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 | ||
159 | if (likely(!gfs2_withdrawn(sdp))) | |
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 | ||
961ae1d8 AG |
169 | void gfs2_glock_free(struct gfs2_glock *gl) |
170 | { | |
171 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
172 | ||
ea4e61c7 | 173 | gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0); |
0515480a AG |
174 | rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms); |
175 | smp_mb(); | |
176 | wake_up_glock(gl); | |
961ae1d8 | 177 | call_rcu(&gl->gl_rcu, gfs2_glock_dealloc); |
bc015cb8 | 178 | if (atomic_dec_and_test(&sdp->sd_glock_disposal)) |
3c69c437 | 179 | wake_up(&sdp->sd_kill_wait); |
b3b94faa DT |
180 | } |
181 | ||
182 | /** | |
183 | * gfs2_glock_hold() - increment reference count on glock | |
184 | * @gl: The glock to hold | |
185 | * | |
186 | */ | |
187 | ||
4ad02083 | 188 | struct gfs2_glock *gfs2_glock_hold(struct gfs2_glock *gl) |
b3b94faa | 189 | { |
e66cf161 SW |
190 | GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref)); |
191 | lockref_get(&gl->gl_lockref); | |
4ad02083 | 192 | return gl; |
b3b94faa DT |
193 | } |
194 | ||
8ff22a6f BM |
195 | /** |
196 | * demote_ok - Check to see if it's ok to unlock a glock | |
197 | * @gl: the glock | |
198 | * | |
199 | * Returns: 1 if it's ok | |
200 | */ | |
201 | ||
202 | static int demote_ok(const struct gfs2_glock *gl) | |
203 | { | |
204 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
205 | ||
206 | if (gl->gl_state == LM_ST_UNLOCKED) | |
207 | return 0; | |
f42ab085 | 208 | if (!list_empty(&gl->gl_holders)) |
8ff22a6f BM |
209 | return 0; |
210 | if (glops->go_demote_ok) | |
211 | return glops->go_demote_ok(gl); | |
212 | return 1; | |
213 | } | |
214 | ||
bc015cb8 | 215 | |
29687a2a SW |
216 | void gfs2_glock_add_to_lru(struct gfs2_glock *gl) |
217 | { | |
7881ef3f RL |
218 | if (!(gl->gl_ops->go_flags & GLOF_LRU)) |
219 | return; | |
220 | ||
29687a2a SW |
221 | spin_lock(&lru_lock); |
222 | ||
38a618db | 223 | list_move_tail(&gl->gl_lru, &lru_list); |
7881ef3f RL |
224 | |
225 | if (!test_bit(GLF_LRU, &gl->gl_flags)) { | |
226 | set_bit(GLF_LRU, &gl->gl_flags); | |
29687a2a | 227 | atomic_inc(&lru_count); |
7881ef3f | 228 | } |
29687a2a | 229 | |
29687a2a SW |
230 | spin_unlock(&lru_lock); |
231 | } | |
232 | ||
8f6cb409 | 233 | static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl) |
f42ab085 | 234 | { |
645ebd49 BP |
235 | if (!(gl->gl_ops->go_flags & GLOF_LRU)) |
236 | return; | |
237 | ||
8f6cb409 | 238 | spin_lock(&lru_lock); |
7881ef3f | 239 | if (test_bit(GLF_LRU, &gl->gl_flags)) { |
f42ab085 SW |
240 | list_del_init(&gl->gl_lru); |
241 | atomic_dec(&lru_count); | |
242 | clear_bit(GLF_LRU, &gl->gl_flags); | |
243 | } | |
244 | spin_unlock(&lru_lock); | |
245 | } | |
246 | ||
6b0c7440 AG |
247 | /* |
248 | * Enqueue the glock on the work queue. Passes one glock reference on to the | |
249 | * work queue. | |
b3b94faa | 250 | */ |
6b0c7440 AG |
251 | static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) { |
252 | if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) { | |
253 | /* | |
254 | * We are holding the lockref spinlock, and the work was still | |
255 | * queued above. The queued work (glock_work_func) takes that | |
256 | * spinlock before dropping its glock reference(s), so it | |
257 | * cannot have dropped them in the meantime. | |
258 | */ | |
259 | GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2); | |
260 | gl->gl_lockref.count--; | |
261 | } | |
262 | } | |
b3b94faa | 263 | |
6b0c7440 AG |
264 | static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) { |
265 | spin_lock(&gl->gl_lockref.lock); | |
266 | __gfs2_glock_queue_work(gl, delay); | |
267 | spin_unlock(&gl->gl_lockref.lock); | |
268 | } | |
269 | ||
270 | static void __gfs2_glock_put(struct gfs2_glock *gl) | |
b3b94faa | 271 | { |
15562c43 | 272 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
bc015cb8 | 273 | struct address_space *mapping = gfs2_glock2aspace(gl); |
b3b94faa | 274 | |
e66cf161 | 275 | lockref_mark_dead(&gl->gl_lockref); |
e66cf161 | 276 | spin_unlock(&gl->gl_lockref.lock); |
228804a3 | 277 | gfs2_glock_remove_from_lru(gl); |
e66cf161 | 278 | GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders)); |
ee1e2c77 BP |
279 | if (mapping) { |
280 | truncate_inode_pages_final(mapping); | |
281 | if (!gfs2_withdrawn(sdp)) | |
7716506a | 282 | GLOCK_BUG_ON(gl, !mapping_empty(mapping)); |
ee1e2c77 | 283 | } |
e66cf161 SW |
284 | trace_gfs2_glock_put(gl); |
285 | sdp->sd_lockstruct.ls_ops->lm_put_lock(gl); | |
b3b94faa DT |
286 | } |
287 | ||
71c1b213 AG |
288 | /* |
289 | * Cause the glock to be put in work queue context. | |
290 | */ | |
291 | void gfs2_glock_queue_put(struct gfs2_glock *gl) | |
292 | { | |
293 | gfs2_glock_queue_work(gl, 0); | |
294 | } | |
295 | ||
6b0c7440 AG |
296 | /** |
297 | * gfs2_glock_put() - Decrement reference count on glock | |
298 | * @gl: The glock to put | |
299 | * | |
300 | */ | |
301 | ||
302 | void gfs2_glock_put(struct gfs2_glock *gl) | |
303 | { | |
304 | if (lockref_put_or_lock(&gl->gl_lockref)) | |
305 | return; | |
306 | ||
307 | __gfs2_glock_put(gl); | |
308 | } | |
309 | ||
6802e340 | 310 | /** |
61444649 | 311 | * may_grant - check if it's ok to grant a new lock |
6802e340 | 312 | * @gl: The glock |
61444649 | 313 | * @current_gh: One of the current holders of @gl |
6802e340 SW |
314 | * @gh: The lock request which we wish to grant |
315 | * | |
61444649 AG |
316 | * With our current compatibility rules, if a glock has one or more active |
317 | * holders (HIF_HOLDER flag set), any of those holders can be passed in as | |
318 | * @current_gh; they are all the same as far as compatibility with the new @gh | |
319 | * goes. | |
320 | * | |
321 | * Returns true if it's ok to grant the lock. | |
6802e340 SW |
322 | */ |
323 | ||
61444649 AG |
324 | static inline bool may_grant(struct gfs2_glock *gl, |
325 | struct gfs2_holder *current_gh, | |
326 | struct gfs2_holder *gh) | |
6802e340 | 327 | { |
61444649 AG |
328 | if (current_gh) { |
329 | GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags)); | |
330 | ||
331 | switch(current_gh->gh_state) { | |
332 | case LM_ST_EXCLUSIVE: | |
333 | /* | |
334 | * Here we make a special exception to grant holders | |
335 | * who agree to share the EX lock with other holders | |
336 | * who also have the bit set. If the original holder | |
337 | * has the LM_FLAG_NODE_SCOPE bit set, we grant more | |
338 | * holders with the bit set. | |
339 | */ | |
340 | return gh->gh_state == LM_ST_EXCLUSIVE && | |
341 | (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) && | |
342 | (gh->gh_flags & LM_FLAG_NODE_SCOPE); | |
343 | ||
344 | case LM_ST_SHARED: | |
345 | case LM_ST_DEFERRED: | |
346 | return gh->gh_state == current_gh->gh_state; | |
347 | ||
348 | default: | |
349 | return false; | |
350 | } | |
06e908cd | 351 | } |
61444649 | 352 | |
6802e340 | 353 | if (gl->gl_state == gh->gh_state) |
61444649 | 354 | return true; |
6802e340 | 355 | if (gh->gh_flags & GL_EXACT) |
61444649 | 356 | return false; |
209806ab | 357 | if (gl->gl_state == LM_ST_EXCLUSIVE) { |
61444649 AG |
358 | return gh->gh_state == LM_ST_SHARED || |
359 | gh->gh_state == LM_ST_DEFERRED; | |
209806ab | 360 | } |
61444649 AG |
361 | if (gh->gh_flags & LM_FLAG_ANY) |
362 | return gl->gl_state != LM_ST_UNLOCKED; | |
363 | return false; | |
6802e340 SW |
364 | } |
365 | ||
366 | static void gfs2_holder_wake(struct gfs2_holder *gh) | |
367 | { | |
368 | clear_bit(HIF_WAIT, &gh->gh_iflags); | |
4e857c58 | 369 | smp_mb__after_atomic(); |
6802e340 | 370 | wake_up_bit(&gh->gh_iflags, HIF_WAIT); |
ad26967b BP |
371 | if (gh->gh_flags & GL_ASYNC) { |
372 | struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd; | |
373 | ||
374 | wake_up(&sdp->sd_async_glock_wait); | |
375 | } | |
6802e340 SW |
376 | } |
377 | ||
d5341a92 SW |
378 | /** |
379 | * do_error - Something unexpected has happened during a lock request | |
c551f66c LJ |
380 | * @gl: The glock |
381 | * @ret: The status from the DLM | |
d5341a92 SW |
382 | */ |
383 | ||
a527b38e | 384 | static void do_error(struct gfs2_glock *gl, const int ret) |
d5341a92 SW |
385 | { |
386 | struct gfs2_holder *gh, *tmp; | |
387 | ||
388 | list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) { | |
ba3e77a4 | 389 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) |
d5341a92 SW |
390 | continue; |
391 | if (ret & LM_OUT_ERROR) | |
392 | gh->gh_error = -EIO; | |
393 | else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) | |
394 | gh->gh_error = GLR_TRYFAILED; | |
395 | else | |
396 | continue; | |
397 | list_del_init(&gh->gh_list); | |
398 | trace_gfs2_glock_queue(gh, 0); | |
399 | gfs2_holder_wake(gh); | |
400 | } | |
401 | } | |
402 | ||
61444649 AG |
403 | /** |
404 | * find_first_holder - find the first "holder" gh | |
405 | * @gl: the glock | |
406 | */ | |
407 | ||
408 | static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl) | |
409 | { | |
410 | struct gfs2_holder *gh; | |
411 | ||
412 | if (!list_empty(&gl->gl_holders)) { | |
413 | gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, | |
414 | gh_list); | |
415 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
416 | return gh; | |
417 | } | |
418 | return NULL; | |
419 | } | |
420 | ||
e6f85600 BP |
421 | /* |
422 | * gfs2_instantiate - Call the glops instantiate function | |
3c5c67ec | 423 | * @gh: The glock holder |
e6f85600 | 424 | * |
53d69132 | 425 | * Returns: 0 if instantiate was successful, or error. |
e6f85600 | 426 | */ |
f2e70d8f | 427 | int gfs2_instantiate(struct gfs2_holder *gh) |
e6f85600 BP |
428 | { |
429 | struct gfs2_glock *gl = gh->gh_gl; | |
430 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
f2e70d8f BP |
431 | int ret; |
432 | ||
433 | again: | |
434 | if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags)) | |
86c30a01 | 435 | goto done; |
e6f85600 | 436 | |
f2e70d8f BP |
437 | /* |
438 | * Since we unlock the lockref lock, we set a flag to indicate | |
439 | * instantiate is in progress. | |
440 | */ | |
7a92deaa | 441 | if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) { |
f2e70d8f BP |
442 | wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG, |
443 | TASK_UNINTERRUPTIBLE); | |
444 | /* | |
445 | * Here we just waited for a different instantiate to finish. | |
446 | * But that may not have been successful, as when a process | |
447 | * locks an inode glock _before_ it has an actual inode to | |
448 | * instantiate into. So we check again. This process might | |
449 | * have an inode to instantiate, so might be successful. | |
450 | */ | |
451 | goto again; | |
452 | } | |
453 | ||
5f38a4d3 | 454 | ret = glops->go_instantiate(gl); |
f2e70d8f BP |
455 | if (!ret) |
456 | clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags); | |
7a92deaa | 457 | clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags); |
86c30a01 AG |
458 | if (ret) |
459 | return ret; | |
460 | ||
461 | done: | |
462 | if (glops->go_held) | |
463 | return glops->go_held(gh); | |
464 | return 0; | |
e6f85600 BP |
465 | } |
466 | ||
6802e340 SW |
467 | /** |
468 | * do_promote - promote as many requests as possible on the current queue | |
469 | * @gl: The glock | |
470 | * | |
de3e7f97 | 471 | * Returns true on success (i.e., progress was made or there are no waiters). |
6802e340 SW |
472 | */ |
473 | ||
de3e7f97 | 474 | static bool do_promote(struct gfs2_glock *gl) |
6802e340 | 475 | { |
6feaec81 | 476 | struct gfs2_holder *gh, *current_gh; |
6802e340 | 477 | |
ba3e77a4 | 478 | current_gh = find_first_holder(gl); |
6feaec81 | 479 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { |
60d8bae9 | 480 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) |
6802e340 | 481 | continue; |
8f0028fc | 482 | if (!may_grant(gl, current_gh, gh)) { |
17a6ecee | 483 | /* |
565f82b5 BP |
484 | * If we get here, it means we may not grant this |
485 | * holder for some reason. If this holder is at the | |
486 | * head of the list, it means we have a blocked holder | |
de3e7f97 | 487 | * at the head, so return false. |
17a6ecee | 488 | */ |
5a27a43e | 489 | if (list_is_first(&gh->gh_list, &gl->gl_holders)) |
de3e7f97 | 490 | return false; |
17a6ecee BP |
491 | do_error(gl, 0); |
492 | break; | |
493 | } | |
17a6ecee BP |
494 | set_bit(HIF_HOLDER, &gh->gh_iflags); |
495 | trace_gfs2_promote(gh); | |
496 | gfs2_holder_wake(gh); | |
ba3e77a4 | 497 | if (!current_gh) |
0befb851 | 498 | current_gh = gh; |
6802e340 | 499 | } |
de3e7f97 | 500 | return true; |
6802e340 SW |
501 | } |
502 | ||
6802e340 SW |
503 | /** |
504 | * find_first_waiter - find the first gh that's waiting for the glock | |
505 | * @gl: the glock | |
506 | */ | |
507 | ||
508 | static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl) | |
509 | { | |
510 | struct gfs2_holder *gh; | |
511 | ||
512 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { | |
513 | if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
514 | return gh; | |
515 | } | |
516 | return NULL; | |
517 | } | |
518 | ||
519 | /** | |
