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 | ||
54a7e755 MF |
70 | static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres); |
71 | static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres); | |
ccd979bd | 72 | |
d680efe9 | 73 | /* |
cc567d89 | 74 | * Return value from ->downconvert_worker functions. |
d680efe9 | 75 | * |
b5e500e2 | 76 | * These control the precise actions of ocfs2_unblock_lock() |
d680efe9 MF |
77 | * and ocfs2_process_blocked_lock() |
78 | * | |
79 | */ | |
80 | enum ocfs2_unblock_action { | |
81 | UNBLOCK_CONTINUE = 0, /* Continue downconvert */ | |
82 | UNBLOCK_CONTINUE_POST = 1, /* Continue downconvert, fire | |
83 | * ->post_unlock callback */ | |
84 | UNBLOCK_STOP_POST = 2, /* Do not downconvert, fire | |
85 | * ->post_unlock() callback. */ | |
86 | }; | |
87 | ||
88 | struct ocfs2_unblock_ctl { | |
89 | int requeue; | |
90 | enum ocfs2_unblock_action unblock_action; | |
91 | }; | |
92 | ||
810d5aeb MF |
93 | static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres, |
94 | int new_level); | |
95 | static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres); | |
96 | ||
cc567d89 MF |
97 | static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, |
98 | int blocking); | |
99 | ||
cc567d89 MF |
100 | static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, |
101 | int blocking); | |
d680efe9 MF |
102 | |
103 | static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, | |
104 | struct ocfs2_lock_res *lockres); | |
ccd979bd | 105 | |
f625c979 MF |
106 | /* |
107 | * OCFS2 Lock Resource Operations | |
108 | * | |
109 | * These fine tune the behavior of the generic dlmglue locking infrastructure. | |
0d5dc6c2 MF |
110 | * |
111 | * The most basic of lock types can point ->l_priv to their respective | |
112 | * struct ocfs2_super and allow the default actions to manage things. | |
113 | * | |
114 | * Right now, each lock type also needs to implement an init function, | |
115 | * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres() | |
116 | * should be called when the lock is no longer needed (i.e., object | |
117 | * destruction time). | |
f625c979 | 118 | */ |
ccd979bd | 119 | struct ocfs2_lock_res_ops { |
54a7e755 MF |
120 | /* |
121 | * Translate an ocfs2_lock_res * into an ocfs2_super *. Define | |
122 | * this callback if ->l_priv is not an ocfs2_super pointer | |
123 | */ | |
124 | struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *); | |
b5e500e2 | 125 | |
0d5dc6c2 MF |
126 | /* |
127 | * Optionally called in the downconvert (or "vote") thread | |
128 | * after a successful downconvert. The lockres will not be | |
129 | * referenced after this callback is called, so it is safe to | |
130 | * free memory, etc. | |
131 | * | |
132 | * The exact semantics of when this is called are controlled | |
133 | * by ->downconvert_worker() | |
134 | */ | |
d680efe9 | 135 | void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *); |
f625c979 | 136 | |
16d5b956 MF |
137 | /* |
138 | * Allow a lock type to add checks to determine whether it is | |
139 | * safe to downconvert a lock. Return 0 to re-queue the | |
140 | * downconvert at a later time, nonzero to continue. | |
141 | * | |
142 | * For most locks, the default checks that there are no | |
143 | * incompatible holders are sufficient. | |
144 | * | |
145 | * Called with the lockres spinlock held. | |
146 | */ | |
147 | int (*check_downconvert)(struct ocfs2_lock_res *, int); | |
148 | ||
5ef0d4ea MF |
149 | /* |
150 | * Allows a lock type to populate the lock value block. This | |
151 | * is called on downconvert, and when we drop a lock. | |
152 | * | |
153 | * Locks that want to use this should set LOCK_TYPE_USES_LVB | |
154 | * in the flags field. | |
155 | * | |
156 | * Called with the lockres spinlock held. | |
157 | */ | |
158 | void (*set_lvb)(struct ocfs2_lock_res *); | |
159 | ||
cc567d89 MF |
160 | /* |
161 | * Called from the downconvert thread when it is determined | |
162 | * that a lock will be downconverted. This is called without | |
163 | * any locks held so the function can do work that might | |
164 | * schedule (syncing out data, etc). | |
165 | * | |
166 | * This should return any one of the ocfs2_unblock_action | |
167 | * values, depending on what it wants the thread to do. | |
168 | */ | |
169 | int (*downconvert_worker)(struct ocfs2_lock_res *, int); | |
170 | ||
f625c979 MF |
171 | /* |
172 | * LOCK_TYPE_* flags which describe the specific requirements | |
173 | * of a lock type. Descriptions of each individual flag follow. | |
174 | */ | |
175 | int flags; | |
ccd979bd MF |
176 | }; |
177 | ||
f625c979 MF |
178 | /* |
179 | * Some locks want to "refresh" potentially stale data when a | |
180 | * meaningful (PRMODE or EXMODE) lock level is first obtained. If this | |
181 | * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the | |
182 | * individual lockres l_flags member from the ast function. It is | |
183 | * expected that the locking wrapper will clear the | |
184 | * OCFS2_LOCK_NEEDS_REFRESH flag when done. | |
185 | */ | |
186 | #define LOCK_TYPE_REQUIRES_REFRESH 0x1 | |
187 | ||
b80fc012 | 188 | /* |
5ef0d4ea MF |
189 | * Indicate that a lock type makes use of the lock value block. The |
190 | * ->set_lvb lock type callback must be defined. | |
b80fc012 MF |
191 | */ |
192 | #define LOCK_TYPE_USES_LVB 0x2 | |
193 | ||
ccd979bd | 194 | static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = { |
54a7e755 | 195 | .get_osb = ocfs2_get_inode_osb, |
f625c979 | 196 | .flags = 0, |
ccd979bd MF |
197 | }; |
198 | ||
199 | static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops = { | |
54a7e755 | 200 | .get_osb = ocfs2_get_inode_osb, |
810d5aeb MF |
201 | .check_downconvert = ocfs2_check_meta_downconvert, |
202 | .set_lvb = ocfs2_set_meta_lvb, | |
b80fc012 | 203 | .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB, |
ccd979bd MF |
204 | }; |
205 | ||
ccd979bd | 206 | static struct ocfs2_lock_res_ops ocfs2_inode_data_lops = { |
54a7e755 | 207 | .get_osb = ocfs2_get_inode_osb, |
cc567d89 | 208 | .downconvert_worker = ocfs2_data_convert_worker, |
f625c979 | 209 | .flags = 0, |
ccd979bd MF |
210 | }; |
211 | ||
212 | static struct ocfs2_lock_res_ops ocfs2_super_lops = { | |
f625c979 | 213 | .flags = LOCK_TYPE_REQUIRES_REFRESH, |
ccd979bd MF |
214 | }; |
215 | ||
216 | static struct ocfs2_lock_res_ops ocfs2_rename_lops = { | |
f625c979 | 217 | .flags = 0, |
ccd979bd MF |
218 | }; |
219 | ||
d680efe9 | 220 | static struct ocfs2_lock_res_ops ocfs2_dentry_lops = { |
54a7e755 | 221 | .get_osb = ocfs2_get_dentry_osb, |
d680efe9 | 222 | .post_unlock = ocfs2_dentry_post_unlock, |
cc567d89 | 223 | .downconvert_worker = ocfs2_dentry_convert_worker, |
f625c979 | 224 | .flags = 0, |
d680efe9 MF |
225 | }; |
226 | ||
ccd979bd MF |
227 | static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres) |
228 | { | |
229 | return lockres->l_type == OCFS2_LOCK_TYPE_META || | |
230 | lockres->l_type == OCFS2_LOCK_TYPE_DATA || | |
231 | lockres->l_type == OCFS2_LOCK_TYPE_RW; | |
232 | } | |
233 | ||
ccd979bd MF |
234 | static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres) |
235 | { | |
236 | BUG_ON(!ocfs2_is_inode_lock(lockres)); | |
237 | ||
238 | return (struct inode *) lockres->l_priv; | |
239 | } | |
240 | ||
d680efe9 MF |
241 | static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres) |
242 | { | |
243 | BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY); | |
244 | ||
245 | return (struct ocfs2_dentry_lock *)lockres->l_priv; | |
246 | } | |
247 | ||
54a7e755 MF |
248 | static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres) |
249 | { | |
250 | if (lockres->l_ops->get_osb) | |
251 | return lockres->l_ops->get_osb(lockres); | |
252 | ||
253 | return (struct ocfs2_super *)lockres->l_priv; | |
254 | } | |
255 | ||
ccd979bd MF |
256 | static int ocfs2_lock_create(struct ocfs2_super *osb, |
257 | struct ocfs2_lock_res *lockres, | |
258 | int level, | |
259 | int dlm_flags); | |
260 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, | |
261 | int wanted); | |
262 | static void ocfs2_cluster_unlock(struct ocfs2_super *osb, | |
263 | struct ocfs2_lock_res *lockres, | |
264 | int level); | |
265 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres); | |
266 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres); | |
267 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres); | |
268 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level); | |
269 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, | |
270 | struct ocfs2_lock_res *lockres); | |
271 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, | |
272 | int convert); | |
273 | #define ocfs2_log_dlm_error(_func, _stat, _lockres) do { \ | |
274 | mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \ | |
275 | "resource %s: %s\n", dlm_errname(_stat), _func, \ | |
276 | _lockres->l_name, dlm_errmsg(_stat)); \ | |
277 | } while (0) | |
278 | static void ocfs2_vote_on_unlock(struct ocfs2_super *osb, | |
279 | struct ocfs2_lock_res *lockres); | |
280 | static int ocfs2_meta_lock_update(struct inode *inode, | |
281 | struct buffer_head **bh); | |
282 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb); | |
283 | static inline int ocfs2_highest_compat_lock_level(int level); | |
ccd979bd | 284 | |
ccd979bd MF |
285 | static void ocfs2_build_lock_name(enum ocfs2_lock_type type, |
286 | u64 blkno, | |
287 | u32 generation, | |
288 | char *name) | |
289 | { | |
290 | int len; | |
291 | ||
292 | mlog_entry_void(); | |
293 | ||
294 | BUG_ON(type >= OCFS2_NUM_LOCK_TYPES); | |
295 | ||
b0697053 MF |
296 | len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x", |
297 | ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD, | |
298 | (long long)blkno, generation); | |
ccd979bd MF |
299 | |
300 | BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1)); | |
301 | ||
302 | mlog(0, "built lock resource with name: %s\n", name); | |
303 | ||
304 | mlog_exit_void(); | |
305 | } | |
306 | ||
34af946a | 307 | static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock); |
ccd979bd MF |
308 | |
309 | static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res, | |
310 | struct ocfs2_dlm_debug *dlm_debug) | |
311 | { | |
312 | mlog(0, "Add tracking for lockres %s\n", res->l_name); | |
313 | ||
314 | spin_lock(&ocfs2_dlm_tracking_lock); | |
315 | list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking); | |
316 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
317 | } | |
318 | ||
319 | static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res) | |
320 | { | |
321 | spin_lock(&ocfs2_dlm_tracking_lock); | |
322 | if (!list_empty(&res->l_debug_list)) | |
323 | list_del_init(&res->l_debug_list); | |
324 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
325 | } | |
326 | ||
327 | static void ocfs2_lock_res_init_common(struct ocfs2_super *osb, | |
328 | struct ocfs2_lock_res *res, | |
329 | enum ocfs2_lock_type type, | |
ccd979bd MF |
330 | struct ocfs2_lock_res_ops *ops, |
331 | void *priv) | |
332 | { | |
ccd979bd MF |
333 | res->l_type = type; |
334 | res->l_ops = ops; | |
335 | res->l_priv = priv; | |
336 | ||
337 | res->l_level = LKM_IVMODE; | |
338 | res->l_requested = LKM_IVMODE; | |
339 | res->l_blocking = LKM_IVMODE; | |
340 | res->l_action = OCFS2_AST_INVALID; | |
341 | res->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
342 | ||
343 | res->l_flags = OCFS2_LOCK_INITIALIZED; | |
344 | ||
345 | ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug); | |
346 | } | |
347 | ||
348 | void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res) | |
349 | { | |
350 | /* This also clears out the lock status block */ | |
351 | memset(res, 0, sizeof(struct ocfs2_lock_res)); | |
352 | spin_lock_init(&res->l_lock); | |
353 | init_waitqueue_head(&res->l_event); | |
354 | INIT_LIST_HEAD(&res->l_blocked_list); | |
355 | INIT_LIST_HEAD(&res->l_mask_waiters); | |
356 | } | |
357 | ||
358 | void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res, | |
359 | enum ocfs2_lock_type type, | |
24c19ef4 | 360 | unsigned int generation, |
ccd979bd MF |
361 | struct inode *inode) |
362 | { | |
363 | struct ocfs2_lock_res_ops *ops; | |
364 | ||
365 | switch(type) { | |
366 | case OCFS2_LOCK_TYPE_RW: | |
367 | ops = &ocfs2_inode_rw_lops; | |
368 | break; | |
369 | case OCFS2_LOCK_TYPE_META: | |
370 | ops = &ocfs2_inode_meta_lops; | |
371 | break; | |
372 | case OCFS2_LOCK_TYPE_DATA: | |
373 | ops = &ocfs2_inode_data_lops; | |
374 | break; | |
375 | default: | |
376 | mlog_bug_on_msg(1, "type: %d\n", type); | |
377 | ops = NULL; /* thanks, gcc */ | |
378 | break; | |
379 | }; | |
380 | ||
d680efe9 | 381 | ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno, |
24c19ef4 | 382 | generation, res->l_name); |
d680efe9 MF |
383 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode); |
384 | } | |
385 | ||
54a7e755 MF |
386 | static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres) |
387 | { | |
388 | struct inode *inode = ocfs2_lock_res_inode(lockres); | |
389 | ||
390 | return OCFS2_SB(inode->i_sb); | |
391 | } | |
