]>
Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * dlmglue.c | |
5 | * | |
6 | * Code which implements an OCFS2 specific interface to our DLM. | |
7 | * | |
8 | * Copyright (C) 2003, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/types.h> | |
27 | #include <linux/slab.h> | |
28 | #include <linux/highmem.h> | |
29 | #include <linux/mm.h> | |
30 | #include <linux/smp_lock.h> | |
31 | #include <linux/crc32.h> | |
32 | #include <linux/kthread.h> | |
33 | #include <linux/pagemap.h> | |
34 | #include <linux/debugfs.h> | |
35 | #include <linux/seq_file.h> | |
36 | ||
37 | #include <cluster/heartbeat.h> | |
38 | #include <cluster/nodemanager.h> | |
39 | #include <cluster/tcp.h> | |
40 | ||
41 | #include <dlm/dlmapi.h> | |
42 | ||
43 | #define MLOG_MASK_PREFIX ML_DLM_GLUE | |
44 | #include <cluster/masklog.h> | |
45 | ||
46 | #include "ocfs2.h" | |
47 | ||
48 | #include "alloc.h" | |
d680efe9 | 49 | #include "dcache.h" |
ccd979bd MF |
50 | #include "dlmglue.h" |
51 | #include "extent_map.h" | |
52 | #include "heartbeat.h" | |
53 | #include "inode.h" | |
54 | #include "journal.h" | |
55 | #include "slot_map.h" | |
56 | #include "super.h" | |
57 | #include "uptodate.h" | |
58 | #include "vote.h" | |
59 | ||
60 | #include "buffer_head_io.h" | |
61 | ||
62 | struct ocfs2_mask_waiter { | |
63 | struct list_head mw_item; | |
64 | int mw_status; | |
65 | struct completion mw_complete; | |
66 | unsigned long mw_mask; | |
67 | unsigned long mw_goal; | |
68 | }; | |
69 | ||
70 | static void ocfs2_inode_ast_func(void *opaque); | |
71 | static void ocfs2_inode_bast_func(void *opaque, | |
72 | int level); | |
d680efe9 MF |
73 | static void ocfs2_dentry_ast_func(void *opaque); |
74 | static void ocfs2_dentry_bast_func(void *opaque, | |
75 | int level); | |
ccd979bd MF |
76 | static void ocfs2_super_ast_func(void *opaque); |
77 | static void ocfs2_super_bast_func(void *opaque, | |
78 | int level); | |
79 | static void ocfs2_rename_ast_func(void *opaque); | |
80 | static void ocfs2_rename_bast_func(void *opaque, | |
81 | int level); | |
82 | ||
d680efe9 MF |
83 | /* |
84 | * Return value from ocfs2_convert_worker_t functions. | |
85 | * | |
86 | * These control the precise actions of ocfs2_generic_unblock_lock() | |
87 | * and ocfs2_process_blocked_lock() | |
88 | * | |
89 | */ | |
90 | enum ocfs2_unblock_action { | |
91 | UNBLOCK_CONTINUE = 0, /* Continue downconvert */ | |
92 | UNBLOCK_CONTINUE_POST = 1, /* Continue downconvert, fire | |
93 | * ->post_unlock callback */ | |
94 | UNBLOCK_STOP_POST = 2, /* Do not downconvert, fire | |
95 | * ->post_unlock() callback. */ | |
96 | }; | |
97 | ||
98 | struct ocfs2_unblock_ctl { | |
99 | int requeue; | |
100 | enum ocfs2_unblock_action unblock_action; | |
101 | }; | |
102 | ||
ccd979bd MF |
103 | /* so far, all locks have gotten along with the same unlock ast */ |
104 | static void ocfs2_unlock_ast_func(void *opaque, | |
105 | enum dlm_status status); | |
ccd979bd | 106 | static int ocfs2_unblock_meta(struct ocfs2_lock_res *lockres, |
d680efe9 | 107 | struct ocfs2_unblock_ctl *ctl); |
ccd979bd | 108 | static int ocfs2_unblock_data(struct ocfs2_lock_res *lockres, |
d680efe9 | 109 | struct ocfs2_unblock_ctl *ctl); |
ccd979bd | 110 | static int ocfs2_unblock_inode_lock(struct ocfs2_lock_res *lockres, |
d680efe9 MF |
111 | struct ocfs2_unblock_ctl *ctl); |
112 | static int ocfs2_unblock_dentry_lock(struct ocfs2_lock_res *lockres, | |
113 | struct ocfs2_unblock_ctl *ctl); | |
ccd979bd | 114 | static int ocfs2_unblock_osb_lock(struct ocfs2_lock_res *lockres, |
d680efe9 MF |
115 | struct ocfs2_unblock_ctl *ctl); |
116 | ||
117 | static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, | |
118 | struct ocfs2_lock_res *lockres); | |
ccd979bd MF |
119 | |
120 | struct ocfs2_lock_res_ops { | |
121 | void (*ast)(void *); | |
122 | void (*bast)(void *, int); | |
123 | void (*unlock_ast)(void *, enum dlm_status); | |
d680efe9 MF |
124 | int (*unblock)(struct ocfs2_lock_res *, struct ocfs2_unblock_ctl *); |
125 | void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *); | |
ccd979bd MF |
126 | }; |
127 | ||
d680efe9 MF |
128 | typedef int (ocfs2_convert_worker_t)(struct ocfs2_lock_res *, int); |
129 | static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb, | |
130 | struct ocfs2_lock_res *lockres, | |
131 | struct ocfs2_unblock_ctl *ctl, | |
132 | ocfs2_convert_worker_t *worker); | |
133 | ||
ccd979bd MF |
134 | static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = { |
135 | .ast = ocfs2_inode_ast_func, | |
136 | .bast = ocfs2_inode_bast_func, | |
137 | .unlock_ast = ocfs2_unlock_ast_func, | |
138 | .unblock = ocfs2_unblock_inode_lock, | |
139 | }; | |
140 | ||
141 | static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops = { | |
142 | .ast = ocfs2_inode_ast_func, | |
143 | .bast = ocfs2_inode_bast_func, | |
144 | .unlock_ast = ocfs2_unlock_ast_func, | |
145 | .unblock = ocfs2_unblock_meta, | |
146 | }; | |
147 | ||
ccd979bd MF |
148 | static struct ocfs2_lock_res_ops ocfs2_inode_data_lops = { |
149 | .ast = ocfs2_inode_ast_func, | |
150 | .bast = ocfs2_inode_bast_func, | |
151 | .unlock_ast = ocfs2_unlock_ast_func, | |
152 | .unblock = ocfs2_unblock_data, | |
153 | }; | |
154 | ||
155 | static struct ocfs2_lock_res_ops ocfs2_super_lops = { | |
156 | .ast = ocfs2_super_ast_func, | |
157 | .bast = ocfs2_super_bast_func, | |
158 | .unlock_ast = ocfs2_unlock_ast_func, | |
159 | .unblock = ocfs2_unblock_osb_lock, | |
160 | }; | |
161 | ||
162 | static struct ocfs2_lock_res_ops ocfs2_rename_lops = { | |
163 | .ast = ocfs2_rename_ast_func, | |
164 | .bast = ocfs2_rename_bast_func, | |
165 | .unlock_ast = ocfs2_unlock_ast_func, | |
166 | .unblock = ocfs2_unblock_osb_lock, | |
167 | }; | |
168 | ||
d680efe9 MF |
169 | static struct ocfs2_lock_res_ops ocfs2_dentry_lops = { |
170 | .ast = ocfs2_dentry_ast_func, | |
171 | .bast = ocfs2_dentry_bast_func, | |
172 | .unlock_ast = ocfs2_unlock_ast_func, | |
173 | .unblock = ocfs2_unblock_dentry_lock, | |
174 | .post_unlock = ocfs2_dentry_post_unlock, | |
175 | }; | |
176 | ||
ccd979bd MF |
177 | static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres) |
178 | { | |
179 | return lockres->l_type == OCFS2_LOCK_TYPE_META || | |
180 | lockres->l_type == OCFS2_LOCK_TYPE_DATA || | |
181 | lockres->l_type == OCFS2_LOCK_TYPE_RW; | |
182 | } | |
183 | ||
184 | static inline int ocfs2_is_super_lock(struct ocfs2_lock_res *lockres) | |
185 | { | |
186 | return lockres->l_type == OCFS2_LOCK_TYPE_SUPER; | |
187 | } | |
188 | ||
189 | static inline int ocfs2_is_rename_lock(struct ocfs2_lock_res *lockres) | |
190 | { | |
191 | return lockres->l_type == OCFS2_LOCK_TYPE_RENAME; | |
192 | } | |
193 | ||
194 | static inline struct ocfs2_super *ocfs2_lock_res_super(struct ocfs2_lock_res *lockres) | |
195 | { | |
196 | BUG_ON(!ocfs2_is_super_lock(lockres) | |
197 | && !ocfs2_is_rename_lock(lockres)); | |
198 | ||
199 | return (struct ocfs2_super *) lockres->l_priv; | |
200 | } | |
201 | ||
202 | static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres) | |
203 | { | |
204 | BUG_ON(!ocfs2_is_inode_lock(lockres)); | |
205 | ||
206 | return (struct inode *) lockres->l_priv; | |
207 | } | |
208 | ||
d680efe9 MF |
209 | static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres) |
210 | { | |
211 | BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY); | |
212 | ||
213 | return (struct ocfs2_dentry_lock *)lockres->l_priv; | |
214 | } | |
215 | ||
ccd979bd MF |
216 | static int ocfs2_lock_create(struct ocfs2_super *osb, |
217 | struct ocfs2_lock_res *lockres, | |
218 | int level, | |
219 | int dlm_flags); | |
220 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, | |
221 | int wanted); | |
222 | static void ocfs2_cluster_unlock(struct ocfs2_super *osb, | |
223 | struct ocfs2_lock_res *lockres, | |
224 | int level); | |
225 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres); | |
226 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres); | |
227 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres); | |
228 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level); | |
229 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, | |
230 | struct ocfs2_lock_res *lockres); | |
231 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, | |
232 | int convert); | |
233 | #define ocfs2_log_dlm_error(_func, _stat, _lockres) do { \ | |
234 | mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \ | |
235 | "resource %s: %s\n", dlm_errname(_stat), _func, \ | |
236 | _lockres->l_name, dlm_errmsg(_stat)); \ | |
237 | } while (0) | |
238 | static void ocfs2_vote_on_unlock(struct ocfs2_super *osb, | |
239 | struct ocfs2_lock_res *lockres); | |
240 | static int ocfs2_meta_lock_update(struct inode *inode, | |
241 | struct buffer_head **bh); | |
242 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb); | |
243 | static inline int ocfs2_highest_compat_lock_level(int level); | |
244 | static inline int ocfs2_can_downconvert_meta_lock(struct inode *inode, | |
245 | struct ocfs2_lock_res *lockres, | |
246 | int new_level); | |
247 | ||
ccd979bd MF |
248 | static void ocfs2_build_lock_name(enum ocfs2_lock_type type, |
249 | u64 blkno, | |
250 | u32 generation, | |
251 | char *name) | |
252 | { | |
253 | int len; | |
254 | ||
255 | mlog_entry_void(); | |
256 | ||
257 | BUG_ON(type >= OCFS2_NUM_LOCK_TYPES); | |
258 | ||
b0697053 MF |
259 | len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x", |
260 | ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD, | |
261 | (long long)blkno, generation); | |
ccd979bd MF |
262 | |
263 | BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1)); | |
264 | ||
265 | mlog(0, "built lock resource with name: %s\n", name); | |
266 | ||
267 | mlog_exit_void(); | |
268 | } | |
269 | ||
34af946a | 270 | static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock); |
ccd979bd MF |
271 | |
272 | static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res, | |
273 | struct ocfs2_dlm_debug *dlm_debug) | |
274 | { | |
275 | mlog(0, "Add tracking for lockres %s\n", res->l_name); | |
276 | ||
277 | spin_lock(&ocfs2_dlm_tracking_lock); | |
278 | list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking); | |
279 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
280 | } | |
281 | ||
282 | static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res) | |
283 | { | |
284 | spin_lock(&ocfs2_dlm_tracking_lock); | |
285 | if (!list_empty(&res->l_debug_list)) | |
286 | list_del_init(&res->l_debug_list); | |
287 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
288 | } | |
289 | ||
290 | static void ocfs2_lock_res_init_common(struct ocfs2_super *osb, | |
291 | struct ocfs2_lock_res *res, | |
292 | enum ocfs2_lock_type type, | |
ccd979bd MF |
293 | struct ocfs2_lock_res_ops *ops, |
294 | void *priv) | |
295 | { | |
ccd979bd MF |
296 | res->l_type = type; |
297 | res->l_ops = ops; | |
298 | res->l_priv = priv; | |
299 | ||
300 | res->l_level = LKM_IVMODE; | |
301 | res->l_requested = LKM_IVMODE; | |
302 | res->l_blocking = LKM_IVMODE; | |
303 | res->l_action = OCFS2_AST_INVALID; | |
304 | res->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
305 | ||
306 | res->l_flags = OCFS2_LOCK_INITIALIZED; | |
307 | ||
308 | ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug); | |
309 | } | |
310 | ||
311 | void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res) | |
312 | { | |
313 | /* This also clears out the lock status block */ | |
314 | memset(res, 0, sizeof(struct ocfs2_lock_res)); | |
315 | spin_lock_init(&res->l_lock); | |
316 | init_waitqueue_head(&res->l_event); | |
317 | INIT_LIST_HEAD(&res->l_blocked_list); | |
318 | INIT_LIST_HEAD(&res->l_mask_waiters); | |
319 | } | |
320 | ||
321 | void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res, | |
322 | enum ocfs2_lock_type type, | |
24c19ef4 | 323 | unsigned int generation, |
ccd979bd MF |
324 | struct inode *inode) |
325 | { | |
326 | struct ocfs2_lock_res_ops *ops; | |
327 | ||
328 | switch(type) { | |
329 | case OCFS2_LOCK_TYPE_RW: | |
330 | ops = &ocfs2_inode_rw_lops; | |
331 | break; | |
332 | case OCFS2_LOCK_TYPE_META: | |
333 | ops = &ocfs2_inode_meta_lops; | |
334 | break; | |
335 | case OCFS2_LOCK_TYPE_DATA: | |
336 | ops = &ocfs2_inode_data_lops; | |
337 | break; | |
338 | default: | |
339 | mlog_bug_on_msg(1, "type: %d\n", type); | |
340 | ops = NULL; /* thanks, gcc */ | |
341 | break; | |
342 | }; | |
343 | ||
d680efe9 | 344 | ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno, |
24c19ef4 | 345 | generation, res->l_name); |
d680efe9 MF |
346 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode); |
347 | } | |
348 | ||
349 | static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres) | |
350 | { | |
351 | __be64 inode_blkno_be; | |
352 | ||
353 | memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], | |
354 | sizeof(__be64)); | |
355 | ||
356 | return be64_to_cpu(inode_blkno_be); | |
357 | } | |
358 | ||
359 | void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl, | |
360 | u64 parent, struct inode *inode) | |
361 | { | |
362 | int len; | |
363 | u64 inode_blkno = OCFS2_I(inode)->ip_blkno; | |
364 | __be64 inode_blkno_be = cpu_to_be64(inode_blkno); | |
365 | struct ocfs2_lock_res *lockres = &dl->dl_lockres; | |
366 | ||
367 | ocfs2_lock_res_init_once(lockres); | |
368 | ||
369 | /* | |
370 | * Unfortunately, the standard lock naming scheme won't work | |
371 | * here because we have two 16 byte values to use. Instead, | |
372 | * we'll stuff the inode number as a binary value. We still | |
373 | * want error prints to show something without garbling the | |
374 | * display, so drop a null byte in there before the inode | |
375 | * number. A future version of OCFS2 will likely use all | |
376 | * binary lock names. The stringified names have been a | |
377 | * tremendous aid in debugging, but now that the debugfs | |
378 | * interface exists, we can mangle things there if need be. | |
379 | * | |
380 | * NOTE: We also drop the standard "pad" value (the total lock | |
381 | * name size stays the same though - the last part is all | |
382 | * zeros due to the memset in ocfs2_lock_res_init_once() | |
383 | */ | |
384 | len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START, | |
385 | "%c%016llx", | |
386 | ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY), | |