520 | * state_change - record that the glock is now in a different state | |
521 | * @gl: the glock | |
c551f66c | 522 | * @new_state: the new state |
6802e340 SW |
523 | */ |
524 | ||
525 | static void state_change(struct gfs2_glock *gl, unsigned int new_state) | |
526 | { | |
527 | int held1, held2; | |
528 | ||
529 | held1 = (gl->gl_state != LM_ST_UNLOCKED); | |
530 | held2 = (new_state != LM_ST_UNLOCKED); | |
531 | ||
532 | if (held1 != held2) { | |
e66cf161 | 533 | GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref)); |
6802e340 | 534 | if (held2) |
e66cf161 | 535 | gl->gl_lockref.count++; |
6802e340 | 536 | else |
e66cf161 | 537 | gl->gl_lockref.count--; |
6802e340 | 538 | } |
7cf8dcd3 BP |
539 | if (new_state != gl->gl_target) |
540 | /* shorten our minimum hold time */ | |
541 | gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR, | |
542 | GL_GLOCK_MIN_HOLD); | |
6802e340 SW |
543 | gl->gl_state = new_state; |
544 | gl->gl_tchange = jiffies; | |
545 | } | |
546 | ||
35b6f8fb AG |
547 | static void gfs2_set_demote(struct gfs2_glock *gl) |
548 | { | |
549 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
550 | ||
551 | set_bit(GLF_DEMOTE, &gl->gl_flags); | |
552 | smp_mb(); | |
553 | wake_up(&sdp->sd_async_glock_wait); | |
554 | } | |
555 | ||
6802e340 SW |
556 | static void gfs2_demote_wake(struct gfs2_glock *gl) |
557 | { | |
558 | gl->gl_demote_state = LM_ST_EXCLUSIVE; | |
559 | clear_bit(GLF_DEMOTE, &gl->gl_flags); | |
4e857c58 | 560 | smp_mb__after_atomic(); |
6802e340 SW |
561 | wake_up_bit(&gl->gl_flags, GLF_DEMOTE); |
562 | } | |
563 | ||
564 | /** | |
565 | * finish_xmote - The DLM has replied to one of our lock requests | |
566 | * @gl: The glock | |
567 | * @ret: The status from the DLM | |
568 | * | |
569 | */ | |
570 | ||
571 | static void finish_xmote(struct gfs2_glock *gl, unsigned int ret) | |
572 | { | |
573 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
574 | struct gfs2_holder *gh; | |
575 | unsigned state = ret & LM_OUT_ST_MASK; | |
576 | ||
f3dd1649 | 577 | spin_lock(&gl->gl_lockref.lock); |
63997775 | 578 | trace_gfs2_glock_state_change(gl, state); |
6802e340 SW |
579 | state_change(gl, state); |
580 | gh = find_first_waiter(gl); | |
581 | ||
582 | /* Demote to UN request arrived during demote to SH or DF */ | |
583 | if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) && | |
584 | state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED) | |
585 | gl->gl_target = LM_ST_UNLOCKED; | |
586 | ||
587 | /* Check for state != intended state */ | |
588 | if (unlikely(state != gl->gl_target)) { | |
1fc05c8d AG |
589 | if (gh && (ret & LM_OUT_CANCELED)) |
590 | gfs2_holder_wake(gh); | |
6802e340 SW |
591 | if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) { |
592 | /* move to back of queue and try next entry */ | |
593 | if (ret & LM_OUT_CANCELED) { | |
0b93bac2 | 594 | list_move_tail(&gh->gh_list, &gl->gl_holders); |
6802e340 SW |
595 | gh = find_first_waiter(gl); |
596 | gl->gl_target = gh->gh_state; | |
66fa9912 BP |
597 | if (do_promote(gl)) |
598 | goto out; | |
6802e340 SW |
599 | goto retry; |
600 | } | |
601 | /* Some error or failed "try lock" - report it */ | |
602 | if ((ret & LM_OUT_ERROR) || | |
603 | (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) { | |
604 | gl->gl_target = gl->gl_state; | |
605 | do_error(gl, ret); | |
606 | goto out; | |
607 | } | |
608 | } | |
609 | switch(state) { | |
610 | /* Unlocked due to conversion deadlock, try again */ | |
611 | case LM_ST_UNLOCKED: | |
612 | retry: | |
613 | do_xmote(gl, gh, gl->gl_target); | |
614 | break; | |
615 | /* Conversion fails, unlock and try again */ | |
616 | case LM_ST_SHARED: | |
617 | case LM_ST_DEFERRED: | |
618 | do_xmote(gl, gh, LM_ST_UNLOCKED); | |
619 | break; | |
620 | default: /* Everything else */ | |
e54c78a2 BP |
621 | fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n", |
622 | gl->gl_target, state); | |
6802e340 SW |
623 | GLOCK_BUG_ON(gl, 1); |
624 | } | |
f3dd1649 | 625 | spin_unlock(&gl->gl_lockref.lock); |
6802e340 SW |
626 | return; |
627 | } | |
628 | ||
629 | /* Fast path - we got what we asked for */ | |
630 | if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) | |
631 | gfs2_demote_wake(gl); | |
632 | if (state != LM_ST_UNLOCKED) { | |
633 | if (glops->go_xmote_bh) { | |
de3f906f AG |
634 | int rv; |
635 | ||
f3dd1649 | 636 | spin_unlock(&gl->gl_lockref.lock); |
f68effb3 | 637 | rv = glops->go_xmote_bh(gl); |
f3dd1649 | 638 | spin_lock(&gl->gl_lockref.lock); |
6802e340 SW |
639 | if (rv) { |
640 | do_error(gl, rv); | |
641 | goto out; | |
642 | } | |
643 | } | |
de3f906f | 644 | do_promote(gl); |
6802e340 SW |
645 | } |
646 | out: | |
647 | clear_bit(GLF_LOCK, &gl->gl_flags); | |
f3dd1649 | 648 | spin_unlock(&gl->gl_lockref.lock); |
6802e340 SW |
649 | } |
650 | ||
865cc3e9 BP |
651 | static bool is_system_glock(struct gfs2_glock *gl) |
652 | { | |
653 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
654 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
655 | ||
656 | if (gl == m_ip->i_gl) | |
657 | return true; | |
658 | return false; | |
659 | } | |
660 | ||
6802e340 SW |
661 | /** |
662 | * do_xmote - Calls the DLM to change the state of a lock | |
663 | * @gl: The lock state | |
664 | * @gh: The holder (only for promotes) | |
665 | * @target: The target lock state | |
666 | * | |
667 | */ | |
668 | ||
86934198 BP |
669 | static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, |
670 | unsigned int target) | |
f3dd1649 AG |
671 | __releases(&gl->gl_lockref.lock) |
672 | __acquires(&gl->gl_lockref.lock) | |
6802e340 SW |
673 | { |
674 | const struct gfs2_glock_operations *glops = gl->gl_ops; | |
15562c43 | 675 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
b58bf407 | 676 | unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0); |
6802e340 SW |
677 | int ret; |
678 | ||
601ef0d5 BP |
679 | if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) && |
680 | gh && !(gh->gh_flags & LM_FLAG_NOEXP)) | |
86934198 BP |
681 | goto skip_inval; |
682 | ||
0b93bac2 | 683 | lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP); |
921169ca SW |
684 | GLOCK_BUG_ON(gl, gl->gl_state == target); |
685 | GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target); | |
6802e340 SW |
686 | if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) && |
687 | glops->go_inval) { | |
d99724c3 BP |
688 | /* |
689 | * If another process is already doing the invalidate, let that | |
690 | * finish first. The glock state machine will get back to this | |
691 | * holder again later. | |
692 | */ | |
693 | if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS, | |
694 | &gl->gl_flags)) | |
695 | return; | |
6802e340 SW |
696 | do_error(gl, 0); /* Fail queued try locks */ |
697 | } | |
47a25380 | 698 | gl->gl_req = target; |
a245769f SW |
699 | set_bit(GLF_BLOCKING, &gl->gl_flags); |
700 | if ((gl->gl_req == LM_ST_UNLOCKED) || | |
701 | (gl->gl_state == LM_ST_EXCLUSIVE) || | |
702 | (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB))) | |
703 | clear_bit(GLF_BLOCKING, &gl->gl_flags); | |
f3dd1649 | 704 | spin_unlock(&gl->gl_lockref.lock); |
1c634f94 BP |
705 | if (glops->go_sync) { |
706 | ret = glops->go_sync(gl); | |
707 | /* If we had a problem syncing (due to io errors or whatever, | |
708 | * we should not invalidate the metadata or tell dlm to | |
709 | * release the glock to other nodes. | |
710 | */ | |
711 | if (ret) { | |
712 | if (cmpxchg(&sdp->sd_log_error, 0, ret)) { | |
713 | fs_err(sdp, "Error %d syncing glock \n", ret); | |
714 | gfs2_dump_glock(NULL, gl, true); | |
715 | } | |
b11e1a84 | 716 | goto skip_inval; |
1c634f94 BP |
717 | } |
718 | } | |
33dbd1e4 BP |
719 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) { |
720 | /* | |
721 | * The call to go_sync should have cleared out the ail list. | |
722 | * If there are still items, we have a problem. We ought to | |
723 | * withdraw, but we can't because the withdraw code also uses | |
724 | * glocks. Warn about the error, dump the glock, then fall | |
725 | * through and wait for logd to do the withdraw for us. | |
726 | */ | |
727 | if ((atomic_read(&gl->gl_ail_count) != 0) && | |
728 | (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) { | |
ea4e61c7 BP |
729 | gfs2_glock_assert_warn(gl, |
730 | !atomic_read(&gl->gl_ail_count)); | |
33dbd1e4 BP |
731 | gfs2_dump_glock(NULL, gl, true); |
732 | } | |
6802e340 | 733 | glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA); |
33dbd1e4 BP |
734 | clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags); |
735 | } | |
6802e340 | 736 | |
b11e1a84 | 737 | skip_inval: |
6802e340 | 738 | gfs2_glock_hold(gl); |
d93ae386 BP |
739 | /* |
740 | * Check for an error encountered since we called go_sync and go_inval. | |
741 | * If so, we can't withdraw from the glock code because the withdraw | |
742 | * code itself uses glocks (see function signal_our_withdraw) to | |
743 | * change the mount to read-only. Most importantly, we must not call | |
744 | * dlm to unlock the glock until the journal is in a known good state | |
745 | * (after journal replay) otherwise other nodes may use the object | |
746 | * (rgrp or dinode) and then later, journal replay will corrupt the | |
747 | * file system. The best we can do here is wait for the logd daemon | |
748 | * to see sd_log_error and withdraw, and in the meantime, requeue the | |
749 | * work for later. | |
750 | * | |
865cc3e9 BP |
751 | * We make a special exception for some system glocks, such as the |
752 | * system statfs inode glock, which needs to be granted before the | |
753 | * gfs2_quotad daemon can exit, and that exit needs to finish before | |
754 | * we can unmount the withdrawn file system. | |
755 | * | |
d93ae386 BP |
756 | * However, if we're just unlocking the lock (say, for unmount, when |
757 | * gfs2_gl_hash_clear calls clear_glock) and recovery is complete | |
758 | * then it's okay to tell dlm to unlock it. | |
759 | */ | |
760 | if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp))) | |
761 | gfs2_withdraw_delayed(sdp); | |
865cc3e9 BP |
762 | if (glock_blocked_by_withdraw(gl) && |
763 | (target != LM_ST_UNLOCKED || | |
764 | test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) { | |
765 | if (!is_system_glock(gl)) { | |
86934198 BP |
766 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); /* sets demote */ |
767 | /* | |
768 | * Ordinarily, we would call dlm and its callback would call | |
769 | * finish_xmote, which would call state_change() to the new state. | |
770 | * Since we withdrew, we won't call dlm, so call state_change | |
771 | * manually, but to the UNLOCKED state we desire. | |
772 | */ | |
773 | state_change(gl, LM_ST_UNLOCKED); | |
774 | /* | |
775 | * We skip telling dlm to do the locking, so we won't get a | |
776 | * reply that would otherwise clear GLF_LOCK. So we clear it here. | |
777 | */ | |
778 | clear_bit(GLF_LOCK, &gl->gl_flags); | |
779 | clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags); | |
d93ae386 BP |
780 | gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD); |
781 | goto out; | |
865cc3e9 BP |
782 | } else { |
783 | clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags); | |
d93ae386 BP |
784 | } |
785 | } | |
786 | ||
921169ca SW |
787 | if (sdp->sd_lockstruct.ls_ops->lm_lock) { |
788 | /* lock_dlm */ | |
789 | ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags); | |
3e11e530 BM |
790 | if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED && |
791 | target == LM_ST_UNLOCKED && | |
792 | test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) { | |
793 | finish_xmote(gl, target); | |
6b0c7440 | 794 | gfs2_glock_queue_work(gl, 0); |
d93ae386 | 795 | } else if (ret) { |
e54c78a2 | 796 | fs_err(sdp, "lm_lock ret %d\n", ret); |
eb43e660 | 797 | GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp)); |
dba2d70c | 798 | } |
921169ca SW |
799 | } else { /* lock_nolock */ |
800 | finish_xmote(gl, target); | |
6b0c7440 | 801 | gfs2_glock_queue_work(gl, 0); |
6802e340 | 802 | } |
d93ae386 | 803 | out: |
f3dd1649 | 804 | spin_lock(&gl->gl_lockref.lock); |
6802e340 SW |
805 | } |
806 | ||
6802e340 SW |
807 | /** |
808 | * run_queue - do all outstanding tasks related to a glock | |
809 | * @gl: The glock in question | |
810 | * @nonblock: True if we must not block in run_queue | |
811 | * | |
812 | */ | |
813 | ||
814 | static void run_queue(struct gfs2_glock *gl, const int nonblock) | |
f3dd1649 AG |
815 | __releases(&gl->gl_lockref.lock) |
816 | __acquires(&gl->gl_lockref.lock) | |
6802e340 SW |
817 | { |
818 | struct gfs2_holder *gh = NULL; | |
819 | ||
820 | if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) | |
821 | return; | |
822 | ||
823 | GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)); | |
824 | ||
825 | if (test_bit(GLF_DEMOTE, &gl->gl_flags) && | |
826 | gl->gl_demote_state != gl->gl_state) { | |
827 | if (find_first_holder(gl)) | |
d8348de0 | 828 | goto out_unlock; |
6802e340 SW |
829 | if (nonblock) |
830 | goto out_sched; | |
831 | set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags); | |