392 | ||
d680efe9 MF |
393 | static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres) |
394 | { | |
395 | __be64 inode_blkno_be; | |
396 | ||
397 | memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], | |
398 | sizeof(__be64)); | |
399 | ||
400 | return be64_to_cpu(inode_blkno_be); | |
401 | } | |
402 | ||
54a7e755 MF |
403 | static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres) |
404 | { | |
405 | struct ocfs2_dentry_lock *dl = lockres->l_priv; | |
406 | ||
407 | return OCFS2_SB(dl->dl_inode->i_sb); | |
408 | } | |
409 | ||
d680efe9 MF |
410 | void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl, |
411 | u64 parent, struct inode *inode) | |
412 | { | |
413 | int len; | |
414 | u64 inode_blkno = OCFS2_I(inode)->ip_blkno; | |
415 | __be64 inode_blkno_be = cpu_to_be64(inode_blkno); | |
416 | struct ocfs2_lock_res *lockres = &dl->dl_lockres; | |
417 | ||
418 | ocfs2_lock_res_init_once(lockres); | |
419 | ||
420 | /* | |
421 | * Unfortunately, the standard lock naming scheme won't work | |
422 | * here because we have two 16 byte values to use. Instead, | |
423 | * we'll stuff the inode number as a binary value. We still | |
424 | * want error prints to show something without garbling the | |
425 | * display, so drop a null byte in there before the inode | |
426 | * number. A future version of OCFS2 will likely use all | |
427 | * binary lock names. The stringified names have been a | |
428 | * tremendous aid in debugging, but now that the debugfs | |
429 | * interface exists, we can mangle things there if need be. | |
430 | * | |
431 | * NOTE: We also drop the standard "pad" value (the total lock | |
432 | * name size stays the same though - the last part is all | |
433 | * zeros due to the memset in ocfs2_lock_res_init_once() | |
434 | */ | |
435 | len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START, | |
436 | "%c%016llx", | |
437 | ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY), | |
438 | (long long)parent); | |
439 | ||
440 | BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1)); | |
441 | ||
442 | memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be, | |
443 | sizeof(__be64)); | |
444 | ||
445 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres, | |
446 | OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops, | |
447 | dl); | |
ccd979bd MF |
448 | } |
449 | ||
450 | static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res, | |
451 | struct ocfs2_super *osb) | |
452 | { | |
453 | /* Superblock lockres doesn't come from a slab so we call init | |
454 | * once on it manually. */ | |
455 | ocfs2_lock_res_init_once(res); | |
d680efe9 MF |
456 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO, |
457 | 0, res->l_name); | |
ccd979bd | 458 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER, |
ccd979bd MF |
459 | &ocfs2_super_lops, osb); |
460 | } | |
461 | ||
462 | static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res, | |
463 | struct ocfs2_super *osb) | |
464 | { | |
465 | /* Rename lockres doesn't come from a slab so we call init | |
466 | * once on it manually. */ | |
467 | ocfs2_lock_res_init_once(res); | |
d680efe9 MF |
468 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name); |
469 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME, | |
ccd979bd MF |
470 | &ocfs2_rename_lops, osb); |
471 | } | |
472 | ||
473 | void ocfs2_lock_res_free(struct ocfs2_lock_res *res) | |
474 | { | |
475 | mlog_entry_void(); | |
476 | ||
477 | if (!(res->l_flags & OCFS2_LOCK_INITIALIZED)) | |
478 | return; | |
479 | ||
480 | ocfs2_remove_lockres_tracking(res); | |
481 | ||
482 | mlog_bug_on_msg(!list_empty(&res->l_blocked_list), | |
483 | "Lockres %s is on the blocked list\n", | |
484 | res->l_name); | |
485 | mlog_bug_on_msg(!list_empty(&res->l_mask_waiters), | |
486 | "Lockres %s has mask waiters pending\n", | |
487 | res->l_name); | |
488 | mlog_bug_on_msg(spin_is_locked(&res->l_lock), | |
489 | "Lockres %s is locked\n", | |
490 | res->l_name); | |
491 | mlog_bug_on_msg(res->l_ro_holders, | |
492 | "Lockres %s has %u ro holders\n", | |
493 | res->l_name, res->l_ro_holders); | |
494 | mlog_bug_on_msg(res->l_ex_holders, | |
495 | "Lockres %s has %u ex holders\n", | |
496 | res->l_name, res->l_ex_holders); | |
497 | ||
498 | /* Need to clear out the lock status block for the dlm */ | |
499 | memset(&res->l_lksb, 0, sizeof(res->l_lksb)); | |
500 | ||
501 | res->l_flags = 0UL; | |
502 | mlog_exit_void(); | |
503 | } | |
504 | ||
505 | static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres, | |
506 | int level) | |
507 | { | |
508 | mlog_entry_void(); | |
509 | ||
510 | BUG_ON(!lockres); | |
511 | ||
512 | switch(level) { | |
513 | case LKM_EXMODE: | |
514 | lockres->l_ex_holders++; | |
515 | break; | |
516 | case LKM_PRMODE: | |
517 | lockres->l_ro_holders++; | |
518 | break; | |
519 | default: | |
520 | BUG(); | |
521 | } | |
522 | ||
523 | mlog_exit_void(); | |
524 | } | |
525 | ||
526 | static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres, | |
527 | int level) | |
528 | { | |
529 | mlog_entry_void(); | |
530 | ||
531 | BUG_ON(!lockres); | |
532 | ||
533 | switch(level) { | |
534 | case LKM_EXMODE: | |
535 | BUG_ON(!lockres->l_ex_holders); | |
536 | lockres->l_ex_holders--; | |
537 | break; | |
538 | case LKM_PRMODE: | |
539 | BUG_ON(!lockres->l_ro_holders); | |
540 | lockres->l_ro_holders--; | |
541 | break; | |
542 | default: | |
543 | BUG(); | |
544 | } | |
545 | mlog_exit_void(); | |
546 | } | |
547 | ||
548 | /* WARNING: This function lives in a world where the only three lock | |
549 | * levels are EX, PR, and NL. It *will* have to be adjusted when more | |
550 | * lock types are added. */ | |
551 | static inline int ocfs2_highest_compat_lock_level(int level) | |
552 | { | |
553 | int new_level = LKM_EXMODE; | |
554 | ||
555 | if (level == LKM_EXMODE) | |
556 | new_level = LKM_NLMODE; | |
557 | else if (level == LKM_PRMODE) | |
558 | new_level = LKM_PRMODE; | |
559 | return new_level; | |
560 | } | |
561 | ||
562 | static void lockres_set_flags(struct ocfs2_lock_res *lockres, | |
563 | unsigned long newflags) | |
564 | { | |
565 | struct list_head *pos, *tmp; | |
566 | struct ocfs2_mask_waiter *mw; | |
567 | ||
568 | assert_spin_locked(&lockres->l_lock); | |
569 | ||
570 | lockres->l_flags = newflags; | |
571 | ||
572 | list_for_each_safe(pos, tmp, &lockres->l_mask_waiters) { | |
573 | mw = list_entry(pos, struct ocfs2_mask_waiter, mw_item); | |
574 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) | |
575 | continue; | |
576 | ||
577 | list_del_init(&mw->mw_item); | |
578 | mw->mw_status = 0; | |
579 | complete(&mw->mw_complete); | |
580 | } | |
581 | } | |
582 | static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or) | |
583 | { | |
584 | lockres_set_flags(lockres, lockres->l_flags | or); | |
585 | } | |
586 | static void lockres_clear_flags(struct ocfs2_lock_res *lockres, | |
587 | unsigned long clear) | |
588 | { | |
589 | lockres_set_flags(lockres, lockres->l_flags & ~clear); | |
590 | } | |
591 | ||
592 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres) | |
593 | { | |
594 | mlog_entry_void(); | |
595 | ||
596 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); | |
597 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); | |
598 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
599 | BUG_ON(lockres->l_blocking <= LKM_NLMODE); | |
600 | ||
601 | lockres->l_level = lockres->l_requested; | |
602 | if (lockres->l_level <= | |
603 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) { | |
604 | lockres->l_blocking = LKM_NLMODE; | |
605 | lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED); | |
606 | } | |
607 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
608 | ||
609 | mlog_exit_void(); | |
610 | } | |
611 | ||
612 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres) | |
613 | { | |
614 | mlog_entry_void(); | |
615 | ||
616 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); | |
617 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); | |
618 | ||
619 | /* Convert from RO to EX doesn't really need anything as our | |
620 | * information is already up to data. Convert from NL to | |
621 | * *anything* however should mark ourselves as needing an | |
622 | * update */ | |
f625c979 MF |
623 | if (lockres->l_level == LKM_NLMODE && |
624 | lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) | |
ccd979bd MF |
625 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); |
626 | ||
627 | lockres->l_level = lockres->l_requested; | |
628 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
629 | ||
630 | mlog_exit_void(); | |
631 | } | |
632 | ||
633 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres) | |
634 | { | |
635 | mlog_entry_void(); | |
636 | ||
637 | BUG_ON((!lockres->l_flags & OCFS2_LOCK_BUSY)); | |
638 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); | |
639 | ||
640 | if (lockres->l_requested > LKM_NLMODE && | |
f625c979 MF |
641 | !(lockres->l_flags & OCFS2_LOCK_LOCAL) && |
642 | lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) | |
ccd979bd MF |
643 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); |
644 | ||
645 | lockres->l_level = lockres->l_requested; | |
646 | lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED); | |
647 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
648 | ||
649 | mlog_exit_void(); | |
650 | } | |
651 | ||
ccd979bd MF |
652 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, |
653 | int level) | |
654 | { | |
655 | int needs_downconvert = 0; | |
656 | mlog_entry_void(); | |
657 | ||
658 | assert_spin_locked(&lockres->l_lock); | |
659 | ||
660 | lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); | |
661 | ||
662 | if (level > lockres->l_blocking) { | |
663 | /* only schedule a downconvert if we haven't already scheduled | |
664 | * one that goes low enough to satisfy the level we're | |
665 | * blocking. this also catches the case where we get | |
666 | * duplicate BASTs */ | |
667 | if (ocfs2_highest_compat_lock_level(level) < | |
668 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) | |
669 | needs_downconvert = 1; | |
670 | ||
671 | lockres->l_blocking = level; | |
672 | } | |
673 | ||
674 | mlog_exit(needs_downconvert); | |
675 | return needs_downconvert; | |
676 | } | |
677 | ||
aa2623ad | 678 | static void ocfs2_blocking_ast(void *opaque, int level) |
ccd979bd | 679 | { |
aa2623ad MF |
680 | struct ocfs2_lock_res *lockres = opaque; |
681 | struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres); | |
ccd979bd MF |
682 | int needs_downconvert; |
683 | unsigned long flags; | |
684 | ||
ccd979bd MF |
685 | BUG_ON(level <= LKM_NLMODE); |
686 | ||
aa2623ad MF |
687 | mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n", |
688 | lockres->l_name, level, lockres->l_level, | |
689 | ocfs2_lock_type_string(lockres->l_type)); | |
690 | ||
ccd979bd MF |
691 | spin_lock_irqsave(&lockres->l_lock, flags); |
692 | needs_downconvert = ocfs2_generic_handle_bast(lockres, level); | |
693 | if (needs_downconvert) | |
694 | ocfs2_schedule_blocked_lock(osb, lockres); | |
695 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
696 | ||
d680efe9 MF |
697 | wake_up(&lockres->l_event); |
698 | ||
ccd979bd | 699 | ocfs2_kick_vote_thread(osb); |
ccd979bd MF |
700 | } |
701 | ||
e92d57df | 702 | static void ocfs2_locking_ast(void *opaque) |
ccd979bd | 703 | { |
e92d57df | 704 | struct ocfs2_lock_res *lockres = opaque; |
ccd979bd MF |
705 | struct dlm_lockstatus *lksb = &lockres->l_lksb; |
706 | unsigned long flags; | |
707 | ||
708 | spin_lock_irqsave(&lockres->l_lock, flags); | |
709 | ||
710 | if (lksb->status != DLM_NORMAL) { | |
711 | mlog(ML_ERROR, "lockres %s: lksb status value of %u!\n", | |
712 | lockres->l_name, lksb->status); | |
713 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
714 | return; | |
715 | } | |
716 | ||
717 | switch(lockres->l_action) { | |
718 | case OCFS2_AST_ATTACH: | |
719 | ocfs2_generic_handle_attach_action(lockres); | |
e92d57df | 720 | lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL); |
ccd979bd MF |
721 | break; |
722 | case OCFS2_AST_CONVERT: | |
723 | ocfs2_generic_handle_convert_action(lockres); | |
724 | break; | |
725 | case OCFS2_AST_DOWNCONVERT: | |
726 | ocfs2_generic_handle_downconvert_action(lockres); | |
727 | break; | |
728 | default: | |
e92d57df MF |
729 | mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u " |
730 | "lockres flags = 0x%lx, unlock action: %u\n", | |
731 | lockres->l_name, lockres->l_action, lockres->l_flags, | |
732 | lockres->l_unlock_action); | |
ccd979bd MF |
733 | BUG(); |
734 | } | |
735 | ||
ccd979bd MF |
736 | /* set it to something invalid so if we get called again we |
737 | * can catch it. */ | |
738 | lockres->l_action = OCFS2_AST_INVALID; | |
ccd979bd MF |
739 | |
740 | wake_up(&lockres->l_event); | |
d680efe9 | 741 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
ccd979bd MF |
742 | } |
743 | ||
ccd979bd MF |
744 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, |
745 | int convert) | |
746 | { | |
747 | unsigned long flags; | |
