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 | 119 | |
f625c979 MF |
120 | /* |
121 | * OCFS2 Lock Resource Operations | |
122 | * | |
123 | * These fine tune the behavior of the generic dlmglue locking infrastructure. | |
124 | */ | |
ccd979bd MF |
125 | struct ocfs2_lock_res_ops { |
126 | void (*ast)(void *); | |
127 | void (*bast)(void *, int); | |
128 | void (*unlock_ast)(void *, enum dlm_status); | |
d680efe9 MF |
129 | int (*unblock)(struct ocfs2_lock_res *, struct ocfs2_unblock_ctl *); |
130 | void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *); | |
f625c979 MF |
131 | |
132 | /* | |
133 | * LOCK_TYPE_* flags which describe the specific requirements | |
134 | * of a lock type. Descriptions of each individual flag follow. | |
135 | */ | |
136 | int flags; | |
ccd979bd MF |
137 | }; |
138 | ||
f625c979 MF |
139 | /* |
140 | * Some locks want to "refresh" potentially stale data when a | |
141 | * meaningful (PRMODE or EXMODE) lock level is first obtained. If this | |
142 | * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the | |
143 | * individual lockres l_flags member from the ast function. It is | |
144 | * expected that the locking wrapper will clear the | |
145 | * OCFS2_LOCK_NEEDS_REFRESH flag when done. | |
146 | */ | |
147 | #define LOCK_TYPE_REQUIRES_REFRESH 0x1 | |
148 | ||
d680efe9 MF |
149 | typedef int (ocfs2_convert_worker_t)(struct ocfs2_lock_res *, int); |
150 | static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb, | |
151 | struct ocfs2_lock_res *lockres, | |
152 | struct ocfs2_unblock_ctl *ctl, | |
153 | ocfs2_convert_worker_t *worker); | |
154 | ||
ccd979bd MF |
155 | static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = { |
156 | .ast = ocfs2_inode_ast_func, | |
157 | .bast = ocfs2_inode_bast_func, | |
158 | .unlock_ast = ocfs2_unlock_ast_func, | |
159 | .unblock = ocfs2_unblock_inode_lock, | |
f625c979 | 160 | .flags = 0, |
ccd979bd MF |
161 | }; |
162 | ||
163 | static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops = { | |
164 | .ast = ocfs2_inode_ast_func, | |
165 | .bast = ocfs2_inode_bast_func, | |
166 | .unlock_ast = ocfs2_unlock_ast_func, | |
167 | .unblock = ocfs2_unblock_meta, | |
f625c979 | 168 | .flags = LOCK_TYPE_REQUIRES_REFRESH, |
ccd979bd MF |
169 | }; |
170 | ||
ccd979bd MF |
171 | static struct ocfs2_lock_res_ops ocfs2_inode_data_lops = { |
172 | .ast = ocfs2_inode_ast_func, | |
173 | .bast = ocfs2_inode_bast_func, | |
174 | .unlock_ast = ocfs2_unlock_ast_func, | |
175 | .unblock = ocfs2_unblock_data, | |
f625c979 | 176 | .flags = 0, |
ccd979bd MF |
177 | }; |
178 | ||
179 | static struct ocfs2_lock_res_ops ocfs2_super_lops = { | |
180 | .ast = ocfs2_super_ast_func, | |
181 | .bast = ocfs2_super_bast_func, | |
182 | .unlock_ast = ocfs2_unlock_ast_func, | |
183 | .unblock = ocfs2_unblock_osb_lock, | |
f625c979 | 184 | .flags = LOCK_TYPE_REQUIRES_REFRESH, |
ccd979bd MF |
185 | }; |
186 | ||
187 | static struct ocfs2_lock_res_ops ocfs2_rename_lops = { | |
188 | .ast = ocfs2_rename_ast_func, | |
189 | .bast = ocfs2_rename_bast_func, | |
190 | .unlock_ast = ocfs2_unlock_ast_func, | |
191 | .unblock = ocfs2_unblock_osb_lock, | |
f625c979 | 192 | .flags = 0, |
ccd979bd MF |
193 | }; |
194 | ||
d680efe9 MF |
195 | static struct ocfs2_lock_res_ops ocfs2_dentry_lops = { |
196 | .ast = ocfs2_dentry_ast_func, | |
197 | .bast = ocfs2_dentry_bast_func, | |
198 | .unlock_ast = ocfs2_unlock_ast_func, | |
199 | .unblock = ocfs2_unblock_dentry_lock, | |
200 | .post_unlock = ocfs2_dentry_post_unlock, | |
f625c979 | 201 | .flags = 0, |
d680efe9 MF |
202 | }; |
203 | ||
ccd979bd MF |
204 | static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres) |
205 | { | |
206 | return lockres->l_type == OCFS2_LOCK_TYPE_META || | |
207 | lockres->l_type == OCFS2_LOCK_TYPE_DATA || | |
208 | lockres->l_type == OCFS2_LOCK_TYPE_RW; | |
209 | } | |
210 | ||
211 | static inline int ocfs2_is_super_lock(struct ocfs2_lock_res *lockres) | |
212 | { | |
213 | return lockres->l_type == OCFS2_LOCK_TYPE_SUPER; | |
214 | } | |
215 | ||
216 | static inline int ocfs2_is_rename_lock(struct ocfs2_lock_res *lockres) | |
217 | { | |
218 | return lockres->l_type == OCFS2_LOCK_TYPE_RENAME; | |
219 | } | |
220 | ||
221 | static inline struct ocfs2_super *ocfs2_lock_res_super(struct ocfs2_lock_res *lockres) | |
222 | { | |
223 | BUG_ON(!ocfs2_is_super_lock(lockres) | |
224 | && !ocfs2_is_rename_lock(lockres)); | |
225 | ||
226 | return (struct ocfs2_super *) lockres->l_priv; | |
227 | } | |
228 | ||
229 | static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres) | |
230 | { | |
231 | BUG_ON(!ocfs2_is_inode_lock(lockres)); | |
232 | ||
233 | return (struct inode *) lockres->l_priv; | |
234 | } | |
235 | ||
d680efe9 MF |
236 | static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres) |
237 | { | |
238 | BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY); | |
239 | ||
240 | return (struct ocfs2_dentry_lock *)lockres->l_priv; | |
241 | } | |
242 | ||
ccd979bd MF |
243 | static int ocfs2_lock_create(struct ocfs2_super *osb, |
244 | struct ocfs2_lock_res *lockres, | |
245 | int level, | |
246 | int dlm_flags); | |
247 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, | |
248 | int wanted); | |
249 | static void ocfs2_cluster_unlock(struct ocfs2_super *osb, | |
250 | struct ocfs2_lock_res *lockres, | |
251 | int level); | |
252 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres); | |
253 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres); | |
254 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres); | |
255 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level); | |
256 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, | |
257 | struct ocfs2_lock_res *lockres); | |
258 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, | |
259 | int convert); | |
260 | #define ocfs2_log_dlm_error(_func, _stat, _lockres) do { \ | |
261 | mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \ | |
262 | "resource %s: %s\n", dlm_errname(_stat), _func, \ | |
263 | _lockres->l_name, dlm_errmsg(_stat)); \ | |
264 | } while (0) | |
265 | static void ocfs2_vote_on_unlock(struct ocfs2_super *osb, | |
266 | struct ocfs2_lock_res *lockres); | |
267 | static int ocfs2_meta_lock_update(struct inode *inode, | |
268 | struct buffer_head **bh); | |
269 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb); | |
270 | static inline int ocfs2_highest_compat_lock_level(int level); | |
271 | static inline int ocfs2_can_downconvert_meta_lock(struct inode *inode, | |
272 | struct ocfs2_lock_res *lockres, | |
273 | int new_level); | |
274 | ||
ccd979bd MF |
275 | static void ocfs2_build_lock_name(enum ocfs2_lock_type type, |
276 | u64 blkno, | |
277 | u32 generation, | |
278 | char *name) | |
279 | { | |
280 | int len; | |
281 | ||
282 | mlog_entry_void(); | |
283 | ||
284 | BUG_ON(type >= OCFS2_NUM_LOCK_TYPES); | |
285 | ||
b0697053 MF |
286 | len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x", |
287 | ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD, | |
288 | (long long)blkno, generation); | |
ccd979bd MF |
289 | |
290 | BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1)); | |
291 | ||
292 | mlog(0, "built lock resource with name: %s\n", name); | |
293 | ||
294 | mlog_exit_void(); | |
295 | } | |
296 | ||
34af946a | 297 | static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock); |
ccd979bd MF |
298 | |
299 | static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res, | |
300 | struct ocfs2_dlm_debug *dlm_debug) | |
301 | { | |
302 | mlog(0, "Add tracking for lockres %s\n", res->l_name); | |
303 | ||
304 | spin_lock(&ocfs2_dlm_tracking_lock); | |
305 | list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking); | |
306 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
307 | } | |
308 | ||
309 | static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res) | |
310 | { | |
311 | spin_lock(&ocfs2_dlm_tracking_lock); | |
312 | if (!list_empty(&res->l_debug_list)) | |
313 | list_del_init(&res->l_debug_list); | |
314 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
315 | } | |
316 | ||
317 | static void ocfs2_lock_res_init_common(struct ocfs2_super *osb, | |
318 | struct ocfs2_lock_res *res, | |
319 | enum ocfs2_lock_type type, | |
ccd979bd MF |
320 | struct ocfs2_lock_res_ops *ops, |
321 | void *priv) | |
322 | { | |
ccd979bd MF |
323 | res->l_type = type; |
324 | res->l_ops = ops; | |
325 | res->l_priv = priv; | |
326 | ||
327 | res->l_level = LKM_IVMODE; | |
328 | res->l_requested = LKM_IVMODE; | |
329 | res->l_blocking = LKM_IVMODE; | |
330 | res->l_action = OCFS2_AST_INVALID; | |
331 | res->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
332 | ||
333 | res->l_flags = OCFS2_LOCK_INITIALIZED; | |
334 | ||
335 | ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug); | |
336 | } | |
337 | ||
338 | void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res) | |
339 | { | |
340 | /* This also clears out the lock status block */ | |
341 | memset(res, 0, sizeof(struct ocfs2_lock_res)); | |
342 | spin_lock_init(&res->l_lock); | |
343 | init_waitqueue_head(&res->l_event); | |
344 | INIT_LIST_HEAD(&res->l_blocked_list); | |
345 | INIT_LIST_HEAD(&res->l_mask_waiters); | |
346 | } | |
347 | ||
348 | void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res, | |
349 | enum ocfs2_lock_type type, | |
24c19ef4 | 350 | unsigned int generation, |
ccd979bd MF |
351 | struct inode *inode) |
352 | { | |
353 | struct ocfs2_lock_res_ops *ops; | |
354 | ||
355 | switch(type) { | |
356 | case OCFS2_LOCK_TYPE_RW: | |
357 | ops = &ocfs2_inode_rw_lops; | |
358 | break; | |
359 | case OCFS2_LOCK_TYPE_META: | |
360 | ops = &ocfs2_inode_meta_lops; | |
361 | break; | |
362 | case OCFS2_LOCK_TYPE_DATA: | |
363 | ops = &ocfs2_inode_data_lops; | |
364 | break; | |
365 | default: | |
366 | mlog_bug_on_msg(1, "type: %d\n", type); | |
367 | ops = NULL; /* thanks, gcc */ | |
368 | break; | |
369 | }; | |
370 | ||
d680efe9 | 371 | ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno, |
24c19ef4 | 372 | generation, res->l_name); |
d680efe9 MF |
373 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode); |
374 | } | |
375 | ||
376 | static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres) | |
377 | { | |
378 | __be64 inode_blkno_be; | |
379 | ||
380 | memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], | |
381 | sizeof(__be64)); | |
382 | ||
383 | return be64_to_cpu(inode_blkno_be); | |
384 | } | |
385 | ||
386 | void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl, | |
387 | u64 parent, struct inode *inode) | |
388 | { | |
389 | int len; | |
390 | u64 inode_blkno = OCFS2_I(inode)->ip_blkno; | |
391 | __be64 inode_blkno_be = cpu_to_be64(inode_blkno); | |
392 | struct ocfs2_lock_res *lockres = &dl->dl_lockres; | |
393 | ||
394 | ocfs2_lock_res_init_once(lockres); | |
395 | ||
396 | /* | |
397 | * Unfortunately, the standard lock naming scheme won't work | |
398 | * here because we have two 16 byte values to use. Instead, | |
399 | * we'll stuff the inode number as a binary value. We still | |
400 | * want error prints to show something without garbling the | |
401 | * display, so drop a null byte in there before the inode | |
402 | * number. A future version of OCFS2 will likely use all | |
403 | * binary lock names. The stringified names have been a | |
404 | * tremendous aid in debugging, but now that the debugfs | |
405 | * interface exists, we can mangle things there if need be. | |
406 | * | |
407 | * NOTE: We also drop the standard "pad" value (the total lock | |
408 | * name size stays the same though - the last part is all | |
409 | * zeros due to the memset in ocfs2_lock_res_init_once() | |
410 | */ | |
411 | len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START, | |
412 | "%c%016llx", | |
413 | ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY), | |
414 | (long long)parent); | |
415 | ||
416 | BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1)); | |
417 | ||
418 | memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be, | |
419 | sizeof(__be64)); | |
420 | ||
421 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres, | |
422 | OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops, | |
423 | dl); | |
ccd979bd MF |
424 | } |
425 | ||
426 | static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res, | |
427 | struct ocfs2_super *osb) | |
428 | { | |
429 | /* Superblock lockres doesn't come from a slab so we call init | |
430 | * once on it manually. */ | |
431 | ocfs2_lock_res_init_once(res); | |
d680efe9 MF |
432 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO, |
433 | 0, res->l_name); | |
ccd979bd | 434 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER, |
ccd979bd MF |
435 | &ocfs2_super_lops, osb); |
436 | } | |
437 | ||
438 | static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res, | |
439 | struct ocfs2_super *osb) | |
440 | { | |
441 | /* Rename lockres doesn't come from a slab so we call init | |
442 | * once on it manually. */ | |
443 | ocfs2_lock_res_init_once(res); | |
d680efe9 MF |
444 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name); |
445 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME, | |
ccd979bd MF |
446 | &ocfs2_rename_lops, osb); |
447 | } | |
448 | ||
449 | void ocfs2_lock_res_free(struct ocfs2_lock_res *res) | |
450 | { | |
451 | mlog_entry_void(); | |
452 | ||
453 | if (!(res->l_flags & OCFS2_LOCK_INITIALIZED)) | |
454 | return; | |
455 | ||
456 | ocfs2_remove_lockres_tracking(res); | |
457 | ||
458 | mlog_bug_on_msg(!list_empty(&res->l_blocked_list), | |
459 | "Lockres %s is on the blocked list\n", | |
460 | res->l_name); | |
461 | mlog_bug_on_msg(!list_empty(&res->l_mask_waiters), | |
462 | "Lockres %s has mask waiters pending\n", | |
463 | res->l_name); | |
464 | mlog_bug_on_msg(spin_is_locked(&res->l_lock), | |
465 | "Lockres %s is locked\n", | |
466 | res->l_name); | |
467 | mlog_bug_on_msg(res->l_ro_holders, | |
468 | "Lockres %s has %u ro holders\n", | |
469 | res->l_name, res->l_ro_holders); | |
