | |
1370 | */ | |
1371 | ||
1372 | int gfs2_glock_holder_ready(struct gfs2_holder *gh) | |
1373 | { | |
1374 | if (gh->gh_error || (gh->gh_flags & GL_SKIP)) | |
1375 | return gh->gh_error; | |
1376 | gh->gh_error = gfs2_instantiate(gh); | |
1377 | if (gh->gh_error) | |
1378 | gfs2_glock_dq(gh); | |
1379 | return gh->gh_error; | |
1380 | } | |
1381 | ||
07a79049 BP |
1382 | /** |
1383 | * gfs2_glock_wait - wait on a glock acquisition | |
1384 | * @gh: the glock holder | |
1385 | * | |
1386 | * Returns: 0 on success | |
1387 | */ | |
1388 | ||
1389 | int gfs2_glock_wait(struct gfs2_holder *gh) | |
da755fdb | 1390 | { |
01123cf1 | 1391 | unsigned long start_time = jiffies; |
7cf8dcd3 | 1392 | |
6802e340 | 1393 | might_sleep(); |
74316201 | 1394 | wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE); |
01123cf1 | 1395 | gfs2_glock_update_hold_time(gh->gh_gl, start_time); |
53d69132 | 1396 | return gfs2_glock_holder_ready(gh); |
da755fdb SW |
1397 | } |
1398 | ||
ad26967b BP |
1399 | static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs) |
1400 | { | |
1401 | int i; | |
1402 | ||
1403 | for (i = 0; i < num_gh; i++) | |
1404 | if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) | |
1405 | return 1; | |
1406 | return 0; | |
1407 | } | |
1408 | ||
1409 | /** | |
1410 | * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions | |
1411 | * @num_gh: the number of holders in the array | |
1412 | * @ghs: the glock holder array | |
1413 | * | |
1414 | * Returns: 0 on success, meaning all glocks have been granted and are held. | |
1415 | * -ESTALE if the request timed out, meaning all glocks were released, | |
1416 | * and the caller should retry the operation. | |
1417 | */ | |
1418 | ||
1419 | int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs) | |
1420 | { | |
1421 | struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd; | |
1422 | int i, ret = 0, timeout = 0; | |
1423 | unsigned long start_time = jiffies; | |
ad26967b BP |
1424 | |
1425 | might_sleep(); | |
1426 | /* | |
1427 | * Total up the (minimum hold time * 2) of all glocks and use that to | |
1428 | * determine the max amount of time we should wait. | |
1429 | */ | |
1430 | for (i = 0; i < num_gh; i++) | |
1431 | timeout += ghs[i].gh_gl->gl_hold_time << 1; | |
1432 | ||
ad26967b | 1433 | if (!wait_event_timeout(sdp->sd_async_glock_wait, |
bdff777c | 1434 | !glocks_pending(num_gh, ghs), timeout)) { |
ad26967b | 1435 | ret = -ESTALE; /* request timed out. */ |
bdff777c AG |
1436 | goto out; |
1437 | } | |
ad26967b | 1438 | |
ad26967b | 1439 | for (i = 0; i < num_gh; i++) { |
bdff777c | 1440 | struct gfs2_holder *gh = &ghs[i]; |
53d69132 | 1441 | int ret2; |
ad26967b | 1442 | |
bdff777c AG |
1443 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) { |
1444 | gfs2_glock_update_hold_time(gh->gh_gl, | |
1445 | start_time); | |
ad26967b | 1446 | } |
53d69132 | 1447 | ret2 = gfs2_glock_holder_ready(gh); |
ad26967b | 1448 | if (!ret) |
53d69132 | 1449 | ret = ret2; |
ad26967b BP |
1450 | } |
1451 | ||
bdff777c AG |
1452 | out: |
1453 | if (ret) { | |
1454 | for (i = 0; i < num_gh; i++) { | |
1455 | struct gfs2_holder *gh = &ghs[i]; | |
ad26967b | 1456 | |
bdff777c AG |
1457 | gfs2_glock_dq(gh); |
1458 | } | |
1459 | } | |
ad26967b BP |
1460 | return ret; |
1461 | } | |
1462 | ||
b3b94faa | 1463 | /** |
6802e340 SW |
1464 | * handle_callback - process a demote request |
1465 | * @gl: the glock | |
1466 | * @state: the state the caller wants us to change to | |
c551f66c LJ |
1467 | * @delay: zero to demote immediately; otherwise pending demote |
1468 | * @remote: true if this came from a different cluster node | |
b3b94faa | 1469 | * |
6802e340 SW |
1470 | * There are only two requests that we are going to see in actual |
1471 | * practise: LM_ST_SHARED and LM_ST_UNLOCKED | |
b3b94faa DT |
1472 | */ |
1473 | ||
6802e340 | 1474 | static void handle_callback(struct gfs2_glock *gl, unsigned int state, |
81ffbf65 | 1475 | unsigned long delay, bool remote) |
b3b94faa | 1476 | { |
35b6f8fb AG |
1477 | if (delay) |
1478 | set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags); | |
1479 | else | |
1480 | gfs2_set_demote(gl); | |
6802e340 SW |
1481 | if (gl->gl_demote_state == LM_ST_EXCLUSIVE) { |
1482 | gl->gl_demote_state = state; | |
1483 | gl->gl_demote_time = jiffies; | |
6802e340 SW |
1484 | } else if (gl->gl_demote_state != LM_ST_UNLOCKED && |
1485 | gl->gl_demote_state != state) { | |
1486 | gl->gl_demote_state = LM_ST_UNLOCKED; | |
b3b94faa | 1487 | } |
b94a170e | 1488 | if (gl->gl_ops->go_callback) |
81ffbf65 | 1489 | gl->gl_ops->go_callback(gl, remote); |
7bd8b2eb | 1490 | trace_gfs2_demote_rq(gl, remote); |
b3b94faa DT |
1491 | } |
1492 | ||
6802e340 | 1493 | void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...) |
7c52b166 | 1494 | { |
5e69069c | 1495 | struct va_format vaf; |
7c52b166 RP |
1496 | va_list args; |
1497 | ||
1498 | va_start(args, fmt); | |
5e69069c | 1499 | |
6802e340 | 1500 | if (seq) { |
1bb49303 | 1501 | seq_vprintf(seq, fmt, args); |
6802e340 | 1502 | } else { |
5e69069c JP |
1503 | vaf.fmt = fmt; |
1504 | vaf.va = &args; | |
1505 | ||
d77d1b58 | 1506 | pr_err("%pV", &vaf); |
6802e340 | 1507 | } |
5e69069c | 1508 | |
7c52b166 RP |
1509 | va_end(args); |
1510 | } | |
1511 | ||
cbe6d257 AG |
1512 | static inline bool pid_is_meaningful(const struct gfs2_holder *gh) |
1513 | { | |
1514 | if (!(gh->gh_flags & GL_NOPID)) | |
1515 | return true; | |
1516 | if (gh->gh_state == LM_ST_UNLOCKED) | |
1517 | return true; | |
1518 | return false; | |
1519 | } | |
1520 | ||
b3b94faa DT |
1521 | /** |
1522 | * add_to_queue - Add a holder to the wait queue (but look for recursion) | |
1523 | * @gh: the holder structure to add | |
1524 | * | |
6802e340 SW |
1525 | * Eventually we should move the recursive locking trap to a |
1526 | * debugging option or something like that. This is the fast | |
1527 | * path and needs to have the minimum number of distractions. | |
1528 | * | |
b3b94faa DT |
1529 | */ |
1530 | ||
6802e340 | 1531 | static inline void add_to_queue(struct gfs2_holder *gh) |
f3dd1649 AG |
1532 | __releases(&gl->gl_lockref.lock) |
1533 | __acquires(&gl->gl_lockref.lock) | |
b3b94faa DT |
1534 | { |
1535 | struct gfs2_glock *gl = gh->gh_gl; | |
15562c43 | 1536 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
6802e340 SW |
1537 | struct list_head *insert_pt = NULL; |
1538 | struct gfs2_holder *gh2; | |
e5dc76b9 | 1539 | int try_futile = 0; |
b3b94faa | 1540 | |
ad26967b | 1541 | GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL); |
fee852e3 | 1542 | if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags)) |
ad26967b | 1543 | GLOCK_BUG_ON(gl, true); |
190562bd | 1544 | |
6802e340 | 1545 | if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) { |
61444649 | 1546 | if (test_bit(GLF_LOCK, &gl->gl_flags)) { |
8f0028fc | 1547 | struct gfs2_holder *current_gh; |
61444649 | 1548 | |
ba3e77a4 | 1549 | current_gh = find_first_holder(gl); |
8f0028fc | 1550 | try_futile = !may_grant(gl, current_gh, gh); |
61444649 | 1551 | } |
6802e340 SW |
1552 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) |
1553 | goto fail; | |
1554 | } | |
1555 | ||
1556 | list_for_each_entry(gh2, &gl->gl_holders, gh_list) { | |
cbe6d257 AG |
1557 | if (likely(gh2->gh_owner_pid != gh->gh_owner_pid)) |
1558 | continue; | |
1559 | if (gh->gh_gl->gl_ops->go_type == LM_TYPE_FLOCK) | |
1560 | continue; | |
cbe6d257 AG |
1561 | if (!pid_is_meaningful(gh2)) |
1562 | continue; | |
1563 | goto trap_recursive; | |
1564 | } | |