748 | ||
749 | mlog_entry_void(); | |
750 | spin_lock_irqsave(&lockres->l_lock, flags); | |
751 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
752 | if (convert) | |
753 | lockres->l_action = OCFS2_AST_INVALID; | |
754 | else | |
755 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
756 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
757 | ||
758 | wake_up(&lockres->l_event); | |
759 | mlog_exit_void(); | |
760 | } | |
761 | ||
762 | /* Note: If we detect another process working on the lock (i.e., | |
763 | * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller | |
764 | * to do the right thing in that case. | |
765 | */ | |
766 | static int ocfs2_lock_create(struct ocfs2_super *osb, | |
767 | struct ocfs2_lock_res *lockres, | |
768 | int level, | |
769 | int dlm_flags) | |
770 | { | |
771 | int ret = 0; | |
772 | enum dlm_status status; | |
773 | unsigned long flags; | |
774 | ||
775 | mlog_entry_void(); | |
776 | ||
777 | mlog(0, "lock %s, level = %d, flags = %d\n", lockres->l_name, level, | |
778 | dlm_flags); | |
779 | ||
780 | spin_lock_irqsave(&lockres->l_lock, flags); | |
781 | if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) || | |
782 | (lockres->l_flags & OCFS2_LOCK_BUSY)) { | |
783 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
784 | goto bail; | |
785 | } | |
786 | ||
787 | lockres->l_action = OCFS2_AST_ATTACH; | |
788 | lockres->l_requested = level; | |
789 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
790 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
791 | ||
792 | status = dlmlock(osb->dlm, | |
793 | level, | |
794 | &lockres->l_lksb, | |
795 | dlm_flags, | |
796 | lockres->l_name, | |
f0681062 | 797 | OCFS2_LOCK_ID_MAX_LEN - 1, |
e92d57df | 798 | ocfs2_locking_ast, |
ccd979bd | 799 | lockres, |
aa2623ad | 800 | ocfs2_blocking_ast); |
ccd979bd MF |
801 | if (status != DLM_NORMAL) { |
802 | ocfs2_log_dlm_error("dlmlock", status, lockres); | |
803 | ret = -EINVAL; | |
804 | ocfs2_recover_from_dlm_error(lockres, 1); | |
805 | } | |
806 | ||
807 | mlog(0, "lock %s, successfull return from dlmlock\n", lockres->l_name); | |
808 | ||
809 | bail: | |
810 | mlog_exit(ret); | |
811 | return ret; | |
812 | } | |
813 | ||
814 | static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres, | |
815 | int flag) | |
816 | { | |
817 | unsigned long flags; | |
818 | int ret; | |
819 | ||
820 | spin_lock_irqsave(&lockres->l_lock, flags); | |
821 | ret = lockres->l_flags & flag; | |
822 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
823 | ||
824 | return ret; | |
825 | } | |
826 | ||
827 | static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres) | |
828 | ||
829 | { | |
830 | wait_event(lockres->l_event, | |
831 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY)); | |
832 | } | |
833 | ||
834 | static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres) | |
835 | ||
836 | { | |
837 | wait_event(lockres->l_event, | |
838 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING)); | |
839 | } | |
840 | ||
841 | /* predict what lock level we'll be dropping down to on behalf | |
842 | * of another node, and return true if the currently wanted | |
843 | * level will be compatible with it. */ | |
844 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, | |
845 | int wanted) | |
846 | { | |
847 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
848 | ||
849 | return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking); | |
850 | } | |
851 | ||
852 | static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw) | |
853 | { | |
854 | INIT_LIST_HEAD(&mw->mw_item); | |
855 | init_completion(&mw->mw_complete); | |
856 | } | |
857 | ||
858 | static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw) | |
859 | { | |
860 | wait_for_completion(&mw->mw_complete); | |
861 | /* Re-arm the completion in case we want to wait on it again */ | |
862 | INIT_COMPLETION(mw->mw_complete); | |
863 | return mw->mw_status; | |
864 | } | |
865 | ||
866 | static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres, | |
867 | struct ocfs2_mask_waiter *mw, | |
868 | unsigned long mask, | |
869 | unsigned long goal) | |
870 | { | |
871 | BUG_ON(!list_empty(&mw->mw_item)); | |
872 | ||
873 | assert_spin_locked(&lockres->l_lock); | |
874 | ||
875 | list_add_tail(&mw->mw_item, &lockres->l_mask_waiters); | |
876 | mw->mw_mask = mask; | |
877 | mw->mw_goal = goal; | |
878 | } | |
879 | ||
880 | /* returns 0 if the mw that was removed was already satisfied, -EBUSY | |
881 | * if the mask still hadn't reached its goal */ | |
882 | static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres, | |
883 | struct ocfs2_mask_waiter *mw) | |
884 | { | |
885 | unsigned long flags; | |
886 | int ret = 0; | |
887 | ||
888 | spin_lock_irqsave(&lockres->l_lock, flags); | |
889 | if (!list_empty(&mw->mw_item)) { | |
890 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) | |
891 | ret = -EBUSY; | |
892 | ||
893 | list_del_init(&mw->mw_item); | |
894 | init_completion(&mw->mw_complete); | |
895 | } | |
896 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
897 | ||
898 | return ret; | |
899 | ||
900 | } | |
901 | ||
902 | static int ocfs2_cluster_lock(struct ocfs2_super *osb, | |
903 | struct ocfs2_lock_res *lockres, | |
904 | int level, | |
905 | int lkm_flags, | |
906 | int arg_flags) | |
907 | { | |
908 | struct ocfs2_mask_waiter mw; | |
909 | enum dlm_status status; | |
910 | int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR); | |
911 | int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */ | |
912 | unsigned long flags; | |
913 | ||
914 | mlog_entry_void(); | |
915 | ||
916 | ocfs2_init_mask_waiter(&mw); | |
917 | ||
b80fc012 MF |
918 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) |
919 | lkm_flags |= LKM_VALBLK; | |
920 | ||
ccd979bd MF |
921 | again: |
922 | wait = 0; | |
923 | ||
924 | if (catch_signals && signal_pending(current)) { | |
925 | ret = -ERESTARTSYS; | |
926 | goto out; | |
927 | } | |
928 | ||
929 | spin_lock_irqsave(&lockres->l_lock, flags); | |
930 | ||
931 | mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING, | |
932 | "Cluster lock called on freeing lockres %s! flags " | |
933 | "0x%lx\n", lockres->l_name, lockres->l_flags); | |
934 | ||
935 | /* We only compare against the currently granted level | |
936 | * here. If the lock is blocked waiting on a downconvert, | |
937 | * we'll get caught below. */ | |
938 | if (lockres->l_flags & OCFS2_LOCK_BUSY && | |
939 | level > lockres->l_level) { | |
940 | /* is someone sitting in dlm_lock? If so, wait on | |
941 | * them. */ | |
942 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); | |
943 | wait = 1; | |
944 | goto unlock; | |
945 | } | |
946 | ||
947 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { | |
948 | /* lock has not been created yet. */ | |
949 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
950 | ||
951 | ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0); | |
952 | if (ret < 0) { | |
953 | mlog_errno(ret); | |
954 | goto out; | |
955 | } | |
956 | goto again; | |
957 | } | |
958 | ||
959 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED && | |
960 | !ocfs2_may_continue_on_blocked_lock(lockres, level)) { | |
961 | /* is the lock is currently blocked on behalf of | |
962 | * another node */ | |
963 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0); | |
964 | wait = 1; | |
965 | goto unlock; | |
966 | } | |
967 | ||
968 | if (level > lockres->l_level) { | |
969 | if (lockres->l_action != OCFS2_AST_INVALID) | |
970 | mlog(ML_ERROR, "lockres %s has action %u pending\n", | |
971 | lockres->l_name, lockres->l_action); | |
972 | ||
973 | lockres->l_action = OCFS2_AST_CONVERT; | |
974 | lockres->l_requested = level; | |
975 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
976 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
977 | ||
978 | BUG_ON(level == LKM_IVMODE); | |
979 | BUG_ON(level == LKM_NLMODE); | |
980 | ||
981 | mlog(0, "lock %s, convert from %d to level = %d\n", | |
982 | lockres->l_name, lockres->l_level, level); | |
983 | ||
984 | /* call dlm_lock to upgrade lock now */ | |
985 | status = dlmlock(osb->dlm, | |
986 | level, | |
987 | &lockres->l_lksb, | |
b80fc012 | 988 | lkm_flags|LKM_CONVERT, |
ccd979bd | 989 | lockres->l_name, |
f0681062 | 990 | OCFS2_LOCK_ID_MAX_LEN - 1, |
e92d57df | 991 | ocfs2_locking_ast, |
ccd979bd | 992 | lockres, |
aa2623ad | 993 | ocfs2_blocking_ast); |
ccd979bd MF |
994 | if (status != DLM_NORMAL) { |
995 | if ((lkm_flags & LKM_NOQUEUE) && | |
996 | (status == DLM_NOTQUEUED)) | |
997 | ret = -EAGAIN; | |
998 | else { | |
999 | ocfs2_log_dlm_error("dlmlock", status, | |
1000 | lockres); | |
1001 | ret = -EINVAL; | |
1002 | } | |
1003 | ocfs2_recover_from_dlm_error(lockres, 1); | |
1004 | goto out; | |
1005 | } | |
1006 | ||
1007 | mlog(0, "lock %s, successfull return from dlmlock\n", | |
1008 | lockres->l_name); | |
1009 | ||
1010 | /* At this point we've gone inside the dlm and need to | |
1011 | * complete our work regardless. */ | |
1012 | catch_signals = 0; | |
1013 | ||
1014 | /* wait for busy to clear and carry on */ | |
1015 | goto again; | |
1016 | } | |
1017 | ||
1018 | /* Ok, if we get here then we're good to go. */ | |
1019 | ocfs2_inc_holders(lockres, level); | |
1020 | ||
1021 | ret = 0; | |
1022 | unlock: | |
1023 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1024 | out: | |
1025 | /* | |
1026 | * This is helping work around a lock inversion between the page lock | |
1027 | * and dlm locks. One path holds the page lock while calling aops | |
1028 | * which block acquiring dlm locks. The voting thread holds dlm | |
1029 | * locks while acquiring page locks while down converting data locks. | |
1030 | * This block is helping an aop path notice the inversion and back | |
1031 | * off to unlock its page lock before trying the dlm lock again. | |
1032 | */ | |
1033 | if (wait && arg_flags & OCFS2_LOCK_NONBLOCK && | |
1034 | mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) { | |
1035 | wait = 0; | |
1036 | if (lockres_remove_mask_waiter(lockres, &mw)) | |
1037 | ret = -EAGAIN; | |
1038 | else | |
1039 | goto again; | |
1040 | } | |
1041 | if (wait) { | |
1042 | ret = ocfs2_wait_for_mask(&mw); | |
1043 | if (ret == 0) | |
1044 | goto again; | |
1045 | mlog_errno(ret); | |
1046 | } | |
1047 | ||
1048 | mlog_exit(ret); | |
1049 | return ret; | |
1050 | } | |
1051 | ||
1052 | static void ocfs2_cluster_unlock(struct ocfs2_super *osb, | |
1053 | struct ocfs2_lock_res *lockres, | |
1054 | int level) | |
1055 | { | |
1056 | unsigned long flags; | |
1057 | ||
1058 | mlog_entry_void(); | |
1059 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1060 | ocfs2_dec_holders(lockres, level); | |
1061 | ocfs2_vote_on_unlock(osb, lockres); | |
1062 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1063 | mlog_exit_void(); | |
1064 | } | |
1065 | ||
da66116e AB |
1066 | static int ocfs2_create_new_lock(struct ocfs2_super *osb, |
1067 | struct ocfs2_lock_res *lockres, | |
1068 | int ex, | |
1069 | int local) | |
ccd979bd | 1070 | { |
d680efe9 | 1071 | int level = ex ? LKM_EXMODE : LKM_PRMODE; |
ccd979bd | 1072 | unsigned long flags; |
24c19ef4 | 1073 | int lkm_flags = local ? LKM_LOCAL : 0; |
ccd979bd MF |
1074 | |
1075 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1076 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); | |
1077 | lockres_or_flags(lockres, OCFS2_LOCK_LOCAL); | |
1078 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1079 | ||
24c19ef4 | 1080 | return ocfs2_lock_create(osb, lockres, level, lkm_flags); |
ccd979bd MF |
1081 | } |
1082 | ||
1083 | /* Grants us an EX lock on the data and metadata resources, skipping | |
1084 | * the normal cluster directory lookup. Use this ONLY on newly created | |
1085 | * inodes which other nodes can't possibly see, and which haven't been | |
1086 | * hashed in the inode hash yet. This can give us a good performance | |
1087 | * increase as it'll skip the network broadcast normally associated | |
1088 | * with creating a new lock resource. */ | |
1089 | int ocfs2_create_new_inode_locks(struct inode *inode) | |
1090 | { | |
1091 | int ret; | |
d680efe9 | 1092 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd MF |
1093 | |
1094 | BUG_ON(!inode); | |
1095 | BUG_ON(!ocfs2_inode_is_new(inode)); | |
1096 | ||
1097 | mlog_entry_void(); | |
1098 | ||
b0697053 | 1099 | mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno); |
ccd979bd MF |
1100 | |
1101 | /* NOTE: That we don't increment any of the holder counts, nor | |
1102 | * do we add anything to a journal handle. Since this is | |
1103 | * supposed to be a new inode which the cluster doesn't know | |
1104 | * about yet, there is no need to. As far as the LVB handling | |
1105 | * is concerned, this is basically like acquiring an EX lock | |
1106 | * on a resource which has an invalid one -- we'll set it | |
1107 | * valid when we release the EX. */ | |
1108 | ||
24c19ef4 | 1109 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1); |
ccd979bd MF |
1110 | if (ret) { |
1111 | mlog_errno(ret); | |
1112 | goto bail; | |