470 | mlog_bug_on_msg(res->l_ex_holders, | |
471 | "Lockres %s has %u ex holders\n", | |
472 | res->l_name, res->l_ex_holders); | |
473 | ||
474 | /* Need to clear out the lock status block for the dlm */ | |
475 | memset(&res->l_lksb, 0, sizeof(res->l_lksb)); | |
476 | ||
477 | res->l_flags = 0UL; | |
478 | mlog_exit_void(); | |
479 | } | |
480 | ||
481 | static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres, | |
482 | int level) | |
483 | { | |
484 | mlog_entry_void(); | |
485 | ||
486 | BUG_ON(!lockres); | |
487 | ||
488 | switch(level) { | |
489 | case LKM_EXMODE: | |
490 | lockres->l_ex_holders++; | |
491 | break; | |
492 | case LKM_PRMODE: | |
493 | lockres->l_ro_holders++; | |
494 | break; | |
495 | default: | |
496 | BUG(); | |
497 | } | |
498 | ||
499 | mlog_exit_void(); | |
500 | } | |
501 | ||
502 | static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres, | |
503 | int level) | |
504 | { | |
505 | mlog_entry_void(); | |
506 | ||
507 | BUG_ON(!lockres); | |
508 | ||
509 | switch(level) { | |
510 | case LKM_EXMODE: | |
511 | BUG_ON(!lockres->l_ex_holders); | |
512 | lockres->l_ex_holders--; | |
513 | break; | |
514 | case LKM_PRMODE: | |
515 | BUG_ON(!lockres->l_ro_holders); | |
516 | lockres->l_ro_holders--; | |
517 | break; | |
518 | default: | |
519 | BUG(); | |
520 | } | |
521 | mlog_exit_void(); | |
522 | } | |
523 | ||
524 | /* WARNING: This function lives in a world where the only three lock | |
525 | * levels are EX, PR, and NL. It *will* have to be adjusted when more | |
526 | * lock types are added. */ | |
527 | static inline int ocfs2_highest_compat_lock_level(int level) | |
528 | { | |
529 | int new_level = LKM_EXMODE; | |
530 | ||
531 | if (level == LKM_EXMODE) | |
532 | new_level = LKM_NLMODE; | |
533 | else if (level == LKM_PRMODE) | |
534 | new_level = LKM_PRMODE; | |
535 | return new_level; | |
536 | } | |
537 | ||
538 | static void lockres_set_flags(struct ocfs2_lock_res *lockres, | |
539 | unsigned long newflags) | |
540 | { | |
541 | struct list_head *pos, *tmp; | |
542 | struct ocfs2_mask_waiter *mw; | |
543 | ||
544 | assert_spin_locked(&lockres->l_lock); | |
545 | ||
546 | lockres->l_flags = newflags; | |
547 | ||
548 | list_for_each_safe(pos, tmp, &lockres->l_mask_waiters) { | |
549 | mw = list_entry(pos, struct ocfs2_mask_waiter, mw_item); | |
550 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) | |
551 | continue; | |
552 | ||
553 | list_del_init(&mw->mw_item); | |
554 | mw->mw_status = 0; | |
555 | complete(&mw->mw_complete); | |
556 | } | |
557 | } | |
558 | static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or) | |
559 | { | |
560 | lockres_set_flags(lockres, lockres->l_flags | or); | |
561 | } | |
562 | static void lockres_clear_flags(struct ocfs2_lock_res *lockres, | |
563 | unsigned long clear) | |
564 | { | |
565 | lockres_set_flags(lockres, lockres->l_flags & ~clear); | |
566 | } | |
567 | ||
568 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres) | |
569 | { | |
570 | mlog_entry_void(); | |
571 | ||
572 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); | |
573 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); | |
574 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
575 | BUG_ON(lockres->l_blocking <= LKM_NLMODE); | |
576 | ||
577 | lockres->l_level = lockres->l_requested; | |
578 | if (lockres->l_level <= | |
579 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) { | |
580 | lockres->l_blocking = LKM_NLMODE; | |
581 | lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED); | |
582 | } | |
583 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
584 | ||
585 | mlog_exit_void(); | |
586 | } | |
587 | ||
588 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres) | |
589 | { | |
590 | mlog_entry_void(); | |
591 | ||
592 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); | |
593 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); | |
594 | ||
595 | /* Convert from RO to EX doesn't really need anything as our | |
596 | * information is already up to data. Convert from NL to | |
597 | * *anything* however should mark ourselves as needing an | |
598 | * update */ | |
f625c979 MF |
599 | if (lockres->l_level == LKM_NLMODE && |
600 | lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) | |
ccd979bd MF |
601 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); |
602 | ||
603 | lockres->l_level = lockres->l_requested; | |
604 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
605 | ||
606 | mlog_exit_void(); | |
607 | } | |
608 | ||
609 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres) | |
610 | { | |
611 | mlog_entry_void(); | |
612 | ||
613 | BUG_ON((!lockres->l_flags & OCFS2_LOCK_BUSY)); | |
614 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); | |
615 | ||
616 | if (lockres->l_requested > LKM_NLMODE && | |
f625c979 MF |
617 | !(lockres->l_flags & OCFS2_LOCK_LOCAL) && |
618 | lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) | |
ccd979bd MF |
619 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); |
620 | ||
621 | lockres->l_level = lockres->l_requested; | |
622 | lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED); | |
623 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
624 | ||
625 | mlog_exit_void(); | |
626 | } | |
627 | ||
628 | static void ocfs2_inode_ast_func(void *opaque) | |
629 | { | |
630 | struct ocfs2_lock_res *lockres = opaque; | |
631 | struct inode *inode; | |
632 | struct dlm_lockstatus *lksb; | |
633 | unsigned long flags; | |
634 | ||
635 | mlog_entry_void(); | |
636 | ||
637 | inode = ocfs2_lock_res_inode(lockres); | |
638 | ||
b0697053 MF |
639 | mlog(0, "AST fired for inode %llu, l_action = %u, type = %s\n", |
640 | (unsigned long long)OCFS2_I(inode)->ip_blkno, lockres->l_action, | |
ccd979bd MF |
641 | ocfs2_lock_type_string(lockres->l_type)); |
642 | ||
643 | BUG_ON(!ocfs2_is_inode_lock(lockres)); | |
644 | ||
645 | spin_lock_irqsave(&lockres->l_lock, flags); | |
646 | ||
647 | lksb = &(lockres->l_lksb); | |
648 | if (lksb->status != DLM_NORMAL) { | |
649 | mlog(ML_ERROR, "ocfs2_inode_ast_func: lksb status value of %u " | |
b0697053 MF |
650 | "on inode %llu\n", lksb->status, |
651 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
652 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
653 | mlog_exit_void(); | |
654 | return; | |
655 | } | |
656 | ||
657 | switch(lockres->l_action) { | |
658 | case OCFS2_AST_ATTACH: | |
659 | ocfs2_generic_handle_attach_action(lockres); | |
660 | lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL); | |
661 | break; | |
662 | case OCFS2_AST_CONVERT: | |
663 | ocfs2_generic_handle_convert_action(lockres); | |
664 | break; | |
665 | case OCFS2_AST_DOWNCONVERT: | |
666 | ocfs2_generic_handle_downconvert_action(lockres); | |
667 | break; | |
668 | default: | |
669 | mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u " | |
670 | "lockres flags = 0x%lx, unlock action: %u\n", | |
671 | lockres->l_name, lockres->l_action, lockres->l_flags, | |
672 | lockres->l_unlock_action); | |
673 | ||
674 | BUG(); | |
675 | } | |
676 | ||
ccd979bd MF |
677 | /* set it to something invalid so if we get called again we |
678 | * can catch it. */ | |
679 | lockres->l_action = OCFS2_AST_INVALID; | |
680 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
681 | wake_up(&lockres->l_event); | |
682 | ||
683 | mlog_exit_void(); | |
684 | } | |
685 | ||
686 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, | |
687 | int level) | |
688 | { | |
689 | int needs_downconvert = 0; | |
690 | mlog_entry_void(); | |
691 | ||
692 | assert_spin_locked(&lockres->l_lock); | |
693 | ||
694 | lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); | |
695 | ||
696 | if (level > lockres->l_blocking) { | |
697 | /* only schedule a downconvert if we haven't already scheduled | |
698 | * one that goes low enough to satisfy the level we're | |
699 | * blocking. this also catches the case where we get | |
700 | * duplicate BASTs */ | |
701 | if (ocfs2_highest_compat_lock_level(level) < | |
702 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) | |
703 | needs_downconvert = 1; | |
704 | ||
705 | lockres->l_blocking = level; | |
706 | } | |
707 | ||
708 | mlog_exit(needs_downconvert); | |
709 | return needs_downconvert; | |
710 | } | |
711 | ||
712 | static void ocfs2_generic_bast_func(struct ocfs2_super *osb, | |
713 | struct ocfs2_lock_res *lockres, | |
714 | int level) | |
715 | { | |
716 | int needs_downconvert; | |
717 | unsigned long flags; | |
718 | ||
719 | mlog_entry_void(); | |
720 | ||
721 | BUG_ON(level <= LKM_NLMODE); | |
722 | ||
723 | spin_lock_irqsave(&lockres->l_lock, flags); | |
724 | needs_downconvert = ocfs2_generic_handle_bast(lockres, level); | |
725 | if (needs_downconvert) | |
726 | ocfs2_schedule_blocked_lock(osb, lockres); | |
727 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
728 | ||
d680efe9 MF |
729 | wake_up(&lockres->l_event); |
730 | ||
ccd979bd MF |
731 | ocfs2_kick_vote_thread(osb); |
732 | ||
ccd979bd MF |
733 | mlog_exit_void(); |
734 | } | |
735 | ||
736 | static void ocfs2_inode_bast_func(void *opaque, int level) | |
737 | { | |
738 | struct ocfs2_lock_res *lockres = opaque; | |
739 | struct inode *inode; | |
740 | struct ocfs2_super *osb; | |
741 | ||
742 | mlog_entry_void(); | |
743 | ||
744 | BUG_ON(!ocfs2_is_inode_lock(lockres)); | |
745 | ||
746 | inode = ocfs2_lock_res_inode(lockres); | |
747 | osb = OCFS2_SB(inode->i_sb); | |
748 | ||
b0697053 MF |
749 | mlog(0, "BAST fired for inode %llu, blocking %d, level %d type %s\n", |
750 | (unsigned long long)OCFS2_I(inode)->ip_blkno, level, | |
751 | lockres->l_level, ocfs2_lock_type_string(lockres->l_type)); | |
ccd979bd MF |
752 | |
753 | ocfs2_generic_bast_func(osb, lockres, level); | |
754 | ||
755 | mlog_exit_void(); | |
756 | } | |
757 | ||
f625c979 | 758 | static void ocfs2_generic_ast_func(struct ocfs2_lock_res *lockres) |
ccd979bd MF |
759 | { |
760 | struct dlm_lockstatus *lksb = &lockres->l_lksb; | |
761 | unsigned long flags; | |
762 | ||
763 | spin_lock_irqsave(&lockres->l_lock, flags); | |
764 | ||
765 | if (lksb->status != DLM_NORMAL) { | |
766 | mlog(ML_ERROR, "lockres %s: lksb status value of %u!\n", | |
767 | lockres->l_name, lksb->status); | |
768 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
769 | return; | |
770 | } | |
771 | ||
772 | switch(lockres->l_action) { | |
773 | case OCFS2_AST_ATTACH: | |
774 | ocfs2_generic_handle_attach_action(lockres); | |
775 | break; | |
776 | case OCFS2_AST_CONVERT: | |
777 | ocfs2_generic_handle_convert_action(lockres); | |
778 | break; | |
779 | case OCFS2_AST_DOWNCONVERT: | |
780 | ocfs2_generic_handle_downconvert_action(lockres); | |
781 | break; | |
782 | default: | |
783 | BUG(); | |
784 | } | |
785 | ||
ccd979bd MF |
786 | /* set it to something invalid so if we get called again we |
787 | * can catch it. */ | |
788 | lockres->l_action = OCFS2_AST_INVALID; | |
ccd979bd MF |
789 | |
790 | wake_up(&lockres->l_event); | |
d680efe9 | 791 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
ccd979bd MF |
792 | } |
793 | ||
794 | static void ocfs2_super_ast_func(void *opaque) | |
795 | { | |
796 | struct ocfs2_lock_res *lockres = opaque; | |
797 | ||
798 | mlog_entry_void(); | |
799 | mlog(0, "Superblock AST fired\n"); | |
800 | ||
801 | BUG_ON(!ocfs2_is_super_lock(lockres)); | |
f625c979 | 802 | ocfs2_generic_ast_func(lockres); |
ccd979bd MF |
803 | |
804 | mlog_exit_void(); | |
805 | } | |
806 | ||
807 | static void ocfs2_super_bast_func(void *opaque, | |
808 | int level) | |
809 | { | |
810 | struct ocfs2_lock_res *lockres = opaque; | |
811 | struct ocfs2_super *osb; | |
812 | ||
813 | mlog_entry_void(); | |
814 | mlog(0, "Superblock BAST fired\n"); | |
815 | ||
816 | BUG_ON(!ocfs2_is_super_lock(lockres)); | |
817 | osb = ocfs2_lock_res_super(lockres); | |
818 | ocfs2_generic_bast_func(osb, lockres, level); | |
819 | ||
820 | mlog_exit_void(); | |
821 | } | |
822 | ||
823 | static void ocfs2_rename_ast_func(void *opaque) | |
824 | { | |
825 | struct ocfs2_lock_res *lockres = opaque; | |
826 | ||
827 | mlog_entry_void(); | |
828 | ||
829 | mlog(0, "Rename AST fired\n"); | |
830 | ||
831 | BUG_ON(!ocfs2_is_rename_lock(lockres)); | |
832 | ||
f625c979 | 833 | ocfs2_generic_ast_func(lockres); |
ccd979bd MF |
834 | |
835 | mlog_exit_void(); | |
836 | } | |
837 | ||
838 | static void ocfs2_rename_bast_func(void *opaque, | |
839 | int level) | |
840 | { | |
841 | struct ocfs2_lock_res *lockres = opaque; | |
842 | struct ocfs2_super *osb; | |
843 | ||
844 | mlog_entry_void(); | |
845 | ||
846 | mlog(0, "Rename BAST fired\n"); | |
847 | ||
848 | BUG_ON(!ocfs2_is_rename_lock(lockres)); | |
849 | ||
850 | osb = ocfs2_lock_res_super(lockres); | |
851 | ocfs2_generic_bast_func(osb, lockres, level); | |
852 | ||
853 | mlog_exit_void(); | |
854 | } | |
855 | ||
d680efe9 MF |
856 | static void ocfs2_dentry_ast_func(void *opaque) |
857 | { | |
858 | struct ocfs2_lock_res *lockres = opaque; | |
859 | ||
860 | BUG_ON(!lockres); | |
861 | ||
f625c979 | 862 | ocfs2_generic_ast_func(lockres); |
d680efe9 MF |
863 | } |
864 | ||
865 | static void ocfs2_dentry_bast_func(void *opaque, int level) | |
866 | { | |
867 | struct ocfs2_lock_res *lockres = opaque; | |
868 | struct ocfs2_dentry_lock *dl = lockres->l_priv; | |
869 | struct ocfs2_super *osb = OCFS2_SB(dl->dl_inode->i_sb); | |
870 | ||
871 | mlog(0, "Dentry bast: level: %d, name: %s\n", level, | |
872 | lockres->l_name); | |
873 | ||
874 | ocfs2_generic_bast_func(osb, lockres, level); | |
875 | } | |
876 | ||
ccd979bd MF |
877 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, |
878 | int convert) | |
879 | { | |
880 | unsigned long flags; | |
881 | ||
882 | mlog_entry_void(); | |
883 | spin_lock_irqsave(&lockres->l_lock, flags); | |
884 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
885 | if (convert) | |
886 | lockres->l_action = OCFS2_AST_INVALID; | |
887 | else | |
888 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
889 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
890 | ||