1565 | list_for_each_entry(gh2, &gl->gl_holders, gh_list) { | |
e5dc76b9 BP |
1566 | if (try_futile && |
1567 | !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) { | |
6802e340 SW |
1568 | fail: |
1569 | gh->gh_error = GLR_TRYFAILED; | |
1570 | gfs2_holder_wake(gh); | |
1571 | return; | |
b4c20166 | 1572 | } |
6802e340 SW |
1573 | if (test_bit(HIF_HOLDER, &gh2->gh_iflags)) |
1574 | continue; | |
6802e340 | 1575 | } |
edae38a6 | 1576 | trace_gfs2_glock_queue(gh, 1); |
a245769f SW |
1577 | gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT); |
1578 | gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT); | |
6802e340 SW |
1579 | if (likely(insert_pt == NULL)) { |
1580 | list_add_tail(&gh->gh_list, &gl->gl_holders); | |
6802e340 SW |
1581 | return; |
1582 | } | |
1583 | list_add_tail(&gh->gh_list, insert_pt); | |
0b93bac2 AG |
1584 | spin_unlock(&gl->gl_lockref.lock); |
1585 | if (sdp->sd_lockstruct.ls_ops->lm_cancel) | |
1586 | sdp->sd_lockstruct.ls_ops->lm_cancel(gl); | |
1587 | spin_lock(&gl->gl_lockref.lock); | |
6802e340 | 1588 | return; |
b3b94faa | 1589 | |
6802e340 | 1590 | trap_recursive: |
e54c78a2 BP |
1591 | fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip); |
1592 | fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid)); | |
1593 | fs_err(sdp, "lock type: %d req lock state : %d\n", | |
6802e340 | 1594 | gh2->gh_gl->gl_name.ln_type, gh2->gh_state); |
e54c78a2 BP |
1595 | fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip); |
1596 | fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid)); | |
1597 | fs_err(sdp, "lock type: %d req lock state : %d\n", | |
6802e340 | 1598 | gh->gh_gl->gl_name.ln_type, gh->gh_state); |
3792ce97 | 1599 | gfs2_dump_glock(NULL, gl, true); |
6802e340 | 1600 | BUG(); |
b3b94faa DT |
1601 | } |
1602 | ||
1603 | /** | |
1604 | * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock) | |
1605 | * @gh: the holder structure | |
1606 | * | |
1607 | * if (gh->gh_flags & GL_ASYNC), this never returns an error | |
1608 | * | |
1609 | * Returns: 0, GLR_TRYFAILED, or errno on failure | |
1610 | */ | |
1611 | ||
1612 | int gfs2_glock_nq(struct gfs2_holder *gh) | |
1613 | { | |
1614 | struct gfs2_glock *gl = gh->gh_gl; | |
f9f229c1 | 1615 | int error; |
b3b94faa | 1616 | |
601ef0d5 | 1617 | if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP)) |
b3b94faa | 1618 | return -EIO; |
b3b94faa | 1619 | |
f9f229c1 AG |
1620 | if (gh->gh_flags & GL_NOBLOCK) { |
1621 | struct gfs2_holder *current_gh; | |
1622 | ||
1623 | error = -ECHILD; | |
1624 | spin_lock(&gl->gl_lockref.lock); | |
1625 | if (find_last_waiter(gl)) | |
1626 | goto unlock; | |
1627 | current_gh = find_first_holder(gl); | |
1628 | if (!may_grant(gl, current_gh, gh)) | |
1629 | goto unlock; | |
1630 | set_bit(HIF_HOLDER, &gh->gh_iflags); | |
1631 | list_add_tail(&gh->gh_list, &gl->gl_holders); | |
1632 | trace_gfs2_promote(gh); | |
1633 | error = 0; | |
1634 | unlock: | |
1635 | spin_unlock(&gl->gl_lockref.lock); | |
1636 | return error; | |
1637 | } | |
1638 | ||
f42ab085 SW |
1639 | if (test_bit(GLF_LRU, &gl->gl_flags)) |
1640 | gfs2_glock_remove_from_lru(gl); | |
1641 | ||
a4e8145e | 1642 | gh->gh_error = 0; |
f3dd1649 | 1643 | spin_lock(&gl->gl_lockref.lock); |
b3b94faa | 1644 | add_to_queue(gh); |
01b172b7 BP |
1645 | if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) && |
1646 | test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) { | |
0809f6ec | 1647 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
01b172b7 | 1648 | gl->gl_lockref.count++; |
1e860444 | 1649 | gfs2_glock_queue_work(gl, 0); |
01b172b7 | 1650 | } |
6802e340 | 1651 | run_queue(gl, 1); |
f3dd1649 | 1652 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa | 1653 | |
f9f229c1 | 1654 | error = 0; |
6802e340 SW |
1655 | if (!(gh->gh_flags & GL_ASYNC)) |
1656 | error = gfs2_glock_wait(gh); | |
b3b94faa | 1657 | |
b3b94faa DT |
1658 | return error; |
1659 | } | |
1660 | ||
1661 | /** | |
1662 | * gfs2_glock_poll - poll to see if an async request has been completed | |
1663 | * @gh: the holder | |
1664 | * | |
1665 | * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on | |
1666 | */ | |
1667 | ||
1668 | int gfs2_glock_poll(struct gfs2_holder *gh) | |
1669 | { | |
6802e340 | 1670 | return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1; |
b3b94faa DT |
1671 | } |
1672 | ||
dc732906 BP |
1673 | static inline bool needs_demote(struct gfs2_glock *gl) |
1674 | { | |
1675 | return (test_bit(GLF_DEMOTE, &gl->gl_flags) || | |
1676 | test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags)); | |
1677 | } | |
b3b94faa | 1678 | |
dc732906 | 1679 | static void __gfs2_glock_dq(struct gfs2_holder *gh) |
b3b94faa DT |
1680 | { |
1681 | struct gfs2_glock *gl = gh->gh_gl; | |
c4f68a13 | 1682 | unsigned delay = 0; |
6802e340 | 1683 | int fast_path = 0; |
b3b94faa | 1684 | |
ba3e77a4 AG |
1685 | /* |
1686 | * This holder should not be cached, so mark it for demote. | |
1687 | * Note: this should be done before the check for needs_demote | |
1688 | * below. | |
1689 | */ | |
1690 | if (gh->gh_flags & GL_NOCACHE) | |
1691 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); | |
dc732906 | 1692 | |
ba3e77a4 AG |
1693 | list_del_init(&gh->gh_list); |
1694 | clear_bit(HIF_HOLDER, &gh->gh_iflags); | |
1695 | trace_gfs2_glock_queue(gh, 0); | |
b3b94faa | 1696 | |
ba3e77a4 AG |
1697 | /* |
1698 | * If there hasn't been a demote request we are done. | |
1699 | * (Let the remaining holders, if any, keep holding it.) | |
1700 | */ | |
1701 | if (!needs_demote(gl)) { | |
1702 | if (list_empty(&gl->gl_holders)) | |
1703 | fast_path = 1; | |
dc732906 | 1704 | } |
08d73666 | 1705 | |
7881ef3f | 1706 | if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl)) |
4abb6ad9 BP |
1707 | gfs2_glock_add_to_lru(gl); |
1708 | ||
6b0c7440 AG |
1709 | if (unlikely(!fast_path)) { |
1710 | gl->gl_lockref.count++; | |
1711 | if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) && | |
1712 | !test_bit(GLF_DEMOTE, &gl->gl_flags) && | |
1713 | gl->gl_name.ln_type == LM_TYPE_INODE) | |
1714 | delay = gl->gl_hold_time; | |
1e860444 | 1715 | gfs2_glock_queue_work(gl, delay); |
6b0c7440 | 1716 | } |
dc732906 BP |
1717 | } |
1718 | ||
1719 | /** | |
1720 | * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock) | |
1721 | * @gh: the glock holder | |
1722 | * | |
1723 | */ | |
1724 | void gfs2_glock_dq(struct gfs2_holder *gh) | |
1725 | { | |
1726 | struct gfs2_glock *gl = gh->gh_gl; | |
6b46a061 | 1727 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
dc732906 BP |
1728 | |
1729 | spin_lock(&gl->gl_lockref.lock); | |
97236ad5 AG |
1730 | if (!gfs2_holder_queued(gh)) { |
1731 | /* | |
1732 | * May have already been dequeued because the locking request | |
1733 | * was GL_ASYNC and it has failed in the meantime. | |
1734 | */ | |
1735 | goto out; | |
1736 | } | |
6b46a061 | 1737 | |
1fc05c8d AG |
1738 | if (list_is_first(&gh->gh_list, &gl->gl_holders) && |
1739 | !test_bit(HIF_HOLDER, &gh->gh_iflags)) { | |
1740 | spin_unlock(&gl->gl_lockref.lock); | |
1741 | gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl); | |
1742 | wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE); | |
1743 | spin_lock(&gl->gl_lockref.lock); | |
1744 | } | |
1745 | ||
6b46a061 AG |
1746 | /* |
1747 | * If we're in the process of file system withdraw, we cannot just | |
1748 | * dequeue any glocks until our journal is recovered, lest we introduce | |
1749 | * file system corruption. We need two exceptions to this rule: We need | |