1113 | } | |
1114 | ||
24c19ef4 MF |
1115 | /* |
1116 | * We don't want to use LKM_LOCAL on a meta data lock as they | |
1117 | * don't use a generation in their lock names. | |
1118 | */ | |
1119 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_meta_lockres, 1, 0); | |
ccd979bd MF |
1120 | if (ret) { |
1121 | mlog_errno(ret); | |
1122 | goto bail; | |
1123 | } | |
1124 | ||
24c19ef4 | 1125 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_data_lockres, 1, 1); |
ccd979bd MF |
1126 | if (ret) { |
1127 | mlog_errno(ret); | |
1128 | goto bail; | |
1129 | } | |
1130 | ||
1131 | bail: | |
1132 | mlog_exit(ret); | |
1133 | return ret; | |
1134 | } | |
1135 | ||
1136 | int ocfs2_rw_lock(struct inode *inode, int write) | |
1137 | { | |
1138 | int status, level; | |
1139 | struct ocfs2_lock_res *lockres; | |
1140 | ||
1141 | BUG_ON(!inode); | |
1142 | ||
1143 | mlog_entry_void(); | |
1144 | ||
b0697053 MF |
1145 | mlog(0, "inode %llu take %s RW lock\n", |
1146 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1147 | write ? "EXMODE" : "PRMODE"); |
1148 | ||
1149 | lockres = &OCFS2_I(inode)->ip_rw_lockres; | |
1150 | ||
1151 | level = write ? LKM_EXMODE : LKM_PRMODE; | |
1152 | ||
1153 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0, | |
1154 | 0); | |
1155 | if (status < 0) | |
1156 | mlog_errno(status); | |
1157 | ||
1158 | mlog_exit(status); | |
1159 | return status; | |
1160 | } | |
1161 | ||
1162 | void ocfs2_rw_unlock(struct inode *inode, int write) | |
1163 | { | |
1164 | int level = write ? LKM_EXMODE : LKM_PRMODE; | |
1165 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres; | |
1166 | ||
1167 | mlog_entry_void(); | |
1168 | ||
b0697053 MF |
1169 | mlog(0, "inode %llu drop %s RW lock\n", |
1170 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1171 | write ? "EXMODE" : "PRMODE"); |
1172 | ||
1173 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
1174 | ||
1175 | mlog_exit_void(); | |
1176 | } | |
1177 | ||
1178 | int ocfs2_data_lock_full(struct inode *inode, | |
1179 | int write, | |
1180 | int arg_flags) | |
1181 | { | |
1182 | int status = 0, level; | |
1183 | struct ocfs2_lock_res *lockres; | |
1184 | ||
1185 | BUG_ON(!inode); | |
1186 | ||
1187 | mlog_entry_void(); | |
1188 | ||
b0697053 MF |
1189 | mlog(0, "inode %llu take %s DATA lock\n", |
1190 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1191 | write ? "EXMODE" : "PRMODE"); |
1192 | ||
1193 | /* We'll allow faking a readonly data lock for | |
1194 | * rodevices. */ | |
1195 | if (ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) { | |
1196 | if (write) { | |
1197 | status = -EROFS; | |
1198 | mlog_errno(status); | |
1199 | } | |
1200 | goto out; | |
1201 | } | |
1202 | ||
1203 | lockres = &OCFS2_I(inode)->ip_data_lockres; | |
1204 | ||
1205 | level = write ? LKM_EXMODE : LKM_PRMODE; | |
1206 | ||
1207 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, | |
1208 | 0, arg_flags); | |
1209 | if (status < 0 && status != -EAGAIN) | |
1210 | mlog_errno(status); | |
1211 | ||
1212 | out: | |
1213 | mlog_exit(status); | |
1214 | return status; | |
1215 | } | |
1216 | ||
1217 | /* see ocfs2_meta_lock_with_page() */ | |
1218 | int ocfs2_data_lock_with_page(struct inode *inode, | |
1219 | int write, | |
1220 | struct page *page) | |
1221 | { | |
1222 | int ret; | |
1223 | ||
1224 | ret = ocfs2_data_lock_full(inode, write, OCFS2_LOCK_NONBLOCK); | |
1225 | if (ret == -EAGAIN) { | |
1226 | unlock_page(page); | |
1227 | if (ocfs2_data_lock(inode, write) == 0) | |
1228 | ocfs2_data_unlock(inode, write); | |
1229 | ret = AOP_TRUNCATED_PAGE; | |
1230 | } | |
1231 | ||
1232 | return ret; | |
1233 | } | |
1234 | ||
1235 | static void ocfs2_vote_on_unlock(struct ocfs2_super *osb, | |
1236 | struct ocfs2_lock_res *lockres) | |
1237 | { | |
1238 | int kick = 0; | |
1239 | ||
1240 | mlog_entry_void(); | |
1241 | ||
1242 | /* If we know that another node is waiting on our lock, kick | |
1243 | * the vote thread * pre-emptively when we reach a release | |
1244 | * condition. */ | |
1245 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) { | |
1246 | switch(lockres->l_blocking) { | |
1247 | case LKM_EXMODE: | |
1248 | if (!lockres->l_ex_holders && !lockres->l_ro_holders) | |
1249 | kick = 1; | |
1250 | break; | |
1251 | case LKM_PRMODE: | |
1252 | if (!lockres->l_ex_holders) | |
1253 | kick = 1; | |
1254 | break; | |
1255 | default: | |
1256 | BUG(); | |
1257 | } | |
1258 | } | |
1259 | ||
1260 | if (kick) | |
1261 | ocfs2_kick_vote_thread(osb); | |
1262 | ||
1263 | mlog_exit_void(); | |
1264 | } | |
1265 | ||
1266 | void ocfs2_data_unlock(struct inode *inode, | |
1267 | int write) | |
1268 | { | |
1269 | int level = write ? LKM_EXMODE : LKM_PRMODE; | |
1270 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_data_lockres; | |
1271 | ||
1272 | mlog_entry_void(); | |
1273 | ||
b0697053 MF |
1274 | mlog(0, "inode %llu drop %s DATA lock\n", |
1275 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1276 | write ? "EXMODE" : "PRMODE"); |
1277 | ||
1278 | if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) | |
1279 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
1280 | ||
1281 | mlog_exit_void(); | |
1282 | } | |
1283 | ||
1284 | #define OCFS2_SEC_BITS 34 | |
1285 | #define OCFS2_SEC_SHIFT (64 - 34) | |
1286 | #define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1) | |
1287 | ||
1288 | /* LVB only has room for 64 bits of time here so we pack it for | |
1289 | * now. */ | |
1290 | static u64 ocfs2_pack_timespec(struct timespec *spec) | |
1291 | { | |
1292 | u64 res; | |
1293 | u64 sec = spec->tv_sec; | |
1294 | u32 nsec = spec->tv_nsec; | |
1295 | ||
1296 | res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK); | |
1297 | ||
1298 | return res; | |
1299 | } | |
1300 | ||
1301 | /* Call this with the lockres locked. I am reasonably sure we don't | |
1302 | * need ip_lock in this function as anyone who would be changing those | |
1303 | * values is supposed to be blocked in ocfs2_meta_lock right now. */ | |
1304 | static void __ocfs2_stuff_meta_lvb(struct inode *inode) | |
1305 | { | |
1306 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1307 | struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres; | |
1308 | struct ocfs2_meta_lvb *lvb; | |
1309 | ||
1310 | mlog_entry_void(); | |
1311 | ||
1312 | lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
1313 | ||
24c19ef4 MF |
1314 | /* |
1315 | * Invalidate the LVB of a deleted inode - this way other | |
1316 | * nodes are forced to go to disk and discover the new inode | |
1317 | * status. | |
1318 | */ | |
1319 | if (oi->ip_flags & OCFS2_INODE_DELETED) { | |
1320 | lvb->lvb_version = 0; | |
1321 | goto out; | |
1322 | } | |
1323 | ||
4d3b83f7 | 1324 | lvb->lvb_version = OCFS2_LVB_VERSION; |
ccd979bd MF |
1325 | lvb->lvb_isize = cpu_to_be64(i_size_read(inode)); |
1326 | lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters); | |
1327 | lvb->lvb_iuid = cpu_to_be32(inode->i_uid); | |
1328 | lvb->lvb_igid = cpu_to_be32(inode->i_gid); | |
1329 | lvb->lvb_imode = cpu_to_be16(inode->i_mode); | |
1330 | lvb->lvb_inlink = cpu_to_be16(inode->i_nlink); | |
1331 | lvb->lvb_iatime_packed = | |
1332 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime)); | |
1333 | lvb->lvb_ictime_packed = | |
1334 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime)); | |
1335 | lvb->lvb_imtime_packed = | |
1336 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime)); | |
ca4d147e | 1337 | lvb->lvb_iattr = cpu_to_be32(oi->ip_attr); |
f9e2d82e | 1338 | lvb->lvb_igeneration = cpu_to_be32(inode->i_generation); |
ccd979bd | 1339 | |
24c19ef4 | 1340 | out: |
ccd979bd MF |
1341 | mlog_meta_lvb(0, lockres); |
1342 | ||
1343 | mlog_exit_void(); | |
1344 | } | |
1345 | ||
1346 | static void ocfs2_unpack_timespec(struct timespec *spec, | |
1347 | u64 packed_time) | |
1348 | { | |
1349 | spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT; | |
1350 | spec->tv_nsec = packed_time & OCFS2_NSEC_MASK; | |
1351 | } | |
1352 | ||
1353 | static void ocfs2_refresh_inode_from_lvb(struct inode *inode) | |
1354 | { | |
1355 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1356 | struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres; | |
1357 | struct ocfs2_meta_lvb *lvb; | |
1358 | ||
1359 | mlog_entry_void(); | |
1360 | ||
1361 | mlog_meta_lvb(0, lockres); | |
1362 | ||
1363 | lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
1364 | ||
1365 | /* We're safe here without the lockres lock... */ | |
1366 | spin_lock(&oi->ip_lock); | |
1367 | oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters); | |
1368 | i_size_write(inode, be64_to_cpu(lvb->lvb_isize)); | |
1369 | ||
ca4d147e HP |
1370 | oi->ip_attr = be32_to_cpu(lvb->lvb_iattr); |
1371 | ocfs2_set_inode_flags(inode); | |
1372 | ||
ccd979bd MF |
1373 | /* fast-symlinks are a special case */ |
1374 | if (S_ISLNK(inode->i_mode) && !oi->ip_clusters) | |
1375 | inode->i_blocks = 0; | |
1376 | else | |
1377 | inode->i_blocks = | |
1378 | ocfs2_align_bytes_to_sectors(i_size_read(inode)); | |
1379 | ||
1380 | inode->i_uid = be32_to_cpu(lvb->lvb_iuid); | |
1381 | inode->i_gid = be32_to_cpu(lvb->lvb_igid); | |
1382 | inode->i_mode = be16_to_cpu(lvb->lvb_imode); | |
1383 | inode->i_nlink = be16_to_cpu(lvb->lvb_inlink); | |
1384 | ocfs2_unpack_timespec(&inode->i_atime, | |
1385 | be64_to_cpu(lvb->lvb_iatime_packed)); | |
1386 | ocfs2_unpack_timespec(&inode->i_mtime, | |
1387 | be64_to_cpu(lvb->lvb_imtime_packed)); | |
1388 | ocfs2_unpack_timespec(&inode->i_ctime, | |
1389 | be64_to_cpu(lvb->lvb_ictime_packed)); | |
1390 | spin_unlock(&oi->ip_lock); | |
1391 | ||
1392 | mlog_exit_void(); | |
1393 | } | |
1394 | ||
f9e2d82e MF |
1395 | static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode, |
1396 | struct ocfs2_lock_res *lockres) | |
ccd979bd MF |
1397 | { |
1398 | struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
1399 | ||
f9e2d82e MF |
1400 | if (lvb->lvb_version == OCFS2_LVB_VERSION |
1401 | && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation) | |
ccd979bd MF |
1402 | return 1; |
1403 | return 0; | |
1404 | } | |
1405 | ||
1406 | /* Determine whether a lock resource needs to be refreshed, and | |
1407 | * arbitrate who gets to refresh it. | |
1408 | * | |
1409 | * 0 means no refresh needed. | |
1410 | * | |
1411 | * > 0 means you need to refresh this and you MUST call | |
1412 | * ocfs2_complete_lock_res_refresh afterwards. */ | |
1413 | static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres) | |
1414 | { | |
1415 | unsigned long flags; | |
1416 | int status = 0; | |
1417 | ||
1418 | mlog_entry_void(); | |
1419 | ||
1420 | refresh_check: | |
1421 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1422 | if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) { | |
1423 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1424 | goto bail; | |
1425 | } | |
1426 | ||
1427 | if (lockres->l_flags & OCFS2_LOCK_REFRESHING) { | |
1428 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1429 | ||
1430 | ocfs2_wait_on_refreshing_lock(lockres); | |
1431 | goto refresh_check; | |
1432 | } | |
1433 | ||
1434 | /* Ok, I'll be the one to refresh this lock. */ | |
1435 | lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING); | |
1436 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1437 | ||
1438 | status = 1; | |
1439 | bail: | |
1440 | mlog_exit(status); | |
1441 | return status; | |
1442 | } | |
1443 | ||
1444 | /* If status is non zero, I'll mark it as not being in refresh | |
1445 | * anymroe, but i won't clear the needs refresh flag. */ | |
1446 | static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres, | |
1447 | int status) | |
1448 | { | |
1449 | unsigned long flags; | |
1450 | mlog_entry_void(); | |
1451 | ||
1452 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1453 | lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING); | |
1454 | if (!status) | |
1455 | lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); | |
1456 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1457 | ||
1458 | wake_up(&lockres->l_event); | |
1459 | ||
1460 | mlog_exit_void(); | |
1461 | } | |
1462 | ||
1463 | /* may or may not return a bh if it went to disk. */ | |
1464 | static int ocfs2_meta_lock_update(struct inode *inode, | |
1465 | struct buffer_head **bh) | |
1466 | { | |
1467 | int status = 0; | |
1468 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1469 | struct ocfs2_lock_res *lockres; | |
1470 | struct ocfs2_dinode *fe; | |
1471 | ||
1472 | mlog_entry_void(); | |
1473 | ||
1474 | spin_lock(&oi->ip_lock); | |
1475 | if (oi->ip_flags & OCFS2_INODE_DELETED) { | |
b0697053 | 1476 | mlog(0, "Orphaned inode %llu was deleted while we " |
ccd979bd | 1477 | "were waiting on a lock. ip_flags = 0x%x\n", |
b0697053 | 1478 | (unsigned long long)oi->ip_blkno, oi->ip_flags); |
ccd979bd MF |
1479 | spin_unlock(&oi->ip_lock); |
1480 | status = -ENOENT; | |
1481 | goto bail; | |
1482 | } | |
1483 | spin_unlock(&oi->ip_lock); | |
1484 | ||
1485 | lockres = &oi->ip_meta_lockres; | |
1486 | ||
1487 | if (!ocfs2_should_refresh_lock_res(lockres)) | |