891 | wake_up(&lockres->l_event); | |
892 | mlog_exit_void(); | |
893 | } | |
894 | ||
895 | /* Note: If we detect another process working on the lock (i.e., | |
896 | * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller | |
897 | * to do the right thing in that case. | |
898 | */ | |
899 | static int ocfs2_lock_create(struct ocfs2_super *osb, | |
900 | struct ocfs2_lock_res *lockres, | |
901 | int level, | |
902 | int dlm_flags) | |
903 | { | |
904 | int ret = 0; | |
905 | enum dlm_status status; | |
906 | unsigned long flags; | |
907 | ||
908 | mlog_entry_void(); | |
909 | ||
910 | mlog(0, "lock %s, level = %d, flags = %d\n", lockres->l_name, level, | |
911 | dlm_flags); | |
912 | ||
913 | spin_lock_irqsave(&lockres->l_lock, flags); | |
914 | if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) || | |
915 | (lockres->l_flags & OCFS2_LOCK_BUSY)) { | |
916 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
917 | goto bail; | |
918 | } | |
919 | ||
920 | lockres->l_action = OCFS2_AST_ATTACH; | |
921 | lockres->l_requested = level; | |
922 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
923 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
924 | ||
925 | status = dlmlock(osb->dlm, | |
926 | level, | |
927 | &lockres->l_lksb, | |
928 | dlm_flags, | |
929 | lockres->l_name, | |
f0681062 | 930 | OCFS2_LOCK_ID_MAX_LEN - 1, |
ccd979bd MF |
931 | lockres->l_ops->ast, |
932 | lockres, | |
933 | lockres->l_ops->bast); | |
934 | if (status != DLM_NORMAL) { | |
935 | ocfs2_log_dlm_error("dlmlock", status, lockres); | |
936 | ret = -EINVAL; | |
937 | ocfs2_recover_from_dlm_error(lockres, 1); | |
938 | } | |
939 | ||
940 | mlog(0, "lock %s, successfull return from dlmlock\n", lockres->l_name); | |
941 | ||
942 | bail: | |
943 | mlog_exit(ret); | |
944 | return ret; | |
945 | } | |
946 | ||
947 | static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres, | |
948 | int flag) | |
949 | { | |
950 | unsigned long flags; | |
951 | int ret; | |
952 | ||
953 | spin_lock_irqsave(&lockres->l_lock, flags); | |
954 | ret = lockres->l_flags & flag; | |
955 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
956 | ||
957 | return ret; | |
958 | } | |
959 | ||
960 | static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres) | |
961 | ||
962 | { | |
963 | wait_event(lockres->l_event, | |
964 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY)); | |
965 | } | |
966 | ||
967 | static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres) | |
968 | ||
969 | { | |
970 | wait_event(lockres->l_event, | |
971 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING)); | |
972 | } | |
973 | ||
974 | /* predict what lock level we'll be dropping down to on behalf | |
975 | * of another node, and return true if the currently wanted | |
976 | * level will be compatible with it. */ | |
977 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, | |
978 | int wanted) | |
979 | { | |
980 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
981 | ||
982 | return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking); | |
983 | } | |
984 | ||
985 | static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw) | |
986 | { | |
987 | INIT_LIST_HEAD(&mw->mw_item); | |
988 | init_completion(&mw->mw_complete); | |
989 | } | |
990 | ||
991 | static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw) | |
992 | { | |
993 | wait_for_completion(&mw->mw_complete); | |
994 | /* Re-arm the completion in case we want to wait on it again */ | |
995 | INIT_COMPLETION(mw->mw_complete); | |
996 | return mw->mw_status; | |
997 | } | |
998 | ||
999 | static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres, | |
1000 | struct ocfs2_mask_waiter *mw, | |
1001 | unsigned long mask, | |
1002 | unsigned long goal) | |
1003 | { | |
1004 | BUG_ON(!list_empty(&mw->mw_item)); | |
1005 | ||
1006 | assert_spin_locked(&lockres->l_lock); | |
1007 | ||
1008 | list_add_tail(&mw->mw_item, &lockres->l_mask_waiters); | |
1009 | mw->mw_mask = mask; | |
1010 | mw->mw_goal = goal; | |
1011 | } | |
1012 | ||
1013 | /* returns 0 if the mw that was removed was already satisfied, -EBUSY | |
1014 | * if the mask still hadn't reached its goal */ | |
1015 | static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres, | |
1016 | struct ocfs2_mask_waiter *mw) | |
1017 | { | |
1018 | unsigned long flags; | |
1019 | int ret = 0; | |
1020 | ||
1021 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1022 | if (!list_empty(&mw->mw_item)) { | |
1023 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) | |
1024 | ret = -EBUSY; | |
1025 | ||
1026 | list_del_init(&mw->mw_item); | |
1027 | init_completion(&mw->mw_complete); | |
1028 | } | |
1029 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1030 | ||
1031 | return ret; | |
1032 | ||
1033 | } | |
1034 | ||
1035 | static int ocfs2_cluster_lock(struct ocfs2_super *osb, | |
1036 | struct ocfs2_lock_res *lockres, | |
1037 | int level, | |
1038 | int lkm_flags, | |
1039 | int arg_flags) | |
1040 | { | |
1041 | struct ocfs2_mask_waiter mw; | |
1042 | enum dlm_status status; | |
1043 | int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR); | |
1044 | int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */ | |
1045 | unsigned long flags; | |
1046 | ||
1047 | mlog_entry_void(); | |
1048 | ||
1049 | ocfs2_init_mask_waiter(&mw); | |
1050 | ||
1051 | again: | |
1052 | wait = 0; | |
1053 | ||
1054 | if (catch_signals && signal_pending(current)) { | |
1055 | ret = -ERESTARTSYS; | |
1056 | goto out; | |
1057 | } | |
1058 | ||
1059 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1060 | ||
1061 | mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING, | |
1062 | "Cluster lock called on freeing lockres %s! flags " | |
1063 | "0x%lx\n", lockres->l_name, lockres->l_flags); | |
1064 | ||
1065 | /* We only compare against the currently granted level | |
1066 | * here. If the lock is blocked waiting on a downconvert, | |
1067 | * we'll get caught below. */ | |
1068 | if (lockres->l_flags & OCFS2_LOCK_BUSY && | |
1069 | level > lockres->l_level) { | |
1070 | /* is someone sitting in dlm_lock? If so, wait on | |
1071 | * them. */ | |
1072 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); | |
1073 | wait = 1; | |
1074 | goto unlock; | |
1075 | } | |
1076 | ||
1077 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { | |
1078 | /* lock has not been created yet. */ | |
1079 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1080 | ||
1081 | ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0); | |
1082 | if (ret < 0) { | |
1083 | mlog_errno(ret); | |
1084 | goto out; | |
1085 | } | |
1086 | goto again; | |
1087 | } | |
1088 | ||
1089 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED && | |
1090 | !ocfs2_may_continue_on_blocked_lock(lockres, level)) { | |
1091 | /* is the lock is currently blocked on behalf of | |
1092 | * another node */ | |
1093 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0); | |
1094 | wait = 1; | |
1095 | goto unlock; | |
1096 | } | |
1097 | ||
1098 | if (level > lockres->l_level) { | |
1099 | if (lockres->l_action != OCFS2_AST_INVALID) | |
1100 | mlog(ML_ERROR, "lockres %s has action %u pending\n", | |
1101 | lockres->l_name, lockres->l_action); | |
1102 | ||
1103 | lockres->l_action = OCFS2_AST_CONVERT; | |
1104 | lockres->l_requested = level; | |
1105 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
1106 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1107 | ||
1108 | BUG_ON(level == LKM_IVMODE); | |
1109 | BUG_ON(level == LKM_NLMODE); | |
1110 | ||
1111 | mlog(0, "lock %s, convert from %d to level = %d\n", | |
1112 | lockres->l_name, lockres->l_level, level); | |
1113 | ||
1114 | /* call dlm_lock to upgrade lock now */ | |
1115 | status = dlmlock(osb->dlm, | |
1116 | level, | |
1117 | &lockres->l_lksb, | |
1118 | lkm_flags|LKM_CONVERT|LKM_VALBLK, | |
1119 | lockres->l_name, | |
f0681062 | 1120 | OCFS2_LOCK_ID_MAX_LEN - 1, |
ccd979bd MF |
1121 | lockres->l_ops->ast, |
1122 | lockres, | |
1123 | lockres->l_ops->bast); | |
1124 | if (status != DLM_NORMAL) { | |
1125 | if ((lkm_flags & LKM_NOQUEUE) && | |
1126 | (status == DLM_NOTQUEUED)) | |
1127 | ret = -EAGAIN; | |
1128 | else { | |
1129 | ocfs2_log_dlm_error("dlmlock", status, | |
1130 | lockres); | |
1131 | ret = -EINVAL; | |
1132 | } | |
1133 | ocfs2_recover_from_dlm_error(lockres, 1); | |
1134 | goto out; | |
1135 | } | |
1136 | ||
1137 | mlog(0, "lock %s, successfull return from dlmlock\n", | |
1138 | lockres->l_name); | |
1139 | ||
1140 | /* At this point we've gone inside the dlm and need to | |
1141 | * complete our work regardless. */ | |
1142 | catch_signals = 0; | |
1143 | ||
1144 | /* wait for busy to clear and carry on */ | |
1145 | goto again; | |
1146 | } | |
1147 | ||
1148 | /* Ok, if we get here then we're good to go. */ | |
1149 | ocfs2_inc_holders(lockres, level); | |
1150 | ||
1151 | ret = 0; | |
1152 | unlock: | |
1153 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1154 | out: | |
1155 | /* | |
1156 | * This is helping work around a lock inversion between the page lock | |
1157 | * and dlm locks. One path holds the page lock while calling aops | |
1158 | * which block acquiring dlm locks. The voting thread holds dlm | |
1159 | * locks while acquiring page locks while down converting data locks. | |
1160 | * This block is helping an aop path notice the inversion and back | |
1161 | * off to unlock its page lock before trying the dlm lock again. | |
1162 | */ | |
1163 | if (wait && arg_flags & OCFS2_LOCK_NONBLOCK && | |
1164 | mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) { | |
1165 | wait = 0; | |
1166 | if (lockres_remove_mask_waiter(lockres, &mw)) | |
1167 | ret = -EAGAIN; | |
1168 | else | |
1169 | goto again; | |
1170 | } | |
1171 | if (wait) { | |
1172 | ret = ocfs2_wait_for_mask(&mw); | |
1173 | if (ret == 0) | |
1174 | goto again; | |
1175 | mlog_errno(ret); | |
1176 | } | |
1177 | ||
1178 | mlog_exit(ret); | |
1179 | return ret; | |
1180 | } | |
1181 | ||
1182 | static void ocfs2_cluster_unlock(struct ocfs2_super *osb, | |
1183 | struct ocfs2_lock_res *lockres, | |
1184 | int level) | |
1185 | { | |
1186 | unsigned long flags; | |
1187 | ||
1188 | mlog_entry_void(); | |
1189 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1190 | ocfs2_dec_holders(lockres, level); | |
1191 | ocfs2_vote_on_unlock(osb, lockres); | |
1192 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1193 | mlog_exit_void(); | |
1194 | } | |
1195 | ||
d680efe9 MF |
1196 | int ocfs2_create_new_lock(struct ocfs2_super *osb, |
1197 | struct ocfs2_lock_res *lockres, | |
24c19ef4 MF |
1198 | int ex, |
1199 | int local) | |
ccd979bd | 1200 | { |
d680efe9 | 1201 | int level = ex ? LKM_EXMODE : LKM_PRMODE; |
ccd979bd | 1202 | unsigned long flags; |
24c19ef4 | 1203 | int lkm_flags = local ? LKM_LOCAL : 0; |
ccd979bd MF |
1204 | |
1205 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1206 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); | |
1207 | lockres_or_flags(lockres, OCFS2_LOCK_LOCAL); | |
1208 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1209 | ||
24c19ef4 | 1210 | return ocfs2_lock_create(osb, lockres, level, lkm_flags); |
ccd979bd MF |
1211 | } |
1212 | ||
1213 | /* Grants us an EX lock on the data and metadata resources, skipping | |
1214 | * the normal cluster directory lookup. Use this ONLY on newly created | |
1215 | * inodes which other nodes can't possibly see, and which haven't been | |
1216 | * hashed in the inode hash yet. This can give us a good performance | |
1217 | * increase as it'll skip the network broadcast normally associated | |
1218 | * with creating a new lock resource. */ | |
1219 | int ocfs2_create_new_inode_locks(struct inode *inode) | |
1220 | { | |
1221 | int ret; | |
d680efe9 | 1222 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd MF |
1223 | |
1224 | BUG_ON(!inode); | |
1225 | BUG_ON(!ocfs2_inode_is_new(inode)); | |
1226 | ||
1227 | mlog_entry_void(); | |
1228 | ||
b0697053 | 1229 | mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno); |
ccd979bd MF |
1230 | |
1231 | /* NOTE: That we don't increment any of the holder counts, nor | |
1232 | * do we add anything to a journal handle. Since this is | |
1233 | * supposed to be a new inode which the cluster doesn't know | |
1234 | * about yet, there is no need to. As far as the LVB handling | |
1235 | * is concerned, this is basically like acquiring an EX lock | |
1236 | * on a resource which has an invalid one -- we'll set it | |
1237 | * valid when we release the EX. */ | |
1238 | ||
24c19ef4 | 1239 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1); |
ccd979bd MF |
1240 | if (ret) { |
1241 | mlog_errno(ret); | |
1242 | goto bail; | |
1243 | } | |
1244 | ||
24c19ef4 MF |
1245 | /* |
1246 | * We don't want to use LKM_LOCAL on a meta data lock as they | |
1247 | * don't use a generation in their lock names. | |
1248 | */ | |
1249 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_meta_lockres, 1, 0); | |
ccd979bd MF |
1250 | if (ret) { |
1251 | mlog_errno(ret); | |
1252 | goto bail; | |
1253 | } | |
1254 | ||
24c19ef4 | 1255 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_data_lockres, 1, 1); |
ccd979bd MF |
1256 | if (ret) { |
1257 | mlog_errno(ret); | |
1258 | goto bail; | |
1259 | } | |
1260 | ||
1261 | bail: | |
1262 | mlog_exit(ret); | |
1263 | return ret; | |
1264 | } | |
1265 | ||
1266 | int ocfs2_rw_lock(struct inode *inode, int write) | |
1267 | { | |
1268 | int status, level; | |
1269 | struct ocfs2_lock_res *lockres; | |
1270 | ||
1271 | BUG_ON(!inode); | |
1272 | ||
1273 | mlog_entry_void(); | |
1274 | ||
b0697053 MF |
1275 | mlog(0, "inode %llu take %s RW lock\n", |
1276 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1277 | write ? "EXMODE" : "PRMODE"); |
1278 | ||
1279 | lockres = &OCFS2_I(inode)->ip_rw_lockres; | |
1280 | ||
1281 | level = write ? LKM_EXMODE : LKM_PRMODE; | |
1282 | ||
1283 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0, | |
1284 | 0); | |
1285 | if (status < 0) | |
1286 | mlog_errno(status); | |
1287 | ||