1750 | * to allow unlocking of nondisk glocks and the glock for our own | |
1751 | * journal that needs recovery. | |
1752 | */ | |
1753 | if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) && | |
1754 | glock_blocked_by_withdraw(gl) && | |
1755 | gh->gh_gl != sdp->sd_jinode_gl) { | |
1756 | sdp->sd_glock_dqs_held++; | |
1757 | spin_unlock(&gl->gl_lockref.lock); | |
1758 | might_sleep(); | |
1759 | wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY, | |
1760 | TASK_UNINTERRUPTIBLE); | |
1761 | spin_lock(&gl->gl_lockref.lock); | |
1762 | } | |
1763 | ||
dc732906 | 1764 | __gfs2_glock_dq(gh); |
97236ad5 | 1765 | out: |
f3dd1649 | 1766 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1767 | } |
1768 | ||
d93cfa98 AD |
1769 | void gfs2_glock_dq_wait(struct gfs2_holder *gh) |
1770 | { | |
1771 | struct gfs2_glock *gl = gh->gh_gl; | |
1772 | gfs2_glock_dq(gh); | |
81e1d450 | 1773 | might_sleep(); |
74316201 | 1774 | wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE); |
d93cfa98 AD |
1775 | } |
1776 | ||
b3b94faa DT |
1777 | /** |
1778 | * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it | |
1779 | * @gh: the holder structure | |
1780 | * | |
1781 | */ | |
1782 | ||
1783 | void gfs2_glock_dq_uninit(struct gfs2_holder *gh) | |
1784 | { | |
1785 | gfs2_glock_dq(gh); | |
1786 | gfs2_holder_uninit(gh); | |
1787 | } | |
1788 | ||
1789 | /** | |
1790 | * gfs2_glock_nq_num - acquire a glock based on lock number | |
1791 | * @sdp: the filesystem | |
1792 | * @number: the lock number | |
1793 | * @glops: the glock operations for the type of glock | |
1794 | * @state: the state to acquire the glock in | |
25985edc | 1795 | * @flags: modifier flags for the acquisition |
b3b94faa DT |
1796 | * @gh: the struct gfs2_holder |
1797 | * | |
1798 | * Returns: errno | |
1799 | */ | |
1800 | ||
cd915493 | 1801 | int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number, |
8fb4b536 | 1802 | const struct gfs2_glock_operations *glops, |
b58bf407 | 1803 | unsigned int state, u16 flags, struct gfs2_holder *gh) |
b3b94faa DT |
1804 | { |
1805 | struct gfs2_glock *gl; | |
1806 | int error; | |
1807 | ||
1808 | error = gfs2_glock_get(sdp, number, glops, CREATE, &gl); | |
1809 | if (!error) { | |
1810 | error = gfs2_glock_nq_init(gl, state, flags, gh); | |
1811 | gfs2_glock_put(gl); | |
1812 | } | |
1813 | ||
1814 | return error; | |
1815 | } | |
1816 | ||
1817 | /** | |
1818 | * glock_compare - Compare two struct gfs2_glock structures for sorting | |
1819 | * @arg_a: the first structure | |
1820 | * @arg_b: the second structure | |
1821 | * | |
1822 | */ | |
1823 | ||
1824 | static int glock_compare(const void *arg_a, const void *arg_b) | |
1825 | { | |
a5e08a9e SW |
1826 | const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a; |
1827 | const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b; | |
1828 | const struct lm_lockname *a = &gh_a->gh_gl->gl_name; | |
1829 | const struct lm_lockname *b = &gh_b->gh_gl->gl_name; | |
b3b94faa DT |
1830 | |
1831 | if (a->ln_number > b->ln_number) | |
a5e08a9e SW |
1832 | return 1; |
1833 | if (a->ln_number < b->ln_number) | |
1834 | return -1; | |
1c0f4872 | 1835 | BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type); |
a5e08a9e | 1836 | return 0; |
b3b94faa DT |
1837 | } |
1838 | ||
1839 | /** | |
44dab005 | 1840 | * nq_m_sync - synchronously acquire more than one glock in deadlock free order |
b3b94faa DT |
1841 | * @num_gh: the number of structures |
1842 | * @ghs: an array of struct gfs2_holder structures | |
c551f66c | 1843 | * @p: placeholder for the holder structure to pass back |
b3b94faa DT |
1844 | * |
1845 | * Returns: 0 on success (all glocks acquired), | |
1846 | * errno on failure (no glocks acquired) | |
1847 | */ | |
1848 | ||
1849 | static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs, | |
1850 | struct gfs2_holder **p) | |
1851 | { | |
1852 | unsigned int x; | |
1853 | int error = 0; | |
1854 | ||
1855 | for (x = 0; x < num_gh; x++) | |
1856 | p[x] = &ghs[x]; | |
1857 | ||
1858 | sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL); | |
1859 | ||
1860 | for (x = 0; x < num_gh; x++) { | |
b3b94faa DT |
1861 | error = gfs2_glock_nq(p[x]); |
1862 | if (error) { | |
1863 | while (x--) | |
1864 | gfs2_glock_dq(p[x]); | |
1865 | break; | |
1866 | } | |
1867 | } | |
1868 | ||
1869 | return error; | |
1870 | } | |
1871 | ||
1872 | /** | |
1873 | * gfs2_glock_nq_m - acquire multiple glocks | |
1874 | * @num_gh: the number of structures | |
1875 | * @ghs: an array of struct gfs2_holder structures | |
1876 | * | |
b3b94faa DT |
1877 | * Returns: 0 on success (all glocks acquired), |
1878 | * errno on failure (no glocks acquired) | |
1879 | */ | |
1880 | ||
1881 | int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs) | |
1882 | { | |
eaf5bd3c SW |
1883 | struct gfs2_holder *tmp[4]; |
1884 | struct gfs2_holder **pph = tmp; | |
b3b94faa DT |
1885 | int error = 0; |
1886 | ||
eaf5bd3c SW |
1887 | switch(num_gh) { |
1888 | case 0: | |
b3b94faa | 1889 | return 0; |
eaf5bd3c | 1890 | case 1: |
b3b94faa | 1891 | return gfs2_glock_nq(ghs); |
eaf5bd3c SW |
1892 | default: |
1893 | if (num_gh <= 4) | |
b3b94faa | 1894 | break; |
6da2ec56 KC |
1895 | pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *), |
1896 | GFP_NOFS); | |
eaf5bd3c SW |
1897 | if (!pph) |
1898 | return -ENOMEM; | |
b3b94faa DT |
1899 | } |
1900 | ||
eaf5bd3c | 1901 | error = nq_m_sync(num_gh, ghs, pph); |
b3b94faa | 1902 | |
eaf5bd3c SW |
1903 | if (pph != tmp) |
1904 | kfree(pph); | |
b3b94faa DT |
1905 | |
1906 | return error; | |
1907 | } | |
1908 | ||
1909 | /** | |
1910 | * gfs2_glock_dq_m - release multiple glocks | |
1911 | * @num_gh: the number of structures | |
1912 | * @ghs: an array of struct gfs2_holder structures | |
1913 | * | |
1914 | */ | |
1915 | ||
1916 | void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs) | |
1917 | { | |
fa1bbdea BP |
1918 | while (num_gh--) |
1919 | gfs2_glock_dq(&ghs[num_gh]); | |
b3b94faa DT |
1920 | } |
1921 | ||
f057f6cd | 1922 | void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state) |
da755fdb | 1923 | { |
c4f68a13 BM |
1924 | unsigned long delay = 0; |
1925 | unsigned long holdtime; | |
1926 | unsigned long now = jiffies; | |
b3b94faa | 1927 | |
f057f6cd | 1928 | gfs2_glock_hold(gl); |
e2c6c8a7 | 1929 | spin_lock(&gl->gl_lockref.lock); |
7cf8dcd3 | 1930 | holdtime = gl->gl_tchange + gl->gl_hold_time; |
e2c6c8a7 | 1931 | if (!list_empty(&gl->gl_holders) && |
7cf8dcd3 | 1932 | gl->gl_name.ln_type == LM_TYPE_INODE) { |
7b5e3d5f SW |
1933 | if (time_before(now, holdtime)) |
1934 | delay = holdtime - now; | |
1935 | if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags)) | |
7cf8dcd3 | 1936 | delay = gl->gl_hold_time; |
7b5e3d5f | 1937 | } |
81ffbf65 | 1938 | handle_callback(gl, state, delay, true); |
1e860444 | 1939 | gfs2_glock_queue_work(gl, delay); |
f3dd1649 | 1940 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1941 | } |
1942 | ||
0809f6ec SW |
1943 | /** |
1944 | * gfs2_should_freeze - Figure out if glock should be frozen | |
1945 | * @gl: The glock in question | |
1946 | * | |
1947 | * Glocks are not frozen if (a) the result of the dlm operation is | |
1948 | * an error, (b) the locking operation was an unlock operation or | |
1949 | * (c) if there is a "noexp" flagged request anywhere in the queue | |
1950 | * | |
1951 | * Returns: 1 if freezing should occur, 0 otherwise | |
1952 | */ | |
1953 | ||
1954 | static int gfs2_should_freeze(const struct gfs2_glock *gl) | |
1955 | { | |
1956 | const struct gfs2_holder *gh; | |