1488 | goto bail; | |
1489 | ||
1490 | /* This will discard any caching information we might have had | |
1491 | * for the inode metadata. */ | |
1492 | ocfs2_metadata_cache_purge(inode); | |
1493 | ||
1494 | /* will do nothing for inode types that don't use the extent | |
1495 | * map (directories, bitmap files, etc) */ | |
1496 | ocfs2_extent_map_trunc(inode, 0); | |
1497 | ||
f9e2d82e | 1498 | if (ocfs2_meta_lvb_is_trustable(inode, lockres)) { |
b0697053 MF |
1499 | mlog(0, "Trusting LVB on inode %llu\n", |
1500 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1501 | ocfs2_refresh_inode_from_lvb(inode); |
1502 | } else { | |
1503 | /* Boo, we have to go to disk. */ | |
1504 | /* read bh, cast, ocfs2_refresh_inode */ | |
1505 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno, | |
1506 | bh, OCFS2_BH_CACHED, inode); | |
1507 | if (status < 0) { | |
1508 | mlog_errno(status); | |
1509 | goto bail_refresh; | |
1510 | } | |
1511 | fe = (struct ocfs2_dinode *) (*bh)->b_data; | |
1512 | ||
1513 | /* This is a good chance to make sure we're not | |
1514 | * locking an invalid object. | |
1515 | * | |
1516 | * We bug on a stale inode here because we checked | |
1517 | * above whether it was wiped from disk. The wiping | |
1518 | * node provides a guarantee that we receive that | |
1519 | * message and can mark the inode before dropping any | |
1520 | * locks associated with it. */ | |
1521 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
1522 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
1523 | status = -EIO; | |
1524 | goto bail_refresh; | |
1525 | } | |
1526 | mlog_bug_on_msg(inode->i_generation != | |
1527 | le32_to_cpu(fe->i_generation), | |
b0697053 | 1528 | "Invalid dinode %llu disk generation: %u " |
ccd979bd | 1529 | "inode->i_generation: %u\n", |
b0697053 MF |
1530 | (unsigned long long)oi->ip_blkno, |
1531 | le32_to_cpu(fe->i_generation), | |
ccd979bd MF |
1532 | inode->i_generation); |
1533 | mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) || | |
1534 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)), | |
b0697053 MF |
1535 | "Stale dinode %llu dtime: %llu flags: 0x%x\n", |
1536 | (unsigned long long)oi->ip_blkno, | |
1537 | (unsigned long long)le64_to_cpu(fe->i_dtime), | |
ccd979bd MF |
1538 | le32_to_cpu(fe->i_flags)); |
1539 | ||
1540 | ocfs2_refresh_inode(inode, fe); | |
1541 | } | |
1542 | ||
1543 | status = 0; | |
1544 | bail_refresh: | |
1545 | ocfs2_complete_lock_res_refresh(lockres, status); | |
1546 | bail: | |
1547 | mlog_exit(status); | |
1548 | return status; | |
1549 | } | |
1550 | ||
1551 | static int ocfs2_assign_bh(struct inode *inode, | |
1552 | struct buffer_head **ret_bh, | |
1553 | struct buffer_head *passed_bh) | |
1554 | { | |
1555 | int status; | |
1556 | ||
1557 | if (passed_bh) { | |
1558 | /* Ok, the update went to disk for us, use the | |
1559 | * returned bh. */ | |
1560 | *ret_bh = passed_bh; | |
1561 | get_bh(*ret_bh); | |
1562 | ||
1563 | return 0; | |
1564 | } | |
1565 | ||
1566 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
1567 | OCFS2_I(inode)->ip_blkno, | |
1568 | ret_bh, | |
1569 | OCFS2_BH_CACHED, | |
1570 | inode); | |
1571 | if (status < 0) | |
1572 | mlog_errno(status); | |
1573 | ||
1574 | return status; | |
1575 | } | |
1576 | ||
1577 | /* | |
1578 | * returns < 0 error if the callback will never be called, otherwise | |
1579 | * the result of the lock will be communicated via the callback. | |
1580 | */ | |
1581 | int ocfs2_meta_lock_full(struct inode *inode, | |
ccd979bd MF |
1582 | struct buffer_head **ret_bh, |
1583 | int ex, | |
1584 | int arg_flags) | |
1585 | { | |
1586 | int status, level, dlm_flags, acquired; | |
1587 | struct ocfs2_lock_res *lockres; | |
1588 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1589 | struct buffer_head *local_bh = NULL; | |
1590 | ||
1591 | BUG_ON(!inode); | |
1592 | ||
1593 | mlog_entry_void(); | |
1594 | ||
b0697053 MF |
1595 | mlog(0, "inode %llu, take %s META lock\n", |
1596 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1597 | ex ? "EXMODE" : "PRMODE"); |
1598 | ||
1599 | status = 0; | |
1600 | acquired = 0; | |
1601 | /* We'll allow faking a readonly metadata lock for | |
1602 | * rodevices. */ | |
1603 | if (ocfs2_is_hard_readonly(osb)) { | |
1604 | if (ex) | |
1605 | status = -EROFS; | |
1606 | goto bail; | |
1607 | } | |
1608 | ||
1609 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) | |
1610 | wait_event(osb->recovery_event, | |
1611 | ocfs2_node_map_is_empty(osb, &osb->recovery_map)); | |
1612 | ||
1613 | acquired = 0; | |
1614 | lockres = &OCFS2_I(inode)->ip_meta_lockres; | |
1615 | level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1616 | dlm_flags = 0; | |
1617 | if (arg_flags & OCFS2_META_LOCK_NOQUEUE) | |
1618 | dlm_flags |= LKM_NOQUEUE; | |
1619 | ||
1620 | status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags); | |
1621 | if (status < 0) { | |
1622 | if (status != -EAGAIN && status != -EIOCBRETRY) | |
1623 | mlog_errno(status); | |
1624 | goto bail; | |
1625 | } | |
1626 | ||
1627 | /* Notify the error cleanup path to drop the cluster lock. */ | |
1628 | acquired = 1; | |
1629 | ||
1630 | /* We wait twice because a node may have died while we were in | |
1631 | * the lower dlm layers. The second time though, we've | |
1632 | * committed to owning this lock so we don't allow signals to | |
1633 | * abort the operation. */ | |
1634 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) | |
1635 | wait_event(osb->recovery_event, | |
1636 | ocfs2_node_map_is_empty(osb, &osb->recovery_map)); | |
1637 | ||
24c19ef4 MF |
1638 | /* |
1639 | * We only see this flag if we're being called from | |
1640 | * ocfs2_read_locked_inode(). It means we're locking an inode | |
1641 | * which hasn't been populated yet, so clear the refresh flag | |
1642 | * and let the caller handle it. | |
1643 | */ | |
1644 | if (inode->i_state & I_NEW) { | |
1645 | status = 0; | |
1646 | ocfs2_complete_lock_res_refresh(lockres, 0); | |
1647 | goto bail; | |
1648 | } | |
1649 | ||
ccd979bd MF |
1650 | /* This is fun. The caller may want a bh back, or it may |
1651 | * not. ocfs2_meta_lock_update definitely wants one in, but | |
1652 | * may or may not read one, depending on what's in the | |
1653 | * LVB. The result of all of this is that we've *only* gone to | |
1654 | * disk if we have to, so the complexity is worthwhile. */ | |
1655 | status = ocfs2_meta_lock_update(inode, &local_bh); | |
1656 | if (status < 0) { | |
1657 | if (status != -ENOENT) | |
1658 | mlog_errno(status); | |
1659 | goto bail; | |
1660 | } | |
1661 | ||
1662 | if (ret_bh) { | |
1663 | status = ocfs2_assign_bh(inode, ret_bh, local_bh); | |
1664 | if (status < 0) { | |
1665 | mlog_errno(status); | |
1666 | goto bail; | |
1667 | } | |
1668 | } | |
1669 | ||
ccd979bd MF |
1670 | bail: |
1671 | if (status < 0) { | |
1672 | if (ret_bh && (*ret_bh)) { | |
1673 | brelse(*ret_bh); | |
1674 | *ret_bh = NULL; | |
1675 | } | |
1676 | if (acquired) | |
1677 | ocfs2_meta_unlock(inode, ex); | |
1678 | } | |
1679 | ||
1680 | if (local_bh) | |
1681 | brelse(local_bh); | |
1682 | ||
1683 | mlog_exit(status); | |
1684 | return status; | |
1685 | } | |
1686 | ||
1687 | /* | |
1688 | * This is working around a lock inversion between tasks acquiring DLM locks | |
1689 | * while holding a page lock and the vote thread which blocks dlm lock acquiry | |
1690 | * while acquiring page locks. | |
1691 | * | |
1692 | * ** These _with_page variantes are only intended to be called from aop | |
1693 | * methods that hold page locks and return a very specific *positive* error | |
1694 | * code that aop methods pass up to the VFS -- test for errors with != 0. ** | |
1695 | * | |
1696 | * The DLM is called such that it returns -EAGAIN if it would have blocked | |
1697 | * waiting for the vote thread. In that case we unlock our page so the vote | |
1698 | * thread can make progress. Once we've done this we have to return | |
1699 | * AOP_TRUNCATED_PAGE so the aop method that called us can bubble that back up | |
1700 | * into the VFS who will then immediately retry the aop call. | |
1701 | * | |
1702 | * We do a blocking lock and immediate unlock before returning, though, so that | |
1703 | * the lock has a great chance of being cached on this node by the time the VFS | |
1704 | * calls back to retry the aop. This has a potential to livelock as nodes | |
1705 | * ping locks back and forth, but that's a risk we're willing to take to avoid | |
1706 | * the lock inversion simply. | |
1707 | */ | |
1708 | int ocfs2_meta_lock_with_page(struct inode *inode, | |
ccd979bd MF |
1709 | struct buffer_head **ret_bh, |
1710 | int ex, | |
1711 | struct page *page) | |
1712 | { | |
1713 | int ret; | |
1714 | ||
4bcec184 | 1715 | ret = ocfs2_meta_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK); |
ccd979bd MF |
1716 | if (ret == -EAGAIN) { |
1717 | unlock_page(page); | |
4bcec184 | 1718 | if (ocfs2_meta_lock(inode, ret_bh, ex) == 0) |
ccd979bd MF |
1719 | ocfs2_meta_unlock(inode, ex); |
1720 | ret = AOP_TRUNCATED_PAGE; | |
1721 | } | |
1722 | ||
1723 | return ret; | |
1724 | } | |
1725 | ||
1726 | void ocfs2_meta_unlock(struct inode *inode, | |
1727 | int ex) | |
1728 | { | |
1729 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1730 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres; | |
1731 | ||
1732 | mlog_entry_void(); | |
1733 | ||
b0697053 MF |
1734 | mlog(0, "inode %llu drop %s META lock\n", |
1735 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1736 | ex ? "EXMODE" : "PRMODE"); |
1737 | ||
1738 | if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) | |
1739 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
1740 | ||
1741 | mlog_exit_void(); | |
1742 | } | |
1743 | ||
1744 | int ocfs2_super_lock(struct ocfs2_super *osb, | |
1745 | int ex) | |
1746 | { | |
1747 | int status; | |
1748 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1749 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; | |
1750 | struct buffer_head *bh; | |
1751 | struct ocfs2_slot_info *si = osb->slot_info; | |
1752 | ||
1753 | mlog_entry_void(); | |
1754 | ||
1755 | if (ocfs2_is_hard_readonly(osb)) | |
1756 | return -EROFS; | |
1757 | ||
1758 | status = ocfs2_cluster_lock(osb, lockres, level, 0, 0); | |
1759 | if (status < 0) { | |
1760 | mlog_errno(status); | |
1761 | goto bail; | |
1762 | } | |
1763 | ||
1764 | /* The super block lock path is really in the best position to | |
1765 | * know when resources covered by the lock need to be | |
1766 | * refreshed, so we do it here. Of course, making sense of | |
1767 | * everything is up to the caller :) */ | |
1768 | status = ocfs2_should_refresh_lock_res(lockres); | |
1769 | if (status < 0) { | |
1770 | mlog_errno(status); | |
1771 | goto bail; | |
1772 | } | |
1773 | if (status) { | |
1774 | bh = si->si_bh; | |
1775 | status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0, | |
1776 | si->si_inode); | |
1777 | if (status == 0) | |
1778 | ocfs2_update_slot_info(si); | |
1779 | ||
1780 | ocfs2_complete_lock_res_refresh(lockres, status); | |
1781 | ||
1782 | if (status < 0) | |
1783 | mlog_errno(status); | |
1784 | } | |
1785 | bail: | |
1786 | mlog_exit(status); | |
1787 | return status; | |
1788 | } | |
1789 | ||
1790 | void ocfs2_super_unlock(struct ocfs2_super *osb, | |
1791 | int ex) | |
1792 | { | |
1793 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1794 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; | |
1795 | ||
1796 | ocfs2_cluster_unlock(osb, lockres, level); | |
1797 | } | |
1798 | ||
1799 | int ocfs2_rename_lock(struct ocfs2_super *osb) | |
1800 | { | |
1801 | int status; | |
1802 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; | |
1803 | ||
1804 | if (ocfs2_is_hard_readonly(osb)) | |
1805 | return -EROFS; | |
1806 | ||
1807 | status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, 0); | |
1808 | if (status < 0) | |
1809 | mlog_errno(status); | |
1810 | ||
1811 | return status; | |
1812 | } | |
1813 | ||
1814 | void ocfs2_rename_unlock(struct ocfs2_super *osb) | |
1815 | { | |
1816 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; | |
1817 | ||
1818 | ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE); | |
1819 | } | |
1820 | ||
d680efe9 MF |
1821 | int ocfs2_dentry_lock(struct dentry *dentry, int ex) |
1822 | { | |
1823 | int ret; | |
1824 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1825 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; | |
1826 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); | |
1827 | ||
1828 | BUG_ON(!dl); | |
1829 | ||
1830 | if (ocfs2_is_hard_readonly(osb)) | |
1831 | return -EROFS; | |
1832 | ||
1833 | ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0); | |
1834 | if (ret < 0) | |
1835 | mlog_errno(ret); | |
1836 | ||
1837 | return ret; | |
1838 | } | |
1839 | ||
1840 | void ocfs2_dentry_unlock(struct dentry *dentry, int ex) | |
1841 | { | |
1842 | int level = ex ? LKM_EXMODE : LKM_PRMODE; | |
1843 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; | |
1844 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); | |