265d529c | 832 | GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE); |
6802e340 SW |
833 | gl->gl_target = gl->gl_demote_state; |
834 | } else { | |
835 | if (test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
836 | gfs2_demote_wake(gl); | |
de3e7f97 | 837 | if (do_promote(gl)) |
d8348de0 | 838 | goto out_unlock; |
6802e340 SW |
839 | gh = find_first_waiter(gl); |
840 | gl->gl_target = gh->gh_state; | |
841 | if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) | |
842 | do_error(gl, 0); /* Fail queued try locks */ | |
843 | } | |
844 | do_xmote(gl, gh, gl->gl_target); | |
845 | return; | |
846 | ||
847 | out_sched: | |
7e71c55e | 848 | clear_bit(GLF_LOCK, &gl->gl_flags); |
4e857c58 | 849 | smp_mb__after_atomic(); |
e66cf161 | 850 | gl->gl_lockref.count++; |
6b0c7440 | 851 | __gfs2_glock_queue_work(gl, 0); |
7e71c55e SW |
852 | return; |
853 | ||
d8348de0 | 854 | out_unlock: |
6802e340 | 855 | clear_bit(GLF_LOCK, &gl->gl_flags); |
4e857c58 | 856 | smp_mb__after_atomic(); |
7e71c55e | 857 | return; |
6802e340 SW |
858 | } |
859 | ||
3781ec9e AG |
860 | /** |
861 | * glock_set_object - set the gl_object field of a glock | |
862 | * @gl: the glock | |
863 | * @object: the object | |
864 | */ | |
865 | void glock_set_object(struct gfs2_glock *gl, void *object) | |
866 | { | |
867 | void *prev_object; | |
868 | ||
869 | spin_lock(&gl->gl_lockref.lock); | |
870 | prev_object = gl->gl_object; | |
871 | gl->gl_object = object; | |
872 | spin_unlock(&gl->gl_lockref.lock); | |
873 | if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == NULL)) { | |
874 | pr_warn("glock=%u/%llx\n", | |
875 | gl->gl_name.ln_type, | |
876 | (unsigned long long)gl->gl_name.ln_number); | |
877 | gfs2_dump_glock(NULL, gl, true); | |
878 | } | |
879 | } | |
880 | ||
881 | /** | |
882 | * glock_clear_object - clear the gl_object field of a glock | |
883 | * @gl: the glock | |
9ffa1888 | 884 | * @object: object the glock currently points at |
3781ec9e AG |
885 | */ |
886 | void glock_clear_object(struct gfs2_glock *gl, void *object) | |
887 | { | |
888 | void *prev_object; | |
889 | ||
890 | spin_lock(&gl->gl_lockref.lock); | |
891 | prev_object = gl->gl_object; | |
892 | gl->gl_object = NULL; | |
893 | spin_unlock(&gl->gl_lockref.lock); | |
9ffa1888 | 894 | if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == object)) { |
3781ec9e AG |
895 | pr_warn("glock=%u/%llx\n", |
896 | gl->gl_name.ln_type, | |
897 | (unsigned long long)gl->gl_name.ln_number); | |
898 | gfs2_dump_glock(NULL, gl, true); | |
899 | } | |
900 | } | |
901 | ||
f286d627 AG |
902 | void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation) |
903 | { | |
904 | struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr; | |
905 | ||
906 | if (ri->ri_magic == 0) | |
907 | ri->ri_magic = cpu_to_be32(GFS2_MAGIC); | |
908 | if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC)) | |
909 | ri->ri_generation_deleted = cpu_to_be64(generation); | |
910 | } | |
911 | ||
912 | bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation) | |
913 | { | |
914 | struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr; | |
915 | ||
916 | if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC)) | |
917 | return false; | |
918 | return generation <= be64_to_cpu(ri->ri_generation_deleted); | |
919 | } | |
920 | ||
9e8990de AG |
921 | static void gfs2_glock_poke(struct gfs2_glock *gl) |
922 | { | |
923 | int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP; | |
924 | struct gfs2_holder gh; | |
925 | int error; | |
926 | ||
b016d9a8 | 927 | __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_); |
c07bfb4d | 928 | error = gfs2_glock_nq(&gh); |
9e8990de AG |
929 | if (!error) |
930 | gfs2_glock_dq(&gh); | |
c07bfb4d | 931 | gfs2_holder_uninit(&gh); |
9e8990de AG |
932 | } |
933 | ||
8c7b9262 AG |
934 | static bool gfs2_try_evict(struct gfs2_glock *gl) |
935 | { | |
936 | struct gfs2_inode *ip; | |
937 | bool evicted = false; | |
938 | ||
939 | /* | |
940 | * If there is contention on the iopen glock and we have an inode, try | |
941 | * to grab and release the inode so that it can be evicted. This will | |
942 | * allow the remote node to go ahead and delete the inode without us | |
943 | * having to do it, which will avoid rgrp glock thrashing. | |
944 | * | |
945 | * The remote node is likely still holding the corresponding inode | |
946 | * glock, so it will run before we get to verify that the delete has | |
947 | * happened below. | |
948 | */ | |
949 | spin_lock(&gl->gl_lockref.lock); | |
950 | ip = gl->gl_object; | |
951 | if (ip && !igrab(&ip->i_inode)) | |
952 | ip = NULL; | |
953 | spin_unlock(&gl->gl_lockref.lock); | |
954 | if (ip) { | |
b0dcffd8 | 955 | gl->gl_no_formal_ino = ip->i_no_formal_ino; |
8c7b9262 AG |
956 | set_bit(GIF_DEFERRED_DELETE, &ip->i_flags); |
957 | d_prune_aliases(&ip->i_inode); | |
958 | iput(&ip->i_inode); | |
959 | ||
960 | /* If the inode was evicted, gl->gl_object will now be NULL. */ | |
961 | spin_lock(&gl->gl_lockref.lock); | |
962 | ip = gl->gl_object; | |
9e8990de | 963 | if (ip) { |
8c7b9262 | 964 | clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags); |
f0c0ade8 AG |
965 | if (!igrab(&ip->i_inode)) |
966 | ip = NULL; | |
9e8990de | 967 | } |
8c7b9262 | 968 | spin_unlock(&gl->gl_lockref.lock); |
f0c0ade8 AG |
969 | if (ip) { |
970 | gfs2_glock_poke(ip->i_gl); | |
971 | iput(&ip->i_inode); | |
9e8990de | 972 | } |
8c7b9262 AG |
973 | evicted = !ip; |
974 | } | |
975 | return evicted; | |
976 | } | |
977 | ||
f0e56edc AG |
978 | bool gfs2_queue_try_to_evict(struct gfs2_glock *gl) |
979 | { | |
980 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
981 | ||
982 | if (test_and_set_bit(GLF_TRY_TO_EVICT, &gl->gl_flags)) | |
983 | return false; | |
984 | return queue_delayed_work(sdp->sd_delete_wq, | |
985 | &gl->gl_delete, 0); | |
986 | } | |
987 | ||
988 | static bool gfs2_queue_verify_evict(struct gfs2_glock *gl) | |
989 | { | |
990 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; | |
991 | ||
992 | if (test_and_set_bit(GLF_VERIFY_EVICT, &gl->gl_flags)) | |
993 | return false; | |
994 | return queue_delayed_work(sdp->sd_delete_wq, | |
995 | &gl->gl_delete, 5 * HZ); | |
996 | } | |
997 | ||
b94a170e BM |
998 | static void delete_work_func(struct work_struct *work) |
999 | { | |
a0e3cc65 AG |
1000 | struct delayed_work *dwork = to_delayed_work(work); |
1001 | struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete); | |
15562c43 | 1002 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
ec5ec66b | 1003 | struct inode *inode; |
044b9414 SW |
1004 | u64 no_addr = gl->gl_name.ln_number; |
1005 | ||
f0e56edc | 1006 | if (test_and_clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags)) { |
8c7b9262 AG |
1007 | /* |
1008 | * If we can evict the inode, give the remote node trying to | |
1009 | * delete the inode some time before verifying that the delete | |
1010 | * has happened. Otherwise, if we cause contention on the inode glock | |
1011 | * immediately, the remote node will think that we still have | |
1012 | * the inode in use, and so it will give up waiting. | |
9e8990de AG |
1013 | * |
1014 | * If we can't evict the inode, signal to the remote node that | |
1015 | * the inode is still in use. We'll later try to delete the | |
1016 | * inode locally in gfs2_evict_inode. | |
1017 | * | |
1018 | * FIXME: We only need to verify that the remote node has | |
1019 | * deleted the inode because nodes before this remote delete | |
1020 | * rework won't cooperate. At a later time, when we no longer | |
1021 | * care about compatibility with such nodes, we can skip this | |
1022 | * step entirely. | |
8c7b9262 AG |
1023 | */ |
1024 | if (gfs2_try_evict(gl)) { | |
e7beb8b6 | 1025 | if (test_bit(SDF_KILL, &sdp->sd_flags)) |
6c0246a9 | 1026 | goto out; |
f0e56edc | 1027 | if (gfs2_queue_verify_evict(gl)) |
8c7b9262 | 1028 | return; |
8c7b9262 | 1029 | } |
88f4a9f8 | 1030 | goto out; |
8c7b9262 AG |
1031 | } |
1032 | ||
f0e56edc AG |
1033 | if (test_and_clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags)) { |
1034 | inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino, | |
1035 | GFS2_BLKST_UNLINKED); | |
1036 | if (IS_ERR(inode)) { | |
1037 | if (PTR_ERR(inode) == -EAGAIN && | |
e7beb8b6 | 1038 | !test_bit(SDF_KILL, &sdp->sd_flags) && |
f0e56edc | 1039 | gfs2_queue_verify_evict(gl)) |
c412a97c | 1040 | return; |
f0e56edc AG |
1041 | } else { |
1042 | d_prune_aliases(inode); | |
1043 | iput(inode); | |
1044 | } | |
b94a170e | 1045 | } |
f0e56edc | 1046 | |
88f4a9f8 | 1047 | out: |
b94a170e BM |
1048 | gfs2_glock_put(gl); |
1049 | } | |
1050 | ||
c4f68a13 BM |
1051 | static void glock_work_func(struct work_struct *work) |
1052 | { | |
6802e340 | 1053 | unsigned long delay = 0; |
c4f68a13 | 1054 | struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work); |
6b0c7440 | 1055 | unsigned int drop_refs = 1; |
c4f68a13 | 1056 | |
26bb7505 | 1057 | if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) { |
6802e340 | 1058 | finish_xmote(gl, gl->gl_reply); |
6b0c7440 | 1059 | drop_refs++; |
26bb7505 | 1060 | } |
f3dd1649 | 1061 | spin_lock(&gl->gl_lockref.lock); |
f90e5b5b | 1062 | if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) && |
265d529c SW |
1063 | gl->gl_state != LM_ST_UNLOCKED && |
1064 | gl->gl_demote_state != LM_ST_EXCLUSIVE) { | |
6802e340 | 1065 | unsigned long holdtime, now = jiffies; |
f90e5b5b | 1066 | |
7cf8dcd3 | 1067 | holdtime = gl->gl_tchange + gl->gl_hold_time; |
6802e340 SW |
1068 | if (time_before(now, holdtime)) |
1069 | delay = holdtime - now; | |
f90e5b5b BP |
1070 | |
1071 | if (!delay) { | |
1072 | clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags); | |
35b6f8fb | 1073 | gfs2_set_demote(gl); |
f90e5b5b | 1074 | } |
6802e340 SW |
1075 | } |
1076 | run_queue(gl, 0); | |
6b0c7440 AG |
1077 | if (delay) { |
1078 | /* Keep one glock reference for the work we requeue. */ | |
1079 | drop_refs--; | |
7cf8dcd3 BP |
1080 | if (gl->gl_name.ln_type != LM_TYPE_INODE) |
1081 | delay = 0; | |
6b0c7440 | 1082 | __gfs2_glock_queue_work(gl, delay); |
7cf8dcd3 | 1083 | } |
6b0c7440 AG |
1084 | |
1085 | /* | |
1086 | * Drop the remaining glock references manually here. (Mind that | |
1087 | * __gfs2_glock_queue_work depends on the lockref spinlock begin held | |
1088 | * here as well.) | |
1089 | */ | |
1090 | gl->gl_lockref.count -= drop_refs; | |
1091 | if (!gl->gl_lockref.count) { | |
1092 | __gfs2_glock_put(gl); | |
1093 | return; | |
1094 | } | |
1095 | spin_unlock(&gl->gl_lockref.lock); | |
c4f68a13 BM |
1096 | } |
1097 | ||
0515480a AG |
1098 | static struct gfs2_glock *find_insert_glock(struct lm_lockname *name, |
1099 | struct gfs2_glock *new) | |
1100 | { | |
1101 | struct wait_glock_queue wait; | |
a91323e2 | 1102 | wait_queue_head_t *wq = glock_waitqueue(name); |
0515480a AG |
1103 | struct gfs2_glock *gl; |
1104 | ||
a91323e2 AG |
1105 | wait.name = name; |
1106 | init_wait(&wait.wait); | |
1107 | wait.wait.func = glock_wake_function; | |
1108 | ||
0515480a | 1109 | again: |
a91323e2 | 1110 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
0515480a AG |
1111 | rcu_read_lock(); |
1112 | if (new) { | |
1113 | gl = rhashtable_lookup_get_insert_fast(&gl_hash_table, | |
1114 | &new->gl_node, ht_parms); | |
1115 | if (IS_ERR(gl)) | |
1116 | goto out; | |
1117 | } else { | |
1118 | gl = rhashtable_lookup_fast(&gl_hash_table, | |
1119 | name, ht_parms); | |
1120 | } | |
1121 | if (gl && !lockref_get_not_dead(&gl->gl_lockref)) { | |
1122 | rcu_read_unlock(); | |
1123 | schedule(); | |
1124 | goto again; | |
1125 | } | |
1126 | out: | |
1127 | rcu_read_unlock(); | |
a91323e2 | 1128 | finish_wait(wq, &wait.wait); |
0515480a AG |
1129 | return gl; |
1130 | } | |
1131 | ||
b3b94faa DT |
1132 | /** |
1133 | * gfs2_glock_get() - Get a glock, or create one if one doesn't exist | |
1134 | * @sdp: The GFS2 superblock | |
1135 | * @number: the lock number | |
1136 | * @glops: The glock_operations to use | |
1137 | * @create: If 0, don't create the glock if it doesn't exist | |
1138 | * @glp: the glock is returned here | |
1139 | * | |
1140 | * This does not lock a glock, just finds/creates structures for one. | |
1141 | * | |
1142 | * Returns: errno | |
1143 | */ | |
1144 | ||
cd915493 | 1145 | int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number, |
8fb4b536 | 1146 | const struct gfs2_glock_operations *glops, int create, |
b3b94faa DT |
1147 | struct gfs2_glock **glp) |
1148 | { | |
009d8518 | 1149 | struct super_block *s = sdp->sd_vfs; |
15562c43 BP |
1150 | struct lm_lockname name = { .ln_number = number, |
1151 | .ln_type = glops->go_type, | |
1152 | .ln_sbd = sdp }; | |
0a52aba7 | 1153 | struct gfs2_glock *gl, *tmp; |
009d8518 | 1154 | struct address_space *mapping; |
0a52aba7 | 1155 | int ret = 0; |
b3b94faa | 1156 | |
0515480a AG |
1157 | gl = find_insert_glock(&name, NULL); |
1158 | if (gl) { | |
1159 | *glp = gl; | |
b3b94faa | 1160 | return 0; |
0515480a | 1161 | } |
64d576ba SW |
1162 | if (!create) |
1163 | return -ENOENT; | |
b3b94faa | 1164 | |
11d8b79e KC |
1165 | if (glops->go_flags & GLOF_ASPACE) { |