1957 | ||
1958 | if (gl->gl_reply & ~LM_OUT_ST_MASK) | |
1959 | return 0; | |
1960 | if (gl->gl_target == LM_ST_UNLOCKED) | |
1961 | return 0; | |
1962 | ||
1963 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { | |
1964 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
1965 | continue; | |
1966 | if (LM_FLAG_NOEXP & gh->gh_flags) | |
1967 | return 0; | |
1968 | } | |
1969 | ||
1970 | return 1; | |
1971 | } | |
1972 | ||
b3b94faa | 1973 | /** |
f057f6cd SW |
1974 | * gfs2_glock_complete - Callback used by locking |
1975 | * @gl: Pointer to the glock | |
1976 | * @ret: The return value from the dlm | |
b3b94faa | 1977 | * |
f3dd1649 | 1978 | * The gl_reply field is under the gl_lockref.lock lock so that it is ok |
47a25380 | 1979 | * to use a bitfield shared with other glock state fields. |
b3b94faa DT |
1980 | */ |
1981 | ||
f057f6cd | 1982 | void gfs2_glock_complete(struct gfs2_glock *gl, int ret) |
b3b94faa | 1983 | { |
15562c43 | 1984 | struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct; |
0809f6ec | 1985 | |
f3dd1649 | 1986 | spin_lock(&gl->gl_lockref.lock); |
f057f6cd | 1987 | gl->gl_reply = ret; |
0809f6ec | 1988 | |
e0c2a9aa | 1989 | if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) { |
0809f6ec | 1990 | if (gfs2_should_freeze(gl)) { |
f057f6cd | 1991 | set_bit(GLF_FROZEN, &gl->gl_flags); |
f3dd1649 | 1992 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa | 1993 | return; |
0809f6ec | 1994 | } |
b3b94faa | 1995 | } |
47a25380 | 1996 | |
e66cf161 | 1997 | gl->gl_lockref.count++; |
f057f6cd | 1998 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
1e860444 | 1999 | gfs2_glock_queue_work(gl, 0); |
f3dd1649 | 2000 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
2001 | } |
2002 | ||
4f0f586b ST |
2003 | static int glock_cmp(void *priv, const struct list_head *a, |
2004 | const struct list_head *b) | |
4506a519 SW |
2005 | { |
2006 | struct gfs2_glock *gla, *glb; | |
2007 | ||
2008 | gla = list_entry(a, struct gfs2_glock, gl_lru); | |
2009 | glb = list_entry(b, struct gfs2_glock, gl_lru); | |
2010 | ||
2011 | if (gla->gl_name.ln_number > glb->gl_name.ln_number) | |
2012 | return 1; | |
2013 | if (gla->gl_name.ln_number < glb->gl_name.ln_number) | |
2014 | return -1; | |
2015 | ||
2016 | return 0; | |
2017 | } | |
2018 | ||
acf1f42f AG |
2019 | static bool can_free_glock(struct gfs2_glock *gl) |
2020 | { | |
2021 | bool held = gl->gl_state != LM_ST_UNLOCKED; | |
2022 | ||
2023 | return !test_bit(GLF_LOCK, &gl->gl_flags) && | |
2024 | gl->gl_lockref.count == held; | |
2025 | } | |
2026 | ||
4506a519 SW |
2027 | /** |
2028 | * gfs2_dispose_glock_lru - Demote a list of glocks | |
2029 | * @list: The list to dispose of | |
2030 | * | |
2031 | * Disposing of glocks may involve disk accesses, so that here we sort | |
2032 | * the glocks by number (i.e. disk location of the inodes) so that if | |
2033 | * there are any such accesses, they'll be sent in order (mostly). | |
2034 | * | |
2035 | * Must be called under the lru_lock, but may drop and retake this | |
2036 | * lock. While the lru_lock is dropped, entries may vanish from the | |
2037 | * list, but no new entries will appear on the list (since it is | |
2038 | * private) | |
2039 | */ | |
2040 | ||
7a1ad9d8 | 2041 | static unsigned long gfs2_dispose_glock_lru(struct list_head *list) |
4506a519 SW |
2042 | __releases(&lru_lock) |
2043 | __acquires(&lru_lock) | |
2044 | { | |
2045 | struct gfs2_glock *gl; | |
7a1ad9d8 | 2046 | unsigned long freed = 0; |
4506a519 SW |
2047 | |
2048 | list_sort(NULL, list, glock_cmp); | |
2049 | ||
2050 | while(!list_empty(list)) { | |
969183bc | 2051 | gl = list_first_entry(list, struct gfs2_glock, gl_lru); |
f3dd1649 | 2052 | if (!spin_trylock(&gl->gl_lockref.lock)) { |
94a09a39 | 2053 | add_back_to_lru: |
7a1ad9d8 | 2054 | list_move(&gl->gl_lru, &lru_list); |
e66cf161 SW |
2055 | continue; |
2056 | } | |
acf1f42f | 2057 | if (!can_free_glock(gl)) { |
f3dd1649 | 2058 | spin_unlock(&gl->gl_lockref.lock); |
94a09a39 SW |
2059 | goto add_back_to_lru; |
2060 | } | |
7a1ad9d8 AG |
2061 | list_del_init(&gl->gl_lru); |
2062 | atomic_dec(&lru_count); | |
2063 | clear_bit(GLF_LRU, &gl->gl_flags); | |
2064 | freed++; | |
e66cf161 | 2065 | gl->gl_lockref.count++; |
4506a519 | 2066 | if (demote_ok(gl)) |
81ffbf65 | 2067 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); |
1e860444 | 2068 | gfs2_glock_queue_work(gl, 0); |
f3dd1649 | 2069 | spin_unlock(&gl->gl_lockref.lock); |
94a09a39 | 2070 | cond_resched_lock(&lru_lock); |
4506a519 | 2071 | } |
7a1ad9d8 | 2072 | return freed; |
4506a519 SW |
2073 | } |
2074 | ||
2a005855 SW |
2075 | /** |
2076 | * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote | |
2077 | * @nr: The number of entries to scan | |
2078 | * | |
4506a519 SW |
2079 | * This function selects the entries on the LRU which are able to |
2080 | * be demoted, and then kicks off the process by calling | |
2081 | * gfs2_dispose_glock_lru() above. | |
2a005855 | 2082 | */ |
b3b94faa | 2083 | |
7a1ad9d8 | 2084 | static unsigned long gfs2_scan_glock_lru(unsigned long nr) |
b3b94faa | 2085 | { |
8fb8f70e | 2086 | struct gfs2_glock *gl, *next; |
4506a519 | 2087 | LIST_HEAD(dispose); |
7a1ad9d8 | 2088 | unsigned long freed = 0; |
b3b94faa | 2089 | |
97cc1025 | 2090 | spin_lock(&lru_lock); |
8fb8f70e | 2091 | list_for_each_entry_safe(gl, next, &lru_list, gl_lru) { |
7a1ad9d8 | 2092 | if (!nr--) |
8fb8f70e | 2093 | break; |
7a1ad9d8 | 2094 | if (can_free_glock(gl)) |
acf1f42f | 2095 | list_move(&gl->gl_lru, &dispose); |
b3b94faa | 2096 | } |
4506a519 | 2097 | if (!list_empty(&dispose)) |
7a1ad9d8 | 2098 | freed = gfs2_dispose_glock_lru(&dispose); |
97cc1025 | 2099 | spin_unlock(&lru_lock); |
1ab6c499 DC |
2100 | |
2101 | return freed; | |
2a005855 SW |
2102 | } |
2103 | ||
1ab6c499 DC |
2104 | static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink, |
2105 | struct shrink_control *sc) | |
2a005855 | 2106 | { |
1ab6c499 DC |
2107 | if (!(sc->gfp_mask & __GFP_FS)) |
2108 | return SHRINK_STOP; | |
2109 | return gfs2_scan_glock_lru(sc->nr_to_scan); | |
2110 | } | |
2a005855 | 2111 | |
1ab6c499 DC |
2112 | static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink, |
2113 | struct shrink_control *sc) | |
2114 | { | |
55f841ce | 2115 | return vfs_pressure_ratio(atomic_read(&lru_count)); |
b3b94faa DT |
2116 | } |
2117 | ||
a304c23c | 2118 | static struct shrinker *glock_shrinker; |
97cc1025 | 2119 | |
b3b94faa | 2120 | /** |
dbffb29d | 2121 | * glock_hash_walk - Call a function for glock in a hash bucket |
b3b94faa DT |
2122 | * @examiner: the function |
2123 | * @sdp: the filesystem | |
b3b94faa | 2124 | * |
98687f42 HX |
2125 | * Note that the function can be called multiple times on the same |
2126 | * object. So the user must ensure that the function can cope with | |
2127 | * that. | |
b3b94faa DT |
2128 | */ |
2129 | ||
88ffbf3e | 2130 | static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp) |
b3b94faa | 2131 | { |
bc015cb8 | 2132 | struct gfs2_glock *gl; |
98687f42 HX |
2133 | struct rhashtable_iter iter; |
2134 | ||
2135 | rhashtable_walk_enter(&gl_hash_table, &iter); | |
2136 | ||
2137 | do { | |
97a6ec4a | 2138 | rhashtable_walk_start(&iter); |
98687f42 | 2139 | |
7427f3bb AG |
2140 | while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) { |
2141 | if (gl->gl_name.ln_sbd == sdp) | |
88ffbf3e | 2142 | examiner(gl); |
7427f3bb | 2143 | } |
98687f42 HX |
2144 | |
2145 | rhashtable_walk_stop(&iter); | |
2146 | } while (cond_resched(), gl == ERR_PTR(-EAGAIN)); | |