1288 | mlog_exit(status); | |
1289 | return status; | |
1290 | } | |
1291 | ||
1292 | void ocfs2_rw_unlock(struct inode *inode, int write) | |
1293 | { | |
1294 | int level = write ? LKM_EXMODE : LKM_PRMODE; | |
1295 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres; | |
1296 | ||
1297 | mlog_entry_void(); | |
1298 | ||
b0697053 MF |
1299 | mlog(0, "inode %llu drop %s RW lock\n", |
1300 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1301 | write ? "EXMODE" : "PRMODE"); |
1302 | ||
1303 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
1304 | ||
1305 | mlog_exit_void(); | |
1306 | } | |
1307 | ||
1308 | int ocfs2_data_lock_full(struct inode *inode, | |
1309 | int write, | |
1310 | int arg_flags) | |
1311 | { | |
1312 | int status = 0, level; | |
1313 | struct ocfs2_lock_res *lockres; | |
1314 | ||
1315 | BUG_ON(!inode); | |
1316 | ||
1317 | mlog_entry_void(); | |
1318 | ||
b0697053 MF |
1319 | mlog(0, "inode %llu take %s DATA lock\n", |
1320 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1321 | write ? "EXMODE" : "PRMODE"); |
1322 | ||
1323 | /* We'll allow faking a readonly data lock for | |
1324 | * rodevices. */ | |
1325 | if (ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) { | |
1326 | if (write) { | |
1327 | status = -EROFS; | |
1328 | mlog_errno(status); | |
1329 | } | |
1330 | goto out; | |
1331 | } | |
1332 | ||
1333 | lockres = &OCFS2_I(inode)->ip_data_lockres; | |
1334 | ||
1335 | level = write ? LKM_EXMODE : LKM_PRMODE; | |
1336 | ||
1337 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, | |
1338 | 0, arg_flags); | |
1339 | if (status < 0 && status != -EAGAIN) | |
1340 | mlog_errno(status); | |
1341 | ||
1342 | out: | |
1343 | mlog_exit(status); | |
1344 | return status; | |
1345 | } | |
1346 | ||
1347 | /* see ocfs2_meta_lock_with_page() */ | |
1348 | int ocfs2_data_lock_with_page(struct inode *inode, | |
1349 | int write, | |
1350 | struct page *page) | |
1351 | { | |
1352 | int ret; | |
1353 | ||
1354 | ret = ocfs2_data_lock_full(inode, write, OCFS2_LOCK_NONBLOCK); | |
1355 | if (ret == -EAGAIN) { | |
1356 | unlock_page(page); | |
1357 | if (ocfs2_data_lock(inode, write) == 0) | |
1358 | ocfs2_data_unlock(inode, write); | |
1359 | ret = AOP_TRUNCATED_PAGE; | |
1360 | } | |
1361 | ||
1362 | return ret; | |
1363 | } | |
1364 | ||
1365 | static void ocfs2_vote_on_unlock(struct ocfs2_super *osb, | |
1366 | struct ocfs2_lock_res *lockres) | |
1367 | { | |
1368 | int kick = 0; | |
1369 | ||
1370 | mlog_entry_void(); | |
1371 | ||
1372 | /* If we know that another node is waiting on our lock, kick | |
1373 | * the vote thread * pre-emptively when we reach a release | |
1374 | * condition. */ | |
1375 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) { | |
1376 | switch(lockres->l_blocking) { | |
1377 | case LKM_EXMODE: | |
1378 | if (!lockres->l_ex_holders && !lockres->l_ro_holders) | |
1379 | kick = 1; | |
1380 | break; | |
1381 | case LKM_PRMODE: | |
1382 | if (!lockres->l_ex_holders) | |
1383 | kick = 1; | |
1384 | break; | |
1385 | default: | |
1386 | BUG(); | |
1387 | } | |
1388 | } | |
1389 | ||
1390 | if (kick) | |
1391 | ocfs2_kick_vote_thread(osb); | |
1392 | ||
1393 | mlog_exit_void(); | |
1394 | } | |
1395 | ||
1396 | void ocfs2_data_unlock(struct inode *inode, | |
1397 | int write) | |
1398 | { | |
1399 | int level = write ? LKM_EXMODE : LKM_PRMODE; | |
1400 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_data_lockres; | |
1401 | ||
1402 | mlog_entry_void(); | |
1403 | ||
b0697053 MF |
1404 | mlog(0, "inode %llu drop %s DATA lock\n", |
1405 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1406 | write ? "EXMODE" : "PRMODE"); |
1407 | ||
1408 | if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) | |
1409 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
1410 | ||
1411 | mlog_exit_void(); | |
1412 | } | |
1413 | ||
1414 | #define OCFS2_SEC_BITS 34 | |
1415 | #define OCFS2_SEC_SHIFT (64 - 34) | |
1416 | #define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1) | |
1417 | ||
1418 | /* LVB only has room for 64 bits of time here so we pack it for | |
1419 | * now. */ | |
1420 | static u64 ocfs2_pack_timespec(struct timespec *spec) | |
1421 | { | |
1422 | u64 res; | |
1423 | u64 sec = spec->tv_sec; | |
1424 | u32 nsec = spec->tv_nsec; | |
1425 | ||
1426 | res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK); | |
1427 | ||
1428 | return res; | |
1429 | } | |
1430 | ||
1431 | /* Call this with the lockres locked. I am reasonably sure we don't | |
1432 | * need ip_lock in this function as anyone who would be changing those | |
1433 | * values is supposed to be blocked in ocfs2_meta_lock right now. */ | |
1434 | static void __ocfs2_stuff_meta_lvb(struct inode *inode) | |
1435 | { | |
1436 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1437 | struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres; | |
1438 | struct ocfs2_meta_lvb *lvb; | |
1439 | ||
1440 | mlog_entry_void(); | |
1441 | ||
1442 | lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
1443 | ||
24c19ef4 MF |
1444 | /* |
1445 | * Invalidate the LVB of a deleted inode - this way other | |
1446 | * nodes are forced to go to disk and discover the new inode | |
1447 | * status. | |
1448 | */ | |
1449 | if (oi->ip_flags & OCFS2_INODE_DELETED) { | |
1450 | lvb->lvb_version = 0; | |
1451 | goto out; | |
1452 | } | |
1453 | ||
4d3b83f7 | 1454 | lvb->lvb_version = OCFS2_LVB_VERSION; |
ccd979bd MF |
1455 | lvb->lvb_isize = cpu_to_be64(i_size_read(inode)); |
1456 | lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters); | |
1457 | lvb->lvb_iuid = cpu_to_be32(inode->i_uid); | |
1458 | lvb->lvb_igid = cpu_to_be32(inode->i_gid); | |
1459 | lvb->lvb_imode = cpu_to_be16(inode->i_mode); | |
1460 | lvb->lvb_inlink = cpu_to_be16(inode->i_nlink); | |
1461 | lvb->lvb_iatime_packed = | |
1462 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime)); | |
1463 | lvb->lvb_ictime_packed = | |
1464 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime)); | |
1465 | lvb->lvb_imtime_packed = | |
1466 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime)); | |
ca4d147e | 1467 | lvb->lvb_iattr = cpu_to_be32(oi->ip_attr); |
f9e2d82e | 1468 | lvb->lvb_igeneration = cpu_to_be32(inode->i_generation); |
ccd979bd | 1469 | |
24c19ef4 | 1470 | out: |
ccd979bd MF |
1471 | mlog_meta_lvb(0, lockres); |
1472 | ||
1473 | mlog_exit_void(); | |
1474 | } | |
1475 | ||
1476 | static void ocfs2_unpack_timespec(struct timespec *spec, | |
1477 | u64 packed_time) | |
1478 | { | |
1479 | spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT; | |
1480 | spec->tv_nsec = packed_time & OCFS2_NSEC_MASK; | |
1481 | } | |
1482 | ||
1483 | static void ocfs2_refresh_inode_from_lvb(struct inode *inode) | |
1484 | { | |
1485 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1486 | struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres; | |
1487 | struct ocfs2_meta_lvb *lvb; | |
1488 | ||
1489 | mlog_entry_void(); | |
1490 | ||
1491 | mlog_meta_lvb(0, lockres); | |
1492 | ||
1493 | lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
1494 | ||
1495 | /* We're safe here without the lockres lock... */ | |
1496 | spin_lock(&oi->ip_lock); | |
1497 | oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters); | |
1498 | i_size_write(inode, be64_to_cpu(lvb->lvb_isize)); | |
1499 | ||
ca4d147e HP |
1500 | oi->ip_attr = be32_to_cpu(lvb->lvb_iattr); |
1501 | ocfs2_set_inode_flags(inode); | |
1502 | ||
ccd979bd MF |
1503 | /* fast-symlinks are a special case */ |
1504 | if (S_ISLNK(inode->i_mode) && !oi->ip_clusters) | |
1505 | inode->i_blocks = 0; | |
1506 | else | |
1507 | inode->i_blocks = | |
1508 | ocfs2_align_bytes_to_sectors(i_size_read(inode)); | |
1509 | ||
1510 | inode->i_uid = be32_to_cpu(lvb->lvb_iuid); | |
1511 | inode->i_gid = be32_to_cpu(lvb->lvb_igid); | |
1512 | inode->i_mode = be16_to_cpu(lvb->lvb_imode); | |
1513 | inode->i_nlink = be16_to_cpu(lvb->lvb_inlink); | |
1514 | ocfs2_unpack_timespec(&inode->i_atime, | |
1515 | be64_to_cpu(lvb->lvb_iatime_packed)); | |
1516 | ocfs2_unpack_timespec(&inode->i_mtime, | |
1517 | be64_to_cpu(lvb->lvb_imtime_packed)); | |
1518 | ocfs2_unpack_timespec(&inode->i_ctime, | |
1519 | be64_to_cpu(lvb->lvb_ictime_packed)); | |
1520 | spin_unlock(&oi->ip_lock); | |
1521 | ||
1522 | mlog_exit_void(); | |
1523 | } | |
1524 | ||
f9e2d82e MF |
1525 | static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode, |
1526 | struct ocfs2_lock_res *lockres) | |
ccd979bd MF |
1527 | { |
1528 | struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
1529 | ||
f9e2d82e MF |
1530 | if (lvb->lvb_version == OCFS2_LVB_VERSION |
1531 | && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation) | |
ccd979bd MF |
1532 | return 1; |
1533 | return 0; | |
1534 | } | |
1535 | ||
1536 | /* Determine whether a lock resource needs to be refreshed, and | |
1537 | * arbitrate who gets to refresh it. | |
1538 | * | |
1539 | * 0 means no refresh needed. | |
1540 | * | |
1541 | * > 0 means you need to refresh this and you MUST call | |
1542 | * ocfs2_complete_lock_res_refresh afterwards. */ | |
1543 | static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres) | |
1544 | { | |
1545 | unsigned long flags; | |
1546 | int status = 0; | |
1547 | ||
1548 | mlog_entry_void(); | |
1549 | ||
1550 | refresh_check: | |
1551 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1552 | if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) { | |
1553 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1554 | goto bail; | |
1555 | } | |
1556 | ||
1557 | if (lockres->l_flags & OCFS2_LOCK_REFRESHING) { | |
1558 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1559 | ||
1560 | ocfs2_wait_on_refreshing_lock(lockres); | |
1561 | goto refresh_check; | |
1562 | } | |
1563 | ||
1564 | /* Ok, I'll be the one to refresh this lock. */ | |
1565 | lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING); | |
1566 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1567 | ||
1568 | status = 1; | |
1569 | bail: | |
1570 | mlog_exit(status); | |
1571 | return status; | |
1572 | } | |
1573 | ||
1574 | /* If status is non zero, I'll mark it as not being in refresh | |
1575 | * anymroe, but i won't clear the needs refresh flag. */ | |
1576 | static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres, | |
1577 | int status) | |
1578 | { | |
1579 | unsigned long flags; | |
1580 | mlog_entry_void(); | |
1581 | ||
1582 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1583 | lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING); | |
1584 | if (!status) | |
1585 | lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); | |
1586 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1587 | ||
1588 | wake_up(&lockres->l_event); | |
1589 | ||
1590 | mlog_exit_void(); | |
1591 | } | |
1592 | ||
1593 | /* may or may not return a bh if it went to disk. */ | |
1594 | static int ocfs2_meta_lock_update(struct inode *inode, | |
1595 | struct buffer_head **bh) | |
1596 | { | |
1597 | int status = 0; | |
1598 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1599 | struct ocfs2_lock_res *lockres; | |
1600 | struct ocfs2_dinode *fe; | |
1601 | ||
1602 | mlog_entry_void(); | |
1603 | ||
1604 | spin_lock(&oi->ip_lock); | |
1605 | if (oi->ip_flags & OCFS2_INODE_DELETED) { | |
b0697053 | 1606 | mlog(0, "Orphaned inode %llu was deleted while we " |
ccd979bd | 1607 | "were waiting on a lock. ip_flags = 0x%x\n", |
b0697053 | 1608 | (unsigned long long)oi->ip_blkno, oi->ip_flags); |
ccd979bd MF |
1609 | spin_unlock(&oi->ip_lock); |
1610 | status = -ENOENT; | |
1611 | goto bail; | |
1612 | } | |
1613 | spin_unlock(&oi->ip_lock); | |
1614 | ||
1615 | lockres = &oi->ip_meta_lockres; | |
1616 | ||
1617 | if (!ocfs2_should_refresh_lock_res(lockres)) | |
1618 | goto bail; | |
1619 | ||
1620 | /* This will discard any caching information we might have had | |
1621 | * for the inode metadata. */ | |
1622 | ocfs2_metadata_cache_purge(inode); | |
1623 | ||
1624 | /* will do nothing for inode types that don't use the extent | |
1625 | * map (directories, bitmap files, etc) */ | |
1626 | ocfs2_extent_map_trunc(inode, 0); | |
1627 | ||
f9e2d82e | 1628 | if (ocfs2_meta_lvb_is_trustable(inode, lockres)) { |
b0697053 MF |
1629 | mlog(0, "Trusting LVB on inode %llu\n", |
1630 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1631 | ocfs2_refresh_inode_from_lvb(inode); |
1632 | } else { | |
1633 | /* Boo, we have to go to disk. */ | |
1634 | /* read bh, cast, ocfs2_refresh_inode */ | |
1635 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno, | |
1636 | bh, OCFS2_BH_CACHED, inode); | |
1637 | if (status < 0) { | |
1638 | mlog_errno(status); | |
1639 | goto bail_refresh; | |
1640 | } | |
1641 | fe = (struct ocfs2_dinode *) (*bh)->b_data; | |
1642 | ||
1643 | /* This is a good chance to make sure we're not | |
1644 | * locking an invalid object. | |
1645 | * | |
1646 | * We bug on a stale inode here because we checked | |
1647 | * above whether it was wiped from disk. The wiping | |
1648 | * node provides a guarantee that we receive that | |
1649 | * message and can mark the inode before dropping any | |
1650 | * locks associated with it. */ | |
1651 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
1652 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
1653 | status = -EIO; | |
1654 | goto bail_refresh; | |
1655 | } | |
1656 | mlog_bug_on_msg(inode->i_generation != | |
1657 | le32_to_cpu(fe->i_generation), | |
b0697053 | 1658 | "Invalid dinode %llu disk generation: %u " |
ccd979bd | 1659 | "inode->i_generation: %u\n", |
b0697053 MF |
1660 | (unsigned long long)oi->ip_blkno, |
1661 | le32_to_cpu(fe->i_generation), | |
ccd979bd MF |
1662 | inode->i_generation); |
1663 | mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) || | |
1664 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)), | |
b0697053 MF |
1665 | "Stale dinode %llu dtime: %llu flags: 0x%x\n", |
1666 | (unsigned long long)oi->ip_blkno, | |
1667 | (unsigned long long)le64_to_cpu(fe->i_dtime), | |
ccd979bd MF |
1668 | le32_to_cpu(fe->i_flags)); |