1166 | struct gfs2_glock_aspace *gla = | |
1167 | kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS); | |
1168 | if (!gla) | |
1169 | return -ENOMEM; | |
1170 | gl = &gla->glock; | |
1171 | } else { | |
1172 | gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS); | |
1173 | if (!gl) | |
1174 | return -ENOMEM; | |
1175 | } | |
dba2d70c | 1176 | memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb)); |
11d8b79e | 1177 | gl->gl_ops = glops; |
dba2d70c DT |
1178 | |
1179 | if (glops->go_flags & GLOF_LVB) { | |
f7be987b | 1180 | gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS); |
4e2f8849 | 1181 | if (!gl->gl_lksb.sb_lvbptr) { |
11d8b79e | 1182 | gfs2_glock_dealloc(&gl->gl_rcu); |
dba2d70c DT |
1183 | return -ENOMEM; |
1184 | } | |
dba2d70c DT |
1185 | } |
1186 | ||
8f05228e | 1187 | atomic_inc(&sdp->sd_glock_disposal); |
88ffbf3e | 1188 | gl->gl_node.next = NULL; |
f2e70d8f | 1189 | gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0; |
b3b94faa | 1190 | gl->gl_name = name; |
515b269d | 1191 | lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass); |
e66cf161 | 1192 | gl->gl_lockref.count = 1; |
b3b94faa | 1193 | gl->gl_state = LM_ST_UNLOCKED; |
6802e340 | 1194 | gl->gl_target = LM_ST_UNLOCKED; |
c4f68a13 | 1195 | gl->gl_demote_state = LM_ST_EXCLUSIVE; |
8b0e1953 | 1196 | gl->gl_dstamp = 0; |
a245769f SW |
1197 | preempt_disable(); |
1198 | /* We use the global stats to estimate the initial per-glock stats */ | |
1199 | gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type]; | |
1200 | preempt_enable(); | |
1201 | gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0; | |
1202 | gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0; | |
c4f68a13 | 1203 | gl->gl_tchange = jiffies; |
ec45d9f5 | 1204 | gl->gl_object = NULL; |
7cf8dcd3 | 1205 | gl->gl_hold_time = GL_GLOCK_DFT_HOLD; |
c4f68a13 | 1206 | INIT_DELAYED_WORK(&gl->gl_work, glock_work_func); |
2ffed529 BP |
1207 | if (gl->gl_name.ln_type == LM_TYPE_IOPEN) |
1208 | INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func); | |
b3b94faa | 1209 | |
009d8518 SW |
1210 | mapping = gfs2_glock2aspace(gl); |
1211 | if (mapping) { | |
1212 | mapping->a_ops = &gfs2_meta_aops; | |
1213 | mapping->host = s->s_bdev->bd_inode; | |
1214 | mapping->flags = 0; | |
1215 | mapping_set_gfp_mask(mapping, GFP_NOFS); | |
252aa6f5 | 1216 | mapping->private_data = NULL; |
009d8518 | 1217 | mapping->writeback_index = 0; |
b3b94faa DT |
1218 | } |
1219 | ||
0515480a | 1220 | tmp = find_insert_glock(&name, gl); |
0a52aba7 | 1221 | if (!tmp) { |
88ffbf3e | 1222 | *glp = gl; |
0a52aba7 | 1223 | goto out; |
b3b94faa | 1224 | } |
0a52aba7 AG |
1225 | if (IS_ERR(tmp)) { |
1226 | ret = PTR_ERR(tmp); | |
1227 | goto out_free; | |
1228 | } | |
0515480a | 1229 | *glp = tmp; |
0a52aba7 AG |
1230 | |
1231 | out_free: | |
11d8b79e | 1232 | gfs2_glock_dealloc(&gl->gl_rcu); |
da7d554f | 1233 | if (atomic_dec_and_test(&sdp->sd_glock_disposal)) |
3c69c437 | 1234 | wake_up(&sdp->sd_kill_wait); |
b3b94faa | 1235 | |
0a52aba7 | 1236 | out: |
88ffbf3e | 1237 | return ret; |
b3b94faa DT |
1238 | } |
1239 | ||
1240 | /** | |
ffd0cd3c | 1241 | * __gfs2_holder_init - initialize a struct gfs2_holder in the default way |
b3b94faa DT |
1242 | * @gl: the glock |
1243 | * @state: the state we're requesting | |
1244 | * @flags: the modifier flags | |
1245 | * @gh: the holder structure | |
1246 | * | |
1247 | */ | |
1248 | ||
b016d9a8 AG |
1249 | void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags, |
1250 | struct gfs2_holder *gh, unsigned long ip) | |
b3b94faa DT |
1251 | { |
1252 | INIT_LIST_HEAD(&gh->gh_list); | |
4ad02083 | 1253 | gh->gh_gl = gfs2_glock_hold(gl); |
b016d9a8 | 1254 | gh->gh_ip = ip; |
b1e058da | 1255 | gh->gh_owner_pid = get_pid(task_pid(current)); |
b3b94faa DT |
1256 | gh->gh_state = state; |
1257 | gh->gh_flags = flags; | |
b3b94faa | 1258 | gh->gh_iflags = 0; |
b3b94faa DT |
1259 | } |
1260 | ||
1261 | /** | |
1262 | * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it | |
1263 | * @state: the state we're requesting | |
1264 | * @flags: the modifier flags | |
1265 | * @gh: the holder structure | |
1266 | * | |
1267 | * Don't mess with the glock. | |
1268 | * | |
1269 | */ | |
1270 | ||
b58bf407 | 1271 | void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh) |
b3b94faa DT |
1272 | { |
1273 | gh->gh_state = state; | |
579b78a4 | 1274 | gh->gh_flags = flags; |
3b8249f6 | 1275 | gh->gh_iflags = 0; |
d29c0afe | 1276 | gh->gh_ip = _RET_IP_; |
30badc95 | 1277 | put_pid(gh->gh_owner_pid); |
1a0eae88 | 1278 | gh->gh_owner_pid = get_pid(task_pid(current)); |
b3b94faa DT |
1279 | } |
1280 | ||
1281 | /** | |
1282 | * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference) | |
1283 | * @gh: the holder structure | |
1284 | * | |
1285 | */ | |
1286 | ||
1287 | void gfs2_holder_uninit(struct gfs2_holder *gh) | |
1288 | { | |
b1e058da | 1289 | put_pid(gh->gh_owner_pid); |
b3b94faa | 1290 | gfs2_glock_put(gh->gh_gl); |
6df9f9a2 | 1291 | gfs2_holder_mark_uninitialized(gh); |
d0dc80db | 1292 | gh->gh_ip = 0; |
b3b94faa DT |
1293 | } |
1294 | ||
01123cf1 AG |
1295 | static void gfs2_glock_update_hold_time(struct gfs2_glock *gl, |
1296 | unsigned long start_time) | |
1297 | { | |
1298 | /* Have we waited longer that a second? */ | |
1299 | if (time_after(jiffies, start_time + HZ)) { | |
1300 | /* Lengthen the minimum hold time. */ | |
1301 | gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR, | |
1302 | GL_GLOCK_MAX_HOLD); | |
1303 | } | |
1304 | } | |
1305 | ||
53d69132 AG |
1306 | /** |
1307 | * gfs2_glock_holder_ready - holder is ready and its error code can be collected | |
1308 | * @gh: the glock holder | |
1309 | * | |
1310 | * Called when a glock holder no longer needs to be waited for because it is | |
1311 | * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has | |
1312 | * failed (gh_error != 0). | |
1313 | */ | |
1314 | ||
1315 | int gfs2_glock_holder_ready(struct gfs2_holder *gh) | |
1316 | { | |
1317 | if (gh->gh_error || (gh->gh_flags & GL_SKIP)) | |
1318 | return gh->gh_error; | |
1319 | gh->gh_error = gfs2_instantiate(gh); | |
1320 | if (gh->gh_error) | |
1321 | gfs2_glock_dq(gh); | |
1322 | return gh->gh_error; | |
1323 | } | |
1324 | ||
07a79049 BP |
1325 | /** |
1326 | * gfs2_glock_wait - wait on a glock acquisition | |
1327 | * @gh: the glock holder | |
1328 | * | |
1329 | * Returns: 0 on success | |
1330 | */ | |
1331 | ||
1332 | int gfs2_glock_wait(struct gfs2_holder *gh) | |
da755fdb | 1333 | { |
01123cf1 | 1334 | unsigned long start_time = jiffies; |
7cf8dcd3 | 1335 | |
6802e340 | 1336 | might_sleep(); |
74316201 | 1337 | wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE); |
01123cf1 | 1338 | gfs2_glock_update_hold_time(gh->gh_gl, start_time); |
53d69132 | 1339 | return gfs2_glock_holder_ready(gh); |
da755fdb SW |
1340 | } |
1341 | ||
ad26967b BP |
1342 | static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs) |
1343 | { | |
1344 | int i; | |
1345 | ||
1346 | for (i = 0; i < num_gh; i++) | |
1347 | if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) | |
1348 | return 1; | |
1349 | return 0; | |
1350 | } | |
1351 | ||
1352 | /** | |
1353 | * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions | |
1354 | * @num_gh: the number of holders in the array | |
1355 | * @ghs: the glock holder array | |
1356 | * | |
1357 | * Returns: 0 on success, meaning all glocks have been granted and are held. | |
1358 | * -ESTALE if the request timed out, meaning all glocks were released, | |
1359 | * and the caller should retry the operation. | |
1360 | */ | |
1361 | ||
1362 | int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs) | |
1363 | { | |
1364 | struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd; | |
1365 | int i, ret = 0, timeout = 0; | |
1366 | unsigned long start_time = jiffies; | |
ad26967b BP |
1367 | |
1368 | might_sleep(); | |
1369 | /* | |
1370 | * Total up the (minimum hold time * 2) of all glocks and use that to | |
1371 | * determine the max amount of time we should wait. | |
1372 | */ | |
1373 | for (i = 0; i < num_gh; i++) | |
1374 | timeout += ghs[i].gh_gl->gl_hold_time << 1; | |
1375 | ||
ad26967b | 1376 | if (!wait_event_timeout(sdp->sd_async_glock_wait, |
bdff777c | 1377 | !glocks_pending(num_gh, ghs), timeout)) { |
ad26967b | 1378 | ret = -ESTALE; /* request timed out. */ |
bdff777c AG |
1379 | goto out; |
1380 | } | |
ad26967b | 1381 | |
ad26967b | 1382 | for (i = 0; i < num_gh; i++) { |
bdff777c | 1383 | struct gfs2_holder *gh = &ghs[i]; |
53d69132 | 1384 | int ret2; |
ad26967b | 1385 | |
bdff777c AG |
1386 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) { |
1387 | gfs2_glock_update_hold_time(gh->gh_gl, | |
1388 | start_time); | |
ad26967b | 1389 | } |
53d69132 | 1390 | ret2 = gfs2_glock_holder_ready(gh); |
ad26967b | 1391 | if (!ret) |
53d69132 | 1392 | ret = ret2; |
ad26967b BP |
1393 | } |
1394 | ||
bdff777c AG |
1395 | out: |
1396 | if (ret) { | |
1397 | for (i = 0; i < num_gh; i++) { | |
1398 | struct gfs2_holder *gh = &ghs[i]; | |
ad26967b | 1399 | |
bdff777c AG |
1400 | gfs2_glock_dq(gh); |
1401 | } | |
1402 | } | |
ad26967b BP |
1403 | return ret; |
1404 | } | |
1405 | ||
b3b94faa | 1406 | /** |
6802e340 SW |
1407 | * handle_callback - process a demote request |
1408 | * @gl: the glock | |
1409 | * @state: the state the caller wants us to change to | |
c551f66c LJ |
1410 | * @delay: zero to demote immediately; otherwise pending demote |
1411 | * @remote: true if this came from a different cluster node | |
b3b94faa | 1412 | * |
6802e340 SW |
1413 | * There are only two requests that we are going to see in actual |
1414 | * practise: LM_ST_SHARED and LM_ST_UNLOCKED | |
b3b94faa DT |
1415 | */ |
1416 | ||
6802e340 | 1417 | static void handle_callback(struct gfs2_glock *gl, unsigned int state, |
81ffbf65 | 1418 | unsigned long delay, bool remote) |
b3b94faa | 1419 | { |
35b6f8fb AG |
1420 | if (delay) |
1421 | set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags); | |
1422 | else | |
1423 | gfs2_set_demote(gl); | |
6802e340 SW |
1424 | if (gl->gl_demote_state == LM_ST_EXCLUSIVE) { |
1425 | gl->gl_demote_state = state; | |
1426 | gl->gl_demote_time = jiffies; | |
6802e340 SW |
1427 | } else if (gl->gl_demote_state != LM_ST_UNLOCKED && |
1428 | gl->gl_demote_state != state) { | |
1429 | gl->gl_demote_state = LM_ST_UNLOCKED; | |
b3b94faa | 1430 | } |
b94a170e | 1431 | if (gl->gl_ops->go_callback) |
81ffbf65 | 1432 | gl->gl_ops->go_callback(gl, remote); |
7bd8b2eb | 1433 | trace_gfs2_demote_rq(gl, remote); |
b3b94faa DT |
1434 | } |
1435 | ||
6802e340 | 1436 | void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...) |
7c52b166 | 1437 | { |
5e69069c | 1438 | struct va_format vaf; |
7c52b166 RP |
1439 | va_list args; |
1440 | ||
1441 | va_start(args, fmt); | |
5e69069c | 1442 | |
6802e340 | 1443 | if (seq) { |
1bb49303 | 1444 | seq_vprintf(seq, fmt, args); |
6802e340 | 1445 | } else { |
5e69069c JP |
1446 | vaf.fmt = fmt; |
1447 | vaf.va = &args; | |
1448 | ||
d77d1b58 | 1449 | pr_err("%pV", &vaf); |
6802e340 | 1450 | } |
5e69069c | 1451 | |
7c52b166 RP |
1452 | va_end(args); |
1453 | } | |
1454 | ||
cbe6d257 AG |
1455 | static inline bool pid_is_meaningful(const struct gfs2_holder *gh) |
1456 | { | |
1457 | if (!(gh->gh_flags & GL_NOPID)) | |
1458 | return true; | |
1459 | if (gh->gh_state == LM_ST_UNLOCKED) | |
1460 | return true; | |
1461 | return false; | |
1462 | } | |
1463 | ||
b3b94faa DT |
1464 | /** |
1465 | * add_to_queue - Add a holder to the wait queue (but look for recursion) | |
1466 | * @gh: the holder structure to add | |
1467 | * | |
6802e340 SW |
1468 | * Eventually we should move the recursive locking trap to a |
1469 | * debugging option or something like that. This is the fast | |
1470 | * path and needs to have the minimum number of distractions. | |
1471 | * | |
b3b94faa DT |
1472 | */ |
1473 | ||
6802e340 | 1474 | static inline void add_to_queue(struct gfs2_holder *gh) |
f3dd1649 AG |
1475 | __releases(&gl->gl_lockref.lock) |
1476 | __acquires(&gl->gl_lockref.lock) | |
b3b94faa DT |
1477 | { |
1478 | struct gfs2_glock *gl = gh->gh_gl; | |
15562c43 | 1479 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
6802e340 SW |
1480 | struct list_head *insert_pt = NULL; |
1481 | struct gfs2_holder *gh2; | |
e5dc76b9 | 1482 | int try_futile = 0; |
b3b94faa | 1483 | |
ad26967b | 1484 | GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL); |
fee852e3 | 1485 | if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags)) |
ad26967b | 1486 | GLOCK_BUG_ON(gl, true); |
190562bd | 1487 | |
6802e340 | 1488 | if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) { |
61444649 | 1489 | if (test_bit(GLF_LOCK, &gl->gl_flags)) { |
8f0028fc | 1490 | struct gfs2_holder *current_gh; |
61444649 | 1491 | |
ba3e77a4 | 1492 | current_gh = find_first_holder(gl); |
8f0028fc | 1493 | try_futile = !may_grant(gl, current_gh, gh); |
61444649 | 1494 | } |
6802e340 SW |
1495 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) |
1496 | goto fail; | |
1497 | } | |
1498 | ||