1845 | ||
1846 | ocfs2_cluster_unlock(osb, &dl->dl_lockres, level); | |
1847 | } | |
1848 | ||
ccd979bd MF |
1849 | /* Reference counting of the dlm debug structure. We want this because |
1850 | * open references on the debug inodes can live on after a mount, so | |
1851 | * we can't rely on the ocfs2_super to always exist. */ | |
1852 | static void ocfs2_dlm_debug_free(struct kref *kref) | |
1853 | { | |
1854 | struct ocfs2_dlm_debug *dlm_debug; | |
1855 | ||
1856 | dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt); | |
1857 | ||
1858 | kfree(dlm_debug); | |
1859 | } | |
1860 | ||
1861 | void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug) | |
1862 | { | |
1863 | if (dlm_debug) | |
1864 | kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free); | |
1865 | } | |
1866 | ||
1867 | static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug) | |
1868 | { | |
1869 | kref_get(&debug->d_refcnt); | |
1870 | } | |
1871 | ||
1872 | struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void) | |
1873 | { | |
1874 | struct ocfs2_dlm_debug *dlm_debug; | |
1875 | ||
1876 | dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL); | |
1877 | if (!dlm_debug) { | |
1878 | mlog_errno(-ENOMEM); | |
1879 | goto out; | |
1880 | } | |
1881 | ||
1882 | kref_init(&dlm_debug->d_refcnt); | |
1883 | INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking); | |
1884 | dlm_debug->d_locking_state = NULL; | |
1885 | out: | |
1886 | return dlm_debug; | |
1887 | } | |
1888 | ||
1889 | /* Access to this is arbitrated for us via seq_file->sem. */ | |
1890 | struct ocfs2_dlm_seq_priv { | |
1891 | struct ocfs2_dlm_debug *p_dlm_debug; | |
1892 | struct ocfs2_lock_res p_iter_res; | |
1893 | struct ocfs2_lock_res p_tmp_res; | |
1894 | }; | |
1895 | ||
1896 | static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start, | |
1897 | struct ocfs2_dlm_seq_priv *priv) | |
1898 | { | |
1899 | struct ocfs2_lock_res *iter, *ret = NULL; | |
1900 | struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug; | |
1901 | ||
1902 | assert_spin_locked(&ocfs2_dlm_tracking_lock); | |
1903 | ||
1904 | list_for_each_entry(iter, &start->l_debug_list, l_debug_list) { | |
1905 | /* discover the head of the list */ | |
1906 | if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) { | |
1907 | mlog(0, "End of list found, %p\n", ret); | |
1908 | break; | |
1909 | } | |
1910 | ||
1911 | /* We track our "dummy" iteration lockres' by a NULL | |
1912 | * l_ops field. */ | |
1913 | if (iter->l_ops != NULL) { | |
1914 | ret = iter; | |
1915 | break; | |
1916 | } | |
1917 | } | |
1918 | ||
1919 | return ret; | |
1920 | } | |
1921 | ||
1922 | static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos) | |
1923 | { | |
1924 | struct ocfs2_dlm_seq_priv *priv = m->private; | |
1925 | struct ocfs2_lock_res *iter; | |
1926 | ||
1927 | spin_lock(&ocfs2_dlm_tracking_lock); | |
1928 | iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv); | |
1929 | if (iter) { | |
1930 | /* Since lockres' have the lifetime of their container | |
1931 | * (which can be inodes, ocfs2_supers, etc) we want to | |
1932 | * copy this out to a temporary lockres while still | |
1933 | * under the spinlock. Obviously after this we can't | |
1934 | * trust any pointers on the copy returned, but that's | |
1935 | * ok as the information we want isn't typically held | |
1936 | * in them. */ | |
1937 | priv->p_tmp_res = *iter; | |
1938 | iter = &priv->p_tmp_res; | |
1939 | } | |
1940 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
1941 | ||
1942 | return iter; | |
1943 | } | |
1944 | ||
1945 | static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v) | |
1946 | { | |
1947 | } | |
1948 | ||
1949 | static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
1950 | { | |
1951 | struct ocfs2_dlm_seq_priv *priv = m->private; | |
1952 | struct ocfs2_lock_res *iter = v; | |
1953 | struct ocfs2_lock_res *dummy = &priv->p_iter_res; | |
1954 | ||
1955 | spin_lock(&ocfs2_dlm_tracking_lock); | |
1956 | iter = ocfs2_dlm_next_res(iter, priv); | |
1957 | list_del_init(&dummy->l_debug_list); | |
1958 | if (iter) { | |
1959 | list_add(&dummy->l_debug_list, &iter->l_debug_list); | |
1960 | priv->p_tmp_res = *iter; | |
1961 | iter = &priv->p_tmp_res; | |
1962 | } | |
1963 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
1964 | ||
1965 | return iter; | |
1966 | } | |
1967 | ||
1968 | /* So that debugfs.ocfs2 can determine which format is being used */ | |
1969 | #define OCFS2_DLM_DEBUG_STR_VERSION 1 | |
1970 | static int ocfs2_dlm_seq_show(struct seq_file *m, void *v) | |
1971 | { | |
1972 | int i; | |
1973 | char *lvb; | |
1974 | struct ocfs2_lock_res *lockres = v; | |
1975 | ||
1976 | if (!lockres) | |
1977 | return -EINVAL; | |
1978 | ||
d680efe9 MF |
1979 | seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION); |
1980 | ||
1981 | if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY) | |
1982 | seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1, | |
1983 | lockres->l_name, | |
1984 | (unsigned int)ocfs2_get_dentry_lock_ino(lockres)); | |
1985 | else | |
1986 | seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name); | |
1987 | ||
1988 | seq_printf(m, "%d\t" | |
ccd979bd MF |
1989 | "0x%lx\t" |
1990 | "0x%x\t" | |
1991 | "0x%x\t" | |
1992 | "%u\t" | |
1993 | "%u\t" | |
1994 | "%d\t" | |
1995 | "%d\t", | |
ccd979bd MF |
1996 | lockres->l_level, |
1997 | lockres->l_flags, | |
1998 | lockres->l_action, | |
1999 | lockres->l_unlock_action, | |
2000 | lockres->l_ro_holders, | |
2001 | lockres->l_ex_holders, | |
2002 | lockres->l_requested, | |
2003 | lockres->l_blocking); | |
2004 | ||
2005 | /* Dump the raw LVB */ | |
2006 | lvb = lockres->l_lksb.lvb; | |
2007 | for(i = 0; i < DLM_LVB_LEN; i++) | |
2008 | seq_printf(m, "0x%x\t", lvb[i]); | |
2009 | ||
2010 | /* End the line */ | |
2011 | seq_printf(m, "\n"); | |
2012 | return 0; | |
2013 | } | |
2014 | ||
2015 | static struct seq_operations ocfs2_dlm_seq_ops = { | |
2016 | .start = ocfs2_dlm_seq_start, | |
2017 | .stop = ocfs2_dlm_seq_stop, | |
2018 | .next = ocfs2_dlm_seq_next, | |
2019 | .show = ocfs2_dlm_seq_show, | |
2020 | }; | |
2021 | ||
2022 | static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file) | |
2023 | { | |
2024 | struct seq_file *seq = (struct seq_file *) file->private_data; | |
2025 | struct ocfs2_dlm_seq_priv *priv = seq->private; | |
2026 | struct ocfs2_lock_res *res = &priv->p_iter_res; | |
2027 | ||
2028 | ocfs2_remove_lockres_tracking(res); | |
2029 | ocfs2_put_dlm_debug(priv->p_dlm_debug); | |
2030 | return seq_release_private(inode, file); | |
2031 | } | |
2032 | ||
2033 | static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file) | |
2034 | { | |
2035 | int ret; | |
2036 | struct ocfs2_dlm_seq_priv *priv; | |
2037 | struct seq_file *seq; | |
2038 | struct ocfs2_super *osb; | |
2039 | ||
2040 | priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL); | |
2041 | if (!priv) { | |
2042 | ret = -ENOMEM; | |
2043 | mlog_errno(ret); | |
2044 | goto out; | |
2045 | } | |
8e18e294 | 2046 | osb = inode->i_private; |
ccd979bd MF |
2047 | ocfs2_get_dlm_debug(osb->osb_dlm_debug); |
2048 | priv->p_dlm_debug = osb->osb_dlm_debug; | |
2049 | INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list); | |
2050 | ||
2051 | ret = seq_open(file, &ocfs2_dlm_seq_ops); | |
2052 | if (ret) { | |
2053 | kfree(priv); | |
2054 | mlog_errno(ret); | |
2055 | goto out; | |
2056 | } | |
2057 | ||
2058 | seq = (struct seq_file *) file->private_data; | |
2059 | seq->private = priv; | |
2060 | ||
2061 | ocfs2_add_lockres_tracking(&priv->p_iter_res, | |
2062 | priv->p_dlm_debug); | |
2063 | ||
2064 | out: | |
2065 | return ret; | |
2066 | } | |
2067 | ||
4b6f5d20 | 2068 | static const struct file_operations ocfs2_dlm_debug_fops = { |
ccd979bd MF |
2069 | .open = ocfs2_dlm_debug_open, |
2070 | .release = ocfs2_dlm_debug_release, | |
2071 | .read = seq_read, | |
2072 | .llseek = seq_lseek, | |
2073 | }; | |
2074 | ||
2075 | static int ocfs2_dlm_init_debug(struct ocfs2_super *osb) | |
2076 | { | |
2077 | int ret = 0; | |
2078 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; | |
2079 | ||
2080 | dlm_debug->d_locking_state = debugfs_create_file("locking_state", | |
2081 | S_IFREG|S_IRUSR, | |
2082 | osb->osb_debug_root, | |
2083 | osb, | |
2084 | &ocfs2_dlm_debug_fops); | |
2085 | if (!dlm_debug->d_locking_state) { | |
2086 | ret = -EINVAL; | |
2087 | mlog(ML_ERROR, | |
2088 | "Unable to create locking state debugfs file.\n"); | |
2089 | goto out; | |
2090 | } | |
2091 | ||
2092 | ocfs2_get_dlm_debug(dlm_debug); | |
2093 | out: | |
2094 | return ret; | |
2095 | } | |
2096 | ||
2097 | static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb) | |
2098 | { | |
2099 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; | |
2100 | ||
2101 | if (dlm_debug) { | |
2102 | debugfs_remove(dlm_debug->d_locking_state); | |
2103 | ocfs2_put_dlm_debug(dlm_debug); | |
2104 | } | |
2105 | } | |
2106 | ||
2107 | int ocfs2_dlm_init(struct ocfs2_super *osb) | |
2108 | { | |
2109 | int status; | |
2110 | u32 dlm_key; | |
2111 | struct dlm_ctxt *dlm; | |
2112 | ||
2113 | mlog_entry_void(); | |
2114 | ||
2115 | status = ocfs2_dlm_init_debug(osb); | |
2116 | if (status < 0) { | |
2117 | mlog_errno(status); | |
2118 | goto bail; | |
2119 | } | |
2120 | ||
2121 | /* launch vote thread */ | |
78427043 | 2122 | osb->vote_task = kthread_run(ocfs2_vote_thread, osb, "ocfs2vote"); |
ccd979bd MF |
2123 | if (IS_ERR(osb->vote_task)) { |
2124 | status = PTR_ERR(osb->vote_task); | |
2125 | osb->vote_task = NULL; | |
2126 | mlog_errno(status); | |
2127 | goto bail; | |
2128 | } | |
2129 | ||
2130 | /* used by the dlm code to make message headers unique, each | |
2131 | * node in this domain must agree on this. */ | |
2132 | dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str)); | |
2133 | ||
2134 | /* for now, uuid == domain */ | |
2135 | dlm = dlm_register_domain(osb->uuid_str, dlm_key); | |
2136 | if (IS_ERR(dlm)) { | |
2137 | status = PTR_ERR(dlm); | |
2138 | mlog_errno(status); | |
2139 | goto bail; | |
2140 | } | |
2141 | ||
2142 | ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb); | |
2143 | ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb); | |
2144 | ||
2145 | dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb); | |
2146 | ||
2147 | osb->dlm = dlm; | |
2148 | ||
2149 | status = 0; | |
2150 | bail: | |
2151 | if (status < 0) { | |
2152 | ocfs2_dlm_shutdown_debug(osb); | |
2153 | if (osb->vote_task) | |
2154 | kthread_stop(osb->vote_task); | |
2155 | } | |
2156 | ||
2157 | mlog_exit(status); | |
2158 | return status; | |
2159 | } | |
2160 | ||
2161 | void ocfs2_dlm_shutdown(struct ocfs2_super *osb) | |
2162 | { | |
2163 | mlog_entry_void(); | |
2164 | ||
2165 | dlm_unregister_eviction_cb(&osb->osb_eviction_cb); | |
2166 | ||
2167 | ocfs2_drop_osb_locks(osb); | |
2168 | ||
2169 | if (osb->vote_task) { | |
2170 | kthread_stop(osb->vote_task); | |
2171 | osb->vote_task = NULL; | |
2172 | } | |
2173 | ||
2174 | ocfs2_lock_res_free(&osb->osb_super_lockres); | |
2175 | ocfs2_lock_res_free(&osb->osb_rename_lockres); | |
2176 | ||
2177 | dlm_unregister_domain(osb->dlm); | |
2178 | osb->dlm = NULL; | |
2179 | ||
2180 | ocfs2_dlm_shutdown_debug(osb); | |
2181 | ||
2182 | mlog_exit_void(); | |
2183 | } | |
2184 | ||
2a45f2d1 | 2185 | static void ocfs2_unlock_ast(void *opaque, enum dlm_status status) |
ccd979bd MF |
2186 | { |
2187 | struct ocfs2_lock_res *lockres = opaque; | |
2188 | unsigned long flags; | |
2189 | ||
2190 | mlog_entry_void(); | |
2191 | ||
2192 | mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name, | |
2193 | lockres->l_unlock_action); | |
2194 | ||
2195 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2196 | /* We tried to cancel a convert request, but it was already | |
2197 | * granted. All we want to do here is clear our unlock | |
2198 | * state. The wake_up call done at the bottom is redundant | |
2199 | * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't | |
2200 | * hurt anything anyway */ | |
2201 | if (status == DLM_CANCELGRANT && | |
2202 | lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { | |
2203 | mlog(0, "Got cancelgrant for %s\n", lockres->l_name); | |
2204 | ||
2205 | /* We don't clear the busy flag in this case as it | |
2206 | * should have been cleared by the ast which the dlm | |
2207 | * has called. */ | |
2208 | goto complete_unlock; | |
2209 | } | |
2210 | ||
2211 | if (status != DLM_NORMAL) { | |
2212 | mlog(ML_ERROR, "Dlm passes status %d for lock %s, " | |
2213 | "unlock_action %d\n", status, lockres->l_name, | |
2214 | lockres->l_unlock_action); | |
2215 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2216 | return; | |
2217 | } | |
2218 | ||
2219 | switch(lockres->l_unlock_action) { | |
2220 | case OCFS2_UNLOCK_CANCEL_CONVERT: | |
2221 | mlog(0, "Cancel convert success for %s\n", lockres->l_name); | |
2222 | lockres->l_action = OCFS2_AST_INVALID; | |
2223 | break; | |
2224 | case OCFS2_UNLOCK_DROP_LOCK: | |
2225 | lockres->l_level = LKM_IVMODE; | |
2226 | break; | |
2227 | default: | |