387 | (long long)parent); | |
388 | ||
389 | BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1)); | |
390 | ||
391 | memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be, | |
392 | sizeof(__be64)); | |
393 | ||
394 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres, | |
395 | OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops, | |
396 | dl); | |
ccd979bd MF |
397 | } |
398 | ||
399 | static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res, | |
400 | struct ocfs2_super *osb) | |
401 | { | |
402 | /* Superblock lockres doesn't come from a slab so we call init | |
403 | * once on it manually. */ | |
404 | ocfs2_lock_res_init_once(res); | |
d680efe9 MF |
405 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO, |
406 | 0, res->l_name); | |
ccd979bd | 407 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER, |
ccd979bd MF |
408 | &ocfs2_super_lops, osb); |
409 | } | |
410 | ||
411 | static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res, | |
412 | struct ocfs2_super *osb) | |
413 | { | |
414 | /* Rename lockres doesn't come from a slab so we call init | |
415 | * once on it manually. */ | |
416 | ocfs2_lock_res_init_once(res); | |
d680efe9 MF |
417 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name); |
418 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME, | |
ccd979bd MF |
419 | &ocfs2_rename_lops, osb); |
420 | } | |
421 | ||
422 | void ocfs2_lock_res_free(struct ocfs2_lock_res *res) | |
423 | { | |
424 | mlog_entry_void(); | |
425 | ||
426 | if (!(res->l_flags & OCFS2_LOCK_INITIALIZED)) | |
427 | return; | |
428 | ||
429 | ocfs2_remove_lockres_tracking(res); | |
430 | ||
431 | mlog_bug_on_msg(!list_empty(&res->l_blocked_list), | |
432 | "Lockres %s is on the blocked list\n", | |
433 | res->l_name); | |
434 | mlog_bug_on_msg(!list_empty(&res->l_mask_waiters), | |
435 | "Lockres %s has mask waiters pending\n", | |
436 | res->l_name); | |
437 | mlog_bug_on_msg(spin_is_locked(&res->l_lock), | |
438 | "Lockres %s is locked\n", | |
439 | res->l_name); | |
440 | mlog_bug_on_msg(res->l_ro_holders, | |
441 | "Lockres %s has %u ro holders\n", | |
442 | res->l_name, res->l_ro_holders); | |
443 | mlog_bug_on_msg(res->l_ex_holders, | |
444 | "Lockres %s has %u ex holders\n", | |
445 | res->l_name, res->l_ex_holders); | |
446 | ||
447 | /* Need to clear out the lock status block for the dlm */ | |
448 | memset(&res->l_lksb, 0, sizeof(res->l_lksb)); | |
449 | ||
450 | res->l_flags = 0UL; | |
451 | mlog_exit_void(); | |
452 | } | |
453 | ||
454 | static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres, | |
455 | int level) | |
456 | { | |
457 | mlog_entry_void(); | |
458 | ||
459 | BUG_ON(!lockres); | |
460 | ||
461 | switch(level) { | |
462 | case LKM_EXMODE: | |
463 | lockres->l_ex_holders++; | |
464 | break; | |
465 | case LKM_PRMODE: | |
466 | lockres->l_ro_holders++; | |
467 | break; | |
468 | default: | |
469 | BUG(); | |
470 | } | |
471 | ||
472 | mlog_exit_void(); | |
473 | } | |
474 | ||
475 | static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres, | |
476 | int level) | |
477 | { | |
478 | mlog_entry_void(); | |
479 | ||
480 | BUG_ON(!lockres); | |
481 | ||
482 | switch(level) { | |
483 | case LKM_EXMODE: | |
484 | BUG_ON(!lockres->l_ex_holders); | |
485 | lockres->l_ex_holders--; | |
486 | break; | |
487 | case LKM_PRMODE: | |
488 | BUG_ON(!lockres->l_ro_holders); | |
489 | lockres->l_ro_holders--; | |
490 | break; | |
491 | default: | |
492 | BUG(); | |
493 | } | |
494 | mlog_exit_void(); | |
495 | } | |
496 | ||
497 | /* WARNING: This function lives in a world where the only three lock | |
498 | * levels are EX, PR, and NL. It *will* have to be adjusted when more | |
499 | * lock types are added. */ | |
500 | static inline int ocfs2_highest_compat_lock_level(int level) | |
501 | { | |
502 | int new_level = LKM_EXMODE; | |
503 | ||
504 | if (level == LKM_EXMODE) | |
505 | new_level = LKM_NLMODE; | |
506 | else if (level == LKM_PRMODE) | |
507 | new_level = LKM_PRMODE; | |
508 | return new_level; | |
509 | } | |
510 | ||
511 | static void lockres_set_flags(struct ocfs2_lock_res *lockres, | |
512 | unsigned long newflags) | |
513 | { | |
514 | struct list_head *pos, *tmp; | |
515 | struct ocfs2_mask_waiter *mw; | |
516 | ||
517 | assert_spin_locked(&lockres->l_lock); | |
518 | ||
519 | lockres->l_flags = newflags; | |
520 | ||
521 | list_for_each_safe(pos, tmp, &lockres->l_mask_waiters) { | |
522 | mw = list_entry(pos, struct ocfs2_mask_waiter, mw_item); | |
523 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) | |
524 | continue; | |
525 | ||
526 | list_del_init(&mw->mw_item); | |
527 | mw->mw_status = 0; | |
528 | complete(&mw->mw_complete); | |
529 | } | |
530 | } | |
531 | static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or) | |
532 | { | |
533 | lockres_set_flags(lockres, lockres->l_flags | or); | |
534 | } | |
535 | static void lockres_clear_flags(struct ocfs2_lock_res *lockres, | |
536 | unsigned long clear) | |
537 | { | |
538 | lockres_set_flags(lockres, lockres->l_flags & ~clear); | |
539 | } | |
540 | ||
541 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres) | |
542 | { | |
543 | mlog_entry_void(); | |
544 | ||
545 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); | |
546 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); | |
547 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
548 | BUG_ON(lockres->l_blocking <= LKM_NLMODE); | |
549 | ||
550 | lockres->l_level = lockres->l_requested; | |
551 | if (lockres->l_level <= | |
552 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) { | |
553 | lockres->l_blocking = LKM_NLMODE; | |
554 | lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED); | |
555 | } | |
556 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
557 | ||
558 | mlog_exit_void(); | |
559 | } | |
560 | ||
561 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres) | |
562 | { | |
563 | mlog_entry_void(); | |
564 | ||
565 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); | |
566 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); | |
567 | ||
568 | /* Convert from RO to EX doesn't really need anything as our | |
569 | * information is already up to data. Convert from NL to | |
570 | * *anything* however should mark ourselves as needing an | |
571 | * update */ | |
572 | if (lockres->l_level == LKM_NLMODE) | |
573 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); | |
574 | ||
575 | lockres->l_level = lockres->l_requested; | |
576 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
577 | ||
578 | mlog_exit_void(); | |
579 | } | |
580 | ||
581 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres) | |
582 | { | |
583 | mlog_entry_void(); | |
584 | ||
585 | BUG_ON((!lockres->l_flags & OCFS2_LOCK_BUSY)); | |
586 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); | |
587 | ||
588 | if (lockres->l_requested > LKM_NLMODE && | |
589 | !(lockres->l_flags & OCFS2_LOCK_LOCAL)) | |
590 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); | |
591 | ||
592 | lockres->l_level = lockres->l_requested; | |
593 | lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED); | |
594 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
595 | ||
596 | mlog_exit_void(); | |
597 | } | |
598 | ||
599 | static void ocfs2_inode_ast_func(void *opaque) | |
600 | { | |
601 | struct ocfs2_lock_res *lockres = opaque; | |
602 | struct inode *inode; | |
603 | struct dlm_lockstatus *lksb; | |
604 | unsigned long flags; | |
605 | ||
606 | mlog_entry_void(); | |
607 | ||
608 | inode = ocfs2_lock_res_inode(lockres); | |
609 | ||
b0697053 MF |
610 | mlog(0, "AST fired for inode %llu, l_action = %u, type = %s\n", |
611 | (unsigned long long)OCFS2_I(inode)->ip_blkno, lockres->l_action, | |
ccd979bd MF |
612 | ocfs2_lock_type_string(lockres->l_type)); |
613 | ||
614 | BUG_ON(!ocfs2_is_inode_lock(lockres)); | |
615 | ||
616 | spin_lock_irqsave(&lockres->l_lock, flags); | |
617 | ||
618 | lksb = &(lockres->l_lksb); | |
619 | if (lksb->status != DLM_NORMAL) { | |
620 | mlog(ML_ERROR, "ocfs2_inode_ast_func: lksb status value of %u " | |
b0697053 MF |
621 | "on inode %llu\n", lksb->status, |
622 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
623 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
624 | mlog_exit_void(); | |
625 | return; | |
626 | } | |
627 | ||
628 | switch(lockres->l_action) { | |
629 | case OCFS2_AST_ATTACH: | |
630 | ocfs2_generic_handle_attach_action(lockres); | |
631 | lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL); | |
632 | break; | |
633 | case OCFS2_AST_CONVERT: | |
634 | ocfs2_generic_handle_convert_action(lockres); | |
635 | break; | |
636 | case OCFS2_AST_DOWNCONVERT: | |
637 | ocfs2_generic_handle_downconvert_action(lockres); | |
638 | break; | |
639 | default: | |
640 | mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u " | |
641 | "lockres flags = 0x%lx, unlock action: %u\n", | |
642 | lockres->l_name, lockres->l_action, lockres->l_flags, | |
643 | lockres->l_unlock_action); | |
644 | ||
645 | BUG(); | |
646 | } | |
647 | ||
648 | /* data and rw locking ignores refresh flag for now. */ | |
649 | if (lockres->l_type != OCFS2_LOCK_TYPE_META) | |
650 | lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); | |
651 | ||
652 | /* set it to something invalid so if we get called again we | |
653 | * can catch it. */ | |
654 | lockres->l_action = OCFS2_AST_INVALID; | |
655 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
656 | wake_up(&lockres->l_event); | |
657 | ||
658 | mlog_exit_void(); | |
659 | } | |
660 | ||
661 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, | |
662 | int level) | |
663 | { | |
664 | int needs_downconvert = 0; | |
665 | mlog_entry_void(); | |
666 | ||
667 | assert_spin_locked(&lockres->l_lock); | |
668 | ||
669 | lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); | |
670 | ||
671 | if (level > lockres->l_blocking) { | |
672 | /* only schedule a downconvert if we haven't already scheduled | |
673 | * one that goes low enough to satisfy the level we're | |
674 | * blocking. this also catches the case where we get | |
675 | * duplicate BASTs */ | |
676 | if (ocfs2_highest_compat_lock_level(level) < | |
677 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) | |
678 | needs_downconvert = 1; | |
679 | ||
680 | lockres->l_blocking = level; | |
681 | } | |
682 | ||
683 | mlog_exit(needs_downconvert); | |
684 | return needs_downconvert; | |
685 | } | |
686 | ||
687 | static void ocfs2_generic_bast_func(struct ocfs2_super *osb, | |
688 | struct ocfs2_lock_res *lockres, | |
689 | int level) | |
690 | { | |
691 | int needs_downconvert; | |
692 | unsigned long flags; | |
693 | ||
694 | mlog_entry_void(); | |
695 | ||
696 | BUG_ON(level <= LKM_NLMODE); | |
697 | ||
698 | spin_lock_irqsave(&lockres->l_lock, flags); | |
699 | needs_downconvert = ocfs2_generic_handle_bast(lockres, level); | |
700 | if (needs_downconvert) | |
701 | ocfs2_schedule_blocked_lock(osb, lockres); | |
702 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
703 | ||
d680efe9 MF |
704 | wake_up(&lockres->l_event); |
705 | ||
ccd979bd MF |
706 | ocfs2_kick_vote_thread(osb); |
707 | ||
ccd979bd MF |
708 | mlog_exit_void(); |
709 | } | |
710 | ||
711 | static void ocfs2_inode_bast_func(void *opaque, int level) | |
712 | { | |
713 | struct ocfs2_lock_res *lockres = opaque; | |
714 | struct inode *inode; | |
715 | struct ocfs2_super *osb; | |
716 | ||
717 | mlog_entry_void(); | |
718 | ||
719 | BUG_ON(!ocfs2_is_inode_lock(lockres)); | |
720 | ||
721 | inode = ocfs2_lock_res_inode(lockres); | |
722 | osb = OCFS2_SB(inode->i_sb); | |
723 | ||
b0697053 MF |
724 | mlog(0, "BAST fired for inode %llu, blocking %d, level %d type %s\n", |
725 | (unsigned long long)OCFS2_I(inode)->ip_blkno, level, | |
726 | lockres->l_level, ocfs2_lock_type_string(lockres->l_type)); | |
ccd979bd MF |
727 | |
728 | ocfs2_generic_bast_func(osb, lockres, level); | |
729 | ||
730 | mlog_exit_void(); | |
731 | } | |
732 | ||
733 | static void ocfs2_generic_ast_func(struct ocfs2_lock_res *lockres, | |
734 | int ignore_refresh) | |
735 | { | |
736 | struct dlm_lockstatus *lksb = &lockres->l_lksb; | |
737 | unsigned long flags; | |
738 | ||
739 | spin_lock_irqsave(&lockres->l_lock, flags); | |
740 | ||
741 | if (lksb->status != DLM_NORMAL) { | |
742 | mlog(ML_ERROR, "lockres %s: lksb status value of %u!\n", | |
743 | lockres->l_name, lksb->status); | |
744 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
745 | return; | |
746 | } | |
747 | ||
748 | switch(lockres->l_action) { | |
749 | case OCFS2_AST_ATTACH: | |
750 | ocfs2_generic_handle_attach_action(lockres); | |
751 | break; | |
752 | case OCFS2_AST_CONVERT: | |
753 | ocfs2_generic_handle_convert_action(lockres); | |
754 | break; | |
755 | case OCFS2_AST_DOWNCONVERT: | |
756 | ocfs2_generic_handle_downconvert_action(lockres); | |
757 | break; | |
758 | default: | |
759 | BUG(); | |
760 | } | |
761 | ||
762 | if (ignore_refresh) | |
763 | lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); | |
764 | ||
765 | /* set it to something invalid so if we get called again we | |
766 | * can catch it. */ | |
767 | lockres->l_action = OCFS2_AST_INVALID; | |
ccd979bd MF |
768 | |
769 | wake_up(&lockres->l_event); | |
d680efe9 | 770 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
ccd979bd MF |
771 | } |
772 | ||
773 | static void ocfs2_super_ast_func(void *opaque) | |
774 | { | |
775 | struct ocfs2_lock_res *lockres = opaque; | |
776 | ||
777 | mlog_entry_void(); | |
778 | mlog(0, "Superblock AST fired\n"); | |
779 | ||
780 | BUG_ON(!ocfs2_is_super_lock(lockres)); | |
781 | ocfs2_generic_ast_func(lockres, 0); | |
782 | ||
783 | mlog_exit_void(); | |
784 | } | |
785 | ||
786 | static void ocfs2_super_bast_func(void *opaque, | |
787 | int level) | |
788 | { | |
789 | struct ocfs2_lock_res *lockres = opaque; | |
790 | struct ocfs2_super *osb; | |
791 | ||
792 | mlog_entry_void(); | |
793 | mlog(0, "Superblock BAST fired\n"); | |
794 | ||
795 | BUG_ON(!ocfs2_is_super_lock(lockres)); | |
796 | osb = ocfs2_lock_res_super(lockres); | |
797 | ocfs2_generic_bast_func(osb, lockres, level); | |
798 | ||
799 | mlog_exit_void(); | |
800 | } | |
801 | ||
802 | static void ocfs2_rename_ast_func(void *opaque) | |
803 | { | |
804 | struct ocfs2_lock_res *lockres = opaque; | |
805 | ||
806 | mlog_entry_void(); | |
807 | ||