1499 | list_for_each_entry(gh2, &gl->gl_holders, gh_list) { | |
cbe6d257 AG |
1500 | if (likely(gh2->gh_owner_pid != gh->gh_owner_pid)) |
1501 | continue; | |
1502 | if (gh->gh_gl->gl_ops->go_type == LM_TYPE_FLOCK) | |
1503 | continue; | |
cbe6d257 AG |
1504 | if (!pid_is_meaningful(gh2)) |
1505 | continue; | |
1506 | goto trap_recursive; | |
1507 | } | |
1508 | list_for_each_entry(gh2, &gl->gl_holders, gh_list) { | |
e5dc76b9 BP |
1509 | if (try_futile && |
1510 | !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) { | |
6802e340 SW |
1511 | fail: |
1512 | gh->gh_error = GLR_TRYFAILED; | |
1513 | gfs2_holder_wake(gh); | |
1514 | return; | |
b4c20166 | 1515 | } |
6802e340 SW |
1516 | if (test_bit(HIF_HOLDER, &gh2->gh_iflags)) |
1517 | continue; | |
6802e340 | 1518 | } |
edae38a6 | 1519 | trace_gfs2_glock_queue(gh, 1); |
a245769f SW |
1520 | gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT); |
1521 | gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT); | |
6802e340 SW |
1522 | if (likely(insert_pt == NULL)) { |
1523 | list_add_tail(&gh->gh_list, &gl->gl_holders); | |
6802e340 SW |
1524 | return; |
1525 | } | |
1526 | list_add_tail(&gh->gh_list, insert_pt); | |
0b93bac2 AG |
1527 | spin_unlock(&gl->gl_lockref.lock); |
1528 | if (sdp->sd_lockstruct.ls_ops->lm_cancel) | |
1529 | sdp->sd_lockstruct.ls_ops->lm_cancel(gl); | |
1530 | spin_lock(&gl->gl_lockref.lock); | |
6802e340 | 1531 | return; |
b3b94faa | 1532 | |
6802e340 | 1533 | trap_recursive: |
e54c78a2 BP |
1534 | fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip); |
1535 | fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid)); | |
1536 | fs_err(sdp, "lock type: %d req lock state : %d\n", | |
6802e340 | 1537 | gh2->gh_gl->gl_name.ln_type, gh2->gh_state); |
e54c78a2 BP |
1538 | fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip); |
1539 | fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid)); | |
1540 | fs_err(sdp, "lock type: %d req lock state : %d\n", | |
6802e340 | 1541 | gh->gh_gl->gl_name.ln_type, gh->gh_state); |
3792ce97 | 1542 | gfs2_dump_glock(NULL, gl, true); |
6802e340 | 1543 | BUG(); |
b3b94faa DT |
1544 | } |
1545 | ||
1546 | /** | |
1547 | * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock) | |
1548 | * @gh: the holder structure | |
1549 | * | |
1550 | * if (gh->gh_flags & GL_ASYNC), this never returns an error | |
1551 | * | |
1552 | * Returns: 0, GLR_TRYFAILED, or errno on failure | |
1553 | */ | |
1554 | ||
1555 | int gfs2_glock_nq(struct gfs2_holder *gh) | |
1556 | { | |
1557 | struct gfs2_glock *gl = gh->gh_gl; | |
b3b94faa DT |
1558 | int error = 0; |
1559 | ||
601ef0d5 | 1560 | if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP)) |
b3b94faa | 1561 | return -EIO; |
b3b94faa | 1562 | |
f42ab085 SW |
1563 | if (test_bit(GLF_LRU, &gl->gl_flags)) |
1564 | gfs2_glock_remove_from_lru(gl); | |
1565 | ||
a4e8145e | 1566 | gh->gh_error = 0; |
f3dd1649 | 1567 | spin_lock(&gl->gl_lockref.lock); |
b3b94faa | 1568 | add_to_queue(gh); |
01b172b7 BP |
1569 | if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) && |
1570 | test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) { | |
0809f6ec | 1571 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
01b172b7 | 1572 | gl->gl_lockref.count++; |
6b0c7440 | 1573 | __gfs2_glock_queue_work(gl, 0); |
01b172b7 | 1574 | } |
6802e340 | 1575 | run_queue(gl, 1); |
f3dd1649 | 1576 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa | 1577 | |
6802e340 SW |
1578 | if (!(gh->gh_flags & GL_ASYNC)) |
1579 | error = gfs2_glock_wait(gh); | |
b3b94faa | 1580 | |
b3b94faa DT |
1581 | return error; |
1582 | } | |
1583 | ||
1584 | /** | |
1585 | * gfs2_glock_poll - poll to see if an async request has been completed | |
1586 | * @gh: the holder | |
1587 | * | |
1588 | * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on | |
1589 | */ | |
1590 | ||
1591 | int gfs2_glock_poll(struct gfs2_holder *gh) | |
1592 | { | |
6802e340 | 1593 | return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1; |
b3b94faa DT |
1594 | } |
1595 | ||
dc732906 BP |
1596 | static inline bool needs_demote(struct gfs2_glock *gl) |
1597 | { | |
1598 | return (test_bit(GLF_DEMOTE, &gl->gl_flags) || | |
1599 | test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags)); | |
1600 | } | |
b3b94faa | 1601 | |
dc732906 | 1602 | static void __gfs2_glock_dq(struct gfs2_holder *gh) |
b3b94faa DT |
1603 | { |
1604 | struct gfs2_glock *gl = gh->gh_gl; | |
c4f68a13 | 1605 | unsigned delay = 0; |
6802e340 | 1606 | int fast_path = 0; |
b3b94faa | 1607 | |
ba3e77a4 AG |
1608 | /* |
1609 | * This holder should not be cached, so mark it for demote. | |
1610 | * Note: this should be done before the check for needs_demote | |
1611 | * below. | |
1612 | */ | |
1613 | if (gh->gh_flags & GL_NOCACHE) | |
1614 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); | |
dc732906 | 1615 | |
ba3e77a4 AG |
1616 | list_del_init(&gh->gh_list); |
1617 | clear_bit(HIF_HOLDER, &gh->gh_iflags); | |
1618 | trace_gfs2_glock_queue(gh, 0); | |
b3b94faa | 1619 | |
ba3e77a4 AG |
1620 | /* |
1621 | * If there hasn't been a demote request we are done. | |
1622 | * (Let the remaining holders, if any, keep holding it.) | |
1623 | */ | |
1624 | if (!needs_demote(gl)) { | |
1625 | if (list_empty(&gl->gl_holders)) | |
1626 | fast_path = 1; | |
dc732906 | 1627 | } |
08d73666 | 1628 | |
7881ef3f | 1629 | if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl)) |
4abb6ad9 BP |
1630 | gfs2_glock_add_to_lru(gl); |
1631 | ||
6b0c7440 AG |
1632 | if (unlikely(!fast_path)) { |
1633 | gl->gl_lockref.count++; | |
1634 | if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) && | |
1635 | !test_bit(GLF_DEMOTE, &gl->gl_flags) && | |
1636 | gl->gl_name.ln_type == LM_TYPE_INODE) | |
1637 | delay = gl->gl_hold_time; | |
1638 | __gfs2_glock_queue_work(gl, delay); | |
1639 | } | |
dc732906 BP |
1640 | } |
1641 | ||
1642 | /** | |
1643 | * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock) | |
1644 | * @gh: the glock holder | |
1645 | * | |
1646 | */ | |
1647 | void gfs2_glock_dq(struct gfs2_holder *gh) | |
1648 | { | |
1649 | struct gfs2_glock *gl = gh->gh_gl; | |
6b46a061 | 1650 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
dc732906 BP |
1651 | |
1652 | spin_lock(&gl->gl_lockref.lock); | |
97236ad5 AG |
1653 | if (!gfs2_holder_queued(gh)) { |
1654 | /* | |
1655 | * May have already been dequeued because the locking request | |
1656 | * was GL_ASYNC and it has failed in the meantime. | |
1657 | */ | |
1658 | goto out; | |
1659 | } | |
6b46a061 | 1660 | |
1fc05c8d AG |
1661 | if (list_is_first(&gh->gh_list, &gl->gl_holders) && |
1662 | !test_bit(HIF_HOLDER, &gh->gh_iflags)) { | |
1663 | spin_unlock(&gl->gl_lockref.lock); | |
1664 | gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl); | |
1665 | wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE); | |
1666 | spin_lock(&gl->gl_lockref.lock); | |
1667 | } | |
1668 | ||
6b46a061 AG |
1669 | /* |
1670 | * If we're in the process of file system withdraw, we cannot just | |
1671 | * dequeue any glocks until our journal is recovered, lest we introduce | |
1672 | * file system corruption. We need two exceptions to this rule: We need | |
1673 | * to allow unlocking of nondisk glocks and the glock for our own | |
1674 | * journal that needs recovery. | |
1675 | */ | |
1676 | if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) && | |
1677 | glock_blocked_by_withdraw(gl) && | |
1678 | gh->gh_gl != sdp->sd_jinode_gl) { | |
1679 | sdp->sd_glock_dqs_held++; | |
1680 | spin_unlock(&gl->gl_lockref.lock); | |
1681 | might_sleep(); | |
1682 | wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY, | |
1683 | TASK_UNINTERRUPTIBLE); | |
1684 | spin_lock(&gl->gl_lockref.lock); | |
1685 | } | |
1686 | ||
dc732906 | 1687 | __gfs2_glock_dq(gh); |
97236ad5 | 1688 | out: |
f3dd1649 | 1689 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1690 | } |
1691 | ||
d93cfa98 AD |
1692 | void gfs2_glock_dq_wait(struct gfs2_holder *gh) |
1693 | { | |
1694 | struct gfs2_glock *gl = gh->gh_gl; | |
1695 | gfs2_glock_dq(gh); | |
81e1d450 | 1696 | might_sleep(); |
74316201 | 1697 | wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE); |
d93cfa98 AD |
1698 | } |
1699 | ||
b3b94faa DT |
1700 | /** |
1701 | * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it | |
1702 | * @gh: the holder structure | |
1703 | * | |
1704 | */ | |
1705 | ||
1706 | void gfs2_glock_dq_uninit(struct gfs2_holder *gh) | |
1707 | { | |
1708 | gfs2_glock_dq(gh); | |
1709 | gfs2_holder_uninit(gh); | |
1710 | } | |
1711 | ||
1712 | /** | |
1713 | * gfs2_glock_nq_num - acquire a glock based on lock number | |
1714 | * @sdp: the filesystem | |
1715 | * @number: the lock number | |
1716 | * @glops: the glock operations for the type of glock | |
1717 | * @state: the state to acquire the glock in | |
25985edc | 1718 | * @flags: modifier flags for the acquisition |
b3b94faa DT |
1719 | * @gh: the struct gfs2_holder |
1720 | * | |
1721 | * Returns: errno | |
1722 | */ | |
1723 | ||
cd915493 | 1724 | int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number, |
8fb4b536 | 1725 | const struct gfs2_glock_operations *glops, |
b58bf407 | 1726 | unsigned int state, u16 flags, struct gfs2_holder *gh) |
b3b94faa DT |
1727 | { |
1728 | struct gfs2_glock *gl; | |
1729 | int error; | |
1730 | ||
1731 | error = gfs2_glock_get(sdp, number, glops, CREATE, &gl); | |
1732 | if (!error) { | |
1733 | error = gfs2_glock_nq_init(gl, state, flags, gh); | |
1734 | gfs2_glock_put(gl); | |
1735 | } | |
1736 | ||
1737 | return error; | |
1738 | } | |
1739 | ||
1740 | /** | |
1741 | * glock_compare - Compare two struct gfs2_glock structures for sorting | |
1742 | * @arg_a: the first structure | |
1743 | * @arg_b: the second structure | |
1744 | * | |
1745 | */ | |
1746 | ||
1747 | static int glock_compare(const void *arg_a, const void *arg_b) | |
1748 | { | |
a5e08a9e SW |
1749 | const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a; |
1750 | const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b; | |
1751 | const struct lm_lockname *a = &gh_a->gh_gl->gl_name; | |
1752 | const struct lm_lockname *b = &gh_b->gh_gl->gl_name; | |
b3b94faa DT |
1753 | |
1754 | if (a->ln_number > b->ln_number) | |
a5e08a9e SW |
1755 | return 1; |
1756 | if (a->ln_number < b->ln_number) | |
1757 | return -1; | |
1c0f4872 | 1758 | BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type); |
a5e08a9e | 1759 | return 0; |
b3b94faa DT |
1760 | } |
1761 | ||
1762 | /** | |
44dab005 | 1763 | * nq_m_sync - synchronously acquire more than one glock in deadlock free order |
b3b94faa DT |
1764 | * @num_gh: the number of structures |
1765 | * @ghs: an array of struct gfs2_holder structures | |
c551f66c | 1766 | * @p: placeholder for the holder structure to pass back |
b3b94faa DT |
1767 | * |
1768 | * Returns: 0 on success (all glocks acquired), | |
1769 | * errno on failure (no glocks acquired) | |
1770 | */ | |
1771 | ||
1772 | static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs, | |
1773 | struct gfs2_holder **p) | |
1774 | { | |
1775 | unsigned int x; | |
1776 | int error = 0; | |
1777 | ||
1778 | for (x = 0; x < num_gh; x++) | |
1779 | p[x] = &ghs[x]; | |
1780 | ||
1781 | sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL); | |
1782 | ||
1783 | for (x = 0; x < num_gh; x++) { | |
b3b94faa DT |
1784 | error = gfs2_glock_nq(p[x]); |
1785 | if (error) { | |
1786 | while (x--) | |
1787 | gfs2_glock_dq(p[x]); | |
1788 | break; | |
1789 | } | |
1790 | } | |
1791 | ||
1792 | return error; | |
1793 | } | |
1794 | ||
1795 | /** | |
1796 | * gfs2_glock_nq_m - acquire multiple glocks | |
1797 | * @num_gh: the number of structures | |
1798 | * @ghs: an array of struct gfs2_holder structures | |
1799 | * | |
b3b94faa DT |
1800 | * Returns: 0 on success (all glocks acquired), |
1801 | * errno on failure (no glocks acquired) | |
1802 | */ | |
1803 | ||
1804 | int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs) | |
1805 | { | |
eaf5bd3c SW |
1806 | struct gfs2_holder *tmp[4]; |
1807 | struct gfs2_holder **pph = tmp; | |
b3b94faa DT |
1808 | int error = 0; |
1809 | ||
eaf5bd3c SW |
1810 | switch(num_gh) { |
1811 | case 0: | |
b3b94faa | 1812 | return 0; |
eaf5bd3c | 1813 | case 1: |
b3b94faa | 1814 | return gfs2_glock_nq(ghs); |
eaf5bd3c SW |
1815 | default: |
1816 | if (num_gh <= 4) | |
b3b94faa | 1817 | break; |
6da2ec56 KC |
1818 | pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *), |
1819 | GFP_NOFS); | |
eaf5bd3c SW |
1820 | if (!pph) |
1821 | return -ENOMEM; | |
b3b94faa DT |
1822 | } |
1823 | ||
eaf5bd3c | 1824 | error = nq_m_sync(num_gh, ghs, pph); |
b3b94faa | 1825 | |
eaf5bd3c SW |
1826 | if (pph != tmp) |
1827 | kfree(pph); | |
b3b94faa DT |
1828 | |
1829 | return error; | |
1830 | } | |
1831 | ||
1832 | /** | |
1833 | * gfs2_glock_dq_m - release multiple glocks | |
1834 | * @num_gh: the number of structures | |
1835 | * @ghs: an array of struct gfs2_holder structures | |
1836 | * | |
1837 | */ | |
1838 | ||
1839 | void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs) | |
1840 | { | |
fa1bbdea BP |