1669 | ||
1670 | ocfs2_refresh_inode(inode, fe); | |
1671 | } | |
1672 | ||
1673 | status = 0; | |
1674 | bail_refresh: | |
1675 | ocfs2_complete_lock_res_refresh(lockres, status); | |
1676 | bail: | |
1677 | mlog_exit(status); | |
1678 | return status; | |
1679 | } | |
1680 | ||
1681 | static int ocfs2_assign_bh(struct inode *inode, | |
1682 | struct buffer_head **ret_bh, | |
1683 | struct buffer_head *passed_bh) | |
1684 | { | |
1685 | int status; | |
1686 | ||
1687 | if (passed_bh) { | |
1688 | /* Ok, the update went to disk for us, use the | |
1689 | * returned bh. */ | |
1690 | *ret_bh = passed_bh; | |
1691 | get_bh(*ret_bh); | |
1692 | ||
1693 | return 0; | |
1694 | } | |
1695 | ||
1696 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
1697 | OCFS2_I(inode)->ip_blkno, | |
1698 | ret_bh, | |
1699 | OCFS2_BH_CACHED, | |
1700 | inode); | |
1701 | if (status < 0) | |
1702 | mlog_errno(status); | |
1703 | ||
1704 | return status; | |
1705 | } | |
1706 | ||
1707 | /* | |
1708 | * returns < 0 error if the callback will never be called, otherwise | |
1709 | * the result of the lock will be communicated via the callback. | |
1710 | */ | |
1711 | int ocfs2_meta_lock_full(struct inode *inode, | |
1712 | struct ocfs2_journal_handle *handle, | |
1713 | struct buffer_head **ret_bh, | |
1714 | int ex, | |
1715 | int arg_flags) | |
1716 | { | |
1717 | int status, level, dlm_flags, acquired; | |
1718 | struct ocfs2_lock_res *lockres; | |
1719 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1720 | struct buffer_head *local_bh = NULL; | |
1721 | ||
1722 | BUG_ON(!inode); | |
1723 | ||
1724 | mlog_entry_void(); | |
1725 | ||
b0697053 MF |
1726 | mlog(0, "inode %llu, take %s META lock\n", |
1727 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1728 | ex ? "EXMODE" : "PRMODE"); |
1729 | ||
1730 | status = 0; | |
1731 | acquired = 0; | |
1732 | /* We'll allow faking a readonly metadata lock for | |
1733 | * rodevices. */ | |
1734 | if (ocfs2_is_hard_readonly(osb)) { | |
1735 | if (ex) | |
1736 | status = -EROFS; | |
1737 | goto bail; | |
1738 | } | |
1739 | ||
1740 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) | |
1741 | wait_event(osb->recovery_event, | |
1742 | ocfs2_node_map_is_empty(osb, &osb->recovery_map)); | |
1743 | ||
1744 | acquired = 0; | |
1745 | lockres = &OCFS2_I(inode)->ip_meta_lockres; | |
1746 | level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1747 | dlm_flags = 0; | |
1748 | if (arg_flags & OCFS2_META_LOCK_NOQUEUE) | |
1749 | dlm_flags |= LKM_NOQUEUE; | |
1750 | ||
1751 | status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags); | |
1752 | if (status < 0) { | |
1753 | if (status != -EAGAIN && status != -EIOCBRETRY) | |
1754 | mlog_errno(status); | |
1755 | goto bail; | |
1756 | } | |
1757 | ||
1758 | /* Notify the error cleanup path to drop the cluster lock. */ | |
1759 | acquired = 1; | |
1760 | ||
1761 | /* We wait twice because a node may have died while we were in | |
1762 | * the lower dlm layers. The second time though, we've | |
1763 | * committed to owning this lock so we don't allow signals to | |
1764 | * abort the operation. */ | |
1765 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) | |
1766 | wait_event(osb->recovery_event, | |
1767 | ocfs2_node_map_is_empty(osb, &osb->recovery_map)); | |
1768 | ||
24c19ef4 MF |
1769 | /* |
1770 | * We only see this flag if we're being called from | |
1771 | * ocfs2_read_locked_inode(). It means we're locking an inode | |
1772 | * which hasn't been populated yet, so clear the refresh flag | |
1773 | * and let the caller handle it. | |
1774 | */ | |
1775 | if (inode->i_state & I_NEW) { | |
1776 | status = 0; | |
1777 | ocfs2_complete_lock_res_refresh(lockres, 0); | |
1778 | goto bail; | |
1779 | } | |
1780 | ||
ccd979bd MF |
1781 | /* This is fun. The caller may want a bh back, or it may |
1782 | * not. ocfs2_meta_lock_update definitely wants one in, but | |
1783 | * may or may not read one, depending on what's in the | |
1784 | * LVB. The result of all of this is that we've *only* gone to | |
1785 | * disk if we have to, so the complexity is worthwhile. */ | |
1786 | status = ocfs2_meta_lock_update(inode, &local_bh); | |
1787 | if (status < 0) { | |
1788 | if (status != -ENOENT) | |
1789 | mlog_errno(status); | |
1790 | goto bail; | |
1791 | } | |
1792 | ||
1793 | if (ret_bh) { | |
1794 | status = ocfs2_assign_bh(inode, ret_bh, local_bh); | |
1795 | if (status < 0) { | |
1796 | mlog_errno(status); | |
1797 | goto bail; | |
1798 | } | |
1799 | } | |
1800 | ||
1801 | if (handle) { | |
1802 | status = ocfs2_handle_add_lock(handle, inode); | |
1803 | if (status < 0) | |
1804 | mlog_errno(status); | |
1805 | } | |
1806 | ||
1807 | bail: | |
1808 | if (status < 0) { | |
1809 | if (ret_bh && (*ret_bh)) { | |
1810 | brelse(*ret_bh); | |
1811 | *ret_bh = NULL; | |
1812 | } | |
1813 | if (acquired) | |
1814 | ocfs2_meta_unlock(inode, ex); | |
1815 | } | |
1816 | ||
1817 | if (local_bh) | |
1818 | brelse(local_bh); | |
1819 | ||
1820 | mlog_exit(status); | |
1821 | return status; | |
1822 | } | |
1823 | ||
1824 | /* | |
1825 | * This is working around a lock inversion between tasks acquiring DLM locks | |
1826 | * while holding a page lock and the vote thread which blocks dlm lock acquiry | |
1827 | * while acquiring page locks. | |
1828 | * | |
1829 | * ** These _with_page variantes are only intended to be called from aop | |
1830 | * methods that hold page locks and return a very specific *positive* error | |
1831 | * code that aop methods pass up to the VFS -- test for errors with != 0. ** | |
1832 | * | |
1833 | * The DLM is called such that it returns -EAGAIN if it would have blocked | |
1834 | * waiting for the vote thread. In that case we unlock our page so the vote | |
1835 | * thread can make progress. Once we've done this we have to return | |
1836 | * AOP_TRUNCATED_PAGE so the aop method that called us can bubble that back up | |
1837 | * into the VFS who will then immediately retry the aop call. | |
1838 | * | |
1839 | * We do a blocking lock and immediate unlock before returning, though, so that | |
1840 | * the lock has a great chance of being cached on this node by the time the VFS | |
1841 | * calls back to retry the aop. This has a potential to livelock as nodes | |
1842 | * ping locks back and forth, but that's a risk we're willing to take to avoid | |
1843 | * the lock inversion simply. | |
1844 | */ | |
1845 | int ocfs2_meta_lock_with_page(struct inode *inode, | |
1846 | struct ocfs2_journal_handle *handle, | |
1847 | struct buffer_head **ret_bh, | |
1848 | int ex, | |
1849 | struct page *page) | |
1850 | { | |
1851 | int ret; | |
1852 | ||
1853 | ret = ocfs2_meta_lock_full(inode, handle, ret_bh, ex, | |
1854 | OCFS2_LOCK_NONBLOCK); | |
1855 | if (ret == -EAGAIN) { | |
1856 | unlock_page(page); | |
1857 | if (ocfs2_meta_lock(inode, handle, ret_bh, ex) == 0) | |
1858 | ocfs2_meta_unlock(inode, ex); | |
1859 | ret = AOP_TRUNCATED_PAGE; | |
1860 | } | |
1861 | ||
1862 | return ret; | |
1863 | } | |
1864 | ||
1865 | void ocfs2_meta_unlock(struct inode *inode, | |
1866 | int ex) | |
1867 | { | |
1868 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1869 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres; | |
1870 | ||
1871 | mlog_entry_void(); | |
1872 | ||
b0697053 MF |
1873 | mlog(0, "inode %llu drop %s META lock\n", |
1874 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1875 | ex ? "EXMODE" : "PRMODE"); |
1876 | ||
1877 | if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) | |
1878 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
1879 | ||
1880 | mlog_exit_void(); | |
1881 | } | |
1882 | ||
1883 | int ocfs2_super_lock(struct ocfs2_super *osb, | |
1884 | int ex) | |
1885 | { | |
1886 | int status; | |
1887 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1888 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; | |
1889 | struct buffer_head *bh; | |
1890 | struct ocfs2_slot_info *si = osb->slot_info; | |
1891 | ||
1892 | mlog_entry_void(); | |
1893 | ||
1894 | if (ocfs2_is_hard_readonly(osb)) | |
1895 | return -EROFS; | |
1896 | ||
1897 | status = ocfs2_cluster_lock(osb, lockres, level, 0, 0); | |
1898 | if (status < 0) { | |
1899 | mlog_errno(status); | |
1900 | goto bail; | |
1901 | } | |
1902 | ||
1903 | /* The super block lock path is really in the best position to | |
1904 | * know when resources covered by the lock need to be | |
1905 | * refreshed, so we do it here. Of course, making sense of | |
1906 | * everything is up to the caller :) */ | |
1907 | status = ocfs2_should_refresh_lock_res(lockres); | |
1908 | if (status < 0) { | |
1909 | mlog_errno(status); | |
1910 | goto bail; | |
1911 | } | |
1912 | if (status) { | |
1913 | bh = si->si_bh; | |
1914 | status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0, | |
1915 | si->si_inode); | |
1916 | if (status == 0) | |
1917 | ocfs2_update_slot_info(si); | |
1918 | ||
1919 | ocfs2_complete_lock_res_refresh(lockres, status); | |
1920 | ||
1921 | if (status < 0) | |
1922 | mlog_errno(status); | |
1923 | } | |
1924 | bail: | |
1925 | mlog_exit(status); | |
1926 | return status; | |
1927 | } | |
1928 | ||
1929 | void ocfs2_super_unlock(struct ocfs2_super *osb, | |
1930 | int ex) | |
1931 | { | |
1932 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1933 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; | |
1934 | ||
1935 | ocfs2_cluster_unlock(osb, lockres, level); | |
1936 | } | |
1937 | ||
1938 | int ocfs2_rename_lock(struct ocfs2_super *osb) | |
1939 | { | |
1940 | int status; | |
1941 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; | |
1942 | ||
1943 | if (ocfs2_is_hard_readonly(osb)) | |
1944 | return -EROFS; | |
1945 | ||
1946 | status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, 0); | |
1947 | if (status < 0) | |
1948 | mlog_errno(status); | |
1949 | ||
1950 | return status; | |
1951 | } | |
1952 | ||
1953 | void ocfs2_rename_unlock(struct ocfs2_super *osb) | |
1954 | { | |
1955 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; | |
1956 | ||
1957 | ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE); | |
1958 | } | |
1959 | ||
d680efe9 MF |
1960 | int ocfs2_dentry_lock(struct dentry *dentry, int ex) |
1961 | { | |
1962 | int ret; | |
1963 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1964 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; | |
1965 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); | |
1966 | ||
1967 | BUG_ON(!dl); | |
1968 | ||
1969 | if (ocfs2_is_hard_readonly(osb)) | |
1970 | return -EROFS; | |
1971 | ||
1972 | ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0); | |
1973 | if (ret < 0) | |
1974 | mlog_errno(ret); | |
1975 | ||
1976 | return ret; | |
1977 | } | |
1978 | ||
1979 | void ocfs2_dentry_unlock(struct dentry *dentry, int ex) | |
1980 | { | |
1981 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1982 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; | |
1983 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); | |
1984 | ||
1985 | ocfs2_cluster_unlock(osb, &dl->dl_lockres, level); | |
1986 | } | |
1987 | ||
ccd979bd MF |
1988 | /* Reference counting of the dlm debug structure. We want this because |
1989 | * open references on the debug inodes can live on after a mount, so | |
1990 | * we can't rely on the ocfs2_super to always exist. */ | |
1991 | static void ocfs2_dlm_debug_free(struct kref *kref) | |
1992 | { | |
1993 | struct ocfs2_dlm_debug *dlm_debug; | |
1994 | ||
1995 | dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt); | |
1996 | ||
1997 | kfree(dlm_debug); | |
1998 | } | |
1999 | ||
2000 | void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug) | |
2001 | { | |
2002 | if (dlm_debug) | |
2003 | kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free); | |
2004 | } | |
2005 | ||
2006 | static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug) | |
2007 | { | |
2008 | kref_get(&debug->d_refcnt); | |
2009 | } | |
2010 | ||
2011 | struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void) | |
2012 | { | |
2013 | struct ocfs2_dlm_debug *dlm_debug; | |
2014 | ||
2015 | dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL); | |
2016 | if (!dlm_debug) { | |
2017 | mlog_errno(-ENOMEM); | |
2018 | goto out; | |
2019 | } | |
2020 | ||
2021 | kref_init(&dlm_debug->d_refcnt); | |
2022 | INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking); | |
2023 | dlm_debug->d_locking_state = NULL; | |
2024 | out: | |
2025 | return dlm_debug; | |
2026 | } | |
2027 | ||
2028 | /* Access to this is arbitrated for us via seq_file->sem. */ | |
2029 | struct ocfs2_dlm_seq_priv { | |
2030 | struct ocfs2_dlm_debug *p_dlm_debug; | |
2031 | struct ocfs2_lock_res p_iter_res; | |
2032 | struct ocfs2_lock_res p_tmp_res; | |
2033 | }; | |
2034 | ||
2035 | static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start, | |
2036 | struct ocfs2_dlm_seq_priv *priv) | |
2037 | { | |
2038 | struct ocfs2_lock_res *iter, *ret = NULL; | |
2039 | struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug; | |
2040 | ||
2041 | assert_spin_locked(&ocfs2_dlm_tracking_lock); | |
2042 | ||
2043 | list_for_each_entry(iter, &start->l_debug_list, l_debug_list) { | |
2044 | /* discover the head of the list */ | |
2045 | if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) { | |
2046 | mlog(0, "End of list found, %p\n", ret); | |
2047 | break; | |
2048 | } | |
2049 | ||
2050 | /* We track our "dummy" iteration lockres' by a NULL | |
2051 | * l_ops field. */ | |
2052 | if (iter->l_ops != NULL) { | |
2053 | ret = iter; | |
2054 | break; | |
2055 | } | |
2056 | } | |
2057 | ||
2058 | return ret; | |
2059 | } | |
2060 | ||
2061 | static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos) | |
2062 | { | |
2063 | struct ocfs2_dlm_seq_priv *priv = m->private; | |
2064 | struct ocfs2_lock_res *iter; | |
2065 | ||
2066 | spin_lock(&ocfs2_dlm_tracking_lock); | |
2067 | iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv); | |
2068 | if (iter) { | |
2069 | /* Since lockres' have the lifetime of their container | |