808 | mlog(0, "Rename AST fired\n"); | |
809 | ||
810 | BUG_ON(!ocfs2_is_rename_lock(lockres)); | |
811 | ||
812 | ocfs2_generic_ast_func(lockres, 1); | |
813 | ||
814 | mlog_exit_void(); | |
815 | } | |
816 | ||
817 | static void ocfs2_rename_bast_func(void *opaque, | |
818 | int level) | |
819 | { | |
820 | struct ocfs2_lock_res *lockres = opaque; | |
821 | struct ocfs2_super *osb; | |
822 | ||
823 | mlog_entry_void(); | |
824 | ||
825 | mlog(0, "Rename BAST fired\n"); | |
826 | ||
827 | BUG_ON(!ocfs2_is_rename_lock(lockres)); | |
828 | ||
829 | osb = ocfs2_lock_res_super(lockres); | |
830 | ocfs2_generic_bast_func(osb, lockres, level); | |
831 | ||
832 | mlog_exit_void(); | |
833 | } | |
834 | ||
d680efe9 MF |
835 | static void ocfs2_dentry_ast_func(void *opaque) |
836 | { | |
837 | struct ocfs2_lock_res *lockres = opaque; | |
838 | ||
839 | BUG_ON(!lockres); | |
840 | ||
841 | ocfs2_generic_ast_func(lockres, 1); | |
842 | } | |
843 | ||
844 | static void ocfs2_dentry_bast_func(void *opaque, int level) | |
845 | { | |
846 | struct ocfs2_lock_res *lockres = opaque; | |
847 | struct ocfs2_dentry_lock *dl = lockres->l_priv; | |
848 | struct ocfs2_super *osb = OCFS2_SB(dl->dl_inode->i_sb); | |
849 | ||
850 | mlog(0, "Dentry bast: level: %d, name: %s\n", level, | |
851 | lockres->l_name); | |
852 | ||
853 | ocfs2_generic_bast_func(osb, lockres, level); | |
854 | } | |
855 | ||
ccd979bd MF |
856 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, |
857 | int convert) | |
858 | { | |
859 | unsigned long flags; | |
860 | ||
861 | mlog_entry_void(); | |
862 | spin_lock_irqsave(&lockres->l_lock, flags); | |
863 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
864 | if (convert) | |
865 | lockres->l_action = OCFS2_AST_INVALID; | |
866 | else | |
867 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
868 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
869 | ||
870 | wake_up(&lockres->l_event); | |
871 | mlog_exit_void(); | |
872 | } | |
873 | ||
874 | /* Note: If we detect another process working on the lock (i.e., | |
875 | * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller | |
876 | * to do the right thing in that case. | |
877 | */ | |
878 | static int ocfs2_lock_create(struct ocfs2_super *osb, | |
879 | struct ocfs2_lock_res *lockres, | |
880 | int level, | |
881 | int dlm_flags) | |
882 | { | |
883 | int ret = 0; | |
884 | enum dlm_status status; | |
885 | unsigned long flags; | |
886 | ||
887 | mlog_entry_void(); | |
888 | ||
889 | mlog(0, "lock %s, level = %d, flags = %d\n", lockres->l_name, level, | |
890 | dlm_flags); | |
891 | ||
892 | spin_lock_irqsave(&lockres->l_lock, flags); | |
893 | if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) || | |
894 | (lockres->l_flags & OCFS2_LOCK_BUSY)) { | |
895 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
896 | goto bail; | |
897 | } | |
898 | ||
899 | lockres->l_action = OCFS2_AST_ATTACH; | |
900 | lockres->l_requested = level; | |
901 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
902 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
903 | ||
904 | status = dlmlock(osb->dlm, | |
905 | level, | |
906 | &lockres->l_lksb, | |
907 | dlm_flags, | |
908 | lockres->l_name, | |
f0681062 | 909 | OCFS2_LOCK_ID_MAX_LEN - 1, |
ccd979bd MF |
910 | lockres->l_ops->ast, |
911 | lockres, | |
912 | lockres->l_ops->bast); | |
913 | if (status != DLM_NORMAL) { | |
914 | ocfs2_log_dlm_error("dlmlock", status, lockres); | |
915 | ret = -EINVAL; | |
916 | ocfs2_recover_from_dlm_error(lockres, 1); | |
917 | } | |
918 | ||
919 | mlog(0, "lock %s, successfull return from dlmlock\n", lockres->l_name); | |
920 | ||
921 | bail: | |
922 | mlog_exit(ret); | |
923 | return ret; | |
924 | } | |
925 | ||
926 | static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres, | |
927 | int flag) | |
928 | { | |
929 | unsigned long flags; | |
930 | int ret; | |
931 | ||
932 | spin_lock_irqsave(&lockres->l_lock, flags); | |
933 | ret = lockres->l_flags & flag; | |
934 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
935 | ||
936 | return ret; | |
937 | } | |
938 | ||
939 | static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres) | |
940 | ||
941 | { | |
942 | wait_event(lockres->l_event, | |
943 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY)); | |
944 | } | |
945 | ||
946 | static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres) | |
947 | ||
948 | { | |
949 | wait_event(lockres->l_event, | |
950 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING)); | |
951 | } | |
952 | ||
953 | /* predict what lock level we'll be dropping down to on behalf | |
954 | * of another node, and return true if the currently wanted | |
955 | * level will be compatible with it. */ | |
956 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, | |
957 | int wanted) | |
958 | { | |
959 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
960 | ||
961 | return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking); | |
962 | } | |
963 | ||
964 | static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw) | |
965 | { | |
966 | INIT_LIST_HEAD(&mw->mw_item); | |
967 | init_completion(&mw->mw_complete); | |
968 | } | |
969 | ||
970 | static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw) | |
971 | { | |
972 | wait_for_completion(&mw->mw_complete); | |
973 | /* Re-arm the completion in case we want to wait on it again */ | |
974 | INIT_COMPLETION(mw->mw_complete); | |
975 | return mw->mw_status; | |
976 | } | |
977 | ||
978 | static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres, | |
979 | struct ocfs2_mask_waiter *mw, | |
980 | unsigned long mask, | |
981 | unsigned long goal) | |
982 | { | |
983 | BUG_ON(!list_empty(&mw->mw_item)); | |
984 | ||
985 | assert_spin_locked(&lockres->l_lock); | |
986 | ||
987 | list_add_tail(&mw->mw_item, &lockres->l_mask_waiters); | |
988 | mw->mw_mask = mask; | |
989 | mw->mw_goal = goal; | |
990 | } | |
991 | ||
992 | /* returns 0 if the mw that was removed was already satisfied, -EBUSY | |
993 | * if the mask still hadn't reached its goal */ | |
994 | static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres, | |
995 | struct ocfs2_mask_waiter *mw) | |
996 | { | |
997 | unsigned long flags; | |
998 | int ret = 0; | |
999 | ||
1000 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1001 | if (!list_empty(&mw->mw_item)) { | |
1002 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) | |
1003 | ret = -EBUSY; | |
1004 | ||
1005 | list_del_init(&mw->mw_item); | |
1006 | init_completion(&mw->mw_complete); | |
1007 | } | |
1008 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1009 | ||
1010 | return ret; | |
1011 | ||
1012 | } | |
1013 | ||
1014 | static int ocfs2_cluster_lock(struct ocfs2_super *osb, | |
1015 | struct ocfs2_lock_res *lockres, | |
1016 | int level, | |
1017 | int lkm_flags, | |
1018 | int arg_flags) | |
1019 | { | |
1020 | struct ocfs2_mask_waiter mw; | |
1021 | enum dlm_status status; | |
1022 | int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR); | |
1023 | int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */ | |
1024 | unsigned long flags; | |
1025 | ||
1026 | mlog_entry_void(); | |
1027 | ||
1028 | ocfs2_init_mask_waiter(&mw); | |
1029 | ||
1030 | again: | |
1031 | wait = 0; | |
1032 | ||
1033 | if (catch_signals && signal_pending(current)) { | |
1034 | ret = -ERESTARTSYS; | |
1035 | goto out; | |
1036 | } | |
1037 | ||
1038 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1039 | ||
1040 | mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING, | |
1041 | "Cluster lock called on freeing lockres %s! flags " | |
1042 | "0x%lx\n", lockres->l_name, lockres->l_flags); | |
1043 | ||
1044 | /* We only compare against the currently granted level | |
1045 | * here. If the lock is blocked waiting on a downconvert, | |
1046 | * we'll get caught below. */ | |
1047 | if (lockres->l_flags & OCFS2_LOCK_BUSY && | |
1048 | level > lockres->l_level) { | |
1049 | /* is someone sitting in dlm_lock? If so, wait on | |
1050 | * them. */ | |
1051 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); | |
1052 | wait = 1; | |
1053 | goto unlock; | |
1054 | } | |
1055 | ||
1056 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { | |
1057 | /* lock has not been created yet. */ | |
1058 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1059 | ||
1060 | ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0); | |
1061 | if (ret < 0) { | |
1062 | mlog_errno(ret); | |
1063 | goto out; | |
1064 | } | |
1065 | goto again; | |
1066 | } | |
1067 | ||
1068 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED && | |
1069 | !ocfs2_may_continue_on_blocked_lock(lockres, level)) { | |
1070 | /* is the lock is currently blocked on behalf of | |
1071 | * another node */ | |
1072 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0); | |
1073 | wait = 1; | |
1074 | goto unlock; | |
1075 | } | |
1076 | ||
1077 | if (level > lockres->l_level) { | |
1078 | if (lockres->l_action != OCFS2_AST_INVALID) | |
1079 | mlog(ML_ERROR, "lockres %s has action %u pending\n", | |
1080 | lockres->l_name, lockres->l_action); | |
1081 | ||
1082 | lockres->l_action = OCFS2_AST_CONVERT; | |
1083 | lockres->l_requested = level; | |
1084 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
1085 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1086 | ||
1087 | BUG_ON(level == LKM_IVMODE); | |
1088 | BUG_ON(level == LKM_NLMODE); | |
1089 | ||
1090 | mlog(0, "lock %s, convert from %d to level = %d\n", | |
1091 | lockres->l_name, lockres->l_level, level); | |
1092 | ||
1093 | /* call dlm_lock to upgrade lock now */ | |
1094 | status = dlmlock(osb->dlm, | |
1095 | level, | |
1096 | &lockres->l_lksb, | |
1097 | lkm_flags|LKM_CONVERT|LKM_VALBLK, | |
1098 | lockres->l_name, | |
f0681062 | 1099 | OCFS2_LOCK_ID_MAX_LEN - 1, |
ccd979bd MF |
1100 | lockres->l_ops->ast, |
1101 | lockres, | |
1102 | lockres->l_ops->bast); | |
1103 | if (status != DLM_NORMAL) { | |
1104 | if ((lkm_flags & LKM_NOQUEUE) && | |
1105 | (status == DLM_NOTQUEUED)) | |
1106 | ret = -EAGAIN; | |
1107 | else { | |
1108 | ocfs2_log_dlm_error("dlmlock", status, | |
1109 | lockres); | |
1110 | ret = -EINVAL; | |
1111 | } | |
1112 | ocfs2_recover_from_dlm_error(lockres, 1); | |
1113 | goto out; | |
1114 | } | |
1115 | ||
1116 | mlog(0, "lock %s, successfull return from dlmlock\n", | |
1117 | lockres->l_name); | |
1118 | ||
1119 | /* At this point we've gone inside the dlm and need to | |
1120 | * complete our work regardless. */ | |
1121 | catch_signals = 0; | |
1122 | ||
1123 | /* wait for busy to clear and carry on */ | |
1124 | goto again; | |
1125 | } | |
1126 | ||
1127 | /* Ok, if we get here then we're good to go. */ | |
1128 | ocfs2_inc_holders(lockres, level); | |
1129 | ||
1130 | ret = 0; | |
1131 | unlock: | |
1132 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1133 | out: | |
1134 | /* | |
1135 | * This is helping work around a lock inversion between the page lock | |
1136 | * and dlm locks. One path holds the page lock while calling aops | |
1137 | * which block acquiring dlm locks. The voting thread holds dlm | |
1138 | * locks while acquiring page locks while down converting data locks. | |
1139 | * This block is helping an aop path notice the inversion and back | |
1140 | * off to unlock its page lock before trying the dlm lock again. | |
1141 | */ | |
1142 | if (wait && arg_flags & OCFS2_LOCK_NONBLOCK && | |
1143 | mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) { | |
1144 | wait = 0; | |
1145 | if (lockres_remove_mask_waiter(lockres, &mw)) | |
1146 | ret = -EAGAIN; | |
1147 | else | |
1148 | goto again; | |
1149 | } | |
1150 | if (wait) { | |
1151 | ret = ocfs2_wait_for_mask(&mw); | |
1152 | if (ret == 0) | |
1153 | goto again; | |
1154 | mlog_errno(ret); | |
1155 | } | |
1156 | ||
1157 | mlog_exit(ret); | |
1158 | return ret; | |
1159 | } | |
1160 | ||
1161 | static void ocfs2_cluster_unlock(struct ocfs2_super *osb, | |
1162 | struct ocfs2_lock_res *lockres, | |
1163 | int level) | |
1164 | { | |
1165 | unsigned long flags; | |
1166 | ||
1167 | mlog_entry_void(); | |
1168 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1169 | ocfs2_dec_holders(lockres, level); | |
1170 | ocfs2_vote_on_unlock(osb, lockres); | |
1171 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1172 | mlog_exit_void(); | |
1173 | } | |
1174 | ||
d680efe9 MF |
1175 | int ocfs2_create_new_lock(struct ocfs2_super *osb, |
1176 | struct ocfs2_lock_res *lockres, | |
24c19ef4 MF |
1177 | int ex, |
1178 | int local) | |
ccd979bd | 1179 | { |
d680efe9 | 1180 | int level = ex ? LKM_EXMODE : LKM_PRMODE; |
ccd979bd | 1181 | unsigned long flags; |
24c19ef4 | 1182 | int lkm_flags = local ? LKM_LOCAL : 0; |
ccd979bd MF |
1183 | |
1184 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1185 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); | |
1186 | lockres_or_flags(lockres, OCFS2_LOCK_LOCAL); | |
1187 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1188 | ||
24c19ef4 | 1189 | return ocfs2_lock_create(osb, lockres, level, lkm_flags); |
ccd979bd MF |
1190 | } |
1191 | ||
1192 | /* Grants us an EX lock on the data and metadata resources, skipping | |
1193 | * the normal cluster directory lookup. Use this ONLY on newly created | |
1194 | * inodes which other nodes can't possibly see, and which haven't been | |
1195 | * hashed in the inode hash yet. This can give us a good performance | |
1196 | * increase as it'll skip the network broadcast normally associated | |
1197 | * with creating a new lock resource. */ | |
1198 | int ocfs2_create_new_inode_locks(struct inode *inode) | |
1199 | { | |
1200 | int ret; | |
d680efe9 | 1201 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd MF |
1202 | |
1203 | BUG_ON(!inode); | |
1204 | BUG_ON(!ocfs2_inode_is_new(inode)); | |
1205 | ||
1206 | mlog_entry_void(); | |
1207 | ||
b0697053 | 1208 | mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno); |
ccd979bd MF |
1209 | |
1210 | /* NOTE: That we don't increment any of the holder counts, nor | |
1211 | * do we add anything to a journal handle. Since this is | |
1212 | * supposed to be a new inode which the cluster doesn't know | |
1213 | * about yet, there is no need to. As far as the LVB handling | |
1214 | * is concerned, this is basically like acquiring an EX lock | |
1215 | * on a resource which has an invalid one -- we'll set it | |