1841 | while (num_gh--) |
1842 | gfs2_glock_dq(&ghs[num_gh]); | |
b3b94faa DT |
1843 | } |
1844 | ||
f057f6cd | 1845 | void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state) |
da755fdb | 1846 | { |
c4f68a13 BM |
1847 | unsigned long delay = 0; |
1848 | unsigned long holdtime; | |
1849 | unsigned long now = jiffies; | |
b3b94faa | 1850 | |
f057f6cd | 1851 | gfs2_glock_hold(gl); |
e2c6c8a7 | 1852 | spin_lock(&gl->gl_lockref.lock); |
7cf8dcd3 | 1853 | holdtime = gl->gl_tchange + gl->gl_hold_time; |
e2c6c8a7 | 1854 | if (!list_empty(&gl->gl_holders) && |
7cf8dcd3 | 1855 | gl->gl_name.ln_type == LM_TYPE_INODE) { |
7b5e3d5f SW |
1856 | if (time_before(now, holdtime)) |
1857 | delay = holdtime - now; | |
1858 | if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags)) | |
7cf8dcd3 | 1859 | delay = gl->gl_hold_time; |
7b5e3d5f | 1860 | } |
81ffbf65 | 1861 | handle_callback(gl, state, delay, true); |
6b0c7440 | 1862 | __gfs2_glock_queue_work(gl, delay); |
f3dd1649 | 1863 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1864 | } |
1865 | ||
0809f6ec SW |
1866 | /** |
1867 | * gfs2_should_freeze - Figure out if glock should be frozen | |
1868 | * @gl: The glock in question | |
1869 | * | |
1870 | * Glocks are not frozen if (a) the result of the dlm operation is | |
1871 | * an error, (b) the locking operation was an unlock operation or | |
1872 | * (c) if there is a "noexp" flagged request anywhere in the queue | |
1873 | * | |
1874 | * Returns: 1 if freezing should occur, 0 otherwise | |
1875 | */ | |
1876 | ||
1877 | static int gfs2_should_freeze(const struct gfs2_glock *gl) | |
1878 | { | |
1879 | const struct gfs2_holder *gh; | |
1880 | ||
1881 | if (gl->gl_reply & ~LM_OUT_ST_MASK) | |
1882 | return 0; | |
1883 | if (gl->gl_target == LM_ST_UNLOCKED) | |
1884 | return 0; | |
1885 | ||
1886 | list_for_each_entry(gh, &gl->gl_holders, gh_list) { | |
1887 | if (test_bit(HIF_HOLDER, &gh->gh_iflags)) | |
1888 | continue; | |
1889 | if (LM_FLAG_NOEXP & gh->gh_flags) | |
1890 | return 0; | |
1891 | } | |
1892 | ||
1893 | return 1; | |
1894 | } | |
1895 | ||
b3b94faa | 1896 | /** |
f057f6cd SW |
1897 | * gfs2_glock_complete - Callback used by locking |
1898 | * @gl: Pointer to the glock | |
1899 | * @ret: The return value from the dlm | |
b3b94faa | 1900 | * |
f3dd1649 | 1901 | * The gl_reply field is under the gl_lockref.lock lock so that it is ok |
47a25380 | 1902 | * to use a bitfield shared with other glock state fields. |
b3b94faa DT |
1903 | */ |
1904 | ||
f057f6cd | 1905 | void gfs2_glock_complete(struct gfs2_glock *gl, int ret) |
b3b94faa | 1906 | { |
15562c43 | 1907 | struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct; |
0809f6ec | 1908 | |
f3dd1649 | 1909 | spin_lock(&gl->gl_lockref.lock); |
f057f6cd | 1910 | gl->gl_reply = ret; |
0809f6ec | 1911 | |
e0c2a9aa | 1912 | if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) { |
0809f6ec | 1913 | if (gfs2_should_freeze(gl)) { |
f057f6cd | 1914 | set_bit(GLF_FROZEN, &gl->gl_flags); |
f3dd1649 | 1915 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa | 1916 | return; |
0809f6ec | 1917 | } |
b3b94faa | 1918 | } |
47a25380 | 1919 | |
e66cf161 | 1920 | gl->gl_lockref.count++; |
f057f6cd | 1921 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
6b0c7440 | 1922 | __gfs2_glock_queue_work(gl, 0); |
f3dd1649 | 1923 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
1924 | } |
1925 | ||
4f0f586b ST |
1926 | static int glock_cmp(void *priv, const struct list_head *a, |
1927 | const struct list_head *b) | |
4506a519 SW |
1928 | { |
1929 | struct gfs2_glock *gla, *glb; | |
1930 | ||
1931 | gla = list_entry(a, struct gfs2_glock, gl_lru); | |
1932 | glb = list_entry(b, struct gfs2_glock, gl_lru); | |
1933 | ||
1934 | if (gla->gl_name.ln_number > glb->gl_name.ln_number) | |
1935 | return 1; | |
1936 | if (gla->gl_name.ln_number < glb->gl_name.ln_number) | |
1937 | return -1; | |
1938 | ||
1939 | return 0; | |
1940 | } | |
1941 | ||
1942 | /** | |
1943 | * gfs2_dispose_glock_lru - Demote a list of glocks | |
1944 | * @list: The list to dispose of | |
1945 | * | |
1946 | * Disposing of glocks may involve disk accesses, so that here we sort | |
1947 | * the glocks by number (i.e. disk location of the inodes) so that if | |
1948 | * there are any such accesses, they'll be sent in order (mostly). | |
1949 | * | |
1950 | * Must be called under the lru_lock, but may drop and retake this | |
1951 | * lock. While the lru_lock is dropped, entries may vanish from the | |
1952 | * list, but no new entries will appear on the list (since it is | |
1953 | * private) | |
1954 | */ | |
1955 | ||
1956 | static void gfs2_dispose_glock_lru(struct list_head *list) | |
1957 | __releases(&lru_lock) | |
1958 | __acquires(&lru_lock) | |
1959 | { | |
1960 | struct gfs2_glock *gl; | |
1961 | ||
1962 | list_sort(NULL, list, glock_cmp); | |
1963 | ||
1964 | while(!list_empty(list)) { | |
969183bc | 1965 | gl = list_first_entry(list, struct gfs2_glock, gl_lru); |
4506a519 | 1966 | list_del_init(&gl->gl_lru); |
1ab19c5d | 1967 | clear_bit(GLF_LRU, &gl->gl_flags); |
f3dd1649 | 1968 | if (!spin_trylock(&gl->gl_lockref.lock)) { |
94a09a39 | 1969 | add_back_to_lru: |
e66cf161 | 1970 | list_add(&gl->gl_lru, &lru_list); |
7881ef3f | 1971 | set_bit(GLF_LRU, &gl->gl_flags); |
e66cf161 SW |
1972 | atomic_inc(&lru_count); |
1973 | continue; | |
1974 | } | |
94a09a39 | 1975 | if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) { |
f3dd1649 | 1976 | spin_unlock(&gl->gl_lockref.lock); |
94a09a39 SW |
1977 | goto add_back_to_lru; |
1978 | } | |
e66cf161 | 1979 | gl->gl_lockref.count++; |
4506a519 | 1980 | if (demote_ok(gl)) |
81ffbf65 | 1981 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); |
4506a519 | 1982 | WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags)); |
6b0c7440 | 1983 | __gfs2_glock_queue_work(gl, 0); |
f3dd1649 | 1984 | spin_unlock(&gl->gl_lockref.lock); |
94a09a39 | 1985 | cond_resched_lock(&lru_lock); |
4506a519 SW |
1986 | } |
1987 | } | |
1988 | ||
2a005855 SW |
1989 | /** |
1990 | * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote | |
1991 | * @nr: The number of entries to scan | |
1992 | * | |
4506a519 SW |
1993 | * This function selects the entries on the LRU which are able to |
1994 | * be demoted, and then kicks off the process by calling | |
1995 | * gfs2_dispose_glock_lru() above. | |
2a005855 | 1996 | */ |
b3b94faa | 1997 | |
1ab6c499 | 1998 | static long gfs2_scan_glock_lru(int nr) |
b3b94faa | 1999 | { |
8fb8f70e | 2000 | struct gfs2_glock *gl, *next; |
4506a519 | 2001 | LIST_HEAD(dispose); |
1ab6c499 | 2002 | long freed = 0; |
b3b94faa | 2003 | |
97cc1025 | 2004 | spin_lock(&lru_lock); |
8fb8f70e AG |
2005 | list_for_each_entry_safe(gl, next, &lru_list, gl_lru) { |
2006 | if (nr-- <= 0) | |
2007 | break; | |
97cc1025 | 2008 | /* Test for being demotable */ |
94a09a39 | 2009 | if (!test_bit(GLF_LOCK, &gl->gl_flags)) { |
228804a3 AG |
2010 | if (!spin_trylock(&gl->gl_lockref.lock)) |
2011 | continue; | |
62862485 BP |
2012 | if (gl->gl_lockref.count <= 1 && |
2013 | (gl->gl_state == LM_ST_UNLOCKED || | |
2014 | demote_ok(gl))) { | |
228804a3 AG |
2015 | list_move(&gl->gl_lru, &dispose); |
2016 | atomic_dec(&lru_count); | |
2017 | freed++; | |
2018 | } | |
2019 | spin_unlock(&gl->gl_lockref.lock); | |
97cc1025 | 2020 | } |
b3b94faa | 2021 | } |
4506a519 SW |
2022 | if (!list_empty(&dispose)) |
2023 | gfs2_dispose_glock_lru(&dispose); | |
97cc1025 | 2024 | spin_unlock(&lru_lock); |
1ab6c499 DC |
2025 | |
2026 | return freed; | |
2a005855 SW |
2027 | } |
2028 | ||
1ab6c499 DC |
2029 | static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink, |
2030 | struct shrink_control *sc) | |
2a005855 | 2031 | { |
1ab6c499 DC |
2032 | if (!(sc->gfp_mask & __GFP_FS)) |
2033 | return SHRINK_STOP; | |
2034 | return gfs2_scan_glock_lru(sc->nr_to_scan); | |
2035 | } | |
2a005855 | 2036 | |
1ab6c499 DC |
2037 | static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink, |
2038 | struct shrink_control *sc) | |
2039 | { | |
55f841ce | 2040 | return vfs_pressure_ratio(atomic_read(&lru_count)); |
b3b94faa DT |
2041 | } |
2042 | ||
a304c23c | 2043 | static struct shrinker *glock_shrinker; |
97cc1025 | 2044 | |
b3b94faa | 2045 | /** |
dbffb29d | 2046 | * glock_hash_walk - Call a function for glock in a hash bucket |
b3b94faa DT |
2047 | * @examiner: the function |
2048 | * @sdp: the filesystem | |
b3b94faa | 2049 | * |
98687f42 HX |
2050 | * Note that the function can be called multiple times on the same |
2051 | * object. So the user must ensure that the function can cope with | |
2052 | * that. | |
b3b94faa DT |
2053 | */ |
2054 | ||
88ffbf3e | 2055 | static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp) |
b3b94faa | 2056 | { |
bc015cb8 | 2057 | struct gfs2_glock *gl; |
98687f42 HX |
2058 | struct rhashtable_iter iter; |
2059 | ||
2060 | rhashtable_walk_enter(&gl_hash_table, &iter); | |
2061 | ||
2062 | do { | |
97a6ec4a | 2063 | rhashtable_walk_start(&iter); |
98687f42 | 2064 | |
7427f3bb AG |
2065 | while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) { |
2066 | if (gl->gl_name.ln_sbd == sdp) | |
88ffbf3e | 2067 | examiner(gl); |
7427f3bb | 2068 | } |
98687f42 HX |
2069 | |
2070 | rhashtable_walk_stop(&iter); | |
2071 | } while (cond_resched(), gl == ERR_PTR(-EAGAIN)); | |
2072 | ||
2073 | rhashtable_walk_exit(&iter); | |
bc015cb8 SW |
2074 | } |
2075 | ||
a0e3cc65 AG |
2076 | void gfs2_cancel_delete_work(struct gfs2_glock *gl) |
2077 | { | |
f0e56edc AG |
2078 | clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags); |
2079 | clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags); | |
3056dc46 | 2080 | if (cancel_delayed_work(&gl->gl_delete)) |
a0e3cc65 | 2081 | gfs2_glock_put(gl); |
a0e3cc65 AG |
2082 | } |
2083 | ||
2084 | static void flush_delete_work(struct gfs2_glock *gl) | |
2085 | { | |
2ffed529 | 2086 | if (gl->gl_name.ln_type == LM_TYPE_IOPEN) { |
0247f4e9 AG |
2087 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
2088 | ||
2ffed529 | 2089 | if (cancel_delayed_work(&gl->gl_delete)) { |
0247f4e9 | 2090 | queue_delayed_work(sdp->sd_delete_wq, |
2ffed529 BP |
2091 | &gl->gl_delete, 0); |
2092 | } | |
34244d71 | 2093 | } |
a0e3cc65 AG |
2094 | } |
2095 | ||
2096 | void gfs2_flush_delete_work(struct gfs2_sbd *sdp) | |
2097 | { | |
2098 | glock_hash_walk(flush_delete_work, sdp); | |
0247f4e9 | 2099 | flush_workqueue(sdp->sd_delete_wq); |
a0e3cc65 AG |
2100 | } |
2101 | ||
f057f6cd SW |
2102 | /** |
2103 | * thaw_glock - thaw out a glock which has an unprocessed reply waiting | |
2104 | * @gl: The glock to thaw | |
2105 | * | |
f057f6cd SW |
2106 | */ |
2107 | ||
2108 | static void thaw_glock(struct gfs2_glock *gl) | |
2109 | { | |
7427f3bb AG |
2110 | if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) |
2111 | return; | |
2112 | if (!lockref_get_not_dead(&gl->gl_lockref)) | |
6b0c7440 | 2113 | return; |
6b0c7440 AG |
2114 | set_bit(GLF_REPLY_PENDING, &gl->gl_flags); |
2115 | gfs2_glock_queue_work(gl, 0); | |
f057f6cd SW |
2116 | } |
2117 | ||
b3b94faa DT |
2118 | /** |
2119 | * clear_glock - look at a glock and see if we can free it from glock cache | |
2120 | * @gl: the glock to look at | |
2121 | * | |
2122 | */ | |
2123 | ||
2124 | static void clear_glock(struct gfs2_glock *gl) | |
2125 | { | |
f42ab085 | 2126 | gfs2_glock_remove_from_lru(gl); |
b3b94faa | 2127 | |
f3dd1649 | 2128 | spin_lock(&gl->gl_lockref.lock); |
7427f3bb AG |
2129 | if (!__lockref_is_dead(&gl->gl_lockref)) { |
2130 | gl->gl_lockref.count++; | |
2131 | if (gl->gl_state != LM_ST_UNLOCKED) | |
2132 | handle_callback(gl, LM_ST_UNLOCKED, 0, false); | |
2133 | __gfs2_glock_queue_work(gl, 0); | |
2134 | } | |
f3dd1649 | 2135 | spin_unlock(&gl->gl_lockref.lock); |
b3b94faa DT |
2136 | } |
2137 | ||
f057f6cd SW |
2138 | /** |
2139 | * gfs2_glock_thaw - Thaw any frozen glocks | |
2140 | * @sdp: The super block | |
2141 | * | |
2142 | */ | |
2143 | ||
2144 | void gfs2_glock_thaw(struct gfs2_sbd *sdp) | |
2145 | { | |
bc015cb8 SW |
2146 | glock_hash_walk(thaw_glock, sdp); |
2147 | } | |
f057f6cd | 2148 | |
3792ce97 | 2149 | static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid) |
bc015cb8 | 2150 | { |
f3dd1649 | 2151 | spin_lock(&gl->gl_lockref.lock); |
3792ce97 | 2152 | gfs2_dump_glock(seq, gl, fsid); |
f3dd1649 | 2153 | spin_unlock(&gl->gl_lockref.lock); |
bc015cb8 SW |
2154 | } |
2155 | ||
2156 | static void dump_glock_func(struct gfs2_glock *gl) | |
2157 | { | |
3792ce97 | 2158 | dump_glock(NULL, gl, true); |
f057f6cd SW |
2159 | } |
2160 | ||
053640a7 BP |
2161 | static void withdraw_dq(struct gfs2_glock *gl) |
2162 | { | |
2163 | spin_lock(&gl->gl_lockref.lock); | |
2164 | if (!__lockref_is_dead(&gl->gl_lockref) && | |
2165 | glock_blocked_by_withdraw(gl)) | |
2166 | do_error(gl, LM_OUT_ERROR); /* remove pending waiters */ | |
2167 | spin_unlock(&gl->gl_lockref.lock); | |
2168 | } | |
2169 | ||
2170 | void gfs2_gl_dq_holders(struct gfs2_sbd *sdp) | |
2171 | { | |
2172 | glock_hash_walk(withdraw_dq, sdp); | |
2173 | } | |
2174 | ||
b3b94faa DT |
2175 | /** |
2176 | * gfs2_gl_hash_clear - Empty out the glock hash table | |