2147 | ||
2148 | rhashtable_walk_exit(&iter); | |
bc015cb8 SW |
2149 | } |
2150 | ||
a0e3cc65 AG |
2151 | void gfs2_cancel_delete_work(struct gfs2_glock *gl) |
2152 | { | |
f0e56edc AG |
2153 | clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags); |
2154 | clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags); | |
3056dc46 | 2155 | if (cancel_delayed_work(&gl->gl_delete)) |
a0e3cc65 | 2156 | gfs2_glock_put(gl); |
a0e3cc65 AG |
2157 | } |
2158 | ||
2159 | static void flush_delete_work(struct gfs2_glock *gl) | |
2160 | { | |
2ffed529 | 2161 | if (gl->gl_name.ln_type == LM_TYPE_IOPEN) { |
0247f4e9 AG |
2162 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
2163 | ||
2ffed529 | 2164 | if (cancel_delayed_work(&gl->gl_delete)) { |
0247f4e9 | 2165 | queue_delayed_work(sdp->sd_delete_wq, |
2ffed529 BP |
2166 | &gl->gl_delete, 0); |
2167 | } | |
34244d71 | 2168 | } |
a0e3cc65 AG |
2169 | } |
2170 | ||
2171 | void gfs2_flush_delete_work(struct gfs2_sbd *sdp) | |
2172 | { | |
2173 | glock_hash_walk(flush_delete_work, sdp); | |
0247f4e9 | 2174 | flush_workqueue(sdp->sd_delete_wq); |
a0e3cc65 AG |
2175 | } |
2176 | ||
f057f6cd SW |
2177 | /** |
2178 | * thaw_glock - thaw out a glock which has an unprocessed reply waiting | |
2179 | * @gl: The glock to thaw | |
2180 | * | |
f057f6cd SW |
2181 | */ |
2182 | ||
2183 | static void thaw_glock(struct gfs2_glock *gl) | |
2184 | { | |
7427f3bb AG |
2185 | if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) |
2186 | return; | |
2187 | if (!lockref_get_not_dead(&gl->gl_lockref)) | |
6b0c7440 | 2188 | return; |
1cd28e15 AG |
2189 | |
2190 | spin_lock(&gl->gl_lockref.lock); | |
6b0c7440 | 2191 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
1e860444 | 2192 | gfs2_glock_queue_work(gl, 0); |
1cd28e15 | 2193 | spin_unlock(&gl->gl_lockref.lock); |
f057f6cd SW |
2194 | } |
2195 | ||
b3b94faa DT |
2196 | /** |
2197 | * clear_glock - look at a glock and see if we can free it from glock cache | |
2198 | * @gl: the glock to look at | |
2199 | * | |
2200 | */ | |
2201 | ||
2202 | static void clear_glock(struct gfs2_glock *gl) | |
2203 | { | |
f42ab085 | 2204 | gfs2_glock_remove_from_lru(gl); |
b3b94faa | 2205 | |
f3dd1649 | 2206 | spin_lock(&gl->gl_lockref.lock); |
7427f3bb AG |
2207 | if (!__lockref_is_dead(&gl->gl_lockref)) { |
2208 | gl->gl_lockref.count++; | |
2209 | if (gl->gl_state != LM_ST_UNLOCKED) | |
2210 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); | |
1e860444 | 2211 | gfs2_glock_queue_work(gl, 0); |
7427f3bb | 2212 | } |
f3dd1649 | 2213 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
2214 | } |
2215 | ||
f057f6cd SW |
2216 | /** |
2217 | * gfs2_glock_thaw - Thaw any frozen glocks | |
2218 | * @sdp: The super block | |
2219 | * | |
2220 | */ | |
2221 | ||
2222 | void gfs2_glock_thaw(struct gfs2_sbd *sdp) | |
2223 | { | |
bc015cb8 SW |
2224 | glock_hash_walk(thaw_glock, sdp); |
2225 | } | |
f057f6cd | 2226 | |
3792ce97 | 2227 | static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid) |
bc015cb8 | 2228 | { |
f3dd1649 | 2229 | spin_lock(&gl->gl_lockref.lock); |
3792ce97 | 2230 | gfs2_dump_glock(seq, gl, fsid); |
f3dd1649 | 2231 | spin_unlock(&gl->gl_lockref.lock); |
bc015cb8 SW |
2232 | } |
2233 | ||
2234 | static void dump_glock_func(struct gfs2_glock *gl) | |
2235 | { | |
3792ce97 | 2236 | dump_glock(NULL, gl, true); |
f057f6cd SW |
2237 | } |
2238 | ||
053640a7 BP |
2239 | static void withdraw_dq(struct gfs2_glock *gl) |
2240 | { | |
2241 | spin_lock(&gl->gl_lockref.lock); | |
2242 | if (!__lockref_is_dead(&gl->gl_lockref) && | |
2243 | glock_blocked_by_withdraw(gl)) | |
2244 | do_error(gl, LM_OUT_ERROR); /* remove pending waiters */ | |
2245 | spin_unlock(&gl->gl_lockref.lock); | |
2246 | } | |
2247 | ||
2248 | void gfs2_gl_dq_holders(struct gfs2_sbd *sdp) | |
2249 | { | |
2250 | glock_hash_walk(withdraw_dq, sdp); | |
2251 | } | |
2252 | ||
b3b94faa DT |
2253 | /** |
2254 | * gfs2_gl_hash_clear - Empty out the glock hash table | |
2255 | * @sdp: the filesystem | |
b3b94faa | 2256 | * |
1bdad606 | 2257 | * Called when unmounting the filesystem. |
b3b94faa DT |
2258 | */ |
2259 | ||
fefc03bf | 2260 | void gfs2_gl_hash_clear(struct gfs2_sbd *sdp) |
b3b94faa | 2261 | { |
fb6791d1 | 2262 | set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags); |
222cb538 | 2263 | flush_workqueue(glock_workqueue); |
bc015cb8 | 2264 | glock_hash_walk(clear_glock, sdp); |
8f05228e | 2265 | flush_workqueue(glock_workqueue); |
3c69c437 | 2266 | wait_event_timeout(sdp->sd_kill_wait, |
2aba1b5b BP |
2267 | atomic_read(&sdp->sd_glock_disposal) == 0, |
2268 | HZ * 600); | |
d98779e6 AG |
2269 | gfs2_lm_unmount(sdp); |
2270 | gfs2_free_dead_glocks(sdp); | |
bc015cb8 | 2271 | glock_hash_walk(dump_glock_func, sdp); |
b3b94faa DT |
2272 | } |
2273 | ||
6802e340 | 2274 | static const char *state2str(unsigned state) |
04b933f2 | 2275 | { |
6802e340 SW |
2276 | switch(state) { |
2277 | case LM_ST_UNLOCKED: | |
2278 | return "UN"; | |
2279 | case LM_ST_SHARED: | |
2280 | return "SH"; | |
2281 | case LM_ST_DEFERRED: | |
2282 | return "DF"; | |
2283 | case LM_ST_EXCLUSIVE: | |
2284 | return "EX"; | |
2285 | } | |
2286 | return "??"; | |
2287 | } | |
2288 | ||
b58bf407 | 2289 | static const char *hflags2str(char *buf, u16 flags, unsigned long iflags) |
6802e340 SW |
2290 | { |
2291 | char *p = buf; | |
2292 | if (flags & LM_FLAG_TRY) | |
2293 | *p++ = 't'; | |
2294 | if (flags & LM_FLAG_TRY_1CB) | |
2295 | *p++ = 'T'; | |
2296 | if (flags & LM_FLAG_NOEXP) | |
2297 | *p++ = 'e'; | |
2298 | if (flags & LM_FLAG_ANY) | |
f057f6cd | 2299 | *p++ = 'A'; |
06e908cd BP |
2300 | if (flags & LM_FLAG_NODE_SCOPE) |
2301 | *p++ = 'n'; | |
6802e340 SW |
2302 | if (flags & GL_ASYNC) |
2303 | *p++ = 'a'; | |
2304 | if (flags & GL_EXACT) | |
2305 | *p++ = 'E'; | |
6802e340 SW |
2306 | if (flags & GL_NOCACHE) |
2307 | *p++ = 'c'; | |
2308 | if (test_bit(HIF_HOLDER, &iflags)) | |
2309 | *p++ = 'H'; | |
2310 | if (test_bit(HIF_WAIT, &iflags)) | |
2311 | *p++ = 'W'; | |
4c69038d BP |
2312 | if (flags & GL_SKIP) |
2313 | *p++ = 's'; | |
6802e340 SW |
2314 | *p = 0; |
2315 | return buf; | |
04b933f2 RP |
2316 | } |
2317 | ||
b3b94faa DT |
2318 | /** |
2319 | * dump_holder - print information about a glock holder | |
6802e340 | 2320 | * @seq: the seq_file struct |
b3b94faa | 2321 | * @gh: the glock holder |
3792ce97 | 2322 | * @fs_id_buf: pointer to file system id (if requested) |
b3b94faa | 2323 | * |
b3b94faa DT |
2324 | */ |
2325 | ||
3792ce97 BP |
2326 | static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh, |
2327 | const char *fs_id_buf) | |
b3b94faa | 2328 | { |
cbe6d257 AG |
2329 | const char *comm = "(none)"; |
2330 | pid_t owner_pid = 0; | |
6802e340 | 2331 | char flags_buf[32]; |
b3b94faa | 2332 | |
0b3a2c99 | 2333 | rcu_read_lock(); |
cbe6d257 AG |
2334 | if (pid_is_meaningful(gh)) { |
2335 | struct task_struct *gh_owner; | |
2336 | ||
2337 | comm = "(ended)"; | |
2338 | owner_pid = pid_nr(gh->gh_owner_pid); | |
b1e058da | 2339 | gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID); |
cbe6d257 AG |
2340 | if (gh_owner) |
2341 | comm = gh_owner->comm; | |
2342 | } | |
3792ce97 BP |
2343 | gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n", |
2344 | fs_id_buf, state2str(gh->gh_state), | |
cc18152e | 2345 | hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags), |
cbe6d257 | 2346 | gh->gh_error, (long)owner_pid, comm, (void *)gh->gh_ip); |
0b3a2c99 | 2347 | rcu_read_unlock(); |
b3b94faa DT |
2348 | } |
2349 | ||