1216 | * valid when we release the EX. */ | |
1217 | ||
24c19ef4 | 1218 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1); |
ccd979bd MF |
1219 | if (ret) { |
1220 | mlog_errno(ret); | |
1221 | goto bail; | |
1222 | } | |
1223 | ||
24c19ef4 MF |
1224 | /* |
1225 | * We don't want to use LKM_LOCAL on a meta data lock as they | |
1226 | * don't use a generation in their lock names. | |
1227 | */ | |
1228 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_meta_lockres, 1, 0); | |
ccd979bd MF |
1229 | if (ret) { |
1230 | mlog_errno(ret); | |
1231 | goto bail; | |
1232 | } | |
1233 | ||
24c19ef4 | 1234 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_data_lockres, 1, 1); |
ccd979bd MF |
1235 | if (ret) { |
1236 | mlog_errno(ret); | |
1237 | goto bail; | |
1238 | } | |
1239 | ||
1240 | bail: | |
1241 | mlog_exit(ret); | |
1242 | return ret; | |
1243 | } | |
1244 | ||
1245 | int ocfs2_rw_lock(struct inode *inode, int write) | |
1246 | { | |
1247 | int status, level; | |
1248 | struct ocfs2_lock_res *lockres; | |
1249 | ||
1250 | BUG_ON(!inode); | |
1251 | ||
1252 | mlog_entry_void(); | |
1253 | ||
b0697053 MF |
1254 | mlog(0, "inode %llu take %s RW lock\n", |
1255 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1256 | write ? "EXMODE" : "PRMODE"); |
1257 | ||
1258 | lockres = &OCFS2_I(inode)->ip_rw_lockres; | |
1259 | ||
1260 | level = write ? LKM_EXMODE : LKM_PRMODE; | |
1261 | ||
1262 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0, | |
1263 | 0); | |
1264 | if (status < 0) | |
1265 | mlog_errno(status); | |
1266 | ||
1267 | mlog_exit(status); | |
1268 | return status; | |
1269 | } | |
1270 | ||
1271 | void ocfs2_rw_unlock(struct inode *inode, int write) | |
1272 | { | |
1273 | int level = write ? LKM_EXMODE : LKM_PRMODE; | |
1274 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres; | |
1275 | ||
1276 | mlog_entry_void(); | |
1277 | ||
b0697053 MF |
1278 | mlog(0, "inode %llu drop %s RW lock\n", |
1279 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1280 | write ? "EXMODE" : "PRMODE"); |
1281 | ||
1282 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
1283 | ||
1284 | mlog_exit_void(); | |
1285 | } | |
1286 | ||
1287 | int ocfs2_data_lock_full(struct inode *inode, | |
1288 | int write, | |
1289 | int arg_flags) | |
1290 | { | |
1291 | int status = 0, level; | |
1292 | struct ocfs2_lock_res *lockres; | |
1293 | ||
1294 | BUG_ON(!inode); | |
1295 | ||
1296 | mlog_entry_void(); | |
1297 | ||
b0697053 MF |
1298 | mlog(0, "inode %llu take %s DATA lock\n", |
1299 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1300 | write ? "EXMODE" : "PRMODE"); |
1301 | ||
1302 | /* We'll allow faking a readonly data lock for | |
1303 | * rodevices. */ | |
1304 | if (ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) { | |
1305 | if (write) { | |
1306 | status = -EROFS; | |
1307 | mlog_errno(status); | |
1308 | } | |
1309 | goto out; | |
1310 | } | |
1311 | ||
1312 | lockres = &OCFS2_I(inode)->ip_data_lockres; | |
1313 | ||
1314 | level = write ? LKM_EXMODE : LKM_PRMODE; | |
1315 | ||
1316 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, | |
1317 | 0, arg_flags); | |
1318 | if (status < 0 && status != -EAGAIN) | |
1319 | mlog_errno(status); | |
1320 | ||
1321 | out: | |
1322 | mlog_exit(status); | |
1323 | return status; | |
1324 | } | |
1325 | ||
1326 | /* see ocfs2_meta_lock_with_page() */ | |
1327 | int ocfs2_data_lock_with_page(struct inode *inode, | |
1328 | int write, | |
1329 | struct page *page) | |
1330 | { | |
1331 | int ret; | |
1332 | ||
1333 | ret = ocfs2_data_lock_full(inode, write, OCFS2_LOCK_NONBLOCK); | |
1334 | if (ret == -EAGAIN) { | |
1335 | unlock_page(page); | |
1336 | if (ocfs2_data_lock(inode, write) == 0) | |
1337 | ocfs2_data_unlock(inode, write); | |
1338 | ret = AOP_TRUNCATED_PAGE; | |
1339 | } | |
1340 | ||
1341 | return ret; | |
1342 | } | |
1343 | ||
1344 | static void ocfs2_vote_on_unlock(struct ocfs2_super *osb, | |
1345 | struct ocfs2_lock_res *lockres) | |
1346 | { | |
1347 | int kick = 0; | |
1348 | ||
1349 | mlog_entry_void(); | |
1350 | ||
1351 | /* If we know that another node is waiting on our lock, kick | |
1352 | * the vote thread * pre-emptively when we reach a release | |
1353 | * condition. */ | |
1354 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) { | |
1355 | switch(lockres->l_blocking) { | |
1356 | case LKM_EXMODE: | |
1357 | if (!lockres->l_ex_holders && !lockres->l_ro_holders) | |
1358 | kick = 1; | |
1359 | break; | |
1360 | case LKM_PRMODE: | |
1361 | if (!lockres->l_ex_holders) | |
1362 | kick = 1; | |
1363 | break; | |
1364 | default: | |
1365 | BUG(); | |
1366 | } | |
1367 | } | |
1368 | ||
1369 | if (kick) | |
1370 | ocfs2_kick_vote_thread(osb); | |
1371 | ||
1372 | mlog_exit_void(); | |
1373 | } | |
1374 | ||
1375 | void ocfs2_data_unlock(struct inode *inode, | |
1376 | int write) | |
1377 | { | |
1378 | int level = write ? LKM_EXMODE : LKM_PRMODE; | |
1379 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_data_lockres; | |
1380 | ||
1381 | mlog_entry_void(); | |
1382 | ||
b0697053 MF |
1383 | mlog(0, "inode %llu drop %s DATA lock\n", |
1384 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1385 | write ? "EXMODE" : "PRMODE"); |
1386 | ||
1387 | if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) | |
1388 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
1389 | ||
1390 | mlog_exit_void(); | |
1391 | } | |
1392 | ||
1393 | #define OCFS2_SEC_BITS 34 | |
1394 | #define OCFS2_SEC_SHIFT (64 - 34) | |
1395 | #define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1) | |
1396 | ||
1397 | /* LVB only has room for 64 bits of time here so we pack it for | |
1398 | * now. */ | |
1399 | static u64 ocfs2_pack_timespec(struct timespec *spec) | |
1400 | { | |
1401 | u64 res; | |
1402 | u64 sec = spec->tv_sec; | |
1403 | u32 nsec = spec->tv_nsec; | |
1404 | ||
1405 | res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK); | |
1406 | ||
1407 | return res; | |
1408 | } | |
1409 | ||
1410 | /* Call this with the lockres locked. I am reasonably sure we don't | |
1411 | * need ip_lock in this function as anyone who would be changing those | |
1412 | * values is supposed to be blocked in ocfs2_meta_lock right now. */ | |
1413 | static void __ocfs2_stuff_meta_lvb(struct inode *inode) | |
1414 | { | |
1415 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1416 | struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres; | |
1417 | struct ocfs2_meta_lvb *lvb; | |
1418 | ||
1419 | mlog_entry_void(); | |
1420 | ||
1421 | lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
1422 | ||
24c19ef4 MF |
1423 | /* |
1424 | * Invalidate the LVB of a deleted inode - this way other | |
1425 | * nodes are forced to go to disk and discover the new inode | |
1426 | * status. | |
1427 | */ | |
1428 | if (oi->ip_flags & OCFS2_INODE_DELETED) { | |
1429 | lvb->lvb_version = 0; | |
1430 | goto out; | |
1431 | } | |
1432 | ||
4d3b83f7 | 1433 | lvb->lvb_version = OCFS2_LVB_VERSION; |
ccd979bd MF |
1434 | lvb->lvb_isize = cpu_to_be64(i_size_read(inode)); |
1435 | lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters); | |
1436 | lvb->lvb_iuid = cpu_to_be32(inode->i_uid); | |
1437 | lvb->lvb_igid = cpu_to_be32(inode->i_gid); | |
1438 | lvb->lvb_imode = cpu_to_be16(inode->i_mode); | |
1439 | lvb->lvb_inlink = cpu_to_be16(inode->i_nlink); | |
1440 | lvb->lvb_iatime_packed = | |
1441 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime)); | |
1442 | lvb->lvb_ictime_packed = | |
1443 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime)); | |
1444 | lvb->lvb_imtime_packed = | |
1445 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime)); | |
ca4d147e | 1446 | lvb->lvb_iattr = cpu_to_be32(oi->ip_attr); |
f9e2d82e | 1447 | lvb->lvb_igeneration = cpu_to_be32(inode->i_generation); |
ccd979bd | 1448 | |
24c19ef4 | 1449 | out: |
ccd979bd MF |
1450 | mlog_meta_lvb(0, lockres); |
1451 | ||
1452 | mlog_exit_void(); | |
1453 | } | |
1454 | ||
1455 | static void ocfs2_unpack_timespec(struct timespec *spec, | |
1456 | u64 packed_time) | |
1457 | { | |
1458 | spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT; | |
1459 | spec->tv_nsec = packed_time & OCFS2_NSEC_MASK; | |
1460 | } | |
1461 | ||
1462 | static void ocfs2_refresh_inode_from_lvb(struct inode *inode) | |
1463 | { | |
1464 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1465 | struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres; | |
1466 | struct ocfs2_meta_lvb *lvb; | |
1467 | ||
1468 | mlog_entry_void(); | |
1469 | ||
1470 | mlog_meta_lvb(0, lockres); | |
1471 | ||
1472 | lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
1473 | ||
1474 | /* We're safe here without the lockres lock... */ | |
1475 | spin_lock(&oi->ip_lock); | |
1476 | oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters); | |
1477 | i_size_write(inode, be64_to_cpu(lvb->lvb_isize)); | |
1478 | ||
ca4d147e HP |
1479 | oi->ip_attr = be32_to_cpu(lvb->lvb_iattr); |
1480 | ocfs2_set_inode_flags(inode); | |
1481 | ||
ccd979bd MF |
1482 | /* fast-symlinks are a special case */ |
1483 | if (S_ISLNK(inode->i_mode) && !oi->ip_clusters) | |
1484 | inode->i_blocks = 0; | |
1485 | else | |
1486 | inode->i_blocks = | |
1487 | ocfs2_align_bytes_to_sectors(i_size_read(inode)); | |
1488 | ||
1489 | inode->i_uid = be32_to_cpu(lvb->lvb_iuid); | |
1490 | inode->i_gid = be32_to_cpu(lvb->lvb_igid); | |
1491 | inode->i_mode = be16_to_cpu(lvb->lvb_imode); | |
1492 | inode->i_nlink = be16_to_cpu(lvb->lvb_inlink); | |
1493 | ocfs2_unpack_timespec(&inode->i_atime, | |
1494 | be64_to_cpu(lvb->lvb_iatime_packed)); | |
1495 | ocfs2_unpack_timespec(&inode->i_mtime, | |
1496 | be64_to_cpu(lvb->lvb_imtime_packed)); | |
1497 | ocfs2_unpack_timespec(&inode->i_ctime, | |
1498 | be64_to_cpu(lvb->lvb_ictime_packed)); | |
1499 | spin_unlock(&oi->ip_lock); | |
1500 | ||
1501 | mlog_exit_void(); | |
1502 | } | |
1503 | ||
f9e2d82e MF |
1504 | static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode, |
1505 | struct ocfs2_lock_res *lockres) | |
ccd979bd MF |
1506 | { |
1507 | struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
1508 | ||
f9e2d82e MF |
1509 | if (lvb->lvb_version == OCFS2_LVB_VERSION |
1510 | && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation) | |
ccd979bd MF |
1511 | return 1; |
1512 | return 0; | |
1513 | } | |
1514 | ||
1515 | /* Determine whether a lock resource needs to be refreshed, and | |
1516 | * arbitrate who gets to refresh it. | |
1517 | * | |
1518 | * 0 means no refresh needed. | |
1519 | * | |
1520 | * > 0 means you need to refresh this and you MUST call | |
1521 | * ocfs2_complete_lock_res_refresh afterwards. */ | |
1522 | static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres) | |
1523 | { | |
1524 | unsigned long flags; | |
1525 | int status = 0; | |
1526 | ||
1527 | mlog_entry_void(); | |
1528 | ||
1529 | refresh_check: | |
1530 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1531 | if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) { | |
1532 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1533 | goto bail; | |
1534 | } | |
1535 | ||
1536 | if (lockres->l_flags & OCFS2_LOCK_REFRESHING) { | |
1537 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1538 | ||
1539 | ocfs2_wait_on_refreshing_lock(lockres); | |
1540 | goto refresh_check; | |
1541 | } | |
1542 | ||
1543 | /* Ok, I'll be the one to refresh this lock. */ | |
1544 | lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING); | |
1545 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1546 | ||
1547 | status = 1; | |
1548 | bail: | |
1549 | mlog_exit(status); | |
1550 | return status; | |
1551 | } | |
1552 | ||
1553 | /* If status is non zero, I'll mark it as not being in refresh | |
1554 | * anymroe, but i won't clear the needs refresh flag. */ | |
1555 | static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres, | |
1556 | int status) | |
1557 | { | |
1558 | unsigned long flags; | |
1559 | mlog_entry_void(); | |
1560 | ||
1561 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1562 | lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING); | |
1563 | if (!status) | |
1564 | lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); | |
1565 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1566 | ||
1567 | wake_up(&lockres->l_event); | |
1568 | ||
1569 | mlog_exit_void(); | |
1570 | } | |
1571 | ||
1572 | /* may or may not return a bh if it went to disk. */ | |
1573 | static int ocfs2_meta_lock_update(struct inode *inode, | |
1574 | struct buffer_head **bh) | |
1575 | { | |
1576 | int status = 0; | |
1577 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1578 | struct ocfs2_lock_res *lockres; | |
1579 | struct ocfs2_dinode *fe; | |
1580 | ||
1581 | mlog_entry_void(); | |
1582 | ||
1583 | spin_lock(&oi->ip_lock); | |
1584 | if (oi->ip_flags & OCFS2_INODE_DELETED) { | |
b0697053 | 1585 | mlog(0, "Orphaned inode %llu was deleted while we " |
ccd979bd | 1586 | "were waiting on a lock. ip_flags = 0x%x\n", |
b0697053 | 1587 | (unsigned long long)oi->ip_blkno, oi->ip_flags); |
ccd979bd MF |
1588 | spin_unlock(&oi->ip_lock); |
1589 | status = -ENOENT; | |
1590 | goto bail; | |
1591 | } | |
1592 | spin_unlock(&oi->ip_lock); | |
1593 | ||
1594 | lockres = &oi->ip_meta_lockres; | |
1595 | ||
1596 | if (!ocfs2_should_refresh_lock_res(lockres)) | |
1597 | goto bail; | |
1598 | ||
1599 | /* This will discard any caching information we might have had | |
1600 | * for the inode metadata. */ | |
1601 | ocfs2_metadata_cache_purge(inode); | |
1602 | ||
1603 | /* will do nothing for inode types that don't use the extent | |
1604 | * map (directories, bitmap files, etc) */ | |
1605 | ocfs2_extent_map_trunc(inode, 0); | |
1606 | ||
f9e2d82e | 1607 | if (ocfs2_meta_lvb_is_trustable(inode, lockres)) { |
b0697053 MF |
1608 | mlog(0, "Trusting LVB on inode %llu\n", |
1609 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1610 | ocfs2_refresh_inode_from_lvb(inode); |
1611 | } else { | |
1612 | /* Boo, we have to go to disk. */ | |
1613 | /* read bh, cast, ocfs2_refresh_inode */ | |
1614 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno, | |
1615 | bh, OCFS2_BH_CACHED, inode); | |
1616 | if (status < 0) { | |
1617 | mlog_errno(status); | |
1618 | goto bail_refresh; | |
1619 | } | |
1620 | fe = (struct ocfs2_dinode *) (*bh)->b_data; | |
1621 | ||
1622 | /* This is a good chance to make sure we're not | |