2177 | * @sdp: the filesystem | |
b3b94faa | 2178 | * |
1bdad606 | 2179 | * Called when unmounting the filesystem. |
b3b94faa DT |
2180 | */ |
2181 | ||
fefc03bf | 2182 | void gfs2_gl_hash_clear(struct gfs2_sbd *sdp) |
b3b94faa | 2183 | { |
fb6791d1 | 2184 | set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags); |
222cb538 | 2185 | flush_workqueue(glock_workqueue); |
bc015cb8 | 2186 | glock_hash_walk(clear_glock, sdp); |
8f05228e | 2187 | flush_workqueue(glock_workqueue); |
3c69c437 | 2188 | wait_event_timeout(sdp->sd_kill_wait, |
2aba1b5b BP |
2189 | atomic_read(&sdp->sd_glock_disposal) == 0, |
2190 | HZ * 600); | |
bc015cb8 | 2191 | glock_hash_walk(dump_glock_func, sdp); |
b3b94faa DT |
2192 | } |
2193 | ||
6802e340 | 2194 | static const char *state2str(unsigned state) |
04b933f2 | 2195 | { |
6802e340 SW |
2196 | switch(state) { |
2197 | case LM_ST_UNLOCKED: | |
2198 | return "UN"; | |
2199 | case LM_ST_SHARED: | |
2200 | return "SH"; | |
2201 | case LM_ST_DEFERRED: | |
2202 | return "DF"; | |
2203 | case LM_ST_EXCLUSIVE: | |
2204 | return "EX"; | |
2205 | } | |
2206 | return "??"; | |
2207 | } | |
2208 | ||
b58bf407 | 2209 | static const char *hflags2str(char *buf, u16 flags, unsigned long iflags) |
6802e340 SW |
2210 | { |
2211 | char *p = buf; | |
2212 | if (flags & LM_FLAG_TRY) | |
2213 | *p++ = 't'; | |
2214 | if (flags & LM_FLAG_TRY_1CB) | |
2215 | *p++ = 'T'; | |
2216 | if (flags & LM_FLAG_NOEXP) | |
2217 | *p++ = 'e'; | |
2218 | if (flags & LM_FLAG_ANY) | |
f057f6cd | 2219 | *p++ = 'A'; |
06e908cd BP |
2220 | if (flags & LM_FLAG_NODE_SCOPE) |
2221 | *p++ = 'n'; | |
6802e340 SW |
2222 | if (flags & GL_ASYNC) |
2223 | *p++ = 'a'; | |
2224 | if (flags & GL_EXACT) | |
2225 | *p++ = 'E'; | |
6802e340 SW |
2226 | if (flags & GL_NOCACHE) |
2227 | *p++ = 'c'; | |
2228 | if (test_bit(HIF_HOLDER, &iflags)) | |
2229 | *p++ = 'H'; | |
2230 | if (test_bit(HIF_WAIT, &iflags)) | |
2231 | *p++ = 'W'; | |
4c69038d BP |
2232 | if (flags & GL_SKIP) |
2233 | *p++ = 's'; | |
6802e340 SW |
2234 | *p = 0; |
2235 | return buf; | |
04b933f2 RP |
2236 | } |
2237 | ||
b3b94faa DT |
2238 | /** |
2239 | * dump_holder - print information about a glock holder | |
6802e340 | 2240 | * @seq: the seq_file struct |
b3b94faa | 2241 | * @gh: the glock holder |
3792ce97 | 2242 | * @fs_id_buf: pointer to file system id (if requested) |
b3b94faa | 2243 | * |
b3b94faa DT |
2244 | */ |
2245 | ||
3792ce97 BP |
2246 | static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh, |
2247 | const char *fs_id_buf) | |
b3b94faa | 2248 | { |
cbe6d257 AG |
2249 | const char *comm = "(none)"; |
2250 | pid_t owner_pid = 0; | |
6802e340 | 2251 | char flags_buf[32]; |
b3b94faa | 2252 | |
0b3a2c99 | 2253 | rcu_read_lock(); |
cbe6d257 AG |
2254 | if (pid_is_meaningful(gh)) { |
2255 | struct task_struct *gh_owner; | |
2256 | ||
2257 | comm = "(ended)"; | |
2258 | owner_pid = pid_nr(gh->gh_owner_pid); | |
b1e058da | 2259 | gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID); |
cbe6d257 AG |
2260 | if (gh_owner) |
2261 | comm = gh_owner->comm; | |
2262 | } | |
3792ce97 BP |
2263 | gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n", |
2264 | fs_id_buf, state2str(gh->gh_state), | |
cc18152e | 2265 | hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags), |
cbe6d257 | 2266 | gh->gh_error, (long)owner_pid, comm, (void *)gh->gh_ip); |
0b3a2c99 | 2267 | rcu_read_unlock(); |
b3b94faa DT |
2268 | } |
2269 | ||
627c10b7 | 2270 | static const char *gflags2str(char *buf, const struct gfs2_glock *gl) |
6802e340 | 2271 | { |
627c10b7 | 2272 | const unsigned long *gflags = &gl->gl_flags; |
6802e340 | 2273 | char *p = buf; |
627c10b7 | 2274 | |
6802e340 SW |
2275 | if (test_bit(GLF_LOCK, gflags)) |
2276 | *p++ = 'l'; | |
6802e340 SW |
2277 | if (test_bit(GLF_DEMOTE, gflags)) |
2278 | *p++ = 'D'; | |
2279 | if (test_bit(GLF_PENDING_DEMOTE, gflags)) | |
2280 | *p++ = 'd'; | |
2281 | if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags)) | |
2282 | *p++ = 'p'; | |
2283 | if (test_bit(GLF_DIRTY, gflags)) | |
2284 | *p++ = 'y'; | |
2285 | if (test_bit(GLF_LFLUSH, gflags)) | |
2286 | *p++ = 'f'; | |
2287 | if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags)) | |
2288 | *p++ = 'i'; | |
2289 | if (test_bit(GLF_REPLY_PENDING, gflags)) | |
2290 | *p++ = 'r'; | |
f057f6cd | 2291 | if (test_bit(GLF_INITIAL, gflags)) |
d8348de0 | 2292 | *p++ = 'I'; |
f057f6cd SW |
2293 | if (test_bit(GLF_FROZEN, gflags)) |
2294 | *p++ = 'F'; | |
e2c6c8a7 | 2295 | if (!list_empty(&gl->gl_holders)) |
7b5e3d5f | 2296 | *p++ = 'q'; |
627c10b7 SW |
2297 | if (test_bit(GLF_LRU, gflags)) |
2298 | *p++ = 'L'; | |
2299 | if (gl->gl_object) | |
2300 | *p++ = 'o'; | |
a245769f SW |
2301 | if (test_bit(GLF_BLOCKING, gflags)) |
2302 | *p++ = 'b'; | |
5deaf1f6 BP |
2303 | if (test_bit(GLF_FREEING, gflags)) |
2304 | *p++ = 'x'; | |
f2e70d8f BP |
2305 | if (test_bit(GLF_INSTANTIATE_NEEDED, gflags)) |
2306 | *p++ = 'n'; | |
2307 | if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags)) | |
2308 | *p++ = 'N'; | |
f0e56edc AG |
2309 | if (test_bit(GLF_TRY_TO_EVICT, gflags)) |
2310 | *p++ = 'e'; | |
2311 | if (test_bit(GLF_VERIFY_EVICT, gflags)) | |
2312 | *p++ = 'E'; | |
6802e340 SW |
2313 | *p = 0; |
2314 | return buf; | |
b3b94faa DT |
2315 | } |
2316 | ||
2317 | /** | |
8eae1ca0 | 2318 | * gfs2_dump_glock - print information about a glock |
6802e340 | 2319 | * @seq: The seq_file struct |
b3b94faa | 2320 | * @gl: the glock |
3792ce97 | 2321 | * @fsid: If true, also dump the file system id |
6802e340 SW |
2322 | * |
2323 | * The file format is as follows: | |
2324 | * One line per object, capital letters are used to indicate objects | |
2325 | * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented, | |
2326 | * other objects are indented by a single space and follow the glock to | |
2327 | * which they are related. Fields are indicated by lower case letters | |
2328 | * followed by a colon and the field value, except for strings which are in | |
2329 | * [] so that its possible to see if they are composed of spaces for | |
2330 | * example. The field's are n = number (id of the object), f = flags, | |
2331 | * t = type, s = state, r = refcount, e = error, p = pid. | |
b3b94faa | 2332 | * |
b3b94faa DT |
2333 | */ |
2334 | ||
3792ce97 | 2335 | void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid) |
b3b94faa | 2336 | { |
6802e340 SW |
2337 | const struct gfs2_glock_operations *glops = gl->gl_ops; |
2338 | unsigned long long dtime; | |
2339 | const struct gfs2_holder *gh; | |
2340 | char gflags_buf[32]; | |
3792ce97 | 2341 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
98fb0574 | 2342 | char fs_id_buf[sizeof(sdp->sd_fsname) + 7]; |
7e901d6e | 2343 | unsigned long nrpages = 0; |
b3b94faa | 2344 | |
7e901d6e BP |
2345 | if (gl->gl_ops->go_flags & GLOF_ASPACE) { |
2346 | struct address_space *mapping = gfs2_glock2aspace(gl); | |
2347 | ||
2348 | nrpages = mapping->nrpages; | |
2349 | } | |
3792ce97 BP |
2350 | memset(fs_id_buf, 0, sizeof(fs_id_buf)); |
2351 | if (fsid && sdp) /* safety precaution */ | |
2352 | sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname); | |
6802e340 SW |
2353 | dtime = jiffies - gl->gl_demote_time; |
2354 | dtime *= 1000000/HZ; /* demote time in uSec */ | |
2355 | if (!test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
2356 | dtime = 0; | |
3792ce97 | 2357 | gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d " |
7e901d6e BP |
2358 | "v:%d r:%d m:%ld p:%lu\n", |
2359 | fs_id_buf, state2str(gl->gl_state), | |
2360 | gl->gl_name.ln_type, | |
2361 | (unsigned long long)gl->gl_name.ln_number, | |
2362 | gflags2str(gflags_buf, gl), | |
2363 | state2str(gl->gl_target), | |
2364 | state2str(gl->gl_demote_state), dtime, | |
2365 | atomic_read(&gl->gl_ail_count), | |
2366 | atomic_read(&gl->gl_revokes), | |
2367 | (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages); | |
b3b94faa | 2368 | |
ac3beb6a | 2369 | list_for_each_entry(gh, &gl->gl_holders, gh_list) |
3792ce97 | 2370 | dump_holder(seq, gh, fs_id_buf); |
ac3beb6a | 2371 | |
6802e340 | 2372 | if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump) |
3792ce97 | 2373 | glops->go_dump(seq, gl, fs_id_buf); |
b3b94faa DT |
2374 | } |
2375 | ||
a245769f SW |
2376 | static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr) |
2377 | { | |
2378 | struct gfs2_glock *gl = iter_ptr; | |
2379 | ||
4d207133 | 2380 | seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n", |
a245769f SW |
2381 | gl->gl_name.ln_type, |
2382 | (unsigned long long)gl->gl_name.ln_number, | |
4d207133 BH |
2383 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT], |
2384 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR], | |
2385 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB], | |
2386 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB], | |
2387 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT], | |
2388 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR], | |
2389 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT], | |
2390 | (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]); | |
a245769f SW |
2391 | return 0; |
2392 | } | |
2393 | ||
2394 | static const char *gfs2_gltype[] = { | |
2395 | "type", | |
2396 | "reserved", | |
2397 | "nondisk", | |
2398 | "inode", | |
2399 | "rgrp", | |
2400 | "meta", | |
2401 | "iopen", | |
2402 | "flock", | |
2403 | "plock", | |
2404 | "quota", | |
2405 | "journal", | |
2406 | }; | |
2407 | ||
2408 | static const char *gfs2_stype[] = { | |
2409 | [GFS2_LKS_SRTT] = "srtt", | |
2410 | [GFS2_LKS_SRTTVAR] = "srttvar", | |
2411 | [GFS2_LKS_SRTTB] = "srttb", | |
2412 | [GFS2_LKS_SRTTVARB] = "srttvarb", | |
2413 | [GFS2_LKS_SIRT] = "sirt", | |
2414 | [GFS2_LKS_SIRTVAR] = "sirtvar", | |
2415 | [GFS2_LKS_DCOUNT] = "dlm", | |
2416 | [GFS2_LKS_QCOUNT] = "queue", | |
2417 | }; | |
2418 | ||
2419 | #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype)) | |
2420 | ||
2421 | static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr) | |
2422 | { | |
81648d04 AG |
2423 | struct gfs2_sbd *sdp = seq->private; |
2424 | loff_t pos = *(loff_t *)iter_ptr; | |
2425 | unsigned index = pos >> 3; | |
2426 | unsigned subindex = pos & 0x07; | |
a245769f SW |
2427 | int i; |
2428 | ||
2429 | if (index == 0 && subindex != 0) | |
2430 | return 0; | |
6802e340 | 2431 | |
a245769f SW |
2432 | seq_printf(seq, "%-10s %8s:", gfs2_gltype[index], |
2433 | (index == 0) ? "cpu": gfs2_stype[subindex]); | |
b3b94faa | 2434 | |
a245769f SW |
2435 | for_each_possible_cpu(i) { |
2436 | const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i); | |
8f7e0a80 AG |
2437 | |
2438 | if (index == 0) | |
2439 | seq_printf(seq, " %15u", i); | |
2440 | else | |
2441 | seq_printf(seq, " %15llu", (unsigned long long)lkstats-> | |
2442 | lkstats[index - 1].stats[subindex]); | |
a245769f SW |
2443 | } |
2444 | seq_putc(seq, '\n'); | |
2445 | return 0; | |
2446 | } | |
8fbbfd21 | 2447 | |
85d1da67 SW |
2448 | int __init gfs2_glock_init(void) |
2449 | { | |
0515480a | 2450 | int i, ret; |
88ffbf3e BP |
2451 | |
2452 | ret = rhashtable_init(&gl_hash_table, &ht_parms); | |
2453 | if (ret < 0) | |
2454 | return ret; | |
8fbbfd21 | 2455 | |
d2115778 | 2456 | glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM | |
58a69cb4 | 2457 | WQ_HIGHPRI | WQ_FREEZABLE, 0); |
88ffbf3e BP |
2458 | if (!glock_workqueue) { |
2459 | rhashtable_destroy(&gl_hash_table); | |
dfc4616d | 2460 | return -ENOMEM; |
88ffbf3e | 2461 | } |
97cc1025 | 2462 | |
a304c23c QZ |
2463 | glock_shrinker = shrinker_alloc(0, "gfs2-glock"); |
2464 | if (!glock_shrinker) { | |
e0d735c1 CY |
2465 | destroy_workqueue(glock_workqueue); |
2466 | rhashtable_destroy(&gl_hash_table); | |
a304c23c | 2467 | return -ENOMEM; |
e0d735c1 | 2468 | } |
c4f68a13 | 2469 | |
a304c23c QZ |
2470 | glock_shrinker->count_objects = gfs2_glock_shrink_count; |
2471 | glock_shrinker->scan_objects = gfs2_glock_shrink_scan; | |
2472 | ||
2473 | shrinker_register(glock_shrinker); | |
2474 | ||
0515480a AG |
2475 | for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++) |
2476 | init_waitqueue_head(glock_wait_table + i); | |
2477 | ||
85d1da67 SW |
2478 | return 0; |
2479 | } | |
2480 | ||
8fbbfd21 SW |
2481 | void gfs2_glock_exit(void) |
2482 | { | |
a304c23c | 2483 | shrinker_free(glock_shrinker); |
88ffbf3e | 2484 | rhashtable_destroy(&gl_hash_table); |
c4f68a13 | 2485 | destroy_workqueue(glock_workqueue); |
8fbbfd21 SW |
2486 | } |
2487 | ||
7ac07fda | 2488 | static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n) |
bc015cb8 | 2489 | { |
3fd5d3ad AG |
2490 | struct gfs2_glock *gl = gi->gl; |
2491 | ||
2492 | if (gl) { | |
2493 | if (n == 0) | |
2494 | return; | |
2495 | if (!lockref_put_not_zero(&gl->gl_lockref)) | |
2496 | gfs2_glock_queue_put(gl); | |
7ac07fda AG |
2497 | } |
2498 | for (;;) { | |
3fd5d3ad AG |
2499 | gl = rhashtable_walk_next(&gi->hti); |
2500 | if (IS_ERR_OR_NULL(gl)) { | |
2501 | if (gl == ERR_PTR(-EAGAIN)) { | |
2502 | n = 1; | |
2503 | continue; | |
7ac07fda | 2504 | } |
3fd5d3ad AG |
2505 | gl = NULL; |
2506 | break; | |