627c10b7 | 2350 | static const char *gflags2str(char *buf, const struct gfs2_glock *gl) |
6802e340 | 2351 | { |
627c10b7 | 2352 | const unsigned long *gflags = &gl->gl_flags; |
6802e340 | 2353 | char *p = buf; |
627c10b7 | 2354 | |
6802e340 SW |
2355 | if (test_bit(GLF_LOCK, gflags)) |
2356 | *p++ = 'l'; | |
6802e340 SW |
2357 | if (test_bit(GLF_DEMOTE, gflags)) |
2358 | *p++ = 'D'; | |
2359 | if (test_bit(GLF_PENDING_DEMOTE, gflags)) | |
2360 | *p++ = 'd'; | |
2361 | if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags)) | |
2362 | *p++ = 'p'; | |
2363 | if (test_bit(GLF_DIRTY, gflags)) | |
2364 | *p++ = 'y'; | |
2365 | if (test_bit(GLF_LFLUSH, gflags)) | |
2366 | *p++ = 'f'; | |
2367 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags)) | |
2368 | *p++ = 'i'; | |
2369 | if (test_bit(GLF_REPLY_PENDING, gflags)) | |
2370 | *p++ = 'r'; | |
f057f6cd | 2371 | if (test_bit(GLF_INITIAL, gflags)) |
d8348de0 | 2372 | *p++ = 'I'; |
f057f6cd SW |
2373 | if (test_bit(GLF_FROZEN, gflags)) |
2374 | *p++ = 'F'; | |
e2c6c8a7 | 2375 | if (!list_empty(&gl->gl_holders)) |
7b5e3d5f | 2376 | *p++ = 'q'; |
627c10b7 SW |
2377 | if (test_bit(GLF_LRU, gflags)) |
2378 | *p++ = 'L'; | |
2379 | if (gl->gl_object) | |
2380 | *p++ = 'o'; | |
a245769f SW |
2381 | if (test_bit(GLF_BLOCKING, gflags)) |
2382 | *p++ = 'b'; | |
5deaf1f6 BP |
2383 | if (test_bit(GLF_FREEING, gflags)) |
2384 | *p++ = 'x'; | |
f2e70d8f BP |
2385 | if (test_bit(GLF_INSTANTIATE_NEEDED, gflags)) |
2386 | *p++ = 'n'; | |
2387 | if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags)) | |
2388 | *p++ = 'N'; | |
f0e56edc AG |
2389 | if (test_bit(GLF_TRY_TO_EVICT, gflags)) |
2390 | *p++ = 'e'; | |
2391 | if (test_bit(GLF_VERIFY_EVICT, gflags)) | |
2392 | *p++ = 'E'; | |
6802e340 SW |
2393 | *p = 0; |
2394 | return buf; | |
b3b94faa DT |
2395 | } |
2396 | ||
2397 | /** | |
8eae1ca0 | 2398 | * gfs2_dump_glock - print information about a glock |
6802e340 | 2399 | * @seq: The seq_file struct |
b3b94faa | 2400 | * @gl: the glock |
3792ce97 | 2401 | * @fsid: If true, also dump the file system id |
6802e340 SW |
2402 | * |
2403 | * The file format is as follows: | |
2404 | * One line per object, capital letters are used to indicate objects | |
2405 | * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented, | |
2406 | * other objects are indented by a single space and follow the glock to | |
2407 | * which they are related. Fields are indicated by lower case letters | |
2408 | * followed by a colon and the field value, except for strings which are in | |
2409 | * [] so that its possible to see if they are composed of spaces for | |
2410 | * example. The field's are n = number (id of the object), f = flags, | |
2411 | * t = type, s = state, r = refcount, e = error, p = pid. | |
b3b94faa | 2412 | * |
b3b94faa DT |
2413 | */ |
2414 | ||
3792ce97 | 2415 | void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid) |
b3b94faa | 2416 | { |
6802e340 SW |
2417 | const struct gfs2_glock_operations *glops = gl->gl_ops; |
2418 | unsigned long long dtime; | |
2419 | const struct gfs2_holder *gh; | |
2420 | char gflags_buf[32]; | |
3792ce97 | 2421 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
98fb0574 | 2422 | char fs_id_buf[sizeof(sdp->sd_fsname) + 7]; |
7e901d6e | 2423 | unsigned long nrpages = 0; |
b3b94faa | 2424 | |
7e901d6e BP |
2425 | if (gl->gl_ops->go_flags & GLOF_ASPACE) { |
2426 | struct address_space *mapping = gfs2_glock2aspace(gl); | |
2427 | ||
2428 | nrpages = mapping->nrpages; | |
2429 | } | |
3792ce97 BP |
2430 | memset(fs_id_buf, 0, sizeof(fs_id_buf)); |
2431 | if (fsid && sdp) /* safety precaution */ | |
2432 | sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname); | |
6802e340 SW |
2433 | dtime = jiffies - gl->gl_demote_time; |
2434 | dtime *= 1000000/HZ; /* demote time in uSec */ | |
2435 | if (!test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
2436 | dtime = 0; | |
3792ce97 | 2437 | gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d " |
7e901d6e BP |
2438 | "v:%d r:%d m:%ld p:%lu\n", |
2439 | fs_id_buf, state2str(gl->gl_state), | |
2440 | gl->gl_name.ln_type, | |
2441 | (unsigned long long)gl->gl_name.ln_number, | |
2442 | gflags2str(gflags_buf, gl), | |
2443 | state2str(gl->gl_target), | |
2444 | state2str(gl->gl_demote_state), dtime, | |
2445 | atomic_read(&gl->gl_ail_count), | |
2446 | atomic_read(&gl->gl_revokes), | |
2447 | (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages); | |
b3b94faa | 2448 | |
ac3beb6a | 2449 | list_for_each_entry(gh, &gl->gl_holders, gh_list) |
3792ce97 | 2450 | dump_holder(seq, gh, fs_id_buf); |
ac3beb6a | 2451 | |
6802e340 | 2452 | if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump) |
3792ce97 | 2453 | glops->go_dump(seq, gl, fs_id_buf); |
b3b94faa DT |
2454 | } |
2455 | ||
a245769f SW |
2456 | static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr) |
2457 | { | |
2458 | struct gfs2_glock *gl = iter_ptr; | |
2459 | ||
4d207133 | 2460 | seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n", |
a245769f SW |
2461 | gl->gl_name.ln_type, |
2462 | (unsigned long long)gl->gl_name.ln_number, | |
4d207133 BH |
2463 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT], |
2464 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR], | |
2465 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB], | |
2466 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB], | |
2467 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT], | |
2468 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR], | |
2469 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT], | |
2470 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]); | |
a245769f SW |
2471 | return 0; |
2472 | } | |
2473 | ||
2474 | static const char *gfs2_gltype[] = { | |
2475 | "type", | |
2476 | "reserved", | |
2477 | "nondisk", | |
2478 | "inode", | |
2479 | "rgrp", | |
2480 | "meta", | |
2481 | "iopen", | |
2482 | "flock", | |
2483 | "plock", | |
2484 | "quota", | |
2485 | "journal", | |
2486 | }; | |
2487 | ||
2488 | static const char *gfs2_stype[] = { | |
2489 | [GFS2_LKS_SRTT] = "srtt", | |
2490 | [GFS2_LKS_SRTTVAR] = "srttvar", | |
2491 | [GFS2_LKS_SRTTB] = "srttb", | |
2492 | [GFS2_LKS_SRTTVARB] = "srttvarb", | |
2493 | [GFS2_LKS_SIRT] = "sirt", | |
2494 | [GFS2_LKS_SIRTVAR] = "sirtvar", | |
2495 | [GFS2_LKS_DCOUNT] = "dlm", | |
2496 | [GFS2_LKS_QCOUNT] = "queue", | |
2497 | }; | |
2498 | ||
2499 | #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype)) | |
2500 | ||
2501 | static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr) | |
2502 | { | |
81648d04 AG |
2503 | struct gfs2_sbd *sdp = seq->private; |
2504 | loff_t pos = *(loff_t *)iter_ptr; | |
2505 | unsigned index = pos >> 3; | |
2506 | unsigned subindex = pos & 0x07; | |
a245769f SW |
2507 | int i; |
2508 | ||
2509 | if (index == 0 && subindex != 0) | |
2510 | return 0; | |
6802e340 | 2511 | |
a245769f SW |
2512 | seq_printf(seq, "%-10s %8s:", gfs2_gltype[index], |
2513 | (index == 0) ? "cpu": gfs2_stype[subindex]); | |
b3b94faa | 2514 | |
a245769f SW |
2515 | for_each_possible_cpu(i) { |
2516 | const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i); | |
8f7e0a80 AG |
2517 | |
2518 | if (index == 0) | |
2519 | seq_printf(seq, " %15u", i); | |
2520 | else | |
2521 | seq_printf(seq, " %15llu", (unsigned long long)lkstats-> | |
2522 | lkstats[index - 1].stats[subindex]); | |
a245769f SW |
2523 | } |
2524 | seq_putc(seq, '\n'); | |
2525 | return 0; | |
2526 | } | |
8fbbfd21 | 2527 | |
85d1da67 SW |
2528 | int __init gfs2_glock_init(void) |
2529 | { | |
0515480a | 2530 | int i, ret; |