2228 | BUG(); | |
2229 | } | |
2230 | ||
2231 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
2232 | complete_unlock: | |
2233 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
2234 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2235 | ||
2236 | wake_up(&lockres->l_event); | |
2237 | ||
2238 | mlog_exit_void(); | |
2239 | } | |
2240 | ||
ccd979bd | 2241 | static int ocfs2_drop_lock(struct ocfs2_super *osb, |
0d5dc6c2 | 2242 | struct ocfs2_lock_res *lockres) |
ccd979bd MF |
2243 | { |
2244 | enum dlm_status status; | |
2245 | unsigned long flags; | |
b80fc012 | 2246 | int lkm_flags = 0; |
ccd979bd MF |
2247 | |
2248 | /* We didn't get anywhere near actually using this lockres. */ | |
2249 | if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) | |
2250 | goto out; | |
2251 | ||
b80fc012 MF |
2252 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) |
2253 | lkm_flags |= LKM_VALBLK; | |
2254 | ||
ccd979bd MF |
2255 | spin_lock_irqsave(&lockres->l_lock, flags); |
2256 | ||
2257 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING), | |
2258 | "lockres %s, flags 0x%lx\n", | |
2259 | lockres->l_name, lockres->l_flags); | |
2260 | ||
2261 | while (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
2262 | mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = " | |
2263 | "%u, unlock_action = %u\n", | |
2264 | lockres->l_name, lockres->l_flags, lockres->l_action, | |
2265 | lockres->l_unlock_action); | |
2266 | ||
2267 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2268 | ||
2269 | /* XXX: Today we just wait on any busy | |
2270 | * locks... Perhaps we need to cancel converts in the | |
2271 | * future? */ | |
2272 | ocfs2_wait_on_busy_lock(lockres); | |
2273 | ||
2274 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2275 | } | |
2276 | ||
0d5dc6c2 MF |
2277 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) { |
2278 | if (lockres->l_flags & OCFS2_LOCK_ATTACHED && | |
2279 | lockres->l_level == LKM_EXMODE && | |
2280 | !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) | |
2281 | lockres->l_ops->set_lvb(lockres); | |
2282 | } | |
ccd979bd MF |
2283 | |
2284 | if (lockres->l_flags & OCFS2_LOCK_BUSY) | |
2285 | mlog(ML_ERROR, "destroying busy lock: \"%s\"\n", | |
2286 | lockres->l_name); | |
2287 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) | |
2288 | mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name); | |
2289 | ||
2290 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { | |
2291 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2292 | goto out; | |
2293 | } | |
2294 | ||
2295 | lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED); | |
2296 | ||
2297 | /* make sure we never get here while waiting for an ast to | |
2298 | * fire. */ | |
2299 | BUG_ON(lockres->l_action != OCFS2_AST_INVALID); | |
2300 | ||
2301 | /* is this necessary? */ | |
2302 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
2303 | lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK; | |
2304 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2305 | ||
2306 | mlog(0, "lock %s\n", lockres->l_name); | |
2307 | ||
b80fc012 | 2308 | status = dlmunlock(osb->dlm, &lockres->l_lksb, lkm_flags, |
2a45f2d1 | 2309 | ocfs2_unlock_ast, lockres); |
ccd979bd MF |
2310 | if (status != DLM_NORMAL) { |
2311 | ocfs2_log_dlm_error("dlmunlock", status, lockres); | |
2312 | mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags); | |
2313 | dlm_print_one_lock(lockres->l_lksb.lockid); | |
2314 | BUG(); | |
2315 | } | |
2316 | mlog(0, "lock %s, successfull return from dlmunlock\n", | |
2317 | lockres->l_name); | |
2318 | ||
2319 | ocfs2_wait_on_busy_lock(lockres); | |
2320 | out: | |
2321 | mlog_exit(0); | |
2322 | return 0; | |
2323 | } | |
2324 | ||
2325 | /* Mark the lockres as being dropped. It will no longer be | |
2326 | * queued if blocking, but we still may have to wait on it | |
2327 | * being dequeued from the vote thread before we can consider | |
2328 | * it safe to drop. | |
2329 | * | |
2330 | * You can *not* attempt to call cluster_lock on this lockres anymore. */ | |
2331 | void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres) | |
2332 | { | |
2333 | int status; | |
2334 | struct ocfs2_mask_waiter mw; | |
2335 | unsigned long flags; | |
2336 | ||
2337 | ocfs2_init_mask_waiter(&mw); | |
2338 | ||
2339 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2340 | lockres->l_flags |= OCFS2_LOCK_FREEING; | |
2341 | while (lockres->l_flags & OCFS2_LOCK_QUEUED) { | |
2342 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0); | |
2343 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2344 | ||
2345 | mlog(0, "Waiting on lockres %s\n", lockres->l_name); | |
2346 | ||
2347 | status = ocfs2_wait_for_mask(&mw); | |
2348 | if (status) | |
2349 | mlog_errno(status); | |
2350 | ||
2351 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2352 | } | |
2353 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2354 | } | |
2355 | ||
d680efe9 MF |
2356 | void ocfs2_simple_drop_lockres(struct ocfs2_super *osb, |
2357 | struct ocfs2_lock_res *lockres) | |
ccd979bd | 2358 | { |
d680efe9 | 2359 | int ret; |
ccd979bd | 2360 | |
d680efe9 | 2361 | ocfs2_mark_lockres_freeing(lockres); |
0d5dc6c2 | 2362 | ret = ocfs2_drop_lock(osb, lockres); |
d680efe9 MF |
2363 | if (ret) |
2364 | mlog_errno(ret); | |
2365 | } | |
ccd979bd | 2366 | |
d680efe9 MF |
2367 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb) |
2368 | { | |
2369 | ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres); | |
2370 | ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres); | |
ccd979bd MF |
2371 | } |
2372 | ||
ccd979bd MF |
2373 | int ocfs2_drop_inode_locks(struct inode *inode) |
2374 | { | |
2375 | int status, err; | |
ccd979bd MF |
2376 | |
2377 | mlog_entry_void(); | |
2378 | ||
2379 | /* No need to call ocfs2_mark_lockres_freeing here - | |
2380 | * ocfs2_clear_inode has done it for us. */ | |
2381 | ||
2382 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
0d5dc6c2 | 2383 | &OCFS2_I(inode)->ip_data_lockres); |
ccd979bd MF |
2384 | if (err < 0) |
2385 | mlog_errno(err); | |
2386 | ||
2387 | status = err; | |
2388 | ||
2389 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
0d5dc6c2 | 2390 | &OCFS2_I(inode)->ip_meta_lockres); |
ccd979bd MF |
2391 | if (err < 0) |
2392 | mlog_errno(err); | |
2393 | if (err < 0 && !status) | |
2394 | status = err; | |
2395 | ||
2396 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
0d5dc6c2 | 2397 | &OCFS2_I(inode)->ip_rw_lockres); |
ccd979bd MF |
2398 | if (err < 0) |
2399 | mlog_errno(err); | |
2400 | if (err < 0 && !status) | |
2401 | status = err; | |
2402 | ||
2403 | mlog_exit(status); | |
2404 | return status; | |
2405 | } | |
2406 | ||
2407 | static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, | |
2408 | int new_level) | |
2409 | { | |
2410 | assert_spin_locked(&lockres->l_lock); | |
2411 | ||
2412 | BUG_ON(lockres->l_blocking <= LKM_NLMODE); | |
2413 | ||
2414 | if (lockres->l_level <= new_level) { | |
2415 | mlog(ML_ERROR, "lockres->l_level (%u) <= new_level (%u)\n", | |
2416 | lockres->l_level, new_level); | |
2417 | BUG(); | |
2418 | } | |
2419 | ||
2420 | mlog(0, "lock %s, new_level = %d, l_blocking = %d\n", | |
2421 | lockres->l_name, new_level, lockres->l_blocking); | |
2422 | ||
2423 | lockres->l_action = OCFS2_AST_DOWNCONVERT; | |
2424 | lockres->l_requested = new_level; | |
2425 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
2426 | } | |
2427 | ||
2428 | static int ocfs2_downconvert_lock(struct ocfs2_super *osb, | |
2429 | struct ocfs2_lock_res *lockres, | |
2430 | int new_level, | |
2431 | int lvb) | |
2432 | { | |
2433 | int ret, dlm_flags = LKM_CONVERT; | |
2434 | enum dlm_status status; | |
2435 | ||
2436 | mlog_entry_void(); | |
2437 | ||
2438 | if (lvb) | |
2439 | dlm_flags |= LKM_VALBLK; | |
2440 | ||
2441 | status = dlmlock(osb->dlm, | |
2442 | new_level, | |
2443 | &lockres->l_lksb, | |
2444 | dlm_flags, | |
2445 | lockres->l_name, | |
f0681062 | 2446 | OCFS2_LOCK_ID_MAX_LEN - 1, |
e92d57df | 2447 | ocfs2_locking_ast, |
ccd979bd | 2448 | lockres, |
aa2623ad | 2449 | ocfs2_blocking_ast); |
ccd979bd MF |
2450 | if (status != DLM_NORMAL) { |
2451 | ocfs2_log_dlm_error("dlmlock", status, lockres); | |
2452 | ret = -EINVAL; | |
2453 | ocfs2_recover_from_dlm_error(lockres, 1); | |
2454 | goto bail; | |
2455 | } | |
2456 | ||
2457 | ret = 0; | |
2458 | bail: | |
2459 | mlog_exit(ret); | |
2460 | return ret; | |
2461 | } | |
2462 | ||
2463 | /* returns 1 when the caller should unlock and call dlmunlock */ | |
2464 | static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, | |
2465 | struct ocfs2_lock_res *lockres) | |
2466 | { | |
2467 | assert_spin_locked(&lockres->l_lock); | |
2468 | ||
2469 | mlog_entry_void(); | |
2470 | mlog(0, "lock %s\n", lockres->l_name); | |
2471 | ||
2472 | if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { | |
2473 | /* If we're already trying to cancel a lock conversion | |
2474 | * then just drop the spinlock and allow the caller to | |
2475 | * requeue this lock. */ | |
2476 | ||
2477 | mlog(0, "Lockres %s, skip convert\n", lockres->l_name); | |
2478 | return 0; | |
2479 | } | |
2480 | ||
2481 | /* were we in a convert when we got the bast fire? */ | |
2482 | BUG_ON(lockres->l_action != OCFS2_AST_CONVERT && | |
2483 | lockres->l_action != OCFS2_AST_DOWNCONVERT); | |
2484 | /* set things up for the unlockast to know to just | |
2485 | * clear out the ast_action and unset busy, etc. */ | |
2486 | lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT; | |
2487 | ||
2488 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY), | |
2489 | "lock %s, invalid flags: 0x%lx\n", | |
2490 | lockres->l_name, lockres->l_flags); | |
2491 | ||
2492 | return 1; | |
2493 | } | |
2494 | ||
2495 | static int ocfs2_cancel_convert(struct ocfs2_super *osb, | |
2496 | struct ocfs2_lock_res *lockres) | |
2497 | { | |
2498 | int ret; | |
2499 | enum dlm_status status; | |
2500 | ||
2501 | mlog_entry_void(); | |
2502 | mlog(0, "lock %s\n", lockres->l_name); | |
2503 | ||
2504 | ret = 0; | |
2505 | status = dlmunlock(osb->dlm, | |
2506 | &lockres->l_lksb, | |
2507 | LKM_CANCEL, | |
2a45f2d1 | 2508 | ocfs2_unlock_ast, |
ccd979bd MF |
2509 | lockres); |
2510 | if (status != DLM_NORMAL) { | |
2511 | ocfs2_log_dlm_error("dlmunlock", status, lockres); | |
2512 | ret = -EINVAL; | |
2513 | ocfs2_recover_from_dlm_error(lockres, 0); | |
2514 | } | |
2515 | ||
2516 | mlog(0, "lock %s return from dlmunlock\n", lockres->l_name); | |
2517 | ||
ccd979bd MF |
2518 | mlog_exit(ret); |
2519 | return ret; | |
2520 | } | |
2521 | ||
b5e500e2 MF |
2522 | static int ocfs2_unblock_lock(struct ocfs2_super *osb, |
2523 | struct ocfs2_lock_res *lockres, | |
2524 | struct ocfs2_unblock_ctl *ctl) | |
ccd979bd MF |
2525 | { |
2526 | unsigned long flags; | |
2527 | int blocking; | |
2528 | int new_level; | |
2529 | int ret = 0; | |
5ef0d4ea | 2530 | int set_lvb = 0; |
ccd979bd MF |
2531 | |
2532 | mlog_entry_void(); | |
2533 | ||
2534 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2535 | ||
2536 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
2537 | ||
2538 | recheck: | |
2539 | if (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
d680efe9 | 2540 | ctl->requeue = 1; |
ccd979bd MF |
2541 | ret = ocfs2_prepare_cancel_convert(osb, lockres); |
2542 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2543 | if (ret) { | |
2544 | ret = ocfs2_cancel_convert(osb, lockres); | |
2545 | if (ret < 0) | |
2546 | mlog_errno(ret); | |
2547 | } | |
2548 | goto leave; | |
2549 | } | |
2550 | ||
2551 | /* if we're blocking an exclusive and we have *any* holders, | |
2552 | * then requeue. */ | |
2553 | if ((lockres->l_blocking == LKM_EXMODE) | |
f7fbfdd1 MF |
2554 | && (lockres->l_ex_holders || lockres->l_ro_holders)) |
2555 | goto leave_requeue; | |
ccd979bd MF |
2556 | |
2557 | /* If it's a PR we're blocking, then only | |
2558 | * requeue if we've got any EX holders */ | |
2559 | if (lockres->l_blocking == LKM_PRMODE && | |
f7fbfdd1 MF |
2560 | lockres->l_ex_holders) |
2561 | goto leave_requeue; | |
2562 | ||
2563 | /* | |
2564 | * Can we get a lock in this state if the holder counts are | |
2565 | * zero? The meta data unblock code used to check this. | |
2566 | */ | |
2567 | if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) | |
2568 | && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) | |
2569 | goto leave_requeue; | |
ccd979bd | 2570 | |
16d5b956 MF |
2571 | new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking); |
2572 | ||
2573 | if (lockres->l_ops->check_downconvert | |
2574 | && !lockres->l_ops->check_downconvert(lockres, new_level)) | |
2575 | goto leave_requeue; | |
2576 | ||
ccd979bd MF |
2577 | /* If we get here, then we know that there are no more |
2578 | * incompatible holders (and anyone asking for an incompatible | |
2579 | * lock is blocked). We can now downconvert the lock */ | |
cc567d89 | 2580 | if (!lockres->l_ops->downconvert_worker) |
ccd979bd MF |
2581 | goto downconvert; |
2582 | ||
2583 | /* Some lockres types want to do a bit of work before | |