2070 | * (which can be inodes, ocfs2_supers, etc) we want to | |
2071 | * copy this out to a temporary lockres while still | |
2072 | * under the spinlock. Obviously after this we can't | |
2073 | * trust any pointers on the copy returned, but that's | |
2074 | * ok as the information we want isn't typically held | |
2075 | * in them. */ | |
2076 | priv->p_tmp_res = *iter; | |
2077 | iter = &priv->p_tmp_res; | |
2078 | } | |
2079 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
2080 | ||
2081 | return iter; | |
2082 | } | |
2083 | ||
2084 | static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v) | |
2085 | { | |
2086 | } | |
2087 | ||
2088 | static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
2089 | { | |
2090 | struct ocfs2_dlm_seq_priv *priv = m->private; | |
2091 | struct ocfs2_lock_res *iter = v; | |
2092 | struct ocfs2_lock_res *dummy = &priv->p_iter_res; | |
2093 | ||
2094 | spin_lock(&ocfs2_dlm_tracking_lock); | |
2095 | iter = ocfs2_dlm_next_res(iter, priv); | |
2096 | list_del_init(&dummy->l_debug_list); | |
2097 | if (iter) { | |
2098 | list_add(&dummy->l_debug_list, &iter->l_debug_list); | |
2099 | priv->p_tmp_res = *iter; | |
2100 | iter = &priv->p_tmp_res; | |
2101 | } | |
2102 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
2103 | ||
2104 | return iter; | |
2105 | } | |
2106 | ||
2107 | /* So that debugfs.ocfs2 can determine which format is being used */ | |
2108 | #define OCFS2_DLM_DEBUG_STR_VERSION 1 | |
2109 | static int ocfs2_dlm_seq_show(struct seq_file *m, void *v) | |
2110 | { | |
2111 | int i; | |
2112 | char *lvb; | |
2113 | struct ocfs2_lock_res *lockres = v; | |
2114 | ||
2115 | if (!lockres) | |
2116 | return -EINVAL; | |
2117 | ||
d680efe9 MF |
2118 | seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION); |
2119 | ||
2120 | if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY) | |
2121 | seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1, | |
2122 | lockres->l_name, | |
2123 | (unsigned int)ocfs2_get_dentry_lock_ino(lockres)); | |
2124 | else | |
2125 | seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name); | |
2126 | ||
2127 | seq_printf(m, "%d\t" | |
ccd979bd MF |
2128 | "0x%lx\t" |
2129 | "0x%x\t" | |
2130 | "0x%x\t" | |
2131 | "%u\t" | |
2132 | "%u\t" | |
2133 | "%d\t" | |
2134 | "%d\t", | |
ccd979bd MF |
2135 | lockres->l_level, |
2136 | lockres->l_flags, | |
2137 | lockres->l_action, | |
2138 | lockres->l_unlock_action, | |
2139 | lockres->l_ro_holders, | |
2140 | lockres->l_ex_holders, | |
2141 | lockres->l_requested, | |
2142 | lockres->l_blocking); | |
2143 | ||
2144 | /* Dump the raw LVB */ | |
2145 | lvb = lockres->l_lksb.lvb; | |
2146 | for(i = 0; i < DLM_LVB_LEN; i++) | |
2147 | seq_printf(m, "0x%x\t", lvb[i]); | |
2148 | ||
2149 | /* End the line */ | |
2150 | seq_printf(m, "\n"); | |
2151 | return 0; | |
2152 | } | |
2153 | ||
2154 | static struct seq_operations ocfs2_dlm_seq_ops = { | |
2155 | .start = ocfs2_dlm_seq_start, | |
2156 | .stop = ocfs2_dlm_seq_stop, | |
2157 | .next = ocfs2_dlm_seq_next, | |
2158 | .show = ocfs2_dlm_seq_show, | |
2159 | }; | |
2160 | ||
2161 | static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file) | |
2162 | { | |
2163 | struct seq_file *seq = (struct seq_file *) file->private_data; | |
2164 | struct ocfs2_dlm_seq_priv *priv = seq->private; | |
2165 | struct ocfs2_lock_res *res = &priv->p_iter_res; | |
2166 | ||
2167 | ocfs2_remove_lockres_tracking(res); | |
2168 | ocfs2_put_dlm_debug(priv->p_dlm_debug); | |
2169 | return seq_release_private(inode, file); | |
2170 | } | |
2171 | ||
2172 | static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file) | |
2173 | { | |
2174 | int ret; | |
2175 | struct ocfs2_dlm_seq_priv *priv; | |
2176 | struct seq_file *seq; | |
2177 | struct ocfs2_super *osb; | |
2178 | ||
2179 | priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL); | |
2180 | if (!priv) { | |
2181 | ret = -ENOMEM; | |
2182 | mlog_errno(ret); | |
2183 | goto out; | |
2184 | } | |
2185 | osb = (struct ocfs2_super *) inode->u.generic_ip; | |
2186 | ocfs2_get_dlm_debug(osb->osb_dlm_debug); | |
2187 | priv->p_dlm_debug = osb->osb_dlm_debug; | |
2188 | INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list); | |
2189 | ||
2190 | ret = seq_open(file, &ocfs2_dlm_seq_ops); | |
2191 | if (ret) { | |
2192 | kfree(priv); | |
2193 | mlog_errno(ret); | |
2194 | goto out; | |
2195 | } | |
2196 | ||
2197 | seq = (struct seq_file *) file->private_data; | |
2198 | seq->private = priv; | |
2199 | ||
2200 | ocfs2_add_lockres_tracking(&priv->p_iter_res, | |
2201 | priv->p_dlm_debug); | |
2202 | ||
2203 | out: | |
2204 | return ret; | |
2205 | } | |
2206 | ||
4b6f5d20 | 2207 | static const struct file_operations ocfs2_dlm_debug_fops = { |
ccd979bd MF |
2208 | .open = ocfs2_dlm_debug_open, |
2209 | .release = ocfs2_dlm_debug_release, | |
2210 | .read = seq_read, | |
2211 | .llseek = seq_lseek, | |
2212 | }; | |
2213 | ||
2214 | static int ocfs2_dlm_init_debug(struct ocfs2_super *osb) | |
2215 | { | |
2216 | int ret = 0; | |
2217 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; | |
2218 | ||
2219 | dlm_debug->d_locking_state = debugfs_create_file("locking_state", | |
2220 | S_IFREG|S_IRUSR, | |
2221 | osb->osb_debug_root, | |
2222 | osb, | |
2223 | &ocfs2_dlm_debug_fops); | |
2224 | if (!dlm_debug->d_locking_state) { | |
2225 | ret = -EINVAL; | |
2226 | mlog(ML_ERROR, | |
2227 | "Unable to create locking state debugfs file.\n"); | |
2228 | goto out; | |
2229 | } | |
2230 | ||
2231 | ocfs2_get_dlm_debug(dlm_debug); | |
2232 | out: | |
2233 | return ret; | |
2234 | } | |
2235 | ||
2236 | static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb) | |
2237 | { | |
2238 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; | |
2239 | ||
2240 | if (dlm_debug) { | |
2241 | debugfs_remove(dlm_debug->d_locking_state); | |
2242 | ocfs2_put_dlm_debug(dlm_debug); | |
2243 | } | |
2244 | } | |
2245 | ||
2246 | int ocfs2_dlm_init(struct ocfs2_super *osb) | |
2247 | { | |
2248 | int status; | |
2249 | u32 dlm_key; | |
2250 | struct dlm_ctxt *dlm; | |
2251 | ||
2252 | mlog_entry_void(); | |
2253 | ||
2254 | status = ocfs2_dlm_init_debug(osb); | |
2255 | if (status < 0) { | |
2256 | mlog_errno(status); | |
2257 | goto bail; | |
2258 | } | |
2259 | ||
2260 | /* launch vote thread */ | |
78427043 | 2261 | osb->vote_task = kthread_run(ocfs2_vote_thread, osb, "ocfs2vote"); |
ccd979bd MF |
2262 | if (IS_ERR(osb->vote_task)) { |
2263 | status = PTR_ERR(osb->vote_task); | |
2264 | osb->vote_task = NULL; | |
2265 | mlog_errno(status); | |
2266 | goto bail; | |
2267 | } | |
2268 | ||
2269 | /* used by the dlm code to make message headers unique, each | |
2270 | * node in this domain must agree on this. */ | |
2271 | dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str)); | |
2272 | ||
2273 | /* for now, uuid == domain */ | |
2274 | dlm = dlm_register_domain(osb->uuid_str, dlm_key); | |
2275 | if (IS_ERR(dlm)) { | |
2276 | status = PTR_ERR(dlm); | |
2277 | mlog_errno(status); | |
2278 | goto bail; | |
2279 | } | |
2280 | ||
2281 | ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb); | |
2282 | ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb); | |
2283 | ||
2284 | dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb); | |
2285 | ||
2286 | osb->dlm = dlm; | |
2287 | ||
2288 | status = 0; | |
2289 | bail: | |
2290 | if (status < 0) { | |
2291 | ocfs2_dlm_shutdown_debug(osb); | |
2292 | if (osb->vote_task) | |
2293 | kthread_stop(osb->vote_task); | |
2294 | } | |
2295 | ||
2296 | mlog_exit(status); | |
2297 | return status; | |
2298 | } | |
2299 | ||
2300 | void ocfs2_dlm_shutdown(struct ocfs2_super *osb) | |
2301 | { | |
2302 | mlog_entry_void(); | |
2303 | ||
2304 | dlm_unregister_eviction_cb(&osb->osb_eviction_cb); | |
2305 | ||
2306 | ocfs2_drop_osb_locks(osb); | |
2307 | ||
2308 | if (osb->vote_task) { | |
2309 | kthread_stop(osb->vote_task); | |
2310 | osb->vote_task = NULL; | |
2311 | } | |
2312 | ||
2313 | ocfs2_lock_res_free(&osb->osb_super_lockres); | |
2314 | ocfs2_lock_res_free(&osb->osb_rename_lockres); | |
2315 | ||
2316 | dlm_unregister_domain(osb->dlm); | |
2317 | osb->dlm = NULL; | |
2318 | ||
2319 | ocfs2_dlm_shutdown_debug(osb); | |
2320 | ||
2321 | mlog_exit_void(); | |
2322 | } | |
2323 | ||
2324 | static void ocfs2_unlock_ast_func(void *opaque, enum dlm_status status) | |
2325 | { | |
2326 | struct ocfs2_lock_res *lockres = opaque; | |
2327 | unsigned long flags; | |
2328 | ||
2329 | mlog_entry_void(); | |
2330 | ||
2331 | mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name, | |
2332 | lockres->l_unlock_action); | |
2333 | ||
2334 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2335 | /* We tried to cancel a convert request, but it was already | |
2336 | * granted. All we want to do here is clear our unlock | |
2337 | * state. The wake_up call done at the bottom is redundant | |
2338 | * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't | |
2339 | * hurt anything anyway */ | |
2340 | if (status == DLM_CANCELGRANT && | |
2341 | lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { | |
2342 | mlog(0, "Got cancelgrant for %s\n", lockres->l_name); | |
2343 | ||
2344 | /* We don't clear the busy flag in this case as it | |
2345 | * should have been cleared by the ast which the dlm | |
2346 | * has called. */ | |
2347 | goto complete_unlock; | |
2348 | } | |
2349 | ||
2350 | if (status != DLM_NORMAL) { | |
2351 | mlog(ML_ERROR, "Dlm passes status %d for lock %s, " | |
2352 | "unlock_action %d\n", status, lockres->l_name, | |
2353 | lockres->l_unlock_action); | |
2354 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2355 | return; | |
2356 | } | |
2357 | ||
2358 | switch(lockres->l_unlock_action) { | |
2359 | case OCFS2_UNLOCK_CANCEL_CONVERT: | |
2360 | mlog(0, "Cancel convert success for %s\n", lockres->l_name); | |
2361 | lockres->l_action = OCFS2_AST_INVALID; | |
2362 | break; | |
2363 | case OCFS2_UNLOCK_DROP_LOCK: | |
2364 | lockres->l_level = LKM_IVMODE; | |
2365 | break; | |
2366 | default: | |
2367 | BUG(); | |
2368 | } | |
2369 | ||
2370 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
2371 | complete_unlock: | |
2372 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
2373 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2374 | ||
2375 | wake_up(&lockres->l_event); | |
2376 | ||
2377 | mlog_exit_void(); | |
2378 | } | |
2379 | ||
2380 | typedef void (ocfs2_pre_drop_cb_t)(struct ocfs2_lock_res *, void *); | |
2381 | ||
2382 | struct drop_lock_cb { | |
2383 | ocfs2_pre_drop_cb_t *drop_func; | |
2384 | void *drop_data; | |
2385 | }; | |
2386 | ||
2387 | static int ocfs2_drop_lock(struct ocfs2_super *osb, | |
2388 | struct ocfs2_lock_res *lockres, | |
2389 | struct drop_lock_cb *dcb) | |
2390 | { | |
2391 | enum dlm_status status; | |
2392 | unsigned long flags; | |
2393 | ||
2394 | /* We didn't get anywhere near actually using this lockres. */ | |
2395 | if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) | |
2396 | goto out; | |
2397 | ||
2398 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2399 | ||
2400 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING), | |
2401 | "lockres %s, flags 0x%lx\n", | |
2402 | lockres->l_name, lockres->l_flags); | |
2403 | ||
2404 | while (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
2405 | mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = " | |
2406 | "%u, unlock_action = %u\n", | |
2407 | lockres->l_name, lockres->l_flags, lockres->l_action, | |
2408 | lockres->l_unlock_action); | |
2409 | ||
2410 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2411 | ||
2412 | /* XXX: Today we just wait on any busy | |
2413 | * locks... Perhaps we need to cancel converts in the | |
2414 | * future? */ | |
2415 | ocfs2_wait_on_busy_lock(lockres); | |
2416 | ||
2417 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2418 | } | |
2419 | ||
2420 | if (dcb) | |
2421 | dcb->drop_func(lockres, dcb->drop_data); | |
2422 | ||
2423 | if (lockres->l_flags & OCFS2_LOCK_BUSY) | |
2424 | mlog(ML_ERROR, "destroying busy lock: \"%s\"\n", | |
2425 | lockres->l_name); | |
2426 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) | |
2427 | mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name); | |
2428 | ||
2429 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { | |
2430 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2431 | goto out; | |
2432 | } | |
2433 | ||
2434 | lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED); | |
2435 | ||
2436 | /* make sure we never get here while waiting for an ast to | |
2437 | * fire. */ | |
2438 | BUG_ON(lockres->l_action != OCFS2_AST_INVALID); | |
2439 | ||
2440 | /* is this necessary? */ | |
2441 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
2442 | lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK; | |
2443 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2444 | ||
2445 | mlog(0, "lock %s\n", lockres->l_name); | |
2446 | ||
2447 | status = dlmunlock(osb->dlm, &lockres->l_lksb, LKM_VALBLK, | |
2448 | lockres->l_ops->unlock_ast, lockres); | |
2449 | if (status != DLM_NORMAL) { | |
2450 | ocfs2_log_dlm_error("dlmunlock", status, lockres); | |
2451 | mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags); | |
2452 | dlm_print_one_lock(lockres->l_lksb.lockid); | |
2453 | BUG(); | |
2454 | } | |
2455 | mlog(0, "lock %s, successfull return from dlmunlock\n", | |
2456 | lockres->l_name); | |
2457 | ||
2458 | ocfs2_wait_on_busy_lock(lockres); | |
2459 | out: | |
2460 | mlog_exit(0); | |
2461 | return 0; | |
2462 | } | |
2463 | ||
2464 | /* Mark the lockres as being dropped. It will no longer be | |
2465 | * queued if blocking, but we still may have to wait on it | |
2466 | * being dequeued from the vote thread before we can consider | |
2467 | * it safe to drop. | |
2468 | * | |
2469 | * You can *not* attempt to call cluster_lock on this lockres anymore. */ | |
2470 | void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres) | |
2471 | { | |
2472 | int status; | |
2473 | struct ocfs2_mask_waiter mw; | |