88ffbf3e BP |
2531 | |
2532 | ret = rhashtable_init(&gl_hash_table, &ht_parms); | |
2533 | if (ret < 0) | |
2534 | return ret; | |
8fbbfd21 | 2535 | |
d2115778 | 2536 | glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM | |
58a69cb4 | 2537 | WQ_HIGHPRI | WQ_FREEZABLE, 0); |
88ffbf3e BP |
2538 | if (!glock_workqueue) { |
2539 | rhashtable_destroy(&gl_hash_table); | |
dfc4616d | 2540 | return -ENOMEM; |
88ffbf3e | 2541 | } |
97cc1025 | 2542 | |
a304c23c QZ |
2543 | glock_shrinker = shrinker_alloc(0, "gfs2-glock"); |
2544 | if (!glock_shrinker) { | |
e0d735c1 CY |
2545 | destroy_workqueue(glock_workqueue); |
2546 | rhashtable_destroy(&gl_hash_table); | |
a304c23c | 2547 | return -ENOMEM; |
e0d735c1 | 2548 | } |
c4f68a13 | 2549 | |
a304c23c QZ |
2550 | glock_shrinker->count_objects = gfs2_glock_shrink_count; |
2551 | glock_shrinker->scan_objects = gfs2_glock_shrink_scan; | |
2552 | ||
2553 | shrinker_register(glock_shrinker); | |
2554 | ||
0515480a AG |
2555 | for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++) |
2556 | init_waitqueue_head(glock_wait_table + i); | |
2557 | ||
85d1da67 SW |
2558 | return 0; |
2559 | } | |
2560 | ||
8fbbfd21 SW |
2561 | void gfs2_glock_exit(void) |
2562 | { | |
a304c23c | 2563 | shrinker_free(glock_shrinker); |
88ffbf3e | 2564 | rhashtable_destroy(&gl_hash_table); |
c4f68a13 | 2565 | destroy_workqueue(glock_workqueue); |
8fbbfd21 SW |
2566 | } |
2567 | ||
7ac07fda | 2568 | static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n) |
bc015cb8 | 2569 | { |
3fd5d3ad AG |
2570 | struct gfs2_glock *gl = gi->gl; |
2571 | ||
2572 | if (gl) { | |
2573 | if (n == 0) | |
2574 | return; | |
ee2be7d7 | 2575 | gfs2_glock_put_async(gl); |
7ac07fda AG |
2576 | } |
2577 | for (;;) { | |
3fd5d3ad AG |
2578 | gl = rhashtable_walk_next(&gi->hti); |
2579 | if (IS_ERR_OR_NULL(gl)) { | |
2580 | if (gl == ERR_PTR(-EAGAIN)) { | |
2581 | n = 1; | |
2582 | continue; | |
7ac07fda | 2583 | } |
3fd5d3ad AG |
2584 | gl = NULL; |
2585 | break; | |
2586 | } | |
2587 | if (gl->gl_name.ln_sbd != gi->sdp) | |
2588 | continue; | |
2589 | if (n <= 1) { | |
2590 | if (!lockref_get_not_dead(&gl->gl_lockref)) | |
2591 | continue; | |
2592 | break; | |
2593 | } else { | |
2594 | if (__lockref_is_dead(&gl->gl_lockref)) | |
2595 | continue; | |
2596 | n--; | |
bc015cb8 | 2597 | } |
14d37564 | 2598 | } |
3fd5d3ad | 2599 | gi->gl = gl; |
7c52b166 RP |
2600 | } |
2601 | ||
6802e340 | 2602 | static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos) |
27c3b415 | 2603 | __acquires(RCU) |
7c52b166 | 2604 | { |
6802e340 | 2605 | struct gfs2_glock_iter *gi = seq->private; |
7ac07fda | 2606 | loff_t n; |
ba1ddcb6 | 2607 | |
7ac07fda AG |
2608 | /* |
2609 | * We can either stay where we are, skip to the next hash table | |
2610 | * entry, or start from the beginning. | |
2611 | */ | |
2612 | if (*pos < gi->last_pos) { | |
2613 | rhashtable_walk_exit(&gi->hti); | |
2614 | rhashtable_walk_enter(&gl_hash_table, &gi->hti); | |
2615 | n = *pos + 1; | |
2616 | } else { | |
2617 | n = *pos - gi->last_pos; | |
2618 | } | |
7c52b166 | 2619 | |
7ac07fda | 2620 | rhashtable_walk_start(&gi->hti); |
7c52b166 | 2621 | |
7ac07fda | 2622 | gfs2_glock_iter_next(gi, n); |
ba1ddcb6 | 2623 | gi->last_pos = *pos; |
6802e340 | 2624 | return gi->gl; |
7c52b166 RP |
2625 | } |
2626 | ||
6802e340 | 2627 | static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr, |
7c52b166 RP |
2628 | loff_t *pos) |
2629 | { | |
6802e340 | 2630 | struct gfs2_glock_iter *gi = seq->private; |
7c52b166 RP |
2631 | |
2632 | (*pos)++; | |
ba1ddcb6 | 2633 | gi->last_pos = *pos; |
7ac07fda | 2634 | gfs2_glock_iter_next(gi, 1); |
6802e340 | 2635 | return gi->gl; |
7c52b166 RP |
2636 | } |
2637 | ||
6802e340 | 2638 | static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr) |
27c3b415 | 2639 | __releases(RCU) |
7c52b166 | 2640 | { |
6802e340 | 2641 | struct gfs2_glock_iter *gi = seq->private; |
bc015cb8 | 2642 | |
88ffbf3e | 2643 | rhashtable_walk_stop(&gi->hti); |
7c52b166 RP |
2644 | } |
2645 | ||
6802e340 | 2646 | static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr) |
7c52b166 | 2647 | { |
3792ce97 | 2648 | dump_glock(seq, iter_ptr, false); |
ac3beb6a | 2649 | return 0; |
7c52b166 RP |
2650 | } |
2651 | ||
a245769f SW |
2652 | static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos) |
2653 | { | |
81648d04 | 2654 | preempt_disable(); |
a245769f SW |
2655 | if (*pos >= GFS2_NR_SBSTATS) |
2656 | return NULL; | |
81648d04 | 2657 | return pos; |
a245769f SW |
2658 | } |
2659 | ||
2660 | static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr, | |
2661 | loff_t *pos) | |
2662 | { | |
a245769f | 2663 | (*pos)++; |
81648d04 | 2664 | if (*pos >= GFS2_NR_SBSTATS) |
a245769f | 2665 | return NULL; |
81648d04 | 2666 | return pos; |
a245769f SW |
2667 | } |
2668 | ||
2669 | static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr) | |
2670 | { | |
2671 | preempt_enable(); | |
2672 | } | |
2673 | ||
4ef29002 | 2674 | static const struct seq_operations gfs2_glock_seq_ops = { |
7c52b166 RP |
2675 | .start = gfs2_glock_seq_start, |
2676 | .next = gfs2_glock_seq_next, | |
2677 | .stop = gfs2_glock_seq_stop, | |
2678 | .show = gfs2_glock_seq_show, | |
2679 | }; | |
2680 | ||
a245769f SW |
2681 | static const struct seq_operations gfs2_glstats_seq_ops = { |
2682 | .start = gfs2_glock_seq_start, | |
2683 | .next = gfs2_glock_seq_next, | |
2684 | .stop = gfs2_glock_seq_stop, | |
2685 | .show = gfs2_glstats_seq_show, | |
2686 | }; | |
2687 | ||
e8a8023e | 2688 | static const struct seq_operations gfs2_sbstats_sops = { |
a245769f SW |
2689 | .start = gfs2_sbstats_seq_start, |
2690 | .next = gfs2_sbstats_seq_next, | |
2691 | .stop = gfs2_sbstats_seq_stop, | |
2692 | .show = gfs2_sbstats_seq_show, | |
2693 | }; | |
2694 | ||
0fe2f1e9 SW |
2695 | #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL) |
2696 | ||
92ecd73a AG |
2697 | static int __gfs2_glocks_open(struct inode *inode, struct file *file, |
2698 | const struct seq_operations *ops) | |
7c52b166 | 2699 | { |
92ecd73a | 2700 | int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter)); |
6802e340 SW |
2701 | if (ret == 0) { |
2702 | struct seq_file *seq = file->private_data; | |
2703 | struct gfs2_glock_iter *gi = seq->private; | |
88ffbf3e | 2704 | |
6802e340 | 2705 | gi->sdp = inode->i_private; |
0fe2f1e9 | 2706 | seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN); |
df5d2f55 | 2707 | if (seq->buf) |
0fe2f1e9 | 2708 | seq->size = GFS2_SEQ_GOODSIZE; |
7ac07fda AG |
2709 | /* |
2710 | * Initially, we are "before" the first hash table entry; the | |
2711 | * first call to rhashtable_walk_next gets us the first entry. | |
2712 | */ | |
2713 | gi->last_pos = -1; | |
88ffbf3e | 2714 | gi->gl = NULL; |
7ac07fda | 2715 | rhashtable_walk_enter(&gl_hash_table, &gi->hti); |
6802e340 SW |
2716 | } |
2717 | return ret; | |
7c52b166 RP |
2718 | } |
2719 | ||
92ecd73a AG |
2720 | static int gfs2_glocks_open(struct inode *inode, struct file *file) |
2721 | { | |
2722 | return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops); | |
2723 | } | |
2724 | ||
88ffbf3e BP |
2725 | static int gfs2_glocks_release(struct inode *inode, struct file *file) |
2726 | { | |
2727 | struct seq_file *seq = file->private_data; | |
2728 | struct gfs2_glock_iter *gi = seq->private; | |
2729 | ||
3fd5d3ad AG |
2730 | if (gi->gl) |
2731 | gfs2_glock_put(gi->gl); | |
7ac07fda | 2732 | rhashtable_walk_exit(&gi->hti); |
88ffbf3e BP |
2733 | return seq_release_private(inode, file); |
2734 | } | |
2735 | ||