1623 | * locking an invalid object. | |
1624 | * | |
1625 | * We bug on a stale inode here because we checked | |
1626 | * above whether it was wiped from disk. The wiping | |
1627 | * node provides a guarantee that we receive that | |
1628 | * message and can mark the inode before dropping any | |
1629 | * locks associated with it. */ | |
1630 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
1631 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
1632 | status = -EIO; | |
1633 | goto bail_refresh; | |
1634 | } | |
1635 | mlog_bug_on_msg(inode->i_generation != | |
1636 | le32_to_cpu(fe->i_generation), | |
b0697053 | 1637 | "Invalid dinode %llu disk generation: %u " |
ccd979bd | 1638 | "inode->i_generation: %u\n", |
b0697053 MF |
1639 | (unsigned long long)oi->ip_blkno, |
1640 | le32_to_cpu(fe->i_generation), | |
ccd979bd MF |
1641 | inode->i_generation); |
1642 | mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) || | |
1643 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)), | |
b0697053 MF |
1644 | "Stale dinode %llu dtime: %llu flags: 0x%x\n", |
1645 | (unsigned long long)oi->ip_blkno, | |
1646 | (unsigned long long)le64_to_cpu(fe->i_dtime), | |
ccd979bd MF |
1647 | le32_to_cpu(fe->i_flags)); |
1648 | ||
1649 | ocfs2_refresh_inode(inode, fe); | |
1650 | } | |
1651 | ||
1652 | status = 0; | |
1653 | bail_refresh: | |
1654 | ocfs2_complete_lock_res_refresh(lockres, status); | |
1655 | bail: | |
1656 | mlog_exit(status); | |
1657 | return status; | |
1658 | } | |
1659 | ||
1660 | static int ocfs2_assign_bh(struct inode *inode, | |
1661 | struct buffer_head **ret_bh, | |
1662 | struct buffer_head *passed_bh) | |
1663 | { | |
1664 | int status; | |
1665 | ||
1666 | if (passed_bh) { | |
1667 | /* Ok, the update went to disk for us, use the | |
1668 | * returned bh. */ | |
1669 | *ret_bh = passed_bh; | |
1670 | get_bh(*ret_bh); | |
1671 | ||
1672 | return 0; | |
1673 | } | |
1674 | ||
1675 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
1676 | OCFS2_I(inode)->ip_blkno, | |
1677 | ret_bh, | |
1678 | OCFS2_BH_CACHED, | |
1679 | inode); | |
1680 | if (status < 0) | |
1681 | mlog_errno(status); | |
1682 | ||
1683 | return status; | |
1684 | } | |
1685 | ||
1686 | /* | |
1687 | * returns < 0 error if the callback will never be called, otherwise | |
1688 | * the result of the lock will be communicated via the callback. | |
1689 | */ | |
1690 | int ocfs2_meta_lock_full(struct inode *inode, | |
1691 | struct ocfs2_journal_handle *handle, | |
1692 | struct buffer_head **ret_bh, | |
1693 | int ex, | |
1694 | int arg_flags) | |
1695 | { | |
1696 | int status, level, dlm_flags, acquired; | |
1697 | struct ocfs2_lock_res *lockres; | |
1698 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1699 | struct buffer_head *local_bh = NULL; | |
1700 | ||
1701 | BUG_ON(!inode); | |
1702 | ||
1703 | mlog_entry_void(); | |
1704 | ||
b0697053 MF |
1705 | mlog(0, "inode %llu, take %s META lock\n", |
1706 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1707 | ex ? "EXMODE" : "PRMODE"); |
1708 | ||
1709 | status = 0; | |
1710 | acquired = 0; | |
1711 | /* We'll allow faking a readonly metadata lock for | |
1712 | * rodevices. */ | |
1713 | if (ocfs2_is_hard_readonly(osb)) { | |
1714 | if (ex) | |
1715 | status = -EROFS; | |
1716 | goto bail; | |
1717 | } | |
1718 | ||
1719 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) | |
1720 | wait_event(osb->recovery_event, | |
1721 | ocfs2_node_map_is_empty(osb, &osb->recovery_map)); | |
1722 | ||
1723 | acquired = 0; | |
1724 | lockres = &OCFS2_I(inode)->ip_meta_lockres; | |
1725 | level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1726 | dlm_flags = 0; | |
1727 | if (arg_flags & OCFS2_META_LOCK_NOQUEUE) | |
1728 | dlm_flags |= LKM_NOQUEUE; | |
1729 | ||
1730 | status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags); | |
1731 | if (status < 0) { | |
1732 | if (status != -EAGAIN && status != -EIOCBRETRY) | |
1733 | mlog_errno(status); | |
1734 | goto bail; | |
1735 | } | |
1736 | ||
1737 | /* Notify the error cleanup path to drop the cluster lock. */ | |
1738 | acquired = 1; | |
1739 | ||
1740 | /* We wait twice because a node may have died while we were in | |
1741 | * the lower dlm layers. The second time though, we've | |
1742 | * committed to owning this lock so we don't allow signals to | |
1743 | * abort the operation. */ | |
1744 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) | |
1745 | wait_event(osb->recovery_event, | |
1746 | ocfs2_node_map_is_empty(osb, &osb->recovery_map)); | |
1747 | ||
24c19ef4 MF |
1748 | /* |
1749 | * We only see this flag if we're being called from | |
1750 | * ocfs2_read_locked_inode(). It means we're locking an inode | |
1751 | * which hasn't been populated yet, so clear the refresh flag | |
1752 | * and let the caller handle it. | |
1753 | */ | |
1754 | if (inode->i_state & I_NEW) { | |
1755 | status = 0; | |
1756 | ocfs2_complete_lock_res_refresh(lockres, 0); | |
1757 | goto bail; | |
1758 | } | |
1759 | ||
ccd979bd MF |
1760 | /* This is fun. The caller may want a bh back, or it may |
1761 | * not. ocfs2_meta_lock_update definitely wants one in, but | |
1762 | * may or may not read one, depending on what's in the | |
1763 | * LVB. The result of all of this is that we've *only* gone to | |
1764 | * disk if we have to, so the complexity is worthwhile. */ | |
1765 | status = ocfs2_meta_lock_update(inode, &local_bh); | |
1766 | if (status < 0) { | |
1767 | if (status != -ENOENT) | |
1768 | mlog_errno(status); | |
1769 | goto bail; | |
1770 | } | |
1771 | ||
1772 | if (ret_bh) { | |
1773 | status = ocfs2_assign_bh(inode, ret_bh, local_bh); | |
1774 | if (status < 0) { | |
1775 | mlog_errno(status); | |
1776 | goto bail; | |
1777 | } | |
1778 | } | |
1779 | ||
1780 | if (handle) { | |
1781 | status = ocfs2_handle_add_lock(handle, inode); | |
1782 | if (status < 0) | |
1783 | mlog_errno(status); | |
1784 | } | |
1785 | ||
1786 | bail: | |
1787 | if (status < 0) { | |
1788 | if (ret_bh && (*ret_bh)) { | |
1789 | brelse(*ret_bh); | |
1790 | *ret_bh = NULL; | |
1791 | } | |
1792 | if (acquired) | |
1793 | ocfs2_meta_unlock(inode, ex); | |
1794 | } | |
1795 | ||
1796 | if (local_bh) | |
1797 | brelse(local_bh); | |
1798 | ||
1799 | mlog_exit(status); | |
1800 | return status; | |
1801 | } | |
1802 | ||
1803 | /* | |
1804 | * This is working around a lock inversion between tasks acquiring DLM locks | |
1805 | * while holding a page lock and the vote thread which blocks dlm lock acquiry | |
1806 | * while acquiring page locks. | |
1807 | * | |
1808 | * ** These _with_page variantes are only intended to be called from aop | |
1809 | * methods that hold page locks and return a very specific *positive* error | |
1810 | * code that aop methods pass up to the VFS -- test for errors with != 0. ** | |
1811 | * | |
1812 | * The DLM is called such that it returns -EAGAIN if it would have blocked | |
1813 | * waiting for the vote thread. In that case we unlock our page so the vote | |
1814 | * thread can make progress. Once we've done this we have to return | |
1815 | * AOP_TRUNCATED_PAGE so the aop method that called us can bubble that back up | |
1816 | * into the VFS who will then immediately retry the aop call. | |
1817 | * | |
1818 | * We do a blocking lock and immediate unlock before returning, though, so that | |
1819 | * the lock has a great chance of being cached on this node by the time the VFS | |
1820 | * calls back to retry the aop. This has a potential to livelock as nodes | |
1821 | * ping locks back and forth, but that's a risk we're willing to take to avoid | |
1822 | * the lock inversion simply. | |
1823 | */ | |
1824 | int ocfs2_meta_lock_with_page(struct inode *inode, | |
1825 | struct ocfs2_journal_handle *handle, | |
1826 | struct buffer_head **ret_bh, | |
1827 | int ex, | |
1828 | struct page *page) | |
1829 | { | |
1830 | int ret; | |
1831 | ||
1832 | ret = ocfs2_meta_lock_full(inode, handle, ret_bh, ex, | |
1833 | OCFS2_LOCK_NONBLOCK); | |
1834 | if (ret == -EAGAIN) { | |
1835 | unlock_page(page); | |
1836 | if (ocfs2_meta_lock(inode, handle, ret_bh, ex) == 0) | |
1837 | ocfs2_meta_unlock(inode, ex); | |
1838 | ret = AOP_TRUNCATED_PAGE; | |
1839 | } | |
1840 | ||
1841 | return ret; | |
1842 | } | |
1843 | ||
1844 | void ocfs2_meta_unlock(struct inode *inode, | |
1845 | int ex) | |
1846 | { | |
1847 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1848 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres; | |
1849 | ||
1850 | mlog_entry_void(); | |
1851 | ||
b0697053 MF |
1852 | mlog(0, "inode %llu drop %s META lock\n", |
1853 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1854 | ex ? "EXMODE" : "PRMODE"); |
1855 | ||
1856 | if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) | |
1857 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
1858 | ||
1859 | mlog_exit_void(); | |
1860 | } | |
1861 | ||
1862 | int ocfs2_super_lock(struct ocfs2_super *osb, | |
1863 | int ex) | |
1864 | { | |
1865 | int status; | |
1866 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1867 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; | |
1868 | struct buffer_head *bh; | |
1869 | struct ocfs2_slot_info *si = osb->slot_info; | |
1870 | ||
1871 | mlog_entry_void(); | |
1872 | ||
1873 | if (ocfs2_is_hard_readonly(osb)) | |
1874 | return -EROFS; | |
1875 | ||
1876 | status = ocfs2_cluster_lock(osb, lockres, level, 0, 0); | |
1877 | if (status < 0) { | |
1878 | mlog_errno(status); | |
1879 | goto bail; | |
1880 | } | |
1881 | ||
1882 | /* The super block lock path is really in the best position to | |
1883 | * know when resources covered by the lock need to be | |
1884 | * refreshed, so we do it here. Of course, making sense of | |
1885 | * everything is up to the caller :) */ | |
1886 | status = ocfs2_should_refresh_lock_res(lockres); | |
1887 | if (status < 0) { | |
1888 | mlog_errno(status); | |
1889 | goto bail; | |
1890 | } | |
1891 | if (status) { | |
1892 | bh = si->si_bh; | |
1893 | status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0, | |
1894 | si->si_inode); | |
1895 | if (status == 0) | |
1896 | ocfs2_update_slot_info(si); | |
1897 | ||
1898 | ocfs2_complete_lock_res_refresh(lockres, status); | |
1899 | ||
1900 | if (status < 0) | |
1901 | mlog_errno(status); | |
1902 | } | |
1903 | bail: | |
1904 | mlog_exit(status); | |
1905 | return status; | |
1906 | } | |
1907 | ||
1908 | void ocfs2_super_unlock(struct ocfs2_super *osb, | |
1909 | int ex) | |
1910 | { | |
1911 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1912 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; | |
1913 | ||
1914 | ocfs2_cluster_unlock(osb, lockres, level); | |
1915 | } | |
1916 | ||
1917 | int ocfs2_rename_lock(struct ocfs2_super *osb) | |
1918 | { | |
1919 | int status; | |
1920 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; | |
1921 | ||
1922 | if (ocfs2_is_hard_readonly(osb)) | |
1923 | return -EROFS; | |
1924 | ||
1925 | status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, 0); | |
1926 | if (status < 0) | |
1927 | mlog_errno(status); | |
1928 | ||
1929 | return status; | |
1930 | } | |
1931 | ||
1932 | void ocfs2_rename_unlock(struct ocfs2_super *osb) | |
1933 | { | |
1934 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; | |
1935 | ||
1936 | ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE); | |
1937 | } | |
1938 | ||
d680efe9 MF |
1939 | int ocfs2_dentry_lock(struct dentry *dentry, int ex) |
1940 | { | |
1941 | int ret; | |
1942 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1943 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; | |
1944 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); | |
1945 | ||
1946 | BUG_ON(!dl); | |
1947 | ||
1948 | if (ocfs2_is_hard_readonly(osb)) | |
1949 | return -EROFS; | |
1950 | ||
1951 | ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0); | |
1952 | if (ret < 0) | |
1953 | mlog_errno(ret); | |
1954 | ||
1955 | return ret; | |
1956 | } | |
1957 | ||
1958 | void ocfs2_dentry_unlock(struct dentry *dentry, int ex) | |
1959 | { | |
1960 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1961 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; | |
1962 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); | |
1963 | ||
1964 | ocfs2_cluster_unlock(osb, &dl->dl_lockres, level); | |
1965 | } | |
1966 | ||
ccd979bd MF |
1967 | /* Reference counting of the dlm debug structure. We want this because |
1968 | * open references on the debug inodes can live on after a mount, so | |
1969 | * we can't rely on the ocfs2_super to always exist. */ | |
1970 | static void ocfs2_dlm_debug_free(struct kref *kref) | |
1971 | { | |
1972 | struct ocfs2_dlm_debug *dlm_debug; | |
1973 | ||
1974 | dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt); | |
1975 | ||
1976 | kfree(dlm_debug); | |
1977 | } | |
1978 | ||
1979 | void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug) | |
1980 | { | |
1981 | if (dlm_debug) | |
1982 | kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free); | |
1983 | } | |
1984 | ||
1985 | static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug) | |
1986 | { | |
1987 | kref_get(&debug->d_refcnt); | |
1988 | } | |
1989 | ||
1990 | struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void) | |
1991 | { | |
1992 | struct ocfs2_dlm_debug *dlm_debug; | |
1993 | ||
1994 | dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL); | |
1995 | if (!dlm_debug) { | |
1996 | mlog_errno(-ENOMEM); | |
1997 | goto out; | |
1998 | } | |
1999 | ||
2000 | kref_init(&dlm_debug->d_refcnt); | |
2001 | INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking); | |
2002 | dlm_debug->d_locking_state = NULL; | |
2003 | out: | |
2004 | return dlm_debug; | |
2005 | } | |
2006 | ||
2007 | /* Access to this is arbitrated for us via seq_file->sem. */ | |
2008 | struct ocfs2_dlm_seq_priv { | |
2009 | struct ocfs2_dlm_debug *p_dlm_debug; | |
2010 | struct ocfs2_lock_res p_iter_res; | |
2011 | struct ocfs2_lock_res p_tmp_res; | |
2012 | }; | |
2013 | ||
2014 | static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start, | |
2015 | struct ocfs2_dlm_seq_priv *priv) | |
2016 | { | |
2017 | struct ocfs2_lock_res *iter, *ret = NULL; | |
2018 | struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug; | |
2019 | ||
2020 | assert_spin_locked(&ocfs2_dlm_tracking_lock); | |