2507 | } | |
2508 | if (gl->gl_name.ln_sbd != gi->sdp) | |
2509 | continue; | |
2510 | if (n <= 1) { | |
2511 | if (!lockref_get_not_dead(&gl->gl_lockref)) | |
2512 | continue; | |
2513 | break; | |
2514 | } else { | |
2515 | if (__lockref_is_dead(&gl->gl_lockref)) | |
2516 | continue; | |
2517 | n--; | |
bc015cb8 | 2518 | } |
14d37564 | 2519 | } |
3fd5d3ad | 2520 | gi->gl = gl; |
7c52b166 RP |
2521 | } |
2522 | ||
6802e340 | 2523 | static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos) |
27c3b415 | 2524 | __acquires(RCU) |
7c52b166 | 2525 | { |
6802e340 | 2526 | struct gfs2_glock_iter *gi = seq->private; |
7ac07fda | 2527 | loff_t n; |
ba1ddcb6 | 2528 | |
7ac07fda AG |
2529 | /* |
2530 | * We can either stay where we are, skip to the next hash table | |
2531 | * entry, or start from the beginning. | |
2532 | */ | |
2533 | if (*pos < gi->last_pos) { | |
2534 | rhashtable_walk_exit(&gi->hti); | |
2535 | rhashtable_walk_enter(&gl_hash_table, &gi->hti); | |
2536 | n = *pos + 1; | |
2537 | } else { | |
2538 | n = *pos - gi->last_pos; | |
2539 | } | |
7c52b166 | 2540 | |
7ac07fda | 2541 | rhashtable_walk_start(&gi->hti); |
7c52b166 | 2542 | |
7ac07fda | 2543 | gfs2_glock_iter_next(gi, n); |
ba1ddcb6 | 2544 | gi->last_pos = *pos; |
6802e340 | 2545 | return gi->gl; |
7c52b166 RP |
2546 | } |
2547 | ||
6802e340 | 2548 | static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr, |
7c52b166 RP |
2549 | loff_t *pos) |
2550 | { | |
6802e340 | 2551 | struct gfs2_glock_iter *gi = seq->private; |
7c52b166 RP |
2552 | |
2553 | (*pos)++; | |
ba1ddcb6 | 2554 | gi->last_pos = *pos; |
7ac07fda | 2555 | gfs2_glock_iter_next(gi, 1); |
6802e340 | 2556 | return gi->gl; |
7c52b166 RP |
2557 | } |
2558 | ||
6802e340 | 2559 | static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr) |
27c3b415 | 2560 | __releases(RCU) |
7c52b166 | 2561 | { |
6802e340 | 2562 | struct gfs2_glock_iter *gi = seq->private; |
bc015cb8 | 2563 | |
88ffbf3e | 2564 | rhashtable_walk_stop(&gi->hti); |
7c52b166 RP |
2565 | } |
2566 | ||
6802e340 | 2567 | static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr) |
7c52b166 | 2568 | { |
3792ce97 | 2569 | dump_glock(seq, iter_ptr, false); |
ac3beb6a | 2570 | return 0; |
7c52b166 RP |
2571 | } |
2572 | ||
a245769f SW |
2573 | static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos) |
2574 | { | |
81648d04 | 2575 | preempt_disable(); |
a245769f SW |
2576 | if (*pos >= GFS2_NR_SBSTATS) |
2577 | return NULL; | |
81648d04 | 2578 | return pos; |
a245769f SW |
2579 | } |
2580 | ||
2581 | static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr, | |
2582 | loff_t *pos) | |
2583 | { | |
a245769f | 2584 | (*pos)++; |
81648d04 | 2585 | if (*pos >= GFS2_NR_SBSTATS) |
a245769f | 2586 | return NULL; |
81648d04 | 2587 | return pos; |
a245769f SW |
2588 | } |
2589 | ||
2590 | static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr) | |
2591 | { | |
2592 | preempt_enable(); | |
2593 | } | |
2594 | ||
4ef29002 | 2595 | static const struct seq_operations gfs2_glock_seq_ops = { |
7c52b166 RP |
2596 | .start = gfs2_glock_seq_start, |
2597 | .next = gfs2_glock_seq_next, | |
2598 | .stop = gfs2_glock_seq_stop, | |
2599 | .show = gfs2_glock_seq_show, | |
2600 | }; | |
2601 | ||
a245769f SW |
2602 | static const struct seq_operations gfs2_glstats_seq_ops = { |
2603 | .start = gfs2_glock_seq_start, | |
2604 | .next = gfs2_glock_seq_next, | |
2605 | .stop = gfs2_glock_seq_stop, | |
2606 | .show = gfs2_glstats_seq_show, | |
2607 | }; | |
2608 | ||
e8a8023e | 2609 | static const struct seq_operations gfs2_sbstats_sops = { |
a245769f SW |
2610 | .start = gfs2_sbstats_seq_start, |
2611 | .next = gfs2_sbstats_seq_next, | |
2612 | .stop = gfs2_sbstats_seq_stop, | |
2613 | .show = gfs2_sbstats_seq_show, | |
2614 | }; | |
2615 | ||
0fe2f1e9 SW |
2616 | #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL) |
2617 | ||
92ecd73a AG |
2618 | static int __gfs2_glocks_open(struct inode *inode, struct file *file, |
2619 | const struct seq_operations *ops) | |
7c52b166 | 2620 | { |
92ecd73a | 2621 | int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter)); |
6802e340 SW |
2622 | if (ret == 0) { |
2623 | struct seq_file *seq = file->private_data; | |
2624 | struct gfs2_glock_iter *gi = seq->private; | |
88ffbf3e | 2625 | |
6802e340 | 2626 | gi->sdp = inode->i_private; |
0fe2f1e9 | 2627 | seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN); |
df5d2f55 | 2628 | if (seq->buf) |
0fe2f1e9 | 2629 | seq->size = GFS2_SEQ_GOODSIZE; |
7ac07fda AG |
2630 | /* |
2631 | * Initially, we are "before" the first hash table entry; the | |
2632 | * first call to rhashtable_walk_next gets us the first entry. | |
2633 | */ | |
2634 | gi->last_pos = -1; | |
88ffbf3e | 2635 | gi->gl = NULL; |
7ac07fda | 2636 | rhashtable_walk_enter(&gl_hash_table, &gi->hti); |
6802e340 SW |
2637 | } |
2638 | return ret; | |
7c52b166 RP |
2639 | } |
2640 | ||
92ecd73a AG |
2641 | static int gfs2_glocks_open(struct inode *inode, struct file *file) |
2642 | { | |
2643 | return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops); | |
2644 | } | |
2645 | ||
88ffbf3e BP |
2646 | static int gfs2_glocks_release(struct inode *inode, struct file *file) |
2647 | { | |
2648 | struct seq_file *seq = file->private_data; | |
2649 | struct gfs2_glock_iter *gi = seq->private; | |
2650 | ||
3fd5d3ad AG |
2651 | if (gi->gl) |
2652 | gfs2_glock_put(gi->gl); | |
7ac07fda | 2653 | rhashtable_walk_exit(&gi->hti); |
88ffbf3e BP |
2654 | return seq_release_private(inode, file); |
2655 | } | |
2656 | ||
a245769f SW |
2657 | static int gfs2_glstats_open(struct inode *inode, struct file *file) |
2658 | { | |
92ecd73a | 2659 | return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops); |
a245769f SW |
2660 | } |
2661 | ||
a245769f SW |
2662 | static const struct file_operations gfs2_glocks_fops = { |
2663 | .owner = THIS_MODULE, | |
2664 | .open = gfs2_glocks_open, | |
2665 | .read = seq_read, | |
2666 | .llseek = seq_lseek, | |
88ffbf3e | 2667 | .release = gfs2_glocks_release, |
a245769f SW |
2668 | }; |
2669 | ||
2670 | static const struct file_operations gfs2_glstats_fops = { | |
7c52b166 | 2671 | .owner = THIS_MODULE, |
a245769f SW |
2672 | .open = gfs2_glstats_open, |
2673 | .read = seq_read, | |
2674 | .llseek = seq_lseek, | |
88ffbf3e | 2675 | .release = gfs2_glocks_release, |
a245769f SW |
2676 | }; |
2677 | ||
4480c27c AG |
2678 | struct gfs2_glockfd_iter { |
2679 | struct super_block *sb; | |
2680 | unsigned int tgid; | |
2681 | struct task_struct *task; | |
2682 | unsigned int fd; | |
2683 | struct file *file; | |
2684 | }; | |
2685 | ||
2686 | static struct task_struct *gfs2_glockfd_next_task(struct gfs2_glockfd_iter *i) | |
2687 | { | |
2688 | struct pid_namespace *ns = task_active_pid_ns(current); | |
2689 | struct pid *pid; | |
2690 | ||
2691 | if (i->task) | |
2692 | put_task_struct(i->task); | |
2693 | ||
2694 | rcu_read_lock(); | |
2695 | retry: | |
2696 | i->task = NULL; | |
2697 | pid = find_ge_pid(i->tgid, ns); | |
2698 | if (pid) { | |
2699 | i->tgid = pid_nr_ns(pid, ns); | |
2700 | i->task = pid_task(pid, PIDTYPE_TGID); | |
2701 | if (!i->task) { | |
2702 | i->tgid++; | |
2703 | goto retry; | |
2704 | } | |
2705 | get_task_struct(i->task); | |
2706 | } | |
2707 | rcu_read_unlock(); | |
2708 | return i->task; | |
2709 | } | |
2710 | ||
2711 | static struct file *gfs2_glockfd_next_file(struct gfs2_glockfd_iter *i) | |
2712 | { | |
2713 | if (i->file) { | |
2714 | fput(i->file); | |
2715 | i->file = NULL; | |
2716 | } | |
2717 | ||
2718 | rcu_read_lock(); | |
2719 | for(;; i->fd++) { | |
2720 | struct inode *inode; | |
2721 | ||
0ede61d8 | 2722 | i->file = task_lookup_next_fdget_rcu(i->task, &i->fd); |
4480c27c AG |
2723 | if (!i->file) { |
2724 | i->fd = 0; | |
2725 | break; | |
2726 | } | |
0ede61d8 | 2727 | |
4480c27c | 2728 | inode = file_inode(i->file); |
0ede61d8 | 2729 | if (inode->i_sb == i->sb) |
4480c27c | 2730 | break; |
0ede61d8 CB |
2731 | |
2732 | rcu_read_unlock(); | |
2733 | fput(i->file); | |
2734 | rcu_read_lock(); | |
4480c27c AG |
2735 | } |
2736 | rcu_read_unlock(); | |
2737 | return i->file; | |
2738 | } | |
2739 | ||
2740 | static void *gfs2_glockfd_seq_start(struct seq_file *seq, loff_t *pos) | |
2741 | { | |
2742 | struct gfs2_glockfd_iter *i = seq->private; | |
2743 | ||
2744 | if (*pos) | |
2745 | return NULL; | |
2746 | while (gfs2_glockfd_next_task(i)) { | |
2747 | if (gfs2_glockfd_next_file(i)) | |
2748 | return i; | |
2749 | i->tgid++; | |
2750 | } | |
2751 | return NULL; | |
2752 | } | |
2753 | ||
2754 | static void *gfs2_glockfd_seq_next(struct seq_file *seq, void *iter_ptr, | |
2755 | loff_t *pos) | |
2756 | { | |
2757 | struct gfs2_glockfd_iter *i = seq->private; | |
2758 | ||
2759 | (*pos)++; | |
2760 | i->fd++; | |
2761 | do { | |
2762 | if (gfs2_glockfd_next_file(i)) | |
2763 | return i; | |
2764 | i->tgid++; | |
2765 | } while (gfs2_glockfd_next_task(i)); | |
2766 | return NULL; | |
2767 | } | |
2768 | ||
2769 | static void gfs2_glockfd_seq_stop(struct seq_file *seq, void *iter_ptr) | |
2770 | { | |
2771 | struct gfs2_glockfd_iter *i = seq->private; | |
2772 | ||
2773 | if (i->file) | |
2774 | fput(i->file); | |
2775 | if (i->task) | |
2776 | put_task_struct(i->task); | |
2777 | } | |
2778 | ||
56535dc6 AG |
2779 | static void gfs2_glockfd_seq_show_flock(struct seq_file *seq, |
2780 | struct gfs2_glockfd_iter *i) | |
2781 | { | |
2782 | struct gfs2_file *fp = i->file->private_data; | |
2783 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; | |
2784 | struct lm_lockname gl_name = { .ln_type = LM_TYPE_RESERVED }; | |
2785 | ||
2786 | if (!READ_ONCE(fl_gh->gh_gl)) | |
2787 | return; | |
2788 | ||
2789 | spin_lock(&i->file->f_lock); | |
2790 | if (gfs2_holder_initialized(fl_gh)) | |
2791 | gl_name = fl_gh->gh_gl->gl_name; | |
2792 | spin_unlock(&i->file->f_lock); | |
2793 | ||
2794 | if (gl_name.ln_type != LM_TYPE_RESERVED) { | |
2795 | seq_printf(seq, "%d %u %u/%llx\n", | |
2796 | i->tgid, i->fd, gl_name.ln_type, | |
2797 | (unsigned long long)gl_name.ln_number); | |
2798 | } | |
2799 | } | |
2800 | ||
4480c27c AG |
2801 | static int gfs2_glockfd_seq_show(struct seq_file *seq, void *iter_ptr) |
2802 | { | |
2803 | struct gfs2_glockfd_iter *i = seq->private; | |
2804 | struct inode *inode = file_inode(i->file); | |
2805 | struct gfs2_glock *gl; | |
2806 | ||
2807 | inode_lock_shared(inode); | |
2808 | gl = GFS2_I(inode)->i_iopen_gh.gh_gl; | |
2809 | if (gl) { | |
2810 | seq_printf(seq, "%d %u %u/%llx\n", | |
2811 | i->tgid, i->fd, gl->gl_name.ln_type, | |
2812 | (unsigned long long)gl->gl_name.ln_number); | |
2813 | } | |
56535dc6 | 2814 | gfs2_glockfd_seq_show_flock(seq, i); |
4480c27c AG |
2815 | inode_unlock_shared(inode); |
2816 | return 0; | |
2817 | } | |
2818 | ||
2819 | static const struct seq_operations gfs2_glockfd_seq_ops = { | |
2820 | .start = gfs2_glockfd_seq_start, | |
2821 | .next = gfs2_glockfd_seq_next, | |
2822 | .stop = gfs2_glockfd_seq_stop, | |
2823 | .show = gfs2_glockfd_seq_show, | |
2824 | }; | |
2825 | ||
2826 | static int gfs2_glockfd_open(struct inode *inode, struct file *file) | |
2827 | { | |
2828 | struct gfs2_glockfd_iter *i; | |
2829 | struct gfs2_sbd *sdp = inode->i_private; | |
2830 | ||
2831 | i = __seq_open_private(file, &gfs2_glockfd_seq_ops, | |
2832 | sizeof(struct gfs2_glockfd_iter)); | |
2833 | if (!i) | |
2834 | return -ENOMEM; | |
2835 | i->sb = sdp->sd_vfs; | |
2836 | return 0; | |
2837 | } | |
2838 | ||
2839 | static const struct file_operations gfs2_glockfd_fops = { | |
2840 | .owner = THIS_MODULE, | |
2841 | .open = gfs2_glockfd_open, | |
2842 | .read = seq_read, | |
2843 | .llseek = seq_lseek, | |
2844 | .release = seq_release_private, | |
2845 | }; | |
2846 | ||
e8a8023e | 2847 | DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats); |
7c52b166 | 2848 | |
2abbf9a4 GKH |
2849 | void gfs2_create_debugfs_file(struct gfs2_sbd *sdp) |
2850 | { | |
2851 | sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root); | |
7c52b166 | 2852 | |
2abbf9a4 GKH |
2853 | debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, |
2854 | &gfs2_glocks_fops); | |
2855 | ||
4480c27c AG |
2856 | debugfs_create_file("glockfd", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, |
2857 | &gfs2_glockfd_fops); | |
2858 | ||
2abbf9a4 GKH |
2859 | debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, |
2860 | &gfs2_glstats_fops); | |
2861 | ||
2862 | debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp, | |
2863 | &gfs2_sbstats_fops); | |
7c52b166 RP |
2864 | } |
2865 | ||
2866 | void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp) | |
2867 | { | |
2abbf9a4 GKH |
2868 | debugfs_remove_recursive(sdp->debugfs_dir); |
2869 | sdp->debugfs_dir = NULL; | |
7c52b166 RP |
2870 | } |
2871 | ||
2abbf9a4 | 2872 | void gfs2_register_debugfs(void) |
7c52b166 RP |
2873 | { |
2874 | gfs2_root = debugfs_create_dir("gfs2", NULL); | |
7c52b166 RP |
2875 | } |
2876 | ||
2877 | void gfs2_unregister_debugfs(void) | |
2878 | { | |
2879 | debugfs_remove(gfs2_root); | |
5f882096 | 2880 | gfs2_root = NULL; |
7c52b166 | 2881 | } |