2584 | * downconverting a lock. Allow that here. The worker function | |
2585 | * may sleep, so we save off a copy of what we're blocking as | |
2586 | * it may change while we're not holding the spin lock. */ | |
2587 | blocking = lockres->l_blocking; | |
2588 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2589 | ||
cc567d89 | 2590 | ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking); |
d680efe9 MF |
2591 | |
2592 | if (ctl->unblock_action == UNBLOCK_STOP_POST) | |
2593 | goto leave; | |
ccd979bd MF |
2594 | |
2595 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2596 | if (blocking != lockres->l_blocking) { | |
2597 | /* If this changed underneath us, then we can't drop | |
2598 | * it just yet. */ | |
2599 | goto recheck; | |
2600 | } | |
2601 | ||
2602 | downconvert: | |
d680efe9 | 2603 | ctl->requeue = 0; |
ccd979bd | 2604 | |
5ef0d4ea MF |
2605 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) { |
2606 | if (lockres->l_level == LKM_EXMODE) | |
2607 | set_lvb = 1; | |
2608 | ||
2609 | /* | |
2610 | * We only set the lvb if the lock has been fully | |
2611 | * refreshed - otherwise we risk setting stale | |
2612 | * data. Otherwise, there's no need to actually clear | |
2613 | * out the lvb here as it's value is still valid. | |
2614 | */ | |
2615 | if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) | |
2616 | lockres->l_ops->set_lvb(lockres); | |
2617 | } | |
2618 | ||
ccd979bd MF |
2619 | ocfs2_prepare_downconvert(lockres, new_level); |
2620 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
5ef0d4ea | 2621 | ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb); |
ccd979bd MF |
2622 | leave: |
2623 | mlog_exit(ret); | |
2624 | return ret; | |
f7fbfdd1 MF |
2625 | |
2626 | leave_requeue: | |
2627 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2628 | ctl->requeue = 1; | |
2629 | ||
2630 | mlog_exit(0); | |
2631 | return 0; | |
ccd979bd MF |
2632 | } |
2633 | ||
d680efe9 MF |
2634 | static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, |
2635 | int blocking) | |
ccd979bd MF |
2636 | { |
2637 | struct inode *inode; | |
2638 | struct address_space *mapping; | |
2639 | ||
ccd979bd MF |
2640 | inode = ocfs2_lock_res_inode(lockres); |
2641 | mapping = inode->i_mapping; | |
2642 | ||
2643 | if (filemap_fdatawrite(mapping)) { | |
b0697053 MF |
2644 | mlog(ML_ERROR, "Could not sync inode %llu for downconvert!", |
2645 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
2646 | } |
2647 | sync_mapping_buffers(mapping); | |
2648 | if (blocking == LKM_EXMODE) { | |
2649 | truncate_inode_pages(mapping, 0); | |
2650 | unmap_mapping_range(mapping, 0, 0, 0); | |
2651 | } else { | |
2652 | /* We only need to wait on the I/O if we're not also | |
2653 | * truncating pages because truncate_inode_pages waits | |
2654 | * for us above. We don't truncate pages if we're | |
2655 | * blocking anything < EXMODE because we want to keep | |
2656 | * them around in that case. */ | |
2657 | filemap_fdatawait(mapping); | |
2658 | } | |
2659 | ||
d680efe9 | 2660 | return UNBLOCK_CONTINUE; |
ccd979bd MF |
2661 | } |
2662 | ||
810d5aeb MF |
2663 | static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres, |
2664 | int new_level) | |
2665 | { | |
2666 | struct inode *inode = ocfs2_lock_res_inode(lockres); | |
2667 | int checkpointed = ocfs2_inode_fully_checkpointed(inode); | |
2668 | ||
2669 | BUG_ON(new_level != LKM_NLMODE && new_level != LKM_PRMODE); | |
2670 | BUG_ON(lockres->l_level != LKM_EXMODE && !checkpointed); | |
2671 | ||
2672 | if (checkpointed) | |
2673 | return 1; | |
2674 | ||
2675 | ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb)); | |
2676 | return 0; | |
2677 | } | |
2678 | ||
2679 | static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres) | |
2680 | { | |
2681 | struct inode *inode = ocfs2_lock_res_inode(lockres); | |
2682 | ||
2683 | __ocfs2_stuff_meta_lvb(inode); | |
2684 | } | |
2685 | ||
d680efe9 MF |
2686 | /* |
2687 | * Does the final reference drop on our dentry lock. Right now this | |
2688 | * happens in the vote thread, but we could choose to simplify the | |
2689 | * dlmglue API and push these off to the ocfs2_wq in the future. | |
2690 | */ | |
2691 | static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, | |
2692 | struct ocfs2_lock_res *lockres) | |
2693 | { | |
2694 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
2695 | ocfs2_dentry_lock_put(osb, dl); | |
2696 | } | |
2697 | ||
2698 | /* | |
2699 | * d_delete() matching dentries before the lock downconvert. | |
2700 | * | |
2701 | * At this point, any process waiting to destroy the | |
2702 | * dentry_lock due to last ref count is stopped by the | |
2703 | * OCFS2_LOCK_QUEUED flag. | |
2704 | * | |
2705 | * We have two potential problems | |
2706 | * | |
2707 | * 1) If we do the last reference drop on our dentry_lock (via dput) | |
2708 | * we'll wind up in ocfs2_release_dentry_lock(), waiting on | |
2709 | * the downconvert to finish. Instead we take an elevated | |
2710 | * reference and push the drop until after we've completed our | |
2711 | * unblock processing. | |
2712 | * | |
2713 | * 2) There might be another process with a final reference, | |
2714 | * waiting on us to finish processing. If this is the case, we | |
2715 | * detect it and exit out - there's no more dentries anyway. | |
2716 | */ | |
2717 | static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, | |
2718 | int blocking) | |
2719 | { | |
2720 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
2721 | struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode); | |
2722 | struct dentry *dentry; | |
2723 | unsigned long flags; | |
2724 | int extra_ref = 0; | |
2725 | ||
2726 | /* | |
2727 | * This node is blocking another node from getting a read | |
2728 | * lock. This happens when we've renamed within a | |
2729 | * directory. We've forced the other nodes to d_delete(), but | |
2730 | * we never actually dropped our lock because it's still | |
2731 | * valid. The downconvert code will retain a PR for this node, | |
2732 | * so there's no further work to do. | |
2733 | */ | |
2734 | if (blocking == LKM_PRMODE) | |
2735 | return UNBLOCK_CONTINUE; | |
2736 | ||
2737 | /* | |
2738 | * Mark this inode as potentially orphaned. The code in | |
2739 | * ocfs2_delete_inode() will figure out whether it actually | |
2740 | * needs to be freed or not. | |
2741 | */ | |
2742 | spin_lock(&oi->ip_lock); | |
2743 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
2744 | spin_unlock(&oi->ip_lock); | |
2745 | ||
2746 | /* | |
2747 | * Yuck. We need to make sure however that the check of | |
2748 | * OCFS2_LOCK_FREEING and the extra reference are atomic with | |
2749 | * respect to a reference decrement or the setting of that | |
2750 | * flag. | |
2751 | */ | |
2752 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2753 | spin_lock(&dentry_attach_lock); | |
2754 | if (!(lockres->l_flags & OCFS2_LOCK_FREEING) | |
2755 | && dl->dl_count) { | |
2756 | dl->dl_count++; | |
2757 | extra_ref = 1; | |
2758 | } | |
2759 | spin_unlock(&dentry_attach_lock); | |
2760 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2761 | ||
2762 | mlog(0, "extra_ref = %d\n", extra_ref); | |
2763 | ||
2764 | /* | |
2765 | * We have a process waiting on us in ocfs2_dentry_iput(), | |
2766 | * which means we can't have any more outstanding | |
2767 | * aliases. There's no need to do any more work. | |
2768 | */ | |
2769 | if (!extra_ref) | |
2770 | return UNBLOCK_CONTINUE; | |
2771 | ||
2772 | spin_lock(&dentry_attach_lock); | |
2773 | while (1) { | |
2774 | dentry = ocfs2_find_local_alias(dl->dl_inode, | |
2775 | dl->dl_parent_blkno, 1); | |
2776 | if (!dentry) | |
2777 | break; | |
2778 | spin_unlock(&dentry_attach_lock); | |
2779 | ||
2780 | mlog(0, "d_delete(%.*s);\n", dentry->d_name.len, | |
2781 | dentry->d_name.name); | |
2782 | ||
2783 | /* | |
2784 | * The following dcache calls may do an | |
2785 | * iput(). Normally we don't want that from the | |
2786 | * downconverting thread, but in this case it's ok | |
2787 | * because the requesting node already has an | |
2788 | * exclusive lock on the inode, so it can't be queued | |
2789 | * for a downconvert. | |
2790 | */ | |
2791 | d_delete(dentry); | |
2792 | dput(dentry); | |
2793 | ||
2794 | spin_lock(&dentry_attach_lock); | |
2795 | } | |
2796 | spin_unlock(&dentry_attach_lock); | |
2797 | ||
2798 | /* | |
2799 | * If we are the last holder of this dentry lock, there is no | |
2800 | * reason to downconvert so skip straight to the unlock. | |
2801 | */ | |
2802 | if (dl->dl_count == 1) | |
2803 | return UNBLOCK_STOP_POST; | |
2804 | ||
2805 | return UNBLOCK_CONTINUE_POST; | |
2806 | } | |
2807 | ||
ccd979bd MF |
2808 | void ocfs2_process_blocked_lock(struct ocfs2_super *osb, |
2809 | struct ocfs2_lock_res *lockres) | |
2810 | { | |
2811 | int status; | |
d680efe9 | 2812 | struct ocfs2_unblock_ctl ctl = {0, 0,}; |
ccd979bd MF |
2813 | unsigned long flags; |
2814 | ||
2815 | /* Our reference to the lockres in this function can be | |
2816 | * considered valid until we remove the OCFS2_LOCK_QUEUED | |
2817 | * flag. */ | |
2818 | ||
2819 | mlog_entry_void(); | |
2820 | ||
2821 | BUG_ON(!lockres); | |
2822 | BUG_ON(!lockres->l_ops); | |
ccd979bd MF |
2823 | |
2824 | mlog(0, "lockres %s blocked.\n", lockres->l_name); | |
2825 | ||
2826 | /* Detect whether a lock has been marked as going away while | |
2827 | * the vote thread was processing other things. A lock can | |
2828 | * still be marked with OCFS2_LOCK_FREEING after this check, | |
2829 | * but short circuiting here will still save us some | |
2830 | * performance. */ | |
2831 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2832 | if (lockres->l_flags & OCFS2_LOCK_FREEING) | |
2833 | goto unqueue; | |
2834 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2835 | ||
b5e500e2 | 2836 | status = ocfs2_unblock_lock(osb, lockres, &ctl); |
ccd979bd MF |
2837 | if (status < 0) |
2838 | mlog_errno(status); | |
2839 | ||
2840 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2841 | unqueue: | |
d680efe9 | 2842 | if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) { |
ccd979bd MF |
2843 | lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED); |
2844 | } else | |
2845 | ocfs2_schedule_blocked_lock(osb, lockres); | |
2846 | ||
2847 | mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name, | |
d680efe9 | 2848 | ctl.requeue ? "yes" : "no"); |
ccd979bd MF |
2849 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
2850 | ||
d680efe9 MF |
2851 | if (ctl.unblock_action != UNBLOCK_CONTINUE |
2852 | && lockres->l_ops->post_unlock) | |
2853 | lockres->l_ops->post_unlock(osb, lockres); | |
2854 | ||
ccd979bd MF |
2855 | mlog_exit_void(); |
2856 | } | |
2857 | ||
2858 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, | |
2859 | struct ocfs2_lock_res *lockres) | |
2860 | { | |
2861 | mlog_entry_void(); | |
2862 | ||
2863 | assert_spin_locked(&lockres->l_lock); | |
2864 | ||
2865 | if (lockres->l_flags & OCFS2_LOCK_FREEING) { | |
2866 | /* Do not schedule a lock for downconvert when it's on | |
2867 | * the way to destruction - any nodes wanting access | |
2868 | * to the resource will get it soon. */ | |
2869 | mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n", | |
2870 | lockres->l_name, lockres->l_flags); | |
2871 | return; | |
2872 | } | |
2873 | ||
2874 | lockres_or_flags(lockres, OCFS2_LOCK_QUEUED); | |
2875 | ||
2876 | spin_lock(&osb->vote_task_lock); | |
2877 | if (list_empty(&lockres->l_blocked_list)) { | |
2878 | list_add_tail(&lockres->l_blocked_list, | |
2879 | &osb->blocked_lock_list); | |
2880 | osb->blocked_lock_count++; | |
2881 | } | |
2882 | spin_unlock(&osb->vote_task_lock); | |
2883 | ||
2884 | mlog_exit_void(); | |
2885 | } | |
2886 | ||
2887 | /* This aids in debugging situations where a bad LVB might be involved. */ | |
2888 | void ocfs2_dump_meta_lvb_info(u64 level, | |
2889 | const char *function, | |
2890 | unsigned int line, | |
2891 | struct ocfs2_lock_res *lockres) | |
2892 | { | |
2893 | struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; | |
2894 | ||
2895 | mlog(level, "LVB information for %s (called from %s:%u):\n", | |
2896 | lockres->l_name, function, line); | |
f9e2d82e MF |
2897 | mlog(level, "version: %u, clusters: %u, generation: 0x%x\n", |
2898 | lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters), | |
2899 | be32_to_cpu(lvb->lvb_igeneration)); | |
b0697053 MF |
2900 | mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n", |
2901 | (unsigned long long)be64_to_cpu(lvb->lvb_isize), | |
2902 | be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid), | |
2903 | be16_to_cpu(lvb->lvb_imode)); | |
2904 | mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, " | |
ca4d147e | 2905 | "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink), |
b0697053 MF |
2906 | (long long)be64_to_cpu(lvb->lvb_iatime_packed), |
2907 | (long long)be64_to_cpu(lvb->lvb_ictime_packed), | |
ca4d147e HP |
2908 | (long long)be64_to_cpu(lvb->lvb_imtime_packed), |
2909 | be32_to_cpu(lvb->lvb_iattr)); | |
ccd979bd | 2910 | } |