2021 | ||
2022 | list_for_each_entry(iter, &start->l_debug_list, l_debug_list) { | |
2023 | /* discover the head of the list */ | |
2024 | if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) { | |
2025 | mlog(0, "End of list found, %p\n", ret); | |
2026 | break; | |
2027 | } | |
2028 | ||
2029 | /* We track our "dummy" iteration lockres' by a NULL | |
2030 | * l_ops field. */ | |
2031 | if (iter->l_ops != NULL) { | |
2032 | ret = iter; | |
2033 | break; | |
2034 | } | |
2035 | } | |
2036 | ||
2037 | return ret; | |
2038 | } | |
2039 | ||
2040 | static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos) | |
2041 | { | |
2042 | struct ocfs2_dlm_seq_priv *priv = m->private; | |
2043 | struct ocfs2_lock_res *iter; | |
2044 | ||
2045 | spin_lock(&ocfs2_dlm_tracking_lock); | |
2046 | iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv); | |
2047 | if (iter) { | |
2048 | /* Since lockres' have the lifetime of their container | |
2049 | * (which can be inodes, ocfs2_supers, etc) we want to | |
2050 | * copy this out to a temporary lockres while still | |
2051 | * under the spinlock. Obviously after this we can't | |
2052 | * trust any pointers on the copy returned, but that's | |
2053 | * ok as the information we want isn't typically held | |
2054 | * in them. */ | |
2055 | priv->p_tmp_res = *iter; | |
2056 | iter = &priv->p_tmp_res; | |
2057 | } | |
2058 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
2059 | ||
2060 | return iter; | |
2061 | } | |
2062 | ||
2063 | static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v) | |
2064 | { | |
2065 | } | |
2066 | ||
2067 | static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
2068 | { | |
2069 | struct ocfs2_dlm_seq_priv *priv = m->private; | |
2070 | struct ocfs2_lock_res *iter = v; | |
2071 | struct ocfs2_lock_res *dummy = &priv->p_iter_res; | |
2072 | ||
2073 | spin_lock(&ocfs2_dlm_tracking_lock); | |
2074 | iter = ocfs2_dlm_next_res(iter, priv); | |
2075 | list_del_init(&dummy->l_debug_list); | |
2076 | if (iter) { | |
2077 | list_add(&dummy->l_debug_list, &iter->l_debug_list); | |
2078 | priv->p_tmp_res = *iter; | |
2079 | iter = &priv->p_tmp_res; | |
2080 | } | |
2081 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
2082 | ||
2083 | return iter; | |
2084 | } | |
2085 | ||
2086 | /* So that debugfs.ocfs2 can determine which format is being used */ | |
2087 | #define OCFS2_DLM_DEBUG_STR_VERSION 1 | |
2088 | static int ocfs2_dlm_seq_show(struct seq_file *m, void *v) | |
2089 | { | |
2090 | int i; | |
2091 | char *lvb; | |
2092 | struct ocfs2_lock_res *lockres = v; | |
2093 | ||
2094 | if (!lockres) | |
2095 | return -EINVAL; | |
2096 | ||
d680efe9 MF |
2097 | seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION); |
2098 | ||
2099 | if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY) | |
2100 | seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1, | |
2101 | lockres->l_name, | |
2102 | (unsigned int)ocfs2_get_dentry_lock_ino(lockres)); | |
2103 | else | |
2104 | seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name); | |
2105 | ||
2106 | seq_printf(m, "%d\t" | |
ccd979bd MF |
2107 | "0x%lx\t" |
2108 | "0x%x\t" | |
2109 | "0x%x\t" | |
2110 | "%u\t" | |
2111 | "%u\t" | |
2112 | "%d\t" | |
2113 | "%d\t", | |
ccd979bd MF |
2114 | lockres->l_level, |
2115 | lockres->l_flags, | |
2116 | lockres->l_action, | |
2117 | lockres->l_unlock_action, | |
2118 | lockres->l_ro_holders, | |
2119 | lockres->l_ex_holders, | |
2120 | lockres->l_requested, | |
2121 | lockres->l_blocking); | |
2122 | ||
2123 | /* Dump the raw LVB */ | |
2124 | lvb = lockres->l_lksb.lvb; | |
2125 | for(i = 0; i < DLM_LVB_LEN; i++) | |
2126 | seq_printf(m, "0x%x\t", lvb[i]); | |
2127 | ||
2128 | /* End the line */ | |
2129 | seq_printf(m, "\n"); | |
2130 | return 0; | |
2131 | } | |
2132 | ||
2133 | static struct seq_operations ocfs2_dlm_seq_ops = { | |
2134 | .start = ocfs2_dlm_seq_start, | |
2135 | .stop = ocfs2_dlm_seq_stop, | |
2136 | .next = ocfs2_dlm_seq_next, | |
2137 | .show = ocfs2_dlm_seq_show, | |
2138 | }; | |
2139 | ||
2140 | static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file) | |
2141 | { | |
2142 | struct seq_file *seq = (struct seq_file *) file->private_data; | |
2143 | struct ocfs2_dlm_seq_priv *priv = seq->private; | |
2144 | struct ocfs2_lock_res *res = &priv->p_iter_res; | |
2145 | ||
2146 | ocfs2_remove_lockres_tracking(res); | |
2147 | ocfs2_put_dlm_debug(priv->p_dlm_debug); | |
2148 | return seq_release_private(inode, file); | |
2149 | } | |
2150 | ||
2151 | static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file) | |
2152 | { | |
2153 | int ret; | |
2154 | struct ocfs2_dlm_seq_priv *priv; | |
2155 | struct seq_file *seq; | |
2156 | struct ocfs2_super *osb; | |
2157 | ||
2158 | priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL); | |
2159 | if (!priv) { | |
2160 | ret = -ENOMEM; | |
2161 | mlog_errno(ret); | |
2162 | goto out; | |
2163 | } | |
2164 | osb = (struct ocfs2_super *) inode->u.generic_ip; | |
2165 | ocfs2_get_dlm_debug(osb->osb_dlm_debug); | |
2166 | priv->p_dlm_debug = osb->osb_dlm_debug; | |
2167 | INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list); | |
2168 | ||
2169 | ret = seq_open(file, &ocfs2_dlm_seq_ops); | |
2170 | if (ret) { | |
2171 | kfree(priv); | |
2172 | mlog_errno(ret); | |
2173 | goto out; | |
2174 | } | |
2175 | ||
2176 | seq = (struct seq_file *) file->private_data; | |
2177 | seq->private = priv; | |
2178 | ||
2179 | ocfs2_add_lockres_tracking(&priv->p_iter_res, | |
2180 | priv->p_dlm_debug); | |
2181 | ||
2182 | out: | |
2183 | return ret; | |
2184 | } | |
2185 | ||
4b6f5d20 | 2186 | static const struct file_operations ocfs2_dlm_debug_fops = { |
ccd979bd MF |
2187 | .open = ocfs2_dlm_debug_open, |
2188 | .release = ocfs2_dlm_debug_release, | |
2189 | .read = seq_read, | |
2190 | .llseek = seq_lseek, | |
2191 | }; | |
2192 | ||
2193 | static int ocfs2_dlm_init_debug(struct ocfs2_super *osb) | |
2194 | { | |
2195 | int ret = 0; | |
2196 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; | |
2197 | ||
2198 | dlm_debug->d_locking_state = debugfs_create_file("locking_state", | |
2199 | S_IFREG|S_IRUSR, | |
2200 | osb->osb_debug_root, | |
2201 | osb, | |
2202 | &ocfs2_dlm_debug_fops); | |
2203 | if (!dlm_debug->d_locking_state) { | |
2204 | ret = -EINVAL; | |
2205 | mlog(ML_ERROR, | |
2206 | "Unable to create locking state debugfs file.\n"); | |
2207 | goto out; | |
2208 | } | |
2209 | ||
2210 | ocfs2_get_dlm_debug(dlm_debug); | |
2211 | out: | |
2212 | return ret; | |
2213 | } | |
2214 | ||
2215 | static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb) | |
2216 | { | |
2217 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; | |
2218 | ||
2219 | if (dlm_debug) { | |
2220 | debugfs_remove(dlm_debug->d_locking_state); | |
2221 | ocfs2_put_dlm_debug(dlm_debug); | |
2222 | } | |
2223 | } | |
2224 | ||
2225 | int ocfs2_dlm_init(struct ocfs2_super *osb) | |
2226 | { | |
2227 | int status; | |
2228 | u32 dlm_key; | |
2229 | struct dlm_ctxt *dlm; | |
2230 | ||
2231 | mlog_entry_void(); | |
2232 | ||
2233 | status = ocfs2_dlm_init_debug(osb); | |
2234 | if (status < 0) { | |
2235 | mlog_errno(status); | |
2236 | goto bail; | |
2237 | } | |
2238 | ||
2239 | /* launch vote thread */ | |
78427043 | 2240 | osb->vote_task = kthread_run(ocfs2_vote_thread, osb, "ocfs2vote"); |
ccd979bd MF |
2241 | if (IS_ERR(osb->vote_task)) { |
2242 | status = PTR_ERR(osb->vote_task); | |
2243 | osb->vote_task = NULL; | |
2244 | mlog_errno(status); | |
2245 | goto bail; | |
2246 | } | |
2247 | ||
2248 | /* used by the dlm code to make message headers unique, each | |
2249 | * node in this domain must agree on this. */ | |
2250 | dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str)); | |
2251 | ||
2252 | /* for now, uuid == domain */ | |
2253 | dlm = dlm_register_domain(osb->uuid_str, dlm_key); | |
2254 | if (IS_ERR(dlm)) { | |
2255 | status = PTR_ERR(dlm); | |
2256 | mlog_errno(status); | |
2257 | goto bail; | |
2258 | } | |
2259 | ||
2260 | ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb); | |
2261 | ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb); | |
2262 | ||
2263 | dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb); | |
2264 | ||
2265 | osb->dlm = dlm; | |
2266 | ||
2267 | status = 0; | |
2268 | bail: | |
2269 | if (status < 0) { | |
2270 | ocfs2_dlm_shutdown_debug(osb); | |
2271 | if (osb->vote_task) | |
2272 | kthread_stop(osb->vote_task); | |
2273 | } | |
2274 | ||
2275 | mlog_exit(status); | |
2276 | return status; | |
2277 | } | |
2278 | ||
2279 | void ocfs2_dlm_shutdown(struct ocfs2_super *osb) | |
2280 | { | |
2281 | mlog_entry_void(); | |
2282 | ||
2283 | dlm_unregister_eviction_cb(&osb->osb_eviction_cb); | |
2284 | ||
2285 | ocfs2_drop_osb_locks(osb); | |
2286 | ||
2287 | if (osb->vote_task) { | |
2288 | kthread_stop(osb->vote_task); | |
2289 | osb->vote_task = NULL; | |
2290 | } | |
2291 | ||
2292 | ocfs2_lock_res_free(&osb->osb_super_lockres); | |
2293 | ocfs2_lock_res_free(&osb->osb_rename_lockres); | |
2294 | ||
2295 | dlm_unregister_domain(osb->dlm); | |
2296 | osb->dlm = NULL; | |
2297 | ||
2298 | ocfs2_dlm_shutdown_debug(osb); | |
2299 | ||
2300 | mlog_exit_void(); | |
2301 | } | |
2302 | ||
2303 | static void ocfs2_unlock_ast_func(void *opaque, enum dlm_status status) | |
2304 | { | |
2305 | struct ocfs2_lock_res *lockres = opaque; | |
2306 | unsigned long flags; | |
2307 | ||
2308 | mlog_entry_void(); | |
2309 | ||
2310 | mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name, | |
2311 | lockres->l_unlock_action); | |
2312 | ||
2313 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2314 | /* We tried to cancel a convert request, but it was already | |
2315 | * granted. All we want to do here is clear our unlock | |
2316 | * state. The wake_up call done at the bottom is redundant | |
2317 | * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't | |
2318 | * hurt anything anyway */ | |
2319 | if (status == DLM_CANCELGRANT && | |
2320 | lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { | |
2321 | mlog(0, "Got cancelgrant for %s\n", lockres->l_name); | |
2322 | ||
2323 | /* We don't clear the busy flag in this case as it | |
2324 | * should have been cleared by the ast which the dlm | |
2325 | * has called. */ | |
2326 | goto complete_unlock; | |
2327 | } | |
2328 | ||
2329 | if (status != DLM_NORMAL) { | |
2330 | mlog(ML_ERROR, "Dlm passes status %d for lock %s, " | |
2331 | "unlock_action %d\n", status, lockres->l_name, | |
2332 | lockres->l_unlock_action); | |
2333 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2334 | return; | |
2335 | } | |
2336 | ||
2337 | switch(lockres->l_unlock_action) { | |
2338 | case OCFS2_UNLOCK_CANCEL_CONVERT: | |
2339 | mlog(0, "Cancel convert success for %s\n", lockres->l_name); | |
2340 | lockres->l_action = OCFS2_AST_INVALID; | |
2341 | break; | |
2342 | case OCFS2_UNLOCK_DROP_LOCK: | |
2343 | lockres->l_level = LKM_IVMODE; | |
2344 | break; | |
2345 | default: | |
2346 | BUG(); | |
2347 | } | |
2348 | ||
2349 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
2350 | complete_unlock: | |
2351 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
2352 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2353 | ||
2354 | wake_up(&lockres->l_event); | |
2355 | ||
2356 | mlog_exit_void(); | |
2357 | } | |
2358 | ||
2359 | typedef void (ocfs2_pre_drop_cb_t)(struct ocfs2_lock_res *, void *); | |
2360 | ||
2361 | struct drop_lock_cb { | |
2362 | ocfs2_pre_drop_cb_t *drop_func; | |
2363 | void *drop_data; | |
2364 | }; | |
2365 | ||
2366 | static int ocfs2_drop_lock(struct ocfs2_super *osb, | |
2367 | struct ocfs2_lock_res *lockres, | |
2368 | struct drop_lock_cb *dcb) | |
2369 | { | |
2370 | enum dlm_status status; | |
2371 | unsigned long flags; | |
2372 | ||
2373 | /* We didn't get anywhere near actually using this lockres. */ | |
2374 | if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) | |
2375 | goto out; | |
2376 | ||
2377 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2378 | ||
2379 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING), | |
2380 | "lockres %s, flags 0x%lx\n", | |
2381 | lockres->l_name, lockres->l_flags); | |
2382 | ||
2383 | while (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
2384 | mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = " | |
2385 | "%u, unlock_action = %u\n", | |
2386 | lockres->l_name, lockres->l_flags, lockres->l_action, | |
2387 | lockres->l_unlock_action); | |
2388 | ||
2389 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2390 | ||
2391 | /* XXX: Today we just wait on any busy | |
2392 | * locks... Perhaps we need to cancel converts in the | |
2393 | * future? */ | |
2394 | ocfs2_wait_on_busy_lock(lockres); | |
2395 | ||
2396 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2397 | } | |
2398 | ||
2399 | if (dcb) | |
2400 | dcb->drop_func(lockres, dcb->drop_data); | |
2401 | ||
2402 | if (lockres->l_flags & OCFS2_LOCK_BUSY) | |
2403 | mlog(ML_ERROR, "destroying busy lock: \"%s\"\n", | |
2404 | lockres->l_name); | |
2405 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) | |
2406 | mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name); | |
2407 | ||
2408 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { | |
2409 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2410 | goto out; | |
2411 | } | |
2412 | ||
2413 | lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED); | |
2414 | ||
2415 | /* make sure we never get here while waiting for an ast to | |
2416 | * fire. */ | |
2417 | BUG_ON(lockres->l_action != OCFS2_AST_INVALID); | |
2418 | ||
2419 | /* is this necessary? */ | |
2420 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
2421 | lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK; | |
2422 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2423 | ||
2424 | mlog(0, "lock %s\n", lockres->l_name); | |
2425 | ||
2426 | status = dlmunlock(osb->dlm, &lockres->l_lksb, LKM_VALBLK, | |
2427 | lockres->l_ops->unlock_ast, lockres); | |
2428 | if (status != DLM_NORMAL) { | |
2429 | ocfs2_log_dlm_error("dlmunlock", status, lockres); | |
2430 | mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags); | |
2431 | dlm_print_one_lock(lockres->l_lksb.lockid); | |
2432 | BUG(); | |
2433 | } | |
2434 | mlog(0, "lock %s, successfull return from dlmunlock\n", | |
2435 | lockres->l_name); | |