a245769f SW |
2736 | static int gfs2_glstats_open(struct inode *inode, struct file *file) |
2737 | { | |
92ecd73a | 2738 | return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops); |
a245769f SW |
2739 | } |
2740 | ||
a245769f SW |
2741 | static const struct file_operations gfs2_glocks_fops = { |
2742 | .owner = THIS_MODULE, | |
2743 | .open = gfs2_glocks_open, | |
2744 | .read = seq_read, | |
2745 | .llseek = seq_lseek, | |
88ffbf3e | 2746 | .release = gfs2_glocks_release, |
a245769f SW |
2747 | }; |
2748 | ||
2749 | static const struct file_operations gfs2_glstats_fops = { | |
7c52b166 | 2750 | .owner = THIS_MODULE, |
a245769f SW |
2751 | .open = gfs2_glstats_open, |
2752 | .read = seq_read, | |
2753 | .llseek = seq_lseek, | |
88ffbf3e | 2754 | .release = gfs2_glocks_release, |
a245769f SW |
2755 | }; |
2756 | ||
4480c27c AG |
2757 | struct gfs2_glockfd_iter { |
2758 | struct super_block *sb; | |
2759 | unsigned int tgid; | |
2760 | struct task_struct *task; | |
2761 | unsigned int fd; | |
2762 | struct file *file; | |
2763 | }; | |
2764 | ||
2765 | static struct task_struct *gfs2_glockfd_next_task(struct gfs2_glockfd_iter *i) | |
2766 | { | |
2767 | struct pid_namespace *ns = task_active_pid_ns(current); | |
2768 | struct pid *pid; | |
2769 | ||
2770 | if (i->task) | |
2771 | put_task_struct(i->task); | |
2772 | ||
2773 | rcu_read_lock(); | |
2774 | retry: | |
2775 | i->task = NULL; | |
2776 | pid = find_ge_pid(i->tgid, ns); | |
2777 | if (pid) { | |
2778 | i->tgid = pid_nr_ns(pid, ns); | |
2779 | i->task = pid_task(pid, PIDTYPE_TGID); | |
2780 | if (!i->task) { | |
2781 | i->tgid++; | |
2782 | goto retry; | |
2783 | } | |
2784 | get_task_struct(i->task); | |
2785 | } | |
2786 | rcu_read_unlock(); | |
2787 | return i->task; | |
2788 | } | |
2789 | ||
2790 | static struct file *gfs2_glockfd_next_file(struct gfs2_glockfd_iter *i) | |
2791 | { | |
2792 | if (i->file) { | |
2793 | fput(i->file); | |
2794 | i->file = NULL; | |
2795 | } | |
2796 | ||
2797 | rcu_read_lock(); | |
2798 | for(;; i->fd++) { | |
2799 | struct inode *inode; | |
2800 | ||
0ede61d8 | 2801 | i->file = task_lookup_next_fdget_rcu(i->task, &i->fd); |
4480c27c AG |
2802 | if (!i->file) { |
2803 | i->fd = 0; | |
2804 | break; | |
2805 | } | |
0ede61d8 | 2806 | |
4480c27c | 2807 | inode = file_inode(i->file); |
0ede61d8 | 2808 | if (inode->i_sb == i->sb) |
4480c27c | 2809 | break; |
0ede61d8 CB |
2810 | |
2811 | rcu_read_unlock(); | |
2812 | fput(i->file); | |
2813 | rcu_read_lock(); | |
4480c27c AG |
2814 | } |
2815 | rcu_read_unlock(); | |
2816 | return i->file; | |
2817 | } | |
2818 | ||
2819 | static void *gfs2_glockfd_seq_start(struct seq_file *seq, loff_t *pos) | |
2820 | { | |
2821 | struct gfs2_glockfd_iter *i = seq->private; | |
2822 | ||
2823 | if (*pos) | |
2824 | return NULL; | |
2825 | while (gfs2_glockfd_next_task(i)) { | |
2826 | if (gfs2_glockfd_next_file(i)) | |
2827 | return i; | |
2828 | i->tgid++; | |
2829 | } | |
2830 | return NULL; | |
2831 | } | |
2832 | ||
2833 | static void *gfs2_glockfd_seq_next(struct seq_file *seq, void *iter_ptr, | |
2834 | loff_t *pos) | |
2835 | { | |
2836 | struct gfs2_glockfd_iter *i = seq->private; | |
2837 | ||
2838 | (*pos)++; | |
2839 | i->fd++; | |
2840 | do { | |
2841 | if (gfs2_glockfd_next_file(i)) | |
2842 | return i; | |
2843 | i->tgid++; | |
2844 | } while (gfs2_glockfd_next_task(i)); | |
2845 | return NULL; | |
2846 | } | |
2847 | ||
2848 | static void gfs2_glockfd_seq_stop(struct seq_file *seq, void *iter_ptr) | |
2849 | { | |
2850 | struct gfs2_glockfd_iter *i = seq->private; | |
2851 | ||
2852 | if (i->file) | |
2853 | fput(i->file); | |
2854 | if (i->task) | |
2855 | put_task_struct(i->task); | |
2856 | } | |
2857 | ||
56535dc6 AG |
2858 | static void gfs2_glockfd_seq_show_flock(struct seq_file *seq, |
2859 | struct gfs2_glockfd_iter *i) | |
2860 | { | |
2861 | struct gfs2_file *fp = i->file->private_data; | |
2862 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; | |
2863 | struct lm_lockname gl_name = { .ln_type = LM_TYPE_RESERVED }; | |
2864 | ||
2865 | if (!READ_ONCE(fl_gh->gh_gl)) | |
2866 | return; | |
2867 | ||
2868 | spin_lock(&i->file->f_lock); | |
2869 | if (gfs2_holder_initialized(fl_gh)) | |
2870 | gl_name = fl_gh->gh_gl->gl_name; | |
2871 | spin_unlock(&i->file->f_lock); | |
2872 | ||
2873 | if (gl_name.ln_type != LM_TYPE_RESERVED) { | |
2874 | seq_printf(seq, "%d %u %u/%llx\n", | |
2875 | i->tgid, i->fd, gl_name.ln_type, | |
2876 | (unsigned long long)gl_name.ln_number); | |
2877 | } | |
2878 | } | |
2879 | ||
4480c27c AG |
2880 | static int gfs2_glockfd_seq_show(struct seq_file *seq, void *iter_ptr) |
2881 | { | |
2882 | struct gfs2_glockfd_iter *i = seq->private; | |
2883 | struct inode *inode = file_inode(i->file); | |
2884 | struct gfs2_glock *gl; | |
2885 | ||
2886 | inode_lock_shared(inode); | |
2887 | gl = GFS2_I(inode)->i_iopen_gh.gh_gl; | |
2888 | if (gl) { | |
2889 | seq_printf(seq, "%d %u %u/%llx\n", | |
2890 | i->tgid, i->fd, gl->gl_name.ln_type, | |
2891 | (unsigned long long)gl->gl_name.ln_number); | |
2892 | } | |
56535dc6 | 2893 | gfs2_glockfd_seq_show_flock(seq, i); |
4480c27c AG |
2894 | inode_unlock_shared(inode); |
2895 | return 0; | |
2896 | } | |
2897 | ||
2898 | static const struct seq_operations gfs2_glockfd_seq_ops = { | |
2899 | .start = gfs2_glockfd_seq_start, | |
2900 | .next = gfs2_glockfd_seq_next, | |
2901 | .stop = gfs2_glockfd_seq_stop, | |
2902 | .show = gfs2_glockfd_seq_show, | |
2903 | }; | |
2904 | ||
2905 | static int gfs2_glockfd_open(struct inode *inode, struct file *file) | |
2906 | { | |
2907 | struct gfs2_glockfd_iter *i; | |
2908 | struct gfs2_sbd *sdp = inode->i_private; | |
2909 | ||
2910 | i = __seq_open_private(file, &gfs2_glockfd_seq_ops, | |
2911 | sizeof(struct gfs2_glockfd_iter)); | |
2912 | if (!i) | |
2913 | return -ENOMEM; | |
2914 | i->sb = sdp->sd_vfs; | |
2915 | return 0; | |
2916 | } | |
2917 | ||
2918 | static const struct file_operations gfs2_glockfd_fops = { | |
2919 | .owner = THIS_MODULE, | |
2920 | .open = gfs2_glockfd_open, | |
2921 | .read = seq_read, | |
2922 | .llseek = seq_lseek, | |
2923 | .release = seq_release_private, | |
2924 | }; | |
2925 | ||
e8a8023e | 2926 | DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats); |
7c52b166 | 2927 | |
2abbf9a4 GKH |
2928 | void gfs2_create_debugfs_file(struct gfs2_sbd *sdp) |
2929 | { | |
2930 | sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root); | |
7c52b166 | 2931 | |
2abbf9a4 GKH |
2932 | debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, |
2933 | &gfs2_glocks_fops); | |
2934 | ||
4480c27c AG |
2935 | debugfs_create_file("glockfd", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, |
2936 | &gfs2_glockfd_fops); | |
2937 | ||
2abbf9a4 GKH |
2938 | debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, |
2939 | &gfs2_glstats_fops); | |
2940 | ||
2941 | debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, | |
2942 | &gfs2_sbstats_fops); | |
7c52b166 RP |
2943 | } |
2944 | ||
2945 | void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp) | |
2946 | { | |
2abbf9a4 GKH |
2947 | debugfs_remove_recursive(sdp->debugfs_dir); |
2948 | sdp->debugfs_dir = NULL; | |
7c52b166 RP |
2949 | } |
2950 | ||
2abbf9a4 | 2951 | void gfs2_register_debugfs(void) |
7c52b166 RP |
2952 | { |
2953 | gfs2_root = debugfs_create_dir("gfs2", NULL); | |
7c52b166 RP |
2954 | } |
2955 | ||
2956 | void gfs2_unregister_debugfs(void) | |
2957 | { | |
2958 | debugfs_remove(gfs2_root); | |
5f882096 | 2959 | gfs2_root = NULL; |
7c52b166 | 2960 | } |