2474 | unsigned long flags; | |
2475 | ||
2476 | ocfs2_init_mask_waiter(&mw); | |
2477 | ||
2478 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2479 | lockres->l_flags |= OCFS2_LOCK_FREEING; | |
2480 | while (lockres->l_flags & OCFS2_LOCK_QUEUED) { | |
2481 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0); | |
2482 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2483 | ||
2484 | mlog(0, "Waiting on lockres %s\n", lockres->l_name); | |
2485 | ||
2486 | status = ocfs2_wait_for_mask(&mw); | |
2487 | if (status) | |
2488 | mlog_errno(status); | |
2489 | ||
2490 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2491 | } | |
2492 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2493 | } | |
2494 | ||
d680efe9 MF |
2495 | void ocfs2_simple_drop_lockres(struct ocfs2_super *osb, |
2496 | struct ocfs2_lock_res *lockres) | |
ccd979bd | 2497 | { |
d680efe9 | 2498 | int ret; |
ccd979bd | 2499 | |
d680efe9 MF |
2500 | ocfs2_mark_lockres_freeing(lockres); |
2501 | ret = ocfs2_drop_lock(osb, lockres, NULL); | |
2502 | if (ret) | |
2503 | mlog_errno(ret); | |
2504 | } | |
ccd979bd | 2505 | |
d680efe9 MF |
2506 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb) |
2507 | { | |
2508 | ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres); | |
2509 | ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres); | |
ccd979bd MF |
2510 | } |
2511 | ||
2512 | static void ocfs2_meta_pre_drop(struct ocfs2_lock_res *lockres, void *data) | |
2513 | { | |
2514 | struct inode *inode = data; | |
2515 | ||
2516 | /* the metadata lock requires a bit more work as we have an | |
2517 | * LVB to worry about. */ | |
2518 | if (lockres->l_flags & OCFS2_LOCK_ATTACHED && | |
2519 | lockres->l_level == LKM_EXMODE && | |
2520 | !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) | |
2521 | __ocfs2_stuff_meta_lvb(inode); | |
2522 | } | |
2523 | ||
2524 | int ocfs2_drop_inode_locks(struct inode *inode) | |
2525 | { | |
2526 | int status, err; | |
2527 | struct drop_lock_cb meta_dcb = { ocfs2_meta_pre_drop, inode, }; | |
2528 | ||
2529 | mlog_entry_void(); | |
2530 | ||
2531 | /* No need to call ocfs2_mark_lockres_freeing here - | |
2532 | * ocfs2_clear_inode has done it for us. */ | |
2533 | ||
2534 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
2535 | &OCFS2_I(inode)->ip_data_lockres, | |
2536 | NULL); | |
2537 | if (err < 0) | |
2538 | mlog_errno(err); | |
2539 | ||
2540 | status = err; | |
2541 | ||
2542 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
2543 | &OCFS2_I(inode)->ip_meta_lockres, | |
2544 | &meta_dcb); | |
2545 | if (err < 0) | |
2546 | mlog_errno(err); | |
2547 | if (err < 0 && !status) | |
2548 | status = err; | |
2549 | ||
2550 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
2551 | &OCFS2_I(inode)->ip_rw_lockres, | |
2552 | NULL); | |
2553 | if (err < 0) | |
2554 | mlog_errno(err); | |
2555 | if (err < 0 && !status) | |
2556 | status = err; | |
2557 | ||
2558 | mlog_exit(status); | |
2559 | return status; | |
2560 | } | |
2561 | ||
2562 | static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, | |
2563 | int new_level) | |
2564 | { | |
2565 | assert_spin_locked(&lockres->l_lock); | |
2566 | ||
2567 | BUG_ON(lockres->l_blocking <= LKM_NLMODE); | |
2568 | ||
2569 | if (lockres->l_level <= new_level) { | |
2570 | mlog(ML_ERROR, "lockres->l_level (%u) <= new_level (%u)\n", | |
2571 | lockres->l_level, new_level); | |
2572 | BUG(); | |
2573 | } | |
2574 | ||
2575 | mlog(0, "lock %s, new_level = %d, l_blocking = %d\n", | |
2576 | lockres->l_name, new_level, lockres->l_blocking); | |
2577 | ||
2578 | lockres->l_action = OCFS2_AST_DOWNCONVERT; | |
2579 | lockres->l_requested = new_level; | |
2580 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
2581 | } | |
2582 | ||
2583 | static int ocfs2_downconvert_lock(struct ocfs2_super *osb, | |
2584 | struct ocfs2_lock_res *lockres, | |
2585 | int new_level, | |
2586 | int lvb) | |
2587 | { | |
2588 | int ret, dlm_flags = LKM_CONVERT; | |
2589 | enum dlm_status status; | |
2590 | ||
2591 | mlog_entry_void(); | |
2592 | ||
2593 | if (lvb) | |
2594 | dlm_flags |= LKM_VALBLK; | |
2595 | ||
2596 | status = dlmlock(osb->dlm, | |
2597 | new_level, | |
2598 | &lockres->l_lksb, | |
2599 | dlm_flags, | |
2600 | lockres->l_name, | |
f0681062 | 2601 | OCFS2_LOCK_ID_MAX_LEN - 1, |
ccd979bd MF |
2602 | lockres->l_ops->ast, |
2603 | lockres, | |
2604 | lockres->l_ops->bast); | |
2605 | if (status != DLM_NORMAL) { | |
2606 | ocfs2_log_dlm_error("dlmlock", status, lockres); | |
2607 | ret = -EINVAL; | |
2608 | ocfs2_recover_from_dlm_error(lockres, 1); | |
2609 | goto bail; | |
2610 | } | |
2611 | ||
2612 | ret = 0; | |
2613 | bail: | |
2614 | mlog_exit(ret); | |
2615 | return ret; | |
2616 | } | |
2617 | ||
2618 | /* returns 1 when the caller should unlock and call dlmunlock */ | |
2619 | static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, | |
2620 | struct ocfs2_lock_res *lockres) | |
2621 | { | |
2622 | assert_spin_locked(&lockres->l_lock); | |
2623 | ||
2624 | mlog_entry_void(); | |
2625 | mlog(0, "lock %s\n", lockres->l_name); | |
2626 | ||
2627 | if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { | |
2628 | /* If we're already trying to cancel a lock conversion | |
2629 | * then just drop the spinlock and allow the caller to | |
2630 | * requeue this lock. */ | |
2631 | ||
2632 | mlog(0, "Lockres %s, skip convert\n", lockres->l_name); | |
2633 | return 0; | |
2634 | } | |
2635 | ||
2636 | /* were we in a convert when we got the bast fire? */ | |
2637 | BUG_ON(lockres->l_action != OCFS2_AST_CONVERT && | |
2638 | lockres->l_action != OCFS2_AST_DOWNCONVERT); | |
2639 | /* set things up for the unlockast to know to just | |
2640 | * clear out the ast_action and unset busy, etc. */ | |
2641 | lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT; | |
2642 | ||
2643 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY), | |
2644 | "lock %s, invalid flags: 0x%lx\n", | |
2645 | lockres->l_name, lockres->l_flags); | |
2646 | ||
2647 | return 1; | |
2648 | } | |
2649 | ||
2650 | static int ocfs2_cancel_convert(struct ocfs2_super *osb, | |
2651 | struct ocfs2_lock_res *lockres) | |
2652 | { | |
2653 | int ret; | |
2654 | enum dlm_status status; | |
2655 | ||
2656 | mlog_entry_void(); | |
2657 | mlog(0, "lock %s\n", lockres->l_name); | |
2658 | ||
2659 | ret = 0; | |
2660 | status = dlmunlock(osb->dlm, | |
2661 | &lockres->l_lksb, | |
2662 | LKM_CANCEL, | |
2663 | lockres->l_ops->unlock_ast, | |
2664 | lockres); | |
2665 | if (status != DLM_NORMAL) { | |
2666 | ocfs2_log_dlm_error("dlmunlock", status, lockres); | |
2667 | ret = -EINVAL; | |
2668 | ocfs2_recover_from_dlm_error(lockres, 0); | |
2669 | } | |
2670 | ||
2671 | mlog(0, "lock %s return from dlmunlock\n", lockres->l_name); | |
2672 | ||
2673 | mlog_exit(ret); | |
2674 | return ret; | |
2675 | } | |
2676 | ||
2677 | static inline int ocfs2_can_downconvert_meta_lock(struct inode *inode, | |
2678 | struct ocfs2_lock_res *lockres, | |
2679 | int new_level) | |
2680 | { | |
2681 | int ret; | |
2682 | ||
2683 | mlog_entry_void(); | |
2684 | ||
2685 | BUG_ON(new_level != LKM_NLMODE && new_level != LKM_PRMODE); | |
2686 | ||
2687 | if (lockres->l_flags & OCFS2_LOCK_REFRESHING) { | |
2688 | ret = 0; | |
2689 | mlog(0, "lockres %s currently being refreshed -- backing " | |
2690 | "off!\n", lockres->l_name); | |
2691 | } else if (new_level == LKM_PRMODE) | |
2692 | ret = !lockres->l_ex_holders && | |
2693 | ocfs2_inode_fully_checkpointed(inode); | |
2694 | else /* Must be NLMODE we're converting to. */ | |
2695 | ret = !lockres->l_ro_holders && !lockres->l_ex_holders && | |
2696 | ocfs2_inode_fully_checkpointed(inode); | |
2697 | ||
2698 | mlog_exit(ret); | |
2699 | return ret; | |
2700 | } | |
2701 | ||
2702 | static int ocfs2_do_unblock_meta(struct inode *inode, | |
2703 | int *requeue) | |
2704 | { | |
2705 | int new_level; | |
2706 | int set_lvb = 0; | |
2707 | int ret = 0; | |
2708 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres; | |
2709 | unsigned long flags; | |
2710 | ||
2711 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2712 | ||
2713 | mlog_entry_void(); | |
2714 | ||
2715 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2716 | ||
2717 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
2718 | ||
2719 | mlog(0, "l_level=%d, l_blocking=%d\n", lockres->l_level, | |
2720 | lockres->l_blocking); | |
2721 | ||
2722 | BUG_ON(lockres->l_level != LKM_EXMODE && | |
2723 | lockres->l_level != LKM_PRMODE); | |
2724 | ||
2725 | if (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
2726 | *requeue = 1; | |
2727 | ret = ocfs2_prepare_cancel_convert(osb, lockres); | |
2728 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2729 | if (ret) { | |
2730 | ret = ocfs2_cancel_convert(osb, lockres); | |
2731 | if (ret < 0) | |
2732 | mlog_errno(ret); | |
2733 | } | |
2734 | goto leave; | |
2735 | } | |
2736 | ||
2737 | new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking); | |
2738 | ||
2739 | mlog(0, "l_level=%d, l_blocking=%d, new_level=%d\n", | |
2740 | lockres->l_level, lockres->l_blocking, new_level); | |
2741 | ||
2742 | if (ocfs2_can_downconvert_meta_lock(inode, lockres, new_level)) { | |
2743 | if (lockres->l_level == LKM_EXMODE) | |
2744 | set_lvb = 1; | |
2745 | ||
2746 | /* If the lock hasn't been refreshed yet (rare), then | |
2747 | * our memory inode values are old and we skip | |
2748 | * stuffing the lvb. There's no need to actually clear | |
2749 | * out the lvb here as it's value is still valid. */ | |
2750 | if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) { | |
2751 | if (set_lvb) | |
2752 | __ocfs2_stuff_meta_lvb(inode); | |
2753 | } else | |
2754 | mlog(0, "lockres %s: downconverting stale lock!\n", | |
2755 | lockres->l_name); | |
2756 | ||
2757 | mlog(0, "calling ocfs2_downconvert_lock with l_level=%d, " | |
2758 | "l_blocking=%d, new_level=%d\n", | |
2759 | lockres->l_level, lockres->l_blocking, new_level); | |
2760 | ||
2761 | ocfs2_prepare_downconvert(lockres, new_level); | |
2762 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2763 | ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb); | |
2764 | goto leave; | |
2765 | } | |
2766 | if (!ocfs2_inode_fully_checkpointed(inode)) | |
2767 | ocfs2_start_checkpoint(osb); | |
2768 | ||
2769 | *requeue = 1; | |
2770 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2771 | ret = 0; | |
2772 | leave: | |
2773 | mlog_exit(ret); | |
2774 | return ret; | |
2775 | } | |
2776 | ||
2777 | static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb, | |
2778 | struct ocfs2_lock_res *lockres, | |
d680efe9 | 2779 | struct ocfs2_unblock_ctl *ctl, |
ccd979bd MF |
2780 | ocfs2_convert_worker_t *worker) |
2781 | { | |
2782 | unsigned long flags; | |
2783 | int blocking; | |
2784 | int new_level; | |
2785 | int ret = 0; | |
2786 | ||
2787 | mlog_entry_void(); | |
2788 | ||
2789 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2790 | ||
2791 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
2792 | ||
2793 | recheck: | |
2794 | if (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
d680efe9 | 2795 | ctl->requeue = 1; |
ccd979bd MF |
2796 | ret = ocfs2_prepare_cancel_convert(osb, lockres); |
2797 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2798 | if (ret) { | |
2799 | ret = ocfs2_cancel_convert(osb, lockres); | |
2800 | if (ret < 0) | |
2801 | mlog_errno(ret); | |
2802 | } | |
2803 | goto leave; | |
2804 | } | |
2805 | ||
2806 | /* if we're blocking an exclusive and we have *any* holders, | |
2807 | * then requeue. */ | |
2808 | if ((lockres->l_blocking == LKM_EXMODE) | |
2809 | && (lockres->l_ex_holders || lockres->l_ro_holders)) { | |
2810 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
d680efe9 | 2811 | ctl->requeue = 1; |
ccd979bd MF |
2812 | ret = 0; |
2813 | goto leave; | |
2814 | } | |
2815 | ||
2816 | /* If it's a PR we're blocking, then only | |
2817 | * requeue if we've got any EX holders */ | |
2818 | if (lockres->l_blocking == LKM_PRMODE && | |
2819 | lockres->l_ex_holders) { | |
2820 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
d680efe9 | 2821 | ctl->requeue = 1; |
ccd979bd MF |
2822 | ret = 0; |
2823 | goto leave; | |
2824 | } | |
2825 | ||
2826 | /* If we get here, then we know that there are no more | |
2827 | * incompatible holders (and anyone asking for an incompatible | |
2828 | * lock is blocked). We can now downconvert the lock */ | |
2829 | if (!worker) | |
2830 | goto downconvert; | |
2831 | ||
2832 | /* Some lockres types want to do a bit of work before | |
2833 | * downconverting a lock. Allow that here. The worker function | |
2834 | * may sleep, so we save off a copy of what we're blocking as | |
2835 | * it may change while we're not holding the spin lock. */ | |
2836 | blocking = lockres->l_blocking; | |
2837 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2838 | ||
d680efe9 MF |
2839 | ctl->unblock_action = worker(lockres, blocking); |
2840 | ||
2841 | if (ctl->unblock_action == UNBLOCK_STOP_POST) | |
2842 | goto leave; | |
ccd979bd MF |
2843 | |
2844 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2845 | if (blocking != lockres->l_blocking) { | |
2846 | /* If this changed underneath us, then we can't drop | |
2847 | * it just yet. */ | |
2848 | goto recheck; | |
2849 | } | |
2850 | ||
2851 | downconvert: | |
d680efe9 | 2852 | ctl->requeue = 0; |
ccd979bd MF |
2853 | new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking); |
2854 | ||
2855 | ocfs2_prepare_downconvert(lockres, new_level); | |
2856 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2857 | ret = ocfs2_downconvert_lock(osb, lockres, new_level, 0); | |
2858 | leave: | |
2859 | mlog_exit(ret); | |
2860 | return ret; | |
2861 | } | |
2862 | ||
d680efe9 MF |
2863 | static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, |
2864 | int blocking) | |
ccd979bd MF |
2865 | { |
2866 | struct inode *inode; | |
2867 | struct address_space *mapping; | |
2868 | ||
ccd979bd MF |
2869 | inode = ocfs2_lock_res_inode(lockres); |