2436 | ||
2437 | ocfs2_wait_on_busy_lock(lockres); | |
2438 | out: | |
2439 | mlog_exit(0); | |
2440 | return 0; | |
2441 | } | |
2442 | ||
2443 | /* Mark the lockres as being dropped. It will no longer be | |
2444 | * queued if blocking, but we still may have to wait on it | |
2445 | * being dequeued from the vote thread before we can consider | |
2446 | * it safe to drop. | |
2447 | * | |
2448 | * You can *not* attempt to call cluster_lock on this lockres anymore. */ | |
2449 | void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres) | |
2450 | { | |
2451 | int status; | |
2452 | struct ocfs2_mask_waiter mw; | |
2453 | unsigned long flags; | |
2454 | ||
2455 | ocfs2_init_mask_waiter(&mw); | |
2456 | ||
2457 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2458 | lockres->l_flags |= OCFS2_LOCK_FREEING; | |
2459 | while (lockres->l_flags & OCFS2_LOCK_QUEUED) { | |
2460 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0); | |
2461 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2462 | ||
2463 | mlog(0, "Waiting on lockres %s\n", lockres->l_name); | |
2464 | ||
2465 | status = ocfs2_wait_for_mask(&mw); | |
2466 | if (status) | |
2467 | mlog_errno(status); | |
2468 | ||
2469 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2470 | } | |
2471 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2472 | } | |
2473 | ||
d680efe9 MF |
2474 | void ocfs2_simple_drop_lockres(struct ocfs2_super *osb, |
2475 | struct ocfs2_lock_res *lockres) | |
ccd979bd | 2476 | { |
d680efe9 | 2477 | int ret; |
ccd979bd | 2478 | |
d680efe9 MF |
2479 | ocfs2_mark_lockres_freeing(lockres); |
2480 | ret = ocfs2_drop_lock(osb, lockres, NULL); | |
2481 | if (ret) | |
2482 | mlog_errno(ret); | |
2483 | } | |
ccd979bd | 2484 | |
d680efe9 MF |
2485 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb) |
2486 | { | |
2487 | ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres); | |
2488 | ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres); | |
ccd979bd MF |
2489 | } |
2490 | ||
2491 | static void ocfs2_meta_pre_drop(struct ocfs2_lock_res *lockres, void *data) | |
2492 | { | |
2493 | struct inode *inode = data; | |
2494 | ||
2495 | /* the metadata lock requires a bit more work as we have an | |
2496 | * LVB to worry about. */ | |
2497 | if (lockres->l_flags & OCFS2_LOCK_ATTACHED && | |
2498 | lockres->l_level == LKM_EXMODE && | |
2499 | !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) | |
2500 | __ocfs2_stuff_meta_lvb(inode); | |
2501 | } | |
2502 | ||
2503 | int ocfs2_drop_inode_locks(struct inode *inode) | |
2504 | { | |
2505 | int status, err; | |
2506 | struct drop_lock_cb meta_dcb = { ocfs2_meta_pre_drop, inode, }; | |
2507 | ||
2508 | mlog_entry_void(); | |
2509 | ||
2510 | /* No need to call ocfs2_mark_lockres_freeing here - | |
2511 | * ocfs2_clear_inode has done it for us. */ | |
2512 | ||
2513 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
2514 | &OCFS2_I(inode)->ip_data_lockres, | |
2515 | NULL); | |
2516 | if (err < 0) | |
2517 | mlog_errno(err); | |
2518 | ||
2519 | status = err; | |
2520 | ||
2521 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
2522 | &OCFS2_I(inode)->ip_meta_lockres, | |
2523 | &meta_dcb); | |
2524 | if (err < 0) | |
2525 | mlog_errno(err); | |
2526 | if (err < 0 && !status) | |
2527 | status = err; | |
2528 | ||
2529 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
2530 | &OCFS2_I(inode)->ip_rw_lockres, | |
2531 | NULL); | |
2532 | if (err < 0) | |
2533 | mlog_errno(err); | |
2534 | if (err < 0 && !status) | |
2535 | status = err; | |
2536 | ||
2537 | mlog_exit(status); | |
2538 | return status; | |
2539 | } | |
2540 | ||
2541 | static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, | |
2542 | int new_level) | |
2543 | { | |
2544 | assert_spin_locked(&lockres->l_lock); | |
2545 | ||
2546 | BUG_ON(lockres->l_blocking <= LKM_NLMODE); | |
2547 | ||
2548 | if (lockres->l_level <= new_level) { | |
2549 | mlog(ML_ERROR, "lockres->l_level (%u) <= new_level (%u)\n", | |
2550 | lockres->l_level, new_level); | |
2551 | BUG(); | |
2552 | } | |
2553 | ||
2554 | mlog(0, "lock %s, new_level = %d, l_blocking = %d\n", | |
2555 | lockres->l_name, new_level, lockres->l_blocking); | |
2556 | ||
2557 | lockres->l_action = OCFS2_AST_DOWNCONVERT; | |
2558 | lockres->l_requested = new_level; | |
2559 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
2560 | } | |
2561 | ||
2562 | static int ocfs2_downconvert_lock(struct ocfs2_super *osb, | |
2563 | struct ocfs2_lock_res *lockres, | |
2564 | int new_level, | |
2565 | int lvb) | |
2566 | { | |
2567 | int ret, dlm_flags = LKM_CONVERT; | |
2568 | enum dlm_status status; | |
2569 | ||
2570 | mlog_entry_void(); | |
2571 | ||
2572 | if (lvb) | |
2573 | dlm_flags |= LKM_VALBLK; | |
2574 | ||
2575 | status = dlmlock(osb->dlm, | |
2576 | new_level, | |
2577 | &lockres->l_lksb, | |
2578 | dlm_flags, | |
2579 | lockres->l_name, | |
f0681062 | 2580 | OCFS2_LOCK_ID_MAX_LEN - 1, |
ccd979bd MF |
2581 | lockres->l_ops->ast, |
2582 | lockres, | |
2583 | lockres->l_ops->bast); | |
2584 | if (status != DLM_NORMAL) { | |
2585 | ocfs2_log_dlm_error("dlmlock", status, lockres); | |
2586 | ret = -EINVAL; | |
2587 | ocfs2_recover_from_dlm_error(lockres, 1); | |
2588 | goto bail; | |
2589 | } | |
2590 | ||
2591 | ret = 0; | |
2592 | bail: | |
2593 | mlog_exit(ret); | |
2594 | return ret; | |
2595 | } | |
2596 | ||
2597 | /* returns 1 when the caller should unlock and call dlmunlock */ | |
2598 | static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, | |
2599 | struct ocfs2_lock_res *lockres) | |
2600 | { | |
2601 | assert_spin_locked(&lockres->l_lock); | |
2602 | ||
2603 | mlog_entry_void(); | |
2604 | mlog(0, "lock %s\n", lockres->l_name); | |
2605 | ||
2606 | if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { | |
2607 | /* If we're already trying to cancel a lock conversion | |
2608 | * then just drop the spinlock and allow the caller to | |
2609 | * requeue this lock. */ | |
2610 | ||
2611 | mlog(0, "Lockres %s, skip convert\n", lockres->l_name); | |
2612 | return 0; | |
2613 | } | |
2614 | ||
2615 | /* were we in a convert when we got the bast fire? */ | |
2616 | BUG_ON(lockres->l_action != OCFS2_AST_CONVERT && | |
2617 | lockres->l_action != OCFS2_AST_DOWNCONVERT); | |
2618 | /* set things up for the unlockast to know to just | |
2619 | * clear out the ast_action and unset busy, etc. */ | |
2620 | lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT; | |
2621 | ||
2622 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY), | |
2623 | "lock %s, invalid flags: 0x%lx\n", | |
2624 | lockres->l_name, lockres->l_flags); | |
2625 | ||
2626 | return 1; | |
2627 | } | |
2628 | ||
2629 | static int ocfs2_cancel_convert(struct ocfs2_super *osb, | |
2630 | struct ocfs2_lock_res *lockres) | |
2631 | { | |
2632 | int ret; | |
2633 | enum dlm_status status; | |
2634 | ||
2635 | mlog_entry_void(); | |
2636 | mlog(0, "lock %s\n", lockres->l_name); | |
2637 | ||
2638 | ret = 0; | |
2639 | status = dlmunlock(osb->dlm, | |
2640 | &lockres->l_lksb, | |
2641 | LKM_CANCEL, | |
2642 | lockres->l_ops->unlock_ast, | |
2643 | lockres); | |
2644 | if (status != DLM_NORMAL) { | |
2645 | ocfs2_log_dlm_error("dlmunlock", status, lockres); | |
2646 | ret = -EINVAL; | |
2647 | ocfs2_recover_from_dlm_error(lockres, 0); | |
2648 | } | |
2649 | ||
2650 | mlog(0, "lock %s return from dlmunlock\n", lockres->l_name); | |
2651 | ||
2652 | mlog_exit(ret); | |
2653 | return ret; | |
2654 | } | |
2655 | ||
2656 | static inline int ocfs2_can_downconvert_meta_lock(struct inode *inode, | |
2657 | struct ocfs2_lock_res *lockres, | |
2658 | int new_level) | |
2659 | { | |
2660 | int ret; | |
2661 | ||
2662 | mlog_entry_void(); | |
2663 | ||
2664 | BUG_ON(new_level != LKM_NLMODE && new_level != LKM_PRMODE); | |
2665 | ||
2666 | if (lockres->l_flags & OCFS2_LOCK_REFRESHING) { | |
2667 | ret = 0; | |
2668 | mlog(0, "lockres %s currently being refreshed -- backing " | |
2669 | "off!\n", lockres->l_name); | |
2670 | } else if (new_level == LKM_PRMODE) | |
2671 | ret = !lockres->l_ex_holders && | |
2672 | ocfs2_inode_fully_checkpointed(inode); | |
2673 | else /* Must be NLMODE we're converting to. */ | |
2674 | ret = !lockres->l_ro_holders && !lockres->l_ex_holders && | |
2675 | ocfs2_inode_fully_checkpointed(inode); | |
2676 | ||
2677 | mlog_exit(ret); | |
2678 | return ret; | |
2679 | } | |
2680 | ||
2681 | static int ocfs2_do_unblock_meta(struct inode *inode, | |
2682 | int *requeue) | |
2683 | { | |
2684 | int new_level; | |
2685 | int set_lvb = 0; | |
2686 | int ret = 0; | |
2687 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres; | |
2688 | unsigned long flags; | |
2689 | ||
2690 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2691 | ||
2692 | mlog_entry_void(); | |
2693 | ||
2694 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2695 | ||
2696 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
2697 | ||
2698 | mlog(0, "l_level=%d, l_blocking=%d\n", lockres->l_level, | |
2699 | lockres->l_blocking); | |
2700 | ||
2701 | BUG_ON(lockres->l_level != LKM_EXMODE && | |
2702 | lockres->l_level != LKM_PRMODE); | |
2703 | ||
2704 | if (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
2705 | *requeue = 1; | |
2706 | ret = ocfs2_prepare_cancel_convert(osb, lockres); | |
2707 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2708 | if (ret) { | |
2709 | ret = ocfs2_cancel_convert(osb, lockres); | |
2710 | if (ret < 0) | |
2711 | mlog_errno(ret); | |
2712 | } | |
2713 | goto leave; | |
2714 | } | |
2715 | ||
2716 | new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking); | |
2717 | ||
2718 | mlog(0, "l_level=%d, l_blocking=%d, new_level=%d\n", | |
2719 | lockres->l_level, lockres->l_blocking, new_level); | |
2720 | ||
2721 | if (ocfs2_can_downconvert_meta_lock(inode, lockres, new_level)) { | |
2722 | if (lockres->l_level == LKM_EXMODE) | |
2723 | set_lvb = 1; | |
2724 | ||
2725 | /* If the lock hasn't been refreshed yet (rare), then | |
2726 | * our memory inode values are old and we skip | |
2727 | * stuffing the lvb. There's no need to actually clear | |
2728 | * out the lvb here as it's value is still valid. */ | |
2729 | if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) { | |
2730 | if (set_lvb) | |
2731 | __ocfs2_stuff_meta_lvb(inode); | |
2732 | } else | |
2733 | mlog(0, "lockres %s: downconverting stale lock!\n", | |
2734 | lockres->l_name); | |
2735 | ||
2736 | mlog(0, "calling ocfs2_downconvert_lock with l_level=%d, " | |
2737 | "l_blocking=%d, new_level=%d\n", | |
2738 | lockres->l_level, lockres->l_blocking, new_level); | |
2739 | ||
2740 | ocfs2_prepare_downconvert(lockres, new_level); | |
2741 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2742 | ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb); | |
2743 | goto leave; | |
2744 | } | |
2745 | if (!ocfs2_inode_fully_checkpointed(inode)) | |
2746 | ocfs2_start_checkpoint(osb); | |
2747 | ||
2748 | *requeue = 1; | |
2749 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2750 | ret = 0; | |
2751 | leave: | |
2752 | mlog_exit(ret); | |
2753 | return ret; | |
2754 | } | |
2755 | ||
2756 | static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb, | |
2757 | struct ocfs2_lock_res *lockres, | |
d680efe9 | 2758 | struct ocfs2_unblock_ctl *ctl, |
ccd979bd MF |
2759 | ocfs2_convert_worker_t *worker) |
2760 | { | |
2761 | unsigned long flags; | |
2762 | int blocking; | |
2763 | int new_level; | |
2764 | int ret = 0; | |
2765 | ||
2766 | mlog_entry_void(); | |
2767 | ||
2768 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2769 | ||
2770 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
2771 | ||
2772 | recheck: | |
2773 | if (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
d680efe9 | 2774 | ctl->requeue = 1; |
ccd979bd MF |
2775 | ret = ocfs2_prepare_cancel_convert(osb, lockres); |
2776 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2777 | if (ret) { | |
2778 | ret = ocfs2_cancel_convert(osb, lockres); | |
2779 | if (ret < 0) | |
2780 | mlog_errno(ret); | |
2781 | } | |
2782 | goto leave; | |
2783 | } | |
2784 | ||
2785 | /* if we're blocking an exclusive and we have *any* holders, | |
2786 | * then requeue. */ | |
2787 | if ((lockres->l_blocking == LKM_EXMODE) | |
2788 | && (lockres->l_ex_holders || lockres->l_ro_holders)) { | |
2789 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
d680efe9 | 2790 | ctl->requeue = 1; |
ccd979bd MF |
2791 | ret = 0; |
2792 | goto leave; | |
2793 | } | |
2794 | ||
2795 | /* If it's a PR we're blocking, then only | |
2796 | * requeue if we've got any EX holders */ | |
2797 | if (lockres->l_blocking == LKM_PRMODE && | |
2798 | lockres->l_ex_holders) { | |
2799 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
d680efe9 | 2800 | ctl->requeue = 1; |
ccd979bd MF |
2801 | ret = 0; |
2802 | goto leave; | |
2803 | } | |
2804 | ||
2805 | /* If we get here, then we know that there are no more | |
2806 | * incompatible holders (and anyone asking for an incompatible | |
2807 | * lock is blocked). We can now downconvert the lock */ | |
2808 | if (!worker) | |
2809 | goto downconvert; | |
2810 | ||
2811 | /* Some lockres types want to do a bit of work before | |
2812 | * downconverting a lock. Allow that here. The worker function | |
2813 | * may sleep, so we save off a copy of what we're blocking as | |
2814 | * it may change while we're not holding the spin lock. */ | |
2815 | blocking = lockres->l_blocking; | |
2816 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2817 | ||
d680efe9 MF |
2818 | ctl->unblock_action = worker(lockres, blocking); |
2819 | ||
2820 | if (ctl->unblock_action == UNBLOCK_STOP_POST) | |
2821 | goto leave; | |
ccd979bd MF |
2822 | |
2823 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2824 | if (blocking != lockres->l_blocking) { | |
2825 | /* If this changed underneath us, then we can't drop | |
2826 | * it just yet. */ | |
2827 | goto recheck; | |
2828 | } | |
2829 | ||
2830 | downconvert: | |
d680efe9 | 2831 | ctl->requeue = 0; |
ccd979bd MF |
2832 | new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking); |
2833 | ||
2834 | ocfs2_prepare_downconvert(lockres, new_level); | |
2835 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2836 | ret = ocfs2_downconvert_lock(osb, lockres, new_level, 0); | |
2837 | leave: | |
2838 | mlog_exit(ret); | |
2839 | return ret; | |