2870 | mapping = inode->i_mapping; | |
2871 | ||
2872 | if (filemap_fdatawrite(mapping)) { | |
b0697053 MF |
2873 | mlog(ML_ERROR, "Could not sync inode %llu for downconvert!", |
2874 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
2875 | } |
2876 | sync_mapping_buffers(mapping); | |
2877 | if (blocking == LKM_EXMODE) { | |
2878 | truncate_inode_pages(mapping, 0); | |
2879 | unmap_mapping_range(mapping, 0, 0, 0); | |
2880 | } else { | |
2881 | /* We only need to wait on the I/O if we're not also | |
2882 | * truncating pages because truncate_inode_pages waits | |
2883 | * for us above. We don't truncate pages if we're | |
2884 | * blocking anything < EXMODE because we want to keep | |
2885 | * them around in that case. */ | |
2886 | filemap_fdatawait(mapping); | |
2887 | } | |
2888 | ||
d680efe9 | 2889 | return UNBLOCK_CONTINUE; |
ccd979bd MF |
2890 | } |
2891 | ||
2892 | int ocfs2_unblock_data(struct ocfs2_lock_res *lockres, | |
d680efe9 | 2893 | struct ocfs2_unblock_ctl *ctl) |
ccd979bd MF |
2894 | { |
2895 | int status; | |
2896 | struct inode *inode; | |
2897 | struct ocfs2_super *osb; | |
2898 | ||
2899 | mlog_entry_void(); | |
2900 | ||
2901 | inode = ocfs2_lock_res_inode(lockres); | |
2902 | osb = OCFS2_SB(inode->i_sb); | |
2903 | ||
b0697053 MF |
2904 | mlog(0, "unblock inode %llu\n", |
2905 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 2906 | |
d680efe9 | 2907 | status = ocfs2_generic_unblock_lock(osb, lockres, ctl, |
ccd979bd MF |
2908 | ocfs2_data_convert_worker); |
2909 | if (status < 0) | |
2910 | mlog_errno(status); | |
2911 | ||
b0697053 | 2912 | mlog(0, "inode %llu, requeue = %d\n", |
d680efe9 | 2913 | (unsigned long long)OCFS2_I(inode)->ip_blkno, ctl->requeue); |
ccd979bd MF |
2914 | |
2915 | mlog_exit(status); | |
2916 | return status; | |
2917 | } | |
2918 | ||
2919 | static int ocfs2_unblock_inode_lock(struct ocfs2_lock_res *lockres, | |
d680efe9 | 2920 | struct ocfs2_unblock_ctl *ctl) |
ccd979bd MF |
2921 | { |
2922 | int status; | |
2923 | struct inode *inode; | |
2924 | ||
2925 | mlog_entry_void(); | |
2926 | ||
2927 | mlog(0, "Unblock lockres %s\n", lockres->l_name); | |
2928 | ||
2929 | inode = ocfs2_lock_res_inode(lockres); | |
2930 | ||
2931 | status = ocfs2_generic_unblock_lock(OCFS2_SB(inode->i_sb), | |
d680efe9 | 2932 | lockres, ctl, NULL); |
ccd979bd MF |
2933 | if (status < 0) |
2934 | mlog_errno(status); | |
2935 | ||
2936 | mlog_exit(status); | |
2937 | return status; | |
2938 | } | |
2939 | ||
d680efe9 MF |
2940 | static int ocfs2_unblock_meta(struct ocfs2_lock_res *lockres, |
2941 | struct ocfs2_unblock_ctl *ctl) | |
ccd979bd MF |
2942 | { |
2943 | int status; | |
2944 | struct inode *inode; | |
2945 | ||
2946 | mlog_entry_void(); | |
2947 | ||
2948 | inode = ocfs2_lock_res_inode(lockres); | |
2949 | ||
b0697053 MF |
2950 | mlog(0, "unblock inode %llu\n", |
2951 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 2952 | |
d680efe9 | 2953 | status = ocfs2_do_unblock_meta(inode, &ctl->requeue); |
ccd979bd MF |
2954 | if (status < 0) |
2955 | mlog_errno(status); | |
2956 | ||
b0697053 | 2957 | mlog(0, "inode %llu, requeue = %d\n", |
d680efe9 | 2958 | (unsigned long long)OCFS2_I(inode)->ip_blkno, ctl->requeue); |
ccd979bd MF |
2959 | |
2960 | mlog_exit(status); | |
2961 | return status; | |
2962 | } | |
2963 | ||
d680efe9 MF |
2964 | /* |
2965 | * Does the final reference drop on our dentry lock. Right now this | |
2966 | * happens in the vote thread, but we could choose to simplify the | |
2967 | * dlmglue API and push these off to the ocfs2_wq in the future. | |
2968 | */ | |
2969 | static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, | |
2970 | struct ocfs2_lock_res *lockres) | |
2971 | { | |
2972 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
2973 | ocfs2_dentry_lock_put(osb, dl); | |
2974 | } | |
2975 | ||
2976 | /* | |
2977 | * d_delete() matching dentries before the lock downconvert. | |
2978 | * | |
2979 | * At this point, any process waiting to destroy the | |
2980 | * dentry_lock due to last ref count is stopped by the | |
2981 | * OCFS2_LOCK_QUEUED flag. | |
2982 | * | |
2983 | * We have two potential problems | |
2984 | * | |
2985 | * 1) If we do the last reference drop on our dentry_lock (via dput) | |
2986 | * we'll wind up in ocfs2_release_dentry_lock(), waiting on | |
2987 | * the downconvert to finish. Instead we take an elevated | |
2988 | * reference and push the drop until after we've completed our | |
2989 | * unblock processing. | |
2990 | * | |
2991 | * 2) There might be another process with a final reference, | |
2992 | * waiting on us to finish processing. If this is the case, we | |
2993 | * detect it and exit out - there's no more dentries anyway. | |
2994 | */ | |
2995 | static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, | |
2996 | int blocking) | |
2997 | { | |
2998 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
2999 | struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode); | |
3000 | struct dentry *dentry; | |
3001 | unsigned long flags; | |
3002 | int extra_ref = 0; | |
3003 | ||
3004 | /* | |
3005 | * This node is blocking another node from getting a read | |
3006 | * lock. This happens when we've renamed within a | |
3007 | * directory. We've forced the other nodes to d_delete(), but | |
3008 | * we never actually dropped our lock because it's still | |
3009 | * valid. The downconvert code will retain a PR for this node, | |
3010 | * so there's no further work to do. | |
3011 | */ | |
3012 | if (blocking == LKM_PRMODE) | |
3013 | return UNBLOCK_CONTINUE; | |
3014 | ||
3015 | /* | |
3016 | * Mark this inode as potentially orphaned. The code in | |
3017 | * ocfs2_delete_inode() will figure out whether it actually | |
3018 | * needs to be freed or not. | |
3019 | */ | |
3020 | spin_lock(&oi->ip_lock); | |
3021 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
3022 | spin_unlock(&oi->ip_lock); | |
3023 | ||
3024 | /* | |
3025 | * Yuck. We need to make sure however that the check of | |
3026 | * OCFS2_LOCK_FREEING and the extra reference are atomic with | |
3027 | * respect to a reference decrement or the setting of that | |
3028 | * flag. | |
3029 | */ | |
3030 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3031 | spin_lock(&dentry_attach_lock); | |
3032 | if (!(lockres->l_flags & OCFS2_LOCK_FREEING) | |
3033 | && dl->dl_count) { | |
3034 | dl->dl_count++; | |
3035 | extra_ref = 1; | |
3036 | } | |
3037 | spin_unlock(&dentry_attach_lock); | |
3038 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
3039 | ||
3040 | mlog(0, "extra_ref = %d\n", extra_ref); | |
3041 | ||
3042 | /* | |
3043 | * We have a process waiting on us in ocfs2_dentry_iput(), | |
3044 | * which means we can't have any more outstanding | |
3045 | * aliases. There's no need to do any more work. | |
3046 | */ | |
3047 | if (!extra_ref) | |
3048 | return UNBLOCK_CONTINUE; | |
3049 | ||
3050 | spin_lock(&dentry_attach_lock); | |
3051 | while (1) { | |
3052 | dentry = ocfs2_find_local_alias(dl->dl_inode, | |
3053 | dl->dl_parent_blkno, 1); | |
3054 | if (!dentry) | |
3055 | break; | |
3056 | spin_unlock(&dentry_attach_lock); | |
3057 | ||
3058 | mlog(0, "d_delete(%.*s);\n", dentry->d_name.len, | |
3059 | dentry->d_name.name); | |
3060 | ||
3061 | /* | |
3062 | * The following dcache calls may do an | |
3063 | * iput(). Normally we don't want that from the | |
3064 | * downconverting thread, but in this case it's ok | |
3065 | * because the requesting node already has an | |
3066 | * exclusive lock on the inode, so it can't be queued | |
3067 | * for a downconvert. | |
3068 | */ | |
3069 | d_delete(dentry); | |
3070 | dput(dentry); | |
3071 | ||
3072 | spin_lock(&dentry_attach_lock); | |
3073 | } | |
3074 | spin_unlock(&dentry_attach_lock); | |
3075 | ||
3076 | /* | |
3077 | * If we are the last holder of this dentry lock, there is no | |
3078 | * reason to downconvert so skip straight to the unlock. | |
3079 | */ | |
3080 | if (dl->dl_count == 1) | |
3081 | return UNBLOCK_STOP_POST; | |
3082 | ||
3083 | return UNBLOCK_CONTINUE_POST; | |
3084 | } | |
3085 | ||
3086 | static int ocfs2_unblock_dentry_lock(struct ocfs2_lock_res *lockres, | |
3087 | struct ocfs2_unblock_ctl *ctl) | |
3088 | { | |
3089 | int ret; | |
3090 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
3091 | struct ocfs2_super *osb = OCFS2_SB(dl->dl_inode->i_sb); | |
3092 | ||
3093 | mlog(0, "unblock dentry lock: %llu\n", | |
3094 | (unsigned long long)OCFS2_I(dl->dl_inode)->ip_blkno); | |
3095 | ||
3096 | ret = ocfs2_generic_unblock_lock(osb, | |
3097 | lockres, | |
3098 | ctl, | |
3099 | ocfs2_dentry_convert_worker); | |
3100 | if (ret < 0) | |
3101 | mlog_errno(ret); | |
3102 | ||
3103 | mlog(0, "requeue = %d, post = %d\n", ctl->requeue, ctl->unblock_action); | |
3104 | ||
3105 | return ret; | |
3106 | } | |
3107 | ||
ccd979bd MF |
3108 | /* Generic unblock function for any lockres whose private data is an |
3109 | * ocfs2_super pointer. */ | |
3110 | static int ocfs2_unblock_osb_lock(struct ocfs2_lock_res *lockres, | |
d680efe9 | 3111 | struct ocfs2_unblock_ctl *ctl) |
ccd979bd MF |
3112 | { |
3113 | int status; | |
3114 | struct ocfs2_super *osb; | |
3115 | ||
3116 | mlog_entry_void(); | |
3117 | ||
3118 | mlog(0, "Unblock lockres %s\n", lockres->l_name); | |
3119 | ||
3120 | osb = ocfs2_lock_res_super(lockres); | |
3121 | ||
3122 | status = ocfs2_generic_unblock_lock(osb, | |
3123 | lockres, | |
d680efe9 | 3124 | ctl, |
ccd979bd MF |
3125 | NULL); |
3126 | if (status < 0) | |
3127 | mlog_errno(status); | |
3128 | ||
3129 | mlog_exit(status); | |
3130 | return status; | |
3131 | } | |
3132 | ||
3133 | void ocfs2_process_blocked_lock(struct ocfs2_super *osb, | |
3134 | struct ocfs2_lock_res *lockres) | |
3135 | { | |
3136 | int status; | |
d680efe9 | 3137 | struct ocfs2_unblock_ctl ctl = {0, 0,}; |
ccd979bd MF |
3138 | unsigned long flags; |
3139 | ||
3140 | /* Our reference to the lockres in this function can be | |
3141 | * considered valid until we remove the OCFS2_LOCK_QUEUED | |
3142 | * flag. */ | |
3143 | ||
3144 | mlog_entry_void(); | |
3145 | ||
3146 | BUG_ON(!lockres); | |
3147 | BUG_ON(!lockres->l_ops); | |
3148 | BUG_ON(!lockres->l_ops->unblock); | |
3149 | ||
3150 | mlog(0, "lockres %s blocked.\n", lockres->l_name); | |
3151 | ||
3152 | /* Detect whether a lock has been marked as going away while | |
3153 | * the vote thread was processing other things. A lock can | |
3154 | * still be marked with OCFS2_LOCK_FREEING after this check, | |
3155 | * but short circuiting here will still save us some | |
3156 | * performance. */ | |
3157 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3158 | if (lockres->l_flags & OCFS2_LOCK_FREEING) | |
3159 | goto unqueue; | |
3160 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
3161 | ||
d680efe9 | 3162 | status = lockres->l_ops->unblock(lockres, &ctl); |
ccd979bd MF |
3163 | if (status < 0) |
3164 | mlog_errno(status); | |
3165 | ||
3166 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3167 | unqueue: | |
d680efe9 | 3168 | if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) { |
ccd979bd MF |
3169 | lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED); |
3170 | } else | |
3171 | ocfs2_schedule_blocked_lock(osb, lockres); | |
3172 | ||
3173 | mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name, | |
d680efe9 | 3174 | ctl.requeue ? "yes" : "no"); |
ccd979bd MF |
3175 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3176 | ||
d680efe9 MF |
3177 | if (ctl.unblock_action != UNBLOCK_CONTINUE |
3178 | && lockres->l_ops->post_unlock) | |
3179 | lockres->l_ops->post_unlock(osb, lockres); | |
3180 | ||
ccd979bd MF |
3181 | mlog_exit_void(); |
3182 | } | |
3183 | ||
3184 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, | |
3185 | struct ocfs2_lock_res *lockres) | |
3186 | { | |
3187 | mlog_entry_void(); | |
3188 | ||
3189 | assert_spin_locked(&lockres->l_lock); | |
3190 | ||
3191 | if (lockres->l_flags & OCFS2_LOCK_FREEING) { | |
3192 | /* Do not schedule a lock for downconvert when it's on | |
3193 | * the way to destruction - any nodes wanting access | |
3194 | * to the resource will get it soon. */ | |
3195 | mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n", | |
3196 | lockres->l_name, lockres->l_flags); | |
3197 | return; | |
3198 | } | |
3199 | ||
3200 | lockres_or_flags(lockres, OCFS2_LOCK_QUEUED); | |
3201 | ||
3202 | spin_lock(&osb->vote_task_lock); | |
3203 | if (list_empty(&lockres->l_blocked_list)) { | |
3204 | list_add_tail(&lockres->l_blocked_list, | |
3205 | &osb->blocked_lock_list); | |
3206 | osb->blocked_lock_count++; | |
3207 | } | |
3208 | spin_unlock(&osb->vote_task_lock); | |
3209 | ||
3210 | mlog_exit_void(); | |
3211 | } | |
3212 | ||
3213 | /* This aids in debugging situations where a bad LVB might be involved. */ | |
3214 | void ocfs2_dump_meta_lvb_info(u64 level, | |
3215 | const char *function, | |
3216 | unsigned int line, | |
3217 | struct ocfs2_lock_res *lockres) | |
3218 | { | |
3219 | struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
3220 | ||
3221 | mlog(level, "LVB information for %s (called from %s:%u):\n", | |
3222 | lockres->l_name, function, line); | |
f9e2d82e MF |
3223 | mlog(level, "version: %u, clusters: %u, generation: 0x%x\n", |
3224 | lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters), | |
3225 | be32_to_cpu(lvb->lvb_igeneration)); | |
b0697053 MF |
3226 | mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n", |
3227 | (unsigned long long)be64_to_cpu(lvb->lvb_isize), | |
3228 | be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid), | |
3229 | be16_to_cpu(lvb->lvb_imode)); | |
3230 | mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, " | |
ca4d147e | 3231 | "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink), |
b0697053 MF |
3232 | (long long)be64_to_cpu(lvb->lvb_iatime_packed), |
3233 | (long long)be64_to_cpu(lvb->lvb_ictime_packed), | |
ca4d147e HP |
3234 | (long long)be64_to_cpu(lvb->lvb_imtime_packed), |
3235 | be32_to_cpu(lvb->lvb_iattr)); | |
ccd979bd | 3236 | } |