2840 | } | |
2841 | ||
d680efe9 MF |
2842 | static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, |
2843 | int blocking) | |
ccd979bd MF |
2844 | { |
2845 | struct inode *inode; | |
2846 | struct address_space *mapping; | |
2847 | ||
ccd979bd MF |
2848 | inode = ocfs2_lock_res_inode(lockres); |
2849 | mapping = inode->i_mapping; | |
2850 | ||
2851 | if (filemap_fdatawrite(mapping)) { | |
b0697053 MF |
2852 | mlog(ML_ERROR, "Could not sync inode %llu for downconvert!", |
2853 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
2854 | } |
2855 | sync_mapping_buffers(mapping); | |
2856 | if (blocking == LKM_EXMODE) { | |
2857 | truncate_inode_pages(mapping, 0); | |
2858 | unmap_mapping_range(mapping, 0, 0, 0); | |
2859 | } else { | |
2860 | /* We only need to wait on the I/O if we're not also | |
2861 | * truncating pages because truncate_inode_pages waits | |
2862 | * for us above. We don't truncate pages if we're | |
2863 | * blocking anything < EXMODE because we want to keep | |
2864 | * them around in that case. */ | |
2865 | filemap_fdatawait(mapping); | |
2866 | } | |
2867 | ||
d680efe9 | 2868 | return UNBLOCK_CONTINUE; |
ccd979bd MF |
2869 | } |
2870 | ||
2871 | int ocfs2_unblock_data(struct ocfs2_lock_res *lockres, | |
d680efe9 | 2872 | struct ocfs2_unblock_ctl *ctl) |
ccd979bd MF |
2873 | { |
2874 | int status; | |
2875 | struct inode *inode; | |
2876 | struct ocfs2_super *osb; | |
2877 | ||
2878 | mlog_entry_void(); | |
2879 | ||
2880 | inode = ocfs2_lock_res_inode(lockres); | |
2881 | osb = OCFS2_SB(inode->i_sb); | |
2882 | ||
b0697053 MF |
2883 | mlog(0, "unblock inode %llu\n", |
2884 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 2885 | |
d680efe9 | 2886 | status = ocfs2_generic_unblock_lock(osb, lockres, ctl, |
ccd979bd MF |
2887 | ocfs2_data_convert_worker); |
2888 | if (status < 0) | |
2889 | mlog_errno(status); | |
2890 | ||
b0697053 | 2891 | mlog(0, "inode %llu, requeue = %d\n", |
d680efe9 | 2892 | (unsigned long long)OCFS2_I(inode)->ip_blkno, ctl->requeue); |
ccd979bd MF |
2893 | |
2894 | mlog_exit(status); | |
2895 | return status; | |
2896 | } | |
2897 | ||
2898 | static int ocfs2_unblock_inode_lock(struct ocfs2_lock_res *lockres, | |
d680efe9 | 2899 | struct ocfs2_unblock_ctl *ctl) |
ccd979bd MF |
2900 | { |
2901 | int status; | |
2902 | struct inode *inode; | |
2903 | ||
2904 | mlog_entry_void(); | |
2905 | ||
2906 | mlog(0, "Unblock lockres %s\n", lockres->l_name); | |
2907 | ||
2908 | inode = ocfs2_lock_res_inode(lockres); | |
2909 | ||
2910 | status = ocfs2_generic_unblock_lock(OCFS2_SB(inode->i_sb), | |
d680efe9 | 2911 | lockres, ctl, NULL); |
ccd979bd MF |
2912 | if (status < 0) |
2913 | mlog_errno(status); | |
2914 | ||
2915 | mlog_exit(status); | |
2916 | return status; | |
2917 | } | |
2918 | ||
d680efe9 MF |
2919 | static int ocfs2_unblock_meta(struct ocfs2_lock_res *lockres, |
2920 | struct ocfs2_unblock_ctl *ctl) | |
ccd979bd MF |
2921 | { |
2922 | int status; | |
2923 | struct inode *inode; | |
2924 | ||
2925 | mlog_entry_void(); | |
2926 | ||
2927 | inode = ocfs2_lock_res_inode(lockres); | |
2928 | ||
b0697053 MF |
2929 | mlog(0, "unblock inode %llu\n", |
2930 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 2931 | |
d680efe9 | 2932 | status = ocfs2_do_unblock_meta(inode, &ctl->requeue); |
ccd979bd MF |
2933 | if (status < 0) |
2934 | mlog_errno(status); | |
2935 | ||
b0697053 | 2936 | mlog(0, "inode %llu, requeue = %d\n", |
d680efe9 | 2937 | (unsigned long long)OCFS2_I(inode)->ip_blkno, ctl->requeue); |
ccd979bd MF |
2938 | |
2939 | mlog_exit(status); | |
2940 | return status; | |
2941 | } | |
2942 | ||
d680efe9 MF |
2943 | /* |
2944 | * Does the final reference drop on our dentry lock. Right now this | |
2945 | * happens in the vote thread, but we could choose to simplify the | |
2946 | * dlmglue API and push these off to the ocfs2_wq in the future. | |
2947 | */ | |
2948 | static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, | |
2949 | struct ocfs2_lock_res *lockres) | |
2950 | { | |
2951 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
2952 | ocfs2_dentry_lock_put(osb, dl); | |
2953 | } | |
2954 | ||
2955 | /* | |
2956 | * d_delete() matching dentries before the lock downconvert. | |
2957 | * | |
2958 | * At this point, any process waiting to destroy the | |
2959 | * dentry_lock due to last ref count is stopped by the | |
2960 | * OCFS2_LOCK_QUEUED flag. | |
2961 | * | |
2962 | * We have two potential problems | |
2963 | * | |
2964 | * 1) If we do the last reference drop on our dentry_lock (via dput) | |
2965 | * we'll wind up in ocfs2_release_dentry_lock(), waiting on | |
2966 | * the downconvert to finish. Instead we take an elevated | |
2967 | * reference and push the drop until after we've completed our | |
2968 | * unblock processing. | |
2969 | * | |
2970 | * 2) There might be another process with a final reference, | |
2971 | * waiting on us to finish processing. If this is the case, we | |
2972 | * detect it and exit out - there's no more dentries anyway. | |
2973 | */ | |
2974 | static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, | |
2975 | int blocking) | |
2976 | { | |
2977 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
2978 | struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode); | |
2979 | struct dentry *dentry; | |
2980 | unsigned long flags; | |
2981 | int extra_ref = 0; | |
2982 | ||
2983 | /* | |
2984 | * This node is blocking another node from getting a read | |
2985 | * lock. This happens when we've renamed within a | |
2986 | * directory. We've forced the other nodes to d_delete(), but | |
2987 | * we never actually dropped our lock because it's still | |
2988 | * valid. The downconvert code will retain a PR for this node, | |
2989 | * so there's no further work to do. | |
2990 | */ | |
2991 | if (blocking == LKM_PRMODE) | |
2992 | return UNBLOCK_CONTINUE; | |
2993 | ||
2994 | /* | |
2995 | * Mark this inode as potentially orphaned. The code in | |
2996 | * ocfs2_delete_inode() will figure out whether it actually | |
2997 | * needs to be freed or not. | |
2998 | */ | |
2999 | spin_lock(&oi->ip_lock); | |
3000 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
3001 | spin_unlock(&oi->ip_lock); | |
3002 | ||
3003 | /* | |
3004 | * Yuck. We need to make sure however that the check of | |
3005 | * OCFS2_LOCK_FREEING and the extra reference are atomic with | |
3006 | * respect to a reference decrement or the setting of that | |
3007 | * flag. | |
3008 | */ | |
3009 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3010 | spin_lock(&dentry_attach_lock); | |
3011 | if (!(lockres->l_flags & OCFS2_LOCK_FREEING) | |
3012 | && dl->dl_count) { | |
3013 | dl->dl_count++; | |
3014 | extra_ref = 1; | |
3015 | } | |
3016 | spin_unlock(&dentry_attach_lock); | |
3017 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
3018 | ||
3019 | mlog(0, "extra_ref = %d\n", extra_ref); | |
3020 | ||
3021 | /* | |
3022 | * We have a process waiting on us in ocfs2_dentry_iput(), | |
3023 | * which means we can't have any more outstanding | |
3024 | * aliases. There's no need to do any more work. | |
3025 | */ | |
3026 | if (!extra_ref) | |
3027 | return UNBLOCK_CONTINUE; | |
3028 | ||
3029 | spin_lock(&dentry_attach_lock); | |
3030 | while (1) { | |
3031 | dentry = ocfs2_find_local_alias(dl->dl_inode, | |
3032 | dl->dl_parent_blkno, 1); | |
3033 | if (!dentry) | |
3034 | break; | |
3035 | spin_unlock(&dentry_attach_lock); | |
3036 | ||
3037 | mlog(0, "d_delete(%.*s);\n", dentry->d_name.len, | |
3038 | dentry->d_name.name); | |
3039 | ||
3040 | /* | |
3041 | * The following dcache calls may do an | |
3042 | * iput(). Normally we don't want that from the | |
3043 | * downconverting thread, but in this case it's ok | |
3044 | * because the requesting node already has an | |
3045 | * exclusive lock on the inode, so it can't be queued | |
3046 | * for a downconvert. | |
3047 | */ | |
3048 | d_delete(dentry); | |
3049 | dput(dentry); | |
3050 | ||
3051 | spin_lock(&dentry_attach_lock); | |
3052 | } | |
3053 | spin_unlock(&dentry_attach_lock); | |
3054 | ||
3055 | /* | |
3056 | * If we are the last holder of this dentry lock, there is no | |
3057 | * reason to downconvert so skip straight to the unlock. | |
3058 | */ | |
3059 | if (dl->dl_count == 1) | |
3060 | return UNBLOCK_STOP_POST; | |
3061 | ||
3062 | return UNBLOCK_CONTINUE_POST; | |
3063 | } | |
3064 | ||
3065 | static int ocfs2_unblock_dentry_lock(struct ocfs2_lock_res *lockres, | |
3066 | struct ocfs2_unblock_ctl *ctl) | |
3067 | { | |
3068 | int ret; | |
3069 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
3070 | struct ocfs2_super *osb = OCFS2_SB(dl->dl_inode->i_sb); | |
3071 | ||
3072 | mlog(0, "unblock dentry lock: %llu\n", | |
3073 | (unsigned long long)OCFS2_I(dl->dl_inode)->ip_blkno); | |
3074 | ||
3075 | ret = ocfs2_generic_unblock_lock(osb, | |
3076 | lockres, | |
3077 | ctl, | |
3078 | ocfs2_dentry_convert_worker); | |
3079 | if (ret < 0) | |
3080 | mlog_errno(ret); | |
3081 | ||
3082 | mlog(0, "requeue = %d, post = %d\n", ctl->requeue, ctl->unblock_action); | |
3083 | ||
3084 | return ret; | |
3085 | } | |
3086 | ||
ccd979bd MF |
3087 | /* Generic unblock function for any lockres whose private data is an |
3088 | * ocfs2_super pointer. */ | |
3089 | static int ocfs2_unblock_osb_lock(struct ocfs2_lock_res *lockres, | |
d680efe9 | 3090 | struct ocfs2_unblock_ctl *ctl) |
ccd979bd MF |
3091 | { |
3092 | int status; | |
3093 | struct ocfs2_super *osb; | |
3094 | ||
3095 | mlog_entry_void(); | |
3096 | ||
3097 | mlog(0, "Unblock lockres %s\n", lockres->l_name); | |
3098 | ||
3099 | osb = ocfs2_lock_res_super(lockres); | |
3100 | ||
3101 | status = ocfs2_generic_unblock_lock(osb, | |
3102 | lockres, | |
d680efe9 | 3103 | ctl, |
ccd979bd MF |
3104 | NULL); |
3105 | if (status < 0) | |
3106 | mlog_errno(status); | |
3107 | ||
3108 | mlog_exit(status); | |
3109 | return status; | |
3110 | } | |
3111 | ||
3112 | void ocfs2_process_blocked_lock(struct ocfs2_super *osb, | |
3113 | struct ocfs2_lock_res *lockres) | |
3114 | { | |
3115 | int status; | |
d680efe9 | 3116 | struct ocfs2_unblock_ctl ctl = {0, 0,}; |
ccd979bd MF |
3117 | unsigned long flags; |
3118 | ||
3119 | /* Our reference to the lockres in this function can be | |
3120 | * considered valid until we remove the OCFS2_LOCK_QUEUED | |
3121 | * flag. */ | |
3122 | ||
3123 | mlog_entry_void(); | |
3124 | ||
3125 | BUG_ON(!lockres); | |
3126 | BUG_ON(!lockres->l_ops); | |
3127 | BUG_ON(!lockres->l_ops->unblock); | |
3128 | ||
3129 | mlog(0, "lockres %s blocked.\n", lockres->l_name); | |
3130 | ||
3131 | /* Detect whether a lock has been marked as going away while | |
3132 | * the vote thread was processing other things. A lock can | |
3133 | * still be marked with OCFS2_LOCK_FREEING after this check, | |
3134 | * but short circuiting here will still save us some | |
3135 | * performance. */ | |
3136 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3137 | if (lockres->l_flags & OCFS2_LOCK_FREEING) | |
3138 | goto unqueue; | |
3139 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
3140 | ||
d680efe9 | 3141 | status = lockres->l_ops->unblock(lockres, &ctl); |
ccd979bd MF |
3142 | if (status < 0) |
3143 | mlog_errno(status); | |
3144 | ||
3145 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3146 | unqueue: | |
d680efe9 | 3147 | if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) { |
ccd979bd MF |
3148 | lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED); |
3149 | } else | |
3150 | ocfs2_schedule_blocked_lock(osb, lockres); | |
3151 | ||
3152 | mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name, | |
d680efe9 | 3153 | ctl.requeue ? "yes" : "no"); |
ccd979bd MF |
3154 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3155 | ||
d680efe9 MF |
3156 | if (ctl.unblock_action != UNBLOCK_CONTINUE |
3157 | && lockres->l_ops->post_unlock) | |
3158 | lockres->l_ops->post_unlock(osb, lockres); | |
3159 | ||
ccd979bd MF |
3160 | mlog_exit_void(); |
3161 | } | |
3162 | ||
3163 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, | |
3164 | struct ocfs2_lock_res *lockres) | |
3165 | { | |
3166 | mlog_entry_void(); | |
3167 | ||
3168 | assert_spin_locked(&lockres->l_lock); | |
3169 | ||
3170 | if (lockres->l_flags & OCFS2_LOCK_FREEING) { | |
3171 | /* Do not schedule a lock for downconvert when it's on | |
3172 | * the way to destruction - any nodes wanting access | |
3173 | * to the resource will get it soon. */ | |
3174 | mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n", | |
3175 | lockres->l_name, lockres->l_flags); | |
3176 | return; | |
3177 | } | |
3178 | ||
3179 | lockres_or_flags(lockres, OCFS2_LOCK_QUEUED); | |
3180 | ||
3181 | spin_lock(&osb->vote_task_lock); | |
3182 | if (list_empty(&lockres->l_blocked_list)) { | |
3183 | list_add_tail(&lockres->l_blocked_list, | |
3184 | &osb->blocked_lock_list); | |
3185 | osb->blocked_lock_count++; | |
3186 | } | |
3187 | spin_unlock(&osb->vote_task_lock); | |
3188 | ||
3189 | mlog_exit_void(); | |
3190 | } | |
3191 | ||
3192 | /* This aids in debugging situations where a bad LVB might be involved. */ | |
3193 | void ocfs2_dump_meta_lvb_info(u64 level, | |
3194 | const char *function, | |
3195 | unsigned int line, | |
3196 | struct ocfs2_lock_res *lockres) | |
3197 | { | |
3198 | struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
3199 | ||
3200 | mlog(level, "LVB information for %s (called from %s:%u):\n", | |
3201 | lockres->l_name, function, line); | |
f9e2d82e MF |
3202 | mlog(level, "version: %u, clusters: %u, generation: 0x%x\n", |
3203 | lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters), | |
3204 | be32_to_cpu(lvb->lvb_igeneration)); | |
b0697053 MF |
3205 | mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n", |
3206 | (unsigned long long)be64_to_cpu(lvb->lvb_isize), | |
3207 | be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid), | |
3208 | be16_to_cpu(lvb->lvb_imode)); | |
3209 | mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, " | |
ca4d147e | 3210 | "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink), |
b0697053 MF |
3211 | (long long)be64_to_cpu(lvb->lvb_iatime_packed), |
3212 | (long long)be64_to_cpu(lvb->lvb_ictime_packed), | |
ca4d147e HP |
3213 | (long long)be64_to_cpu(lvb->lvb_imtime_packed), |
3214 | be32_to_cpu(lvb->lvb_iattr)); | |
ccd979bd | 3215 | } |